text
stringlengths 4
2.78M
| meta
dict |
|---|---|
---
abstract: 'We introduce a data-driven method and shows its skills for spatiotemporal prediction of high-dimensional chaotic dynamics and turbulence. The method is based on a finite-dimensional approximation of the Koopman operator where the observables are vector-valued and delay-embedded, and the nonlinearities are treated as external forcings. The predictive capabilities of the method are demonstrated for well-known prototypes of chaos such as the Kuramoto-Sivashinsky equation and Lorenz-96 system, for which the data-driven predictions are accurate for several Lyapunov timescales. Similar performance is seen for two-dimensional lid-driven cavity flows at high Reynolds numbers.'
author:
- 'M. A. Khodkar$^1$'
- 'Pedram Hassanzadeh$^{1,2}$'
- 'Athanasios Antoulas$^{3, 4, 5}$'
bibliography:
- 'Main.bib'
title: 'A Koopman-based framework for forecasting the spatiotemporal evolution of chaotic dynamics with nonlinearities modeled as exogenous forcings'
---
[^1]
[^2]
[^1]: [email protected]
[^2]: [email protected]
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'A numerical investigation of a non-commutative field theory defined via the spectral action principle is conducted. The construction of this triple relies on an 8-dimensional Clifford algebra. Following to the standard procedure of non-commutative geometry, the spectral action is computed for the product of the triple $(\mathcal{A}_4,\mathcal{H}_4,\mathcal{D}_4)$ with a matrix-valued spectral triple. Using Monte Carlo simulation we study various quantities such as the energy density, the specific heat density and some order parameters varying the matrix size and the independent parameters of the model.'
---
\
[**First numerical approach to a Grosse-Wulkenhaar model** ]{}
0.5 cm
[ Bernardino Spisso]{}\
*Mathematisches Institut der Westfälischen Wilhelms-Universität*\
*Einsteinstraße 62, D-48149 Münster, Germany*\
*e-mail:[email protected]*
1 cm
Introduction
=============
The main object of this work is a particular non-commutative field theory which is derived using the spectral action principle and then treated numerically. Non-commutativity can be found in many fields of physics like quantum field theories, string theory, condensed matter physics. The first application of non-commutativity into physics is dated from the middle of the last century inspired by the ideas of quantum mechanics, where starting from classical mechanics, the commutative algebra of functions on the phase space is replaced by a non-commutative operator algebra on a Hilbert space. The duality between ordinary spaces $M$ and proper commutative algebras is expressed by the Gel’fand-Naimark theorem which states the fact that the algebra of all continuous functions on $M$ is the only possible type of commutative $C^*$-algebra. Additionally, given a commutative $C^*$-algebra $C$, it is possible to reconstruct a Hausdorff topological space $M$ in order to obtain that $C$ is the algebra of continuous functions on $M$. The study of commutative $C^*$-algebras is equivalent to the study of topological Hausdorff spaces. The previous duality has inspired the identification, in non-commutative geometry, of some algebraical objects as a category of non-commutative topological spaces. Alain Connes [@Connes], one of the founders of non-commutative geometry, has proposed a candidate for the objects of such category, the spectral triples $(\mathcal{A},\mathcal{H}, \mathcal{D})$ [@Connes-1], composed by an algebra $\mathcal{A}$, an Hilbert space $\mathcal{H}$ on which $\mathcal{A}$ is represented and an selfadjoint operator $\mathcal{D}$. In fact, every compact oriented Riemannian manifold can be used to define a spectral triple, this kind of manifold $M$ characterizes a Dirac operator on self-adjoint Clifford module bundles over $M$. Connes, after a conjecture in 1996 [@Connes3] and some considerable attempts of Rennie and Varilly [@Rennie-vari], proved the so called reconstruction theorem [@ConnesRec] for commutative spectral triples satisfying various axioms, showing that exists a compact oriented smooth manifold $X$ such that $A = C^\infty(X)$ is the algebra of smooth functions on $X$ and every compact oriented smooth manifold emerges in this way. Pushed by the aim of reformulating the standard model of particles in a non-commutative way [@Connes2; @Connes3], Connes has introduced the almost-commutative spectral triple extending the axioms of the reconstruction theorem to a non-commutative algebra. A first attempt to formulate a field theory for a truly non-commutative algebra was obtained replacing in the usual field theory action the point-wise multiplication of the fields with a non-commutative one, namely a $\star$-product. The fields now belongs to $\mathbb{R}^4_\Theta$, a vector space defined by an enough regular class functions on $\mathbb{R}^4_\Theta$ equipped with the Moyal product: $$(f \star g)(x) = \int \int d^4 y \frac{d^4 k}{(2\pi)^4} f(x+\frac{1}{2}\Theta \cdot k) g(x+y) e^{i\langle k,y\rangle}$$ Where $\Theta$ is a skew-symmetric matrix. A very important question about non-commutative quantum field theory [@ncft3], is whether or not the quantum theory is well-defined or in other words if it is renormalizable or not. At first sight, the non-locality of the non-commutate action induced by the $\star$-product in the position space can induce us to fear some problems for the renormalization. In fact it was discovered [@ncft3] that after computing the Feynman rules for such theory and deriving the loop amplitude we find that the non-local interaction terms in the action induce an oscillatory factors (involving loop momenta) in the Feynman integrals. Studying the structure of Feynman diagrams for the action, Filk [@one-loop-nc3] has showed that are present two types of loop diagram the planar and non-planar diagrams. The planar diagrams do not have this oscillatory factors coming from the non-local interaction terms, and therefore the corresponding integrals are the same as in usual quantum field theory. On the other hand, all non-planar diagrams have the oscillatory factors involving loop momenta. Due to this terms the renormalization of quantum field theories on the non-commutative $\mathbb{R}^n$ is not achieved and these models show a phenomenon called UV/IR-mixing [@UV-mix]. Chepelev and Roiban [@UV-mix1] analyses UV/IR-mixing to all orders, the conclusion of the power-counting theorem is that field theories on non-commutative $\mathbb{R}^n$ are not renormalizable if the divergence of their commutative counterparts are higher than logarithmic.
A great step towards the non-commutative field theory was made when H.Grosse and R.Wulkenhaar [@phi4-non], found a non-commutative $\varphi^4$-theory renormalizable action which develops additional marginal coupling, corresponding to an harmonic oscillator potential for the real-valued free field $\varphi$ on $\mathbb{R}^4_\Theta$ : $$S[\varphi]=\int d^4x \left(\frac{1}{2} \varphi \star (-\Delta+\Omega^2 \tilde{x}^2 + \mu^2)\star\varphi + \frac{\lambda}{4}\varphi \star \varphi \star \varphi \star \varphi\right)(x) \nonumber$$ Where $x = 2\Theta^{-1} \cdot x, \ \lambda \in \mathbb{R} \ \Omega \in [0,1]$, and $\mu$ is a real parameter. Using the Moyal matrix base, which turns the $\star$-product into a standard (infinite) matrix product, H.Grosse and R.Wulkenhaar were able to prove the perturbative renormalizability of the theory [@matrix-renorm]. Afterward, R.Wulkenhaar et al. [@nc-renorm] found an alternative simpler normalization proof using multi-scale analysis in matrix base, showing the equivalence of various renormalization schemes. A last, but useful, renormalization proof was formulated using Symanzik type hyperbolic polynomials [@nc-renorm2].
The non-commutative model treated in this work is a sort of extension, via spectral action principle, of the scalar W-G model, in which we are interested to formulate a Yang-Mills theory in renormalizable way on Moyal space. We can expect that usual Yang-Mills theory on Moyal space without modifications of the action by something similar to an oscillator potential, to be not renormalizable [@UV-mix]. Additionally, the Moyal space with usual Dirac operator is a spectral triple, the corresponding spectral action was computed in [@nc-S], with the result that it is the usual not renormalizable action on Moyal plane. In [@8-dim] H.Grosse and R.Wulkenhaar, in order to obtain a gauge theory with an oscillator potential via the spectral action principle, used a Dirac operator constructed using the statement $ \mathcal{D}^2= H$ where the four dimensional Laplacian is substituted by the four dimensional oscillator Hamiltonian $H = -\Delta+\Omega^2 ||x||^2$. The idea behind is that the spectral dimension is defined through the Dirac operator so the spectral dimension defined by such Dirac operator is related to the harmonic oscillator phase space dimension. It turns out that to write down an Dirac operator, so that its square equals the 4D harmonic oscillator Hamiltonian, is an easy task using eight dimension Clifford algebra. In addition, can be shown that using this Dirac operator on 4D-Moyal space, is possible define an eight-dimensional spectral triple. After defined the Dirac operator with the desired spectrum it is considered the total spectral triple as the tensor product of the “oscillating” spectral triple $(\mathcal{A}_4, \mathcal{H}_4, \mathcal{D}_4 )$ with an almost-commutative triple and then is perform the previous described procedure of non-commutative geometry to compute the spectral action. We notice that matrix algebra introduces an extension of the standard potential in the commutative case, in fact the scalar field $\phi$ and the $X_\mu$ fields are present together in a potential of the form[^1] $(\alpha X_\mu \star X^\mu + \beta \bar{\varphi} \star \varphi - 1)^2$, with $\alpha, \beta \in \mathbb{R}^+$ and $X_\mu(x) = (\Theta^{-1} )_{\mu}^\nu x_\nu + A_\mu(x)$ is a covariant coordinate.
The high non-triviality of the vacuum makes very difficult to explicit the vacuum configuration of the system in [@Goursac1] A. de Goursac, J.C. Wallet, and R. Wulkenhaar, using the matrix base formalism, have found an expressions from vacuum solutions deriving them from the relevant solutions of the equations of motion. Although, the complexity of the vacuum configuration makes the perturbative approach very complicated, in order to conduct some investigations will be considered a non-perturbative scheme using a discretized matrix model of the action in which the fields become matrices, the star product become the matrix multiplication and the integral turns in a matrix trace.
Now comes in to play the numerical treatment, the standard method is to approximate the space by discrete points, for example using a lattice approximation and then calculate the observables over that set of points [@lattice]. Since an approximation in the position space is not suitable due to the oscillator factor of the Moyal product, instead the lattice approximation, will be used the matrix Moyal base, which was already used in the first renormalization proof of $\varphi^4$-model restricted to finite matrices. Hence, will be performed a Monte Carlo simulation studying some statistical quantity such the energy density and specific heat varying the parameters $\Omega, \frac{\chi_{-1}}{\chi_0}, \alpha$ and gathering some informations on the various contributions of the fields to the action. The simulations are quite cumbersome due the complexity of the action and the number of independent matrices to handle but we are able to get an acceptable balance between the computation precision and the computation time. For the simulations is applied a standard Metropolis-Monte Carlo algorithm [@Metro] with various estimators for the error and for the autocorrelation time of the samples. In general we chose the range of parameters in order to avoid problems with the thermalization process, obtaining numerical simulations where is enough to wait a relative small number of Monte Carlo steps to compute independent results from the initial conditions. we are interested on the continuous limit that correspond to matrices of infinite size. We will consider various size of the matrices expecting a stabilization of the values of observables like the energy density, increasing the matrix size. In order to find same possible phase transitions will be used the specific heat which is a measure of the dispersion of the energy. The phase transitions are registered as peaks of the specific heat, increasing the matrices size.
8-dim spectral action
======================
In this section will be computed a spectral action starting from a non-commutative spectral triple. The feature of this particular triple is the choice of a 4-dimension Harmonic Dirac operator. The idea behind this construction [@8-dim] is to relate the Dirac operator with the oscillator Hamiltonian operator. Roughly speaking, we look at the Dirac operator as a generalization of the Laplace operator so we have $\mathcal{D}^2\approx H$. Considering the spectrum $\omega(2n+1)$ $n \in N$ of the one-dimensional harmonic oscillator Hamiltonian $H$, can be deduced that $H^{-1}$ is a non-commutative infinitesimal of order one. The non-commutative dimension of a spectral triple, equipped with the 4D Harmonic Dirac operator $ \mathcal{D}^2_4= H_4$, is fixed by the non-commutative order of the inverse operator $\mathcal{D}^{-1}$ which is eight not four. This occurrence connects the spectral dimension to the phase space dimension instead the one of the configuration space [@con-phase]. In order to construct such Harmonic Dirac operator and the spectral triple we will work in the framework of the generalized $n$-dimensional harmonic operators. Will be studied the 4-dimensional case in order to construct the non-commutative spectral triple which is starting point for the field theory we are interested in. Having the 4-dimensional Harmonic Dirac operator with harmonic oscillator spectrum, to implement the Higgs mechanism we will consider the tensor product of the non-commutative triple with a finite Connes-Lott type spectral triple [@con-lot]. We will fluctuate the total Dirac operator following the standard machinery [@S-A; @Connes3] of non-commutative geometry to get “Gauged” Dirac operator. Thus we will proceed to compute the spectral action in which are present two U(1)-Moyal Yang-Mills fields unified with a complex Higgs field.
Harmonic Dirac operators
=========================
**1**
The Harmonic Dirac operator in $n$-dimensions can be defined using the Clifford algebra of $\mathbb{R}^{2n}$ represented on the Hilbert space $ \mathbb{C}^{2^n}$, it is very useful to consider $n$-dimensional fermionic annihilation and creation operators $b_\mu$, $b_\nu^\dagger$ and $n$-dimensional bosonic annihilation and creation operators $a_\mu$, $a^\dagger_\nu$ satisfying for $\mu,\nu=1,\cdots,n$: $$\begin{aligned}
&[a_\mu,a_\nu]=[a^\dagger_\mu,a^\dagger_\nu]=0, & [a_\mu,a^\dagger_\nu]=\delta_{\mu\nu} \label{com1} \\
&\{b_\mu,b_\nu\}=\{b_\mu^\dagger ,b_\nu^\dagger\}=0, & \{b_\mu,b_\nu^\dagger\}=\delta_{\mu\nu} \label{anti1}\end{aligned}$$ Where $a_\mu =\frac{1}{\sqrt{2\omega}} (\omega x_\mu + \partial_\mu ), \ a_\mu^\dagger = \frac{1}{\sqrt{2\omega}}(\omega x_\mu-\partial_\mu ) $. Using this operators is possible to construct a Dirac operator as: $$\mathcal{D}_n=-i\sqrt{2\omega}\delta^{\mu\nu}a_\mu^\dagger\otimes b_\nu + i\sqrt{2\omega}\delta^{\mu\nu}a_\mu\otimes b_\nu^\dagger=i\frac{d}{dx_\mu}\otimes(b_\mu+b_\mu^\dagger)+ i\omega x^\mu\otimes(b_\mu-b^\dagger_\mu) \label{dirac2}$$ summed over repeated index. We can define the fermionic part of the Hilbert space on which the Dirac operator acts starting from the vacuum state by subsequent applications of the fermionic creation operators $b_\nu^\dagger$ on the vacuum $b|0\rangle=0$, using the anti-commutation relations defining $\Lambda(\mathbb{C}^n)$. The complete Hilbert space is $\mathcal{H}_n=S(\mathbb{R}^n)\otimes\Lambda(\mathbb{C}^n)$. Beside, we can define a grading operator $\chi_n $ as: $$\begin{aligned}
\chi_n= \textbf{I}\otimes \prod^n_{\mu=1}(b_\mu b^\dagger_\mu-b_\mu b^\dagger_\mu )\end{aligned}$$ Using the relations - we can compute the square the Dirac operator as: $$\mathcal{D}_n^2=2\omega a^\dagger_\mu a^\mu\otimes \textbf{I} -2\omega \textbf{I} \otimes b^\dagger_\mu b^\mu =2\omega N_B\otimes \textbf{I} -2\omega \textbf{I}\otimes N_F$$ Where $N_F$ and $N_B$ are the number operators. In this form it easy to see that $\mathcal{D}^2_n $, being a “difference” between fermionic and bosonic number operator, has only one zero mode corresponding to the vacuum state. For practical reasons it is convenient write $\mathcal{D}^2_n $ as: $$\mathcal{D}_n^2=\omega \delta^{\mu\nu}(a^\dagger_\mu a_\nu + a_\nu a^\dagger_\mu)\otimes \textbf{I} -2\omega \textbf{I} \otimes \delta^{\mu\nu}(b^\dagger_\mu b_\nu - b_\nu b^\dagger_\mu)=H_n\otimes \textbf{I} +\omega\otimes \Sigma_n$$ where in $H_n$ we can recognize the harmonic oscillator Hamiltonian and the spin operator $\Sigma_n$. The universality property of the Clifford algebra grants the existence of an isomorphism between the 2$n$-dimensional Clifford algebra and the Hilbert space $ \mathcal{H}_n=S(\mathbb{R}^n)\otimes \mathbb{C}^{2^n}$. In this representation the Dirac operator is: $$\mathcal{D}_n=i\Gamma^\mu\partial_\mu +\omega\Gamma^{\mu+n}x_\mu \label{Diracn}$$ Where $\Gamma^\mu$ turns to be $\Gamma^\mu=(b_\mu+b^\dagger_\mu)$, $ \Gamma^{\mu+n}=i(b^\dagger_\mu-b_\mu)$ which satisfy the relations: $$\Gamma_a\Gamma_b+\Gamma_b\Gamma_a=2\delta_{ab} \ \textrm{with} \ a,b=1,\cdots,2n \label{Gamman}$$ Beside, the grading operator is represented as: $$\chi_n=(-i)^n(-1)^{\frac{n(n-1)}{2}}\otimes \Gamma_1\cdots\Gamma_{2n}$$
An harmonic spectral triple for the Moyal plane
===============================================
In the framework of non-commutative field theories on 4-dimensional Moyal plane has been proved [@phi4-non; @matrix-renorm] that the introduction of an harmonic oscillator term makes a $\phi^4$-model on 4-dimensional Moyal plane renormalizable. Such oscillator term can be written as: $$H_m=-\frac{\partial^2}{\partial x_\mu \partial x^\mu}+\Omega^2 \tilde{x}_\mu\tilde{x}^\mu +m^2$$ where $\tilde{x}_\mu:=2(\Theta^{-1})_{\mu\nu} x^\nu$, $\Theta$ can be chosen as two copies of the Pauli matrix $\Theta = i\theta\sigma_2 \otimes \textbf{I}_2$ or explicitly: $$\Theta=\left(
\begin{array}{cccc}
0& \theta &0&0 \\
-\theta &0&0&0 \\
0&0&0&\theta \\
0&0&-\theta&0
\end{array}
\right), \ \theta\in\mathbb{R}$$ With this choice we have $\Theta^{-1} = -\frac{i\sigma_2 }{\theta}\otimes \textbf{I}_2 $. Quantum mechanics tell us that in the Hilbert space $L^2(\mathbb{R}^4)$ exists an orthonormal basis $\psi_s ,\ s\in\mathbb{R}^4$ of eigenfunctions of $H_m$ with eigenvalues $$\lambda_s(m)=\left(\frac{4\Omega}{\theta} \left(s+2+ \frac{\theta m^2}{2\Omega}\right)\right), \ s\in\mathbb{N}$$ The inverse $H^{-1}_m$ extends to a selfadjoint compact operator on $L^2(\mathbb{R}^4)$ with eigenvalues $\lambda^{-1}_s(m)$. If we look at the trace the operator $H^{-4}_m$ we find: $$\operatorname{Tr}(H^{-s}_m)=\sum_{n=0}^\infty(n + 3)(n + 2)(n + 1)(\lambda_n (m))^s$$ Which is derived simply from the number of possibilities to express $s$ as a sum of four ordered natural numbers. This means that $H^{-4}$ belongs to the Dixmier trace ideal $L^{(1,\infty)}(L^2(\mathbb{R}^4))$ of compact operators and the relation $\mathcal{D}^2=H$ implies that the 4-dimensional Moyal space has spectral dimension 8. From the previous section, we can define a proper Dirac operator just considering the 4-dimensional case obtaining a Dirac operator built from a 8-dimensional Cifford algebra: $$\mathcal{D}_4 = i\Gamma_\mu \partial_\mu + \Omega \Gamma_{\mu+4} \tilde{x}_\mu$$ Here, the $\Gamma_k \in M_{16}(\mathbb{C}), k = 1, . . . , 8$ are the generators of the 8-dimensional real Clifford algebra, satisfying $$\Gamma_k \Gamma_l + \Gamma_l\Gamma_k = 2\delta_{kl} \textbf{I}$$ We take the Hilbert space $\mathcal{H}_4 = S^2(\mathbb{R}^4,\mathcal{S})=L^2(\mathbb{R}^4)\otimes\mathbb{C}^{16} $ of Schwartz functions of spinors over 4-dimensional euclidean space. Accordingly with for $\psi \in \mathcal{H}_4$ we obtain: $$\mathcal{D}^2_4 \psi= \left((-\Delta + \Omega^2 \tilde{x}_\mu \tilde{x}^\mu )\textbf{I} + \Sigma\right)\psi \ , \ \Sigma:= -i\Omega(\Theta^{-1})^{\mu\nu} [\Gamma_\mu ,\Gamma_{\nu+4}] \label{Sigma1}$$ with $\Delta = \partial^\mu \partial_\mu$. As algebra we chose the Moyal algebra $\mathbb{R}^4_\Theta$: $$\mathcal{A}_4 = \mathbb{R}_\Theta^4=(\mathcal{S}(\mathbb{R}^4),\star)$$ where $(\mathcal{S}(\mathbb{R}^4),\star)$ is the algebra of the Schwartz functions on $\mathbb{R}^4$, with the Moyal product $$(f \star g)(x) = \int \int d^4 y \frac{d^4 k}{(2\pi)^4} f(x+\frac{1}{2}\Theta \cdot k) g(x+y) e^{i\langle k,y\rangle} \ ,\ f,g \in \mathcal{A}_4 \label{star}$$ The representation of the algebra $\mathcal{A}_4$ on $\mathcal{H}_4$ is by component-wise diagonal Moyal product [@moyal-triple] $\star : \mathcal{A}_4 \times \mathcal{H}_4 \to \mathcal{H}_4$. The Moyal product can be extended to constant functions using another representation of the product with the integral representation of the Dirac distribution. Taking in account, for smooth spinors, the identity $2x^\mu\psi=x\star\psi+\psi\star x$ and the relation $$[x^\nu,f]_\star = i\Theta^{\nu\rho} \partial_\rho \label{comxf}$$ we compute the commutator of that action with the Dirac operator $$\begin{array}{l}
\mathcal{D}_4(f\star\psi)-f\star(\mathcal{D}_4 \psi) \\
= i\Gamma^\mu ((\partial_\mu f) \star\psi + f\star \partial_\mu \psi) + \frac{1}{2} \Omega\Gamma^{\mu+4} (\tilde{x}_\mu\star(f \star \psi) + (f \star \psi) \star \tilde{x}_\mu) \\
-i\Gamma^\mu f \star \partial_\mu \psi - \frac{1}{2} \Omega\Gamma^{\mu+4}(f\star( \tilde{x}_\mu\star\psi)+f\star(\psi \star \tilde{x}_\mu))\\
= i(\Gamma^\mu + \Omega\Gamma^{\mu+4} )(\partial_\mu f ) \star \psi .
\end{array} \label{d4-com}$$ The previous commutator confirms that $(\mathcal{A}_4,\mathcal{H}_4,\mathcal{D}_4)$ satisfy the main[^2] axioms of spectral triple, in fact the commutator is bounded and due to its commutation with Moyal multiplication, order-one condition is fulfilled. Now we introduce a very useful relation connected to the heat kernel type expansion associated to a regular spectral triple taken from [@non-com-tri]. This relation will be used later in order to compute the spectral action. Considering a regular non-unital spectral triple $(\mathcal{A},\mathcal{H},\mathcal{D},J)$ and two pseudo-differential operator $A_0 \in \Psi_0(\mathcal{A}) $ $A_1\in \Psi_1(\mathcal{A})$ of order respectively 0 and 1. We consider the following decomposition: $$e^{-t(\mathcal{D}^2+A_0+A_1)}=\sum_{j=0}^4(-1)^jE_j(t)-t^5R$$ Using Duhamel principle [@Duhamel] $$e^{-t(A+B)}= e^{-tA}-t\int_0^1 ds \ e^{-st(A+B)}Be^{-(1-s)tA}$$ we can identify: $$\begin{aligned}
E_0(t)&=&e^{-t\mathcal{D}^2}\nonumber \\ \
E_j(t)&=&\int_{\Delta_j} d^js \ e^{-s_1t\mathcal{D}^2}(A_0+A_1)e^{-(s_2-s_1)t\mathcal{D}^2}\cdots(A_0+A_1)e^{-(1-s_j)t\mathcal{D}^2} \nonumber \\ \
&&\end{aligned}$$ and $$\begin{aligned}
&&R=\int_{\Delta_5}ds_1ds_2ds_3ds_4ds_5 e^{-s_1t(\mathcal{D}^2+A_0+A_1)}(A_0+A_1)e^{-(s_2-s_1)t\mathcal{D}^2}(A_0+A_1) \nonumber\\
&& \times e^{-(s_3-s_2)t\mathcal{D}^2}(A_0+A_1)e^{-(s_4-s_3)t\mathcal{D}^2}(A_0+A_1)e^{-(s_5-s_4)t\mathcal{D}^2}(A_0+A_1)e^{-(1-s_5)t\mathcal{D}^2}\nonumber\\
&&\end{aligned}$$ The domains of the integrals $\Delta_j$ are the $j$-simplex: $$\Delta_j=\{s\in\mathbb{R}^j;0\leq s_1\leq s_2\leq s_1\leq\cdots\leq s_j\leq 1\} \simeq \{s\in\mathbb{R}^{j+1};s_i\geq 0,\sum_{i=0}^j s_i=1\}$$ Taking in account the relation: $$\left[ e^{-t\mathcal{D}^2},A \right]=\int_0^1 ds \frac{d}{ds}\left(e^{-ts\mathcal{D}^2}Ae^{-t(1-s)\mathcal{D}^2}\right) =-t\int_0^1 ds e^{-ts\mathcal{D}^2}\left[\mathcal{D}^2,A\right]e^{-t(1-s)\mathcal{D}^2}$$
and considering the trace in [@non-com-tri] is computed the leading term of $e^{-t(\mathcal{D}^2+A_0+A_1)}$ for $t\to 0$: $$\begin{aligned}
&&\operatorname{Tr}(e^{-t(\mathcal{D}^2+A_0+A_1)}) \nonumber \\
&&=\textrm{Tr}\Big(\Big\{1-t(A_0+A_1)+\frac{t^2}{2}(A^2_0+A_1A_0+A_0A_1+A^2_1) \nonumber \\
&&-\frac{t^3}{6}(A_0\left[\mathcal{D}^2,A_1\right]+A_1\left[\mathcal{D}^2,A_0\right]+A_1\left[\mathcal{D}^2,A_1\right]+A_0A^2_1+A_1A_0A_1+A_1^2A_0\nonumber \\
&&+A^3_1)+\frac{t^4}{24}(A_1\left[\mathcal{D}^2 \left[\mathcal{D}^2,A_0\right]\right]+2A^2_1\left[\mathcal{D}^2 ,A_1\right]+A_1\left[\mathcal{D}^2 ,A_1\right]A_1+A_1^4\Big)e^{-t\mathcal{D}^2}\Big\}\nonumber \\
&&+\mathcal{O}(\sqrt{t}) \label{heat1}\end{aligned}$$
4-dimensional harmonic Yang-Mills model
=======================================
Following the Connes-Lott models, in order to implement the Higgs mechanism, we consider the total spectral triple as the tensor product of the 8-dimensional spectral triple $(\mathcal{A}_4,\mathcal{H}_4,\mathcal{D}_4,\Gamma_9)$ with the two point Connes-Lott like spectral triple $(\mathbb{C}\otimes \mathbb{C},\mathbb{C}^2, M\sigma_1 )$. The total Dirac operator of the product triple is: $$\mathcal{D}_T = \mathcal{D}_4 \otimes \textbf{I} + \Gamma_9 \otimes M\sigma_1$$ Or explicitly: $$\mathcal{D}_T= \left(\begin{array}{cc}
\mathcal{D}_4& M\Gamma_9 \\
M\Gamma_9& \mathcal{D}_4
\end{array}\right)$$ The algebra becomes $ \mathcal{A}_T=\mathcal{A}_4 \oplus \mathcal{A}_4 $ and acts by diagonal star multiplication on $\mathcal{H}_T= \mathcal{H}_4 \oplus \mathcal{H}_4$. The fluctuated Dirac operator is found using $\mathcal{D}_A = \mathcal{D}_T + \Sigma_i a_i[\mathcal{D}_T,b_i]$ with $a_i,b_i\in\mathcal{A}_T $ of the form $(f,g)$, the computation of the commutator $\mathcal{D}_T$ with $(f,g)$ gives: $$[\mathcal{D}_T,(f, g)] =
\left(\begin{array}{cc}
i(\Gamma^\mu + \Omega \Gamma^{\mu+4} )L_\star(\partial_\mu f) & M\Gamma_9 L_\star(f-g) \\
M\Gamma_9 L_\star(g-f) & i(\Gamma^\mu + \Omega\Gamma^{\mu+4} )L_\star(\partial_\mu g)
\end{array}\right)$$ $L_\star(f)\psi = f \star \psi$ is the left Moyal multiplication. From the commutator we deduce that the form of selfadjoint fluctuated Dirac has to be: $$\mathcal{D}_A =\left(\begin{array}{cc}
\mathcal{D}_4 + (\Gamma_\mu + \Omega \Gamma_{\mu+4})L_\star(A^\mu )& \Gamma_9 L_\star(\phi) \\
\Gamma_9 L_\star (\bar{\phi}) &\mathcal{D}_4 + (\Gamma_\mu + \Omega \Gamma_{\mu+4} )L_\star (B^\mu ) \end{array}\right)$$ Where $\phi \in \mathcal{A}_4$ is the Higgs complex field and $A_\mu, B_\mu \in \mathcal{A}_4$ are real fields. The spectral action computation needs the square of $\mathcal{D}_A$: $$\mathcal{D}^2_A =\left(\begin{array}{cc}
(H^2_0 + L_\star (\phi \star \varphi))1 + \Sigma + F_A
&i(\Gamma_\mu + \Omega\Gamma_{\mu+4})\Gamma_9 L_\star (D^\mu \phi)\\
i(\Gamma_\mu + \Omega \Gamma_{\mu+4} )\Gamma_9 L_\star (\overline{D^\mu \phi})&
(H^2 + L_\star (\phi \star \phi))1 + \Sigma + F_B
\end{array}\right)$$ with $$\begin{aligned}
D_\mu \phi &=& \partial_\mu\phi - iA_\mu \star \phi + i\phi\star B_\mu\\
F_A &=& \{\mathcal{D}_4 , (\Gamma_\mu + \Omega \Gamma_{\mu+4} )L_\star (A^\mu )\} + (\Gamma_\mu + \Omega\Gamma_{\mu+4} )(\Gamma_\nu + \Omega \Gamma_{\nu+4} )L_\star (A^\mu \star A^\nu)\nonumber \\
&=& \{ L_\star(A^\mu ), i\partial_\mu + \Omega^2 M_\bullet(x_\mu)\} + (1 +\Omega^2)L_\star (A_\mu \star A^\mu ) \nonumber \\
&+& i \left(\frac{1}{4} [\Gamma_\mu, \Gamma_\nu]+ \frac{1}{4} \Omega^2 [\Gamma_{\mu+4} , \Gamma_{\nu+4} ] + \Omega\Gamma_\mu \Gamma_{\nu+4} \right) L_\star (F^{\mu\nu}_A ) ,\end{aligned}$$ $(M_\bullet(\tilde{x}_\mu)\psi)(x) = \tilde{x}_\mu\psi(x)$ is ordinary pointwise multiplication and $F_B$ is obtained just replacing $A$ with $B$. We can recognize in previous expression the field strength $F^A_{\mu\nu} = \partial_\mu A_\nu - \partial_\nu A_\mu -i(A_\mu\star A_\nu - A_\nu\star A_\mu )$
Spectral action
---------------
Recalling the spectral action principle, the bosonic action can be defined exclusively by the spectrum of the Dirac operator. The general form for such bosonic action is: $$S(\mathcal{D}_A) = \textrm{Tr}\chi(\mathcal{D}^2_A) \label{Action}$$ Where $\chi$ is a regularization function $\chi: R_+\to R_+$ for which trace exists.\
The trace in is defined on $\mathcal{B}(L^2(\mathbb{R}^4))$ by $$\operatorname{Tr}(A) =\int_{\mathbb{R}^4} dx \ A(x,x)$$ together with the matrix trace including the Clifford algebra. By Laplace transformation one has $$S(\mathcal{D}_A) =\int_0^\infty dt \operatorname{Tr}(e^{-t\mathcal{D}^2_A})\tilde{\chi}(t) \label{Action1}$$ where $\tilde{\chi}$ is the inverse Laplace transform of $\chi(s)$, $$\chi(s) = \int_0^\infty dt e^{-st}\tilde{\chi}(t).$$ The trace in is given by: $$\operatorname{Tr}(e^{-t\mathcal{D}^2_A}) =\int_{\mathbb{R}^4} dx \ \textrm{tr}(e^{-t\mathcal{D}^2_A})(x,x)$$ Assuming the trace of the heat kernel $e^{-t\mathcal{D}^2_A}$ has an asymptotic expansion $$\operatorname{Tr}(e^{-t\mathcal{D}^2_A})= \sum^\infty_{n=-\delta} a_n(\mathcal{D}^2_A)t^{n} \ , \ \delta \in \mathbb{N} \label{heat}$$ we obtain replacing the previous expansion into $$S(\mathcal{D}_A) =\sum_{n=-\delta}^\infty a_n(\mathcal{D}^2_A)\int_0^\infty dt \ t^{n} \tilde{\chi}(t) \label{heatd}$$ To compute the integrals we have to consider separately the cases $ n\notin\mathbb{N}$ and $ n\in\mathbb{N}$: $$\chi_n =\Bigg\{
\begin{array}{ll}
\frac{1}{\Gamma(-n)}\int_0^\infty ds \ s^{-n-1} \chi(s) & \textrm{for} \ n\notin\mathbb{N} \\
(-1)^{n-\delta} \chi^{(n)}(0) & \textrm{for} \ n\in\mathbb{N}
\end{array}\label{chi}$$ Due to the nature of the $\chi(t)$ function (usually one chose a characteristic function), we can assume $\chi(0)$ much bigger then the derivatives $\chi^{(m)}(0)$ for any $m>0$ appearing in . Consequently in the expansion we will take in account only the finite or singular part for $t\to 0$
Our strategy to compute the action is to use the relation , therefore after explicitly expressed $A_0$ and $A_1$ we proceed to the calculus of the traces and in the end we will identify the leading part of the action comparing the result with the expansions -. We can identify the operators $A_0$ and $A_1$ appearing in the as follow: $$A_0 =\left(\begin{array}{cc}
L_\star (V_{A,\phi})\textbf{I} + L_\star(F_A^{\mu\nu})\Gamma^\Omega_{\mu\nu}
&i(\Gamma^\mu + \Omega\Gamma^{\mu+4})\Gamma_9 L_\star (D_\mu \phi)\\
i(\Gamma^\mu + \Omega \Gamma^{\mu+4} )\Gamma_9 L_\star (\overline{D_\mu \phi})&
L_\star (V_{B,\phi})\textbf{I} + L_\star(F_B^{\mu\nu})\Gamma^\Omega_{\mu\nu}
\end{array}\right) \label{A0}$$ $$A_1 =\left(\begin{array}{cc}
2i(1+\Omega^2)L_\star(A^\mu)\nabla_\mu^{(\Omega)} &0\\
0& 2i(1+\Omega^2)L_\star(B^\mu)\nabla_\mu^{(\Omega)}
\end{array}\right)\label{A1}$$ with $$V_{A,\phi}= \phi\star\bar{\phi}+(1+\Omega^2)(i \partial_\mu A^\mu +A_\mu\star A^\mu) , \ V_{B,\phi}= \bar{\phi}\star\phi+(1+\Omega^2)(i\partial_\mu B^\mu+B_\mu\star A^\mu)$$ $\nabla_\mu^{(\Omega)}$ are define as $\nabla_\mu^{(\Omega)}=\frac{1}{1+\Omega^2}\left(\partial_\mu-i\Omega^2 M_\bullet(\tilde{x}_\mu)\right) $. We are allowed to split the traces in two parts a matrix trace and the continuous one. After the matrices trace computations we obtain [@non-com-tri] for the $A$ field: $$\begin{aligned}
&&\textrm{Tr}(e^{-t\mathcal{D}_A^2})\nonumber \\
&&=\Bigg\{16\cosh^4(t\Omega)\textrm{tr}(e^{-tH_4^2})-tT(16V_{A,\phi})-t\mathcal{T}_\mu(32i(1+\Omega^2)A^\mu)\nonumber \\
&&+\frac{t^2}{2}\mathcal{T}\Big(16V_{A,\phi}\star V_{A,\phi}+16(1+\Omega^2)D_\mu\phi\star\overline{D^\mu\phi}+8(1+\Omega^2)F_{\mu\nu}^A F^{\mu\nu}_A\nonumber \\
&&+ 32i(1+\Omega^2)A^\mu\star\partial_\mu V_{A,\phi}\Big)\nonumber \\
&&+ \frac{t^2}{2}\mathcal{T}_\mu\Big(32i(1+\Omega^2)A^\mu\star V_{A,\phi}+32i(1+\Omega^2)V_{A,\phi}\star A^\mu\nonumber \\
&&-64(1+\Omega^2)^2A^\nu\star\partial_\nu A^\mu\Big)+\frac{t^2}{2}\mathcal{T}_{\mu\nu}\left(-64(1+\Omega^2)^2 A^\mu\star A^\nu \right)\nonumber \\
&&- \frac{t^3}{6}\mathcal{T}_{\mu\nu}\Big(-64(1+\Omega^2)^2V_{A,\phi}\star \partial^\mu A^\nu -64i(1+\Omega^2)^2A^\nu\star\partial^\nu V_{A,\phi}\nonumber \\
&&+64(1+\Omega^2)^3A_\rho\star(\delta^{\rho\mu}\triangle A^\nu +2\partial^\rho\partial^\mu A^\nu)\nonumber \\
&&- 64(1+\Omega^2)^2(V_{A,\phi}\star A^\mu\star A^\nu + A^\mu\star V_{A,\phi}\star A^\nu +A^\mu\star A^\nu\star V_{A,\phi})\nonumber \\
&&-128i(1+\Omega^2)^3\left(A^\rho\star(\partial_\rho A^\mu)\star A^\nu +A^\rho\star A^\mu(\partial_\rho A^\nu\right)\Big)\nonumber \\
&&- \frac{t^3}{6}\mathcal{T}_{\mu\nu\rho}\left(128(1+\Omega^2)^3A^\mu\star\partial^\nu A^\rho -128i(1+\Omega^2)^3A^\nu\star A^\mu\star A^\rho\right)\nonumber \\
&&+ \frac{t^3}{6}\tilde{\mathcal{T}}_{\mu\nu}\left(512i\Omega^2(1+\Omega^2)(\Theta^{-1})^{\rho\nu}A^\mu\star A_\rho\right)\nonumber \\
&&+ \frac{t^4}{24}\mathcal{T}_{\mu\nu\rho\sigma}\Big(-256(1+\Omega^2)^4A^\mu\star\partial^\nu\partial^\rho A^\sigma +512i(1+\Omega^2)^3 A^\mu\star A^\nu\star \partial^\rho A^\sigma\nonumber \\
&&+256i(1+\Omega^2)^4 A^\mu\star(\partial^\nu A^\rho)\star A^\sigma +256(1+\Omega^2)^4 A^\mu\star A^\nu\star A^\rho\star A^\sigma\Big)\Bigg\}\nonumber \\
&& + \mbox{$B$ field contribution} \ +\mathcal{O}(\sqrt{t})\end{aligned}$$ Where in order to simplify the notation we introduce the functions: $$\begin{aligned}
\mathcal{T}(f)&=&\textrm{Tr}_{L^2(\mathbb{R}^4)}\left(L_\star(f)e^{-tH_4}\right) \nonumber \\
\mathcal{T}_{\mu_1\cdots\mu_k}(f)&=&\textrm{Tr}_{L^2(\mathbb{R}^4)}\left(L_\star(f)\nabla^{(\Omega)}_{\mu_1}\cdots\nabla^{(\Omega)}_{\mu_k}e^{-tH_4}\right) \nonumber \\
\tilde{\mathcal{T}}_{\mu\nu}(f)&=&\textrm{Tr}_{L^2(\mathbb{R}^4)}\left(L_\star(f)\nabla^{(\Omega)}_{\mu_1}\nabla^{(1)}_{\mu_k}e^{-tH_4}\right) \end{aligned}$$
The contributions for the $B$ fields are obtained operating the following substitutions: $$\left\{ A_\mu \to B_\mu, \ F^A_{\mu\nu}\to F^B_{\mu\nu}, \ V_{A,\phi} \to V_{B,\phi}, \ D_\mu\phi \leftrightarrow \overline{D^\mu\phi}\right\}$$ After the computation of the $\mathcal{T},\mathcal{T}_{\mu_1\cdots\mu_k}$ [^3], we have all the ingredients required to compute the leading part of the action replacing all the traces into the . Using the trace property of the star product and the identities $$\begin{aligned}
\tilde{z}_\mu\star f=\frac{1}{2}\{\tilde{z}_\mu,f\} +\frac{1}{2}[\tilde{z}_\mu,f]=\frac{1}{2}\{\tilde{z}_\mu,f\}+i\partial_\mu f \\
\{\tilde{z}_\mu , f\star g \}_\star=\{\tilde{z}_\mu,f \}\star g -if\star \partial_\mu g \end{aligned}$$ we get after some manipulations: $$\begin{aligned}
&&\textrm{Tr}(e^{-t\mathcal{D}^2})=2 \cosh^4(\tilde{\Omega}t) \nonumber \\
&&+\frac{1}{\pi^2(1+\Omega^2)^2} \int d^4z \Bigg\{ -\frac{1}{t}\Bigg(\left(\phi\star\bar{\phi}+ \frac{4\Omega^2}{1+\Omega^2}(\tilde{X}^A_\mu\star\tilde{X}_A^\mu -\tilde{X}^0_{\mu}\star\tilde{X}_0^\mu)\right)\nonumber \\
&&+\left(\bar{\phi}\star\phi+\frac{4\Omega^2}{1+\Omega^2}(\tilde{X}^B_\mu\star\tilde{X}_B^\mu -\tilde{X}^0_{\mu}\star\tilde{X}_0^\mu)\right)\Bigg)\nonumber \\
&&+\frac{1}{2}\left( \phi\star\bar{\phi}+\frac{4\Omega^2}{1+\Omega^2}\tilde{X}^A_\mu\star\tilde{X}_A^\mu\right)^2-\frac{1}{2}\left(\frac{4\Omega^2}{1+\Omega^2}\tilde{X}^\mu_0\star\tilde{X}^0_\mu\right)^2\nonumber \\
&&+\frac{1}{2}\left( \bar{\phi}\star\phi+\frac{4\Omega^2}{1+\Omega^2}\tilde{X}^B_\mu\star\tilde{X}_B^\mu\right)^2-\frac{1}{2}\left(\frac{4\Omega^2}{1+\Omega^2}\tilde{X}^\mu_0\star\tilde{X}^0_\mu\right)^2\nonumber \\
&&+\frac{1}{2}\left(\frac{(1-\Omega^2)^2}{2}-\frac{(1-\Omega^2)^4}{6(1+\Omega^2)^2}\right)(F^A_{\mu\nu}\star F_A^{\mu\nu}+F^B_{\mu\nu}\star F_B^{\mu\nu})\nonumber \\
&&+\frac{1+\Omega^2}{2}\left(D_\mu\phi\star\overline{D^\mu\phi}+\overline{D_\mu\phi}\star D^\mu\phi\right)\Bigg\}(z)+\mathcal{O}(\sqrt{t})\end{aligned}$$ where $$\tilde{X}^A_\mu=\frac{\tilde{z}_\mu}{2}+A_\mu, \ \tilde{X}^B_\mu=\frac{\tilde{z}_\mu}{2}+B_\mu, \ \tilde{X}^0_\mu=\frac{\tilde{z}_\mu}{2}$$ Using the Laurent expansion of $ \coth^4(t^\prime)=t^{\prime-4}+ \frac{4}{3}t^{\prime-2}+\frac{26}{45}+\mathcal{O}(t^{\prime 2})$ and comparing the previous expression to the expansion and putting $\chi_0=\chi(0)$ we are finally able to write the spectral action as: $$\begin{aligned}
&&S(\mathcal{D}_A)= \frac{\theta^4\chi_{-4}}{8\Omega^4}+\frac{2\theta^2\chi_{-2}}{3\Omega^2}+\frac{52\chi_{0}}{45} \nonumber \\
&&+\frac{\chi_0}{\pi^2(1+\Omega^2)^2}+\frac{1}{\pi^2(1+\Omega^2)^2} \int d^4z \Bigg\{\left(\frac{(1-\Omega^2)^2}{2}-\frac{(1-\Omega^2)^4}{6(1+\Omega^2)^2}\right)(F^A_{\mu\nu}\star F_A^{\mu\nu}+F^B_{\mu\nu}\star F_B^{\mu\nu})\nonumber \\
&&+\left(\bar{\phi}\star\phi+ \frac{4\Omega^2}{1+\Omega^2}\tilde{X}^A_\mu\star\tilde{X}_A^\mu-\frac{\chi_{-1}}{\chi_0}\right)^2\nonumber \\
&&+\left(\phi\star\bar{\phi}+\frac{4\Omega^2}{1+\Omega^2}\tilde{X}^B_\mu\star\tilde{X}_B^\mu-\frac{\chi_{-1}}{\chi_0}\right)^2\nonumber \\
&&-2\left(\frac{4\Omega^2}{1+\Omega^2}\tilde{X}^\mu_0\star\tilde{X}^0_\mu-\frac{\chi_{-1}}{\chi_0} \right)^2+ 2(1+\Omega^2)D_\mu\phi\star\overline{D^\mu\phi}\Bigg\}(z)+\mathcal{O}(\chi_1)\label{faction}\end{aligned}$$ We notice that Higgs mechanism introduces an extension of the standard Higgs potential in the commutative case, in fact the Higgs scalar field $\phi$ and the $\tilde{X}_A^\mu$, $\tilde{X}_B^\mu$ fields are present together in the potential. In this way the gauge field takes part in the definition of the vacuum. Another important property of the action, considering the $\tilde{X}_A^\mu$, $\tilde{X}_B^\mu$ as independent, is the invariance under the translations: $$\phi(x)\to\phi(x+a),\ X^\mu_A(x)\to X_A^\mu(x+a), \ X^\mu_B(x)\to X^\mu_B(x+a) , \ X^\mu_0(x)\to X^\mu_0(x+a)$$ which in other $\phi^4$-renormalizable theory is broken. Beside, the action is invariant under $U(1) \times U(1)$ transformations: $$\phi \to u_A\star\phi\star\overline{u_B}, \ \tilde{X}\to u_A\star \tilde{X}^\mu_A\star \overline{u_A}, \ \tilde{X}^\mu_B\to u_B\star\tilde{X}^\mu_B\star \overline{u_B} \label{gauge}$$ In field theory the ground state can be defined through the minimum of the action, the relevant part of the for the minimization is: $$\begin{aligned}
&&S(\mathcal{D}_A)= \nonumber \\
&&+\frac{1}{(1+\Omega^2)^2} \int d^4z \Bigg\{\left(\frac{(1-\Omega^2)^2}{2}-\frac{(1-\Omega^2)^4}{6(1+\Omega^2)^2}\right)(F^A_{\mu\nu}\star F_A^{\mu\nu}+F^B_{\mu\nu}\star F_B^{\mu\nu})\nonumber \\
&&+\left(\bar{\phi}\star\phi+ \frac{4\Omega^2}{1+\Omega^2}\tilde{X}^A_\mu\star\tilde{X}_A^\mu-\frac{\chi_{-1}}{\chi_0}\right)^2\nonumber \\
&&+\left(\phi\star\bar{\phi}+ \frac{4\Omega^2}{1+\Omega^2}\tilde{X}^B_\mu\star\tilde{X}_B^\mu-\frac{\chi_{-1}}{\chi_0}\right)^2\nonumber \\
&& -2\left(\frac{4\Omega^2}{1+\Omega^2}\tilde{X}^\mu_0\star\tilde{X}^0_\mu-\frac{\chi_{-1}}{\chi_0} \right)^2+2(1+\Omega^2)D_\mu\phi\star\overline{D^\mu\phi}\Bigg\}(z)+\mathcal{O}(\chi_1) \label{faction1}\end{aligned}$$ Where we have omitted the constant part and we have rescaled the coefficient in front of the integral. Considering the fields $X_A^\mu$, $X_B^\mu$ as fields variables instead $A^\mu$, $B^\mu$ we can state that each terms of the action is semi-positive defined, so in order to find the minimum it is sufficient to minimize them separately. There are the two possible minimum for the field strength part and for the covariant derivative part: the trivial solution with $\phi$ and $\tilde{X}_A^\mu$, $\tilde{X}_B^\mu$ equal to the null fields and the solution with $\phi$, $X_A^\mu$, $X_B^\mu$ proportional to the identity. In each cases both the field strength part and the covariant derivative part disappear. For the potential parts we have: $$\begin{aligned}
V_A=V_B=\left(\frac{\chi_{-1}}{\chi_0}\right)^2 && \textrm{for} \ \phi=\tilde{X}^\mu_A=\tilde{X}^\mu_B=0 \\
V_A=(\alpha^2_\phi+4\frac{4\Omega^2}{1+\Omega^2}\alpha^2_A-\frac{\chi_{-1}}{\chi_0})^2, && \textrm{for} \ \phi=\alpha_\phi \textbf{I}, \ \tilde{X}^\mu_A=\alpha_A \textbf{I}^\mu, \ \tilde{X}^\mu_B=\alpha_B \textbf{I}^\mu \\
V_B=(\alpha^2_\phi+4\frac{4\Omega^2}{1+\Omega^2}\alpha^2_B-\frac{\chi_{-1}}{\chi_0})^2 &&\end{aligned}$$ Referring to the second case and minimizing the potentials, the minimum seems to be for $$\phi_0=\frac{\chi_{-1}}{\chi_0}\cos\alpha \textbf{I},\ \tilde{X}^\mu_{A0}=\tilde{X}^\mu_{B0}= \frac{1}{2}\sqrt{\frac{\chi_{-1}}{\chi_0}}\sqrt{ \frac{1+\Omega^2}{4\Omega^2}} \sin\alpha \textbf{I}^\mu \label{Vacuum}$$ However, the previous position is not allowed because the identity does not belong to the algebra under consideration. In general the non-triviality of the vacuum makes very difficult to explicit the vacuum configuration of the system in [@Goursac1] A. de Goursac, J.C. Wallet, and R. Wulkenhaar, using the matrix base formalism, have found an expressions from vacuum solutions deriving them from the relevant solutions equations of motion. Although, the complexity of the vacuum configuration makes the perturbative approach very complicated, in order to conduct some investigation in the next section will be consider a non-perturbative approach using a discretized matrix model of the action obtained using a Moyal base. In this setting the action reduces to $$\begin{aligned}
&&S(\mathcal{D}_A)= \nonumber \\
&&+\frac{1}{(1+\Omega^2)^2} \int d^4z \Bigg\{\left(\frac{(1-\Omega^2)^2}{2}-\frac{(1-\Omega^2)^4}{6(1+\Omega^2)^2}\right)(F^A_{\mu\nu}\star F_A^{\mu\nu}+F^B_{\mu\nu}\star F_B^{\mu\nu})\nonumber \\
&&+\left(\bar{\phi}\star\phi+\frac{4\Omega^2}{1+\Omega^2}\tilde{X}^A_\mu\star\tilde{X}_A^\mu-\frac{\chi_{-1}}{\chi_0}\right)^2 \nonumber \\
&&+\left(\phi\star\bar{\phi}+ \frac{4\Omega^2}{1+\Omega^2}\tilde{X}^B_\mu\star\tilde{X}_B^\mu-\frac{\chi_{-1}}{\chi_0}\right)^2\nonumber \\
&&+ 2(1+\Omega^2)D_\mu\phi\star\overline{D^\mu\phi}\Bigg\}(z)+\mathcal{O}(\chi_1) \label{S}\end{aligned}$$ The omitted factor for the finite matrix model of size $N$ becomes constant so can be ignored. The minimum is obtained like before and formally is in this case the identity, of course, belongs to the matrix space. It is interesting to notice that the vacuum of the finite model, due to the Higgs field, is no longer invariant under the transformations , but is invariant under a subgroup of $U(N)\times U(N)$: $$u_A=\overline{u_B} \longrightarrow u_A\star \textbf{I} \star\overline{u_B}=u_A\star \overline{u_A}=\textbf{I}$$ Having discretized the model will be performed a Monte Carlo simulation studying some statistical quantity such the energy density, specific heat, varying the parameters $\Omega, \frac{\chi_{-1}}{\chi_0}, \alpha$ and gathering some informations on the various contributions of the fields to the action. The simulations are quite cumbersome due to the complexity of the action and to the number of independent matrix to handle.
Discretization of the action
============================
The first step across the numerical analysis is to apply a discretization scheme. Various schemes can be used like lattice approximation, but the nature of star product due to its oscillator exponential, makes the lattice approach not suitable without adaptations. We will use another discretization scheme in which our fields are approximated by finite matrices and the star product becomes the standard matrix multiplication. Using the identity $D_\mu\phi =\phi\star\tilde{X}_{B\mu}-\tilde{X}_{A\mu}\star\phi$ we can recast the action in the following form: $$\begin{aligned}
&S(\phi,\tilde{X}_A,\tilde{X}_B)&= \frac{1}{(1+\Omega^2)^2}\int d^4z\Bigg\{\left(\frac{\left(1-\Omega^2\right)^2}{2}-\frac{\left(1-\Omega^2\right)^4}{6\left(1+\Omega^2\right)^2}\right)\Big(\left[\tilde{X}_{A\mu},\tilde{X}_{A\nu}\right]_\star\left[\tilde{X}_{A}^\mu,\tilde{X}_{A}^\nu\right]_\star \nonumber \\
& &+\left[\tilde{X}_{B\mu},\tilde{X}_{B\nu}\right]_\star\left[\tilde{X}_{B}^\mu,\tilde{X}_{B}^\nu\right]_\star\Big)
+ \left(\phi\star\bar{\phi}+\frac{4\Omega^2}{1+\Omega^2}\tilde{X}_A^\mu\star\tilde{X}_{A\mu}-\frac{\chi_{-1}}{\chi_0}\right)^2 \nonumber \\
& & +\left(\bar{\phi}\star\phi +\frac{4\Omega^2}{1+\Omega^2}\tilde{X}_B^\mu\star\tilde{X}_{B\mu}-\frac{\chi_{-1}}{\chi_0} \right)^2 \nonumber \\
& &+2(1+\Omega^2)\left(\phi\star\tilde{X}_{B\mu}-\tilde{X}_{A\mu}\star\phi\right)
\left(\bar{\phi}\star\tilde{X}_{A}^\mu-\tilde{X}_{B}^\mu\star\bar{\phi}\right)\Bigg\}(z) \nonumber \\
& &+\mathcal{O}(\chi_1)\label{S0}\end{aligned}$$As a first approach to the numerical simulation and forced by limited computation resource, we will consider the Monte Carlo simulation of the previous action around its minimum and the simulation will take $\sqrt{\frac{\chi_{-1}}{\chi_0}} $ as a positive parameter. In this setting the behavior of the simulations will be identical for the negative case and avoiding $\frac{\chi_{-1}}{\chi_0}$ to be negative we have not any problems about the thermalization. In order to define the previous action around the minimum we translate the fields $\phi$, $\tilde{X}_{A\mu}$, $\tilde{X}_{B\mu}$ using the following translated fields: $$\begin{aligned}
\phi &=& \psi +\sqrt{\frac{\chi_{-1}}{\chi_0}}\cos\alpha\textbf{I} \ \\
\tilde{X}_{A\mu}&=& Y_{A\mu} +\frac{1}{2}\sqrt{\frac{\chi_{-1}}{\chi_0}}\sqrt{\frac{2\Omega^2}{(1+\Omega^2)}}\textbf{I}_\mu\sin\alpha \\
\tilde{X}_{B\mu}&=& Y_{B\mu} +\frac{1}{2}\sqrt{\frac{\chi_{-1}}{\chi_0}}\sqrt{\frac{2\Omega^2}{(1+\Omega^2)}}\textbf{I}_\mu\sin\alpha \end{aligned}$$ Substituting the previous fields into we get a positive action with minimum in zero: $$\begin{aligned}
&S(\psi,Y_A,Y_B)&= +\frac{1}{(1+\Omega^2)^2}\int d^4z\Bigg\{D\Big(\left[Y_{A\mu},Y_{A\nu}\right]_\star\left[Y_{A}^\mu,Y_{A}^\nu\right]_\star
+ \left[Y_{B\mu},Y_{B\nu}\right]_\star\left[Y_{B}^\mu,Y_{B}^\nu\right]_\star\Big) \nonumber \\
& &+ \left(\psi\star\bar{\psi}+\mu\cos\alpha(\psi+\bar{\psi})+ CY_A^\mu\star Y_{A\mu} +\mu \textbf{I}^\mu Y_{A\mu}\sin\alpha \right)^2 \nonumber \\
& &+ \left(\bar{\psi}\star\psi +\mu\cos\alpha(\psi+\bar{\psi})+ CY_B^\mu\star Y_{B\mu} +\mu \textbf{I}^\mu Y_{B\mu}\sin\alpha \right)^2 \nonumber \\ \nonumber \\
& &+2(1+\Omega^2)\left((Y_{B\mu}-Y_{A\mu})\mu\cos\alpha +\psi\star Y_{B\mu}-Y_{A\mu}\star\psi\right)\nonumber \\
& &\star\left((Y_{A}^\mu-Y_{B}^\mu)\mu\cos\alpha+\bar{\psi}\star Y_{A}^\mu -Y_{B}^\mu\star\bar{\psi}\right)\Bigg\}(z) +\mathcal{O}(\chi_1)
\label{Sf} \end{aligned}$$ Where for simplicity we put: $$\begin{aligned}
C=\frac{1+\Omega^2}{4\Omega^2},& D=\frac{\left(1-\Omega^2\right)^2}{2}-\frac{\left(1-\Omega^2\right)^4}{6\left(1+\Omega^2\right)^2} , & \frac{\chi_{-1}}{\chi_0}=\mu^2\end{aligned}$$
Discretization by Moyal base
----------------------------
The following treatment is mainly taken from [@moyal-triple; @bondia2] as introduction to the Moyal base which will be used later. We can define on the algebra $\mathbb{R}_\Theta^2$ a natural basis of eigenfunctions $f_{mn}$ of the harmonic oscillator, where $m, n \in \mathbb{N}$. This base satisfy the $\star$-multiplication rule: $$(f_{mn} \star f_{kl} )(x) = \delta_{nk} f_{ml} (x) \label{star-rule}$$ and this useful property: $$\begin{aligned}
\int d^2 x f_{mn}(x)&=& \delta_{mn} \int d^2x f_0 =2\pi\theta\delta_{mn}\end{aligned}$$ The previous multiplication rule associates the $\star$-product between $f_{ml}$ with the ordinary matrix product: In this base we can write any elements of $ \mathbb{R}_\Theta^2$ but we have to require the rapid decay [@bondia2] of the sequences of coefficients $\{a_{mn}\}$: $$\sum^\infty_{m,n=0} a_{mn} f_{mn}(x) \in \mathbb{R}_\Theta^2, \ \textrm{if}, \ \sum^\infty_{m,n=0} \left((2m+1)^{2k} (2n+1)^{2k} |a_{mn}|^2 \right)^\frac{1}{2} < \infty \ \textrm{for all} \ k .$$ The eigenfunctions $f_{nm}$ can be expressed with the help of Laguerre functions [@bondia2; @phi4-non; @moy-base]: $$f_{mn}(\rho,\varphi)= 2(-1)^m \sqrt{\frac{m!}{n!}} e^{i\varphi(n-m)} \left(\sqrt{\frac{2}{\theta}}\rho\right)^{n-m} e^{-\frac{\rho^2}{\theta}}L_{m}^{n-m}\left(\frac{2}{\theta}\rho^2\right) \label{mb}$$ Our fields can be expanded in this base as: $$X^\mu(x) =\sum_{m_i,n_i \in \mathbb{N}} X^\mu_{\genfrac{}{}{0pt}{}{m_1n_1}{m_2n_2}} f_{m_1n_1}(x_0,x_1)f_{m_2n_2}(x_2,x_3 )\label{X-exp}$$ and $$\psi(x) =\sum_{m_i,n_i \in \mathbb{N}} \psi_{\genfrac{}{}{0pt}{}{m_1n_1}{m_2n_2}} f_{m_1n_1}(x_0,x_1)f_{m_2n_2}(x_2,x_3 )\label{psi-exp}$$ Using this base we can forget the Moyal product in this way the model becomes to 9-matrix model. A $\star$-product between two fields using can be written as $$\begin{aligned}
\Psi(x)\star\Phi(x)&=&\sum_{m_i,n_i,k_1,l_1\in \mathbb{N}}\Psi_{\genfrac{}{}{0pt}{}{m_1n_1}{m_2n_2}}\Phi_{\genfrac{}{}{0pt}{}{k_1l_1}{k_2l_2}} f_{m_1n_1}(x_0,x_1)\star f_{k_1l_1}(x_0,x_1) \nonumber \\
&\times & f_{m_2n_2}(x_2,x_3)\star f_{k_2l_2}(x_2,x_3) \nonumber \\
&=& \sum_{m_i,n_i,k_1,l_1\in \mathbb{N}}\Psi_{\genfrac{}{}{0pt}{}{m_1n_1}{m_2n_2}}\Phi_{\genfrac{}{}{0pt}{}{k_1l_1}{k_2l_2}}\delta_{n_1k_1}\delta_{n_2k_2}f_{m_1l_1}(x_0,x_1)f_{m_2l_2}(x_2,x_3)\nonumber \\
&=& \sum_{m_i,l_1\in \mathbb{N}}\Psi\Phi_{\genfrac{}{}{0pt}{}{m_1l_1}{m_2l_2}} f_{m_1l_1}(x_0,x_1)f_{m_2l_2}(x_2,x_3)\end{aligned}$$ where $$\Psi\Phi_{\genfrac{}{}{0pt}{}{m_1l_1}{m_2l_2}}=\sum_{n_1,n_2 \in \mathbb{N}}\Psi_{\genfrac{}{}{0pt}{}{m_1n_1}{m_2n_2}}\Phi_{\genfrac{}{}{0pt}{}{n_1l_1}{n_2l_2}}$$ So the star product became a “double” matrix multiplication, the action, the equations of field and all treatments can be conducted on the infinite matrices instead directly on the continues fields.
Beside, for finite matrices, the $\mathbb{N}^2$-indexed double sequences can be written as tensor products of ordinary matrices, $$X_{\genfrac{}{}{0pt}{}{m_1n_1}{m_2n_2}}= \sum_{i=1}^K
X^i_{m_1n_1} \otimes X^i_{m_2n_2} \;.
\label{tensorproduct}$$ Since the matrix product and trace also factor into these independent components, the action factors into $S=\sum_{i=1}^K
S(\psi^{1i},Y_A^{1i},Y_B^{1i})S(\psi^{2i},Y_A^{2i},Y_B^{2i})$. Then, regarding all $\psi^{1i},Y_A^{1i}$, $Y_B^{1i},\psi^{2i},Y_A^{2i},Y_B^{2i}$ as random variables over which to integrate in the partition function, the partition function factors, too: $$\begin{aligned}
&\int
\mathcal{D}(\psi^{11},Y_A^{11},Y_B^{11},\psi^{21},Y_A^{21},Y_B^{21})
\cdots
\mathcal{D}(\psi^{1K},Y_A^{1K},Y_B^{1K},\psi^{2K},Y_A^{2K},Y_B^{2K})
\;e^{-S}
\nonumber
\\
&=\bigg(
\int \mathcal{D}(\psi^{1i},Y_A^{1i},Y_B^{1i},\psi^{2i},Y_A^{2i},Y_B^{2i})
\;e^{-S(\psi^{1i},Y_A^{1i},Y_B^{1i})\cdot S(\psi^{2i},Y_A^{2i},Y_B^{2i})}
\bigg)^K\;.\end{aligned}$$ We may therefore restrict ourselves to $K=1$. Using this approximations the calculus will be performed just on standard infinite matrix, but is not enough to be handled numerically. We have to perform a truncation in order to obtain finite matrices, this truncation will consist in a maximum $m,n<N$ in the expansion -. It is easy to verify that this kind of approximation corresponds in a cut in energy, in fact from the definition of $f_{mn}$ we have: $$\{H, f_{mn}\}_\star= \theta(m+n+\frac{1}{2} )f_{mn}(\varphi,\rho)$$ Beside, can be proved [@moy-base] that the functions $f_{mn}$ with $m, n < N$ induce a cut-off in position space and momentum space: $$\rho_{max} \sim \sqrt{2\theta} \ ,\ \textrm{for} \ m, n < N$$ and $$p_{max} \sim \sqrt{\frac{8N}{\theta}} \,\ \textrm{for} \ m, n < N$$ Summarizing, to operate the discretization we have the following correspondences: $$\begin{aligned}
\phi(x) \in \mathbb{R}_\Theta^4 &\rightarrow & \hat{\phi} \in \mathbb{M}_N \\
Y^A_\mu(x) \in \mathbb{R}_\Theta^4 &\rightarrow & \hat{Y}^A_\mu \in \mathbb{M}_N \\
Y^B_\mu(x) \in \mathbb{R}_\Theta^4 &\rightarrow & \hat{Y}^B_\mu \in \mathbb{M}_N \\
\int a(x)dx &\rightarrow & \textrm{Tr}(\hat{a}) \end{aligned}$$ After truncating the representative matrices is convenient to operate another substitution [@Goursac1]: $$\begin{aligned}
Z_0=\hat{Y}^A_0+i\hat{Y}^A_1, & \bar{Z}_0=\hat{Y}^A_0-i\hat{Y}^A_1 \nonumber \\
Z_1=\hat{Y}^B_0+i\hat{Y}^B_1, & \bar{Z}_1=\hat{Y}^B_0-i\hat{Y}^B_1 \nonumber \\
Z_2=\hat{Y}^A_2+i\hat{Y}^A_2, & \bar{Z}_2=\hat{Y}^A_2-i\hat{Y}^A_3 \nonumber \\
Z_3=\hat{Y}^B_2+i\hat{Y}^B_3, & \bar{Z}_3=\hat{Y}^B_2-i\hat{Y}^B_3 \label{Z-sub}
\end{aligned}$$ In the end the discretized action is: $$S_4=\frac{1}{1+\Omega^2}\operatorname{Tr}\left(\mathcal{L}_F+\mathcal{L}_{V_0}+\mathcal{L}_{V_1}+\mathcal{L}_{D_0}\bar{\mathcal{L}}_{D_0} +\mathcal{L}_{D_1}\bar{\mathcal{L}}_{D_1}+\mathcal{L}_{D_2}\bar{\mathcal{L}}_{D_2}+\mathcal{L}_{D_3}\bar{\mathcal{L}}_{D_3}\right) \label{S4}$$ With $$\begin{aligned}
\mathcal{L}_{4F}&=&\frac{D}{2}\Big(\left[\bar{Z}_0,Z_0\right]^2 +\left[\bar{Z}_1,Z_1\right]^2 +
\frac{1}{4}\Big(\left[Z_0+\bar{Z}_0,Z_2-\bar{Z}_2\right]^2-\left[Z_0+\bar{Z}_0,Z_2+\bar{Z}_2\right]^2\nonumber \\
&+& \left[Z_0-\bar{Z}_0,Z_2+\bar{Z}_2\right]^2- \left[Z_0-\bar{Z}_0,Z_2-\bar{Z}_2\right]^2 -\left[Z_1+\bar{Z}_1,Z_3+\bar{Z}_3\right]^2\nonumber \\
&+& \left[Z_1+\bar{Z}_1,Z_3-\bar{Z}_3\right]^2 +\left[Z_1-\bar{Z}_1,Z_3+\bar{Z}_3\right]^2 -\left[Z_1-\bar{Z}_1,Z_3-\bar{Z}_3\right]^2\Big)\Big)\nonumber \\
\mathcal{L}_{4V_0}&=&\big(\psi\bar{\psi}+\mu\cos\alpha(\psi+\bar{\psi})+ \frac{1}{2}\left(\left\{\bar{Z}_0,Z_0\right\} +\left\{\bar{Z}_2,Z_2\right\}\right)\nonumber \\
&+&\frac{\mu\sin\alpha}{2\sqrt{C}}((-1+i)(Z_0+Z_2)+(1+i)(\bar{Z}_0+\bar{Z}_2))\big)^2 \nonumber \\
\mathcal{L}_{4V_1}&=&\big(\bar{\psi}\psi+\mu\cos\alpha(\psi+\bar{\psi})+ \frac{1}{2}\left(\left\{\bar{Z}_1,Z_1\right\} +\left\{\bar{Z}_3,Z_3\right\}\right)\nonumber \\
&+& \frac{\mu\sin\alpha}{2\sqrt{C}}((-1+i)(Z_1+Z_3)+(1+i)(\bar{Z}_1+\bar{Z}_3))\big)^2 \nonumber \\
\mathcal{L}_{4D_0}&=& \sqrt{2(1+\Omega^2)}\left(\mu\cos\alpha(Z_1+\bar{Z}_1-Z_0-\bar{Z}_0 ) + \psi(Z_1+\bar{Z}_1)-(Z_0+\bar{Z}_0)\psi\right)\nonumber \\
\mathcal{L}_{4D_1}&=& \sqrt{2(1+\Omega^2)}\left(\mu\cos\alpha(Z_1-\bar{Z}_1-Z_0+\bar{Z}_0 ) + \psi(Z_1-\bar{Z}_1)-(Z_0-\bar{Z}_0)\psi\right)\nonumber \\
\mathcal{L}_{4D_2}&=& \sqrt{2(1+\Omega^2)}\left(\mu\cos\alpha(Z_3+\bar{Z}_3-Z_2-\bar{Z}_2 ) + \psi(Z_3+\bar{Z}_3)-(Z_2+\bar{Z}_2)\psi\right)\nonumber \\
\mathcal{L}_{4D_3}&=& \sqrt{2(1+\Omega^2)}\left(\mu\cos\alpha(Z_3-\bar{Z}_1-Z_2+\bar{Z}_2 ) + \psi(Z_3-\bar{Z}_3)-(Z_2-\bar{Z}_2)\psi\right)\nonumber \\
\nonumber \end{aligned}$$ Where for simplicity we have omitted the hat on the matrices and the bars stand for the hermitian conjugate. In this case the action becomes 5 complex matrix model instead eight real matrices $Y_{A\mu}$, $Y_{B\mu}$ and one complex matrix $\psi$. This form may seem cumbersome but it is more comfortable for numerical simulations. The next step is to define the estimator for the average values of interest and to develop some numerical parameters in order to analyze the numerical results.
Definition of the observables
=============================
Following Monte Carlo methods, will be produced a sequence of configurations $\{(\psi,Z_i)_j \}, j = 1, 2,\cdots,T_{MC}$ and evaluated the average of the observables over that set of configurations. The sequences of configurations obtained, a Monte Carlo chain, are representations of the configuration space at the given parameters. In this frame the expectation value is approximated as $$\langle O\rangle \approx \frac{1}{T_{MC}}\sum_{j=1}^{T_{MC}}O_j$$ where $O_j$ is the value of the observable $O$ evaluated in the $j$-sampled configuration, $(\psi,Z_i)_j$, $O_i= O[(\psi,Z_i)_j]$. The internal energy is defined as: $$E(\Omega,\mu,\alpha)= \langle S\rangle$$ and the specific heat takes the form $$C(\Omega,\mu,\alpha)= \langle S^2\rangle - \langle S\rangle^2$$ These terms correspond to the usual definitions for energy $$E(\Omega,\mu,\alpha) = -\frac{1}{\mathcal{Z}}\frac{\partial\mathcal{Z}}{\partial\beta}$$ and specific heat $$C(\Omega,\mu,\alpha) =\frac{\partial E}{\partial\beta}$$ where $\mathcal{Z}$ is the partition function. It is very useful to compute separately the average values of the four contributions: $$\begin{aligned}
F(\Omega,\mu,\alpha)&=& \langle \operatorname{Tr} \mathcal{L}_F \rangle \\
V_0(\Omega,\mu,\alpha) &=& \langle \operatorname{Tr} \mathcal{L}_{V_0} \rangle \\
V_1(\Omega,\mu,\alpha) &=& \langle \operatorname{Tr} \mathcal{L}_{V_1} \rangle \\
D(\Omega,\mu,\alpha)&=& \langle \operatorname{Tr}\left(\mathcal{L}_{D_j}\bar{\mathcal{L}}_{D^j}\right) \rangle \\
V&=&V_0+V_1\end{aligned}$$ Where $i,j=1,2$ or $i,j=1,\cdots,4$.
Order parameters
----------------
The previous quantities are not enough if we want to measure the various contributions of different modes of the fields to the configuration $\psi,Z_i$. Therefore we need some control parameters usually called order parameters. As a first idea we can think about a quantity related to the norms of the fields for example the sums $\sum_{nm} |\psi_{nm}|^2$, $\sum_{nm} |Z_{inm}|^2$ of all squared entries of our matrices, this quantity is called the full power of the field [@order-par; @order-par1] and it can be computed as the trace of the square: $$\begin{aligned}
\varphi^2_a &=& \operatorname{Tr}(|\psi|^2) \label{vara} \\
Z^2_{ia} &=& \operatorname{Tr}(|Z_i|^2) \end{aligned}$$ However, $\langle\varphi_a\rangle$ alone is not a real order parameter because does not distinguish contributions from the different modes but we can use it as a reference to define the quantities: $$\begin{aligned}
\varphi^2_0 &=& \sum^N_{n=0} |a_{nn}|^2 \nonumber \\
Z^2_{i0} &=& \sum^N_{n=0} |z_{inn}|^2 \label{var0}\end{aligned}$$ Referring to the base it is easy to see that such parameters are connected to the pure spherical contribution. This quantity will be used to analyze the spherical contribution to the full power of the field. We can generalize the previous quantity defining some parameters $\varphi_l$ in such a way they form a decomposition of the full power of the fields. $$\varphi^2_a=\varphi^2_0+\sum_l \varphi^2_l , \ Z_{ia}^2=Z_{i0}^2+\sum_l Z_{il}^2$$ Following this prescription the other quantity for $l>0$ can be defined as: $$\varphi^2_l =\sum^l_{n,m=0} |a_{nm}(1-\delta_{nm})|^2 , \ Z_{il}^2 =\sum^l_{n,m=0} |z_{lnm}(1-\delta_{nm})|^2 \label{varl}$$ If the contribution is dominated from the spherical symmetric parameter we expect to have $\langle\varphi^2_a\rangle \sim \langle\varphi^2_0\rangle$, $\langle Z_{ia}^2\rangle \sim \langle Z^2_{i0}\rangle$. With we can define the order parameters $\varphi^2_1$, $Z_{i1}^2$ as a particular case $l = 1$ we have $$\varphi^2_1 =|a_{10}|^2+ |a_{01}|^2, \ Z_{i1}^2 =|z_{i10}|^2+ |z_{i01}|^2$$ In the next simulations we be evaluated the quantities related to $l = 0$ and to $l = 1$ as representative of those contribution where the rotational symmetry is broken. Using higher $l$ in we can analyze the contributions of the remaining modes, turns out that the measurements of the first two modes are enough to characterize the behavior of the system.
Numerical results
==================
Now we discuss the results of the Monte Carlo simulation on the approximated spectral model. As a first approach we use some restrictions on the parameters. Will be considered the spectral action around its minimum in order to simplify the calculation, in this frame in the action is symmetric under the transformation $\mu\to-\mu$ so will be used the condition $\mu\geq0 $ and $\mu^2\geq0$. Will be explored the range $\mu\in[0,3.1]$, this interval was chosen to show a particular behavior of the system for fixed $\Omega$. The parameter $\Omega$ appears only with its square and is defined as a real parameter, therefore for the $\Omega$ too we require $\Omega\geq0$. Beside, if we refer to the scalar model, is possible to prove [@phi4-non] using the Langmann-Szabo duality [@L-S-dual], that the model can be fully described varying $\Omega$ in the range $[0,1]$, for higher $\Omega$ the system can be remap inside the previous interval. In the present model the L-S duality does not hold any more, but forced by limited resource, we conjecture that the interval $\Omega\in[0,1]$ still enough to describe the system. The last parameter to consider is $\alpha$, it is connected to the choice of the vacuum state, the range of $\alpha$ is $[0,2\pi]$. The study of the system varying this parameter is quite important from a theoretical point of view because is related to the vacuum invariance. Beside, in the action appear some contributions proportional to $(\sin\alpha)/\Omega$ which seem to diverge for $\Omega=0$, numerically we have verified that is an eliminable divergence and the curves of the observables can be extended in zero by continuity. Studying the dependence on $\alpha$ we can conclude that in the limit $N\to\infty$ the observables are independent from $\alpha$, therefore for our purposes $\alpha$ will be fixed equal to zero avoiding the annoying terms. In general for each observable are computed the graphs for $N=$5, 10, 15, 20 matrix size.
Varying $\alpha$
-----------------
We start looking at the variation of energy density and the full power of the fields density for fixed $\mu$ and $\Omega$, varying $\alpha \in [0,2\pi]$. As representative here will be presented the graphs for $\mu=1$, $\Omega=\{1,0.5\}$ but we obtain the same behaviors for any other choice of the parameters allowed in the considered range.
![Total energy density and full power of the fields density for $\varphi^2_a$, $Z_{0a}^2$ (from the left to the right) fixing $\mu=1$, $\Omega=1$, varying $\alpha$ and $N$. $N=5$ (circle), $N=10$ (square), $N=15$ (triangle), $N=20$ (cross). []{data-label="Figure 1"}](graphs/mu=1-Om=1/Eng.eps "fig:") ![Total energy density and full power of the fields density for $\varphi^2_a$, $Z_{0a}^2$ (from the left to the right) fixing $\mu=1$, $\Omega=1$, varying $\alpha$ and $N$. $N=5$ (circle), $N=10$ (square), $N=15$ (triangle), $N=20$ (cross). []{data-label="Figure 1"}](graphs/mu=1-Om=1/APhi.eps "fig:") ![Total energy density and full power of the fields density for $\varphi^2_a$, $Z_{0a}^2$ (from the left to the right) fixing $\mu=1$, $\Omega=1$, varying $\alpha$ and $N$. $N=5$ (circle), $N=10$ (square), $N=15$ (triangle), $N=20$ (cross). []{data-label="Figure 1"}](graphs/mu=1-Om=1/AZ0.eps "fig:")
All tree graphs show an oscillating behavior of the values, this oscillation is present in all other quantities measured. The amplitude of this oscillation becomes smaller and smaller increasing the size of the matrix and this is true for all the quantities measured. The same trend is described in fig.\[Figure 2\] in which position of the maximum are different but the amplitudes becomes smaller increasing $N$
![Total energy density and full power of the fields density for $\varphi^2_a$, $Z_{0a}^2$ (from the left to the right) fixing $\mu=1$, $\Omega=0.5$, varying $\alpha$ and $N$. []{data-label="Figure 2"}](graphs/mu=1-Om=0.5/Eng.eps "fig:") ![Total energy density and full power of the fields density for $\varphi^2_a$, $Z_{0a}^2$ (from the left to the right) fixing $\mu=1$, $\Omega=0.5$, varying $\alpha$ and $N$. []{data-label="Figure 2"}](graphs/mu=1-Om=0.5/APhi.eps "fig:") ![Total energy density and full power of the fields density for $\varphi^2_a$, $Z_{0a}^2$ (from the left to the right) fixing $\mu=1$, $\Omega=0.5$, varying $\alpha$ and $N$. []{data-label="Figure 2"}](graphs/mu=1-Om=0.5/AZ0.eps "fig:")
This results allow us to consider $\alpha=0$ for all next graphs, since we are interested in the behavior of the system for $N \to \infty $. This occurrence simplify all the next simulations thanks to the vanishing the of terms $\sim(\sin\alpha)/\Omega $ appearing in the discretized action. Beside, such results induce us to reckon the parameter $\alpha$ as connected to the remaining invariance of the vacuum state for the exact model.
Varying $\Omega$
-----------------
Now we will analyze three cases in which $\mu$ is fixed to $0,1,3$, $\alpha$ is zero and we vary $\Omega\in[0,1]$. In the rest of this section we ignore for the computation of $E,D,V,F$ the prefactor $(1+\Omega^2)$. In this way we focus our attention to the integral as the source of possible phase transitions. The graphs in fig.\[Figure 4\] show the total energy density $\langle S \rangle / N^2$ and the various contributions: the potential $V / N^2$, the Yang-Mills part $F / N^2$ and the covariant derivative part $D / N^2$, for $\mu=1$. There is no evident discontinuity or peak and increasing the size of the matrices the curves remain smooth.
![Total energy density and the various contributions for $\mu=1$, $\alpha=0$ varying $\Omega$ and $N$. From the left to the right $E$, $V$, $D$, $F$ with $N=5$ (circle), $N=10$ (square), $N=15$ (triangle), $N=20$ (cross).[]{data-label="Figure 4"}](graphs/mu=1/Eng-N=Tot.eps "fig:") ![Total energy density and the various contributions for $\mu=1$, $\alpha=0$ varying $\Omega$ and $N$. From the left to the right $E$, $V$, $D$, $F$ with $N=5$ (circle), $N=10$ (square), $N=15$ (triangle), $N=20$ (cross).[]{data-label="Figure 4"}](graphs/mu=1/V-N=Tot.eps "fig:") ![Total energy density and the various contributions for $\mu=1$, $\alpha=0$ varying $\Omega$ and $N$. From the left to the right $E$, $V$, $D$, $F$ with $N=5$ (circle), $N=10$ (square), $N=15$ (triangle), $N=20$ (cross).[]{data-label="Figure 4"}](graphs/mu=1/D-N=Tot.eps "fig:") ![Total energy density and the various contributions for $\mu=1$, $\alpha=0$ varying $\Omega$ and $N$. From the left to the right $E$, $V$, $D$, $F$ with $N=5$ (circle), $N=10$ (square), $N=15$ (triangle), $N=20$ (cross).[]{data-label="Figure 4"}](graphs/mu=1/F-N=Tot.eps "fig:")
Comparing the energy density and the various contributions fig.\[Figure 5\] we notice that the contributions between $F$ and $V$ balance each other and the total energy follows the slope of $D$, this behavior continues increasing the size of the matrices fig.\[Figure 5\].\
![Comparison of the total energy density and the various contributions for $\mu=1$, $\alpha=0$. $E$ (circle), $F$ (triangle), $D$ (cross), $V$ (square). With $N=5$ (left) and $N=20$ (right). []{data-label="Figure 5"}](graphs/mu=1/EngTot-N=5.eps "fig:") ![Comparison of the total energy density and the various contributions for $\mu=1$, $\alpha=0$. $E$ (circle), $F$ (triangle), $D$ (cross), $V$ (square). With $N=5$ (left) and $N=20$ (right). []{data-label="Figure 5"}](graphs/mu=1/EngTot-N=20.eps "fig:")
![Specific heat for $\mu=1$.[]{data-label="Figure 6"}](graphs/mu=1/C-N=Tot.eps)
The specific heat density shows fig.\[Figure 6\] a peak in $\Omega=1$, this peak increases as $N$ increase. This behavior is typical of a phase transition, the peak is not clear for small $N$ due to the finite volume effect. In order to gain some informations on the composition of the fields we look at the order parameters defined in the previous chapter. Starting from $\psi$ field, in the figure \[Figure 6\] it is showed the graphs for $\varphi_a^2 $, $\varphi^2_0 $ and $\varphi^2_1 $ for $N=5$.\
The three values $\varphi_a^2 $, $\varphi^2_0 $ and $\varphi^2_1 $ seem essentially constant, comparing the three graphs fig.\[Figure 8\] it is easy to see the dominance of the spherical contribution $\varphi^2_0 $ to the full power of the field.
![On the left comparison of $\varphi_a^2 $ (circle), $\varphi^2_0 $ (square) and $\varphi^2_1 $ (triangle) density. On the right comparison of $Z_{0a}^2$ (circle), $Z_{00}^2$ (square) and $Z_{01}^2$ (triangle) density.[]{data-label="Figure 8"}](graphs/mu=1/TotPhi-N=5.eps "fig:") ![On the left comparison of $\varphi_a^2 $ (circle), $\varphi^2_0 $ (square) and $\varphi^2_1 $ (triangle) density. On the right comparison of $Z_{0a}^2$ (circle), $Z_{00}^2$ (square) and $Z_{01}^2$ (triangle) density.[]{data-label="Figure 8"}](graphs/mu=1/TotZ0-N=5.eps "fig:")
The behavior of the $Z_0$ fields is different, the spherical contribution becomes dominant approaching to $\Omega=0$ starting from a zone in which the contribution of $Z_{00}^2$ and $Z_{01}^2 $ are comparable. For brevity will not be showed only the graphs for $Z_{0a}^2 $, $Z_{00}^2$ and $Z_{01}^2 $ but taking in account the statistical errors the other $Z_i$ related graphs appear compatible to the $Z_0$ case. The dependence of the previous quantities on $N$ are showed in the following graphs fig.\[Figure 9\].
![Starting from the up left corner and from the left to the right the densities for $\varphi_a^2 $, $\varphi^2_0 $, $\varphi^2_1 $, $Z_{0a}^2$, $Z_{00}^2 $ and $Z_{01}^2$ for $\mu=1$ varying $\Omega$ and $N$. []{data-label="Figure 9"}](graphs/mu=1/APhi-N=Tot.eps "fig:") ![Starting from the up left corner and from the left to the right the densities for $\varphi_a^2 $, $\varphi^2_0 $, $\varphi^2_1 $, $Z_{0a}^2$, $Z_{00}^2 $ and $Z_{01}^2$ for $\mu=1$ varying $\Omega$ and $N$. []{data-label="Figure 9"}](graphs/mu=1/A0Phi-N=Tot.eps "fig:") ![Starting from the up left corner and from the left to the right the densities for $\varphi_a^2 $, $\varphi^2_0 $, $\varphi^2_1 $, $Z_{0a}^2$, $Z_{00}^2 $ and $Z_{01}^2$ for $\mu=1$ varying $\Omega$ and $N$. []{data-label="Figure 9"}](graphs/mu=1/A1Phi-N=Tot.eps "fig:") ![Starting from the up left corner and from the left to the right the densities for $\varphi_a^2 $, $\varphi^2_0 $, $\varphi^2_1 $, $Z_{0a}^2$, $Z_{00}^2 $ and $Z_{01}^2$ for $\mu=1$ varying $\Omega$ and $N$. []{data-label="Figure 9"}](graphs/mu=1/AZ0-N=Tot.eps "fig:") ![Starting from the up left corner and from the left to the right the densities for $\varphi_a^2 $, $\varphi^2_0 $, $\varphi^2_1 $, $Z_{0a}^2$, $Z_{00}^2 $ and $Z_{01}^2$ for $\mu=1$ varying $\Omega$ and $N$. []{data-label="Figure 9"}](graphs/mu=1/A0Z0-N=Tot.eps "fig:") ![Starting from the up left corner and from the left to the right the densities for $\varphi_a^2 $, $\varphi^2_0 $, $\varphi^2_1 $, $Z_{0a}^2$, $Z_{00}^2 $ and $Z_{01}^2$ for $\mu=1$ varying $\Omega$ and $N$. []{data-label="Figure 9"}](graphs/mu=1/A1Z0-N=Tot.eps "fig:")
The values of the quantity of all the previous parameters decreases with $N$ but the dominance of the $\varphi_0$ on the total power of the field is independent by $N$. The peak related to $Z_0$ decrease, but if look at the single graph for $N=20$ the spherical contribution approaching the point $\Omega=0$ it features a peak.
Now we will analyze the model for $\mu=0$; fig.\[Figure 10\] shows the graphs for total energy density and the contributions $V$, $D$, $F$. The slope of the total energy density seems to be constant. The $D$ contribution and the $F$ do not balance each other like in the previous case, but all the three contributions balance among them self to produce a constant sum.
![Total energy density, various contributions and the comparison among them for $\mu=0$ varying $\Omega$ and $N$. From the left to the right $E$, $V$, $D$, $F$ and comparison.[]{data-label="Figure 10"}](graphs/mu=0/Eng-N=Tot.eps "fig:") ![Total energy density, various contributions and the comparison among them for $\mu=0$ varying $\Omega$ and $N$. From the left to the right $E$, $V$, $D$, $F$ and comparison.[]{data-label="Figure 10"}](graphs/mu=0/V-N=Tot.eps "fig:") ![Total energy density, various contributions and the comparison among them for $\mu=0$ varying $\Omega$ and $N$. From the left to the right $E$, $V$, $D$, $F$ and comparison.[]{data-label="Figure 10"}](graphs/mu=0/D-N=Tot.eps "fig:") ![Total energy density, various contributions and the comparison among them for $\mu=0$ varying $\Omega$ and $N$. From the left to the right $E$, $V$, $D$, $F$ and comparison.[]{data-label="Figure 10"}](graphs/mu=0/F-N=Tot.eps "fig:") ![Total energy density, various contributions and the comparison among them for $\mu=0$ varying $\Omega$ and $N$. From the left to the right $E$, $V$, $D$, $F$ and comparison.[]{data-label="Figure 10"}](graphs/mu=0/TotEng-N=20.eps "fig:")
![Specific heat density for $\mu=0$ varying $\Omega$ and $N$.[]{data-label="Figure 11"}](graphs/mu=0/C-N=Tot.eps)
The specific heat density shows fig.\[Figure 11\] again the peak in $\Omega=1$ as $N$ increase. For the other quantities $\varphi_a^2 $, $\varphi^2_0 $, $\varphi^2_1 $ and $Z_{0a}^2 $, $Z_{00}^2$, $Z_{01}^2$ we have the same behavior fig.\[Figure 14\] of $\mu=1$ case, except for the oscillation appearing in the $Z_{0a}^2 $, $Z_{00}^2$ graphs close to zero, anyway it appears only for $N=5$.
![Starting from the up left corner and from the left to the right the densities for $\varphi_a^2 $, $\varphi^2_0 $, $\varphi^2_1 $, $Z_{0a}^2$, $Z_{00}^2 $ and $Z_{01}^2$ for $\mu=0$ varying $\Omega$ and $N$. []{data-label="Figure 12"}](graphs/mu=0/APhi-N=Tot.eps "fig:") ![Starting from the up left corner and from the left to the right the densities for $\varphi_a^2 $, $\varphi^2_0 $, $\varphi^2_1 $, $Z_{0a}^2$, $Z_{00}^2 $ and $Z_{01}^2$ for $\mu=0$ varying $\Omega$ and $N$. []{data-label="Figure 12"}](graphs/mu=0/A0Phi-N=Tot.eps "fig:") ![Starting from the up left corner and from the left to the right the densities for $\varphi_a^2 $, $\varphi^2_0 $, $\varphi^2_1 $, $Z_{0a}^2$, $Z_{00}^2 $ and $Z_{01}^2$ for $\mu=0$ varying $\Omega$ and $N$. []{data-label="Figure 12"}](graphs/mu=0/A1Phi-N=Tot.eps "fig:") ![Starting from the up left corner and from the left to the right the densities for $\varphi_a^2 $, $\varphi^2_0 $, $\varphi^2_1 $, $Z_{0a}^2$, $Z_{00}^2 $ and $Z_{01}^2$ for $\mu=0$ varying $\Omega$ and $N$. []{data-label="Figure 12"}](graphs/mu=0/AZ0-N=Tot.eps "fig:") ![Starting from the up left corner and from the left to the right the densities for $\varphi_a^2 $, $\varphi^2_0 $, $\varphi^2_1 $, $Z_{0a}^2$, $Z_{00}^2 $ and $Z_{01}^2$ for $\mu=0$ varying $\Omega$ and $N$. []{data-label="Figure 12"}](graphs/mu=0/A0Z0-N=Tot.eps "fig:") ![Starting from the up left corner and from the left to the right the densities for $\varphi_a^2 $, $\varphi^2_0 $, $\varphi^2_1 $, $Z_{0a}^2$, $Z_{00}^2 $ and $Z_{01}^2$ for $\mu=0$ varying $\Omega$ and $N$. []{data-label="Figure 12"}](graphs/mu=0/A1Z0-N=Tot.eps "fig:")
A complete different response of the system is described in the graphs for $\mu=3$, as we can see from fig.\[Figure 13\]. The slope of total energy density is very similar to the $F$ component instead $D$. Beside, appears a maximum around $\Omega=0.4$ for $N\to\infty$. This dramatic change of the graphs might be interpreted as consequence of a phase transition ruled by the parameter $\mu$, actually in the next section we will find a peak in the specific heat density for some fixed $\Omega$ and varying $\mu\in[0,3]$.
![Total energy density, various contributions and the comparison among them for $\mu=3$ varying $\Omega$ and $N$. From the left to the right $E$, $V$, $D$, $F$ and comparison. []{data-label="Figure 13"}](graphs/mu=3/Eng-N=Tot.eps "fig:") ![Total energy density, various contributions and the comparison among them for $\mu=3$ varying $\Omega$ and $N$. From the left to the right $E$, $V$, $D$, $F$ and comparison. []{data-label="Figure 13"}](graphs/mu=3/V-N=Tot.eps "fig:") ![Total energy density, various contributions and the comparison among them for $\mu=3$ varying $\Omega$ and $N$. From the left to the right $E$, $V$, $D$, $F$ and comparison. []{data-label="Figure 13"}](graphs/mu=3/D-N=Tot.eps "fig:") ![Total energy density, various contributions and the comparison among them for $\mu=3$ varying $\Omega$ and $N$. From the left to the right $E$, $V$, $D$, $F$ and comparison. []{data-label="Figure 13"}](graphs/mu=3/F-N=Tot.eps "fig:") ![Total energy density, various contributions and the comparison among them for $\mu=3$ varying $\Omega$ and $N$. From the left to the right $E$, $V$, $D$, $F$ and comparison. []{data-label="Figure 13"}](graphs/mu=3/TotEng-N=20.eps "fig:")
Specific heat density displays fig.\[Figure 14\] a strong change too, in fact instead the peak in $\Omega=1$, it appears in the opposite side of the studied interval in $\Omega=0$. This peak too, due to its grows increasing $N$ could indicate a phase transition.
![Specific heat density for $\mu=3$ varying $\Omega$ and $N$.[]{data-label="Figure 14"}](graphs/mu=3/C-N=Tot.eps)
![Starting from the up left corner and from the left to the right the densities for $\varphi_a^2 $, $\varphi^2_0 $, $\varphi^2_1 $, $Z_{0a}^2$, $Z_{00}^2 $ and $Z_{01}^2$ for $\mu=3$ varying $\Omega$ and $N$.[]{data-label="Figure 15"}](graphs/mu=3/APhi-N=Tot.eps "fig:") ![Starting from the up left corner and from the left to the right the densities for $\varphi_a^2 $, $\varphi^2_0 $, $\varphi^2_1 $, $Z_{0a}^2$, $Z_{00}^2 $ and $Z_{01}^2$ for $\mu=3$ varying $\Omega$ and $N$.[]{data-label="Figure 15"}](graphs/mu=3/A0Phi-N=Tot.eps "fig:") ![Starting from the up left corner and from the left to the right the densities for $\varphi_a^2 $, $\varphi^2_0 $, $\varphi^2_1 $, $Z_{0a}^2$, $Z_{00}^2 $ and $Z_{01}^2$ for $\mu=3$ varying $\Omega$ and $N$.[]{data-label="Figure 15"}](graphs/mu=3/A1Phi-N=Tot.eps "fig:") ![Starting from the up left corner and from the left to the right the densities for $\varphi_a^2 $, $\varphi^2_0 $, $\varphi^2_1 $, $Z_{0a}^2$, $Z_{00}^2 $ and $Z_{01}^2$ for $\mu=3$ varying $\Omega$ and $N$.[]{data-label="Figure 15"}](graphs/mu=3/AZ0-N=Tot.eps "fig:") ![Starting from the up left corner and from the left to the right the densities for $\varphi_a^2 $, $\varphi^2_0 $, $\varphi^2_1 $, $Z_{0a}^2$, $Z_{00}^2 $ and $Z_{01}^2$ for $\mu=3$ varying $\Omega$ and $N$.[]{data-label="Figure 15"}](graphs/mu=3/A0Z0-N=Tot.eps "fig:") ![Starting from the up left corner and from the left to the right the densities for $\varphi_a^2 $, $\varphi^2_0 $, $\varphi^2_1 $, $Z_{0a}^2$, $Z_{00}^2 $ and $Z_{01}^2$ for $\mu=3$ varying $\Omega$ and $N$.[]{data-label="Figure 15"}](graphs/mu=3/A1Z0-N=Tot.eps "fig:")
The fig.\[Figure 15\] describes the behavior of the order parameters densities $\varphi_a^2 $, $\varphi^2_0 $, $\varphi^2_1 $ and $Z_{0a}^2$, $Z_{00}^2 $, $Z_{01}^2$, they have a similar aspect to the previous relative graphs. For the $\psi$ field the spherical contribution remains dominant, beside in the $\varphi^2_1 $ graph appears a deviation from the constant slope this deviation is evident for $N=5$ but still there for higher $N$. The order parameters for $Z_0$ display a peak close to the origin without oscillations even for $N=5$. This maximum for higher $N$ does not move closer to the origin, in other words, this shift in not due to the finite volume effect. Even for $Z^2_{01} $ graph appears a deviation from the constant slope, a small peak which becomes shifted and smoother for higher $N$
Varying $\mu$
--------------
In this section is analyzed the response of the system varying $\mu\in[0,3]$ where $\Omega$ is fixed to $0,0.5,1$ and $\alpha$ is always zero. We start displaying the graphs fig.\[Figure 16\] of the total energy density and of various contributions for $\Omega=0$. There is no evident discontinuity but appears a peak in the total energy density around $\mu\approx 2.5$ for $N=20$. Comparing all the contributions is easy to notice that the slope of the total energy is dictated by the curve $V$ of the potential part.
![The total energy density and the various contributions for $\Omega=0$ varying $\mu$ and $N$. From the left to the right $E$, $V$, $D$, $F$ an comparison with $N=5$ (circle), $N=10$ (square), $N=15$ (triangle), $N=20$ (cross). For the comparison: $E$ (circle), $V$ (square), $D$ (triangle), $F$ (cross).[]{data-label="Figure 16"}](graphs/Om=0/Eng-N=Tot.eps "fig:") ![The total energy density and the various contributions for $\Omega=0$ varying $\mu$ and $N$. From the left to the right $E$, $V$, $D$, $F$ an comparison with $N=5$ (circle), $N=10$ (square), $N=15$ (triangle), $N=20$ (cross). For the comparison: $E$ (circle), $V$ (square), $D$ (triangle), $F$ (cross).[]{data-label="Figure 16"}](graphs/Om=0/V-N=Tot.eps "fig:") ![The total energy density and the various contributions for $\Omega=0$ varying $\mu$ and $N$. From the left to the right $E$, $V$, $D$, $F$ an comparison with $N=5$ (circle), $N=10$ (square), $N=15$ (triangle), $N=20$ (cross). For the comparison: $E$ (circle), $V$ (square), $D$ (triangle), $F$ (cross).[]{data-label="Figure 16"}](graphs/Om=0/D-N=Tot.eps "fig:") ![The total energy density and the various contributions for $\Omega=0$ varying $\mu$ and $N$. From the left to the right $E$, $V$, $D$, $F$ an comparison with $N=5$ (circle), $N=10$ (square), $N=15$ (triangle), $N=20$ (cross). For the comparison: $E$ (circle), $V$ (square), $D$ (triangle), $F$ (cross).[]{data-label="Figure 16"}](graphs/Om=0/F-N=Tot.eps "fig:") ![The total energy density and the various contributions for $\Omega=0$ varying $\mu$ and $N$. From the left to the right $E$, $V$, $D$, $F$ an comparison with $N=5$ (circle), $N=10$ (square), $N=15$ (triangle), $N=20$ (cross). For the comparison: $E$ (circle), $V$ (square), $D$ (triangle), $F$ (cross).[]{data-label="Figure 16"}](graphs/Om=0/TotEng-N=Tot.eps "fig:")
![Specific heat density for $\Omega=0$ varying $\mu$ and $N$.[]{data-label="Figure 17"}](graphs/Om=0/C-N=Tot.eps)
As mentioned before, the specific heat density fig.\[Figure 17\] features a peak around $\mu\approx2.5$ for $N=20$ and again, due to this behavior as $N$ increase, we could relate this peak to a phase transition. The plots for the quantities $\varphi_a^2 $ and $\varphi^2_0 $ denote a strong dependence on $\mu$, in particular the slope of $\varphi^2_0 $ seems mostly linear, $\varphi^2_1 $ related graphs feature a similar behavior but the slope is no longer linear. Comparing the three graphs fig.\[Figure 18\] we deduce that close to the origin the non spherical contribution is bigger the spherical one, increasing $\mu$ this situation capsizes and $\varphi^2_0 $ becomes dominant respect $\varphi^2_1 $.
![Starting from the up left corner and from the left to the right the densities for $\varphi_a^2 $, $\varphi^2_0 $, $\varphi^2_1 $, $Z_{0a}^2$, $Z_{00}^2 $ and $Z_{01}^2$ for $\Omega=0$ varying $\mu$ and $N$. []{data-label="Figure 18"}](graphs/Om=0/APhi-N=Tot.eps "fig:") ![Starting from the up left corner and from the left to the right the densities for $\varphi_a^2 $, $\varphi^2_0 $, $\varphi^2_1 $, $Z_{0a}^2$, $Z_{00}^2 $ and $Z_{01}^2$ for $\Omega=0$ varying $\mu$ and $N$. []{data-label="Figure 18"}](graphs/Om=0/A0Phi-N=Tot.eps "fig:") ![Starting from the up left corner and from the left to the right the densities for $\varphi_a^2 $, $\varphi^2_0 $, $\varphi^2_1 $, $Z_{0a}^2$, $Z_{00}^2 $ and $Z_{01}^2$ for $\Omega=0$ varying $\mu$ and $N$. []{data-label="Figure 18"}](graphs/Om=0/A1Phi-N=Tot.eps "fig:") ![Starting from the up left corner and from the left to the right the densities for $\varphi_a^2 $, $\varphi^2_0 $, $\varphi^2_1 $, $Z_{0a}^2$, $Z_{00}^2 $ and $Z_{01}^2$ for $\Omega=0$ varying $\mu$ and $N$. []{data-label="Figure 18"}](graphs/Om=0/AZ0-N=Tot.eps "fig:") ![Starting from the up left corner and from the left to the right the densities for $\varphi_a^2 $, $\varphi^2_0 $, $\varphi^2_1 $, $Z_{0a}^2$, $Z_{00}^2 $ and $Z_{01}^2$ for $\Omega=0$ varying $\mu$ and $N$. []{data-label="Figure 18"}](graphs/Om=0/A0Z0-N=Tot.eps "fig:") ![Starting from the up left corner and from the left to the right the densities for $\varphi_a^2 $, $\varphi^2_0 $, $\varphi^2_1 $, $Z_{0a}^2$, $Z_{00}^2 $ and $Z_{01}^2$ for $\Omega=0$ varying $\mu$ and $N$. []{data-label="Figure 18"}](graphs/Om=0/A1Z0-N=Tot.eps "fig:")
The behavior of the $Z_0$ fields is quite different, referring to figure \[Figure 18\], the spherical contribution is always dominant for the all interval $\mu\in[0,3]$. The curves for $Z_{0a}^2$, $Z_{00}^2 $ are compatible to the constant slope, for $Z_{01}^2 $ we have the same dependence on $\mu$ in particular there is a smooth descending step, however this step becomes smoother for bigger $N$. It is behooves to say that due to some cancellations effects the statistical errors are quite big and they can hide some dependence, anyway this results tell us about the dependence of the order parameter for $Z_i$ and in general of the system, on the two choice $\Omega=0$ or $\Omega\neq0$.
Now we will analyze the model for $\Omega=0.5$; as fig.\[Figure 19\] shows the graphs have a different slope comparing to the previous case, the maximum of total energy density follow the one of the $V$ component. If we focus ourself only on the total energy graph and we compare it with the one for $\Omega=0$, we notice a shift of the maximum for each $N$. In particular in fig.\[Figure 19\] some maximum are moved outside the considered interval.
![Total energy density and contributions for $\Omega=0.5$ varying $\mu$ and $N$. From the left to the right $E$, $V$, $D$, $F$.[]{data-label="Figure 19"}](graphs/Om=0.5/Eng-N=Tot.eps "fig:") ![Total energy density and contributions for $\Omega=0.5$ varying $\mu$ and $N$. From the left to the right $E$, $V$, $D$, $F$.[]{data-label="Figure 19"}](graphs/Om=0.5/V-N=Tot.eps "fig:") ![Total energy density and contributions for $\Omega=0.5$ varying $\mu$ and $N$. From the left to the right $E$, $V$, $D$, $F$.[]{data-label="Figure 19"}](graphs/Om=0.5/D-N=Tot.eps "fig:") ![Total energy density and contributions for $\Omega=0.5$ varying $\mu$ and $N$. From the left to the right $E$, $V$, $D$, $F$.[]{data-label="Figure 19"}](graphs/Om=0.5/F-N=Tot.eps "fig:")
We can find this shift very clearly looking at specific heat density graph fig.\[Figure 20\], we find again the peak as $N$ increase but it is shifted around $\mu\approx 3.3$.
![Specific heat density for $\Omega=0.5$ varying $\mu$ and $N$.[]{data-label="Figure 20"}](graphs/Om=0.5/C-N=Tot.eps)
The graphs fig.\[Figure 23\] for $\varphi_a^2 $, $\varphi^2_0 $ have the same behavior of $\Omega=0.5$ case, excluding some fluctuations close to the origin for $N=5$ due to the finite volume effect, $\varphi^2_1 $ graph displays an almost constant curve. However, close to the origin, the spherical contribution and the first non spherical one are comparable.
![Starting from the up left corner and from the left to the right the densities for $\varphi_a^2 $, $\varphi^2_0 $, $\varphi^2_1 $, $Z_{0a}^2$, $Z_{00}^2 $ and $Z_{01}^2$ for $\Omega=0.5$ varying $\mu$ and $N$.[]{data-label="Figure 21"}](graphs/Om=0.5/APhi-N=Tot.eps "fig:") ![Starting from the up left corner and from the left to the right the densities for $\varphi_a^2 $, $\varphi^2_0 $, $\varphi^2_1 $, $Z_{0a}^2$, $Z_{00}^2 $ and $Z_{01}^2$ for $\Omega=0.5$ varying $\mu$ and $N$.[]{data-label="Figure 21"}](graphs/Om=0.5/A0Phi-N=Tot.eps "fig:") ![Starting from the up left corner and from the left to the right the densities for $\varphi_a^2 $, $\varphi^2_0 $, $\varphi^2_1 $, $Z_{0a}^2$, $Z_{00}^2 $ and $Z_{01}^2$ for $\Omega=0.5$ varying $\mu$ and $N$.[]{data-label="Figure 21"}](graphs/Om=0.5/A1Phi-N=Tot.eps "fig:") ![Starting from the up left corner and from the left to the right the densities for $\varphi_a^2 $, $\varphi^2_0 $, $\varphi^2_1 $, $Z_{0a}^2$, $Z_{00}^2 $ and $Z_{01}^2$ for $\Omega=0.5$ varying $\mu$ and $N$.[]{data-label="Figure 21"}](graphs/Om=0.5/AZ0-N=Tot.eps "fig:") ![Starting from the up left corner and from the left to the right the densities for $\varphi_a^2 $, $\varphi^2_0 $, $\varphi^2_1 $, $Z_{0a}^2$, $Z_{00}^2 $ and $Z_{01}^2$ for $\Omega=0.5$ varying $\mu$ and $N$.[]{data-label="Figure 21"}](graphs/Om=0.5/A0Z0-N=Tot.eps "fig:") ![Starting from the up left corner and from the left to the right the densities for $\varphi_a^2 $, $\varphi^2_0 $, $\varphi^2_1 $, $Z_{0a}^2$, $Z_{00}^2 $ and $Z_{01}^2$ for $\Omega=0.5$ varying $\mu$ and $N$.[]{data-label="Figure 21"}](graphs/Om=0.5/A1Z0-N=Tot.eps "fig:")
The introduction of $\Omega\neq 0$ creates, in the $Z_0$ fields order parameters fig.\[Figure 21\], a dependence similar to the graphs for the $\psi$; the full power of the field density and the spherical contribution are no more constant and they grow increasing $\mu$. Even in this case the spherical contribution is always dominant excluding the region around $\mu=0$.
The last set of graphs for the 4-dimensional model are obtained fixing $\Omega=1$, due to the vanishing of prefactor in front of the Yang-Mills part of the action the $F$ contribution is always zero. The following diagrams for the energy and contributions show the absence of the previous peak and comparing again them with former graphs they seem a sort dilatation.
![Total energy density and the various contributions for $\Omega=1$ varying $\mu$ and $N$. From the left to the right $E$, $V$, $D$. []{data-label="Figure 22"}](graphs/Om=1/Eng-N=Tot.eps "fig:") ![Total energy density and the various contributions for $\Omega=1$ varying $\mu$ and $N$. From the left to the right $E$, $V$, $D$. []{data-label="Figure 22"}](graphs/Om=1/V-N=Tot.eps "fig:") ![Total energy density and the various contributions for $\Omega=1$ varying $\mu$ and $N$. From the left to the right $E$, $V$, $D$. []{data-label="Figure 22"}](graphs/Om=1/D-N=Tot.eps "fig:")
The specific heat density does not show the peak in zero any more fig.\[Figure 23\] and the curves does not show any particular point as $N$ increase, actually the peak can be found for higher $\mu$.
![Specific heat density for $\Omega=1$ varying $\mu$ and $N$.[]{data-label="Figure 23"}](graphs/Om=1/C-N=Tot.eps)
![Starting from the up left corner and from the left to the right the densities for $\varphi_a^2 $, $\varphi^2_0 $, $\varphi^2_1 $, $Z_{0a}^2$, $Z_{00}^2 $ and $Z_{01}^2$ for $\Omega=1$ varying $\mu$ and $N$.[]{data-label="Figure 24"}](graphs/Om=1/APhi-N=Tot.eps "fig:") ![Starting from the up left corner and from the left to the right the densities for $\varphi_a^2 $, $\varphi^2_0 $, $\varphi^2_1 $, $Z_{0a}^2$, $Z_{00}^2 $ and $Z_{01}^2$ for $\Omega=1$ varying $\mu$ and $N$.[]{data-label="Figure 24"}](graphs/Om=1/A0Phi-N=Tot.eps "fig:") ![Starting from the up left corner and from the left to the right the densities for $\varphi_a^2 $, $\varphi^2_0 $, $\varphi^2_1 $, $Z_{0a}^2$, $Z_{00}^2 $ and $Z_{01}^2$ for $\Omega=1$ varying $\mu$ and $N$.[]{data-label="Figure 24"}](graphs/Om=1/A1Phi-N=Tot.eps "fig:") ![Starting from the up left corner and from the left to the right the densities for $\varphi_a^2 $, $\varphi^2_0 $, $\varphi^2_1 $, $Z_{0a}^2$, $Z_{00}^2 $ and $Z_{01}^2$ for $\Omega=1$ varying $\mu$ and $N$.[]{data-label="Figure 24"}](graphs/Om=1/AZ0-N=Tot.eps "fig:") ![Starting from the up left corner and from the left to the right the densities for $\varphi_a^2 $, $\varphi^2_0 $, $\varphi^2_1 $, $Z_{0a}^2$, $Z_{00}^2 $ and $Z_{01}^2$ for $\Omega=1$ varying $\mu$ and $N$.[]{data-label="Figure 24"}](graphs/Om=1/A0Z0-N=Tot.eps "fig:") ![Starting from the up left corner and from the left to the right the densities for $\varphi_a^2 $, $\varphi^2_0 $, $\varphi^2_1 $, $Z_{0a}^2$, $Z_{00}^2 $ and $Z_{01}^2$ for $\Omega=1$ varying $\mu$ and $N$.[]{data-label="Figure 24"}](graphs/Om=1/A1Z0-N=Tot.eps "fig:")
At last in fig.\[Figure 24\] we found a behavior of the order parameters density for $Z_0$ and $\psi$ fields similar to the former graphs for $\Omega=0.5$ and they are compatible with a dilatation of the previous diagrams.
Conclusions and prospectives {#conclusions-and-prospectives .unnumbered}
============================
We have presented a study of a spectral action model constructed in order to extend to the Yang-Mills theories the non-commutative Wulkenhaar-Grosse model. The main aim of this work was to test a first Monte Carlo approach based on a non-perturbative regularization method. We have performed Monte Carlo simulations and obtained the values of the defined observables varying the parameters of the system. Despite the complexity of the approximated spectral action considered here we were able to obtain some reliable numerical results, we can conclude that a numerical approach of this kind of model using the matrix Moyal base seems feasible. The specific heat density shows various peaks indicating phase transitions, in particular studying the behaviors for some fixed $\mu$ we found a peak around $\Omega=0$ for $\mu=\{0,1\}$ and peak in $\Omega=1$ for $ \mu=3$, beside we notice a huge change in the energy density and in its contributions between the cases $\mu=\{0,1\}$ and $\mu=3$. Other peaks in the specific heat density can be found varying $\mu$ and fixing $\Omega$, the graphs show that increasing $\Omega$ the peak in the specific heat, starting from $\mu\approx 2.4$ for $\Omega=0$, is moved towards higher $\mu$. The order parameters introduced show a strong dependence on the occurrence of $\Omega=0$ or $\Omega\neq0$. Referring to the fixed $\mu$ graphs we found a peak in the spherical contribution for the gauge fields $Z_i$, we can interpreted this slope as a sort of symmetry breaking introduced by $\Omega\neq0$. Additionally, varying $\mu$ and fixing $\Omega$ the other parameters display an increasing slope with $\mu$ for all fields and all situation but one; the graphs of the order parameters concerning $Z_{0a}$, $Z_{00}$ for $\Omega=0$ show a constant behavior. The natural next steps in the numerical study of this model, could be the computation of the transition curves in order to separate the phase regions and classify them using eventually some additional order parameters. Our treatment, forced by limited resource, was conducted conjecturing that the system can fully described varying $\Omega$ in the range $[0,1]$ but since the L-S duality does not hold any more in our case, will be very useful to extend this range. Actually, the computed graphs does not show any periodicity in $ \Omega\in[0,1]$ so there are some hints to infer that the range $[0,1]$ is not enough, however only a direct computation will clarify this point. Will be very interesting to do not require any more the condition $\mu^2>0$, in order to implement this change we have to conduct the calculation no more around the minimum of the actions. The expansion of the parameters space, together with the classification of the different phase regions allows us to compare our model with the results of the simulation conducted on the fuzzy spaces, looking in particular about the occurrence of the so called non-uniformly ordered phase which is connected with the UV/IR mixing. Since we have constructed our model starting from a renormalizable one this study is very desirable.
Acknowledgements {#acknowledgements .unnumbered}
================
This work has been supported by the Marie Curie Research Training Network MRTN-CT-2006-031962 in Noncommutative Geometry, EU-NCG. I wish here to acknowledge all those who contributed and help me making this work possible. In particular to the Dublin Institute for Advanced Studies To Prof. Raimar Wulkenhaar for all the guidance, and support, I am very grateful for his careful review of my work and his very useful comments.
[99]{}
A. Connes, *Noncommutative Geometry*, Academic Press, Inc. (1994). A. Connes, *Geometry from the spectral point of view*, Lett. Math. Phys. 34 (1995), no. 3, 203–238. A. Connes, *On the spectral characterization of manifolds*,\
(2008) arXiv:0810.2088v1\[math.OA\]. A. Connes, *Noncommutative geometry and reality*, J. Math. Phys. 36 (1995) 6194. V. Gayral, J. H. Jureit, T. Krajewski and R. Wulkenhaar, *Quantum field theory on projectivemodules*, CPT-P67-2006 \[hep-th/0612048\]. A. Connes, *Gravity coupled with matter and the foundation of noncommutative geometry*, Comm. Math. Phys. 155 (1996) 109. A. Rennie Joseph and C. Varilly *Reconstruction of manifolds in noncommutative* *geometry*, (2006) arXiv:math/0610418v4\[math.OA\].
M.R. Douglas, *D-Geometry and Noncommutative Geometry*, \[hep-th/9901146\]. C.P. Martin and D. Sanchez-Ruiz, *The One-loop UV Divergent Structure of U(1) Yang-Mills Theory on Noncommutative $ \mathbb{R}^4$*, Phys.Rev.Lett. 83 (1999) 476-479, \[hep-th/9903077\].A. Connes and J. Lott, *Particle models and noncommutative geometry (expanded version)*, Nucl. Phys. Proc. Suppl. 18B (1991) 29. A. Matusis, L. Susskind and N. Toumbas, *The IR/UV connection in the non-commutative gauge theories*, JHEP 0012 (2000) 002 \[arXiv:hep-th/0002075\]. I. Chepelev, R. Roiban,*Convergence theorem for non-commutative Feynman graphs and renormalization*, JHEP 0103 (2001) 001 \[arXiv:hep-th/0008090\]. T. Krajewski, R. Wulkenhaar,*Perturbative quantum gauge fields on the noncommutative torus*, \[hep-th/9903187\]. V. Rivasseau, F. Vignes-Tourneret, R. Wulkenhaar, *Renormalization of non-commutative $\varphi^4$-theory by multi-scale analysis*, Phys. Lett. B533 (2002) 168–177, \[hep-th/0202039\]. V. Rivasseau, *Constructive Matrix Theory*, arXiv:0706.1224 \[hep-th\]. R. Gurau and V. Rivasseau, *Parametric representation of noncommutative field theory*,Commun. Math. Phys. 272 (2007) 811 \[arXiv:math-ph/0606030\]. V. Gayral, B. Iochum, *The spectral action for Moyal planes*, J. Math. Phys. 46 (2005). 043503 \[arXiv:hep-th/0402147\]. H. Grosse and R. Wulkenhaar, *$8D$-spectral triple on $4D$-Moyal space and the vacuum of noncommutative gauge theory*, arXiv:0709.0095\[hep-ht\]. V. Gayral and R. Wulkenhaar *In preparation*. A. de Goursac, J.-C. Wallet, and R. Wulkenhaar, *On the vacuum states for noncommutative gauge theory*, Eur. Phys. J. C56 (2008) 293–304,arXiv:0803.3035 \[hep-th\]. I. Montvay and G. Münster, *Quantum Field Theory on a Lattice*, Cambridge University Press, (1997). M. E. Newman and G. T. Barkema,*Monte Carlo Methods in Statistical Physics*Oxford University Press (2002). H. Chamseddine and A. Connes, *The Spectral Action Principle*, Math.Phys.186:731-750 (1997). H. Grosse and R. Wulkenhaar, *Renormalisation of $\phi^4$-theory on noncommutative $ \mathbb{R}^4$ in the matrix base*, Commun. Math. Phys. 256 (2005) 305 \[arXiv:hep-th/0401128\]. H. Grosse and R. Wulkenhaar, *Power-counting theorem for non-local matrix models and renormalisation*, \[arXiv:hep-th/0305066\]. V. Gayral, J. M. Gracia-Bondia. Iochum, T. Schücker and J. C. Varilly, *Moyal planes are spectral triples*, Commun. Math. Phys. 246 (2004) 569 \[arXiv:hep-th/0307241\]. V. Gayral, J. M. Gracia-Bondia and F. Ruiz, *Position-dependent noncommutative products: Classical construction and field theory*, Nucl. Phys. B 727 (2005) 513 \[arXiv:hep-th/0504022\]. J.M. Gracia-Bondia, J.C. Varilly, H. Figueroa, *Algebras of distributions suitable for phase-space quantum mechanics*, I, J. Math. Phys. 29, (1988). X. Martin, *A matrix phase for the $\phi^4$ scalar field on the fuzzy sphere*, JHEP 0404 (2004) 077 \[hep-th/0402230\]. J. Medina, *Fuzzy Scalar Field Theories: Numerical* *and Analytical Investigations*, arXiv:0801.1284v1 \[hep-th\]. I. Gradshteyn and I. M. Ryzhik, *Table of integrals, series, and products*, Academic Press, San Diego, 6th ed., (2000). E. Langmann and R. J. Szabo, *Duality in scalar field theory on noncommutative phase spaces*, Phys. Lett. B533 (2002) 168–177, \[hep-th/0202039\].
[^1]: Einstein notation on repeated indices is used.
[^2]: Orientability axiom and Poincaré duality will be not considered
[^3]: $\tilde{\mathcal{T}}_{\mu\nu}$ can be neglected
| {
"pile_set_name": "ArXiv"
} |
IFT-UAM/CSIC-06-29\
[**hep-th/0606201**]{}\
October $19^{\rm th}$, $2006$
[*Jorge Bellorín*]{} [^1], [*Patrick Meessen*]{} [^2], [*and Tom[á]{}s Ortín*]{} [^3]
[*Instituto de Física Teórica UAM/CSIC\
Facultad de Ciencias C-XVI, C.U. Cantoblanco, E-28049-Madrid, Spain*]{}\
[**Abstract**]{}
> We show that requiring unbroken supersymmetry *everywhere* in black-hole-type solutions of $N=2,d=4$ supergravity coupled to vector supermultiplets ensures in most cases absence of naked singularities. We formulate three specific conditions which we argue are equivalent to the requirement of global supersymmetry. These three conditions can be related to the absence of sources for NUT charge, angular momentum, scalar hair and negative energy, although the solutions can still have globally defined angular momentum and non-trivial scalar fields, as we show in an explicit example. Furthermore, only the solutions satisfying these requirements seem to have a microscopic interpretation in String Theory since only they have supersymmetric sources. These conditions exclude, for instance, singular solutions such as the Kerr-Newman with $M=|q|$, which fails to be everywhere supersymmetric.
>
> We also present a re-derivation of several results concerning attractors in $N=2,d=4$ theories based on the explicit knowledge of the most general solutions in the timelike class.
Introduction
============
In spite of the impressive progress made during the last few years in the study of supersymmetric black-hole solutions, there are important questions that remain unanswered or whose answer is unclear. For instance, we know how to construct many supersymmetric black-hole-type solutions, but many of them are singular. Some of these become regular when string corrections are taken into account and for all the regular black hole solutions we seem to have a String Theory model that accounts for its entropy. How are the other singular solutions to be understood? How can it be that they are supersymmetric and yet there is no String Theory model for them? Or, if there is, why are they singular?
The main goal of this paper is to try to answer this question by giving a set of conditions that supersymmetric black-hole-type solutions must satisfy in order to be admissible in the context of $N=2,d=4$ supergravity coupled to vector supermultiplets. Admissible solutions will be regular and will describe one or several black holes in static equilibrium, even though the system may have a finite global angular momentum, as is for example the case in the solution constructed in Ref. [@Elvang:2005sa]. Furthermore, we expect only admissible solutions to have a miscroscopic String Theory model. We will argue that the non-admissible solutions are, in general, not truly supersymmetric in the sense that will be explained later on and the conditions of admissibility can be seen as conditions for a solution to be everywhere supersymmetric. For instance: the Kerr-Newman solution with equal charge and mass, which is singular but nevertheless commonly believed to be supersymmetric, is non-admissible according to our criteria. We will show that it fails to be supersymmetric at the singularity, where the sources might be located. Equivalently we can say that the Kerr-Newman field with $M=|q|$ is caused by non-supersymmetric sources. This explains why it is not described by any supersymmetric String Theory model. We will also show that, generically, rotating sources are not allowed by supersymmetry and that regular, supersymmetric solutions with angular momentum are always composite objects made out of several static black holes in equilibrium. The angular momentum has its origin in the dipole momenta of the electromagnetic fields corresponding to the distribution of charged black holes. Something similar happens for scalar fields: supersymmetric configurations satisfying our conditions can have non-trivial scalar fields but cannot have sources.
In order to prove these results, we will make use of the explicit knowledge of the most general solutions of $N=2,d=4$ supergravity coupled to vector multiplets, which have recently been classified in Ref. [@Meessen:2006tu][^4] . All the asymptotically flat supersymmetric black hole solutions seem to belong to the timelike class, and, although they coincide with the solutions found in Ref. [@Behrndt:1997ny], the general formalism will allow us to make further progress in their understanding. In particular, we will use the *Killing Spinor Identities* (KSIs) [@Kallosh:1993wx; @Bellorin:2005hy], which can be understood as integrability conditions for the Killing spinor equations, in order to study supersymmetry at the singular points where the sources of these solutions should be located.
The final ingredient will be the attractor equations of $N=2,d=4$ supergravity [@Ferrara:1995ih; @Strominger:1996kf; @Ferrara:1996dd; @Ferrara:1996um]: these provide us with information about the sources thought of as being placed at the attractor points. In fact, we will find interesting relations between KSIs and attractor equations, the former showing explicitly that
1. supersymmetry always requires the absence of the kind of scalar hair called *primary* in Ref. [@Coleman:1991ku], and that
2. when the attractor equations are satisfied there are no sources whatsoever for scalar hair.[^5]
These results can be viewed as an extension of those of Ref. [@Kallosh:1992ii] in which it was observed that supersymmetry seems to act as a cosmic censor for static black-hole-type configurations but not for the stationary ones, such as the Kerr-Newman $M=|q|$ solution.
This paper is organized as follows: in Section \[sec-d4BPS\] we review the timelike class of supersymmetric solutions of $N=2,d=4$ supergravity coupled to vector multiplets. First, we review how all the solutions in this class can be constructed from a symplectic vector of real harmonic functions and then in Section \[sec-KSI\] we derive the KSIs that, by assumption of supersymmetry, all solutions must satisfy. In Section \[sec-attractors\] a re-derivation of some of the main results involving $N=2,d=4$ supersymmetric black-hole attractors, taking advantage of the actual and explicit knowledge of all the solutions of this kind, which has helped us to improve some of the presentations existing in the literature and prove some new results.
In Section \[sec-n2d4\] we study how the KSIs constrain the possible sources and singularities of black-hole-type solutions and the interplay with the attractor equations in a general way. The main result of this section will be the formulation of three conditions that express the existence of supersymmetry everywhere in the solutions, including, particularly, the locations of the sources. These conditions should ensure the regularity of the admissible solutions and we study in very close detail several examples in Section \[sec-n2d4bhs\]. Section \[sec-conclusion\] contains our conclusions.
Timelike BPS solutions of $N=2,d=4$ SUEGRA {#sec-d4BPS}
==========================================
It was recently shown in Ref. [@Meessen:2006tu] that all the supersymmetric solutions in the timelike class of $N=2,d=4$ supergravity coupled to $n$ vector multiplets[^6] can be constructed by setting the $2\bar{n}=2(n+1)$ components of a real, symplectic vector $\mathcal{I}=(\mathcal{I}^{\Lambda},\mathcal{I}_{\Lambda})$ equal to $2\bar{n}=2(n+1)$ real functions harmonic on 3-dimensional Euclidean space[^7]
$$\label{eq:stabilization}
\mathcal{I}
\equiv
\left(
\begin{array}{c}
\mathcal{I}^{\Lambda} \\
\mathcal{I}_{\Lambda} \\
\end{array}
\right)
\, ,
\hspace{1cm}
\partial_{m}\partial_{m} \mathcal{I}^{\Lambda}
=
\partial_{m}\partial_{m} \mathcal{I}_{\Lambda}=0\, ,
\hspace{1cm}
\Lambda=0,1,\cdots,n\, .$$
This real section $\mathcal{I}$ enters the theory as the imaginary part of the section $\mathcal{V}/X$, where $\mathcal{V}$ is the covariantly-holomorphic canonical section defining special geometry:
$$\label{eq:SGDefFund}
\mathcal{V} =
\left(
\begin{array}{c}
\mathcal{L}^{\Lambda}\\
\mathcal{M}_{\Sigma}\\
\end{array}
\right) \;\; \rightarrow \;\;
\left\{
\begin{array}{lcl}
\langle \mathcal{V}\mid\mathcal{V}^{*}\rangle
& \equiv &
\mathcal{L}^{*\, \Lambda}\mathcal{M}_{\Lambda}
-\mathcal{L}^{\Lambda}\mathcal{M}^{*}_{\Lambda}
= -i\, , \\
& & \\
\mathfrak{D}_{i^{*}}\mathcal{V} & = & (\partial_{i^{*}}-
{\textstyle\frac{1}{2}}\partial_{i^{*}}\mathcal{K})\mathcal{V} =0 \, ,\\
& & \\
\langle\mathfrak{D}_{i}\mathcal{V}\mid\mathcal{V}\rangle & = & 0 \, .
\end{array}
\right.$$
$X$ on the other hand is proportional to the complex, scalar bilinear constructed out of the Killing spinors: supersymmetry and consistency of the solutions imply that it can be expressed in terms of $\mathcal{I}$, see e.g. Ref. [@Meessen:2006tu] or Eq. (\[eq:MRI\]).
Eqs. (\[eq:stabilization\]) are sometimes known as the *generalized stabilization equations*, the standard stabilization equations having the same form but with the harmonic functions $(\mathcal{I}^{\Lambda},\mathcal{I}_{\Lambda})$ replaced by magnetic and electric charges, [*e.g.*]{} $(p^{\Lambda},q_{\Lambda})$.
The real part of $\mathcal{V}/X$, denoted by $\mathcal{R}\equiv
(\mathcal{R}^{\Lambda},\mathcal{R}_{\Lambda})$ can, in principle, be written in terms of the real harmonic functions, which is usually referred to as “solving the stabilization equations”. In theories with a prepotential, the homogeneity properties of the prepotential allow us to write
$$\label{eq:MdF}
\mathcal{M}_{\Lambda}/X =
\frac{\partial \mathcal{F}(\mathcal{L}^{\cdot}/X)}{\partial (\mathcal{L}^{\Lambda}/X)}\, .$$
Taking the imaginary part of this equation, we have
$$\label{eq:IRHH}
\mathcal{I}_{\Lambda} (\mathcal{R}^{\cdot},\mathcal{I}^{\cdot})
=\mathcal{I}_{\Lambda}\, ,$$
which implicitly defines $\mathcal{R}^{\Lambda}(\mathcal{I}^{\cdot},\mathcal{I}_{\cdot})$, although solving these equations can be extremely hard and in general the explicit solution is unknown.
The real part of Eqs. (\[eq:MdF\]) and the above solutions give straightforwardly the functions $R_{\Lambda}(R^{\cdot}(\mathcal{I}^{\cdot},\mathcal{I}_{\cdot}),
\mathcal{I}^{\cdot})$.
Having the complete symplectic section $\mathcal{V}/X$ entirely given in terms of the real harmonic functions, one can construct the fields of the solutions as follows:
1. The $n$ complex scalar fields $Z^{i}$ are given by the quotients
$$\label{eq:scalars}
Z^{i}= \frac{\mathcal{L}^{i}/X}{\mathcal{L}^{0}/X}=
\frac{\mathcal{R}^{i}+i\mathcal{I}^{i}}{\mathcal{R}^{0}+i\mathcal{I}^{0}}\, .$$
2. The metric has the form
$$\label{eq:metricd4}
ds^{2} = 2|X|^{2}(dt+\omega)^{2} -\frac{1}{2|X|^{2}}dx^{i}dx^{i}\, ,
\hspace{1cm}
i,j=1,2,3\, ,$$
where
$$\label{eq:MRI}
\frac{1}{2|X|^{2}}= \langle\,\mathcal{R}\mid \mathcal{I}\, \rangle\, ,$$
and $\omega$ is a time-independent 1-form on Euclidean 3-dimensional space satisfying the equation
$$\label{eq:oidi}
(d\omega)_{mn} =2\epsilon_{mnp}
\langle\,\mathcal{I}\mid \partial_{p}\mathcal{I}\, \rangle\, .$$
3. The symplectic vector of field strengths and their duals $F=(F^{\Lambda},\tilde{F}_{\Lambda})$ is given by
$$\label{eq:esas}
F = -{\textstyle\frac{1}{2}} \{d[\mathcal{R} \hat{V}]
-{}^{\star}[d\mathcal{I}\wedge \hat{V}] \}\, ,
\hspace{1cm}
\hat{V}=2\sqrt{2}|X|^{2}(dt+\omega)\, .$$
The Killing spinors of these solutions have the form
$$\epsilon_{I}=X^{1/2}\epsilon_{I\, 0}\, ,
\hspace{1cm}
\partial_{\mu}\epsilon_{I\, 0}=0\, ,
\hspace{1cm}
\epsilon_{I\, 0} +i\gamma_{0}
\epsilon_{IJ}\epsilon^{J}{}_{0}=0\, ,$$
which implies
$$\label{eq:constraint}
\epsilon_{I} +i\gamma_{0}e^{i\alpha} \epsilon_{IJ}\epsilon^{J}=0\, ,
\hspace{1cm}
e^{i\alpha}= (X/X^{*})^{1/2}\, .$$
Observe that we can write
$$X = \frac{\mathcal{L}^{\Lambda}(Z,Z^{*})}{\mathcal{R}^{\Lambda}
+i\mathcal{I}^{\Lambda}}\, ,$$
for any $\Lambda$.
Killing Spinor Identities {#sec-KSI}
-------------------------
All supersymmetric configurations satisfy the *Killing Spinor Identities* relating the Einstein equations $\mathcal{E}^{\mu\nu}$, the Maxwell equations $\mathcal{E}_{\Lambda}{}^{\mu}$, the Bianchi identities $\mathcal{B}^{\Lambda\, \mu}$ and the scalar equations of motion $\mathcal{E}^{i}$ [@Kallosh:1993wx; @Bellorin:2005hy; @Meessen:2006tu]
$$\begin{aligned}
\label{eq:ksipsi}
\mathcal{E}_{a}{}^{\mu}\gamma^{a}\epsilon_{I} -4i
\langle\, \mathcal{E}^{\mu} \mid \, \mathcal{V}\, \rangle
\epsilon_{IJ}\epsilon^{J} & = & 0\, ,\\
& & \nonumber \\
\label{eq:ksilambda}
\mathcal{E}^{i} \epsilon^{I} +2i
\langle\, \not\!\mathcal{E} \mid \, \mathcal{U}^{*\, i}\, \rangle
\epsilon^{IJ} \epsilon_{J}& = & 0\, ,\end{aligned}$$
where $\mathcal{E}^{\mu}$ is the symplectic vector $(\mathcal{B}^{\Lambda\,
\mu}, \mathcal{E}_{\Lambda}{}^{\mu})$.
In the timelike case, they lead to the following identities in an orthonormal frame
$$\begin{aligned}
\label{eq:ksi1}
\mathcal{E}^{ab} & = & \eta^{a}{}_{0} \eta^{b}{}_{0}\mathcal{E}^{00}\, ,\\
& & \nonumber \\
\label{eq:ksi2}
\langle\, \mathcal{V}/X \mid \, \mathcal{E}^{a} \, \rangle & = &
{\textstyle\frac{1}{4}}|X|^{-1} \mathcal{E}^{00}\delta^{a}{}_{0}\, ,\\
& & \nonumber \\
\label{eq:ksi3}
\langle\, \mathcal{U}^{*}_{i^{*}}\mid \, \mathcal{E}^{a} \, \rangle & = &
{\textstyle\frac{1}{2}} e^{-i\alpha}\mathcal{E}_{i^{*}}\delta^{a}{}_{0}\, .\end{aligned}$$
These equations imply directly
$$\mathcal{E}^{0m}=0\, ,\,\,\,\,\,
\mathcal{E}^{mn}=0\, ,\,\,\,\,\,
\langle\, \mathcal{V}\mid \, \mathcal{E}^{m} \, \rangle =0\, ,\,\,\,\,\,
\langle\, \mathcal{U}^{*}_{i^{*}}\mid \, \mathcal{E}^{m} \, \rangle=0\, .$$
Further, the r.h.s. of Eq. (\[eq:ksi1\]) is real, and this leads to two important identities:
$$\begin{aligned}
\label{eq:ksi2-2}
\langle\, \mathcal{I} \mid \, \mathcal{E}^{0} \, \rangle & = & 0\, ,\\
& & \nonumber \\
\label{eq:ksi2-1}
\mathcal{E}^{00} & = & \pm 4|\langle\, \mathcal{V} \mid \, \mathcal{E}^{0}\,
\rangle|\, .\end{aligned}$$
Attractor equations {#sec-attractors}
-------------------
It is well-known that, in general, the scalar fields of the black-hole solutions of these theories have certain attractor values that depend solely on the electric and magnetic charges and which are attained at the event horizons irrespectively of the chosen asymptotic values [@Ferrara:1995ih; @Strominger:1996kf].[^8] The attractor values are those which extremize a specific function; furthermore, the absolute value squared of the central charge for the attractor values is essentially the horizon area [@Ferrara:1996dd; @Ferrara:1996um]. Here we are going to rederive these results using our notation and to relate them to the KSIs. We also want to improve the previous derivations by making explicit use of the knowledge of all the supersymmetric configurations.
Let us consider single, static, asymptotically flat, spherically symmetric, black-hole-type solutions of $N=2,d=4$ supergravity coupled to vector multiplets: they are given by real harmonic functions of the form
$$\label{eq:Iforspherical}
\mathcal{I} = \mathcal{I}_{\infty} +\frac{q}{r}\, ,$$
which is the general choice compatible with the assumptions. The metric can be conveniently written in spherical coordinates as
$$ds^{2} = 2|X|^{2}dt^{2} -\frac{1}{2|X|^{2}}[dr^{2}+r^{2}d\Omega^{2}_{(2)}]\, .$$
This metric describes black holes if
$$-g_{rr}=\frac{1}{2|X|^{2}}\stackrel{r\rightarrow \infty}{\longrightarrow}
1+\frac{2M}{r}\, ,$$
is always finite for finite $r$, whence $M$, which is the mass, must be positive. Further, we have to require
$$\frac{1}{2|X|^{2}} \stackrel{r\rightarrow 0}{\longrightarrow}
\frac{A}{4\pi r^{2}}>0\, ,$$
which imposes the existence of an event horizon with area $A> 0$ at $r=0$ instead of a naked singularity.
The existence of attractors (fixed points) of the scalar fields follows from the fact that in supersymmetric configurations, the scalars satisfy first-order differential equations, as follows immediately from the Killing spinor equations associated to the gaugino supersymmetry transformation rule:
$$\label{eq:gaugsusyrule}
\delta_{\epsilon}\lambda^{Ii} =
i\not\!\partial Z^{i}\epsilon^{I}
+\epsilon^{IJ}\not\!G^{i\, +}\epsilon_{J}=0\, .$$
To derive the needed first-order equations, we first use the time-independence of the solutions
$$i\gamma^{m}\partial_{m} Z^{i}\epsilon^{I}
-4\epsilon^{IJ}G^{i\, +}{}_{0m}\gamma^{m}\gamma^{0}\epsilon_{J}=0\, ,$$
and then the known constraint Eq. (\[eq:constraint\]) as to obtain
$$(\partial_{m} Z^{i} -4e^{i\alpha}G^{i\, +}{}_{0m})
\gamma^{m}\epsilon^{I}=0\, ,\,\,\,\,
\Rightarrow \partial_{m} Z^{i} =4e^{i\alpha}G^{i\, +}{}_{0m}\, .$$
Going over to curved indices, the equation takes the form
$$\frac{dZ^{i}}{dr} =2\sqrt{2}G^{i\, +}{}_{tr}/X^{*}\, .$$
The self-duality of $G^{i\, +}$ allows us to express the $G^{i\, +}{}_{tr}$ component in terms of the $G^{i\, +}{}_{\theta\phi}$:
$$G^{i\, +}{}_{tr}= i({}^{\star}G^{i\, +}){}_{\theta\phi}=
-i\frac{2|X|^{2}}{r^{2}\sin{\theta}} G^{i\, +}{}_{\theta\phi}\, ,$$
which combined with
$$G^{i\, +} =\mathcal{T}^{i}{}_{\Lambda}F^{\Lambda\, +}
=
{\textstyle\frac{i}{2}}
\mathcal{G}^{ij^{*}}
\langle\, \mathfrak{D}_{j^{*}}\mathcal{V}^{*} \mid F\, \rangle
=
{\textstyle\frac{i}{2}}
\mathcal{G}^{ij^{*}}
\mathfrak{D}_{j^{*}}\langle\, \mathcal{V}^{*} \mid F\, \rangle\, ,$$
leads to
$$\label{eq:esaecuacion}
\frac{dZ^{i}}{dr} =2\sqrt{2} \frac{X}{r^{2}\sin{\theta}}
\mathcal{G}^{ij^{*}}
\mathfrak{D}_{j^{*}}\langle\, \mathcal{V}^{*} \mid F_{\theta\phi}\, \rangle \, .$$
Since the form of all the fields in terms of $\mathcal{I}(r)$ is in principle known, we can try to find a more explicit form for this equation: using the general form of the vector fields Eq. (\[eq:esas\]) and of $\mathcal{I}(r)$, Eq. (\[eq:Iforspherical\]), we find
$$F_{\theta\phi}
=
{\textstyle \frac{1}{\sqrt{2}}}r^{2}\sin{\theta}\frac{d\mathcal{I}}{dr}
=
-\frac{q}{\sqrt{2}}\sin{\theta}\, .$$
After substituting this into Eq. (\[eq:esaecuacion\]), one ends up with
$$\frac{dZ^{i}}{d\rho} =2 X \mathcal{G}^{ij^{*}}
\mathfrak{D}_{j^{*}}\mathcal{Z}^{*} \, ,$$
where $\rho\equiv 1/r$ and where
$$\mathcal{Z}(Z,q) \equiv \langle\, \mathcal{V} \mid q \, \rangle\, ,$$
is the [*central charge of the theory*]{} [@Ceresole:1995jg]. Observe that the presence of the factor $X$ in the r.h.s. is crucial for it to have zero global Kähler weight, just as the l.h.s. Further observe that the $r$-dependence is only through the scalars $Z^{i}(r)$!
The r.h.s. of this system of differential equations depends only on the scalar fields $Z^{i}$, and, thus, it is an autonomous system of ordinary differential equations[^9] that has fixed points $Z^{i}_{\rm
fix}$ at the values at which the r.h.s. vanishes
$$\label{eq:attractorequations}
\left. \mathfrak{D}_{i}\mathcal{Z}\right|_{Z^{i}=Z^{i}_{\rm fix}}=0\, .$$
If the solution of this system of equations exists, it gives the fixed values of the scalars $Z^{i}_{\rm fix}$ as functions of the electric and magnetic charges only
$$Z^{i}_{\rm fix}= Z^{i}_{\rm fix}(q)\, ,$$
since the asymptotic values (moduli) $Z^{i}_{\infty}$ do not occur in the above differential equation. The fixed values are reached by the scalars at the value $\rho=\infty$, i.e. $r=0$, which is where the event horizon would be, as discussed at the beginning of this section and in what follows.
The fixed values may or may not be admissible, i.e. they may or may not belong to the definition domain of the complex coordinates $Z^{i}$. If the asymptotic values $Z^{i}_{\infty}$ are admissible and the fixed values $Z^{i}_{\rm fix}(q)$ are not, there must be a singularity between $r=\infty$ and $r=0$, which will induce a curvature singularity. We will require both the asymptotic and the fixed values to be admissible. These aspects will be discussed in Section \[sec-n2d4\].
Black-hole solutions whose scalars take the asymptotic values $Z^{i}_{\infty}=Z^{i}_{\rm fix}$ have constant scalar fields, and are called *doubly extreme black holes*. These values are the ones that extremize, not the central charge, but the zero-Kähler-weight combination $e^{\mathcal{K}/2}\mathcal{Z}$:
$$\left. \mathfrak{D}_{i}\mathcal{Z}\right|_{Z^{i}=Z^{i}_{\rm fix}}=
\left.
e^{-\mathcal{K}/2}
\partial_{i}\left(e^{\mathcal{K}/2}\mathcal{Z}\right)\right|_{Z^{i}=
Z^{i}_{\rm fix}}=0\, .$$
### Consequences of the existence of attractors
There are no more scalar fields in the theory, but in the timelike supersymmetric solutions there is another scalar object[^10] that satisfies a first-order differential equation: $X$. From the Killing spinor equation associated to the gravitino supersymmetry transformation rule it is possible to derive [@Meessen:2006tu]
$$\mathfrak{D}_{\mu}X = -i T^{+}{}_{\mu\nu}V^{\nu}\, ,$$
where $V^{\mu}$ is the timelike Killing vector constructed from the Killing spinor. The graviphoton field strength can be written in the form
$$T^{+} = \langle\, \mathcal{V} \mid F \, \rangle\, ,$$
and, together with
$$V^{\nu}F_{\nu\mu}=2 \nabla_{\mu}(|X|^{2}\mathcal{R})\, ,$$
the equation for $X$ becomes
$$\mathfrak{D}_{\mu}X =2 i \langle\, \mathcal{V} \mid
\nabla_{\mu}(|X|^{2}\mathcal{R}) \, \rangle\, .$$
Dividing both sides by $X$ and expanding the r.h.s. using $\mathcal{V}/X=\mathcal{R}+i\mathcal{I}$ we get
$$\frac{\mathfrak{D}_{\mu}X}{X} =2 i |X|^{2} \langle\, \mathcal{R} \mid
\nabla_{\mu} \mathcal{R} \, \rangle
-2\nabla_{\mu}|X|^{2} \langle\, \mathcal{I} \mid \mathcal{R} \, \rangle
-2|X|^{2} \langle\, \mathcal{I} \mid \nabla_{\mu} \mathcal{R} \, \rangle\, .$$
Now, from Eq. (\[eq:MRI\])
$$-2 \langle\, \mathcal{I} \mid \nabla_{\mu} \mathcal{R} \, \rangle =
\nabla_{\mu}|X|^{-2} -2
\langle\, \mathcal{R} \mid \nabla_{\mu} \mathcal{I} \, \rangle\, ,$$
and we get
$$\frac{\mathfrak{D}_{\mu}X}{X} =2 i |X|^{2} \langle\, \mathcal{R} \mid
\nabla_{\mu} \mathcal{R} \, \rangle
-2|X|^{2} \langle\, \mathcal{R} \mid \nabla_{\mu} \mathcal{I} \, \rangle\, .$$
Finally, using
$$\label{eq:RdR=IdI}
\langle\, \mathcal{R} \mid
\nabla_{\mu}\mathcal{R} \, \rangle =
\langle\, \mathcal{I} \mid
\nabla_{\mu}\mathcal{I} \, \rangle\, ,$$
which is proved in Appendix \[sec-proofs\], we arrive at[^11]
$$\label{eq:dxZ}
\mathfrak{D}_{\mu}X^{-1} =2 \langle\, \mathcal{V}^{*} \mid
\nabla_{\mu} \mathcal{I} \, \rangle\, .$$
This equation is valid for all supersymmetric configurations in the timelike class. For those considered in this section we arrive at the equation we were looking for:
$$\mathfrak{D}_{\rho}X^{-1} =2 \mathcal{Z}^{*}\, .$$
The real and imaginary parts of this equation are
$$\begin{aligned}
\label{eq:dgrr}
\frac{d(-g_{rr})}{d\rho} & = & 2 \Re{\rm e}(\mathcal{Z}^{*}/X^{*})
=2 \langle\, \mathcal{R} \mid q \, \rangle\, ,\\
& & \nonumber \\
\frac{d\alpha}{d\rho} +\mathcal{Q}_{\rho} & = & |X|^{2}
-2\Im{\rm m}(\mathcal{Z}^{*}/X^{*})
=2 \langle\, \mathcal{I} \mid q \, \rangle
=2 \langle\, \mathcal{I}_{\infty} \mid q \, \rangle\, .\end{aligned}$$
For the spherically symmetric solutions under consideration $\omega$ vanishes and this requires the phase of $X$ to be covariantly constant, i.e.
$$\label{eq:noNUT}
\langle\, \mathcal{I} \mid q \, \rangle=
\langle\, \mathcal{I}_{\infty} \mid q \, \rangle=0\, .$$
We will later show that this is equivalent to the requirement that the NUT charge vanishes. Since there is only dependence on $\rho$, the phase of $X$ can simply be gauged away by means of a K" ahler transformations. The phase of $\mathcal{Z}$ is then also constant, whence $\mathcal{Z}/X$ is real, which can be used to write
$$\label{eq:dX-1}
\frac{d |X|^{-1}}{d\rho} =\pm 2 |\mathcal{Z}|\, .$$
The $\pm$ sign is the sign of $\langle\, \mathcal{R} \mid q \, \rangle$ and we can argue that it has to be positive if the mass is going to be positive: if we take Eq. (\[eq:dgrr\]) at $\rho=0$ ($r=\infty$), we find that the mass of the solution is given by the linear combination of charges and moduli
$$\label{eq:linearBPSmass}
M= \langle\, \mathcal{R}_{\infty} \mid q \, \rangle\, .$$
Observe that there is no [*a priori*]{} guarantee that $M>0$: this is a condition that has to be imposed independently as to avoid singularities. We will do so and will only consider the positive sign above; Eq. (\[eq:dX-1\]) is then the expression found in the literature.
If we take another derivative of Eq. (\[eq:dgrr\]) and use Eq. (\[eq:dX-1\]), we find
$$\frac{d^{2}(-g_{rr})}{d\rho^{2}} =
2\frac{d |X^{-1}|}{d\rho}|\mathcal{Z}| +2|X|^{-1}\frac{d|\mathcal{Z}|}{d\rho}=
4|\mathcal{Z}|^{2} +2|X|^{-1}\left(\frac{dZ^{i}}{d\rho}\partial_{i}|\mathcal{Z}|
+{\rm c.c.}\right)\, .$$
Now, at $\rho=\rho_{\rm fix}=0$ we have $Z^{i}=Z_{\rm fix}$ and $dZ^{i}/d\rho=0$, and the above equation takes on the form
$$\label{eq:AZfix}
\frac{A}{2\pi} = 4|\mathcal{Z}_{\rm fix}|^{2}\, .$$
Again, there is no [*a priori*]{} guarantee that $|\mathcal{Z}_{\rm fix}|\neq 0$, which therefore is another condition that has to be imposed independently as to avoid singularities. Actually, even though in this expression $A$ is basically an absolute value, the positivity of $A$ is only guaranteed if the scalar fields take admissible values, the mass is positive etc.
These identities allow us to find two interesting expressions for $|\mathcal{Z}_{\rm fix}|$. Expanding the two sides of Eq. (\[eq:dgrr\]) as a power series in $\rho$ we find
$$\frac{A}{2\pi} =
2 \langle\, \left.\frac{d\mathcal{R}}{d\rho}\right|_{\rho=0}
\mid q \, \rangle\, .$$
Using the expressions in Appendix \[sec-proofs\] we get [@Ceresole:1995ca; @Ferrara:1996dd]
$$|\mathcal{Z}_{\rm fix}|^{2}
={\textstyle\frac{1}{2}}
\langle\, \left.\frac{d\mathcal{R}}{d\rho}\right|_{\rho=0}
\mid q \, \rangle
=-{\textstyle\frac{1}{2}}
q^{T}\mathcal{M}(\mathcal{F}_{\rm fix})q\, ,$$
where
$$\mathcal{M}(\mathcal{F}) \equiv
\left(
\begin{array}{cc}
\Im{\rm m}\mathcal{F}
+\Re{\rm e}\mathcal{F}\Im{\rm m}\mathcal{F}^{-1}\Re{\rm e}\mathcal{F} &
-\Im{\rm m}\mathcal{F}^{-1}\Re{\rm e}\mathcal{F} \\
& \\
-\Re{\rm e}\mathcal{F}\Im{\rm m}\mathcal{F}^{-1} & \Im{\rm m}\mathcal{F}^{-1} \\
\end{array}
\right)\, .$$
A direct computation of $|\mathcal{Z}_{\rm fix}|^{2}$ gives
$$|\mathcal{Z}_{\rm fix}|^{2} =
|\langle\, \mathcal{V}_{\rm fix} \mid q \, \rangle|^{2}
=-\langle\, q \mid \mathcal{V}_{\rm fix}\, \rangle
\langle\, \mathcal{V}^{*}_{\rm fix}\mid q\, \rangle\, .$$
The matrix of this bilinear is
$$\mid \mathcal{V}_{\rm fix}\, \rangle
\langle\, \mathcal{V}^{*}_{\rm fix}\mid =
-\left(
\begin{array}{cc}
\mathcal{M}_{\Lambda}\mathcal{M}^{*}_{\Sigma} &
-\mathcal{M}_{\Lambda}\mathcal{L}^{*\, \Sigma} \\
& \\
-\mathcal{L}^{\Lambda}\mathcal{M}^{*}_{\Sigma} &
\mathcal{L}^{\Lambda}\mathcal{L}^{*\, \Sigma} \\
\end{array}
\right)_{\rm fix}\, .$$
We can use the relation
$$\mathcal{L}^{*\, \Lambda}\mathcal{L}^{\Sigma}
=
-\textstyle{1\over 2}\Im{\rm m}(\mathcal{N})^{-1|\Lambda\Sigma}
-f^{\Lambda}{}_{i}\mathcal{G}^{ii^{*}}
f^{*\, \Sigma}{}_{i^{*}}\, ,$$
taking into account that at the fixed point the second term in the r.h.s. will not contribute, and that only its symmetric part will contribute, to get [@Ceresole:1995ca; @Ferrara:1996dd]
$$|\mathcal{Z}_{\rm fix}|^{2}
=-{\textstyle\frac{1}{2}}
q^{T}\mathcal{M}(\mathcal{N}_{\rm fix})q\, .$$
So far we have checked that the coefficient of the $\rho^{2}$ term of $-g_{rr}$ is given by the value of the central charge at the fixed point but, if there are terms of higher order in $\rho$ in $-g_{rr}$ there will not be a regular horizon. We can, however, see that taking another derivative of $-g_{rr}$ w.r.t. $\rho$ at $\rho=0$ will give zero if the attractor equations (\[eq:attractorequations\]) are satisfied and the same will happen for higher derivatives.
Summarizing we can say that the attractor equations (plus the positivity of the mass, which is not guaranteed) seem to be sufficient conditions to have regular, static, spherically symmetric black holes.
Finally, observe that Eq. (\[eq:linearBPSmass\]) plus the identification, which will be established later on, between the NUT charge and the linear expression of the charges
$$\label{eq:linearBPSNUT}
N= \langle\, \mathcal{I}_{\infty} \mid q \, \rangle\, ,$$
lead to a complex BPS relation
$$\label{eq:complexlinearBPS}
M+iN= \langle\, (\mathcal{V}/X)_{\infty} \mid q \, \rangle\, .$$
We will argue that supersymmetry requires $N$ to vanish, whence the above relation reads
$$M= \pm\sqrt{2} |\mathcal{Z}_{\infty}|\, ,$$
which is the standard BPS relation between mass and central charge. Of course, only the positive sign will be admissible.
Relations between the $N=2,d=4$ KSIs, attractors and sources {#sec-n2d4}
============================================================
The equations of motion[^12] for supersymmetric configurations of supergravity theories satisfy certain relations known as *Killing spinor identities* (*KSI*s), which can also be derived from the integrability conditions of the Killing spinor equations [@Kallosh:1993wx; @Bellorin:2005hy]. We have unbroken supersymmetry wherever the Killing spinors exist, and these exist, locally, wherever the KSIs are satisfied. Thus, if we are to have unbroken supersymmetry everywhere we must demand the KSIs to be satisfied everywhere. In this section we are going to study the consequences of demanding the black-hole solutions of $N=2,d=4$ supergravity to be everywhere supersymmetric.
The KSIs of $N=2,d=4$ supergravity are given in Eqs. (\[eq:ksipsi\]) and (\[eq:ksilambda\]) and they lead to Eqs. (\[eq:ksi1\])-(\[eq:ksi2-1\]) for configurations in the timelike class. Since we are going to consider configurations that solve the equations of motion, it may seem that the KSIs are automatically satisfied. However, most solutions have singularities at which the equations of motion are not satisfied, i.e. one has $\mathcal{E}(\phi)=\mathcal{J}(\phi)$. The r.h.s. of the equations of motion at the singularities can be associated to sources for the corresponding fields and the KSIs are then understood as relations between the possible sources of supersymmetric solutions: the KSIs put constraints on possible sources of supersymmetric solutions.
Let us consider from this point of view the KSIs Eqs. (\[eq:ksi1\])-(\[eq:ksi2-1\]): the first of them, Eq. (\[eq:ksi1\]), tells us that the components $\mathcal{E}^{0m}$ and $\mathcal{E}^{mn}$ of the Einstein equations must vanish automatically for supersymmetric configurations and they must do so everywhere if the solutions are everywhere supersymmetric. This means that the sources $\mathcal{J}^{0m}$ and $\mathcal{J}^{mn}$ of the Einstein equation must vanish identically everywhere
$$\label{eq:J0mJmn}
\mathcal{J}^{0m}=\mathcal{J}^{mn}=0\, .$$
Hence, singular (delta-like) sources are not allowed, and in particular this means that no localized sources of angular momentum are allowed.
Any singular contributions to $\mathcal{J}^{0m}$ and $\mathcal{J}^{mn}$ must originate in the $R^{0m}$ components of the Ricci tensor; more precisely, they come from the term $\partial_{\underline{m}}(d\omega)_{\underline{m}\underline{n}}$, where $\omega$ is the 1-form that appears off-diagonally in the metric of the timelike supersymmetric solutions of $N=2,d=4$ supergravity Eq. (\[eq:metricd4\]). Therefore, using Eq. (\[eq:oidi\]) and defining the complex 3-dimensional vector $\vec{\mathcal{W}}$
$$\vec{\mathcal{W}}=
(\mathcal{W}_{\underline{m}})\equiv
(\langle\, \mathcal{V}/X \mid \,
\partial_{\underline{m}}\mathcal{I} \, \rangle)\, ,
\hspace{1cm}
\Im{\rm m}\, (\mathcal{W}_{\underline{m}})=
{\textstyle\frac{1}{4}}
\epsilon_{mnp}
(d\omega)_{\underline{n}\underline{p}}
=\langle\, \mathcal{I} \mid \,
\partial_{\underline{m}}\mathcal{I} \, \rangle\, ,$$
we can translate the above KSIs, Eqs. (\[eq:J0mJmn\]), to the condition
$$\Im{\rm m}\, (\vec{\nabla}\times \vec{\mathcal{W}})=0\, ,$$
which has to be imposed everywhere. Actually, only the singular parts of this equation have to be taken into account since, dealing with solutions, the finite parts must be canceled in the equations of motion by other finite contributions. Therefore, from now on we will ignore all finite contributions to this equation.
Let us consider the real and imaginary parts of Eq. (\[eq:ksi2\]), namely Eq. (\[eq:ksi2-1\]) and (\[eq:ksi2-2\]). The real part gives us two important pieces of information: first, it tells us that the component $\mathcal{J}^{00}$ of the source of the Einstein equation is related to component $\mathcal{J}^{0}$ of the source of the combined Maxwell and Bianchi equations $\mathcal{E}^{a}$. If the electromagnetic fields have only one static point-like source at $r=0$, $\mathcal{E}^{t}\sim
\frac{1}{\sqrt{2}}q\delta^{(3)}(\vec{x})/ \sqrt{|g|}$, then using the fact that $\mathcal{Z}/X$ is real (see Eq. (\[eq:noNUT\]) and the previous discussion)
$$\mathcal{E}^{0t}= \pm 2\sqrt{2} \left.|\mathcal{Z}|\right|_{r=0}
\delta^{(3)}(\vec{x})/ \sqrt{|g|}\, ,$$
which shows that, if the attractor equations are satisfied, the source for the Einstein equations is just $\pm |\mathcal{Z}_{\rm fix}(q)|$. The sign is related to the positivity of $\langle\, \mathcal{R} \mid q \, \rangle$, which is, as was discussed before, associated to the positivity of the mass etc. This is the only value admissible by supersymmetry, since we can understand this source as a source of energy. However, if the scalars take non-admissible values we will find the wrong sign or a zero at $r=0$ and supersymmetry will be broken at the source: we will have to require that the attractor equations are solved by admissible values of the scalars.
The second piece of information we can obtain from the real part concerns the spacelike components of the electromagnetic sources. Combined with the spacelike components of the imaginary part, Eq. (\[eq:ksi2-2\]), we get the condition
$$\label{eq:VJm}
\langle\, \mathcal{V}/X \mid \, \mathcal{J}^{\underline{m}} \, \rangle =0\, .$$
Let us now consider the time component of the imaginary part of the KSI Eq. (\[eq:ksi2\]), Eq. (\[eq:ksi2-2\]):
$$\label{eq:IJt}
\langle\, \mathcal{I} \mid \, \mathcal{J}^{t} \, \rangle= 0\, .$$
To find the physical meaning of this condition we use the explicit form of the symplectic vector of vector field strengths $F$ for timelike BPS solutions Eq. (\[eq:esas\]):
$$\label{eq:jotamu}
\mathcal{J}^{\mu}=
\mathcal{E}^{\mu}=
-({}^{\star}dF)^{\mu}= |X|^{2}\,
(\partial_{\underline{m}}\partial_{\underline{m}}\mathcal{I})\, V^{\mu}=
\frac{\delta^{\mu}{}_{t}}{\sqrt{2}}
\frac{\partial_{\underline{m}}\partial_{\underline{m}}\mathcal{I}}{\sqrt{|g|}}\, .$$
This result tells us that the KSIs Eq. (\[eq:VJm\]) are always satisfied and that the KSI Eq. (\[eq:IJt\]) is equivalent to the condition
$$\label{eq:iddi}
\langle\, \mathcal{I} \mid \,
\partial_{\underline{m}}\partial_{\underline{m}}\mathcal{I} \, \rangle
=
\Im{\rm m}\, (\partial_{\underline{m}}\mathcal{W}_{\underline{m}})
= 0\, ,$$
which is nothing but the integrability condition for the equation determining $\omega$, which now has to be satisfied everywhere as a consequence of demanding unbroken supersymmetry *everywhere*. For the point-like sources considered above, these equations take the form
$$\sum_{A}\langle\, \mathcal{I} \mid \, q_{A} \, \rangle
\delta^{(3)}(\vec{x}-\vec{x}^{A})/ \sqrt{|g|}=0\, .$$
The consequences of imposing this condition were first studied by Denef and Bates in Refs. [@Denef:2000nb; @Bates:2003vx] in the context of general $N=2,d=4$ supergravity, but was studied earlier by Hartle and Hawking in Ref. [@Hartle:1972ya] in the context of Israel-Wilson-Perjés (IWP) solutions of the Einstein-Maxwell theory. As shown by Tod in Ref. [@Tod:1983pm] these are precisely the timelike solutions of pure $N=2,d=4$ supergravity and a special case of the general problem that we are going to study. Hartle and Hawking were motivated, not by supersymmetry, but rather by the prospect of finding regular solutions describing more than one black hole. They were, in particular, worried about possible string singularities related to NUT charges. These singularities can be eliminated by compactifying the time coordinate with certain period [@kn:M], but at the price of losing asymptotic flatness. Let us consider a possible string singularity parametrized by $z$ and choose polar coordinates $\rho,\phi$ around it. If one considers the integral of the 1-form $\omega$ that appears in the metric along a loop of radius $R$ enclosing the possible string singularity at two different points $z_{1}$ and $z_{2}$, denoted by $I(R,z_{1,2})$, one can use Stokes’ theorem to derive
$$I(R,z_{1})-I(R,z_{2}) = \int_{\Sigma^{2}} d\omega = 2\int_{\Sigma^{2}}
\star_{3}\Im{\rm m}\mathcal{W}\, ,$$
where $\Sigma^{2}$ is a surfaces whose boundaries are the loops of radius $R$ at $z_{1,2}$. In the zero radius limit $\Sigma^{2}$ is a closed surface that crosses the possible string singularity at $z_{1}$ and $z_{2}$ and we have
$$2\pi \lim_{R\rightarrow 0} R [\omega_{\phi}(R,z_{1})-\omega_{\phi}(R,z_{2})]
= 2\int_{\Sigma^{2}}
\star_{3}\Im{\rm m}\mathcal{W}= \int_{\Sigma^{3}}
d\star_{3}\Im{\rm m}\mathcal{W}= 2\int_{\Sigma^{3}} d^{3}x
\Im{\rm m}\, (\partial_{\underline{m}}\mathcal{W}_{\underline{m}})\, ,$$
where $\partial\Sigma^{3}=\Sigma^{2}$. Thus, $\Im{\rm m}\,
(\partial_{\underline{m}}\mathcal{W}_{\underline{m}})\neq 0$ implies that $\omega_{\phi}$ is singular on the string somewhere between $z_{1}$ and $z_{2}$. These singularities are related to the presence of NUT sources, since we can define the NUT charge contained in $\Sigma^{3}$ as the integral of $d\omega$ over $\Sigma^{2}=\partial\Sigma^{3}$:
$$-8\pi N_{\Sigma}= \int_{\Sigma^{2}}d\omega = \int_{\Sigma^{3}}d^{2}\omega =
2\int_{\Sigma^{3}} d^{3}x
\Im{\rm m}\, (\partial_{\underline{m}}\mathcal{W}_{\underline{m}})\, .$$
Thus, the condition $\Im{\rm m}\,
(\partial_{\underline{m}}\mathcal{W}_{\underline{m}})=0$, required by supersymmetry, is equivalent to the absence of sources of NUT charge.
Hartle and Hawking argued that the only solutions in the IWP class with no NUT charge (and no singularities) were the Majumdar-Papapetrou solutions [@Majumdar:1947eu; @kn:Pa] which are regular and static. We will review their arguments in Section \[sec-twobhsouren2sugra\] and show that there are indeed non-trivial solutions that satisfy the KSIs and have no NUT charges, apart from the Majumdar-Papapetrou ones; they all have negative total mass, which causes other naked singularities to appear.
Thus, if we include positivity of all masses among the requirements necessary to have supersymmetry, the only supersymmetric black-holes-type solutions of pure $N=2,d=4$ supergravity will indeed be the Majumdar-Papapetrou solutions. We will have to consider more general $N=2,d=4$ theories in order to be able to have stationary solutions such as the one found in Ref. [@Elvang:2005sa], that satisfy the KSIs and have positive mass. This will be done in Section \[sec-generaln2d4\].
Next, let us consider the KSI Eq. (\[eq:ksi3\]) which relates the sources of the scalar fields with those of the vector fields. If we consider only point-like sources and call $\Sigma_{A}$ the scalar charge at $\vec{x}_{A}$, this equation implies, at each sources
$$\Sigma_{A}=2e^{-i\alpha}\left.\mathfrak{D}_{i}\mathcal{Z}\right|_{\vec{x}_{A}}\, .$$
As mentioned before, the scalar sources are completely determined by the electric and magnetic charges and the asymptotic values of the scalar fields. This is known as *secondary scalar hair* [@Coleman:1991ku]. Primary scalar hair correspond to completely free parameters as in the Einstein-scalar solutions of Ref. [@kn:JNW] or in the solutions of Ref. [@Agnese:1994zx] which may be embedded in $N=4,d=4$ supergravity. Neither of these solutions is supersymmetric (nor regular) and the above KSI explains just why.
But there is more to the above KSI: it shows that the existence of attractors at the sources implies total absence of scalar sources, either of primary or secondary type. Since this seems to be necessary in order to have regular event horizons, this KSI implies that there will not be supersymmetric black holes with scalar hair in these theories. Unfortunately, it seems possible to have singular supersymmetric solutions with primary scalar hair.
We can summarize the results obtained in this section as follows: we have identified a series of requirements necessary to avoid singularities in supersymmetric black-hole-type solutions of $N=2,d=4$ supergravity coupled to vector multiplets, which can be associated to having unbroken supersymmetry everywhere (including the sources).
- The conditions
$$\begin{aligned}
\label{eq:dW}
\Im{\rm m}\, (\vec{\nabla}\times \vec{\mathcal{W}})
& = & 0\, ,\\
& & \nonumber \\
\label{eq:d*W}
\Im{\rm m}\, (\vec{\nabla}\cdot \vec{\mathcal{W}})
& = & 0\, , \end{aligned}$$
have to be satisfied everywhere in order to have supersymmetry everywhere. They ensure the absence of string singularities associated to source of NUT charge and other singularities associated to sources of angular momentum We stress that, when dealing with solutions, all finite contributions to the first equation should be ignored and the second equation can only have singular terms in the l.h.s.
- The mass has to be positive. Actually, the masses of each of the sources of the solutions should be positive. They cannot be rigorously defined in general (for multi-black-hole solutions), but they can be identified with certain confidence in the supersymmetric configurations at hands [@kn:BrilLin].
- The attractor equations (\[eq:attractorequations\]) must be satisfied at each of the sources for admissible values of the scalars and the value of the central charge at each of them must be finite. As we have seen, the first condition is equivalent to the total absence of scalar sources.
The last two conditions are associated to the finiteness and positivity of $-g_{rr}$ outside the sources. Since $-g_{rr}\sim e^{-\mathcal{K}}$, it would be finite and positive as long as the scalar fields take admissible values within their domain of definition. All the zeroes of $-g_{rr}$ can be related to singularities of the scalar fields. Imposing that the scalar fields take admissible values everywhere is too strong a condition, since it is almost equivalent to directly impose absence of singularities in the metric.
The conditions that we have imposed are, however, heuristically equivalent: for a single black-hole solution the conditions of asymptotic flatness and positivity of the masses ensure positivity of $-g_{rr}$ in the limit $r\rightarrow \infty$. The third condition ensures positivity in the $r\rightarrow 0$ limit and, furthermore, ensures that there will be a horizon of finite area. Since there are no reasons to expect singularities at finite values of $r$, the positivity and finiteness should hold for all finite values of $r$. The same should happen in multi-black-hole solutions.
$N=2,d=4$ attractors, KSIs and BPS black-hole sources {#sec-n2d4bhs}
=====================================================
Now we want to apply the results of the previous sections to several examples of black-hole-type solutions of $N=2,d=4$ supergravity theories, demanding the three conditions formulated in the introduction and checking the regularity of those solutions that satisfy them. We are going to start with the simplest theory.
Pure $N=2,d=4$ supergravity {#sec-puren2d4}
---------------------------
This theory has $\bar{n}=1$, no scalar fields, and it is given by the prepotential
$$\label{eq:prepotentialpure}
\mathcal{F}=-{\textstyle\frac{i}{2}} (\mathcal{X}^{0})^{2}\, ,\,\,\,
\Rightarrow \mathcal{F}_{0}=-i\mathcal{X}^{0}\, .$$
This implies that the components of the symplectic section $\mathcal{V}$ are constant
$$\mathcal{L}^{0}=i\mathcal{M}_{0}=e^{i\gamma}/\sqrt{2}\, ,$$
and $X$ is not related to any Kähler potential, but
$$X = \frac{e^{i\gamma}}{\sqrt{2}}(\mathcal{L}^{0}/X)^{-1}
= \frac{e^{i\gamma}}{\sqrt{2}(\mathcal{R}^{0}+i\mathcal{I}^{0})}\, .$$
The central charge is constant and given by
$$\mathcal{Z}=-\frac{ie^{i\gamma}}{\sqrt{2}}(p^{0}-iq_{0})\equiv
-\frac{ie^{i\gamma}}{\sqrt{2}}\tilde{q}\, .$$
The attractor equations do not make sense because $\mathcal{Z}$ is already moduli-independent.
The timelike supersymmetric configurations of this theory were first found by Tod in his pioneering paper Ref. [@Tod:1983pm], belong to the family of solutions found by Perjés, Israel and Wilson (IWP) [@Perjes:1971gv; @kn:IW]; they are completely determined by the choice of a single complex, harmonic function that we denote by $\tilde{\mathcal{I}}$. In the framework of general $N=2,d=4$ theories, the solutions of pure $N=2,d=4$ supergravity are given by just two real harmonic functions $\mathcal{I}^{0}$ and $\mathcal{I}_{0}$, the components of the real symplectic vector $\mathcal{I}$. The relation between $\mathcal{I}$ and $\tilde{\mathcal{I}}$ is
$$\tilde{\mathcal{I}}= \mathcal{I}^{0}-i\mathcal{I}_{0}\, .$$
Observe that
$$X = -\frac{ie^{i\gamma}}{\sqrt{2}\tilde{I}}\, ,$$
and therefore $\sqrt{2}X$ coincides with the function $V$ of Ref. [@Tod:1983pm] and is the inverse of the complex harmonic function.
It is convenient to use the complex formulation of this theory. In it, the symplectic product of two real symplectic vectors $x,y$ can be written in the form $\langle\, x\mid y\, \rangle =
\Im{\rm m}\, (\tilde{x}^{*}\tilde{y})$ where the tilde indicates complexification ($\tilde{x}=x^{0}-ix_{0}$ etc.). Further, electric-magnetic duality rotations of the symplectic vectors is equivalent to multiplication by a global phase $\tilde{x}^{\prime}=e^{i\gamma}\tilde{x}$. We would like to stress that the metric is invariant under these transformations.
Using Eq. (\[eq:prepotentialpure\]) one finds that $\mathcal{R}$, the real part of $\mathcal{V}/X$ is the symplectic vector
$$\mathcal{R}=
\left(
\begin{array}{c}
-\mathcal{I}_{0} \\ \mathcal{I}^{0} \\
\end{array}
\right)\, ,\,\,\, \Rightarrow
\tilde{\mathcal{R}} = -i\tilde{\mathcal{I}}\, ,
\,\,\, \Rightarrow -g_{rr}= \frac{1}{2|X|^{2}}=
\langle\, \mathcal{R} \mid \mathcal{I}\, \rangle =
|\tilde{\mathcal{I}}|^{2}\, .$$
Finally,
$$\vec{\mathcal{W}} = \tilde{\mathcal{I}}^{*}\vec{\nabla}\tilde{\mathcal{I}}\, .$$
It was argued by Hartle and Hawking [@Hartle:1972ya] that the only regular black hole solutions in the IWP family are the static Majumdar-Papapetrou solutions that describe several charged black holes in static equilibrium. We are going to see that these are in fact the only solutions which are everywhere supersymmetric (condition I) and that demanding positivity of the masses of the components (condition II) is enough to have regular black holes (condition III plays no rôle here).
### Single, static black hole solutions
The complex harmonic function $\tilde{\mathcal{I}}$ adequate to describe a static, spherically symmetric, extreme black hole with magnetic and electric charges $p^{0}$ and $q_{0}$ is
$$\tilde{\mathcal{I}}=\tilde{\mathcal{I}}_{\infty} +\frac{\tilde{q}}{r}\, ,
\hspace{1cm}
\tilde{q}\equiv p^{0}-iq_{0}\, ,$$
and asymptotic flatness requires $|\tilde{\mathcal{I}}_{\infty}|=1$. Since $\tilde{\mathcal{I}}_{\infty}$ is just a phase that can be taken to be unity by an electric-magnetic duality rotation. Then,
$$-g_{rr}=|\tilde{\mathcal{I}}|^{2}=1
+\frac{2\Re{\rm e}(\tilde{\mathcal{I}}^{*}_{\infty}\tilde{q} )}{r}
+\frac{|\tilde{q}|^{2}}{r^{2}}\, .$$
The mass is given by
$$\label{eq:themass}
M= \Re{\rm e} (\tilde{\mathcal{I}}_{\infty}^{*}\tilde{q}) =
\langle\, \mathcal{R}_{\infty}\mid q\, \rangle\, ,$$
and the equations of motion and supersymmetry seem to allow for it to be positive or negative. When $M$ is negative $|\tilde{I}|^{2}$ will vanish for some finite value of $r$, giving rise to a naked singularity. In the limit $r\rightarrow 0$, which makes sense if $M$ is positive, we find that the area of the 2-spheres of constant $t$ and $r$ is finite and equal to
$$A = 4\pi |\tilde{q}|^{2}=8\pi |\mathcal{Z}|^{2}\, .$$
Observe that, in general,
$$|M| \neq \sqrt{2}|\mathcal{Z}|\, ,$$
even though these solutions are usually understood to be supersymmetric.
For this solution Eq. (\[eq:dW\]) is automatically satisfied, while Eq. (\[eq:d\*W\]) takes the form
$$\Im{\rm m}\, (\vec{\nabla}\cdot \vec{\mathcal{W}})
= -4\pi \Im{\rm m}\, (\tilde{\mathcal{I}}_{\infty}^{*}\tilde{q})\,
\delta^{(3)}(\vec{x})=0\, .$$
We can, either
1. Adopt the point of view proposed in this paper that the integrability condition has to be satisfied everywhere (condition I), whence impose the condition
$$\label{eq:hq}
\Im{\rm m}\, (\tilde{\mathcal{I}}_{\infty}^{*}\tilde{q})=
\langle\, \mathcal{I}_{\infty} \mid q\, \rangle = 0\, .$$
$\tilde{\mathcal{I}}_{\infty}$ is just a phase and this condition determines it: $\tilde{\mathcal{I}}_{\infty}= \pm
\tilde{q}/|\tilde{q}|\equiv e^{i\beta}$. The complex harmonic function becomes
$$\tilde{\mathcal{I}}
= e^{i\beta}\left(1 \pm \frac{|\tilde{q}|}{r} \right)\, ,$$
The overall phase $e^{i\beta}$ is irrelevant for our problem (it can always be eliminated by an electric-magnetic duality rotation that does not change the metric), but the relative sign between the two terms, which is the sign of the mass,
$$M= \pm |\tilde{q}|=\pm |\mathcal{Z}|\, ,$$
is important since the minus sign leads to naked singularities. We take the positive sign as to comply with condition II. We can the integrate the equation for $\omega$ everywhere. The above condition, however, implies the vanishing of the r.h.s. of the equation and, therefore, also that of $\omega$. Thus, after imposing conditions I and II we obtain a solution which is static and spherically symmetric and has a regular horizon if $M>0$; Or
2. We can accept this singularity, ignoring condition I, arguing that, after all, the harmonic functions are already singular at that point[^13] and proceed to integrate the equation and obtain $\omega$ which, in spherical coordinates, takes the form
$$\omega= 2N \cos{\theta}d\phi\, ,$$
where $N$ is NUT charge and it is given by
$$\label{eq:NUT}
N= \Im{\rm m}\, (\tilde{\mathcal{I}}_{\infty}^{*}\tilde{q})=
\langle\, \mathcal{I}_{\infty} \mid q\, \rangle\, ,\,\,\,
\Rightarrow |M+iN|=\sqrt{2} |\mathcal{Z}_{\infty}|\, .$$
The metric is no longer static, but stationary, and contains either wire singularities or closed timelike curves plus Taub-NUT asymptotics.
It is clear that by imposing conditions I and II, these pathologies are avoided. Furthermore, in the microscopic models of black holes constructed in the framework of String Theory there seem to be no configurations that give rise to macroscopic NUT charge (nor to negative masses). The agreement between spacetime supersymmetry and the microscopic String Theory models on this point, together with the elimination of pathologies is encouraging and we will see that it applies to more cases.
### Single black hole solutions with a dipole term
Let us now consider harmonic functions adequate to describe rotating supersymmetric black holes. We can add angular momentum to the previous solution by adding a dipole term to its complex harmonic function which becomes:
$$\tilde{\mathcal{I}}=\tilde{\mathcal{I}}_{\infty} +\frac{\tilde{q}}{r}
+(\vec{\tilde{m}}\cdot\vec{\nabla})\frac{1}{r}\, ,$$
where $\vec{m}=(\vec{m}^{0},\vec{m}_{0})$ is a symplectic vector of dipole magnetic and electric momenta. When they are parallel we can take them to have only $z$ component and, then, in spherical coordinates
$$\tilde{\mathcal{I}}=\tilde{\mathcal{I}}_{\infty} +\frac{\tilde{q}}{r}
-\frac{\tilde{m}\cos{\theta}}{r^{2}}\, .$$
The corresponding $\omega$ (which exists except at the singularities of $\tilde{\mathcal{I}}$) is
$$\omega= \left[2N \cos{\theta} +2J\frac{\sin^{2}{\theta}}{r^{2}}
+\Im{\rm m}(\tilde{q}^{*}\tilde{m}) \frac{\sin^{2}{\theta}}{r^{3}} \right] d\phi\, .$$
$N$ is the NUT charge and is given again by Eq. (\[eq:NUT\]). The new features are $J$, the $z$ component of the angular momentum, given by
$$\label{eq:J}
J = \Im{\rm m}\, (\tilde{\mathcal{I}}_{\infty}^{*}\tilde{m})=
\langle\, \mathcal{I}_{\infty} \mid m, \rangle\, ,$$
and $\Im{\rm m}(\tilde{q}^{*}\tilde{m})$ which does not have a conventional name but vanishes when $N=J=0$.
Let us now analyze the KSIs Eqs. (\[eq:dW\]) and (\[eq:d\*W\]) (condition I). In the general case they take, respectively, the form
$$\begin{aligned}
2\left[\Im{\rm m}\,(\tilde{q}^{*} \nabla_{m})\vec{\nabla}\frac{1}{r}\right]
\times\vec{\nabla}\frac{1}{r}
-i\left(\nabla_{m^{*}}\vec{\nabla}\frac{1}{r}\right)
\times \left(\nabla_{m}\vec{\nabla}\frac{1}{r}\right)
& = & 0\, ,\\
& & \nonumber \\
\Im{\rm m}\, (\tilde{\mathcal{I}}_{\infty}^{*}\tilde{q})\, \delta^{(3)}(\vec{x})
+
\Im{\rm m}\,(\tilde{\mathcal{I}}_{\infty}^{*}\nabla_{m})\,
\delta^{(3)}(\vec{x})
+\frac{1}{r}\Im{\rm m}\,(\tilde{q}^{*}\nabla_{m})\,
\delta^{(3)}(\vec{x})
+ & & \nonumber \\
& & \nonumber \\
+\delta^{(3)}(\vec{x})
\Im{\rm m}\,(\tilde{q}\nabla_{m^{*}})\frac{1}{r}
+\Im{\rm m}\,\left\{\left(\nabla_{m^{*}}\frac{1}{r}\right)
\left(\nabla_{m}\delta^{(3)}(\vec{x})\right)\right\} & = & 0\, ,\end{aligned}$$
and are satisfied if
$$\begin{aligned}
N & = & \Im{\rm m}\, (\tilde{\mathcal{I}}_{\infty}^{*}\tilde{q})=
\langle\, \mathcal{I}_{\infty} \mid q\, \rangle =0\, , \\
& & \nonumber \\
\vec{J} & = & \Im{\rm m}\, (\tilde{\mathcal{I}}_{\infty}^{*}\vec{\tilde{m}})=
\langle\, \mathcal{I}_{\infty} \mid \vec{m}\, \rangle =0\, ,\\
& & \nonumber \\
& &
\Im{\rm m}\, (\tilde{q}^{*}\vec{\tilde{m}})=
\langle\, q \mid \vec{m}\, \rangle =0\, ,\\
& & \nonumber \\
& &
\Im{\rm m}\, (\tilde{m}^{*}_{[\underline{m}}\tilde{m}_{\underline{n}]})=
\langle\, m_{[\underline{m}} \mid m_{\underline{n}]}\, \rangle =0\, ,\end{aligned}$$
where we have defined the differential operator $\nabla_{m}\equiv
\vec{\tilde{m}}\cdot\vec{\nabla}$ and where we have taken into account Eq. (\[eq:J\]) to identify the angular momentum.
The first condition is, again, the absence of sources of NUT charge. The second condition is the absence of sources of angular momentum. The third and fourth conditions are automatically satisfied in this theory if the first two are.
In this case, these conditions are not enough to eliminate all the singularities introduced by the dipole term since the above conditions do not cancel terms like $|\vec{\tilde{m}}\cdot\vec{\nabla}\frac{1}{r}|^{2}$ in the $g_{rr}$ component of the metric and we no longer find a regular 2-sphere in the $r\rightarrow 0$ limit. However, we are going to argue that, although technically possible, dipole terms should not be allowed in $\mathcal{I}$ because their only possible origin is a distribution of point-like charges and it is the fundamental distribution of point-like charges that we have to consider in the above equations and not the field they produce at distances larger than its size. It is in these conditions that imposing supersymmetry everywhere is equivalent to cosmic censorship.
Indeed, from the point of view of the electromagnetic fields, the magnetic dipole momenta, for instance, can have two fundamental origins: dipole momenta in a distribution of magnetic monopoles or fundamental dipole momenta that can be seen as stationary electric currents. In standard electrodynamics the first possibility is experimentally excluded (see, e.g. Ref. [@Jackson:1977iu]) but in $N=2,d=4$ supersymmetric configurations it is the only one allowed (see Eq. (\[eq:jotamu\])).
### The supersymmetric Kerr-Newman solution
Therefore we must only consider distributions of static point-like charges. We will do so in a moment, but there is an interesting example of rotating black-hole-type solution which must be considered before: it is given by the complex harmonic function
$$\tilde{\mathcal{I}}= \tilde{\mathcal{I}}_{\infty}
+\frac{\tilde{q}}{\tilde{r}}\, ,
\hspace{1cm}
\tilde{r}\equiv \sqrt{x^{2}+y^{2} +(z-i\alpha)^{2}}\, ,$$
which is known to lead to the (“ultra-extreme”) supersymmetric Kerr-Newman solution with angular momentum around the $z$ axis; as is known it has naked singularities, as all 4-dimensional supersymmetric rotating “black-holes” [@Bergshoeff:1996gg]. This is the prototype of solution for which supersymmetry does not act as a “cosmic censor” as proposed in [@Kallosh:1992ii]. Generalizations of this solution in some other $N=2,d=4$ theories have been constructed in Ref. [@Behrndt:1997ny].
The asymptotic expansion of $\tilde{I}$
$$\tilde{\mathcal{I}} \sim \tilde{\mathcal{I}}_{\infty}
+\frac{\tilde{q}}{r} -\frac{i\alpha\tilde{q}z}{r^{3}}+\cdots\, ,$$
corresponds to a charge distribution with only two independent parameters: $\alpha$ and $\tilde{q}$. The magnetic (electric) dipole momentum is equal to the product of $\alpha$ and the electric (magnetic) charge and the infinite number of non-vanishing higher momenta depend also on these few parameters.
According to the point of view advocated here this solution should not be considered because it corresponds to the far field of a very charge distribution. As we are going to see, condition I is enough to exclude it.
Finding the sources of the solution associated to the above complex harmonic function is very complicated. To start with, $\tilde{\mathcal{I}}$ is singular on the ring $x^{2}+y^{2}=\alpha^{2}\, ,\,\,\,z=0$ but it is also discontinuous on a disk bounded by the ring (see e.g. [@Kaiser:2001yn], whose results we are going to use here. See also Refs. [@Newman:2002mk; @Gsponer:2004gv].).
Eqs. (\[eq:dW\]) and (\[eq:d\*W\]), which express condition I, take, respectively, the form
$$\begin{aligned}
\Im{\rm m}\, (\tilde{\mathcal{I}}_{\infty}^{*}\tilde{q})
\Re{\rm e}\, (\vec{\nabla}\times \vec{C})
+\Re{\rm e}\, (\tilde{\mathcal{I}}_{\infty}^{*}\tilde{q})
\Im{\rm m}\, (\vec{\nabla}\times \vec{C})
+|\tilde{q}|^{2} \Im{\rm m}\,\left(\frac{1}{\tilde{r}^{*}}
\vec{\nabla}\times \vec{C}\right)
& = & 0\, ,\\
& & \nonumber \\
\Im{\rm m}\, (\tilde{\mathcal{I}}_{\infty}^{*}\tilde{q})
\Re{\rm e}\, (\vec{\nabla}\cdot\vec{C})
+\Re{\rm e}\, (\tilde{\mathcal{I}}_{\infty}^{*}\tilde{q})
\Im{\rm m}\, (\vec{\nabla}\cdot \vec{C})
+|\tilde{q}|^{2} \Im{\rm m}\,\left(\frac{1}{\tilde{r}^{*}}
\vec{\nabla}\cdot \vec{C}\right)
& = & 0\, , \end{aligned}$$
where we have defined
$$\vec{C}\equiv \frac{(x,y,z-i\alpha)}{[x^{2}+y^{2} +(z-i\alpha)^{2}]^{3/2}} \, .$$
The curl and divergence of $\vec{C}$ have been carefully computed in Ref. [@Kaiser:2001yn] in a distributional sense, i.e. as integrals of their products with test functions. For us it is enough to known that
$$\Re{\rm e}\, (\vec{\nabla}\times \vec{C})= \Im{\rm m}\, (\vec{\nabla}\cdot
\vec{C})=0\, ,$$
and that $\Im{\rm m}\, (\vec{\nabla}\times \vec{C})$ vanishes for vanishing $\alpha$. We are left with
$$\begin{aligned}
\left[\Re{\rm e}\, (\tilde{\mathcal{I}}_{\infty}^{*}\tilde{q})
+|\tilde{q}|^{2} \Re{\rm e}\, \frac{1}{\tilde{r}}\right]
\Im{\rm m}\, (\vec{\nabla}\times \vec{C})
& = & 0\, ,\\
& & \nonumber \\
\left[\Im{\rm m}\, (\tilde{\mathcal{I}}_{\infty}^{*}\tilde{q})
-|\tilde{q}|^{2} \Im{\rm m}\,\frac{1}{\tilde{r}}
\right]
\Re{\rm e}\, (\vec{\nabla}\cdot\vec{C})
& = & 0\, .\end{aligned}$$
The only way to satisfy the first condition is to have $\Im{\rm m}\,
(\vec{\nabla}\times \vec{C})=0$, which requires $\alpha=0$ (no sources of angular momentum). Since $\Re{\rm e}\,
(\vec{\nabla}\cdot\vec{C})\neq 0$ always, the only way to satisfy the second condition is to have $\Im{\rm m}\,
(\tilde{\mathcal{I}}_{\infty}^{*}\tilde{q})=0$ as before (no sources of NUT charge) and $\Im{\rm m}\,\frac{1}{\tilde{r}}=0$ which also requires $\alpha=0$.
Thus, imposing supersymmetry everywhere is equivalent, yet again, to requiring absence of sources of NUT charge and angular momentum. In the supersymmetric Kerr-Newman solution all the angular momentum originates in that source[^14] and, thus, that solution and its naked singularities can be excluded from the class of everywhere supersymmetric solutions of $N=2,d=4$ supergravity. Again, supersymmetry acts as a cosmic censor and, most importantly, there is agreement between the macroscopic description of black holes provided by Supergravity and the microscopic models provided by String Theory in which there seems to be no way of having angular momentum without breaking supersymmetry.
Therefore, we must only consider distributions of point-like charges, which correspond to complex harmonic functions of the form
$$\tilde{\mathcal{I}}=\tilde{\mathcal{I}}_{\infty}
+\sum_{A}\frac{\tilde{q}_{A}}{|\vec{x}-\vec{x}_{A}|}\, ,$$
from which dipole (and higher) momenta arise only in asymptotic expansions:
$$\tilde{\mathcal{I}}\sim \tilde{\mathcal{I}}_{\infty}
+\frac{\sum_{A}\tilde{q}_{A}}{|\vec{x}|}
+\frac{(\sum_{A}\tilde{q}_{A}\vec{x}_{A})\cdot\vec{x}}{|\vec{x}|^{3}}+\cdots\, ,$$
and may give rise to non-vanishing angular momentum
$$\vec{J} = \Im{\rm m}\, (\tilde{\mathcal{I}}_{\infty}^{*}\vec{\tilde{m}})=
\langle\, \mathcal{I}_{\infty} \mid \vec{m}\, \rangle \, ,
\hspace{1cm}
\vec{m} = -\sum_{A}q_{A}\vec{x}_{A}\, ,$$
but not to non-vanishing NUT charge.
$$N = \Im{\rm m}\, (\tilde{\mathcal{I}}_{\infty}^{*}\tilde{q})=
\langle\, \mathcal{I}_{\infty} \mid q \, \rangle =0\, ,
\hspace{1cm}
q = \sum_{A}q_{A}\, .$$
We are going to look for this kind of solutions in pure $N=2,d=4$ supergravity next, recovering the (negative) Hartle and Hawking result [@Hartle:1972ya]. We will have to look for them in more general $N=2,d=4$ theories.
### Solutions with two black holes {#sec-twobhsouren2sugra}
Let us consider, to start with, just two poles
$$\tilde{\mathcal{I}}= \tilde{\mathcal{I}}_{\infty}
+\frac{\tilde{q}_{1}}{|\vec{x}-\vec{x}_{1}|}
+\frac{\tilde{q}_{2}}{|\vec{x}-\vec{x}_{2}|}\, .$$
Asymptotic flatness requires $|\tilde{\mathcal{I}}_{\infty}|=1$. The condition Eq. (\[eq:dW\]) is automatically satisfied and (\[eq:d\*W\]) takes the form
$$\left[\langle\, \mathcal{I}_{\infty}\mid q_{1}\, \rangle\,
+\frac{\langle\, q_{2}\mid q_{1}\, \rangle\,}{|\vec{x}_{1}
-\vec{x}_{2}|}\right]\delta^{(3)}(\vec{x}-\vec{x}_{1})
+
\left[\langle\, \mathcal{I}_{\infty} \mid q_{2}\, \rangle\,
+\frac{\langle\, q_{1}\mid q_{2}\, \rangle\,}{|\vec{x}_{2}
-\vec{x}_{1}|}\right]\delta^{(3)}(\vec{x}-\vec{x}_{2}) = 0\, ,$$
which leads to the two equations
$$\label{eq:twoequations}
\begin{array}{rcl}
\langle\, \mathcal{I}_{\infty}\mid q_{1}\, \rangle\,
+{\displaystyle\frac{\langle\, q_{2}\mid q_{1}\, \rangle\,}{|\vec{x}_{1}
-\vec{x}_{2}|}} & = & 0\, ,\\
& & \\
\langle\, \mathcal{I}_{\infty} \mid q_{2}\, \rangle\,
+{\displaystyle\frac{\langle\, q_{1}\mid q_{2}\, \rangle\,}{|\vec{x}_{2}
-\vec{x}_{1}|}} & = & 0\, , \\
\end{array}$$
each of which expresses the absence of sources of NUT charge at $\vec{x}_{1}$ and $\vec{x}_{2}$. The antisymmetry of the symplectic product implies the consistency condition
$$\label{eq:hq1q2}
\langle\, \mathcal{I}_{\infty}\mid q_{1}+q_{2}\, \rangle\, =0\, ,$$
which means that the total charge of the two objects satisfies the same condition (no global NUT charge) as the charge of just one.
Expanding asymptotically $\mathcal{I}$ and using the above constraints we find that this two-body system has a total mass and angular momentum given by
$$\begin{aligned}
M & = & \sum_{A} \langle\, \mathcal{R}_{\infty}\mid q_{A}\, \rangle \equiv \sum_{A} M_{A}\, ,\\
& & \nonumber \\
\label{eq:angularmomentum}
\vec{J}
& = &
\langle\, \mathcal{I}_{\infty} \mid \vec{m}\, \rangle
=
\langle\, q_{1} \mid q_{2}\, \rangle \frac{(\vec{x}_{2}-\vec{x}_{1})}{|\vec{x}_{2}-\vec{x}_{1}|}\, .\end{aligned}$$
Observe that there is total angular momentum even though there are no sources of angular momentum.
There are two types of solutions to these equations required by condition I:
1. Each object’s charge satisfies the condition for single independent objects $\langle\, \mathcal{I}_{\infty} \mid q_{A}\, \rangle =0$ which requires $\langle\, q_{2}\mid q_{1}\, \rangle=0$. In this theory this means that the phases of $\tilde{\mathcal{I}}_{\infty},\tilde{q}_{1}$ and $\tilde{q}_{2}$ are such that
$$\tilde{I} = e^{i\beta}\left(1
+\sum_{A}\frac{s_{A}|\tilde{q}_{A}|}{|\vec{x}-\vec{x}_{A}|}
\right)\, ,$$
where $s_{A}=\pm 1$. The total mass is given by the formula Eq. (\[eq:themass\])
$$M= \Re{\rm e} (\tilde{\mathcal{I}}_{\infty}^{*}\sum_{A}\tilde{q}_{A}) =
\langle\, \mathcal{R}_{\infty}\mid \sum_{A}q_{A}\, \rangle
= \sum_{A}s_{A}|\tilde{q}_{A}|\, ,$$
and the angular momentum vanishes ($\omega$ vanishes).
These are the Majumdar-Papapetrou solutions [@Majumdar:1947eu; @kn:Pa]. Only the solutions with all $s_{A}=+1$ are regular, but one could argue that only those correspond to objects that would have positive masses $M_{A}=|\tilde{q}_{A}|$ if they were isolated [@kn:BrilLin]. This is the meaning of condition II.
These solutions describe two charged, static black holes in equilibrium with their event horizons placed at $\vec{x}_{1}$ and $\vec{x}_{2}$ which are really 2-spheres of finite areas equal to $4\pi|\tilde{q}_{1}|^{2}$ and $4\pi|\tilde{q}_{2}|^{2}$. They are, as argued by Hartle and Hawking, and as we are going to see, the only regular black-hole-type solutions in the whole IWP family [@Hartle:1972ya]
2. $\langle\, \mathcal{I}_{\infty} \mid q_{A}\, \rangle \neq 0$ and we have two objects that cannot exist independently in the vacuum $\mathcal{I}_{\infty}$ (i.e. we have a bound state). The distance between them is fixed by the condition of absence of sources of NUT charge to be
$$|\vec{x}_{2}-\vec{x}_{1}| =
{\displaystyle\frac{\langle\, q_{1}\mid q_{2}\, \rangle}
{\langle\, \mathcal{I}_{\infty}\mid q_{1}\, \rangle}}\, .$$
The sign of the r.h.s. can always be made positive by flipping the sign of $\mathcal{I}_{\infty}$, which is irrelevant for the moduli and for solving Eq. (\[eq:hq1q2\]). Thus, this equation always has a solution. However, when all the above conditions have been satisfied, the total mass of the solution is negative. The simplest way to see this is by first making $\tilde{\mathcal{I}}_{\infty}=1$ by a duality rotation that does not change the metric. After the duality rotation one finds $\tilde{q}^{\prime}_{A}=M_{A}+iN_{A}$, meaning that they are complex combinations of the masses and NUT charges of each object. Using $N_{2}=-N_{1}$, the above condition takes the form
$$N_{1} +\frac{N_{1}M_{2}-N_{2}M_{1}}{|\vec{x}_{2}-\vec{x}_{1}|}=
N_{1}\left(1 +\frac{M_{1}+M_{2}}{|\vec{x}_{2}-\vec{x}_{1}|}\right)=0\, ,$$
which has solution only for vanishing NUT charges or for negative total mass $M_{1}+M_{2}$ which violates condition II and produces naked singularities. Thus, we cannot simultaneously satisfy conditions I and II for bound states with $\langle\, q_{1}\mid q_{2}\, \rangle \neq 0$.
This result can be generalized to solutions with more poles: let us consider first the 3-pole harmonic function
$$\tilde{\mathcal{I}}= \tilde{\mathcal{I}}_{\infty}
+\frac{\tilde{q}_{1}}{|\vec{x}-\vec{x}_{1}|}
+\frac{\tilde{q}_{2}}{|\vec{x}-\vec{x}_{2}|}
+\frac{\tilde{q}_{3}}{|\vec{x}-\vec{x}_{3}|}\, .$$
The $\omega$ integrability condition leads to three equations (one to cancel the NUT charge at each pole) which can be written as a linear system for the $N_{A}$s:
$$\left(
\begin{array}{ccc}
\left(1+\frac{M_{2}}{r_{12}}+\frac{M_{3}}{r_{14}}\right) & -\frac{M_{1}}{r_{12}} & -\frac{M_{1}}{r_{13}} \\[3mm]
-\frac{M_{2}}{r_{12}} & \left(1+\frac{M_{1}}{r_{12}}+\frac{M_{3}}{r_{23}}\right) & -\frac{M_{2}}{r_{23}} \\[3mm]
-\frac{M_{3}}{r_{13}} & -\frac{M_{3}}{r_{23}} & \left(1+\frac{M_{1}}{r_{13}}+\frac{M_{2}}{r_{23}}\right) \\
\end{array}
\right)
\left(
\begin{array}{c}
N_{1} \\ \\ N_{2} \\ \\ N_{3} \\
\end{array}
\right) =0\, .$$
It is easy to see that the determinant of the matrix is $+1$ plus terms linear and quadratic in the masses, all with positive sign. It will never vanish if all the masses are positive. This argument can be easily generalized to a higher number of poles and, therefore we conclude that the only solutions satisfying conditions I and II are the Majumdar-Papapetrou solutions. This result should be read in a positive sense: no singular solutions are allowed by the conditions proposed in the introduction, even if only static solutions are allowed in this simple theory. To find solutions with angular momentum satisfying conditions I-III we need to consider theories with scalars.
General $N=2,d=4$ supergravity {#sec-generaln2d4}
------------------------------
The setup of our problem in general $N=2,d=4$ theories is similar to pure supergravity case. Let us first consider spherically-symmetric, static, single black-hole-type solutions with magnetic and electric charges $p^{\Lambda}$ and $q_{\Lambda}$. They are determined by a symplectic vector of $2\bar{n}$ real harmonic functions
$$\mathcal{I}=
\left(
\begin{array}{c}
\mathcal{I}^{\Lambda} \\ \mathcal{I}_{\Lambda} \\
\end{array}
\right)
=
\mathcal{I}_{\infty} +\frac{q}{r}\, ,\
\hspace{1cm}
q\equiv
\left(
\begin{array}{c}
p^{\Lambda} \\ q_{\Lambda} \\
\end{array}
\right)\, ,
\hspace{1cm}
\mathcal{I}_{\infty}
\equiv
\left(
\begin{array}{c}
\mathcal{I}_{\infty}^{\Lambda} \\ \mathcal{I}_{\Lambda\, \infty} \\
\end{array}
\right)\, .$$
We assume that the stabilization equations have been solved and $\mathcal{R}(\mathcal{I})$ has been found in order to be able to construct the fields of the solutions.
The $n$ complex scalars are constructed using the general formula Eq. (\[eq:scalars\]). The moduli (the values of the $n$ complex scalars $Z^{i}$ at infinity, $Z^{i}_{\infty}$) are complicated functions $Z^{i}_{\infty}(\mathcal{I}_{\infty})$ of these $2n+2$ real constant components of $\mathcal{I}_{\infty}$. One of the components of $\mathcal{I}_{\infty}$ can be determined as a function of the remaining $2n+1$ by imposing asymptotic flatness of the metric, that is, $\langle\,
\mathcal{R}_{\infty} \mid \mathcal{I}_{\infty}\, \rangle=1$, and another one can be determined by imposing condition I, since Eq. (\[eq:iddi\]) implies
$$N=\langle\, \mathcal{I}_{\infty}\mid q\, \rangle=0\, .$$
It should always be possible to give the $2n$ real moduli any admissible value within their definition domain with the remaining $2n$ unconstrained real components of $\mathcal{I}_{\infty}$. This is difficult to prove explicitly due to the complicated and theory-dependent relations between $\mathcal{I}_{\infty}$ and the moduli $Z^{i}_{\infty}$, but it is safe to assume that in general it is possible.
Let us turn to condition II. The positivity of the masses, which is given by the general expression Eq. (\[eq:linearBPSmass\]) has to be imposed by hand and, although this can always be done, it is a non-trivial constraint on the charges and moduli. The positivity of the masses can be also understood as part of a stronger requirement that the scalar fields take values only within their definition domain for all values of $r$. Actually, this requirement should suffice to ensure the finiteness of $-g_{rr}$ for $r\neq 0$.
The finiteness of $-g_{rr}$ for $r\neq 0$ is not enough to have a black hole and condition III has to be imposed to find a finite horizon area at $r=0$.
If we want to describe more than one black hole we have to use harmonic functions with two point-like singularities:
$$\mathcal{I}=\mathcal{I}_{\infty} +\frac{q_{1}}{|\vec{x}-\vec{x}_{1}|}
+\frac{q_{2}}{|\vec{x}-\vec{x}_{2}|}\, .$$
Again, one of the components of $\mathcal{I}_{\infty}$ is determined by imposing asymptotic flatness. Condition I now leads to the two equations Eqs. (\[eq:twoequations\]) which should determine another component of $\mathcal{I}_{\infty}$ and the parameter $|\vec{x}_{1}-\vec{x}_{2}|$ if $\langle q_{2} \mid q_{1}\rangle \neq 0$. The question now is whether these solutions can be obtained while maintaining the positivity of the masses (condition II)
$$M_{i} \equiv \langle\, \mathcal{R}_{\infty} \mid q_{i}\, \rangle >0\, ,$$
and solving the attractor equations for each of the singularities of the harmonic functions. We have no general answer to these questions and, what we are going to do is to study how the three conditions can actually be imposed in a particularly simple example and suffice to ensure regularity of the solutions.
### A toy model with a complex scalar field {#sec-toy}
We are going to consider the $\bar{n}=2$ theory with prepotential
$$\mathcal{F}=-i\mathcal{X}^{0}\mathcal{X}^{1}\, .$$
This theory has only one complex scalar
$$\tau\equiv i\mathcal{X}^{1}/\mathcal{X}^{0}\, ,$$
in terms of which the period matrix is given by
$$(\mathcal{N}_{\Lambda\Sigma})=
\left(
\begin{array}{cc}
-\tau & 0 \\
0 & 1/\tau \\
\end{array}
\right)$$
and, in the $\mathcal{X}^{0}=i/2$ gauge, the Kähler potential and metric are
$$\mathcal{K}=-\ln{\Im{\rm m}\tau}\, ,
\hspace{1cm}
\mathcal{G}_{\tau\tau^{*}} = (2\Im{\rm m}\tau)^{-2}\, .$$
The reality of the Kähler potential requires the positivity of $\Im{\rm
m}\tau$. Therefore, $\tau$ parametrizes the coset $SL(2,\mathbb{R})/SO(2)$ and can be identified with the *axidilaton* and this theory is a truncation of the $SO(4)$ formulation of $N=4,d=4$ supergravity.
The symplectic section $\mathcal{V}$ is
$$\mathcal{V}=\frac{1}{2(\Im {\rm m}\tau)^{1/2}}
\left(
\begin{array}{c}
i \\ \tau \\ -i\tau \\ 1 \\
\end{array}
\right)\, ,$$
and the central charge is
$$\mathcal{Z}(\tau,\tau^{*},q)= \langle \mathcal{V} \mid q \rangle =
\frac{1}{2(\Im {\rm m}\tau)^{1/2}}[(p^{1}-iq_{0}) -(q_{1}+ip^{0})\tau]\, .$$
The attractor equation is
$$\left. \frac{d}{d\tau} \frac{1}{\Im {\rm m}\tau}[(p^{1}-iq_{0})
-(q_{1}+ip^{0})\tau]\right|_{\tau=\tau_{\rm fix}}=0\, ,$$
and has the general solution
$$\tau_{\rm fix} = \frac{p^{1}+iq_{0}}{q_{1}-ip^{0}}\, ,$$
which is admissible (belongs to the definition domain of $\tau$) if
$$\Im{\rm m}\tau_{\rm fix} = p^{0}p^{1}+q_{0}q_{1} > 0\, .$$
The central charge at the fixed point of the scalar takes the value
$$\mathcal{Z}_{\rm fix} = -i \frac{q_{1}+ip^{0}}{|q_{1}+ip^{0}|}
\sqrt{p^{0}p^{1}+q_{0}q_{1}}\, ,$$
and it is always finite for $\tau_{\rm fix} \neq 0$.
Let us now consider solutions with
$$\mathcal{I}=\mathcal{I}_{\infty} +\frac{q}{r}\, .$$
In this theory the stabilization equations can be easily solved and they lead to
$$\mathcal{R}=
\left(
\begin{array}{cc}
0 & -\sigma^{1} \\
\sigma^{1} & 0 \\
\end{array}
\right)\mathcal{I}\, ,
\,\,\,\,
\Rightarrow
\,\,\,\,
-g_{rr}=\langle\, \mathcal{R} \mid \mathcal{I}\, \rangle =
2(\mathcal{I}^{0}\mathcal{I}^{1}+\mathcal{I}_{0}\mathcal{I}_{1})\, ,$$
which shows that the area of the horizon (if any) is related to $|\mathcal{Z}_{\rm fix}|^{2}$ above according to the general formula Eq. (\[eq:AZfix\]).
We also have
$$\tau = i\frac{\mathcal{L}^{1}/X}{\mathcal{L}^{0}/X}=
\frac{\mathcal{I}^{1}+i\mathcal{I}_{0}}{\mathcal{I}_{1}-i\mathcal{I}^{0}}\, ,$$
which implies that the 4 harmonic functions are not entirely independent but have to satisfy
$$\label{eq:esacondicion}
\Im{\rm m}\tau = \mathcal{I}^{0} \mathcal{I}^{1}
+\mathcal{I}_{0} \mathcal{I}_{1} > 0\, ,$$
which ensures that, if there are no pathologies that make a black-hole interpretation of the solution impossible, the attractor equations will always have solutions and $\mathcal{Z}_{\rm fix}\neq 0$. Thus, we will not have to worry about condition III but only about the positive definiteness of $\Im{\rm
m}\tau$.
The only possible pathologies (negative mass and presence of NUT charge) are clearly avoided by imposing conditions I and II, which is always possible and presents no difficulties.
Let us now consider solutions of the form
$$\mathcal{I}=\mathcal{I}_{\infty} +\frac{q_{1}}{r_{1}}
+\frac{q_{2}}{r_{2}}\, ,
\hspace{1cm}
r_{i}\equiv |\vec{x}-\vec{x}_{i}|\, .$$
Our goal is to find a configuration (i.e. a set of asymptotic values $\mathcal{I}_{\infty}$ and charges $q_{1,2}$) that satisfy conditions I-III. The previous discussions indicate how this has to be done and which formulas need to be applied. There is no systematic procedure to find such a configuration but it is not too difficult to find one:
$$\begin{array}{rcl}
\mathcal{I}^{0} & = & \,\,\,\,\,\,
{\displaystyle
\frac{1}{\sqrt{2}}
+\frac{q}{r_{1}}+\frac{q}{r_{2}}
}\, ,\\
& & \\
\mathcal{I}^{1} & = & \,\,\,\,\,\,
{\displaystyle
\frac{1}{\sqrt{2}}
+\frac{8q}{r_{1}}+\frac{8q}{r_{2}}
}\, ,\\
& & \\
\mathcal{I}_{0} & = & \hspace{2.2cm}
{\displaystyle-\frac{4q}{r_{2}}
}\, ,\\
& & \\
\mathcal{I}_{1} & = &
{\displaystyle
-\frac{1}{4\sqrt{2}}
-\frac{q}{r_{1}}+\frac{q}{r_{2}}
}\, ,\\
\end{array}$$
where $q>0$ in order to guarantee Eq. (\[eq:esacondicion\]). The metric component
$$-g_{rr}= 1 +\frac{9\sqrt{2}q}{r_{1}}
+\frac{10\sqrt{2}q}{r_{2}}
+\frac{16q^{2}}{r_{1}^{2}}
+\frac{8q^{2}}{r_{2}^{2}}
+\frac{40q^{2}}{r_{1}r_{2}}\, ,$$
is finite everywhere outside $r_{1,2}=0$, and therefore, so is $\Im{\rm
m}\tau$. In particular the “mass” of each of the two objects is positive
$$M_{1}=9q/\sqrt{2}\, ,
\hspace{1cm}
M_{2}=5\sqrt{2}q\, ,
\hspace{1cm}
M=M_{1}+M_{2}=19q/\sqrt{2}\, ,$$
and in the $r_{1,2}\rightarrow 0$ limits we find spheres of finite areas
$$\frac{A_{1}}{4\pi} = 16q^{2}= 2|\mathcal{Z}_{\rm fix,1}|^{2}\, ,
\hspace{1cm}
\frac{A_{2}}{4\pi} = 8q^{2} = 2|\mathcal{Z}_{\rm fix,2}|^{2}\, .$$
The total horizon area is
$$\frac{A}{4\pi}=\frac{A_{1}}{4\pi}+\frac{A_{2}}{4\pi} = 24q^{2} <
2|\mathcal{Z}_{\rm fix, tot}|^{2}=64q^{2}\, ,$$
which is the area of the horizon of a single black hole having the sum of the charges of the two black holes.
For this configuration
$$\langle\, \mathcal{I}_{\infty} \mid q_{1}\, \rangle =
-\langle\, \mathcal{I}_{\infty} \mid q_{2}\, \rangle =-q/\sqrt{2}\, ,
\hspace{1cm}
\langle\, q_{2} \mid q_{1}\, \rangle= 12q^{2}\, ,$$
so, choosing
$$r_{12}=|\vec{x}_{2}-\vec{x}_{1}|=12\sqrt{2}q\, ,$$
we satisfy condition I (no NUT charges). The system has nevertheless angular momentum given by the general formula Eq. (\[eq:angularmomentum\]):
$$|J|= |\langle\, q_{2} \mid q_{1}\, \rangle| = 12q^{2}\, .$$
Conclusions {#sec-conclusion}
===========
We have formulated three conditions that supersymmetric black-hole-type solutions have to satisfy in order to be supersymmetric everywhere, including at the sources. We have shown how these conditions constrain the possible sources by, basically, excluding those with NUT charge, angular momentum, negative energy and scalar hair, which seemingly cannot be modeled in String Theory. We arrived at a picture in which if an observer far away from one of the globally supersymmetric configurations we have considered, detects angular momentum and non-trivial scalar fields he/she will only find static electromagnetic sources in equilibrium when approaching the system.
These conditions and this picture should be improved by considering quantum corrections. Another interesting course of action would be to consider regularity of black-hole solutions in $N>2$ theories, e.g. [@Cvetic:1995uj], and investigate the rôle played by the attractor [@Ferrara:1996um].
It is also clear that the situation in $d=5$ is completely different as there are regular rotating supersymmetric black holes for which microscopic String Theory models are known [@Breckenridge:1996is]. Work on these issues is already in progress [@kn:BMO].
Acknowledgments {#acknowledgments .unnumbered}
===============
T.O. would like to thank Renata Kallosh for pointing the authors towards references [@Denef:2000nb] and [@Bates:2003vx] and for useful conversations, Roberto Emparan for his explanations concerning rotating black-hole configurations and many other useful comments and, finally, M.M. Fernández for her long standing support.
This work has been supported in part by the Spanish Ministry of Science and Education grant BFM2003-01090 and the Comunidad de Madrid grant HEPHACOS P-ESP-00346.
Proofs of some identities {#sec-proofs}
=========================
Let us consider the generalized stabilization equations derived from Eq. (\[eq:MdF\]). Differentiating the imaginary part of that equation (i.e. Eq.(\[eq:IRHH\]), we get
$$d\mathcal{I}_{\Lambda}= d\Im{\rm m}\mathcal{F}_{\Lambda} =
{\textstyle\frac{1}{2i}}(d\mathcal{X}^{\Lambda}\mathcal{F}_{\Sigma\Lambda}
-d\mathcal{X}^{*\, \Lambda}\mathcal{F}^{*}_{\Sigma\Lambda})=
d\mathcal{R}^{\Sigma}\Im{\rm m}\mathcal{F}_{\Sigma\Lambda}
+d\mathcal{I}^{\Sigma}\Re{\rm e}\mathcal{F}_{\Sigma\Lambda}\, ,$$
where we have used $\mathcal{X}^{\Lambda}=\mathcal{R}^{\Lambda}+i\mathcal{I}^{\Lambda}$. Using the invertibility of the imaginary part of $\mathcal{F}_{\Sigma\Lambda}$ we get
$$d\mathcal{R}^{\Sigma} =
\Im{\rm m}\mathcal{F}^{\Sigma\Lambda}d\mathcal{I}_{\Lambda}
-\Im{\rm m}\mathcal{F}^{\Sigma\Omega}
\Re{\rm e}\mathcal{F}_{\Omega\Lambda}d\mathcal{I}^{\Lambda}\, .$$
On the other hand, differentiating the real part of Eq. (\[eq:MdF\])
$$d\mathcal{R}_{\Lambda}= d\Re{\rm e}\mathcal{F}_{\Lambda}=
{\textstyle\frac{1}{2}}(d\mathcal{X}^{\Lambda}\mathcal{F}_{\Sigma\Lambda}
+d\mathcal{X}^{*\, \Lambda}\mathcal{F}^{*}_{\Sigma\Lambda})=
d\mathcal{R}^{\Sigma}\Re{\rm e}\mathcal{F}_{\Sigma\Lambda}
-d\mathcal{I}^{\Sigma}\Im{\rm m}\mathcal{F}_{\Sigma\Lambda}\, ,$$
and, substituting our previous result for $d\mathcal{R}^{\Lambda}$
$$d\mathcal{R}_{\Sigma}= \Re{\rm e}\mathcal{F}_{\Sigma\Omega}
\Im{\rm m}\mathcal{F}^{\Omega\Lambda}dH_{\Lambda}
-(\Im{\rm m}\mathcal{F}_{\Sigma\Lambda}
+\Re{\rm e}\mathcal{F}_{\Sigma\Omega}\Im{\rm m}\mathcal{F}^{\Omega\Delta}
\Re{\rm e}\mathcal{F}_{\Delta\Lambda})d\mathcal{I}^{\Lambda}\, .$$
We can write all these results in the form
$$\begin{aligned}
d\mathcal{R} & = &
\left(
\begin{array}{cc}
-\Im{\rm m}\mathcal{F}^{-1}\Re{\rm e}\mathcal{F} & \Im{\rm m}\mathcal{F}^{-1}
\\
& \\
-(\Im{\rm m}\mathcal{F}
+\Re{\rm e}\mathcal{F}\Im{\rm m}\mathcal{F}^{-1}\Re{\rm e}\mathcal{F}) &
\Re{\rm e}\mathcal{F}\Im{\rm m}\mathcal{F}^{-1}
\end{array}
\right)
d\mathcal{I}\, , \\
& & \nonumber \\
& & \nonumber \\
d\mathcal{I} & = &
\left(
\begin{array}{cc}
\Im{\rm m}\mathcal{F}^{-1}\Re{\rm e}\mathcal{F} & -\Im{\rm m}\mathcal{F}^{-1}
\\
& \\
\Im{\rm m}\mathcal{F}
+\Re{\rm e}\mathcal{F}\Im{\rm m}\mathcal{F}^{-1}\Re{\rm e}\mathcal{F} &
-\Re{\rm e}\mathcal{F}\Im{\rm m}\mathcal{F}^{-1}
\end{array}
\right)
d\mathcal{R}\, , \end{aligned}$$
from which we can read identities such as
$$\begin{array}{rclclrclcl}
{\displaystyle
\frac{\partial \mathcal{R}^{\Sigma}}{\partial \mathcal{I}_{\Lambda}}
}
& = &
{\displaystyle
\frac{\partial \mathcal{R}^{\Lambda}}{\partial \mathcal{I}_{\Sigma}}
}
& = &
{\displaystyle
\frac{\partial \mathcal{I}^{\Lambda}}{\partial \mathcal{R}_{\Sigma}}\, ,
}
\hspace{1.5cm} &
{\displaystyle
\frac{\partial \mathcal{R}_{\Sigma}}{\partial \mathcal{I}^{\Lambda}}
}
& = &
{\displaystyle
\frac{\partial \mathcal{R}_{\Lambda}}{\partial \mathcal{I}^{\Sigma}}
}
& = &
{\displaystyle
-\frac{\partial \mathcal{I}_{\Lambda}}{\partial \mathcal{R}^{\Sigma}}
\, ,
}\\
& & & & & & & \\
{\displaystyle
\frac{\partial \mathcal{R}^{\Sigma}}{\partial \mathcal{I}^{\Lambda}}
}
& = &
{\displaystyle
-\frac{\partial \mathcal{R}_{\Lambda}}{\partial \mathcal{I}_{\Sigma}}
}
& = &
{\displaystyle
\frac{\partial \mathcal{I}_{\Lambda}}{\partial \mathcal{R}_{\Sigma}}
\, ,
}
&
{\displaystyle
\frac{\partial \mathcal{R}_{\Sigma}}{\partial \mathcal{I}_{\Lambda}}
}
& = &
{\displaystyle
-\frac{\partial \mathcal{R}^{\Lambda}}{\partial \mathcal{I}^{\Sigma}}
}
& = &
{\displaystyle
\frac{\partial \mathcal{I}^{\Lambda}}{\partial \mathcal{R}^{\Sigma}}\, .
}
\\
\end{array}$$
We can now prove Eq. (\[eq:dRI=RdI\]): taking the derivative of $\mathcal{R}$ as a function of $\mathcal{I}$ we have
$$\begin{array}{rcl}
\langle\, \nabla_{\mu}\mathcal{R} \mid
\mathcal{I} \, \rangle & = &
{\displaystyle
\langle\,
\frac{\partial \mathcal{R}}{\partial \mathcal{I}^{\Lambda}}\nabla_{\mu} \mathcal{I}^{\Lambda}
+\frac{\partial \mathcal{R}}{\partial \mathcal{I}_{\Lambda}}\nabla_{\mu} \mathcal{I}_{\Lambda}
\mid \mathcal{I} \, \rangle
}
\\
& & \\
& = &
{\displaystyle
\nabla_{\mu}\mathcal{I}^{\Lambda}
\left(
\mathcal{I}^{\Sigma}
\frac{\partial \mathcal{R}_{\Sigma}}{\partial \mathcal{I}^{\Lambda}}
-
\mathcal{I}_{\Sigma}
\frac{\partial \mathcal{R}^{\Sigma}}{\partial \mathcal{I}^{\Lambda}}
\right)
+
\nabla_{\mu}\mathcal{I}_{\Lambda}
\left(
\mathcal{I}^{\Sigma}
\frac{\partial \mathcal{R}_{\Sigma}}{\partial \mathcal{I}_{\Lambda}}
-
\mathcal{I}_{\Sigma}
\frac{\partial \mathcal{R}^{\Sigma}}{\partial \mathcal{I}_{\Lambda}}
\right)\, ,
}
\\
\end{array}$$
and using now the above relations between partial derivatives
$$\langle\, \nabla_{\mu}\mathcal{R} \mid
\mathcal{I} \, \rangle =
\nabla_{\mu}\mathcal{I}^{\Lambda}
\left(
\mathcal{I}^{\Sigma}
\frac{\partial \mathcal{R}_{\Lambda}}{\partial \mathcal{I}^{\Sigma}}
+
\mathcal{I}_{\Sigma}
\frac{\partial \mathcal{R}_{\Lambda}}{\partial \mathcal{I}_{\Sigma}}
\right)
-
\nabla_{\mu}\mathcal{I}_{\Lambda}
\left(
\mathcal{I}^{\Sigma}
\frac{\partial \mathcal{R}^{\Lambda}}{\partial \mathcal{I}^{\Sigma}}
+
\mathcal{I}_{\Sigma}
\frac{\partial \mathcal{R}^{\Lambda}}{\partial \mathcal{I}_{\Sigma}}
\right)\, .$$
Given that the real section $\mathcal{R}$ is homogeneous of first order in the $\mathcal{I}$’s
$$\mathcal{I}^{\Sigma}
\frac{\partial \mathcal{R}_{\Lambda}}{\partial \mathcal{I}^{\Sigma}}
+
\mathcal{I}_{\Sigma}
\frac{\partial \mathcal{R}_{\Lambda}}{\partial \mathcal{I}_{\Sigma}}
= \mathcal{R}_{\Lambda}\, ,
\hspace{2cm}
\mathcal{I}^{\Sigma}
\frac{\partial \mathcal{R}^{\Lambda}}{\partial \mathcal{I}^{\Sigma}}
+
\mathcal{I}_{\Sigma}
\frac{\partial \mathcal{R}^{\Lambda}}{\partial \mathcal{I}_{\Sigma}}
= \mathcal{R}^{\Lambda}\, ,$$
which proves the identity.
Similarly, expanding the r.h.s. of Eq. (\[eq:RdR=IdI\]) we get
$$\langle\, \mathcal{R} \mid \nabla_{\mu}\mathcal{R} \, \rangle =
\left(
\frac{\partial\mathcal{R}^{\Lambda}}{\partial\mathcal{I}^{\Sigma}}
R_{\Lambda}
-
\frac{\partial\mathcal{R}_{\Lambda}}{\partial\mathcal{I}^{\Sigma}}
R^{\Lambda}
\right)d\mathcal{I}^{\Sigma}
+
\left(
\frac{\partial\mathcal{R}^{\Lambda}}{\partial\mathcal{I}_{\Sigma}}
R_{\Lambda}
-
\frac{\partial\mathcal{R}_{\Lambda}}{\partial\mathcal{I}_{\Sigma}}
R^{\Lambda}
\right)d\mathcal{I}_{\Sigma}\, ,$$
and using the identities between partial derivatives and the fact that the real section $\mathcal{I}$ is homogeneous of first order in $\mathcal{R}$, we arrive at the result we wanted.
[99]{}
H. Elvang, R. Emparan, D. Mateos and H.S. Reall, JHEP [**0508**]{} (2005) 042 \[[arXiv:[hep-th/0504125]{}]{}\].
P. Meessen and T. Ortín, Nucl. Phys. B [**749**]{} (2006) 291 \[[arXiv:[hep-th/0603099]{}]{}\].
M. Hübscher, P. Meessen and T. Ortín, arXiv:hep-th/0606281.
K. Behrndt, D. Lüst and W.A. Sabra, Nucl. Phys. B [**510**]{} (1998) 264 \[[arXiv:[hep-th/9705169]{}]{}\]. R. Kallosh and T. Ortín, [arXiv:[hep-th/9306085]{}]{}. J. Bellorín and T. Ortín, Phys. Lett. B [**616**]{} (2005) 118 \[[arXiv:[hep-th/0501246]{}]{}\]. S. Ferrara, R. Kallosh and A. Strominger, Phys. Rev. D [**52**]{} (1995) 5412 \[[arXiv:[hep-th/9508072]{}]{}\]. A. Strominger, Phys. Lett. B [**383**]{} (1996) 39 \[[arXiv:[hep-th/9602111]{}]{}\]. A. Ceresole, R. D’Auria and S. Ferrara, Nucl. Phys. Proc. Suppl. [**46**]{} (1996) 67 \[[arXiv:[hep-th/9509160]{}]{}\]. S. Ferrara and R. Kallosh, Phys. Rev. D [**54**]{} (1996) 1514 \[[arXiv:[hep-th/9602136]{}]{}\]. S. Ferrara and R. Kallosh, Phys. Rev. D [**54**]{} (1996) 1525 \[[arXiv:[hep-th/9603090]{}]{}\]. S.R. Coleman, J. Preskill and F. Wilczek, Nucl. Phys. B [**378**]{} (1992) 175 \[[arXiv:[hep-th/9201059]{}]{}\]. G.W. Moore, [arXiv:[hep-th/9807087]{}]{}.
R. Kallosh, A.D. Linde, T. Ortín, A.W. Peet and A. Van Proeyen, Phys. Rev. D [**46**]{} (1992) 5278 \[[arXiv:[hep-th/9205027]{}]{}\]. A. Ceresole, R. D’Auria, S. Ferrara and A. Van Proeyen, Nucl. Phys. B [**444**]{} (1995) 92 \[[arXiv:[hep-th/9502072]{}]{}\]. F. Denef, JHEP [**0008**]{} (2000) 050 \[arXiv:[arXiv:[hep-th/0005049]{}]{}\]. B. Bates and F. Denef, [arXiv:[hep-th/0304094]{}]{}. J.B. Hartle and S.W. Hawking, Commun. Math. Phys. [**26**]{} (1972) 87. K.P. Tod, Phys. Lett. B [**121**]{} (1983) 241. C. Misner, J. Math. Phys. **4** (1963) 924.
S.D. Majumdar, Phys. Rev. [**72**]{} (1947) 390. A. Papapetrou, Proc. Roy. Irish. Acad. **A51** (1947) 191.
A.I. Janis, E.T. Newman and J. Winicour, Phys. Rev. Lett. **20** (1968) 878.
A.G. Agnese and M. La Camera, Phys. Rev. D [**49**]{} (1994) 2126. D.R. Brill and R.W. Lindquist, Phys. Rev. **131** (1963) 471-476.
Z. Perjés, Phys. Rev. Lett. [**27**]{} (1971) 1668. W. Israel and G.A. Wilson, J. Math. Phys. **13**, (1972) 865.
J.D. Jackson, Yellow Report CERN-77-17 [SPIRES entry](http://www.slac.stanford.edu/spires/find/hep/www?r=cern-77-17) Reprinted in V. Stefan and V.F. Weisskopf, eds., Physics and Society: Essays in Honor of Victor Frederick Weisskopf (AIP Press, New York, Springer, Berlin, 1998) 236pp.
E. Bergshoeff, R. Kallosh and T. Ortín, Nucl. Phys. B [**478**]{} (1996) 156 \[[arXiv:[hep-th/9605059]{}]{}\]. G. Kaiser, J. Phys. A [**37**]{} (2004) 8735 \[[arXiv:[gr-qc/0108041]{}]{}\].
E.T. Newman, Phys. Rev. D [**65**]{} (2002) 104005 \[[arXiv:[gr-qc/0201055]{}]{}\]. A. Gsponer, [arXiv:[gr-qc/0405046]{}]{}. M. Cvetič and D. Youm, Phys. Rev. D [**53**]{} (1996) 584 \[[arXiv:[hep-th/9507090]{}]{}\]: M. Cvetič and A.A. Tseytlin, Phys. Rev. D [**53**]{} (1996) 5619 \[Erratum-ibid. D [**55**]{} (1997) 3907\] \[[arXiv:[hep-th/9512031]{}]{}\]: M. Cvetič and C.M. Hull, Nucl. Phys. B [**480**]{} (1996) 296 \[[arXiv:[hep-th/9606193]{}]{}\]. J.C. Breckenridge, R. C. Myers, A. W. Peet and C. Vafa, Phys. Lett. B [**391**]{} (1997) 93 \[[arXiv:[hep-th/9602065]{}]{}\]. J. Bellorín, P. Meessen and T. Ortín, (to appear).
[^1]: E-mail: [[email protected]]{}
[^2]: E-mail: [[email protected]]{}
[^3]: E-mail: [[email protected]]{}
[^4]: In this paper we will not consider the coupling to hypermultiplets. The classification of the supersymmetric solutions with both vector multiplets and hypermultiplets is considered in Ref. [@kn:HMO].
[^5]: If there is more than one basin of attraction, contrary to what is assumed in this article, this last conclusion might change due to the [*area codes*]{} [@Moore:1998pn].
[^6]: These solutions were first found in slightly different form in Ref. [@Behrndt:1997ny] and the procedure followed in Ref [@Meessen:2006tu] shows that they are the only solutions in this class.
[^7]: If the functions are not harmonic, the field configurations are still supersymmetric, but are [*not*]{} solutions of the equations of motion.
[^8]: If there are multiple attractor regions, it might happen that there is some residual dependency on the asymptotic values. Here we assume there to be only one attractor region.
[^9]: The use of the variable $\rho=1/r$ is essential in this argument. it is easy to see that the derivatives of the scalar fields of typical black-hole solutions w.r.t. to $r$ do not vanish at $r=0$, while their derivatives w.r.t. $\rho$ do..
[^10]: In previous derivations in the literature the absolute value $|X|= e^{U}$ is considered, but then the Kähler weights and the reality properties of the two sides of the equations derived are different.
[^11]: Observe that the compatibility between Eq. (\[eq:MRI\]) and the following equations requires the identity $$\label{eq:dRI=RdI}
\langle\, \nabla_{\mu}\mathcal{R} \mid
\mathcal{I} \, \rangle =
\langle\, \mathcal{R} \mid
\nabla_{\mu}\mathcal{I} \, \rangle\, ,$$ to hold. For theories admitting a prepotential, this is done in Appendix \[sec-proofs\].
[^12]: By *equation of motion* $\mathcal{E}(\phi)$ of a given field $\phi$ we will mean here the l.h.s. of the equation of motion $\delta S/\delta\phi=\mathcal{E}(\phi)=0$. This slight abuse of language should lead to no confusions.
[^13]: We have seen that the solution can, nevertheless, be regular at that point, which is the event horizon.
[^14]: We are going to see that there are solutions with angular momentum and no elementary sources of angular momentum.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Online social systems are multiplex in nature as multiple links may exist between the same two users across different social networks. In this work, we introduce a framework for studying links and interactions between users beyond the individual social network. Exploring the cross-section of two popular online platforms - Twitter and location-based social network Foursquare - we represent the two together as a composite *multilayer online social network*. Through this paradigm we study the interactions of pairs of users differentiating between those with links on one or both networks. We find that users with multiplex links, who are connected on both networks, interact more and have greater neighbourhood overlap on both platforms, in comparison with pairs who are connected on just one of the social networks. In particular, the most frequented locations of users are considerably closer, and similarity is considerably greater among multiplex links. We present a number of structural and interaction features, such as the multilayer Adamic/Adar coefficient, which are based on the extension of the concept of the node neighbourhood beyond the single network. Our evaluation, which aims to shed light on the implications of multiplexity for the link generation process, shows that multilayer features, constructed from properties across social networks, perform better than their single network counterparts in predicting links across networks. We propose that combining information from multiple networks in a multilayer configuration can provide new insights into user interactions on online social networks, and can significantly improve link prediction overall with valuable applications to social bootstrapping and friend recommendations.'
author:
- |
Desislava Hristova$^*$, Anastasios Noulas$^*$, Chloë Brown$^*$, Mirco Musolesi$^{**}$, Cecilia Mascolo$^*$\
$^*$Computer Laboratory, University of Cambridge\
$^{**}$School of Computer Science, University of Birmingham\
`dh475,an346,clb76,cm542 @ cam.ac.uk; m.musolesi @ cs.bham.ac.uk`
title: |
A Multilayer Approach to Multiplexity and Link Prediction in\
Online Geo-Social Networks
---
Introduction
============
Online social media has become an ecosystem of overlapping and complementary social networking services, inherently multiplex in nature, as multiple links may exist between the same pair of users [@kivela2013]. Multiplexity is a well studied property in the social sciences [@haythornthwaite1998] and it has been explored in social networks from Renaissance Florence [@Padgett06organizationalinvention] to the Internet age [@hay2005]. Despite the broad contextual differences, multi-relational ties are consistently found to exhibit greater intensity of interactions across different communication channels, and therefore a stronger bond [@haythornthwaite1998; @me2014]. Nevertheless, *there is a lack of research about online social networks and their value from a multiplex perspective*.
Recently, empirical models of multilayer networks have emerged to address the multi-relational nature of social networks [@kivela2013; @Szell19072010]. In such models, interactions are considered as layers in a systemic view of the social network. We adopt such a model in our analysis, where we shift the concept of a link and neighbourhood to encompass more than one network. This allows us to study interactions and structural properties across online social networks (OSNs), addressing the need for further understanding of their complimentary and overlapping nature, and multiplexity online. Although there have been some recent comparative studies of multiple online social networks [@ICWSM148059; @lee2014], and their intersection [@Szell19072010], the applications of multiplex network properties to OSNs is yet to be substantially addressed.
In this work, we explore intersecting networks, multiplex ties, and their application to link prediction across OSNs. Link prediction systems are key components of social networking services due to their practical applicability to friend recommendations and social network bootstrapping, as well as to understanding the link generation process. Link prediction is a well-studied problem, explored in the context of both OSNs and location-based social networks (LBSNs) [@liben2007link; @menon2011link; @crandall2010; @scellato2011exploiting]. However, only very few link prediction works tackle multiple networks at a time [@lee2014; @Tang2012], while *most link prediction systems only employ features internal to the network under prediction*, without considering additional link information from other OSNs.
Our main contributions can be summarised as follows:
- We generalise the notion of a *multilayer online social network*, and extend definitions of neighbourhood to span multiple networks, adapting measures of overlap such as the Adamic/Adar coefficient in social networks to the multilayer context.\
- We find that *pairs with links on both Twitter and Foursquare exhibit significantly higher interaction on both social networks* in terms of number of mentions and colocation within the same venues, as well as a lower distance and higher number of common hashtags in their tweets.
- A significantly *higher overlap can be observed between the neighbourhoods of nodes with links on both networks*, in particular with relation to the Adamic/Adar measure of neighbourhood overlap, which is significantly more expressed in the multilayer neighbourhood.
- In our evaluation, *we predict Twitter links from Foursquare features and vice versa*, and we achieve this with AUC scores up to 0.86 on the different datasets. In predicting links which span both networks, we achieve the highest AUC score of 0.88 from our multilayer features set, *proving the multilayer construct a useful tool for social bootstrapping and friend recommendations*.
The remainder of this work details these contributions, and summarises related work, concluding with a discussion of the implications, limitations, and applications of the proposed framework.
Related Work
============
Our work identifies with three main areas: multi-relational social networks, media multiplexity, and link prediction in online social networks. We summarise the state of the art in these areas in the following sections.
Multilayer Social Networks
--------------------------
Multi-relational or multilayer networks have been explored in the context of a wide range of systems from global air transportation [@cardillo2013] to massive online multiplayer games [@Szell19072010]. A comprehensive review of multilayer network models can be found in [@kivela2013]. In the context of social networks, it is generally accepted that the more information we can obtain about the relationship between people, the more insight we can gain. A recent large-scale study on the subject has demonstrated the need for multi-channel data when comprehensively studying social networks [@lehman2014]. Despite the observable multilayer nature of the composite OSNs of users [@kivela2013; @kazienko2010; @brodka2012], most research efforts have been focused on theoretical modelling [@kivela2013], with little to no empirical work exploiting data-driven applications in the domain of multilayer OSNs, especially with respect to how location-based and social interactions are coupled in the online social space. We attempt to fill these gaps in the present work by presenting a generalisable online multilayer framework applied to classic problems such as link prediction in OSNs. Our framework is strongly motivated by the theory of media multiplexity, which we review next.
Media Multiplexity
------------------
Media multiplexity [@hay2005] is the principle that tie strength is observed to be greater when the number of media channels used to communicate between two people is greater (higher multiplexity). In [@haythornthwaite1998] the authors studied the effects of media use on relationships in an academic organisation and found that those pairs of participants who utilised more types of media (including email and videoconferencing) interacted more frequently and therefore had a closer relationship, such as friendship. More recently, multiplexity has been studied in light of multilayer communication networks, where the intersection of the layers was found to indicate a strong tie, while single-layer links were found to denote a weaker relationship [@me2014]. The strength of social ties is an important consideration in friend recommendations and link prediction [@Gilbert2009], and we employ the previously understudied multiplex properties of OSNs to such ends in this work.
Link Prediction
---------------
The problem of link prediction was first introduced in the seminal work of Kleinberg et al. [@liben2007link] and since then, has been applied in various network domains. For instance, in [@scellato2011exploiting] the authors exploit place features in location-based services to recommend friendships, and in [@backstrom2011supervised] a new model based on supervised random walks is proposed to predict new links in Facebook. Most of these works build on features that are endogenous to the system that hosts the social network of users. In our evaluation, however, we train and test on heterogeneous networks. In a similar spirit, the authors in [@Sadilek2012] show how using both location and social information from the same network significantly improves link prediction. Our approach differs in that it frames the link prediction task in the context of multilayer networks and empirically shows the relationship between two different systems - Foursquare and Twitter - by mining features from both. Before presenting our framework and analysis, we will next state the research questions we are interested in answering through this work.
Research Questions
==================
In light of the related work presented above, our goal is to mend the gap between multilayer network models, media multiplexity properties, and link prediction systems. More specifically, we address the following research questions in this work:\
**RQ1:** *How do structural properties such as degree extend into the multilayer neighbourhood?* We propose a multilayer version of the network neighbourhood, which extends it to multiple networks (layers) and observe how such structural properties are manifested across Twitter and Foursquare.\
\
**RQ2:** *What are the structural and behavioural differences between single network and multiplex links?* In order to understand the value of multiplex links (users connected on more than one network), we observe how they compare to single network links in terms of neighbourhood overlap, Twitter interaction, similarity and mobility in Foursquare.\
\
**RQ3:** *Can we use information about links from one layer to predict links on the other?* Many online social systems suffer from a lack of initial user adoption. Although many social networks nowadays incorporate the option of importing contacts from another pre-existing network and copying links, this method does not offer a ranking of users by relevance targeted towards the specific platform.\
\
**RQ4:** *Can we predict links which exist on more than one network (i.e., multiplex links)?* Media multiplexity is a valuable source of tie strength information, and has further structural implications, which are of interest to OSN services and link prediction systems. We would like to explore the potential of identifying such links for building more successful online communities.\
\
We will next present our multilayer framework for OSNs, and study user behaviour and properties across Twitter and Foursquare, extending our analysis to multiplex links in comparison with single-layer links. We finally integrate this into a link prediction system for OSNs, where we evaluate the utility of the metrics and features described in this work in hope to answer the above posed questions.
Multi-relational Framework
==========================
The network of human interactions is usually represented by a graph $G$ where the nodes in set $V$ represent people and the edges $E$ represent interactions. While this representation has been immensely helpful for the uncovering of many social phenomena, it is focused on a single-layer abstraction of human relations. In this section, we describe a model, which represents the multiplexity of OSNs by supporting multiple friendship and interaction links.
Multilayer Online Social Network
--------------------------------
We represent the parallel interactions between nodes across OSNs as a *multilayer network* $\cal{M}$, an ensemble of $M$ graphs, each corresponding to an OSN. We indicate the $\alpha$-th layer of the multilayer as $G^\alpha(V^\alpha, E^\alpha)$, where $V^{\alpha}$ and $E^{\alpha}$ are the sets of vertices and edges of the graph $G^{\alpha}$. We can then denote the sequences of graphs composing the $M$-layer multilayer graph as ${\cal{M}} = \{G^1,...,G^\alpha,...,G^M\}$. The graphs are brought together as a multilayer system by the common members across layers as illustrated in Figure \[fig:mdat\].
Multilayer social networks are a natural representation of media multiplexity, as each layer can depict an OSN. Figure \[fig:multi\] illustrates the case at hand, where there are two OSN platforms represented by $G^\alpha$ and $G^\beta$. Members need not be present at all layers and the multilayer network is not limited to two layers. While each platform can be explored separately as a network in its own right, this does not capture the dimensionality of online social life, which spans across multiple OSNs.
[0.24]{} ![Multilayer model of OSNs with [slowromancap1@]{}. Multiplex link; [slowromancap2@]{}. Single-layer link on $G^\alpha$; and [slowromancap3@]{}. Single-layer link on $G^\beta$.[]{data-label="fig:mdat"}](fig1a.pdf "fig:"){width="\textwidth"}
[0.26]{} ![Multilayer model of OSNs with [slowromancap1@]{}. Multiplex link; [slowromancap2@]{}. Single-layer link on $G^\alpha$; and [slowromancap3@]{}. Single-layer link on $G^\beta$.[]{data-label="fig:mdat"}](fig1b.pdf "fig:"){width="\textwidth"}
Figure \[fig:links\] illustrates three link types as observed in Figure \[fig:multi\] for the case of a two layer network. Firstly, we define a *multiplex link* between two nodes $i$ and $j$ as a link that exists between them *at least in two layers* $\alpha, \beta \in \cal{M}$. Second, we say that a *single-layer link* between two nodes $i$ and $j$ exists if the link appears *only in one layer* in the multilayer social network. In systems with more layers, multiplexity can take on a value depending on how many layers the link is present on. In the case at hand, given layer $\alpha$ and layer $\beta$, we denote the set of all links present in the multilayer network as $E^{\alpha \cup \beta}$, which yields the global connectivity.
We also define the set of multiplex links as $E^{\alpha \cap \beta}$ and the set of all single-layer links on layer $\alpha$ only as $E^{\alpha \backslash \beta}$. These multilayer edge sets can be further extended to the $M$ layer network by considering more layers $\{1, \dots, M\}$ as part of the intersection or union of graphs. The presence of multiplex and single-layer links in the above edge sets defines the multilayer neighbourhood of nodes in the network, as expanded upon next.
The Multilayer Neighbourhood
----------------------------
Following our definition of a multilayer online social network, we can redefine the ego network of a node as the *multilayer neighbourhood*. While the simple node neighbourhood is the collection of nodes one hop away from the ego, the multilayer global neghbourhood (denoted by $GN$) of a node $i$ can be derived by the total number of unique neighbours across layers:
$$\Gamma_{GNi} = \{j \in V^{\cal{M}} : e_{i,j} \in E^{\alpha \cup \beta}\}$$
and their global multilayer degree as:
$$k_{GNi} = | \Gamma_{GNi}|$$
which provides insight into the entire connectivity of nodes across layers, and can therefore be interpreted as a global measure of the immediate degree of a node. We can similarly define the core neighbourhood (denoted by $CN$) of a node $i$ across layers of the multilayer network as:
$$\Gamma_{CNi} = \{j \in V^{\cal{M}} : e_{i,j} \in E^{\alpha \cap \beta}\}$$
and their core multilayer degree as:
$$k_{CNi} = | \Gamma_{CNi}|$$
where we only consider neighbours which exist across all layers. This simple formulation allows for powerful extensions of existing metrics of local neighbourhood similarity. We can define the overlap (Jaccard similarity) of two users $i$ and $j$’s global neighbourhoods as:
$$sim_{GNij} = \frac{|\Gamma_{GNi} \bigcap \Gamma_{GNj}|}{|\Gamma_{GNi} \bigcup \Gamma_{GNj}|}$$
where the number of common friends is divided by the number of total friends of $i$ and $j$. The same can be done for the core degree of two users. The Jaccard coefficient, often used in information retrieval, has also been widely used in link prediction [@liben2007link].
We can further extend our definition of the multilayer neighbourhood to the Adamic/Adar coefficient for link likelihood [@adamic2001], which considers the overlap of two neighbourhoods based on the popularity of common friends (originally through web pages) in a single-layer network as:
$$aa\_sim_{GNij} = \sum_{z \in \Gamma_{GNi} \bigcap \Gamma_{GNj}}
\frac{1}{log(|\Gamma_{GNz}|)}$$
where it is applied to the global common neighbours between two nodes but can be equally applied to their core neighbourhoods. This metric has shown to be successful in the link prediction in its original single-layer form in both social networks and location-based networks [@liben2007link; @scellato2011exploiting]. In the present work, we aim to show its applicability to the multilayer space in predicting online social links across and between Twitter and Foursquare. We will next describe the specific datasets, which we apply this framework to.
Dataset
=======
Twitter and Foursquare are two of the most popular social networks, both with respect to research efforts and user base. They have distinct broadcasting functionalities - microblogging and check-ins. While Twitter can reveal a lot about user interests, Foursquare check-ins provide a proxy for human mobility. In Foursquare users check-in to venues that they visit through their location enabled devices, and share their visit or opinion of a place with their friends. Foursquare is two years younger than Twitter and its broadcasting functionality is exclusively for mobile users (50M to date[^1]), while 80% of Twitter’s 284M users are active on mobile[^2]. Twitter generally allows anyone to “follow" and be “followed", where followers and followed do not necessarily know one another. On the other hand, Foursquare supports undirected links, referred to as “friendship". A similar undirected relationship can be constructed from Twitter, where a link can be considered between two users if they both follow each other reciprocally [@Kwak2010]. Since we are interested in ultimately in predicting friendship, we consider only reciprocal Twitter links throughout this work.
![Social network graph for San Francisco. Blue edges are single-layer edges, while pink edges are multiplex edges. The node size is proportional to the global degree of that node.[]{data-label="fig:sf"}](Untitled2.png)
Our dataset was collected from Twitter and Foursquare in the United States between May and September 2012, where tweets and check-ins were downloaded for users who had checked-in during that time, and where those check-ins were shared on Twitter. This allows us to study the intersection of the two networks through a subset of users who have accounts and are active on both Twitter and Foursquare, and have chosen to share their check-ins to Twitter.
\[h!\]
[l\*[6]{}[l]{}r]{} Property & New York & Chicago & SF & All\
$|V^{\cal{M}}|$ & 6,401 & 2,883 & 1,705 & 10,989\
$|E^{T \cap F}|$ & 9,101 & 5,486 & 1,517 & 16,104\
$|E^{T \backslash F}|$ & 13,623 & 7,949 & 1,776 & 23,348\
$|E^{F \backslash T}|$ & 6,394 & 4,202 & 863 & 11,459\
$<k_{GN}>$ & 4.55 & 6.12 & 2.44 & 4.63\
$<k_{CN}>$ & 1.42 & 1.9 & 0.89 & 1.47\
$tweets$ & 2,509,802 & 1,288,865 & 632,780 & 4,431,447\
$checkins$ & 228,422 & 105,250 & 46,823 & 380,495\
$venues$ & 24,110 & 11,773 & 6,934 & 42,817\
[0.245]{} {width="\textwidth"}
[0.245]{} {width="\textwidth"}
[0.245]{} {width="\textwidth"}
[0.245]{} {width="\textwidth"}
We focus our analysis on the top three cities in terms of activity during the period. Table \[tab:datt\] shows the details for each city, in terms of activity and venues, multilayer edges and degrees for each network, where $E^{T \cap F}$ denotes the set of edges, which exist on both Twitter and Foursquare, $E^{T \backslash F}$ and $E^{F \backslash T}$ are the sets of edges on Twitter only and Foursquare only respectively.
Figure \[fig:sf\] additionally illustrates the case of San Francisco, where blue edges represent single-layer links on either Foursquare or Twitter, and pink edges represent multiplex links on both. We use a Fruchterman Reingold graph layout [@Fruchterman1991] to show the core-periphery structure of the network, with larger nodes having a higher global degree $k_{GN}$. In the following section, we discuss the implications of these sets in detail, where we consider all three cities together, and later evaluate each one separately.
Multilayer Analysis
===================
We begin our analysis by exploring the intersection between the Twitter and Foursquare social networks. We observe user the degree properties across the two networks at a larger scale for all three cities, while later we perform our evaluation on each city separately.
RQ1: Multilayer Degrees
-----------------------
We introduced two degree metrics based on the multilayer neighbourhood of a node in Equations 2 and 4, where the *global neighbourhood* is equivalent to the union of neighbours on both networks, and the *core neighbourhood* is equivalent to the intersection of neighbours across both networks. In this section we consider how the degrees relate to user activity and each other.
In both cases (Figures \[fig:ideg\] and \[fig:udeg\]), users with high activity on both networks, and in particular with high Twitter activity, have the highest degrees in both the core and global neighbourhoods. When we compare the two in Figure \[fig:uvi\], we observe that their joint distribution follows the long-tail exhibited in single-layer social networks as well. Further, we observe the multiplex overlap ratio of the core to global neighbourhood degrees in Figure \[fig:odeg\]. This is simply the core over the global degree:
$$mor_{i} = \frac{k_{CNi}}{k_{GNi}}$$
which indicates the percent of multiplex links in $i$’s multilayer neighbourhood. High activity nodes across both layers at the centre of Figure \[fig:odeg\] have the highest overlap.
In Figure \[fig:uvi\], we compare the two multilayer degrees. We note that the majority of users have a low degree in both, and there is a relationship between the two. The core degree is bound by the global degree and is always a fraction of it, while the global degree may never exceed the sum of the individual layer degrees. This relationship is apparent in the figure, where *the highest degree users are those who have a large number of links which overlap (multiplex links)*. This can be due to the fact that these users are more engaged across the two platforms. We further explore the value of link multiplexity in the following section.
RQ2: Link Multiplexity
----------------------
[0.24]{} {width="\textwidth"}
[0.24]{} {width="\textwidth"}
[0.24]{} {width="\textwidth"}
[0.24]{} {width="\textwidth"}
We study the three types of links as described in our multilayer model above: multiplex links on both Twitter and Foursquare, which we denote as *tf* for simplicity; single-layer links on Foursquare only (denoted as *fo*); single-layer links on Twitter only (denoted as *to*), and compare these to unconnected pairs of users (denoted as *na*). We consider reciprocal Twitter links only, where $e_{ij}, e_{ji} \in E^T$. Reciprocal relationships in Twitter have been considered as equivalent to undirected ones in other OSNs [@Kwak2010].
Multiplexity and Neighbourhoods
-------------------------------
The number of common friends has been shown to be an important indicator of a link in social networks [@liben2007link]. Moreover, the neighbourhood overlap weighted on the popularity of common links between two users has been shown to be a good predictor of friendship in online networks [@adamic2001]. Figure \[fig:structure\] shows the Adamic/Adar metric of neighbourhood similarity across the various single and multilayer neighbourhoods described in Section 3, and the four link types.
The Adamic/Adar metric is distinctly higher for multiplex links. In agreement with previous studies of tie strength [@Gilbert2009], we observe that multiplex links share a greater overlap in all single and multilayer neighbourhoods. In single-layer neighbourhoods (Figure \[fig:aat\] and \[fig:aaf\]) we observe that after multiplex links, those links internal to the network under consideration have a higher overlap than exogenous ones (*to* in Figure \[fig:aat\] and *fo* in Figure \[fig:aaf\]), followed by unconnected pairs, which have the least overlap.
With respect to the multilayer neighbourhoods, we can observe a much more pronounced overlap across the link types. While the global neighbourhood overlap follows a similar distribution to the single-layer neighbourhoods but at a much lower scale, in Figure \[fig:aai\] we can observe more clearly that unconnected pairs share little if any neighbours, while multiplex links have a significant overlap. With respect to the global neighbourhood (Figure \[fig:aau\]), both Foursquare only and Twitter only links share significantly more overlap (scale is higher on x axis) than when observing the single-layer neighbourhoods in Figures \[fig:aat\] and \[fig:aaf\]. This indicates that some common neighbours lie across layers, and not just within, *the global neighbourhood revealing a more complete image of connectivity, which stretches beyond the single network*.
The core neighbourhood overlap is most prominent for multiplex links (Figure \[fig:aai\]), which indicates that they share more friends across networks than any other type of link. While this is expected, it *confirms that the neighbourhood overlap is a good indicator of multiplexity in ties*, and is particularly strengthened in its weighted form through the Adamic/Adar metric of neighbourhood similarity.
Multiplexity and Interaction
----------------------------
The volume of interactions between users is often used as a measure of tie strength [@onnela2007]. In this section we compare how the volume of interactions reflects on multiplex and single-layer links. We consider the following interactions on Twitter and Foursquare:\
**Number of mentions:** This interaction feature simply measures the number of times user $i$ has mentioned user $j$ on Twitter during the period. Any user on Twitter can mention any other user and need not have a directed or undirected link to the user he is mentioning.\
**Number of common hashtags:** Similarity between users on Twitter can be captured through common interests. Topics are commonly expressed on Twitter with hashtags using the \# symbol. Similar individuals have been shown to have a greater likelihood of forming a tie through the principles of homophily [@mcpherson2001].\
**Number of colocations:** The number of times two users have checked-in to the same venue within a given time window. In order to reduce false positives, we consider a shorter time window of 1 hour only. Two users at the same place, at the same time on multiple occasions, increases the likelihood of them knowing each other (and having a link on social media). We weight each colocation on the popularity of a place in terms of the total user visits, to reduce the probability that colocation is by chance at a large hub venue such an airport or train station.\
**Distance:** Human mobility and distance play an important role in the formation of links, both online and offline, and have been shown to be highly indicative of social ties and useful for link prediction [@Wang2011]. We calculate the distance between the geographic coordinates of two users’ most frequent check-in locations as the Haversine distance, the most common measure of great-circle spherical distance: $$dist_{ij} = haversine(lat_i,lon_i, lat_j,lon_j)$$
where the coordinate pairs for $i,j$ are of the places where those users have checked-in most frequently, equivalent to the mode in the multiset of venues where they have checked-in. We only consider users who have more than two check-ins over the whole period, and resolve ties by picking an arbitrary venue location from the top ranked venues of a user.
In Figures \[fig:men\] to \[fig:dist\], we observe four types of geographic and social interaction on the two social networking services, where each box-and-whiskers plot represents an interaction between multiplex links (*tf*), Twitter only (*to*), Foursquare only (*fo*), and unconnected pairs (*na*) on the x axis. On the y axis we can observe the distribution in four quartiles, representing 25% of values each. The dark line in the middle of the box represents the median of the distribution, while the dots are the outliers. The “whiskers" represent the top and bottom quartiles, while the boxes are the middle quartiles of the distribution.
In terms of Twitter mentions (Figure \[fig:men\]), multiplex ties and non-connected pairs of users exhibit an overall greater number of mentions than any other group, including the Twitter only group. It is uncommon that pairs connected on Foursquare only mention each other. Mentions are quite common between users who are not connected on any network, which may be as a result of mentioning celebrities and other commercial accounts. This is not the case for hashtags, where we find that almost all of unconnected users share 10 or less hashtags with the exception of outliers. Hashtags distinguish the link type between users better than mentions.
With regards to Foursquare interaction, multiplex ties have the highest probability of multiple colocations, with Foursquare and Twitter only ties having less, and unconnected pairs more so with the exception of some outliers. In terms of distance, Twitter only and unconnected pairs are the furthest apart in terms of most frequented location, making multiplex and Foursquare links more distinguishable through this feature, as those pairs have less distance between their most frequented locations.
Although *there is certainly greater interaction between multiplex links*, followed by Twitter only and Foursquare only links, we would like to eliminate the randomness introduced by the positive results for unconnected pairs (*na*). We propose two multilayer interaction metrics combining heterogenous features from both networks in order to better distinguish between the different link types. Firstly, we define the global similarity as the Twitter similarity over Foursquare distance as:
$$sim_{GNij} = \frac{sim_{ij}^a}{dist_{ij}^b}$$
where $sim$ can be replaced with any type of similarity, which is the mass or sum of that similarity for a pair of users, and $a,b$ are exponents which can be tuned to optimise the features. Figure \[fig:mult2\] shows how this feature captures the different levels of links (a=2, b=1). We additionally frame a feature which captures the complete interaction across layers of social networks:
$$int_{GNij} = \sum_\alpha^M |int^\alpha_{ij}|$$
where $int$ can be any type of interaction of layer $\alpha$, this can be further refined by giving a weight to each interaction but in our case, we consider the coefficient to be equal to 1 and use colocations from the Foursquare layer and mentions from the Twitter layer to express the global interaction of two users in the multilayer network. This feature allows us to capture the levels of different link types significantly better as shown in Figure \[fig:mult1\].
[0.22]{} ![Interaction features for the different link types.[]{data-label="fig:interactions"}](log_men.pdf "fig:"){width="\textwidth"}
[0.22]{} ![Interaction features for the different link types.[]{data-label="fig:interactions"}](log_hasht.pdf "fig:"){width="\textwidth"}
[0.22]{} ![Interaction features for the different link types.[]{data-label="fig:interactions"}](log_colocs.pdf "fig:"){width="\textwidth"}
[0.22]{} ![Interaction features for the different link types.[]{data-label="fig:interactions"}](dist_df.pdf "fig:"){width="\textwidth"}
[0.22]{} ![Interaction features for the different link types.[]{data-label="fig:interactions"}](ints.pdf "fig:"){width="\textwidth"}
[0.22]{} ![Interaction features for the different link types.[]{data-label="fig:interactions"}](dist_hasht.pdf "fig:"){width="\textwidth"}
Although we base our analysis on only two of many possible communication channels online, we are nonetheless able to observe the greater overlap of neighbourhoods and higher intensity of interaction characteristic of multiplex links, which is in consistency with the theory of media multiplexity [@hay2005]. We evaluate the predictive performance of the union of the features presented in the following section.
Multiplexity & Link Prediction
==============================
In this section we address the link prediction problem across layers of social networks, and aim to answer our final two research questions: *Can we predict one network using information from the other?*, and *Can we predict multiplex links in OSNs?* We evaluate the likelihood of forming a social tie as a process that depends on a union of factors, using the Foursquare, Twitter, and multilayer features we have defined up until now in a supervised learning approach, and comparing their predictive power in terms of AUC scores for each feature set respectively.
[0.19]{} {width="\textwidth"}
[0.19]{} {width="\textwidth"}
[0.19]{} {width="\textwidth"}
[0.19]{} {width="\textwidth"}
[0.19]{} {width="\textwidth"}
Prediction Space
----------------
The main motivation for considering multiple social networks in a multilayer construct is that each layer carries with it additional information about the links between the same users, which can potentially enhance the predictive model. In light of the multilayer nature of OSNs, we are also interested in whether we can achieve better prediction by combining features from multiple networks.
Formally, for two users $i,j \in \cal{M}$, where $V^{\cal{M}}$ are the nodes (users) that are present in any layer of the multilayer network, we employ a set of features that output a score $r_{ij}^\alpha$ so that all possible pairs $V^{\cal{M}} \times V^{\cal{M}}$ are ranked according to their expectation of having a link $e_{ij}^\alpha$ on a specific layer $\alpha$ in the network. We specify and evaluate two distinct prediction tasks.
Our first goal is to rank pairs of users based on their interaction on one social network in order to predict a link on the other. This entails using mobility interactions to predict social links on Twitter, and using social interactions on Twitter to predict links on Foursquare. Subsequently, we are interested in predicting the multiplex links at the cross-section of the two networks using multilayer features. This type of links have both structural and social tie implications as we have demonstrated in this work, which makes them desirable to identify.
We perform our evaluation on three datasets described at the start of this work in Section 5, where we have Twitter, Foursquare, and the derived multilayer features for the cities of San Francisco, Chicago, and New York. We adopt a supervised learning approach for the prediction tasks, and for each city, which is considered as an independent multilayer network, where we train and test on different layers. Supervised learning methodologies have been proposed as a better alternative to unsupervised models for link prediction [@lichtenwalter2010new].
We compare the performance of feature sets using the Random Forest classifier [@rf2001] with $10$-fold cross-validation testing strategy: for each test we train on $90\%$ of the data and test on the remaining $10\%$. For every test case the user pairs in the test set were ranked according to the scores returned by the classifiers for the positive class label (i.e., for an existing link), and subsequently, Area Under the Curve (AUC) scores were calculated by averaging the results across all folds. We use AUC scores as a measure of performance because it considers all possible thresholds of probability in terms of true positive (TP) and false positive (FP) values rate, which are computed by comparing the predicted output against the target labels of the test data.
In terms of algorithmic implementation, we have used public versions of the algorithms available in [@pedregosa2011scikit]. The features presented earlier in this work, of which each feature set comprises are summarised in Table \[tab:feats\]. We denote the Twitter neighbourhood as $\Gamma^T$ and the Foursquare neighbourhood as $\Gamma^F$. Next, we specify each prediction task and present the results of the supervised learning evaluation in terms of the predictive power of each feature set in both tasks.\
[|r|l|]{}\
$mentions$ & $|mentions_{ij}|$\
$hashtags$ & $|hashtags_{ij}|$\
$overlap$ & $\frac{|\Gamma_i^T \cap \Gamma_j^T|}{|\Gamma_i^T \cup \Gamma_j^T|}$\
$aa\_sim$&$ \sum\limits_{z \in \Gamma_i^T \cap \Gamma_j^T} \frac{1}{log(|\Gamma_z^T|)}$\
\
$colocs$ & $|colocations_{ij}|$\
$dist$ & $haversine(lat_i,lon_i, lat_j,lon_j)$\
$overlap$ & $ \frac{|\Gamma_i^F \cap \Gamma_j^F|}{|\Gamma_i^F \cup \Gamma_j^F|}$\
$aa\_sim$ & $\sum\limits_{z \in \Gamma_i^F \cap \Gamma_j^F} \frac{1}{log(|\Gamma_z^F|)}$\
\
$int_{GNij} $ & $\sum\limits_{\alpha}^{M} |int^\alpha_{ij}|$\
$sim_{GNij}$ & $\frac{sim_{ij}^a}{dist_{ij}^b}$\
$overlap$ & $ \frac{|\Gamma_{CNi} \cap \Gamma_{CNj}|}{|\Gamma_{CNi} \cup \Gamma_{CNj}|}$\
$aa\_sim$ & $\sum\limits_{z \in \Gamma_{CNi} \cap \Gamma_{CNj}} \frac{1}{log(|\Gamma_{CNz}|)}$\
RQ3: Cross-network prediction
-----------------------------
The Receiver Operating Characteristic (ROC) curves (defined as the True Positive versus False Positive Rate for varying decision thresholds) and the corresponding Area Under the Curve (AUC) scores are shown in Figure \[fig:roc1\] for the three datasets. We now discuss these results with respect to each task. In the first prediction task, for a pair of users $i$ and $j$ we define a feature vector $\mathbf{x_{ij}^\alpha}$ encoding the values of the users’ feature scores on layer $\alpha$ in the multilayer network. We also specify a target label $y_{ij}^\beta \in \{-1,+1\}$ representing whether the user pair is connected on the $\beta$ layer under prediction.
We use the supervised Random Forest classifier (45 trees, optimised with tree depth = 25) to predict links from one layer using features from the other. Figure \[fig:roc\_fsq\] shows the ROC curves and respective AUC scores for each dataset in predicting Foursquare links from Twitter features, ranging between 0.7 for the New York dataset to 0.81 for Chicago, and 0.73 for San Francisco. We compare this to the reverse task of predicting Twitter links using Foursquare features in Figure \[fig:roc\_mfsq0\], where we obtain AUC scores of 0.86, 0.73, and 0.79 for the three cities respectively. We observe slightly higher results for Twitter links, and we note that this may be as a result of the higher number of Twitter links in our dataset or as a result of the greater difficulty of the inverse task.\
RQ4: Multiplex link prediction
------------------------------
In our second prediction task, we are interested in evaluating the performance of each feature set in predicting link multiplexity. Given a feature vector $\mathbf{x_{ij}}$, we would like to predict a target label $y_{ij} \in \{-1,+1\}$, where a link exists on both layers (+1) or none (-1). We compare performance of the multilayer features to the Twitter and Foursquare sets.
In this task, we use all three feature sets to predict multiplex links, which generally exhibit signs of a stronger online bond through interaction and structural properties as we have seen in the first part of this work. In Figures \[fig:m\_fsq\] and \[fig:rm\_twt\], we observe how Twitter and Foursquare features perform in predicting multiplex links using the Random Forest algorithm again, with the highest AUC scores of 0.82 and 0.84 for each set respectively. The Foursquare feature set performs better in terms of AUC scores but the multilayer feature set outperforms both (AUC = 0.88 for Chicago), due to its inclusion of features from different layers and cross-layer structural properties.
In conclusion, it is possible to predict links between heterogeneous social networks and to predict multiplex links spanning multiple networks using multilayer features as we have seen in our subset of users. We discuss the applications of these results in the following section.
Discussion & Conclusions
========================
In this work we have demonstrated the structural and interaction properties of links across two online social networks and have also shown the value of multilayer features in predicting links on both Twitter and Foursquare, and multiplex links. We believe that the primary contribution is methodological, since it provides a novel framework for investigating multiplexity across different social networks. The techniques discussed in this work are general and can be potentially used to investigate other scenarios for which datasets containing information about social interactions across multiple networks are available. In this section, we discuss the implications, limitations and real-world applications of these results.
Implications
------------
Recently, social media has been increasingly alluded to as an *ecosystem*. The parallel comes after the emergence of multiple OSNs, interacting as a system, while competing for the same resources - users and their attention. We have addressed this system aspect by modelling multiple social networks as a multilayer online social network in this work. We have also identified two extensions of the node neighbourhood. The global neighbourhood or degree gives insight into a users’ full connectivity across services, this is especially important when considering users with asymmetric activity and degree across networks since their centrality in the online ecosystem can be under or over-estimated. We additionally defined the core degree, which on the other hand reveals the intersection across networks, and therefore the stronger online ties - those relevant on multiple networks.
The strength of ties manifested through multiplexity is expressed through a greater intensity of interactions and greater similarity across attributes both the offline [@hay2005; @me2014], and in the online context as we have seen in this work. We have introduced a number of features, which take into consideration the multilayer neighbourhood of users in OSNs. The Adamic/Adar coefficient of neighbourhood similarity in its core neighbourhood version proved to be a strong indicator of multiplex ties. Additionally, we introduced combined features, such as the global interaction and similarity over distance, which reflect more distinctively the type of link, which exists between two users, than its single-layer counterparts. These features can be applied across multiple networks and can be flexible in their construction according to the context of the OSNs under consideration.
Limitations
-----------
Media multiplexity is fascinating from the social networks perspective as it can reveal the strength and nature of a social tie given the full communication profile of people across all media they use [@hay2005]. Unfortunately, full online and offline communication profiles of individuals were not available and our analysis is limited to two social networks. Nevertheless, we have observed some evidence of media multiplexity manifested in the greater intensity and structural overlap of multiplex links and have gained insight into how we can utilise these properties for link prediction. Certainly, considering more OSNs and further relating media multiplexity to its offline manifestation is one of our future goals, and we believe that with the further integration of social media services and availability of data this will be possible in the near future.
Our data is limited to a sub-sample of users who we know have active accounts on both networks in three US cities, Foursquare check-ins also being limited to those posted on Twitter. This excludes a number of users who may have Foursquare accounts but have not linked them on Twitter. Nevertheless, we were able to show that it is possible to predict one social network from the other in a cross-network manner and we hope to extend our prediction and analysis to a greater scale and geographical scope in the future.\
Applications
------------
Most new OSNs use contact list integration with external existing networks, such as copying friendships from Facebook through the open graph protocol.[^3] Copying links from pre-existing social networks to new ones results in higher social interaction between copied links than between links created natively in the platform [@Zhong2014]. We propose that extending this copied network with a rank of relevance of contacts using multiplexity can provide even further benefits for newly launched services.
In addition to fostering multiplexity, however, new OSNs and especially interest-driven ones such as Pinterest for example, may benefit from similarity-based friend recommendations. In this work, we apply mobility features and neighbourhood similarity from Foursquare to predict links on Twitter and vice versa, highlighting the relationship between similar users across heterogeneous platforms. Similarly in [@Tang2012], the authors infer types of relationships across different domains such as mobile and co-author networks. Although using a transfer knowledge framework, and not exogenous interaction features like we do, the authors also agree that integrating social theory in the prediction framework can greatly improve results. The present work is a step towards understanding the composite nature of online social network services and hopefully towards enhancing their functionality and purpose.
Acknowledgements
================
This work was supported by the Project LASAGNE, Contract No. 318132 (STREP), funded by the European Commission.
[^1]: <https://foursquare.com/about>
[^2]: <https://about.twitter.com/company>
[^3]: <https://developers.facebook.com/docs/opengraph>
| {
"pile_set_name": "ArXiv"
} |
Introduction
============
Conventional quantum field theory is a set of quantum mechanics for $N$-particle states and transitions between them. Its starting point is a Fock space spanned by the states of (approximately) free particles with different values of energies and momenta. Having at first a quantum mechanical description of these particles, we construct their Hilbert space and then take into account the processes of creation and annihilation, thus secondary quantizing the theory. But it is now well known that $N$-particle states at the Planckian scale could be rather different from free particle states. Indeed, there is no free particle state for particles with trans-planckian energies moving in different directions. The Compton wavelength in this case is smaller than the gravitational radius of the particle. Such particles will inevitably form black holes (see e.g. [@thooft]).
So it is natural, when dealing with quantum field theory at the Planckian scale, first to try to construct a Hilbert space containing $N$ gravitating particles before second quantization.
At first sight, this $N$-particle quantum mechanics is introduced only as a technical tool regularizing quantum field theory at the Planckian scale. This could be done in different ways. In particular, there could be different models of quantum mechanics of gravitating particles. But fortunately there is a language for studying the physical content of these models at already the quantum-mechanical level. Namely, since the “black hole states” should be present in Hilbert space, this first-quantized quantum mechanics should be suitable for describing such physical phenomena as the Hawking radiation, mass spectrum, black hole entropy. Analyzing certain predictions about these phenomena (like the Hawking radiation spectrum or the black hole entropy formula) given by different quantum mechanics, one could choose among them. But in gravitational models we have some special property: the gravitational field is not just the field responsible for the gravitational interaction between particles, but it also determines the global structure of the space-time. Usually the space-time itself is the configuration space for particle dynamics. So, in general, when we take into account the gravitational interaction, the structure of the “configuration space” becomes dependent on the initial conditions, as could be observed by analyzing classical solutions. Hence it is impossible to define a configuration space for gravitating particles naively. On the quantum-mechanical level another problem arises. In quantum mechanics we are supposed to have a superposition of different classical states. It means that in the case of a self-gravitating system we are forced to work in terms of superposition of different space-time geometries. There have been plenty of attempts to describe this picture but it still remains a puzzle how to construct any field theory on such “quantum space-time”.
This problem is the most important if we want to construct a field theory taking into account the gravitational interaction at the Planckian scale. As a matter of fact, it has very little to do with the problem of scattering, creation and annihilation of gravitons. Being present in field theory, the problem of “quantum space-time” arises already at the quantum-mechanical level, i.e. for systems with a finite number of degrees of freedom. Thus we can try to solve it in this simplest case when it is easier to obtain a correct definition of the configuration space.
How can we construct a quantum mechanics of gravitating particles without facing the problems of creation and annihilation of gravitons? A possible approach to this problem is to take into account only some global degrees of freedom of the gravitational field relevant for the dynamics of gravitating particles. These degrees of freedom are topological in the absence of sources.
It is well known that gravitational theories contain topological degrees of freedom. For example, 2+1 gravity is pure topological and equivalent to the Chern-Simons theory with the gauge group $ISO(2,1)$ [@2+1]. Spherically symmetric 3+1 gravity without matter is also topological due to the Birkhoff uniqueness theorem.
On the other hand, in some complicated cases, when the theory is not topological, one can look for its topological sectors. If we are interested only in some particular observables, it could be enough to study quantum mechanics of these global degrees of freedom. For example, we could separate the dynamics of global degrees of freedom from the dynamics in volume by extracting some important surface terms from the action functional thus obtaining a field theory on the boundary surface [This idea is in good agreement with the so-called ’t Hooft “holographic principle” which states that there should exist some surface, 2+1-dimensional, field theory, whose degrees of freedom are relevant to the black hole entropy ([@holpr]). An attempt to describe these quantum black hole degrees of freedom by a Chern-Simons field theory on the horizon (or at infinity) was recently made in [@ashtekar; @carlip; @solodukhin]. For a recent review see [@banados].]{}.
The corresponding field theories in such topological models should have a finite number of physical degrees of freedom. Thus the quantum-mechanical phase space is present in the problem from the very beginning. Then one has to add some sources and the proper Hamiltonian which results in the transition from topological to physical theory, as could be easily seen, for example, in the 2D Yang-Mills case (cf. [@gorsk-nekr] and refs. therein). As a result, one has a quantum-mechanical problem (it is often integrable) describing the dynamics of these topological degrees of freedom. Mathematically this finite-dimensional system is obtained by Hamiltonian reduction.
Let us try to apply this approach to gravitating particles. To construct a quantum mechanics for such a topological field theory with a particle we need first to clarify what is the “classical phase space". The latter can be defined as the space of all classical solutions modulo gauge transformations. For the case of self-gravitating particles in $2+1$ dimensions this approach was used in [@matchull-welling].
In what follows (as in our previous papers) we will study maybe the simplest gravitational model of the type described above — spherically-symmetric gravity with a self-gravitating thin dust shell (see [@ber-kuz; @prd57; @prl]).
In this model we see among the classical solutions space-time manifolds with different geometries. In Fig.\[fig:kruskal\] the geometry of the complete Schwarzschild space-time is shown. It contains two isometric regions with two singularities (future and past) at $R=0$ and two infinities in the left and right asymptotically flat regions $R_+$ and $R_-$. In Fig. \[fig:trajectories\] different types of the corresponding Carter-Penrose diagrams are presented. In the “black hole case" (a) the turning point of the shell lies in the $R_+$-region and at this point it can be seen by an observer at the right infinity. In the “wormhole case" (b) the turning point of the shell is on the opposite side of the Einstein-Rosen bridge in the $R_-$-region and could not be seen from $R_+$-infinity. In the case of unbounded motion (c) the shell starts from the infinity $R=\infty $ in the $R_+$-region and collapses to $R=0$ forming a singularity.
All these classical configurations should be present in the finite-dimensional phase space of the gravitating shell.
Due to the high symmetry it is possible in this model to fulfil the above reduction explicitly. In the case of spherically symmetric gravity without matter such a reduction was made by K. Kuchař ([@kuchar]). The resulting “topological” gauge-invariant degree of freedom is a variable $m$ (and the corresponding momentum) which is defined by the boundary conditions at the infinity and is nothing but the Schwarzschild mass measured by an observer at infinity. If a thin dust shell is included as a source, we have (after reduction) another “physical” degree of freedom, describing the shell motion. The variable $m$ enters into the resulting equation for the gravitating shell and can be formally considered as a parameter. There is another parameter, the bare mass of the shell $M$. In classical mechanics, cases (a) and (b) are realized for bounded motion of the shell when $M/m< 2$ and $M/m> 2$, respectively.
There were several attempts to construct both classical and quantum mechanics for such a system. For example, in [@prd56] the reduced phase space for a self-gravitating shell was constructed as a set of initial data for the black hole case (a) only. In [Ref.[@kr-wilch]]{} the local wave function of a self-gravitating null shell was found to describe the effects of back reaction in non-thermal corrections to the spectrum of the Hawking radiation. But in quantum mechanics the wave function should be defined over the whole configuration space and it is of crucial importance to construct a global picture for the dynamics of the system in hand. Some important results, such as the quantization conditions, can be obtained only from global properties of the wave function. So it is necessary in our case to construct a global configuration space taking into account all classical solutions, in particular, the “wormhole” case (b) as well as the “black hole” case (a) for any value of the ratio $M/m$. But for each particular value of this ratio the part of the global configuration space representing classical solutions with the opposite sign of $(M/m-2)$ should be classically forbidden in an effective quantum mechanics of the self-gravitating shell. For example, in the black hole case $(M/m<2)$ the part of the configuration space representing wormhole classical solutions is classically forbidden.
This is the first qualitative difference of the self-gravitating shells motion from test particle (shell) motion on a fixed Kruskal background when the shell can move in all parts of the Carter-Penrose diagram irrespective of the value of $M/m$. The appearance of an additional classically forbidden region in the configuration space, where the wave function should exponentially decrease, results in a new quantization condition for the parameters $M$ and $m$ ([@prd57; @prl]). This effect (whose physical consequences are discussed in [@prl] and will be discussed below) can be illustrated by the following simple quantum-mechanical example. Let us consider the following radial Schrödinger equation: ++( E- +) R(r)=0. Then, let us suppose that the asymptotic behaviour of the wave function at negative infinity $r\to -\infty$ along the real line is also important for some physical reasons (for example, the true configurations variable is the area $s=r^2$, and the classical configuration space is a positive semi-axis $s>0$). In this case it is easy to see from the exact solution (see [@prl]) that this new requirement, together with usual ones at $r\to \infty$ and $r\to 0$, gives not only a quantization condition for the parameter $E$, but also one more quantization condition, so that the parameter $A$ is quantized as well.
In [@prd57] the formalism was constructed to describe global properties of the configuration space and globally-defined quantum mechanics for the case of a self-gravitating shell. In this case, after reduction, using the Kuchař gauge-invariant variables, we are left with the only nontrivial equation describing the dynamics of the shell. Formally this dynamics is one-dimensional. The variable $\hat R$ which describes the position of the shell is gauge-invariant and has the meaning of the shell radius. But to parameterize the whole configuration space it is not enough to have $R$ varying from $0$ to $\infty$. This can be easily seen from the observation that it is impossible to distinguish black-hole-type classical solutions from the wormhole-type solutions in terms of the variable $R$ (see Fig.\[fig:trajectories\]). This variable covers the configuration space twice. Fortunately there is a way to avoid this difficulty. The equation which governs the dynamics of the shell is an equation in finite differences. The shift of the argument of the wave function in this equation occurs along the imaginary axis, and this means that the equation is actually defined not on the real line, but over some complex manifold. The equation for the shell dynamics contains the square root of the “Schwarzschild factors” $\sqrt{1-\frac{2m}{R}}$. So the natural complex manifold for the equation is the Riemannian surface $S_F$ on which the coefficients of the equation are analytical functions. This Riemannian surface is just a two-(real)dimensional sphere obtained after gluing two complex planes along the sides of the cuts made on each plane along the interval between the branching points of the coefficients on the real line. The configuration space for our self-gravitating shell is the real section of $S_F$. This configuration space properly represents different classical solutions for the self-gravitating shell (for a detailed analysis see [@prl] and [@grr]). All main results from our models are due to the non-trivial structure of the configuration space. This real section $\im(\rho)=0$ of the Riemannian surface $S_F$ covers the real line twice, as is shown in Fig.\[fig:sf\]. It consists of the $V_{\pm}, T_{\pm}$ and $R_{\pm}$ intervals. The sign of the branching function is taken to be $+$ on the intervals $V_+$, $R_+$ and $-$ on the intervals $T_-$, $R_-$ and $(F_{\In}F_{\out})^{1/2}=\pm i\sqrt{\left|F_{\In}F_{\out}\right|}$ on $T_\pm$, respectively.
In [@prd57; @prl; @exact] the first step to the construction of a global wave function on such a complicated configuration space was made. The asymptotic behaviour of the analytical wave functions at the infinities in $R_{+}$ and $R_{-}$ regions was analyzed. As a result, a discrete mass spectrum of bound states and a discrete spectrum for infinite motion of the system were found. Analyzing these two spectra, the Bekenstein-Mukhanov mass spectrum for black holes [@bekenstein] was obtained.
In the present paper we study a global quasiclassical solution for the $V_{+}$, $T_{+}$ and $R_{+}$ regions. The result is rather unexpected — for a quasiclassical solution with two waves of equal amplitudes under the horizon we obtain, after analytical continuation in the $R_{+}$ region, ingoing and outgoing waves with the amplitudes $Z_{\In}$ and $Z_{\out}$. Namely, $$Z_{\In}^2/Z_{\out}^2=\exp\{-\delta A/(4\mpl^2)\}$$ where $A$ is the black hole horizon area. This exactly coincides with the main result of the Hartle and Hawking consideration [@h-h] from which one can derive the values of the black hole temperature and entropy.
Quantum mechanics of self-gravitating massless particles
========================================================
As was shown in [@prd57; @prl], the radial relativistic Schrödinger equation for the massless self-gravitating null-dust shell has the form \[eq:main\] (S+i)+(S-i)=. Here S= R\^2/ (4G\^2m\^2) is a dimensionless variable which measures the area of the shell ($G$ is the gravitational constant and $m=m_{\out}$ is the Schwarzschild mass of the black hole as seen by an observer at infinity). The dimensionless shift parameter is =\^2/(2m\^2);= ($\mpl $ is the Planck mass). The functions $F_{\In,\out}$ are just the coefficients of the Schwarzschild metric inside and outside the shell: F\_=1- 1/ , F\_=1-/ ($\mu=m_{\In}/m_{\out}$ is the quotient of the Schwarzschild masses inside and outside the shell). If we suppose that the energy of the null shell \[eq:eps\] =m\_-m\_ is much smaller than the black hole mass $m_{\out}$, then m\_m\_=m; 1-/m . So in the limit of test particles (when the back reaction is not taken into account) we can expand all the quantities in powers of the small parameter $\epsilon / m$.
Another limiting situation is when the shift parameter is small compared to a characteristic scale on which the wave function varies significantly, then one can approximate the shifted wave function $\Psi(S\pm i\zeta)$ by its Tailor expansion near $S$, so that (\[eq:main\]) takes the form \[eq:ode\] -\^2”(S)+2(S) =(S), which is just the usual Schroedinger equation of nonrelativistic quantum mechanics. This is natural because the limit $\zeta\to 0$ (or $\mpl \to
0$) is either nonrelativistic, or classical, or the limit of weak gravitational field. Thus we see that the coefficient in the right-hand side of (\[eq:main\]) plays the role of a potential term for the motion of the null shell in a gravitational field.
One important note is that the “truncated” equation (\[eq:ode\]) is certainly not valid in the vicinity of the in- and out- horizons — they are singular points of (\[eq:main\]) and we cannot assume that $\Psi$ varies slowly there.
The shift of the argument of the wave function in (\[eq:main\]) is along the imaginary axis, this means that the equation is actually defined not on the real line, but over some complex manifold. The natural complex manifold for (\[eq:main\]) is the Riemannian surface $S_F$ of the branching function \[eq:sf\] (F\_F\_)\^[1/2]{} = ($\rho=\sqrt{S}$) on which the coefficients of the equation are analytical functions. This Riemannian surface $S_F$ is just a two-(real)dimensional sphere obtained after gluing two complex planes along the sides of the cuts made on each of them along the interval $\rho\in (\mu ,1)$ of the real line. The configuration space for our self-gravitating shell is the real section of $S_F$. This configuration space represents properly a different classical solution for a self-gravitating shell (for a detailed analysis see [@prl] and [@grr]). All main results in our models are due to a non-trivial structure of the configuration space. This real section $\im(\rho)=0$ of the Riemannian surface $S_F$ covers the real line twice, as is shown in Fig. \[fig:sf\]. It consists of the intervals $V_{\pm}, T_{\pm}$ and $R_{\pm}$. The sign of the branching function (\[eq:sf\]) is taken to be $+$ on the intervals $V_+,R_+$, $-$ on intervals $T_-,R_-$, and $(F_{\In}F_{\out})^{1/2}=\pm i\sqrt{\left|F_{\In}F_{\out}\right|}$ on $T_\pm$, respectively.
Quasiclassical wave function
============================
In our previous papers the main attention was devoted to the behaviour of the wave function at the infinities in the $R_+$ and $R_-$ regions. Here we study the properties of the quasiclassical wave function near the horizons $s=1$ and $s=\mu$. Consider the quasiclassical solutions of (\[eq:main\]) in the form \[eq:anzatz\] ={}(\_0+\_1+ …). Substituting (\[eq:anzatz\]) into (\[eq:main\]), one gets in the zero order in $\zeta$ the Hamilton-Jacobi equation for $P_S=\d \Omega/ \d S$: \[eq:hj\] {P\_S}= whence it follows \[eq:ps\] P\_S =()\^[1/2]{}=( )\^[1/2]{}. The $\pm$ signs correspond to expanding and collapsing trajectories of the null shell.
The points $S=S_0$ where $P_S=0$ are the turning points of the shell classical motion. From (\[eq:ps\]) it is easy to see that there are no such points for the null-shell motion. But the points $S=1$ and $S=\mu^2$ (the points where the apparent horizons of the internal and external Schwarzschild metrics are situated) are singular points of (\[eq:main\]). The quasiclassical anzatz (\[eq:anzatz\]) is not a good approximation for solving (\[eq:main\]) near these points.
In the region $S\in (\mu^2, 1)$ between the horizons the momentum $P_S$ (\[eq:ps\]) has an imaginary part, so this is an analogue of the “classically forbidden” region if we use the analogy with nonrelativistic quantum mechanics provided by the form (\[eq:ode\]) of (\[eq:main\]) [It is interesting to note that these singular points could be treated as turning points. (\[eq:main\]) is defined over the Riemannian surface $S_F$. The points $\mu^2$ and $1$ are branching points of this Riemannian surface and the coordinate $S$ is not a regular coordinate on $S_F$ in the neighbourhoods of these points. The regular coordinate in the vicinity of $S=\mu^2$ is $u=\sqrt{\rho-\mu}$. The momentum conjugate to the coordinate $u$ is $$P_u=P_S\,\frac{dS}{du}=\pm
2u(u^2+\mu)\ln \frac{u}{(1-\mu)-u^2}$$ which is equal to zero at the point $u=0$. This means that in terms of the regular coordinate $u$ on $S_F$ the point $S=\mu^2$ is a turning point of the classical motion rather than a singular point of the classical dynamical system. The singularity of the coefficients of (\[eq:main\]) originates actually from the singularity of the coordinate $S$ on the Riemannian surface near $S=\mu^2$ which in turn is caused by an irregular behaviour of the radial coordinate $R=2Gm\sqrt{S}$ on the horizon $R=2Gm_{\In}$ of the Kruskal manifold. ]{}. But it should be noted that this analogy is not direct because this region is not entirely forbidden classically: of course, we have in this region a trajectory of A particle falling to the black hole singularity. The true origin of this “special region” is that it is indeed classically forbidden for a particle trajectory going outside the black hole. In some sense, the origin of this ban is not dynamical but casual — a particle should go faster than light to get out of the horizon from the black hole interior. Quantum-mechanically this situation was analyzed in detail for the second-order equation (\[eq:ode\]) (for a bare mass of the shell $M$ not necessarily equal to zero) in [@prd57]. This analysis is valid in the $T$-region far from the horizon $S=1$. The result is that we have only an ingoing (or outgoing) wave in $T_{\pm}$-regions, respectively. The wave in the opposite direction is enormously damped near the horizon relative to the “correct” quasiclassical waves in each regions. This quasiclassical picture reproduces the classical behaviour of the shell.
If the energy $\epsilon$ of the shell is small as compared to the mass of black hole, then $\mu$ is close to $1$ and the region situated between $S=\mu^2$ and $S=1$ is very narrow and the contribution of the damped waves is not negligible. In what follows we will try to take this contribution into account and to determine its physical meaning.
States inside and outside the horizon {#sec:resonant}
-------------------------------------
Another important feature in the present consideration which differs from the situation described in [@prd57], when $\mu$ was equal to zero, is the appearance of the region $V_{+}$. In this region we also have two real solution of (\[eq:hj\]) and thus two different quasiclassical waves. But the nature of this region is quite different from that of the region $R_+$. We should stress that the coordinate $R$ under the horizon (including the region $V_+$ as well!) is actually a time coordinate, and the quasiclassical wave function \[quasi\] \~{ \_S P\_[S]{}dS} with $P_S>0$ represents a wave moving forward in time, while the solution (\[quasi\]) with $P_S<0$ represents a wave moving backward in time.
The classical particle trajectory under the horizon which starts near $R=0$, propagates backward in time up to the horizon, then is reflected from it and then propagates forward in time back to the singularity $R=0$.
According to the usual interpretation of waves propagating backward in time we might treat them as antiparticles which propagate forward in time, but with the opposite sign of energy. This is clear from (\[eq:hj\]). The energy of the particle is $\epsilon=\delta m =m_{\out}-m_{\In}$. If we take the solution of (\[eq:hj\]) with the minus sign before the logarithm in the r.h.s., we can write \[under1\] P\_S=- = + = + where $\tilde F_{\In}=F_{\out}$ and $\tilde F_{\out}=F_{\In}$. This means that $\tilde m_{\out}=m_{\In}$, $\tilde m_{\In}=m_{\out}$ and $\tilde E=-E$ — instead of treating part of the trajectory as a the trajectory of a particle of energy $E$ propagating back in time, we can treat as that of a particle of energy $-E$ propagating forward in time. Each solution describes the situation when either a particle or an antiparticle eventually falls into the singularity because there is a probability flow directed to $R=0$. Now we have to make the first step in the construction of the global quasiclassical solution in the configuration space: we should glue the waves in the $V_+$ region with the waves in the $R_+$ region. This may be done as usual by analytical continuation through the complexified configuration space. We will not consider below the continuation of the solutions to the $R_-$ and $V_-$ regions, so the above continuation will actually be made through the ordinary complex plane.
The integral in (\[quasi\]) can be calculated explicitly in the case in question and takes the form \_S P\_[S]{}dS = \_x x dx = + a(x-a)\[(x-a)-1\] |\^[a=1]{}\_[a=]{} (here $x=\sqrt{S}$).
We can continue this expression analytically from $V$ to the $R$ region along the contour situated far from the branching points $x=1$ and $x=\mu$. The only result of such a continuation is the appearance of the additional terms $i\pi$ in each logarithm: \[delta\] ((x-a)-+ i) + a(x-a)\[(x-a)-1+i\] |\^[a=1]{}\_[a=]{}.
As a result, we obtain that the quasiclassical wave function (\[quasi\]) acquires, after the analytical continuation, an additional factor in its amplitude \[cont\] A { \_S P\_[S]{}dS} A { \_S P\_[S]{}dS} ().
The second solution acquires the reversed factor \[cont1\] A {-\_S P\_[S]{}dS} A{ -\_S P\_[S]{}dS} (-).
So for a state with both waves having equal amplitudes in the $V_+$ region we have in the $R_+$ region outgoing and ingoing waves with amplitudes related to each other as in (\[cont\]), (\[cont1\]). The validity of the above consideration depends now on the following important problem to be analyzed. The quasiclassical anzatz (\[quasi\]) is not a good approximation for the true wave function not only at the points $S=1$ and $S=\mu^2$ — the branching points of the momentum $P_S$ — but also at the so-called Stokes lines (see e.g. [@wazov]). Thse lines are solutions of the equation \[stoks\] P\_S dS=0. To take seriously the above analytical continuation, we must be sure that the path in the complex plane along which we continue our quasiclassical solution does not intersect Stokes lines. Otherwise we can lose some important part of the quasiclassical solution which is exponentially small compared with the wave (\[quasi\]) before the Stokes line but becomes large after the intersection. But fortunately in our case it can be easily seen (analytically as well as numerically) that we can reach the region $(1,\infty)$ from $(0,\mu^2)$ through the complex plane without intersecting the Stokes lines.
Hawking radiation spectrum {#sec:hawking}
==========================
(\[eq:main\]) is a field equation for first-quantized self-gravitating massless particles in the field of a black hole. (We suppose that it is this equation that must replace the radial Klein-Gordon equation for the $s$-modes of a scalar field if we want to take into account its back reaction onto the gravitational field of the black hole.)
We suppose that the vacuum state of second quantized theory inside the horizon consists of zero-mode oscillations of particles with different energies. The natural property of the vacuum would be that all the possible zero modes with different energies $\epsilon$ are present with the same amplitude. A particle-antiparticle pair with the energy $\epsilon$ falling into the singularity is presented in our model as a solution which consists, under the horizon, of both quasiclassical waves (forward and backward with respect to the variable $R$ which is time-like in the $V$-region) with equal amplitudes.
Now from the previous section we know that such a state under the horizon gives us the ingoing and outgoing waves with the amplitudes $Z_{\In}$ and $Z_{\out}$, \[k\^2\] P=={-(1-\^2)}.
Let us look at the last formula in more details. We must recall the definition of $\zeta$ and rewrite (\[k\^2\]) in a more convenient form: \[P\] P= {-} = {- }. Introducing the area of the horizon $A$, we obtain finally \[P1\] P= {- }.
This result precisely coincides with the main result of Hartle and Hawking in [@h-h]. Following their line of reasoning, we can treat this probability distribution as the Gibbs distribution \[eq:temp\] P={-m/T}. We see that it follows from the comparison of the two distributions that the correct mass formula for the Schwarzschild black hole is valid: \[eq:mass\] m=T[A]{}/4.
Thus we have arrived at the conclusion that our first-quantized model for a self-gravitating particle describes such an important phenomenon of black-hole physics as the Hawking radiation.
[[startsection[subsection]{}[2]{}[0pt]{} [-3ex plus -1ex minus -.2ex]{}[1.4ex plus .2ex]{} [****]{}]{}\*[Acknowledgement]{} The authors are grateful to Russian Basic Research Foundation for financial support (Grant No 99-02-18524). The work of A.B. and A.N. was also supported by Soros Educational Programme for postgraduate students.]{}
[99]{}
G. ’t Hooft, [*Class. Quant. Grav. **16***]{}, 395–405 (1999); gr-qc/9805079.
A. Achucarro and P.K. Townsend, [*Phys. Lett. **B 180***]{}, 89 (1986);\
E. Witten, [*Commun. Math. Phys. **104***]{}, 207 (1986).
G.’t Hooft, “Dimensional Reduction in Quantum Gravity”, [*in:*]{} Salam-Festschrift: A Collection of Talks, World Scientific Series in 20th Century Physics, vol. 4, ed. A. Ali, J. Ellis and S. Randjbar-Daemi (World Scientific, 1993), THU-93/26, gr-qc/9310026;\
L. Susskind, [*J. Math. Phys.* ]{} [**36**]{}, 6337 (1995).
S. Carlip, “Black Hole Entropy from Conformal Field Theory in Any Dimension”, hep-th/9812013.
S.N. Solodukhin, “Conformal Description of Horizon’s States”, hep-th/9812056.
A. Ashtekar, J. Baez, A. Corichi and K. Krasnov, “Quantum Geometry and Black Hole Entropy”, gr-qc/9710007.
M. Banados, “Three-Dimensional Quantum Geometry and Black Holes”, hep-th/9901148.
A. Gorsky and N. Nekrasov, [*Nucl. Phys. **B436***]{}, 582 (1995), hep-th/9401017.
H.-J. Matschull and M. Welling, [*Class. Quant. Grav. **15***]{}, 2981-3030 (1998), gr-qc/9708054.
V.A. Berezin, A.M. Boyarsky and A.Yu. Neronov, [*Phys. Rev. **D 57***]{}, 1118 (1998).
V.A. Berezin, V.A. Kuzmin and I.I. Tkachev, [*Phys. Rev. **D 36***]{}, 2919 (1987).
V.A. Berezin, [*Phys. Rev. **D 55***]{}, 2139 (1997).
K.V. Kuchař, [*Phys. Rev. **D 50***]{}, 3961–3981 (1994); gr-qc/9403003.
A.Yu. Neronov, [*Phys. Rev. **D 59***]{}, 044023 (1999).
V.A. Berezin, A.M. Boyarsky and A.Yu.Neronov, “Black Hole Mass Spectrum vs. Spectrum of Hawking Radiation”, gr-qc/9808027.
J.L. Friedman, J. Louko and S.N. Winters-Hilt, “Reduced Phase Space Formalism for Spherically Symmetric Geometry with a Massive Dust Shell”, [*Phys. Rev. **D 56***]{}, 7674–7691 (1997).
P. Kraus and F. Wilczek, [*Nucl. Phys. **B 433***]{} 403 (1995).
J. Bekenstein and V. Mukhanov, [*Phys. Lett.* ]{} [**B 360**]{}, 7 (1995)
V.A. Berezin, A.M. Boyarsky and A.Yu. Neronov, “Quantum Mechanics of Self-Gravitating Particles and Quantum Black Hole Models”, [*preprint* ]{} INR RAS, 1999.
V.A. Berezin, A.M. Boyarsky and A.Yu. Neronov, “On the Exact Spectra of Relativistic Schroedinger Equation in Finite Differences”, gr-qc/9902028.
S.W. Hawking, [*Commun. Math. Phys. **43***]{}, 199 (1975); Nature [**248**]{}, 30 (1974).
J.B. Hartle and S.W. Hawking, [*Phys. Rev. **D13***]{}, 2188 (1976).
W. Wazov, “Asymptotic Expansions for Ordinary Differential Equations”, J. Wiley & Sons Inc., 1965.
![Penrose diagram for a Schwarzchild black hole. Dashed lines are curves of constant radius[]{data-label="fig:kruskal"}](fig2){width="\columnwidth"}
![Different space-times with a self-gravitating shell[]{data-label="fig:trajectories"}](fig1){width="\columnwidth"}
![The real section of the Riemannian surface $S_F$ (\[eq:sf\]) covers the real line twice[]{data-label="fig:sf"}](fig8){width="\columnwidth"}
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We report comprehensive (magneto)transport studies of the two-phase state in (TMTSF)$_2$ClO$_4$, where superconducting (SC) phase coexists with spin-density wave insulator (SDW). By tuning the degree of ClO$_4$ anion ordering in controlled manner we smoothly suppress the SDW state and study resulting evolution of the SC phase spatial texture. We find that as SDW is suppressed, SC regions initially appear inside the SDW insulator in a form of filaments extended in the interlayer direction and further merge into the two-dimensional sheets across the most conducting axis of the crystal. We demonstrate that almost all our results can be explained within the soliton phase model, though with several assumptions they can also be related with the creation of non-uniform deformations. We believe that the anisotropy is intrinsic to SC/SDW coexistence in various quasi one-dimensional superconductors.'
author:
- 'Ya.A. Gerasimenko$^1$'
- 'S.V. Sanduleanu$^{1,2}$'
- 'V.A. Prudkoglyad$^1$'
- 'A.V. Kornilov$^1$'
- 'J. Yamada$^3$'
- 'J.S. Qualls$^4$'
- 'V.M. Pudalov$^{1,2}$'
title: 'Coexistence of Superconductor and Spin-Density Wave in (TMTSF)$_2$ClO$_4$: Spatial Structure of the Two-phase State'
---
Introduction
============
Many low-dimensional and layered unconventional superconductors have similar phase diagrams in the sense that the superconductor phase neighbors the insulating (often magnetically ordered) phase, and the two phases may coexist with each other in a finite interval of doping or pressure [@Lebed08; @Stewart11; @Johnston10; @Chubukov12]. The well-known example is the quasi one-dimensional (Q1D) organic (TMTSF)$_2$PF$_6$[@Lebed08] compound (hereafter denoted as PF6), where superconductivity (SC) coexists with antiferromagnetically ordered spin-density wave (SDW) state in a narrow pressure region on the $T-P$ phase diagram. This interesting region was extensively studied using macroscopic (transport[@Vuletic02; @Kornilov04; @Lee05]) and local (NMR[@Yu02; @Lee05]) probes. Both types of measurements provide solid evidence for segregation of SC and SDW phases in real space. The phase segregation at the first glance can be explained within the framework of simple thermodynamic arguments, as a result of interplay between elastic and SDW energies [@Vuletic02; @Kornilov04; @Lee05] at the first order transition. This is in agreement with the transport measurements suggesting a large macroscopic spatial size of the metal(SC) regions [@Vuletic02; @Lee05]. NMR measurements however show deviations from this simple picture [@Yu02; @Lee05]. An order of magnitude increase of the SC critical field in the two-phase state, as determined from transport measurements [@Lee02], also suggests a smaller size of the SC domains. The above inconsistency of the experimental data suggests that the spatial structure of the two phase state may be highly anisotropic and non-trivial, and therefore requires more targeted studies.
A number of theories was put forward to explain the coexistence of metal(SC) and SDW phases in PF6. However, some of them do not envisage any specific spatial texture. Among them is the SO(4) symmetry treatment [@Podolsky04] of the specific problem of SDW coexisting with triplet SC, similar to that for the purely 1D systems. In another approach [@Grigoriev08] a semimetallic state of a non-fully gapped SDW is considered, where SC pairing occurs within electron and hole pockets.
Some other theories inferred spatial texture specifically for the SC and SDW coexistence and their reasoning is not applicable to metal/SDW separation at temperatures above the SC onset. In one of them the shape of SC domains is determined by the effective mass anisotropy [@Lee02]. Another theory infers square lattice of domains from the entanglement of SDW wave vector and the magnetic moment of triplet SC [@ZhangDeMelo06].
An alternative scenario is based on the SDW order parameter variation in soliton phase (SP)[@GorkovGrigoriev05; @GorkovGrigoriev07; @Grigoriev09]. The SP theory suggests that SDW order parameter can become nonuniform with metallic(SC) domain walls emerging as sheets normal to the most conducting chain direction $\mathbf{a}$. However, recent transport anisotropy measurements in PF6[@Kang10] revealed that SC develops in a different manner, first in a form of filaments along the least conducting (interlayer) $\mathbf{c}$ direction and only for higher pressures transforms to 2D sheets. Whereas this observation can resolve the inconsistency between the transport and NMR data, the origin of the spatial segregation and its anisotropy still remains unknown.
The emergence of the metallic(SC) phase along with SDW suppression is intimately related to the creation of ungapped carriers. The latter is governed by the Fermi surface (FS) nesting, which easily occurs in Q1D systems, where FS is a pair of slightly warped sheets. In experiments with PF6 the FS nesting is altered by pressure, which also affects lattice parameters. Indeed, under pressure nesting is spoiled due to FS warping caused by transfer integrals increase [@Yoshino01; @Yamaji82]. This entanglement prevents one from separating the influence of variations of unnested carriers density from that of the lattice spacing changes with pressure. Consequently, to address this problem it is necessary to study the SC and SDW coexistence at a fixed pressure by changing some other external parameter controlling the FS nesting.
Such an opportunity is provided in (TMTSF)$_2$ClO$_4$ (hereafter denoted as ClO4), where the SC and SDW phases are known to coexist at ambient pressure [@Schwenk84; @Gerasimenko13]. In this material, instead of pressure one can vary the degree of dipole ordering of ClO$_4$ tetrahedral anions. Due to slow kinetics of anion ordering[@Pouget90], the latter can be varied over wide range by choosing the appropriate cooling rate in the vicinity of ordering transition temperature $T_{AO}=24$K. This possibility opens a way of exploring the remarkable phase diagram with a continuous transition from the SDW state at high disorders to the homogeneous SC state at low disorders [@Schwenk84; @Gerasimenko13].
Although the effect of anion ordering on conduction electrons is complicated, the most pronounced is the bandstructure folding in the interchain direction caused by the doubling of lattice period along the $\mathbf{b}$-axis in the ordered state. As a result, in the ordered state due to folding the FS splits in four sheets and nesting is spoiled. When anions are disordered the FS is not different from that in PF6 and is almost perfectly nested. It was suggested therefore, that SC/SDW coexistence in this system is due to the loss of long-range anion order as cooling rate is increased[@Schwenk84].
An alternative explanation[@ZanchiBjelis01] suggests that the degree of ordering affects the magnitude of the FS splitting, the so-called dimerization gap $V$. It was shown theoretically[@ZanchiBjelis01; @SenguptaDupuis01], that the SDW onset temperature gradually decreases as the gap $V$ grows. This suggestion is supported by our recent angular magnetoresistance experiments[@Gerasimenko13], where we traced the FS evolution with disorder. Particularly, upon increasing cooling rate, we observed gradual growth of interchain bandwidth (and hence decrease of the gap $V$) ending eventually with SDW onset. Furthermore, in the two-phase SC/SDW state the temperature dependences of resistance showed hysteresis[@Gerasimenko13], which could not be ascribed to the loss of anion order. These observations allowed us to conclude, that the anion ordering induced FS splitting controls the phase coexistence.
The above results motivated us to perform measurements of the spatial texture of the SC phase embedded in the SDW background in ClO4 when only a single band parameter, the dimerization gap $V$, was varied. By tuning $V$ we drove the system through the SC/SDW phase boundary at the $T-V$ phase diagram, keeping the dispersion in the interlayer direction unaffected. We found that the SC regions arise first as filaments along $\mathbf{c}$ axis and then merge into sheets in the $\mathbf{b-c}$ plane. Our observations reproduce qualitatively the behavior observed in PF6 under pressure[@Kang10]. We conclude, therefore, that the SC spatial texture depends neither on details of band structure, as it was proposed in Ref. , nor on character of lattice deformations under pressure. Such texture represents then the intrinsic way of allocating unnested carriers that appear with suppression of SDW state irrespective of their exact origin.
The observed anisotropy can be explained qualitatively within the soliton phase scenario for strongly anisotropic lattice, with no additional assumptions. Indeed, within the SP theory the energy gain comes from alignment of solitons in walls across the molecular chains, i.e. the creation of soliton band. The energy cost is primarily caused by the growing repulsion of solitons on the same chain as the number of walls increases[@BrazovskiiGorkovLebed82]. When one constructs a wall by adding another soliton in it, the width of the resulting soliton band will be much bigger for $\mathbf{b}$-axis due to larger transfer integral compared to that for $\mathbf{c}$-axis. This allows one to create longer walls along the $\mathbf{c}$-axis before the soliton band becomes larger than the SDW gap. This process leads to stronger repulsion (i.e., energy cost) of walls oriented along $\mathbf{b}$-axis, thus making the $\mathbf{c}$-axis alignment of solitons more favorable. As the number of unnested carriers increases, the $\mathbf{c}$-axis soliton band will become fully occupied; further, the solitons will start aligning also along the $\mathbf{b}$-axis thus forming the $\mathbf{b-c}$-plane domains.
Experimental
============
The major experimental idea of our paper is to probe the SC/SDW phase diagram using the controlled variation of the degree of anion ordering. In this way we want to determine how the zero-resistance SC state emerges within the insulating SDW state in real space. Thus, the proper starting point for such experiment is to prepare our samples in the most homogeneous SDW state, i.e. to prevent anions from ordering as much as possible. This is achieved by rapid cooling, or “quenching”, of samples from some temperature $T_Q>T_{AO}$ above anion ordering temperature. Smaller degrees of ordering can be achieved by slower cooling, however such procedure would require careful quantification of the resulting disorders. Instead, we have chosen to anneal the quenched sample at some temperature below $T_{AO}$, which is also known to cause gradual ordering of anions[@Pouget90]. In this way we know exactly that each consecutive disorder is weaker. Furthermore, the use of annealing prevents us from affecting other sample properties except the degree of ordering after the initial disorder was created. In contrast, consecutive rapid coolings could result in uncontrollable change of sample state (e.g. create defects).
It should be emphasized though that quenched samples are essentially spatially inhomogeneous on a small scale. A prerequisite for this is the slow anion ordering [@Pouget90], much slower than cooling rates used in experiment. As a result, small anion-ordered inclusions are formed in a disordered background. Whereas we cannot avoid creating them during cooldown, their number can be adjusted by the sample preparation. In what follows we present the results for various quench parameters and show the insignificant influence of these inclusions on anisotropy of the spatial texture of the SC phase.
In our experiment we vary the two independent parameters: quench rate and quench temperature. The former determine predominantly the degree of ordering, since the anion-ordered volume fraction tends to saturate for higher rates[@Pouget90]. The role of $T_Q$ is more complicated. On the one hand, in Ref. the degree of ordering was found to increase as the quench temperature decreased. On the other hand, varying $T_Q$ changes fraction of the ordered inclusions due to fluctuations of anion ordering noticeable below $\sim 30$K as observed in Ref. .
To characterize the resulting disorder we measured two parameters: the transition temperature and the width of the SDW transition. The former depends on the ordering-induced band-splitting $V$, namely, SDW onset at high temperatures reflects low degree of ordering [@Gerasimenko13; @ZanchiBjelis01]. Typical values of transition temperatures achieved after quenching in our measurements, $6-6.5$K, are close to those observed experimentally by other groups [@Schwenk84; @Qualls00]. The width of SDW transition is small for high quench rates and $T_Q$-temperatures; it increases as $T_Q$ decreases. Therefore, narrow and high-temperature SDW onset is the signature of weakly ordered state with low amount of anion-ordered inclusions. Wide and high-temperature onset indicates weakly ordered state with high amount of inclusions.
We deduce the spatial structure of the two-phase state from resistance measurements. In our experiment we prepare samples in such a way, that two resistivity components can be measured on the same sample (see Fig. \[fig\_sample\]). In this way we extract the anisotropy for the two of three planes by performing simultaneous measurements on different contacts for the same disorders. In the case of the third plane we quench two samples simultaneously to achieve close initial conditions and compare the results. To provide more support for these results we also measure SC critical fields, $H_{c2}$, and deduce from these data additional information on the spatial texture[@Lee02].
(TMTSF)$_2$ClO$_4$ single crystals of typical dimensions $x\times y\times z=3\times0.1\times0.03$mm were synthesized by a conventional electrochemical technique. Three resistance components, $R_{xx}$, $R_{yy}$ and $R_{zz}$, along the orthorhombic principal axes, $\mathbf{a}$, $\mathbf{b}^\prime$ and $\mathbf{c}^*$, were measured using conventional 4-wire low-frequency AC-technique. In order to measure $R_{xx}$ and $R_{zz}$ ($R_{xx}$ and $R_{yy}$) components eight annealed 10um Pt wires were glued with a conducting graphite paint on the opposing faces normal to the $\mathbf{c}^*$ ($\mathbf{b}^\prime$) axis, correspondingly (see Fig. \[fig\_sample\]). To relieve stress that could appear during cooldown, samples were left suspended above the holder on the long Pt wires. Since the crystals are small, measurements in $xx-zz$ and $xx-yy$ configurations were performed on separate samples, in order to avoid conduction shunting with contact pads. Each of two resistivity components on a single sample were measured sequentially. Care was taken to align contact pads opposite to each other to minimize the admixture of other resistivity components.
Resistances measured in this way do not represent precisely the corresponding bulk resistivities, because of somewhat inhomogeneous current distribution in the sample bulk. The latter affects mostly the $R_{xx}$ resistance in the $xx-zz$ sample. Indeed, in that case due to high resistivity anisotropy current spreads into the bulk only partially, which leads to the admixture of $\rho_{zz}$ component [@Buravov94]. In contrast, for $xx-yy$ sample $\rho_{zz}$ is shunted by the contact pads (see Fig. \[fig\_sample\]), which allows us to separate intrinsic features of $\rho_{xx}$ temperature dependence from $\rho_{zz}$ ones. As we will show below, the $\rho_{zz}$ potential admixture does not affect the results of our observations.
![Two resistivity measurement configurations used in this experiment: $xx-zz$ (top) and $xx-yy$ (bottom). Black rectangles show schematically contact pads (or graphite paint droplets) to which the Pt wires were glued.[]{data-label="fig_sample"}](fig1.eps){width="47.00000%"}
Samples were mounted on a top-loading $^3$He probe and aligned at room temperature under microscope to within a few degrees for measurements in magnetic field $\mathbf{H\|c^*}$. To improve temperature homogeneity during cooldown, holders with samples were put inside the copper casing. The probe with the sample was initially cooled from room temperature to a certain quench temperature $T_Q \gtrsim T_{AO} =24$K. The cooling rate at this stage was chosen $0.2-0.3$K/min in order to avoid cracks. Further the samples were quenched from $T_Q$ by rapidly bringing the sample holder in contact with the 1K stage. The quench rates were determined from either readings of the thermometer installed on the copper casing or from comparison of the sample resistance with the equilibrium $R(T)$ dependence, both demonstrating similar results. The cooling rates obtained this way were as high as 600K/min in the vicinity of $T_{AO}$. Somewhat smaller quench rates, e.g. 100K/min, were achieved after rapid heating the cold copper casing above $T_{AO}$ by its subsequent free cooling. After the sample was quenched, weaker disorders were obtained by annealing. Annealing temperatures were gradually increased from 15 to 23K in our measurements, causing progressively weaker disorders; annealing below 15K had negligibly small effect on $T_{SDW}$. To achieve the fully anion-ordered state, samples were cooled from 40K to 20K for 12 hours.
Quench cooling could in principle produce different kinds of inhomogeneities or strains in the sample. To see how this affects our data, we have measured low-temperature $R_{xx}(T)$-dependences on the opposite surfaces and, in addition, $R_{zz}(T)$-ones using various opposing contact pairs (see Fig. \[fig\_sample\]). Although the resistance values were slightly different, all the essential features of the corresponding dependences, including the SDW onset, were reproduced with high accuracy on different contact pairs. Since SDW transition temperature is defined by the band splitting $V$, we deduce that there are no large-scale spatial inhomogeneities of anion ordering. This result is also in accord with the previous direct measurements of $V$ from angular dependences of magnetoresistance [@Gerasimenko13].
The potential strains might originate from different thermal expansion coefficients of the sample and conducting paint used for making contacts. From the Debye temperature estimates of $\theta_D=213$K[@Garoche82], we conclude that for $T_Q\le40$K the thermal expansion of a sample is negligibly small, according to the Grüneisen formula[@Ziman72], and almost no additional strain is produced by quenching. Rapid cooling might also relieve the strain generated at higher temperatures. It is believed to create microcracks in a sample, which in their turn can create parallel conduction channels predominantly along the interlayer $\mathbf{c}$-axis[@Oh04]. Such channel was suggested to alter $R_{zz}(\mathbf{H\|b^\prime})$ dependence, leading to saturation of magnetoresistance in high fields[@Oh04]. However, in our samples we did not observe significant shunting in fields up to 13Tesla, indicating the insignificance or absence of microcracks.
Our studies have been done with six single crystals from two batches; all of them demonstrated qualitatively similar behavior. Below we present the results for the samples cut from three different crystals. The data for each of the three crystals is shown in Figs. \[fig\_xxzz\]–\[fig\_zzyy\] correspondingly. The curves within a set of disorders are labeled relatively to each other in such a way, that \#1 corresponds to the most ordered state, and the highest number corresponds to the quenched state.
Results
=======
Anisotropy in the $\mathbf{a-c}$ plane
--------------------------------------
The set of temperature dependences of $R_{xx}$ and $R_{zz}$ resistances for the same sample at various disorders corresponding to different parts of the SC-SDW phase diagram is presented in Fig. \[fig\_xxzz\]. This set covers the full phase diagram (Fig. \[fig\_xxzz\]a) where the $R_{xx}(T)$ temperature dependence gradually changes its character from insulating to metallic and then to superconducting one. For strong disorders ($\#$6-10) the sample undergoes SDW transition, followed by insulating temperature dependence characteristic of a gapped SDW state. The dependence is not activated, suggesting that the insulating background contains metallic inclusions. However, their fraction is apparently too small, below the percolation threshold, since no downturn in $R(T)$ (signalling the superconductivity onset) is observed at low temperatures, except for the kink at $T\approx1$K, where the $R_{xx}(T)$ slope becomes smaller. Remarkably, for the same disorders (\#\#6–10) $R_{zz}(T)$ demonstrates complete superconducting transition with $T_c \approx1$K (see Fig. \[fig\_xxzz\]b). Moreover, the SDW onset seen in $R_{xx}(T)$ coincides with the steep increase of $R_{zz}(T)$ resistance at high temperatures, however below $T_{SDW}$ it shows a downturn (\#\#6–10) in contrast to $R_{xx}(T)$.
These observations show that at strong disorders the sample consists of metallic(SC) regions embedded in the SDW insulating background. Whereas they form a continuous SC path in the interlayer direction, inside the layers they are essentially separated from each other. The change of the $R_{xx}(T)$ slope below 1K is observed for all the measured samples and is clearly connected with the SC onset. Indeed, the superconducting state emerging inside the metallic regions is expected to improve shunting of the SDW background at low temperatures due to emerging Josephson coupling between the neighboring regions.
The major effect of disorder in ClO4 is the variation of the band splitting $V$ - the deviation from perfect FS nesting, as was discussed above. Therefore, as disorder weakens, the corresponding growth of the number of unnested carriers is expected to increase the metallic(SC) fraction; such behavior was indeed observed in PF6 upon approaching SDW endpoint[@Vuletic02].
In accord with the latter expectation, one can see that superconductivity smoothly emerges on the $R_{xx}(T)$-dependence for weaker disorders (\#\#5–1), similar to the behavior observed in granular superconductors [@Deutscher80]. More detailed analysis indeed shows signatures of the percolation type transition. As disorder weakens, the dip appears at about 1K on the insulating $R_{xx}(T)$ dependence (curve \#5). Such reentrant behavior indicates the increase of metallic(SC) fraction, though it is still lower than percolation threshold. The dip is followed by metallic (\#4, \#3) temperature dependence, which suggests emergence of superconducting paths along the $\mathbf{a}$ axis. Indeed, both the downturn and the metallic behavior disappear in magnetic field (see Fig. \[fig\_xxzz\]c), which unambiguously indicates their superconducting nature.
As disorder weakens, the superconductivity onset temperature remains almost constant, $T_c\sim 1$K, whereas the concomitant resistivity drop gradually enhances (\#2, \#3), until $R_{xx}$ vanishes to zero along both axes, $\mathbf{a}$ and $\mathbf{c}$ (\#1). The latter is consistent with anion ordered ClO4 being a homogeneous superconductor [@Lebed08; @Pesty88]. We note that such behavior of $T_c$ and $R(T)$ is similar to smooth development of the percolation transition in granular superconductors[@Deutscher80].
![Temperature dependences of (a) intralayer, $R_{xx}$, and (b) interlayer, $R_{zz}$, resistances for a set of disorders. Sample was cooled at the rate of approx. 100K/min from $T_Q=27$K. Arrows in panels (a) and (b) indicate SDW transition temperatures, $T_{SDW}$. The latter were determined as a peak value of $\mathrm{d}\ln R_{xx}/\mathrm{d}(1/T)$, as shown for the strongest disorder (\#10) in panel (d). Panel (c) shows magnetic field $H||c$ effect on $R_{xx}(T)$ dependence for one of disorders (\#4). []{data-label="fig_xxzz"}](fig2.eps){width="45.00000%"}
![Temperature dependences of intralayer (a) $R_{xx}$ and (b) $R_{yy}$ resistances for a set of disorders. Sample was cooled at the rate of approx. 600K/min from $T_Q=30$K. Inset in panel (a) shows the Arrhenius plot of $R_{yy}(T)$ for the strongest disorder. Inset in panel (b) shows the hysteresis $R_{yy}(T)$-dependence for disorder $\#7$. Two trajectories of temperature sweep: the large loop (a-e-f-b-c-d-a) includes $T_{SDW}\approx 2$K, whereas (c-d-e-f-c) takes place completely at $T<T_{SDW}$.[]{data-label="fig_xxyy"}](fig3.eps){width="45.00000%"}
Anisotropy in the $\mathbf{a-b}$ plane
--------------------------------------
Figure \[fig\_xxyy\] shows disorder evolution of temperature dependences of the intralayer resistances $R_{xx}$ and $R_{yy}$. The SDW transition is much sharper for strong disorders and the resistivity temperature dependence is closer to the activated one, at least above $T\approx1$K (cf. inset on Fig. \[fig\_xxyy\]a). Below 1K the slope of $R_{yy}(T)$ decreases; this decrease is lifted in small magnetic field of the order of the expected $H_{c2} \sim 0.1$T. We ascribe therefore the slowdown of $R_{yy}(T)$ to shunting caused by small amount of metallic(SC) regions, same as discussed above for $xx-zz$ sample. This effect is more pronounced for the latter sample (e.g. \#10 in Fig. \[fig\_xxzz\]a) which indicates larger amount of the metallic(SC) regions. This difference is connected with the lower quench rate in the vicinity of $T_{AO}$ as will be discussed in the following sections.
For the strongest disorder, both resistivity components show qualitatively similar temperature dependences (curves \#12 in Figs. \[fig\_xxyy\]a,b). However, for somewhat weaker disorders (\#\#10–7) the $R_{yy}(T)$-dependence is non-monotonic; there firstly a dip (\#10, \#9) and then a downturn (\#8,\#7) appears below 1K on the insulating background. For these same disorders, $R_{xx}(T)$-dependence remains monotonic and insulating. The non-monotonic features observed in $R_{yy}(T)$ indicate the superconductivity onset, similar to that described above for $xx-zz$ sample. As disorder weakens further (\#\#6–3), the downturn appears also in $R_{xx}(T)$, and finally both resistivity components show complete superconducting transition (\#2, \#1), as expected in the anion-ordered case. We emphasize, that the change of slope below 1K and appearance of the downturn in $R_{xx}(T)$ does not depend on the measurement configuration, $xx-zz$ or $xx-yy$, and thus adequately represents the $\mathbf{a}$-axis spatial texture.
Therefore, the results shown in figures \[fig\_xxzz\] and \[fig\_xxyy\] demonstrate that the metallic or superconducting paths emerge normal to the chains $\mathbf{a}$ axis.
Hysteresis behavior
-------------------
One of the most remarkable features of transport properties in the region of metal(SC) and SDW coexistence is the hysteresis in temperature dependence of resistances between cooling and heating curves (cf. \#10 on Fig. \[fig\_xxzz\]b and \#7 on the inset in Fig. \[fig\_xxyy\]b). When a sample is heated from the SC state, the resistance is larger compared to that observed on its cooling from the high-temperature metallic state. Furthermore, the hysteresis vanishes only in the vicinity of the SDW transition. The hysteresis is a stationary effect and does not depend on the temperature sweep rate. Both its width $\delta T\equiv T_\uparrow(R) -T_\downarrow(R)$ and amplitude $\delta R/R(T=\rm const)$ increase as disorder weakens. The hysteresis is observed only when the $R(T)$-dependence is insulating (i.e. the SDW state manifests). When $R(T)$ is metallic the cooling and heating traces coincided within experimental uncertainties (see, e.g. curves \#\#3–1 in Fig. \[fig\_xxyy\]a). Such a behavior clearly indicates history effects in the spatial distribution of the two-phase state and will be discussed in more details later.
Quench rate effect
------------------
Although the maximum $T_{SDW}$ value for a given disorder is almost the same for $xx-zz$ and $xx-yy$ samples, somewhat different $R_{xx}(T)$ behaviors impede their direct comparison. For example, the dip on the $R_{xx}(T)$ dependences which is present on curve \#5 in Fig. \[fig\_xxzz\]a for $xx-zz$ sample is absent on curve $\#$10 in Fig. \[fig\_xxyy\] for $xx-yy$ one with similar $T_{SDW}$ value. This fact indicates that the former sample has larger metallic(SC) phase fraction, as discussed above for the $xx-zz$ sample. This prevents one from figuring out the anisotropy of the SC phase spatial texture in the $\mathbf{b-c}$-plane. The only parameter different for the two samples is the quench rate, which was lower for $xx-zz$ one. Since the amount of metallic phase is lower for higher quench rate, the most direct comparison of the differences in the SC onset can be obtained for samples quenched (i) rapidly and (ii) at the same rate.
![Temperature dependences of (a) interlayer $R_{zz}$ and (b) intralayer $R_{yy}$ resistances for a set of disorders. Samples were cooled several times at the rate of approx. 600K/min from $T_Q=40$K. The downturn on $R_{yy}(T)$ for curves \#15, \#16 was absent for the first cooling (see inset in panel (a)) and appeared after numerous rapid coolings, and therefore might be an artifact. It is also not pronounced at lower disorders. Inset in panel (a) shows the curves obtained after the first cooling of this sample at the rate of $\approx600$K/min. Inset in panel (b) shows the behavior of $R_{yy}$ and $R_{zz}$ on the same sample for moderate cooling rates. Dashed line indicates the position of the SC transition onset seen in $R_{zz}$ and the corresponding dip in $R_{yy}$. []{data-label="fig_zzyy"}](fig4.eps){width="45.00000%"}
Anisotropy in the $\mathbf{b-c}$ plane
--------------------------------------
To figure out the spatial anisotropy in the $\mathbf{b-c}$ plane we cut a sample in two parts to measure $R_{yy}$ and $R_{zz}$ separately and quenched both parts simultaneously several times at very high rates (up to 600K/min in the vicinity of T$_{\rm AO}$) to achieve similar conditions. The results of the $R_{yy}(T)$ and $R_{zz}(T)$ simultaneous measurements are presented in Fig. \[fig\_zzyy\]. The abrupt SDW onset along with the pronounced activated temperature dependence indicate that the amount of the ordered phase is small, as discussed above in relation with the quench rate effect. Already at this strongest disorder the $R_{yy}$ and $R_{zz}$ curves are notably different, though the anion-ordered fraction is low and dimerization gap is small. Indeed whereas the $R_{yy}(T)$ dependence deviates from activated behavior in the vicinity of the SC onset, the $R_{zz}(T)$ dependence deviates at all temperatures indicating the substantial metallic shunting in the interlayer direction.
The curves at intermediate disorders show distinct difference between the $R_{yy}(T)$- and $R_{zz}(T)$-behaviors in the vicinity of SC onset (see inset in Fig. \[fig\_zzyy\]b). The interlayer $R_{zz}(T)$ exhibits the downturn associated with the SC onset (see also Fig. \[fig\_xxzz\]c); in contrast the intralayer $R_{yy}(T)$ retains the insulating character with only minor deflection at $T_c$. Similar deflection transforming to a dip was observed also in $R_{yy}(T)$ for $xx-yy$ sample (cf. \#11-9 in Fig. \[fig\_xxyy\]a) and indicates the emergence of continuous SC paths along $\mathbf{b}$-axis. Thus, the above measurements demonstrate that SC regions arise first along the $\mathbf{c}$-axis.
![Temperature dependences of critical field $H_{c2}||c^*$ for a set of disorders from Fig. \[fig\_zzyy\]. $H_{c2}$ is determined as the onset ($99\%$) of resistive $R_{zz}(H)$ transition. Inset shows disorder dependence of critical field $H_{c2}$ at normalized temperature $t=T/T_c=0.3$. Various disorders on the horizontal axis are characterized by the corresponding resistance $R_{zz}(T=8\,K)$ normalized to that in the anion-ordered state. []{data-label="critfields"}](fig5.eps){width="45.00000%"}
Superconducting critical field
------------------------------
An independent probe for the in-plane texture of a superconducting phase is the critical field $H_{c2}||c^*$. Both the critical field and the slope $dH_{c2}/dT$ for this direction are inversely proportional to the in-plane coherence lengths $\xi_x \xi_y$. In the limit of small-sized SC regions, $d_i\le\xi_i$, the size of the region $d_i$ can be substituted for the coherence length $\xi_i$. Such a substitution can be justified at least for strong disorders [@note1]. Since we have figured out above that the SC paths emerge first across the $\mathbf{a}$-axis, we conclude that $d_y$ grows much faster than $d_x$ with decreasing disorder. Therefore the growth of the critical field with disorder (at strong disorders) can be associated with the domain size, predominantly along the $\mathbf{b}$-axis, $d_y$.
There is no complete SC transition observed for strong disorders and the SC onset manifests only in the $R_{zz}(T)$ downturn (seen, e.g. in Fig.3a). We use the disappearance of this downturn in magnetic field as an indicator for SC suppression and extract $H_{c2}$ in this way. The $H_{c2}(T)$-dependences are presented on Fig. \[critfields\]. One can see that both the slope $dH_{c2}/dT$ and $H_{c2}$ at low temperatures are approximately constant for strong disorders \#\#11–14 (see the inset in Fig. \[critfields\]). These observations demonstrate that for strong disorders \#\#11–14 the SC regions do not grow significantly in the $\mathbf{a-b}$-plane. In contrast, the enlargement of the regions along $\mathbf{c}$-axis is observed, as evidenced by the downturn that appears on insulating $R_{zz}(T)$-dependence below 1K for disorder \#14 but is absent for stronger ones (cf. \#16 on Fig. \[fig\_zzyy\]).
For weaker disorders \#\#9–7 critical field decreases, indicating increase of $d_y$. This is also supported by the development of the deflection in $R_{yy}(T)$-dependence in the vicinity of $T_c$ (cf. inset in Fig. \[fig\_zzyy\]b). Upon further decrease of disorder the low-temperature $H_{c2}$ tends to saturate (disorders \#\#7–2 in the inset in Fig. \[critfields\]), which is expected for $d_y>\xi_y$. The overall above behavior demonstrates, that as disorder weakens the superconducting regions grow first in the interlayer direction, and further, at intermediate disorders, start increasing along $\mathbf{b}$-axis.
Interestingly, $H_{c2}$ sharply increases again in the anion ordered state (see \#1 in Fig. \[critfields\]). Obviously, this effect has nothing to do with the spatial dimensions of SC regions, because the ordered state is believed to be the homogeneous superconductor[@Pesty88]. Similar behavior was also observed for weak disorders in (TMTSF)$_2$ClO$_4$[@Joo05] and (TMTSF)$_2$(ClO$_4$)$_{1-x}$(ReO$_4$)$_x$[@Joo04]. This intriguing issue however is out of the scope of the present paper and obviously requires further investigation.
Discussion
==========
From our measurements of resistivity components and critical fields we conclude that (i) the spatial texture of the metallic(SC) phase emerging in the background SDW state evolves from strongly anisotropic to almost isotropic, and (ii) its relative fraction depends on prehistory. In this section we will discuss both these results within two competing approaches. One of them is the emergence of soliton phase [@BrazovskiiGorkovLebed82; @GorkovGrigoriev05], which has intrinsic spatial anisotropy, and another is based on phase segregation [@Vuletic02; @Lee05] in anisotropic lattice. Firstly we will address the question of whether the existence of anion-ordered (AO) regions can be responsible for the observed anisotropy.
Role of the AO
--------------
Having no our own quantitative data on the fraction of the anion-ordered regions (where the metal(SC) phase is favorable), we can however make an estimate based on other available results. Earlier X-ray diffraction measurements[@Pouget90] showed, that at a 100K/min quench rate, about 20% of anion-ordered phase survives in small regions with average dimensions of $l_x=30$, $l_y=60$ and $l_z=50$nm along $\mathbf{a}$, $\mathbf{b}$ and $\mathbf{c}$ axes, respectively. As disorder weakens, the number of AO inclusions grows while their size remains almost the same. Assuming that the anion ordered inclusions have similar $l_y$ and $l_z$ size, we conclude that this 20% fraction is well below the 2D percolation threshold in the $\mathbf{b-c}$-plane. Furthermore, were the superconductivity onset a percolation transition in the $\mathbf{b-c}$-plane, it should be almost isotropic, which contradicts the present results. Finally, such transition obviously cannot account for the hysteresis of the $R(T)$ dependence.
At the same time, a larger number of AO inclusions, obtained at lower quench rates, apparently increases the number of metallic(SC) regions without affecting the whole spatial texture. Indeed, in this case we observe a drastic difference in $R_{zz}(T)$ behavior: for rapidly quenched sample ($\sim600$K/min in the vicinity of $T_{AO}$) the transport along $\mathbf{c}$-axis is insulating (e.g. \#16 on Fig. \[fig\_zzyy\]a), whereas at lower quench rates ($\sim100$K/min) it shows signatures of a metallic behavior (e.g. \#9 on Fig. \[fig\_xxzz\]b). Our simple explanation is that the AO inclusions may bridge the emerging metallic(SC) regions; the former are almost isotropic [@Pouget90] and do not affect the anisotropy of the overall spatial texture.
SDW suppression by imperfect nesting
------------------------------------
Since anion ordering can account for neither the anisotropy, nor hysteresis, we conclude that the observed two-phase state is intimately related to SDW properties. The SDW is being suppressed along with the creation of unnested carriers. When pressure is a driving parameter (as in PF$_6$), nesting is most strongly deteriorated by 2D dispersion, essentially the increase of interchain ($b$-axis) hopping. Within the tight-binding model the nesting deviations are represented predominantly by the next-to-nearest interchain transfer integral, $t_b^\prime$[@Yamaji82].
On the other hand, for Q1D systems it is known that magnetic field $\mathbf{H\|c}$ tends to reduce the size of electrons wave function to a single chain (see e.g. [@Lebed08]), thus acting opposite to pressure. This effect was predicted to restore SDW onset at higher temperatures[@Montambaux88]. The predictions of this so-called imperfect nesting theory were indeed observed in various $T_{SDW}(H,P)$ measurements [@Danner96; @Gerasimenko09; @Biskup95; @Matsunaga01]. In the same way, measurements of $T_{SDW}$ behavior in magnetic field $\mathbf{H\|b}$ allows one to investigate the effect of interlayer dispersion. The latter due to its small value was observed to play minor role in SDW suppression [@Danner96; @Gerasimenko09].
The above effect of pressure is observed for both PF6 and ClO4[@Kang93], but for the latter compound nesting can be spoiled also by increasing the band splitting parameter $V$. It originates from charge disproportion between the neighboring chains in the $\mathbf{a-b}$ plane caused by displacement of ClO$_4^-$ tetrahedra [@Pouget90; @Zhang05; @LePevelen01] with respect to the TMTSF molecules. Thus, the effect of the band splitting is quite similar to that of doping, but instead of changing number of carriers in a fixed bandstructure, the anion ordering shifts bands with respect to a fixed chemical potential. This is reminiscent of “physical doping” in layered cuprates where under applied electric field electrons are transferred between neighboring layers [@Koval10].
As the deviations from nesting increase, the indirect gap $\Delta_0-\epsilon(\mathbf{k}_\perp)$ may become negative, e.g. the bottom of conduction band in the center of Brillouin zone may become lower than the top of valence band at its border [@BrazovskiiGorkovSchrieffer82]. As discussed in Refs. , at this point either electron and hole pockets will appear atop of the gapped Fermi surface, or the excess carriers can be trapped into solitonic midgap states resulting in a non-uniform SDW order parameter. The former scenario can naturally lead to the phase separation if the SDW energy depends on deformation[@Lee05], whereas the latter one incorporates a built-in spatial texture coming from alignment of solitons.
Soliton phase scenario
----------------------
The energy cost of a single soliton creation in a molecular chain is fixed and depends only on the SDW direct gap, $2\Delta_0/\pi$. The alignment of individual solitons in a wall across the chains allows double occupation of soliton levels and gains kinetic energy. The former makes the soliton band on average half-filled [@BrazovskiiGorkovLebed82]. Kinetic energy is obviously larger for longer walls and for stronger hopping within them. Therefore, the soliton phase becomes favorable, if the density of unnested carriers, i.e. the length of a wall becomes high enough, so that the energy gain overcomes the cost of creating solitons[@BrazovskiiGorkovSchrieffer82; @BrazovskiiGorkovLebed82]. This process was predicted to happen in PF6 before the semimetallic state emerges[@GorkovGrigoriev05].
In three-dimensional case, the soliton walls can be created along either $\mathbf{b}$ or $\mathbf{c}$-axis, or as 2D sheets in a $\mathbf{b-c}$-plane. The energy gain is the largest for 2D sheets, whereas 1D walls along $\mathbf{c}$-axis are the least favorable, because of small hopping integral in this direction. The above reasoning clearly contradicts spatial anisotropy observed in this paper.
The case of ClO4 is especially interesting in this context, since the emerging anisotropy is not the effect of the interplay of band parameters. The only quantity that grows as band splitting $V$ is increased is the number of unnested carriers. We observe that metallic(SC) domains are formed first along $\mathbf{c}$-axis and then merge/grow in the $\mathbf{b-c}$-plane sheets. This leads therefore to the quite clear scenario: with the increase of $V$ most of the unnested carriers are first allocated in the band of $\mathbf{c}$-axis quasi-1D filaments, and further they merge in 2D sheets in the $\mathbf{b-c}$-plane.
The major question following from the above discussion is why the soliton walls initially emerge along the $\mathbf{c}$- rather than the $\mathbf{b}$-axis? To answer this question, an interaction of solitons inside the molecular chains should be compared for these directions. The energy losses increase strongly (at first, exponentially) as the inverse distance between the solitons within the same chain increases [@BrazovskiiGorkovLebed82], i.e. the number of walls grows. In contrast, the energy gain increases linearly with a number of SWs. Therefore, creating longer walls is energetically favorable, though their length is limited by the maximum bandwidth, the SDW gap. At this point the difference between $\mathbf{b}$ and $\mathbf{c}$-axis walls comes in play. For a 1D filament the density of states increases with the distance between solitons within the wall, and thus is larger for $\mathbf{c}$-axis band due to the larger lattice spacing. Therefore, for the same width of soliton band the number of solitons within the $\mathbf{c}$-axis wall will be larger compared to that for $\mathbf{b}$-axis. This implies the larger number of walls in the latter case and hence the stronger interaction, which makes the alignment in $\mathbf{b}$ direction less favorable. We note here, that SWs are not necessary, in theory, to be strictly 1D, but their length along $\mathbf{c}$-axis should be still much larger than that along $\mathbf{b}$-axis.
The described situation is somewhat different from that considered in Ref. , where the SP energy gain was calculated for the limit of small soliton bandwidth and, as a result, small number of walls. In experiment we observe the spatial anisotropy of metallic(SC) phase when SDW is nearly suppressed and $\epsilon$ is large[@Qualls00]. Therefore, both the soliton bandwidth and the number of SWs are implied to be large, and interaction between walls should be taken into account.
We would like to emphasize, that the above conclusion that the $\mathbf{c}$-axis is a preferable direction, is inherent to the strongly-anisotropic systems with $t_b\gg t_c$, and should also apply to (TMTSF)$_2$PF$_6$ and other Bechgaard salts.
Phase segregation scenario
--------------------------
One of the ways to allocate the unnested carriers is to create nonuniform deformation of the lattice, which (in the case of constant volume) will result in spatially separated metallic (compressed) and SDW (expanded) regions. The spatial dimensions of the regions in this case are defined by the interplay between the elastic and SDW energies. The direct way to explain the spatial texture is then to take into account the anisotropy of the lattice.
The major reason for this segregation to be favorable was pointed out by Lee et al.[@Lee05]: if the SDW energy gain depends on deformation $\delta x$, then the emergence of tricritical point and the first order transition follow from large $\partial T_{SDW}/\partial x$ value in the vicinity of the SDW endpoint. In case of ClO4 we can track the evolution of spatial texture for fixed $b,c$ and $t_b,t_c$, as we get closer to the SDW endpoint and increase $\partial T_{SDW}/\partial x$ by changing $V$.
The uniaxial lattice deformation along the $\mathbf{c}$-axis has the least energy cost, however the effect of hopping in this direction on nesting was shown to be small[@Danner96; @Gerasimenko09]. Therefore, the spatial anisotropy within this scenario is not straightforward and more detailed analysis is required. Below we utilize the model for phase segregation developed by Vuletic et al.[@Vuletic02] to estimate the fraction of metallic and SDW regions. To simplify the equations we treat the deformations along different axes independently, keeping in mind that due to Poisson ratio deforming e.g. $\mathbf{c}$-axis will result in much smaller effect on $\mathbf{a}$ and $\mathbf{b}$ axes. In this case the fraction of metallic phase along $i$-th axis is defined by[@Vuletic02]: $$R_i=\frac{1}{2}-K_i|F_{SDW}(t_i)|\left(\frac{\partial F_{SDW}}{\partial t_i}\frac{\partial t_i}{\partial \delta_i}\right)^{-2},
\label{eq_psb}$$ where $K_i$ and $\delta_i$ are elastic constant and the deformation along $i$-th axis and $F_{SDW}$ is the SDW energy gain. As $V$ increases, the only quantity affected in this relation is $|F_{SDW}(t_i,V)|(\partial F_{SDW}/\partial t_i)^{-2}$, whereas $K_i (\partial t_i/\partial \delta_i)^{-2}$ depends only on crystal structure and is therefore fixed. The former is smaller for $i=b$ as was shown above, whereas the latter is likely to be smaller for $i=c$. Indeed, $K_c<K_b$ and quasi-uniaxial stress experiments performed on PF6[@Guo00] showed large effect of $\mathbf{c}$-axis stress on interlayer transport, which indicates large value of $\partial t_c/\partial \delta_c$.
To account for the observed spatial anisotropy of the metallic phase we have to assume $R_c>R_b$ for strong disorders and $R_c\approx R_b$ for weaker disorders, close to the SDW endpoint. Qualitatively, for almost perfect nesting both $\partial F_{SDW}/\partial t_b$, $\partial F_{SDW}/\partial t_c$ are small, whereas for almost suppressed SDW the former is expected to be larger, reproducing the desired behavior. However, for this scenario to work, one requires a certain relation between the elastic SDW properties entering the Eq. (\[eq\_psb\]), and is not a general property of an anisotropic system.
Hysteresis
----------
Hysteresis in ClO4 (and also in PF6[@Vuletic02]) is quite different from the one associated with the existence of metastable phases at the first order transition. The latter occurs upon crossing the phase boundary by varying e.g. temperature. However, in our measurements the hysteresis is observed not only when the phase boundary is crossed, but also when the temperature is cycled well below $T_{SDW}$ (see inset in Fig. \[fig\_xxyy\]b). These observations suggest that the hysteresis is linked rather with temperature dependent redistribution of phases.
Within the phase segregation scenario, such hysteresis can be ascribed to changes of the spatial distribution of strain. Non-uniform lattice deformations are favorable as long as their cost is smaller than the overall energy gain, $F_\mathrm{metal}-F_\mathrm{SDW}\gtrsim K\delta_x^2$. In other words, the cost of local strain variations should be less than that of metal-SDW transition. Thus, upon sweeping the temperature one can expect transitions between almost degenerate local free energy minima which correspond to different strain distributions. The hysteresis then is caused by an effective “dry friction” associated with small energy barriers between the above minima. Within the SW scenario, the hysteresis might, in principle, originate from pinning of the walls by defects, though no detailed treatment of the SP behavior in the vicinity of its boundary with metal was given up to date.
Superconducting state
---------------------
Finally, we would like to comment briefly on the nature of the SC state coexisting with the SDW phase in ClO4. In this paper we use the SC transition simply as a tool to determine the spatial texture of the two-phase state, whereas the mere fact of the SC survival inside the disordered regions is intriguing itself. Indeed, it was suggested[@Yonezawa12] that in the anion-ordered state ClO4 is a nodal d-wave superconductor. Furthermore, random anion potential was shown to decrease $T_c$ at low disorders and therefore was suggested to act as a pair-breaking mechanism [@Pesty88; @Joo05]. At higher disorders, where the two-phase state emerges, $T_c$ was noted to saturate [@Joo05]; it was suggested therefore that the SC phase is preserved in the anion-ordered inclusions[@Joo05] as SDW phase emerges inside disordered regions. The latter picture however contradicts our angular magnetoresistance measurements [@Gerasimenko13] and as was shown above cannot account for the anisotropy of the SC spatial texture.
It is also interesting to note, that somewhat similar saturation of $T_c$ with disorder was observed in quasi-2D superconductors, $\kappa$-(BEDT-TTF)$_2$X with X=Cu(NCS)$_2$[@Analytis06; @Sano10] or Cu\[N(CN)$_2$\]Br[@Sano10]. The disorder in that case was created by either X-ray or proton irradiation. It is even more interesting that $T_c$ decrease in these compounds was also observed with varying cooling rate[@Su98]. The latter was suggested to freeze intrinsic disorder in BEDT-TTF molecule’s methyl group orientations.
Despite these similarities, the underlying physics in these compounds is quite different from that in ClO4: the ground state competing with SC in them is Mott insulator. Therefore, the unifying property is likely the built-in molecular or anionic degrees of freedom. Indeed, these systems demonstrate explicitly the interaction between the conduction electrons and either molecular[@Kuwata11] or anion[@Zhang05] surrounding, as was deduced from NMR measurements. The peculiar nature of this interaction with additional degrees of freedom could be responsible then for the SC preservation even at high disorders. Obviously, more detailed studies are required to clarify the nature of superconductivity both with respect to disorder and inside the two-phase state.
Conclusions
===========
In summary, we have studied the evolution of the spatial texture of metallic(SC) phase emerging inside the SDW phase as the latter is being suppressed by anion ordering in (TMTSF)$_2$ClO$_4$. The degree of ordering affects, on average, the band splitting $V$ responsible for SDW suppression[@ZanchiBjelis01; @Gerasimenko13], i.e. acts similar to doping. This is in contrast to the case of other TMTSF-based superconductors, where SDW vanishes due to increasing FS warping under pressure. We used controlled variation of anion disorder to disentangle the effect of unnested carriers from that of the lattice deformation and to determine the intrinsic spatial texture of the two-phase SC/SDW state.
We established that (i) metallic(SC) regions initially arise inside the majority SDW phase in a form of quasi-1D filaments elongated in the interlayer, $\mathbf{c}$-axis, direction; (ii) these regions merge in the sheets in the $\mathbf{b-c}$-plane as SDW is further suppressed by weakening disorder; (iii) although temperature dependence of resistivity exhibits hysteresis in the coexistence region, the spatial texture is independent of temperature sweep direction, i.e. prehistory. Furthermore, (iv) small anion-ordered regions embedded into the homogeneously disordered background increase the relative fraction of metallic(SC) phase (as evidenced e.g. by the violation of activated behavior for strong disorders) but do not affect its spatial anisotropy. The similarity of anisotropy of the spatial texture in our measurements in (TMSTF)$_2$ClO$_4$ to that observed by Kang et al.[@Kang10] in (TMTSF)$_2$PF$_6$ suggests that the properties (i)-(iii) are intrinsic to metal(SC) and SDW coexistence in various organic quasi-1D systems.
We explained qualitatively the observed evolution of the spatial texture within the soliton phase theory[@BrazovskiiGorkovLebed82; @GorkovGrigoriev05] in a strongly anisotropic lattice by taking into account the interaction between soliton walls. Still, the origin of hysteresis within this approach remains under question. The alternative phase segregation scenario turned out less promising, because in order to account for the anisotropy it requires some special assumptions on elastic properties of the lattice.
Recent generalization of soliton phase to quasi two-dimensional systems[@GorkovTeitelbaum10] could make our observations applicable also to the SC/SDW phase separation in iron pnictide superconductors.
Acknowledgements
================
The authors would like to thank P.D. Grigoriev, A.G. Lebed and S.E. Brown for discussions. The work was supported by RFBR, Programs of the Russian Academy of Sciences, by Russian Ministry for Education and Science (grant No 8375), and using research equipment of the Shared Facilities Center at LPI.
[99]{} For review see: “The Physics of Organic Conductors and Superconductors”, ed. A. Lebed, Springer-Verlag (2008) G.R. Stewart, Rev. Mod. Phys., [**83**]{}, 1589 (2011). D.C. Johnston, Adv. Phys., [**59**]{}(6), 803–1061 (2010). A. Chubukov, Ann. Rev. Cond. Mat. Phys. [**3**]{}, 57–92 (2012) T. Vuletic, P. Auban-Senzier, C. Pasquier, S. Tomic, D. Jérome, M. Héritier, K. Bechgaard, Eur. Phys. J. B **25**, 319 (2002); A.V. Kornilov, V.M. Pudalov, Y. Kitaoka, K. Ishida, G.-q. Zheng, T. Mito, J. S. Qualls, Phys. Rev. B **69**, 224404 (2004); I.J. Lee, S.E. Brown, W. Yu, M.J. Naughton, P.M. Chaikin, Phys. Rev. Lett. **94**, 197001 (2005) W. Yu, S.E. Brown, F. Zamborszky, I.J. Lee, P.M. Chaikin, Int. J. Mod. Phys. B **16**, 3090 (2002) I.J. Lee, P.M. Chaikin, M.J. Naughton, Phys. Rev. Lett. **88**, 207002 (2002) D. Podolsky, E. Altman, T. Rostunov, E. Demler, Phys. Rev. Lett. **93**, 246402 (2004) P.D. Grigoriev, Phys. Rev. B **77**, 224508 (2008) W. Zhang and C.A.R. Sá de Melo, Phys. Rev. Lett. **97**, 047001 (2006) L.P. Gor’kov and P.D. Grigoriev, Europhys. Lett. **71**, 425 (2005) L.P. Gor’kov and P.D. Grigoriev, Phys. Rev. B **75**, 020507 (2007) P.D. Grigoriev, Physica B **404**, 513 (2009) N. Kang, B. Salameh, P. Auban-Senzier, D. Jérome, C.R. Pasquier, S. Brazovskii, Phys. Rev. B **81**, 100509 (2010) H. Yoshino, A. Oda, K. Murata, H. Nishikawa, K. Kikuchi, I. Ikemoto, Synth. Met. **120**, 885 (2001) K. Yamaji, J. Phys. Soc. Jap. **51**, 2787 (1982) H. Schwenk, K. Andres, F. Wudl, Phys. Rev. B, **27**, 5846 (1983); **29**, 500, (1984) Ya.A. Gerasimenko, V.A. Prudkoglyad, A.V. Kornilov, S.V. Sanduleanu, J.S. Qualls, V.M. Pudalov, JETP Lett. **97**, 419 (2013) J.-P. Pouget, S. Kagoshima, T. Tamegai, Y. Nogami, K. Kubo, T. Nakajima, K. Bechgaard, J. Phys. Soc. Japan **59**, 2036 (1990) D. Zanchi and A. Bjelis, Europhys. Lett. **56**, 596 (2001) K. Sengupta and N. Dupuis, Phys. Rev. B **65**, 035108 (2001) L.I. Buravov, N.D. Kushch, V.N. Laukhin, A.G. Khomenko, E.B. Yagubskii, M.V. Kartsovnik, A.E. Kovalev, L.P. Rozenberg, R.P. Shibaeva, M.A. Tanatar, V.S. Yefanov, V.V. Dyakin, V.A. Bondarenko, J. Phys. I France **4**, 441-451 (1994) P. Garoche, R. Brusetti, D. Jérome, J. Phys. Lett. **43**, L147 (1982) J. M. Ziman, Principles of the Theory of Solids, Cambridge University Press (1972) J.I. Oh and M.J. Naughton, Phys. Rev. Lett. **92**, 067001 (2004) F. Zhang, Y. Kurosaki, J. Shinagawa, B. Alavi, S.E. Brown, Phys. Rev. B **72**, 060501 (2005) F. Pesty, K. Wang, P. Garoche, Synth. Met. **27**, 137 (1988) G. Deutscher, O. Entin-Wohlman, S. Fishman, Y. Shapira, Phys. Rev. B **21**, 5041 (1980) The $a$-axis BCS superconducting coherence length is $\xi_x=\hbar v_F/1.76 T_c\approx400$nm with $v_F=10^7$cm/s and $T_c=0.96\,K$. The charachteristic size of SC regions, $d_x$ for strong disorders can be estimated either from 30-60nm sizes of AO inclusions or the width of the assumed soliton wall along $a$-axis. The latter is of the order of SDW coherence length $\xi_{SDW}=\hbar v_F/1.76 T_{SDW}\approx70$nm. Both these estimates for $d_x$ are smaller than $\xi_x$. S. A. Brazovskii, L. P. Gor’kov, J. S. Schrieffer, Phys. Scr. **25**, 423 (1982) S. A. Brazovskii, L. P. Gor’kov, A. G. Lebed, JETP **56**, 683 (1982) G. Montambaux, Phys. Rev. B **38**, 4788 (1988) G.M. Danner, P.M. Chaikin and S.T. Hannahs, Phys. Rev. B **53**, 2727 (1996) Ya.A. Gerasimenko, V.A. Prudkoglyad, A.V. Kornilov, V.M. Pudalov, V.N. Zverev, A.-K. Klehe, J.S. Qualls, Phys. Rev. B **80**, 184417 (2009) N. Biškup, S. Tomic, D. Jérome, Phys. Rev. B **51**, 17972 (1995) N. Matsunaga, K. Yamashita, H. Kotani, K. Nomura, T. Sasaki, T. Hanajiri, J. Yamada, S. Nakatsuji, H. Anzai, Phys. Rev. B **64**, 052405 (2001) D. Le Pévelen, J. Gaultier, Y. Barrans, D. Chasseau, F. Castet, L. Ducasse, Eur. Phys. J. B **19**, 363 (2001) Y. Koval, X. Jin, C. Bergmann, Y. Simsek, L. Özyüzer, P. Müller, H. Wang, G. Behr, B. Büchner, Appl. Phys. Lett. **96**, 082507 (2010) J.S. Qualls, C.H. Mielke, J.S. Brooks, L.K. Montgomery, D.G. Rickel, N. Harrison, S.Y. Han, Phys. Rev. B **62**, 12680 (2000) W. Kang, S.T. Hannahs and P.M. Chaikin, Phys. Rev. Lett **70**, 3091 (1993) F. Guo, K. Murata, A. Oda, Y. Mizuno, H. Yoshino, J. Phys. Soc. Jap. **69**, 2164 (2000) N. Joo, P. Auban-Senzier, C.R. Pasquier, D. Jérome, K. Bechgaard, Europhys. Lett., **72**, 645 (2005) N. Joo, P. Auban-Senzier, C. Pasquier, P. Monod, D. Jérome, K. Bechgaard, Eur. Phys. J. B **40**, 43 (2004) S. Yonezawa, Y. Maeno, K. Bechgaard and D. Jérome, Phys. Rev. B **85**, 140502(R) (2012) J.G. Analytis, A. Ardavan, S.J. Blundell, R.L. Owen, E.F. Garman, C. Jeynes, B.J. Powell, Phys. Rev. Lett. **96**, 177002 (2006) K. Sano, T. Sasaki, N. Yoneyama, N. Kobayashi, Phys. B **405**, S279 (2010) X. Su, F. Zuo, J.A. Schlueter, M.E. Kelly, J.M. Williams, Phys. Rev. B **57**, R14056 (1998) Y. Kuwata, M. Itaya, A. Kawamoto, Phys. Rev. B **83**, 144505 (2011) L.P. Gor’kov and G.B. Teitel’baum, Phys. Rev. B **82**, 020510 (2010)
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
Goodness-of-fit tests gauge whether a given set of observations is consistent (up to expected random fluctuations) with arising as independent and identically distributed (i.i.d.) draws from a user-specified probability distribution known as the “model.” The standard gauges involve the discrepancy between the model and the empirical distribution of the observed draws. Some measures of discrepancy are cumulative; others are not. The most popular cumulative measure is the Kolmogorov-Smirnov statistic; when all probability distributions under consideration are discrete, a natural noncumulative measure is the Euclidean distance between the model and the empirical distributions. In the present paper, both mathematical analysis and its illustration via various data sets indicate that the Kolmogorov-Smirnov statistic tends to be more powerful than the Euclidean distance when there is a natural ordering for the values that the draws can take — that is, when the data is ordinal — whereas the Euclidean distance is more reliable and more easily understood than the Kolmogorov-Smirnov statistic when there is no natural ordering (or partial order) — that is, when the data is nominal.
[*Keywords:*]{} significance, hypothesis, chi-square, root-mean-square, mean-square
author:
- 'Jacob Carruth, Mark Tygert, and Rachel Ward'
bibliography:
- 'stat.bib'
title: 'A comparison of the discrete Kolmogorov-Smirnov statistic and the Euclidean distance'
---
Introduction {#intro}
============
Testing goodness-of-fit is one of the foundations of modern statistics, as elucidated by [@rao], for example. The formulation in the discrete setting involves $n$ independent and identically distributed (i.i.d.) draws from a probability distribution over $m$ bins (“categories,” “cells,” and “classes” are common synonyms for “bins”). In accordance with the standard conventions, we will use $p$ to denote the actual (unknown) underlying distribution of the draws; $p = (p^{(1)}, p^{(2)}, \dots, p^{(m)})$, with $p^{(1)}$, $p^{(2)}$, …, $p^{(m)}$ being nonnegative and $$\label{prob}
\sum_{j=1}^m p^{(j)} = 1.$$ We will use $p_0$ to denote a user-specified distribution, usually called the “model”; again $p_0 = (p_0^{(1)}, p_0^{(2)}, \dots, p_0^{(m)})$, with $p_0^{(1)}$, $p_0^{(2)}$, …, $p_0^{(m)}$ being nonnegative and $$\label{prob0}
\sum_{j=1}^m p_0^{(j)} = 1.$$ A goodness-of-fit test produces a value — the “P-value” — that gauges the consistency of the observed data with the assumption that $p = p_0$. In many formulations, the user-specified model $p_0$ consists of a family of probability distributions parameterized by $\theta$, where $\theta$ can be integer-valued, real-valued, complex-valued, vector-valued, matrix-valued, or any combination of the many possibilities. In such cases, the P-value gauges the consistency of the observed data with the assumption that $p = p_0(\hat\theta)$, where $\hat\theta$ is an estimate (taken to be the maximum-likelihood estimate throughout the present paper). We now review the definition of P-values.
P-values are defined via the empirical distribution $\hat{p}$, where $\hat{p} = (\hat{p}^{(1)}, \hat{p}^{(2)}, \dots, \hat{p}^{(m)})$, with $\hat{p}^{(j)}$ being the proportion of the $n$ observed draws that fall in the $j$th bin, that is, $\hat{p}^{(j)}$ is the number of draws falling in the $j$th bin, divided by $n$. P-values involve a hypothetical experiment taking $n$ i.i.d. draws from the assumed actual underlying distribution $p = p_0(\hat\theta)$. We denote by $\hat{P}$ the empirical distribution of the draws from the hypothetical experiment; we denote by $\hat\Theta$ a maximum-likelihood estimate of $\theta$ obtained from the hypothetical experiment. The P-value is then the probability that the discrepancy between the random variables $\hat{P}$ and $p_0(\hat\Theta)$ is at least as large as the observed discrepancy between $\hat{p}$ and $p_0(\hat\theta)$, calculating the probability under the assumption that $p = p_0(\hat\theta)$.
To complete the definition of P-values, we must choose a measure of discrepancy. In the present paper, we consider the (discrete) Kolmogorov-Smirnov and Euclidean distances, $$\label{dKolmogorov-Smirnov}
d_1(a,b) = \max_{1 \le k \le m}
\left| \sum_{j=1}^k a^{(j)} - \sum_{j=1}^k b^{(j)} \right|$$ and $$\label{dEuclidean}
d_2(a,b) = \sqrt{\sum_{j=1}^m (a^{(j)} - b^{(j)})^2},$$ respectively. The P-value for the Kolmogorov-Smirnov statistic is the probability that $d_1(\hat{P},p_0(\hat\Theta)) \ge d_1(\hat{p},p_0(\hat\theta))$; the P-value for the Euclidean distance is the probability that $d_2(\hat{P},p_0(\hat\Theta)) \ge d_2(\hat{p},p_0(\hat\theta))$. When evaluating the probabilities, we view $\hat{P}$ and $\hat\Theta$ as random variables, constructed with i.i.d. draws from the assumed distribution $p = p_0(\hat\theta)$, while viewing the observed $\hat{p}$ and $\hat\theta$ as fixed, not random.
If a P-value is very small, then we can be confident that the given observed draws are inconsistent with the assumed model, are not i.i.d., or are both inconsistent and not i.i.d.
Needless to say, the Kolmogorov-Smirnov distance defined in (\[dKolmogorov-Smirnov\]) is the maximum absolute difference between cumulative distribution functions. The Kolmogorov-Smirnov statistic depends on the ordering of the bins, unlike the Euclidean distance.
As supported by the investigations below, we recommend using the Kolmogorov-Smirnov statistic when there is a natural ordering of the bins, while the Euclidean distance is more reliable and more easily understood than the Kolmogorov-Smirnov statistic when there is no natural ordering (or partial order). Unlike the Euclidean distance, the Kolmogorov-Smirnov statistic utilizes the information in a natural ordering of the bins, when the latter is available. [@horn] gave similar recommendations when comparing the $\chi^2$ and Kolmogorov-Smirnov statistics. Detailed comparisons between the Euclidean distance and $\chi^2$ statistics are available in [@perkins-tygert-ward3].
The Kolmogorov-Smirnov statistic is cumulative; it accentuates low-frequency differences between the model and the empirical distribution of the draws, but tends to average away and otherwise obscure high-frequency differences. Similar observations have been made by [@pettitt-stephens], [@dagostino-stephens], [@choulakian-lockhart-stephens], [@from], [@best-rayner], [@haschenburger-spinelli], [@steele-chaseling], [@lockhart-spinelli-stephens], [@ampadu], and [@ampadu-wang-steele], among others. Our suggestions appear to be closest to those of [@horn].
There are many cumulative approaches similar to the Kolmogorov-Smirnov statistic. These include the Cramér–von-Mises, Watson, Kuiper, and Rényi statistics, as well as their Anderson-Darling variants; Section 14.3.4 of [@press-teukolsky-vetterling-flannery], [@stephens2], and [@renyi] review these statistics. We ourselves are fond of the Kuiper approach. However, the present paper focuses on the popular Kolmogorov-Smirnov statistic; the Cramér–von-Mises, Watson, and Kuiper variants are very similar.
The remainder of the present paper has the following structure: Section \[nonatorder\] describes how the Euclidean distance is generally preferable to the Kolmogorov-Smirnov statistic when there is no natural ordering (or partial order) of the bins. Section \[natorder\] describes how the Kolmogorov-Smirnov statistic is generally preferable to the Euclidean distance when there is a natural ordering of the bins. Section \[data\_analysis\] illustrates both cases with examples of data sets and the associated P-values, computing the P-values via Monte-Carlo simulations with guaranteed error bounds. The reader may wish to begin with Section \[data\_analysis\], referring back to earlier sections as needed.
The case when the bins do not have a natural order {#nonatorder}
==================================================
The Euclidean distance is generally preferable to the Kolmogorov-Smirnov statistic when there is no natural ordering (or partial order) of the bins. As discussed by [@perkins-tygert-ward2], the interaction of parameter estimation and the Euclidean distance is easy to understand and quantify, at least asymptotically, in the limit of large numbers of draws. In contrast, the interaction of parameter estimation and the Kolmogorov-Smirnov statistic can be very complicated, though [@choulakian-lockhart-stephens] and [@lockhart-spinelli-stephens] have pointed out that the interaction is somewhat simpler with Cramér’s and von Mises’, Watson’s, and some of Anderson’s and Darling’s very similar statistics. That said, the Euclidean distance can be more reliable even when there are no parameters in the model, that is, when the model $p_0$ is a single, fixed, fully specified probability distribution; the remainder of the present section describes why.
The basis of the analysis is the following lemma, a reformulation of the fact that the expected maximum absolute deviation from zero of the standard Brownian bridge is $\sqrt{\pi/2} \cdot \ln(2) \approx .8687$ [see, for example, Section 3 of @marsaglia-tsang-wang].
\[bridge\] Suppose that $m$ is even and that $D^{(1)}$, $D^{(2)}$, …, $D^{(m)}$ form a randomly ordered list of $m/2$ positive ones and $m/2$ negative ones (with the ordering drawn uniformly at random). Then, $$\E \max_{1 \le k \le m} \left| \sum_{j=1}^k D^{(j)} \right| \Bigg/ \sqrt{m}
\quad \longrightarrow \quad \sqrt{\pi/2} \cdot \ln(2)$$ in the limit that $m \to \infty$, where (as usual) $\E$produces the expected value.
We denote by $p$ the actual underlying distribution of the $n$ observed i.i.d. draws. We denote by $p_0$ the model distribution. We denote by $\hat{P}$ the empirical distribution of the $n$ draws. These are all probability distributions, that is, $p^{(j)} \ge 0$, $p_0^{(j)} \ge 0$, and $\hat{P}^{(j)} \ge 0$ for $j = 1$, $2$, …, $m$, and (\[prob\]) and (\[prob0\]) hold.
Suppose that the actual underlying distribution $p^{(1)}$, $p^{(2)}$, …, $p^{(m)}$ of the draws is the same as the model distribution $p_0^{(1)}$, $p_0^{(2)}$, …, $p_0^{(m)}$; the random variables $\hat{P}^{(1)}$, $\hat{P}^{(2)}$, …, $\hat{P}^{(m)}$ are then the proportions of $n$ i.i.d. draws from $p_0$ that fall in the respective $m$ bins. The Euclidean distance is $$U = \sqrt{\sum_{j=1}^m (\hat{P}^{(j)} - p_0^{(j)})^2}.$$ The Kolmogorov-Smirnov statistic is $$V = \max_{1 \le k \le m} \left|\sum_{j=1}^k (\hat{P}^{(j)} - p_0^{(j)})\right|.$$ The expected value of the square of the Euclidean distance is $$\label{expectedx}
\E U^2 = \sum_{j=1}^m \E(\hat{P}^{(j)} - p_0^{(j)})^2
= \sum_{j=1}^m \frac{p_0^{(j)}}{n}
= \frac{1}{n}.$$ As shown, for example, by [@durbin] using Lemma \[bridge\] above, the expected value of $\sqrt{n}$ times the Kolmogorov-Smirnov statistic is $$\label{expecteds}
\E V\sqrt{n} \to \sqrt{\pi/2} \cdot \ln(2) \approx .8687$$ in the limit that $n \to \infty$ and $\max_{1 \le j \le m} p_0^{(j)} \to 0$. Comparing (\[expectedx\]) and (\[expecteds\]), we see that $U$ and $V$ are roughly the same size (inversely proportional to $\sqrt{n}$) when the actual underlying distribution of the draws is the same as the model distribution.
However, when the actual underlying distribution of the draws differs from the model distribution, the Euclidean distance and the Kolmogorov-Smirnov statistic can be very different. If the number $n$ of draws is large, then the empirical distribution $\hat{P}$ will be very close to the actual distribution $p$. Therefore, to study the performance of the goodness-of-fit statistics as $n \to \infty$ when the actual distribution $p$ differs from the model distribution $p_0$ (and both are independent of $n$), we can focus on the difference between $p$ and $p_0$ (rather than the difference between $\hat{P}$ and $p_0$). We now define and study the difference $$\label{diffs}
d^{(j)} = p^{(j)} - p_0^{(j)}$$ for $j = 1$, $2$, …, $m$. The Euclidean distance between $p$ and $p_0$ (the root-sum-square difference) is $$\label{Euclidean}
u = \sqrt{\sum_{j=1}^m (d^{(j)})^2}.$$ The Kolmogorov-Smirnov statistic (the maximum absolute cumulative difference) is $$\label{KS}
v = \max_{1 \le k \le m} \left| \sum_{j=1}^k d^{(j)} \right|.$$
For simplicity (and because the following analysis generalizes straightforwardly), let us consider the illustrative case in which $|d^{(1)}| = |d^{(2)}| = \dots = |d^{(m)}|$, that is, $$\label{equal}
|d^{(j)}| = c_m$$ for all $j = 1$, $2$, …, $m$, where $c_m$ is a positive real number ($c_m$ must always satisfy $m \cdot c_m \le 2$, since $m \cdot c_m = \sum_{j=1}^m c_m = \sum_{j=1}^m |d^{(j)}|
\le \sum_{j=1}^m [p^{(j)} + p_0^{(j)}] = 2$). Combining (\[diffs\]), (\[prob\]), and (\[prob0\]) yields that $$\label{zero}
\sum_{j=1}^m d^{(j)} = 0.$$ Together, (\[zero\]) and (\[equal\]) imply that $m$ is even and that half of $d^{(1)}$, $d^{(2)}$, …, $d^{(m)}$ are equal to $+c_m$, and the other half are equal to $-c_m$.
Combining (\[equal\]) and (\[Euclidean\]) yields that the Euclidean distance is $$u = \sqrt{m} \cdot c_m.$$ The fact that half of $d^{(1)}$, $d^{(2)}$, …, $d^{(m)}$ are equal to $+c_m$, and the other half are equal to $-c_m$, yields that the Kolmogorov-Smirnov statistic $v$ defined in (\[KS\]) could be as small as $c_m$ or as large as $m \cdot c_m/2$, depending on the ordering of the signs in $d^{(1)}$, $d^{(2)}$, …, $d^{(m)}$. If all orderings are equally likely (which is equivalent to ordering the bins uniformly at random), then by Lemma \[bridge\] the mean value for $v$ is $\sqrt{m\pi/2} \cdot \ln(2) \cdot c_m \approx \sqrt{m} \cdot .8687 \cdot c_m$ in the limit that $m$ is large (this is the expected maximum absolute deviation from zero of a tied-down random walk with $m$ steps, each of length $c_m$, that starts and ends at zero; the random walk ends at zero due to (\[zero\])).
Thus, in the limit that the number $n$ of draws is large (and $\max_{1 \le j \le m} p_0^{(j)} \to 0$, while both the model $p_0$ and the alternative distribution $p$ are independent of $n$), the Euclidean distance and the Kolmogorov-Smirnov statistic have similar statistical power on average, if all orderings of the bins are equally likely. However, the Euclidean distance is the same for any ordering of the bins, whereas the power of the Kolmogorov-Smirnov statistic depends strongly on the ordering. We see, then, that the Euclidean distance is more reliable than the Kolmogorov-Smirnov statistic when there is no especially natural ordering for the bins.
\[l1\] It is possible to use an ordering for which the Kolmogorov-Smirnov statistic attains its greatest value (this corresponds to renumbering the bins such that the differences $D^{(j)} = \hat{P}^{(j)}-p_0^{(j)}$ satisfy $D^{(1)} \ge D^{(2)} \ge \dots \ge D^{(m)}$ or $D^{(1)} \le D^{(2)} \le \dots \le D^{(m)}$). However, this data-dependent ordering produces a statistic which is proportional to the $l^1$ distance $\sum_{j=1}^m |D^{(j)}|$ (whereas the Euclidean distance is the $l^2$ distance), as remarked at the top of page 396 of [@hoeffding]. The resulting statistic is no longer cumulative.
The case when the bins have a natural order {#natorder}
===========================================
The Kolmogorov-Smirnov statistic is often preferable to the Euclidean distance when there is a natural ordering of the bins. In fact, the Kolmogorov-Smirnov statistic is always preferable when the data is very sparse and there is a natural ordering of the bins. In the limit that the maximum expected number of draws per bin tends to zero, the Euclidean distance always takes the same value under the null hypothesis, providing no discriminative power: indeed, when the draws producing the empirical distribution $\hat{P}$ are taken from the model distribution $p_0$, the Euclidean distance is almost surely $1/\sqrt{n}$, $$\sqrt{\sum_{j=1}^m (\hat{P}^{(j)}-p_0^{(j)})^2} = \frac{1}{\sqrt{n}},$$ in the limit that $n \cdot \max_{1 \le j \le m} p_0^{(j)} \to 0$ (the reason is that, in this limit, $\max_{1 \le j \le m} p_0^{(j)} \to 0$ and moreover almost every realization of the experiment satisfies that, for all $j = 1$, $2$, …, $m$, $\hat{P}^{(j)} = 0$ or $\hat{P}^{(j)} = 1/n$, that is, there is at most one observed draw per bin). In contrast, the Kolmogorov-Smirnov statistic is nontrivial even in the limit that the maximum expected number of draws per bin tends to zero — in fact, this is exactly the continuum limit for the original Kolmogorov-Smirnov statistic involving continuous cumulative distribution functions (as opposed to the discontinuous cumulative distribution functions arising from the discrete distributions considered in the present paper). Furthermore, the Kolmogorov-Smirnov statistic is sensitive to symmetry (or asymmetry) in a distribution, and can detect other interesting properties of distributions that depend on the ordering of the bins.
Data analysis {#data_analysis}
=============
This section gives four examples illustrating the performance of the Kolmogorov-Smirnov statistic and the Euclidean distance in various circumstances. The Kolmogorov-Smirnov statistic is more powerful than the Euclidean distance in the first two examples, for which there are natural orderings of the bins. The Euclidean distance is more reliable than the Kolmogorov-Smirnov statistic in the last two examples, for which any ordering of the bins is necessarily rather arbitrary. We computed all P-values via Monte-Carlo simulations with guaranteed error bounds, as in Remark 3.3 of [@perkins-tygert-ward3]. Remark 3.4 of [@perkins-tygert-ward3] proves that the standard error of the estimate for a P-value $P$ is $\sqrt{P(1-P)/\ell}$, where $\ell$ is the number of simulations conducted to calculate the P-value.
A test of randomness
--------------------
A particular random number generator is supposed to produce an integer from 1 to $2^{32}$ uniformly at random. The model distribution for such a generator is $$\label{simplemod}
p_0^{(j)} = 2^{-32}$$ for $j = 1$, $2$, …, $2^{32}$. We test the (obviously poor) generator which produces the numbers 1, 2, 3, …, $n$, in that order, so that the observed distribution of the generated numbers is $$\label{baddata}
\hat{p}^{(j)} = \left\{ \begin{array}{rl} 1/n, & j = 1,\ 2,\ \dots,\ n \\
0, & j = n+1,\ n+2,\ \dots,\ 2^{32}
\end{array} \right.$$ for $j = 1$, $2$, …, $2^{32}$. For these observations, the P-value for the Euclidean distance is 1 to several digits of precision, while the P-value for the Kolmogorov-Smirnov statistic is 0 to several digits, at least for $n$ between a hundred and a million. So, as expected, the Euclidean distance has almost no discriminative power for such sparse data, whereas the Kolmogorov-Smirnov statistic easily discerns that the data (\[baddata\]) is inconsistent with the model (\[simplemod\]).
Like the Euclidean distance, classical goodness-of-fit statistics such as $\chi^2$, $G^2$ (the log–likelihood-ratio), and the Freeman-Tukey/Hellinger distance are invariant to the ordering of the bins, and also produce P-values that are equal to 1 to several digits of precision, at least for $n$ between a hundred and a million. For definitions and further discussion of the $\chi^2$, $G^2$, and Freeman-Tukey statistics, see Section 2 of [@perkins-tygert-ward3].
A test of Poissonity
--------------------
A Poisson-distributed random number generator with mean $100$ is supposed to produce a nonnegative integer according to the model $$\label{models}
p_0^{(j)} = \frac{100^j}{j! \cdot \exp(100)}$$ for $j = 0$, $1$, $2$, $3$, …. We test the (obviously poor) generator which produces the numbers 100, 101, 102, …, 109, so that the observed distribution of the numbers is $$\label{observations}
\hat{p}^{(j)}
= \left\{ \begin{array}{rl} 1/10, & j = 100, 101, 102, \dots, 109 \\
0, & \hbox{otherwise}
\end{array} \right.$$ for $j = 0$, $1$, $2$, $3$, …. The P-values, each computed via 4,000,000 simulations, are
- Kolmogorov-Smirnov: .0075
- Euclidean distance: .998
- $\chi^2$: .999
- $G^2$ (the log–likelihood-ratio): .999
- Freeman-Tukey (the Hellinger distance): .998
For definitions and further discussion of the $\chi^2$, $G^2$, and Freeman-Tukey statistics, see Section 2 of [@perkins-tygert-ward3]. The Kolmogorov-Smirnov statistic is far more powerful for this example, in which the bins have a natural ordering (in this example the bins are the nonnegative integers).
Figure \[observedpmf\] plots the model probabilities $p_0^{(0)}$, $p_0^{(1)}$, $p_0^{(2)}$, … defined in (\[models\]) along with the observed proportions $\hat{p}^{(0)}$, $\hat{p}^{(1)}$, $\hat{p}^{(2)}$, … defined in (\[observations\]). Figure \[simulatedpmf\] plots the model probabilities $p_0^{(0)}$, $p_0^{(1)}$, $p_0^{(2)}$, …along with analogues of the proportions $\hat{p}^{(0)}$, $\hat{p}^{(1)}$, $\hat{p}^{(2)}$, …for a simulation generating 10 i.i.d. draws according to the model.
Figure \[observedcmf\] plots the cumulative model probabilities $p_0^{(0)}$, $p_0^{(0)}+p_0^{(1)}$, $p_0^{(0)}+p_0^{(1)}+p_0^{(2)}$, … along with the cumulative observed proportions $\hat{p}^{(0)}$, $\hat{p}^{(0)}+\hat{p}^{(1)}$, $\hat{p}^{(0)}+\hat{p}^{(1)}+\hat{p}^{(2)}$, …. Figure \[simulatedcmf\] plots the cumulative model probabilities $p_0^{(0)}$, $p_0^{(0)}+p_0^{(1)}$, $p_0^{(0)}+p_0^{(1)}+p_0^{(2)}$, … along with analogues of the cumulative proportions $\hat{p}^{(0)}$, $\hat{p}^{(0)}+\hat{p}^{(1)}$, $\hat{p}^{(0)}+\hat{p}^{(1)}+\hat{p}^{(2)}$, …for the simulation generating 10 i.i.d. draws according to the model.
\
\
\
\
A test of Hardy-Weinberg equilibrium
------------------------------------
In a population with suitably random mating, the proportions of pairs of Rhesus haplotypes in members of the population (each member has one pair) can be expected to follow the Hardy-Weinberg law discussed by [@guo-thompson], namely to arise via random sampling from the model $$\label{hw}
p_0^{(j,k)}(\theta_1, \theta_2, \dots, \theta_9)
= \left\{ \begin{array}{cl}
2 \cdot \theta_j \cdot \theta_k, & j > k \\
(\theta_k)^2, & j = k
\end{array} \right.$$ for $j,k = 1$, $2$, …, $9$ with $j \ge k$, under the constraint that $$\sum_{j=1}^9 \theta_j = 1,$$ where the parameters $\theta_1$, $\theta_2$, …, $\theta_9$ are the proportions of the nine Rhesus haplotypes in the population (naturally, their maximum-likelihood estimates are the proportions of the haplotypes in the given data). For $j,k = 1$, $2$, …, $9$ with $j \ge k$, therefore, $p_0^{(j,k)}$ is the expected probability that the pair of haplotypes in the genome of an individual is the pair $j$ and $k$, given the parameters $\theta_1$, $\theta_2$, …, $\theta_9$.
In this formulation, the hypothesis of suitably random mating entails that the members of the sample population are i.i.d. draws from the model specified in (\[hw\]); if a goodness-of-fit statistic rejects the model with high confidence, then we can be confident that mating has not been suitably random.
Table \[hwt\] provides data on $n = 8297$ individuals; we duplicated Figure 3 of [@guo-thompson] to obtain Table \[hwt\]. Figure \[phwt\] plots the associated P-values, each computed via 90,000 Monte-Carlo simulations. The Kolmogorov-Smirnov statistic depends on the ordering of the bins; for the first trial $t=1$ in Figure \[phwt\], the order of the bins is the lexicographical ordering, namely $(1,1)$, $(2,1)$, $(2,2)$, $(3,1)$, $(3,2)$, $(3,3)$, …, $(9,9)$. The nine trials $t = 2$, $3$, …, $10$ displayed in Figure \[phwt\] use pseudorandom orderings of the bins. Please note that the Euclidean distance does not depend on the ordering.
Generally, a more powerful statistic produces lower P-values. In Figure \[phwt\], the P-values for the Kolmogorov-Smirnov statistic are sometimes lower, sometimes higher than the P-values for the Euclidean distance. There is no particularly natural ordering of the bins for Figure \[phwt\]; Figure \[phwt\] displays 10 different orderings corresponding to 10 different trials. Figure \[phwt\] demonstrates that the Euclidean distance is more reliable than the Kolmogorov-Smirnov statistic when there is no natural ordering (or partial order) for the bins.
The P-values for classical goodness-of-fit statistics are substantially higher; the classical statistics are less powerful for this example. The P-values, each computed via 4,000,000 Monte-Carlo simulations, are
- Euclidean distance: .039
- $\chi^2$: .693
- $G^2$ (the log–likelihood-ratio): .600
- Freeman-Tukey (the Hellinger distance): .562
For definitions and further discussion of the $\chi^2$, $G^2$, and Freeman-Tukey statistics, see Section 4.5 of [@perkins-tygert-ward3]. Like the Euclidean distance, the $\chi^2$, $G^2$, and Freeman-Tukey statistics are all invariant to the ordering of the bins.
$k$\
$j$
$_{j\hspace{-.3pc}}\diagdown{}^{\hspace{-.35pc}k}$ 1 2 3 4 5 6 7 8 9
---------------------------------------------------- ------ ----- ---- ------ ---- ------ --- --- ---
1 1236
2 120 3
3 18 0 0
4 982 55 7 249
5 32 1 0 12 0
6 2582 132 20 1162 29 1312
7 6 0 0 4 0 4 0
8 2 0 0 0 0 0 0 0
9 115 5 2 53 1 149 0 0 4
: Frequencies of pairs of Rhesus haplotypes[]{data-label="hwt"}
\
A test of uniformity
--------------------
Table \[skittlest\] duplicates Table 1 of [@gilchrist], giving the colors of the $n = 62$ pieces of candy in a 2.17 ounce bag. Figure \[pskittlest\] plots the P-values for Table \[skittlest\] to be consistent up to expected random fluctuations with Table \[skittlestu\], the model of uniform proportions. We computed each P-value via 4,000,000 Monte-Carlo simulations. The Kolmogorov-Smirnov statistic depends on the ordering of the bins; the ten trials $t = 1$, $2$, …, $10$ displayed in Figure \[pskittlest\] use pseudorandom orderings of the bins. The Euclidean distance does not depend on the ordering.
Generally, a more powerful statistic produces lower P-values. In Figure \[pskittlest\], the P-values for the Kolmogorov-Smirnov statistic are sometimes lower, sometimes higher than the P-values for the Euclidean distance. There is no particularly natural ordering of the bins for Table \[skittlestu\]; Figure \[pskittlest\] displays 10 different pseudorandom orderings corresponding to 10 different trials. Figure \[pskittlest\] illustrates that the Euclidean distance is more reliable than the Kolmogorov-Smirnov statistic when there is no natural ordering (or partial order) for the bins.
Table \[skittlest\] provides a possible means for ordering the bins. However, such an ordering will depend on the observed data. Using a data-dependent ordering can profoundly alter the nature of the goodness-of-fit statistic; see Remark \[l1\].
Like the Euclidean distance, many classical goodness-of-fit statistics are invariant to the ordering of the bins. The following are P-values, each computed via 4,000,000 Monte-Carlo simulations:
- Euclidean distance: .770
- $\chi^2$: .770
- $G^2$ (the log–likelihood-ratio): .766
- Freeman-Tukey (the Hellinger distance): .755
For definitions and further discussion of the $\chi^2$, $G^2$, and Freeman-Tukey statistics, see Section 2 of [@perkins-tygert-ward3]. For this example, the Euclidean distance and the $\chi^2$ statistic produce exactly the same P-values: for the model of homogeneous proportions, displayed in Table \[skittlestu\], the Euclidean distance is directly proportional to the square root of the $\chi^2$ statistic, and hence the Euclidean distance is a strictly increasing function of $\chi^2$.
-------------- -- ----- -------- -------- ------- --------
[*color*]{} red orange yellow green violet
[*number*]{} 15 9 14 11 13
-------------- -- ----- -------- -------- ------- --------
: Observed frequencies of colors of candies in a 2.17 ounce bag[]{data-label="skittlest"}
-------------- -- ------ -------- -------- ------- --------
[*color*]{} red orange yellow green violet
[*number*]{} 12.4 12.4 12.4 12.4 12.4
-------------- -- ------ -------- -------- ------- --------
: Expected frequencies of colors of candies in a 2.17 ounce bag[]{data-label="skittlestu"}
\
Acknowledgements {#acknowledgements .unnumbered}
================
We would like to thank Alex Barnett, Gérard Ben Arous, James Berger, Tony Cai, Sourav Chatterjee, Ronald Raphael Coifman, Ingrid Daubechies, Jianqing Fan, Jiayang Gao, Andrew Gelman, Leslie Greengard, Peter W. Jones, Deborah Mayo, Peter McCullagh, Michael O’Neil, Ron Peled, William Perkins, William H. Press, Vladimir Rokhlin, Joseph Romano, Gary Simon, Amit Singer, Michael Stein, Stephen Stigler, Joel Tropp, Larry Wasserman, and Douglas A. Wolfe. This work was supported in part by Alfred P. Sloan Research Fellowships, a Donald D. Harrington Faculty Fellowship, and a DARPA Young Faculty Award.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The solution of eigenproblems is often a key computational bottleneck that limits the tractable system size of numerical algorithms, among them electronic structure theory in chemistry and in condensed matter physics. Large eigenproblems can easily exceed the capacity of a single compute node, thus must be solved on distributed-memory parallel computers. We here present GPU-oriented optimizations of the ELPA two-stage tridiagonalization eigensolver (ELPA2). On top of its existing cuBLAS-based GPU offloading, we add a CUDA kernel to speed up the back-transformation of eigenvectors, which can be the computationally most expensive part of the two-stage tridiagonalization algorithm. We demonstrate the performance of this GPU-accelerated eigensolver by benchmark calculations on two hybrid CPU-GPU architectures, namely a compute cluster based on Intel Xeon Gold CPUs and NVIDIA Volta GPUs, and the Summit supercomputer based on IBM POWER9 CPUs and NVIDIA Volta GPUs. Consistent with previous benchmarks on CPU-only architectures, the GPU-accelerated two-stage solver exhibits a parallel performance superior to the one-stage solver.'
address:
- 'Department of Mechanical Engineering and Materials Science, Duke University, Durham, NC 27708, USA'
- 'Molecular Sciences Software Institute, Blacksburg, VA 24060, USA'
- 'Max Planck Computing and Data Facility, Garching D-85747, Germany'
- 'Institute of Mathematics of the Czech Academy of Sciences, Žitná 25, 115 67 Prague, Czech Republic'
- NVIDIA Corporation
- 'Center for Nanophase Materials Sciences, Oak Ridge National Laboratory, Oak Ridge, TN 37830, USA'
author:
- 'Victor Wen-zhe Yu'
- Jonathan Moussa
- Pavel Ks
- Andreas Marek
- Peter Messmer
- Mina Yoon
- Hermann Lederer
- Volker Blum
bibliography:
- 'elpa2gpu.bib'
title: 'GPU-Acceleration of the ELPA2 Distributed Eigensolver for Dense Symmetric and Hermitian Eigenproblems'
---
Eigensolver ,dense linear algebra ,parallel computing ,high-performance computing ,GPU ,CUDA
Introduction {#sec:intro}
============
Finding the eigenvalues and eigenvectors of large dense matrices is a frequent problem in computational science and engineering. For example, in Kohn-Sham density-functional theory (KS-DFT) [@dft_hohenberg_1964; @dft_kohn_1965], the many-electron problem for the Born-Oppenheimer electronic ground state is reduced to a system of single particle equations that can be discretized into a generalized eigenproblem in the following matrix form $$\label{eq:gevp}
\boldsymbol{H} \boldsymbol{C} = \boldsymbol{S} \boldsymbol{C} \boldsymbol{\Sigma} .$$
Here the Hamiltonian matrix $\boldsymbol{H}$ and the overlap matrix $\boldsymbol{S}$ are real symmetric or complex Hermitian. The matrix $\boldsymbol{C}$ and the diagonal matrix $\boldsymbol{\Sigma}$ are the eigenvectors and eigenvalues, respectively, of this eigensystem. In the framework of KS-DFT, $\boldsymbol{C}$ and $\boldsymbol{\Sigma}$ (or the information they carry, at least) are needed for the construction of $\boldsymbol{H}$. Therefore, Eq. \[eq:gevp\] is a non-linear problem and must be solved self-consistently. It is possible, and has already been implemented in various codes [@conquest_gillan_2007; @fhiaims_blum_2009; @onetep_skylaris_2005; @siesta_soler_2002], to restrict the computational cost of the construction of $\boldsymbol{H}$ to scale linearly with respect to the system size $N$ for any semi-local and hybrid exchange-correlation functional. In contrast, the solution of a dense eigenproblem (“diagonalization”) scales as $\mathcal{O}(N^3)$, quickly growing to become prohibitive as $N$ increases to large values.
Various strategies exist to facilitate or circumvent the solution of Eq. \[eq:gevp\] [@elsi_yu_2018; @elsi_yu_2019]. (1) The key idea in a conventional dense eigensolver [@numerical_press_2007; @matrix_golub_2013] is tridiagonalization, which brings the original matrix to a tridiagonal form by a series of Householder transformations. This algorithm suffers from the inefficiency of BLAS level-2 matrix-vector operations. New algorithms such as pentadiagonalization [@eigenexa_imamura_2011] and two-stage tridiagonalization [@2stage_lang_1993; @2stage_bischof_1994; @elpa_auckenthaler_2011; @elpa_marek_2014] have been developed, leading to enhanced performance over the conventional one-stage tridiagonalization approach. (2) Iterative eigensolvers [@davidson_davidson_1975; @davidson_sleijpen_1996; @iterative_payne_1992; @iterative_kresse_1996] are commonly employed by DFT codes, particularly those based on plane-wave (PW) basis functions and pseudopotentials. In that case, because of the large number of basis functions, i.e. the large dimension of the matrices in Eq. \[eq:gevp\], needed in an accurate calculation, a direct solution of Eq. \[eq:gevp\] is rather infeasible. Iterative eigensolvers are well suited to find a small fraction of low-lying eigenstates, commensurate with the needs of a PW-based code. (3) When using spatially localized basis functions, such as linear combination of atomic orbitals (LCAO), the fraction of needed eigenpairs out of the full matrix dimension can be fairly large. In this scenario, iterative solvers no longer have an advantage over direct eigensolvers. With localized basis functions, locality in the physical system can be translated to sparsity in the $\boldsymbol{H}$ and $\boldsymbol{S}$ matrices. Methods exploiting this sparsity can be formulated as $\mathcal{O}(N) \sim \mathcal{O}(N^2)$ [@linear_goedecker_1999; @linear_bowler_2012; @linear_moussa_2019; @feast_polizzi_2009; @pexsi_lin_2013; @ntpoly_dawson_2018] by circumventing the explicit solution of Eq. \[eq:gevp\]. In particular, linear scaling algorithms in a density matrix formalism have been successfully applied to simulations of one million atoms [@million_bowler_2010; @million_vandevondele_2012]. Despite the success in extreme-scale simulations, reduced scaling methods come with a computational prefactor that is much larger than that of the $\mathcal{O}(N^3)$ diagonalization method. Moreover, the applicability and optimal performance of reduced scaling methods are often limited to some certain problem types. As of today, dense eigensolvers, with their small computational prefactor and general applicability, remain the default method in most LCAO codes. Even in PW codes, the performance of a dense eigensolver is still crucial, because at some stage of an iterative solver there will be a dense eigenproblem, the size of which scales with the number of valence electrons in the system being simulated. Therefore, any improvement made to a dense eigensolver would benefit the entire electronic structure community, and the broader field of computational science in general.
The ubiquitous adoption of graphics processing units (GPU) in high-performance computing opens up new opportunities to accelerate dense eigensolvers. A GPU device consists of hundreds to thousands of parallel cores operating at a relatively low frequency. These cores are naturally suited for parallel computational tasks, such as vector and matrix operations found in dense linear algebra. On top of that, GPUs typically have a power efficiency superior to traditional CPUs, and therefore play an important role in supercomputing towards the exascale. According to the latest release of the TOP500 list at the time of writing (November 2019) [@top500], five of the top ten machines are GPU-accelerated, including Summit, the world’s fastest computer based on IBM POWER9 CPUs and NVIDIA Tesla Volta V100 GPUs.
Eigensolvers operating on hybrid CPU-GPU machines have long been available in GPU-oriented linear algebra packages such as cuSOLVER [@cusolver] and MAGMA [@magma_tomov_2010; @magma_dongarra_2014]. These packages are designed and optimized for shared-memory host architectures. They can be very fast, but the problem size they can tackle is limited by the memory capacity of a single compute node. Fully exploiting the power of GPU-accelerated supercomputers would require a distributed-memory implementation. The MPI-parallel, distributed-memory ELPA library implements the conventional one-stage diagonalization method and the two-stage diagonalization proposed in Refs. [@2stage_lang_1993; @2stage_bischof_1994], known as the “ELPA1” and “ELPA2” solvers, respectively. Both of them have been ported to GPUs by substituting BLAS calls with the corresponding cuBLAS functions [@elpa_kus_2019a], making ELPA the only publicly available, distributed-memory, GPU-accelerated eigensolver to our knowledge. The CPU-only version of ELPA1 and ELPA2 and the GPU-accelerated version of ELPA1 and ELPA2 are hereafter referred to as CPU-ELPA1, CPU-ELPA2, GPU-ELPA1, and GPU-ELPA2, respectively.
In single-node tests published by Ks et al. [@elpa_kus_2019a; @elpa_kus_2019b], the performance of GPU-ELPA1 is better than that of GPU-ELPA2. When using 2 IBM POWER8 CPUs (24 cores in total) and 4 NVIDIA Pascal P100 GPUs, GPU-ELPA1 delivers up to 11.9x performance boost compared to CPU-ELPA1 using 24 CPU cores. The efficiency of CPU-ELPA2 on various distributed-memory CPU platforms [@elpa_marek_2014; @elpa_gutheil_2014; @elpa_cook_2018] indicates that GPU-ELPA2 may outperform GPU-ELPA1 on multiple nodes. Historically, the GPU porting of ELPA2 dates back to 2013, when Peter Messmer from NVIDIA programmed the first version of GPU-ELPA2. Then the code was refactored and merged into the mainline version of ELPA, and has been available in released versions of the ELPA eigensolver library since 2016. In this paper, we report our latest optimizations and developments that enable a performance improvement on distributed-memory, GPU-accelerated architectures. Specifically, several synchronizations and memory transfers between CPUs and GPUs have been optimized. Additionally, some kernels in one of the major computational steps, the tridiagonal-to-banded back-transformation of eigenvectors (Eq. \[eq:2stage\_bkwd1\]), have been rewritten. The rest of this paper is organized as follows. First, we briefly review the two-stage diagonalization algorithm, in particular the tridiagonal-to-banded back-transformation of eigenvectors and its CPU implementation in the ELPA library. Next, we outline the GPU acceleration strategies employed in GPU-ELPA2, and elaborate on our CUDA implementation of the tridiagonal-to-banded back-transformation of eigenvectors, which is essentially a GPU extension of the algorithm in Refs. [@elpa_auckenthaler_2011; @elpa_auckenthaler_2013]. Finally, we demonstrate the performance and scalability of the GPU-ELPA2 solver by a set of benchmarks on two GPU-accelerated computers, namely the Talos cluster at Max Planck Computing and Data Facility in Garching, Germany, based on Intel Xeon Gold CPUs and NVIDIA Volta GPUs, and the Summit supercomputer at Oak Ridge National Laboratory in Tennessee, USA, based on IBM POWER9 CPUs and NVIDIA Volta GPUs.
Two-Stage Tridiagonalization in ELPA2 {#sec:review_2stage}
=====================================
Overview of the Two-Stage Tridiagonalization {#subsec:cpu_2stage}
--------------------------------------------
The textbook procedure [@numerical_press_2007; @matrix_golub_2013] to solve a dense generalized eigenproblem, like the one in Eq. \[eq:gevp\], first computes the Cholesky factorization of $\boldsymbol{S}$ $$\label{eq:cholesky}
\boldsymbol{S} = \boldsymbol{L} \boldsymbol{L}^* ,$$
then uses $\boldsymbol{L}$ to transform Eq. \[eq:gevp\] to a standard eigenproblem $$\label{eq:evp}
\boldsymbol{\tilde{H}} \boldsymbol{\tilde{C}} = \boldsymbol{\tilde{C}} \boldsymbol{\Sigma} .$$
$\boldsymbol{\tilde{H}}$ is $$\label{eq:gevp2evp}
\boldsymbol{\tilde{H}} = \boldsymbol{L}^{-1} \boldsymbol{H} (\boldsymbol{L}^*)^{-1} ,$$
and the eigenvectors $\boldsymbol{\tilde{C}}$ must be back-transformed in order to retrieve the eigenvectors of Eq. \[eq:gevp\], i.e. $$\label{eq:evp2gevp}
\boldsymbol{C} = (\boldsymbol{L}^*)^{-1} \boldsymbol{\tilde{C}} .$$
The direct solution of Eq. \[eq:evp\] is based on tridiagonalization, that is, the full matrix $\boldsymbol{\tilde{H}}$ is transformed to a tridiagonal matrix $\boldsymbol{T}$. This is typically accomplished by individual Householder transformations, which take the shape of matrix-vector operations. Eigenvalues and eigenvectors of $\boldsymbol{T}$ can be easily (compared to the original problem) solved. Then, eigenvectors of $\boldsymbol{T}$ are back-transformed to obtain eigenvectors of $\boldsymbol{\tilde{H}}$. This algorithm is adopted by a variety of dense linear algebra packages, such as LAPACK [@lapack_anderson_1999] targeting sequential and shared-memory parallel architectures; cuSOLVER [@cusolver] and MAGMA [@magma_tomov_2010; @magma_dongarra_2014] targeting shared-memory architectures with GPU accelerators; ScaLAPACK [@scalapack_blackford_1997] and Elemental [@elemental_poulson_2013] targeting distributed-memory parallel architectures. However, as mentioned above, the tridiagonalization step makes extensive use of memory-bound, BLAS level-2 matrix-vector operations, whose performance is limited on modern computer architectures.
The two-stage tridiagonalization algorithm proposed by Bischof, Sun, and Lang [@2stage_lang_1993; @2stage_bischof_1994] is an established alternative to the conventional one-stage method. As shown in Eq. \[eq:2stage\] below and further illustrated in Fig. \[fig:2stage\], the tridiagonalization of the full matrix $\boldsymbol{\tilde{H}}$ is carried out in two transformations. The first transformation $\boldsymbol{P}$ reduces $\boldsymbol{\tilde{H}}$ to a banded matrix $\boldsymbol{B}$, and the second transformation $\boldsymbol{Q}$ reduces $\boldsymbol{B}$ to a tridiagonal matrix $\boldsymbol{T}$. The eigenvalues $\boldsymbol{\Sigma}$ and eigenvectors $\boldsymbol{X}$ of $\boldsymbol{T}$ are solved as done in the one-stage method. The back-transformation of eigenvectors is also carried out in two steps. $\boldsymbol{X}$ is first back-transformed to $\boldsymbol{Y}$, the eigenvectors of $\boldsymbol{B}$, then to $\boldsymbol{\tilde{C}}$, the eigenvectors of $\boldsymbol{\tilde{H}}$.
\[eq:2stage\] $$\begin{aligned}
\boldsymbol{B} & = \boldsymbol{P} \boldsymbol{\tilde{H}} \boldsymbol{P}^* , \label{eq:2stage_fwd1} \\
\boldsymbol{T} & = \boldsymbol{Q} \boldsymbol{B} \boldsymbol{Q}^* , \label{eq:2stage_fwd2} \\
\boldsymbol{T} \boldsymbol{X} & = \boldsymbol{X} \boldsymbol{\Sigma} , \label{eq:2stage_solve} \\
\boldsymbol{Y} & = \boldsymbol{Q}^* \boldsymbol{X} , \label{eq:2stage_bkwd1} \\
\boldsymbol{\tilde{C}} & = \boldsymbol{P}^* \boldsymbol{Y} . \label{eq:2stage_bkwd2}\end{aligned}$$
{width="99.00000%"}
This two-stage tridiagonalization approach is implemented in several linear algebra software packages [@elpa_auckenthaler_2011; @elpa_marek_2014; @magma_tomov_2010; @lapack_anderson_1999; @mkl_arturov_2018], including a high-performance distributed-memory implementation in the ELPA library [@elpa_auckenthaler_2011; @elpa_marek_2014]. The introduction of the banded matrix stage leads to faster computation compared to the one-stage method, for the transformation in Eq. \[eq:2stage\_fwd1\] mostly involves highly efficient BLAS level-3 matrix-matrix operations, and the transformation in Eq. \[eq:2stage\_fwd2\] only works on a sparse banded matrix $\boldsymbol{B}$ instead of a full matrix. The solution of Eq. \[eq:2stage\_solve\] is accelerated in ELPA by extending the divide-and-conquer symmetric tridiagonal eigensolver [@dc_cuppen_1980; @dc_gu_1995; @dc_tisseur_1999] such that unwanted eigenvectors are not computed [@elpa_auckenthaler_2011; @elpa_auckenthaler_2013]. Regarding the back-transformation of eigenvectors, it is rather difficult to directly use BLAS routines for Eq. \[eq:2stage\_bkwd1\]. Manually optimized “kernels” written in architecture-specific instruction sets are employed for this particular step [@elpa_auckenthaler_2011; @elpa_kus_2019b; @elpa_auckenthaler_2013]. The ELPA2 solver is highly scalable on massively parallel, distributed-memory architectures. It avoids global MPI communications by using a 2-dimensional (2D) process grid and restricting the communication to take place within either the row direction or the column direction. Depending on the size of the matrix, ELPA2 scales to at least tens of thousands of CPU cores [@elpa_marek_2014; @elpa_gutheil_2014; @elpa_cook_2018].
Tridiagonal-to-Banded Back-Transformation of Eigenvectors {#subsec:cpu_step4}
---------------------------------------------------------
We now summarize the tridiagonal-to-banded back-transformation algorithm and its CPU implementation in ELPA2. The reader is also referred to Refs. [@2stage_lang_1993; @2stage_bischof_1994; @elpa_auckenthaler_2011; @elpa_auckenthaler_2013]. ELPA2 relies on the “bulge chasing” algorithm [@2stage_lang_1993; @2stage_bischof_1994] to reduce a banded matrix $\boldsymbol{B}$ to a tridiagonal matrix $\boldsymbol{T}$. Let $N$ and $b$ denote the dimension and semi-bandwidth of $\boldsymbol{B}$, respectively. The banded-to-tridiagonal transformation is done in $N-2$ stages, with $(N-i)/b$ sweeps in the $i^\text{th}$ stage. The first sweep of the $i^\text{th}$ stage reduces the $i^\text{th}$ column of $\boldsymbol{B}$ to the target tridiagonal form, at the same time introducing fill-ins (“bulges”) to the remainder of $\boldsymbol{B}$. From the second sweep on, the first column of the fill-ins introduced in the previous sweep is eliminated, while introducing new fill-ins further down the matrix. The transformation applied in the $j^\text{th}$ sweep of the $i^\text{th}$ stage can be written as $$\label{eq:householder}
\boldsymbol{Q}_{(i,j)} = \boldsymbol{Q}^*_{(i,j)} = \boldsymbol{I} - \tau_{(i,j)} \boldsymbol{v}_{(i,j)} \boldsymbol{v}_{(i,j)}^* ,$$
where $\boldsymbol{Q}_{(i,j)}$ is a Householder transformation matrix; $\boldsymbol{I}$ is the identity matrix; the scalar $\tau_{(i,j)}$ and vector $\boldsymbol{v}_{(i,j)}$ are computed following the standard Householder method [@numerical_press_2007; @householder_householder_1958]. Most Householder vectors $\boldsymbol{v}_{(i,j)}$ have a length equal to $b$, except that $\boldsymbol{v}_{(i,j)}$ generated in the last sweep of each stage may be shorter. Matrix-vector operations are still needed in order to apply these Householder transformations, but the computational cost is much less than in the one-stage tridiagonalization algorithm, as the vectors are much shorter. The left panel of Fig. \[fig:step4\] shows where vectors $\boldsymbol{v}_{(i,j)}$ are generated with an example of $N=17$, $b=4$, and $N-2=15$ stages of bulge chasing. In the actual code, the Householder matrices $\boldsymbol{Q}_{(i,j)}$ are never explicitly constructed. Instead, $\tau_{(i,j)}$ and $\boldsymbol{v}_{(i,j)}$ are stored and used for the transformations in Eqs. \[eq:2stage\_fwd2\] and \[eq:2stage\_bkwd1\], where $\boldsymbol{Q}$ is the product of all $\boldsymbol{Q}_{(i,j)}$.
{width="99.00000%"}
After the tridiagonal eigenproblem (Eq. \[eq:2stage\_solve\]) is solved, we get the eigenvalues $\boldsymbol{\Sigma}$ and eigenvectors $\boldsymbol{X}$ of the tridiagonal matrix $\boldsymbol{T}$. The main task of the tridiagonal-to-banded back-transformation is to apply all $\boldsymbol{Q}^*_{(i,j)}$ to $\boldsymbol{X}$ (Eq. \[eq:2stage\_bkwd1\]). It is obvious that the eigenvectors can be back-transformed independently, leading to a trivial parallelism over eigenvectors. Moreover, as the Householder vectors $\boldsymbol{v}_{(i,j)}$ are shorter than the eigenvectors $\boldsymbol{X}$, a single transformation $\boldsymbol{Q}^*_{(i,j)}$ only alters a few rows of $\boldsymbol{X}$ where $\boldsymbol{v}_{(i,j)}$ has non-zero values. For instance, the Householder vector $\boldsymbol{v}_{(1,4)}$ at the left bottom corner of Fig. \[fig:step4\] only alters the last four rows of $\boldsymbol{X}$. This leads to another level of parallelism to be exploited within an individual eigenvector.
In ELPA2, the $N_p$ processes are organized in an $N_{pr}$ by $N_{pc}$ grid. The $N_{ev}$ eigenvectors are uniformly distributed across the $N_{pc}$ process columns. Within a process column, the eigenvectors are distributed across the $N_{pr}$ processes in a block manner. Each process applies a series of Householder transformations to its local part of the eigenvector matrix $\boldsymbol{X}$. The colored part in Fig. \[fig:step4\] shows an example of the local computation by a process. This process applies four Householder transformations, $\boldsymbol{Q}^*_{(8,1)}$, $\boldsymbol{Q}^*_{(7,1)}$, $\boldsymbol{Q}^*_{(6,1)}$, and $\boldsymbol{Q}^*_{(5,1)}$, to its local part of the eigenvector matrix, referred to as $\boldsymbol{X}_\text{local}$ hereafter. Apparently, $\boldsymbol{Q}^*_{(4,1)}$, $\boldsymbol{Q}^*_{(3,1)}$, and $\boldsymbol{Q}^*_{(2,1)}$ also modify the top part of $\boldsymbol{X}_\text{local}$. These three transformations are however applied by the upper adjacent process in the same process column. Therefore, the top three rows of $\boldsymbol{X}_\text{local}$ must be exchanged with the upper adjacent process. Likewise, the bottom three rows of $\boldsymbol{X}_\text{local}$ must be exchanged with the lower adjacent process in the same process column. The middle part of $\boldsymbol{X}_\text{local}$ is not involved in any data exchange. It appears in Fig. \[fig:step4\] that almost the entire $\boldsymbol{X}_\text{local}$ matrix needs to be exchanged between adjacent process rows, as there is only one row in the middle part. However, the height of the middle part actually increases with the matrix size $N$, whereas the height of the top and bottom parts can never exceed the semi-bandwidth $b$. Therefore, the amount of data that needs to be exchanged, i.e., data in the top and bottom parts, is limited by $b$, usually accounting for a small fraction of $\boldsymbol{X}_\text{local}$.
GPU Acceleration of ELPA2 {#sec:gpu_2stage}
=========================
In ELPA2, the two-stage tridiagonalization is implemented as five separate subroutines corresponding to the five steps in Eq. \[eq:2stage\]. The input, output, and internal working matrices of ELPA2 are distributed across CPU cores. ELPA2 relies on its own explicit MPI calls to handle distributed linear algebra operations. The efficient MPI communication pattern in the CPU version of ELPA2 is not altered in the GPU version, where GPU acceleration is mainly realized by substituting local BLAS calls with the corresponding cuBLAS functions, as is done for the full-to-banded transformation (Eq. \[eq:2stage\_fwd1\]), the solution of the tridiagonal system (Eq. \[eq:2stage\_solve\]), and the banded-to-full back-transformation (Eq. \[eq:2stage\_bkwd2\]). The tridiagonal-to-banded back-transformation (Eq. \[eq:2stage\_bkwd1\]) is GPU accelerated by a CUDA implementation of Eq. \[eq:householder\]. The banded-to-tridiagonal transformation (Eq. \[eq:2stage\_fwd2\]) has not been ported to GPUs, because of its low computational cost as shown in Fig. \[fig:decomp5\_summit\].
GPU Offloading via cuBLAS {#subsec:gpu_cublas}
-------------------------
The API of cuBLAS is designed to be almost identical to that of the standard CPU BLAS, making cuBLAS-based GPU offloading straightforward. Here we only comment on two technical aspects. First, the communication and synchronization between CPUs and GPUs should be avoided as much as possible. Before calling a cuBLAS function, the input arrays must reside on the GPU memory, which often requires a copy of the data from CPU to GPU. In order for the CPU to access the result of cuBLAS, another copy from GPU to CPU is needed. In our GPU porting of ELPA2, CPU-GPU memory transfers are reduced to minimum by leaving data on the GPU as long as possible. Most often, GPU data is copied to the CPU in order to participate in an MPI communication call. In the version discussed in this paper, we have not yet explored GPU-aware MPI to directly communicate data on the GPU. In each of the GPU-accelerated subroutines, the allocation of GPU memory, which implies a CPU-GPU synchronization, is performed before the main work begins by precomputing the size of the allocation. The allocated GPU memory is reused wherever possible, and is deallocated after the main work finishes. Avoiding frequent allocation and deallocation of GPU memory in a loop or in a recursive routine greatly reduces the amount of implicit CPU-GPU synchronization.
Second, the CPU code of the GPU-accelerated version of ELPA2 operates in a pure MPI mode without threading, i.e., one MPI task for each CPU core. As most (if not all) mainstream computers today have more CPU cores than GPU devices, several MPI tasks would have to share one GPU device. When the size of an eigenproblem is relatively small, the amount of work assigned to each individual MPI task may not be able to fully saturate the GPU. In such cases, the NVIDIA Multi-Process Service (MPS) transparently allows work from different MPI tasks to be executed concurrently on the GPU, increasing the overall GPU utilization. It is thus recommended to use ELPA with MPS switched on.
CUDA Implementation of Parallel Householder Transformations {#subsec:gpu_step4}
-----------------------------------------------------------
In this section, we present our CUDA implementation of the local computation in Fig. \[fig:step4\], which is the key step in the tridiagonal-to-banded back-transformation (Eq. \[eq:2stage\_bkwd1\]). In a CUDA program, the large number of GPU cores are arranged into a grid of blocks, each of which in turn comprises a grid of threads. All the GPU cores work in a single instruction, multiple thread (SIMT) fashion, i.e., a single instruction is simultaneously executed by multiple threads with different data [@cuda_nickolls_2008]. As explained in Sec. \[subsec:cpu\_step4\], in a given step of the tridiagonal-to-banded back-transformation, a process is responsible for applying $N_v$ Householder transformations to its local eigenvector matrix $\boldsymbol{X}_\text{local}$, which has $N_R$ rows and $N_C$ columns. A Householder transformation defined by $\tau$ and $\boldsymbol{v}$ is applied to an eigenvector $\boldsymbol{x}$ by $$\label{eq:kernel}
(\boldsymbol{I} - \tau \boldsymbol{v} \boldsymbol{v}^*) \boldsymbol{x} = \boldsymbol{x} - \tau \boldsymbol{v} (\boldsymbol{v}^* \boldsymbol{x}) .$$
The $N_v$ Householder transformations are applied sequentially. The order in which they are applied in the tridiagonal-to-banded back-transformation is the reverse of the order in which they are generated in the banded-to-tridiagonal transformation. Consider the example in Fig. \[fig:step4\] again, the first transformation would be $\boldsymbol{Q}^*_{(8,1)}$, then $\boldsymbol{Q}^*_{(7,1)}$, $\boldsymbol{Q}^*_{(6,1)}$, and finally $\boldsymbol{Q}^*_{(5,1)}$. It is obvious that from one transformation to the next, the rows of $\boldsymbol{X}_\text{local}$ modified by the transformation are shifted upward by one row. This is also seen in Fig. \[fig:step4\], where $\boldsymbol{Q}^*_{(8,1)}$ modifies the 4$^\text{th}$ to 7$^\text{th}$ rows of $\boldsymbol{X}_\text{local}$, $\boldsymbol{Q}^*_{(7,1)}$ modifies the 3$^\text{rd}$ to 6$^\text{th}$ rows of $\boldsymbol{X}_\text{local}$, and so forth. Each individual $\boldsymbol{Q}^*_{(i,j)}$ actually only modifies $b$ rows of $\boldsymbol{X}_\text{local}$.
In order to map this local computation to the GPU cores, we choose a 1D block grid with a 1D thread grid within each block. The CUDA kernel is launched with $N_C$ blocks and $b$ threads, so that each block works on an eigenvector, and each thread works on an element of this eigenvector. Specifically, the Householder transformations are applied to $\boldsymbol{X}_\text{local}$ as follows:
1. Copy $\boldsymbol{X}_\text{local}$, as well as all $\boldsymbol{v}$ and $\tau$ that ever update $\boldsymbol{X}_\text{local}$, from CPU to GPU. Each GPU block is responsible for one column of $\boldsymbol{X}_\text{local}$, denoted as $\boldsymbol{x}$.
2. Compute Eq. \[eq:kernel\] for the first Householder transformation.
1. Compute dot product $\boldsymbol{v}^* \boldsymbol{x}$. Only $b$ elements of $\boldsymbol{x}$ contribute to the dot product. Each of the $b$ GPU threads loads one element of $\boldsymbol{v}$ and one element of $\boldsymbol{x}$, multiplies them together, and stores the thread-local result in shared memory. The final dot product is obtained by a parallel reduction involving all the threads in a block.
2. Update $\boldsymbol{x}$ by $\boldsymbol{x} = \boldsymbol{x} - \tau (\boldsymbol{v}^* \boldsymbol{x}) \boldsymbol{v}$. Again, only $b$ elements of $\boldsymbol{x}$ are modified. Since the dot product $\boldsymbol{v}^* \boldsymbol{x}$ has already been computed and $\tau$ is a scalar, this update can be done in a straightforward element-wise fashion, i.e., each thread updates one element of $\boldsymbol{x}$.
3. Among the $b$ elements of $\boldsymbol{x}$ that are updated by the first Householder transformation, only the lowest element will not be affected by the next Householder transformation. Before applying the next transformation, the last thread in each block writes its element of $\boldsymbol{x}$ into $\boldsymbol{X}_\text{local}$. Then thread $i$ ($i>=1$) takes the element of $\boldsymbol{x}$ from thread ($i-1$), while thread 0 loads a new element from $\boldsymbol{X}_\text{local}$.
4. Now all threads are ready to repeat steps 1, 2, and 3 for the next Householder transformation.
5. The CUDA kernel finishes when all transformations are applied. The updated $\boldsymbol{X}_\text{local}$ is then copied back to CPU.
An example of this procedure is given in Fig. \[fig:kernel\], where four Householder transformations are applied to a local eigenvector matrix with $N_R=7$ rows and $N_C=6$ columns. The CUDA kernel is launched with six blocks and four threads, as indicated by the block id and thread id (both zero-based) in the figure. Throughout the execution of the kernel, block $i$ is responsible for the $(i+1)^\text{th}$ eigenvector. The four Householder transformations need to be applied in reverse order, i.e. from right to left. In the application of the rightmost transformation, threads 0, 1, 2, 3 in each block are responsible for the 4$^\text{th}$, 5$^\text{th}$, 6$^\text{th}$, and 7$^\text{th}$ elements, respectively, of the eigenvector this block is responsible for. After computing Eq. \[eq:kernel\], thread 3 immediately writes the 7$^\text{th}$ element back to the $\boldsymbol{X}_\text{local}$ array. Threads 1, 2, 3 take the updated 4$^\text{th}$, 5$^\text{th}$, and 6$^\text{th}$ elements from threads 0, 1, 2, respectively. Then thread 0 loads the 3$^\text{rd}$ element of the eigenvector from $\boldsymbol{X}_\text{local}$, and all the threads are ready for the next transformation. It is obvious that all the transformations can be applied in exactly the same way. After the kernel finishes, the final $\boldsymbol{X}_\text{local}$ is copied back from GPU to CPU.
{width="99.00000%"}
Performance and Scalability {#sec:results}
===========================
The performance of the GPU-accelerated ELPA2 solver is benchmarked on the Talos cluster at Max Planck Computing and Data Facility and the Summit supercomputer at Oak Ridge National Laboratory. Each node of Talos has 2 Intel Xeon Gold 6148 CPUs (40 cores in total) and 2 NVIDIA Tesla Volta V100 (each has 32 GB high-bandwidth memory, double precision peak 7.0 TFLOP/s, PCIe 32 GB/s interconnect) GPUs. Benchmarks presented in Fig. \[fig:all\_talos\] are performed on Talos. The ELPA code is compiled with the Intel 19.0.5 compiler suite, Intel MPI 2019.5, Intel MKL 2019.5, and CUDA 10.1. Each node of Summit has 2 IBM POWER9 CPUs (44 cores in total, of which 42 are for running applications) and 6 NVIDIA Tesla Volta V100 (each has 16 GB high-bandwidth memory, double precision peak 7.8 TFLOP/s, NVLink 300 GB/s interconnect) GPUs. Benchmarks presented in Figs. \[fig:all\_summit\], \[fig:decomp3\_summit\], \[fig:decomp5\_summit\] are performed on Summit. The ELPA code is compiled with the IBM XL 16.1.1 compiler suite, IBM Spectrum MPI 10.3, IBM ESSL 6.1, and CUDA 10.1.
Fig. \[fig:all\_talos\] shows the total time to solution of CPU-ELPA1, CPU-ELPA2, GPU-ELPA1, and GPU-ELPA2 on the Talos cluster. All eigenvalues and eigenvectors of randomly generated real and complex matrices of size $N$ = 40,000 to 100,000 are computed with up to 64 nodes. As already demonstrated in published benchmarks [@elpa_marek_2014; @elpa_gutheil_2014; @elpa_cook_2018; @elpa_auckenthaler_2013], CPU-ELPA2 greatly outperforms CPU-ELPA1 in terms of performance and scalability. The performance difference between GPU-ELPA1 and GPU-ELPA2 is rather small. For small node counts, GPU-ELPA1 is marginally faster than GPU-ELPA2, which agrees with the published single-node tests [@elpa_kus_2019b]. What has not been tested previously is the performance on multiple nodes. It turns out that GPU-ELPA2 becomes faster than GPU-ELPA1 as the node count increases. The crossover point depends on the dimension of the problem, e.g., 4 nodes in the $N$ = 40,000 real case, and 16 nodes in the $N$ = 100,000 complex case. When using 64 nodes, the speedup of GPU-ELPA2 over GPU-ELPA1, averaged over all matrix sizes in Fig. \[fig:all\_talos\], is 2.2x.
In Fig. \[fig:all\_talos\], the speedup enabled by the GPUs ranges from no speedup at all to 3.3x. Three general trends emerge: (1) For the same matrix size, the speedup becomes smaller as the node count increases. For small $N$, CPU-ELPA2 can even be faster than the GPU-accelerated solvers, thanks to the near-optimal strong scaling of CPU-ELPA2. (2) For the same number of nodes, the speedup becomes larger as the matrix size increases. (3) For the same node count and the same matrix size, the speedup is larger for complex matrices than for real matrices.
{width="99.00000%"}
The timing experiment in Fig. \[fig:all\_talos\] is repeated on the Summit supercomputer. The results are shown in Fig. \[fig:all\_summit\]. CPU-ELPA1 is omitted for simplicity. On Summit, the GPU-accelerated solvers GPU-ELPA1 and GPU-ELPA2 are always faster than the CPU-ELPA2 solver, with a maximum speedup of over 20x. The speedup of GPU-ELPA2 over CPU-ELPA2 remains 2.2x ($N$ = 40,000 real) to 6.7x ($N$ = 100,000 complex) even for 64 nodes. For the same matrix size and the same node count, the speedup on Summit appears greater than on Talos, which can be partially attributed to the difference in hardware. Summit has 6 GPUs per node, whereas Talos only has two GPUs per node. Data transfers on Summit take advantage of the NVLink technology [@nvlink_foley_2017] for high-bandwidth interconnect between CPUs and GPUs. Besides, a high-performance CPU kernel for the tridiagonal-to-banded back-transformation, written in AVX512 instructions [@elpa_kus_2019b], is employed for the Intel Xeon Gold CPUs on Talos, rendering better performance of CPU-ELPA2 on Talos. Overall, we therefore observe that the absolute per-node timings of CPU-ELPA2 in Figs. \[fig:all\_talos\] and \[fig:all\_summit\] are already lower on Talos than on Summit. This difference in the CPU-only results, which are the baseline of the reported speedups, probably exaggerates the comparison of GPU-ELPA2 and CPU-ELPA2 on Summit somewhat, relative to Talos. Nevertheless, the three trends summarized from the tests on Talos are still valid on Summit, that is, the GPU speedup is larger for (1) larger matrix size, (2) fewer nodes, and (3) solving a complex problem instead of a real one. On both computers, the strong scaling of the GPU solvers is never as good as that of the CPU solvers. This can be explained by the workload assigned to the individual nodes. When using a large number of nodes or solving a small matrix, the workload on each node becomes so little that the many GPUs cannot be saturated, and the cost of CPU-GPU communications cannot be amortized. In contrast, when solving a large matrix or using a small number of nodes, a large amount of local work is offloaded to the GPUs, resulting in a significant speedup.
{width="99.00000%"}
Note that in Figs. \[fig:all\_talos\] and \[fig:all\_summit\], all eigenvalues and eigenvectors are computed. In applications such as materials simulations based on KS-DFT, only a portion of the eigenspectrum, e.g., typically 20% to 60% for LCAO basis sets, is of interest. In this case, the advantage of ELPA2 over ELPA1 should be more significant, as the computational complexity of the back-transformation is proportional to the number of eigenvectors to be calculated. This is demonstrated in Fig. \[fig:decomp3\_summit\], where the total timings (red circles) of the GPU-ELPA1 (solid) and GPU-ELPA2 (dashed) are decomposed into the forward tridiagonalization (blue squares), the solution of the tridiagonal problem (yellow diamonds), and the backward transformation (violet triangles). Two test cases, namely $N$ = 40,000 real and $N$ = 100,000 real, are shown as examples for a single node on Summit, i.e., the worst-case scenario for ELPA2. The two-stage tridiagonalization in ELPA2 is always faster than the one-stage tridiagonalization in ELPA1 by a factor of approximately three. The back-transformation accounts for a small fraction of the total time in ELPA1, but is the most time-consuming part in ELPA2 when all the eigenvectors are computed. When computing fewer eigenvectors, the burden of the two-stage back-transformation in ELPA2 can be greatly alleviated, making ELPA2 more favorable than ELPA1.
{width="99.00000%"}
Given that the optimal performance may be achieved with GPU-ELPA1, GPU-ELPA2, or CPU-ELPA2, depending on the specifics of the problem and the architecture, we highlight the auto-tuning feature in the ELPA library [@elpa_kus_2019b]. When ELPA is called repeatedly, like e.g. in a self-consistent KS-DFT calculation, this auto-tuning feature automatically iterates over possible combinations of solvers and runtime parameters. The best solver for a given problem can be identified and utilized without any additional input from the user.
In Fig. \[fig:decomp5\_summit\], we further decompose the timings of CPU-ELPA2 and GPU-ELPA2 into the five computational steps defined in Eq. \[eq:2stage\]. Again, the $N$ = 40,000 real and $N$ = 100,000 real tests are shown as examples. Steps that have been GPU-accelerated display an excellent speedup, namely 5.8x, 6.1x, 17.1x, and 11.3x (averaged over all data points in Fig. \[fig:all\_summit\]) for the full-to-banded transformation (Eq. \[eq:2stage\_fwd1\]), the solution of the tridiagonal problem (Eq. \[eq:2stage\_solve\]), the tridiagonal-to-banded back-transformation (Eq. \[eq:2stage\_bkwd1\]), and the banded-to-full back-transformation (Eq. \[eq:2stage\_bkwd2\]), respectively. The banded-to-tridiagonal transformation step (Eq. \[eq:2stage\_fwd2\]) is not yet GPU-accelerated, as it never stands as a bottleneck. The tridiagonal-to-banded back-transformation, which uses the newly developed CUDA kernel described in Sec. \[subsec:gpu\_step4\], shows a strong scaling that is close to ideal. The scaling of the full-to-banded transformation and the banded-to-full back-transformation is not as good as the other steps. These two steps limit the overall parallel efficiency of GPU-ELPA2, therefore they would be the first target for further algorithmic and technical optimization.
{width="99.00000%"}
Conclusions {#sec:conclude}
===========
In this paper, we report our GPU-oriented optimizations of the two-stage tridiagonalization eigensolver ELPA2 in the ELPA library, the only publicly available, distributed-memory, GPU-accelerated eigensolver library at the time of writing. The local BLAS operations in ELPA2 are offloaded to GPUs via the cuBLAS library. The tridiagonal-to-banded back-transformation of eigenvectors, which cannot be easily written as BLAS operations, is GPU-accelerated by a CUDA implementation. The overall performance of the GPU-accelerated ELPA2 solver is promising. It delivers a significant performance boost over the CPU-only version of ELPA2, as demonstrated by benchmarks on the Talos and Summit computers. Owing to the advanced two-stage tridiagonalization algorithm, the parallel scaling of the GPU-ELPA2 solver is superior to that of the GPU-ELPA1 solver. Based on an analysis of the individual computational steps in ELPA2, we identify the full-to-banded transformation and banded-to-full back-transformation steps as the next target for future optimization, as their strong scaling is not yet optimal. Nevertheless, the GPU-ELPA2 solver in its current form already unlocks a significant potential of GPU computations.
Acknowledgments {#sec:thanks .unnumbered}
===============
This research was supported by the National Science Foundation (NSF) under Award 1450280. Yu was supported by a fellowship from the Molecular Sciences Software Institute under NSF Award 1547580. This research used resources of the Oak Ridge Leadership Computing Facility at the Oak Ridge National Laboratory, which is supported by the Office of Science of the U.S. Department of Energy under Contract DE-AC05-00OR22725. We thank Dr. Christoph Angerer and Scott Biersdorff from NVIDIA, Dr. Björn Lange, Dr. William Huhn, and Dr. Wenhui Mi from Duke University, Dr. Jack Deslippe, Dr. Mauro Del Ben, and Dr. Charlene Yang from National Energy Research Scientific Computing, Dr. Eduardo D’Azevedo from Oak Ridge Leadership Computing Facility (OLCF), Dr. Markus Rampp from Max Planck Computing and Data Facility, Garching, Germany, and Dr. Christian Carbogno and Dr. Matthias Scheffler from Fritz Haber Institute of the Max Planck Society, Berlin, Germany for fruitful discussions. Part of the optimization presented in this paper was the outcome of the 2019 OLCF GPU Hackathon. We gratefully acknowledge the organizer Dr. Thomas Papatheodore (OLCF) and mentors Brent Leback (NVIDIA) and Dr. Markus Eisenbach (OLCF) for their advice during this event.
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'N.González, A.SmithCastelli, F.Faifer, C.Escudero, S.A.Cellone,'
bibliography:
- 'biblio.bib'
title: 'Galaxias de bajo brillo superficial: análogas a los satélites de Andrómeda en Pegasus I?'
---
Introducción {#S_intro}
============
En el trabajo de @2018arXiv181000710G se plantea una posible correlación entre la distancia de los cúmulos/grupos que albergarían galaxias de bajo brillo superficial (LSB) y sus radios efectivos. Esta correlación parece indicar que las galaxias LSB poseen un tamaño real más grande a mayores distancias. En cambio, sus tamaños aparentes resultan más pequeños a distancia lejanas. Este efecto podría deberse solamente a las limitaciones que resultan de observar estos tipos de objetos de bajo brillo superficial. Esto es, las LSB pequeñas observadas a grandes distancias podrían confundirse con objetos de fondo, mientras que aquellas galaxias LSB cercanas podrían no detectarse debido a sus tamaños angulares extremadamente grandes. Sin embargo, la probabilidad de encontrar galaxias LSB grandes en un volumen pequeño alrededor del Grupo Local resultaría baja. Recientemente, @Muller2018 encontraron posibles galaxias ultra difusas (UDGs) extremadamente grandes en el grupo de Leo-I (D $\sim$ 10.7 Mpc). De confirmarse su pertenencia en este grupo, estos objetos pasarán a ser las UDGs más cercanas.
![Mosaico de 2.85 $\times$ 2.85 arcmin en el filtro $g'$ en el cual se muestra dos de las galaxias LSB detectadas en PegasusI (círculos) reportadas en @2018arXiv181000710G. Superpuesto en este mosaico, a modo de comparación se muestra la apariencia que tendrían las dSphs de Andrómeda llevadas a la distancia de PegasusI (cuadrados).[]{data-label="Figura1"}](dSph_And_Peg-rot-ejes_color.png){width="41.00000%"}
{width="87.00000%"}
![Mosaico de 30.5$\times$30.5 arcmin en el filtro $r'$ de SDSS DR12 mostrando la región central del grupo de PegasusI. Los marcos azules corresponden a los campos de GEMINI- GMOS utilizados en este trabajo ($\sim$ 5.5 arcmin de lado). Los círculos rojos indican la ubicación de las candidatas LSB [@2018arXiv181000710G]. Los cuadrados violetas indican la ubicación de galaxias candidatas cuya morfología resultan similares a las galaxias satélites observadas en Andrómeda.[]{data-label="Figura3"}](satelites_andromeda_LSB-sochias.jpeg){width="40.50000%"}
![En los paneles izquierdos se muestran las imágenes de 10$\times$10 arcsec en el filtro $g'$ de cuatro de los objetos encontrados en la primera búsqueda con características similares a los satélites de Andrómeda a la distacia de PegasusI. En los paneles derechos se muestran sus respectivas imágenes en el filtro $i'$. En ambos filtros se considero los mismos niveles de despliegue. Los cuadros negros señalan los objetos petencialmente canditados, los cuadros blancos indican los objetos que podrían ser descartados.[]{data-label="Figura4"}](c1_ID505-g_ims-proc.jpeg "fig:"){width="24.00000%"} ![En los paneles izquierdos se muestran las imágenes de 10$\times$10 arcsec en el filtro $g'$ de cuatro de los objetos encontrados en la primera búsqueda con características similares a los satélites de Andrómeda a la distacia de PegasusI. En los paneles derechos se muestran sus respectivas imágenes en el filtro $i'$. En ambos filtros se considero los mismos niveles de despliegue. Los cuadros negros señalan los objetos petencialmente canditados, los cuadros blancos indican los objetos que podrían ser descartados.[]{data-label="Figura4"}](c1_ID505-i_ims-proc.jpeg "fig:"){width="24.00000%"}\
![En los paneles izquierdos se muestran las imágenes de 10$\times$10 arcsec en el filtro $g'$ de cuatro de los objetos encontrados en la primera búsqueda con características similares a los satélites de Andrómeda a la distacia de PegasusI. En los paneles derechos se muestran sus respectivas imágenes en el filtro $i'$. En ambos filtros se considero los mismos niveles de despliegue. Los cuadros negros señalan los objetos petencialmente canditados, los cuadros blancos indican los objetos que podrían ser descartados.[]{data-label="Figura4"}](c1_ID3941-g_ims-proc.jpeg "fig:"){width="24.00000%"} ![En los paneles izquierdos se muestran las imágenes de 10$\times$10 arcsec en el filtro $g'$ de cuatro de los objetos encontrados en la primera búsqueda con características similares a los satélites de Andrómeda a la distacia de PegasusI. En los paneles derechos se muestran sus respectivas imágenes en el filtro $i'$. En ambos filtros se considero los mismos niveles de despliegue. Los cuadros negros señalan los objetos petencialmente canditados, los cuadros blancos indican los objetos que podrían ser descartados.[]{data-label="Figura4"}](c1_ID3941-i_ims-proc.jpeg "fig:"){width="24.00000%"}\
![En los paneles izquierdos se muestran las imágenes de 10$\times$10 arcsec en el filtro $g'$ de cuatro de los objetos encontrados en la primera búsqueda con características similares a los satélites de Andrómeda a la distacia de PegasusI. En los paneles derechos se muestran sus respectivas imágenes en el filtro $i'$. En ambos filtros se considero los mismos niveles de despliegue. Los cuadros negros señalan los objetos petencialmente canditados, los cuadros blancos indican los objetos que podrían ser descartados.[]{data-label="Figura4"}](c1_ID3762-g_ims-proc.jpeg "fig:"){width="24.00000%"} ![En los paneles izquierdos se muestran las imágenes de 10$\times$10 arcsec en el filtro $g'$ de cuatro de los objetos encontrados en la primera búsqueda con características similares a los satélites de Andrómeda a la distacia de PegasusI. En los paneles derechos se muestran sus respectivas imágenes en el filtro $i'$. En ambos filtros se considero los mismos niveles de despliegue. Los cuadros negros señalan los objetos petencialmente canditados, los cuadros blancos indican los objetos que podrían ser descartados.[]{data-label="Figura4"}](c1_ID3762-i_ims-proc.jpeg "fig:"){width="24.00000%"}\
![En los paneles izquierdos se muestran las imágenes de 10$\times$10 arcsec en el filtro $g'$ de cuatro de los objetos encontrados en la primera búsqueda con características similares a los satélites de Andrómeda a la distacia de PegasusI. En los paneles derechos se muestran sus respectivas imágenes en el filtro $i'$. En ambos filtros se considero los mismos niveles de despliegue. Los cuadros negros señalan los objetos petencialmente canditados, los cuadros blancos indican los objetos que podrían ser descartados.[]{data-label="Figura4"}](c2_ID4547-g_ims-proc.jpeg "fig:"){width="24.00000%"} ![En los paneles izquierdos se muestran las imágenes de 10$\times$10 arcsec en el filtro $g'$ de cuatro de los objetos encontrados en la primera búsqueda con características similares a los satélites de Andrómeda a la distacia de PegasusI. En los paneles derechos se muestran sus respectivas imágenes en el filtro $i'$. En ambos filtros se considero los mismos niveles de despliegue. Los cuadros negros señalan los objetos petencialmente canditados, los cuadros blancos indican los objetos que podrían ser descartados.[]{data-label="Figura4"}](c2_ID4547-i_ims-proc.jpeg "fig:"){width="24.00000%"}\
Selección de las candidatas a LSB {#S_seleccion}
=================================
Con respecto a la posible presencia de un sesgo en la detección de galaxias LSB, nos preguntamos cuántos objetos similares a los satélites dSph de Andrómeda ($D~\sim~0.784$ Mpc) aparecerían a la distancia de PegasusI ($D~\sim~50$ Mpc). En la Fig. \[Figura1\] mostramos los modelos de seis de estás galaxias obtenidas de los perfiles de brillo reportados por [@1992AJ....103..840C] y [@1999AJ....118.1230C], sobrepuesto en uno de los campos GEMINI-GMOS en PegasusI.
Adicionalmente, comparamos los perfiles de brillo superficial de las galaxias LSB con los de los satélites de Andrómeda escaleados a la distancia de PegasusI (ver Fig. \[Figura2\]). Se puede ver que las dSphs de Andrómeda muestran tamaños entre $2~\textless~r_{\rm
tot}~\textless~8$ arcsec y brillos superficiales centrales entre $24.5~\textless~\mu_{0,g'}~\textless~26$ mag arcsec${^{-2}}$. Su brillo superficial central es bastante bajo, pero aún más brillante que los de las galaxias LSB detectadas en PegasusI. Sin embargo, en comparación, sus tamaños aparentes resultan más pequeños.
En este contexto, la pregunta que surge es, ¿Cuántas LSB de tipo temprano se podrían perder en la inspección visual porque son muy pequeñas y fáciles de confundir con los objetos de fondo? Para buscar estos objetos decidimos utilizar el software SExtractor adoptando los siguientes criterios: $$\begin{aligned}
&{\rm CLASS\_STAR} \leqslant0.2, \\
&{\rm FLAGS} \leqslant 2, \\
&{\rm Elipticidad} \leqslant 0.2,\\
&2~\textless~r_{\rm tot}~\textless~8~{\rm arcsec},\\
&24.5 \lesssim \mu_{0_{(g',r',i')}} \lesssim~26~{\rm mag~arcsec}^{-2},\\
&0~\lesssim~g'-i'~\lesssim~1.3,\\
&{\rm y}~0~\lesssim~g'-r'~\lesssim~1.2\\\end{aligned}$$ Se adoptaron estos rangos de colores considerando los valores mostrados para las galaxias LSB de tipo temprano en la sección 5.1 de @2018arXiv181000710G.
Datos fotométricos
==================
Este trabajo está basado en ocho campos profundos tomados en los filtros $g'$, $r'$ e $i'$ [@1996AJ....111.1748F], empleando la cámara GMOS de Gemini Norte (Programa GN-2008B-Q-14, PI: F. Faifer; Programa GN-2012A-Q-55, PI: A. Smith Castelli; Programa GN-2012B-Q-69, PI: F. Faifer ; Programa GN-2014A-Q-70, PI: F. Faifer; Programa GN-2014B-Q-17, PI: N. González; Programa GN-2015B-Q-13, PI: N. González). Estas imágenes cubren el entorno cercano a las dos galaxias dominantes de PegasusI: NGC7626 y NCG7619, y se utilizaron para obtener la selección preliminar de de noventa galaxias candidatas cuya morfología resultan similares a las galaxias satélites de Andrómeda. La Fig. \[Figura3\] muestra la orientación de los diferentes campos analizados y la ubicación de estos objetos. Por su parte, en la Fig. \[Figura4\], se muestran las imágenes de cuatro de estás candidatas en los filtros $g'$ e $i'$.
Dado que las galaxias LSB con tamaños similares a las de los satélites de Andrómeda serán difíciles de detectar en una inspección visual a la distancia de PegasusI, la identificación de las mismas se realizó utilizando SExtractor en las imágenes GEMINI-GMOS. Para ello, utilizando las tareas [*ellipse*]{} y [*bmodel*]{} de IRAF, se procedió a modelar las distribuciones de brillo superficial de las galaxias elípticas NGC7626 y NGC7619 y sus respectivos halos, incluyendo varios objetos extendidos. Luego, se restaron dichos modelos con el fin de poder identificar y medir las candidatas a galaxias LSB (para más detalles, remitimos al lector a la sección3 de @2018arXiv181000710G). Posteriormente realizamos la fotometría con SExtractor utilizando el criterio de selección mencionado en la Sec. \[S\_seleccion\], el cual permitió identificar noventa objetos con características similares a los satélites de Andrómeda a la distancia de PegasusI.
Resultados preliminares y trabajo a futuro
==========================================
En este trabajos se realizó la detección de objetos similares a las galaxias satélites dSph de Andrómeda en el grupo PegasusI. Como resultado, se detectaron en el grupo noventa objetos con estas características. El primer paso, será realizar una revisión detallada de estos objetos, y analizar si los mismos presentan algún tipo de subestructura y/o posible formación estelar, ya que no es esperable que estas características esten presentes en galaxias de tipo temprano. Para ello, se comenzará inspeccionando las imágenes de cada objeto en los filtros $g'$ e $i'$. Como ejemplo, en la Fig. \[Figura4\], se muestran en los paneles superiores dos objetos que son potencialmente candidatos a ser galaxias similares a las satélites de Andrómeda. En contraparte, en los paneles inferiores se muetran dos objetos que podrían ser descartados de la primera selección por mostrar un cierta subestructura.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'By repeated trials, one can determine the fairness of a classical coin with a confidence which grows with the number of trials. A quantum coin can be in a superposition of heads and tails and its state is most generally a density matrix. Given a string of qubits representing a series of trials, one can measure them individually and determine the state with a certain confidence. We show that there is an improved strategy which measures the qubits after entangling them, which leads to a greater confidence. This strategy is demonstrated on the simulation facility of IBM quantum computers.'
address:
- '${}$ Indian Institute of Technology, Kharagpur'
- '${}$Raman Research Institute, Bangalore 560080, India. '
author:
- 'Arpita Maitra$^{}$'
- 'Joseph Samuel$^{}$'
- 'Supurna Sinha$^{}$'
title: 'Likelihood Theory in a Quantum World: tests with Quantum coins and computers'
---
[*[Introduction:]{}*]{} When testing a theory against experimental data, the Bayesian approach gives us a rational way of revising our theoretical expectations in the light of new data. To take the simple and familiar example of coin tossing, let us start with the belief that our coin is fair. If we then toss the coin ten times and turn up nine heads, our belief in fairness will be shaken but not destroyed: it is still possible that the nine heads were generated by chance. If the run of heads continues, we would be hard pressed to cling to our belief in the fairness of the coin. Revising our beliefs in the light of new data is an essential component of the scientific method, a point which was strikingly brought out in the Monty Hall problem, which occupied the community in the 1990’s. Bayesian theory is routinely used in testing the efficacy of drugs, where one starts with the “null hypothesis” that the drug being tested is no more effective than a placebo. If the drug fares better than a placebo over a sufficiently large number of trials, we are more inclined to believe in its efficacy.
Let us now return to coin tossing, which captures the idea of Bayesian inference in its simplest form. The intuitive idea is quantitatively captured in likelihood theory. Our initial belief (prior) is that the probability of heads is $q_{H}$ and tails $q_{T}$. If we see a particular string ${\cal S}_N=\{HTHHHHTTHTTT.....\}$, in $N$ coin tosses, the likelihood of such a string emerging from the distribution $P_B=\{q_H,q_T\}$ is given by (see below for a derivation) $$L({\cal S}_N|P_B)=\frac{1}{\sqrt{2\pi Np_H p_T}} \exp{-\{ND_{KL}(P_A\|P_B)\}}
\label{firstequation}$$ where $P_A=\{p_H,p_T\}=\{N_H/N,N_T/N\}$ is the observed frequency distribution in a string of $N$ tosses. $D_{KL}(P_A\|P_B)$ is the relative entropy or the Kullback-Leibler divergence between the distributions $P_B$ and $P_A$.
When we pass from classical coins to quantum coins, a new possibility emerges: the coins may be in a superposition of heads $\ket{H}$ and tails $\ket{T}$. Even more generally, the state of the system need not even be pure but a density matrix. The probability distributions $P_B, P_A$ describing classical coins are replaced by density matrices $\rho_B$ and $\rho_A$. One of the central problems in quantum information is quantum state discrimination. Given two quantum states, how easily can we tell them apart? This question has led to the issue of distinguishability measures on the space of quantum states [@shunichi; @statest; @anthony; @osaki; @barnett]. More recently in [@entropy] it has been noticed that quantum entanglement leads to a significant improvement in distinguishing two quantum states. In this Letter, we go beyond [@entropy] by actually implementing our theoretical model on a state of the art quantum computer, the IBM quantum computer and checking our theoretical ideas against quantum simulations and experiments.
Our goal in this Letter is to describe an experiment which demonstrates the use of entanglement as a resource in state discrimination. We have successfully implemented our ideas on the ibm quantum simulator. In experimental runs, it appears that the quantum advantage we seek to demonstrate is swamped by noise. The experiment may well be within the reach of more sophisticated quantum computers which are currently in use and so provides a short term goal possibly within reach.
The outline of this Letter is as follows. We first review classical likelihood theory to set the background for our work. We then describe how likelihood theory is modified in a quantum world and show how these abstract ideas can be translated into reality using existing quantum computers. We discuss the simplest direct measurement strategy where one measures qubits one by one. We then show that there exist improved strategies which exploit quantum entanglement in an essential way. We translate our abstract ideas into quantum scores, combinations of quantum logic gates, which can be run on quantum computers. Finally, we present the results of running our programs on existing IBM quantum computers and end with some concluding remarks.
[*[Review of Classical Likelihood Theory:]{}*]{} Let us consider a biased coin for which the probability of getting a head is $p_H= 1/3$ and that of getting a tail is $p_T= 2/3$. Suppose we incorrectly assume that the coin is fair and assign probabilities $q_H= 1/2$ and $q_T= 1/2$ for getting a head and a tail respectively. The question of interest is the number of trials needed to be able to distinguish (at a given confidence level) between our assumed probability distribution and the measured probability distribution. A popular measure for distinguishing between the expected distribution and the measured distribution is given by the relative entropy or the KL divergence (KLD) which is widely used in the context of distinguishing classical probability distributions [@jon]. Let us consider $N$ independent tosses of a coin leading to a string ${\cal S}_N=\{HTHHTHTHHTTTTT......\}$. What is the probability that the string ${\cal S}$ is generated by the model distribution $P_B=\{q,1-q\}$? The observed frequency distribution is $P_A=\{p,1-p\}$. If there are $N_H$ heads and $N_T$ tails in the string ${\cal S}_N$ then the probability of getting such a string is $\frac{N!}{N_H!N_T!}q^{N_H}(1-q)^{N_T}$ which we call the likelihood function $L({\cal S}_N|P_B)$. If we take the average of the logarithm of this likelihood function and use Stirling’s approximation for large $N$ we get the following expression: $$\frac{1}{N}\log{L(N|{P_B)}}=-D_{KL}(P_A\|P_B)+\frac{1}{N}\log{\frac{1}{\sqrt{2\pi Np(1-p)}}},
\label{likelihood}$$ where $p=\frac{N_H}{N}$ and $D_{KL}(P_A\|P_B)=p\log{\frac{p}{q}}+(1-p)\log{\frac{1-p}{1-q}}$. The second term in (\[likelihood\]) is due to the sub-leading term $\frac{1}{2}\log{2\pi N}$ of Stirling’s approximation. If $D_{KL}(P_A\|P_B)\neq 0$ then the likelihood of the string $S$ being produced by the $P_B$ distribution decreases exponentially with $N$. This results in eq.(\[firstequation\]). Thus $D_{KL}(P_A\|P_B)$ gives us the divergence of the measured distribution from the model distribution. The KL divergence is positive and vanishes if and only if the two distributions $P_A$ and $P_B$ are equal.
In this limit, we find that the exponential divergence gives way to a power law divergence, as shown by the subleading term in (\[likelihood\]). The arguments above generalize appropriately to an arbitrary number of outcomes (instead of two) and also to continuous random variables.
[*[Quantum Likelihood Theory:]{}*]{} In passing from classical likelihood theory described above to the quantum one, there are two additional features to be considered. First, as mentioned earlier, because of the superposition principle, a quantum coin would in general be described by a density matrix rather than a probability distribution. Second, in measuring a quantum system, one has an additional freedom: the choice of measurement basis. It is evident that the choice of measurement basis can affect our discriminating power between competing quantum states. We would be well advised to choose the basis wisely, so as to maximise our discriminating power between two given density matrices.
Consider two mixed density matrices $\rho_A$ and $\rho_B$. Suppose that the (prior) quantum distribution is described by $\rho_B$. For a choice $\{m_i\}$ ($i=1...n$) of orthonormal basis, the probable outcomes have the probabilities $q_i=\bra{m_i}\rho_B\ket{m_i}$. Similarly, if the density matrix is $\rho_A$, we have the probabilities $p_i=\bra{m_i}\rho_A\ket{m_i}$. Both $\{q_i\}$ and $\{p_i\}$ are normalised probability distributions. These can now be regarded as classical probability distributions and we can compute their relative entropy, defined as $D_{KL}(m) = \sum_{i=1..n} p_i Log(p_i/q_i)$. The argument $m$ on the LHS indicates that the relative entropy [*depends*]{} on the measurement basis $m$, since the $p$s and $q$s do. Let us now choose $m=m^*$ optimally so as to maximise $D_{KL}(m)$ which is a measure of our discriminating power. The basis optimised value of the Kulback-Leibler divergence is denoted by $S(A||B)= D_{KL}(m^*)$. For example, we could be detecting a weak signal (as in quantum metrology) using a quantum detector, which would remain in the state $\rho_B$ in the absence of an incoming signal, but get excited to state $\rho_A$ when the signal is present. If the state is $\rho_A$, our confidence $C$ that the state is in fact $\rho_B$ will decrease at the rate $$C=\frac{1}{\sqrt{2\pi Np(1-p)}} \exp{-\{N S(A||B)\}}.$$ where $N$ is the number of measured systems.
[*[Tossing Quantum Coins:]{}*]{} The abstract theory of the last section can be converted into concrete experiments using the recent progress in constructing quantum computers in the laboratory. Choosing the optimal basis is effected by a Unitary transformation, which can be implemented using quantum gates. The quantum circuit that we implement on the quantum computer can be divided into three parts. The first part is state preparation, the second is Unitary transformation and the third is measurement. We explore two strategies, which we call the direct and the entangling strategy. The two strategies differ only in the second part. In the direct strategy, we measure the qubits one at a time, taking care to choose the measurement basis so as to maximise our ability to distinguish the two states. In the entangling strategy, we perform an entangling unitary transformation on a pair of qubits before performing the measurement.
[*State Preparation:*]{} Without loss of generality, we can assume the state $\rho_A$ to be along the $z$ direction of the Bloch sphere and $\rho_B$ to lie in the $x-z$ plane of the Bloch sphere, making an angle $\delta$ with the $z$ axis. To prepare the initial quantum state $\rho_A$, which is a density matrix, we begin with two qubits (say a,b) both in the state $\ket{0}$: $\ket{0}_a\otimes\ket{0}_b$. We then perform a local unitary rotation through an angle $\beta$ in the $x-z$ plane on one of them (b) and arrive at $\ket{0}_a\otimes(\cos{\beta/2}\ket{0}_b+\sin{\beta/2}\ket{1}_b)$. We then apply a CNOT gate with b as the control and a as the target. The total system is now in an entangled state. If we ignore (trace over) qubit a, we get an impure density matrix on qubit b. $\rho_A=(\cos{\beta/2})^2\ket{0}\bra{0}+(\sin{\beta/2})^2\ket{1}\bra{1}$ This is our initial state $\rho_A$, whose purity depends on the entangling angle $\beta$. In runs we have used the value of $\beta=0.2$ and $\delta=1.8$.
In practice (and for easier comparison with the entangling strategy) we repeat the arrangement with two more qubits with qubits a and b replaced respectively by d and c. We get identical states in qubits b and c. The rest of the circuit performs an unitary transformation on the qubits b and c and then performs a measurement in the computational basis.
[*[Direct strategy:]{}*]{} In the direct measurement strategy we perform measurements on individual qubits. One can show [@supplementary] that the optimal basis consists of two antipodal points in the $x-z$ plane of the Bloch sphere. This determines a direction in the $x-z$ plane which is characterised by an angle $\varphi$ in the range $0-\pi$. We optimise over $\varphi$ by choosing $\varphi$, the measurement basis so that $D_{KL}(\varphi)$ defined in the last section, is as large as possible. Then $$S_{rel}=D_{KL}(\varphi^{*})$$ is the maximum value where $\varphi^{*}$ is the value of $\varphi$ corresponding to the maximum value of $D_{KL}$. This gives us the optimal one-qubit strategy for distinguishability of $\rho_A$ and $\rho_B$. This is our direct measurement strategy. The optimal unitary transformation for the direct strategy is easy to find numerically. The unitary transformations of interest are in fact orthogonal [@supplementary], and belong to $SO(2)$ (because the states are assumed to be in the $x-z$ plane) and characterised by a single angle $\varphi$. We numerically evaluate the relative entropy as a function of $\varphi$ and find the maximum $\varphi^*$. We use this unitary transformation $U_2$ to get the optimised measurement basis, which is implemented as a local transformation on qubits b and c. The logic circuit which implements the direct strategy on a quantum computer is displayed in Fig. 1.
[*[Entangling strategy:]{}*]{} The entangling strategy for discriminating qubits gives us an advantage over the Direct Measurement Strategy, in quantum state discrimination. Consider $2N$ idependent and identically distributed qubits. Let us group these $2N$ qubits into $N$ pairs. We now perfom measurements on each pair, choosing our measurement basis to give us optimal results for state distinguishability. This is effected by performing an unitary transformation $U_4$ [*in the two qubit Hilbert space*]{} followed by measurement in the computational basis. By grouping the qubits in pairs, we have increased the freedom in the choice of measurement basis. For, we can choose entangled bases as well as separable ones. This greater freedom means that we can improve on the direct strategy. Maximising $S_{rel}$ over [*all*]{} two qubit bases, is clearly an improvement over maximising over separable ones.
We maximise the relative entropy in the space of two qubit bases, by doing a numerical search. Computer experiments reveal that one does indeed gain by using the entangling strategy. Our initial aim was to show that this entangling strategy indeed gives us an improvement over the Direct Measurement Strategy which we have described in the last section, by explicitly implementing it on a quantum computer.
For the entangled strategy, we find it advantageous to simplify our search by searching within $SO(4)$. We perform a random walk in the $SO(4)$ space to find the optimal $U_4$. This yields a relative entropy per qubit which is higher than that obtained by the direct strategy. It is known from the theory of Makhlin transforms [@makhlin; @shende] how to decompose any $U_4$ into local unitary gates and CNOT gates. It turns out that the unitary can be implemented (up to a phase) by using just two CNOT gates and six unitary gates. Our mathematica programs compute the required unitaries and produce a qasm program that can be implemented on the ibm machines. The quantum logic circuit implementing the entangling strategy is shown in Fig. 2.
![Figure shows the log of the likelihood as a function of the number or runs. The red dots represent the result of using the entanglement strategy and the blue ones use the direct strategy. The red dots show a steeper negative slope. This shows that the likelihood of the state being $\rho_B$ decreases faster as the number of qubits measured increases. []{data-label="3"}](plotsrel.pdf){width="40.00000%"}
[*[Results:]{}*]{} Let us summarize the results of the study. We have implemented our idea at three levels:
\(a) a mathematica program which numerically computes and compares the relative entropy for the two computational strategies - the Direct and the Entangled, (b) a simulation on the ibmqx4 computer which mimics the actual quantum experiment by incorporating some of the specific engineering details, and (c) an actual quantum experiment on the ibmqx4 computer. Below is a table which summarizes our results. The relative entropies corresponding to the direct and entangling strategies $S_{\mathrm{reldir}}$ and $S_{\mathrm{relent}}$ and their difference $S_{\mathrm{reldiff}}=S_{\mathrm{relent}}-S_{\mathrm{reldir}}$ are displayed corresponding to the mathematica program (Theory), ibm simulator (Sim) and ibm experiment (Expt) in the table.
[ | m[5em]{} | m[2cm]{}| m[2cm]{} | m[2cm]{}| ]{} & Theory& Sim &Expt\
Srel dir & 4.506& 4.519&4.043\
Srel ent & 4.723 & 4.696&1.660\
Srel diff& 0.217& 0.177&-2.382\
We notice that our mathematica programs are in close agreement with the results of the simulations on the ibmqx4 quantum computer. However, there is a considerable discrepancy between the results of the real quantum ibmqx4 experiments and the ibmqx4 simulations or the results of the mathematica programs. This discrepancy could be due to limitations coming from gate infidelity and decoherence. Thus the noise from these two sources masks the quantum advantage effect that we expect from our theoretical studies [@entropy].
We graphically capture the main result of our study in Fig.3 where we plot the Log\[Likelihood\] versus the number of qubits which clearly demonstrates the superiority of the Entangled Strategy over the Direct Strategy. Thus there is a clear quantum advantage in choosing the Entangled Strategy instead of the Direct one.
[*[Conclusion:]{}*]{} In this Letter, we described how Likelihood theory works in the quantum regime and show how these ideas can be implemented on a quantum computer. The main message is that entanglement can be used as a resource to improve our power to discriminate between quantum states of qubits. Entanglement is crucially used here, since it is our ability to use non separable bases that gives us the quantum advantage.
The scheme we propose has been implemented on the ibmqx4 computer. We have performed both simulations and real experiments on this computer. We notice that our theoretical expectation based on Mathematica programs agrees very well with results of the simulations. However, decoherence and limitations in gate fidelity mask the effect of quantum advantage stemming from entanglement that we notice in our Mathematica programs and simulations on ibmqx4. We have also tested our ideas on the Melbourne quantum computer IBM Q16, which has provided some marginal improvement in the experimental results.
Our study therefore has two significant aspects to it. It has provided a way to check theoretical expectations of quantum advantage from entanglement in the context of Quantum Likelihood Theory. In addition it has brought out the limitations of the present day ibmqx4 computer so far as experiments go. This would motivate researchers to improve the quality of the experiments by improving gate fidelity and by reducing noise due to decoherence.
[*[Acknowledgement]{}*]{} We acknowledge the use of the ibmqx4 quantum computer and the IBM Q16 computer in Melbourne.
[10]{} natexlab\#1[\#1]{}bibnamefont \#1[\#1]{}bibfnamefont \#1[\#1]{}url \#1[`#1`]{}urlprefix\[2\][\#2]{} \[1\][\#1]{} \[1\][\#1]{} temptokena[ citenamefont citenamefont citenamefont ]{}@fileswauxout[temptokena]{}temptokena
, **, Applied Mathematical Sciences (, ), ISBN , <https://books.google.co.in/books?id=UkSFCwAAQBAJ>.
, ** (, ), ISBN .
, ****(), (), <http://dx.doi.org/10.1080/00107510010002599>, .
, , , ****, (), <http://link.aps.org/doi/10.1103/PhysRevA.54.1691>.
, ****(), (), <http://aop.osa.org/abstract.cfm?URI=aop-1-2-238>.
, , , , ****, ().
, **** (), <http://arxiv.org/abs/1404.2000>.
().
, ****, ().
, , , ****, (), <https://link.aps.org/doi/10.1103/PhysRevA.69.062321>.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The Airy gas model of the edge electron gas is used to construct an exchange-energy functional which is an alternative to those obtained in the local density and generalized gradient approximations. Test calculations for rare gas atoms, molecules, solids and surfaces show that the Airy gas functional performs better than the local density approximation in all cases and better than the generalized gradient approximation for solids and surfaces.'
address:
- |
Condensed Matter Theory Group, Physics Department,\
Uppsala University, S-75121 Uppsala, Sweden
- |
Research Institute for Solid State Physics,\
H-1525 Budapest, P.O.Box 49, Hungary
- |
Center for Atomic-scale Materials Physics and Department of Physics,\
Technical University of Denmark, DK-2800 Lyngby, Denmark
author:
- 'L. Vitos, B. Johansson'
- 'J. Kollár'
- 'H. L. Skriver'
date: 4 May 2000
title: The exchange energy in the local Airy gas approximation
---
15.6cm
Since the pioneering papers on density functional theory (DFT) [@hohenberg64; @kohn65] there has been a constant search for exchange-correlation functionals of chemical accuracy. This includes the works on the generalized gradient approximation (GGA) [@perdew86; @perdew91; @perdew96; @zhang98; @perdew98] which are dedicated efforts to construct local functionals for inhomogeneous systems ranging from atoms to solids based on the uniform electron gas, i.e., the local density approximation (LDA), and density gradient corrections, as well as the development of a number gradient level, semiempirical functionals [@lee88; @becke88; @hamprecht98; @hammer99]. The GGA functionals have had a considerable impact upon the fields of quantum chemistry and solid state physics because they reduce the LDA overbinding and generally improve the calculated properties, relative to experiments, of molecules [@perdew96; @perdew99; @kurth99; @becke86p] and bulk solids [@perdew92p; @vitos97; @ozolins93; @kollar97; @soderlind96; @cho96]. However, they perform less well for the bulk properties of late transition metals and semiconductors [@khein95; @corso96; @filippi94], and the underestimate of the exchange energies of surfaces [@perdew92p] as well as the overestimate of the dissociation energies of the multiply bonded molecules [@perdew96; @kurth99] indicate the necessity to go beyond the gradient level approximations and develop functionals that depend upon other inhomogeneity parameters, e.g., higher derivatives of the charge density or the Kohn-Sham kinetic energy density. One step in this direction is the meta generalized gradient approximation (meta-GGA) of Perdew, Kurth, Zupan, and Blaha [@perdew99] which proves highly promising for both finite and extended systems [@kurth99].
In the present work we introduce and apply a new gradient level exchange energy functional based on the concept of the edge electron gas [@kohn98]. Besides the formal interest in the development of density based, orbital independent functionals there are several reasons why in applications of DFT the focus is on the approximate, local exchange-correlation schemes. Within the Kohn-Sham approach to DFT the Kohn-Sham exchange energy may be determined exactly and as demostrated recently [@stadele97; @gorling99; @ivanov99] so may the corresponding local exchange potential. However, the exact Kohn-Sham exchange formalism is non-local and orbital-based, i.e. both the exchange energy and potential are highly complicated non-local functionals of the Kohn-Sham orbitals. In consequence, the application of exact exchange is computationally demanding. Furthermore, when exchange is treated exactly the error cancellations between the exchange and correlation energies on which all approximate schemes depend are lost owing to the poor description of correlation effects and, as a result, the total energies worsen [@perdew92p; @gorling99]. For these reasons the exact Kohn-Sham exchange energy has only been used in practice in connection with semi-empirical, hybrid approximations [@becke].
The concept of the edge electron gas was put forward by Kohn and Mattsson [@kohn98] as an appropriate basis for the treatment of systems with edge surface outside of which all Kohn-Sham orbitals decay exponentially. Its simplest realization, the Airy gas model, is based on the linear potential approximation and may serve as the starting point for the construction of functionals which are alternative to the GGA. The Airy gas model has recently been used to construct an explicit kinetic energy functional for inhomogeneous systems [@vitos99] which for atoms and surfaces has the accuracy of functionals based on a second order gradient expansion.
Here we have taken the exchange energy of the Airy gas model derived by Kohn and Mattsson [@kohn98] and cast it in a form amenable to a simple, accurate parametrization. The procedure may be viewed as local mapping of the real system described by its density and scaled gradient onto the Airy gas model and represents one possible solution to the joining of the interior to the edge regions. The parametrized functional which we refer to as the local Airy gas (LAG) functional is tested in calculations of the exchange energies of rare gas atoms and of metallic surfaces within the jellium model where the exact results are known [@lang70]. In addition, we apply the LAG exchange functional in conjunction with the LDA for correlation [@perdew92] in calculations of the molecular binding energies and bulk properties of solids.
The present LAG exchange functional has a number of advantages over previous GGA functionals: i) it explicitly includes the properties of the edge region where much interesting physics occurs, ii) its accuracy may be systematically improved by including higher order expansions of the effective potential of the model system, and iii) the resulting exchange-energy functional is as simple and well-defined as that of the standard LDA. i.e., it has no adjustable parameters.
The starting point for the Airy gas exchange energy functional is the potential
$$\begin{aligned}
\label{eq:linearpot}
v_{eff}(z) = \left\{
\begin{array}{lll}
\infty\ &\mbox{for} & z \le -L \\
Fz &\mbox{for} & -L < z < \infty
\end{array}\right.,\end{aligned}$$
which is linear in $z$, independent of $x$ and $y$, and has a hard wall at $-L$ far from the electronic edge at $z=0$. The slope of the effective potential $F = dv_{eff}/dz$ leads to a characteristic length scale
$$\label{eq:l}
l\equiv \left(\frac{\hbar^2}{2mF}\right)^{1/3} ,$$
and the electron and exchange-energy densities are then given by
$$n(z) = l^{-3}\;n(\zeta),$$
and
$$\label{eq:exdef}
\varepsilon_x(z) =-\frac{e^2}{2} l^{-4}\;\varepsilon_x(\zeta),$$
where $\zeta=z/l$,
$$n(\zeta) = \;\frac{1}{2\pi}\int_0^{\infty} Ai^2(\zeta+\zeta')\ \zeta'd\zeta',$$
and $$\begin{aligned}
\label{eq:exairy}
\varepsilon_x(\zeta)& = & \frac{1}{\pi}\int_{-\infty}^{\infty}
\int_0^{\infty}\int_0^{\infty} Ai(\zeta+\epsilon)
Ai(\zeta'+\epsilon) Ai(\zeta+\epsilon') Ai(\zeta'+\epsilon')
\nonumber \\
&\times& |\zeta'- \zeta|^{-3}
g(\sqrt{\epsilon}|\zeta'- \zeta|,\sqrt{\epsilon'}|\zeta'- \zeta|)
d\zeta' \; d\epsilon \; d\epsilon'.\end{aligned}$$ A contour plot of the universal function $g(s,s')$ may be found in Ref.[@kohn98]. The exchange energy (\[eq:exdef\]) may be written in the form
$$\label{eq:exF}
\varepsilon_x(z)=\varepsilon_x^{LDA}(z)F_x[s(z)],$$
where $\varepsilon_x^{LDA}(z)$ is the exchange energy density of the uniform electron gas. The enhancement function
$$\label{eq:F}
F_x(\zeta) \equiv \frac{2}{3}\left (\frac{\pi}{3}\right )^{1/3}
\frac{\varepsilon_x(\zeta)}{n^{4/3}(\zeta)},$$
is the unique function $(F_x(\zeta),s(\zeta))$ of the scaled gradient
$$\label{eq:gradient}
s(\zeta) \equiv \frac{n'(\zeta)}{2(3\pi^2)^{1/3}n^{4/3}(\zeta)},$$
plotted in Fig. \[fig1\]. For comparison we also present results obtained by the GGA of Perdew, Burke, and Ernzerhof (PBE) as defined in Ref.[@perdew96], and the second order gradient expansion (GEA) [@gea]. It follows from the figure that the exchange density (\[eq:exF\]) in the low gradient limit of the Airy gas model reduces to $\varepsilon_x^{LDA}(z)$ as it should. In the large gradient limit $\varepsilon_x(\zeta) \approx -\frac{n(\zeta)}{2\zeta}$, Ref. [@kohn98], and similar to the case of the kinetic energy density [@vitos99] we use the properties of the Airy gas to find the following explicit asymptotic expression
$$\label{eq:large}
\varepsilon_x[n(z)] \approx -\frac{e^2}{2}\frac{n(z)}{2}\left[ n(z)
\frac{\partial^3 n(z)}{\partial z^3}
-\frac{\partial n(z)} {\partial z}\frac{\partial^2 n(z)}{\partial z^2}
\right] \left [
\frac{\partial n(z)} {\partial z}\right]^{-2},$$
in terms of the density and its derivatives.
The density of the exchange energy per electron of the Airy gas is plotted as a function of the distance $z$ from the electronic edge in Fig. \[fig2\]. It is seen that the large gradient expression (\[eq:large\]) is accurate for $z/l > -1.4$ corresponding to $s > 0.5$. It is also seen that neither the LDA nor the PBE GGA [@perdew96] leads to the correct behaviour near and beyond the electronic edge at $z=0$.
The scaled gradient is conserved when going from the real electron gas to the Airy gas model [@vitos99] and therefore the enhancement function $F_x(s)$ parametrized, for instance, in a modified Becke form [@becke86]
$$\label{eq:exp}
F^{LAG}_x(s) = 1+\beta\frac{ s^{\alpha}}
{(1+\gamma s^{\alpha})^{\delta}},$$
which includes the proper LDA limit, may be used to obtain the exchange energy density of the real electron gas from the local, scaled gradient $s[n(z)]$. For $\alpha= 2.626712, \beta=0.041106, \gamma= 0.092070$, and $\delta= 0.657946$ we find that the local deviation between the exact result (\[eq:F\]) and the parametrized form (\[eq:exp\]) integrated over the range $0<s<20$ is less than $0.3 \%$. We note that the present parametrization, being an overall fit, does not reduce to the GEA [@perdew92p; @dreizler90] in the low gradient limit. In contrast to the case of the kinetic energy [@vitos99] we have not been able to find an explicit, analytical expression for the exchange energy for small $s$ values, and to establish the behaviour numerically has not been attempted because the $s \rightarrow 0$ limit is reached only at $z\rightarrow -\infty$ as seen in Fig. \[fig2\]. The exact behaviour of the LAG exchange functional at $s \rightarrow 0$ is therefore not know at present.
In the following we report the results of applying the LAG exchange functional to four test systems: i) rare gas atoms, ii) diatomic molecules, iii) jellium surfaces, and iv) solids. In all cases the total energy is calculated using self-consistent LDA densities. For molecules and solids the LAG exchange energy is combined with the LDA correlation energy [@perdew92], since correlation effects has not been worked out in the Airy gas model. The motivation of this combination is given in terms of the enhancement function over the local exchange energy [@perdew96], defined as $F_{xc}(s)\equiv\epsilon_{xc}[n]/\epsilon^{LDA}(n)$, where $\epsilon_{xc}[n]$ denotes the exchange-correlation energy density.
Most of the currently applied approximate density functionals are based on error cancellations between the exchange and correlation energies [@kurth99; @gorling99]. For physically interesting densities this cancellation leads to $F_{xc}(s)$ with negligible slope up to $s\approx 1$. Plots of the enhancement function over the local exchange energy for gradient level and meta-GGA approximations can be found in Refs. [@perdew96; @kurth99]. In the present LAG exchange plus LDA correlation scheme this function becames $F^{LAG}_{xc}(s) = F^{LAG}_x(s) + \epsilon^{LDA}_c(n)/\epsilon^{LDA}_x(n)$, where $\epsilon^{LDA}_c(n)$ is the correlation energy density of the uniform electron gas. Thus, the $F^{LAG}_{xc}(s)$ is determined only by the LAG enhancement function (\[eq:exp\]), which, for $s<1$, is a slowly increasing function of $s$. Therefore, we expect the present exchange-correlation scheme to preserve the excellent cancellation properties of the LDA and PBE GGA, and, at the same time, to bring the calculated properties in closer agreement with experiment than conventional LDA.
For the rare gas atoms included in Table \[table1\] the GEA, PBE, and LAG functionals yield exchange energies which are, on the average, 6.4 %, 8.5 %, and 1.8 %, respectively, larger than those obtained in the LDA. The PBE values are in very good agreement with the exact Kohn-Sham results [@kurth99; @dreizler90], which are given relative to the LDA energies in the last column of the table. The LAG approximation represents only a minor improvement relative to the LDA total atomic exchange energies.
The effect of the gradient correction to the LDA atomization energies for a few selected diatomic molecules is shown in Table \[table2\] which also includes the relative difference between the LDA results and experimental data [@kurth99]. Here, the LDA charge densities for the molecules have been generated using the full charge density (FCD) technique in conjunction with the exact muffin-tin orbital method (EMTO) . It is seen that the LAG approximation (i.e. LAG exchange and LDA correlation energy) and PBE GGA have comparable accuracy: Both functionals reduce the LDA overbinding, and yield atomization energies which are, on the average, 16.8 % (PBE) and 16.2 % (LAG) smaller than the LDA values.
In Fig. \[fig3\] we compare four exchange functionals applied to the jellium model of metallic surfaces [@lang70]. The fact that for a given $r_s$-value the exchange energies become increasingly negative in the order LDA, LAG, GEA, and PBE is a simple consequence of the enhancement functions shown in Fig. \[fig1\] and in agreement with the observation that the GGA significantly underestimate surface exchange energies [@perdew92p]. We note that LAG approximation represents an improvement over both the LDA and PBE and vary less with $r_s$ than either of the other two approximations.
As a final test of the LAG approximation we have calculated the atomic volumes and bulk moduli of several metals and semiconductors in their observed low temperature crystal structures by means of the FCD-EMTO method [@kollar00; @andersen94; @vitos00]. The results for the equilibrium atomic radii are plotted in Fig. \[fig4\]. For some selected metals and semiconductors, for which accurate LDA, PBE, and meta-GGA results have been published [@kurth99], the atomic radii and bulk moduli are presented in Table \[table3\]. The comparison of our LDA [@perdew92] atomic radii for the transition metal series with those obtained by the full-potential linear muffin-tin orbital [@ozolins93] and linear augmented plane wave [@khein95] methods using the same LDA gives mean deviations of 0.33 %, 0.43 %, and 0.49 % for the $3d, 4d$, and $5d$ series, respectively. For Li and Na the present LDA results agree within 0.07 % with the full-potential values from Ref. [@perdew92p]. We therefore expect that the results of the present LAG and PBE calculations shown in Fig. \[fig4\] will deviate less than 0.5 % from full-potential calculations. The mean deviations between the present atomic radii and bulk moduli listed in Table \[table3\] and those of Ref. [@kurth99] obtained using the linear augmented plane wave method are 0.20 % and 3.28 % for the LDA and 0.27 % and 3.26 % for the PBE functionals.
The LDA atomic radii shown in Fig. \[fig4\] deviate, on average, by 2.26 % from the experimental values [@young91; @radii0], while those calculated in the LAG model and the PBE deviate by $0.83 \%$ and $0.91 \%$, respectively. Among the energy functionals considered in Table \[table3\] the LAG is found to give the lowest mean deviations for both atomic radii and bulk moduli. We note that for these solids the LAG approximation achieves the accuracy of the recently developed meta-GGA [@perdew99; @kurth99].
We have used the Airy gas model of the edge electron gas that is equivalent to the linear potential approximation to develop an exchange energy functional which may serve as an alternative to the functionals based on the generalized gradient appoximation, e.g., PBE GGA. Test calculations for finite and extended systems show that the LAG approximation is more accurate than the local density approximation in all cases. While the LAG results for atoms are very close to the LDA results and, hence inferior to the PBE GGA results, its accuracy for the atomization energies of diatomic molecules is similar to that of the PBE GGA. In bulk systems the LAG results are, on average, closer to the experimental values than those obtained in the PBE GGA. These results are very satisfactory in view of the fact that the LAG exchange functional is derived solely form the properties of the Airy gas, and, hence, with no [*a priory*]{} assumptions concerning the exchange enhancement factor. In this sense it is truely [*ab initio*]{} but for the correlation effects which needs to be worked out in the Airy gas model.
[*Acknowledgments:*]{} We gratefully acknowledge interesting and fruitful discussions with Professor Walter Kohn. L.V. acknowledges the valuable observations by Professor J. P. Perdew and Dr. Á. Nagy. L.V. and B.J. are grateful to the Swedish Natural Science Research Council and the Swedish Foundation for Strategic Research for financial support. Part of this work was supported by the research project OTKA 23390 of the Hungarian Scientific Research Fund. The Center for Atomic-scale Materials Physics is sponsored by the Danish National Research Foundation.
P. Hohenberg, and W. Kohn, Phys. Rev. [**136**]{},B 864 (1964).
W. Kohn and L.J. Sham, Phys. Rev. [**140 A**]{}, 1133 (1965).
J.P. Perdew and Y. Wang, , 8800 (1986); J.P. Perdew, , 8822 (1986).
J.P. Perdew, in [*Electronic Structure of Solids ’91*]{}, edited by P. Ziesche and H. Eschrig, Akademie Verlag, Berlin, (1991).
J.P. Perdew, K. Burke, and M. Ernzerhof, , 3865 (1996).
Y. Zhang and W. Yang, , 890 (1998).
J.P. Perdew, K. Burke, and M. Ernzerhof, , 891 (1998).
C. Lee, W. Yang, and R. G. Parr, , 785 (1988).
A.D. Becke, , 3098 (1988).
F. A. Hamprecht, A. J. Cohen, D. J. Tozer, and N. C. Handy, J. Chem. Phys. [**109**]{}, 6264 (1998).
B. Hammer, L. B. Hansen, and J. K. Norskov, , 7413 (1999).
J.P. Perdew, S. Kurth, A. Zupan, and P. Blahya, , 2544 (1999).
S. Kurth, J. P. Perdew, and P. Blaha, Int. J. Quant. Chem. [**75**]{}, 889 (1999).
A.D. Becke, J. Chem. Phys. [**84**]{}, 4524 (1986).
J.P. Perdew, J.A. Chevary, S.H. Vosko, K.A. Jackson, M.R. Pederson, D.J. Singh, and C. Fiolhais, , 6671 (1992).
L. Vitos, J. Kollár, and H. L. Skriver, , 13521 (1997).
V. Ozilins and M. Körling, , 18304 (1993).
J. Kollár, L. Vitos, and H. L. Skriver, , 15353 (1997).
P. Söderlind, J. A. Moriarty, and J. M. Wills, , 14063 (1996).
J. H. Cho and M. Scheffler, , 10685 (1996).
A. Khein, D. J. Singh, and C. J. Umrigar, , 4105 (1995).
A. D. Corso, A. Pasquarello, A. Baldereschi, and R. Car, , 1180 (1996).
C. Filippi, D. J. Singh, and C. J. Umrigar, , 14947 (1994).
W. Kohn and A. E. Mattsson, , 3487 (1998).
M. Städele, J. A. Majewski, P. Vogl, and A. Görling, , 2089 (1997); M. Städele, M. Moukara, J. A. Majewski, P. Vogl, and A. Görling, , 10031 (1999).
A. Görling, , 5459 (1999).
S. Ivanov, S. Hirata, and R. J. Barlett, , 5455 (1999).
A.D. Becke, J. Chem. Phys. [**98**]{}, 1372, (1993); A.D. Becke, J. Chem. Phys. [**98**]{}, 5648, (1993); A.D. Becke, J. Chem. Phys. [**109**]{}, 8188 (1998).
L. Vitos, B. Johansson, J. Kollár, and H.L. Skriver, , [**61**]{}, 52511 (2000).
N. D. Lang and W. Kohn, , 4555 (1970).
J. Perdew and Y. Wang, , 13244 (1992).
Here we use the second order gradient expansion enhancement factor, $1 + C_xs^2$ with $C_x =0.1234$ which according to Ref. [@dreizler90], pg. 87, should be preferred over the original $C_x=0.0864$. In fact, we find that the original value performes better for both the Airy gas and jellium models.
A. D. Becke, J. Chem. Phys. [**85**]{}, 7184 (1986).
R.M. Dreizler and E.K.U. Gross, [*Density Functional Theory*]{}, Springer-Verlag, Berlin (1990), and references therein.
J. Kollár, L. Vitos, and H. L. Skriver, in “Ab-initio electronic properties of solids”, Lectures Notes in Physics, Springer-Verlag, Berlin, (2000).
O. K. Andersen, O. Jepsen, and G. Krier, in [*Lectures on Methods of Electronic Structure Calculations*]{}, edited by V. Kumar, O. K. Andersen, and A. Mookerjee, World Scientific Publishing Co., Singapore, pp. 63-124 (1994).
L. Vitos, H.L. Skriver, B. Johansson, J. Kollár, Comp. Mat. Sci. XXXX (2000).
D.A. Young, [*Phase Diagrams of the Elements*]{}, University of California Press, Berkeley, (1991)
The radii taken from Ref. [@young91] are reduced to $T=0 K$ using the linear thermal expansion coefficients from: W.B. Pearson, [*A handbook of lattice spacings and structures of metals and alloys*]{}, Pergamon Press, London (1958).
The exact Kohn-Sham jellium surface exchange energies are estimated using the LDA and the exact results from Table II of: A. R. E. Mohammed and V. Sahni, , 4879 (1985).
ATOM GEA PBE LAG KS
------ ------ ------ ----- ------
He 13.9 15.0 4.2 16.0
Ne 6.7 9.4 1.9 9.7
Ar 5.1 7.7 1.4 8.3
Kr 3.5 5.5 0.9 5.9
Xe 2.9 4.7 0.7 5.0
: The effect of GEA [@gea], PBE [@perdew96], and LAG gradient corrections (in percentage) on the LDA atomic exchange energies. All functionals are evaluated from the self-consistent LDA [@perdew92] Kohn-Sham densities. KS denotes the relative difference of the exact and LDA exchange energies from Ref. [@kurth99].
\[table1\]
MOLECULE PBE LAG Expt.
---------- ------- ------- ------- --
Li$_2$ -22.6 -17.8 2.1
Be$_2$ -25.9 -32.8 -76.6
CO -9.8 -8.6 -13.3
N$_2$ -12.2 -10.7 -14.5
NO -14.2 -11.2 -23.0
O$_2$ -16.1 -16.0 -31.1
: The effect of PBE [@perdew96] and LAG gradient corrections (in percentage) on the LDA atomization energies for diatomic molecules. Both functionals are evaluated from the self-consistent LDA [@perdew92] Kohn-Sham densities generated by the FCD-EMTO [@kollar00; @andersen94; @vitos00]. Expt. denotes the relative difference of the experimental and LDA atomization energies from Ref. [@kurth99].
\[table2\]
S$_{LDA}$ S$_{PBE}$ S$_{LAG}$ S$_{PKZB}$ S$_{Expt.}$ B$_{LDA}$ B$_{PBE}$ B$_{LAG}$ B$_{PKZB}$ B$_{Expt.}$
------ ----------- ----------- ----------- ------------ ------------- ------------- ------------ ------------ ------------ -------------
Na 3.769 3.916 3.927 4.019 3.936 8.2 7.6 7.3 7.0 6.9
Al 2.947 2.989 2.977 2.966 2.991 81.2 75.2 76.2 90.5 77.3
Fe 2.565 2.645 2.604 2.627 2.667 253 178 209 198 172
Cu 2.602 2.684 2.656 2.656 2.658 193 137 157 154 138
Pd 2.846 2.916 2.883 2.888 2.873 235 184 203 181 181
W 2.929 2.977 2.953 2.946 2.940 312 292 299 311 310
Pt 2.888 2.943 2.916 2.908 2.892 304 244 268 267 283
Au 2.998 3.081 3.043 3.041 2.997 194 134 156 153 172
Si 3.163 3.198 3.189 3.200 3.182 100 92.8 94.0 93.6 98.8
Ge 3.303 3.384 3.354 3.349 3.318 71.6 61.2 64.0 64.6 76.8
GaAs 3.296 3.375 3.346 3.347 3.312 73.0 62.0 72.1 65.1 74.8
NaCl 3.202 3.346 3.337 3.284 3.306 32.9 23.0 21.7 28.1 24.5
(1.48 %) (1.28 %) (0.80%) (0.88 %) ($17.2 \%$) ($9.2 \%$) ($9.1 \%$) (9.3 %)
: Theoretical equilibrium atomic radii (in $Bohr$) and bulk moduli (in $GPa$) for some selected solids. The present calculations have been performed for crystallographic $\alpha$ phases using the FCD-EMTO method [@kollar00; @andersen94; @vitos00]. The results obtained by the meta-GGA of Perdew, Kurth, Zupan and Blaha (PKZB) and the experimental values are from Ref. [@kurth99]. The mean absolute values of the relative errors are shown in parenthesis.
\[table3\]
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We present striking results from [*Rossi X-ray Timing Explorer (RXTE)*]{} observations of the 21-hr low mass X-ray binary , showing five complex dips in unprecedented detail. For the first time, dipping is detected up to 15 keV. Prominent flares are also observed in the light curves, limited to energies above $\sim$8 keV. Spectra selected by intensity during dip episodes can be well fit with a two-component model consisting of a point-like blackbody from the neutron star and progressive covering of an extended Comptonized region, presumably an accretion disk corona (ADC), corrected for photons scattered into and out of the X-ray beam by a interstellar dust halo. We find that the outer regions of the absorber are highly ionized and that electron scattering is totally responsible for the X-ray attenuation during shallow dipping. The timescales of dip ingress and egress indicate that the envelope of material absorbing the ADC has smaller angular size than the ADC itself, and that the ADC is likely limited to a height-to-radius ratio of 10%, rather than being spherical in extent. In addition, we have analyzed $\sim$4.5 yrs of [*RXTE*]{} All Sky Monitor (ASM) coverage to derive the first accurate orbital ephemeris for , with phase zero (the time of dip centers) well-described by the relation 245 0088.63918(69) $+$ N $\times$ 0.869907(12) (JD).'
author:
- 'Alan P. Smale'
- 'Michael J. Church, Monika Bałucińska-Church'
title: The Ephemeris and Dipping Spectral Behavior of
---
Introduction
============
The study of X-ray dipping sources has played a critical role in the understanding and visualization of how a low mass X-ray binary (LMXB) might look to a nearby astronautical observer; it is generally accepted that the dips are due to occultations of the central source by a thickened region of the accretion disk rim where the gas stream from the companion impacts upon the outer disk. However, the dips themselves also provide a wealth of physical information not limited to their role in defining the system geometry. From their irregular, ragged appearance and cycle-to-cycle differences in width and depth we know that the disk structure is dynamic. Also, spectral fitting results show that dips are associated with, but cannot be completely described by, increases in absorption of $N_H$ $\approxgt$ 10$^{23}$ cm$^{-2}$.
is one of the most unusual members of this class; its persistent emission is the brightest (6$\times$10$^{37}$ erg s$^{-1}$), its dip profiles seemingly the most erratic, and its 21-hr orbital period is the longest – $\sim$5–25 times longer than other dipping sources, corresponding to a much greater stellar separation and a larger accretion disk radius. Dipping is deep, $\sim$ 75% in the 1–10 keV band, and can at times reach a stable lower level (Watson et al. 1985; Church & Bałucińska-Church 1995) suggesting that one spectral component is entirely removed. The source also exhibits strong flaring in which the X-ray flux can increase by 30% or more over timescales of a few thousand seconds. In this regard, it may provide an interesting overlap between the dipping LMXBs, generally assumed to be atoll sources by virtue of their persistent emission level and noise characteristics, and the Z-track sources with their well-known flaring branches (e.g. Hasinger & van der Klis 1989).
The depth, duration, and spectral evolution in dipping vary considerably between members of the dipping class. An additional complication is the fact that several sources exhibit an unabsorbed component of the spectrum, e.g. XB1916$-$053 (Smale et al.1988, 1992), XB0748$-$676 (Parmar et al. 1986) and XB1254$-$690 (Courvoisier et al. 1986). More recently, it has been demonstrated that a unified approach describes the dipping sources (Church & Bałucińska-Church 1995; Church et al. 1997, Bałucińska-Church et al. 1999). Their model consists of point-like blackbody emission from the neutron star, plus extended Comptonized emission from the Accretion Disk Corona (ADC), and a “progressive covering” description of absorption. In this description, the absorber on the outer edge of the accretion disk (White & Swank 1982) moves progressively across the emission regions, so that at any stage of dipping, part of the Comptonized emission is absorbed and part is not, giving rise to the observed unabsorbed component. As a point source, the blackbody is rapidly absorbed once the envelope of the absorber reaches the neutron star.
In previous observations of with [*EXOSAT*]{} and [*Ginga*]{}, a blackbody plus bremsstrahlung model was used to parameterize the spectra (Jones & Watson 1989). Subsequently, Church & Bałucińska-Church (1995) showed that the light curve at energies $>$5 keV was dominated by flaring, which can strongly modify the spectrum and make the spectral investigation of dipping difficult. By selecting sections of non-dip and dip data without apparent flaring, the above two-component model was found to fit the data well, showing that in deep dipping the blackbody component was totally absorbed, and the Comptonized component relatively little absorbed. However, with only one non-dip and one deep dip spectrum it was not possible to determine the extent of absorption of the Comptonized component.
A more recent [*BeppoSAX*]{} observation revealed the dust scattered halo of the source demonstrated by Angelini et al. (1997). Modeling of the radial distribution of count rate in the MECS at a series of energies below 5 keV allowed an optical depth to dust scattering of $\rm {2.4\pm 0.4}$ at 1 keV to be derived, the data being consistent with a cross-section varying as $E^{-2}$ (Bałucińska-Church et al. 2000a). Spectral models of non-dip and dip data included the effects of this interstellar dust scattering, and good fits were obtained with the above two-component model, in which the blackbody had temperature $kT$ = 1.31$\pm$0.07 keV, and the Comptonized emission was approximated by a cut-off power law with photon index of $\Gamma$ = 2.0$^{+0.5}_{-0.7}$ and cut-off energy $\sim$12 keV. While more sensitive than previous studies, the [*BeppoSAX*]{} observation also included only a single dip episode.
Previous observations of have been limited either by the short duration of the observations or by the signal-to-noise ratio of the data; a thorough investigation requires a baseline of several orbits, coupled with a collecting area and efficiency large enough to study the spectral variability on short timescales. In this paper, we present light curves obtained with [*RXTE*]{} covering 4.5 contiguous orbits of , during which both strong dipping and strong flaring took place. We derive an accurate system ephemeris, and perform a detailed study of the spectral variability in dips. A companion paper will contain our analysis of the evolution and physics of the flares (Bałucińska-Church et al. 2000b, hereafter Paper II).
[c]{}
Observations
============
We observed with [*RXTE*]{} (Bradt, Rothschild, & Swank, 1993) from 1999 September 27 18:22 UT – October 1 13:11 UT, for an on-source total good time of 212 ksec. The X-ray data presented here were obtained using the PCA instrument with the Standard 2 and Good Xenon configurations, with time resolutions of 16 sec and $<1\mu$sec respectively. The PCA consists of five Xe proportional counter units (PCUs) numbered PCUs 0 through 4, with a combined total effective area of about 6500 cm$^2$ (Jahoda et al 1996). Only PCUs 0 and 2 were reliably on throughout our observations, and we limit our analysis to data from these detectors. Data were also obtained with the HEXTE phoswich detectors, which are sensitive over the energy range 15–250 keV (Rothschild et al. 1998). However, because of the cut-off spectral shape of , the counts detected were limited to energies $<$25 keV, and the HEXTE data did not in this case provide any information additional to that gained using the PCA.
Based on a preliminary determination of the source ephemeris, derived from the above observations and the [*RXTE*]{} All Sky Monitor (ASM) light curve, we scheduled the balance of our [*RXTE*]{} pointed observing time to capture a complete dip episode, uninterrupted by Earth occultations and South Atlantic Anomaly passages. These observations were performed on 1999 November 12:20–16:21 UT, for a total good time of 14.3 ksec. For this observation, PCUs 0, 2, and 3 were reliably on throughout.
[c]{}
We performed our data analysis using the [*RXTE*]{} standard analysis software, FTOOLS 5.0. Background subtraction of the PCA data was performed utilizing the “skyvle/skyactiv 20000131” models generated by the [*RXTE*]{} PCA team. The quality of the background subtraction was checked in two ways: (i) by comparing the source and background spectra and light curves at high energies (50–100 keV) where itself no longer contributes detectable events; and (ii) by using the same models to background-subtract the data obtained during slews to and from the source. We conclude that our background subtractions in the 2–20 keV energy range are accurate to a fraction of a count per second. Since one of our goals was a sensitive spectral analysis, we took pains to check the quality of both the background subtraction and the response matrices used, including a careful analysis of Crab calibration observations on 1999 September 26, October 13, November 8, and November 23, bracketing our observations of . Based on this work, we added 2% systematic errors to the spectral fits.
Our orbital ephemeris calculations made use of two standard tools for establishing periodicities: a discrete Fourier transform technique optimized for unevenly-sampled time series (Scargle 1989), and a period-folding code utilizing the L-statistic (Davies 1990). Errors were calculated using the formulas intrinsic to these methods, and corroborated by fitting periodic functions to the unfolded light curve data and calculating the formal errors on the orbital period and epoch directly.
The search for fast variability in utilized the high time resolution PCA data obtained in the Good Xenon mode. Discrete power spectral density distributions (PSDs) were calculated by dividing the data into segments of uniform length, performing fast Fourier transforms of each, and averaging the results. The PSDs were normalized such that their integral gives the squared RMS fractional variability (Miyamoto et al 1991; van der Klis 1989). We subtracted the Poisson noise level from the power spectra, taking into account the modifications expected from PCA detector deadtime.
Results
=======
Overall source behavior
-----------------------
In Figure 1, we show the background-subtracted 2–20 keV light curve for the 1999 September observation. Superimposed on the mean persistent flux levels, the most prominent features of the observation are four regular dips separated from each other by $\sim$21 hrs, representing the orbital period of the system, and strong irregular flaring, most evident at phases $\phi$=0.30–0.75 (assigning $\phi$=0 to the dip center). The 1999 November observation is shown in Figure 2, and provides a good illustration of the prodigious amount of detail in an individual dip from this source. The complex phenomenology seen in the overall light curve becomes more amenable to examination when we divide the 1999 September data into four energy bands covering 2–6, 6–10, 10–15, and 15–20 keV (Figure 3). (This is, incidentally, the first time the variability of the source above 10 keV has been accessible with good signal to noise.) From this Figure, we can come to the following qualitative conclusions:
$\bullet$ Dips are deepest at lower energies, but are still (barely) discernable above 15 keV.
$\bullet$ The flaring is a higher-energy effect, not visible in the 2–6 keV range, but becoming the dominant feature in the 10–15 keV and 15–20 keV light curves. Examining light curves with even finer energy resolution, we can determine that flaring first becomes significant above 8 keV.
[c]{}
The ephemeris of
-----------------
Applying the period search algorithms to the 1999 September PCA observation, we find a best-fitting period of 20.98$\pm$0.14 hrs. This is consistent with previous estimates from datasets of limited duration, but for a useful long-term ephemeris, we turn to the publicly-available data from the [*RXTE*]{} ASM. The ASM (Levine et al. 1996) consists of three similar Scanning Shadow Cameras, sensitive to X-rays in the 2–12 keV energy band, that cover $\sim$80% of the sky every 90 minutes, with each datapoint representing a flux measurement from a single 90s dwell. is detected in the ASM with a weighted mean net count rate of 3.61$\pm$0.01 count sec$^{-1}$. (For comparison, the Crab Nebula produces a count rate of $\sim$75 ASM count sec$^{-1}$.) For this work, we analyzed data spanning the 1616 days from 1996 January 5 to 2000 June 8. The 21-hr orbital periodicity shows up strongly even in a rudimentary Fourier transform of the data at a frequency of (1.3306$\pm$0.0002)$\times$10$^{-5}$ Hz. Combining Fourier and period folding analysis, we derive an orbital ephemeris for of
245 0088.63918(69) $+$ N$\times$0.869907(12) (JD).
This orbital period (20.8778$\pm$0.0003 hrs) is a two-significant-figure improvement over previous derivations, with a predictive value limited only by the phase jitter and irregularity of individual dips. It is sufficiently accurate to extend back to e.g.the EXOSAT observations of obtained in 1985 March 25-27, with no cycle-count ambiguity.
A QPO search
------------
Recently, a new type of quasi-periodic oscillation (QPO) has been discovered in three of the dipping sources. The feature was first seen in X1323$-$619, where a $\sim$1 Hz QPO was detected with an rms amplitude of 9%, remaining roughly constant in the persistent emission, the dips, and during the bursts (Jonker, van der Klis & Wijnands 1999). Similar features have been observed in X0748$-$676, with a central frequency $\nu$ that varied between 0.6 and 2.4 Hz and rms amplitudes of 8–12%, and in X1746$-$370, with $\nu$=1.0–1.6 Hz and rms amplitude 7% (Homan et al. 1999; Jonker et al. 2000). The QPO in X0748$-$676 are present in all the observations analyzed by Homan et al. (1999) except the pointing with the highest countrate. For X1746$-$370 the picture is even more clear-cut; the atoll-source “island” and “banana” (Hasinger & van der Klis 1989) branches are well defined by the data, and the QPOs are absent during the latter.
For all three sources, there is little or no dependence of the QPO strength on the photon energy. This is very unusual: the amplitude of all other types of QPO seen in $Z$ sources, atoll sources, and black hole candidates (BHC) from 0.01–1200 Hz shows a strong dependence on photon energy. Clearly the dipper QPO are a different phenomenon, related to the high source inclination, and perhaps linked directly to the modulating effect of material in the accretion stream or at the disk edge; the exact mechanism is still open to conjecture. As this QPO is not observed in other LMXBs at lower inclinations, variations in the mass accretion [*rate*]{} are unlikely to be directly responsible, although the disappearance of the QPO at higher inferred mass accretion rates may be due to changes in the accretion [*geometry*]{}. The most likely candidate is some modulating effect caused by the accretion stream, or partial covering of an extended X-ray source by a near-opaque medium; this structure may disappear, change size, or change in optical depth at higher accretion rates.
We have searched the data for evidence of similar QPO activity, and find none. The extensive data set at our disposal allows us to place a rather stringent upper limit of 1% on the rms amplitude of a feature between 0.5–2 Hz with a FWHM of 0.4 Hz.
The absence of such features in is perhaps not surprising; 1 Hz QPOs are observed in XB1323$-$62 and XB1746$-$37 in the low-intensity island state of these atoll sources, but are not visible in the higher-intensity banana state. The derived hardness-intensity and color-color diagrams for the current observations of are dominated by the dip and flare activity and give no direct indication of which of the two states the source may be in, however the high overall source luminosity (see below) implies a high accretion rate, perhaps incompatible with the production of such QPO.
Spectral evolution during dipping
---------------------------------
The energy bands in Figure 3 were chosen to emphasise and isolate the dipping activity, which dominates the light curve in the lowest band, and flaring, which dominates in the two bands above 10 keV. From Figures 2 and 3, it can be seen that there is a high degree of fast variability in dipping on timescales of $\sim $32 s, and that there is a tendency for dipping to saturate at a constant low level. Moreover, the 2–6 keV light curve shows that dipping consists of two stages: deep dipping with associated rapid variability, preceded and followed by relatively shallow shoulders to the dipping which last $\sim
$12,000 s. Flaring, although apparently random in occurrence, is generally not observed during dip episodes at the high level seen outside dips. However, there appears to be a section of flaring data at $\rm {\sim 1.4\times 10^5}$ s, where the flaring coincides with the weak shoulder of the second dip.
We chose to study the spectral evolution through the dip episodes in the 1999 September observation in two stages. First, we accumulated spectra from a limited subset of the data, ruthlessly eliminating any data sections with even a suggestion of flare contamination. This approach produced 4 spectra, one for the non-dip emission, two at intermediate dipping levels, and a saturated dip spectrum. We fit these simultaneously using a wide range of models to allow elimination of most of these, and determine the probable form of the best-fit model. In the second stage, we divided the bulk of the observation into intensity bands and fit these jointly using the model determined from the first stage. This approach gave us confidence in our data selection criteria and enabled us to perform a sensitive search for subtle features.
### Stage 1: Simultaneous fits to 4 spectra
Because of the strong spectral changes associated with flares, it was essential to avoid any possibility of data contamination with flare activity in our first pass. We thus selected data from the region around $\sim1.7\times 10^5$ s (in Figure 1), consisting of a complete satellite orbit of data during which no flaring is detected. A deep dip spectrum was obtained from that part of the third dip which reached a saturated, stable lower level in the low-energy light curve. All traces of dip ingress and egress were removed by selecting 320 seconds of data beginning at $\rm {2.1678\times 10^5}$ s from the start of the observation. Other dip spectra were selected in intensity bands at 180–190 count s$^{-1}$ and 130–140 count s$^{-1}$. The total source counts contained in the 4 spectra selected, from non-dip to the deepest dipping respectively, were $\sim
$7.4$\times$10$^5$, 3.5$\times$10$^4$, 4.4$\times$10$^4$, and 2.2$\times$10$^4$.
These four spectra were first fitted simultaneously using simple, one-component models. In all cases, the parameters specifying the emission process such as the power law photon index were chained between the four spectra, since these cannot change during dipping. Results for all model fitting are shown in Table 1. (All values of the hydrogen column density quoted in the tables are in units of 10$^{22}$ atoms cm$^{-2}$, and errors are quoted to 90% confidence.)
Simple models were unable to fit the spectra: an absorbed cut-off power law gave a $\chi^2$ per degree of freedom $(dof)$ of 8104/188; an absorbed bremsstrahlung model gave a $\chi^2/dof$ of 8789/189 and an absorbed blackbody model gave a $\chi^2/dof$ of 12666/189. Next, the two-component model consisting of a blackbody plus cut-off power law was tried, first with the non-dip spectrum alone. A good fit was obtained with $\chi^2/dof$ = 20/46, and so this model was applied to simultaneous fitting of all four spectra, producing an overall fit statistic of $\chi^2/dof$ = 231/180. The two-component model has the form: AG(AB.BB $+$ PCF.CPL), where AG represents Galactic absorption, BB the blackbody component, CPL the cut-off power law, AB additional absorption during dipping and PCF the progressive covering factor. This is the model shown to be able to describe spectral evolution in dipping in other LMXB dipping sources (see Introduction), and in this model the point-source blackbody is understood to be absorbed rapidly when the envelope of the absorber covers the neutron star, whereas the extended Comptonized emission is covered progressively as the absorber moves across the extended source.
Strong broad, residuals at $\sim$6.4 keV were found, at the level of 4% in the non-dip spectrum, too high to be due to residual uncertainty in the instrument response function, and so a Gaussian line was included to the above model as a source term. Because of a tendency for widths of broad lines to increase in spectral fitting by absorption of continuum emission into the line, it was necessary to fix the width, and tests showed that 0.4 keV was appropriate. Fitting this model to 4 spectra simultaneously required the normalization of the line to decrease systematically in dipping, from 2.59$\times$10$^{-3}$ photon cm$^{-2}$ s$^{-1}$ in non-dip to 5.17$\times$10$^{-4}$ photon cm$^{-2}$ s$^{-1}$ in deep dipping. Fitting the deep dip with the latter normalization gave $\chi^2/dof$ = 38/45, whereas fixing the normalization at the non-dip value gave $\chi^2/dof$ = 130/46, demonstrating the significance of the decrease. Thus, we included the line within the progressive covering term to give a model: AG(AB.BB + PCF(CPL + GAU)). Using this approach, no change in the line normalization was necessary, showing that the line was well modeled by assuming that it originates in the same region as the Comptonized emission, i.e. the ADC, and so is subject to the same covering factor.
At this stage, the fits still exhibited excesses at low energy, resulting in a quality of fit that was still not acceptable. Consequently, we next added the same interstellar dust scattering terms to the spectral model that were used in the [*BeppoSAX*]{} study (see Bałucińska-Church et al. 2000a, and above), including terms both for scattering out of the beam and into the beam, giving the final form of the fitted model:
$\rm {AG\;e^{-\tau}(\;AB\; .\; BB + PCF\;(\;CPL + GAU\; ))\;+}$$\rm {AG\;(\; 1\; -\;e^{-\tau})(\;BB\; +\; CPL\; +\; GAU\;)}$.
The factor $\rm {e^{-\tau}}$ represents scattering out of the beam, while the $\rm {(1\;-\; e^{-\tau})}$ term is scattering into the beam. For non-dip emission, we make the usual assumption that the intensities of these components are equal (e.g. Martin 1970). (Non-standard spectral components $\rm
{e^{-\tau}}$ and $\rm {(1\,-\,e^{-\tau})}$ were produced for inclusion in spectral fitting with the XSPEC package.) With the addition of the dust scattering terms, the low energy excesses in the spectra were removed and a good fit obtained to all 4 spectra simultaneously with a $\chi^2/dof$ = 127/179.
While our fitted model may appear over-constrained at first sight, it is crucial to note that in our simultaneous fits, [*all*]{} emission terms (blackbody $kT$ and normalization, $\Gamma$, cutoff and normalization values) are fixed at their non-dip values, and the dust scattering parameters are fixed at their SAX MEC values. Only the column densities and covering fraction are free parameters, in a joint fit to 4 spectra of differing shapes and over a fivefold range of intensities.
[c]{}
In addition, to demonstrate that the inclusion of the line and dust scattering do not favor the two-component model over the other candidate models, the same terms were added to the one-component models fits in the most favorable way, i.e. with the line being subject to progressive covering to account for its variation in intensity during dipping. The results of fitting all models are included in Table 1. It can be seen that adding the line and dust scattering to simple models results in some improvement, but the fits are still unacceptable. Although the overall fit of the best-fit model to the 4 spectra was acceptable, the spectra (non-dip to deepest dip) having individual values of $\chi^2/dof$ of 20/46, 74/46, 60/46 and 38/46, the intermediate dip spectra had broad excesses at $\sim $5 keV. This was investigated further in our next step, which was to analyze spectra from the whole observation, with much improved statistics at the intermediate levels.
### Stage 2: Spectra from the complete observation
Spectra were selected in intensity bands 20 count s$^{-1}$ wide, with a non-dip spectrum taken from the band 300-320 count s$^{-1}$. Five dip spectra were accumulated corresponding to intensity bands: 280–300, 260–280, 220–240, 100–120 and 60–80 count s$^{-1}$, such that the non-dip spectrum plus the first two of these provide good coverage of the shoulders of dipping, and the remaining spectra span the main, deeper parts of dips. We excluded a section of data 60 ksec in length at the beginning of the observation, in which the 15–20 keV light curve (Figure 3) shows unusual behavior. The six spectra thus derived have a considerably higher total count rate than the 4 spectra previously used, allowing the relatively small intensity changes in the shoulders of dipping to be investigated. We applied our best fit model first to these shoulders.
In Figure 4, the unfolded spectrum of the non-dip level, the deepest spectrum in the shoulders of dipping, and the deepest dip spectrum of all are shown. The dip shoulder spectrum can be approximately fitted as a constant vertical shift downwards of the non-dip spectrum, indicating that electron scattering is the dominant absorption process, and that photoelectric absorption is negligible. From the fitting it is also clear that the blackbody requires zero column density, for all spectra in the shoulders of dipping. Thus, the intensity reduction in the shoulders must correspond to the absorber overlapping the extended ADC but not the neutron star, and the outer regions of the absorber which are involved in this overlap must be highly ionized. Comparison of the non-dip and deepest-dip spectra also reveals a vertical shift between the spectra at 20 keV showing the effects of electron scattering. This effect, clear in intensity-selected data with good statistics, was not visible in Stage 1. Consequently, it becomes necessary to upgrade the PCF model previously used to a progressive covering model in which X-rays are attenuated by two processes: electron scattering and possible photoelectric absorption expressed in the form:
$$(f\, e^{-N_e \sigma_T} e^{-N_H \sigma_{PE}} \,+\, (1-f))$$
where $N_e$ is the electron column density, $\sigma_T$ is the Thomson cross-section, and $\sigma_{PE}$ the photoelectric cross-section, and the contributions of scattering and absorption are allowed to be independent.
[c]{}
[c]{}
Good fits were obtained with this model, as shown in Figures 5 and 6. The emission parameters used are shown in Table 2, and we present the parameters of the best fits to the complete set of intensity-selected spectra in Table 3. The values of the emission parameters used for the complete set agree well within the errors with the values shown for the best-fit model of Table 1.
Initially the fit to the intermediate dip spectrum at 220–240 c s$^{-1}$ was relatively poor with an excess seen in the spectrum at $\sim $5 keV and $\chi ^2/dof$ = 145/45. This spectrum was investigated further. A concern throughout this work was the possibility of dip spectra being contaminated by flaring, given the strong spectral changes we show take place in flaring in Paper II. Deep dip spectra will not be contaminated as flaring is associated with the neutron star, but we cannot rule out the possibility for the intermediate spectra may be. However, it appears that this excess has another cause. If the absorber is highly ionized, residual photoelectric absorption will be due to the ionized absorber, not neutral absorption. In the early stages of dipping, this is not a problem since attenuation of the Comptonized ADC emission is caused by the large electron column density $N_e^{CPL}$ whereas the absorption column $N_H^{CPL}$ remains zero (Table 3) until dipping becomes deep. In deep dipping, a combination of neutral absorber and electron scattering provides satisfactory fits, as could be expected since the degree of ionization must decrease inside the absorber. Similarly, in shallow dipping the blackbody is not absorbed, and in deep dipping totally absorbed, so the question of ionized absorber does not arise. The exception is the intermediate spectrum with $N_H^{BB}$ = 20$\times$10$^{22}$ atom cm$^{-2}$ (Table 3). Refitting this spectrum with an ionized absorber for the blackbody resulted in a much-improved fit with $\chi^2/dof$ = 53/44 (as shown in the Table), and a value of $\xi$=118 for the ionization parameter. Strictly speaking, even this model is not totally satisfactory since the ‘[*absori*]{}’ model of XSPEC that was used is based on a power law source whereas is dominated by blackbody emission. However, we believe we are now at the limit of what can usefully be learned through spectral fitting, and further refinements will produce diminishing returns.
Discussion
==========
We have shown that spectral evolution during dipping in is well-described by a combination of the point-source blackbody plus extended Comptonization emission model, with a “progressive covering” treatment of the absorption process. To this basic model it was necessary to add a broad Gaussian line at $\sim $6.4 keV, and terms to describe interstellar dust scattering both out of and into the line-of-sight. The blackbody temperature $kT$, and power law index $\Gamma$, and cut-off energy $E_{CO}$ from these [*RXTE*]{} fits (Table 2) are in good agreement with the values determined from the [*BeppoSAX*]{} fitting, of $kT$ = $\rm {1.37\pm 0.07}$ keV, $\Gamma $ = $\rm {2.0^{+0.5}_{-0.7}}$ and $E_{CO}$ $\sim $12 keV (Bałucińska-Church et al. 2000a). During the [*RXTE*]{} observations, the total luminosity in the 1–30 keV band was $\rm {1.47\times
10^{38}}$ erg s$^{-1}$ at a distance of 15 kpc (Christian & Swank, 1997), approximately twice that seen during the [*BeppoSAX*]{} observation, and thus complete consistency of emission parameters should not necessarily be expected. The bolometric blackbody luminosity was $\rm {2.3\times 10^{37}}$ erg s$^{-1}$, or 15% of the total luminosity, smaller than the 36% measured by [*BeppoSAX*]{} (Bałucińska-Church et al. 2000a). Luminosity considerations are discussed further in Paper II.
In dipping, the blackbody component is rapidly and completely removed once the neutron star is covered. The clear occurrence of saturated dipping in most of the dips observed indicates that in addition to the removal of the blackbody, absorption of the Comptonized emission also reaches a stable level. This must be due either to due to an absorber of smaller angular extent than the ADC by a fixed fraction, or to the presence of blobs within the absorber envelope such that $\sim $20% of the absorber is transmitting in less dense regions in between blobs of higher density. The Gaussian line varied in dipping in a way that could be described by giving it the same covering factor as the ADC. There are two contributions to the excess at low energies: the dust-scattered halo components which are shown separately in Figure 6, and also the part of the extended Comptonized emission that is not covered at any level of dipping. Table 3 shows the covering fraction rising to a maximum value of $\sim $80% in the best-fit model, unlike other dipping sources such as XB1916$-$053 (Church et al. 1997) where dipping reaches a depth of 100% at all energies below 10 keV and the covering fraction similarly reaches 100%, proving that the angular extent of all emission regions is less than that of the absorber. In the present case, there may be two reasons why this does not happen: either the absorber is somewhat smaller in angular extent than the ADC extended emission, or more likely, the absorber in the outer accretion disk is not uniformly dense within its envelope, but blobby, as indicated by the strong variability within dipping. We investigate this possibility further below.
We can obtain further information on the ADC from the duration of dip ingress and egress. The sharp ingress to deep dipping, and the fast variability within dipping, take place on timescales of $\sim $32 s, and this is clearly associated with absorption of the point-source blackbody component. However, the long, shallow shoulders of the dips map a gradual process of dip ingress and egress which is consistent with the extended ADC being covered by relatively less dense absorber before the absorber envelope reaches and obscures the neutron star. The duration of the ingress shoulder is $\rm {\sim 12\pm
2}$ ks from the present observation, and we can also derive a duration of $\rm {\sim 13\pm 2}$ ks from studying the EXOSAT light curve (Church & Bałucińska-Church 1995). Some asymmetry can be seen in the shoulders, and we use the shorter ingress time corresponding to the leading edge of the bulge on the accretion disk, not the trailing edge, which would give a longer shoulder depending on the absorber dimensions rather than the dimensions of the ADC. For an absorber of larger angular extent than the ADC, the ingress time $\Delta t$ is the time taken by the leading edge to cross the ADC given by the velocity of material in the outer disk: $$\rm {2\,\pi\, r_{disk}/P \;=\;
d_{ADC}/\Delta t,}$$
where $\rm {r_{disk}}$ is the disk radius, P is the orbital period and $\rm {d_{ADC}}$ is the diameter of the ADC. In the event that the absorber has somewhat smaller angular extent than the ADC, as is a possible interpretation of spectral fitting, the ingress time gives the diameter of the absorber, and the source region could be as much as 20% larger. Using a period of 20.87 hr from our ephemeris work above, and a mass of $\rm {1.4 M_{\sun}}$, a value of $\rm {1.0\times
10^{11}}$ cm can be calculated for the accretion disk radius (Frank et al. 1987). Using our observed $\Delta t$, we derive from this an ADC radius of $\rm {5.0\times 10^{10}}$ cm. The average ingress time from the EXOSAT and [*RXTE*]{} data gives $\rm {5.3\times 10^{10}}$ cm. If the absorber has smaller angular size than the ADC, then the above calculation provides the absorber radius, so that the ADC could be up to 20% larger than this, on the basis of the covering fraction only reaching $\sim $80% in deepest dipping.
We can use the above radius of the ADC to make some comparisons with simple ADC theory. The maximum radius of an ADC that can be supported in hydrostatic equilibrium $\rm {r_{eq}}$ is given by the condition that kT $<$ $\rm {GM_xm/r_{eq}}$ where $\rm {M_x}$ is the mass of the neutron star and m the mass of a proton. This provides a simple formula: $$\rm {r_{eq} \simeq {1.6\times 10^{11}\; M_x\over T_{ADC}\;
M_{\sun}}\;\; (cm) }$$ where $\rm {T_{ADC}}$ is the temperature of the ADC in units of 10$^7$ K. We can use the Comptonization cut-off energy $E_{CO}$ obtained in the present analysis of 12 keV to place limits on the electron temperature $kT_e$ of 4–12 keV, corresponding to an optical depth to electron scattering $\tau <$1 or $\tau >$1 respectively. For low optical depth, $kT_e \sim E_{CO}$, but for higher optical depth, even if Comptonization is not saturated, $kT_e$ will be a factor of 2.5–3.0 smaller than $E_{CO}$ (e.g. Dove et al. 1997). Using these limits on electron temperature, we find $\rm {r_{eq}}$ = $\rm {1.6-5.3\times
10^{10}}$ cm, the smaller value corresponding to the higher temperature. Comparing these with $\rm {r_{ADC}}$ $\sim $ $\rm
{5\times 10^{10}}$ cm from our dip ingress calculation, we can see that there is better agreement with $\rm {r_{eq}}$ calculated assuming a higher optical depth to electron scattering. We thus have evidence that the ADC is maintained in hydrostatic equilibrium. The ADC is likely to be “thin”, with a height-to-radius ratio of the order of 10% at its outer edge, as opposed to being spherical. For example, for an ADC radius of $\rm {5\times 10^{10}}$ cm, it is unlikely that the absorber on the outer disk would be able to cover a spherical ADC extending to this distance in the vertical direction.
This paper utilizes [*RXTE*]{} All Sky Monitor results made publicly available by the ASM/[*RXTE*]{} Team, including members at MIT and NASA/Goddard Space Flight Center, and also archival data obtained through the High Energy Astrophysics Science Archive Research Center Online Service, also at NASA/GSFC.
Angelini, L., Parmar, A.N., White, N.E., 1997, Proc. IAU Colloquium 163, Eds. D.T. Wickramasinghe, G.V. Bicknell, L. Ferrario
Bałucińska-Church, M., Church, M.J., Oosterbroek, T., Segreto, A., Morley, R., & Parmar, A.N., 1999, A&A, 349, 495 Bałucinska-Church, M., Humphrey, P.J., Church, M.J, & Parmar, A.N., 2000a, A&A in press Bałucinska-Church, M., Barnard, R., Church, M.J, & Smale, A.P., 2000b, ApJ in preparation.
Bradt, H.V., Rothschild, R.E., & Swank, J.H., 1993, A&AS, 97, 355
Christian, D.J., & Swank, J.H., 1997, ApJS, 109, 177 Church, M.J., & Bałucińska-Church, M., 1995, A&A, 300, 441 Church, M.J., Dotani, T., Bałucińska-Church, M., Mitsuda, K., Takahashi, T., Inoue, H., & Yoshida, K., 1997, ApJ, 491, 388 Courvoisier, T.J.-L., Parmar, A.N., Peacock, A., & Pakull, M., 1986, ApJ, 309, 265 Davies, S.R., 1990, MNRAS, 244, 93
Dove J. B., Wilms J., Maisack M., & Begelman M. C., 1997, ApJ, 487, 759
Frank, J., King, A.R., & Lasota, J.-P., 1987, A&A, 178, 137
Hasinger, G., & van der Klis, M., 1989, A&A, 225, 79 Homan, J., Jonker, P.G., Wijnands, R., van der Klis, M., & van Paradijs, J., 1999, ApJ, 516, L91
Jahoda, K., Swank, J. H., Giles, A. B., Stark, M. J., Strohmayer, T., Zhang, W., & Morgan, E. H., 1996, in EUV, X-ray and Gamma-Ray Instrumentation for Astronomy VII, ed O. H. Siegmund (Bellingham, WA: SPIE), 59
Jones M.H., & Watson M.G., 1989, Proc. 23rd ESLAB Symposium, ESA SP-296, p. 439 Jonker, P.G., van der Klis, M., Homan, J., Wijnands, R., van Paradijs, J., Méndez, M., Kuulkers, E., & Ford, Eric C., 2000, ApJ, 531, 453 Jonker, P.G., van der Klis, M., & Wijnands, R., 1999, ApJ, 511, L41
Levine, A.M., Bradt, H., Cui, W., Jernigan, J.G., Morgan, E.H., Remillard, R., Shirey, R.E., & Smith, D.A., 1996, ApJ, 469, L33
Martin, P.G., 1970, MNRAS 149, 221
Miyamoto, S., Kimura, K., Kitamoto, S., Dotani, T., & Ebisawa, K., 1991, ApJ, 383, 784
Parmar, A.N., White, N.E., Giommi, P., & Gottwald, M., 1986, ApJ, 308, 199 Rothschild, R. E., Blanco, P. R., Gruber, D. E., Heindl, W. A., MacDonald, D. R., Marsden, D. C., Pelling, M. R., & Wayne, L. R., 1998, ApJ, 496, 538
Scargle, J.D., 1989, ApJ, 343, 874
Smale, A.P., Mason, K.O., White, N.E., & Gottwald, M., 1988, MNRAS, 232, 647
Smale, A.P., Mukai, K., Williams, O.R., Jones, M.H. & Corbet, R.H.D., 1992, ApJ, 400, 330 van der Klis, M., 1989, in Timing Neutron Stars, H. Ogelman & E.P.J. van den Heuvel (eds), Cambridge University Press, p. 252
Watson, M.G., Willingale, R., King, A.R., Grindlay, J.E., & Halpern, J., 1985, Space Sci Rev., 40, 195 White, N.E., & Swank, J.H., 1982, ApJ, 253, L61
[lllll]{} Blackbody & 0.0$^{+0.1}_{-0.0}$ & 1.57$\pm$0.01 & …& 12666/189Bremsstrahlung & 4.48$\pm$0.26 & 4.79$\pm$0.60 & $\dots$ & 8789/189Cut-off power law & 12.5$^{+1.1}_{-0.3}$ & $\dots$ & 3.31$^{+0.19}_{-0.06}$ & 8104/188 Two-cpt (prog. cov.) & 11.6$\pm$0.7 & 1.55$\pm$0.04 &3.00$\pm$0.14 & 231/180 & & & &
Blackbody, line, dust & 0.0$^{+0.1}_{-0.0}$ & 1.56$\pm$0.01 & … & 6677/182 Bremsstrahlung, line, dust & 6.0$\pm$0.1 & 4.46$\pm$0.10 & …& 3433/182 Cut-off power law, line, dust & 12.5$\pm$0.1 & …&2.99$\pm$0.12 & 2810/181 Two-cpt, line, dust (prog. cov.) & 10.1$\pm$0.8 &1.52$\pm$0.05 & 2.74$\pm$0.17 & 127/179
[lrlll]{}
8.6$\pm$3.1 & 1.32$\pm$0.17 &2.29$^{+0.66}_{-1.10}$ & $\sim$12 & 6.4$^{+0.19}_{-0.49}$
[rlllcr]{}
Non-dip &8.6$\pm$3.1 & 8.6$\pm$3.1 &0.0 &0.0&54/45280–300 c s$^{-1}$ & 8.6$\pm$3.1&8.6$^{+77}_{-3.1}$&286$^{+89}_{-59}$ &$0.123\pm0.009$ &19/45260–280 c s$^{-1}$ & 8.6$\pm$3.1 & 8.6$^{+3.4}_{-3.1}$&148$\pm$10&$0.300\pm0.012$&63/45 220–240 c s$^{-1}$ &23$^{+15}_{-12}$&8.6$^{+3.2}_{-3.1}$&120$\pm$6&$0.504\pm0.005$ &53/44100–120 c s$^{-1}$ & $>10^3$ &51$\pm$2&53$\pm$3&$0.793\pm0.006$&44/45 80–100 c s$^{-1}$ & $>10^3$ &154$\pm$5&60$\pm$6&$0.828\pm0.006$&28/45
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In this paper we consider the question of existence of trigonometric series universal in weighted $L^1_{\mu}[0,2\pi]$ spaces with respect to rearrangements and in usual sense.'
address: 'Department of Physics, Chair of Higher Mathematics, Yerevan State University, Yerevan, Al. Manoogian st.1, 375049, Armenia'
author:
- 'S. A. Episkoposian'
title: 'Universal series by trigonometric system in weighted $L^1_{\mu}$ spaces'
---
[^1]
Introduction
============
Let $X$ be a Banach space.
A series $$\sum_{k=1}^\infty f_k,\ \ f_k \in X$$ is said to be universal in $X$ with respect to rearrangements, if for any $f \in X$ the members of (1.1) can be rearranged so that the obtained series $\displaystyle \sum_{k=1}^\infty
f_{\sigma(k)}$ converges to $f$ by norm of $X$.
The series (1.1) is said to be universal (in $X$) in the usual sense, if for any $f \in X$ there exists a growing sequence of natural numbers $n_k$ such that the sequence of partial sums with numbers $n_k$ of the series (1.1) converges to $f$ by norm of $X$.
The series (1.1) is said to be universal (in $X$) concerning partial series, if for any $f \in X$ it is possible to choose a partial series $\displaystyle {\sum_{k=1}^\infty f_{n_k}}$ from (1.1), which converges to the $f$ by norm of $X$.
Note, that many papers are devoted (see \[1\]- \[9\]) to the question on existence of variouse types of universal series in the sense of convergence [*almost everywhere and on a measure*]{}.
The first usual universal in the sense of convergence almost everywhere trigonometric series were constructed by D.E.Menshov \[1\] and V.Ya.Kozlov \[2\]. The series of the form $$\frac {1}{2}+\sum_{k=1}^\infty a_k cos{kx}+b_k sin{kx}$$ was constructed just by them such that for any measurable on $[0,2\pi]$ function $f(x)$ there exists the growing sequence of natural numbers $n_k$ such that the series (1.2) having the sequence of partial sums with numbers $n_k$ converges to $f(x)$ almost everywhere on $[0,2\pi]$. (Note here, that in this result, when $f(x)\in{L^1_{[0,2\pi]}} $, it is impossible to replace convergence almost everywhere by convergence in the metric ${L^1_{[0,2\pi]}}$).
This result was distributed by A.A.Talalian on arbitrary orthonormal complete systems (see \[3\]). He also established (see \[4\]), that if $\{\phi_n(x)\}_{n=1}^\infty
$ - the normalized basis of space ${L^p_{[0,1]}},p>1 $, then there exists a series of the form $$\sum_{k=1}^\infty{a_k\phi_k(x)},\ \ a_k \to 0.$$ which has property: for any measurable function $f(x)$ the members of series (1.3) can be rearranged so that the again received series converge on a measure on \[0,1\] to $f(x)$.
W. Orlicz \[5\] observed the fact that there exist functional series that are universal with respect to rearrangements in the sense of a.e. convergence in the class of a.e. finite measurable functions.
It is also useful to note that even Riemann proved that every convergent numerical series which is not absolutely convergent is universal with respect to rearrangements in the class of all real numbers.
Let $\mu(x)$ be a measurable on $[0,2\pi]$ function with $
0<\mu(x) \le1, x\in[0,2\pi]$ and let $L_\mu^1[0,2\pi]$ be a space of mesurable functions $f(x),\ \ x\in [0,2\pi]$ with $$\int_0^{2\pi} |f(x)| \mu(x) dx<\infty.$$
M.G.Grigorian constructed a series of the form (see \[6\]), $$\sum_{k=-\infty}^\infty{C_ke^{ikx} \ \ with \ \
\sum_{k=-\infty}^\infty \left | {C_k} \right|^q <\infty},\ \
\forall q>2$$ which is universal in $L_\mu^1[0,2\pi]$ concerning partial series for some weighted function $\mu(x),\ \ 0<\mu(x) \le1,
x\in[0,2\pi]$.
In \[9\] it is proved that for any given sequence of natural numbers $\{\lambda_m \}_{m=1}^\infty$ with $\lambda_m \nearrow^\infty$ there exists a series by trigonometric system of the form $$\sum_{k=1}^\infty C_k e^{ikx}, \ \ C_{-k}=\overline{C}_k,$$ with $$\left| \sum_{k=1}^m C_ke^{ikx} \right| \leq \lambda_m,\ \ x\in [0,2\pi],
\ \ ,\ \ m=1,2,...,$$ so that for each $\varepsilon>0$ a weighted function $\mu(x)$, $$0<\mu(x) \le1, \left | \{ x\in[0,2\pi]: \mu(x)\not =1 \} \right |
<\varepsilon$$ can be constructed, so that the series (1.4) is universal in the weighted space $L_\mu^1[0,2\pi]$ with respect simultaneously to rearrangements as well as to subseries.
In this paper we prove the following results.
There exists a series of the form $$\sum_{k=-\infty}^\infty{C_ke^{ikx} \ \ with \ \
\sum_{k=-\infty}^\infty \left | {C_k} \right|^q <\infty},\ \
\forall q>2$$ such that for any number $\varepsilon>0$ a weighted function $\mu(x)$, $ 0<\mu(x) \leq 1$, with $$\left | \{ x\in[0,2\pi]:\mu(x)\not =1 \} \right | <\varepsilon$$ can be constructed, so that the series $(1.5)$ is universal in $L_{\mu}^1 [0,2\pi]$ with respect to rearrangements $.$
There exists a series of the form $(1.5)$ such that for any number $\epsilon>0$ a weighted function $\mu(x)$ with $(1.6)$ can be constructed, so that the series $(1.5)$ is universal in $L_\mu^1[0,2\pi]$ in the usual sense $.$
BASIC LEMMA
============
For any given numbers $0<\varepsilon <{\frac {1}{ 2}}$, $N_0>2$ and a step function $$f(x)= \sum_{s=1}^q \gamma_s \cdot \chi_{\Delta_s} (x),$$ where $\Delta_s$ is an interval of the form $\Delta_m^{(i)}=
\left[ {\frac {i-1} {2^m}},{\frac {i} {2^m}} \right] $, $ 1\leq i
\leq 2^m $ and $$|\gamma_s| \cdot \sqrt {|\Delta_s|}<\epsilon^3 \cdot \left( 8
\cdot \int_0^{2\pi} f^2(x)dx \right)^{-1},\ \ s=1,2,...,q.$$ there exists a measurable set $E \subset [0,2\pi]$ and a polynomial $P(x)$ of the form $$P(x)= \sum_{N_0 \leq |k|<N} C_ke^{ikx}$$ which satisfy the conditions: $$|E|> 2\pi- \varepsilon,$$ $$\int_E |P(x)-f(x)|dx<\varepsilon,$$ $$\sum_{N_0 \leq |k|<N} |C_k|^{2+\varepsilon}< \varepsilon,\ \
C_{-k}=\overline {C}_k$$ $$\max_{N_0 \leq m<N} \left[ \int_e \left | \sum_{N_0 \leq |k|\leq
m} C_k e^{ikx} \right | dx \right] <\varepsilon+\int_e |f_(x)|dx,$$ for every measurable subset $e$ of $E$.
[**Proof** ]{} Let $0<\epsilon<{\frac {1}{ 2}}$ be an arbitrary number.
Set $$g(x)=1,\ \ if \ \ x \in [0,2\pi]\setminus \biggl[ {\frac
{\varepsilon \cdot \pi}{2}}, {\frac{3\varepsilon \cdot
\pi}{2}}\biggr] ;$$ $$g(x)=1-{\frac{2} {\varepsilon}},\ \ if \ \ x \in \biggl[ {\frac { \varepsilon \cdot \pi}{ 2}},{\frac {3\varepsilon \cdot \pi}{ 2}}\biggr];$$ We choose natural numbers $\nu_1$ and $N_1$ so large that the following inequalities be satisfied: $${\frac {1} {2\pi}}\left|\int_0^{2\pi} g_1(t)e^{-ikt}dt \right|<
\frac {\varepsilon}{16 \cdot \sqrt{N_0}},\ \ |k|<N_0,$$ where $$g_1(x)=\gamma_1\cdot g(\nu_1\cdot x)\cdot \chi_{\Delta_1}(x).$$ (By $\chi_E(x)$ we denote the characteristic function of the set $E$.) We put $$E_1=\{ x\in \Delta_s:\ \ g_s(x)=\gamma_s \},$$ By (2.3), (2.5) and (2.6) we have $$|E_1|>2\pi\cdot (1-\epsilon)\cdot |\Delta_1|;\ \ g_1(x)=0,\ \
x\notin \Delta_1,$$ $$\int_0^{2\pi}g_1^2(x)dx<{\frac {2}{ \epsilon}}\cdot
|\gamma_1|^2\cdot|\Delta_1|.$$ Since the trigonometric system $\{e^{ikx}\}_{k=-\infty}^\infty$ is complete in $L^2[0,2\pi]$, we can choose a natural number $N_{1}>N_{0}$ so large that $$\int_0^{2\pi}\left| \sum_{0\leq |k|<N_1}C_k^{(1)}e^{ikx}-
g_1(x)\right| dx \leq {\frac {\varepsilon} { 8}},$$ where $$C_k^{(1)}={\frac {1} {2\pi}}\int_0^{2\pi} g_1(t)e^{-ikt}dt .$$ Hence by (2.4),(2.5) and (2.9) we obtain $$\int_0^{2\pi}\left| \sum_{N_0\leq |k|<N_1}C_k^{(1)}e^{ikx}-
g_1(x)\right| dx\leq {\frac {\varepsilon} { 8}}+\left[ \sum_{0\leq
|k|<N_0} |C_k^{(1)}|^2 \right]^{\frac {1} { 2}}< {\frac
{\varepsilon}{4}};$$ Now assume that the numbers $\nu_1<\nu_2<...\nu_{s-1}$, $N_1<N_2<...<N_{s-1}$, functions $g_1(x),g_2(x),...,g_{s-1}(x)$ and the sets $E_1,E_2,....,E_{s-1}$ are defined. We take sufficiently large natural numbers $\nu_s>\nu_{s-1}$ and $N_s>N_{s-1}$ to satisfy $${\frac {1} {2\pi}}\left|\int_0^{2\pi} g_s(t)e^{-ikt}dt \right|<
\frac {\varepsilon} {16\cdot \sqrt{N_{s-1}}},\ \ \ \ 1\leq s \leq
q,\ \ \ \ |k|<N_{s-1},$$ $$\int_0^{2\pi}\left| \sum_{0\leq |k|<N_s}C_k^{(s)}e^{ikx}-
g_s(x)\right| dx \leq {\frac {\varepsilon} { 4^{s+1}}},$$ where $$g_s(x)=\gamma_s\cdot g(\nu_s\cdot x)\cdot \chi_{\Delta_s}(x),\ \ \
\ C_k^{(s)}={\frac {1} {2\pi}}\int_0^{2\pi} g_s(t)e^{-ikt}dt .$$ Set $$E_s=\{ x\in \Delta_s:\ \ g_s(x)=\gamma_s \},$$ Using the above arguments (see (2.16)-(2.18)), we conclude that the function $g_s(x)$ and the set $E_s$ satisfy the conditions: $$|E_s|>2\pi\cdot (1-\epsilon)\cdot |\Delta_s|;\ \ g_s(x)=0,\ \
x\notin \Delta_s,$$ $$\int_0^{2\pi}g_s^2(x)dx<{\frac {2} {\epsilon}}\cdot
|\gamma_s|^2\cdot|\Delta_s|.$$ $$\int_0^{2\pi}\left| \sum_{N_{s-1}\leq |k|<N_s}C_k^{(s)}e^{ikx}-
g_1(x)\right| dx < {\frac {\varepsilon} { 2^{s+1}}}.$$ Thus, by induction we can define natural numbers $\nu_1<\nu_2<...\nu_q$, $N_1<N_2<...<N_q$, functions $g_1(x),g_2(x),...,g_q(x)$ and sets $E_1,E_2,....,E_q$ such that conditions (2.14)- (2.16) are satisfied for all $s,\ \ 1\leq s
\leq q$. We define a set $E$ and a polynomial $P(x)$ as follows: $$E=\bigcup_{s=1}^q E_s,$$ $$P(x)=\sum_{N_0 \leq |k|< N} C_k e^{ikx}=\sum_{s=1}^q \left[
\sum_{N_{s-1}\leq |k|<N_s} C_k^{(s)}e^{ikx} \right],\ \
C_{-k}=\overline {C}_k,$$ where $$C_k=C_k^{(s)} \ \ for\ \ N_{s-1}\leq |k|<N_s,\ \ s=1,2,...,q, \ \ N=N_q-1.$$ By Bessel’s inequality and (2.3), (2.14) for all $s\in [1,q]$ we get $$\left[ \sum_{N_{s-1}\leq |k|<N_s} |C_k^{(s)}|^2 \right]^{\frac
{1}{2}} \leq \left[ \int_o^{2\pi}g_s^2(x)dx \right]^{\frac
{1}{2}}\leq$$ $$\leq {\frac {2} { \sqrt{\varepsilon}}}\cdot |\gamma_s| \cdot \sqrt
{|\Delta_s|},\ \ s=1,2,...,q.$$ From (2.3), (2.12) and (2.13) it follows that $$|E|> 2\pi- \varepsilon.$$ Taking relations (2.1), (2.3), (2.10), (2.12), (2.18) - (2.21) we obtain $$\int_E |P(x)-f(x)|dx \leq \sum_{s=1}^q \left[ \int_E \left|
\sum_{N_{s-1}\leq |k|<N_s} C_k^{(s)}e^{ikx}-g_s(x) \right|
dx\right]<\varepsilon$$ By (2.1), (2.2), (2.20) -(2.21)for any $k\in [N_0,N]$ we have $$\sum_{N_0\leq |k|< N} |C_k|^{2+ \epsilon} \leq \max_{N_0 \leq k
\leq N} |C_k|^\epsilon \cdot \sum_{k=N_0}^N |C_k|^2 \leq$$ $$\leq \max_{1 \leq s \leq q}\left[\sqrt {\frac {8} {\epsilon}}
\cdot |\gamma_s| \cdot \sqrt {|\Delta_s|}\right] \cdot
\sum_{s=1}^q \left[ \sum_{N_{s-1}\leq |k|<N_s} |C_k^{(s)}|^2
\right] \leq$$ $$\leq \max_{1 \leq s \leq q}\left[ \sqrt {\frac {8} {\epsilon}} \cdot |\gamma_s| \cdot \sqrt {|\Delta_s|}\right] \cdot {\frac {8} { \epsilon}} \cdot \sum_{s=1}^q |\gamma_s|^2 \cdot|\Delta_s| \leq$$ $$\leq\max_{1 \leq s \leq q}\left[ \sqrt {\frac {8} {\epsilon}} \cdot |\gamma_s| \cdot \sqrt {|\Delta_s|}\right] \cdot {\frac {8} {\epsilon}} \cdot \left[ \int_0^1 f^2(x)dx \right]< \epsilon;$$ That is, the statements 1) - 3) of Lemma are satisfied. Now we will check the fulfillment of statement 4) of Lemma. Let $N_0 \leq
m<N$, then for some $s_0,\ \ 1 \leq s_0 \leq q, \ \ \left( N_{s_0}
\leq m< N_{s_0+1} \right) $ we will have (see (2.20) and (2.21)) $$\sum_{N_0 \leq |k|\leq m} C_k e^{ikx}=\sum_{s=1}^{s_0} \left[
\sum_{N_{s-1}\leq |k|<N_s} C_k^{(s)}e^{ikx}
\right]+\sum_{N_{s_0-1}\leq |k|\leq m} C_k^{(s_0+1)} e^{ikx}.$$ Hence and from (2.1), (2.2), (2.3), (2.18), (2.19) and (2.22) for any measurable set $e \subset E$ we obtain $$\int_e \left | \sum_{N_{s-1}\leq |k|\leq m} C_k e^{ikx} \right |
dx \leq$$ $$\leq \sum_{s=1}^{s_0} \left[\int_e \left | \sum_{N_{s-1}\leq
|k|<N_s} C_k^{(s)}e^{ikx}-g_s(x) \right|dx \right] +$$ $$+\sum_{s=1}^{s_0} \int_e |g_s(x) | dx +\int_e \left|
\sum_{N_{s_0-1}\leq |k|\leq m} C_k^{(s_0+1)} e^{ikx} \right|dx<$$ $$<\sum_{s=1}^{s_0} {\frac {\varepsilon} {2^{s+1}}}+\int_e
|f(x)|dx+{\frac {2} {\sqrt {\varepsilon}}} \cdot |\gamma_{s_0+1}|
\cdot \sqrt {|\Delta_{s_0+1}|} <$$ $$< \int_e |f(x)|dx+ \varepsilon.$$
PROOF OF THEOREMS
==================
[**Proof of Theorem 1.4** ]{} Let $$f_1(x), f_2(x),...,f_n(x), \ \ x \in [0,2\pi]$$ be a sequence of all step functions, values and constancy interval endpoints of which are rational numbers. Applying Lemma consecutively, we can find a sequence $\{ E_s\}_{s=1}^\infty $ of sets and a sequence of polynomials $$P_s(x)=\sum_{N_{s-1}\leq |k|<N_s} C_k^{(s)}e^{ikx}$$ $$1=N_0<N_1<...<N_s<....,\ \ s=1,2,....,$$ which satisfy the conditions: $$| E_s| >1-2^{-2(s+1)} ,\ \ E_s \subset [0,2\pi],$$ $$\int_{E_s}|P_s(x)-f_s(x)|dx<2^{-2(s+1)},$$ $$\sum_{N_{s-1}\leq |k|<N_s}\left |C_k^{(s)}\right|^{2+2^{-2s}}< 2^{-2s},\ \ C_{-k}^{(s)}=\overline {C}_k^{(s)}$$ $$\max_{N_{s-1}\leq p<{N_s}} \left[ \int_e \left | \sum_{N_{s-1}\leq
|k|\leq p}C_k e^{ikx} \right | dx \right] <2^{-2(s+1)}+\int_e
|f_s(x)|dx,$$
for every measurable subset $e$ of $E_s$.
Denote $$\sum_{k=-\infty}^\infty C_k e^{ikx}=\sum_{s=1}^\infty \left[
\sum_{N_{s-1}\leq |k|<N_s} C_k^{(s)}e^{ikx} \right],$$ where $ C_k=C_k^{(s)}$ for $N_{s-1}\leq |k|<N_s$, $s=1,2,...$.
Let $\varepsilon$ be an arbitrary positive number. Setting $$\Omega_n = \bigcap_{s=n}^\infty E_s,\ \ n=1,2,....;$$ $$E=\Omega_{n_0} = \bigcap_{s=n_0}^\infty E_s,\ \ n_0=[\log_{1/2} \varepsilon]+1;$$ $$B=\bigcup _{n=n_0} ^\infty \Omega_n =\Omega_{n_0} \bigcup \left(
\bigcup _{n=n_0+1}^ \infty \Omega_n \setminus \Omega_{n-1}
\right).$$
It is clear (see (3.3)) that $| B |=2\pi$ and $| E | >2\pi-
\varepsilon$.
We define a function $\mu(x)$ in the following way: $$\mu(x)=1\ \ for \ \ x \in E \cup ([0,2\pi] \setminus B);$$ $$\mu(x)= \mu_{ n} \ \ for \ \ x \in \Omega_{n} \setminus
\Omega_{n-1},\ \ n\geq n_0+1,$$ where $$\mu_n=\left[ 2^{4n}\cdot \prod_{s=1}^n h_s \right]^{-1};$$ $$h_s=|| f_s(x)||_C+ \max_{N_{s-1}\leq p<{N_s}} \Vert \sum_{N_{s-1}\leq |k|\leq p} C_k^{(s)}e^{ikx} \Vert_{C}+1,$$
where $$||g(x)||_C=\max_{x\in [0,2\pi]} |g(x)|,$$ $g(x)$ is a continuous function on $[0,2\pi]$.
From (3.5),(3.7)-(3.10) we obtain
\(A) – $ 0<\mu(x) \le1, \mu(x)$ is a measurable function and $$\left | \{x\in[0,2\pi]:\mu(x)\not =1\} \right|<\varepsilon.$$
\(B) – $\displaystyle \sum_{k=1}^\infty |C_k|^q<\infty,\ \
\forall q>2.$
Hence, obviously we have $$\lim_{k\to \infty}{C_k}=0.$$ It follows from (3.8)-(3.10) that for all $s \geq n_0$ and $p
\in \left[ N_{s-1},N_s \right)$ $$\int_{[0,2\pi] \setminus \Omega_s} \left| \sum_{N_{s-1}\leq
|k|\leq p} C_k^{(s)}e^{ikx} \right| \mu(x) dx=$$ $$=\sum_{n=s+1}^ \infty \left[\int_{\Omega_n \setminus \Omega_{n-1}}
\left| \sum_{N_{s-1}\leq |k|\leq p} C_k^{(s)}e^{ikx}\right|
\mu_n dx \right] \leq$$ $$\leq \sum_{n=s+1}^ \infty2^{-4n} \left[\int_0^ {2\pi} \left|
\sum_{N_{s-1}\leq |k|\leq p} C_k^{(s)}e^{ikx}\right| h_s^{-1} dx
\right]<2^{-4s}.$$ By (3.4), (3.8)-(3.10) for all $s \geq n_0$ we have $$\int_0^{2\pi} \left| P_s(x)-f_s(x) \right|\mu(x)dx=\int_{\Omega_s}
\left| P_s(x)-f_s(x) \right|\mu(x)dx+$$ $$+\int_{[0,2\pi] \setminus \Omega_{s}} \left| P_s(x)-f_s(x) \right|\mu(x)dx =2^{-2(s+1)}+$$ $$+\sum_{n=s+1}^\infty \left[\int_{\Omega_n \setminus \Omega_{n-1}}
\left| P_s(x)-f_s(x) \right| \mu_n dx\right] \leq 2^{-2(s+1)}+$$ $$+ \sum_{n=s+1}^ \infty 2^{-4s}\left[ \int_0^ {2\pi} \left(\left|
f_s(x) \right| +\left| \sum_{N_{s-1}\leq |k|<N_s}
C_k^{(s)}e^{ikx} \right| \right) h_s^{-1}dx \right ]<$$ $$<2^{-2(s+1)}+2^{-4s}<2^{-2s}.$$ Taking relations (3.6), (3.8)- (3.10) and (3.12) into account we obtain that for all $p \in \left[ N_{s-1},N_s \right)$ and $s \geq
n_0+1$ $$\int_0^{2\pi} \left| \sum_{N_{s-1}\leq |k|\leq p}
C_k^{(s)}e^{ikx} \right| \mu(x) dx=$$ $$=\int_{\Omega_s} \left| \sum_{N_{s-1}\leq |k|\leq p}
C_k^{(s)}e^{ikx} \right| \mu(x) dx+$$ $$+\int_{[0,2\pi] \setminus \Omega_s} \left| \sum_{N_{s-1}\leq
|k|\leq p} C_k^{(s)}e^{ikx} \right| \mu(x) dx<$$ $$< \sum_{n=n_0+1}^ s \left[\int_{\Omega_n \setminus \Omega_{n-1}}
\left| \sum_{N_{s-1}\leq |k|\leq p} C_k^{(s)}e^{ikx} \right| dx
\right]\cdot \mu_n+2^{-4s}<$$ $$<\sum_{n=n_0+1}^ s \left( 2^{-2(s+1)}+\int_{\Omega_n \setminus
\Omega_{n-1}} |f_s(x)|dx \right) \mu_n +2^{-4s} =$$ $$=2^{-2(s+1)} \cdot \sum_{n=n_0+1}^ s \mu_n+\int_{\Omega_s}
|f_s(x)|\mu(x)dx +2^{-4s}<$$ $$<\int_0^{2\pi} |f_s(x)|\mu(x)dx +2^{-4s}.$$ Let $ f(x) \in L_{\mu}^1 [0,2\pi]$ , i. e.$ \int_0^{2\pi} |f(x)|
\mu(x) dx<\infty$ .
It is easy to see that we can choose a function $f_{\nu_1}(x)$ from the sequence (3.1) such that $$\int_0^{2\pi} \left| f(x)- f_{\nu_1}(x) \right|\mu(x)dx<2^{-2},\ \
\nu_1 > n_0+1.$$ Hence, we have $$\int_0^{2\pi} \left| f_{\nu_1}(x)
\right|\mu(x)dx<2^{-2}+\int_0^{2\pi} |f(x)|\mu(x)dx.$$ From (2.1), (A), (3.13) and (3.15) we obtain with $m_1=1$ $$\int_0^{2\pi} \left| f(x)- \left [ P_{\nu_1}(x)+C_{m_1}e^{im_1x}
\right] \right|\mu(x)dx \leq$$ $$\leq \int_0^{2\pi} \left| f(x)- f_{\nu_1}(x) \right|\mu(x)dx+$$ $$+\int_0^{2\pi} \left| f_{\nu_1}(x)-P_{\nu_1}(x) \right|\mu(x)dx+$$ $$+\int_0^{2\pi} \left| C_{m_1}e^{im_1x}\right|\mu(x)dx <2\cdot
2^{-2}+2\pi\cdot\left| C_{m_1}\right|.$$ Assume that numbers $\nu_1<\nu_2<...<\nu_{q-1};m_1<m_2<...<m_{q-1}$ are chosen in such a way that the following condition is satisfied: $$\int_0^{2\pi} \left| f(x)- \sum_{s=1}^j \left [
P_{\nu_s}(x)+C_{m_s}e^{im_sx} \right] \right|\mu(x)dx<$$ $$<2\cdot 2^{-2j}+2\pi \cdot\left|C_{m_j}\right|, \ \ 1\leq j \leq
q-1 .$$ We choose a function $f_{\nu_q}(x)$ from the sequence (3.1) such that $$\int_0^{2\pi} \left| \left( f(x)- \sum_{s=1}^{q-1} \left [
P_{\nu_s}(x)+C_{m_s}e^{im_sx} \right] \right)-f_{n_q}(x)\right|
\mu(x)dx< 2^{-2q},$$ where $ \nu_q>\nu_{q-1};\ \ \nu_q>m_{q-1}$
This with (3.18) imply $$\int_0^{2\pi} \left| f_{\nu_q}(x) \right| \mu(x)dx<2^{-2q}+2\cdot
2^{-2(q-1)}+2\pi\cdot\left|C_{m_{q-1}} \right| =$$ $$=9 \cdot 2^{-2q}+ 2\pi\cdot\left|C_{m_{q-1}} \right|.$$
By (3.13), (3.14) and (3.20) we obtain $$\int_0^{2\pi} \left| f_{\nu_q}(x)- P_{\nu_q}(x)
\right|\mu(x)dx<2^{-2\nu_q},$$ $$\ \ P_{\nu_q}(x)=\sum_{N_{\nu_q-1}\leq |k|<N_{\nu_q}}
C_k^{(\nu_q)}e^{ikx}.$$ $$\max_{N_{\nu_q-1} \leq p<N{\nu_q}} \int_0^{2\pi} \left|
\sum_{k=N_{\nu_q-1}}^ p C_k^{(\nu_q)}e^{ikx} \right | \mu(x) dx<10
\cdot 2^{-2q}+2\pi\cdot\left | C_{m_{q-1}} \right |.$$ Denote $$m_q= \min \left\{ n \in N: n \notin \left\{ \left\{ \{ k
\}_{k=N_{\nu_s-1}}^{N_{\nu_s}-1} \right\}_{s=1}^q \cup \{
m_s\}_{s=1}^{q-1}\right\} \right\}.$$ From (2.1), (A), (3.19) and (3.21) we have $$\int_0^{2\pi} \left| f(x)- \sum_{s=1}^q \left [
P_{\nu_s}(x)+C_{m_s}e^{im_sx} \right] \right| \mu(x)dx\leq$$ $$\leq \int_0^{2\pi} \left| \left( f(x)- \sum_{s=1}^{q-1} \left [
P_{\nu_s}(x)+C_{m_s}e^{im_sx} \right] \right)-f_{\nu_q}(x)\right|
\mu(x)dx+$$ $$+\int_0^{2\pi} \left| f_{\nu_q}(x)-P_{\nu_q}(x)\right| \mu(x)dx+$$ $$+\int_0^{2\pi} \left| C_{m_q}e^{im_qx}\right| \mu(x)dx<2 \cdot
2^{-2q}+2\pi\cdot\left| C_{m_q} \right|.$$ Thus, by induction we on $q$ can choose from series (3.7) a sequence of members $$C_{m_q}e^{im_qx} ,\ \ q=1,2,...,$$ and a sequence of polynomials $$P_{\nu_q}(x)=\sum_{N_{\nu_q-1}\leq |k|<N_{\nu_q}}
C_k^{(\nu_q)}e^{ikx},\ \ N_{n_q-1}>N_{n_{q-1}},\ \ q=1,2,....$$ such that conditions (3.22) - (3.24) are satisfied for all $q\geq
1.$
Taking account the choice of $P_{\nu_q}(x)$ and $C_{m_q}e^{im_qx}$ (see (3.23) and (3.25)) we conclude that the series $$\sum_{q=1}^\infty \left[ \sum_{N_{\nu_q-1}\leq |k|<N_{\nu_q}}
C_k^{(\nu_q)}e^{ikx}+C_{m_q}e^{iqx} \right ]$$ is obtained from the series (3.7) by rearrangement of members. Denote this series by $\sum C_{\sigma(k)}e^{i\sigma(k) x}.$
It follows from (3.11), (3.22) and (3.24) that the series $\sum
C_{\sigma(k)}e^{i \sigma(k)x}$ converges to the function $f(x)$ in the metric $L_{\mu}^1[0,2\pi]$, i.e. the series (3.7) is universal with respect to rearrangements (see Definition 1.1).
[**The Theorem 1.4 is proved.**]{} $$$$ [**Proof of the Theorem 1.5**]{}
Applying Lemma consecutively, we can find a sequence $\{
E_s\}_{s=1}^\infty $ of sets and a sequence of polynomials $$P_s(x)=\sum_{N_{s-1}\leq |k|<N_s} C_k^{(s)}e^{ikx} ,\ \
C_{-k}^{(s)}=\overline {C}_k^{(s)}$$ $$1=N_0<N_1<...<N_s<....,\ \ s=1,2,....,$$
which satisfy the conditions: $$\left| E_s\right| >1-2^{-2(s+1)} ,\ \ E_s\subset [0,2\pi],$$ $$\sum_{N_{s-1}\leq |k|<N_s}\left |C_k^{(s)}\right|^{2+2^{-2s}}< 2^{-2s},$$ $$\int_{E_n} \left | f_n(x)-\sum_{s=1}^n P_s(x)\right | dx <2^{-n},\ \ n=1,2,...,$$ where $\{ f_n(x)\}_{ n=1}^\infty,\ \ x\in [0,2\pi]$ be a sequence of all step functions, values and constancy interval endpoints of which are rational numbers.
Denote $$\sum_{k=-\infty}^\infty C_k e^{ikx}=\sum_{s=1}^\infty \left[
\sum_{N_{s-1}\leq |k|<N_s} C_k^{(s)}e^{ikx} \right],$$ where $ C_k=C_k^{(s)}$ for $N_{s-1}\leq |k|<N_s$, $s=1,2,...$.
It is clear (see (3.28)) that $$\displaystyle \sum_{k=1}^\infty |C_k|^q<\infty, \ \ \forall q>2.$$
Repeating reasoning of Theorem 1 a weighted function $\mu (x),\ \
0<\mu(x)\leq 1$ can constructed so that the following condition is satisfied: $$\int_0^{2\pi} \left| f_n(x)-\sum_{s=1}^n P_s(x) \right|\cdot \mu
(x) dx< 2^{-2n},\ \ n=1,2,...$$ For any function $f(x) \in L_{\mu}^1[0,1]$ we can choose a subsystem $\{f_{n_\nu}(x)\}_{\nu=1}^\infty $ from the sequence (3.1) such that $$\int_0^{2\pi} \left| f(x)- f_{n_\nu}(x) \right|\mu(x)dx<2^{-2\nu}.$$ From (3.30)-(3.32) we conclud $$\int_0^{2\pi} \left| f(x)- \sum_{|k|\leq M_\nu}C_ke^{ikx}
\right| \mu(x)dx =$$ $$\int_0^{2\pi} \left| f(x)- \sum_{s=1}^{n_\nu} \left[
\sum_{N_{s-1}\leq |k|<N_s} C_k^{(s)}e^{ikx} \right] \right|
\mu(x)dx \leq$$ $$\leq \int_0^{2\pi} \left| f(x)- f_{\nu_k}(x) \right| \cdot
\mu(x)dx+$$ $$+\int_0^{2\pi} \left| f_{\nu_k}(x)-\sum_{s=1}^{\nu_k} P_s(x)
\right|\cdot \mu (x) dx<2^{-2k}+2^{-2{\nu_k}}$$ where $M_{\nu}=N_{n_{\nu}}-1$.
Thus, the series (3.30) is universal in $L_\mu^1[0,1]$ in the sense of usual (see Definition 1.2).
[**The Theorem 1.5 is proved.**]{}
The author thanks Professor M.G.Grigorian for his attention to this paper.
[9]{} D.E.Menshov, *On the partial summs of trigonometric series.*Mat. Sb. 20(1947), 197-238 V.Ya.Kozlov, *On the complete systems of orthogonal functions*, Mat. Sb. 26(1950), 351-364. A.A.Talalian , *On the convergence almost everywhere the sumbsequence of partial summs of general orthogonal series*, Izv. Ak. Nauk Arm. SSR ser. Math. 10(1957), 17-34. —— , *On the universal series with respect to rearrangements*, Izv. AN. SSSR ser. Math. 24(1960), 567-604. W.Orlicz, *Uber die unabhangig von der Anordnung fast uberall kniwergenten Reihen*, Bull. de l’Academie Polonaise des Sciences, 81 (1927), p. 117-125; M. G. Grigorian , *On the representation of functions by orthogonal series in weighted $L^p$ spaces*, Studia. Math. 134(3)1999, 211-237. P.L.Ul’janov , *On the unconditional convergence and the summmability*, Izv. AN. SSSR ser. Math. 22(1958), 811-840; F.G.Arutunian, *To representation of functions by multiplaer series*D.A.N. Arm. SSR 64(1976), 72-76; S.A.Episkoposian, M.G.Grigorian, *Representation of functions in weighted spaces $L_\mu^1[0,2\pi]$ by trigonometric and Walsh series* , Analysis Mathematica, 27 (2001), 261-277;
[^1]: The author was supported in part by Grant- 01-000 from the Government of Armenia .
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'S. Wedemeyer'
bibliography:
- 'p13cat\_sven.bib'
title: The CO5BOLD Analysis Tool
---
Introduction
============
The growing computational resources allow for increasingly larger and more detailed numerical simulations of stellar atmospheres, resulting in a considerably large amount of data. The production of advanced comprehensive models must therefore be accompanied by the development of efficient analysis and visualization software that is capable of handling the produced large data sets. Here, the Analysis Tool (abbreviated CAT, Fig. \[fig:logo\]) is described. It is designed for an interactive analysis of 2D and 3D model atmospheres, which are produced with – a widely used state-of-the-art code for the simulation of stellar atmospheres [@2012JCoPh.231..919F].
Program overview
================
CAT has an interactive graphical user interface, which is is programmed in IDL (see Fig. \[fig:screensht1\]). The primary mode of operation is the display and analysis of slices through 3D (or 2D) models. CAT can handle files with multiple snapshots (*.full*).
#### Settings and sessions.
After the installation, adjustments such as the default window size and paths can be made and saved. These standard settings will be restored every time CAT is started. Settings that are connected to a particular analysis session can be saved and restored as session files. It is possible to define a default session, which is automatically loaded when CAT starts.
#### Slice display
The projection plane can be chosen perpendicular to the axes, resulting in the three planes , , and in a 3D model. The position within a plane can be changed either by clicking in the displayed image or through the widgets in the control panel. CAT now also provides a zoom function in a separate tab, in which the magnification and the currently displayed region can be selected (see Fig. \[fig:screensht2\] for a close-up region).
Different physical quantities can be chosen in the *data tab* (see Fig. \[fig:screensht1\]), ranging from the basic quantities contained in the model file to more advanced quantities like, e.g., the spatial components of the electric current density [e.g., @2005ESASP.596E..65S]. The chosen quantity can then be combined with a mathematical operation, e.g. the logarithm, which is useful for quantities that cover many orders of magnitude in the displayed slice (e.g., the mass density in the plane). By default, a color-coded 2D slice is displayed. Alternatively, the data in the current projection plane can be shown as surface plot, contour plot or line plot. Color-coded slices can be overlaid with contours (incl. a reference contour at optical depth unity) and/or a vector field. For the latter the spatial components of the velocity or the magnetic field in the selected projection plane can be drawn as vectors or streamlines, while the perpendicular component is not considered. The length of the vectors can be chosen directly in the *data tab* with more options being available in the *options menu*.
#### Measurements.
The new *measure tab* provides different tools for (i) measuring distances between two points or along a polygon path, (ii) defining markers, and (iii) producing and exporting contour paths. The markers, which are sets of coordinates, and the contours can be interactively selected, modified and exported for further processing. These tools can be used for, e.g., defining 3D structures or for manually tracking features in space and time.
#### Images and animation.
The displayed images can be output in various image formats. The animation tool provides an interactive way to produce MPEG videos or image sequences. It can either be a time animation for a series of model snapshots contained in one or more multi-snapshot file(s) or a spatial scan through a selected model snapshot along one of the spatial axes.
#### Export.
CAT offers several options to export model data. The currently loaded simulation snapshot can be written out in UIO format, even when it is part of a multi-snapshot (*.full*) file. This feature is useful if a simulation terminated without producing an *.end* file or if this file is corrupted for some reason. In this case CAT can be used to extract the last time step from the corresponding *.full* file for using it as start model for the next simulation run. Model snapshots can also be saved as plane-parallel models. In a coming version, it will be possible to save only parts of the model and to extract data into IDL savefiles for further analysis.
The 3D visualization tool, which is integrated in CAT, can give a first 3D impression but has limited functionality (see Fig. \[fig:screensht3\]). The VAPOR tool [@vapor_clyne2007; @vapor_clyne2005] allows for a more comprehensive analysis of the 3D structure, in particular for vector fields like velocity and magnetic field. For that reason, CAT offers the possibility to export data into VAPOR format.
Data analysis examples
======================
CAT has been repeatedly proved useful in the past. It provides an easy way of monitoring the evolution of an ongoing simulation. CAT can also help to quickly reveal what is wrong in cases where a simulation terminates due to problems at a specific grid cell. Next to these “care-taking” tasks, CAT as been essential for the detailed scientific analysis in many cases.
#### Swirls and tornadoes.
One example is the discovery of ring-like structures with increased horizontal velocity at chromospheric heights in 3D magnetohydrodynamic (MHD) models of the Sun. Similar rings were previously observed as so-called “chromospheric swirls” with the Swedish 1-m Solar Telescope . The rings of enhanced velocity, which appeared prominently in CAT (see Fig. \[fig:screensht2\]), suggested that chromospheric swirls are the observational signature of rotating magnetic field structures. This finding finally led to a more comprehensive study of this phenomenon, then known as “magnetic tornadoes” [@2012Natur.486..505W].
#### M-type dwarf stars.
CAT has been extensively used for the development of 3D MHD atmosphere models of M-type dwarf stars that extend from the upper convection zone into the chromosphere [@2012arXiv1207.2342W]. The resulting set of models with different initial magnetic fields with different field strengths and different topologies (e.g., homogeneous vertical or mixed polarities) will be analysed in more detail in forthcoming publications.
Outlook
=======
Since the first version in 2002, the development of CAT has been driven by the need for additional functionality for the analysis of increasingly complex models. Also in the future, CAT will be continuously extended. The feedback of users is therefore very welcome. Possible extensions in future releases could include the ability to read the data formats of other MHD codes and to display and analyse slices at arbitrary angles.
The author likes to thank the organisers of the Workshop (), which was held in Heidelberg, Germany, in 2012.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The Meissner effect in diffusive normal metal / insulator / $d$-wave superconductor junctions is studied theoretically in the framework of the Usadel equation under the generalized boundary condition. The effect of midgap Andreev resonant states (MARS) formed at the interface of $d$-wave superconductor is taken into account. It is shown that the formation of MARS suppresses the susceptibility of the diffusive normal metal.'
address:
- |
Department of Applied Physics, Nagoya University, Nagoya 464-8603, Japan.\
CREST Japan Science and Technology Corporation (JST), Nagoya 464-8603, Japan.
- ' Faculty of Science and Technology, University of Twente, 7500 AE, Enschede, The Netherlands.'
- ' Department of Applied Physics, Hokkaido University, Sapporo 060-8628, Japan.'
author:
- Takehito Yokoyama
- Yukio Tanaka
- Alexander Golubov
- 'Jun-ichiro Inoue'
- Yasuhiro Asano
title: 'Meissner effect in diffusive normal metal / $d$-wave superconductor junctions'
---
,
,
,
,
Meissner effect; proximity effect; midgap Andreev resonant states; $d$-wave superconductor
Introduction
============
In diffusive normal metal / superconductor (DN/S) junctions, the DN acquires induced superconductivity, i.e. Cooper pairs penetrate into the DN. This proximity effect has been studied since the BCS theory was established. The proximity induced Meissner demagnetization in DN/S junctions was measured experimentally by Oda and Nagano[@Oda] and Mota et al.[@Mota1]. It has $T^{-1/2}$ dependence in the dirty limit. The quasiclassical Green’s function theory was used earlier to study the Meissner effect in proximity structures.
The quasiclassical Green’s function theory was developed by Eilenberger [@Eilenberger] and was generalized by Eliashberg [@Eliashberg], Larkin and Ovchinnikov [@Larkin] in order to study the nonequilibrium state. This theory was applied by Zaikin[@Zaikin] and Kieselmann[@Kieselmann] to studing the Meissner effect in DN/S junctions. Narikiyo and Fukuyama [@Narikiyo] calculated the Meissner screening length in a semi-infinite system containing an Anderson impurity. Higashitani and Nagai studied the Meissner effect in the clean limit [@Higashitani]. Belzig et al. [@Bel1; @Bel2] have considered more realistic systems by assuming a perfectly transparent N/S interface. Up to now the boundary conditions derived by Kupriyanov and Lukichev (KL) [@KL] were widely used to study proximity effect in DN/S structures.
A more general boundary conditions was derived by Nazarov [@Nazarov2] based on the Keldysh-Nambu Green’s function formalism [@Zaitsev] within the framework of the Landauer-Büttiker scattering formalism. The merit of this boundary condition is that the BTK theory[@BTK] is reproduced in the ballistic limit while in the diffusive limit with a low transmissivity of the interface, the KL boundary condition is reproduced. Although almost all previous papers on Meissner effect in mesoscopic NS junctions are either based on the KL boundary conditions or on the BTK model, in the actual junctions, the transparency of the junction is not always small and the impurity scattering in the DN cannot be neglected. Tanaka et al.[@TGK] and Yokoyama et al.[@Yoko] calculated tunneling conductance by using the Nazarov’s boundary condition.
It is well known in $d$-wave superconductors that the midgap Andreev resonant states (MARS) are formed at the interface of $d$-wave superconductor. The MARS crucially influence various physical quantities [@TK]. One of the authors (Y.T.) recently generalized the boundary condition of the Keldysh-Nambu Green’s function formalism to unconventional superconductor junctions [@TNGK; @pwave]. It is revealed that in DN/$d$-wave superconductor junctions the proximity effect and the MARS strongly compete with each other [@TNGK], while they coexist in DN/triplet superconductor junctions. The newly obtained boundary conditions expressed in the Keldysh-Nambu Green’s function are useful for the calculation of various physical quantities. The timely problem is to study theoretically the Meissner effect in DN / $d$-wave S junctions using the new boundary conditions [@TNGK]. In the present paper, we calculate the susceptibility of the DN layer in DN/ $d$-wave S junctions for various junction parameters such as the height of the insulating barrier at the interface and the angle between the normal to the interface and the crystal axis of a $d$-wave superconductor.
The organization of the paper is as follows. In section 2, we will provide the derivation of the expression for the susceptibility of the DN. In section 3, the results of calculation are presented for various types of junction. In section 4, the summary of the obtained results is given. In the present paper we set $c=k_B=\hbar=1$.
Formulation
===========
In this section, we introduce the model and the formalism. We consider a junction consisting of vacuum (VAC) and superconducting reservoirs connected by a quasi-one-dimensional diffusive conductor (DN) with a length $L$ much larger than the mean free path. We assume that the interface between the DN conductor and the S electrode at $x=L$ has a resistance $R_{b}$, the DN/VAC interface at $x=0$ is specular, and we apply the generalized boundary conditions by Tanaka [@TNGK] to treat the interface between DN and S. A weak external magnetic field $H$ is applied in $z$-direction (see Fig. 1). The vector potential can be chosen to have only the $y$ component which depends on $x$.
We describe the insulating barrier between DN and S by using the $\delta$-function (i.e., $U(x)=H\delta(x-L)$), which provides the transparency of the junction $T_{m}=4\cos ^{2}\phi /(4\cos ^{2}\phi +Z^{2})$, where $Z=2H/v_{F}$ is a dimensionless constant, $\phi $ is the injection angle of a quasiparticle measured from the interface normal to the junction and $v_{F}$ is Fermi velocity. In the following, we solve the Usadel equations [@Usadel] with using the standard $\theta$-parameterization. The parameter $\theta (x)$ is a measure of the proximity effect in DN and obey the following equation $$D\frac{\partial ^{2}}{\partial x^{2}}\theta (x)-2\omega_n\sin [\theta (x)]=0, \label{Usa1}$$ where $D$ and $\omega_n$ denote the diffusion constant and the Matsubara frequency, respectively. The boundary condition for $\theta(x)$ at the DN/S interface is given in Ref. [@TNGK]. The interface resistance $R_{b}$ is given by $$R_{b}=R_{0} \frac{2} {\int_{-\pi/2}^{\pi/2} d\phi T(\phi)\cos\phi}$$ with $ T(\phi)=4\cos ^{2}\phi /(4\cos ^{2}\phi +Z^{2})$. Here $R_{0}$ is Sharvin resistance $R_{0}^{-1}=e^{2}k_{F}^2S_c/(4\pi^{2})$, where $k_{F}$ is the Fermi wave number and $S_c$ is the constriction area. The current distribution is given by $$j(x) = - 8\pi e^2 N\left( 0 \right)DT\sum\limits_{\omega _n > 0}
{\sin ^2 \theta \left( x \right)} A\left( x \right),$$ where $A(x)$, $N(0)$ and $T$ denote the vector potential, the density of states at the Fermi energy and the temperature of the system respectively. The Maxwell equation reads $$\frac{{d^2 }}{{dx^2 }}A\left( x \right) = - 4\pi j\left( x \right).$$ The boundary conditions for $A(x)$ are given by $$\begin{aligned}
\frac{d}{{dx}}A\left( 0 \right) = H, \qquad A\left( L \right) = 0,\end{aligned}$$ where we have neglected the penetration of magnetic fields into the superconductor by assuming a small penetration depth in S.
Finally we obtain the expression of the susceptibility, $$- 4\pi \chi = 1 + \frac{{A\left( 0 \right)}}{{HL}}.$$ The $d$-wave pair potentials in directional space are given by $\Delta_{\pm} = \Delta(T)\cos2(\phi \mp \alpha)$, where $\Delta(T)$ is the magnitude of pair potential at a given temperature $T$ and $\alpha$ denotes the angle between the normal to the interface and the crystal axis of a $d$-wave superconductor.
Results
=======
In the following, we focus on the magnitude of the diamagnetic susceptibility $\chi$ induced by the proximity effect. Figs. 2 and 3 show the susceptibility for $Z=10$ and $Z=0$ respectively where $K =16\pi e^2 N\left( 0 \right)D^2$. For $\alpha=0$, the temperature dependencies of $-4\pi\chi$ are not much different. For $\alpha=0.125\pi$, the magnitude of $\chi$ for $Z=10$ is much stronger suppressed than that for $Z=0$. At the same time, we find that the magnitude of $\chi$ decreases with increasing $\alpha$. We note in the case of $\alpha=0.25\pi$ that the susceptibility completely vanishes, (i.e., $-4\pi\chi=0$). This is because the proximity effect is absent in diffusive metals due to angular averaging[@TNGK]. The absence of the proximity effect is a significant feature specific for junctions containing unconventional superconductors.
We also plot the $\alpha$ dependencies of the susceptibility at $T/T_C=0.01$ and $T/T_C=0.1$ in Fig. 4. For all cases, $\chi$ is a decreasing function of $\alpha$. At $T/T_C=0.01$, the magnitude of $\chi$ for $Z=10$ rapidly decreases with the increase of $\alpha$. The results imply that the MARS suppresses the proximity effect in low transparent junctions and low temperatures.
Conclusions
===========
In the present paper, we have calculated the induced Meissner effect by the proximity effect in DN region of DN/$d$-wave superconductor junctions. We have solved the Usadel equation under a general boundary condition [@TNGK] in which the formation of the MARS is fully taken into account [@TK]. The magnitude of $\chi$ decreases with the increase of $\alpha$ up to $0.25\pi$. At $\alpha=0.25\pi$, where all quasiparticles feel MARS, the $\chi$ becomes zero. It might be interesting to check experimentally such an anomalous proximity effect in DN. Another future problem is a similar calculation of the induced Meissner effect with a $p$-wave triplet superconductor instead of a $d$-wave one, since dramatic new phenomena are recently predicted in DN/ triplet junctions [@pwave].
The authors appreciate useful and fruitful discussions with Yu. Nazarov and H. Itoh. This work was supported by the Core Research for Evolutional Science and Technology (CREST) of the Japan Science and Technology Corporation (JST). The computational aspect of this work has been performed at the facilities of the Supercomputer Center, Institute for Solid State Physics, University of Tokyo and the Computer Center.
[22]{} Y. Oda, H. Nagano, Solid State Commun. **35** (1980) 631.
A. C. Mota, D. Marek, J. C. Weber, Helv. Phys. Acta **55** (1982) 647.
G.Eilenberger,Z.Phys.**214** (1968) 195.
G.M. Eliashberg, Sov. Phys. JETP **34** (1971) 668.
A.I. Larkin, Yu. V. Ovchinnikov, Sov. Phys. JETP ** 41** (1975) 960.
A. D. Zaikin, Solid State Commun. **41** (1982) 533.
G. Kieselmann, Phys. Rev. B **35** (1987) 6762.
O. Narikiyo, H. Fukuyama, J. Phys. Soc. Jpn. **58** (1989) 4557.
S. Higashitani, K. Nagai, J. Phys. Soc. Jpn. **64** (1995) 549.
W. Belzig, C. Bruder, G. Schön, Phys. Rev. B **53** (1996) 5727.
W. Belzig, C. Bruder, A. L. Fauchère, Phys. Rev. B **58** (1998) 14531.
M.Yu. Kupriyanov, V. F. Lukichev, Zh. Exp. Teor. Fiz. **94** (1988) 139 \[Sov. Phys. JETP **67** (1988) 1163\].
Yu. V. Nazarov, Superlattices and Microstructuctures **25** (1999) 1221, cond-mat/9811155.
A. V. Zaitsev, Sov. Phys. JETP **59** (1984) 1163.
G.E. Blonder, M. Tinkham, T.M. Klapwijk, Phys. Rev. B **25** (1982) 4515.
Y. Tanaka, A. A. Golubov, S. Kashiwaya, Phys. Rev. B **68** (2003) 054513.
T. Yokoyama, Y. Tanaka, A. A. Golubov, J. Inoue, Y. Asano, cond-mat/0406745.
Y. Tanaka, S. Kashiwaya Phys. Rev. Lett. **74** (1995) 3451; Y. Tanaka, S. Kashiwaya, Phys. Rev. B **56** (1997) 892; S. Kashiwaya, Y. Tanaka Rep. Prog Phys **63** (2000) 1641; Yu. S. Barash, M. S. Kalenkov, J. Kurkijärvi Phys. Rev. B **62** (2000) 6665.
Y. Tanaka, Yu. V. Nazarov, S. Kashiwaya, Phys. Rev. Lett. **90** (2003) 167003; Y. Tanaka, Yu. V. Nazarov, A. A. Golubov, S. Kashiwaya, Phys. Rev. B **69** (2004) 144519.
Y. Tanaka, S. Kashiwaya, Phys. Rev. B **70** (2004) 012507.
K.D. Usadel Phys. Rev. Lett. **25** (1970) 507.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Nd$_{0.05}$Ce$_{0.95}$CoIn$_5$ features a magnetic field-driven quantum phase transition that separates two antiferromagnetic phases with an identical magnetic structure inside the superconducting condensate. Using neutron diffraction we demonstrate that the population of the two magnetic domains in the two phases is affected differently by the rotation of the magnetic field in the tetragonal basal plane. In the low-field SDW-phase the domain population is only weakly affected while in the high-field Q-phase they undergo a sharp switch for fields around the $a$-axis. Our results provide evidence that the anisotropic spin susceptibility in both phases arises ultimately from spin-orbit interactions but are qualitatively different in the two phases. This provides evidence that the electronic structure is changed at the quantum phase transition, which yields a modified coupling between magnetism and superconductivity in the Q-phase.'
author:
- 'D. G. Mazzone'
- 'R. Yadav'
- 'M. Bartkowiak'
- 'J. L. Gavilano'
- 'S. Raymond'
- 'E. Ressouche'
- 'G. Lapertot'
- 'M. Kenzelmann'
title: 'Distinct domain switching in Nd$_{0.05}$Ce$_{0.95}$CoIn$_5$ at low and high fields'
---
Introduction {#introduction .unnumbered}
============
Strongly correlated electron systems can feature electronic ground states, in which different electronic charge, spin, orbital and lattice degrees of freedom such as phonons, defects or strain are coupled. Such couplings can trigger novel quantum phenomena, such as unconventional superconductivity, topological and multiferroic phases or heavy-fermion ground states. The understanding of cooperative phenomena is particularly challenging in unconventional superconductors where Cooper pairs are thought to arise from magnetic fluctuations [@Monthoux2007]. Antisymmetric spin-orbit interactions can lead to novel phases with uncommon symmetry such as triplet superconductivity. In CePt$_3$Si, for instance, it is believed that such interactions generate an anomalous spin susceptibility that triggers superconductivity with mixed spin-singlet and triplet Cooper pairs [@Frigeri2004; @Fak2014].
A direct way to gain insight in non-phonon driven superconductivity is to study the coupling of magnetic order and superconductivity. Magnetic superconductors feature a variety of different behavior when tuned via external parameters, such as pressure, chemical substitution or magnetic fields [@Norman2011; @Pfleiderer2009; @White2015]. In most materials a competition between both phenomena is observed [@Norman2011; @Pfleiderer2009; @White2015], but there also exist cases in which magnetic order and superconductivity cooperate [@Huxley2015; @Kenzelmann2008; @Kenzelmann2017]. Examples include the heavy-fermion compound UGe$_2$, where superconductivity is only stable in the presence of ferromagnetic order [@Huxley2015], or CeCoIn$_5$ where magnetism only appears inside the superconducting phase [@Kenzelmann2008; @Kenzelmann2017]. The latter phenomena has been discussed theoretically already for some time [@Fulde1964; @Larkin1965; @Psaltakis1983; @Shimahara2000; @Lebed2006; @Aperis2008; @Yanase2009; @Kato2011; @Suzuki2011; @Agterberg2009; @Michal2011; @Mineev2016; @Kim20172; @Hatakeyama2015; @Hosoya2017], but it remains an open question how magnetic order can emerge from superconductivity.
The series Nd$_{1-x}$Ce$_x$CoIn$_5$ reveals a competition between static magnetic order and superconductivity for $x$ $>$ 0.78 at zero field [@Hu2008]. 5% Nd doped CeCoIn$_5$, however, features an auxiliary field-induced magnetic phase (Q-phase) that is only stable within the superconducting condensate and that collapses in a first-order transition at the upper critical field [@Mazzone2017]. This behavior provides evidence for a cooperative magneto-superconducting ground state in the Q-phase with a coupling that is similar to the one of undoped CeCoIn$_5$. The high-field Q-phase of 5% Nd doped CeCoIn$_5$ is separated from its low-field SDW-phase via a quantum phase transition at $\mu_0H^*\approx$ 8 T [@Mazzone2017]. Both magnetic phases feature the same amplitude modulated spin-density wave (SDW) order with an ordered magnetic moment, $\mu\approx$ 0.15$\mu_B$, that is oriented along the $c$-axis. The magnetic propagation vector, **Q**$_{1,2}$ = ($q$, $\pm q$, 0.5) with $q\approx$ 0.445, is similar to the one of the Q-phase in CeCoIn$_5$ and directed along the nodal direction of the $d_{x^2-y^2}$-superconducting order parameter [@Mazzone2017]. However, it is currently an open question how the symmetry of superconductivity is affected in the Q-phase.
{width="\linewidth"}
Here, we study the population of the two magnetic domains, **Q**$_{1,2}$, in the SDW- and Q-phase of Nd$_{0.05}$Ce$_{0.95}$CoIn$_5$ by means of neutron diffraction for magnetic fields oriented close to **H**$||$\[0 1 0\] and along the \[1 $\bar{1}$ 0\]-direction in reciprocal lattice units (r.l.u.). We find a field-induced redistribution of the domain-population for **H**$||$\[1 $\bar{1}$ 0\], where the intensity in **Q**$_{2}$ is suppressed at $\mu_0H_d$ = 3.6(6) T. The Q-phase features a single spin-density modulation direction except for a small field range of $\pm$2.5$^\circ$ around **H**$||$\[0 1 0\], where a continuous crossover of populated domains is observed. The behavior is different in the SDW-phase, where the magnetic domains remain equally populated for magnetic fields close to the $a$-axis.
Results {#results .unnumbered}
=======
Fig. \[fig1\] displays two magnetic Bragg peaks that belong to the magnetic domains **Q$_1$** and **Q$_2$**. The diffracted neutron intensity was measured for wave-vector transfers, ($\pm q$, $q$, $\mp$0.5), along the tetragonal plane in the Q-phase of Nd$_{0.05}$Ce$_{0.95}$CoIn$_5$. The magnetic field equals $\mu_0H$ = 10.4 T and is oriented along $\psi$ = 2.5 and -2$^\circ$ relative to the $a$-axis. The neutron diffraction results show a populated **Q$_1$**-domain for $\psi$ = 2.5$^\circ$, while **Q$_2$** is suppressed. When rotating the field to $\psi$ = -2$^\circ$ the domain population is switched. A tunable mono-domain state is also found in the Q-phase of CeCoIn$_5$ [@Gerber2014].
![**Domain population.** Background subtracted peak intensity as a function of $\psi$ for representatives of **Q$_2$** in orange and **Q$_1$** in blue measured at $\mu_0H$ = 10.4 T. Solid lines are guide lines to the eyes. Dashed line represent the fully suppressed and fully populated domains.[]{data-label="fig2"}](Fig2.eps){width="\linewidth"}
Fig. \[fig2\] depicts the angular dependence of the background subtracted peak intensity of both domains inside the Q-phase at $\mu_0H$ = 10.4 T. When rotating the magnetic field through \[0 1 0\] one domain continuously depopulates while the other one is populated. The neutron diffraction results reveal a crossover region of $\Delta\psi\approx$ 5$^\circ$ where both domains are at least partly populated. Although this is much broader than in undoped CeCoIn$_5$, this is relatively sharp considering that the Q-domain must be pinned to the Nd dopants [@Gerber2014].
![**Switching in the SDW-phase.** Diffracted neutron intensity along the tetragonal plane of **Q$_2$** in orange and **Q$_1$** in blue, measured at $\mu_0H$ = 2 T and $T$ = 40 mK for $\psi$ = $\pm$2.5$^\circ$. Background in gray was measured at $\mu_0H$ = 11.9 T. Solid lines represent Gaussian fits on a sloping background.[]{data-label="fig3"}](Fig3.eps){width="\linewidth"}
We further studied the magnetic domain population in the SDW-phase. Diffracted neutron intensity along the tetragonal plane of **Q$_1$** and **Q$_2$** at $\mu_0H$ = 2 T is shown in Fig. \[fig3\]. The gray dots denote the background that was measured at $\mu_0H$ = 11.9 T $>$ $\mu_0H_{c_2}$. Within the SDW-phase no change in the domain population is observed in the vicinity of **H**$||$\[0 1 0\]. This is in strong contrast to the Q-phase, where only one of both domains is populated for $\psi$ = $\pm$2.5$^\circ$.
The field dependence of the integrated Bragg peak intensity at (0.56, 0.44, 0.5) is represented by orange circles in Fig. \[fig4\] for **H**$||$\[1 $\bar{1}$ 0\]. It demonstrates that the magnetic **Q**$_2$-domain gradually weakens with increasing magnetic field and is suppressed at $\mu_0H_d$ = 3.6(6) T. This is in strong contrast to the field dependence of the **Q**$_1$-domain, whose intensity increases at small fields and reveals a broad maximum around $\mu_0H\approx$ 4 T [@Mazzone2017].
In addition, Fig. \[fig4\] compares the scaled, integrated intensity of the two magnetic domains, $I_{Q_1}$ and $I_{Q_2}$, with the total integrated intensity $I_{tot}$ = $I_{Q_1}$ + $I_{Q_2}$ for **H**$||$\[1 $\bar{1}$ 0\]. This plot combines new measurements with those published earlier [@Mazzone2017]. The normalization of the integrated intensity was chosen to have equal population at zero field, respecting the tetragonal symmetry.
The field dependence of $I_{Q_1}$ and $I_{Q_2}$ show that fields smaller than $H_d$ trigger a redistribution of the domain population. The intensity in **Q**$_1$ is enhanced, while the magnetic domain in the plane along the field **Q**$_2$ is reduced, such that the total integrated intensity remains constant. Increasing the field further yields a mono-domain state, where only the **Q**$_1$-domain is populated.
![**Field-induced Domain imbalance.** Field dependence of the two magnetic domains, **Q**$_1$ and **Q**$_2$, in Nd$_{0.05}$Ce$_{0.95}$CoIn$_5$ for **H**$||$\[1 $\bar{1}$ 0\] and the total intensity $I_{tot}$ = $I_{Q_1}$ + $I_{Q_2}$. The curve is reconstructed from the data presented here and the ones reported in Ref. [@Mazzone2017]. The Solid lines represent weighted linear fits. []{data-label="fig4"}](Fig4.eps){width="\linewidth"}
Discussion {#discussion .unnumbered}
==========
5% Nd substituted CeCoIn$_5$ features magnetic order with a moment orientation along the tetragonal $c$-axis [@Mazzone2017]. The Zeeman coupling (**MH**) vanishes for fields applied in the tetragonal basal plane and cannot drive the field and angular dependent population of **Q**$_1$ and **Q**$_2$.
An anisotropic spin susceptibility can originate from spin-orbit interactions as observed, for instance, in the non-centrosymmetric superconductor CePt$_3$Si [@Frigeri2004; @Fak2014]. It has been suggested that a similar phenomena can appear in a multiband metal with tetragonal symmetry, such as CeCoIn$_5$, when the spin-orbit coupling reduces the space group symmetry by a basal in-plane field [@Mineev2016]. Based on this microscopic theory , Kim $et.$ $al.$ developed a phenomenological Landau model, which includes a weak spin-orbit coupling term in its free energy density [@Kim20172]. It considers a magnetic moment arrangement perpendicular to the basal plane, as found in the two antiferromagnetic phases of Nd$_{0.05}$Ce$_{0.95}$CoIn$_5$. Depending on the Landau parameters, the model predicts either (A) the coexistence of the Q-domains whose population is tuned with the field direction, or (B) the presence of only one Q-domain with a sharp switching [@Kim20172].
Scenario A yields an equal domain population for $\psi$ = 0$^\circ$ and a maximal difference for $\psi$ = 45$^\circ$. This is consistent with the neutron diffraction results at $\mu_0H$ = 2 T, where two populated domains are found that feature a comparable intensity for **H**$||$\[0 1 0\] and a 80% suppression of **Q**$_2$ for **H**$||$\[1 $\bar{1}$ 0\]. Our data in the SDW phase can thus be explained with spin-orbit couplings mediating a field-induced repopulation of the Q-domains. The much sharper switching at high fields, however, is not consistent with scenario A and suggest qualitatively different behavior more consistent with scenario B. This means that the Landau theory by Kim $et.$ $al.$ [@Kim20172] can describe the SDW and Q-phase separately, but not both phases simultaneously, and different phenomenological parameters would be needed for the low- and high-field phases. This points towards a qualitative change of the electronic structure at $H^*$.
Alternatively it has been suggested that $d$-wave superconductivity and magnetic order is coupled via a spatially modulated spin-triplet superconducting order parameter [@Agterberg2009; @Hosoya2017]. The formation of triplet superconductivity also relies on spin-orbit interactions and the additional order parameter supports a sharp switch of the modulation direction for a field around the $a$-axis [@Agterberg2009; @Gerber2014]. A supplementary superconducting gap is consistent with a non-magnetic primary order parameter in the Q-phase that is postulated by the identical magnetic symmetry in the SDW- and the Q-phase [@Mazzone2017]. Such a scenario can also account for the thermal conductivity results of the Q-phase in CeCoIn$_5$ that reveal a reduced quasiparticle excitation spectrum perpendicular to the populated **Q**-domain [@Kim2016]. A symmetry analysis suggests that the two modulation directions belong to different irreducible representations and that a $p$-wave order parameter is aligned along the suppressed domain [@Gerber2014; @Kim2016]. As a result, the switching has to be sharp, albeit it may be broadened by pinning to Nd-ions.
It has been suggested that the Q-phase arises from the condensation of a superconducting exciton that creates a novel superconducting state [@Michal2011]. The superconducting spin resonance that is observed in CeCoIn$_5$ at zero field appears at the same wave-vector as static magnetic order in the Q-phase [@Raymond2015]. Under magnetic field the resonant splits and the lowest mode may condense into the ground state at the Q-phase boundary [@Michal2011; @Stock2012; @Raymond2012; @Akbari2012]. In 5% Nd doped CeCoIn$_5$ the resonance is not affected by the onset of static magnetic order at zero field, which provides evidence for a decoupling of these fluctuations from magnetic order in the SDW-phase [@Mazzone20172].
In summary, we demonstrate that an in-plane rotation of the magnetic field in Nd$_{0.05}$Ce$_{0.95}$CoIn$_5$ triggers a magnetic domain imbalance that is distinct in the SDW- and the Q-phase. We find two domains with no preferable spin-density modulation direction in the low-field SDW-phase, for fields close to the $a$-axis. At low fields, a field-induced change of the relative domain population is observed when the field is applied along the diagonal direction of the tetragonal plane. The selection of a mono-domain state becomes relatively sharp in the high-field Q-phase, where the spin-density modulation direction can be switched by a few-degree rotation of the field around the $a$-axis. The low- and high-field behavior cannot be simultaneously explained by the available phenomenological theories, and requires a modification in the coupling between superconductivity and magnetic order at $H^*$. We suggest that an additional superconducting order parameter of $p$-wave symmetry emerges at high fields and intertwines magnetic order with $d$-wave superconductivity.
Methods {#methods .unnumbered}
=======
The neutron diffraction experiments at $T$ = 40 mK and up to $\mu_0H$ = 11.9 T were carried out on the thermal neutron lifting-counter two-axis spectrometer D23 at the Institut Laue Langevin, Grenoble France. The single crystal ($m$ = 64 mg) was placed in a vertical-field magnet with dilution insert and exposed to an incident neutron wavelength of $\lambda$ = 1.27 Å. The crystal was oriented either with the vertical axis parallel to \[1 $\bar{1}$ 0\] or along \[0 1 0\]. In the latter setup the tetragonal $a$-axis was tilted into the basal plane using a non-magnetic piezoelectric sample rotator (type ANGt50 from attocube system AG) inside the dilution refrigerator (see Supplementary Materials of Ref. [@Gerber2014]). The relative field direction was directly measured via the vertical tilt of the structural (2, 0, 0) Bragg peak. The relative angle between the magnetic field direction **H** and the tetragonal $a$-axis, \[0 1 0\], is denoted as $\psi$
The solid lines in Fig. \[fig1\] and \[fig3\] represent Gaussian fits to the neutron diffraction data. Magnetic Bragg peaks at ($q$, $q$, -0.5) were fitted using a Gaussian line shape on a linear background. The background at (-$q$, $q$, 0.5) and $\psi$ $\leq$ -2$^\circ$ could not be described satisfactorily by a linear behavior solely. We used an additional Gaussian component centered at $q$ = 0.427. The width of the magnetic Bragg peaks in each domain was fixed to the mean value of all corresponding fits.
The integrated intensity at (0.56, 0.44, 0.5) for $\mu_0H$ = 0, 1, 2, 3 and 4 T and $T$ = 40 mK along **H**$||$\[1 $\bar{1}$ 0\] was obtained from scans along ($q$ + $h$, $q$, 0.5), where $h$ was chosen such that the scan was centered at (1, 0, 1) - **Q$_2$**. The integrated intensity at $\mu_0H$ = 0.5, 1.5, 2.5 and 3.5 T was determined from the background subtracted Bragg peak intensity multiplied with the averaged peak width found in the $q$-scans. A weighted linear fit to these results yields $\mu_0H_d$ = 3.6(6) T.
All data needed to evaluate the conclusions in the paper are present in the paper. Additional data available from authors upon request. Correspondence and requests for materials should be addressed to D. G. M. (email: [email protected])
[20]{}
Monthoux, P., Pines, D. & Lonzarich, G. G. Superconductivity without phonons. *Nature* **450**, 1177-1183 (2007).
Frigeri, P. A., Agterberg, D. F. , Koga, A. & Sigrist, M. Superconductivity without Inversion Symmetry: MnSi versus CePt$_3$Si. *Phys. Rev. Lett.* **92**, 097001 (2004).
Fk, B., $et$ $al.$ Anomalous Spin Response in the Non-Centrosymmetric Metal CePt$_3$Si. *J. Phys. Soc. Jpn.* **83**, 063703 (2014).
Norman, M. R. . , **332**, 196-200 (2011).
Pfleiderer, C. Superconducting phases of $f$-electron compounds. *Rev. Mod. Phys.* **81**, 1551 (2009).
White, B. D., Thompson, J. D. & Maple, M. B. Unconventional superconductivity in heavy-fermion compounds. *Physica C* **514**, 246-278 (2015).
Huxley, A. D. Ferromagnetic superconductors. *Physica C* **514**, 368-377 (2015).
Kenzelmann, M. $et$ $al.$ . , **321**, 1652 (2008).
Kenzelmann, M. . **80**, 034501 (2017).
Fulde, P. & Ferrell, R. A. Superconductivity in Strong Spin-Exchange Field. *Phys. Rev.* **135**, A550 (1964).
Larkin, A. I. & Ovchinnikov, Y. N. Inhomogeneous state of superconductors. *Soviet Physics JEPT-USSR* **20**, 762 (1965).
Psaltakis, G. C. & Fenton, E. C. Superconductivity and spin-density waves: organic superconductors. *J. Phys. C: Solid State Phys.* **16**, 3913-3932 (1983).
Shimahara, H. Coexistence of singlet and triplet at- tractive channels in the pairing interactions mediated by antiferromagnetic fluctuations. *J. Phys. Soc. Jpn.* **69**, 1966-1969 (2000).
Lebed, A. G. Cooper Pairs with Broken Parity and Spin-Rotational Symmetries in d-Wave Superconductors. *Phys. Rev. Lett.* **96**, 037002 (2006).
Aperis, A., Varelogiannis, G., Littlewood, P. B. & Simons, B. D. Coexistence of spin density wave, d-wave singlet and staggered $\pi$-triplet superconductivity. *J. Phys.: Condens. Matter.* **20**, 434235 (2008).
Yanase, Y. & Sigrist, M. Antiferromagnetic order and $\pi$-triplet pairing in the Fulde-Ferrell-Larkin-Ovchinnikov state. *J. Phys. Soc. Jpn.* **78**, 114715 (2009).
Kato, Y., Batista, C. D. & Vekhter, I. Antiferromagnetic Order in Pauli-Limited Unconventional Superconductors. *Phys. Rev. Lett.* **107**, 096401 (2011).
Hatakeyama, Y. & Ikeda, R. Antiferromagnetic order oriented by Fulde-Ferrell-Larkin-Ovchinnikov superconducting order. *Phys. Rev. B* **91**, 094504 (2015).
Hosoya, K. -I. & Ikeda, R. Possible Triplet Superconducting Order in Magnetic Superconducting Phase induced by Paramagnetic Pair-Breaking. *Phys. Rev. B* **95**, 224513 (2017).
Agterberg, D. F., Sigrist, M. & Tsunetsugu, H. Order Parameter and Vortices in the Superconducting Q phase of CeCoIn$_5$. *Phys. Rev. Lett.* **102**, 207004 (2009).
Suzuki, K. M., Ichioka, M. & Machida, K. Theory of an inherent spin-density-wave instability due to vortices in superconductors with strong Pauli effects. *Phys. Rev. B* **83**, 140503(R) (2011).
Michal, V. P. & Mineev, V. P. Field-induced spin-excitation, condensation in the $d_{x^2-y^2}$-wave superconductor CeCoIn$_5$. *Phys. Rev. B* **84**, 052508 (2011).
Mineev, V. P. Antiferromagnetic order in CeCoIn$_5$ oriented by spin-orbital coupling. *Low Temp. Phys.* **43**, 11 (2017).
Kim, D. Y. $et$ $al.$ Switching dynamics of the spin density wave in superconducting CeCoIn$_5$. *Phys. Rev. B* **95**, 241110(R) (2017).
Hu, R., Lee, Y., Hudis, J., Mitrovic, V. F. & Petrovic, C. Composition and field-tuned magnetism and superconductivity in Nd$_{1-x}$Ce$_x$CoIn$_5$. *Phys. Rev. B* **77**, 165129 (2008).
Mazzone, D. G. $et$ $al.$ Field-induced magnetic instability within a superconducting condensate *Sci. Adv.* **3**, e1602055 (2017).
Gerber, S. $et$ $al.$ Switching of magnetic domains reveals evidence for spatially inhomogeneous superconductivity. *Nat. Phys.* **10**, 126-129 (2014).
Kim, D. Y. $et$ $al.$ Intertwined Orders in Heavy-Fermion Superconductor CeCoIn$_5$. *Phys. Rev. X* **6**, 041059 (2016).
Raymond, S. & Lapertot, G. Ising Incommensurate Spin Resonance of CeCoIn$_5$: A Dynamical Precursor of the Q Phase. *Phys. Rev. Lett.* **115**, 037001 (2015).
Stock, C. $et$ $al.$ Magnetic Field Splitting of the Spin Resonance in CeCoIn$_5$. *Phys. Rev. Lett.* **109**, 167207 (2012).
Raymond, S., Kaneko, K., Hiess, A., Steffens, P. & Lapertot, G. Evidence for Three Fluctuation Channels in the Spin Resonance of the Unconventional Superconductor CeCoIn$_5$. *Phys. Rev. Lett.* **109**, 237210 (2012).
Akbari, A. & Thalmeier, P. Field-induced spin exciton doublet splitting in $d_{x^2 - y^2}$-wave CeMIn$_5$ (M = Rh, Ir, Co) heavy-electron superconductors. *Phys. Rev. B* **86**, 134516 (2012).
Mazzone, D. G. $et$ $al.$ Spin Resonance and Magnetic Order in an Unconventional Superconductor. *Phys. Rev. Lett.* **119**, 187002 (2017).
Acknowledgements {#acknowledgements .unnumbered}
================
We acknowledge the Institut Laue-Langevin for the allocated beam time on D23. We thank P. Fouilloux for technical assistance. Discussions with Duk Young Kim and Roman Movshovich are acknowledged. In addition, we thank the Swiss National Foundation (grant No. 200021\_147071, 206021\_139082 and 200021\_138018). This work was also supported by the Swiss State Secretariat for Education, Research and Innovation (SERI) through a CRG-grant.
Author contributions statement {#author-contributions-statement .unnumbered}
==============================
D. G. M., S. R., J. L. G and M. K. planed and led the project. The experiments were carried out by D. G. M., R. Y., M. B., J. L. G., S. R., E. R., and M. K. The sample was grown by G. L and the data analyzed by D. G. M. The manuscript was written by D. G. M., S. R., J. L. G. and M. K. with the input of all co-authors.
Additional information {#additional-information .unnumbered}
======================
**Competing financial interests**: The authors declare that they have no competing interests.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Surprisingly, the relict cosmological constant has a crucial influence on properties of accretion discs orbiting black holes in quasars and active galactic nuclei. We show it by considering basic properties of both the geometrically thin and thick accretion discs in the Kerr–de Sitter black-hole (naked-singularity) spacetimes. Both thin and thick discs must have an outer edge allowing outflow of matter into the outer space, located nearby the so called static radius, where the gravitational attraction of a black hole is balanced by the cosmological repulsion. Jets produced by thick discs can be significantly collimated after crossing the static radius. Extension of discs in quasars is comparable with extension of the associated galaxies, indicating a possibility that the relict cosmological constant puts an upper limit on extension of galaxies.'
address: |
Institute of Physics, Faculty of Philosophy and Science, Silesian University in Opava\
Bezručovo nám. 13, CZ-74601 Opava, Czech Republic\
[email protected]
author:
- ZDENĚK STUCHLÍK
title: |
INFLUENCE OF THE RELICT COSMOLOGICAL CONSTANT ON\
ACCRETION DISCS
---
Introduction
============
Recent data from a wide variety of independent cosmological tests indicate convincingly that within the framework of the inflationary cosmology a non-zero, although very small, vacuum energy density, i.e., a relict repulsive cosmological constant (RRCC), $\Lambda > 0$, or some kind of similarly acting quintessence, has to be invoked in order to explain the dynamics of the recent Universe.[@Bah-etal:1999:SCIEN:; @Kol-Tur:1990:EarUni:]
There is a strong “concordance” indication[@Spe-etal:2003:ASTJS:] that the observed value of the vacuum energy density is $$\varrho_{\mathrm{vac(0)}}\approx 0.73 \varrho_{\rm crit(0)}$$ with present values of the critical energy density $\varrho_{\rm crit(0)}$, and the Hubble parameter $H_{0}$ given by $$\varrho_{\rm crit(0)}=\frac{3H_{0}^2}{8\pi},\quad H_{0}=100h\
\rm{km}\ \rm{s}^{-1}\ \rm{Mpc}^{-1}.$$ Taking value of the dimensionless parameter $h\approx 0.7$, we obtain the RRCC to be $$\Lambda_{0}=8\pi\varrho_{\mathrm{vac(0)}}\approx 1.3 \times 10^{-56}\
\rm{cm}^{-2}.
\label{E3}$$
It is well known that the RRCC strongly influences expansion of the Universe, leading finally to an exponentially accelerated stage.[@Car-Ost:1996:ModAst:] However, surprisingly enough, the RRCC can be relevant for accretion processes in the field of central black holes in quasars and active galactic nuclei.
Basic properties of geometrically thin accretion discs with low accretion rates and negligible pressure are given by the circular geodetical motion in the black-hole backgrounds,[@Nov-Tho:1973:BlaHol:436] while for geometrically thick discs with high accretion rates and relevant pressure they are determined by equipotential surfaces of test perfect fluid rotating in the backgrounds.[@Koz-Jar-Abr:1978:ASTRA:; @Abr-Jar-Sik:1978:ASTRA:] The presence of the RRCC changes substantially the asymptotic structure of the black-hole (naked-singularity) backgrounds as they become asymptotically de Sitter and contain a cosmological event horizon behind which the spacetime is dynamic. Properties of the circular geodesic orbits in the Schwarzschild–de Sitter (SdS) and Reissner–Nordström–de Sitter (RNdS) spacetimes show that due to the presence of the RRCC, the thin discs have not only an inner edge determined (approximately) by the radius of the innermost stable circular orbit, but also an outer edge given by the radius of the outermost stable circular orbit, located nearby the static radius.[@Stu-Hle:1999:PHYSR4:; @Stu-Hle:2002:ACTPS2:] The vicinity of the static radius can be considered as a counterpart to the asymptotically flat region of the Kerr spacetimes, as can be demonstrated by the embedding diagrams of the equatorial plane of both the directly projected geometry and the optical reference geometry reflecting some hidden properties of the geodesic motion.[@Abr-Pra:1990:MONNR:; @Stu-Hle:2000:CLAQG:; @Hle:2002:JB60:] The analysis of equilibrium configurations of perfect fluid orbiting in the SdS black-hole backgrounds shows a possible existence of thick discs with outflow of matter through an outer cusp of the equilibrium configuration due to violation of mechanical equilibrium.[@Stu-Sla-Hle:2000:ASTRA:] Such an outflow can represent a strong stabilizing effect[@Rez-Zan-Fon:2003:ASTRA:] against the runaway instability[@Abr-Cal-Nob:1983:NATURE:] of thick discs orbiting the SdS black holes.
However, it is crucial to understand the role of the RRCC in astrophysically most relevant, rotating Kerr backgrounds. In the Kerr–de Sitter (KdS) backgrounds, we shall consider circular equatorial motion of test particles, relevant for thin discs, and equilibrium configurations of perfect fluid, relevant for thick discs. We shall focus attention on the black-hole backgrounds, but some results related to the naked-singularity backgrounds will be mentioned because of increasing theoretical evidence on possible existence of naked singularities.[@deFel-Yun:2001:CLAQG:]
Kerr–de Sitter spacetimes {#KdS spacetimes}
=========================
In the standard Boyer–Lindquist coordinates ($t,r,\theta, \phi$) and the geometric units ($c=G=1$), the Kerr–(anti-)de Sitter geometry is given by the line element $$\begin{aligned}
\d s^2 = & - & \frac{\Delta_r}{I^2 \rho^2}(\d t-a\sin^2 \theta
\d\phi)^2
+ \frac{\Delta_{\theta}\sin^2 \theta}{I^2 \rho^2}
\left[a\d t- \left(r^2+a^2 \right) \d\phi \right]^2 \nonumber\\
& + & \frac{\rho^2}{\Delta_r} \d r^2 +
\frac{\rho^2}{\Delta_{\theta}} \d\theta^2,
\label{e1}\end{aligned}$$ where $$\begin{aligned}
\Delta_r & = & -\frac{1}{3}\Lambda r^2 \left(r^2+a^2 \right)
+r^2 -2Mr + a^2, \\ \label{e2}
\Delta_{\theta} & = & 1+ \frac{1}{3} \Lambda a^2 \cos^2 \theta,
\quad I = 1 + \frac{1}{3} \Lambda a^2,
\quad \rho^2 = r^2 +a^2 \cos^2 \theta. \label{e3}\end{aligned}$$ The parameters of the spacetime are: mass ($M$), specific angular momentum ($a$), cosmological constant ($\Lambda$). It is convenient to introduce a dimensionless cosmological parameter $$\label{e6}
y = \frac{1}{3} \Lambda M^2.$$ For simplicity, we put $M=1$ hereafter. Equivalently, also the coordinates $t,\ r$, the line element $\d s$, and the rotational parameter of the spacetime $a$, being expressed in units of $M$, become dimensionless. We focus our attention to the case $y>0$ corresponding to the repulsive cosmological constant; then (\[e1\]) describes a KdS spacetime.
The event horizons of the spacetime are given by the pseudosingularities of the line element (\[e1\]), determined by the condition $\Delta_r =0$. The loci of the event horizons are implicitly determined by the relation $$\label{e7}
a^2 = a^2_{\mathrm{h}}(r;y) \equiv \frac{r^2 -2r -yr^4}{yr^2 -1}.$$ It can be shown[@Stu-Sla:2004:PHYSR4:] that a critical value of the cosmological parameter exists $$\label{e11}
y_{\mathrm{c(KdS)}} = \frac{16}{(3+2\sqrt{3})^3} \doteq 0,05924,$$ such that for $y>y_{\mathrm{c(KdS)}}$, only naked-singularity backgrounds exist for $a^2>0$. There is another critical value $y_{\mathrm{c(SdS)}} = 1/27 \doteq 0.03704$, which is limiting the existence of SdS black holes.[@Stu-Hle:1999:PHYSR4:] In the RNdS spacetimes, the critical value is[@Stu-Hle:2002:ACTPS2:] $y_{\mathrm{c(RNdS)}} = 2/27 \doteq 0.07407$.
If $y = y_{\mathrm{c(KdS)}}$, the function $a^2_{\mathrm{h}}(r;y)$ has an inflex point corresponding to a critical value of the rotation parameter of the KdS spacetimes $$\label{e13}
a^2_{\rm crit} = \frac{3}{16} (3+ 2\sqrt{3}) \doteq 1,21202.$$ KdS black holes can exist for $a^2 < a^2_{\rm crit}$ only, while KdS naked singularities can exist for both $a^2 <
a^2_{\rm crit}$ and $a^2 > a^2_{\rm crit}$.
Separation of the KdS black-hole and naked-singularity spacetimes in the parameter space $y$–$a^2$ is shown in Fig. \[f1\]. In the black-hole spacetimes there are two black-hole horizons and the cosmological horizon, with $r_{\mathrm{h}-} < r_{\mathrm{h}+}
< r_{\mathrm{c}}$. In the naked-singularity spacetimes, there is the cosmological horizon $r_{\mathrm{c}}$ only.
![Classification of the Kerr–de Sitter spacetimes. Dashed curves separate black holes and naked singularities. Full curves divide the parametric space by properties of the stable circular orbits relevant for Keplerian accretion discs. Spacetimes with both plus-family and minus-family stable circular orbits ([**I**]{} and [**V**]{}). Spacetimes with no minus-family stable circular orbits ([**II**]{} and [**VI**]{}). Spacetimes with no stable circular orbits ([**III**]{} and [**IV**]{}). Dashed-dotted curve defines the subregion of the naked-singularity spacetimes, where the plus-family circular orbits could be stable and counter-rotating (from the point of view of the LNRF), shaded is the subregion allowing stable circular orbits with $E_{+}<0$! (Taken from Ref. .)[]{data-label="f1"}](class.eps){width=".6\hsize"}
The extreme cases, when two (or all three) horizons coalesce, were discussed in detail for the case of RNdS spacetimes.[@Bri-Hay:1994:CLAQG:; @Hay-Nak:1994:PHYSR4:] In the KdS spacetimes, the situation is analogical. If $r_{\mathrm{h}-}=r_{\mathrm{h}+} <
r_{\mathrm{c}}$, the extreme black-hole case occurs, if $r_{\mathrm{h}-} <
r_{\mathrm{h}+} = r_{\mathrm{c}}$, the marginal naked-singularity case occurs, if $r_{\mathrm{h}-} = r_{\mathrm{h}+} = r_{\mathrm{c}}$, the “ultra-extreme” case occurs corresponding to a naked singularity.
Thin discs {#Thin}
==========
Basic properties of thin accretion discs are determined by equatorial circular motion of test particles because any tilted disc has to be driven to the equatorial plane of the rotating spacetimes due to the dragging of inertial frames.[@Bar-Pet:1975:ASTRJ2L:]
The motion of a test particle with rest mass $m$ is given by the geodesic equations. In a separated and integrated form, the equations were obtained by Carter.[@Car:1973:BlaHol:] For the motion restricted to the equatorial plane ($\d\theta/\d \lambda = 0$, $\theta = \pi/2$) of the KdS spacetime, the Carter equations take the following form $$\begin{aligned}
r^2 \oder{r}{\lambda} & = & \pm R^{1/2} (r), \\
\label{e14}
r^2 \oder{\phi}{\lambda} & = & - IP_{\theta}+
\frac{a I P_r}{\Delta_r}, \\
\label{e15}
r^2 \oder{t}{\lambda} & = & -a I P_{\theta} +
\frac{(r^2 +a^2) I P_r}{\Delta_r},
\label{e16}\end{aligned}$$ where $$\begin{aligned}
& & R(r) = P^2_r -\Delta_r \left(m^2r^2 +K \right), \\
\label{e17}
& & P_r = I {\cal{E}} \left(r^2 +a^2 \right) - I a \Phi,
\quad P_{\theta} = I (a {\cal{E}} - \Phi),
\quad K = I^2 (a {\cal{E}} - \Phi)^2. \label{e18}\end{aligned}$$ The proper time of the particle $\tau$ is related to the affine parameter $\lambda$ by $\tau = m \lambda$. The constants of motion are: energy ($\cal{E}$), related to the stationarity of the geometry, axial angular momentum ($\Phi$), related to the axial symmetry of the geometry, ‘total’ angular momentum ($K$), related to the hidden symmetry of the geometry. For the equatorial motion, $K$ is restricted through Eq. (\[e18\]) following from the conditions on the latitudinal motion.[@Stu:1983:BULAI:] Notice that $\cal{E}$ and $\Phi$ cannot be interpreted as energy and axial angular momentum at infinity, since the spacetime is not asymptotically flat.
The equatorial motion is governed by the constants of motion ${\cal{E}},\ \Phi$. Its properties can be conveniently determined by an “effective potential” given by the condition $R(r) = 0$ for turning points of the radial motion. It is useful to define specific energy and specific angular momentum by the relations $$\label{e21}
E \equiv \frac{I {\cal{E}}}{m}, \ L \equiv \frac{I \Phi}{m}.$$ Solving the equation $R(r) = 0$, we find the effective potential in the form $$\begin{aligned}
E_{(\pm)}(r;L,a,y) &\equiv&
\left[\left(1+y a^2 \right)r\left(r^2 +a^2 \right)
+ 2a^2 \right]^{-1}\nonumber\\
&\times&
\bigg\{a\left[yr\left(r^2+a^2 \right)+2
\right]L\nonumber\\
&\pm& \left.\Delta^{1/2}_r
\left\{r^2 L^2 +r\left[\left(1+ya^2\right)r\left(r^2+a^2
\right)+ 2a^2 \right] \right\}^{1/2}\right\}. \label{e23}\end{aligned}$$ In the stationary regions ($\Delta_r \geq 0$), the motion is allowed where[@Bic-Stu-Bal:1989:BULAI:] $$\label{e24}
E \geq E_{(+)}(r;L,a,y).$$
The equatorial circular orbits can be determined by solving simultaneously the equations $R(r) = 0, dR/dr = 0$. The specific energy of the orbits is given by $$\label{e38}
E_{\pm}(r;a,y) = \frac{1-\frac{2}{r}- \left(r^2+ a^2 \right)y
\pm a \left(\frac{1}{r^3}- y\right)^{1/2}}
{\left[1- \frac{3}{r}- a^2y\pm 2a \left(\frac{1}{r^3}- y
\right)^{1/2} \right]^{1/2}},$$ while the specific angular momentum of the orbits is determined by $$L_{\pm}(r;a,y) =
- \frac{2a +ar\left(r^2 +a^2 \right)y
\mp r\left(r^2 +a^2\right)
\left(\frac{1}{r^3} -y \right)^{1/2}}
{r \left[1 -\frac{3}{r}- a^2y\pm 2a
\left(\frac{1}{r^3} -y\right)^{1/2}
\right]^{1/2}}.
\label{e40}$$ The relations (\[e38\])–(\[e40\]) determine two families of the circular orbits. We call them plus-family orbits and minus-family orbits[@Stu-Sla:2004:PHYSR4:] according to the $\pm$ sign in the relations (\[e38\])–(\[e40\]). Inspecting expressions (\[e38\]) and (\[e40\]), we find two reality restrictions on the circular orbits. The first one is given by the relation $$\label{e49}
y \leq y_{\mathrm{s}} \equiv \frac{1}{r^3},$$ which introduces the notion of the “static radius”, given by the formula $r_{\mathrm{s}} = y^{-1/3}$ independently of the rotational parameter $a$. It can be compared with formally identical result in the Schwarzschild–de Sitter spacetimes.[@Stu-Hle:1999:PHYSR4:] A “free” or “geodetical” observer on the static radius has only $U^t$ component of 4-velocity being non-zero. The position on the static radius is unstable relative to radial perturbations. The second restriction is given by the condition $$\label{e50}
1 -\frac{3}{r} -a^2y \pm 2a\left(\frac{1}{r^3}
-y \right)^{1/2} \geq 0;$$ the equality determines photon circular orbits with $E \to \infty$ and $L \to
\pm \infty$. The photon circular orbits of the plus-family are given by the relation $$\label{e51}
a = a^{(+)}_{\mathrm{ph(1,2)}}(r;y) \equiv \frac{\left(1 -yr^3\right)^{1/2}
\pm \left(1 -3yr^2 \right)^{1/2}}
{yr^{3/2}},$$ while for the minus-family orbits they are given by $$\label{e52}
a = a^{(-)}_{\mathrm{ph(1,2)}}(r;y) \equiv \frac{-\left(1 -yr^3 \right)^{1/2}
\pm \left(1 -3yr^2 \right)^{1/2}}
{yr^{3/2}}.$$ A detailed discussion of the photon circular orbits can be found in Refs. .
The behaviour of circular orbits in the field of Kerr black holes ($y=0$) suggests that the plus-family orbits correspond to the co-rotating orbits, while the minus-family circular orbits correspond to the counter-rotating ones. However, this statement is not correct even for Kerr naked-singularity spacetimes with the rotational parameter low enough, where counter-rotating plus-family orbits could exist nearby the ring singularity.[@Stu:1980:BULAI:] In the KdS spacetimes we cannot identify the plus-family circular orbits with purely co-rotating orbits even in the black-hole spacetimes; moreover, it is not possible to define the co-rotating (counter-rotating) orbits in relation to stationary observers at infinity, as can be done in the Kerr spacetimes, since the KdS spacetimes are not asymptotically flat.
Orientation of the circular orbits in the KdS spacetimes must be related to locally non-rotating frames (LNRF), similarly to the case of asymptotically flat Kerr spacetimes. In the KdS spacetimes, the tetrad of 1-forms corresponding to the LNRF is given by Ref. :
$$\begin{aligned}
\omega^{(t)}&\equiv&\left
(\frac{\Delta_{r}\Delta_{\theta}\varrho^{2}}{I^{2}A}\right )^{1/2}{\rm d}t,
\nonumber \\
\omega^{(r)}&\equiv&\left (\frac{\varrho^{2}}{\Delta_{r}}\right )^{1/2}{\rm
d}r, \nonumber\end{aligned}$$
$$\begin{aligned}
\omega^{(\phi)}&\equiv&\left (\frac{A\sin^{2}\theta}{I^{2}\varrho^{2}}\right
)^{1/2}({\rm d}\phi-\Omega {\rm d}t), \\
\omega^{(\theta)}&\equiv&\left (\frac{\varrho^{2}}{\Delta_{\theta}}\right
)^{1/2}{\rm d}\theta, \end{aligned}$$
2exwith the angular velocity of the LNRF being given by $$\Omega\equiv\frac{{\rm d}\phi}{{\rm d}t}=\frac{a}{A}\left
[-\Delta_{r}+(r^{2}+a^{2})\Delta_{\theta}\right ];\quad
A\equiv (r^{2}+a^{2})^{2}-a^{2}\Delta_{r}.$$
Locally measured components of 4-momentum in the LNRF are given by the projection of a particle’s 4-momentum onto the tetrad $$p^{(\alpha)}=p^{\mu}\omega^{(\alpha)}_{\mu}, \quad
p^{\mu}=m\frac{{\rm d}x^{\mu}}{{\rm d}\tau}\equiv m\dot{x}^{\mu}=\frac{{\rm
d}x^{\mu}}{{\rm d}\lambda}.$$ A simple calculation reveals the intuitively anticipated relation $$p^{(\phi)}=\frac{mr}{A^{1/2}}L.$$ We can see that the sign of the azimuthal component of the 4-momentum measured in the LNRF is given by the sign of the specific angular momentum of a particle on the orbit of interest. Therefore the circular orbits with $p^{(\phi)}>0,\ (L>0)$, we call co-rotating, and the circular orbits with $p^{(\phi)}<0,\ (L<0)$ we call counter-rotating, in agreement with the case of asymptotically flat Kerr spacetimes.
The circular geodesics can be astrophysically relevant, if they are stable with respect to radial perturbations. The loci of the stable circular orbits are given by the condition $$\frac{\d^2 R}{\d r^2} \geq 0
\label{e56}$$ that has to be satisfied simultaneously with the conditions $R(r) =0$ and $\d R/\d r = 0$ determining the circular orbits. The radii of the stable orbits of both families are restricted by the condition[@Stu-Sla:2004:PHYSR4:] $$r \left[6 -r +r^3 (4r -15)y \right] \mp
8a \left[r\left(1 -yr^3 \right)^3\right]^{1/2} +
a^2 \left[3 +r^2y\left(1 -4yr^3 \right)\right] \geq 0.
\label{e57}$$ The marginally stable orbits of both families are described by the relation $$\begin{aligned}
a^2 &=& a^2_{\mathrm{ms(1,2)}}(r;y) \equiv
\left[3 +r^2y \left(1-4yr^3\right)\right]^{-2}r
\bigg\{\left[r -6 -r^3 (4r -15)y \right]\nonumber \\
&\times&
\left[3 +r^2y \left(1 -4yr^3 \right)\right]
+32 \left(1 -yr^3 \right)^3 \pm 8 \left(1-yr^3\right)^{3/2}
\left(1 -4yr^3 \right)^{1/2}\nonumber \\
&\times& \left.\left\{r \left[3 -ry \left(6 +10r -15yr^3\right) \right]
-2 \right \}^{1/2}\right \}. \label{e58}\end{aligned}$$ The ($\pm$) sign in Eq. (\[e58\]) is not directly related to the plus-family and the minus-family orbits. The function $a^2_{\mathrm{ms(1)}}$, corresponding to the $(+)$ sign in Eq. (\[e58\]), determines marginally stable orbits of the plus-family, while the function $a^2_{\mathrm{ms(2)}}$, corresponding to the $(-)$ sign in Eq. (\[e58\]), is relevant for both the plus-family and minus-family orbits. A detailed analysis shows that the critical value of the cosmological parameter for the existence of the stable (plus-family) orbits is given by $$y_{\mathrm{crit(ms+)}} = \frac{100}{(5 + 2\sqrt{10})^{3}}\doteq 0.06886.
\label{ee1}$$ No stable circular orbits (of any family) exist for $y>y_{\mathrm{crit(ms+)}}$. The critical value of $y$ for the existence of the minus-family stable circular orbits is given by $$y_{\mathrm{crit(ms-)}} = \frac{12}{15^{4}}. \label{ee3}$$ It coincides with the limit on the existence of the stable circular orbits in the SdS spacetimes.[@Stu-Hle:1999:PHYSR4:] In the parameter space $y$–$a^{2}$, separation of the KdS spacetimes according to the existence of stable circular orbits is given in Fig. \[f1\].
Behaviour of the effective potential (\[e23\]) enables us to introduce the notion of marginally bound orbits, i.e., unstable circular orbits where a small radial perturbation causes infall of a particle from the orbit to the centre, or its escape to the cosmological horizon. For some special value of the axial parameter $X = L - aE$, denoted as $X_{\mathrm{mb}}$, the effective potential has two local maxima related by $$E_{(+)}(r_{\mathrm{mb(i)}};X_{\mathrm{mb}},a,y)
=E_{(+)}(r_{\mathrm{mb(o)}};X_{\mathrm{mb}},a,y),$$ corresponding to both the inner and outer marginally bound orbits, see Fig. \[f2\]. For completeness, the figure includes the effective potentials defining both the inner and outer marginally stable orbits. The search for the marginally bound orbits in a concrete KdS spacetime admitting stable circular orbits must be realized in a numerical way. Clearly, in the spacetimes with $y \geq 12/15^4$, the minus-family marginally bound orbits do not exist. In the spacetimes admitting stable plus-family orbits, there is $r_{\rm mb(o)} \sim r_{\rm s}$ but $r_{\rm ms(o)} \sim 0.7 r_{\rm s}$.
![Effective potentials of equatorial radial motion of test particles in black-hole spacetime ($y=10^{-4},\ a=0.6$) allowing stable circular orbits for co-rotating particles. Marginally bound (mb) orbits are given by the solid curve corresponding to $X=X_{mb+}\doteq
2.38445$. The curve has two local maxima of the same value, $E_{mb}\doteq 0.93856$, leading to inner (mb(i)) and outer (mb(o)) marginally bound orbits. The dashed effective potential defines inner marginally stable orbit (ms(i)) by coalescing local minimum and (inner) local maximum, and corresponds to $X=X_{ms(i)+}\doteq
2.20307$ with energy $E_{ms(i)+}\doteq 0.90654$. In an analogous manner the dashed-dotted potential defines outer marginally stable orbit (ms(o)) with energy $E_{ms(o)+}\doteq 0.94451$ corresponding to $X=X_{ms(o)+}\doteq 2.90538$. (Taken from Ref. .)[]{data-label="f2"}](effpots.eps){width=".6\hsize"}
In comparison with the asymptotically flat Kerr spacetimes, where the effect of spacetime rotation vanishes for asymptotically large values of the radius, in the KdS spacetimes the properties of the circular orbits must be treated more carefully, because the rotational effect is relevant in the whole region where the circular orbits are allowed and it survives even at the cosmological horizon.
The minus-family orbits have $L_-<0$ in each KdS spacetime and such orbits are counter-rotating relative to the LNRF.
In the black-hole spacetimes, the plus-family orbits are co-rotating in almost all radii where the circular orbits are allowed except some region in vicinity of the static radius, where they become to be counter-rotating. However, these orbits are unstable. The specific angular momentum and energy of particles located at the static radius, where the plus-family and minus-family orbits coalesce, are given by $$L_{\mathrm{s}} = -a\frac{3y^{1/3}+a^{2}y}{\left (1-3y^{1/3}-a^{2}y
\right )^{1/2}}, \quad
E_{\mathrm{s}} = (1-3y^{1/3}-a^2 y)^{1/2}.$$
In the naked-singularity spacetimes, the plus-family orbits behave in a more complex way. They are always counter-rotating in vicinity of the static radius. Moreover, in the naked singularity spacetimes with the rotational parameter low enough ($a<3\sqrt{3}/4,\ y=0$), stable counter-rotating plus-family circular orbits exist. When $a$ is very close to the extreme hole state ($a<4\sqrt{2}/(3\sqrt{3}),\
y=0$), even stable plus-family orbits with $E < 0$ can exist,[@Stu-Sla:2004:PHYSR4:] see Fig. \[f1\].
Angular velocity $\Omega=\d\phi/\d t$ of a thin, Keplerian accretion disc is given by $$\label{e5}
\Omega_{\rm K\pm} = \pm\frac{1}{r^{3/2}/(1-yr^3)^{1/2} \pm a}.$$ Matter in the thin disc spirals from the outer marginally stable orbit through the sequence of stable circular orbits down to the inner marginally stable orbit losing energy and angular momentum due to the viscosity. The necessary conditions for such a differential rotation $$\label{E6}
\frac{\d\Omega_{\rm K+}}{\d r}<0 \quad \frac{\d L_+}{\d r}\geq 0
\qquad\mbox{or}\qquad
\frac{\d\Omega_{\rm K-}}{\d r}>0 \quad \frac{\d L_-}{\d r}\leq 0,$$ are fulfilled by the relations (\[e40\]) and (\[e5\]). The efficiency of accretion, i.e., the efficiency of conversion of rest mass into heat energy of any element of matter transversing the discs from their outer edge located on the outer marginally stable orbit to their inner edge located on the inner marginally stable orbit is given by $$\eta \equiv E_{\mathrm{ms(o)}}-E_{\mathrm{ms(i)}}.$$ For Keplerian discs co-rotating extreme KdS black holes, the accretion efficiency reaches maximum value of $\eta \sim 0.43$ for the pure Kerr case ($y=0$) and tends to zero for $y \to y_{\rm c(KdS)}\doteq 0.059$, the maximum value of $y$ admitting black holes[^1].
Thick discs
===========
Basic properties of thick discs are determined by equilibrium configurations of perfect fluid. Stress-energy tensor of perfect fluid is given by $$T^\mu_{\hphantom{\mu}\nu} = (p+\epsilon) U^\mu U_\nu + p\,\delta^\mu_\nu
\label{eq1}$$ where $\epsilon$ and $p$ denote total energy density and pressure of the fluid, $U^{\mu}$ is its four velocity. We shall consider test perfect fluid rotating in the $\phi$ direction, i.e., $U^{\mu} = \left( U^{t}, U^{\phi}, 0, 0\right)$. The rotating fluid can be characterized by the vector fields of the angular velocity $\Omega \left( r, \theta \right)$ and the angular momentum density $\ell \left( r, \theta \right)$, defined by $$\Omega = \frac{U^{\phi}}{U^{t}}, \qquad \ell = - \frac{U_{\phi}}{U_{t}}.
\label{eq2}$$ The vector fields are related by the metric coefficients of the KdS spacetime $$\Omega = - \frac{g_{t\phi} + \ell g_{tt}}{g_{\phi\phi} + \ell g_{t\phi}}.
\label{eq3}$$
Projecting the energy-momentum conservation law $T^{\mu\nu}_{\hphantom{\mu\nu};\nu} = 0$ onto the hypersurface orthogonal to the four velocity $U^\mu$ by the projection tensor $h_{\mu\nu} = g_{\mu\nu} +
U_\mu U_\nu$, we obtain the relativistic Euler equation in the form $$\frac{\partial_{\mu} p}{p+\epsilon} =
-\partial_{\mu} (\ln U_t) +
\frac{\Omega\,\partial_{\mu} \ell}{1- \Omega \ell}, \label{eqv6}$$ where $$(U_t)^2 =
\frac{g^2_{t\phi} - g_{tt}\,g_{\phi\phi}}%
{g_{\phi\phi} + 2 \ell g_{t\phi} + \ell^2 g_{tt}}. \label{eqv7}$$
For barytropic perfect fluid, i.e., the fluid with an equation of state $p=p(\epsilon)$, the solution of the relativistic Euler equation can be given by Boyer’s condition determining the surfaces of constant pressure through the “equipotential surfaces” of the potential $W(r,\theta)$ by the relations[@Abr-Jar-Sik:1978:ASTRA:] $$\begin{aligned}
&&\int_0^p \frac{\d p}{p+ \epsilon} = W_\mathrm{in} - W =
\ln (U_t)_\mathrm{in} - \ln (U_t) +
\int_{\ell_\mathrm{in}}^{\ell}
\frac{\Omega\,\d\ell}{1- \Omega \ell}; \label{eqv9}\end{aligned}$$ the subscript “in” refers to the inner edge of the disc.
The equipotential surfaces are determined by the condition $$W(r,\theta) = \mathrm{const},
\label{eqv10}$$ and in a given spacetime can be found from Eq. (\[eqv9\]), if a rotation law $\Omega = \Omega(\ell)$ is given. Equilibrium configurations of test perfect fluid are determined by the equipotential surfaces which can be closed or open. Moreover, there is a special class of critical, self-crossing surfaces (with a cusp), which can be either closed or open. The closed equipotential surfaces determine stationary toroidal configurations. The fluid can fill any closed surface – at the surface of the equilibrium configuration pressure vanishes, but its gradient is non-zero.[@Koz-Jar-Abr:1978:ASTRA:] On the other hand, the open equipotential surfaces are important in dynamical situations, [*e.g.*]{}, in modeling of jets.[@LyB:1969:NATURE:; @Bla:1987:300YoG:] The critical, self-crossing closed equipotential surfaces $W_\mathrm{cusp}$ are important in the theory of thick accretion discs, because accretion onto the black hole through the cusp of the equipotential surface, located in the equatorial plane, is possible due to a little overcoming of the critical equipotential surface by the surface of the disc (Paczyński mechanism). Accretion is thus driven by a violation of the hydrostatic equilibrium, rather than by viscosity of the accreting matter.[@Koz-Jar-Abr:1978:ASTRA:]
It is well known that all characteristic properties of the equipotential surfaces for a general rotation law are reflected by the equipotential surfaces of the simplest configurations with uniform distribution of the angular momentum density $\ell$, see Ref. . Moreover, these configurations are very important astrophysically, because they are marginally stable.[@Seg:1975:ASTRJ2:] Under the condition $$\ell(r,\theta) = \mathrm{const},
\label{eqv11}$$ a simple relation for the equipotential surfaces follows from Eq. (\[eqv9\]): $$W(r, \theta) = \ln U_t (r, \theta). \label{eqv12}$$
The equipotential surfaces $\theta = \theta(r)$ are given by the relation $$\oder{\theta}{r} =
-\frac{\partial p/\partial r}{\partial p/\partial \theta}, \label{eqv13}$$ which for the configurations with $\ell = \mathrm{const}$ reduces to $$\oder{\theta}{r} =
-\frac{\partial U_t/\partial r}{\partial U_t/\partial\theta}.\label{eqv14}$$ In the KdS spacetimes there is $$W \left(r, \theta \right) = \ln \left\{ \frac{\rho}{I} \cdot
\frac{\Delta^{1/2}_r \Delta^{1/2}_{\theta}
\sin \theta}{\left[\Delta_{\theta}
\sin^{2}\theta
\left(r^2 + a^2 - a\ell \right)^{2} -
\Delta_{r}
\left(\ell - a\sin^{2}\theta
\right)^{2}\right]^{1/2}}\right\}.$$ The best insight into the $\ell = \mbox{const}$ configurations is given by properties of $W\left(r, \theta\right)$ in the equatorial plane ($\theta =
\pi/2$). The reality conditions of $W\left(r, \theta = \pi/2\right)$ imply $$\ell_{\rm ph-} < \ell < \ell_{\rm ph+},
\label{eq5}$$ where the functions $\ell_{\rm ph \pm} \left(r; a, y\right)$, given by $$\ell_{\rm ph\pm}(r;a,y) = a + \frac{r^2}{a\pm\sqrt{\Delta_r}},
\label{eq7}$$ determine the photon geodesic motion.[@Stu-Hle:2000:CLAQG:; @Stu-Sla:2004:PHYSR4:]
Condition for the local extrema of the potential $W\left(r, \theta =
\pi/2\right)$ is identical with the condition of vanishing of the pressure gradient $\left(\lpder{U_t}{r} = 0 =
\lpder{U_t}{\theta}\right)$. Since in the equatorial plane there is $\lpder{U_t}{\theta} = 0$, independently of $\ell = \mbox{const}$, the only relevant condition is $\lpder{U_t}{r} = 0$, which implies the relation $$\ell=\ell_{\rm K \pm}(r;a,y)$$ with $\ell_{\rm K \pm}$ being the angular momentum density of the geodetical Keplerian orbits $$\ell_{\rm K \pm} \left(r; a, y\right) \equiv \pm
\frac{(r^{2}+a^{2})(1-yr^{3})^{1/2}\mp ar^{1/2}[2+yr(r^{2}+a^{2})]}%
{r^{3/2}[1-y(r^{2}+a^{2})]-2r^{1/2}\pm a(1-yr^{3})^{1/2}}.
\label{eq12}$$ The closed equipotential surfaces, and surfaces with a cusp allowing the outflow of matter from the disc, are permitted in those parts of the functions $\ell_{\rm K \pm} \left(r; a, y\right)$ enabling the existence of stable circular geodesics corresponding to the centre of the equilibrium configurations. Stationary toroidal configurations exist if $\ell\in (\ell_{\rm ms(i)},\ell_{\rm ms(o)})$. We can distinguish three kinds of discs (Fig. \[f3\]):
accretion discs:
: $\ell\in (\ell_{\rm ms(i)},\ell_{\rm mb})$; the last closed surface is self-crossing in the inner cusp, another critical surface self-crossing in the outer cusp is open.
marginally bound accretion discs:
: $\ell=\ell_{\rm mb}$; the last closed surface is self-crossing in both the inner and the outer cusp.
excretion discs:
: $\ell\in (\ell_{\rm mb},\ell_{\rm ms(o)})$; the last closed surface is self-crossing in the outer cusp, another critical surface self-crossing in the inner cusp is open.
![Typical behaviour of equipotential surfaces (meridional sections) in the KdS black-hole spacetimes. Light gray region contains closed equipotential surfaces. The last closed surface is self-crossing in the cusp(s). Possible toroidal configurations correspond to: (a) accretion discs, (b) marginally bound accretion discs and (c) excretion discs.[]{data-label="f3"}](disk.eps){width="1\hsize"}
Conclusions
===========
For astrophysically relevant black holes ($M < 10^{12}M_{\odot}$) and the observed RRCC (\[E3\]), the cosmological parameter is so small ($y < 10^{-22}$) that both co-rotating and counter-rotating discs can exist around KdS black holes. The efficiency of the accretion process is then extremely close to the values relevant for Kerr black holes. The efficiency is strongest for thin, Keplerian discs orbiting extreme black holes. It is suppressed for $a$ descending and/or for $\ell = \mbox{const}$ growing from $\ell_{\rm ms(i)}$ up to $\ell_{\rm mb}$. Notice that the co-rotating toroidal discs are steeper and more extended than the counter-rotating discs.
The crucial effects caused by the RRCC are illustrated in (Fig. \[f4\]).
- The outer edge of the discs. The presence of an outer cusp of toroidal discs nearby the static radius enables outflow of mass and angular momentum from the discs due to a violation of mechanical equilibrium. Recall that such an outflow is impossible from discs around isolated black holes in asymptotically flat spacetimes.[@Koz-Jar-Abr:1978:ASTRA:]
- Strong collimation effect on jets escaping along the rotational axis of toroidal discs indicated by open equipotential surfaces that are narrowing strongly after crossing the static radius.
![Shapes of thick discs and collimation of jets due to a cosmic repulsion. The effect of collimation is relevant near the static radius and further. Left picture depicts thick accretion discs orbiting the Kerr black hole ($y=0,\ a^2=0.99;\ \ell\approx\ell_{\rm mb}$) and the Schwarzschild black hole ($y=0,\ a=0;\ \ell\approx\ell_{\rm mb}$), right picture depicts thick marginally bound accretion discs orbiting the KdS black hole ($y=10^{-6},\ a^2=0.99;\ \ell=\ell_{\rm mb}$) and the SdS black hole ($y=10^{-6},\ a=0;\ \ell=\ell_{\rm mb}$).[]{data-label="f4"}](disks-jets.eps){width=".9\hsize"}
We can give to our results proper astrophysical relevance by presenting numerical estimates for observationally established current value of the RRCC[^2]. Having the value of $\Lambda_{0} \approx 1.3 \times 10^{-56}{\rm cm}^{-2}$, we can determine the mass parameter of the spacetime corresponding to any value of $y$, parameters of the equatorial circular geodesics and basic characteristics of both the thin and thick accretion discs (Table \[t1\]). Outer edge of the marginally bound thick accretion disc is determined by the outer marginally bound circular orbit which is located very close to, and for presented values of $y$ almost at the static radius of a given spacetime.
\[t1\]
It is well known[@Car-Ost:1996:ModAst:] that dimensions of accretion discs around stellar-mass black holes ($M
\sim 10 M_\odot$) in binary systems are typically $10^{-3}$pc, dimensions of large galaxies with central black-hole mass $M \sim 10^8 M_\odot$, of both spiral and elliptical type, are in the interval 50–100 kpc, and the extremely large elliptical galaxies of cD type with central black-hole mass $M \sim 3\times 10^9 M_\odot$ extend up to 1 Mpc. Therefore, we can conclude that the influence of the RRCC is quite negligible in the accretion discs in binary systems of stellar-mass black holes as the static radius exceeds in many orders dimension of the binary systems. But it can be relevant for accretion discs in galaxies with large active nuclei as the static radius puts limit on the extension of the discs well inside the galaxies. Moreover, the agreement (up to one order) of the dimension of the static radius related to the mass parameter of central black holes at nuclei of large galaxies with extension of such galaxies suggests that the RRCC could play an important role in formation and evolution of such galaxies. Of course, the first step in confirming such a suggestion is modelling of the influence of the RRCC on self-gravitating accretion discs.
Acknowledgments {#acknowledgments .unnumbered}
===============
The present work was supported by the Czech grant MSM 4781305903 and by the Committee for Collaboration of Czech Republic with CERN. The author would like to acknowledge Drs Stanislav Hledík and Petr Slaný for collaboration and the excellent working conditions at the CERN’s Theory Division and SISSA’s Astrophysics Sector, respectively, where part of the work was realized.
[10]{}
N. Bahcall, J. P. Ostriker, S. Perlmutter and P. J. Steinhardt, , 1481 (1999).
E. W. Kolb and M. S. Turner, (Addison-Wesley, Redwood City, California, 1990), The Advanced Book Program.
D. N. Spergel [*et al.*]{}, , 175 (2003).
B. W. Carroll and D. A. Ostlie, (Addison-Wesley, Reading, Massachusetts, 1996).
I. D. Novikov and K. S. Thorne, in [*[Black Holes]{}*]{}, eds. C. De Witt and B. S. De Witt (Gordon and Breach, New York–London–Paris, 1973), pp. 291–450.
M. Koz[ł]{}owski, M. Jaroszy[ń]{}ski and M. A. Abramowicz, , 209 (1978).
M. A. Abramowicz, M. Jaroszy[ń]{}ski and M. Sikora, , 221 (1978).
Z. Stuchl[í]{}k and S. Hled[í]{}k, , 044006 (1999).
Z. Stuchl[í]{}k and S. Hled[í]{}k, , 363 (2002).
M. A. Abramowicz and A. R. Prasanna, , 720 (1990).
Z. Stuchl[í]{}k and S. Hled[í]{}k, , 4541 (2000).
S. Hled[í]{}k, in [ *Gravitation: [F]{}ollowing the [P]{}rague [I]{}nspiration ([A]{} [V]{}olume in [C]{}elebration of the 60th [B]{}irthday of [J]{}i[ř]{}[í]{} [B]{}i[č]{}[á]{}k)*]{}, eds. O. Semer[á]{}k, J. Podolsk[ý]{} and M. [Ž]{}ofka (World Scientific, New Jersey, London, Singapore, Hong Kong, 2002), pp. 161–192.
Z. Stuchl[í]{}k, P. Slan[ý]{} and S. Hled[í]{}k, , 425 (2000).
L. Rezzolla, O. Zanotti and J. A. Font, , 603 (2003).
M. A. Abramowicz, M. Calvani and L. Nobili, , 597 (1983).
F. de Felice and Y. Yunqiang, , 1235 (2001).
Z. Stuchl[í]{}k and P. Slan[ý]{}, , 064001 (2004).
D. R. Brill and S. A. Hayward, , 359 (1994).
S. A. Hayward and K.-I. Nakao, , 5080 (1994).
J. M. Bardeen and J. A. Petterson, , L65 (1975).
B. Carter, in [*[Black Holes]{}*]{}, eds. C. De Witt and B. S. De Witt (Gordon and Breach, New York–London–Paris, 1973), pp. 57–214.
Z. Stuchl[í]{}k, , 129 (1983).
J. Bi[č]{}[á]{}k, Z. Stuchl[í]{}k and V. Balek, , 65 (1989).
Z. Stuchl[í]{}k, , 129 (1980).
D. Lynden-Bell, , 690 (1969).
R. D. Blandford, in [*Three hundred years of gravitation*]{}, eds. S. W. Hawking and W. Israel (Cambridge University Press, 1987), p. 277.
M. Jaroszy[ń]{}ski, M. A. Abramowicz and B. Paczy[ń]{}ski, , 1 (1980).
F. H. Seguin, , 745 (1975).
[^1]: Notice that for the plus-family discs orbiting Kerr naked singularities with $a \sim 1$, the efficiency $\eta \sim 1.57$, exceeding strongly the annihilation efficiency. This is caused by strong discontinuity in properties of the plus-family orbits for extreme black holes and naked singularities with $a \to 1$. Conversion of a naked singularity into an extreme black hole leads to an abrupt instability of the innermost parts of the plus-family discs that can have strong observational consequences.[@Stu-Sla:2004:PHYSR4:]
[^2]: For more detailed information in the case of thick discs around Schwarzschild–de Sitter black holes see Ref. , where the estimates for primordial black holes in the early universe with a repulsive cosmological constant related to a hypothetical vacuum energy density connected with the electroweak symmetry breaking or the quark confinement are presented.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: '[In this paper we represent a new form of condition for the consistency of the matrix equation $AXB=C$. If the matrix equation $AXB=C$ is consistent, we determine a form of general solution which contains both reproductive and non-reproductive solutions. Also, we consider applications of the concept of reproductivity for obtaining general solutions of some matrix systems which are in relation to the matrix equation $AXB=C$.]{}'
address:
- '$\,$[Faculty of Civil Engineering, University of Belgrade, Serbia]{}'
- '$\,$[Faculty of Electrical Engineering, University of Belgrade, Serbia]{}'
author:
- '[Biljana Radiči'' c, Branko Maleševi'' c]{}'
title: |
**Some considerations in relation to the\
matrix equation ***A****X****B***$\,$=$\,$***C*****
---
$\!\!\!\!$ [Matrix equation $AXB=C$; reproductive equation; reproductive solution]{}. $\,$[*MSC (2010): $\,15A24$.*]{}
The reproductive equations
==========================
The general concept of the reproductive equations was introduced by S.B. Preši' c [@Presic68] in 1968. In this part of the paper we give the definition of reproductive equations and the most important statements related to the reproductive equations. Using the concept of reproductivity, in the next section, we obtain the general solutions of some matrix systems which are in relation to the matrix equation $AXB=C$.
Let $S$ be a given non-empty set and $J$ be a given unary relation of $S$. Then an equation $J(x)$ is [*consistent*]{} if there is at least one element $x_{0} \in S$, so-called [*the solution*]{}, such that $J(x_{0})$ is true. A formula $x=\phi(t)$, where $\phi: S \longrightarrow S$ is a given function, represents of the equation $J(x)$ if and only if $$(\forall \, t) J(\phi(t)) \wedge (\forall \, x)( J(x) \Longrightarrow (\exists \, t) x
= \phi(t) ).$$
In this part of the paper we give the definition of reproductive equations and the fundamental statements related to the reproductive equations.
*The reproductive equations* are the equations of the following form: $$x=\varphi(x),$$
where $x$ is a unknown, $S$ is a given set and $\varphi:S \longrightarrow S$ is a given function which satisfies the following condition: $$\label{UR}
\varphi\circ\varphi=\varphi.$$
The condition (\[UR\]) is called *the condition of reproductivity*. The fundamental properties of the reproductive equations are given by the following two statements S.B. Prešić [@Presic68] (see also [@Bozic75], and [@Tribute01]).
\[T11\] For any consistent equation $J(x)$ there is an equation of the form $x=\varphi(x)$, which is equivalent to $J(x)$ being in the same time reproductive as well.
\[T12\] If a certain equation $J(x)$ is equivalent to the reproductive one $x=\varphi(x)$, the general solution is given by the formula $x=\varphi(y)$, for any value $y\in S$.
Let us remark that a formula $x=\phi(t)$, where $\phi \!:\!S \longrightarrow S$ is a given function, represents [@Bankovic11] of the equation $J(x)$ if and only if $$(\forall \, t) J(\phi(t)) \wedge (\forall \, t)( J(t) \Longrightarrow t = \phi(t) ).$$
S.B. Preši' c was the first one who considered implementations of reproductivity on some matrix equations [@Presic68] (see also [@Haveric83], [@Haveric84] and [@Presic63]). The concept of reproductivity allows us to analyse various forms of the solution. General applications of the concept of reproductivity were also considered by J.D. Kečki' c in [@Keckic82], [@Keckic83], J.D. Kečki' c and S.B. Preši' c in [@KeckicPresic97], S. Rudeanu in [@Rudeanu78]-[@Rudeanu01] and D. Bankovi' c in [@Bankovic79]-[@Bankovic11].
The matrix equation ***A****X****B***$\,$=$\,$***C***
=====================================================
Let $m, n \in \mathbb{N}$ be natural numbers and $\mathbb{C}$ is the field of complex numbers. The set of all matrices of order $ m \times n $ over $\mathbb{C}$ is denoted by $\mathbb{C}^{m \times n}$. By $\mathbb{C}_{a}^{m \times n}$ we denote the set of all $m \times n$ complex matrices of rank $a$. For $A \in
\mathbb{C}^{m \times n}$, the rank of $A$ is denoted by $\mbox{rank}(A)$. The unit matrix of order $m$ is denoted by $I_m$ (if the dimension of unit matrix is known from the context, we shall omit the index which indicates the dimension and use the symbol $I$). Let $A=[a_{i,j}]\in \mathbb{C}^{m \times n}$. By $A_{i\rightarrow}$ and $A_{\downarrow j}$ we denote the $i$-th row of $A$ and the $j$-th column of $A$, respectively. Therefore, $$A_{i \rightarrow }= ( a_{i,1}, a_{i,2},...,a_{i,n}),\, i=1,...,m$$ and
$$A_{ \downarrow j}=(a_{1,j},a_{2,j},...,a_{m,j})^{T}, \, j=1,...,n .$$
The matrix equation $$\label{AXBC}
AXB=C$$
was considered by many authors ([@Cvetkovic06]-[@Cvetkovic08], [@Haveric83], [@Haveric84], [@Keckic85]-[@MalesevicRadicic12], [@Presic63], [@Presic00], [@Tian10] and [@Tian12]). In the papers [@ADajicJJKoliha]-[@Haveric84], [@Keckic85] and [@Keckic97] the matrix equation (\[AXBC\]) was studied as a part of different matrix systems or as a special case of corresponding matrix equations. Special case of the matrix equation (\[AXBC\]) is the following matrix equation: $$\label{AXAA}
AXA=A.$$ Any solution of this equation is called [*$\{1\}$-inverse*]{} of $A$ and is denoted by $A^{(1)}$. The set of all $\lbrace 1\rbrace$-inverses of $A$ is denoted by $A\{1\}$.
For the matrix $A$, let regular matrices $Q \!\in\! \mathbb{C}^{m \times m}$ and $P \!\in\! \mathbb{C}^{n \times n}$ be determined such that the following equality is true: $$\label{EA}
QAP=E_{a}=
\left[
\begin{array}{c|c}
I_{a} & 0\\\hline
0 & 0
\end {array}
\right],$$ where $a=\mbox{rank}(A)$. In [@Rohde64] C. Rohde showed that the general form of $\{1\}$-inverse $A^{(1)}$ can be represented as: $$\label{AJEDAN}
A^{(1)} =
P
\left[
\begin{array}{c|c}
I_{a} & X_{1}\\\hline
X_{2} & X_{3}
\end {array}
\right]
Q \, ,$$ where $X_{1}$, $X_{2}$ and $X_{3}$ are arbitrary matrices of suitable sizes (see also [@Ben-IsraelGreville03] and [@CampbellMeyer09]). Considerations which follows are described in terms of $\{1\}$-inverse of matrices.
This section of the paper is organized as follows: In subsection [*2.1.*]{} we represent a new form of condition for the consistency of the matrix equation (\[AXBC\]). An extension of Penrose’s theorem related to the general solution of the matrix equation (\[AXBC\]) is given in subsection [*2.2.*]{} Namely, we represent the formula of general solution of the matrix equation (\[AXBC\]) if any particular solution $X_{0}$ is known. In subsection [*2.3.*]{} we give a form of particular solution $X_{0}$ such that the formula of general solution of the matrix equation (\[AXBC\]), which is given in subsection [*2.2.*]{}, is reproductive. The main results of this paper are obtained in subsections [*2.1.*]{}$-$[*2.3.*]{} and additionally in subsection [*2.4.*]{} we give two applications of the concept of reproductivity to some matrix systems which are in relation to the matrix equation (\[AXBC\]).
***2.1.*** Let $A\!\in\!\mathbb{C}_{a}^{m \times n}$, $B\!\in \!\mathbb{C}_{b}^{p \times q}$ and $C\!\in\!\mathbb{C}^{m \times q}$. The matrix $A \!\in \!\mathbb{C}_{a}^{m \times n}$ has $a$ linearly independent rows and $a$ linearly independent columns. Let $T_{A_{r}}$ be a $m \times m$ permutation matrix such that multiplying the matrix $A$ by the matrix $T_{A_{r}}$ on the left, we can permute the rows of the matrix $A$ and let $T_{A_{c}}$ be a $n \times n$ permutation matrix such that multiplying the matrix $A$ by the matrix $T_{A_{c}}$ on the right, we can permute the columns of the matrix $A$. Then, for the matrix $A$ there are permutation matrices $T_{A_{r}}$ and $T_{A_{c}}$ such that the matrix $$\label{AKAPA}
\widehat{A}=T_{A_{r}}AT_{A_{c}}$$ has linearly independent rows and linearly independent columns *at the first $a$ positions*. Analogously, for the matrix $B$ there are permutation matrices $T_{B_{r}}$ and $T_{B_{c}}$ such that the matrix $$\label{BKAPA}
\widehat{B}=T_{B_{r}}BT_{B_{c}}
\vspace*{-0.7 mm}$$ has linearly independent rows and linearly independent columns *at the first $b$ positions.*
The considerations which follow are valid for any choice of matrices $ T_{A_{r}}$, $T_{A_{c}}$, $T_{B_{r}}$ and $T_{B_{c}}$ such that $\widehat{A}$ has linearly independent rows and linearly independent columns at the first $a$ positions and $\widehat{B}$ has linearly independent rows and linearly independent columns at the first $b$ positions. Let $$\label{CKAPA}
\widehat{C}=T_{A_{r}}CT_{B_{c}}.$$ Next, let for the matrices $A$ and $B$ regular matrices $Q_{1}, \, P_{1}$ and $Q_{2},
\, P_{2}$ be determined such that the following equalities are true: $$\label{EAEB}
Q_{1}AP_{1}=
E_{a}=
\left[
\begin{array}{c|c}
I_{a} & 0\\\hline
0 & 0
\end {array}
\right]
\qquad \mbox{and} \qquad
Q_{2}BP_{2}=E_{b}=\left[
\begin{array}{c|c}
I_{b} & 0\\\hline
0 & 0
\end {array}
\right]$$ i.e. $$\label{AB}
A=Q_{1}^{-1}E_{a}P_{1}^{-1}
\qquad \mbox{and} \qquad
B=Q_{2}^{-1}E_{b}P_{2}^{-1}.$$
Then, from (\[AKAPA\]), (\[BKAPA\]) and (\[AB\]) we get that $$\qquad \widehat{A}=T_{A_{r}}Q_{1}^{-1}E_{a}P_{1}^{-1}T_{A_{c}}
\qquad \mbox{and} \qquad
\widehat{B}=T_{B_{r}}Q_{2}^{-1}E_{b}P_{2}^{-1}T_{B_{c}} \qquad$$ i.e. $$\quad \widehat{A}=(Q_{1}T_{A_{r}}^{-1})^{-1}E_{a}(T_{A_{c}}^{-1}P_{1})^{-1}
\quad \mbox{and} \quad
\widehat{B}=(Q_{2}T_{B_{r}}^{-1})^{-1}E_{b}(T_{B_{c}}^{-1}P_{2})^{-1}.\quad$$ If we introduce the following notations: $$\label{QPKAPA}
\widehat{Q_{1}} = Q_{1}T_{A_{r}}^{-1}, \;\; \widehat{P_{1}} = T_{A_{c}}^{-1}P_{1}
\quad \mbox{and} \quad
\widehat{Q_{2}} = Q_{2}T_{B_{r}}^{-1}, \;\; \widehat{P_{2}} = T_{B_{c}}^{-1}P_{2}$$ we get that $$\label{ABKAPA}
\widehat{A}=\widehat{Q_{1}}^{-1}E_{a} \widehat{P_{1}}^{-1} \qquad \mbox {and} \qquad
\widehat{B}=\widehat{Q_{2}}^{-1}E_{b}\widehat{P_{2}}^{-1}.$$ Considering Rohde’s general form of $\{1\}$-inverses $A^{(1)}$ and $B^{(1)}$: $$\label{ABJEDAN}
A^{(1)}=
P_{1}
\left[
\begin{array}{c|c}
I_{a} & X_{1}\\\hline
X_{2} & X_{3}
\end {array}
\right]
Q_{1}
\qquad \mbox{and} \qquad
B^{(1)}=
P_{2}
\left[
\begin{array}{c|c}
I_{b} & Y_{1}\\\hline
Y_{2} & Y_{3}
\end{array}
\right]
Q_{2} \, ,$$ where $X_{1}$, $X_{2}$, $X_{3}$ and $Y_{1}$, $Y_{2}$, $Y_{3}$ are arbitrary matrices of suitable sizes, we obtain that: $$\qquad \widehat{A}^{(1)}=
\widehat{P_{1}}
\left[
\begin{array}{c|c}
I_{a} & X_{1}\\\hline
X_{2} & X_{3}
\end{array}
\right]
\widehat{Q_{1}}
\qquad \mbox{and} \qquad
\widehat{B}^{(1)}=
\widehat{P_{2}}
\left[
\begin{array}{c|c}
I_{b} & Y_{1}\\\hline
Y_{2} & Y_{3}
\end {array}
\right]\widehat{Q_{2}} \qquad$$ i.e. $$\label{AAJEDANKAPA}
\widehat{A}\widehat{A}^{(1)}=
\widehat{Q_{1}}^{-1}E_{a}\widehat{P_{1}}^{-1}\widehat{P_{1}}
\left[
\begin{array}{c|c}
I_{a} & X_{1}\\\hline
X_{2} & X_{3}
\end{array}
\right]
\widehat{Q_{1}}=
\widehat{Q_{1}}^{-1}
\left[
\begin{array}{c|c}
I_{a} & X_{1}\\\hline
0 & 0
\end{array}
\right]
\widehat{Q_{1}}$$
and $$\label{BJEDANBKAPA}
\widehat{B}^{(1)}\widehat{B}=
\widehat{P_{2}}
\left[
\begin{array}{c|c}
I_{b} & Y_{1}\\\hline
Y_{2} & Y_{3}
\end {array}
\right]
\widehat{Q_{2}}\widehat{Q_{2}}^{-1}E_{b}\widehat{P_{2}}^{-1}=
\widehat{P_{2}}
\left[
\begin{array}{c|c}
I_{b} & 0\\\hline
Y_{2} & 0
\end {array}
\right]\widehat{P_{2}}^{-1}.$$
As we mentioned, the matrix $\widehat{A}$ has linearly independent rows and linearly independent columns at the first $a$ positions and the matrix $\widehat{B}$ has linearly independent rows and linearly independent columns at the first $b$ positions.
Let $$\label{VRSTEAKAPA}
\widehat{A}_{i \rightarrow}= \sum_{l=1}^{a}\alpha_{i,l} \widehat{A}_{l \rightarrow},
\quad i=a+1,...,m \, ,$$
$$\label{KOLONEAKAPA}
\widehat{A}_{\downarrow j}= \sum_{k=1}^{a}\alpha'_{k,j}\widehat{A}_{\downarrow k},
\quad j=a+1,...,n \, ,$$
and
$$\label{VRSTEBKAPA}
\widehat{B}_{i \rightarrow}=\sum_{l=1}^{b}\beta'_{i,l}\widehat{B}_{l \rightarrow},
\qquad i=b+1,...,p \, ,$$
$$\label{KOLONEBKAPA}
\widehat{B}_{\downarrow j}=\sum_{k=1}^{b}\beta_{k,j}\widehat{B}_{\downarrow k},\qquad
j=b+1,...,q \, ;$$
for some scalars $\alpha_{i,l}$, $\alpha'_{k,j}$ and $\beta'_{i,l}$, $\beta_{k,j}$. As we know, the matrices $\widehat{Q_{1}}$, $\widehat{P_{1}}$ and $\widehat{Q_{2}}$, $\widehat{P_{2}}$ are not uniquely determined, but we shall use, without loss of generality, their following forms: $$\label{OQPKAPAA}
\widehat{Q_{1}}=
\left[
\begin{array}{c|c}
I_{a} & 0 \\\hline
L_{1} & I_{m-a}
\end{array}
\right], \qquad
\widehat{P_{1}}=
\left[
\begin{array}{c|c}
W^{-1} & L'_{1} \\\hline
0 & I_{n-a}
\end{array}
\right]$$ and $$\label{OQPKAPAB}
\widehat{Q_{2}}=
\left[
\begin{array}{c|c}
U^{-1}& 0 \\\hline
L'_{2} & I_{p-b}
\end{array}
\right], \qquad
\widehat{P_{2}}=
\left[
\begin{array}{c|c}
I_{b} & L_{2} \\\hline
0 & I_{q-b}
\end{array}
\right]$$
for
$\quad L_{1}=\left[
\begin{array}{ccc}
-\alpha_{a+1,1}&...& -\alpha_{a+1,a}\\
.&...& .\\
.&...& .\\
.&...& .\\
-\alpha_{m,1}&...& -\alpha_{m,a}\\
\end{array}
\right]$, $L'_{1}=\left[
\begin{array}{ccc}
-\alpha'_{1,a+1}&...& -\alpha'_{1,n}\\
.&...& .\\
.&...& .\\
.&...& .\\
-\alpha'_{a,a+1}&...& -\alpha'_{a,n}\\
\end{array}
\right]$,
$\quad L'_{2}=\left[
\begin{array}{ccc}
-\beta'_{b+1,1}&...& -\beta'_{b+1,b}\\
.&...& .\\
.&...& .\\
.&...& .\\
-\beta'_{p,1}&...& -\beta'_{p,b}\\
\end{array}
\right]$, $L_{2}=\left[
\begin{array}{ccc}
-\beta_{1,b+1}&...& -\beta_{1,q}\\
.&...& .\\
.&...& .\\
.&...& .\\
-\beta_{b,b+1}&...& -\beta_{b,q}\\
\end{array}
\right]$
and where $W$ is a $a \times a $ submatrix of $\widehat{A}$ such that $\widehat{A}=\!\left[
\begin{array}{c|c}
W & \widehat{A}_{2} \\ \hline
\widehat{A}_{3} & \widehat{A}_{4}
\end{array}
\right]$ and $U$ is a $b \times b$ submatrix of $\widehat{B}$ such that $\widehat{B}=\!\left[
\begin{array}{c|c}
U & \widehat{B}_{2} \\ \hline
\widehat{B}_{3} & \widehat{B}_{4}
\end{array}
\right]$.
Let us emphasize that the following statement is true.
\[L21\] Let $A\!\in\!\mathbb{C}_{a}^{m \times n}$, $B\!\in \!\mathbb{C}_{b}^{p \times q}$, $C\!\in\!\mathbb{C}^{m \times q}$. Suppose that $\widehat{A}$ and $\widehat{B}$ are determined by (\[AKAPA\]) and (\[BKAPA\]). Then, the conditions $$\label{UKABC}
AA^{(1)}CB^{(1)}B=C$$ and $$\label{UKABCKAPA}
\widehat{A}\widehat{A}^{(1)}\widehat{C}\widehat{B}^{(1)}\widehat{B}=\widehat{C}$$ are equivalent.
[**Proof.**]{}$\,$The following equivalences are true $\!\Longleftrightarrow\!$ $\!\Longleftrightarrow\!$ $\!\Longleftrightarrow\!$ . $\diamondsuit$
Let us remark that (\[UKABC\]) is [*Penrose’s condition of consistency*]{} for the matrix equation (\[AXBC\]), [@Penrose55], and if the matrix equation (\[AXBC\]) is consistent, then the formulas of general solution are given in Theorem \[T22\].$\;$and$\;$\[T23\].$\;$In the following statement we give a condition which is equivalent to Penrose’s condition of consistency for the matrix equation (\[AXBC\]). So, we can use this new condition to test the consistency of the matrix equation (\[AXBC\]).
\[T21\] Let $A\!\in \!\mathbb{C}_{a}^{m \times n}$,$B\!\in\!\mathbb{C}_{b}^{p \times q}$, $C\!\in \!\mathbb{C}^{m \times q}$. Suppose that $\widehat{A}$ and $\widehat{B}$ are determined by (\[AKAPA\]) and (\[BKAPA\]) and that (\[VRSTEAKAPA\])–(\[KOLONEBKAPA\]) are satisfied. Then, the condition (\[UKABC\]) is true for any choice of *{*$\!$1*}*-inverses $A^{(1)}$ and $B^{(1)}$ iff $$\label{MatrixC}
\mbox{
$\widehat{C}$=$\left[
\begin{array}{cccccc}
c_{1,1}\!&\!...\!&\!c_{1,b}\!&\!\mbox{\scriptsize
$\displaystyle\sum_{k=1}^{b}$}\beta_{k,b+1}c_{1,k}\!&\!...\!&\mbox{\scriptsize
$\displaystyle\sum_{k=1}^{b}$}\beta_{k,q}c_{1,k}\\[-0.75 ex]
.\!&\!...\!&.&.&\!...\!&.\\[-0.75 ex]
.\!&\!...\!&.&.&\!...\!&.\\[-0.75 ex]
.\!&\!...\!&.&.&\!...\!&.\\[-0.75 ex]
c_{a,1}\!&\!...\!&c_{a,b}&\mbox{\scriptsize
$\displaystyle\sum_{k=1}^{b}$}\beta_{k,b+1}c_{a,k}&\!...\!&\mbox{\scriptsize
$\displaystyle\sum_{k=1}^{b}$}\beta_{k,q}c_{a,k} \\[1.75 ex]
\mbox{\scriptsize $\displaystyle\sum_{l=1}^{a}$}\alpha_{a+1,l}c_{l,1}\!&\!...\!&
\mbox{\scriptsize $\displaystyle\sum_{l=1}^{a}$}\alpha_{a+1,l}c_{l,b}\!&
\begin{small}\mbox{\scriptsize $\displaystyle\sum_{l=1}^{a}$}\mbox{\scriptsize
$\displaystyle\sum_{k=1}^{b}$}\alpha_{a+1,l}\beta_{k,b+1}c_{l,k}\end{small} &\!...\!&
\begin{small}\mbox{\scriptsize $\displaystyle\sum_{l=1}^{a}$}\mbox{\scriptsize
$\displaystyle\sum_{k=1}^{b}$} \alpha_{a+1,l}\beta_{k,q}c_{l,k}\end{small}\\[-0.75 ex]
.&\!...\!& .& .&\!...\!& .\\[-0.75 ex]
.&\!...\!& .& .&\!...\!& .\\[-0.75 ex]
.&\!...\!& .& .&\!...\!& .\\[-0.75 ex]
\mbox{\scriptsize $\displaystyle\sum_{l=1}^{a}$}\alpha_{m,l}c_{l,1}&\!...\!&
\mbox{\scriptsize $\displaystyle\sum_{k=1}^{b}$}\alpha_{m,l}c_{l,b}&
\begin{small}\mbox{\scriptsize $\displaystyle\sum_{l=1}^{a}$}\mbox{\scriptsize
$\displaystyle\sum_{k=1}^{b}$}\alpha_{m,l}\beta_{k,b+1}c_{l,k} \end{small}&\!...\!&
\begin{small} \mbox{\scriptsize $\displaystyle\sum_{l=1}^{a}$}\mbox{\scriptsize
$\displaystyle\sum_{k=1}^{b}$}\alpha_{m,l}\beta_{k,q}c_{l,k} \end{small}
\end {array}
\right]$,}$$ where $c_{i,j}$ are arbitrary elements of $\mathbb{C}$.
[**Proof.**]{} [$(\Longrightarrow)$]{}: Suppose that the condition (\[UKABC\]) is valid for any choice of $\{1\}$-inverses $A^{(1)}$ and $B^{(1)}.$ Based on Lemma \[L21\]. the condition (\[UKABCKAPA\]) is also valid. Then, considering the equalities (\[AAJEDANKAPA\]) and (\[BJEDANBKAPA\]), we get the following equality $$\widehat{Q_{1}}^{-1}\left[ \begin{array}{c|c}
I_{a} & X_{1}\\\hline
0 & 0
\end {array}
\right]\widehat{Q_{1}}\widehat{C}\widehat{P_{2}}
\left[
\begin{array}{c|c}
I_{b} & 0\\\hline
Y_{2} & 0
\end {array} \right] \widehat{P_{2}}^{-1}
=
\widehat{C}.$$ By multiplying the previous equality by $ \widehat{Q_{1}}$ on the left and by $\widehat{P_{2}}$ on the right we get $$\label{QCPKAPA}
\left[
\begin{array}{c|c}
I_{a} & X_{1}\\\hline
0 & 0
\end {array}
\right]
\widehat{Q_{1}}\widehat{C}\widehat{P_{2}}
\left[
\begin{array}{c|c}
I_{b} & 0\\\hline
Y_{2} & 0
\end {array}
\right]=
\widehat{Q_{1}}\widehat{C}\widehat{P_{2}}.$$
Suppose that
$$\widehat{C}=
\left[\begin{array}{ccc}
c_{1,1}&...& c_{1,q}\\
.&...& .\\
.&...& .\\
.&...& .\\
c_{m,1}&...& c_{m,q}\\
\end {array}\right] \! .$$
We are going to show that $\widehat{C}$ has the form (\[MatrixC\]). Let $$\label{EF}
E=\widehat{Q_{1}}\widehat{C}\widehat{P_{2}} \qquad \mbox {and} \qquad
F=\left[ \begin{array} {c|c}
I_{a} & X_{1}\\\hline
0 & 0
\end {array}
\right]E\left[
\begin{array}{c|c}
I_{b} & 0 \\\hline
Y_{2} & 0
\end {array}
\right].$$ From (\[OQPKAPAA\]) and (\[OQPKAPAB\]) we obtain that
for $i=1,...,a, j=1,...,b$ $E_{i,j}=c_{i,j},$
for $i=1,...,a, j=b+1,...,q$ $E_{i,j}=c_{i,j}-\displaystyle\sum_{k=1}^{b}\beta_{k,j}c_{i,k},$
for $i=a+1,...,m, j=1,...,b$ $E_{i,j}=c_{i,j}-\displaystyle\sum_{l=1}^{a}\alpha_{i,l}c_{l,j},$
for $i=a+1,...,m, j=b+1,...,q$ $E_{i,j}=c_{i,j}-\displaystyle\sum_{l=1}^{a}\alpha_{i,l}c_{l,j}
-\displaystyle\sum_{k=1}^{b}\beta_{k,j}(c_{i,k}-\displaystyle\sum_{l=1}^{a}\alpha_{i,l}c_{l,k})$
and
for $ i=1,...,a, j=1,...,b$ $F_{i,j}=c_{i,j}+
\displaystyle \sum_{\overline{l}=a+1}^{m}x_{i,
\overline{l}}(c_{\overline{l},j}-\displaystyle\sum_{l=1}^{a}\alpha_{\overline{l},l}c_{l,j})$
$+\displaystyle\sum_{\overline{k}=b+1}^{q}y_{\overline{k},j}[ c_{i,
\overline{k}}-\displaystyle\sum_{k=1}^{b}\beta_{k,\overline{k}}c_{i,k}$
$+\displaystyle\sum_{\overline{l}=a+1}^{m}x_{i,\overline{l}}
\lbrace c_{\overline{l},\overline{k}}-\displaystyle\sum_{l=1}^{a}\alpha_{\overline{l},l}c_{l,\overline{k}}$
$-\displaystyle\sum_{k=1}^{b}\beta_{k,\overline{k}}
(c_{\overline{l},k}-\displaystyle\sum_{l=1}^{a}\alpha_{\overline{l},l}c_{l,k})\rbrace ] ,$
for $i=1,...,a, j=b+1,...,q$ $F_{i,j}=0,$
for $ i=a+1,...,m, j=1,...,b$ $F_{i,j}=0,$
for $ i=a+1,...,m, j=b+1,...,q$ $F_{i,j}=0.$
Finally, from (\[QCPKAPA\]) and (\[EF\]) i.e. $E=F$ we get that
for $ i=1,...,a, j=1,...,b$ $c_{i,j}$ are arbitrary elements of $\mathbb{C},$
for $ i=a+1,...,m, j=1,...,b$ $c_{i,j}=\displaystyle\sum_{l=1}^{a}\alpha_{i,l}c_{l,j},$
for $ i=1,...,a, j=b+1,...,q$ $c_{i,j}=\displaystyle\sum_{k=1}^{b}\beta_{k,j}c_{i,k},$
for $ i=a+1,...,m, j=b+1,...,q$ $c_{i,j}=\displaystyle\sum_{l=1}^{a}\sum_{k=1}^{b}\alpha_{i,l}\beta_{k,j}c_{l,k}.$
[$(\Longleftarrow)$]{}: Suppose that the matrix $\widehat{C}$ has the form (\[MatrixC\]). Then, $$\label{OQCPKAPA}
\qquad \widehat{Q_{1}}\widehat{C} \widehat{P_{2}}=...=\left[
\begin{array}{cccccc}
c_{1,1}&...&c_{1,b}&0&...& 0 \\[-0.5 ex]
.&...& .&.&...& . \\[-0.5 ex]
.&...& .&.&...& . \\[-0.5 ex]
.&...& .&.&...& . \\[-0.5 ex]
c_{a,1}&...&c_{a,b}&0&...& 0 \\[-0.0 ex]
0&...& 0&0 &...& 0 \\[-0.5 ex]
.&...& .& .&...& . \\[-0.5 ex]
.&...& .& .&...& . \\[-0.5 ex]
.&...& .& .&...& . \\[-0.5 ex]
0&...& 0&0 &...& 0
\end{array}
\right]$$
and $$\label{POQCPKAPA}
\left[ \begin{array}{c|c}
I_{a} & X_{1}\\\hline
0 & 0
\end {array}
\right]\widehat{Q_{1}}\widehat{C} \widehat{P_{2}}\left[
\begin{array}{c|c}
I_{b}& 0 \\\hline
Y_{2}& 0
\end {array}
\right]=...=\left[
\begin{array}{cccccc}
c_{1,1}&...&c_{1,b}&0&...& 0\\
.&...& .&.&...& .\\
.&...& .&.&...& .\\
.&...& .&.&...& .\\
c_{a,1}&...&c_{a,b}&0&...& 0\\
0&...& 0&0 &...& 0\\
.&...& .& .&...& .\\
.&...& .& .&...& .\\
.&...& .& .&...& .\\
0&...& 0&0 &...& 0
\end{array}
\right].$$ From (\[OQCPKAPA\]) and (\[POQCPKAPA\]) we conclude that
$$\left[ \begin{array}{c|c}
I_{a} & X_{1}\\\hline
0 & 0
\end {array}
\right]\widehat{Q_{1}}\widehat{C}
\widehat{P_{2}}\left[
\begin{array}{c|c}
I_{b}& 0 \\\hline
Y_{2}& 0
\end {array}
\right]= \widehat{Q_{1}}
\widehat{C}
\widehat{P_{2}}.$$
By multiplying the previous equality by $\widehat{Q_{1}}^{-1} $ on the left and by $\widehat{P_{2}}^{-1}$ on the right we obtain the following equality:
$$\label{NOCKAPA}
\widehat{Q_{1}}^{-1}
\left[
\begin{array}{c|c}
I_{a} & X_{1}\\\hline
0 & 0
\end{array}
\right]
\widehat{Q_{1}}\widehat{C}\widehat{P_{2}}
\left[
\begin{array}{c|c}
I_{b}& 0 \\\hline
Y_{2}& 0
\end {array}
\right]\widehat{P_{2}}^{-1}=\widehat{C}.$$
From (\[NOCKAPA\]), considering the equalities (\[AAJEDANKAPA\]) and (\[BJEDANBKAPA\]), we see that the condition (\[UKABCKAPA\]) is true. Based on Lemma \[L21\]. we conclude that the condition (\[UKABC\]) is true. $\diamondsuit$
Let us remark that the general form of matrix $C$, such that the matrix equation (\[AXBC\]) is consistent, always exists. The matrix equation (\[AXBC\]) is consistent for an arbitrary matrix $C$ iff a matrix $A$ has full row rank and a matrix $B$ has full column rank (see also Exercises 10.50 from [@KMAbadirJRMagnus]).
In the paper [@Wang04] author considered some forms which are equivalent to Penrose’s condition of consistency for matrix equation (\[AXBC\]).
The application of Theorem \[T21\] will be illustrated by the following examples.
\[E21\] Let be given the following matrices:
$A= \left[
\begin{array}{rrr}
0 & 0 & 0\\
1 & -3 & 2\\
2 & 1 & -1\\
-1 & -4 & 3\\
3 & -2 & 1
\end{array}
\right]$ and $\;B= \left[
\begin{array}{rrrrr}
0 & 1 & 2 & 3 \!&\!-1\\
0 & 3 & 1 & 4 \!&\! 2\\
0 & 4 & 1 & 5 \!&\! 3\\
0 & 2 & 3 & 5 \!&\!-1
\end {array}
\right].$
Then, $rank(A)$=2, $rank(B)$=2 and for
$T_{A_{r}}=
\left[
\begin{array}{ccccc}
0&1&0&0&0\\
0&0&1&0&0\\
0&0&0&1&0\\
0&0&0&0&1\\
1&0&0&0&0
\end{array}
\right]$ and $\;
T_{B_{c}}= \left[
\begin{array}{ccccc}
0&0&1&0&0\\
1&0&0&0&0\\
0&1&0&0&0\\
0&0&0&1&0\\
0&0&0&0&1
\end{array}
\right]$
we get that $$\widehat{A}=T_{A_{r}}A
=...=
\left[
\begin{array}{rrr}
1 & -3 & 2\\
2 & 1 & -1\\
-1 & -4 & 3\\
3 & -2 & 1\\
0 & 0 & 0
\end {array}
\right]\!,
\;\;
\widehat{B}=BT_{B_{c}}
=... =
\left[
\begin{array}{rrrrr}
1 & 2 & 0 & 3 & -1\\
3 & 1 & 0 & 4 & 2\\
4 & 1 & 0 & 5 & 3\\
2 & 3 & 0 & 5 & -1
\end{array}
\right]\!.$$
Therefore,
$ \widehat{A}_{3\rightarrow}=\widehat{A}_{1\rightarrow}-\widehat{A}_{2\rightarrow}$, $ \widehat{A}_{4\rightarrow}=\widehat{A}_{1\rightarrow}+\widehat{A}_{2\rightarrow}$, $\widehat{A}_{5\rightarrow}=0\widehat{A}_{1\rightarrow}+0\widehat{A}_{2\rightarrow}$
and
$\widehat{B}_{\downarrow 3}=0\widehat{B}_{\downarrow 1}+0\widehat{B}_{\downarrow 2}$, $ \widehat{B}_{\downarrow 4}=\widehat{B}_{\downarrow 1}+\widehat{B}_{\downarrow 2}$, $\widehat{B}_{\downarrow 5}=\widehat{B}_{\downarrow 1}-\widehat{B}_{\downarrow 2}.$
From this we get that
$ \alpha_{3,1}=1,$ $\alpha_{3,2}=-1,$ $ \alpha_{4,1}=1,$ $ \alpha_{4,2}=1,$ $\alpha_{5,1}=0,$ $\alpha_{5,2}=0 $
and
$ \beta_{1,3}=0, $ $ \beta_{2,3}=0 ,$ $ \beta_{1,4}=1, $ $ \beta_{2,4}=1, $ $ \beta_{1,5}=1, $ $ \beta_{2,5}=-1. $
Based on Theorem \[T21\]. each matrix $\widehat{C}$ which has the following form $$\widehat{C}
=
\mbox{
\scriptsize $\left[
\begin{array}{ccccc}
c_{1,1}&c_{1,2}&0&c_{1,1}+c_{1,2}& c_{1,1}-c_{1,2}\\
c_{2,1}&c_{2,2}&0&c_{2,1}+c_{2,2}&c_{2,1}-c_{2,2}\\
c_{1,1}-c_{2,1}&c_{1,2}-c_{2,2}&0&c_{1,1}-c_{2,1}+c_{1,2}-c_{2,2}&c_{1,1}-c_{2,1}-
c_{1,2}+c_{2,2}\\
c_{1,1}+c_{2,1}&c_{1,2}+c_{2,2}&0&c_{1,1}+c_{2,1}+c_{1,2}+c_{2,2}&c_{1,1}+c_{2,1}-
c_{1,2}-c_{2,2}\\
0&0& 0& 0& 0
\end{array}
\right]$
}$$ satisfies the condition (\[UKABCKAPA\]). From that we conclude that each matrix $C=T_{A_{r}}^{-1}\widehat{C}T_{B_{c}}^{-1}$ which has the following form $$C
=
\mbox{
\scriptsize $\left[
\begin{array}{ccccc}
0&0&0&0& 0\\
0&c_{1,1}&c_{1,2}&c_{1,1}+c_{1,2}& c_{1,1}-c_{1,2}\\
0&c_{2,1}&c_{2,2}&c_{2,1}+c_{2,2}& c_{2,1}-c_{2,2}\\
0&c_{1,1}-c_{2,1}&c_{1,2}-c_{2,2}&c_{1,1}-c_{2,1}+c_{1,2}-c_{2,2}&c_{1,1}-c_{2,1}-
c_{1,2}+c_{2,2}\\
0&c_{1,1}+c_{2,1 }&c_{1,2}+c_{2,2}&c_{1,1}+c_{2,1 }+c_{1,2}+c_{2,2}&c_{1,1}+c_{2,1}-
c_{1,2}-c_{2,2}
\end {array}
\right]$
}$$ satisfies the condition (\[UKABC\]). $\blacklozenge$
\[E22\] Let $A$ and $B$ be the matrices as in Example \[E21\]. and
a\) $\;
C=\left[
\begin{array}{rrrrr}
0 & 0 & 0 & 0 & 0 \\
0 & 1 & 0 & 1 & 1 \\
0 &-2 & 2 & 0 &-4 \\
0 & 3 &-2 & 1 & 5 \\
0 &-1 & 2 & 1 &-3
\end {array}
\right]\!,$ b) $\;
C=\left[
\begin{array}{rrrrr}
0 & 0 & 0 & 0 & 0 \\
0 & 1 & 0 & 1 & 1 \\
0 &-2 & 2 & 0 & 4 \\
0 & 3 &-2 & 1 & 5 \\
0 &-1 & 2 & 1 &-3
\end {array}
\right]\!.$
If we compare the matrix $C$ from a) and from b) with the general form of matrix $C$ which satisfies the condition (\[UKABC\]) (see Example \[E21\].) we see that the matrix $C$ from a) satisfies the condition (\[UKABC\]) and the matrix $C$ from b) does not satisfy the condition (\[UKABC\]). Therefore, the matrix equation (\[AXBC\]) is consistent for the matrix $C$ from a), but it is not consistent for the matrix $C$ from b). $\blacklozenge $
***2.2.*** Recall that the matrix equation $AXB=C$ is marked with (\[AXBC\]) for $A\!\in\!
\mathbb{C}_{a}^{m \times n}$, , $C\!\in\!
\mathbb{C}^{m \times q}$. Methods for solving the consistent matrix equation (\[AXBC\]) are considered in the book [@Cullis1913] (Chapter X). In the paper [@Penrose55] R. Penrose proved the following theorem related to the matrix equation (\[AXBC\]).
\[T22\] The matrix equation (\[AXBC\]) is consistent iff for some choice of *{*$\!
$1*}*-inverses $A^{(1)}$ and $B^{(1)}$ of the matrices $A$ and $B$ the condition (\[UKABC\]) is true. The general solution of the matrix equation (\[AXBC\]) is given by the formula
$$\label{ORR}
X=f(Y)=A^{(1)}CB^{(1)}+Y-A^{(1)}AYBB^{(1)},$$
where $Y\!\in\!\mathbb{C}^{n \times p}$ is an arbitrary matrix.
\[R21\] If the matrix equation (\[AXBC\]) is consistent, the equivalence
$$AXB=C
\;\Longleftrightarrow\;
X=f(X)=X-A^{(1)}(AXB-C)B^{(1)}$$
is true. Therefore, the starting equation is equivalent to some reproductive equation. Based on Theorem \[T12\]. we can also conclude that (\[ORR\]) is the general solution of the matrix equation (\[AXBC\]).
In this paper we give a simple extension of Theorem \[T22\].
\[T23\] If $X_{0}$ is any particular solution of the matrix equation (\[AXBC\]), the general solution of the matrix equation (\[AXBC\]) is given by the formula $$\label{OR}
X=g(Y)=X_{0}+Y-A^{(1)}AYBB^{(1)},$$ where $Y\!\in\!\mathbb{C}^{n \times p}$ is an arbitrary matrix. The function $g$ satisfies the condition of reproductivity (\[UR\]) iff $X_{0}=A^{(1)}CB^{(1)}$.
[**Proof.**]{} It is easily to see that the solution of the matrix equation (\[AXBC\]) is given by (\[OR\]). On the contrary, let $ X $ is any solution of the matrix equation (\[AXBC\]), then $$\begin{aligned}
X \!&\!=\!&\! X-A^{(1)}CB^{(1)}+A^{(1)}CB^{(1)} \\
\!&\!=\!&\! X-A^{(1)}AXBB^{(1)}+A^{(1)}AX_{0}BB^{(1)} \\
\!&\!=\!&\! X-A^{(1)}A(X-X_{0})B B^{(1)} \\
\!&\!=\!&\! X_{0}+(X-X_{0})-A^{(1)}A(X-X_{0})BB^{(1)} \\
\!&\!=\!&\! X_{0}+Y-A^{(1)}AYBB^{(1)}=g(Y) \, ,\end{aligned}$$ where $Y=X-X_{0}$. From this we see that every solution $X$ of the matrix equation (\[AXBC\]) can be represented in the form (\[OR\]). Based on the following matrix equality: $$g^{2}(Y)
=
g(Y)+(X_{0}-A^{(1)}CB^{(1)})$$ we see that the function $g$ satisfies the condition (\[UR\]) iff $X_{0}=A^{(1)}CB^{(1)}$. $\diamondsuit$
Using the previous theorem and the appropriate choice of particular solution $X_{0}$ we can obtain the general solutions for different cases of the matrix equation (\[AXBC\]). It was considered in the papers [@Haveric83] and [@Presic63].
The general solution (\[OR\]) of the matrix equation (\[AXBC\]) is reproductive iff $X_{0}=A^{(1)}CB^{(1)}$. Therefore, [*Penrose’s general solution*]{} (\[ORR\]) of the matrix equation (\[AXBC\]) is the reproductive solution. If the condition (\[UKABC\]) is not true, the matrix equation (\[AXBC\]) is solved approximately as described in the paper [@Penrose55] and books [@KMAbadirJRMagnus], [@Ben-IsraelGreville03] and [@CampbellMeyer09].
***2.3.*** Using the obtained form of matrix $\widehat{C}$ we obtain the form of particular solution $X_{0}$ of the matrix equation (\[AXBC\]) such that the general solution (\[OR\]) of the matrix equation (\[AXBC\]) is reproductive.
\[T24\] Let $X_{0}$ any particular solution of the matrix equation (\[AXBC\]). The general solution (\[OR\]) of the matrix equation (\[AXBC\]) is reproductive iff $$\label{Form_X_0}
X_{0}=P_{1}
\left[
\begin{array}{c|c}
C_{1} & C_{1}Y_{1}\\\hline
X_{2}C_{1} & X_{2}C_{1}Y_{1}
\end {array}
\right]Q_{2}$$ where $P_{1}$, $Q_{2}$, $ X_{2} $, $ Y_{1}$ are the matrices from (\[ABJEDAN\]) and $C_{1} $ is the submatrix of the matrix $\widehat{C}$ and it has the following form: $$C_{1}=\left[
\begin{array}{ccc}
c_{1,1}&...& c_{1,b}\\
.&...& .\\
.&...& .\\
.&...& .\\
c_{a,1}&...& c_{a,b}\\
\end {array}
\right],$$ where $c_{i,j}$ are some elements of $\mathbb{C}$.
[**Proof.**]{} For general $\{1\}$-inverses $A^{(1)}$ and $B^{(1)}$ the statement follows from Theorem \[T23\]. because $$\begin{array}{rcl}
X_{0}
&\!\!\!=\!\!\!&
A^{(1)}C B^{(1)} \mathop{=} \limits_{(\ref{ABJEDAN})} P_{1}
\left[
\begin{array}{c|c}
I_{a} & X_{1}\\\hline
X_{2} & X_{3}
\end {array}
\right]
Q_{1}CP_{2}
\left[
\begin{array}{c|c}
I_{b} & Y_{1}\\\hline
Y_{2} & Y_{3}
\end {array}
\right]Q_{2} \\[1.5 ex]
&\!\!\!\mathop{=} \limits_{(\ref{CKAPA})}\!\!\!&
P_{1}
\left[
\begin{array}{c|c}
I_{a} & X_{1}\\\hline
X_{2} & X_{3}
\end {array}
\right]Q_{1}T_{A}^{-1}\widehat{C}T_{B}^{-1} P_{2}
\left[
\begin{array}{c|c}
I_{b} & Y_{1}\\\hline
Y_{2} & Y_{3}
\end {array}
\right]Q_{2} \\[1.5 ex]
&\!\!\!\mathop{=} \limits_{(\ref{QPKAPA})}\!\!\!&
P_{1}
\left[
\begin{array}{c|c}
I_{a} & X_{1}\\\hline
X_{2} & X_{3}
\end {array}
\right]\widehat{Q_{1}}\widehat{C}\widehat{ P_{2}}
\left[
\begin{array}{c|c}
I_{b} & Y_{1}\\\hline
Y_{2} & Y_{3}
\end {array}
\right]Q_{2} \\[1.5 ex]
&\!\!\!\mathop{=} \limits_{(\ref{OQCPKAPA})}\!\!\!&
P_{1}
\left[
\begin{array}{c|c}
I_{a} & X_{1}\\\hline
X_{2} & X_{3}
\end {array}
\right]
\left[\begin{array}{c|c}
C_{1} & 0\\\hline
0 & 0
\end {array}
\right]
\left[
\begin{array}{c|c}
I_{b} & Y_{1}\\\hline
Y_{2} & Y_{3}
\end {array}
\right]Q_{2} \\[1.5 ex]
&\!\!\!=\!\!\!&
P_{1}
\left[
\begin{array}{c|c}
C_{1} & C_{1}Y_{1}\\\hline
X_{2}C_{1} & X_{2}C_{1}Y_{1}
\end {array}
\right]Q_{2} \,.\quad \diamondsuit
\end{array}$$
\[X\_0\_Full\_rank\] ***(i)*** If a matrix $A$ has full row rank and a matrix $B$ has full column rank, then parameters from submatrices $X_2$ and $Y_1$ don’t exist and don’t appear in the matrix $X_{0}$ of form (\[Form\_X\_0\]). ***(ii)*** If either a matrix $A$ has full row rank or a matrix $B$ has full column rank, then the matrix $X_0$ of form (\[Form\_X\_0\]) has the structure of an affine linear space with parameters from either submatrix $X_2$ or submatrix $Y_1$, respectively. ***(iii)*** If a matrix $A$ doesn’t have full row rank and a matrix $B$ doesn’t have full column rank, then the matrix $X_0$ of form (\[Form\_X\_0\]) doesn’t have the structure of an affine linear space relative to parameters from submatrices $X_2$ and $Y_1$.
\[R23\] In the paper [@MalesevicRadicic11] authors proved that there is a matrix equation (\[AXBC\]) and its particular solution $X_{1}$ such that $ X_{1}\neq A^{(1)}CB^{(1)}$ for any choice of *{*$\!$1*}*-inverses $A^{(1)}$ and $B^{(1)}$.
\[R24\] According to Theorem VI, pp.$\,$345-346, from [@Cullis1913], it is possible to extract $a \!\cdot\! b$ parameters in a matrix $Y$ such that, these parameters are expressed, in the solution (\[OR\]), as a non-homogeneous linear functions of the other $n\!\cdot\! p - a \!\cdot\! b$ independent parameters.
***2.4.*** In this part of the paper we analysed two applications the concept of reproductivity on some matrix systems which are in relation to the matrix equation (\[AXBC\]).
In [@Penrose55] R. Penrose studied a matrix system $$\label{S1}
(\ref{S1}a) \quad AX=B
\qquad \wedge \qquad
(\ref{S1}b) \quad XD=E,
\qquad$$ where $A,$ $B,$ $D$ and $E$ are given complex matrices corresponding dimensions. He proved that $$X_{1}=A^{(1)}B+ED^{(1)}-A^{(1)}AED^{(1)}$$ is one common solution of the matrix equations (\[S1\]a) and (\[S1\]b) if $AE=BD$ and the matrix equations (\[S1\]a) and (\[S1\]b) are consistent.
In [@Ben-IsraelGreville03] A. Ben-Israel and T.N.E. Greville proved that the matrix equations (\[S1\]a) and (\[S1\]b) have a common solution iff each equation separately has a solution and $AE=BD$. Also, they proved that if $X_{0}$ is any common solution of the matrix equations (\[S1\]a) and (\[S1\]b), the general solution of the matrix system (\[S1\]) is given by the formula $$X=g(Y)=X_{0}+(I-A^{(1)}A)Y(I-DD^{(1)}),$$ where $ Y $ is an arbitrary matrix corresponding dimensions.
We will prove that if the matrix system (\[S1\]) is consistent, the general reproductive solution is given by the formula $$\label{ORRS1}
X=f(Y)=A^{(1)}B+ED^{(1)}-A^{(1)}AED^{(1)}+(I-A^{(1)}A)Y(I-DD^{(1)}),$$ where $Y$ is an arbitrary matrix corresponding dimensions.
If the matrix system (\[S1\]) is consistent, the following equivalence is true $$\label{App_1}
{\big (}AX \!=\!B \; \; \wedge \; \; XD \!=\! E{\big )}
\;\Longleftrightarrow\;
X=f(X).$$ The direct implication of (\[App\_1\]) follows by implications (see Remark \[R21\]. in the subsection 2.1):
$$AX \!=\! B \;\Longrightarrow\; X \!=\! f_{1}(X) \!=\! A^{(1)}B \!+\! X \!-\! A^{(1)}AX \, ,$$ $$XD \!=\! E \;\Longrightarrow\; X \!=\! f_{2}(X) \!=\! ED^{(1)} \!+\! X \!-\!XDD^{(1)} \, ,$$ $$AXD \!=\! BD \!=\! AE \;\Longrightarrow\; X \!=\! f_{3}(X) \!=\! A^{(1)}AED^{(1)} \!+\! X \!-\! A^{(1)}AXDD^{(1)} \, .$$
From the previous implications we can conclude $${\big (} AX \!=\! B \; \; \wedge \; \; XD \!=\! E {\big )}
\;\Longrightarrow\;
X \!=\! f(X) \!=\! f_{1}(X) \!+\! f_{2}(X) \!-\! f_{3}(X).$$
The reverse implication of (\[App\_1\]) is trivial. Notice that the function $f$ is reproductive. Therefore, if the matrix system (\[S1\]) is consistent, it is equivalent to the reproductive matrix equation . Based on Theorem \[T12\]. we conclude that $X\!=\!f(Y)$ is the general reproductive solution of the matrix system (\[S1\]). If there is a particular solution $X_0$ of the matrix system (\[S1\]) so that $X_0 \!\neq\! X_1$, then $X \!=\! g(Y)$ is the general non-reproductive solution. At the end of these application let us remark that equality $X \!=\! g(Y\!-\!X_0) \!=\! f(Y)$ also represents one simple proof of the Statement 1 from [@Haveric84].
\[A22\]Let $A \in \mathbb{C}^{n \times n}$ be a singular matrix. In this section we consider a matrix system $$\label{Sys2}
AXA=A \quad \wedge \quad AX=XA.$$
The consistency of the matrix system (\[Sys2\]) is determined by Theorem 1 in [@Keckic85] (see also [@Ruski_Gruv_84_85] and [@Keckic97]). Let $\bar{A}$ is commutative *{*1*}*-inverse, [@Keckic85]. Based on the reproductivity, we give a new proof that the formula from [@Keckic85]: $$\label{fSolA2}
X
=
f(Y)
=
\bar{A} A \bar{A} + Y - \bar{A} A Y - Y A \bar{A} + \bar{A} A Y A \bar{A},$$ where $Y$ is an arbitrary matrix corresponding dimensions, represents the general solution of the consistent matrix system (\[Sys2\]).
Namely, if the matrix system (\[Sys2\]) is consistent, the equivalence $$\label{App_2}
{\big (} AXA=A \; \; \wedge \; \; AX=XA {\big )}
\;\Longleftrightarrow\;
X=f(X)$$ is true. The direct implication of (\[App\_2\]) is based on the following simple matrix equalities: $$\bar{A} \!\!\mathop{\underbrace{\!\!A\!\!}}\limits_{\mbox{\scriptsize $(=\!A\!X\!A)$}}\!\! \bar{A}
=
\bar{A}AX \!\!\mathop{\underbrace{A\bar{A}}} \limits_{\mbox{\scriptsize $(=\bar{A}\!A)$}}
=
\bar{A} \!\!\mathop{\underbrace{AX}}\limits_{\mbox{\scriptsize $(=XA)$}}\!\! \bar{A}A
=
\bar{A}X \!\!\mathop{\underbrace{A\bar{A}A}}\limits_{\mbox{\scriptsize $(=A)$}}
=
\bar{A} \!\!\mathop{\underbrace{XA}}\limits_{\mbox{\scriptsize $(=AX)$}}
=
\bar{A}AX$$ and $$\mathop{\underbrace{\bar{A} A}}\limits_{\mbox{\scriptsize $(=A\bar{A})$}}\! X A \bar{A}
=
A\bar{A} \!\mathop{\underbrace{X A}}\limits_{\mbox{\scriptsize $(=AX)$}}\! \bar{A}
=
\mathop{\underbrace{A\bar{A} A}}\limits_{\mbox{\scriptsize $(=A)$}}\! X\bar{A}
=
\mathop{\underbrace{A X}}\limits_{\mbox{\scriptsize $(=XA)$}}\! \bar{A}
=
XA\bar{A}.$$
From this we get that $X \!=\! X \!+\! \bar{A} A \bar{A} \!-\! \bar{A}AX \!+ \! \bar{A} A X A \bar{A} \!-\! X A \bar{A} \!=\! f(X)$. The reverse implication of (\[App\_2\]) is trivial. Notice that the function $f$ is reproductive. Therefore, if the matrix system (\[Sys2\]) is consistent, it is equivalent to the reproductive matrix equation $X \!=\! f(X)$. Based on Theorem \[T12\]. we conclude that is the general reproductive solution of the matrix system (\[Sys2\]). If $X_{0}$ is any solution of the matrix system (\[Sys2\]), the formula $$\label{gSolA2}
X = g(Y) = X_{0} + Y - \bar{A} A Y - Y A \bar{A} + \bar{A} A Y A \bar{A},$$ also determines a form of the general solution of the matrix system (\[Sys2\]) because the equality is true. If there is a particular solution $X_0$ of the matrix system (\[Sys2\]) such that $X_0 \!\neq\! \bar{A} A \bar{A}$, then $X \!=\!g(Y)$ is the general non-reproductive solution. Additional applications of the concept of reproductivity for some matrix equations and systems were considered in the paper [@MalesevicRadicic12].
[**Acknowledgment.**]{} Research is partially supported by the Ministry of Science and Education of the Republic of Serbia, Grant No. ON 174032.
[20]{}
, [*Matrix Algebra*]{}, Econometric exercises, Volume **1**, Cambridge, 2005.
, [*On general and reproductive solutions of arbitrary equations*]{}, Publications de l’institut math' ematique, Nouvelle serie, tome **26** (40), Beograd 1979, 31$\;$-$\,$33.
, [*All solutions of finite equations*]{}, Discrete Mathematics Vol. [**137**]{} (1-3), 1995, 1$\;$-$\,$6.
, [*General reproductive solutions of Postian equations*]{}, Discrete Mathematics Vol. [**169**]{} (1-3), 1997, 163$\;$-$\,$168.
, [*All general solutions of Preši' c’s equation*]{}, Facta universitatis, Ser. Math. Inform. Vol. [**17**]{}, Niš 2002, 1$\;$-$\,$4.
, [*General Solutions of System of Finite Equations*]{}, Scientific Publications of the State University of Novi Pazar Ser. A: Appl. Math. Inform. and Mech. vol. **3**, 2 (2011), 117$\;$-$\,$121.
, [*Generalized Inverses: Theory and Applications*]{}, Springer, 2003.
, [*A Note On Reproductive Solutions*]{}, Publications de l’institut math' ematique, Nouvelle serie, tome **19** (33), Beograd 1975, 33$\;$-$\,$35. ([http:/$\!$/publications.mi.sanu.ac.rs/]{})
, [*Generalized Inverses of Linear Transformations*]{}, Society for Industrial and Applied Mathematics, 2009.
, [*Matrices and determinoids$\,$-$\!\,$Volume $I$*]{}, Cambridge, University Press Publ. 1913. ([http:/$\!$/archive.org/details/matricesdetermin01cull]{})
, [*The reflexive solutions of the matrix equation $AXB=C$*]{}, Comp. Math. Appl., **51** (2006), 897$\;$-$\,$902.
, [*Positive and real-positive solutions to the equation $axa^{\ast}\!=\!c$ in $C^{\ast}-\,$algebra*]{}, Linear & Multilinear algebra, **55**, (6) (2007), 535$\;$-$\,$543.
, [*Re-nnd solutions of the matrix equation $AXB=C$*]{}, Journal of the Australian Mathematical Society, **84** (2008), 63$\;$-$\,$72.
, [*Equations $ax=c$ and $xb=d$ in rings and rings with involution with applications to Hilbert space operators*]{}, Lin. Alg. and its Appl. [**429**]{} (2008) 1779$\;$-$\,$1809.
, [*Commuting generalized inverse matrices*]{}, Mathematical Notes, Volume **36**, Number 1, 500$\;$-$\,$504, 1985 (Translated from Matematicheskie Zametki, Vol. **36**, No. 1, 17$\,$-$\,$23, 1984.).
, [*On the functional equation $f \phi f \!=\! f$*]{}, Publications de l’institut math' ematique, Nouvelle serie, tome **29** (43), Beograd 1981, 61$\,$-$\,$64.
, [*Formulae for general reproductive solutions of certain matrix equations*]{}, Publications de l’institut math' ematique, Nouvelle serie, tome **34** (48), Beograd 1983, 81$\,$-$\,$84.
, [*On solutions of a matrix equations system $AX=B$ and $XD=E$*]{}, Matematički Vesnik **36** (1), Beograd 1984, 11$\,$-$\,$16.
, [*Reproductivity of some equations of analysis I*]{}, Publications de l’institut math' ematique, Nouvelle serie, tome **31**(45), Beograd 1982, 73$\;$-$\,$81.
, [*Reproductivity of some equations of analysis II*]{}, Publications de l’institut math' ematique, Nouvelle serie, tome **33**(47), Beograd 1983, 109$\;$-118.
, [*Commutative weak generalized inverses of a square matrix and some related matrix equations*]{}, Publications de l’institut math' ematique, Nouvelle serie, tome **38** (52), Beograd 1985, 39$\;$-$\,$44.
, [*On some generalized inverses of matrices and some linear matrix eguations*]{}, Publications de l’institut math' ematique, Nouvelle serie, tome **45** (59), Beograd 1989, 57$\,$-$\,$63.
, [*Some remarks on possible generalized inverses in semigroups*]{}, Publications de l’institut math' ematique, Nouvelle serie, tome **61** (75), Beograd 1997, 33$\;$-$\,$40.
, [*Reproductivity - A general approach to equations*]{}, Facta universitatis, Ser. Math. Inform. Vol. [**12**]{}, Niš 1997, 157$\;$-$\,$184.
, [*Non-reproductive and reproductive solutions of some matrix equations*]{}, Proceedings of the International conference [*Mathematical and Informational Technologies, MIT$\,$-$\,$2011*]{}, Vrnjačka Banja, Serbia, 2011, 246$\;$-$\,$251. ([http:/$\!$/mit.rs/]{})
, [*Some considerations of matrix equations using the concept of reproductivity*]{}, Kragujevac Journal of Mathematics, **36**(1) (2012), 151$\;$-$\,$161.
, [*A generalized inverses for matrices*]{}, Math. Proc. Cambridge Philos. Soc. **51**(1955), 406$\;$-$\,$413.
, [*Methode de resolution d’une classe d’equations fonctionnelles lineaires*]{}, Comptes rendus de l’Acad' emie des Sciences Paris, [**257**]{} (1963), 2224$\;$-$\,$2226.
, [*Certaines ' equations matricielles*]{}, Publ. Elektrotehn. Fak. Ser. Mat.-Fiz., $\mathit{N}^{\underline{o}}$ 121, Beograd 1963. ([http:/$\!$/pefmath.etf.rs/]{})
, [*Une classe d’' equations matricielles et l’' equation fonctionnelle $f^{2}\!=\!f$*]{}, Publications de l’institut math' ematique, Nouvelle serie, tome **8** (22), Beograd 1968, 143$\;$-$\,$148.
, [*Une methode de resolution des equations dont toutes les solutions appartiennent a un ensemble fini donne*]{}, Comptes rendus de l’Acad' emie des Sciences Paris, [**272**]{} (1971), .
, [*Ein Satz " Uber Reproduktive L" osungen*]{}, Publications de l’institut math' ematique, Nouvelle serie, tome **14** (28), Beograd 1972, 133$\;$-136.
, [*All reproductive solutions of finite equations*]{}, Publications de l’institut math' ematique, Nouvelle serie, tome **44** (58), Beograd 1988, 3$\,$-7.
, [*A generalization of the notion of reproductivity*]{}, Publications de l’institut math' ematique, Nouvelle serie, tome **67** (81), Beograd 2000, 76$\;$-$\,$84.
, [*Contribution to the theory, computation and application of generalized inverses*]{}, Doctoral dissertation, University of North Carolina at Releigh, May 1964.
, [*On general solutions of arbitrary equations*]{}, Publications de l’institut math' ematique, Nouvelle serie, tome **24** (38), Beograd 1978, 143$\;$-$\,$145.
, [*On general and reproductive solutions of finite equations*]{}, Publications de l’institut math' ematique, Nouvelle serie, tome **63** (77), Beograd 1998, 26$\;$-$\,$30.
, [*Lattice Functions and Equations*]{}, Springer, 2001.
, [*On additive decompositions of solutions of the matrix equation $AXB=C$*]{}, Calcolo, Vol. **47** (4), 2010, 193$\;$-$\,$209.
, [*On Additive Decomposition of the Hermitian Solution of the Matrix Equation $AXA^{\ast}=B$*]{}, Mediterranean Journal of Mathematics, **9** (2012), 47$\;$-$\,$60.
, [*A system of matrix equations and a linear matrix equation over arbitrary regular rings with identity*]{}, Lin. Alg. and its Appl. **384**, (2004), 43$\;$-$\,$54.
, [*A tribute to S.$\,$B. Preši' c$\,:$*]{} [*Papers Celebrating his 65$\,$-$\,$th *]{}, Mathematical Institute of the Serbian Academy of Sciences and Arts, Belgrade, publ. 2001. ([http:/$\!$/elibrary.matf.bg.ac.rs/handle/123456789/448]{})
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
We study a symmetric diffusion $X$ on ${\mathbb{R}}^d$ in divergence form in a stationary and ergodic environment, with measurable unbounded and degenerate coefficients $a^\omega$. The diffusion is formally associated with $L^\omega u = \nabla\cdot(a^\omega\nabla u)$, and we make sense of it through Dirichlet forms theory. We prove for $X$ a quenched invariance principle, under some moment conditions on the environment; the key tool is the sublinearity of the corrector obtained by Moser’s iteration scheme.\
\
Nous étudions une diffusion symétrique $X$ sur $R^d$ en forme de divergence dans un environnement aléatoire stationnaire et ergodique, dont les coefficients $a^\omega$ sont mesurables et dégénérés. Cette diffusion qui est formellement engendrée par l’opérateur $L^\omega u=\nabla\cdot(a^\omega\nabla u)$, peut être définie à l’aide de la théorie des formes de Dirichlet. Nous démontrons pour $X$ un principe d’invariance presque sûr sous des conditions de moment de l’environnement; l’outil crucial est la sous-linéarité du correcteur obtenu à l’aide de l’ itération introduite par J. Moser.
address: 'Department of Mathematics, Technische Universität zu Berlin, Straße des 17. Juni 136, 10623 Berlin, Germany'
author:
- 'Alberto Chiarini[^1]'
- 'Jean-Dominique Deuschel'
title: 'Invariance Principle for symmetric Diffusions in a degenerate and unbounded stationary and ergodic Random Medium.'
---
invariance principle ,homogenization ,Moser’s iteration ,reversible dynamics ,Dirichlet forms. 60K37, 60F17
Description of the Main Result
==============================
We are interested in the study of reversible diffusions in a random environment. Namely, we are given an infinitesimal generator $L^\omega$ in divergence form $$\label{eq:generator}
L^\omega u(x) = \nabla \cdot ( a^\omega (x)\nabla u(x)),\quad x\in{\mathbb{R}}^d$$ where $a^\omega(x)$ is a symmetric $d$-dimensional matrix depending on a parameter $\omega$ which describes a random realization of the environment.
We model the environment as a probability space $(\Omega,\mathcal{G},\mu)$ on which a measurable group of transformations $\{\tau_x\}_{x\in{\mathbb{R}}^d}$ is defined. One may think of $\tau_x\omega$ as a translation of the environment $\omega\in\Omega$ in the direction $x\in{\mathbb{R}}^d$. The random field $\{a^\omega(x)\}_{x\in{\mathbb{R}}^d}$ will then be constructed simply by taking a random variable $a:\Omega\to{\mathbb{R}}^{d\times d}$ and by defining $a^\omega(x)\operatorname{\mathrel{\mathop:}=}a(\tau_x \omega)$, we will often use the notation $a(x;\omega)$ for $a^\omega(x)$ as well. We assume that the random environment $(\Omega,\mathcal{G},\mu)$, $\{\tau_x\}_{x\in{\mathbb{R}}^d}$ is stationary and ergodic. A precise formulation of the setup is given in section 3.
It is well known that when $x\to a^\omega(x)$ is bounded and uniformly elliptic, uniformly in $\omega$, then a quenched invariance principle holds for the diffusion process $X_t^\omega$ associated with $L^\omega$. This means that, for $\mu$-almost all $\omega\in\Omega$, the scaled process $X^{{\varepsilon},\omega}_t\operatorname{\mathrel{\mathop:}=}{\varepsilon}X^{\omega}_{t/{\varepsilon}^2}$ converges in distribution to a Brownian motion with a non-trivial covariance structure as ${\varepsilon}$ goes to zero; this is known as diffusive limit. See for example the classic result of Papanicolau and Varadhan [@papvar] where the coefficients are assumed to be differentiable, and [@Osada1983] for measurable coefficients and more general operators.
Recently, a lot of efforts has been put into extending this result beyond the uniform elliptic case. For example [@fannjiang1997] consider a non-symmetric situation with uniformly elliptic symmetric part and unbounded antisymmetric part and the recent paper [@BaMathieu] proves an invariance principle for divergence form operators $L u= e^V \nabla\cdot(e^{-V} \nabla u)$ where $V$ is periodic and measurable. They only assume that $e^V+e^{-V}$ is locally integrable. For what concerns ergodic and stationary environment a recent result has been achieved in the case of random walk in random environment in [@deuschelslowikandresharnack], [@deuschelslowikandres]. In these works moments of order greater than one are needed to get an invariance principle in the diffusive limit; [@deuschelslowikandres] and the techniques therein are the main inspiration for our paper.
The aim of our work is to prove a quenched invariance principle for an operator $L^\omega$ of the form with a random field $a^\omega(x)$ which is ergodic, stationary and possibly unbounded and degenerate. Denote by $a:\Omega\to{\mathbb{R}}^{d\times d}$ the $\mathcal{G}$-measurable random variable which describes the field through $a^\omega(x)=a(\tau_x\omega)$. We assume that $a$ is symmetric and that there exist $\Lambda,\lambda$, $\mathcal{G}$-measurable, positive and finite, such that:
- for $\mu$-almost all $\omega\in\Omega$ and all $\xi\in{\mathbb{R}}^d$ $$\lambda(\omega)|\xi|^2\leq \langle a(\omega) \xi,\xi\rangle \leq \Lambda(\omega)|\xi|^2;$$
- there exist $p,q\in[1,\infty]$ satisfying $1/p+1/q<2/d$ such that $${\mathbb{E}}_\mu [\lambda^{-q}]<\infty,\quad{\mathbb{E}}_\mu [\Lambda^{p}]<\infty,$$
- as functions of $x$, $\lambda^{-1}(\tau_x\omega),\Lambda(\tau_x\omega)\in L^\infty_{loc}({\mathbb{R}}^d)$ for $\mu$-almost all $\omega\in\Omega$.
Since $a^\omega(x)$ is meant to model a random field, it is not natural to assume its differentiability in $x\in{\mathbb{R}}^d$. Accordingly, the operator defined in does not make any sense, and the techniques coming from Stochastic differential equations and Itô calculus are not very helpful neither in constructing the diffusion process, nor in performing the relevant computation.
The theory of Dirichlet forms is the right tool to approach the problem of constructing a diffusion. Instead of the operator $L^\omega$ we shall consider the bilinear form obtained by $L^\omega$, formally integrating by parts, namely $$\label{eq:df}
{\mathcal{E}^\omega}(u,v)\operatorname{\mathrel{\mathop:}=}\sum_{i,j}\int_{{\mathbb{R}}^d} a^\omega_{ij}(x)\partial_i u(x)\partial_j v(x) dx$$ for a proper class of functions $u,v\in{\mathcal{F}}^\omega\subset L^2({\mathbb{R}}^d,dx)$, more precisely ${\mathcal{F}}^\omega$ is the closure of $C_0^\infty({\mathbb{R}}^d)$ in $L^2({\mathbb{R}}^d,dx)$ with respect to ${\mathcal{E}}+(\cdot,\cdot)_{L^2}$. It is a classical result of Fukushima [@fukushima1994dirichlet Theorem 7.2.2] and [@rockner Ch. II example 3b] that it is possible to associate to a diffusion process $\{X^\omega, {\mathbb{P}}_x^\omega, x\in {\mathbb{R}}^d\}$ as soon as $(\lambda^\omega)^{-1}$ and $\Lambda^\omega$ are locally integrable. It is well known that there is a properly exceptional[^2] set $\mathcal{N}^\omega\subset {\mathbb{R}}^d$ of $X^\omega$ such that the associated process is uniquely determined up to the ambiguity of starting points in $\mathcal{N}^\omega$, in our situation the set of exceptional points may depend on the realization of the environment. Assumption \[ass:a.3\] is designed to remove the ambiguity about the properly exceptional set $\mathcal{N}^\omega$. We will then prove that assumption \[ass:a.2\] and ergodicity of the environment are enough to grant that the process $X^\omega$ starting from any $x\in {\mathbb{R}}^d$ does not explode for almost all realization of the environment.
Moment conditions on the environment are a very natural assumption in order to achieve a quenched invariance principle for symmetric diffusions, indeed at least the first moment of $\Lambda$ and $\lambda^{-1}$ is required to obtain the result. As a counterexample one can consider a periodic environment, namely the $d$-dimensional torus $\mathbb{T}^d$, and the following generator in divergence form $$L f(x)\operatorname{\mathrel{\mathop:}=}\frac{1}{{\varphi}(x)}\nabla\cdot ({\varphi}(x)\nabla f(x)),$$ where ${\varphi}:\mathbb{T}^d\to {\mathbb{R}}$ is defined by ${\varphi}(x)\operatorname{\mathrel{\mathop:}=}1_B(x) |x|^{-d}+1_{B^c}(x)$ being $B\subset \mathbb{T}^d$ a ball of radius one centered in the origin. It is clear that ${\varphi}^\alpha\in L^1(\mathbb{T}^d)$ for all $\alpha<1$ but not for $\alpha=1$. If we look for example to $d=2$, then the radial part of the process associated to $L$, for the radius less than one, will be a Bessel process with parameter $\delta=0$ which is known to have a trap in the origin.
As observed in the previous remark, if we want to prove an invariance principle, dealing with symmetric diffusions forces the degeneracy of the diffusion coefficient not to be too strong. Namely, the diffusion coefficient can eventually be zero only on a set of null Lebesgue measure. On the other hand, in the case of non-symmetric diffusions the diffusion coefficient is allowed to vanish in open sets, as was proved in the periodic environment by [@Hairer20082462] and further extended and generalized in [@delaruerhodes], [@sow2009homogenization], [@pardoux2011homogenization]. In these works the strong degeneracy of the diffusion coefficient is compensated by the drift through the Hörmander’s condition; as a result and in contrast with our setting, the coefficients need to be smooth enough.
Once the diffusion process $X^\omega$ is constructed, the standard approach to diffusive limit theorems consists in showing the weak compactness of the rescaled process and in the identification of the limit. In the case of bounded and uniformly elliptic coefficients the compactness is readily obtained by the Aronson-Nash estimates for the heat kernel. In order to identify the limit, we use the standard technique used in [@fannjiang1997], [@kozlov1985] and [@Osada1983]; namely, we decompose the process $X_t^{\varepsilon}$ into a martingale part, called the *harmonic coordinates* and a fluctuation part, called the *correctors*. The martingale part is supposed to capture the long time asymptotic of $X_t^{\varepsilon}$, and will characterize the diffusive limit.
The challenging part is to show that the correctors are uniformly small for almost all realization of the environment, this is attained generalizing Moser’s arguments [@moser1964] to get a maximal inequality for positive subsolutions of uniformly elliptic, divergence form equations. In this sense the relation $1/p+1/q<2/d$ is designed to let the Moser’s iteration scheme work. This integrability assumption firstly appeared in [@edmundspeletier] in order to extend the results of De Giorgi and Nash to degenerate elliptic equations. A similar condition was also recently exploited in [@Zhikov] to obtain estimates of Nash - Aronson type for solutions to degenerate parabolic equations. They look to generator of the form $\mathcal{L} u = \partial_t u - e^{-V}\nabla \cdot ( e^V \nabla u)$, with the assumption that $\sup_{r\geq 1}|r|^{-d}\int_{|x|\leq r} e^{pV}+e^{-qV} dx <\infty$.
We want to stress out that condition \[ass:a.3\] is needed to prove neither the sublinearity of the corrector nor its existence, we used it only to have a more regular density of the semigroup associated to $X^\omega$ and avoid some technicalities due to exceptional sets in the framework of Dirichlet form theory.
Once the correctors are shown to be sublinear, the standard invariance principle for martingales [@helland1982] gives the almost sure convergence to the Wiener measure.
\[thm:invprinc\] Assume \[ass:a.1\], \[ass:a.2\] and \[ass:a.3\] are satisfied. Let $\mathbf{M}^\omega\operatorname{\mathrel{\mathop:}=}(X_t^\omega, {\mathbb{P}}_x^\omega)$, $x\in{\mathbb{R}}^d$, be the minimal diffusion process associated to $({\mathcal{E}^\omega},{\mathcal{F}}^\omega)$ on $L^2({\mathbb{R}}^d,dx)$. Then the following hold
- For $\mu$-almost all $\omega\in\Omega$ the limits $$\lim_{t\to\infty} \frac{1}{t}{\mathbb{E}}_0^\omega[X^\omega_t(i) X^\omega_t(j)] =\mathbf{d}_{ij}\quad i,j=1,...,d$$ exist and are deterministic constants.
- For $\mu$-almost all $\omega\in\Omega$, the laws of the processes $X_t^{\omega,{\varepsilon}}\operatorname{\mathrel{\mathop:}=}{\varepsilon}X^\omega_{t/{\varepsilon}^2}$, ${\varepsilon}>0$ over $C([0,+\infty),{\mathbb{R}}^d)$ converge weakly as ${\varepsilon}\to 0$ to a Wiener measure having the covariance matrix equal to $\mathbf{D}=[\mathbf{d}_{ij}]$. Moreover $\mathbf{D}$ is a positive definite matrix.
#### Description of the method
One of the main objective of the paper is to show that the correctors $\chi = (\chi^1,\dots,\chi^d):{\mathbb{R}}^d\times \Omega \to {\mathbb{R}}^d$ are locally sublinear, namely that $$\limsup_{{\varepsilon}\to 0} \sup_{|x|\leq R} {\varepsilon}|\chi(x/{\varepsilon},\omega)| = 0,\quad \forall R>0,\,\mu\mbox{-a.s.}$$
To obtain a priori estimates on the correctors $\chi$ we exploit the fact that they are constructed in such a way that they are solutions of a Poisson’s equation, which is formally given by $$\label{eq:correctoreq}
\nabla \cdot (a^\omega(x) \nabla \chi^k(x,\omega)) = \nabla \cdot (a^\omega(x) \nabla \pi^k(x)),$$ where $\pi^k(x) := x_k$ is the projection to the $k$th-coordinate.
The equation above has been studied extensively and generalized in many directions, also beyond the linear case, for an introduction, see for example the monographs [@evans2010partial], [@gilbarg2001elliptic] and for recent developments in the theory see [@heinonen2006nonlinear]. When the matrix $a^\omega$ is uniformly elliptic and bounded, uniformly in $\omega\in\Omega$, namely if $$c^{-1}|\xi|^2 \leq \langle a^\omega(x)\xi,\xi \rangle \leq c |\xi|^2$$ for some $c\geq 1$, it is natural to look for weak solutions to in the classical Sobolev space of square integrable functions with square integrable weak derivatives. It is a classical result due to Moser [@moser1964] that an elliptic Harnack inequality holds and a result from Nash [@nash1958] and De Giorgi [@zbMATH03138423] that solutions are Hölder continuous.
The situation changes dramatically when the coefficients are degenerate. In the most typical situation there is a positive weight $\theta:{\mathbb{R}}^d\to {\mathbb{R}}$ and a constant $c>1$ such that $$\theta(x)|\xi|^2 \leq \langle a^\omega(x)\xi,\xi \rangle \leq c\,\theta(x)|\xi|^2.$$ In this setting one looks for solutions to equation in the weighted Sobolev space $W^{1,2}({\mathbb{R}}^d,\theta)$ which is the set of weakly differentiable functions $u:{\mathbb{R}}^d\to {\mathbb{R}}$ such that $$\int_{{\mathbb{R}}^d} |u|^2 \theta dx<\infty,\mbox{ and } \int_{{\mathbb{R}}^d} |\nabla u|^2 \theta dx<\infty,$$ we refer to [@heinonen2006nonlinear], [@zhikov1998weighted] for more information on weighted Sobolev spaces. It was shown in [@fabes1982local] that in order to have local regularity of solutions to it is enough to have weights which are volume doubling, namely such that there exists a constant $C>0$ for which $$\int_{B_{2R}(x)} \theta(y)\,dy \leq C \int_{B_R(x)}\theta(y)\,dy,\quad \forall R>0,\,\forall x\in {\mathbb{R}}^d,$$ and which satisfy weighted Sobolev and Poincaré inequalities. This weights are known in general as $p$-admissible (See [@heinonen2006nonlinear]), but for our discussion of the linear operator $L^\omega=\nabla\cdot(a^\omega\nabla\,)$ it is enough to look at $2$-admissible weights.
\[rem:voldoubling\]In our setting it is not possible to expect the volume doubling property for small balls. The ergodic theorem ensures only that for all $x\in {\mathbb{R}}^d$ and $\mu$-almost all $\omega\in\Omega$ there exist $R_0^\omega(x)>0$ and a dimensional constant $C>0$ such that for all $R>R_0^\omega(x)$ $$\int_{B_{2R}(x)} \Lambda^\omega(y)\,dy \leq C \int_{B_R(x)}\Lambda^\omega(y)\,dy,$$ being $B_R(x)$ the ball of center $x$ and radius $R$. We remark that the constant $R_0^\omega(x)$ cannot be taken uniformly in $x\in {\mathbb{R}}^d$, and $\sup_{x\in {\mathbb{R}}^d} R_0^\omega(x)$ may be infinite.
Examples of $2$-admissible weights are the functions in the Muckenhaupt’s class $A_2$, we refer to [@fabes1982local], [@heinonen2006nonlinear], [@torchinsky2012real] and to the original research paper [@muckenhoupt1972weighted] for an exhaustive treatment on the subject. Here we briefly recall that the class $A_2$ is the set of all non negative functions $\theta:{\mathbb{R}}^d\to [0,\infty]$ for which there exists a constant $C>0$ such that $$\label{eq:muckenhaupt}
\sup_{R>0} \sup_{x\in{\mathbb{R}}^d} \biggr(\frac{1}{|B_R(x)|}\int_{B_R(x)} \theta(y)\,dy\biggl)
\biggr(\frac{1}{|B_R(x)|}\int_{B_R(x)} \theta^{-1}(y)\,dy\biggl) \leq C.$$ It is well known that weights in the class $A_2$ are volume doubling and satisfy a weighted Sobolev inequality. To be more precise, denote by $\theta(B):=\int_B\theta dx$, then there exist constants $C,\delta>0$ such that for all $1\leq k\leq d/(d-1)+\delta$ $$\label{ineq:muck}
\biggl(\frac{1}{\theta(B)}\int_{B}|u|^{2k} \theta dx\biggr)^\frac{1}{k}\leq C |B|^\frac{2}{d} \frac{1}{\theta(B)}\int_{B}|\nabla u|^{2} \theta dx\quad \biggl( \leq C |B|^\frac{2}{d} \frac{{\mathcal{E}}(u,u)}{\theta(B)}\biggr)$$ being $B$ any ball in ${\mathbb{R}}^d$ and $u\in C_0^\infty(B)$.
Working with admissible weights has the advantage of being able to state Hölder continuity results for weak solutions to . It is still an open problem to identify the optimal conditions that a weight has to satisfy in order to grant continuity of weak solutions, see the survey paper [@cavalheiro2008weighted] for details.
Many authors relied on Muckenhaupt’s classes and weighted Sobolev spaces to prove homogenization results. We quote [@de1992homogenization] for the periodic case and [@engstrom2006homogenization] for the ergodic case. In the latter the weights are assumed to belong to a Muckenhaupt class for almost all the realizations of the environment.
In our paper, to prove the sublinearity of the corrector, we assume that the coefficient $a^\omega(x)$ satisfies $$\lambda^\omega(x)|\xi|^2 \leq \langle a^\omega(x)\xi,\xi \rangle \leq \Lambda^\omega(x)|\xi|^2,\quad \mu\mbox{-a.s.}$$ and ${\mathbb{E}}_\mu[\lambda^{-q}]$, ${\mathbb{E}}_\mu[\lambda^{-q}]<\infty$ with $1/p+1/q<2/d$. In this case, the weights $\lambda^\omega(x) := \lambda(\tau_x\omega)$ and $\Lambda^\omega(x):= \Lambda(\tau_x\omega)$ do not belong to any of the classes mentioned above, since, as explained in Remark \[rem:voldoubling\], in general the measures $\lambda^\omega(x)dx$ and $\Lambda^\omega(x)dx$ are not volume doubling. The ergodicity of the environment and the fact that ${\mathbb{E}}_\mu[\lambda^{-1}],{\mathbb{E}}_\mu[\Lambda]$ are finite ensure only that $$\sup_{x\in {\mathbb{R}}^d}\limsup_{R\to\infty}\frac{1}{|B_R(x)|}\int_{B_R(x)} \frac{1}{\lambda^\omega(y)}\,dy<\infty,\quad\sup_{x\in {\mathbb{R}}^d}\limsup_{R\to\infty}\frac{1}{|B_R(x)|}\int_{B_R(x)}\Lambda^\omega(y) dy <\infty,$$ $\mu$-almost surely, and, contrary to , it is not possible to interchange the supremum and the limit staying finite.
Another characterizing feature of our model is that we don’t assume $\Lambda^\omega\leq c \lambda^\omega$. We cannot expect regularity for solutions to , however, we show that the ergodicity of the environment and the moment conditions \[ass:a.2\] are enough to obtain the sublinearity of the correctors; this is done in the same spirit of [@fannjiang1997] where an unbounded but uniformly bounded away from zero non-symmetric case is considered.
Moser’s method to derive a maximal inequality for solutions to is based on two steps. One wants first to get a Sobolev inequality to control some $L^\rho$-norm in terms of the Dirichlet form and then control the Dirichlet form of any solution by a lower moment. This sets up an iteration which leads to control the supremum of the solution on a ball by a lower norm on a slightly bigger ball. In the uniform elliptic and bounded case this is rather standard and it is possible to control the $L^{2d/(d-2)}$-norm of a solution by its $L^2$-norm through the classical Sobolev inequality. In the case of Muckenhaupt’s weights the iteration can be set using the Sobolev inequality on the weighted Sobolev space.
In our paper we are able to control locally on balls the $\rho$-norm of a solution by its $2 p^*$-norm, with $\rho = 2qd/(q(d-2)+d)$ and $p^*=p/(p-1)$. For the Moser iteration we need $\rho>2 p^*$ which is equivalent to condition $1/p+1/q<2/d$. Indeed, by means of Hölder’s inequality and the standard Sobolev inequality, for a ball $B$ of radius $R>0$ and center $x\in {\mathbb{R}}^d$, we can write $$\biggl(\frac{1}{|B_R(x)|}\int_{B_R(x)}|u|^{\rho/p^*}\Lambda^\omega dy\biggr)^\frac{2 p^*}{\rho}\leq C(\lambda,\Lambda,x,R)|B_R(x)|^\frac{2}{d}\frac{{\mathcal{E}}(u,u)}{|B_R(x)|} .$$ where $$C(\lambda,\Lambda,x,R) := C(d) \biggl(\frac{1}{|B_R(x)|}\int_{B_R(x)} (\lambda^\omega(y))^{-q}dy\biggr)^\frac{1}{q}\biggl( \frac{1}{|B_R(x)|}\int_{B_R(x)} (\Lambda^\omega(y))^{p}dy\biggr)^\frac{2}{\rho(p-1)},$$ being $C(d)>0$ a constant depending only on the dimension. The Sobolev inequality above must be compared with . In opposition to , the constant in front of the inequality is strongly dependent on $x\in{\mathbb{R}}^d$ and $R>0$. Therefore, the estimates we derive in Section 2 to control the Dirichlet form of a solution by its $2p/(p-1)$-norm, although following from very well established arguments, are a necessary step in order to clarify the dependence of the constants on $$\frac{1}{|B_R(x)|}\int_{B_R(x)} (\lambda^\omega(y))^{-q}dy,\quad \frac{1}{|B_R(x)|}\int_{B_R(x)} (\Lambda^\omega(y))^{p}dy.$$ The maximal inequality which we obtain in Section 2.3 behaves nicely in the scaling limit, due to the ergodic theorem, and is enough to state the sublinearity of the corrector.
It is believed that the optimal condition for a quenched invariance principle to hold is ${\mathbb{E}}_\mu[\lambda^{-1}]$, ${\mathbb{E}}_\mu[\Lambda]<\infty$. In periodic environment this has been proven recently in [@BaMathieu] using ideas coming from harmonic analysis and Muckenhaupt’s weights. The authors consider a generator in divergence form given by $L u= e^V \nabla\cdot(e^{-V} \nabla u)$, where $V:{\mathbb{R}}^d\to {\mathbb{R}}$ is periodic and measurable such that $e^V+e^{-V}$ is locally integrable. Their argument relies on a time change and on the Sobolev inequality $$\biggl(
\int_{\mathbb{T}^d}|u|^{r} w\, dx\biggr)^\frac{2}{r}\leq C \int_{\mathbb{T}^d}|\nabla u|^{2} e^{-V} dx\quad$$ where $\mathbb{T}^d$ is the $d$-dimensional torus, $u\in C^1(\mathbb{T}^d)$ centered, $r>2$ and $w$ is expressed as an Hardy-Littlewood maximal function.
In this setting it is not possible to use Moser’s iteration technique to prove the sublinearity of the corrector on balls, since to bound the right hand side by the $L^s(\mathbb{T}^d,w)$ norm for some $s<r$ would require further assumptions on the integrability of $e^V+e^{-V}$. In fact, they don’t prove sublinearity of the correctors on balls but along the path of the process. This approach relies on a global uniform upper bound for the density of the process, which can be established due to the compactness of the periodic environment, and the fact that the process of the environment seen from the particle is just the projection of the diffusion on the torus $\mathbb{T}^d$.
Under the conditions \[ass:a.1\], \[ass:a.2\] and that a quenched invariance principle holds, Moser’s method can be successfully applied to obtain a quenched local central limit theorem for the process associated to $({\mathcal{E}}^\omega,{\mathcal{F}}^{\Lambda,\omega})$ on $L^2({\mathbb{R}}^d,\Lambda^\omega dx)$, being ${\mathcal{F}}^{\Lambda,\omega}$ the closure of $({\mathcal{E}}^\omega,C_0^\infty({\mathbb{R}}^d))$ in $L^2({\mathbb{R}}^d,\Lambda^\omega dx)$, see [@deuschelslowikandresharnack], [@chiarinideuschelLCLT]. In these papers, the proof relies on a parabolic Harnack inequality, whose constant depends strongly on the space-time cylinder considered. Thus, it cannot be applied to obtain Hölder continuity of the density. Nevertheless, it is shown that in the diffusive limit it is possible to control oscillations by means of the ergodic theorem.
Despite the fact that a quenched invariance principle is believed to hold for ${\mathbb{E}}_\mu[\lambda^{-1}]$, ${\mathbb{E}}_\mu[\Lambda]<\infty$, it was shown in [@deuschelslowikandresharnack] that the condition ${\mathbb{E}}_\mu[\lambda^{-q}]$, ${\mathbb{E}}_\mu[\Lambda^p]<\infty$, with $1/p+1/q<2/d$ is sharp, for general stationary and ergodic random environment, for a quenched local central limit theorem to hold.
A summary of the paper is the following. In Section \[sec:sobmos\] we develop a priori estimates for solutions to elliptic equations, following Moser’s scheme. In this section the random environment plays no role, and accordingly we have deterministic inequalities in a fairly general framework. Also, we construct a minimal diffusion process associated to the deterministic version of and we discuss its properties.
In Section \[sec:diffre\] we apply the results obtained in Section \[sec:sobmos\] to construct a diffusion process for almost all $\omega\in\Omega$, we define the environment process, and we show how to use it in order to prove that the diffusion is non-explosive.
In Section \[sec:corr\] we prove the existence of the harmonic coordinates and of the corrector. In particular we prove that we can decompose our process in the sum of a martingale part, of which we can compute exactly the quadratic variation, and a fluctuation part.
In Section \[sec:proof\] we use the results of the previous Sections in order to prove the sublinearity of the correctors and, given that, Theorem \[thm:invprinc\].
Sobolev’s inequality and Moser’s iteration scheme {#sec:sobmos}
=================================================
Notation and Basic Definitions
------------------------------
In this section we forget about the random environment. With a slight abuse of notation we will note with $a(x)$, $\lambda(x)$ and $\Lambda(x)$ the deterministic versions of $a(\tau_x\omega)$, $\lambda(\tau_x \omega)$ and $\Lambda(\tau_x \omega)$.
We are given a symmetric matrix $a:{\mathbb{R}}^d\to {\mathbb{R}}^{d\times d}$ such that
- there exist $\lambda,\Lambda:{\mathbb{R}}^d\to{\mathbb{R}}$ non-negative such that for almost all $x\in{\mathbb{R}}^d$ and $\xi\in{\mathbb{R}}^d$ $$\lambda(x)|\xi|^2\leq \langle a(x) \xi,\xi\rangle \leq \Lambda(x)|\xi|^2,$$
- there exist $p,q\in[1,\infty]$ satisfying $1/p+1/q<2/d$ such that $$\sup_{r\geq 1} \frac{1}{|B_r|} \int_{B_r} \Lambda^p +\lambda^{-q} \,dx <\infty.$$
By means of the ergodic theorem, \[ass:a.1\] and \[ass:a.2\] imply that the function $x\to a(\tau_x\omega)$ satisfies \[ass:b.1\] and \[ass:b.2\] for $\mu$-almost all $\omega\in\Omega$.
\[rem:embedding\] Let $B\subset{\mathbb{R}}^d$ be a ball. Assumptions \[ass:b.1\] and $\ref{ass:b.2}$ imply that, for $u\in C_0^\infty(B)$, $$\|1_B\lambda^{-1}\|_q^{-1}\|\nabla u\|^2_{2q/q+1}\leq \int_{{\mathbb{R}}^d}\langle a \nabla u,\nabla u\rangle dx\leq \|1_B \Lambda\|_p \|\nabla u\|_{2p^*}^2,$$ where $p^* = p/(p-1)$. The relation $1/p+1/q<2/d$ is designed in such a way that the Sobolev’s conjugate of $2q/(q+1)$ in ${\mathbb{R}}^d$, which is given by $$\label{eq:rho}
\rho(q,d):= \frac{2qd}{q(d-2)+d},$$ satisfies $\rho(q,d) > 2 p^*$, which implies that the Sobolev space $W^{1,2q/(q+1)}(B)$ is compactly embedded in $L^{2p^*}(B)$, see for example Chapter 7 in [@gilbarg2001elliptic].
Since the generator given in is not well defined, in order to construct a process formally associated to it, we must exploit Dirichlet forms theory. We shall here present some basic definitions coming from the Dirichlet forms theory; for a complete treatment on the subject see [@fukushima1994dirichlet].
Let $X$ be a locally compact metric separable space, and $m$ a positive Radon measure on $X$ such that $\operatorname{supp}[m]=X$. Consider the Hilbert space $L^2(X,m)$ with scalar product $\langle \cdot,\cdot\rangle$. We call a *symmetric* form, a non-negative definite bilinear form ${\mathcal{E}}$ defined on a dense subset $\mathcal{D}({\mathcal{E}})\subset L^2(X,m)$. Given a symmetric form $({\mathcal{E}},\mathcal{D}({\mathcal{E}}))$ on $L^2(X,m)$, the form ${\mathcal{E}}_\beta\operatorname{\mathrel{\mathop:}=}{\mathcal{E}}+\beta \langle \cdot,\cdot\rangle$ defines itself a symmetric form on $L^2(X,m)$ for each $\beta>0$. Note that $\mathcal{D}({\mathcal{E}})$ is a pre-Hilbert space with inner product ${\mathcal{E}}_\beta$. If $\mathcal{D}({\mathcal{E}})$ is complete with respect to ${\mathcal{E}}_\beta$, then ${\mathcal{E}}$ is said to be *closed*.
A closed symmetric form $({\mathcal{E}},\mathcal{D}({\mathcal{E}}))$ on $L^2(X,m)$ is called a *Dirichlet form* if it is Markovian, namely if for any given $u\in \mathcal{D}({\mathcal{E}})$, then $v=(0\vee u)\wedge1$ belongs to $\mathcal{D}({\mathcal{E}})$ and ${\mathcal{E}}(v,v)\leq {\mathcal{E}}(u,u)$.
We say that the Dirichlet form $({\mathcal{E}},\mathcal{D}({\mathcal{E}}))$ on $L^2(X,m)$ is *regular* if there is a subset $\mathcal{H}$ of $\mathcal{D}({\mathcal{E}})\cap C_0(X)$ dense in $\mathcal{D}({\mathcal{E}})$ with respect to ${\mathcal{E}}_1$ and dense in $C_0(X)$ with respect to the uniform norm. $\mathcal{H}$ is called a *core* for $\mathcal{D}({\mathcal{E}})$.
We say that the Dirichlet form $({\mathcal{E}},\mathcal{D}({\mathcal{E}}))$ is *local* if for all $u,v\in\mathcal{D}({\mathcal{E}})$ with disjoint compact support ${\mathcal{E}}(u,v)=0$. ${\mathcal{E}}$ is said *strongly local* if $u,v\in \mathcal{D}({\mathcal{E}})$ with compact support and $v$ constant on a neighborhood of $\operatorname{supp}u$ implies ${\mathcal{E}}(u,v)=0$.\
Let $\theta:{\mathbb{R}}^d\to{\mathbb{R}}$ be a non-negative function such that $\theta^{-1},\theta$ are locally integrable on ${\mathbb{R}}^d$. Consider the symmetric form ${\mathcal{E}}$ on $L^2({\mathbb{R}}^d,\theta dx)$ with domain $C_0^\infty({\mathbb{R}}^d)$ defined by $$\label{eq:DF}
{\mathcal{E}}(u,v)\operatorname{\mathrel{\mathop:}=}\sum_{i,j}\int_{{\mathbb{R}}^d} a_{ij}(x)\partial_i u(x)\partial_j v(x)\,dx.$$ Then, $({\mathcal{E}},C_0^\infty({\mathbb{R}}^d))$ is *closable* in $L^2({\mathbb{R}}^d,\theta dx)$ thanks to [@rockner]\[Ch. II example 3b\], since $\lambda^{-1},\Lambda\in L^1_{loc}({\mathbb{R}}^d)$ by \[ass:b.2\]. We shall denote by $({\mathcal{E}},{\mathcal{F}}^\theta)$ such a closure; it is clear that ${\mathcal{F}}^\theta$ is the completion of $C_0^\infty({\mathbb{R}}^d)$ in $L^2({\mathbb{R}}^d,\theta dx)$ with respect to ${\mathcal{E}}_1$. If $u\in {\mathcal{F}}^\theta$, then $u$ is weakly differentiable with derivatives in $L^1_{loc}({\mathbb{R}}^d)$ and ${\mathcal{E}}(u,u)$ takes the form with $\partial_i u$, $i=1,...,d$ being the weak derivative of $u$ in direction $i$. Observe that $({\mathcal{E}},{\mathcal{F}}^\theta)$ is a strongly local regular Dirichlet form, having $C_0^\infty({\mathbb{R}}^d)$ as a core. In the case that $\theta\equiv 1$ we will simply write ${\mathcal{F}}$.
The Dirichlet forms theory [@fukushima1994dirichlet Theorem 7.2.2] allows to construct a diffusion process $\mathbf{M}^{\theta}\operatorname{\mathrel{\mathop:}=}(X_t^\theta,{\mathbb{P}}^\theta_x,\zeta^\theta)$, associated to $({\mathcal{E}},{\mathcal{F}}^\theta)$, starting from all points outside a properly exceptional set. Since we shall work with random media, the set of exceptional points may depend on the particular realization of the environment. In Section \[sec:mindiff\] we shall construct a diffusion process starting for all $x\in{\mathbb{R}}^d$ at the price of local boundedness of the coefficients.
Fix a ball $B\subset{\mathbb{R}}^d$ and consider ${\mathcal{E}}$ as defined in but on $L^2(B,\theta dx)$, and with domain $C_0^\infty(B)$, then clearly $({\mathcal{E}},C_0^\infty(B))$ is closable in $L^2(B,\theta dx)$. We denote by $({\mathcal{E}},{\mathcal{F}}^\theta_B)$ the closure, which also in this case is a strongly local regular Dirichlet form.
Sobolev’s inequalities
----------------------
Let us introduce some notation. Let $B\subset {\mathbb{R}}^d$ be an open bounded set. For a function $u:{\mathbb{R}}^d\to{\mathbb{R}}$ and for $r\geq 1$ we note $$\|u\|_{r} \operatorname{\mathrel{\mathop:}=}\biggl(\int_{{\mathbb{R}}^d} |u(x)|^r dx\biggr)^\frac{1}{r},\quad
\|u\|_{r,\Lambda} \operatorname{\mathrel{\mathop:}=}\biggl(\int_{{\mathbb{R}}^d} |u(x)|^r\Lambda(x)dx\biggr)^\frac{1}{r},\quad\|u\|_{B,r} \operatorname{\mathrel{\mathop:}=}\biggl(\frac{1}{|B|}\int_B |u(x)|^r\,dx\biggr)^\frac{1}{r}.$$
In the next proposition it is enough to assume the local integrability of $\Lambda$ and the $q$-local integrability of $\lambda^{-1}$.
Fix a ball $B\subset{\mathbb{R}}^d$. Then there exists a constant $C_{sob}>0$, depending only on the dimension $d\geq2$, such that for all $u\in{\mathcal{F}}_B$ $$\label{eq:localsobolev}
\|u\|_\rho^2\leq C_{sob}\|1_B \lambda^{-1}\|_q\, {\mathcal{E}}(u,u).$$
We start proving for $u\in C_0^\infty(B)$. Since $\rho$ as defined in is the Sobolev conjugate of $2q/(q+1)$ in ${\mathbb{R}}^d$, by the classical Sobolev’s inequality there exists $C_{sob}>0$ depending only on $d$ such that $$\|u\|_\rho\leq C_{sob} \|\nabla u\|_{2q/(q+1)},$$ where it is clear that we are integrating over $B$. By Hölder’s inequality and \[ass:b.1\] we can estimate the right hand side as follows $$\|\nabla u\|_{2q/(q+1)}^2=\Bigl(\int_B|\nabla u|^\frac{2q}{q+1}\lambda^\frac{q}{q+1}\lambda^{-\frac{q}{q+1}}\,dx\Bigr)^\frac{q+1}{q}\leq \|1_B\lambda^{-1}\|_{q}\,{\mathcal{E}}(u,u),$$ which leads to for $u\in C_0^\infty(B)$. By approximation, the inequality is easily extended to $u\in{\mathcal{F}}_B$.
Fix a ball $B\subset{\mathbb{R}}^d$. Then there exists a constant $C_{sob}>0$, depending only on the dimension $d\geq2$, such that for all $u\in{\mathcal{F}}_B^\Lambda$ $$\label{eq:localweightedsobolev}
\|u\|_{\rho/p^*,\Lambda}^2\leq C_{sob}\|1_B \lambda^{-1}\|_q\|1_B\Lambda\|_p^{2p^*/\rho}\, {\mathcal{E}}(u,u).$$
The proof easily follows from Hölder’s inequality $$\|u\|_{\rho/p^*,\Lambda}^2\leq \|u\|_\rho^2 \|1_B\Lambda\|_p^{2p^*/\rho}$$ and the previous proposition.
From these two Sobolev’s inequalities it follows that the domains ${\mathcal{F}}_B$ and ${\mathcal{F}}_B^\Lambda$ coincide for all balls $B\subset {\mathbb{R}}^d$. Indeed, from and , since $\rho,\rho/p^*>2$, we get that $({\mathcal{F}}_B,{\mathcal{E}})$ and $({\mathcal{F}}_B^\Lambda,{\mathcal{E}})$ are two Hilbert spaces; therefore ${\mathcal{F}}_B,{\mathcal{F}}_B^\Lambda$ coincide with their extended Dirichlet space which by [@Fukushima1987 page 324] is the same, hence ${\mathcal{F}}_B = {\mathcal{F}}_B^\Lambda$.
#### Cutoffs
Since we want to get apriori estimates for solutions to elliptic partial differential equations in the spirit of the classical theory, we will need to work with functions that are locally in ${\mathcal{F}}$ or ${\mathcal{F}}^\Lambda$ and with cutoffs.
Let $B\subset{\mathbb{R}}^d$ be a ball, a cutoff on $B$ is a function $\eta\in C_0^\infty(B)$, such that $0\leq \eta\leq 1$. Given $\theta:{\mathbb{R}}^d\to{\mathbb{R}}$ as before, we say that $u\in{\mathcal{F}}_{loc}^\theta$, if for all balls $B\subset{\mathbb{R}}^d$ there exists $u_B\in{\mathcal{F}}^\theta$ such that $u = u_B$ almost surely on $B$.
In view of these notations, for $u,v\in {\mathcal{F}}^\theta_{loc}$ we define the bilinear form $$\label{eq:dfcutoff}
{\mathcal{E}}_\eta(u,v)=\sum_{i,j}\int_{{\mathbb{R}}^d} a_{ij}(x)\partial_i u(x)\partial_j v(x)\eta^2(x)\,dx.$$
\[lem:cutoff\] Let $B\subset {\mathbb{R}}^d$ and consider a cutoff $\eta\in C_0^\infty(B)$ as above. Then, $u\in{\mathcal{F}}_{loc}\cup {\mathcal{F}}^\Lambda_{loc}$ implies $\eta u\in {\mathcal{F}}_B$.
Take $u\in{\mathcal{F}}_{loc}^\Lambda$, then there exists $\bar{u}\in{\mathcal{F}}^\Lambda$ such that $u=\bar{u}$ on $2B$. Let $\{f_n\}_{\mathbb{N}}\subset C_0^\infty({\mathbb{R}}^d)$ be such that $f_n\to \bar{u}$ with respect to ${\mathcal{E}}+\langle\cdot,\cdot\rangle_\Lambda$. Clearly $\eta f_n\in{\mathcal{F}}_B^\Lambda$ and $\eta f_n\to \eta \bar{u}=\eta u$ in $L^2(B,\Lambda dx)$. Moreover $${\mathcal{E}}(\eta f_n-\eta f_m)\leq2{\mathcal{E}}(f_n-f_m)+\|\nabla\eta\|_\infty^2\int_B|f_n-f_m|^2\Lambda dx.$$ Hence $\eta f_n$ is Cauchy in $L^2(B,\Lambda dx)$ with respect to ${\mathcal{E}}+\langle\cdot,\cdot\rangle_\Lambda$, which implies that $\eta u\in {\mathcal{F}}^\Lambda_B = {\mathcal{F}}_B$. If $u\in{\mathcal{F}}_{loc}$ the proof is similar, and one has only to observe that $\{f_n\}$ is Cauchy in $W^{2q/(q+1)}(B)$, which by Sobolev’s embedding theorem implies that $\{f_n\}$ is Cauchy in $L^2(B,\Lambda dx)$.
Fix a ball $B\subset{\mathbb{R}}^d$ and a cutoff function $\eta\in C_0^\infty (B)$ as above. Then there exists a constant $C_{sob}>0$, depending only on the dimension $d\geq2$, such that for all $u\in{\mathcal{F}}_{loc}^\Lambda\cup{\mathcal{F}}_{loc}$ $$\label{eq:localsobolev-cutoff}
\|\eta u\|_{\rho}^2\leq 2 C_{sob}\|1_B \lambda^{-1}\|_q \Bigl[{\mathcal{E}}_\eta(u,u)+\|\nabla \eta\|_\infty^2\|1_B u\|^2_{2,\Lambda}\Bigr],$$ and $$\label{eq:localweightedsobolev-cutoff}
\|\eta u\|_{\rho/p^*,\Lambda}^2\leq 2 C_{sob}\|1_B \lambda^{-1}\|_q \|1_B\Lambda\|_p^{2p^*/\rho} \Bigl[{\mathcal{E}}_\eta(u,u)+\|\nabla \eta\|_\infty^2\|1_B u\|^2_{2,\Lambda}\Bigr].$$
We prove only $\eqref{eq:localsobolev-cutoff}$, being analogous. Take $u\in {\mathcal{F}}_{loc}\cup\,{\mathcal{F}}^\Lambda_{loc}$, by Lemma \[lem:cutoff\], $\eta u\in{\mathcal{F}}_B$, therefore we can apply and get $$\|\eta u\|_{\rho}^2\leq C_{sob}\|1_B \lambda^{-1}\|_q\, {\mathcal{E}}(\eta u,\eta u).$$ To get we compute $\nabla(\eta u) = u \nabla \eta+\eta\nabla u$ and we easily estimate $$\begin{aligned}
{\mathcal{E}}(\eta u,\eta u) &= \int_{{\mathbb{R}}^d} \langle a \nabla(\eta u),\nabla(\eta u)\rangle dx\\&\leq 2\int_{{\mathbb{R}}^d} \langle a \nabla u,\nabla u\rangle \eta^2 dx + 2\int_{{\mathbb{R}}^d} \langle a \nabla \eta,\nabla \eta\rangle |u|^2 dx
\\&\leq2{\mathcal{E}}_\eta(u,u)+2\|\nabla\eta\|_\infty^2\|1_B u\|_{2,\Lambda}^2.\end{aligned}$$
Maximal inequality for Poisson’s equation
-----------------------------------------
Let $f:{\mathbb{R}}^d\to{\mathbb{R}}$ be some function with essentially bounded weak derivatives. We say that $u\in{\mathcal{F}}_{loc}$ is a solution (subsolution or supersolution) of the Poisson equation, if $$\label{eq:Poisson}
{\mathcal{E}}(u,{\varphi})= -\int_{{\mathbb{R}}^d} \langle a \nabla f, \nabla{\varphi}\rangle dx\quad(\leq\mbox{ or }\geq)$$ for all ${\varphi}\in C_0^\infty({\mathbb{R}}^d)$, ${\varphi}\geq 0$. For a ball $B\subset{\mathbb{R}}^d$, we say that $u\in{\mathcal{F}}_{loc}$ is a solution (subsolution or supersolution) of the Poisson equation in $B$ if $\eqref{eq:Poisson}$ is satisfied for all ${\varphi}\in {\mathcal{F}}_B$, ${\varphi}\geq 0$.
Given a positive subsolution $u\in{\mathcal{F}}_{loc}$ of , we would like to test for ${\varphi}=u^{2\alpha-1} \eta^2$ with $\alpha>1$ and $\eta$ a cutoff function in $B$. The aim is to get a priori estimates for $u$. One must be careful with powers of the function $u$. Indeed, in general $u^{2\alpha-1}$ is not a weakly differentiable function, and therefore it is not clear that ${\varphi}\in {\mathcal{F}}$. The following Lemma is needed to address such a problem
\[lemma:moser\]Let $G:(0,\infty)\to (0,\infty)$ be a Lipschitz function with Lipschitz constant $L_G>0$. Assume also that $G(0+)=0$. Take $u\in {\mathcal{F}}$, $u\geq{\varepsilon}$, for some ${\varepsilon}>0$ then $G(u)\in {\mathcal{F}}$.
The result follows observing that $G(u)/L_G$ is a normal contraction of $u\in {\mathcal{F}}$, and by standard Dirichlet form theory, see [@fukushima1994dirichlet Ch. 1] for details.
Let $u\in {\mathcal{F}}_{loc}$ be a subsolution of in $B$. Let $\eta\in C_0^\infty(B)$ be a cutoff function, $0\leq\eta\leq1$. Then there exists a constant $C_1>0$ such that for all $\alpha\geq 1$ $$\label{eq:meanvalue}
\|\eta u^+\|_{B,\alpha\rho}^{2\alpha} \leq \alpha^2 C_1\|\lambda^{-1}\|_{B,q}\|\Lambda\|_{B,p}|B|^\frac{2}{d}\Bigl[\|\nabla\eta\|^2_\infty\| u^+\|^{2\alpha}_{B,2\alpha p_*}+\|\nabla f\|^2_\infty \|u^+\|^{2\alpha-2}_{B,2\alpha p_*}\Bigr].$$
We can assume $u\in {\mathcal{F}}_{2B}$ since we shall look only inside $B$ and $u\in{\mathcal{F}}_{loc}$. We build here a function $G$ to be a prototype for a power function. Let $G:(0,\infty)\to (0,\infty)$ be a piecewise $C^1$ function such that $G'(s)$ is bounded by a constant say $C>0$. Assume also that $G$ has a non-negative, non-decreasing derivative $G'(x)$ and $G(0+)=0$. Define $H(s)\geq 0$ by $H'(s)=\sqrt{G'(s)}$, $H(0+)=0$. Observe that we have $G(s)\leq sG'(s)$, $H(s)\leq sH'(s)$. Let $\eta$ be a cutoff in $B$ as above. Then, we have by Lemma \[lemma:moser\] and Lemma \[lem:cutoff\] that $${\varphi}=\eta^2 (G(u^++{\varepsilon})-G({\varepsilon}))\in {\mathcal{F}}_{B}.$$ In particular, ${\varphi}$ is a proper test function. In order to lighten the notation we denote $G_{\varepsilon}(x) \operatorname{\mathrel{\mathop:}=}G(x^++{\varepsilon})-G({\varepsilon})$ and $H_{\varepsilon}(x) \operatorname{\mathrel{\mathop:}=}H(x^++{\varepsilon})-H({\varepsilon})$. Since $u$ is a subsolution to in $B$, we have $$\label{eq:comp1}
{\mathcal{E}^\omega}(u,\eta^2 G_{\varepsilon}(u))\leq -\int_{{\mathbb{R}}^d} \langle a \nabla f, \nabla (\eta^2 G_{\varepsilon}(u)) \rangle dx.$$ Consider first the left hand side and observe that $${\mathcal{E}}(u,\eta^2 G_{\varepsilon}(u)) =\int_{{\mathbb{R}}^d}\langle a \nabla u^+,\nabla u^+\rangle G'_{\varepsilon}(u) \eta^2dx+2\int_{{\mathbb{R}}^d} \langle a \nabla u,\nabla \eta\rangle G_{\varepsilon}(u)\eta dx.$$ Since $$\int_{{\mathbb{R}}^d} \langle a \nabla u^+,\nabla u^+\rangle G_{\varepsilon}'(u)\eta^2dx={\mathcal{E}}_\eta(H_{\varepsilon}(u),H_{\varepsilon}(u)),$$ moving everything on the right hand side of , and taking the absolute value, we have $$\label{eq:comp2}
{\mathcal{E}}_\eta(H_{\varepsilon}(u),H_{\varepsilon}(u)) \leq 2 \int_{{\mathbb{R}}^d}| \langle a \nabla u,\nabla \eta\rangle G_{\varepsilon}(u)\eta| dx +\int_{{\mathbb{R}}^d} |\langle a \nabla f, \nabla (G_{\varepsilon}(u)\eta^2) \rangle | dx.$$ The first term is estimated using $G_{\varepsilon}(u)\leq u^+ G'_{\varepsilon}(u)$ and by Cauchy Schwartz inequality. (We use also the fact that $u^+\nabla u = u^+\nabla u^+$). $$\begin{aligned}
\int_{{\mathbb{R}}^d} |\langle a \nabla u,\nabla \eta\rangle G_{\varepsilon}(u)\eta |dx\leq & \int_{{\mathbb{R}}^d} |\langle a \nabla u^+,\nabla \eta\rangle G'_{\varepsilon}(u) u^+\eta |dx \\ \leq&{\mathcal{E}}_\eta(H_{\varepsilon}(u),H_{\varepsilon}(u))^\frac{1}{2}\|G_{\varepsilon}'(u)(u^+)^2\|_{1,\Lambda}^\frac{1}{2}\|\nabla \eta\|_\infty .\end{aligned}$$ The second term, after using Leibniz rule, is controlled by $$\int_{{\mathbb{R}}^d} |\langle a\nabla f, \nabla u^+\rangle G_{\varepsilon}'(u)\eta^2 |dx+ 2 \int_{{\mathbb{R}}^d} |\langle a\nabla f, G_{\varepsilon}(u)\eta \nabla\eta \rangle|dx$$ whose terms can be estimated by $$\int_{{\mathbb{R}}^d} |\langle a\nabla f, \nabla u^+\rangle G'_{\varepsilon}(u)\eta^2 |dx \leq \|\nabla f\|_\infty \|1_B G'_{\varepsilon}(u)\|_{1,\Lambda}^\frac{1}{2} {\mathcal{E}}_\eta(H_{\varepsilon}(u),H_{\varepsilon}(u))^\frac{1}{2}$$ and by $$\int_{{\mathbb{R}}^d} |\langle a\nabla f,\nabla\eta \rangle G_{\varepsilon}(u)\eta |dx \leq \|\nabla\eta\|_\infty \|\nabla f\|_\infty \|G_{\varepsilon}(u)1_B\|_{1,\Lambda}.$$ Putting everything together in we end up with the estimate $$\begin{aligned}
{\mathcal{E}}_\eta(H_{\varepsilon}(u),H_{\varepsilon}(u)) &\leq 2 \|G_{\varepsilon}'(u)(u^+)^2\|_{1,\Lambda}^\frac{1}{2}\|\nabla \eta\|_\infty{\mathcal{E}}_\eta(H_{\varepsilon}(u),H_{\varepsilon}(u))^\frac{1}{2}\\ &+ \|\nabla f\|_\infty \|1_B G'_{\varepsilon}(u)\|_{1,\Lambda}^\frac{1}{2} {\mathcal{E}}_\eta(H_{\varepsilon}(u),H_{\varepsilon}(u))^\frac{1}{2}
\\& +2\|\nabla\eta\|_\infty \|\nabla f\|_\infty \|G_{\varepsilon}(u)1_B\|_{1,\Lambda},\end{aligned}$$ which finally gives, up to a universal constant $c>0$, $$\begin{aligned}
{\mathcal{E}}_\eta(H_{\varepsilon}(u),H_{\varepsilon}(u))\leq & c\Bigl[ \|G'_{\varepsilon}(u)(u^+)^2\|_{1,\Lambda}\|\nabla\eta\|^2_\infty+\|\nabla f\|^2_\infty \|1_B G_{\varepsilon}'(u)\|_{1,\Lambda} \\&+\|\nabla\eta\|_\infty \|\nabla f\|_\infty \|G_{\varepsilon}(u)1_B\|_{1,\Lambda}\Bigr].\end{aligned}$$ At this point, it is important to observe that $H_{\varepsilon}(u)\in {\mathcal{F}}$ so that we can apply the Sobolev’s inequality with cut-off function $\eta$, namely $$\|\eta H_{\varepsilon}(u)\|^2_\rho\leq 2C_{sob}\|1_B\lambda^{-1}\|_{q} \Bigl[{\mathcal{E}^\omega}_\eta (H_{\varepsilon}(u),H_{\varepsilon}(u))+ \|\nabla \eta\|^2_\infty \|1_B H_{\varepsilon}(u)\|^2_{2,\Lambda}\Bigr].$$ Concatenating the two inequalities yields $$\begin{gathered}
\|\eta H_{\varepsilon}(u)\|^2_\rho\leq 2 C_1\|1_B\lambda^{-1}\|_{q}\Bigl[\|H_{\varepsilon}'(u)^2 u^2\|_{1,\Lambda}\|\nabla\eta\|^2_\infty+\|\nabla f\|^2_\infty \|1_B H_{\varepsilon}'(u)^2\|_{1,\Lambda} \\+\|\nabla\eta\|_\infty \|\nabla f\|_\infty \|G_{\varepsilon}(u)1_B\|_{1,\Lambda}+\|\nabla \eta\|^2_\infty \|1_B H_{\varepsilon}(u)\|^2_{2,\Lambda}\Bigr]\end{gathered}$$ Finally it is time to fix a $H,G$ as power-like function. Namely we take, for $\alpha>1$ $$H_N(x)\operatorname{\mathrel{\mathop:}=}\begin{cases}x^\alpha & x\leq N\\
\alpha N^{\alpha-1}x+(1-\alpha)N^{\alpha} & x>N
\end{cases}$$ which corresponds in taking $$G_N(x)=\int_0^x H_N'(s)^2\,ds.$$ The function $G_N(x)$ has the right properties, moreover $H_N(x)\uparrow x^\alpha$ and $G_N(x)\uparrow \frac{\alpha^2}{2\alpha-1} x^{2\alpha-1}$ as $N$ goes to infinity. Therefore, letting $N\to\infty$, and using the monotone convergence theorem, we obtain $$\begin{gathered}
\|\eta (u^++{\varepsilon})^\alpha\|^2_\rho\leq 2 C_1\|1_B\lambda^{-1}\|_{q}\Bigl[(\alpha^2+1)\| 1_B (u^++{\varepsilon})^{2\alpha}\|_{1,\Lambda}\|\nabla\eta\|^2_\infty\\+\|\nabla f\|^2_\infty \alpha^2 \|1_B (u^++{\varepsilon})^{2\alpha-2}\|_{1,\Lambda} +\frac{\alpha^2}{2\alpha-1}\|\nabla\eta\|_\infty \|\nabla f\|_\infty \|u^{2\alpha-1}1_B\|_{1,\Lambda}\Bigr].\end{gathered}$$ Taking the limit as ${\varepsilon}\to 0$ and averaging over balls we get $$\begin{gathered}
\|\eta (u^+)^\alpha\|^2_{B,\rho} \leq 2 C_1\|\lambda^{-1}\|_{B,q}\|\Lambda\|_{B,p}|B|^\frac{2}{d}\Bigl[(\alpha^2+1)\| (u^+)^{2\alpha}\|_{B,p_*}\|\nabla\eta\|^2_\infty\\+\|\nabla f\|^2_\infty \alpha^2 \|(u^+)^{2\alpha-2}\|_{B,p_*} +\frac{\alpha^2}{2\alpha-1}\|\nabla\eta\|_\infty \|\nabla f\|_\infty \|(u^+)^{2\alpha-1}\|_{B,p_*}\Bigr].\end{gathered}$$ By Jensen’s inequality it holds $$\|u^+\|_{B,(2\alpha-2)p_*}\leq \|u^+\|_{B,2\alpha p_*},\quad\|u^+\|_{B,(2\alpha-1)p_*}\leq \|u^+\|_{B,2\alpha p_*},$$ therefore we can rewrite and get the desired result $$\begin{gathered}
\|\eta u^+\|_{B,\alpha\rho}^{2\alpha} \leq 2 C_1\|\lambda^{-1}\|_{B,q}\|\Lambda\|_{B,p}|B|^\frac{2}{d}\Bigl[(\alpha^2+1)\| u^+\|^{2\alpha}_{B,2\alpha p_*}\|\nabla\eta\|^2_\infty\\+\|\nabla f\|^2_\infty \alpha^2 \|u^+\|^{2\alpha-2}_{B,2\alpha p_*} +\frac{\alpha^2}{2\alpha-1}\|\nabla\eta\|_\infty \|\nabla f\|_\infty \|u^+\|^{2\alpha-1}_{B,2\alpha p_*}\Bigr].\end{gathered}$$ Finally, absorbing the mixed product in the two squares we obtain .
Clearly the same result holds, with the same constant, also for supersolutions with $u^+$ replaced by $u^-$. It is then clear that we can get the same type of inequality for solutions to . This is the content of the next corollary.
\[cor\] Let $u\in {\mathcal{F}}_{loc}$ be a solution of in $B$. Let $\eta\in C_0^\infty(B)$ be a cut-off function. Then there exists a constant $C_1>0$ such that for all $\alpha\geq 1$ $$\label{eq:corollary}
\|\eta u\|_{B,\alpha\rho}^{2\alpha} \leq \alpha^2 C_1\|\lambda^{-1}\|_{B,q}\|\Lambda\|_{B,p}|B|^\frac{2}{d}\Bigl[\|\nabla\eta\|^2_\infty\| u\|^{2\alpha}_{B,2\alpha p_*}+\|\nabla f\|^2_\infty \|u\|^{2\alpha-2}_{B,2\alpha p_*}\Bigr].$$
The proof is trivial, since $u$ is both a subsolution and a supersolution of . Moreover, $u=u^+-u^-$ and $\|u^+\|_r\vee \|u^-\|_r\leq\|u\|_r$.
\[thm:meanvalueinequality\] Fix a point $x_0\in {\mathbb{R}}^d$ and $R>0$. Denote by $B(R)$ the ball of center $x_0$ and radius $R$. Suppose that $u$ is a solution in $B(R)$ of , and assume that $|\nabla f|\leq c_f/R$. Then for any $p,q\in (1,\infty]$ such that $1/p+1/q<2/d$, $d\geq 2$, there exist $\kappa\operatorname{\mathrel{\mathop:}=}\kappa(q,p,d)\in (1,\infty)$, $\gamma\operatorname{\mathrel{\mathop:}=}\gamma(q,p,d)\in (0,1]$ and $C_2\operatorname{\mathrel{\mathop:}=}C_2(q,p,d,c_f)>0$ such that $$\label{eq:moser2}
\|u\|_{B(\sigma'R),\infty}\leq C_2
\biggl(\frac{1\vee\|\lambda^{-1}\|_{B(R),q}\|\Lambda\|_{ B(R),p}}{(\sigma-\sigma')^2 }\biggr)^\kappa \|u\|_{B(\sigma R),\rho}^\gamma\vee\|u\|_{B(\sigma R),\rho},$$ for any fixed $1/2\leq \sigma'<\sigma\leq 1$.
We are going to apply inequality iteratively. For fixed $1/2\leq \sigma'<\sigma\leq 1$, and $k\in{\mathbb{N}}$ define $$\sigma_k=\sigma'+2^{-k+1}(\sigma-\sigma').$$ It is immediate that $\sigma_k-\sigma_{k+1}=2^{-k+1}(\sigma-\sigma')$ and that $\sigma_1 = \sigma$, furthermore $\sigma_k\downarrow \sigma'$. We have already observed that $\rho>2p^*$, where $p^*$ is the Hölder’s conjugate of $p$. Set $\alpha_k \operatorname{\mathrel{\mathop:}=}(\rho/2 p^*)^k$, $k\geq1$, clearly $\alpha_k>1$ for all $k\geq 1$. Finally consider a cutoff $\eta_k$ which is identically $1$ on $B(\sigma_{k+1} R)$ and $\eta_k = 0$ on $\partial B(\sigma_k R)$, assume that $\eta_k$ has a linear decay on $B(\sigma_k R)\setminus B(\sigma_{k+1} R)$, i.e. chose $\eta_k$ in such a way that $\|\nabla \eta_k\|_\infty \leq 2^{k} / (\sigma-\sigma')R$.
An application of Corollary \[cor\] and of the relation $\alpha_k\rho = 2\alpha_{k+1}p^*$, yields $$\begin{aligned}
\|u&\|_{B(\sigma_{k+1}R),2\alpha_{k+1}p^*}\\
&\leq \biggl(C\frac{2^{2k}\alpha_k^2|B(\sigma_k R)|^\frac{2}{d}}{(\sigma-\sigma')^2 R^2}\|\lambda^{-1}\|_{B(\sigma_k R),q}\|\Lambda\|_{ B(\sigma_k R),p}\biggr)^\frac{1}{2\alpha_k} \|u\|_{B(\sigma_k R),2\alpha_k p_*}^{\gamma_k}\\
&\leq
\biggl(C\frac{2^{2k}\alpha_k^2}{(\sigma-\sigma')^2 }\|\lambda^{-1}\|_{B(R),q}\|\Lambda\|_{ B(R),p}\biggr)^\frac{1}{2\alpha_k} \|u\|_{B(\sigma_k R),2\alpha_k p^*}^{\gamma_k},\end{aligned}$$ where $\gamma_k = 1$ if $\|u\|_{B(\sigma_k R),2\alpha_k p^*}\geq1$ and $\gamma_k = 1-1/\alpha_k$ otherwise. We can iterate the inequality above and stop at $k=1$, so that we get $$\|u\|_{B(\sigma_{j+1}R),2\alpha_{j+1}p^*}\leq
\prod_{k=1}^j\biggl(C\frac{(\rho/p^*)^{2k}}{(\sigma-\sigma')^2 }\|\lambda^{-1}\|_{B(R),q}\|\Lambda\|_{ B(R),p}\biggr)^\frac{1}{2\alpha_k} \|u\|_{B(\sigma R),\rho}^{\prod_{k=1}^j\gamma_k}.$$ Observe that $\kappa\operatorname{\mathrel{\mathop:}=}\frac{1}{2}\sum1/\alpha_k<\infty$, $\sum k/\alpha_k<\infty$ and that $$\|u\|_{B(\sigma'R),2\alpha_j p^*}\leq \biggl(\frac{|B(\sigma_kR)|}{|B(\sigma'R)|}\biggr)^\frac{1}{2\alpha_j p^*}\|u\|_{B(\sigma_jR),2\alpha_j p^*}\leq K\|u\|_{B(\sigma_jR),2\alpha_j p^*},$$ for some $K$ and all $j\geq 1$. Hence, taking the limit as $j\to\infty$, gives the inequality $$\|u\|_{B(\sigma'R),\infty}\leq C_2
\biggl(\frac{1\vee\|\lambda^{-1}\|_{B(R),q}\|\Lambda\|_{ B(R),p}}{(\sigma-\sigma')^2 }\biggr)^\kappa \|u\|_{B(\sigma R),\rho}^{\prod_{k=1}^\infty \gamma_k}.$$ Define $\gamma\operatorname{\mathrel{\mathop:}=}\prod_{k=1}^\infty (1-1/\alpha_k) \in (0,1]$. Then, $0<\gamma \leq \prod_{k=1}^\infty \gamma_k\leq 1$ and the above inequality can be written as $$\|u\|_{B(\sigma'R),\infty}\leq C_2
\biggl(\frac{1\vee\|\lambda^{-1}\|_{B(R),q}\|\Lambda\|_{ B(R),p}}{(\sigma-\sigma')^2 }\biggr)^\kappa \|u\|_{B(\sigma R),\rho}^\gamma\vee\|u\|_{B(\sigma R),\rho}.$$ which is the desired inequality.
The previous inequality can be improved. This is what the next Corollary is about. For the proof we follow the argument of [@saloff2002aspects]\[Theorem 2.2.3\].
Suppose that $u$ satisfies the assumptions of Theorem \[thm:meanvalueinequality\]. Then, for all $\alpha\in(0, \infty)$ and for any $1/2\leq \sigma'<\sigma < 1$ there exist $C_3:=C_3(q,p,d,c_f)>0$, $\gamma'\operatorname{\mathrel{\mathop:}=}\gamma'(\gamma,\alpha,\rho)$ and $\kappa'\operatorname{\mathrel{\mathop:}=}\kappa'(\kappa,\alpha,\rho)$, such that $$\label{cor:moser}
\|u\|_{B(\sigma'R),\infty}\leq C_3
\biggl(\frac{1\vee\|\lambda^{-1}\|_{B(R),q}\|\Lambda\|_{ B(R),p}}{(\sigma-\sigma')^2 }\biggr)^{\kappa'} \|u\|_{B(\sigma R),\alpha}^{\gamma'}\vee\|u\|_{B(\sigma R),\alpha}.$$
From inequality we get $$\|u\|_{B(\sigma'R),\infty}\leq C_2
\biggl(\frac{1\vee\|\lambda^{-1}\|_{B(R),q}\|\Lambda\|_{ B(R),p}}{(\sigma-\sigma')^2 }\biggr)^\kappa \|u\|_{B(\sigma R),\rho}^{\gamma}\vee \|u\|_{B(\sigma R),\rho}.$$ Hence, the result follows immediately for $\alpha>\rho$ by means of Jensen’s inequality. For $\alpha\in(0,\rho)$ we use again an iteration argument. Consider $\sigma_k=\sigma-2^{-k}(\sigma-\sigma')$. By Hölder’s inequality we get $$\|u\|_{B(\sigma_k R),\rho}\leq \|u\|_{B(\sigma_k R),\alpha}^\theta\|u\|_{B(\sigma_k R),\infty}^{1-\theta}$$ with $\theta= \alpha/\rho$. An application of inequality gives $$\|u\|_{B(\sigma_{k-1}R),\infty}\leq\\ 2^{2\kappa k} J \|u\|_{B(\sigma R),\alpha}^{\gamma_k\theta}\|u\|_{B(\sigma_k R),\infty}^{\gamma_k-\gamma_k\theta},$$ here $\gamma_k=1$ if $\|u\|_{B(\sigma_k R),\rho}\geq 1$, $\gamma_k = \gamma$ otherwise and $J = c(1\vee\|\lambda^{-1}\|_{B(R),q}\|\Lambda\|_{ B(R),p}/(\sigma-\sigma')^2 )^{\kappa}$, where $c$ is a constant that can be taken greater than one.
By iteration from $k=1$ up to $i>1$, via similar computations as the Theorem \[thm:meanvalueinequality\], we get $$\|u\|_{B(\sigma' R),\infty}\leq (J 2^{2\kappa})^{\sum_{k=1}^{i}k(1-\theta)^{k-1}}\Bigl(\|u\|_{B(\sigma R),\alpha}^{\gamma\theta\sum_{k=1}^{i}(\gamma-\gamma\theta)^{k-1}}\vee \|u\|_{B(\sigma R),\alpha}^{\theta\sum_{k=1}^{i}(1-\theta)^{k-1}}\Bigr) \|u\|_{B(\sigma R),\infty}^{\beta_i}$$ where $\beta_i\to 0$ as $i\to \infty$. which gives the desired result taking the limit as $i\to\infty$. In particular we get $\gamma' = \gamma \theta / (1-\gamma+\gamma \theta)$.
Existence of the Minimal Diffusion {#sec:mindiff}
----------------------------------
In the context of diffusions in random environment we would like to be able to fix a common starting position for almost all realizations of the environment, or alternatively to start the process from all possible positions $x\in{\mathbb{R}}^d$. To achieve this aim we assume the following:
- $\lambda^{-1}(x),\Lambda(x)\in L^\infty_{loc}({\mathbb{R}}^d)$.
Recall that the resolvent $G_\alpha^{B,\theta}$ restricted to $B$ of a diffusion process $\mathbf{M}^{\theta}\operatorname{\mathrel{\mathop:}=}(X_t^\theta,{\mathbb{P}}^\theta_x,\zeta^\theta)$ is defined by $$G_\alpha^{B,\theta} f(x)\operatorname{\mathrel{\mathop:}=}{\mathbb{E}}^\theta_x\biggl[\int_0^{\tau_B} e^{-\alpha t}f(X^\theta_t)\,dt\biggr],\quad f\geq0$$ being $\tau_B = \inf\{t>0: X^\theta_t\in B^c\}$. When $\theta\equiv 1$ we will drop it from the notation.
\[thm:min\_diff\] Assume \[ass:b.1\], \[ass:b.2\], \[ass:b.3\], and $\theta,\theta^{-1}\in L^\infty_{loc}({\mathbb{R}}^d)$. Denote by $C_\infty(B)$ the set of continuous functions vanishing at the boundary. Then, there exists a unique standard diffusion process $\mathbf{M}^{\theta}\operatorname{\mathrel{\mathop:}=}(X_t^\theta,{\mathbb{P}}^\theta_x,\zeta^\theta)$, $x\in {\mathbb{R}}^d$ whose resolvent $G^{B,\theta}_\alpha$ restricted to any open bounded set $B$ satisfies $$G_\alpha^{B,\theta} f \in C_\infty(B),\quad f\in L^p(B,\theta dx),\quad p>d$$ and $G_\alpha^{B,\theta} C_\infty(B)$ is dense in $C_\infty(B)$.
For a proof see for example [@ichihara], [@kunita1970], [@tomisaki].
We will consider from now on only the process $\mathbf{M}^\theta$ constructed in Theorem \[thm:min\_diff\]. Fix a ball $B\subset {\mathbb{R}}^d$ and consider the semigroup associated to the process above killed when exiting from $B$, then its semigroup is given by $$\mathcal{P}_t^{B,\theta} f(x)\operatorname{\mathrel{\mathop:}=}{\mathbb{E}}_x[f(X^{\theta}_t),t<\tau_B],$$ By Theorem \[thm:min\_diff\] and Hille-Yoshida’s Theorem, $\mathcal{P}_t^{B,\theta} C_\infty(B)\subset C_\infty(B)$. Such a property turns out to be very handy to remove all the ambiguities about exceptional sets and to construct a transition kernel $p_t^{B,\theta}(x,y)$ for $\mathcal{P}_t^{B,\theta}$ which is jointly continuous in $x,y$. This is the content of the next theorem whose proof is a slight variation of [@Barlow99non-localdirichlet Theorem 2.1] since we assume to have a Feller semigroup.
\[jointcont\] Let $B\subset {\mathbb{R}}^d$ a ball and $\mathcal{P}_t$ be a Feller semigroup on $L^2(B,m)$, i.e. $\mathcal{P}_t C_\infty(B)\subset C_\infty(B)$. Assume that $$\label{ass:ultracontractivity}
\|\mathcal{P}_t f\|_\infty\leq M(t)\|f\|_1,$$ for all $f\in L^1(B,m)$ and $t>0$ and some lower semicontinuous function $M(t)$ on $(0,\infty)$. Then there exists a positive symmetric kernel $p_t(x,y)$ defined on $(0,\infty)\times B\times B$ such that
- $\mathcal{P}_t(x,dy) = p_t(x,y) m(dy)$, for all $x\in B$, $t>0$,
- for every $t,s>0$ and $x,y\in B$ $$p_{t+s}(x,y)=\int_B p_t(x,z)p_s(z,y)m(dz),$$
- $p_t(x,y)\leq M(t)$ for every $t>0$ and $x,y\in B$,
- for every fixed $t>0$, $p_t(x,y)$ is jointly continuous in $x,y\in B$.
We see that if we choose $m(dx)= \theta(x) dx$ and we assume \[ass:b.1\], \[ass:b.2\], \[ass:b.3\] we immediately get the existence of a transition kernel $p^{B,\theta}_t(x,y)$ for the semigroup $\mathcal{P}_t^{B,\theta}$, jointly continuous in $x,y\in B$. Indeed assumption is easily satisfied by \[ass:b.3\]. In the next proposition we prove the existence of a transition kernel $p_t^\theta(x,y)$ for the semigroup $\mathcal{P}^\theta_t$ of $\mathbf{M}^\theta$ by a localization argument.
Assume \[ass:b.3\] and $\theta,\theta^{-1}\in L^\infty_{loc}({\mathbb{R}}^d)$. Consider the semigroup $\mathcal{P}_t^\theta$ associated to the minimal diffusion $\mathbf{M}^\theta$. Then, there exists a transition kernel $p_t^{\theta}(x,y)$ defined on $(0,\infty)\times {\mathbb{R}}^d\times {\mathbb{R}}^d$ associated to $\mathcal{P}_t^\theta$, $$\mathcal{P}_t^\theta f(x)=\int_{{\mathbb{R}}^d}f(y)p_t^{\theta}(x,y)\theta(y)\,dy,\quad \forall x\in{\mathbb{R}}^d,\,t>0.$$ Moreover, for all $t>0$ and $x,y\in {\mathbb{R}}^d$ $$p^{B_R,\theta}_t(x,y) \nearrow p^\theta_t(x,y),\quad R\to\infty,$$ being the limit increasing in $R$.
The proof comes from the the fact that for all balls $B\subset{\mathbb{R}}^d$ the semigroup $\mathcal{P}_t^{B,\theta}$ satisfies , which means that $\mathcal{P}_t^{\theta}$ is locally ultracontractive and from Theorem 2.12 of [@grigor'yan2012].
As a further consequence of assumption \[ass:b.3\], more precisely from the fact that $\lambda$ is locally bounded from below we can prove that $\mathbf{M}^\theta$ is an irreducible process.
\[prop:irr\] Assume \[ass:b.3\] and assume $\theta^{-1},\theta\in L^\infty_{loc}({\mathbb{R}}^d)$. Then the process $\mathbf{M}^\theta$ is irreducible.
It follows immediately from Corollary 4.6.4. in [@fukushima1994dirichlet].
In the next theorem we clarify the relation between $\mathbf{M}$ and $\mathbf{M}^\theta$, namely, we show that $\mathbf{M}^\theta$ can be obtained by $\mathbf{M}$ through a time change.
\[thm:time\_change\] Assume \[ass:b.3\] and assume $\theta^{-1},\theta\in L^\infty_{loc}({\mathbb{R}}^d)$. Define $\hat{\mathbf{M}}=(\hat{X}_t,{\mathbb{P}}_x)$ by $$\hat{X}_t\operatorname{\mathrel{\mathop:}=}X_{\tau_t},\quad\tau_t=\inf\{s>0;\int_0^s\theta(X_u)\,du>t\},$$ then $\hat{\mathcal{P}}_tf(x)={\mathbb{E}}_x[f(X_{\tau_t})]=\mathcal{P}_t^\theta f(x)$ for almost all $x\in{\mathbb{R}}^d$, $t>0$ and $f:{\mathbb{R}}^d\to{\mathbb{R}}$ positive and measurable.
According to Theorem 6.2.1 of [@fukushima1994dirichlet], $\hat{\mathcal{P}}_tf(x)=\mathcal{P}_t^\theta f(x)$ coincide for almost all $x\in{\mathbb{R}}^d$ and $t>0$.
There is a natural time change $\theta:{\mathbb{R}}^d\to{\mathbb{R}}_{\geq0}$ which makes the process $\mathbf{M}^\theta$ conservative. Namely we pick $\theta\equiv \Lambda$. The condition we give will be suitable in the setting of Ergodic environment, and in particular, is a consequence of \[ass:b.2\].
\[prop:explosion\] Assume that $$\limsup_{R\to\infty} \frac{1}{|B_R|}\int_{B_R} \Lambda(x)\,dx<\infty.$$ Then the process $\mathbf{M}^\Lambda$ is conservative.
The proof is an application of Theorem 5.7.3 of [@fukushima1994dirichlet].
Diffusions in Random Environment {#sec:diffre}
================================
Construction of the Process in Random Environment
-------------------------------------------------
By a stationary and ergodic random environment $(\Omega,\mathcal{G},\mu,\{\tau_x\}_{{\mathbb{R}}^d})$, we mean a probability space $(\Omega,\mathcal{G},\mu)$ on which is defined a group of transformations $\{\tau_x\}_{x\in{\mathbb{R}}^d}$ acting on $\Omega$ such that
- $\mu(\tau_x A) = \mu(A)$ for all $A\in\mathcal{G}$ and any $x\in{\mathbb{R}}^d$;
- if $\tau_x A = A$ for all $x\in{\mathbb{R}}^d$, then $\mu(A)\in\{0,1\}$;
- the function $(x,\omega)\to \tau_x\omega$ is $\mathcal{B}({\mathbb{R}}^d)\otimes\mathcal{G}$-measurable.
Let us consider the following bilinear form $${\mathcal{E}^\omega}(u,v)\operatorname{\mathrel{\mathop:}=}\sum_{i,j}\int_{{\mathbb{R}}^d} a^\omega_{ij}(x)\partial_i u(x)\partial_j v(x) dx,\quad u,v\in C_0^\infty({\mathbb{R}}^d),$$ where $a^\omega_{ij}(x)$ satisfies \[ass:a.1\], \[ass:a.2\] and \[ass:a.3\] of Section 1.
Throughout this section we will look at two Dirichlet forms determined by ${\mathcal{E}}^\omega$ above. One is the Dirichlet form $({\mathcal{E}}^\omega,{\mathcal{F}}^{\omega})$ on $L^2({\mathbb{R}}^d,dx)$ where ${\mathcal{F}}^\omega$ is the completion of $C_0^\infty({\mathbb{R}}^d)$ in $L^2({\mathbb{R}}^d,dx)$ with respect to ${\mathcal{E}}^\omega_1 := {\mathcal{E}}^\omega+(\cdot,\cdot)$. The second is the Dirichlet form $({\mathcal{E}}^\omega,{\mathcal{F}}^{\Lambda,\omega})$ on $L^2({\mathbb{R}}^d,\Lambda^\omega dx)$ where ${\mathcal{F}}^{\Lambda,\omega}$ is the completion of $C_0^\infty({\mathbb{R}}^d)$ in $L^2({\mathbb{R}}^d,\Lambda^\omega dx)$ with respect to ${\mathcal{E}}^\omega_1 := {\mathcal{E}}^\omega+(\cdot,\cdot)_\Lambda$.
We have already observed that \[ass:a.1\], \[ass:a.2\] and \[ass:a.3\] imply \[ass:b.1\], \[ass:b.2\] and \[ass:b.3\] of Section \[sec:sobmos\], for $\mu$-almost all $\omega\in\Omega$. In particular, by Theorem \[thm:min\_diff\], we have the existence, for $\mu$-almost all $\omega\in\Omega$, of two minimal diffusion processes, $\mathbf{M}^{\omega} = (X_t^{\omega},{\mathbb{P}}_x^{\omega},\zeta^\omega)$ and $\mathbf{M}^{\Lambda,\omega} = (X_t^{\Lambda,\omega},{\mathbb{P}}_x^{\Lambda,\omega})$, respectively associated to $({\mathcal{E}}^\omega,{\mathcal{F}}^{\omega})$ and $({\mathcal{E}}^\omega,{\mathcal{F}}^{\Lambda,\omega})$. Denote by $\mathcal{P}_t^\omega$ the semigroup associated to $\mathbf{M}^\omega$ and by $p_t^\omega(x,y)$ its transition kernel with respect to $dx$. Analogously, denote by $\mathcal{Q}_t^\omega$ the semigroup associated to $\mathbf{M}^{\Lambda,\omega}$ and by $q_t^\omega(x,y)$ its transition kernel with respect to $\Lambda^\omega(x)dx$.
Fix a ball $B\subset{\mathbb{R}}^d$. Then for $\mu$-almost all $\omega\in\Omega$ $$\begin{aligned}
\label{eq:homogeneity}
p_t^{B-z,\tau_z\omega}(x-z,y-z)&=p_t^{B,\omega}(x,y),\\q_t^{B-z,\tau_z\omega}(x-z,y-z)&=q_t^{B,\omega}(x,y),\notag\end{aligned}$$ for all $t\geq0$, $x,y\in B$ and $z\in{\mathbb{R}}^d$.
We prove property only for the semigroup $\mathcal{Q}^{B,\omega}_t$, being the other equivalent. It is known in [@fukushima1994dirichlet] that the resolvent $G_\alpha^{B,\omega}$ is uniquely determined by the following equation $${\mathcal{E}}^\omega_\alpha(G^{B,\omega}_\alpha f,v)=\int_B f(x)v(x) \Lambda(x;\omega)\,dx,$$ for all $f\in L^2(B)$, $v\in W^2_0(B)$. On the other hand $$\begin{aligned}
{\mathcal{E}}^\omega_\alpha(G^{B,\omega}_\alpha f,v) &=\int_{B-z} f(x+z)v(x+z) \Lambda(x;\tau_z\omega)\,dx\\
&={\mathcal{E}}^{\tau_z\omega}_\alpha([G^{B-z,\tau_z\omega}_\alpha f(\cdot+z)],v(\cdot+z))\\
&= {\mathcal{E}}^\omega_\alpha([G^{B-z,\tau_z\omega}_\alpha f(\cdot+z)](\cdot-z),v),\end{aligned}$$ for all $f\in L^2(B)$, $v\in W^2_0(B)$. Hence, for $\mu$-almost all $\omega\in\Omega$ $$[G^{B-z\tau_z\omega}_\alpha f(\cdot+z)](x-z)=G^{B,\omega}_\alpha f(x),\quad \mbox{a.a }x\in B, \forall z\in{\mathbb{R}}^d.$$ Moving from the resolvent to the semigroup we get the relation $$[\mathcal{Q}^{B-z,\tau_z\omega}_t f(\cdot+z)](x-z)=\mathcal{Q}^{B,\omega}_t f(x),$$ for all $f\in C_\infty(B)$. The equality is true for all $x\in B$ and for all $z\in{\mathbb{R}}^d$ by the Feller property, $\mu$-almost surely. Finally it is easy to derive the equality for the transition kernel and get $$\label{eq:hom}
q_t^{B-z,\tau_z\omega}(x-z,y-z)=q_t^{B,\omega}(x,y),$$ for all $z\in{\mathbb{R}}^d$, and almost all $x,y\in B$, $\mu$-almost surely. Using the joint continuity of $q_t^{B,\omega}(x,y)$ in $x$ and $y$ (cf. $(iv)$ Theorem \[jointcont\]) we get for all $z\in{\mathbb{R}}^d$, $x,y\in B$, $\mu$-almost surely.
For $\mu$-almost all $\omega\in\Omega$ $$\begin{aligned}
\label{eq:homogeneity2}
p_t^{\tau_z\omega}(x-z,y-z)&=p_t^{\omega}(x,y),\\q_t^{\tau_z\omega}(x-z,y-z)&=q_t^{\omega}(x,y),\notag\end{aligned}$$ for all $t\geq 0$ and $x,y,z\in{\mathbb{R}}^d$
It follows from the previous lemma, passing to the limit. Namely, take an increasing sequence of balls $B_n\uparrow {\mathbb{R}}^d$, then we have $$\begin{aligned}
p_t^{\tau_z\omega}(x-z,y-z) & = \lim_{n\to\infty}p_t^{B_n-z,\tau_z\omega}(x-z,y-z)\\& =\lim_{n\to\infty}p_t^{B_n,\omega}(x,y)=p_t^{\omega}(x,y).\end{aligned}$$
Environment Process
-------------------
We shall first construct the environment process for $\mathbf{M}^{\Lambda.\omega}=(X^{\Lambda,\omega}_t,{\mathbb{P}}^{\Lambda,\omega}_x)=:(Y^{\omega}_t,\mathbb{Q}^{\omega}_x)$, $x\in{\mathbb{R}}^d$, since we know that it is conservative $\mu$-almost surely by Proposition \[prop:explosion\]. From this construction and the Ergodic theorem we will prove that also the process $\mathbf{M}^\omega$ is conservative $\mu$-almost surely.
For a fixed $\omega\in\Omega$, we define a stochastic process on $\Omega$ by $$\eta_t^\omega(\tilde{\omega})\operatorname{\mathrel{\mathop:}=}\tau_{Y_t^\omega(\tilde{\omega})}\omega,\quad t\geq 0$$ where $\tilde{\omega}$ is a point of the sample space of the diffusion $\mathbf{M}^{\Lambda,\omega}$. The process $\eta_t^\omega$ under the measure $\mathbb{Q}^{\omega}_x$ is $\Omega$ valued and it is known as the environment process. First, we describe the semigroup associated to $\eta_t^\omega$ under $\mathbb{Q}^{\omega}_0$. Take any positive and bounded $\mathcal{G}$-measurable function $f:\Omega\to {\mathbb{R}}$ and observe that $$\mathbf{Q}_t f(\omega)\operatorname{\mathrel{\mathop:}=}{\mathbb{E}}_0^{\omega} [f(\tau_{Y^\omega_t}\omega)] = \mathcal{Q}_t^\omega f(\tau_.\omega)(0)=\int_{{\mathbb{R}}^d}f(\tau_y\omega) q^\omega_t(0,y)\Lambda(\tau_y \omega)\,dy.$$
\[prop:symmetric\] $\{\mathbf{Q}_t\}_{t\geq 0}$ defines a symmetric strongly continuous semigroup on $L^2(\Omega,\Lambda d\mu)$, the process $t\to\eta_t^\omega$ is ergodic with respect to $\mu$.
The proof of the contractivity, the symmetry and the strong continuity of $\{\mathbf{Q}_t\}_{t\geq 0}$ on $L^2(\Omega,\Lambda d\mu)$ follows from the stationarity of the environment and by , it is standard and can be found in [@Osada1983], [@zhikov1994homogenization]. The proof ot the ergodicity of the process $t\to\eta_t^\omega$ with respect to $\Lambda d\mu$ can also be found in [@Osada1983] and it is based on the irreducibility of the process $Y^\omega_t$, which was proven in Proposition \[prop:irr\].
\[prop:ergodic\] For all functions $f\in L^p(\Omega,\Lambda d\mu)$, $p\geq 1$, set $f(x;\omega)=f(\tau_x\omega)$, then $$\lim_{t\to\infty}\frac{1}{t}\int_0^t f(Y^\omega_s;\omega)\,ds = {\mathbb{E}}_\mu[f\Lambda],\quad \mathbb{Q}_x^\omega\mbox{-a.s}, \mbox{ a.a. }x\in{\mathbb{R}}^d,$$ for $\mu$-almost all $\omega\in\Omega$.
In order to have the result stated, observe that the measure $Q^{\tau_x\omega}_0$ induced by $\mathbb{Q}_0^{\tau_x\omega}$ through $\eta_t^{\tau_x\omega}$ on the space of $\Omega$-valued trajectories coincides with the measure $Q^{\omega}_x$ induced by $\mathbb{Q}_x^\omega$ through $\eta^{\omega}_t$ . It is then easy to show that for any ball $B\subset{\mathbb{R}}^d$ the two measures $$\int_{\Omega}Q_0^\omega(\cdot) d\mu=\frac{1}{|B|}\int_{B\times \Omega}Q_0^{\tau_x\omega}(\cdot) dx d\mu = \frac{1}{|B|}\int_{\Omega\times B}Q_x^\omega(\cdot) d\mu dx$$ coincide; in the first equality we used the stationarity of the environment. The fact that the limiting relation hold $\int Q_0^\omega(\cdot) d\mu$-almost surely follows immediately from Proposition \[prop:symmetric\], then the result follows.
We use Proposition \[prop:ergodic\] to control the explosion time of the process $\mathbf{M^\omega}=(X_t^\omega,{\mathbb{P}}_x^\omega,\zeta^\omega)$ in terms of the time changed process $\mathbf{M}^{\Lambda,\omega}$. Indeed consider the time change $$\tau_t\operatorname{\mathrel{\mathop:}=}\inf\Bigl\{s>0:\int_{0}^{s}\frac{1}{\Lambda(Y_u^\omega,\omega)}\,du>t\Bigr\},$$ and define the process $\hat{Y}^\omega_t=Y_{\tau_t}^\omega$. We know, by Theorem \[thm:time\_change\] that $\hat{Y}^\omega_t$ is a version of $X^\omega_t$. It is not difficult to see that the explosion time of $\hat{Y}^\omega_t$ equals $\int_{0}^{\infty}\frac{1}{\Lambda(Y_u^\omega,\omega)}\,du$ [@fukushima1994dirichlet see chapter 6]. By Proposition \[prop:ergodic\], $$\lim_{t\to\infty}\frac{1}{t}\int_0^t \frac{1}{\Lambda(Y^\omega_s;\omega)}\,ds = {\mathbb{E}}_\mu[\Lambda^{-1}\Lambda] = 1,\quad\mathbb{Q}_x^\omega\mbox{-a.s}, \mbox{ a.a. }x\in{\mathbb{R}}^d,$$ for $\mu$-almost all $\omega\in\Omega$. It follows that $\hat{Y}^\omega_t$ is conservative for almost all starting points $x\in{\mathbb{R}}^d$, $\mu$-almost surely. This, together with Theorem \[thm:time\_change\] leads to the following result.
Let $\mathbf{M}^\omega=(X_t^\omega, {\mathbb{P}}_x^\omega,\zeta^\omega)$, $x\in{\mathbb{R}}^d$, be the minimal diffusion constructed in section 3.1. Then such a diffusion is conservative.
By Theorem \[thm:time\_change\], $\mathcal{P}_t^\omega 1(x) = \hat{\mathcal{P}}_t^\omega 1(x) = 1$ for almost all $x\in{\mathbb{R}}^d$, and since $\mathbf{M}^\omega$ is our minimal diffusion, then $\mathcal{P}_t^\omega 1(x)=1$ for all $x\in{\mathbb{R}}^d$. We can pass from almost all to all $x\in{\mathbb{R}}^d$ since the minimal diffusion satisfies property (4.2.9) in [@fukushima1994dirichlet], namely $\mathcal{P}_t^\omega(x,dy)$ is absolutely continuous with respect to the Lebesgue measure for each $t>0$ and each $x\in{\mathbb{R}}^d$ (see Theorem 4.5.4 in [@fukushima1994dirichlet]).
From now on we will completely forget about the time changed process. Following the construction in this section it is possible to obtain an environment process for the minimal diffusion $\mathbf{M}^\omega=(X_t^\omega, {\mathbb{P}}_x^\omega)$, namely the process $t\to\tau_{X^\omega_t}\omega=:\psi_t^\omega$, with semigroup $\mathbf{P}_t$, which is precisely given by $$\mathbf{P}_t f(\omega)\operatorname{\mathrel{\mathop:}=}\int_{{\mathbb{R}}^d}f(\tau_y\omega) p^\omega_t(0,y)\,dy.$$
$\{\mathbf{P}_t\}_{t\geq 0}$ defines a symmetric strongly continuous semigroup on $L^2(\Omega,d\mu)$, and $t\to\psi_t^\omega$ is ergodic with respect to $\mu$.
Analogous to Proposition \[prop:symmetric\].
Corrector and Harmonic coordinates {#sec:corr}
==================================
\[sec:corrector\]
Space $L^2(a)$ and Weyl’s decomposition.
----------------------------------------
Fix a stationary and ergodic random medium $(\Omega,\mathcal{G},\mu,\tau_x)$. In this section we rely only on assumption \[ass:a.1\] and ${\mathbb{E}}_\mu[\lambda^{-1}]$, ${\mathbb{E}}_\mu[\Lambda]$ finite.
In order to construct the corrector, we introduce the following space $$L^2(a)\operatorname{\mathrel{\mathop:}=}\bigl\{ V:\Omega\to {\mathbb{R}}^d : {\mathbb{E}}_\mu[\langle a V,V\rangle]<\infty\bigr\}.$$ Such a space is clearly a pre-Hilbert space with the scalar product $$\Theta(U,V)\operatorname{\mathrel{\mathop:}=}{\mathbb{E}}_{\mu}[\langle aU,V\rangle].$$ $L^2(a)$ is isometric to $L^2(\Omega,\mu)^d$ through the map $\Psi: L^2(\Omega,\mu)^d\to L^2(a)$ given by $\Psi(V)=a^{-1/2}V$. In particular $L^2(a)$ is an Hilbert space. Notice that as a consequence of \[ass:a.1\], ${\mathbb{E}}_\mu[\lambda^{-1}],\,{\mathbb{E}}_\mu[\Lambda]<\infty$ and Hölder’s inequality we have that $L^2(a)\subset L^1(\Omega,\mu)$.
The group $\{\tau_x\}_{{\mathbb{R}}^d}$ on $\Omega$ defines a group of strongly continuous unitary operators $\{T_x\}_{{\mathbb{R}}^d}$ on $L^r(\Omega,\mu)$ for any $r>1$, by the position $T_x (V)=V\circ \tau_x$, see [@zhikov1994homogenization Chapter 7]. Therefore, $\{T_x\}_{x\in{\mathbb{R}}^d}$ on $L^2(\Omega,\mu)$ defines the closed operators $D_i$ for $i=1,...,d$, by $$D_i U \operatorname{\mathrel{\mathop:}=}\lim_{h\to 0}\frac{T_{he_i}U-U}{h},$$ where the limit is taken in $L^2(\Omega,\mu)$. Denote by $\mathcal{D}(D_i)$ the domain of $D_i$. We shall consider the following class of smooth functions $$\label{eq:setC}
\mathcal{C}\operatorname{\mathrel{\mathop:}=}\Bigl\{\int_{{\mathbb{R}}^d}f(\tau_x\omega) {\varphi}(x)dx\, |\, f\in L^\infty(\Omega), {\varphi}\in C_0^\infty({\mathbb{R}}^d)\Bigr\}.$$ It can be proved that if $v\in\mathcal{C}$, $$v(\omega)=\int_{{\mathbb{R}}^d}f(\tau_x\omega) {\varphi}(x)dx \Rightarrow D_i v(\omega)=-\int_{{\mathbb{R}}^d}f(\tau_x\omega)\partial_i{\varphi}(x)dx.$$ In particular, $v\in\bigcap_{i=1}^d\mathcal{D}(D_i)$. It is also clear that $\nabla v=(D_1v,\dots,D_dv)\in L^2(a)$ and that $x\to v(\tau_x\omega)\in C^\infty({\mathbb{R}}^d)$ for $\mu$-almost all $\omega\in\Omega$. We define the space of potential $L^2_{pot}$ to be the closure of $\{\nabla v | v\in\mathcal{C}\}$ in $L^2(a)$.
\[lemma:L2pot\] Let $U\in L^2_{pot}$. Then $U$ satisfies the following properties
- ${\mathbb{E}}_\mu[U_i]=0$ for all $i=1,\dots,d$.
- for all $\eta\in C^\infty_0({\mathbb{R}}^d)$ and $i,j=1,\dots,d$ $$\int_{{\mathbb{R}}^d} U_i(\tau_x\omega) \partial_j\eta(x)\,dx=\int_{{\mathbb{R}}^d}
U_j(\tau_x\omega) \partial_i\eta(x)\,dx,$$ for $\mu$-almost all $\omega\in\Omega$.
In both cases the proof follows simply by considering functions of the type $\nabla f$ such that $f\in\mathcal{C}$. Then conclude by density.
Let start with $(i)$. Observe that if $f\in\mathcal{C}$ then $${\mathbb{E}}_\mu[D_i f] = \lim_{h\to 0} {\mathbb{E}}_\mu\Bigl[\frac{T_{he_i} f-f}{h}\Bigr]= \lim_{h\to 0} \frac{{\mathbb{E}}_\mu[T_{he_i} f]-{\mathbb{E}}_\mu[f]}{h}=0.$$ If $U\in L^2_{pot}$, we find $f_n\in \mathcal{C}$ such that $\nabla f_n\to U$ in $L^2(a)$, hence in $L^1(\Omega,\mu)^d$. It follows $${\mathbb{E}}_\mu[U]=\lim_{n\to\infty}{\mathbb{E}}_\mu[\nabla f_n] = 0.$$
We now prove $(ii)$. Consider again $f\in \mathcal{C}$. Then $x\to f(x;\omega)$ is infinitely many times differentiable, $\mu$-almost surely. Integrating by parts we get $$\int_{{\mathbb{R}}^d} D_i f(x;\omega) \partial_j\eta(x)\,dx = -\int_{{\mathbb{R}}^d} f(x;\omega) \partial_i\partial_j\eta(x)\,dx,$$ finally switch the partials and conclude $$\int_{{\mathbb{R}}^d} D_i f(x;\omega) \partial_j\eta(x)\,dx = \int_{{\mathbb{R}}^d} D_j f(x;\omega) \partial_i\eta(x)\,dx.$$ For a general $U\in L^2_{pot}$ take approximations and use the fact that $\nabla f_n\to U$ in $L^2(a)$ implies $D_i f_n(\cdot;\omega)\to U_i(\cdot;\omega)$ in $L^1_{loc}({\mathbb{R}}^d)$ $\mu$-almost surely.
#### Weyl’s decomposition
Since $L^2(a)$ is an Hilbert space and $L^2_{pot}$ is by construction a closed subspace, we can write $$L^2(a) = L^2_{pot}\oplus (L^2_{pot})^\perp.$$ We want to decompose the bounded functions $\{\pi^k\}_{k=1}^d$, where $\pi^k$ is the unit vector in the kth-direction. Since $\pi_k\in L^2(a)$, for each $k=1,\dots,d$, there exist functions $U^k\in L^2_{pot}$ and $R^k\in (L^2_{pot})^\perp$ such that $\pi^k=U^k+R^k$. By definition of orthogonal projection we have $${\mathbb{E}}_\mu[\langle a U^k,V \rangle]={\mathbb{E}}_\mu[\langle a \pi^k,V \rangle],\quad\forall V\in
L^2_{pot}.$$
By definition of $L^2_{pot}$ and orthogonal projection it follows in particular that $${\mathbb{E}}_\mu[\langle a (U^k-\pi_k),U^k-\pi_k \rangle]=\inf_{f\in\mathcal{C}}{\mathbb{E}}_\mu[\langle a (\nabla f-\pi_k),\nabla f-\pi_k \rangle].$$
\[prop:pos\] Set $\mathbf{d}_{ij}\operatorname{\mathrel{\mathop:}=}2{\mathbb{E}}_\mu[\langle a (U^i-\pi_i),U^j-\pi_j \rangle]$. Then the matrix $\{\mathbf{d}_{ij}\}_{i,j}$ is positive definite.
Take any $\xi\in{\mathbb{R}}^d$, then $$\sum_{i,j} \mathbf{d}_{ij} \xi_i \xi_j = 2 {\mathbb{E}}_\mu\Bigl[\langle a \Bigl(\sum_i \xi_i U^i-\xi\Bigr),\sum_j\xi_j U^j-\xi \rangle\Bigr].$$ Since $\sum_i \xi_i U^i\in L^2_{pot}$ is the orthogonal projection of the function $\pi_\xi:\omega\to\xi$, and $\pi_\xi\in L^2(a)$, we have $$\begin{aligned}
\label{eq:cp_1}
\sum_{i,j} \mathbf{d}_{ij} \xi_i \xi_j &=\inf_{{\varphi}\in\mathcal{C}}2{\mathbb{E}}_\mu[\langle a( \nabla{\varphi}-\xi),\nabla {\varphi}-\xi \rangle]\geq \sum_{i=1}^d\inf_{{\varphi}\in\mathcal{C}} 2{\mathbb{E}}_\mu[\lambda| D_i{\varphi}-\xi_i|^2] \notag\\
&=\sum_{i=1}^d|\xi_i|^2\inf_{{\varphi}\in\mathcal{C}}2{\mathbb{E}}_\mu[\lambda|D_i{\varphi}-1|^2]\end{aligned}$$ we end up with a basic one dimensional problem. Observe that by Hölder’s inequality we have $${\mathbb{E}}_\mu[\lambda|D_i{\varphi}-1|^2] \geq {\mathbb{E}}[\lambda^{-1}]^{-1}{\mathbb{E}}_\mu[(D_i{\varphi}-1)]^2 = {\mathbb{E}}[\lambda^{-1}]^{-1}$$ for all ${\varphi}\in C^\infty_b(\Omega)$ since by Lemma \[lemma:L2pot\] we have that ${\mathbb{E}}_\mu[D_i {\varphi}]=0$. Therefore is bounded from below by $\sum_{i=1}^d|\xi_i|^2{\mathbb{E}}_\mu[\lambda^{-1}]^{-1}=|\xi|^2{\mathbb{E}}_\mu[\lambda^{-1}]^{-1}$ and we get the bound $$\sum_{i,j} \mathbf{d}_{ij} \xi_i \xi_j\geq 2 {\mathbb{E}}_\mu[\lambda^{-1}]^{-1} |\xi|^2$$ which is what we wanted to proof.
At this point we build the corrector starting from the functions $U^k\in L^2_{pot}$. For $k=1,...,d$ we define the *corrector* to be the function $\chi^k :{\mathbb{R}}^d\times\Omega\to{\mathbb{R}}$ such that $$\chi^k(x,\omega)\operatorname{\mathrel{\mathop:}=}\sum_{j=1}^d \int_0^1 x_j U_j^k(\tau_{tx}\omega)\,dt.$$ It is not hard to prove that $\chi^k$ is well defined, and taking expectation it follows that ${\mathbb{E}}_\mu[\chi^k(x,\omega)]=0$. The key result about the corrector is listed here below
(Weak differentiability) For $k=1,...,d$ the function $x\to\chi^k(x,\omega)$ is in $L^1_{loc}({\mathbb{R}}^d)$, weakly differentiable $\mu$-almost surely and $\partial_i \chi^k(x,\omega)= U_i^k(\tau_{x}\omega)$.
Let $\eta\in C^\infty_0({\mathbb{R}}^d)$ be any test function and calculate $$\int_{{\mathbb{R}}^d} \chi^k(x,\omega) \partial_i \eta(x)\,dx = \int_{{\mathbb{R}}^d}\sum_{j=1}^d\int_0^1 x_j
U_j^k(\tau_{tx}\omega)\,dt\, \partial_i \eta(x)\,dx.$$ By changing the order of integration and applying the change of variables $y=tx$ we get $$\int_0^1\sum_{j=1}^d\int_{{\mathbb{R}}^d}
U_j^k(\tau_{y}\omega)\frac{y_j}{t^{d+1}}\partial_i\eta\Bigl(\frac{y}{t}\Bigr)\,dx\,dt.$$ Next observe that for $j\neq i$, $$\frac{y_j}{t^{d+1}}\partial_i\eta\Bigl(\frac{y}{t}\Bigr) =
\partial_i\Bigl(\frac{y_j}{t^{d}}\eta\Bigl(\frac{y}{t}\Bigr)\Bigr),$$ which together with property $(ii)$ of Lemma \[lemma:L2pot\] gives. $$\int\chi^k(x,\omega) \partial_i \eta(x)\,dx = \int
U_i^k(\tau_{y}\omega)\int_0^1\sum_{j\neq i}\partial_j\Bigl(\frac{y_j}{t^{d}}\eta\Bigl(\frac{y}{t}\Bigr)\Bigr)+
\frac{y_i}{t^{d+1}}\partial_i\eta\Bigl(\frac{y}{t}\Bigr)dt\,dx.$$ Finally, observe that for $y\neq 0$ $$\int_0^1\sum_{j\neq i}\partial_j\Bigl(\frac{y_j}{t^{d}}\eta\Bigl(\frac{y}{t}\Bigr)\Bigr)+
\frac{y_i}{t^{d+1}}\partial_i\eta\Bigl(\frac{y}{t}\Bigr)dt= -\int_0^1\frac{d}{dt} \Bigl(\eta\Bigl(\frac{y}{t}\Bigr)\frac{1}{t^{d-1}}\Bigr)\,dt = -\eta(y).$$ This ends the proof since it follows that $$\label{eq:weak}
\int_{{\mathbb{R}}^d} \chi^k(x,\omega) \partial_i \eta(x)\,dx =-\int_{{\mathbb{R}}^d} U^k_i(x;\omega) \eta(x) \,dx.$$ One may think that the set of $\omega$ for which holds, depends on $\eta$. Since $C^\infty_0({\mathbb{R}}^d)$ is separable we can remove such ambiguity considering a countable dense subset $\{\eta_n\}_{n\in{\mathbb{N}}}$ of $C^\infty_0({\mathbb{R}}^d)$.
So far we did not need more than the first moment for $\lambda^{-1}$ and $\Lambda$. To get more regularity and exploit the power of Sobolev’s embedding theorems, we shall now assume \[ass:a.2\], namely, for $1/p+1/q<2/d$ we suppose that ${\mathbb{E}}_\mu[\lambda^{-q}]$, ${\mathbb{E}}_\mu[\Lambda^{p}]<\infty$. Such an assumption has the following consequence.
\[prop:integrability\] Assume \[ass:a.1\] and \[ass:a.2\], then the corrector $\chi^k(\cdot,\omega)\in{\mathcal{F}}_{loc}^\omega$ for $\mu$-almost all $\omega\in\Omega$.
By construction, there exists $\{f_n\}_{{\mathbb{N}}}\subset\mathcal{C}$ such that $\nabla f_n\to U^k$ in $L^2(a)$. This implies that for any ball $B\subset{\mathbb{R}}^d$ $$\int_B \langle a(x;\omega)\nabla f_n(x;\omega)-\nabla \chi^k(x,\omega), f_n(x;\omega)-\nabla \chi^k(x,\omega)\rangle\,dx\to 0.$$ Observe that $g_n(x,\omega)=f_n(x;\omega)-f_n(\omega)$ belongs to $C^\infty({\mathbb{R}}^d)$ and satisfies $$g_n(x,\omega)=\sum_{i=1}^d\int_0^1 x_j\partial_j f_n(tx;\omega)\,dt.$$ By means of \[ass:a.2\] it is immediate to prove that $g_n\to \chi^k$ in $W^{1,2q/(q+1)}(B)$ for any ball $B\subset{\mathbb{R}}^d$. We claim that $\eta g_n\to \eta \chi^k$ in $L^2({\mathbb{R}}^d)$ with respect to ${\mathcal{E}^\omega}_1$, for any cut-off $\eta$ and $\mu$-almost surely, which by definition proves $\chi^k(\cdot,\omega)\in{\mathcal{F}}_{loc}^\omega$. Indeed $$\begin{gathered}
\int_{{\mathbb{R}}^d} \langle a \nabla (\eta g_n)-\nabla (\eta\chi^k),\nabla (\eta g_n)-\nabla (\eta\chi^k)\rangle\,dx\leq\\
2\int_{B} \langle a \nabla g_n-\nabla \chi^k,\nabla g_n-\nabla \chi^k\rangle\,dx+2\|\nabla \eta\|_\infty^2\int_{B} \Lambda |g_n-\chi^k|^2\,dx\to 0\end{gathered}$$ where the last integral goes to zero by $g_n\to \chi^k$ in $W^{1,2q/(q+1)}(B)$, and by means of the Sobolev’s embedding theorem $W^{1,2q/(q+1)}(B)\hookrightarrow L^{2p^*}(B)$.
Harmonic coordinates and Poisson equation
-----------------------------------------
Now that we have the corrector we want to construct a weak solution to the Poisson equation $\mathcal{L^\omega} u = 0$ for $\mu$-almost all $\omega$. Consider, for $k=1,...,d$, the *harmonic coordinates* to be the functions $y^k : {\mathbb{R}}^d\times \Omega\to {\mathbb{R}}$ defined by $y^k(x,\omega)\operatorname{\mathrel{\mathop:}=}x_k-\chi^k(x,\omega)$.
We say that a function $u\in{\mathcal{F}}_{loc}$ is ${\mathcal{E}^\omega}$-harmonic if ${\mathcal{E}^\omega}(u,{\varphi})=0$, $\forall {\varphi}\in C_0^\infty({\mathbb{R}}^d)$. The next proposition justifies the name harmonic coordinates.
\[prop:solution\] For $k=1,..,d$, the harmonic coordinates $x\to y ^k(x,\omega)$ are ${\mathcal{E}^\omega}$-harmonic $\mu$-almost surely.
We have to prove that $\mu$-almost surely, for all ${\varphi}\in C_0^\infty({\mathbb{R}}^d)$ $${\mathcal{E}^\omega}(y^k,{\varphi})= \sum_{i,j}\int_{{\mathbb{R}}^d} a_{ij}(x;\omega)\partial_iy ^k(x,\omega)\partial_j{\varphi}(x)\,dx = 0.$$ By construction of the corrector, the stationarity of the environment and the fact that $T_x \mathcal{C}= \mathcal{C}$, we have that $$\sum_{i,j}{\mathbb{E}}_\mu[a_{ij}(x;\omega)\partial_iy ^k(x,\omega)D_j f(\omega)] = 0,\quad \forall x\in{\mathbb{R}}^d,\forall f\in\mathcal{C}.$$ Now fix ${\varphi}\in C_0^\infty({\mathbb{R}}^d)$ and integrate against it, we get that for all $f\in \mathcal{C}$ $$\begin{aligned}
0&=\sum_{i,j}\int_{{\mathbb{R}}^d}{\varphi}(x){\mathbb{E}}_\mu[a_{ij}(x;\omega)\partial_iy ^k(x,\omega)D_j f(\omega)]\,dx\\
&=\sum_{i,j}{\mathbb{E}}_\mu\Bigl[a_{ij}(0;\omega)\partial_iy ^k(0,\omega)\int_{{\mathbb{R}}^d}D_j f(\tau_{-x}\omega){\varphi}(x)\,dx\Bigr]\\
&={\mathbb{E}}_\mu\Bigl[f(\omega)\sum_{i,j}\int_{{\mathbb{R}}^d}a_{ij}(x;\omega)\partial_iy ^k(x,\omega)\partial_j{\varphi}(x)\,dx\Bigr].\end{aligned}$$ Since $\mathcal{C}\subset L^{p}(\Omega,\mu)$ for all $p\geq 1$ densely, it follows that $$\label{eq:poisson}
\sum_{i,j}\int_{{\mathbb{R}}^d}a_{ij}(x;\omega)\partial_iy ^k(x,\omega)\partial_j{\varphi}(x)\,dx = 0,\quad\mu\mbox{-a.s.}$$ this ends the proof. To be precise, one should observe that $C_0^\infty({\mathbb{R}}^d)$ is separable, which ensures that is satisfied for all ${\varphi}\in C_0^\infty({\mathbb{R}}^d)$, $\mu$-almost surely.
Observe that neither \[ass:a.2\] nor \[ass:a.3\] is used in the construction of the harmonic coordinates.
\[rem:bound\] If we define $y_{\varepsilon}^k(x,\omega)\operatorname{\mathrel{\mathop:}=}{\varepsilon}y^k(x/{\varepsilon},\omega)$, then an application of the ergodic theorem yields $$\label{eq:bound}
\lim_{{\varepsilon}\to 0}\int_{B_R}\langle a(x/{\varepsilon};\omega)\nabla_x y_{\varepsilon}^k(x;\omega),\nabla_x y_{\varepsilon}^k(x;\omega) \rangle\,dx = {\mathbb{E}}_\mu[\langle a(\pi_k-U^k),\pi_k-U^k\rangle]|B_R|<\infty.$$ which in view of \[ass:a.2\] and the Sobolev’s embedding theorem implies that $$\limsup_{{\varepsilon}\to 0}\|1_{B_R} y^k_{\varepsilon}\|_{\rho}<\infty,$$ where both limits hold $\mu$-almost surely.
Martingales and Harmonic coordinates
------------------------------------
We will assume as usual \[ass:a.1\], \[ass:a.2\] and \[ass:a.3\].
In a situation where $L^\omega = \nabla\cdot( a^\omega \nabla\,)$ is well defined and associated to the process $X^\omega_t$, the fact that $L^\omega y(x,\omega) = 0$, would imply that $y(X_t^\omega,\omega)$ is a martingale by Itô’s formula. In our case we lack the regularity to use the theory coming from stochastic differential equations and we must rely on Dirichlet Forms technique. We know that $y^k(x,\omega)$ is ${\mathcal{E}}^\omega$-harmonic, which in a weaker sense, is analogous to say that $y^k$ is $L^\omega$-harmonic.
We will use the following theorem due to Fukushima, [@Fukushima1987]\[ Theorem 3.1\].
\[thm:martingale\] Fix a point $x_0$ and assume the following conditions for a process $\mathbf{N}=(Z_t,{\mathbb{P}}_x)$ associated to $({\mathcal{E}},{\mathcal{F}})$ on $L^2({\mathbb{R}}^d,dx)$, and for a function $u:{\mathbb{R}}^d\to{\mathbb{R}}$.
- For all $t>0$ the transition semigroup $\mathcal{P}_t$ of $\mathbf{N}$ satisfies $\mathcal{P}_t\mathds{1}_A(x_0)=0$ whenever $Cap(A)=0$.
- $u\in \mathcal{F}_{loc}$, u is continuous and ${\mathcal{E}}$-harmonic.
- Let $\nu_{\langle u \rangle}$ be the energy measure of $u$, namely the only measure such that $$\int_{{\mathbb{R}}^d} v(x)\,d\nu_{\langle u \rangle}(dx)=2{\mathcal{E}}(uv,v)-{\mathcal{E}}(u^2,v),\quad v\in C^\infty_0({\mathbb{R}}^d).$$ We assume that $\nu_{\langle u \rangle}$ is absolutely continuous with respect to the Lebesgue measure $\nu_{\langle u \rangle} = fdx$ and that the density function $f$ satisfies $${\mathbb{E}}_{x_0}\biggl[\int_0^t f(Z_s)\,ds\biggr]<\infty,\quad t>0.$$
Then $M_t=u(Z_t)-u(Z_0)$ is a ${\mathbb{P}}_{x_0}$-square integrable martingale with $$\langle M \rangle_t = \int_0^t f(Z_s)\,ds,\quad t>0,\quad {\mathbb{P}}_{x_0}\mbox{-a.s.}$$
We want to apply Theorem \[thm:martingale\] to the function $u(x,\omega)=\sum_k \lambda_k y^k(x,\omega)$, being an ${\mathcal{E}}^\omega$-harmonic function, and to the minimal process $\mathbf{M}^\omega = (X^\omega_t,{\mathbb{P}}^\omega_x)$, $x\in{\mathbb{R}}^d$. We fix the starting point to be $x_0=0$. Some attention is required to check that every assumption of Theorem \[thm:martingale\] is satisfied for $\mu$-almost all $\omega\in\Omega$.
By construction, since $\mathbf{M}^\omega = (X^\omega_t,{\mathbb{P}}^\omega_x)$, $x\in{\mathbb{R}}^d$ is the minimal diffusion for almost all $\omega\in\Omega$, it follows that $\mathcal{P}_t 1_A(0)=\int_{A} p_t^\omega(0,y)\,dy=0$ whenever $Cap(A)=0$, so that $(i)$ is satisfied. Indeed $Cap(A)=0$ implies that the Lebesgue measure of $A$ is zero [@fukushima1994dirichlet Page 68].
Assumption $(ii)$ is satisfied for almost all $\omega$ in view of Proposition \[prop:solution\], Proposition \[prop:integrability\] and \[ass:a.3\] which assures the continuity of $x\to y^k(x,\omega)$ for $\mu$-almost all $\omega\in\Omega$ by classical results in elliptic partial differential equations with locally uniformly elliptic coefficients [@gilbarg2001elliptic Gilbarg and Trudinger].
In order to check assumption $(iii)$ we have first to understand $\nu_{\langle u\rangle}$. According to [@fukushima1994dirichlet Theorem 3.2.2] and using the fact that $y^k$ are weakly differentiable, the density $f(x,\omega)$ of $\nu_{\langle u\rangle}$ with respect to the Lebesgue measure is given by $$f(x,\omega)=2\sum_{i,j} \partial_i u(x;\omega)\partial_j u(x;\omega)\, a_{ij}(x;\omega) = 2\sum_{k,h} \lambda_k \lambda_h \Bigl(\sum_{i,j} \partial_i y^k(x;\omega)\partial_j y^h(x;\omega)\, a_{ij}(x;\omega)\Bigr)$$ which we can rewrite as $f(x,\omega)= 2\langle q(x,\omega)\lambda,\lambda\rangle$, with $$q^{hk}(\omega)\operatorname{\mathrel{\mathop:}=}\sum_{i,j} \partial_i y^k(0;\omega)\partial_j y^h(0;\omega)\, a_{ij}(\omega)=\sum_{i,j} (U_i^k(\omega)-\delta_{ik})(U_j^h(\omega)-\delta_{jh})\, a_{ij}(\omega).$$ Next we compute, using the stationarity of the environment process $$\int_{\Omega } {\mathbb{E}}_0^\omega\biggl[\int_0^t f(X^\omega_s;\omega)\,ds\biggr]\,d\mu =2\int_{\Omega } {\mathbb{E}}_0^\omega\biggl[\int_0^t \langle q(\psi^\omega_s)\lambda,\lambda\rangle\,ds\biggr]\,d\mu=2t\int_{\Omega}\langle q(\omega)\lambda,\lambda\rangle\,d\mu,$$ which is finite by construction, since $U\in L^2(a)$. In particular $(iii)$ is satisfied. It follows the following theorem:
\[thm:harmoniccoho\] Assume \[ass:a.1\],\[ass:a.2\] and \[ass:a.3\]. Then $y(X_t^\omega,\omega)$ is a ${\mathbb{P}}^\omega_0$-square integrable martingale with covariation given by $$\langle y^k(X_t^\omega,\omega), y^h(X_t^\omega,\omega)\rangle_t =2 \int_0^t \sum_{i,j}a_{ij}(X^\omega_s,\omega) (\partial_i\chi^k(X^\omega_s,\omega)-\delta_{ik})(\partial_j\chi^h(X^\omega_s,\omega)-\delta_{jh})\, \,ds,$$ for $\mu$-almost all $\omega\in\Omega$.
Above.
Proof of the Invariance Principle {#sec:proof}
=================================
In Section \[sec:corrector\] we constructed the function $\chi,y:{\mathbb{R}}^d\times\Omega\to{\mathbb{R}}^d$ in a way that we can decompose the process $X^\omega$ as $$X_t^\omega = y(X_t^\omega,\omega)+\chi(X_t^\omega,\omega),$$ in particular, we proved in Theorem \[thm:harmoniccoho\] that $y(X_t^\omega,\omega)$ is a martingale. In order to get a quenched invariance principle for the process $X_t^{{\varepsilon},\omega} = {\varepsilon}X_{t/{\varepsilon}^2}^\omega$ we will need to prove that ${\varepsilon}\chi(X_t^{{\varepsilon},\omega}/{\varepsilon},\omega)$ is converging to zero in law and that the quadratic variation of the martingale is converging to a constant.
As first result on the decay of the corrector as ${\varepsilon}\to 0$ we have the following Lemma.
\[lem:control\] For all $R>0$ and for $\mu$-almost all $\omega\in\Omega$ $$\lim_{{\varepsilon}\to0}\|y^k_{\varepsilon}(x;\omega)-x_k\|_{2 p^*,B_R}=\lim_{{\varepsilon}\to0}\|\chi^k_{\varepsilon}(x;\omega)\|_{2 p^*,B_R}=0.$$
It is enough to show that for any $\eta\in C_0^\infty(B_R)$ we have $$\lim_{{\varepsilon}\to 0} \int_{{\mathbb{R}}^d} y^k_{\varepsilon}(x;\omega)\eta(x)\,dx=\int_{{\mathbb{R}}^d} x_k \eta(x)\,dx.$$ Indeed the above property implies the weak convergence $y_{\varepsilon}^k\rightharpoonup
x_k$ in $L^2(B_R)$. This gives the strong convergence in $L^{2p^*}(B_R)$, because $W^{1,2q/(q+1)}(B_R)$ is compactly embedded in $L^{2p^*}(B_R)$ and the sequence $\{y_{\varepsilon}\}_{{\varepsilon}>0}$ is bounded in $W^{1,2q/(q+1)}(B_R)$ by .
Since $\partial_j y^k(x;\omega) =\delta_{jk}- U_j^k(\tau_x\omega)$ and ${\mathbb{E}}_\mu[U_j^k]=0$, the ergodic theorem implies that for each $\delta>0$ arbitrary, $\mu$-almost surely, there exists ${\varepsilon}(\omega)>0$ such that for all ${\varepsilon},s >0$ with $s > {\varepsilon}/{\varepsilon}(\omega)$ $$\label{eq:c1}
\biggl|\sum_j\int_{B_R} \partial_j y_{\varepsilon}^k(sx;\omega) x_j \eta(x)\,dx-\int_{{\mathbb{R}}^d} x_k \eta(x)\,dx\biggr|\leq \delta.$$ Notice that $$\begin{aligned}
\label{eq:1}
\int_{{\mathbb{R}}^d} y^k_{\varepsilon}(x;\omega)\eta(x)\,dx &= \sum_j\int_{B_R}\int_0^1 \partial_j y_{\varepsilon}^k(tx;\omega) x_j \eta(x)\, dt \,dx \notag\\
&=\sum_j\int_0^1\int_{B_R} \partial_j y_{\varepsilon}^k(tx;\omega) x_j \eta(x)\,dx\, dt.
\end{aligned}$$ We split the integral in as the sum $$\sum_j\int_0^{{\varepsilon}/{\varepsilon}(\omega)}\int_{B_R} \partial_j y_{\varepsilon}^k(tx) x_j \eta(x)\,dx\, dt + \sum_j\int_{{\varepsilon}/{\varepsilon}(\omega)}^1\int_{B_R} \partial_j y_{\varepsilon}^k(tx) x_j \eta(x)\,dx\, dt,$$ now we estimate each of the two terms. We can rewrite the second term as $$(1-{\varepsilon}/{\varepsilon}(\omega))\int_{B_R} x_j \eta(x)\,dx+\int_{{\varepsilon}/{\varepsilon}(\omega)}^1 r_{{\varepsilon}/t}\,dt,$$ where the second integral is bounded by $\delta$, in view of . For what concerns the first part, we can easily compute $$\begin{aligned}
\sum_{j}\int_{0}^{{\varepsilon}/{\varepsilon}(\omega)}\int_{B_R} \partial_j y_{\varepsilon}^k(tx) x_j \eta(x)\,dx = {\varepsilon}/{\varepsilon}(\omega) \int_{B_R} {\varepsilon}(\omega)y^k(x/{\varepsilon}(\omega)) \eta(x)\,dx.
\end{aligned}$$ Hence the first part is bounded by $c\cdot ({\varepsilon}/{\varepsilon}(\omega))$ for a constant $c>0$. Finally this yields $$\limsup_{{\varepsilon}\to 0}\biggl|\int_{{\mathbb{R}}^d} y^k_{\varepsilon}(x;\omega)\eta(x)\,dx-\int_{{\mathbb{R}}^d} x_k \eta(x)\,dx\biggr|\leq \delta$$ with $\delta$ arbitrarily chosen.
\[prop:sublinearity\] For all $R>0$, $$\label{eq:sublin}
\lim_{{\varepsilon}\to 0} \sup_{|x|\leq R} {\varepsilon}|\chi(x/{\varepsilon},\omega)| = 0,\quad \mu\mbox{-almost surely.}$$
Observe that $\chi_{\varepsilon}^k(x,\omega)\operatorname{\mathrel{\mathop:}=}{\varepsilon}\chi(x/{\varepsilon},\omega)$ is a solution on $B=B(R)$ for all ${\varepsilon}>0$ of $$\sum_{i,j}\int_{B} a_{ij}^\omega(x/{\varepsilon})\partial_i\chi_{\varepsilon}^k(x;\omega)\partial_j {\varphi}(x)\,dx=\sum_{i,j}\int_B a_{ij}^\omega(x/{\varepsilon})\partial_i f_k(x)\partial_j {\varphi}(x)\,dx,$$ where $f_k(x) = x_k$ and ${\varphi}\in C^\infty_0(B)$. Clearly $|\nabla f_k(x)|\leq 1$ for all $x\in {\mathbb{R}}^d$ and ${\varepsilon}>0$. By Lemma \[lem:control\], we get that $$\lim_{{\varepsilon}\to0}\|\chi^k_{\varepsilon}(x;\omega)\|_{2p^*,B_R}=0$$ Therefore, we can obtain \[eq:sublin\] applying with $\alpha = 2 p^*$. $$\|\chi_{\varepsilon}^k\|_{B(R),\infty}\leq \\ C_3
\biggl(1\vee\|(\lambda^\omega)^{-1}\|_{B(2R/{\varepsilon}),q}\|\Lambda^\omega\|_{ B(2R/{\varepsilon}),p} \biggr)^{\kappa'} \|\chi_{\varepsilon}^k\|_{B(2 R),2p^*}^{\gamma'}\vee\|\chi_{\varepsilon}^k\|_{B(2 R),2p^*}$$ which goes to zero as ${\varepsilon}\to 0$ by Lemma \[lem:control\]. Notice that we can bound $\|\lambda^{-1}\|_{B(2R/{\varepsilon}),q}\|\Lambda\|_{ B(2R/{\varepsilon}),p}$ by a constant, by means of \[ass:a.2\] and the ergodic theorem.
We can now turn to the proof of Theorem \[thm:invprinc\], namely the quenched invariance principle for the diffusions ${\varepsilon}X^\omega_{t/{\varepsilon}^2}$.
With the help of Proposition \[prop:sublinearity\] the proof of this theorem is identical to [@fannjiang1997 Theorem 1], with only a minor difference, namely, the limiting matrix $\mathbf{D}=[\mathbf{d}_{ij}]$ is given by $$\mathbf{d}_{ij} = 2{\mathbb{E}}_\mu[\langle a(\omega) \nabla y^i(0,\omega),\nabla y^j(0,\omega) \rangle]$$ being $y^i(x,\omega)$ the harmonic coordinates as constructed in Section \[sec:corrector\].
For completeness we put her the proof of part (ii) of the theorem and we refer to [@fannjiang1997] for the first part. We make use of the decomposition $${\varepsilon}X^\omega_{t/{\varepsilon}^2} = {\varepsilon}y( X^\omega_{t/{\varepsilon}^2},\omega) + {\varepsilon}\chi( X^\omega_{t/{\varepsilon}^2},\omega).$$ and the fact that $M^{{\varepsilon},\omega} = {\varepsilon}y( X^\omega_{t/{\varepsilon}^2},\omega)$ is a ${\mathbb{P}}_0^\omega$-square integrable continuous martingale $\mu$-almost surely by Theorem \[thm:martingale\]. Its quadratic variation is given by $$\langle M_h^{{\varepsilon},\omega}, M_k^{{\varepsilon},\omega}\rangle_t = {\varepsilon}\int_0^{t/{\varepsilon}^2} 2\sum_{i,j}a_{ij}(X^\omega_s,\omega) (\partial_i\chi^k(X^\omega_s,\omega)-\delta_{ik})(\partial_j\chi^h(X^\omega_s,\omega)-\delta_{jh})\, \,ds.$$ An application of the ergodic theorem for the environmental process shows that $$\lim_{{\varepsilon}\to 0}\langle M_h^{{\varepsilon},\omega}, M_k^{{\varepsilon},\omega}\rangle_t = \mathbf{d}_{hk}t,$$ ${\mathbb{P}}^\omega_0$-almost surely, but also in the $L^1$ sense for almost all $\omega\in\Omega$. We can now apply the central limit for martingales [@helland1982 Theorem 5.4] to conclude that the martingale $M^{{\varepsilon},\omega}$ converges in distribution over $C([0,\infty),{\mathbb{R}}^d)$ under ${\mathbb{P}}_0^\omega$ to a Wiener measure with covariances given by $\mathbf{D}$. The matrix is non degenerate by Proposition \[prop:pos\].
It remains to show that the correctors ${\varepsilon}\chi(X^{\omega}_{t/{\varepsilon}^2},\omega)$ converge to zero in distribution. For that the sublinearity of the corrector will play a major role.
Let $T>0$ be a fixed time horizon. We claim that for all $\delta>0$ $$\label{eq:claim}
\lim_{{\varepsilon}\to 0} {\mathbb{P}}_0^\omega\Bigl(\sup_{0\leq t \leq T} |{\varepsilon}\chi(X^{\omega}_{t/{\varepsilon}^2},\omega)| >\delta \Bigr) = 0.$$ Denote by $\tau_R^{{\varepsilon},\omega}$ the exit time of ${\varepsilon}X^{\omega}_{t/{\varepsilon}^2}$ from the ball $B$ of radius $R>1$ centered at the origin. Observe that $$\begin{aligned}
\limsup_{{\varepsilon}\to 0} {\mathbb{P}}_0^\omega&\Bigl(\sup_{0\leq t \leq T} |{\varepsilon}\chi(X^{\omega}_{t/{\varepsilon}^2},\omega)| >\delta \Bigr)\\
& \leq \limsup_{{\varepsilon}\to 0} {\mathbb{P}}_0^\omega\Bigl(\sup_{0\leq t \leq \tau_R^{{\varepsilon},\omega}} |{\varepsilon}\chi(X^{\omega}_{t/{\varepsilon}^2},\omega)| >\delta \Bigr)+\limsup_{{\varepsilon}\to 0} {\mathbb{P}}_0^\omega\Bigl(\sup_{0\leq t \leq T} |{\varepsilon}X^{\omega}_{t/{\varepsilon}^2}| > R \Bigr).\end{aligned}$$ *First addendum*: By Proposition \[prop:sublinearity\] $$\lim_{{\varepsilon}\to 0} \sup_{0\leq t \leq \tau_R^{{\varepsilon},\omega}} |{\varepsilon}\chi(X^{\omega}_{t/{\varepsilon}^2},\omega)| = 0.$$ and therefore $\mu$-almost surely $$\limsup_{{\varepsilon}\to 0} {\mathbb{P}}_0^\omega\Bigl(\sup_{0\leq t \leq \tau_R^{{\varepsilon},\omega}} |{\varepsilon}\chi(X^{\omega}_{t/{\varepsilon}^2},\omega)| >\delta \Bigr)=0.$$ *Second addendum*: we use again Proposition \[prop:sublinearity\] to say that there exists $\bar{{\varepsilon}}(\omega) >0$, which may depend on $\omega$ such that for all ${\varepsilon}<\bar{{\varepsilon}}(\omega)$ we have $\sup_{0\leq t \leq \tau_R^{{\varepsilon},\omega}} |{\varepsilon}\chi(X^{\omega}_{t/{\varepsilon}^2},\omega)|<1$. For such ${\varepsilon}$ we have $\mu$-almost surely $$\begin{aligned}
{\mathbb{P}}_0^\omega\Bigl(\sup_{0\leq t \leq T} |{\varepsilon}X^{\omega}_{t/{\varepsilon}^2}|\geq R\Bigr) &= {\mathbb{P}}_0^\omega\Bigl(\tau_R^{{\varepsilon},\omega}\leq T\Bigr)\\
&={\mathbb{P}}_0^\omega\Bigl(\tau_R^{{\varepsilon},\omega}\leq T,\sup_{0\leq t \leq \tau_R^{{\varepsilon},\omega}} |{\varepsilon}y(X^{\omega}_{t/{\varepsilon}^2},\omega)|>R-1\Bigr)\\
&\leq{\mathbb{P}}_0^\omega\Bigl(\sup_{0\leq t \leq T} |{\varepsilon}y(X^{\omega}_{t/{\varepsilon}^2},\omega)|>R-1\Bigr)\end{aligned}$$ Since ${\varepsilon}y(X^{\omega}_{\cdot/{\varepsilon}^2},\omega)$ converges in distribution under ${\mathbb{P}}_0^\omega$ to a non-degenerate Brownian motion with deterministic covariance matrix given by $\mathbf{D}$ we have that there exists positive constants $c_1,c_2$ independent on ${\varepsilon}$ and $\omega$ such that $$\limsup_{{\varepsilon}\to 0} {\mathbb{P}}_0^\omega\Bigl(\sup_{0\leq t \leq T} |{\varepsilon}y(X^{\omega}_{t/{\varepsilon}^2},\omega)|>R-1\Bigr)\leq c_1 e^{-c_2 R},$$ from which it follows $$\limsup_{{\varepsilon}\to 0} {\mathbb{P}}_0^\omega\Bigl(\sup_{0\leq t \leq T} |{\varepsilon}X^{\omega}_{t/{\varepsilon}^2}|>r\Bigr)\leq c_1 e^{-c_2 R}.$$ Therefore $$\limsup_{{\varepsilon}\to 0} {\mathbb{P}}_0^\omega\Bigl(\sup_{0\leq t \leq T} |{\varepsilon}\chi(X^{\omega}_{t/{\varepsilon}^2},\omega)| >\delta \Bigr) \leq c_1 e^{-c_2 R}$$ and since $R>1$ was arbitrary, the claim follows, namely the corrector converges to zero in law under ${\mathbb{P}}_0^\omega$, $\mu$-almost surely.
The convergence to zero in law of the correctors ${\varepsilon}\chi(X_{\cdot/{\varepsilon}^2},\omega)$, combined with the fact that ${\varepsilon}y(X_{\cdot/{\varepsilon}^2},\omega)$ satisfies an invariance principle $\mu$-almost surely and that ${\varepsilon}X^\omega_{\cdot/{\varepsilon}^2} ={\varepsilon}\chi(X_{\cdot/{\varepsilon}^2},\omega)+ {\varepsilon}y(X_{\cdot/{\varepsilon}^2},\omega)$, implies that also the family ${\varepsilon}X^\omega_{\cdot/{\varepsilon}^2}$ under ${\mathbb{P}}_0^\omega$ over $C([0,\infty),{\mathbb{R}}^d)$ satisfies an invariance principle $\mu$-almost surely with the same limiting law.
Let $\theta:\Omega\to{\mathbb{R}}$ be a $\mathcal{G}$-measurable function and assume that $\theta(\tau_.\omega)$, $\theta(\tau_.\omega)^{-1}\in L^\infty_{loc}({\mathbb{R}}^d)$ for $\mu$-almost all $\omega\in\Omega$ and that ${\mathbb{E}}_\mu[\theta],{\mathbb{E}}_\mu[\theta^{-1}]<\infty$. Let $\mathbf{M}^{\theta,\omega}\operatorname{\mathrel{\mathop:}=}(X_t^{\theta,\omega},{\mathbb{P}}^{\theta,\omega}_x)$, $x\in{\mathbb{R}}^d$ the minimal diffusion process associated to $({\mathcal{E}}^\omega,{\mathcal{F}}^{\theta,\omega})$ on $L^2({\mathbb{R}}^d,\theta dx)$. Then, for $\mu$-almost all $\omega\in\Omega$, the laws of the processes ${\varepsilon}X^{\theta,\omega}_{t/{\varepsilon}^2}$ over $C([0,\infty),
{\mathbb{R}}^d)$ converge weakly as ${\varepsilon}\to 0$ to a Wiener measure with covariance matrix given by $\mathbf{D}/{\mathbb{E}}_\mu[\theta]$, where $\mathbf{D}$ was given in Theorem \[thm:invprinc\].
Let us define the time change $$\hat{X}^\omega_t\operatorname{\mathrel{\mathop:}=}X_{\tau^\omega_t}^\omega,\quad\tau^\omega_t=\inf\{s>0;\,A_s^\omega\operatorname{\mathrel{\mathop:}=}\int_0^s\theta(X^\omega_u,\omega)\,du>t\}$$ To get asymptotic for ${\varepsilon}^2 A_{t/{\varepsilon}^2}$ it is easy by means of the ergodic theorem for the environmental process. We can prove as in [@BaMathieu Lemma 15] that $$\label{eq:timechange}
\lim_{{\varepsilon}\to 0}\sup_{s\in[0,t]}|{\varepsilon}^2 A^\omega_{s/{\varepsilon}^2}-s {\mathbb{E}}_\mu[\theta]|=0,\quad {\mathbb{P}}_x^\omega\mbox{-a.s}, \mbox{ a.a. }x\in{\mathbb{R}}^d,$$ for $\mu$-almost all $\omega\in\Omega$. Observe that ${\varepsilon}\hat{X}^\omega_{A^\omega(t/{\varepsilon}^2)} = {\varepsilon}X_{t/{\varepsilon}^2}^\omega$, then the convergence for ${\varepsilon}\hat{X}^\omega_{t/{\varepsilon}^2}$ ${\mathbb{P}}_x^\omega$-a.s, for almost all $x\in{\mathbb{R}}^d$, for $\mu$-almost all $\omega\in\Omega$ follows from Theorem 1.1 and . On the other hand the processes $\hat{X}_t^\omega$ and $X_t^{\theta,\omega}$ are equivalent, since they possess the same Dirichlet form, see Theorem 6.2.1 in [@fukushima1994dirichlet]. Hence the same convergence holds for ${\varepsilon}X_{t/{\varepsilon}^2}^{\theta,\omega}$.
References {#references .unnumbered}
==========
[10]{}
S. Andres, J. D. Deuschel, and M. Slowik. Harnack inequalities on weighted graphs and some applications to the random conductance model. . [**0178-8051**]{} (2015) 1–47.
S. Andres, J. D. Deuschel, and M. Slowik. Invariance principle for the random conductance model in a degenerate ergodic environment. (2014). To Appear.
M. Ba and P. Mathieu. A [S]{}obolev inequality and the individual invariance principle for diffusions in a periodic potential. Submitted (2013). Available at arXiv:1312.4817.
M. T. Barlow, R. F. Bass, Z. Chen, and M. Kassmann. Non-local dirichlet forms and symmetric jump processes. (2009) 1963–1999.
A. C. Cavalheiro. Weighted Sobolev spaces and degenerate elliptic equations. (2008) 117–132.
A. Chiarini and J. D. Deuschel. Local central limit theorem for diffusions in a degenerate and unbounded random medium. Submitted (2015). Available at arXiv:1501.03476v1.
R. De Arcangelis and F. Serra Cassano. On the homogenization of degenerate elliptic equations in divergence form. (1992) 119–138.
E. [De Giorgi]{}. (1957) 25–43.
F. Delarue and R. Rhodes. Stochastic homogenization of quasilinear PDEs with a spatial degeneracy. (2009) 61 – 90.
D. E. Edmunds and L. A. Peletier. A Harnack inequality for weak solutions of degenerate quasilinear elliptic equations. (1972) 21–31.
J. Engstr[ö]{}m, L. E. Persson, A. Piatnitski, and P. Wall. Homogenization of random degenerated nonlinear monotone operators. , [**41**]{} (2006) 101–114.
L. C. Evans. . , 2010.
E. B. Fabes, C. E. Kenig, and R. P. Serapioni. The local regularity of solutions of degenerate elliptic equations. (1982) 77–116.
A. Fannjiang and T. Komorowski. A martingale approach to homogenization of unbounded random flows. , [**25**]{} (1997) 1872–1894.
M. Fukushima, S. Nakao, and M. Takeda. On dirichlet forms with random data – recurrence and homogenization. In [*Stochastic Processes – Mathematics and Physics II*]{}, [**1250**]{}. [*Lecture Notes in Mathematics*]{}. Springer Berlin, Heidelberg, 1987.
M. Fukushima, Y. Oshima, and M. Takeda. . De Gruyter studies in mathematics. W. de Gruyter, 1994.
D. Gilbarg and N.S. Trudinger. . Classics in Mathematics. U.S. Government Printing Office, 2001.
A. Grigor’yan and A. Telcs. Two-sided estimates of heat kernels on metric measure spaces. , [**40**]{} (2012) 1212–1284.
M. Hairer and E. Pardoux. Homogenization of periodic linear degenerate PDEs. (2008) 2462–2487.
J. Heinonen, T. Kilpelainen, and O. Martio. . Dover Books on Mathematics Series. Dover Publications, 2006.
I. S. Helland. Central limit theorems for martingales with discrete or continuous time. (1982) 79–94.
K. Ichihara. Some global properties of symmetric diffusion processes. (1978) 441–486, .
S. M. Kozlov. The method of averaging and walks in inhomogeneous environments. (1985) 73-145.
H. Kunita. General boundary conditions for multi-dimensional diffusion processes. (1970) 273–335.
J. Moser. On Harnack’s theorem for elliptic differential equations. (1961) 577–591.
B. Muckenhoupt. Weighted norm inequalities for the Hardy maximal function. (1972) 207–226.
J. Nash. Continuity of solutions of parabolic and elliptic equations. , [**80**]{} (1958) 931–954.
H. Osada. Homogenization of diffusion processes with random stationary coefficients. (1983) 507–517.
G. C. Papanicolaou and S. R. S Varadhan. Boundary value problems with rapidly oscillating random coefficients. , (1979) 835–873.
E. Pardoux and A. B. Sow. Homogenization of a periodic semilinear elliptic degenerate PDE. (2011) 475–493.
M. R[ö]{}ckner. General theory of dirichlet forms and applications. , [**1563**]{} 129–193. [*Lecture Notes in Mathematics*]{}. Springer Berlin Heidelberg, 1993.
L. Saloff-Coste. . London Mathematical Society Lecture Note Series. Cambridge University Press, 2002.
A. B. Sow, R. Rhodes, and E. Pardoux. Homogenization of periodic semilinear parabolic degenerate PDEs. (2009) 979–998.
M. Tomisaki. Dirichlet forms associated with direct product diffusion processes. In [*Functional Analysis in Markov Processes*]{} [**923**]{} 76-119. [ *Lecture Notes in Mathematics*]{}. Springer Berlin Heidelberg, 1982.
A. Torchinsky. . Dover Books on Mathematics Series. Dover Publications, Incorporated, 2012.
V. V. Zhikov. Weighted sobolev spaces. (1998) 1139-1170.
V.V. Zhikov. Estimates of the [N]{}ash – [A]{}ronson type for degenerating parabolic equations. (2013) 66–79.
V.V. Zhikov, S.M. Kozlov, and O.A. Ole[ĭ]{}nik. . Springer-Verlag, 1994.
[^1]: The first author is supported by RTG 1845.
[^2]: A set $\mathcal{N}\subset {\mathbb{R}}^d$ is called properly exceptional if $\mathcal{N}$ is Borel, it has Lebesgue measure zero, and ${\mathbb{P}}_x(X_t\in\mathcal{N}\mbox{ or }X_{t-}\in\mathcal{N} \mbox{ for some }t\geq0)=0$ for all $x\in {\mathbb{R}}^d\setminus \mathcal{N}$.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We study the transversal magnetoconductivity and magnetoresistance of a massive Dirac fermion gas. This can be used as a simple model for gapped Dirac materials. In the zero mass limit the case of gapless Dirac semimetals is also studied. In the case of Weyl semimetals, to reproduce the non-saturating linear magnetoresistance seen in experiments, the use of screened charged impurities is inevitable. In this paper these are included using the first Born approximation for the self-energy. The screening wavenumber is calculated using the random phase approximation with the polarization function taking into account the electron-electron interaction. The Hall conductivity is calculated analytically in the case of no impurities and is shown to be perfectly inversely proportional to the magnetic field. Thus, the magnetic field dependence of the magnetoresistance is mainly determined by $\sigma_{xx}$. We show that in the extreme quantum limit at very high magnetic fields the gapped Dirac materials are expected to have $\sigma_{xx}\propto B^{-3}$ leading to $\varrho_{xx}\propto B^{-1}$, in contrast with the gapless case where $\sigma_{xx}\propto B^{-1}$ and $\varrho_{xx}\propto B$. At lower fields we find that the effect of the mass term is negligible and in the region of the Shubnikov-de Haas oscillations the two systems behave almost identically. We suggest a phenomenological scattering rate that is able to reproduce the linear behavior at the oscillating region. We show that in the case of the scattering rate calculated using the Born approximation the strength of the relative permittivity and the density of impurities affects the magnetic field dependence of the conductivity significantly.'
author:
- Viktor Könye
- Masao Ogata
bibliography:
- 'bibliography.bib'
title: 'Magnetoresistance of a Three-dimensional Dirac Gas'
---
Introduction
============
After the isolation of graphene in 2004[@Novoselov2004] the study of massless Dirac fermions in condensed matter systems became prominent over the last few decades. Following the theoretical and experimental discovery of several other two-dimensional massless fermions[@Wang2015], they were also theoretically proposed in three dimensions[@Young2012; @Wang2013a] and later found experimentally. Three-dimensional Dirac materials studied extensively are for example $\mathrm{Cd}_3\mathrm{As}_2$[@Liu2014; @Borisenko2014], $\mathrm{Na}_3\mathrm{Bi}$[@Liu2014a] and $\mathrm{TaAs}$[@Xu2015]. These materials are topological phases of matter and are classified as Dirac or Weyl semimetals[@Chiu2016]. The simplest model to describe a single independent Weyl node is the $H={\bm{\sigma}}{\bm{p}}$ Weyl Hamiltonian [@Armitage2018]. In the case of Dirac semimetals there are two degenerated Weyl nodes and the Hamiltonian becomes the $4\times4$ Dirac Hamiltonian.
Recently, gapped Dirac semimetals have attracted a lot of attention, since they are expected to have very valuable applications in advanced electronic devices[@Zhu2017; @Song2016]. Experimental realizations were found both in two dimensions[@Hunt2013; @Ye2018] and three dimensions[@Zhu2017; @Chen2017]. A simple effective model for these materials is the general $4\times 4$ Dirac Hamiltonian with a finite mass term[@Kariyado2011; @Kariyado2012; @Kariyado2017]. The dispersion relation is equivalent to that of relativistic fermions, but with effective values for the mass of the fermion and speed of light. In the limit of zero mass term the excitations become massless Dirac fermions as in Dirac semimetals.
Three-dimensional Dirac materials show a lot of exotic phenomena that are not present in usual systems nor in two-dimensional Dirac systems. One of these interesting features is the chiral anomaly and as a consequence negative magnetoresistance in Weyl semimetals. In Weyl semimetals Weyl nodes come in pairs with opposite chirality. In a magnetic field parallel to the electric field the chiral symmetry is broken leading to the chiral anomaly[@Nielsen1983] (also called Adler-Bell-Jackiw anomaly). In transport measurements this leads to a negative longitudinal magnetoresistance[@Huang2015; @Niemann2017]. A very interesting and unique feature that seems to be present in all three-dimensional gapless Dirac materials is a non-saturating linear transverse magnetoresistance[@Liang2015; @Narayanan2015; @Niemann2017; @He2014; @Feng2015]. The present paper focuses on the theoretical description of this phenomenon. At the moment a complete consistent understanding is still not available for this effect. As a generalization we also study the effects of a finite mass term, how does it change the magnetoresistance and how robust is the linear behavior against the gap opening.
The transverse magnetoconductivity and magnetoresistance of Dirac materials are widely studied both in two-dimensional[@Shon1998a] and three-dimensional[@Canuto1970; @Kaminker1981; @Abrikosov1998; @Klier2015a; @Xiao2017; @Klier2017a; @Wang2018; @Suetsugu2018] materials. The more recent calculations are carried out using the Kubo formula and either the first Born or the self-consistent Born approximation[@Bruus2004].
A formalism that was able to describe the linear behavior for Weyl semimetals was proposed by Abrikosov[@Abrikosov1998]. He used the first Born approximation to calculate the scattering rate and the Kubo formula for the conductivity. The important assumption that led to his result was that the impurity potential is a screened Coulomb potential. He only studied the case with zero chemical potential and at very high magnetic fields where only the zeroth Landau level contributes to the conductivity. More recent studies[@Xiao2017; @Klier2015a; @Klier2017a] revisited this calculation in more detail. It was shown that the screened Coulomb impurities used by Abrikosov are crucial for reproducing the linear behavior[@Klier2015a], since a completely different behavior is achieved for the simple case of short-range scatterers. In Ref. Xiao et al. calculated the scattering rates for different Landau levels, and they showed that there is Landau level dependence of the scattering rate. In their result they recovered the linear magnetoresistance for high magnetic fields, but at low fields they obtained a $B^{1/3}$ behavior. Also the effect of the first Landau level is very strong and gives a strong jump in the magnetoresistance. Klier et al.[@Klier2015a; @Klier2017a] gave an analytic argument using the self-consistent Born approximation and several approximations. They determined the scaling of the conductivity in the different magnetic field regimes. For the high field they recovered the linear behavior. For the low fields they got $B^{4/3}$ behavior. This is not consistent with the result of Ref. and the inconsistency is not yet understood.
However, the above studies only discussed the massless case. In the case of massive fermions, some older references[@Canuto1970; @Kaminker1981] discussed the problem in the context of astrophysics and thus the formalism and approximations are not exactly applicable for solid state systems. On the other hand, a recent study of the transverse magnetoresistance in gapped Dirac semimetals only used short-range scatterers and a very simple model for the scattering rate[@Wang2018]. In the case of longitudinal magnetoresistance the self-consistent Born approximation was discussed in the case of short-range impurities in Ref. . However, as shown above, the choice of impurity potential is crucial. Therefore, in order to have a proper description for the massive Dirac fermions, the inclusion of the screening is inevitable.
In this paper, we calculate the magnetoconductivity and magnetoresistivity using the first Born approximation for the massive Dirac Hamiltonian and assuming screened charged impurities. The screening is calculated taking into account the electron-electron interaction through the polarization diagram using the random phase approximation (bubble diagram). For the massless case our result is consistent to that in Ref. . In the massless case we study the effects of different scattering rate choices phenomenologically, and give a scenario where the linear behavior is recovered at low fields. We calculate in detail the case of massive Dirac materials and show that the behavior is very different from the massless case at high magnetic fields.
Model
=====
We study a three-dimensional relativistic electron gas in a constant magnetic field. The one particle dynamics is described through the Dirac equation and the one particle Hamiltonian is: $$\label{eq:ham}
{\bm{H}}{\vcentcolon =}{\bm{\gamma}}^0\left[\sum\limits_{i=1}^3v{\bm{\gamma}}^i\left(p_i+eA_i\right)+\Delta\right],$$ where $v$ replaces $c$ and $\Delta$ replaces $mc^2$ in the usual Dirac Hamiltonian. For the Dirac matrices (${\bm{\gamma}}^\mu$) the usual Dirac representation will be used. The external uniform magnetic field is assumed to point in the $z$ direction, and the Landau gauge ${\bm{A}}=(0,Bx,0)$ is used. From now on $v=1$ and $\hbar=1$ is used without losing generality. With these the Hamiltonian can be expressed as: $${\bm{H}} = \begin{pmatrix}
\Delta & 0 & \pi_z & \pi_x-i\pi_y \\
0 & \Delta & \pi_x+i\pi_y & -\pi_z \\
\pi_z & \pi_x-i\pi_y & -\Delta & 0 \\
\pi_x+i\pi_y & -\pi_z & 0 & -\Delta
\end{pmatrix}{\,,}$$ where $\pi_i{\vcentcolon =}p_i+eA_i$. We can define the following bosonic ladder operators[@Ashby2013] that satisfy $[a,a^\dag]=1$: $$\begin{aligned}
a&{\vcentcolon =}\frac{\ell_B}{\sqrt{2}}(\pi_x- i\pi_y){\,,}& a^\dag&{\vcentcolon =}\frac{\ell_B}{\sqrt{2}}(\pi_x- i\pi_y) {\,,}\end{aligned}$$ where $\ell_B{\vcentcolon =}\sqrt{1/eB}$ is the magnetic length. This length will be used as a natural length scale in the following (At $B=\SI{1}{\tesla}$, $\ell_B\approx\SI{25.66}{\nano\metre}$). Using these the Hamiltonian can be expressed as: $${\bm{H}} = \begin{pmatrix}
\Delta & 0 & p_z & \frac{\sqrt{2}}{\ell_B} a \\
0 & \Delta & \frac{\sqrt{2}}{\ell_B} a^\dag & -p_z \\
p_z & \frac{\sqrt{2}}{\ell_B} a& -\Delta & 0 \\
\frac{\sqrt{2}}{\ell_B} a^\dag & -p_z & 0 & -\Delta
\end{pmatrix}{\,.}$$
This can be solved using the eigenstates $\ket{n}$ of the $a^\dag a$ operator ($a^\dag a\ket{n}=n\ket{n}$). The energy eigenvalues and thus the Landau levels are given by: $$\label{eq:landau}
E_{n\lambda s}(p_z)=\lambda\sqrt{2neB+\Delta^2+p_z^2}{\,,}$$ where $n=0,1,2,\dots$ is the Landau index, $\lambda=\pm1$ represents the band index and $s=\pm1$ represents the two-fold degeneracy (for $n\neq0$ levels). The obtained Landau levels are shown on Fig. \[fig:landau\_levels\]. Compared to the Weyl Hamiltonian[@Abrikosov1998; @Ashby2013] the main difference is the gap present in the energy spectrum. The $n=0$ Landau level is no longer completely linear as in the case of $\Delta=0$.
![\[fig:landau\_levels\] Landau levels (\[eq:landau\]) of the Dirac Hamiltonian. The degeneracy of each level is 2 except the $n=0$ levels. $\Delta=0.5/\ell_B$ and $\ell_B=\sqrt{1/eB}$ are used.](plot_landau_levels){width=".45\textwidth"}
The eigenstates are[@Kaminker1981]: $$\label{eq:eigs1}
\ket{{\bm{\Phi}}_{n\lambda s}} = \begin{pmatrix*}[l] \phantom{-\lambda} u_{n,\lambda,s}&\ket{n-1} \\ \phantom{-\lambda}\llap{$-s$}u_{n,\lambda,-s}&\ket{n} \\ \phantom{-\lambda}\llap{$s\lambda$} u_{n,-\lambda,s}&\ket{n-1} \\ -\lambda u_{n,-\lambda,-s}&\ket{n} \end{pmatrix*}$$ for $n\neq0$ and $$\label{eq:eigs2}
\ket{{\bm{\Phi}}_{0\lambda}} = \begin{pmatrix*}[l] \phantom{-\lambda} 0& \\ \phantom{-\lambda}\llap{$-\tilde{s}$}u_{n,\lambda,-\tilde{s}}&\ket{0} \\ \phantom{-\lambda} 0& \\ -\lambda u_{n,-\lambda,-\tilde{s}}&\ket{0} \end{pmatrix*}$$ for $n=0$ where $\tilde{s}=-\mathrm{sgn}(p_z)$ and $u_{n\lambda s}$ is given by: $$\label{eq:un}
u_{n\lambda s}=\frac{1}{2}\sqrt{\left( 1+\frac{sp_z}{\sqrt{E_n^2-\Delta^2}} \right)\left( 1+\lambda\frac{\Delta}{E_n} \right)}{\,,}$$ with $E_n\equiv E_{n11}(p_z)$. The quantum numbers describing these states are $a\equiv(n,\lambda,s,p_z,p_y)$. The dispersion relation only depends on $n$, $\lambda$ and $p_z$. Each Landau level is $L^2/2\pi\ell_B^2$-fold degenerate in $p_y$ ($L$ is the length of the system) and twofold degenerate in $s$ (for $n\neq0$). The $n=0$ Landau level must be treated with caution since there is no twofold degeneracy in $s$. The wave function of the state $\ket{n}$ can be expressed with the orthonormal Hermite-functions: $$h_n(x;\ell_B) {\vcentcolon =}\frac{(\ell_B^2\pi)^{-1/4}}{\sqrt{2^nn!}}\exp(-\frac{x^2}{2\ell_B^2})H_n\left(\frac{x}{\ell_B}\right){\,,}$$ where $H_n(x)$ are the Hermite-polynomials. With these the eigenfunctions are: $$\braket{x}{n} = \frac{i^n}{L}h_n(x+\ell_B^2p_y;\ell_B){\mathrm{e}^{ip_y y}}{\mathrm{e}^{ip_z z}}{\,.}$$
Using the eigenstates $\ket{{\bm{\Phi}}_a}$ of the Hamiltonian, the Matsubara Green’s function can be expressed as[@Bruus2004]: $${\bm{G}}^{(0)}(i\omega_m) = \sum\limits_{a}\frac{\ket{{\bm{\Phi}}_{a}}\bra{{\bm{\Phi}}_{a}}}{i\omega_m+\mu-E_{a}}{\,.}$$ For practical reasons we will use several representations in the following. Using the wave functions defined as ${\bm{\phi}}_a({\bm{x}}){\vcentcolon =}\braket{{\bm{x}}}{{\bm{\Phi}}_a}$, the non-interacting Green’s function in the coordinate representation becomes: $${\bm{G}}^{(0)}({\bm{x}},{\bm{x}}',i\omega_m) = \sum\limits_{a}\frac{{\bm{\phi}}_{a}{^{\phantom{\dag}}}({\bm{x}}){\bm{\phi}}_{a}^\dag({\bm{x}}')}{i\omega_m+\mu-E_{a}}{\,.}$$ Later, the impurity averaging will be carried out in the momentum representation given by: $${\bm{G}}^{(0)}_{{\bm{k}}{\bm{k}}'}(i\omega_m) = \int{\mathop{}\!\mathrm{d}}^3 x {\mathop{}\!\mathrm{d}}^3 x' {\mathrm{e}^{-i{\bm{k}}{\bm{x}}}}{\bm{G}}^{(0)}({\bm{x}},{\bm{x}}',i\omega_m){\mathrm{e}^{i{\bm{k}}'{\bm{x}}'}}{\,,}$$ $${\bm{G}}^{(0)}(x,x',i\omega_m) = \frac{1}{V^2}\sum\limits_{{\bm{k}},{\bm{k}}'}{\mathrm{e}^{i{\bm{kx}}}}{\bm{G}}^{(0)}_{{\bm{kk}}'}(i\omega_m){\mathrm{e}^{-i{\bm{k}}'{\bm{x}}'}}{\,,}$$ with $V=L^3$. This ${\bm{G}}^{(0)}_{{\bm{k}}{\bm{k}}'}(i\omega_m)$ can be expressed using the Fourier transformed wave functions ${\bm{\phi}}_a({\bm{k}})$ as: $${\bm{G}}^{(0)}_{{\bm{k}}{\bm{k}}'}(i\omega_m) = \sum\limits_{a}\frac{{\bm{\phi}}_{a}{^{\phantom{\dag}}}({\bm{k}}){\bm{\phi}}_{a}^\dag({\bm{k}}')}{i\omega_m+\mu-E_{a}}~.$$
It is important to note here that in usual systems the Green’s function is diagonal in the momentum space, but in the current case the position dependence of the vector potential in the Hamiltonian breaks the translational invariance thus the diagonality in the momentum space is not true. Although for $k_y$ and $k_z$ the diagonality still holds, we keep both ${\bm{k}}$ and ${\bm{k}}'$ for the sake of simplicity and generality.
The final representation is the Landau level representation where the Green’s function is expressed using the eigenstates in Eqs. (\[eq:eigs1\]) and (\[eq:eigs2\]): $$\begin{aligned}
G_{ba} &= \sum\limits_{{\bm{k}},{\bm{k}}'}{\bm{\phi}}_b^\dag({\bm{k}}) {\bm{G}}_{{\bm{k}}{\bm{k}}'}{\bm{\phi}}_a({\bm{k}}'){\,.}\end{aligned}$$ Since these are the eigenstates of the Hamiltonian the non-interacting Green’s function is diagonal: $$G^{(0)}_{ba}(i\omega_m) = \frac{\delta_{ab}}{i\omega_m+\mu-E_a}{\,.}$$
Formalism
=========
Chemical potential
------------------
The chemical potential is obtained by fixing the number density of charge carriers. Similarly to Ref. the density of charge carriers can be expressed as the difference of the density of electrons and holes: $$\label{eq:carrdens}
n_e=\frac{1}{2\pi\ell_B^2}\int\limits_{-\infty}^\infty \frac{{\mathop{}\!\mathrm{d}}p_z}{2\pi} \sum\limits_{n=0}^\infty(2-\delta_{n0}) \left[f(E_n-\mu)-f(E_n+\mu) \right]{\,,}$$ where the factor $(2-\delta_{n0})$ is taking care of the different degeneracy of $s$ for the zeroth Landau level. Using the density of states $D(\varepsilon)$, Eq. (\[eq:carrdens\]) can be expressed as: $$\begin{aligned}
\label{eq:chemdos}
n_e&=\int\limits_{0}^\infty{\mathop{}\!\mathrm{d}}\varepsilon D(\varepsilon)\left[f(\varepsilon-\mu)-f(\varepsilon+\mu) \right]{\,,}\\
\label{eq:dos}
D(\varepsilon)&=\frac{1}{2\pi\ell_B^2}\int\limits_{-\infty}^\infty \frac{{\mathop{}\!\mathrm{d}}p_z}{2\pi} \sum\limits_{n=0}^\infty(2-\delta_{n0})\left[\delta(\varepsilon-E_n)+\delta(\varepsilon+E_n)\right]{\,.}\end{aligned}$$ In Eq. (\[eq:chemdos\]) we have used $D(\varepsilon)=D(-\varepsilon)$. In the following the chemical potential is calculated implicitly solving one of the above equations.
Self-energy {#sec:selfen}
-----------
The Green’s function with impurities will be approximated using the first-order Born approximation[@Bruus2004]. The self energy is obtained using the Feynman diagram shown in Fig. \[fig:feyn\_born\] similarly to previous studies[@Abrikosov1998; @Xiao2017]: $$\label{eq:born}
{\bm{\Sigma}}^B_{{\bm{k}}{\bm{k}}'}(i\omega_m) = n_i\frac{1}{V}\sum\limits_{{\bm{q}}}u_{{\bm{q}}}^2{\bm{G}}^{(0)}_{{\bm{k-q}},{\bm{k'-q}}}(i\omega_m){\,,}$$ where $n_i$ is the number density of the impurities and $u_{{\bm{q}}}$ is the Fourier transform of the effective impurity potential.
![\[fig:feyn\_born\] Feynman diagram for the first-order Born approximation of the self-energy (see Eq. (\[eq:born\])).](fig_feyn_born){width=".45\textwidth"}
Since the translational invariance is broken the self-energy is not diagonal in the momentum space. The self-energy in the Landau level representation is calculated as: $$\label{eq:selfen}
\Sigma_{ba}(i\omega_m) = \sum\limits_{{\bm{k}},{\bm{k}}'}{\bm{\phi}}_b^\dag({\bm{k}}) {\bm{\Sigma}}_{{\bm{k}}{\bm{k}}'}(i\omega_m){\bm{\phi}}_a({\bm{k}}'){\,.}$$
Using the Landau level representation of the self-energy the Dyson equation is (at $i\omega_m$ frequency): $$G_{ab} = \delta_{ab}G^{(0)}_a + G^{(0)}_a\sum\limits_{c}\Sigma_{ac}G_{cb}{\,.}$$ At this point we will assume that the self-energy is diagonal in the Landau level representation. This is not proven analytically, but checking several non-diagonal elements numerically we find that the difference between diagonal and non diagonal elements is several orders of magnitude (in the magnetic field ranges used in following sections), thus the diagonality is a valid assumption. Also we will assume that the real part of the self energy is renormalized into the chemical potential and only use the scattering rate $\Gamma_a=-\Im{\Sigma_a}$. With these, the Green’s function becomes diagonal and using the Dyson equation it is simply given as: $$\label{eq:green}
G_a(i\omega_m) = \frac{1}{i\omega_m+\mu-E_a+i\Gamma_a(i\omega_m)}{\,.}$$
Impurity potential
------------------
The effective impurity potential $u_{{\bm{q}}}$ will be calculated using the so called Random Phase Approximation[@Bruus2004; @Mahan2000] (RPA). Since the Green’s function is not diagonal in the momentum space, the treatment of the RPA must be performed with caution. Diagrammatically the screened impurity potential is expressed as in Fig. \[fig:feyn\_screen\].
![\[fig:feyn\_screen\] Feynman diagram for the RPA of the impurity potential. The double dashed lines are the effective impurity potentials, the single dashed lines are the bare impurity potentials and the wavy line represents the electron-electron interaction.](fig_feyn_screen){width=".45\textwidth"}
Starting from a charged impurity in a dielectric medium $v_{{\bm{q}}}=u_i/q^2$ (single dashed line) and the electron-electron interaction $w_{{\bm{q}}}=u_e/q^2$ (wavy line) the screened impurity potential is expressed implicitly as: $$u_{{\bm{q}}{\bm{q}}'}(i\omega_\lambda) = v_{{\bm{q}}}\delta_{{\bm{q}}{\bm{q}}'}+\frac{1}{V}\sum\limits_{{\bm{q}}''}w_{{\bm{q}}}\Pi^0_{{\bm{q}}{\bm{q}}''}(i\omega_\lambda)u_{{\bm{q}}''{\bm{q}}'}(i\omega_\lambda){\,,}$$ where $\Pi^0$ is the bubble diagram for electrons. For simplicity we study the static and long-wave limit $i\omega_\lambda,{\bm{q}},{\bm{q}}'\to 0$. In this limit we assume that the bubble diagram is diagonal and constant $\Pi^0_{{\bm{qq}}'}(i\omega_m)\approx\delta_{{\bm{qq}}'}\Pi^0_{{\bm{00}}}(0)$. With this the effective screening will also be diagonal and simply expressed as: $$u_{{\bm{q}}}=\frac{u_i}{q^2-u_e\Pi^0_{{\bm{00}}}(0)}\equiv\frac{u_i}{q^2+\kappa^2}{\,.}$$
![\[fig:feyn\_bubble\] The bubble diagram used to calculate the screening wavenumber in Eq. (\[eq:kappa\]).](fig_feyn_bubble){width=".45\textwidth"}
Here $\kappa$ is the screening wavenumber and it is obtained using the static and long-wave limit of the bubble diagram (see Fig. \[fig:feyn\_bubble\]) as: $$\label{eq:kappa}
\kappa^2=-\frac{u_e}{\beta V^2}\sum\limits_{m,{\bm{k}},{\bm{k}}'}\Tr\left[{\bm{G}}_{{\bm{kk}}'}(i\omega_m){\bm{G}}_{{\bm{k}}'{\bm{k}}}(i\omega_m)\right]{\,.}$$
Linear response theory {#sec:lrt}
----------------------
The one particle current operator of the system is: $$\label{eq:curr}
{\bm{J}}_i = {\bm{\gamma}}^0{\bm{\gamma}}^i={\begin{pmatrix} 0 & {\bm{\sigma}}^i \\ {\bm{\sigma}}^i & 0 \end{pmatrix}}{\,.}$$ In the Landau level representation, this is: $$J^{(i)}_{ab} = \int{\mathop{}\!\mathrm{d}}^3 x {\bm{\phi}}_a^\dag({\bm{x}}) {\bm{J}}_i {\bm{\phi}}_b({\bm{x}}){\,.}$$ The optical conductivity is calculated using the Matsubara current-current correlation function[@Bruus2004] as: $$\sigma_{ij}(\omega) = \frac{ie^2}{\omega} \lim\limits_{\delta\to0^+}\Pi_{ij}(i\omega_\lambda=\omega+i\delta){\,,}$$ where the correlation function is calculated as: $$\begin{aligned}
\Pi_{ij}(i\omega_\lambda) &= \frac{1}{V} \sum\limits_{a,b}\frac{1}{\beta}\sum\limits_n J^{(i)}_{ab} G_b(i\omega_n+i\omega_\lambda) J^{(j)}_{ba} G_a(i\omega_n){\,,}\end{aligned}$$ with the Green’s functions taken from Eq. (\[eq:green\]). The vertex correction is neglected in this formula. Based on the results obtained for the Weyl Hamiltonian in Ref. we assume that the effect of the vertex correction is a magnetic field independent renormalization of the scattering rate. The Matsubara sum can be transformed into an integral[@Abrikosov1965] (keeping in mind that due to $\Gamma\propto\mathrm{sgn}(\omega_m)$ there is a branch cut): $$\begin{aligned}
\Pi_{ij}(\omega) &= \frac{1}{V} \sum\limits_{a,b} J^{(i)}_{ab} J^{(j)}_{ba} C_{ba}(\omega){\,,}\end{aligned}$$ where $$\begin{aligned}
\nonumber
C_{ba}(\omega)= -2&\int\limits_{-\infty}^\infty \frac{{\mathop{}\!\mathrm{d}}\varepsilon}{2\pi}\bigg[f(\varepsilon)G_b^R(\varepsilon+\omega)\Im{G^R_a(\varepsilon)}+\\
&+f(\varepsilon+\omega)\Im{G^R_b(\varepsilon+\omega)}G^A_a(\varepsilon)\bigg]{\,,}\end{aligned}$$ where $f(\varepsilon)$ is the Fermi-Dirac distribution and $$\begin{aligned}
G^{R/A}_a(\varepsilon)&{\vcentcolon =}\frac{1}{\varepsilon-E_a+\mu\pm i\Gamma_a(\varepsilon)}{\,.}\end{aligned}$$
The DC conductivity is calculated by taking $\omega\to 0$: $$\label{eq:sigma}
\sigma_{ij} = -\lim\limits_{\omega\to0}\frac{e^2}{\omega}\Im{\Pi_{ij}(\omega)}{\,.}$$
Results
=======
Chemical potential
------------------
As a realistic charge carrier density, $n_e=\SI{1e18}{\per\cubic\centi\metre}$ will be used[@Liang2015]. The density of states is calculated using Eq. (\[eq:dos\]) as: $$\label{eq:dosres}
D(\varepsilon)=\frac{1}{2\pi^2\ell_B^2}\hspace{-5pt}\sum\limits_{n=0}^{\left\lfloor\frac{(\varepsilon^2-\Delta^2)\ell_B^2}{2}\right\rfloor}\hspace{-5pt}(2-\delta_{n0})\frac{|\varepsilon|}{\sqrt{\varepsilon^2-\Delta^2-\frac{2n}{\ell_B^2}}}{\,.}$$ In the case of $\Delta=0$ we recover the result obtained in Ref. . The density of states is shown in Fig. \[fig:dos\].
![\[fig:dos\] Density of states calculated from Eq. (\[eq:dosres\]). The mass term is chosen as $\ell_B\Delta=1$.](plot_dos){width=".45\textwidth"}
Substituting this in Eq. (\[eq:chemdos\]) we obtain the expression for the charge carrier density. At zero temperature after integration: $$\label{eq:chempotsimp}
n_e=\frac{1}{2\pi^2\ell_B^2}\sum\limits_{n=0}^{\left\lfloor\frac{(\mu^2-\Delta^2)\ell_B^2}{2}\right\rfloor}(2-\delta_{n0})\sqrt{\mu^2-\Delta^2-\frac{2n}{\ell_B^2}}{\,.}$$ This is consistent with the $\Delta=0$ result in Ref. . In the high magnetic field limit ($\ell_B\to 0$) only the zeroth Landau level contributes and the equation simply yields $$\label{eq:chemhigh}
\mu = \sqrt{4\pi^4n_e^2\ell_B^4+\Delta^2}{\,.}$$ As we can see at high magnetic fields $\mu\to\Delta$, and for $\Delta=0$ $\mu\propto 1/B$ as in Ref. . In the low magnetic field limit ($\ell_B\to\infty$) the summation can be substituted with an integral and we obtain: $$\label{eq:chemB0}
n_e=\frac{(\mu^2-\Delta^2)^\frac{3}{2}}{3\pi^2}{\,,}$$ which reproduces the zero magnetic field result.
Solving Eq. (\[eq:chempotsimp\]) for zero temperature or Eq. (\[eq:chemdos\]) for finite temperature numerically we obtain the chemical potential as shown in Fig. \[fig:chem\].
![\[fig:chem\] Magnetic field dependence of the chemical potential for two mass terms $\Delta_0=0$ and $\Delta_1=2/\ell_{\SI{1}{\tesla}}\approx\SI{50}{\milli\electronvolt}$ and for three temperatures $T_0=\SI{0}{\kelvin}$, $T_1=\SI{30}{\kelvin}$ and $T_2=\SI{100}{\kelvin}$. The carrier density is fixed at $n_e=10^{18}\si{\per\cubic\centi\metre}$. (a) is for the small magnetic fields with oscillating behavior and (b) is for the large magnetic field limit. The vertical lines show the magnetic fields where a new Landau level crosses the chemical potential as calculated from Eq. (\[eq:Bpeak\]).](plot_chem "fig:"){width=".45\textwidth"} ![\[fig:chem\] Magnetic field dependence of the chemical potential for two mass terms $\Delta_0=0$ and $\Delta_1=2/\ell_{\SI{1}{\tesla}}\approx\SI{50}{\milli\electronvolt}$ and for three temperatures $T_0=\SI{0}{\kelvin}$, $T_1=\SI{30}{\kelvin}$ and $T_2=\SI{100}{\kelvin}$. The carrier density is fixed at $n_e=10^{18}\si{\per\cubic\centi\metre}$. (a) is for the small magnetic fields with oscillating behavior and (b) is for the large magnetic field limit. The vertical lines show the magnetic fields where a new Landau level crosses the chemical potential as calculated from Eq. (\[eq:Bpeak\]).](plot_chem2 "fig:"){width=".45\textwidth"}
For the high and low magnetic field limits we see the behavior explained using Eq. (\[eq:chempotsimp\]). In the case of finite temperature we see some minor deviation but the main behavior remains the same. Between the two limits we see oscillations caused by the singularities present in the density of states. At zero temperature the oscillations are more prominent with sharp changes in the chemical potential. Finite temperature smoothens the curves and at high temperatures the oscillation almost completely disappears.
The oscillation occurs when a new Landau level crosses the chemical potential. At zero temperature the magnetic fields where the oscillation occurs can be calculated from the condition of $(\mu^2-\Delta^2)\ell_B^2/2\in \mathbb{N}$. Solving Eq. (\[eq:chempotsimp\]) with this condition yields: $$\begin{aligned}
\label{eq:Bpeak}
B_m=\left(\frac{\sqrt{2}\pi^2n_e}{A(m)}\right)^{\frac{2}{3}},~A(m){\vcentcolon =}\sum\limits_{n=0}^m(2-\delta_{n0})\sqrt{m-n}{\,,}\end{aligned}$$ where $m\in\mathbb{N}$ denotes the $m^{\text{th}}$ Landau level that crosses the chemical potential. We can see from the formula and the numerical results as well that the peaks occur at the same magnetic field independently of $\Delta$.
Scattering rate {#sec:gamma}
---------------
The screening wavenumber using Eq. (\[eq:kappa\]) becomes: $$\label{eq:wave}
\kappa^2 =\frac{-u_e}{2\pi\ell_B^2}\int\limits_{-\infty}^\infty \frac{{\mathop{}\!\mathrm{d}}p_z}{2\pi} \sum\limits_{\substack{n=0\\\lambda=\pm1}}^\infty(2-\delta_{n0})\frac{\partial f(\lambda E_n-\mu)}{\partial \lambda E_n}{\,.}$$ At zero temperature this formula becomes the same formula as the expression for the density of states (\[eq:dos\]) evaluated at the chemical potential $\kappa^2=D(\mu)$. This can be calculated as explained in the previous section:
$$\label{eq:waveo}
\kappa^2=\frac{u_e}{2\pi^2\ell_B^2}\sum\limits_{n=0}^{\left\lfloor\frac{(\mu^2-\Delta^2)\ell_B^2}{2}\right\rfloor}(2-\delta_{n0})\frac{|\mu|}{\sqrt{\mu^2-\Delta^2-\frac{2n}{\ell_B^2}}}{\,.}$$
For high magnetic fields ($\ell_B\to 0$) only the zeroth Landau level contributes to the screening. Using the high magnetic field dependence of $\mu$ (\[eq:chemhigh\]), we obtain: $$\kappa^2\sim\frac{u_e}{2\pi^2\ell_B^2}\frac{\sqrt{4\pi^4n_e^2\ell_B^4+\Delta^2}}{2\pi^2n_e\ell_B^2}{\,.}$$ For $B\to\infty$ $\kappa^2\propto B$ for $\Delta=0$ as used in Ref. and $\kappa^2\propto B^2$ for $\Delta\neq0$. In the zero field limit similarly to Eq. (\[eq:chemB0\]) the sum can be substituted with an integral and $$\kappa^2\sim\frac{u_e}{\pi^2}\sqrt{3\pi^2n_e+\left(3\pi^2n_e\right)^{\frac{1}{3}}\Delta^2}$$ goes to a constant. The screening wavenumber as a function of magnetic field is shown in Fig. \[fig:screen\].
![\[fig:screen\] Magnetic field dependence of the screening wavenumber for several cases as in Fig. \[fig:chem\].](plot_screen){width=".45\textwidth"}
From now on we introduce the dimensionless quantities using $\ell_B =\sqrt{1/eB}$: $$\begin{aligned}
{\mathcal{P}} &{\vcentcolon =}\ell_B {\bm{p}}{\,,}& \mathcal{E}&{\vcentcolon =}\ell_B\varepsilon{\,,}& \mathcal{M}&{\vcentcolon =}\ell_B\mu{\,,}& \mathcal{D}&{\vcentcolon =}\ell_B\Delta{\,.}\end{aligned}$$ We have to keep in mind that $\mu$ and $\Delta$ are not integration variables but parameters thus $\mathcal{M}$ and $\mathcal{D}$ are functions of $B$. Using Eq. (\[eq:selfen\]) for the scattering rate we obtain:
$$\begin{aligned}
\label{eq:gam}
\Gamma_{n\lambda s}({\mathcal{E}},{\mathcal{P}}_z,B) &=\frac{n_i\pi}{\ell_B^2}\sum\limits_{\ell=0}^{\left\lfloor\frac{({\mathcal{E}}+{\mathcal{M}})^2-\mathcal{D}^2}{2} \right\rfloor}\sum\limits_{\substack{\alpha = \pm 1 \\ t=\pm1}} \int\frac{{\mathop{}\!\mathrm{d}}{\mathcal{Q}}_x{\mathop{}\!\mathrm{d}}{\mathcal{Q}}_y}{(2\pi)^3}u_{{\bm{{\mathcal{Q}}}}_{\ell\alpha}}^2(B) \left|\frac{{\mathcal{E}}+{\mathcal{M}}}{\sqrt{({\mathcal{E}}+{\mathcal{M}})^2-2\ell-\mathcal{D}^2}} \right|
\left|F_{n\lambda s,\ell\gamma_ot}({\bm{{\mathcal{Q}}}}_{\ell\alpha},{\mathcal{P}}_z)\right|^2{\,,}\end{aligned}$$
$$\begin{aligned}
\label{eq:four}
F_{n\lambda s,\ell\gamma t}({\bm{{\mathcal{Q}}}},{\mathcal{P}}_z) &{\vcentcolon =}\int{\mathop{}\!\mathrm{d}}{\mathcal{X}}{\bm{\phi}}_{n\lambda s}^\dag({\mathcal{X}};0,{\mathcal{P}}_z) {\bm{\phi}}_{\ell\gamma t}({\mathcal{X}};{\mathcal{Q}}_y,{\mathcal{P}}_z-{\mathcal{Q}}_z){\mathrm{e}^{i{\mathcal{Q}}_x{\mathcal{X}}}}{\,,}\end{aligned}$$
$$\begin{aligned}
{\bm{{\mathcal{Q}}}}_{\ell\pm} &{\vcentcolon =}({{\mathcal{Q}}_x},{{\mathcal{Q}}_y},{{\mathcal{Q}}_{\ell\pm}}){\,,}&{\mathcal{Q}}_{\ell\pm} &{\vcentcolon =}{\mathcal{P}}_z\pm\sqrt{({\mathcal{E}}+{\mathcal{M}})^2-2\ell-{\mathcal{D}}^2}{\,.}\end{aligned}$$
where $\gamma_o = \mathrm{sgn}(\varepsilon+\mu)$ and for the impurity potential $$u_{{\bm{{\mathcal{Q}}}}}(B) = \frac{u_i\ell_B^2}{{\bm{{\mathcal{Q}}}}^2+\ell_B^2\kappa^2(B)}$$ is used, where the screening is calculated from Eq. (\[eq:wave\]). Note here that the summation over $t$ is only for $l\neq0$.
![\[fig:Gamma\_B\] Scattering rate calculated from Eq. (\[eq:gam\]) with indices $n=0$ and $\lambda=1$ as a function of magnetic field at zero temperature. The screening wavenumber is calculated using Eq. (\[eq:wave\]). The density of charge carriers is $n_e=10^{18}\si{\per\cubic\centi\metre}$. $\Delta_0=0$ and $\Delta_1=2/\ell_{\SI{1}{\tesla}}$.](plot_Gamma_B){width=".45\textwidth"}
In the high magnetic field limit at $\mathcal{E}=0$ (we will see later that this is the important energy at high magnetic fields) only the zeroth Landau level plays role. Using Eq. (\[eq:chemhigh\]) at zero temperature in the case of $\Delta=0$ gives $\Gamma\propto 1/B$ for high magnetic fields. In the case of finite mass term at high fields we get $\Gamma\propto1/B^2$. In Fig. \[fig:Gamma\_B\] we show the scattering rate as a function of the magnetic field at ${\mathcal{E}}=0$ and ${\mathcal{P}}=0$. For the numerical calculations $u_e=e^2/\varepsilon_0$ is used and the energy scale is set using $v=\SI{1e6}{\metre\per\second}$ based on Refs. and . The Fourier transformation in Eq (\[eq:four\]) is calculated using the fractional Fourier transform[@Bailey1994]. The ${\mathcal{Q}}_x$ integral is done using the Simpson’s rule on the result of the fractional Fourier transform and finally the ${\mathcal{Q}}_y$ integral is done through Gaussian quadrature.
![\[fig:Gamma\_Bs\] Scattering rate calculated from Eq. (\[eq:gam\]) with indices $n=0$ and $\lambda=1$ as a function of magnetic field. The density of charge carriers is $n_e=10^{18}\si{\per\cubic\centi\metre}$. Different type of screening wavenumbers are used: a constant, a linear in magnetic field and the one calculated from Eq. (\[eq:waveo\]). $\Delta=0$ and $T=\SI{0}{\kelvin}$.](plot_Gamma_Bs){width=".45\textwidth"}
As shown in Fig. \[fig:Gamma\_B\], at the high magnetic field limit we see the behavior described above. At low fields SdH oscillations can be seen. The effect of the mass term is only relevant in the extremely high magnetic field limit. As a function of the magnetic field the scattering rate first has an increasing background (with SdH oscillations) then after reaching the quantum limit it starts to decrease.
This behavior is strongly dependent on the choice of the magnetic field dependence of the screening wavenumber. In Fig. \[fig:Gamma\_Bs\] different wavenumber choices are shown for the zero temperature, zero mass term case. If we assume that $\kappa$ is independent of $B$ (red line in Fig. \[fig:Gamma\_Bs\]) we get a monotonic increase at high magnetic fields and at lower fields there are divergent peaks with a constant background. If we assume that $\kappa^2$ is proportional to $B$ (green line), $\Gamma\propto 1/B$ at high magnetic fields and at lower fields the divergent peaks are on a decreasing background (This is the screening used in Ref. ). Using our screening calculated as Eq. (\[eq:wave\]) (blue line in Fig. \[fig:Gamma\_Bs\]) we obtain a magnetic field dependence of $\Gamma$ with a maximum at the quantum limit. We will see that this behavior of increasing then decreasing scattering rate is important to reproduce the linear magnetoresistance.
Hall Conductivity
-----------------
The conductivity is calculated through the steps described in Sec. \[sec:lrt\]. In the case of the Hall conductivity we will neglect the effect of impurities since they are expected not to have a large influence on the result[@Abrikosov1998; @Xiao2017].
The matrix elements of the current operator using Eq. (\[eq:curr\]) are:
$$\begin{aligned}
\label{eq:jx}
J_{ab}^{(x)} &= -\phantom{i}\delta_{p_yp_y'}\delta_{p_zp_z'}\left[\delta_{n,n'-1}(\lambda u_{n,-\lambda,-s}u_{n',\lambda',s'}+ss'\lambda'u_{n,\lambda,-s}u_{n',-\lambda',s'})+(n\leftrightarrow n',\lambda\leftrightarrow\lambda',s\leftrightarrow s')\right]{\,,}\\
J_{ab}^{(y)} &= \phantom{-}i\delta_{p_yp_y'}\delta_{p_zp_z'}\left[\delta_{n,n'-1}(\lambda u_{n,-\lambda,-s}u_{n',\lambda',s'}+ss'\lambda'u_{n,\lambda,-s}u_{n',-\lambda',s'})-(n\leftrightarrow n',\lambda\leftrightarrow\lambda',s\leftrightarrow s')\right]{\,.}\end{aligned}$$
In the absence of impurities, the imaginary part of the Green’s function can be substituted with a Dirac delta and $\Im{G_a^R(\varepsilon)}=-i\pi\delta(\varepsilon-E_a+\mu)$. Using Eq. (\[eq:sigma\]) and evaluating the integral and the DC limit, the formula for the Hall conductivity becomes: $$\label{eq:hall}
\sigma_{xy} =2\frac{\sigma_0}{\ell_B}\sum\limits_{n=0}^{\infty}\sum\limits_{\substack{\lambda,\lambda' = \pm 1 \\ s,s'=\pm1}} \int {\mathop{}\!\mathrm{d}}{\mathcal{P}}_z \left(\lambda u_{n,-\lambda,-s}u_{n+1,\lambda',s'}+ss'\lambda'u_{n,\lambda,-s}u_{n+1,-\lambda',s'}\right)^2 \frac{f(\lambda E_n-{\mathcal{M}})-f(\lambda'E_{n+1}-{\mathcal{M}})}{(\lambda E_n-\lambda' E_{n+1})^2}{\,,}$$ where $\sigma_0=e^2/h$ is the inverse of the von Klitzing constant. The summation over $s$ is taken only for $n\neq0$. This formula can be simplified by using the properties of the Fermi distribution ( $f(-{\mathcal{E}}-{\mathcal{M}})=1-f({\mathcal{E}}+{\mathcal{M}})$ ), the definition of $u_{n\lambda s}$ (\[eq:un\]) and the explicit form of $E_n$. After the summations over the $\lambda,\lambda',s,s'$ indices, we can show that the formula becomes: $$\label{eq:hall2}
\sigma_{xy} =\frac{\sigma_0}{\ell_B}\sum\limits_{n=0}^{\infty}\int {\mathop{}\!\mathrm{d}}{\mathcal{P}}_z (1+2n) \big\{\left[f(E_n-{\mathcal{M}})-f(E_{n}+{\mathcal{M}})\right]-\left[f(E_{n+1}-{\mathcal{M}})-f(E_{n+1}+{\mathcal{M}})\right]\big\}{\,.}$$
After rearranging the summation over $n$ we can see that Eq. (\[eq:hall2\]) is proportional to Eq. (\[eq:carrdens\]) and the Hall conductivity can be expressed using the carrier density as: $$\label{eq:hallsimp}
\sigma_{xy}=\sigma_02\pi\ell_B^2n_e=\frac{en_e}{B}{\,.}$$
![\[fig:condxy\] Hall conductivity $\sigma_{xy}$ calculated from Eq. (\[eq:hall\]) as a function of magnetic field at $T=\SI{30}{\kelvin}$. The density of charge carriers is $n_e=10^{18}\si{\per\cubic\centi\metre}$. $\sigma_0/\ell_{\SI{1}{\tesla}}\approx \SI{15}{\per\ohm\per\centi\metre}$.](plot_condxy){width=".45\textwidth"}
Since the charge carrier density is constant the Hall conductivity is exactly inversely proportional to the magnetic field as in usual systems. To check the validity of Eq. (\[eq:hallsimp\]) the Hall conductivity is calculated numerically from Eq. (\[eq:hall\]). The numerical results at different mass terms can be seen in Fig. \[fig:condxy\] (the results are only shown at one finite temperature, but at different temperatures we get exactly the same result). As we can see the Hall conductivity does not depend on the mass term nor the temperature (in the case of no impurities) and it exactly satisfies Eq. (\[eq:hallsimp\]).
Diagonal conductivity
---------------------
In the case of the diagonal component including impurities is necessary in order to get finite conductivity. The impurity is included in the Green’s function as explained in Sec. \[sec:selfen\]. From Eq. (\[eq:jx\]) we see that the matrix elements of the $x$ component of the current operator are all real. Thus, in the Eq. (\[eq:sigma\]) when taking the imaginary part we only need the imaginary part of $C_{ba}(\omega)$. Similarly to Ref. with our notations, we obtain:
\[eq:sigxx\] $$\sigma_{xx} =\frac{2}{\pi}\frac{\sigma_0}{\ell_B}\sum\limits_{n=0}^{\infty}\sum\limits_{\substack{\lambda,\lambda' = \pm 1 \\ s,s'=\pm1}} \int {\mathop{}\!\mathrm{d}}{\mathcal{P}}_z \left(\lambda u_{n,-\lambda,-s}u_{n+1,\lambda',s'}+ss'\lambda'u_{n,\lambda,-s}u_{n+1,-\lambda',s'}\right)^2 C_{n\lambda s;n+1\lambda's'}{\,,}$$ $$C_{ab}{\vcentcolon =}\frac{\beta}{4\ell_B}\int\limits_{-\infty}^\infty \frac{{\mathop{}\!\mathrm{d}}{\mathcal{E}}}{2\pi}\frac{1}{\cosh^2\left(\frac{\beta{\mathcal{E}}}{2\ell_B}\right)}\Im{G_a^R({\mathcal{E}})}\Im{G^R_b({\mathcal{E}})} {\,.}$$
The summation over $s$ is again only taken for $n\neq0$.
First, let us discuss analytically the behavior of $\sigma_{xx}$ in high magnetic fields. In the high magnetic field limit ($\ell_B\to0$) the formula for $C_{ab}$ becomes equivalent to the zero temperature formula since the $T\ell_B$ combination goes to zero. Thus, we have: $$C_{ab}=\frac{1}{2\pi}\Im{G_a^R(0)}\Im{G^R_b(0)}{\,,}$$ where $$\Im{G_a^R(0)}=-\frac{\ell_B\Gamma_a({\mathcal{E}}=0,B)}{(E_a-{\mathcal{M}})^2+(\ell_B\Gamma_a({\mathcal{E}}=0,B))^2}{\,.}$$ In high magnetic fields ${\mathcal{M}}\to{\mathcal{D}}$, so we will use ${\mathcal{M}}\approx{\mathcal{D}}$. Using the high magnetic field limit of the scattering rate derived in Sec. \[sec:gamma\] we see that in both cases the scattering rate is a power function $\Gamma\propto \ell_B^q$ ($q=2$ for ${\mathcal{D}}=0$ and $q=4$ for ${\mathcal{D}}\neq0$) and $\ell_B\Gamma\to0$. For $n>0$ Landau levels, $|E_a-{\mathcal{M}}|>0$ thus $C_{ab}\propto \ell_B^{(2q+2)}$ and as a consequence $\sigma_{xx}\propto \ell_B^{2q+1}$. The case of $n=0$ is more delicate since $|E_0-{\mathcal{M}}|\geq0$. In this case the limit gives a Dirac delta for the imaginary part of the Green’s function $\Im{G_a^R(0)}\sim -\pi\delta(E_a-{\mathcal{M}})$. As a function of ${\mathcal{P}}_z$ this becomes: $$\label{eq:deltajac}
-\pi\delta(E_a-{\mathcal{M}}) = -\pi\frac{\sqrt{{\mathcal{M}}-{\mathcal{D}}}}{{\mathcal{M}}}\delta({\mathcal{P}}_z\pm\sqrt{{\mathcal{M}}-{\mathcal{D}}})~{\,.}$$ In the case of ${\mathcal{D}}=0$ this has no magnetic field dependence and thus $C_{ab}\propto\ell_B^{q+1}$ leading to $\sigma_{xx}\propto\ell_B^{q}$. But in the case of ${\mathcal{D}}\neq0$ Eq. (\[eq:deltajac\]) is proportional to $B^{-1}$ and thus $\sigma_{xx}\propto\ell_B^{q+2}$. Since the $n=0$ case decays the least rapidly it will be the dominant at high magnetic fields so the overall magnetic field dependence of the conductivity becomes: $\sigma_{xx}\propto B^{-1}$ for $\Delta=0$ and $\sigma_{xx} \propto B^{-3}$ for $\Delta\neq0$.
Next, we calculate $\sigma_{xx}$ assuming several choices of magnetic field dependence of the scattering rate in order to clarify its effects on $\sigma_{xx}$. The scattering rate is assumed to be independent of Landau levels and other variables except the magnetic field. The obtained results of $\sigma_{xx}$ for $\Delta=0$ and $\Delta=2/\ell_{\SI{1}{\tesla}}$ are shown in Fig. \[fig:condxx\]. The impurity density is chosen in a way that the ratio of $\sigma_{xx}$ to $\sigma_{xy}$ in our results is similar to the experimental results[@Liang2015].
![\[fig:condxx\] Transverse diagonal conductivity $\sigma_{xx}$ calculated from Eq. (\[eq:sigxx\]) as a function of magnetic field at different temperatures. $\Delta=0$ (top plot) and $\Delta=2/\ell_{\SI{1}{\tesla}}$ (bottom plot). The scattering rate is chosen phenomenologically based on the numerical results in Fig. \[fig:Gamma\_B\]. The inset shows the scattering rates used (no Landau level dependence is assumed). $T_0=\SI{0}{\kelvin}$, $T_1=\SI{50}{\kelvin}$, the density of charge carriers is $n_e=10^{18}\si{\per\cubic\centi\metre}$ and $\sigma_0/\ell_{\SI{1}{\tesla}}\approx \SI{15}{\per\ohm\per\centi\metre}$.](plot_condxx "fig:"){width=".45\textwidth"} ![\[fig:condxx\] Transverse diagonal conductivity $\sigma_{xx}$ calculated from Eq. (\[eq:sigxx\]) as a function of magnetic field at different temperatures. $\Delta=0$ (top plot) and $\Delta=2/\ell_{\SI{1}{\tesla}}$ (bottom plot). The scattering rate is chosen phenomenologically based on the numerical results in Fig. \[fig:Gamma\_B\]. The inset shows the scattering rates used (no Landau level dependence is assumed). $T_0=\SI{0}{\kelvin}$, $T_1=\SI{50}{\kelvin}$, the density of charge carriers is $n_e=10^{18}\si{\per\cubic\centi\metre}$ and $\sigma_0/\ell_{\SI{1}{\tesla}}\approx \SI{15}{\per\ohm\per\centi\metre}$.](plot_condxx_D "fig:"){width=".45\textwidth"}
When we assume that the scattering rate has the same magnetic-field dependence as described in Sec. \[sec:gamma\] (i.e., $\Gamma\propto B^{-1}$ for $\Delta=0$ and $\Gamma \propto B^{-2}$ for $\Delta \neq 0$, green lines in the insets of Fig. \[fig:condxx\]), the analytic behaviors in high magnetic fields are reproduced. (Note that the green lines in the main figures of Fig. \[fig:condxx\] overlap with blue lines in the high field region.) However, in the low field region, we get a faster decrease than $B^{-1}$. This scattering rate is the same as used by Abrikosov[@Abrikosov1998].
On the other hand, when we assume $\Gamma\propto B$ (cyan lines in Fig. \[fig:condxx\]), we obtain $\sigma_{xx}\propto B^{-1}$ in the low field region. However, in this case, the analytic behaviors in high magnetic fields are not reproduced.
As shown in Fig. \[fig:Gamma\_B\] in Sec. \[sec:gamma\], the numerically obtained scattering rate roughly behaves as $\Gamma\propto B$ in the low field region, and $\Gamma\propto B^{-1}$ for $\Delta=0$ and $\Gamma \propto B^{-2}$ for $\Delta \ne 0$ in the high field region. Therefore, we connect these dependencies phenomenologically as shown with blue and red curves in Fig. \[fig:condxx\]. In these cases, we obtain a $\sigma_{xx}\propto B^{-1}$ background with SdH oscillations superimposed in all the magnetic field region for the $\Delta=0$ case, while $\sigma_{xx}\propto B^{-1}$ in the low field region and $\sigma_{xx}\propto B^{-3}$ in the high field region for $\Delta\ne 0$ case. As shown in Fig. \[fig:Gamma\_B\], there is no significant difference between $\Gamma_0$ for $\Delta=0$ and $\Gamma_0$ for $\Delta\ne 0$ in the low field region. The conductivity also behaves similarly, and the two cases behave differently only in the quantum limit where only the lowest Landau level is important.
About the temperature dependence, we can see it is negligible at high fields. This is because the temperature is only present in the $T\ell_B$ combination which goes to zero as the magnetic field gets higher. The effect of temperature is the suppression of the SdH oscillations.
A more precise numerical result can be achieved using the scattering rate calculated from Eq. (\[eq:gam\]). However, the exact numerical integration of $\Gamma$ is a very heavy calculation. Therefore, we assume that the scattering rate is independent of momentum and energy (${\mathcal{P}}_z=0$ and ${\mathcal{E}}=0$) and only the Landau level dependence and magnetic field dependence are kept. For the strength of the interactions we assume $u_e=u_i=e^2/\varepsilon_0\varepsilon_r$ considering different relative permittivities. The results for both the massless and massive cases are shown in Fig. \[fig:condxx\_kap\].
![\[fig:condxx\_kap\] Transverse diagonal conductivity $\sigma_{xx}$ calculated from Eq. (\[eq:sigxx\]) as a function of magnetic field at zero temperature. $\Delta=0$ (top plot) and $\Delta=2/\ell_{\SI{1}{\tesla}}$ (bottom plot). The scattering rate is calculated using Eq. (\[eq:gam\]) using screening wavenumbers calculated through Eq. (\[eq:wave\]). The inset figure shows the scattering rates used (n=1 Landau level). The density of charge carriers is $n_e=10^{18}\si{\per\cubic\centi\metre}$ and $\sigma_0/\ell_{\SI{1}{\tesla}}\approx \SI{15}{\per\ohm\per\centi\metre}$.](plot_condxx_kap "fig:"){width=".45\textwidth"} ![\[fig:condxx\_kap\] Transverse diagonal conductivity $\sigma_{xx}$ calculated from Eq. (\[eq:sigxx\]) as a function of magnetic field at zero temperature. $\Delta=0$ (top plot) and $\Delta=2/\ell_{\SI{1}{\tesla}}$ (bottom plot). The scattering rate is calculated using Eq. (\[eq:gam\]) using screening wavenumbers calculated through Eq. (\[eq:wave\]). The inset figure shows the scattering rates used (n=1 Landau level). The density of charge carriers is $n_e=10^{18}\si{\per\cubic\centi\metre}$ and $\sigma_0/\ell_{\SI{1}{\tesla}}\approx \SI{15}{\per\ohm\per\centi\metre}$.](plot_condxx_kappa_D "fig:"){width=".45\textwidth"}
In the high magnetic field region, we recover the magnetic field dependencies discussed above. In the low field region, $\sigma_{xx}\propto B^{-5/3}$ and the effect of the first Landau level appears as a very strong jump similarly to what was found in Ref. . We see that changing the relative permittivity changes the height of this jump. In the inset, we also show $\Gamma_1$, since this determines mainly the conductivity in high fields. In this system higher scattering rate means higher conductivity (contrary to normal system where the opposite is true). This means that if we increase the density of impurities the conductivity is also increased.
Magnetoresistance
-----------------
The longitudinal resistivity is calculated as: $$\label{eq:rhoxx}
\varrho_{xx}=\frac{\sigma_{xx}}{\sigma_{xx}^2+\sigma_{xy}^2}{\,.}$$
![\[fig:mag\] Magnetoresistance $\varrho_{xx}$ calculated from Eq. (\[eq:rhoxx\]) as a function of magnetic field. $\Delta=0$ (top panel) and $\Delta=2/\ell_{\SI{1}{\tesla}}$ (bottom panel). The resistivity calculated from the phenomenological result (red line) and the resistivity calculated microscopically using the first Born approximation with $\varepsilon=1$ (blue). The density of charge carriers is $n_e=10^{18}\si{\per\cubic\centi\metre}$ and $\sigma_0/\ell_{\SI{1}{\tesla}}\approx \SI{15}{\per\ohm\per\centi\metre}$.](plot_mag "fig:"){width=".45\textwidth"} ![\[fig:mag\] Magnetoresistance $\varrho_{xx}$ calculated from Eq. (\[eq:rhoxx\]) as a function of magnetic field. $\Delta=0$ (top panel) and $\Delta=2/\ell_{\SI{1}{\tesla}}$ (bottom panel). The resistivity calculated from the phenomenological result (red line) and the resistivity calculated microscopically using the first Born approximation with $\varepsilon=1$ (blue). The density of charge carriers is $n_e=10^{18}\si{\per\cubic\centi\metre}$ and $\sigma_0/\ell_{\SI{1}{\tesla}}\approx \SI{15}{\per\ohm\per\centi\metre}$.](plot_magD "fig:"){width=".45\textwidth"}
First, we discuss the magnetoresistance calculated from the phenomenological scattering rate represented by the red line in the insets of Fig. \[fig:condxx\]. In this case, the obtained $\sigma_{xx}$ is proportional to $B^{-1}$ with SdH oscillations for the $\Delta=0$ case. The magnetoresistance calculated in Eq. (\[eq:rhoxx\]) becomes $\rho_{xx} \propto B$ since both $\sigma_{xx}$ and $\sigma_{xy}$ are proportional to $B^{-1}$. This is shown with the red line in the top panel of Fig. \[fig:mag\]. For the case of finite $\Delta$ (bottom panel), the lower field region (oscillating region) behaves similarly to the massless case as explained previously. The main difference is at high fields at the quantum limit. Since $\sigma_{xx} \propto B^{-3}$ the resistivity will be $\rho_{xx} \propto B^{-1}$ at high fields. This means that after the initial increase the magnetoresistance decreases at higher magnetic fields. However, we note that these results depend on the ratio of Hall conductivity and diagonal conductivity. We use that the diagonal component is smaller than the Hall conductivity. This is an experimentally reasonable assumption also done in Ref. 29.
The magnetoresistances calculated using the numerically calculated scattering rates (corresponding to the case with $\varepsilon_r =1$ of Fig. \[fig:condxx\_kap\]) are shown with blue lines in Fig. \[fig:mag\]. In the high field region, they behave similarly to those calculated using the phenomenological scattering rates. However, the low field behavior depends on the exact number of impurities, since the $B$ dependence of the conductivity is no longer $B^{-1}$ as shown in Fig. \[fig:condxx\_kap\].
Summary and Discussions
=======================
We studied the $4\times4$ massive Dirac Hamiltonian in a constant magnetic field which can be used as a simple continuum model for the gapped Dirac semimetals. This model shows certain similarities to the $2\times2$ Weyl Hamiltonian[@Abrikosov1998; @Xiao2017] (i.e. massless case), but it contains several crucial differences.
The chemical potential was calculated implicitly fixing carrier density as a function of the magnetic field. For the gapless case we recover the result obtained in Ref. . We show that for $B\to\infty$ $\mu\to\Delta$. This behavior causes an important difference between the massive and massless case in both the scattering rate and the conductivity.
As we have seen in Sec. \[sec:gamma\] the choice of screening wavenumber in the impurity potential greatly affects the magnetic field dependence of the scattering rate. We have calculated the screening caused by the electron-electron interaction through the random phase approximation (RPA). With this for high magnetic fields we have shown that the screening wavenumber increases as $\kappa^2\propto B$ for $\Delta=0$ and $\kappa^2\propto B^2$ for $\Delta\neq0$. Using the first Born approximation we have studied the scattering rate. For high magnetic fields $\Gamma\propto B^{-1}$ for $\Delta=0$ and $\Gamma\propto B^{-2}$ for $\Delta\neq0$.
The Hall conductivity is shown to be inversely proportional to the magnetic field in the case of no impurities. In the case of a single Weyl node in Ref. Xiao et al. observed a small deviation from this behavior after the first Landau level crosses the chemical potential. In our case the symmetry in the quantum numbers and the explicit form of the eigenfunctions lead to the exact result in which $\sigma_{xy}$ behaves completely classically.
For calculating the diagonal transverse conductivity we discussed several possible magnetic field dependencies for the scattering rate. We have seen that the overall magnetoresistance is very sensitive to this choice. For high magnetic fields we show that $\sigma_{xx}\propto B^{-1}$ for $\Delta=0$ and $\sigma_{xx}\propto B^{-3}$ for $\Delta\neq0$. For lower fields at the oscillating region the massive and massless cases behave very similarly. It is shown that if the scattering rate is proportional to the magnetic field in this region, the conductivity becomes inversely proportional to $B$. We have seen that the temperature dependence is not so relevant. It decreases the SdH oscillations but it does not affect the overall magnetic field dependence. This is consistent with experimental results[@Narayanan2015; @Niemann2017; @He2014; @Feng2015]. The temperature dependence in the experimental results is mainly caused by the normalization using the zero field conductivity (which is strongly temperature dependent).
For the magnetoresistance we recover the linear dependence for $\Delta=0$ at high fields. For the $\Delta\neq0$ case we see a decrease in the magnetoresistance as $\varrho_{xx}\propto B^{-1}$ at high fields. The main difference comes from the bottom of the lowest Landau level. In the massless case this level is linear and gapless, while in the massive case it is quadratic and gapped. At high magnetic fields when the chemical potential is close to the bottom of the lowest Landau level the difference becomes relevant. At low fields, $\varrho_{xx}$ for the massive and massless cases behave very similarly. In addition to the SdH oscillations, $\varrho_{xx}$ is proportional to $B$ if we assume a phenomenological scattering rate as $\Gamma\propto B$. On the other hand, if we use the scattering rate calculated from the Born approximation we get $\varrho_{xx}\propto B^{1/3}$ as in Ref. . Experimentally[@Canuto1970; @Kaminker1981; @Abrikosov1998; @Klier2015a; @Xiao2017; @Klier2017a; @Wang2018; @Suetsugu2018], the results are more consistent with the phenomenological case.
In Fig. \[fig:mag\], we showed an example of the behavior of $\varrho_{xx}$. This behavior depends on the choices of the carrier density $n_e$, the mass term $\Delta$, and the ratio between $\sigma_{xx}$ and $\sigma_{xy}$ which originates from the magnitude of $\Gamma$. As we can see in Fig. \[fig:mag\] for the case of $\Delta\neq 0$, $\rho_{xx}$ changes its behavior from $\propto B$ in lower fields to $\propto B^{-1}$ in higher fields. The magnetic field at which this crossover occurs depends on the choice of $n_e$, $\Delta$, and $\Gamma$. The crossover magnetic field increases when $\Delta$ decreases. On the other hand, when $n_e$ becomes larger, the quantum limit occurs at a higher field and thus the crossover field also becomes higher. In experimental results for massive Dirac electrons only the linear behavior[@Suetsugu2018] is seen. This is consistent with our result, since the effective mass is small and the highest magnetic field used in the experiment is not high enough to get to the quantum limit.
It is natural to assume that the massive relativistic electron gas behaves similarly to the nonrelativistic electron gas at low energies. In the case of the nonrelativistic electron gas the diagonal conductivity is expected to saturate at high fields[@Abrikosov1988]. We do not see this behavior in the case of the gapped Dirac semimetal. The reason is because the two systems are similar only in the quantum limit when only the lowest Landau level is important. The nonrelativistic electron systems usually have a very high Fermi energy and the saturation is only valid when the chemical potential is high.
Finally, let us discuss the possible improvements of the present calculation. An important effect that was neglected in our formalism is the broadening of the density of states due to disorder. If we have used the impurity Green’s function self-consistently the divergent peaks in the density of states at the bottom of the Landau levels would have been suppressed. As a consequence the oscillating behavior in both the chemical potential and the screening wavelength would have been modified. Also, because of self-consistency the scattering rate and conductivity would have been affected. Since the number of impurities is assumed to be small and since we are far from the charge neutrality point, similarly to Ref. we expect that this effect will not change the qualitative behaviors described in the paper.
Using the static and long-wave limit in the RPA is the simplest way to include the electron-electron interaction for the screening. An improvement would be to take into account the frequency and momentum dependence of the polarization function and thus the screening wavenumber. The assumption of a simple screened Coulomb potential might not be sufficient to describe the effect properly.
The proper evaluation of the transport scattering rate through the vertex correction would further improve our calculation. In this paper we assumed that the approximative results obtained for the Weyl Hamiltonian[@Klier2015a] hold in our case as well. This should be revisited in greater detail.
At lower fields we see very strong oscillations in the scattering rate, which are caused by the simple Born approximation. An important improvement would be the self consistent Born approximation. This is numerically very challenging.
We thank H. Matsuura, and H. Maebashi for fruitful discussions. This work was supported by a Grant-in-Aid for Scientific Research (B) on Multiband effects in magnetic responses and transport properties” (No. 18H01162) from the MEXT of the Japanese Government.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Recently, the minimum number of reticulation events that is required to simultaneously embed a collection ${{\mathcal P}}$ of rooted binary phylogenetic trees into a so-called temporal network has been characterized in terms of cherry-picking sequences. Such a sequence is a particular ordering on the leaves of the trees in ${{\mathcal P}}$. However, it is well-known that not all collections of phylogenetic trees have a cherry-picking sequence. In this paper, we show that the problem of deciding whether or not ${{\mathcal P}}$ has a cherry-picking sequence is NP-complete for when ${{\mathcal P}}$ contains at least eight rooted binary phylogenetic trees. Moreover, we use automata theory to show that the problem can be solved in polynomial time if the number of trees in ${{\mathcal P}}$ and the number of cherries in each such tree are bounded by a constant.'
address:
- 'Department of Computer Science, University of Tübingen, Germany'
- 'Department of Computer Science, University of Auckland, New Zealand'
author:
- Janosch Döcker
- Simone Linz
title: 'On the existence of a cherry-picking sequence'
---
2P2N-3-SAT,cherry,cherry-picking sequence,Intermezzo,phylogenetic tree,temporal phylogenetic network
Introduction
============
To represent evolutionary relationships among species, phylogenetic trees have long been a powerful tool. However, as we now not only acknowledge speciation but also non-tree-like processes such as hybridization and lateral gene transfer to be driving forces in the evolution of certain groups of organisms (e.g. bacteria, plants, and fish) [@mallet16; @soucy15], phylogenetic networks become more widely used to represent ancestral histories. A phylogenetic network is a generalization of a rooted phylogenetic tree. More precisely, such a network is a rooted directed acyclic graph whose leaves are labeled [@huson10].
The following optimization problem, which is biologically relevant and mathematically challenging, motivates much of the theoretical work that has been done in reconstructing phylogenetic networks from phylogenetic trees. Given a collection ${{\mathcal P}}$ of rooted binary phylogenetic trees on a set of species such that ${{\mathcal P}}$ correctly represents the tree-like evolution of different parts of the species’ genomes, what is the smallest number of reticulation events that is required to simultaneously embed the trees in ${{\mathcal P}}$ into a phylogenetic network? Here, reticulation events are collectively referring to all non-tree-like events and they are represented by vertices in a phylogenetic network whose in-degree is at least two. Without any structural constraints on a phylogenetic network, it is well-known that ${{\mathcal P}}$ can always be embedded into such a network [@baroni05; @semple07] and, hence, the optimization problem is well-defined. Moreover, despite the problem being NP-hard [@bordewich07], even for when $|{{\mathcal P}}|=2$, several exact algorithms have been developed that, given two rooted phylogenetic trees, construct a phylogenetic network whose number of reticulation events is minimized over the space of all networks that embed both trees [@albrecht12; @chen13; @piovesan12; @wu10].
Motivated by the introduction of temporal networks [@baroni06; @moret04], which are phylogenetic networks that satisfy several time constraints, Humphries et al. [@humphries13; @humphries13a] recently investigated the special case of the aforementioned optimization problem for when one is interested in minimizing the number of reticulation events over the smaller space of all temporal networks that embed a given collection of rooted binary phylogenetic trees. More precisely, in the context of their two papers, the authors considered [*temporal networks*]{} to be phylogenetic networks that satisfy the following three constraints:
(1) speciation events occur successively,
(2) reticulation events occur instantaneously, and
(3) each non-leaf vertex has a child whose in-degree is one.
The second constraint implies that the three species that are involved in a reticulation event, i.e. the new species resulting from this event and its two distinct parents, must coexist in time. Moreover, a phylogenetic network that satisfies the third constraint (but not necessarily the first two constraints) is referred to as a [*tree-child*]{} network in the literature [@cardona12]. Intuitively, if a phylogenetic network ${{\mathcal N}}$ is temporal, then one can assign a time stamp to each of its vertices such that the following holds for each edge $(u,v)$ in ${{\mathcal N}}$. If $v$ is a reticulation, then the time stamp assigned to $u$ is the same as the time stamp assigned to $v$. Otherwise, the time stamp assigned to $v$ is strictly greater than that assigned to $u$. Baroni et al. [@baroni06] showed that it can be checked in polynomial time whether or not a given phylogenetic network satisfies the first two constraints. Humphries et al. [@humphries13] have established a new characterization to compute the minimum number of reticulation events that is needed to simultaneously embed an arbitrarily large collection ${{\mathcal P}}$ of rooted binary phylogenetic trees into a temporal network. This characterization, which is formally defined in Section \[sec:prelim\], is in terms of [*cherries*]{}, and the existence of a particular type of sequence on the leaves of the trees, called a [*cherry-picking sequence*]{}. It was shown that such a sequence for ${{\mathcal P}}$ exists if and only if the trees in ${{\mathcal P}}$ can simultaneously be embedded into a temporal network [@humphries13 Theorem 1]. Moreover, a cherry-picking sequence for ${{\mathcal P}}$ can be exploited further to compute the minimum number of reticulation events that is needed over all temporal networks. Importantly, not every collection ${{\mathcal P}}$ is guaranteed to have a solution, i.e. there may be no cherry-picking sequence for ${{\mathcal P}}$ and, hence no temporal network that embeds all trees in ${{\mathcal P}}$. It was left as an open problem by Humphries et al. [@humphries13] to analyze the computational complexity of deciding whether or not ${{\mathcal P}}$ has a cherry-picking sequence for when $|{{\mathcal P}}|=2$.
In this paper, we make progress towards this question and show that it is NP-complete to decide if ${{\mathcal P}}$ has a cherry-picking sequence for when $|{{\mathcal P}}|\geq 8$. Translated into the language of phylogenetic networks, this result directly implies that it is computationally hard to decide if a collection of at least eight rooted binary phylogenetic trees can simultaneously be embedded into a temporal network. To establish our result, we use a reduction from a variant of the [Intermezzo]{} problem [@guttmann06]. On a more positive note, we show that deciding if ${{\mathcal P}}$ has a cherry-picking sequence can be done in polynomial time if the number of trees and the number of cherries in each such tree are bounded by a constant. To this end, we explore connections between phylogenetic trees and automata theory and show how the problem at hand can be solved by using a deterministic finite automaton.
The remainder of the paper is organized as follows. The next section contains notation and terminology that is used throughout the paper. Section \[sec:intermezzo\] establishes NP-completeness of a variant of the [Intermezzo]{} problem which is then, in turn, used in Section \[sec:cps\] to show that it is NP-complete to decide if ${{\mathcal P}}$ has a cherry-picking sequence [for when $|{{\mathcal P}}|\geq 8$]{}. In Section \[sec:cherries\], we show that deciding if ${{\mathcal P}}$ has a cherry-picking sequence is polynomial-time solvable if the number of cherries in each tree and the size of ${{\mathcal P}}$ are bounded by a constant. We finish the paper with some concluding remarks in Section \[sec:con\].
Preliminaries {#sec:prelim}
=============
This section provides notation and terminology that is used in the subsequent sections. Throughout this paper, $X$ denotes a finite set.
[**Phylogenetic trees.**]{} A [*rooted binary phylogenetic $X$-tree*]{} ${{\mathcal T}}$ is a rooted tree with leaf set $X$ and, apart from the root which has degree two, all interior vertices have degree three. Furthermore, a pair of leaves $\{a,b\}$ of ${{\mathcal T}}$ is called a [*cherry*]{} if $a$ and $b$ are leaves that are adjacent to a common vertex. Note that every rooted binary phylogenetic tree has at least one cherry. We denote by $c_{{\mathcal T}}$ the number of cherries in ${{\mathcal T}}$. We now turn to a rooted binary phylogenetic tree with exactly one cherry. More precisely, we call ${{\mathcal T}}$ a [*caterpillar*]{} if $|X|=n\geq 2$ and the elements in $X$ can be ordered, say $x_1, x_2, \ldots, x_n$, so that $\{x_1, x_2\}$ is a cherry and, if $p_i$ denotes the parent of $x_i$, then, for all $i\in \{3, 4, \ldots, n\}$, we have $(p_i, p_{i-1})$ as an edge in ${{\mathcal T}}$, in which case we denote the caterpillar by $(x_1, x_2, \ldots, x_n)$. To illustrate, Figure \[fig:caterpillar\] shows the caterpillar $(D_1,D_2,\ldots,D_{|A'|})$ with cherry $\{D_1,D_2\}$. Two rooted binary phylogenetic $X$-trees ${{\mathcal T}}$ and ${{\mathcal T}}'$ are said to be [*isomorphic*]{} if the identity map on $X$ induces a graph isomorphism on the underlying trees.
[**Subtrees.**]{} Now, let ${{\mathcal T}}$ be a rooted binary phylogenetic $X$-tree, and let $X'=\{x_1,x_2,\ldots,x_k\}$ be a subset of $X$. The minimal rooted subtree of ${{\mathcal T}}$ that connects all vertices in $X'$ is denoted by ${{\mathcal T}}(X')$. Furthermore, the rooted binary phylogenetic tree obtained from ${{\mathcal T}}(X')$ by contracting all non-root degree-$2$ vertices is the [*restriction of ${{\mathcal T}}$ to $X'$*]{} and is denoted by ${{\mathcal T}}|X'$. We also write ${{\mathcal T}}[-x_1,x_2,\ldots,x_k]$ or ${{\mathcal T}}[-X']$ for short to denote ${{\mathcal T}}|(X-X')$. For a set ${{\mathcal P}}=\{{{\mathcal T}}_1,{{\mathcal T}}_2,\ldots,{{\mathcal T}}_m\}$ of rooted binary phylogenetic $X$-trees, we write ${{\mathcal P}}|X'$ (resp. ${{\mathcal P}}[-X']$) when referring to the set $\{{{\mathcal T}}_1|X',{{\mathcal T}}_2|X',\ldots,{{\mathcal T}}_m|X'\}$ (resp. $\{{{\mathcal T}}_1[-X'],{{\mathcal T}}_2[-X'],\ldots,{{\mathcal T}}_m[-X']\}$). Lastly, a rooted binary phylogenetic tree is [*pendant*]{} in ${{\mathcal T}}$ if it can be detached from ${{\mathcal T}}$ by deleting a single edge.
[**Cherry-picking sequences.**]{} Let ${{\mathcal P}}$ be a set of rooted binary phylogenetic $X$-trees with $|X|=n$. We say that an ordering of the elements in $X$, say $(x_1,x_2,\ldots,x_n)$, is a [*cherry-picking sequence*]{} for ${{\mathcal P}}$ precisely if each $x_i$ with $i\in\{1,2,\ldots,n-1\}$ labels a leaf of a cherry in each tree that is contained in ${{\mathcal P}}[-x_1,x_2,\ldots,x_{i-1}]$. Clearly, if $|{{\mathcal P}}|=1$, then ${{\mathcal P}}$ has a cherry-picking sequence. However, if $|{{\mathcal P}}|>1$, then ${{\mathcal P}}$ may or may not have a cherry-picking sequence.
We now formally state the decision problem that this paper is centered around.
[CPS-Existence]{}\
[**Instance.**]{} A collection ${{\mathcal P}}$ of rooted binary phylogenetic $X$-trees.\
[**Question.**]{} Does there exist a cherry-picking sequence for ${{\mathcal P}}$?
The significance of [CPS-Existence]{} is the problem’s equivalence to the question whether or not all trees in ${{\mathcal P}}$ can simultaneously be embedded into a rooted phylogenetic network that satisfies the three temporal constraints as alluded to in the introduction.
[**Automata and languages.**]{} Let $\Sigma$ be an alphabet. A [*language*]{} ${{\mathcal L}}$ is a subset of all possible strings (also called [*words*]{}) whose symbols are in $\Sigma$. More precisely, ${{\mathcal L}}$ is a subset of $\Sigma^*$, where the operator $*$ is the Kleene star. A [*deterministic finite automaton*]{} (or short [*automaton*]{}) is a tuple ${{\mathcal A}}= (Q, \Sigma, \delta, q_{\text{ini}}, F)$, where
(i) $Q$ is a finite set of states,
(ii) $\Sigma$ is a finite alphabet,
(iii) $\delta \colon Q \times \Sigma \rightarrow Q$ is a transition relation,
(iv) $q_{\text{ini}}$ is the initial state, and
(v) $F \subseteq Q$ are final states.
A given automaton ${{\mathcal A}}$ [*accepts*]{} a word $w = a_1a_2\ldots a_{n}$ if and only if ${{\mathcal A}}$ is in a final state after having read all symbols from left to right, i.e. $$\delta(\ldots\delta(\delta(q_{\text{ini}},a_1),a_2),\ldots a_{n})\in F.$$ The language ${{\mathcal L}}({{\mathcal A}}) \subseteq \Sigma^*$ that is [*recognized*]{} by ${{\mathcal A}}$ is defined as the set of words that ${{\mathcal A}}$ accepts. For the automata constructed in this paper, we have $|F| = 1$ and $\delta$ being a total function that maps each pair of a state in $Q$ and a symbol in $\Sigma$ to a state in $Q$. For a detailed introduction to automata theory and languages, see the book by Hopcroft and Ullman [@hopcroft79].
A variant of the [Intermezzo]{} problem {#sec:intermezzo}
=======================================
In this section, we establish NP-completeness of a variant of the ordering problem [Intermezzo]{}. Let $A$ be a finite set, and let ${{\mathcal O}}$ be an ordering on the elements in $A$. For two elements $a$ and $b$ in $A$, we write $a<b$ precisely if $a$ precedes $b$ in ${{\mathcal O}}$. With this notation in hand, we now formally state [Intermezzo]{} which was shown to be NP-complete via reduction from [3-SAT]{} [@guttmann06 Lemma 1].
[Intermezzo]{}\
[**Instance.**]{} A finite set $A$, a collection $B$ of pairs from $A$, and a collection $C$ of pairwise-disjoint triples of distinct elements in $A$.\
[**Question.**]{} Does there exist a total linear ordering on the elements in $A$ such that $a_i<a_j$ for each $(a_i,a_j)$ in $B$, and $a_i<a_j<a_k$ or $a_j<a_k<a_i$ for each $(a_i,a_j,a_k)$ in $C$?
[**Example.**]{} Consider the following instance of [Intermezzo]{} with three pairs and two disjoint triples (when viewed as sets): $$\begin{aligned}
A &= \{a_1,\, a_2,\, a_3,\, a_4,\, a_5, a_6\}, \\
B &= \{(a_1,\, a_6),\, (a_4,\, a_1),\, (a_4,\, a_3)\}, \\
C &= \{(a_1,\,a_2,\,a_3),\,(a_4,\,a_5,\,a_6)\}.\end{aligned}$$ A total linear ordering on the elements in $A$ that satisfies all constraints defined by $B$ and $C$ is $$\mathcal{O} = (a_2,\, a_4,\, a_3,\, a_1,\, a_5,\, a_6).$$
While each element $a_i\in A$ can appear an unbounded number of times in the input of a given [Intermezzo]{} instance, this number is bounded from above by $N$ in the following [Intermezzo]{} variant.
[$N$-Disjoint-Intermezzo]{}\
[**Instance.**]{} A finite set $A$, collections $B_1,B_2,\ldots,B_N$ of pairs from $A$, and collections $C_1,C_2,\ldots,C_N$ of triples of distinct elements in $A$ such that, for each $\ell \in\{1,2,\ldots,N\}$, the elements in $B_\ell\cup C_\ell$ are pairwise disjoint.\
[**Question.**]{} Does there exist a total linear ordering on the elements in $A$ such that $$a_i<a_j\textnormal{ for each }(a_i,a_j)\in \bigcup_{1 \leq \ell \leq N} B_\ell,$$ and $$a_i<a_j<a_k \textnormal{ or } a_j<a_k<a_i\textnormal{ for each }(a_i,a_j,a_k)\in \bigcup_{1 \leq \ell \leq N} C_\ell\rm{?}$$\
Let $I$ be an instance of [$N$-Disjoint-Intermezzo]{}, and let ${{\mathcal O}}$ be an ordering on the elements of $A$ that satisfies the two ordering constraints for each pair and triple in the statement of [$N$-Disjoint-Intermezzo]{}. We say that ${{\mathcal O}}$ is an [*[$N$-Disjoint-Intermezzo]{} ordering*]{} for $I$. We next show that [$4$-Disjoint-Intermezzo]{} is NP-complete via reduction from the following restricted version of [3-SAT]{}.
[2P2N-3-SAT]{}\
[**Instance.**]{} A set $U$ of variables, and a set ${{\mathcal C}}$ of clauses, where each clause is a disjunction of exactly three literals, such that each variable appears negated exactly twice and unnegated exactly twice in ${{\mathcal C}}$.\
[**Question.**]{} Does there exist a truth assignment for $U$ that satisfies each clause in ${{\mathcal C}}$?
Berman et al. [@berman03 Theorem 1] established NP-completeness for [2P2N-3-SAT]{}.
\[t:4-DI\] [4-Disjoint-Intermezzo]{} is [*NP*]{}-complete.
We show that the construction by Guttmann and Maucher [@guttmann06 Lemma 1], that was used to show that [Intermezzo]{} is NP-complete via reduction from [3-SAT]{}, yields an instance of [4-Disjoint-Intermezzo]{} if we reduce from [2P2N-3-SAT]{}. Using the same notation as Guttmann and Maucher [@guttmann06 Lemma 1], their construction is as follows. Let $I$ be an instance of [2P2N-3-SAT]{} that is given by a set of variables $U = \{u_1,\,\ldots,\,u_n\}$ and a set of clauses $${{\mathcal C}}=\{(c_{1,1} \vee c_{1,2} \vee c_{1,3}),\,\ldots,\,(c_{m,1} \vee c_{m,2} \vee c_{m,3})\},$$ where each $c_{i,j} \in \{u_1, \bar{u}_1,u_2, \bar{u}_2,\ldots,u_n, \bar{u}_n\}$. Furthermore, for $a,b\in\mathbb{N}$, let $a \oplus b$ denote the number $c \in \{1,\,2,\,3\}$ such that $a + b \equiv c \pmod 3$. We define the following three sets: $$\begin{aligned}
A &= \{u_{k,l},\,\bar{u}_{k,l} \mid 1 \leq k \leq n \wedge 1 \leq l \leq 3\} \cup {}\\
&\hphantom{{} = {}}\{c_{i,j}^l \mid 1 \leq i \leq m \wedge 1 \leq j \leq 3 \wedge 1 \leq l \leq 3\},\\[5pt]
B &= \{(u_{k,1},\,\bar{u}_{k,3}),\,(\bar{u}_{k,1},\,u_{k,3}) \mid 1 \leq k \leq n\} \cup {}\\
&\hphantom{{} = {}}\{(c_{i,j,2},\,c_{i,j}^{1}),\,(c_{i,j}^2,\,c_{i,j,1}) \mid 1 \leq i \leq m \wedge 1 \leq j \leq 3\} \cup {}\\
&\hphantom{{} = {}}\{(c_{i,j \oplus 1}^1,\,c_{i,j}^3) \mid 1 \leq i \leq m \wedge 1 \leq j \leq 3\},\\[5pt]
C &= \{(u_{k,1},\,u_{k,2},\,u_{k,3}),\,(\bar{u}_{k,1},\,\bar{u}_{k,2},\,\bar{u}_{k,3}) \mid 1 \leq k \leq n\} \cup {}\\
&\hphantom{{} = {}}\{(c_{i,j}^1,\,c_{i,j}^2,\,c_{i,j}^3) \mid 1 \leq i \leq m \wedge 1 \leq j \leq 3\},\end{aligned}$$ where $c_{i,j,l}$ is an abbreviation of $u_{k,l}$ with $u_k = c_{i,j}$. By construction, the elements in $C$ are pairwise-disjoint triples of distinct elements in $A$ and, so, the three sets $A$, $B$, and $C$ form an instance of [Intermezzo]{}.
Now, we show how the pairs and triples in $B \cup C$ can be partitioned into sets $B_\ell \cup C_\ell$ with $B_\ell \subseteq
B$, $C_\ell \subseteq C$, and $1 \leq \ell \leq 4$ such that the elements in $B_\ell\cup C_\ell$ are pairwise disjoint. Recalling that $C$ is a set of pairwise-disjoint triples, we start by setting $B_1 = \emptyset$ and $C_1 = C$. Furthermore, we set $$\begin{aligned}
B_2 &= \{(u_{k,1},\,\bar{u}_{k,3}),\,(\bar{u}_{k,1},\,u_{k,3}) \mid 1 \leq k \leq n\} \cup {}\\
&\hphantom{{} = {}}\{(c_{i,j \oplus 1}^1,\,c_{i,j}^3) \mid 1 \leq i \leq m \wedge 1 \leq j \leq 3\}\end{aligned}$$ and $C_2 = \emptyset$. By construction, it is easy to check that the pairs in $B_2$ are pairwise disjoint. Lastly, consider the remaining pairs $$B \setminus B_2 = \{(c_{i,j,2}, c_{i,j}^1), (c_{i,j}^2, c_{i,j,1}) \mid 1 \leq i \leq m \wedge 1 \leq j \leq 3\}$$ and observe that the only possibility for two pairs in $B \setminus B_2$ to have a non-empty intersection is to have an element $c_{i,j,l}$ with $l\in\{1,2\}$ in common. Now, since each $c_{i,j,l}$ is equal to an element in $$U'=\{u_{k,l}, \bar{u}_{k,l}\mid 1 \leq k \leq n\wedge 1 \leq l \leq 3\},$$ and each element $u_k$ appears exactly twice negated and twice unnegated in ${{\mathcal C}}$, it follows that there is a partition of $B\setminus B_2$ into $B_3$ and $B_4$ so that all pairs in the resulting two sets are pairwise disjoint. Setting $C_3 = C_4 = \emptyset$ completes the construction of an instance of [4-Disjoint-Intermezzo]{}. Noting that it is straightforward to compute the partition $$B \cup C = \bigcup_{1 \leq \ell \leq 4} \left ( B_\ell \cup C_\ell \right )$$ in polynomial time and that we did not modify the construction described by Guttmann and Maucher [@guttmann06 Lemma 1] itself, it follows from the same proof that $I$ has a satisfying truth assignment if and only if $\bigcup_{1 \leq \ell \leq 4} \left ( B_\ell \cup C_\ell \right )$ has a [4-Disjoint-Intermezzo]{} ordering.
[**Remark.**]{} By the construction of an instance of [4-Disjoint-Intermezzo]{} in the proof of Theorem \[t:4-DI\], we note that no pair or triple occurs twice and that, for each $\ell\in\{1,2,3,4\}$, we have $B_\ell\cup C_\ell\ne\emptyset$. We will freely use these facts throughout the remainder of the paper.
Hardness of [CPS-Existence]{} {#sec:cps}
=============================
In this section, we show that the decision problem [CPS-Existence]{} is NP-complete for any collection of rooted binary phylogenetic trees on the same leaf set that consists of a constant number $m$ of trees with $m\geq 8$. To establish the result, we use a reduction from [4-Disjoint-Intermezzo]{}.
![A caterpillar on $|A'|$ leaves and with cherry $\{D_1,D_2\}$.[]{data-label="fig:caterpillar"}](CaterpillarSkeleton){width=".35\textwidth"}
Let $I$ be an instance of [4-Disjoint-Intermezzo]{}. Using the same notation as in the definition of [$N$-Disjoint-Intermezzo]{}, let $$A'=A\cup\left\{c_r^1,c_r^2,c_r^3,c_r^4\mid c_r\in\bigcup_{1 \leq \ell \leq 4} C_\ell\right\},$$ and let $D=\{d_1,d_2,\ldots,d_{|A'|}\}$. For each $\ell\in\{1,2,3,4\}$, we next construct two rooted binary phylogenetic trees. Let $A_\ell$ be the subset of $A'$ that precisely contains each element of $A'$ that is neither contained in an element of $B_\ell$ nor contained in an element of $$C_\ell\cup \{c_r^1,c_r^2,\ldots,c_r^4 \mid c_r\in C_\ell\}.$$ Furthermore, let ${{\mathcal S}}_\ell$ and ${{\mathcal S}}_\ell'$ both be the caterpillar shown in Figure \[fig:caterpillar\]. Setting $q=1$, let ${{\mathcal T}}_\ell$ and ${{\mathcal T}}_\ell'$ be the two rooted binary phylogenetic trees obtained from ${{\mathcal S}}_\ell$ and ${{\mathcal S}}_\ell'$ that result from the following four-step process.
(i) For each $(a_i,a_j)\in B_\ell$ in turn, replace the leaf $D_q$ in ${{\mathcal S}}_\ell$ (resp. ${{\mathcal S}}_\ell'$) with the 3-taxon tree on the top left (resp. bottom left) in Figure \[fig:gadgets\] and increment $q$ by one.
(ii) For each $c_r\in C_\ell$ with $c_r=(a_i,a_j,a_k)$ in turn, replace the leaf $D_q$ in ${{\mathcal S}}_\ell$ (resp. ${{\mathcal S}}_\ell'$) with the 8-taxon tree on the top right (resp. bottom right) in Figure \[fig:gadgets\] and increment $q$ by one.
(iii) For each $a_i\in A_\ell$ in turn, replace the leaf $D_q$ in ${{\mathcal S}}_\ell$ and ${{\mathcal S}}_\ell'$ with the cherry $\{a_i,d_q\}$ and increment $q$ by one.
(iv) For each element in $\{q,q+1,\ldots,|A'|\}$, replace the leaf label $D_q$ in ${{\mathcal S}}_\ell$ and ${{\mathcal S}}_\ell'$ with $d_q$.
![Gadgets for a pair $(a_i,a_j)$ (left) and gadgets for a triple $(a_i,a_j,a_k)$ (right) that are used in the reduction from [4-Disjoint-Intermezzo]{} to [CPS-Existence]{}.[]{data-label="fig:gadgets"}](gadgets){width=".6\textwidth"}
We call ${{\mathcal P}}_I=\{{{\mathcal T}}_\ell,{{\mathcal T}}_\ell' \mid 1 \leq \ell \leq 4\}$ the set of [*intermezzo trees*]{} associated with $I$. The next observation is an immediate consequence from the above construction and the fact that, for each $1 \leq \ell \leq 4$, the elements in $B_\ell$ and $C_\ell$ are pairwise disjoint.
For an instance $I$ of [4-Disjoint-Intermezzo]{}, the set of intermezzo trees associated with $I$ consists of eight pairwise non-isomorphic rooted binary phylogenetic trees whose set of leaves is $A'\cup D$.
We now establish the main result of this section.
\[lem:CPS-Existence\] Let ${{\mathcal P}}=\{{{\mathcal T}}_1,{{\mathcal T}}_2,\ldots,{{\mathcal T}}_m\}$ be a collection of rooted binary phylogenetic $X$-trees. [CPS-Existence]{} is [NP]{}-complete for $m=8$.
Clearly, [CPS-Existence]{} for $m=8$ is in NP because, given an ordering ${{\mathcal O}}$ on the elements in $X$, we can decide in polynomial time if ${{\mathcal O}}$ is a cherry-picking sequence for ${{\mathcal P}}$. Let $I$ be an instance of [4-Disjoint-Intermezzo]{}, and let ${{\mathcal P}}_I=\{{{\mathcal T}}_\ell,{{\mathcal T}}_\ell'\mid 1 \leq \ell \leq 4\}$ be the set of eight intermezzo trees that are associated with $I$. Note that each tree in ${{\mathcal P}}_I$ can be constructed in polynomial time and has a size that is polynomial in $|A|$. The remainder of the proof essentially consists of establishing the following claim.
[**Claim.**]{} $I$ is a ‘yes’-instance of [4-Disjoint-Intermezzo]{} if and only if ${{\mathcal P}}_I$ has a cherry-picking sequence.
First, suppose that ${{\mathcal P}}_I$ has a cherry-picking sequence. Let ${{\mathcal O}}$ be a cherry-picking sequence for ${{\mathcal P}}_I$, and let ${{\mathcal O}}'$ be the subsequence of ${{\mathcal O}}$ of length $|A|$ that contains each element in $A$. We next show that ${{\mathcal O}}'$ is a [$4$-Disjoint-Intermezzo]{} ordering for $I$. Let $(a_i,a_j)$ be an element of some $B_\ell$ with $1 \leq \ell \leq 4$, and let $d_q$, with $q\in\{1,2,\ldots,|A'|\}$, be the unique leaf label of ${{\mathcal T}}_\ell$ and ${{\mathcal T}}_\ell'$ such that $\{a_i,a_j,d_q\}$ is the leaf set of a pendant subtree of ${{\mathcal T}}_\ell$ and ${{\mathcal T}}_\ell'$. By construction of ${{\mathcal T}}_\ell$ and ${{\mathcal T}}_\ell'$, it is easily seen that $d_q$ exists and $a_i<a_j$ in ${{\mathcal O}}$. Hence, $a_i<a_j$ in ${{\mathcal O}}'$. Turning to the triples, let $c_r=(a_i,a_j,a_k)$ be an element of some $C_\ell$ with $1 \leq \ell \leq 4$, and let $d_q$, with $q\in\{1,2,\ldots,|A'|\}$, be the unique leaf label of ${{\mathcal T}}_\ell$ and ${{\mathcal T}}_\ell'$ such that $\{a_i,a_j,a_k,c_r^1,c_r^2,c_r^3,c_r^4,d_q\}$ is the leaf set of a pendant subtree of ${{\mathcal T}}_\ell$ and ${{\mathcal T}}_\ell'$. Again, by construction, $d_q$ exists. Let ${{\mathcal S}}_\ell={{\mathcal T}}_\ell|\{a_i,a_j,a_k,c_r^1,c_r^2,c_r^3,c_r^4,d_q\}$ and, similarly, let ${{\mathcal S}}_\ell'={{\mathcal T}}_\ell'|\{a_i,a_j,a_k,c_r^1,c_r^2,c_r^3,c_r^4,d_q\}$. It is straightforward to check that each cherry-picking sequence for ${{\mathcal S}}_\ell$ and ${{\mathcal S}}_\ell'$ satisfies either $$a_i<a_j<a_k,\textnormal{ or }a_j<a_k<a_i.$$ Hence, as ${{\mathcal S}}_\ell$ and ${{\mathcal S}}_\ell'$ are pendant in ${{\mathcal T}}_\ell$ and ${{\mathcal T}}_\ell'$, respectively, we have $a_i<a_j<a_k$, or $a_j<a_k<a_i$ in ${{\mathcal O}}$ and, consequently, in ${{\mathcal O}}'$. Since the above argument holds for each pair and each triple, it follows that ${{\mathcal O}}'$ is a [4-Disjoint-Intermezzo]{} ordering for $I$ and, so, $I$ is a ‘yes’-instance.
Conversely, suppose that $I$ is a ‘yes’-instance of [4-Disjoint-Intermezzo]{}. Let ${{\mathcal O}}'$ be a [4-Disjoint-Intermezzo]{} ordering on the elements of $A$. To ease reading, let $$C=\bigcup_{1 \leq \ell \leq 4}C_\ell.$$ Modify ${{\mathcal O}}'$ as follows to obtain an ordering ${{\mathcal O}}$.
(1) Concatenate ${{\mathcal O}}'$ with the sequence $(d_1,d_2,\ldots,d_{|A'|})$.
(2) For each $c_r=(a_i,a_j,a_k)$ in $C$, do one of the following two depending on the order of $a_i$, $a_j$, and $a_k$ in ${{\mathcal O}}'$. If $a_i<a_j<a_k$ in ${{\mathcal O}}'$, then replace $a_i$ with $a_i,c_r^2$ and replace $a_k$ with $a_k,c_r^3,c_r^1,c_r^4$. Otherwise, if $a_j<a_k<a_i$, replace $a_k$ with $a_k,c_r^3$ and replace $a_i$ with $a_i,c_r^2,c_r^1,c_r^4$.
Since ${{\mathcal O}}'$ is a [4-Disjoint-Intermezzo]{} ordering with $a_i<a_j<a_k$ or $a_j<a_k<a_i$ for each $(a_i,a_j,a_k)\in C$, it follows from the construction of ${{\mathcal O}}$ from ${{\mathcal O}}'$ that ${{\mathcal O}}$ is an ordering on the elements in $A'\cup D$. It remains to show that ${{\mathcal O}}$ is a cherry-picking sequence for ${{\mathcal P}}_I$. First, consider a pendant subtree with leaf set $\{a_i,a_j,d_q\}$ in ${{\mathcal T}}_\ell$ and ${{\mathcal T}}_\ell'$ for some $1 \leq \ell \leq 4$. By construction, $(a_i,a_j)$ is a pair in $B_\ell$ and, so, we have $a_i<a_j$ in ${{\mathcal O}}'$ and $a_i<a_j<d_q$ in ${{\mathcal O}}$. Second, consider a pendant subtree with leaf set $\{a_i,a_j,a_k,c_r^1,c_r^2,c_r^3,c_r^4,d_q\}$ in ${{\mathcal T}}_\ell$ and ${{\mathcal T}}_\ell'$ for some $1 \leq \ell \leq 4$. By construction, $(a_i,a_j,a_k)$ is a triple in $C_\ell$ and, so, we have either $a_i<a_j<a_k$ in ${{\mathcal O}}'$ and $$a_i<c_r^2<a_j<a_k<c_r^3<c_r^1<c_r^4<d_q$$ in ${{\mathcal O}}$, or $a_j<a_k<a_i$ in ${{\mathcal O}}'$ and $$a_j<a_k<c_r^3<a_i<c_r^2<c_r^1<c_r^4<d_q$$ in ${{\mathcal O}}$. Third, consider a pendant subtree with leaf set $\{a_i,d_q\}$ in ${{\mathcal T}}_\ell$ and ${{\mathcal T}}_\ell'$ for some $1 \leq \ell \leq 4$. By construction, we have $a_i<d_q$ in ${{\mathcal O}}$. Fourth, if $(a_i,a_j,a_k)\in C$, then, as $I$ has a [4-Disjoint-Intermezzo]{} ordering, there does not exist a pair $(a_k,a_j)$ in $B_\ell$ for some $1 \leq \ell \leq 4$. Lastly, observe that $(d_1,d_2,\ldots,d_{|A'|})$ is a suffix of ${{\mathcal O}}$ and that, for any two trees, say ${{\mathcal S}}$ and ${{\mathcal S}}'$ in ${{\mathcal P}}_I$, we have that ${{\mathcal S}}|D$ and ${{\mathcal S}}'|D$ are isomorphic. Since ${{\mathcal O}}'$ is a [4-Disjoint-Intermezzo]{} ordering, it is now straightforward to check that ${{\mathcal O}}$ is a cherry-picking sequence of ${{\mathcal P}}_I$. This establishes the proof of the claim and, thereby, the theorem.
The next corollary shows that [CPS-Existence]{} is not only NP-complete for a collection of eight rooted binary phylogenetic trees on the same leaf set, but for any such collection with a fixed number $m$ of trees with $m\geq 8$.
Let ${{\mathcal P}}=\{{{\mathcal T}}_1,{{\mathcal T}}_2,\ldots,{{\mathcal T}}_m\}$ be a collection of rooted binary phylogenetic $X$-trees. [CPS-Existence]{} is [NP]{}-complete for any fixed $m$ with $m\geq 8$.
Clearly, [CPS-Existence]{} for $m= t+8$ with $t \geq 0$ is in NP. To establish the corollary, we show how one can modify the reduction that is described prior to Theorem \[lem:CPS-Existence\] to obtain a set of $t+8$ rooted binary phylogenetic trees from an instance of [4-Disjoint-Intermezzo]{}. Let $I$ be an instance of [4-Disjoint-Intermezzo]{}. Throughout the remainder of the proof, we assume that there exists an $1 \leq \ell \leq 4$ such that $|B_\ell \cup C_\ell| > t$. Otherwise, since $t = m - 8$ and $m$ is fixed, it follows that $I$ has a constant number $c$ of pairs and triples with $c\leq 4(m-8)$ and is solvable in polynomial time.
Now, let $B_i$ and $C_i$ with $i \in \{1,\,2,\,3,\,4\}$ be a collection of pairs and triples, respectively, such that $|B_i \cup C_i| > t$. Theorem \[lem:CPS-Existence\] establishes the result for when $t=0$. We may therefore assume that $t>0$ and consider two cases. First, suppose that $t$ is even. Replace $B_i$ and $C_i$ in $I$ with a partition of $B_i \cup C_i$ into $\frac{t}{2} + 1$ sets. Each of the resulting new sets can be split naturally into a collection of pairs and a collection of triples of which at most one is empty. This results in $$\left (\frac t 2+1\right )+(4-1)=\frac t 2 +4$$ collections of pairs and triples, respectively. Now, for each $B_\ell$ and $C_\ell$ with $\ell\in\{1,2,\ldots,\frac t 2+4\}$, construct two rooted binary phylogenetic trees as described in the definition of the set of intermezzo trees associated with $I$. This yields $$2\left (\frac {t} 2 +4\right )=t+8=m$$ pairwise non-isomorphic trees. Second, suppose that $t$ is odd. Replace $B_i$ and $C_i$ in $I$ with a partition of $B_i \cup C_i$ into $\frac{t-1}{2} + 1$ sets. Additionally, add $B_*=\emptyset$ and $C_*=\emptyset$. Analogous to the first case, this results in $$\left (\frac {t-1} 2+1+1\right )+(4-1)=\frac {t-1} 2 +5$$ collections of pairs and triples, respectively. Again, for each $B_\ell$ and $C_\ell$ with $\ell\in\{1,2,\ldots,\frac {t-1} 2+5\}$ construct two rooted binary phylogenetic trees as described in the definition of the set of intermezzo trees associated with $I$. Noting that the two trees for $B_*$ and $C_*$ are isomorphic, it follows that the construction yields $$2\left (\frac {t-1} 2 +5\right )-1=\frac{2t-2}2+10-1=t+8=m$$ pairwise non-isomorphic trees. Since the proof of Theorem \[lem:CPS-Existence\] generalizes to a set of $m$ intermezzo trees, the corollary now follows for both cases.
Bounding the number of cherries {#sec:cherries}
===============================
The main result of this section is the following theorem.
\[t:cherry\] Let ${{\mathcal P}}=\{{{\mathcal T}}_1,{{\mathcal T}}_2,\ldots,{{\mathcal T}}_m\}$ be a collection of rooted binary phylogenetic $X$-trees. Let $c$ be the maximum element in $\{c_{{{\mathcal T}}_1},c_{{{\mathcal T}}_2},\ldots,c_{{{\mathcal T}}_m}\}$. Then solving [CPS-Existence]{} for ${{\mathcal P}}$ takes time $$O\left (|X|^{m(4c-2)+1} + \sum_{i=1}^m f_i(|X|, c_{{{\mathcal T}}_i})\right ),$$ where $f_i(|X|, c_{{{\mathcal T}}_i}) \in |X|^{O(c_{{{\mathcal T}}_i})}$. In particular, the running time is polynomial in $|X|$ if $c$ and $m$ are constant.
Let ${{\mathcal T}}$ be a rooted binary phylogenetic $X$-tree. We denote by ${{\mathcal C}}({{\mathcal T}})$ the recursively defined set of trees that contains ${{\mathcal T}}$ and $\emptyset$, and that satisfies the following property.
[**(P)**]{} If a tree ${{\mathcal T}}'$ is in ${{\mathcal C}}({{\mathcal T}})$ and $\{a,b\}$ is a cherry in ${{\mathcal T}}'$, then ${{\mathcal T}}'[-a]$ and ${{\mathcal T}}'[-b]$ are also contained in ${{\mathcal C}}({{\mathcal T}})$.
We refer to ${{\mathcal C}}({{\mathcal T}})$ as the [*set of cherry-picked trees*]{} of ${{\mathcal T}}$. Intuitively, ${{\mathcal C}}({{\mathcal T}})$ contains each tree that can be obtained from ${{\mathcal T}}$ by repeatedly deleting a leaf of a cherry.
To establish Theorem \[t:cherry\], we consider the set ${{\mathcal C}}({{\mathcal T}})$ of cherry-picked trees of ${{\mathcal T}}$. First, we develop a new vector representation for each tree in ${{\mathcal C}}({{\mathcal T}})$ and show that the size of ${{\mathcal C}}({{\mathcal T}})$ is at most $(|X|+1)^{O(c_{{\mathcal T}})}$. We then construct an automaton whose number of states is $|{{\mathcal C}}({{\mathcal T}})|+1$ and that recognizes whether or not a word that contains each element in $X$ precisely once is a cherry-picking sequence for ${{\mathcal T}}$. Lastly, we show how to use a product automaton construction to solve [CPS-Existence]{} for a set of rooted binary phylogenetic $X$-trees in time that is polynomial if the number of cherries and the number of trees in ${{\mathcal P}}$ is bounded by a constant.
We start with a simple lemma, which shows that deleting a leaf of a cherry never increases the number of cherries.
\[lem:cherries\] Let ${{\mathcal T}}$ be a rooted binary phylogenetic $X$-tree, and let $a$ be an element of a cherry in ${{\mathcal T}}$. Then, $$c_{{{\mathcal T}}}-c_{{{\mathcal T}}[-a]}\in\{0,1\}.$$
Let $b$ be the unique element in $X$ such that $\{a,b\}$ is a cherry in ${{\mathcal T}}$. Observe that each cherry of ${{\mathcal T}}$ other than $\{a,b\}$ is also a cherry of ${{\mathcal T}}[-a]$. Now, let $p$ be the parent of the parent of $a$ in ${{\mathcal T}}$, and let $c$ be the child of $p$ that is not the parent of $a$. If $c$ is a leaf, then it is easily checked that $\{b,c\}$ is a cherry in ${{\mathcal T}}[-a]$ and, so $c_{{{\mathcal T}}}-c_{{{\mathcal T}}[-a]}=0$. On the other hand, if $c$ is not a leaf, then $b$ is not part of a cherry in ${{\mathcal T}}[-a]$ and, so, $c_{{{\mathcal T}}}-c_{{{\mathcal T}}[-a]}=1$.
We now define a labeled tree that will play an important role throughout the remainder of this section. Let ${{\mathcal T}}$ be a rooted binary phylogenetic $X$-tree with cherries $\{\{a_1,b_1\},\{a_2,b_2\},\ldots,\{a_{c_{{\mathcal T}}},b_{c_{{\mathcal T}}}\}\}$. Obtain a tree ${{\mathcal T}}_{{\mathcal I}}$ from ${{\mathcal T}}$ as follows.
1. Set ${{\mathcal T}}_\mathcal{I}$ to be ${{\mathcal T}}$.
2. Delete all leaves of ${{\mathcal T}}_\mathcal{I}$ that are not part of a cherry.
3. Suppress any resulting degree-2 vertex.
4. If the root, say $\rho$, has degree one, delete $\rho$.
5. For each cherry $\{a_i, b_i\}$ with $i\in\{1,2,\ldots,c_{{\mathcal T}}\}$, label the parent of $a_i$ and $b_i$ with $i$, and delete the two leaves $a_i$ and $b_i$.
6. Bijectively label the non-leaf vertices of ${{\mathcal T}}_\mathcal{I}$ with $c_{{\mathcal T}}+1,c_{{\mathcal T}}+2,\ldots,2c_{{\mathcal T}}-1$.
We call ${{\mathcal T}}_{{\mathcal I}}$ the [*index tree*]{} of ${{\mathcal T}}$. By construction, ${{\mathcal T}}_{{\mathcal I}}$ is a labeled rooted binary tree that is unique up to relabeling the internal vertices. To illustrate, an example of the construction of an index tree is shown in Figure \[fig:index\]. The next observation follows immediately from the construction of an index tree.
Let ${{\mathcal T}}$ be a rooted binary phylogenetic tree, and let ${{\mathcal T}}_{{\mathcal I}}$ be the index tree associated with ${{\mathcal T}}$. The size of ${{\mathcal T}}_{{\mathcal I}}$ is $O(c_{{\mathcal T}})$. In particular, if the number of cherries in ${{\mathcal T}}$ is constant, the size of ${{\mathcal T}}_{{\mathcal I}}$ is $O(1)$.
![An example of the construction of an index tree. Steps (1) to (6) refer to the corresponding steps in the definition of an index tree. For simplicity, in Step (1), we have only indicated the leaf labels of leaves that are part of a cherry.[]{data-label="fig:index"}](indexTree){width="80.00000%"}
We next define a particular vector relative to a given set. Let $S$ be a finite set, let $\epsilon$ be an element that is not in $S$, and let $n$ be a non-negative integer. We call $$\begin{aligned}
v=&(&[\xi_1](x_{1}^1,x_{1}^2,\ldots,x_{1}^{q_1},\epsilon),\nonumber\\
&&[\xi_2](x_{2}^1,x_{2}^2,\ldots,x_{2}^{q_2},\epsilon),\nonumber\\
&&\vdots \nonumber\\
&&[\xi_n](x_n^1,x_n^2,\ldots,x_n^{q_n},\epsilon)\hspace{0.3cm})\nonumber\end{aligned}$$ an [*$S$-vector*]{} if each element in $S$ appears at most once in $v$, each $\xi_i$ is an element in $S\cup\{\epsilon\}$, and each $x_i^j$ is an element in $S$. Now consider the following two $S$-vectors: $$\begin{aligned}
v=&(&[\xi_1](x_{1}^1,x_{1}^2,\ldots,x_{1}^{q_1},\epsilon),\nonumber\\
&&[\xi_2](x_{2}^1,x_{2}^2,\ldots,x_{2}^{q_2},\epsilon),\nonumber\\
&&\vdots \nonumber\\
&&[\xi_n](x_n^1,x_n^2,\ldots,x_n^{q_n},\epsilon)\hspace{0.3cm})\nonumber\end{aligned}$$ and $$\begin{aligned}
v'=&(&[\psi_1](y_{1}^1,y_{1}^2,\ldots,y_{1}^{r_1},\epsilon),\nonumber\\
&&[\psi_2](y_{2}^1,y_{2}^2,\ldots,y_{2}^{r_2},\epsilon),\nonumber\\
&&\vdots \nonumber\\
&&[\psi_n](y_n^1,y_n^2,\ldots,y_n^{r_n},\epsilon)\hspace{0.3cm})\nonumber\end{aligned}$$ We say that $v'$ has the [*suffix-property*]{} relative to $v$ if, for each $s\in\{1,2,\ldots,n\}$, the vector component $[\psi_s](y_{s}^1,y_{s}^2,\ldots,y_{s}^{r_s},\epsilon)$ is equal to $[\psi_s](\epsilon)$ or satisfies each of the following equations $$y_{s}^{r_s} = x_{s}^{q_s}, \; y_{s}^{r_s-1} = x_{s}^{q_s-1}, \; \ldots, \; y_{s}^{1} = x_{s}^{q_s-r_s+1}.$$ Lastly, if $v'$ has the property that $[\psi_i](y_{i}^1,y_{i}^2,\ldots,y_{i}^{r_i},\epsilon)=[\epsilon](\epsilon)$ for each $i\in\{1,2,\ldots,n\}$, we call $v'$ the [*empty vector*]{}. Note that the empty vector satisfies the suffix-property relative to every $S$-vector.
Building on the definition of an $S$-vector, we now describe a vector representation of a rooted binary phylogenetic tree that can be constructed by using its index tree as a guide. Roughly, the representation associates a caterpillar-type structure to each vertex in the index tree. Let ${{\mathcal T}}$ be a rooted binary phylogenetic $X$-tree, let $X'\subseteq X$, and let $\epsilon\notin X$. For two vertices $u$ and $v$ in ${{\mathcal T}}$, we say that $u$ (resp. $v$) is an [*ancestor*]{} (resp. [*descendant*]{}) of $v$ (resp. $u$) if there is a directed path from $u$ to $v$ in ${{\mathcal T}}$. Throughout this section, we regard a vertex $v$ of ${{\mathcal T}}$ to be an ancestor and a descendant of itself. The [*most recent common ancestor*]{} of $X'$ is the vertex $v$ in ${{\mathcal T}}$ whose set of descendants contains $X'$ and no descendant of $v$, except $v$ itself, has this property. We denote $v$ by ${{\rm mrca}}_{{{\mathcal T}}}(X')$. Now, let $\{\{a_1,b_1\},\{a_2,b_2\},\ldots,\{a_{c_{{\mathcal T}}},b_{c_{{\mathcal T}}}\}\}$ be the set of all cherries in ${{\mathcal T}}$. First, for each leaf $i\in\{1,2,\ldots,c_{{\mathcal T}}\}$ in ${{\mathcal T}}_{{\mathcal I}}$, let $(a_i,x_{i}^1,x_{i}^2,\ldots,x_{i}^q)$ be the maximal pendant caterpillar in ${{\mathcal T}}$ with cherry $\{a_i,b_i\}$. We denote this by $$[\xi_i](x_{i}^1,x_{i}^2,\ldots,x_{i}^q,\epsilon),$$ where $\xi_i=a_i$ and $x_{i}^1=b_i$. Second, for each non-leaf vertex labeled $i$ in ${{\mathcal T}}_{{\mathcal I}}$ with $i\in\{c_{{\mathcal T}}+1,c_{{\mathcal T}}+2,\ldots,2c_{{\mathcal T}}-1\}$, let $v_i$ be the vertex in ${{\mathcal T}}$ such that $$v_i={\rm mrca}_{{\mathcal T}}(\{a_j,b_j\mid j \textnormal{ is a descendant of } i \textnormal{ in } {{\mathcal T}}_{{\mathcal I}}\}),$$ and let ${{\mathcal T}}_i$ be the rooted binary phylogenetic tree obtained from ${{\mathcal T}}$ by replacing the pendant subtree rooted at $v_i$ with a leaf labeled $v_i$. Now, if $v_i$ is a leaf of a cherry in ${{\mathcal T}}_i$, let $(v_i,x_{i}^1,x_{i}^2,\ldots,x_{i}^q)$ be the maximal pendant caterpillar in ${{\mathcal T}}_i$ with cherry $\{v_i,x_{i}^1\}$. We denote this by $$[\epsilon](x_{i}^1,x_{i}^2,\ldots,x_{i}^q,\epsilon).$$ Otherwise, if $v_i$ is not a leaf of a cherry in ${{\mathcal T}}_i$, we denote this by $$[\epsilon](\epsilon).$$ Now, recall that $2c_{{\mathcal T}}-1$ is the number of vertices in ${{\mathcal T}}_{{\mathcal I}}$. Setting $n=2c_{{\mathcal T}}-1$, we call $$\begin{aligned}
v_{{\mathcal T}}=&(&[\xi_1](x_{1}^1,x_{1}^2,\ldots,x_{1}^{q_1},\epsilon),\nonumber\\
&&[\xi_2](x_{2}^1,x_{2}^2,\ldots,x_{2}^{q_2},\epsilon),\nonumber\\
&&\vdots\nonumber\\
&&[\xi_{n}](x_{n}^1,x_{n}^2,\ldots,x_{n}^{q_{n}},\epsilon)\hspace{0.3cm})\nonumber
$$ the [*vector representation of ${{\mathcal T}}$ relative to ${{\mathcal T}}_{{\mathcal I}}$*]{}, and note that $$\xi_{c_{{\mathcal T}}+1}=\xi_{c_{{\mathcal T}}+2}=\cdots=\xi_{2c_{{\mathcal T}}-2}=\xi_{n}=\epsilon.$$ An example of a tree and its vector representation is shown in Figure \[fig:vector\].
![A rooted binary phylogenetic tree ${{\mathcal T}}$ whose index tree ${{\mathcal T}}_{{\mathcal I}}$ is shown in Step (6) of Figure \[fig:index\]. The vector representation of ${{\mathcal T}}$ relative to ${{\mathcal T}}_{{\mathcal I}}$ is $([x_1](x_2,x_3,\epsilon),[x_6],(x_7,\epsilon),[x_9](x_{10},x_8,\epsilon),[\epsilon](x_5,x_4,\epsilon),[\epsilon](x_{11}, x_{12},\epsilon))$.[]{data-label="fig:vector"}](vector){width="35.00000%"}
Let ${{\mathcal T}}$ be a rooted binary phylogenetic $X$-tree with $\epsilon\notin X$, and let ${{\mathcal T}}_{{\mathcal I}}$ be the index tree of ${{\mathcal T}}$. Let $v_{{\mathcal T}}$ be the vector representation relative to ${{\mathcal T}}_{{\mathcal I}}$. Furthermore, let ${{\mathcal T}}'$ be an element in ${{\mathcal C}}({{\mathcal T}})$, and let $$\begin{aligned}
v_{{{\mathcal T}}'}=&(&[\psi_1](y_{1}^1,y_{1}^2,\ldots,y_{1}^{r_1},\epsilon),\nonumber\\
&&[\psi_2](y_{2}^1,y_{2}^2,\ldots,y_{2}^{r_2},\epsilon),\nonumber\\
&&\vdots \nonumber\\
&&[\psi_n](y_n^1,y_n^2,\ldots,y_n^{r_n},\epsilon)\hspace{0.3cm})\nonumber\end{aligned}$$ be an $X$-vector for ${{\mathcal T}}'$. We say that $v_{{{\mathcal T}}'}$ has the [*cherry-property relative to $v_{{\mathcal T}}$*]{} if, for each cherry $\{a,b\}$ in ${{\mathcal T}}'$, exactly one of the following conditions holds:
(i) There is an index $s \in \{1,\,2,\,\ldots,\,n\}$ such that $\{\psi_s, y_{s}^1\} = \{a,b\}$.
(ii) There are two distinct indices $s,t \in \{1,\,2,\,\ldots,\,n\}$ such that $\{\psi_s, \psi_t\} = \{a,b\}$, the two corresponding vector components are $[\psi_s](\epsilon)$ and $[\psi_t](\epsilon)$, respectively, there is a vertex labeled $u$ in ${{\mathcal T}}_{{\mathcal I}}$ whose two children are labeled $s$ and $t$, and $\psi_u = \epsilon$.
To establish Theorem \[t:cherry\], we next prove three lemmas.
\[l:map\] Let ${{\mathcal T}}$ be a rooted binary phylogenetic $X$-tree, and let $v_{{\mathcal T}}$ be the vector representation of ${{\mathcal T}}$ relative to an index tree of ${{\mathcal T}}$. Then each tree ${{\mathcal T}}''$ in ${{\mathcal C}}({{\mathcal T}})$ can be mapped to an $X$-vector that satisfies the suffix-property and the cherry-property relative to $v_{{\mathcal T}}$. Moreover, the mapping is one-to-one.
Set $n=2c_{{\mathcal T}}-1$. We define a mapping $f$ from the elements in ${{\mathcal C}}({{\mathcal T}})$ into the set of all $X$-vectors that satisfy the suffix-property and the cherry-property relative to $v_{{\mathcal T}}$. First, we map ${{\mathcal T}}$ to $v_{{\mathcal T}}$ and note that $v_{{\mathcal T}}$ satisfies the suffix-property and the cherry-property relative to $v_{{\mathcal T}}$. Second, we map the element $\emptyset$ in ${{\mathcal C}}({{\mathcal T}})$ to the empty vector, say $v_\emptyset$, with $n$ vector components. Again, $v_\emptyset$ satisfies the suffix-property and the cherry-property relative to $v_{{\mathcal T}}$. Now, let ${{\mathcal T}}''$ be an element in ${{\mathcal C}}({{\mathcal T}})\setminus\{{{\mathcal T}},\emptyset\}$. Recalling the recursive definition of ${{\mathcal C}}({{\mathcal T}})$, there exists a tree ${{\mathcal T}}'$ in ${{\mathcal C}}({{\mathcal T}})$ with cherry $\{a,b\}$ such that ${{\mathcal T}}'[-a]$ is isomorphic to ${{\mathcal T}}''$. Suppose that $f$ (defined below) has already mapped ${{\mathcal T}}'$ to the $X$-vector $$\begin{aligned}
v_{{{\mathcal T}}'}=&(&[\psi_1](y_{1}^1,y_{1}^2,\ldots,y_{1}^{r_1},\epsilon),\nonumber\\
&&[\psi_2](y_{2}^1,y_{2}^2,\ldots,y_{2}^{r_2},\epsilon),\nonumber\\
&&\vdots \nonumber\\
&&[\psi_n](y_n^1,y_n^2,\ldots,y_n^{r_n},\epsilon)\hspace{0.3cm}),\nonumber\end{aligned}$$ that satisfies the suffix-property as well as the cherry-property relative to $v_{{{\mathcal T}}}$. Then $f$ maps ${{\mathcal T}}''$ to a vector that can be obtained from $v_{{{\mathcal T}}'}$ in one of the following two cases.
1. If there is an index $s \in \{1,\,2,\,\ldots,\,n\}$ such that $\{\psi_s, y_{s}^1\} = \{a, b\}$, then $f$ maps ${{\mathcal T}}''$ to a vector $v_{{{\mathcal T}}''}$ that is obtained from $v_{{{\mathcal T}}'}$ by replacing the vector component $$[\psi_s](y_{s}^1,y_{s}^2,\ldots,y_{s}^{r_s},\epsilon) \text{ with } [b](y_{s}^2,y_{s}^3\ldots,y_{s}^{r_s},\epsilon).$$
2. Otherwise, there are two indices $s,t \in \{1,\,2,\,\ldots,\,n\}$ with $s \neq t$ such that $\{\psi_s, \psi_t\} = \{a, b\}$, where the two corresponding components have the form $[\psi_s](\epsilon)$ and $[\psi_t](\epsilon)$, respectively. Furthermore, by construction, there is a vertex labeled $u$ in ${{\mathcal T}}_{{\mathcal I}}$ whose two children are the vertices labeled $s$ and $t$ and $\psi_u=\epsilon$. Then, $f$ maps ${{\mathcal T}}''$ to a vector $v_{{{\mathcal T}}''}$ that is obtained from $v_{{{\mathcal T}}'}$ by replacing each of the two vector components $$[\psi_s](\epsilon)\text{ and } [\psi_t](\epsilon)\text{ with } [\epsilon](\epsilon),$$ and replacing the vector component $$[\epsilon](y_{u}^1,y_{u}^2,\ldots,y_{u}^{r_u},\epsilon) \text{ with } [b](y_{u}^1,y_{u}^2\ldots,y_{u}^{r_u},\epsilon).$$
For both cases, it is easily checked that $v_{{{\mathcal T}}''}$ is an $X$-vector that satisfies the suffix-property relative to $v_{{\mathcal T}}$.
We next show that $v_{{{\mathcal T}}''}$ satisfies the cherry-property relative to $v_{{\mathcal T}}$. By Lemma \[lem:cherries\], we have $c_{{{\mathcal T}}'}-c_{{{\mathcal T}}''}\in\{0,1\}$. If $c_{{{\mathcal T}}'}-1=c_{{{\mathcal T}}''}$ then, by construction, each cherry in ${{\mathcal T}}''$ is a cherry in ${{\mathcal T}}'$. Hence, as $v_{{{\mathcal T}}'}$ satisfies the cherry-property relative to $v_{{\mathcal T}}$, we have that $v_{{{\mathcal T}}''}$ satisfies the cherry-property relative to $v_{{\mathcal T}}$. Otherwise, if $c_{{{\mathcal T}}'}=c_{{{\mathcal T}}''}$, then the cherry $\{a,b\}$ in ${{\mathcal T}}'$ is replaced with a new cherry that contains $b$, while all other cherries in ${{\mathcal T}}'$ are also cherries in ${{\mathcal T}}''$ . First, suppose that ${{\mathcal T}}''$ is obtained from ${{\mathcal T}}'$ according to mapping (M1). Observe that $r_s\geq 1$. If $r_s\geq 2$, then $\{b,y_s^2\}$ is the new cherry and, thus, $v_{{{\mathcal T}}''}$ satisfies the cherry-property relative to $v_{{\mathcal T}}$. On the other hand, if $r_s=1$, let $[\psi_t](y_t^1,y_t^2,\ldots,y_t^{r_t},\epsilon)$ be the vector component in $v_{{{\mathcal T}}'}$ such that the vertices labeled $s$ and $t$ in ${{\mathcal T}}_{{\mathcal I}}$ have the same parent. Note that $t$ exists because, otherwise, $s$ is the root of ${{\mathcal T}}_{{\mathcal I}}$ and so the existence of a cherry in ${{\mathcal T}}''$ that is not a cherry in ${{\mathcal T}}'$ implies that $r_s\geq 2$; a contradiction. If $r_t\geq 1$, then $\{\psi_t,y_t^1\}$ is a cherry in ${{\mathcal T}}'$ and ${{\mathcal T}}''$. Thus, the sibling of $b$ in ${{\mathcal T}}''$ is not a leaf, thereby contradicting that $b$ is a leaf of a cherry in ${{\mathcal T}}''$. Hence, $[\psi_t](y_t^1,y_t^2,\ldots,y_t^{r_t},\epsilon)=[\psi_t](\epsilon)$. Now, as $[b](\epsilon)$ and $[\psi_t](\epsilon)$ are two vector components of $v_{{{\mathcal T}}''}$, it again follows that $v_{{{\mathcal T}}''}$ satisfies the cherry-property relative to $v_{{\mathcal T}}$. Second, suppose that ${{\mathcal T}}''$ is obtained from ${{\mathcal T}}'$ according to mapping (M2). Noting that $b$ is an element of the vector component $[\psi_u](y_u^1,y_u^2,\ldots,y_t^{r_u},\epsilon)$ with $\psi_u=b$ in $v_{{{\mathcal T}}''}$, the result can be established by using an argument that is similar to the previous case.
It remains to show that the mapping is one-to-one. Let ${{\mathcal T}}'$ and ${{\mathcal T}}''$ be two distinct elements in ${{\mathcal C}}({{\mathcal T}})\setminus\{\emptyset\}$. Since each element in ${{\mathcal C}}({{\mathcal T}})\setminus\{{{\mathcal T}}, \emptyset\}$ can be obtained from ${{\mathcal T}}$ by repeatedly deleting a leaf of a cherry and suppressing the resulting degree-2 vertex, there exists an element $\ell$ in $X$ that is a leaf in ${{\mathcal T}}'$ and not a leaf in ${{\mathcal T}}''$. Let $X'$ and $X''$ be the leaf set of ${{\mathcal T}}'$ and ${{\mathcal T}}''$, respectively. Noting that $v_{{\mathcal T}}$ is an $X$-vector that contains each element in $X$ exactly once, it follows from construction of the mapping that $v_{{{\mathcal T}}'}$ is an $X'$-vector that contains each element in $X'$ exactly once and that $v_{{{\mathcal T}}''}$ is an $X''$-vector that contains each element in $X''$ exactly once. Hence $v_{{{\mathcal T}}'}\ne v_{{{\mathcal T}}''}$. Moreover, since no element in ${{\mathcal C}}({{\mathcal T}})\setminus\{\emptyset\}$ is mapped to $v_\emptyset$, the mapping is one-to-one. This completes the proof of the lemma.
\[lem:|C(T)|\] Let ${{\mathcal T}}$ be a rooted binary phylogenetic $X$-tree. Then $$|{{\mathcal C}}({{\mathcal T}})|\leq (|X|+1)^{4c_{{\mathcal T}}-2}.$$
Let $$\begin{aligned}
v_{{{\mathcal T}}}=&(&[\xi_1](x_{1}^1,x_{1}^2,\ldots,x_{1}^{q_1},\epsilon),\nonumber\\
&&[\xi_2](x_{2}^1,x_{2}^2,\ldots,x_{2}^{q_2},\epsilon),\nonumber\\
&&\vdots \nonumber\\
&&[\xi_n](x_n^1,x_n^2,\ldots,x_n^{q_n},\epsilon)\hspace{0.3cm})\nonumber\end{aligned}$$ be the vector representation of ${{\mathcal T}}$ relative to an index tree of ${{\mathcal T}}$, where $n=2c_{{\mathcal T}}-1$. We first derive an upper bound on the number of $X$-vectors that satisfy the suffix-property relative to $v_{{\mathcal T}}$. For each $i\in\{1,2,\ldots,n\}$, consider the vector component $[\xi_i](x_{i}^1,x_{i}^2,\ldots,x_{i}^{q_i},\epsilon)$. Then each $X$-vector that satisfies the suffix-property relative to ${{\mathcal T}}$ has an $i$th vector component, say $[\psi_i](y_{i}^1,y_{i}^2,\ldots,y_{i}^{r_i},\epsilon)$, such that $\psi_i\in X\cup\{\epsilon\}$ and $(y_{i}^1,y_{i}^2,\ldots,y_{i}^{r_i},\epsilon)$ is a suffix of $(x_{i}^1,x_{i}^2,\ldots,x_{i}^{q_i},\epsilon)$. Since there are at most $|X|+1$ such suffixes, it follows that there are at most $(|X|+1)^2$ variations of $[\psi_i](y_{i}^1,y_{i}^2,\ldots,y_{i}^{r_i},\epsilon)$. Hence, there are at most $$((|X| + 1)^2)^n = (|X| + 1)^{4c_{{\mathcal T}}-2}.$$ $X$-vectors that satisfy the suffix-property relative to $v_{{\mathcal T}}$. By Lemma \[l:map\], each tree in ${{\mathcal C}}({{\mathcal T}})$ can be mapped to one such vector and, as the map is one-to-one, it follows that ${{\mathcal C}}({{\mathcal T}})$ contains at most $(|X| + 1)^{4c_{{\mathcal T}}-2}$ trees.
For a rooted binary phylogenetic $X$-tree ${{\mathcal T}}$, the next lemma constructs an automaton that recognizes whether or not a word that contains each element in $X$ precisely once is a cherry-picking sequence for ${{\mathcal T}}$.
\[lem:automaton\] Let ${{\mathcal T}}$ be a rooted binary phylogenetic $X$-tree. There is a deterministic finite automaton ${{\mathcal A}}_{{\mathcal T}}$ with $O(|X|^{4c_{{\mathcal T}}-2})$ states that recognizes the language $${{\mathcal L}}_{X}({{\mathcal T}}) = \{x_1x_2\ldots x_{|X|} \mid (x_1,\,x_2,\,\ldots,\,x_{|X|}) \text{ is a cherry-picking sequence for } {{\mathcal T}}\}.$$ Moreover, the automaton ${{\mathcal A}}_{{\mathcal T}}$ can be constructed in time $f(|X|, c_{{\mathcal T}}) \in |X|^{O(c_{{\mathcal T}})}$.
Throughout this proof, we denote the tree without a vertex by $\emptyset$. Let $M$ and $M'$ be two sets. Setting $M = \{{{\mathcal T}}\}$ and $M' = \emptyset$, we construct ${{\mathcal A}}_{{\mathcal T}}$ as follows.
(1) Create the states $q_{{\mathcal T}}$, $q_\emptyset$, and $q_e$. For each $a \in X$, set $\delta(q_e, a) = \delta(q_\emptyset, a) = q_e$.
(2) For each ${{\mathcal T}}\in M$ and each $a \in X$ do the following.
(a) If $a$ is a leaf of a cherry in ${{\mathcal T}}$ or $a$ is the only vertex of ${{\mathcal T}}$, then
(i) create the state $q_{{{\mathcal T}}[-a]}$ if ${{\mathcal T}}[-a]$ is not isomorphic to a tree in $M'$,
(ii) set $M' = M' \cup {{\mathcal T}}[-a]$, and
(iii) set $\delta(q_{{{\mathcal T}}}, a) = q_{{{\mathcal T}}[-a]}$.
(b) Otherwise, set $\delta(q_{{{\mathcal T}}}, a) = q_e$.
(3) Set $M = M'$ and, subsequently, set $M' = \emptyset$. If $M \neq \{\emptyset\}$, continue with $(2)$.
We set the initial state of ${{\mathcal A}}_{{\mathcal T}}$ to be $q_{{\mathcal T}}$ and the final state to be $q_\emptyset$. To illustrate, the construction of ${{\mathcal A}}_{{\mathcal T}}$ is shown in Figure \[fig:automaton\] for a phylogenetic tree on four leaves.
![Construction of an automaton that recognizes the language ${{\mathcal L}}_X({{\mathcal T}}_1)$ as described in the statement of Lemma \[lem:automaton\] and with ${{\mathcal T}}_1$ shown in the top left of this figure. Each vertex (resp. edge) represents a state (resp. transition). The vertex $q_{{{\mathcal T}}_1}$ indicates the initial state whereas the final state is $q_\emptyset$ as indicated by a double circle. To increase readability, most transitions to $q_e$ are omitted. In row $i$, the figure shows $M$, $M'$, and the automaton after the $i$th execution of the for-loop as described in Step (2) in the proof of Lemma \[lem:automaton\].[]{data-label="fig:automaton"}](automatonConstruction){width=".78\textwidth"}
By construction, we have ${{\mathcal A}}_{{\mathcal T}}=({{\mathcal C}}({{\mathcal T}}),X,\delta,q_{{\mathcal T}},\{q_\emptyset\})$. As each cherry-picked tree in ${{\mathcal C}}({{\mathcal T}})$ is mapped to a unique state, it follows from Lemma \[lem:|C(T)|\] that the number of states of ${{\mathcal A}}_{{\mathcal T}}$ is $O(|X|^{4c_{{\mathcal T}}-2})$. Moreover, for each $a\in X$ and each pair of two distinct states ${{\mathcal T}}',{{\mathcal T}}'' \in {{\mathcal C}}({{\mathcal T}})$, there is a transition $\delta(q_{{{\mathcal T}}'}, a) = q_{{{\mathcal T}}''}$ if and only if ${{\mathcal T}}'[-a] = {{\mathcal T}}''$ and $a$ is either a leaf of a cherry in ${{\mathcal T}}'$ or ${{\mathcal T}}'$ consists of the single vertex $a$. The state $q_e$ collects all inputs that do not correspond to the continuation of a cherry-picking sequence. More precisely, there is a transition $\delta(q_{{{\mathcal T}}'}, a) = q_{e}$ if and only if $a$ is not a leaf of a cherry in ${{\mathcal T}}'$ and ${{\mathcal T}}'$ does not consist of the single vertex $a$. It now follows that there is a one-to-one correspondence between the directed paths from $q_{{\mathcal T}}$ to $q_\emptyset$ in ${{\mathcal A}}_{{\mathcal T}}$ and the cherry-picking sequences of ${{\mathcal T}}$ and, hence, ${{\mathcal A}}_{{\mathcal T}}$ recognizes ${{\mathcal L}}_{X}({{\mathcal T}})$.
The time taken to construct ${{\mathcal A}}_{{\mathcal T}}$ is dominated by the number of iterations of the for-loop in Step $(2)$. Since $|M| < |{{\mathcal C}}({{\mathcal T}})|$ and $|{{\mathcal C}}({{\mathcal T}})| \in O(|X|^{4c_{{\mathcal T}}- 2})$, the number of iterations in Step (2) is $O(|{{\mathcal C}}({{\mathcal T}})|\cdot|X|) \subseteq O(|X|^{4c_{{\mathcal T}}- 1})$. Moreover, since Step (2) is executed $|X|$ times, each operation of the for-loop is executed $O(|X|^{4c_{{\mathcal T}}})$ times in total. While the complexity of these operations depend on the implementation and data structure, they can clearly be implemented such that ${{\mathcal A}}_{{\mathcal T}}$ can be constructed in time $|X|^{O(c_{{\mathcal T}})}$. This establishes the lemma.
Generalizing the language that is described in the statement of Lemma \[lem:automaton\], the next straightforward observation describes a language for the decision problem [CPS-Existence]{}.
\[obs:CPS\_language\_problem\] Let ${{\mathcal P}}=\{{{\mathcal T}}_1,{{\mathcal T}}_2,\ldots,{{\mathcal T}}_m\}$ be a collection of rooted binary phylogenetic $X$-trees. Then, solving [CPS-Existence]{} for ${{\mathcal P}}$ is equivalent to deciding if $$\bigcap_{1\leq i\leq m} {{\mathcal L}}_X({{\mathcal T}}_i) \neq \emptyset.$$
We are now in a position to establish Theorem \[t:cherry\].
[*Proof of Theorem \[t:cherry\].*]{} By Observation \[obs:CPS\_language\_problem\], it follows that there is a cherry-picking sequence for ${{\mathcal P}}$ if and only if $
\bigcap_{{{\mathcal T}}_i \in {{\mathcal P}}} {{\mathcal L}}_X({{\mathcal T}}_i) \neq \emptyset,
$ where $${{\mathcal L}}_{X}({{\mathcal T}}) = \{x_1x_2\ldots x_{|X|} \mid (x_1,\,x_2,\,\ldots,\,x_{|X|}) \text{ is a cherry-picking sequence for } {{\mathcal T}}\}.$$
For each ${{\mathcal T}}_i \in {{\mathcal P}}$ with $1 \leq i \leq m$, we follow the notation and construction that is described in the proof of Lemma \[lem:automaton\] to obtain an automaton ${{\mathcal A}}_{{{\mathcal T}}_i}$ with $O(|X|^{4c_{{{\mathcal T}}_i}-2})$ states that recognizes the language ${{\mathcal L}}_X({{\mathcal T}}_i)$. To solve the question whether or not the intersection of these $m$ languages is empty, we use the well-known construction of a product automaton [@kozen97] as follows.
For each ${{\mathcal T}}_i \in {{\mathcal P}}$, let $Q_{{{\mathcal T}}_i}$ be set of states, and let $\delta_{{{\mathcal T}}_i}$ be the transition relation of ${{\mathcal A}}_{{{\mathcal T}}_i}$. We construct a new automaton ${{\mathcal A}}_{{{\mathcal P}}}$, where the set of states $Q_{{\mathcal P}}$ is the cartesian product $Q_{{{\mathcal T}}_1} \times Q_{{{\mathcal T}}_2}\times \ldots \times Q_{{{\mathcal T}}_{m}}$. Furthermore, the alphabet of ${{\mathcal A}}_{{{\mathcal P}}}$ is $X$ and the transition relation $\delta_{{\mathcal P}}\colon Q_{{\mathcal P}}\times X \rightarrow Q_{{\mathcal P}}$ is defined as $$\delta_{{\mathcal P}}((q_1,\ldots,q_m), a) = (\delta_{{{\mathcal T}}_1}(q_1, a),\, \ldots,\, \delta_{{{\mathcal T}}_m}(q_m, a)).$$ Lastly, the initial (resp. final) state of ${{\mathcal A}}_{{\mathcal P}}$ is $(q_1,\ldots,q_m)$ where, for all $i\in\{1,2,\ldots,m\}$, $q_i$ is the initial (resp. final) state of ${{\mathcal A}}_{{{\mathcal T}}_i}$. Intuitively, ${{\mathcal A}}_{{\mathcal P}}$ simulates the parallel execution of the automata ${{\mathcal A}}_{{{\mathcal T}}_1},{{\mathcal A}}_{{{\mathcal T}}_2},\ldots,{{\mathcal A}}_{{{\mathcal T}}_m}$. By construction, an input sequence is accepted by ${{\mathcal A}}_{{\mathcal P}}$ if and only if it is accepted by each automaton ${{\mathcal A}}_{{{\mathcal T}}_i}$. It now follows that there is a cherry-picking sequence for ${{\mathcal P}}$ if and only if the final state of ${{\mathcal A}}_{{\mathcal P}}$ can be reached from the initial state of ${{\mathcal A}}_{{\mathcal P}}$ and, hence, ${{\mathcal L}}({{\mathcal A}}_{{\mathcal P}}) \neq \emptyset$.
It remains to show that the computational complexity is as claimed in the statement of the theorem. Viewing ${{\mathcal A}}_{{\mathcal P}}$ as a directed graph, each directed path from the initial to the final state of ${{\mathcal A}}_{{\mathcal P}}$ has length $|X|$. We can therefore decide whether the final state from ${{\mathcal A}}_{{\mathcal P}}$ is reachable by using breadth-first search [@cormen09] in time $O(|Q_{{\mathcal P}}| + |X| \cdot |Q_{{\mathcal P}}|)$, where $|X| \cdot |Q_{{\mathcal P}}|$ is the number of transitions in ${{\mathcal A}}_{{\mathcal P}}$. By construction and Lemma \[lem:automaton\], it follows that ${{\mathcal A}}_{{\mathcal P}}$ has $$O\left(\prod_{i=1}^m |X|^{4c_{{{\mathcal T}}_i}-2}\right) \subseteq O(|X|^{m(4c-2)})$$ states, i.e. $|Q_{{\mathcal P}}| \in O(|X|^{m(4c-2)})$. Hence, we can decide in time $O(|X|^{m(4c-2)+1})$ whether ${{\mathcal L}}({{\mathcal A}}_{{\mathcal P}}) \neq \emptyset$. By Lemma \[lem:automaton\], it takes time $f_i(|X|, c_{{{\mathcal T}}_i}) \in |X|^{O(c_{{{\mathcal T}}_i})}$ to construct each automaton ${{\mathcal A}}_{{{\mathcal T}}_i}$ and, thus, it follows that deciding if there is a cherry-picking sequence for ${{\mathcal P}}$ can be done in time $$O\left (|X|^{m(4c-2)+1} + \sum_{i=1}^m f_i(|X|, c_{{{\mathcal T}}_i})\right ),$$ which is polynomial in $|X|$ if $c$ and $m$ are constant.\
Concluding remarks {#sec:con}
==================
In this paper, we have shown that [CPS-Existence]{}, a problem of relevance to the construction of phylogenetic networks from a set of phylogenetic trees, is NP-complete for all sets ${{\mathcal P}}$ of rooted binary phylogenetic trees with $|{{\mathcal P}}|\geq 8$. This result partially answers a question posed by Humphries et al. [@humphries13]. They asked if [CPS-Existence]{} is computationally hard for $|{{\mathcal P}}|=2$. To establish our result, we first showed that [$4$-Disjoint-Intermezzo]{}, which is a variant of the [Intermezzo]{} problem that is new to this paper, is NP-complete. Subsequently, we established a reduction from an instance $I$ of [$4$-Disjoint-Intermezzo]{} to an instance $I'$ of [CPS-Existence]{} with $|{{\mathcal P}}|=8$. Since each of the four collections of pairs and triples in $I$ reduces to two trees in $I'$, a possible approach to obtain a stronger hardness result for [CPS-Existence]{} with $|{{\mathcal P}}|<8$ is to show that [$N$-Disjoint-Intermezzo]{} is NP-complete for $N<4$. However, it seems likely that such a result can only be achieved by following a strategy that is different from the one that we used in this paper. In particular, there is no obvious reduction from [2P2N-3-SAT]{} to [$3$-Disjoint-Intermezzo]{}. Moreover, [$1$-Disjoint-Intermezzo]{} is solvable in polynomial time since all pairs and triples are pairwise disjoint and, so, it cannot be used for a reduction even if [CPS-Existence]{} turns out to be NP-complete for $|{{\mathcal P}}|=2$.
In the second part of the paper, we have translated [CPS-Existence]{} into an equivalent problem on languages and used automata theory to show that [CPS-Existence]{} can be solved in polynomial time if the number of trees in ${{\mathcal P}}$ and the number of cherries in each such tree are bounded by a constant. There are currently only a small number of other problems in phylogenetics that have been solved with the help of automata theory (e.g. [@hall10; @westesson12]) and it is to be hoped that the results presented in this paper will stimulate further research to explore connections between combinatorial problems in phylogenetics and automata theory.
[**Acknowledgements**]{}
We thank Britta Dorn for insightful comments on a draft version of this paper. The second author was supported by the New Zealand Marsden Fund.
[**References**]{}
[00]{} B. Albrecht, C. Scornavacca, A. Cenci, and D. H. Huson (2012). Fast computation of minimum hybridization networks. [*Bioinformatics*]{}, 28, 191–197.
M. Baroni, S. Gr[ü]{}newald, V. Moulton, and C. Semple, (2005), Bounding the number of hybridization events for a consistent evolutionary history, [*Journal of Mathematical Biology*]{}, 51, 171–182.
M. Baroni, C. Semple, and M. Steel (2006). Hybrids in real time. [*Systematic Biology*]{}, 55, 46–56.
M. Bordewich and C. Semple (2007). Computing the minimum number of hybridization events for a consistent evolutionary history. [*Discrete Applied Mathematics*]{}, 155, 914–928.
P. Berman, M. Karpinski, and A. D. Scott (2003). Approximation hardness of short symmetric instances of MAX-3SAT. [*Electronic Colloquium on Computational Complexity*]{}, Report No. 49.
G. Cardona, F. Rosselló, and G. Valiente (2009). Comparison of tree-child phylogenetic networks. [*IEEE/ACM Transactions on Computational Biology and Bioinformatics*]{}, 6, 552–569.
Z. Z. Chen and L. Wang (2013). An ultrafast tool for minimum reticulate networks. [*Journal of Computational Biology*]{}, 20, 38–41.
T. H. Cormen, C. E. Leierson, R. L. Rivest, and C. Stein (2009). Introduction to Algorithms. The MIT Press.
W. Guttmann and M. Maucher (2006). Variations on an ordering theme with constraints. In [*Fourth IFIP International Conference on Theoretical Computer Science*]{}, Springer, pp. 77–90.
D. Hall and D. Klein (2010). Finding cognate groups using phylogenies. In [*Proceedings of the 48th Annual Meeting of the Association for Computational Linguistics*]{}, pp. 1030–1039.
J. E. Hopcroft and J. D. Ullman (1979). Introduction to Automata Theory, Languages, and Computation. Addison-Wesley.
P. J. Humphries, S. Linz, and C. Semple (2013). Cherry picking: a characterization of the temporal hybridization number for a set of phylogenies. [*Bulletin of Mathematical Biology*]{}, 75, 1879–1890.
P. J. Humphries, S. Linz, and C. Semple (2013). On the complexity of computing the temporal hybridization number for two phylogenies. [*Discrete Applied Mathematics*]{}, 161, 871–880.
D. H. Huson, R. Rupp, C. Scornavacca (2010). Phylogenetic networks: concepts, algorithms and applications. Cambridge University Press.
D. C. Kozen (1997). Automata and Computability. Undergraduate Texts in Computer Science. Springer.
J. Mallet, N. Besansky, and M. W. Hahn (2016). How reticulated are species? [*BioEssays*]{}, 38, 140–149.
B. M. E. Moret, L. Nakhleh, T. Warnow, C. R. Linder, A. Tholse, A. Padolina, J. Sun, and R. Timme (2004). Phylogenetic networks: modeling, reconstructibility, and accuracy. [*IEEE/ACM Transactions on Computational Biology and Bioinformatics*]{}, 1, 13–23.
T. Piovesan and S. Kelk: A simple fixed parameter tractable algorithm for computing the hybridization number of two (not necessarily binary) trees, [*IEEE/ACM Transactions on Computational Biology and Bioinformatics*]{}, 10, 18–25.
C. Semple (2007). Hybridization networks, pages 277–314. In Gascuel, O. and Steel, M., eds., [*Reconstructing Evolution: New Mathematical and Computational Advances*]{}, Oxford University Press.
S. M. Soucy, J. Huang, and J. P. Gogarten (2015). Horizontal gene transfer: building the web of life. [*Nature Reviews Genetics*]{}, 16, 472–482.
O. Westesson, G. Lunter, B. Paten, and Ian Holmes (2012). Accurate reconstruction of insertion-deletion histories by statistical phylogenetics. [*PLoS One*]{}, 7(4): e34572.
Y. Wu and J. Wang (2010). Fast computation of the exact hybridization number of two phylogenetic trees. In [*International Symposium on Bioinformatics Research and Applications*]{}, Springer, pp. 203–214.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Encouraged by the recent real-time renormalization group results we carried out a detailed analysis of the nonequilibrium Kondo conductance observed in an InAs nanowire-based quantum dot and found them to be in excellent agreement. We show that in a wide range of bias the Kondo conductance zero-bias anomaly is scaled by the Kondo temperature to a universal lineshape predicted by the numerical study. The lineshape can be approximated by a phenomenological expression of a single argument $eV_{sd}/k_{\rm{B}}T_{\rm K}$. The knowledge of an analytical expression for the lineshape provides an alternative way for estimation of the Kondo temperature in a real experiment, with no need for time consuming temperature dependence measurements of the linear conductance.'
author:
- 'Andrey V. Kretinin'
- Hadas Shtrikman
- Diana Mahalu
bibliography:
- 'manuscript.bib'
title: 'Universal lineshape of the Kondo zero-bias anomaly in a quantum dot'
---
In quantum dots an unpaired spin of the localized electron may act as a single magnetic impurity, which interacts with electrons in metallic leads forming a Kondo-correlated state [@Glazman1988; @*Ng1988]. The quantum dots have been recognized as a versatile system for probing the many-body nature of the Kondo effect [@Goldhaber-Gordon1998a; @*Cronenwett1998; @Goldhaber-Gordon1998]. The main advantage of quantum dot hosting the Kondo state is the ability to tune its main characteristic parameter, the Kondo temperature ($T_{\rm{K}}$). Thereby, it is possible to study the reaction of the many-body system to various external perturbations, such as temperature $T$, bias $V$ and magnetic field.
One of the remarkable properties of the Kondo effect is that the response of the Kondo-enhanced conductance to a perturbation is governed by a set of universal laws independent of the physical system in which the Kondo state is realized. It has been experimentally verified that in the low-energy limit ($\{T, e|V|/k_{\rm{B}}\}<<T_{\rm K}$, where $k_{B}$ is the Boltzmann constant and $e$ is the elementary charge) the Kondo conductance as a function of $T$ and $V$ is described by a universal quadratic law [@Grobis2008; @*Scott2009; @*Yamauchi2011; @Kretinin2011] expected from the Fermi-liquid theory [@Costi1994; @Nozi'eres1974; @*Schiller1995]. At higher temperatures ($T \sim T_{K}$) the Kondo conductance has been shown to scale to a universal dependence of a single parameter $T/T_{K}$ found from numerical renormalization group (NRG) study [@Costi1994]. These NRG calculations were later approximated by the phenomenological expression [@Goldhaber-Gordon1998] of a single parameter $T/T_{K}$, which became the main method for estimation of the Kondo temperature in real systems. Recently, we also attempted to check the universal behavior of the magnetic field dependence in both the low- and intermediate-field range [@Kretinin2011]. Despite all the theoretical advances, the problem of the nonequilibrium Kondo model at intermediate bias ($|V| \sim k_{\rm{B}}T_{\rm{K}}/e$) remained unsolved. As a result, the lineshape of one of the main hallmarks of the Kondo effect in quantum dots, known as zero-bias conductance anomaly (ZBA) [@Wingreen1994] had not been fully described by the theory. Only very recently, the 2-loop real-time renomalization group (RTRG) calculations developed for the Kondo model [@Pletyukhov2012] provided a numerical description of the nonequilibrium Kondo conductance in a wide range of biases. The result of these calculations agreed very well with the experimental data and linked previously developed analytical theories made in the low- [@Costi1994] and high-energy [@Haldane1978] limits. The development of the above mentioned RTRG calculations gave us an additional impulse to study the Kondo ZBA in more detail and test the universality of its lineshape.
In this Letter we present a detailed study of ZBA associated with the Kondo effect observed in an InAs nanowire-based quantum dot. The analysis of our experimental data shows that in a wide range of applied bias the Kondo ZBA can be scaled into a universal lineshape predicted by the RTRG calculations [@Pletyukhov2012]. We found our experiment to be well described by the proposed phenomenological expression, which accurately approximates the RTRG results in the experimentally relevant range of bias. We also show that this expression can be employed as an alternative method of quick and accurate estimation of the Kondo temperature.
![(a) The temperature dependence of the linear conductance observed in the InAs nanowire-based quantum dot. (b) The two-dimensional gray-scale plot of the nonequilibrium conductance measured in $V_{g}-V_{sd}$ plain at $T=T_{\rm{base}}$ for the same range of $V_{g}$ as in (a). The colored vertical dashed lines mark the conductance traces at fixed values of $V_{g}$ used for the scaling analysis ($V_{g} = -2.85$ V brown, $V_{g} = -2.845$ V red, $V_{g} = -2.84$ V orange, $V_{g} = -2.835$ V yellow, $V_{g} = -2.83$ V green, $V_{g} = -2.825$ V cyan, $V_{g} = -2.82$ V light blue, $V_{g} = -2.815$ V dark blue).[]{data-label="Fig1"}](Fig1 "fig:"){width="1\columnwidth"}\
*Experiment.*–The quantum dot used in the experiment was formed in a 50 nm-diameter high-quality InAs nanowire grown by the vapor-liquid-solid method on a (011) InAs substrate using molecular-beam epitaxy. The as-grown nanowires were randomly deposited onto a $p^{+}$-Si/SiO$_{2}$ substrate and individually connected by ohmic (Ni/Au) source and drain electrodes using e-beam lithography. To avoid undesirable effects of the underlying substrate the nanowires were suspended in vacuum over grooves predefined in the substrate, and fixed from the sides by the contacts. The lateral size of the dot was defined by the contacts separation ($\sim$450 nm) due to the electrons in nanowire being localized between two Schottky barriers formed at the nanowire-contact interface. The experiment was performed in a He$^{3}$-He$^{4}$ dilution refrigerator with the base temperature $T_{\rm{base}} \approx $10 mK. The transport measurements were made by a standard lock-in technique. Depending on the temperature the ac excitation bias was kept smaller or equal to $k_{B}T$. The dc bias $V_{sd}$ was applied to the source with respect to the drain, “virtually” grounded through the transimpedance preamplifier. The dot occupancy was tuned by the backgate voltage, $V_{g}$, applied to the $p^{+}$-Si substrate. More details on growth, fabrication and experimental set-up can be found in Ref. .
 for color code\], approximated with Eq. (\[DGG\_formula\]) (solid black curve). (b) The dependence of the estimated value of $T_{\rm{K}}$ on $V_{g}$. The error bars here represent the 68% confidence interval. (c) The normalized conductance $G/G_{0}$ as a function of $V_{sd}$ measure at different $V_{g}$ and $T=T_{\rm{base}}$. (d) The same normalized conductance traces as in (c), but replotted as a function of $eV_{sd}/k_{B}T_{K}$ (colored curves). The traces are shown to collapse onto the same universal dependence given by RTRG calculations [@Pletyukhov2012] (black dashed curve). []{data-label="Fig2"}](Fig2 "fig:"){width="1\columnwidth"}\
*Large bias scaling.*–To investigate the Kondo ZBA in more details we used the experimental results previously examined in our study of the low-bias and magnetic field scaling of the spin-1/2 Kondo conductance \[see Ref. , Fig.1(c)\]. Figure \[Fig1\](a) presents the linear differential conductance $G$ through the quantum dot as a function of $V_{g}$ taken at different $T$. The region of $V_{g}$ at around $-2.83$ V corresponds to the conductance enhanced at lower temperatures due to the presence of the many-body Kondo state. There are two Coulomb blockade peaks emerging at higher temperatures, which mark the region of $V_{g}$ with an odd dot occupancy. To identify the relevant Kondo ZBA in Fig. \[Fig1\](b) we show the gray-scale plot of the nonequilibrium differential conductance measured for the same range of $V_{g}$. The ZBA is seen here as a black horizontal line at around $V_{sd} =$ 0. To test the theoretically predicted scaling of the Kondo ZBA we chose several nonequilibrium conductance traces taken at fixed values of $V_{g}$ as shown in Fig. \[Fig1\](b) by colored dashed lines. For each of the chosen values of $V_{g}$ we estimated $T_{\rm{K}}$ by fitting the temperature dependence of the linear conductance $G(T)$ with the phenomenological expression [@Goldhaber-Gordon1998] $$\textstyle
G(T)/G_{0} = \left[1+\left(T/T'_{\rm{K}}\right)^{2}\right]^{-s},
\label{DGG_formula}$$ where $G_{0} = G(T = 0, V_{sd} = 0)$, $T'_{\rm{K}} = T_{\rm{K}}/(2^{1/s} - 1)^{1/2}$ and the parameter $s =$ 0.22. Here the definition of $T_{\rm{K}}$ is such that $G(T=T_{\rm{K}},V_{sd}=0)=1/2G_{0}$. The experimental $G(T)$ dependencies fitted with Eq. (\[DGG\_formula\]) with $T_{\rm{K}}$ used as a fitting parameter are shown in Fig. \[Fig2\](a). The circles represent the experimental points taken at the values of $V_{g}$ color coded in Fig. \[Fig1\](b), and the solid black curve corresponds to Eq. (\[DGG\_formula\]). The deviation from the theoretical curve is related to additional mechanisms engaged in the transport at higher temperature. To avoid the effect of additional mechanisms the fitting procedure was made only for $T \leq$ 200 mK. The fact that at $T <$ 200 mK all the experimental data plotted as a function of $T/T_{\rm{K}}$ collapse on the same theoretical curve reflects the universality of the temperature dependence of the Kondo conductance [@Costi1994; @Goldhaber-Gordon1998]. The estimated value of $T_{\rm{K}}$ at chosen $V_{g}$ is shown in Fig. \[Fig2\](b) and follows the previously observed [@Wiel2000; @Kretinin2011] parabolic-like dependence [@Haldane1978].
As noticed in the early experiments [@Goldhaber-Gordon1998a; @Wiel2000; @Sasaki2000; @*Nygard2000; @*Jespersen2006; @*Csonka2008] the width of the Kondo ZBA appears to be proportional to $T_{\rm{K}}$; the same qualitative behavior is observed in our experiment. Figure \[Fig2\](c) shows the normalized nonequilibrium conductance $G/G_{0}$ at different $V_{g}$ with a well-pronounced ZBA maximum at $V_{sd} =$ 0. The widest ZBA (dark blue curve) is associated with the highest $T_{\rm{K}} \approx$ 900 mK, and the narrowest ZBA (orange and yellow curves) corresponds to the lowest $T_{\rm{K}} \approx$ 300 mK \[see Fig. \[Fig2\](b)\]. In accordance with the RTRG calculations [@Pletyukhov2012] the nonequilibrium conductance in the Kondo regime is scaled by $T_{\rm{K}}$ into a universal dependence (these calculations were made in the zero-temperature limit, as for our experiment, $T/T_{\rm{K}} \leq$ 0.03, which was theoretically checked [@Pletyukhov2012] to be close enough to zero). To verify this prediction we plotted the normalized conductance $G/G_{0}$ as a function of the scaled bias $eV_{sd}/k_{\rm{B}}T_{\rm{K}}$ in Fig. \[Fig2\](d). It is evident from Fig. \[Fig2\](d) that all curves collapse onto the same universal dependence, which is given by the RTRG calculations and plotted as the dashed black curve. The deviations from the prediction observed at higher biases occur due to the approaching resonant level of the dot. In this case the system switches from the Kondo to the mixed-valence regime where the scaling is no longer valid [@Goldhaber-Gordon1998; @Schoeller2000].
![(a) The RTRG calculations of the nonequilibrium Kondo conductance as a function of bias taken from Ref. (blue circles) and its approximation with Eq. (\[HS&MP\_formula\]) (red solid curve) and Eq. (\[simple\_formula\]) (black dashed curve). (b) Extraction of $T_{K}$ using two different measurements, $G(T)$ (blue squares) and $G(V)$ (red circles) fitted with Eq. (\[DGG\_formula\]) (black solid curve) and Eq. (\[simple\_formula\]) (black dashed curve), correspondingly. The two extracted values $T^{(T)}_{\rm{K}}$ and $T^{(V)}_{\rm{K}}$ are the same within the statistical error. (c) The experimental Kondo ZBA measured at $V_{g}=$ -2.83 V \[see green dashed line in Fig. \[Fig1\](b)\] (red circles) and its theoretical approximation made with Eq. (\[simple\_formula\]) (black dashed curve). When the bias is such that $e|V_{sd}|=k_{\rm{B}}T_{\rm{K}}$ the Kondo conductance decreases to 2/3 of its zero-bias value $G_{0}$.[]{data-label="Fig3"}](Fig3 "fig:"){width="1\columnwidth"}\
*Phenomenological formula.*–As shown above the ZBA in the Kondo regime appears to have a universal dependence on bias, scaled by $T_{\rm{K}}$. This property of ZBA makes it potentially valuable for extracting $T_{\rm{K}}$ from the bias dependence of the conductance $G(V_{sd})$, in the same way as for $G(T)$ by approximating it with Eq. (\[DGG\_formula\]). Here the use of numerical calculations is quite cumbersome, and the authors of Ref. suggested a phenomenological formula for describing their RTRG results, which we utilized for our definition of $T_{\rm{K}}$ [^1] $$ G(T=0, \nu)/G_{0} = \left[1+\frac{\left(2^{1/s_{1}}-1\right)\nu^{2}}{\pi+b\left(|\nu|^{s_{2}}-1\right)}\right]^{-s_{1}},
\label{HS&MP_formula}$$ where $\nu = eV_{sd}/k_{\rm{B}}T_{\rm{K}}$. The best fit of Eq. (\[HS&MP\_formula\]) to the RTRG results gives $b = 0.05 \pm 0.01$, $s_{1} = 0.32 \pm 0.01$ and $s_{2} = 1.28 \pm 0.03$, Fig. \[Fig3\](a) illustrates the quality of this fit. As seen from the plot, Eq. (\[HS&MP\_formula\]) (red solid curve) approximates the numerical results (blue circles) very well in a wide range of bias. However, in real experimental situation the Kondo ZBA rarely extends beyond $\pm 2k_{\rm{B}}T_{\rm{K}}/e$ due to a dominating background, thus for the narrower range of bias ($e|V_{sd}|/k_{\rm{B}}T_{\rm{K}} < 10$) one can use a simplified expression $$ G(T=0, \nu)/G_{0} = \left[1+\alpha \nu^{2}\right]^{-s_{1}},
\label{simple_formula}$$ where $\alpha = (2^{1/s_{1}}-1)/\pi$. The simplified formula is identical to the one in Ref. . As can be seen from Fig. \[Fig3\](a) the simplified expression, shown by the black dashed curve, approximates the numerical result adequately up to $eV_{sd}/k_{B}T_{\rm{K}}\sim10$. Note that in the low-bias limit $e|V_{sd}|/k_{\rm{B}}T_{\rm{K}}\ll1$ Eq. (\[simple\_formula\]) reduces to the quadratic Fermi-liquid dependence $G(V_{sd})/G_{0} \approx 1 - s_{1}\alpha(eV_{sd}/k_{\rm{B}}T_{\rm{K}})^2 = 1 - c_{V}(eV_{sd}/k_{\rm{B}}T_{\rm{K}})^2$, where the coefficient $c_{V}\approx0.78$ is in a good agreement with both theory and experiment [@Kretinin2011].
Since the Kondo ZBA is scaled by $T_{\rm{K}}$ into a universal lineshape, which is described by a phenomenological formula, it should be possible to perform the reverse operation and extract the value of $T_{\rm{K}}$ by fitting the experimental $G(V_{sd})$ dependence with Eq. (\[simple\_formula\]). Figure \[Fig3\](b) shows a semilogarithmic plot of the Kondo conductance measured as a function of $T$ (blue circle) and the modified bias $eV_{sd}/k_{\rm{B}}$ (red circles). Each set of data is approximated with its own phenomenological formula using $T_{\rm{K}}$ as a fitting parameter; $G(T)$ with Eq. (\[DGG\_formula\]) for $T\leq$ 200 mK and $G(V_{sd})$ with Eq. (\[simple\_formula\]) for $|V_{sd}|\leq$ 30 $\mu$V, correspondingly [^2]. The two values of the Kondo temperature were found to be $T^{(T)}_{\rm{K}} = 314\pm28$ mK for $G(T)$ and $T^{(V)}_{\rm K} = 293\pm6$ mK for $G(V_{sd})$ and they are equal within the statistical error. Also, fitting of $G(V_{sd})$ provides statistically more accurate value of $T_{\rm K}$ due to a larger number of experimental points available for analysis. In our opinion this method of estimating $T_{K}$ may be advantageous to the traditional one involving measurement of the $G(T)$ dependence and approximating it with Eq. (\[DGG\_formula\]). First, it is much easier to reliably measure the $G(V_{sd})$ dependence at the lowest possible temperature, rather than performing time-consuming measurements of $G(V_{g})$ in the linear regime at multiple temperatures. Normally, the low temperature ($T/T_{\rm K}<<1$), required for Eq. (\[simple\_formula\]) is easily satisfied, especially at submillikelvin temperatures, since for most systems $T_{\rm K}\geq$ 200 mK. Second, as pointed out, the value of $T_{K}$ extracted from the experimental $G(V_{sd})$ dependence is potentially more accurate.
*Width of ZBA.*–Finally, we would like to discuss an important practical implication of Eq. (\[simple\_formula\]). So far, the width of the Kondo ZBA has been used as a rough experimental estimate for $T_{\rm K}$ [@Goldhaber-Gordon1998a; @*Cronenwett1998; @Wiel2000; @Sasaki2000; @*Nygard2000; @*Jespersen2006; @*Csonka2008] and the full width at half maximum (FWHM) of the ZBA peak was assumed to be approximately equal to $2k_{\rm B}T_{\rm K}$. However, more careful measurements demonstrated an overestimation of $T_{\rm K}$ determined by this method, if compared to the values extracted from the $G(T)$ dependence [@Wiel2000]. To clarify the issue of the ZBA width we did a simple mathematical analysis of Eq. (\[simple\_formula\]), which revealed that $G(V_{sd}=\pm k_{\rm B}T_{\rm K}/e) = 0.67G_{0} \approx 2/3G_{0}$ and the ZBA peak FWHM $=2k_{\rm B}T^{*}_{\rm K} \approx 2\sqrt{\pi}k_{\rm B}T_{\rm K}$ [@Note1]. By means of these simple relations a quick and accurate estimation of $T_{\rm K}$ from the raw data is straightforward. As an example Fig. \[Fig3\](c) shows the Kondo ZBA peak measured at $V_{g}=$ -2.83 V and plotted as a function of the normalized bias $eV_{sd}/k_{\rm B}T_{\rm K}$, where $T_{\rm K}$ is found from fitting of the same data with Eq. (\[simple\_formula\]) \[see Fig. \[Fig3\](b)\]. The universal lineshape is shown in Fig. \[Fig3\](c) by the black dashed curve, illustrating an excellent agreement with the experiment. It is seen that the width of the Kondo ZBA is $2k_{\rm B}T_{\rm K}$ at two third of its total magnitude, contrary to the earlier assumed half magnitude [@Wiel2000; @Nygard2000; @*Jespersen2006; @*Csonka2008]. It also suggests an explanation to the discrepancy between the FWHM and the value of $T_{\rm K}$ reported in Ref. , alternative to the dephasing by bias [@Wingreen1994; @Kaminski2000]. Unfortunately, we were unable to measure the FWHM reliably because in our experiment the Kondo ZBA at $0.5G_{0}$ was already broadened by the background.
In conclusion, using recent results of RTRG calculations of the nonequilibrium Kondo model we experimentally verified the universal scaling of the Kondo conductance at intermediate ($e|V_{sd}|\sim k_{\rm B}T_{\rm K}$) bias. We established that the Kondo ZBA in a quantum dot can be scaled by $T_{\rm K}$ to a universal dependence predicted by the numerical calculations and approximated by the phenomenological formula. An excellent agreement with the experiment allowed us to use this formula to extract the value of $T_{\rm K}$ solely from the analysis of the Kondo ZBA lineshape at the lowest temperature. This method appears to be quicker and more statistically accurate compared to the traditional one involving the measurement of the $G(T)$ dependence. Also, a closer look at the phenomenological formula revealed that when the applied bias is equivalent to $k_{\rm B}T_{\rm K}/e$ the Kondo conductance is at two thirds of its zero-bias value. At the same time the FWHM is about $2\sqrt{\pi}k_{\rm B}T_{\rm K}$, which is larger than it was previously thought to be. We demonstrated that those relations can provide an immediate and accurate estimate of the experimental $T_{\rm K}$.
The authors would like to thank Moty Heiblum for making this research possible and for suggestions and critical remarks. We acknowledge Herbert Schoeller and Mikhail Pletyukhov for providing the results of RTRG calculations prior to their publication and for critical reading of the early version of the manuscript. We also thank Yuval Oreg and David Goldhaber-Gordon for enlightening discussions. This work was partially supported by the EU FP6 Program Grant 506095, by the Israeli Science Foundation Grant 530-08 and Israeli Ministry of Science Grant 3-66799.
[^1]: The original formula in Ref. is derived for Kondo temperature $T^{*}_{\rm{K}}$ defined as $G(eV_{sd}=k_{\rm{B}}T^{*}_{\rm{K}})=1/2G_{0}$ and $(T^{*}_{\rm{K}}/T_{\rm{K}})^{2}\approx \pi$. One can obtain the formula for $T^{*}_{\rm{K}}$ by substituting $\pi$ with 1 in the denominator of Eq. (\[HS&MP\_formula\]).
[^2]: Approximation of $G(V_{sd})$ with the unsimplified Eq. (\[HS&MP\_formula\]) gives the same value of $T_{K}$ as Eq. (\[simple\_formula\]) within its statistical error.
| {
"pile_set_name": "ArXiv"
} |
[**Quantum discord for multipartite coherent states interpolating between Werner and Greenberger-Horne-Zeilinger states** ]{}\
[**M. Daoud**]{}$^{a}$[[^1]]{} and [**R. Ahl Laamara**]{}$^{b,c}$ [[^2]]{}
$^{a}$[*Department of Physics , Faculty of Sciences, University Ibnou Zohr,\
Agadir , Morocco*]{}\
$^{b}$[*LPHE-Modeling and Simulation, Faculty of Sciences, University Mohammed V,\
Rabat, Morocco*]{}\
$^{c}$[*Centre of Physics and Mathematics, CPM, CNESTEN,\
Rabat, Morocco*]{}\
The quantum discord is used as measure of quantum correlations for two families of multipartite coherent states. The first family interpolates between generalized ${\rm
GHZ}$ states and generalized Werner states. The second one is an interpolation between generalized ${\rm GHZ}$ and the ground state of the multipartite quantum system. Two inequivalent ways to split the system in a pair of qubits are introduced. The explicit expressions of quantum quantum discord in multipartite coherent states are derived. Its evaluation uses the Koashi-Winter relation in optimizing the conditional entropy. The temporal evolution of quantum correlations (quantum discord and entanglement) is also discussed.
Introduction
============
The characterization of quantum correlations in a bipartite or multipartite quantum system has been intensively investigated in the context of quantum information science [@Horodecki-RMP-2009; @Guhne]. Different methods for quantifying these correlations were reported in many works (for a recent review see [@vedral]). Entanglement, one of the prominent of quantum correlations, constitutes a typical resource to manage information in several ways [@NC-QIQC-2000; @Vedral-RMP-2002]. In this respect, a considerable attention has been paid to develop a quantitative theory of entanglement (concurrence, entanglement of formation and linear entropy [@Rungta; @Ben3; @Wootters; @Coffman]). Entanglement was for a long time viewed as synonymous with quantum correlation. However, some recent studies showed that entanglement is only a special kind of quantum correlations. Indeed, unentangled quantum states can also possess quantum correlations which play a relevant role in improving quantum communication and information protocols better than their classical counter-parts [@Knill]-[@Datta-PRL100-2008]. Therefore, as the non-classicality of correlations present in quantum states is not due solely to the presence of entanglement, there was a need of a measure to characterize and quantify the non-classicality or quantumness of correlations which goes beyond entanglement. The first attempt in this direction was made in the works [@Vedral-et-al] and [@Ollivier-PRL88-2001] where the authors concluded that when entanglement is subtracted from total quantum correlation, there remain correlations that are not entirely classical of origin. Nowadays, it is commonly accepted that the most promising candidate to measure quantum correlations is the so-called quantum discord. It has attracted considerable attention and continues to be intensively investigated in many contexts such as quantum decoherence, quantum computation and phase transition as well as other related fields (a detailed list of references can be found in [@vedral]). Many efforts were deployed to get a closed analytical expression of quantum discord. Only some partial results were derived for some special two-qubit systems [@Vedral-et-al; @Ollivier-PRL88-2001; @Luo08; @Mazhar10]. The extension of the notion of the discord to continuous variable systems and the discussion of its properties were considered in [@Giorda]. It must be emphasized that there is now an intense recent research activity to demonstrate experimentally the advantages of the use of quantum discord without entanglement in quantum protocols. In a recent work [@Dakic], the authors experimentally show, using a variety of polarization-correlated photon pairs, non-zero quantum discord is the optimal resource for remote state preparation. More interestingly, it is found that unentangled state with non-zero quantum discord provides better performance than entangled state by achieving a higher fidelity (which is directly related to quantum discord) in remote state preparation.\
In this paper we shall be mainly concerned with the pairwise quantum correlations, especially quantum discord, present in multipartite coherent states. It must be noticed that the entangled coherent states have received a special attention in the last two decades (for a recent review, see [@Sanders3]). Therefore, paralleling the treatment of entanglement in coherent states systems, the main of this work is to investigate the pairwise quantum discord present in multipartite nonorthogonal states. In fact, entangled coherent states, which are typical examples of entangled nonorthogonal states, have attracted much attention in the literature. One may quote for instance entangled Glauber coherent states [@Sanders], $SU(2)$ and $SU(1,1)$ entangled coherent states [@Wang]. The term ‘entangled coherent state’ was introduced by Sanders in a study concerning production of entangled coherent states by using a nonlinear Mach-Zehnder interferometer [@Sanders; @Sanders2]. Entangled coherent states were initially treated as bimodal states but later generalized to superpositions of multimode coherent states [@Jex; @Zheng; @Wang1]. Generalizations to multimode systems allow the intricacies of multipartite entanglement to become manifest in entangled coherent states, as for instance in [GHZ]{} (Greenberger-Horne-Zeilinger), [W]{} (Werner) states [@Jeong1; @Li] and entangled coherent state versions of cluster states [@Munhoz; @Wang-WF; @Becerra]. In this sense, the characterization of the quantum discord in nonorthogonal multipartite states constitutes an important issue which deserves the same degree of attention as entanglement.\
This paper is organized as follows. In the first section, we introduce a special instance of superpositions of multipartite coherent states. They involve Weyl-Heisenberg, $SU(2)$ and $SU(1,1)$ coherent states labeled by a single complex parameter $z$. We shall focus on balanced superpositions which are symmetric or antisymmetric under the parity transformation $ z \longrightarrow -
z$. The symmetric superposition is interpolating between the ground state of the multipartite system and ${\rm GHZ}$ state. The antisymmetric superposition provides a continuous interpolation between the Werner state and ${\rm GHZ}$ state. To study the bipartite quantum correlations present in the system, two different qubit mapping are considered. Section 3 is devoted to the derivation of the explicit expression of the quantum discord. The optimization over all the measurement is performed by combining the purification method and the Koashi-Winter relation . Finally, in the last section we discuss the dynamical evolution of entanglement and quantum discord under a dephasing channel. Concluding remarks close this paper.
Multipartite coherent states and qubit mapping
===============================================
For several quantum systems, the coherent states can be obtained by exploiting the structure relations of the relevant dynamical group structure $G$. Therefore, any quantum state in the corresponding Hilbert space can be expanded as a sum of the coherent states associated with the group $G$. In this respect, a widespread interest was devoted to the theoretical as well experimental studies of superpositions of coherent states. Evidences of such superpositions first appeared in a study of a certain type of nonlinear Hamiltonian evolution in [@Milburn1; @Milburn2]. Also, a detailed analysis concerning the manifestation of superpositions of coherent states was reported in [@Yurke1; @Yurke2] (see also [@Buzek]). However, it is important to stress that superpositions of coherent states are experimentally difficult to produce, and fundamentally this could be due to extreme sensitivity to environmental decoherence. Some experimental efforts to create superpositions of coherent states were reported in [@Brune].
Basic of coherent states
------------------------
A system of coherent states is defined to be a set $\{ \vert z
\rangle; z \in {\cal D} \}$ of quantum states, in some Hilbert space ${\cal H}$, parameterized by the variable $z$ belonging to the set ${\cal D}$ such that: $ z \longrightarrow \vert z \rangle$ is continuous and the system is over-complete; i.e. $$\label{mesure}
\int \vert z \rangle \langle z \vert~ d\mu(z)= I_{{\cal H}}.$$ The continuity and the resolution of the unity (\[mesure\]) form the minimal set of requirements to define a system of coherent states. They are not orthogonal to each other with respect to the positive measure in (\[mesure\]). It should be noticed that the explicit form of the coherent state $\vert z \rangle$ in not required for our purpose. However, to illustrate some interesting limiting situations and for the sake of completeness, we consider the familiar sets of Perelomov coherent states [@Perelomov] associated with Weyl-Heisenberg, $SU(2)$ and $SU(1,1)$ groups (the extension to other groups and other coherent states definitions is straightforward). A simple way to deal with these symmetries in a unified scheme can be achieved by using the so-called generalized Weyl-Heisenbeg algebra ${\cal A}$ (see [@daoud] and references therein) spanned by an annihilation operator $a^-$, a creation operator $a^+$ and a number operator $N \not= a^+ a^-$ satisfying the structure relations $$\begin{aligned}
[a^- , a^+] = G(N), \quad [N, a^{-}] = - a^{-}, \quad [N, a^{+}] =
+ a^{+}, \label{generalized W-H algebra}\end{aligned}$$ with $$\begin{aligned}
a^+ = (a^-)^{\dagger}, \quad N = N^{\dagger}, \quad G(N) = F(N+1) -
F(N), \label{hermiticity relations}\end{aligned}$$ where the $F$ structure function characterizes the deviation from the usual harmonic oscillator. A representation of the algebra ${\cal A}$ which extends that of Weyl-Heisenberg, is defined through the actions $$\begin{aligned}
&& a^-\vert n \rangle = \sqrt{F(n)} \vert n - 1 \rangle, \quad a^-\vert 0 \rangle = 0,
\label{rep1} \\
&& a^+\vert n \rangle = \sqrt{F(n+1)} \vert n + 1 \rangle, \quad N \vert n \rangle = n \vert n \rangle
\label{rep3}
\end{aligned}$$ on the Hilbert space spanned by the eigenvectors of $N$. This formally defines a representation of ${\cal A}$. Note that $ a^+
a^- = F(N),$ a relation that generalizes $N = a^+ a^-$ for the harmonic oscillator. The operator $F(N)$ can be considered as the Hamiltonian for a quantum system. This provides a physical significance to the structure function $F$. The representation (Fock-Hilbert) space ${\cal H}$ is generated by the orthonormal set $ \{ | n \rangle : n \ {\rm ranging} \}$. The dimension of the representation afforded by (\[rep1\]) and (\[rep3\]) is controlled by the positiveness of the eigenvalues $F(n)$ of the operator $F(N)$. The usual Weyl-Heisenberg algebra is recovered for $$F(N) = N.$$ The algebra ${\cal A}$ coincides with $su(1,1)$ Lie algebra for $$F(N) = N(2k-1+N)$$ where the number $k$, which acquires discrete values $k=\frac{1}{2},1,\frac{3}{2},2,\ldots$, characterizes the $SU(1,1)$ discrete-series representations. The algebra ${\cal A}$ reproduces the spin algebra $su(2)$ when $$F(N) = N(2j+1-N),$$ and the creation and annihilation generators coincides with the usual raising and lowering operators acting in the spin representation of dimension $2j+1$.\
The Perelomov coherent states for the algebra ${\cal A}$ can be constructed via a simple strategy based on the use of a Fock-Bargmann space associated with ${\cal A}$ [@daoud]. This is sketched briefly as follows. Let us look for states in the form $$\begin{aligned}
\vert z \rangle = \sum_{n} a_n z^n \vert n \rangle, \quad a_n \in \mathbb{R}, \quad z \in \mathbb{C},
\label{forme des ec}
\end{aligned}$$ where the sum on $n$ is finite or infinite according to as ${\cal
A}$ admits a finite- or infinite-dimensional representation. The $a_n$ coefficients can then be determined from the correspondence rules $$\begin{aligned}
\vert n \rangle \longrightarrow a_{n} z^{n}, \qquad
a^-\longrightarrow \frac{d}{dz}
\label{correspondance pour Perelomov}
\end{aligned}$$ applied to relations (\[rep1\]) and (\[rep3\]). This analytical realization leads to the following recurrence relation $$\begin{aligned}
n a_{n}= \sqrt{F(n)}a_{n-1},
\label{recur}
\end{aligned}$$ which can be iterated to give $$\begin{aligned}
a_{n}= \frac{\sqrt{F(n)!}}{n!},
\label{recur2}
\end{aligned}$$ (by taking $a_0 = 1$). In Eq. (\[recur2\]), the generalized factorials are defined by $$F(0)! = 1, \qquad F(n)! = F(1) F(2) \ldots F(n).$$ This yields the following normalized coherent states $$\vert z \rangle = N(\vert z \vert)\sum_{n}
\frac{\sqrt{F(n)!}}{n!} z^n \vert n \rangle
\label{cs-main}$$ where the normalization factor is $$\bigg[N(\vert z \vert)\bigg]^{-2} = \sum_{n} \frac{F(n)!}{(n!)^2} \vert z\vert^{2n}.$$ subject to convergence. They satisfy $\vert z \rangle = \exp( z a^+)
\vert 0 \rangle$ and are thus coherent states in the Perelomov sense. For the usual harmonic oscillator ($F(n) = n$), the equation (\[cs-main\]) reads $$\vert z \rangle =e^{-\frac{\vert z \vert^2}{2}} \sum_{n=0}^{\infty}
\frac{z^n}{\sqrt{n!}} \vert n \rangle. \label{glauber-cs}$$ The overlap between two Glauber coherent states is $$\langle z_1 \vert z_2 \rangle = \exp(-\frac{1}{2} (\vert z_1 \vert^2
+ \vert z_2 \vert^2 - 2 \bar z_1 z_2)).\label{overlap-glauber}$$ The standard set of the $SU(2)$ coherent state is obtained from (\[cs-main\]) for $F(n) = n(2j+1-n)$. They are given by $$| z\rangle = (1+ \vert z \vert^{2})^{-j} \sum_{n=0}^{2j} \left[
\frac{(2j)!}{n!(2j-n)!} \right]^{1/2} z^{n} |n\rangle.
\label{su2-cs}$$ The parameter $z$ can acquire any complex value. The $SU(2)$ coherent states are normalized but they are not orthogonal to each other: $$\langle z_{1}|z_{2}\rangle = (1+|z_{1}|^{2})^{-j}
(1+|z_{2}|^{2})^{-j} (1 + \bar z_{1} z_{2})^{2j}.
\label{overlap-su2}$$ Finally, for $F(n)= n(2k-1+N)$, one gets the $SU(1,1)$ coherent states $$|z \rangle = (1-|z|^{2})^{k} \sum_{n=0}^{\infty}
\left[\frac{(2k-1+n)!}{n!(2k-1)!}\right]^{1/2} z^{n} |n\rangle.
\label{su11-cs}$$ where the complex variable $z$ belongs to the unit disc $\{ |z| \in
\mathbb{C}, \quad |z|<1\}$. The kernel of two $SU(1,1)$ coherent states reads $$\langle z_{1}|z_{2}\rangle = (1-|z_{1}|^{2})^{k} (1-|z_{2}|^{2})^{k}
(1 - \bar z_{1} z_{2})^{-2k}. \label{ovelap-su11}$$
Superpositions of multipartite coherent states
----------------------------------------------
The over-completeness relation makes possible the expansion of an arbitrary state of the Hilbert space ${\cal H}$ in terms of the coherent states of the system under consideration. It follows that when one considers a collection of $n$ noninteracting identical particles, the whole Hilbert space is a tensor product and any multipartite coherent state can be written as a superposition of tensorial products of the form $\vert z_1 \rangle\otimes\vert z_2
\rangle\cdots \otimes\vert z_n \rangle \equiv \vert z_1, z_2, \cdots
z_n \rangle$. In fact, the resolution of identity allows us to expand any state $\vert \psi \rangle $ in the space ${\cal H}
\otimes{\cal H} \otimes \cdots {\cal H}$ as $$\label{ncs}
\vert \psi \rangle = \int ~ d\mu(z_1) ~ d\mu(z_2)\cdots ~ d\mu(z_n)\vert z_1, z_2, \cdots z_n \rangle \langle z_1, z_2, \cdots z_n
\vert \psi \rangle$$ reflecting that any multipartite state can be viewed as a superposition of the coherent states $\vert z_1, z_2, \cdots z_n
\rangle$. The multipartite state (\[ncs\]) can be reduced to a sum if the function $ \psi (z_1, z_2, \cdots z_n) = \langle z_1,
z_2, \cdots z_n
\vert \psi \rangle$ can be expressed as a sum of delta functions. Indeed, setting $$\psi (z_1, z_2, \cdots z_n) = (\delta(z - z_1) + e^{im \pi} \delta(z + z_1)) \delta(z_1 - z_2) \delta(z_2 - z_3)\cdots \delta(z_{n-1} - z_n),$$ one gets the following balanced or equally weighted superposition of multipartite coherent states $$\vert \psi \rangle \equiv \vert z, m, n \rangle ={\cal N}(\vert z
\rangle\otimes \vert z \rangle\otimes \cdots\otimes \vert z \rangle
+e^{im\pi}|- z \rangle\otimes |- z \rangle\otimes \cdots\otimes |- z
\rangle) \label{eq:main}$$ where $m \in \mathbb{Z}$. The normalization factor ${\cal N}$ is $${\cal N} = \big[ 2 + 2 p^n \cos m \pi\big]^{-1/2}$$ where $p$ denotes the overlap between the states $\vert z \rangle$ and $\vert - z \rangle$. It is given $$p = \langle z \vert - z \rangle.$$ It is real as it can be verified from (\[cs-main\]). For Weyl-Heisenberg, $SU(2)$ and $SU(1,1)$ coherent states, the quantity $p$ is obtainable from the expressions (\[overlap-glauber\]), (\[overlap-su2\]) and (\[ovelap-su11\]), respectively. Two interesting limits of superpositions of the form (\[eq:main\]) arise when $ p \rightarrow 0$ and $ p \rightarrow 1$.
We first consider the asymptotic limit $p \rightarrow 0 $. In this limit the two states $|z \rangle $ and $|- z \rangle $ approach orthogonality, and an orthogonal basis can be constructed such that $\vert {\bf 0}\rangle\equiv \vert z \rangle$ and $\vert{\bf
1}\rangle \equiv \vert - z \rangle$. Thus, the state $ \vert z , m
,n \rangle$ approaches a multipartite state of ${\rm GHZ}$ type $$\vert z , m, n \rangle \sim \vert {\rm GHZ}\rangle_n = \frac
1{\sqrt{2}}(\vert {\bf 0}\rangle \otimes |{\bf 0}\rangle \otimes
\cdots \otimes\vert {\bf 0}\rangle
+e^{i m \pi}\vert {\bf 1}\rangle \otimes
\vert {\bf 1}\rangle \otimes \cdots \otimes
\vert {\bf 1}\rangle).\label{GHZ}$$ In the situation where $p \rightarrow 1$ (or $ z \rightarrow 0$ ), one should distinguish separately the cases $m = 0 ~({\rm mod}~2)$ and $m = 1 ~({\rm mod}~2)$. For $m$ even, the multipartite superposition (\[eq:main\]) reduces to ground state $$\vert 0 , 0 ~({\rm mod}~ 2) , n \rangle \sim \vert 0\rangle
\otimes\vert 0 \rangle \otimes \cdots \otimes \vert 0 \rangle,$$ and for $m$ odd, the state $\vert z , 1 ~({\rm mod}~ 2) ,n \rangle$ reduces to a multipartite state of Werner type [@Dur00] $$\vert 0 , 1 ~({\rm mod}~ 2), n \rangle \sim \vert\text{\rm
W}\rangle_n
= \frac{1}{\sqrt{n}}(\vert 1\rangle \otimes\vert 0 \rangle \otimes \cdots\otimes
\vert0\rangle +\vert 0\rangle \otimes\vert 1\rangle \otimes\ldots\otimes \vert0\rangle
+\cdots
+ \vert 0\rangle \otimes\vert 0\rangle \otimes \cdots\otimes \vert 1\rangle)~.
\label{Wstate}$$ Here $\vert n\rangle $ $(n=0,1)$ denote the Fock-Hilbert states.
It follows that the states $\vert z, m = 0 ~({\rm mod}~2), n\rangle$ interpolate between states of ${\rm GHZ}$ type $(p \rightarrow 0)$ and the separable state $\vert 0\rangle \otimes\vert 0 \rangle
\otimes \cdots \otimes \vert 0 \rangle$ $(p \rightarrow 1)$. In other hand, the states $\vert z , m = 1 ~({\rm mod}~2), n\rangle$ may be viewed as interpolating between states of ${\rm GHZ}$ type $(p \rightarrow 0)$ and states of Werner type $(p \rightarrow 1)$.
To close this subsection, it is important to emphasize that the main concern in investigating the properties of superpositions of coherent states is how to produce such states. Their experimental production is fundamentally difficult to achieve. This is especially due to extreme sensitivity to environmental decoherence. Experimental efforts to create superpositions of coherent states with the present day technology, are encouraging. Indeed, superpositions of weak coherent states with opposite phase, resembling to a small “Schrödinger’s cat” state (or “Schrödinger’s kitten”), were produced by photon subtraction from squeezed vacuum [@Ourjoumtsev]. Also, the experimental generation of arbitrarily large squeezed Schrodinger cat states, using homodyne detection and photon number states (two photons) as resources was reported in [@Ourjoumtsev]. Very recently, creation of coherent state superpositions, by subtracting up to three photons from a pulse of squeezed vacuum light, is reported in [@Gerrits]. The mean photon number of such coherent states produced by three-photon subtraction is of 2:75. The production of cat states especially ones of high amplitude or mean number of photons remains an experimental challenge. Considering the fast technical progress and the increasing number of groups working in this field, we expect that the generation of cat states (and Bell states) is a goal that is achievable in the near future.
Qubit mapping
-------------
### Pure states
To study the bipartite quantum correlations present in (\[eq:main\]), the whole system can be partitioned in two different ways. We first consider bipartite splitting of the multipartite system, i.e., splitting the entire system into two subsystems, one subsystem containing any $k$ $(1\le k\le n-1)$ particles and the other containing the remaining $n-k$ particles. Accordingly, one writes the state $\vert z, m, n \rangle$ as $$\label{partition1}
\vert z, m, n
\rangle = {\cal N} (\vert z \rangle_k \otimes \vert z \rangle_{n-k}
+ e^{im\pi } \vert -z \rangle_k \otimes \vert -z \rangle_{n-k})$$ where $$\vert \pm z \rangle_l = \vert \pm z \rangle_1\otimes
\vert \pm z \rangle_2\otimes \cdots\otimes \vert \pm z \rangle_l ,
\qquad l = k , n-k.$$ The multi-particle state $\vert z, m , n \rangle$ can be expressed as a state of two logical qubits. For this end, we introduce, for the first subsystem, the orthogonal basis $\{ \vert 0 \rangle_k ,
\vert 1 \rangle_k\}$ defined as $$\label{base1}
\vert 0 \rangle_k = \frac{ \vert z \rangle_k + \vert -z
\rangle_k}{\sqrt{2(1 + p^k)}}
\qquad \vert 1 \rangle_k = \frac{\vert z \rangle_k - \vert - z
\rangle_k}{{\sqrt{2(1- p^{k})}}}.$$ Similarly, we introduce, for the second subsystem, the orthogonal basis $\{ \vert 0 \rangle_{n-k} , \vert 1 \rangle_{n-k}\}$ given by $$\label{base2}
\vert 0 \rangle_{n-k} = \frac{ \vert z \rangle_{n-k} + \vert -z
\rangle_{n-k}}{\sqrt{2(1 + p^{n-k})}}
\qquad \vert 1 \rangle_{n-k} = \frac{\vert z \rangle_{n-k} - \vert - z
\rangle_{n-k}}{{\sqrt{2(1- p^{n-k})}}}.$$ Reporting the equations (\[base1\]) and (\[base2\]) in (\[partition1\]), one has the explicit form of the pure state $\vert z, m, n \rangle$ in the basis $\{ \vert 0 \rangle_{k} \otimes
\vert 0 \rangle_{n-k} ,
\vert 0 \rangle_{k} \otimes \vert 1 \rangle_{n-k} , \vert 1 \rangle_{k}
\otimes \vert 0 \rangle_{n-k} , \vert 1 \rangle_{k} \otimes \vert 1
\rangle_{n-k}\}$. It is given by $$\vert z, m, n \rangle = \sum_{\alpha= 0,1} \sum_{\beta= 0,1}
C_{\alpha,\beta} \vert \alpha \rangle_k \otimes \vert \beta
\rangle_{n-k}\label{mapping1}$$ where the coefficients $C_{\alpha,\beta}$ are $$C_{0,0} = {\cal N}(1 + e^{im\pi}) a_{k}a_{n-k} , \qquad C_{0,1} = {\cal N} (1 -e^{im\pi}) a_{k}b_{n-k}$$ $$C_{1,0} = {\cal N} (1 - e^{im\pi}) a_{n-k}b_{k} , \qquad C_{1,1} = {\cal N} (1 + e^{im\pi}) b_{k}b_{n-k}.$$ in term of the quantities $$a_l =\sqrt{\frac{1+p^l}{2}} , \qquad b_l = \sqrt{\frac{1-p^l}{2}} \qquad {\rm for} ~ l = k, n-k$$ involving the overlap $p$ between two coherent states of equal amplitude and opposite phase.
### Mixed states
The second partition can be realized by considering the bipartite reduced density matrix $\rho_{kl}$ which is obtained by tracing out all other subsystems except ones labeled by the indices $k$ and $l$. There are $n(n-1)/2$ different density matrices $\rho_{kl}$. It is simple to see that all the reduced density matrices $\rho_{kl}$ are identical. Therefore, it is sufficient to consider $\rho_{12}$ and to generalize from this case. Then, by tracing out systems $3,4,\ldots,n$ in the state $\vert z , m, n \rangle$, we obtain the reduced density matrix describing the particles or modes 1 and 2 as $$\begin{aligned}
\rho_{12} &=&\text{Tr}_{3,4,\ldots,n}(\vert z, m, n\rangle \langle
z, m,n |) \nonumber \\
&=&{\cal N}^2(\vert z , z \rangle \langle z , z \vert +\vert - z , -
z \rangle \langle - z , - z | + e^{i m \pi } q |- z , - z \rangle
\langle z , z \vert +e^{-i m \pi }q\vert z , z \rangle \langle - z,
- z \vert ) \label{rho12}\end{aligned}$$ with $q \equiv p^{n-2}$. To study the correlations of the system described by the density matrix $\rho_{12}$, we convert it into a two-qubit system. Thus, we choose an orthogonal pair $\{\vert {\bf
0}\rangle ,\vert {\bf 1}\rangle \}$ as $$\vert z \rangle \equiv a \vert {\bf 0} \rangle + b \vert {\bf 1}
\rangle ~,\; \vert - z \rangle\equiv a \vert {\bf 0} \rangle - b
\vert {\bf 1} \rangle,\label{base}$$ where $$a = \sqrt{\frac{1+p}{2}} \qquad b = \sqrt{\frac{1-p}{2}}.$$ The logical qubits $\vert {\bf 0} \rangle $ and $\vert {\bf 1}
\rangle $ given by $$\vert {\bf 0} \rangle = \frac{1}{\sqrt{2+2p}}(\vert z \rangle +
\vert -z \rangle) ~ \qquad \vert {\bf 1} \rangle =
\frac{1}{\sqrt{2-2p}}(\vert z \rangle - \vert -z
\rangle).\label{qubit-mixed}$$ coincide with even and odd coherent states, respectively. Substituting the equation (\[base\]) into (\[rho12\]), we obtain the density matrix $$\rho_{12} = {\cal N}^2 \left( \begin{smallmatrix} 2a^4(1+q\cos
m\pi)& 0
& 0& 2a^2b^2(1+q\cos m\pi
)\\
0 & 2a^2b^2(1-q\cos m\pi ) & 2a^2b^2(1-q\cos m\pi
) & 0 \\
0 & 2a^2b^2(1-q\cos m\pi ) & 2a^2b^2(1-q\cos m\pi
) & 0 \\
2a^2b^2(1+q\cos m\pi ) & 0 & 0 & 2b^4(1+q\cos m\pi)
\end{smallmatrix}
\right) \label{rho12-matrix}$$ in the basis $\{\vert{\bf 00}\rangle ,\vert{\bf 01}\rangle
,\vert{\bf 10}\rangle ,
\vert{\bf 11}\rangle \}$. It is remarkable that the obtained density belongs to the set of the so-called $X$ states. It can be also written, in the Bloch representation, as $$\rho_{12} = \sum_{\alpha = 0}^{3}\sum_{\beta = 0}^{3}
R_{\alpha,\beta} \sigma^{\alpha} \otimes \sigma^{\beta}$$ where the correlation matrix $R$ is given by $$R = \left(\begin{array}{cccc}
1 & 0 & 0 & 2{\cal N}^2p(1+ q\cos m\pi) \\
0 & 2{\cal N}^2(1-p^2) & 0 & 0 \\
0 & 0 & -2{\cal N}^2(1-p^2)q\cos m\pi & 0 \\
2{\cal N}^2p( 1+ q \cos m\pi) & 0 & 0 & 2{\cal N}^2(p^2+ q\cos m\pi) \\
\end{array}\right)\label{R-matrix}$$ and $\sigma^0$ and $\sigma^i$ ($i = 1,2,3$) stand for identity and the usual Pauli matrices, respectively.
Quantifying the quantum discord
===============================
So far, quantum discord has been calculated explicitly only for a rather limited set of two-qubit quantum states and analytical expressions for more general quantum states are not known. In this section, we give another instance of states for which quantum discord can be evaluated explicitly. Indeed, we shall derive the explicit form of this kind of bipartite correlation in the state (\[eq:main\]) by making use of the qubit mapping corresponding to the partitioning schemes (\[partition1\]) and (\[rho12\]) discussed in the previous section.
Definitions
-----------
The quantum discord is defined as the difference between total correlation and classical correlation. The total correlation is usually quantified by the mutual information $I$ $$\label{def: mutual information}
I(\rho_{AB})=S(\rho_A)+S(\rho_B)-S(\rho_{AB}),$$ where $\rho_{AB}$ is the state of a bipartite quantum system composed of the subsystems $A$ and $B$, the operator $\rho_{A(B)}={\rm Tr}_{B(A)}(\rho_{AB})$ is the reduced state of $A$($B$) and $S(\rho)$ is the von Neumann entropy of a quantum state $\rho$. Suppose that a positive operator valued measure (POVM) measurement is performed on subsystem $A$. The set of POVM elements is denoted by $\mathcal{M}=\{M_k\}$ with $M_k\geqslant 0$ and $\sum_k M_k= \mathbb{I} $. In this paper, we deal with two-qubit rank two states (with only two nonzero eigenvalues). In this case, the generalized positive operator valued measurement is not required. Indeed, it has been clearly shown in [@Hamieh; @Galve] (see also the recent review [@Modi]) that for a bipartite mixed state of two qubits of rank two , the optimal measurement giving the quantum discord is a two element POVM. The elements of such POVM are orthogonal projectors. Thus, it is natural to consider only projective measurements for the subsystem $A$. The von Neumann measurement (from now on just a measurement) on the subsystem $A$ project the system into a statistical ensemble $\{ p_{B,k} ,
\rho_{B,k}\}$ such that $$\rho_{AB} \longrightarrow \frac{(M_k \otimes \mathbb{I})\rho_{AB}(M_k \otimes \mathbb{I})}{p_{B,k}}$$ where the measurement operation is written as [@Luo08] $$\begin{aligned}
M_k = U \, \Pi_k \, U^\dagger \label{Eq:VNmsur}\end{aligned}$$ with $\Pi_k = |k\rangle\langle k| ~ (k = 0,1)$ is the projector for subsystem $A$ along the computational base $|k\rangle$, $U \in
SU(2)$ is a unitary operator and $$p_{B,k} = {\rm tr} \bigg[ (M_k \otimes \mathbb{I})\rho_{AB}(M_k \otimes \mathbb{I}) \bigg].$$ The amount of information acquired about particle $B$ is then given by $$S(\rho_B)-\sum_k ~p_{B,k} ~S(\rho_{B,k}),$$ which depends on measurement $\mathcal{M}$. This dependence can be removed by doing maximization over all the measurements, which gives rise to the definition of classical correlation $$\begin{aligned}
C(\rho_{AB})& =\max_{\mathcal{M}}
\Big[S(\rho_B)-\sum_k ~p_{B,k} ~S(\rho_{B,k})\Big] \nonumber \\
& =S(\rho^B) - \widetilde{S}_{\rm min}
\label{def: classical correlation}\end{aligned}$$ where $\widetilde{S}_{\rm min}$ denotes the minimal value of the conditional entropy
$$\widetilde{S} = \sum_k ~p_{B,k}
~S(\rho_{B,k}).\label{condit-entropy}$$
It follows that quantum discord is then given by $$\label{def: discord}
D(\rho_{AB})= I(\rho_{AB}) - C(\rho_{AB})
=S(\rho_A)+\widetilde{S}_{\rm min}-S(\rho_{AB}).$$ The main step in evaluating the quantum discord is the minimization of conditional entropy to get an explicit expression of the quantum discord in the multipartite system (\[eq:main\]).
Quantum discord for pure bipartite coherent states
---------------------------------------------------
According to the first partitioning scheme (\[partition1\]), the entire $n$ particles system is split into two components $A$, containing $k$ particles, and $B$, containing $n-k$ particles. The bipartite density state $\rho_{k,n-k} = \vert z, m, n \rangle
\langle z, m, n \vert$ is pure and the conditional density is also a pure state. This implies that the quantum conditional entropy is zero. Then, the quantum discord for the pure state $\rho_{AB}\equiv
\rho_{k,n-k}$ coincides with the von Neumann entropy of the subsystem $A$: $$D(\rho_{k,n-k}) = S(\rho_{k})$$ where $\rho_{k}$ is the reduced density of the subsystem $A$. In this scheme the quantum discord can be computed easily. It is given by $$D(\rho_{k,n-k}) = - \lambda_+ \log_2 \lambda_+ - \lambda_- \log_2
\lambda_-$$ in term of the eigenvalues of the reduced density matrix $\rho_{k}$ given by $$\lambda_{\pm}= \frac{1}{2}\bigg( 1 \pm \sqrt{1 - {\cal C}^2_{k,n-k}}
\bigg)$$ where ${\cal C}_{k,n-k}$ is the concurrence between the subsystems $A$ and $B$: $${\cal C}_{k,n-k} = 2 \vert C_{0,0} C_{1,1} - C_{1,0}C_{0,1} \vert
=\frac{\sqrt{1-p^{2k}}\sqrt{1-p^{2(n-k)}}}{1+p^n\cos
m\pi}\label{concurence1}$$ where we used the mapping (\[mapping1\]). Note that the entanglement of formation given by $$E(\rho_{k,n-k}) = H \bigg(\frac{1}{2} + \frac{1}{2}\frac{p^k +
p^{n-k}\cos m\pi}{1 + p^n\cos m\pi}\bigg),\label{qdpure}$$ is nothing but the von Neumann entropy of the subsystem $A$. Here $H$ stands for the binary entropy $ H(x)=h(x)+h(1-x)$ with $ h\left(
x\right) =-x\log_{2}x$. Hence, we have the following closed relation between quantum discord and entanglement of formation $$D(\rho_{k,n-k}) = E(\rho_{k,n-k}).$$ This agrees with the fact that quantum discord and entanglement of formation are identical for pure states and amount to the same set of correlations.\
In the limiting case $p \rightarrow 0$, the state (\[partition1\]) is of ${\rm GHZ}$ type having a maximal bipartite entanglement $({\cal C}_{k,n-k} = 1)$ and we obtain $$D(\rho_{k,n-k})= 1.$$ The situation becomes different when $p \rightarrow 1$. In this case, we have $$D(\rho_{k,n-k})= 0$$ for symmetric pure states (i.e. $m$ even) as expected (in this limit the state (\[eq:main\]) is a $n$ tensorial product of the ground state $\vert 0 \rangle$). In the limit $p \rightarrow 1$, the antisymmetric states (i.e. $m$ odd) are of Werner type. The bipartite concurrence is $${\cal C}_{k,n-k} =
\frac{2}{n}\sqrt{k(n-k)},$$ and the corresponding pairwise quantum discord takes the form $$D(\rho_{k,n-k})=
H\bigg(\frac{1}{2} + \frac{1}{2}\frac{\vert n-2k\vert}{n} \bigg).$$ The quantum correlations (entanglement and quantum discord) in multipartite Werner states vanish as $n$ becomes large.
Quantum discord for mixed bipartite coherent states
----------------------------------------------------
### Mutual information entropy
The subsystems $A$ and $B$ of the subsection 3.1 correspond here to the modes or particles 1 and 2 described by the density operator $\rho_{12}$ (\[rho12-matrix\]). The non vanishing eigenvalues of the density matrix $\rho_{12}$ are $$\lambda_{\pm} = \frac{1}{2} \frac{(1\pm p^2)(1 \pm q\cos(m\pi))}{1 +
p^n\cos(m\pi)},\label{lambda+-}$$ and the joint entropy is $$\label{entropy12}
S(\rho_{12}) = h(\lambda_+) + h(\lambda_-) = H(\lambda_+).$$ The eigenvalues of the marginal $\rho_1 = {\rm Tr}_2
\rho_{12}$ are $$\lambda_{1,\pm} = \frac{1}{2} \frac{(1\pm p)(1 \pm pq\cos(m\pi))}{1 + p^n\cos(m\pi)},$$ and the marginal entropy reads $$\label{entropy1}
S(\rho_{1}) = h(\lambda_{1,+}) + h(\lambda_{1,-}) =
H(\lambda_{1,+}).$$ The eigenvalues of the marginal $\rho_2 = {\rm Tr}_1 \rho_{12}$ are $$\lambda_{2,\pm} = \frac{1}{2} \frac{(1\pm p)(1 \pm pq\cos(m\pi))}{1 + p^n\cos(m\pi)},$$ and the corresponding entropy is given by $$\label{entropy2}
S(\rho_{2}) = h(\lambda_{2,+}) + h(\lambda_{2,-}) =
H(\lambda_{2,+}).$$ Note that for the mixed density under consideration the marginal densities $\rho_1$ and $\rho_2$ are identical. The explicit form of the mutual information (\[def: mutual information\]) writes $$I(\rho_{12})= 2 H\bigg( \frac{1}{2} \frac{(1 + p)(1 + p^{n-1}\cos(m\pi))}{1 + p^n\cos(m\pi)}\bigg)
- H\bigg( \frac{1}{2} \frac{(1 + p^2)(1 + p^{n-2}\cos(m\pi))}{1 +
p^n\cos(m\pi)}\bigg).$$
### Conditional entropy
After computing the quantum mutual information, we next compute the classical correlation $C(\rho_{12}) \equiv C(\rho_{AB})$. This requires an explicit expression of the unitary operator $U$ occurring in (\[Eq:VNmsur\]). So, we parameterizes $U$ as follows $$U = \exp(\eta \sigma_+ - \bar\eta \sigma_-) \exp(i\phi \sigma_3) \qquad \eta \in \mathbb{C}, ~ \phi \in \mathbb{R}.$$ Using this parametrization, one can verify that the quantities defined by $$\langle \sigma_i \rangle_ k = \langle k \vert U^\dagger \sigma_i U \vert k \rangle, \qquad i = 1,2, 3 \quad {\rm and} \quad k = 0,1$$ are given by $$\begin{aligned}
\langle \sigma_3 \rangle_ k = (-)^k \frac{1-\bar \alpha \alpha
}{1+\bar \alpha \alpha},\qquad \langle \sigma_1 \rangle_ k = (-)^k
\frac{\bar \alpha + \alpha}{1+\bar \alpha \alpha} \qquad , \langle
\sigma_2 \rangle_ k = i (-)^k \frac{\bar \alpha - \alpha}{1+\bar
\alpha \alpha}\label{val-moy-z}\end{aligned}$$ where $ \alpha = - i \frac{\eta}{\sqrt{\bar\eta \eta}} \tan
\sqrt{\bar\eta \eta} $. They can be also written as $$\begin{aligned}
\langle \sigma_3 \rangle_ k = (-)^k \cos \theta,\qquad \langle
\sigma_1 \rangle_ k = (-)^k \sin \theta \cos \varphi \qquad ,
\langle \sigma_2 \rangle_ k = (-)^k \sin \theta \sin \varphi
\label{val-moy-theta}\end{aligned}$$ where $\frac{\theta}{2}e^{i\varphi} = -i \eta$. The conditional density operators $\rho_{B,k} \equiv \rho_{2,k}$ are $$\begin{aligned}
\rho_{2,k} = \frac{1}{p_{2,k}}\left(\begin{array}{cc}
(1+ R_{03}) + (R_{30}+ R_{33})\langle \sigma_3\rangle_k & R_{11}\langle \sigma_1\rangle_k - i R_{22}\langle \sigma_2\rangle_k\\
R_{11}\langle \sigma_1\rangle_k + i R_{22}\langle \sigma_2\rangle_k & (1- R_{03}) + (R_{30}- R_{33})\langle \sigma_3\rangle_k \\
\end{array}\right) \label{rho2k}\end{aligned}$$ where the matrix elements $R_{\alpha \beta}$ are giving by (\[R-matrix\]) and $$\begin{aligned}
p_{2,k} = \frac{1}{2} ( 1 + R_{30}~\langle
\sigma_3\rangle_k).\label{p2k}\end{aligned}$$ It follows that the conditional entropy given by $$\begin{aligned}
\widetilde{S} = \sum_{k = 0 , 1} p_{2,k} S(\rho_{2,k})\end{aligned}$$ can be cast in the following form $$\begin{aligned}
\widetilde{S} = \sum_{k = 0 , 1} p_{2,k} ~ H \bigg(\frac{1}{2} +
\frac{1}{2} \sqrt{1 - 4\det
\rho_{2,k}}\bigg).\label{condit-entrop-tilde}\end{aligned}$$ Evidently, the quantities $p_{2,k}$ and $\det \rho_{2,k}$ can be expressed in terms of the angular directions $\theta$ and $\varphi$ by making use of equations (\[val-moy-theta\]), (\[rho2k\]) and (\[p2k\]). Consequently, we can explicitly write $\widetilde{S} =
\widetilde{S}(\theta, \varphi )$ and perform the minimization over the azimuthal and polar angles $\theta$ and $\varphi$. However, there exists another elegant method to perform such an optimization. This is presented in the following subsection.
### Minimization of conditional entropy
To minimize the conditional entropy (\[condit-entrop-tilde\]), we shall use the purification method and the Koashi-Winter relation [@Koachi-Winter] (see also [@Shi]). This relation establishes the connection between the classical correlation of a bipartite state $\rho_{AB} \equiv \rho_{12}$ and the entanglement of formation of its complement $\rho_{BC} \equiv \rho_{23}$. This connection will be clarified hereafter. The density $\rho_{12}$ (\[rho12-matrix\]) is a two-qubit state of rank two. It decomposes as $$\begin{aligned}
\rho_{12} = \lambda_+ \vert \phi_+ \rangle \langle \phi_+ \vert +
\lambda_- \vert \phi_- \rangle \langle \phi_- \vert\end{aligned}$$ where the eigenvalues $\lambda_+$ and $\lambda_-$ are given by (\[lambda+-\]) and the corresponding eigenstates $\vert \phi_+
\rangle$ and $\vert \phi_- \rangle$ are given by $$\begin{aligned}
\vert \phi_+ \rangle = \frac{1}{\sqrt{2(1+p^2)}}\bigg[ (1+p)\vert
{\bf 0} , {\bf 0} \rangle + (1-p)\vert {\bf 1} , {\bf 1} \rangle
\bigg] \qquad \vert \phi_- \rangle = \frac{1}{\sqrt{2}} \bigg[\vert
{\bf 0} , {\bf 1} \rangle +
\vert {\bf 1} , {\bf 0} \rangle \bigg]\end{aligned}$$ in the basis (\[qubit-mixed\]). Attaching a qubit $3$ to the two-qubit system $1$ and $2$, we write the purification of $\rho_{12}$ as $$\begin{aligned}
\vert \phi \rangle = \sqrt{\lambda_1} \vert \phi_+ \rangle \otimes
\vert {\bf 0} \rangle + \sqrt{\lambda_-} \vert \phi_- \rangle
\otimes \vert {\bf 1} \rangle\end{aligned}$$ such that the whole system $123$ is described by the pure density state $\rho_{123} = \vert \phi \rangle \langle \phi \vert $ from which one has the bipartite densities $\rho_{12} = {\rm Tr}_3
\rho_{123}$ and $\rho_{23} = {\rm Tr}_1 \rho_{123}$.
Suppose now that a von Neumann measurement $\{ M_0 , M_1\}$ is performed on the subsystem $1$ (here also we need positive operator valued measurement of rank one that is proportional the one dimensional projector). From the viewpoint of the whole system in the pure state $\vert \phi \rangle$, the measurement gives rise to an ensemble for $\rho_{23}$ that we denote by $${\cal E}_{23} = \{ p_k , \vert \phi_{23,k} \rangle \}$$ where $$p_k = \langle \psi \vert M_k \otimes \mathbb{I} \otimes \mathbb{I} \vert \psi \rangle\quad,
\quad \vert \phi_{23,k} \rangle \langle \phi_{23,k} \vert =
\frac{1}{p_k} {\rm Tr}_1 \bigg[ (M_k \otimes \mathbb{I} \otimes
\mathbb{I}) \vert \psi \rangle \langle \psi \vert \bigg].$$
On the other hand, from the viewpoint of the state $\rho_{12}$, the von Neuman measurement on $1$ gives rise to the ensemble for $\rho_2$ that it was defined previously as ${\cal E}_{2} = \{
p_{2,k} , \rho_{2,k} \}$ (note that $p_k = p_{2,k}$). It is simple to check that the ensemble ${\cal E}_{2} $ can be induced from ${\cal E}_{23}$ by tracing out the qubit $3$, namely $$\rho_{2,k} = {\rm Tr}_3 \bigg[\vert \phi_{23,k} \rangle \langle \phi_{23,k} \vert \bigg].$$ We denote by $E(\vert \phi_{23,k} \rangle )$ the measure of entanglement for pure states. It is given by the von Neumann entropy of the reduced subsystem $ \rho_{2,k} = {\rm Tr}_3 (\vert
\phi_{23,k} \rangle \langle \phi_{23,k} \vert)$ $$E (\vert \phi_{23,k} \rangle ) = S (\rho_{2,k}).$$ It follows that the average of entanglement of formation over the ensemble ${\cal E}_{23}$ $$\overline{E}_{23} = \sum_{k = 0, 1} p_k E (\vert \phi_{23,k} \rangle )$$ coincides with the conditional entropy (\[condit-entrop-tilde\]). At this point, it is important to note that Koachi and Winter have pointed out that the minimum value of $\overline{E}_{23}$ is exactly the entanglement of formation of $\rho_{23}$. Consequently, the minimal value of the conditional entropy coincides with the entanglement of formation of $\rho_{23}$: $$\widetilde{S}_{\rm min} = E(\rho_{23}),$$ which is easy to evaluate. Indeed, we get $$\label{stild-min}
\widetilde{S}_{\rm min} = E(\rho_{23}) = H(\frac{1}{2} + \frac{1}{2} \sqrt{1 - \vert C(\rho_{23})\vert^2})$$ where the concurrence of the density $\rho_{23}$ is $$\vert C(\rho_{23})\vert^2 = \frac{p^{2}(1 - p^{2})(1 - p^{2n-4})}{(1+p^{n}\cos m\pi)^2}$$ Using the equation (\[condit-entrop-tilde\]), one can verify the relation $$\widetilde{S}_{\rm min} = \widetilde{S} (\theta = \frac{\pi}{2}, \phi = 0)$$ which reflects that the minimal value of conditional entropy is reached for $\theta = \frac{\pi}{2}$ and $\phi = 0$. Finally, reporting (\[entropy12\]), (\[entropy1\]) and (\[stild-min\]) in the definition (\[def: discord\]), the explicit expression of quantum discord for the density $\rho_{12}$ is $$\begin{aligned}
\label{qdfinal-mixe}
D(\rho_{12}) = H\bigg(\frac{1}{2} \frac{(1+ p)(1 +
p^{n-1}\cos(m\pi))}{1 + p^n\cos(m\pi)}\bigg) - H\bigg(\frac{1}{2}
\frac{(1 + p^2)(1 + p^{n-2}\cos(m\pi))}{1 +
p^n\cos(m\pi)}\bigg)\label{qdmixte}\\\nonumber + H\bigg(\frac{1}{2}
+ \frac{1}{2} \sqrt{1 - \frac{p^{2}(1 - p^{2})(1 -
p^{2n-4})}{(1+p^{n}\cos m\pi)^2}})\bigg)\end{aligned}$$ in term of the the overlap $p$. Clearly, this result can be also derived using the method developed (independently) in [@Mazhar10] and [@Fanchini] (see also [@B.Li]) to obtain the quantum discord for some special instances of the so-called $X$ states. However, one should stress that our method based on the purification trick combined with the Koashi-Winter relation reduce drastically the optimization process of the conditional entropy. It provides a direct relation between quantum discord and entanglement of formation reflecting how these correlations are distributed in a given tripartite pure system. It has been used in several papers dealing with quantum correlations as for instance in [@Fanchini2] where the authors emphasized its crucial role in exploring the distribution of entanglement in the deterministic quantum computation with one single pure qubit and a collection of an arbitrary number of mixed states.
### Some particular cases
We start with the special case $n = 2$. The state (\[rho12\]) is pure and $q = 1$. The quantum discord (\[qdmixte\]) reads $$D(\rho_{12}) = H\bigg(\frac{(1+p)(1 + p\cos m\pi)}{2(1 + p^2 \cos m\pi)}\bigg).$$ It coincides with the quantum discord $D(\rho_{k,n-k})$ given by the equation (\[qdpure\]) for $n = 2$ and $k=1$. In particular, for symmetric states ($m$ even), one obtains $$D(\rho_{12}) = D(\rho_{1,1})= H\bigg(\frac{(1 + p)^2}{2(1 + p^2 )}\bigg)$$ and $ D(\rho_{12}) = D(\rho_{1,1}) = 1$ for antisymmetric states ($m$ odd) as shown in the figure 1.
Here also, it is interesting to consider the limiting cases $p
\rightarrow 0$ and $p \rightarrow
1$ as it was done previously for pure states. For $n > 2$, the quantum discord vanishes (\[qdfinal-mixe\]) when $p \rightarrow
0$. It vanishes also when $p \rightarrow 1$ for $m$ even. However, for $m$ odd, the quantum discord (\[qdfinal-mixe\]) reduces to the special form $$D(\rho_{12}) = H\bigg(1 - \frac{1}{n}\bigg) + H\bigg(\frac{1}{2} +
\frac{1}{2} \frac{\sqrt{n^2 - 4n + 8}}{n}\bigg) - H\bigg(1 -
\frac{2}{n}\bigg)\label{qd-Wstates}$$ and goes to zero for $n$ large.
To corroborate our analysis, we give in the figures 2 and 3 the behavior of the quantum discord (\[qdfinal-mixe\]). Figure 2 gives a plot of quantum discord versus the overlap $p$ for the mixed state $\rho_{12}$ with $m$ even (the symmetric case). As seen from the figure, after an initial increasing, the quantum discord decreases to vanish when $p \rightarrow 1$. The maximum of quantum discord depends on the number of particles contained in the system. It is remarkable that for $n = 25$ the maximum is larger than ones obtained for $n = 4$ and $n = 5$. Beside the numerical results reported in the figure 2, we have also studied the behavior of quantum discord for other values of $n$. This study shows that for $
6 \leq n \leq 25 $ the maximum of quantum discord is greater than one reached for $n = 5$. Also, the behavior of quantum discord for $
n \geq 20$ is very close to the case $n = 25$ presented in the figure 2. In figure 3, we give a plot of the quantum discord (\[qdfinal-mixe\]) for $m$ odd (the antisymmetric case) and different values of $n$. In this case the quantum discord increases as $p$ increases for a small number of particles and the maximal value of quantum discord is reached for $p \rightarrow 1$. However, for a higher number of particles $n$ ($n = 25$ for instance), the maximum is reached for $p < 1$. In the limit $p \rightarrow 1$, we have a Werner state $\vert W \rangle_n$ and the pairwise quantum discord given by (\[qd-Wstates\]) decreases to vanish in the limit of a large number of particles ($ n \rightarrow \infty $). For $p
\rightarrow 0$, corresponding to ${\rm GHZ}$ type states, the quantum discord is zero.
{width="4in"}\
FIG. 1: [The pairwise quantum discord $D$ versus the overlap $p$ for $n=2$.]{}
{width="4in"}\
FIG. 2: [The pairwise quantum discord $D$ versus the overlap $p$ for symmetric states.]{}
{width="4in"}\
FIG. 3: [The pairwise quantum discord $D$ versus the overlap $p$ for anti-symmetric states.]{}
Dynamics of quantum correlations under dephasing channel
========================================================
The sudden disappearance of entanglement is one of the most intriguing features in quantum mechanics. In fact, it has been observed that in a pair of entangled qubits, interacting with noisy environments, entanglement can disappear in a finite time [@Yu]. This phenomenon, termed in the literature “entanglement sudden death”, was experimentally confirmed [@Almeida].\
In this section, we investigate the dynamics of bipartite quantum correlations (entanglement and quantum discord) of the multipartite coherent states $\vert z , m, n\rangle$ ($n>2$). We focus on the second partitioning scheme (\[rho12\]) where mixed states emerge. The quantum discord and entanglement are two different correlations for mixed states (contrarily to the pure case). We consider the evolution of a mixed bipartite system under a dephasing dissipative channel. In this order, we use the Kraus operator approach [@NC-QIQC-2000] which describes conveniently the dynamics of two qubits interacting independently with individual environments. The time evolution of the bipartite density $\rho_{12} \equiv
\rho_{12}(0)$ (\[rho12-matrix\]) can be written compactly as $$\rho_{12}(t) = \sum_{\mu, \nu}E_{\mu , \nu}(t) ~\rho_{12}(0)~ E_{\mu , \nu}^{\dagger}(t)$$ where the so-called Kraus operators $$E_{\mu , \nu}(t) = E_{\mu}(t)\otimes E_{\nu}(t) \qquad \sum_{\mu,
\nu}E_{\mu , \nu}^{\dagger} E_{\mu , \nu} = \mathbb{I}.$$ The operators $E_{\mu}$ describe the one-qubit quantum channel effects. The non-zero Kraus operators for a dephasing channel are given by $$E_0 = {\rm diag}( 1 , \sqrt{1 - \gamma}) \qquad E_1 = {\rm diag}( 0 , \sqrt{\gamma})$$ with $ \gamma = 1 - e^{-\Gamma t}$ and $\Gamma$ denoting the decay rate. It is easy to check that the density matrix (\[rho12-matrix\]) evolves as $$\rho_{12} (t) = {\cal N}^2 \left( \begin{smallmatrix} 2a^4(1+q\cos
m\pi)& 0
& 0& 2 (1 -
\gamma)a^2b^2(1+q\cos m\pi
)\\
0 & 2a^2b^2(1-q\cos m\pi ) & 2 (1 - \gamma)a^2b^2(1-q\cos m\pi
) & 0 \\
0 & 2 (1 - \gamma)a^2b^2(1-q\cos m\pi ) & 2a^2b^2(1-q\cos m\pi
) & 0 \\
2 (1 - \gamma)a^2b^2(1+q\cos m\pi ) & 0 & 0 & 2b^4(1+q\cos m\pi)
\end{smallmatrix}
\right)$$ Using the prescription provided in the works [@Wootters98] and [@Hil97] to measure the amount of entanglement in bipartite quantum states, one can check that the concurrence is given by $$C(t) = 2~ {\rm max} \{ 0 , \Lambda_1(t) , \Lambda_2(t) \}$$ where $$\Lambda_1(t) = 2 {\cal N}^2 a^2b^2 \bigg[ (1 - \gamma)(1+q\cos m\pi ) - (1-q\cos m\pi )\bigg],$$ $$\Lambda_2(t) = 2 {\cal N}^2 a^2b^2 \bigg[ (1 - \gamma)(1-q\cos m\pi ) - (1+q\cos m\pi )\bigg].$$ It follows that the concurrence is given by $$C (t)= \frac{1}{2}~ \frac{1 - p^2}{1 + p^n\cos m\pi}\bigg[ e^{-\Gamma t}(1 + p^{n-2}) - (1-p^{n-2})\bigg]$$ for $$t < t_0 = \frac{1}{\Gamma} [ ~\ln(1 + p^{n-2}) - \ln(1 - p^{n-2})]$$ and the system is entangled. However, for $ t > t_0$, the concurrence is zero and the entanglement disappears, i.e. the system is separable. This clearly reflects that under dephasing channel, the entanglement suddenly vanishes. Note that the bipartite system under consideration is initially (in the absence of an external interaction) entangled. Indeed, for $ t = 0$, the concurrence is $$C(0) = \frac{p^{n-2} - p^n}{1 + p^n\cos m\pi},$$ and is always non zero except in the limiting cases $ p \rightarrow
0$ or $ p \rightarrow 1$ for $m$ even. Notice that for $m$ odd, the concurrence is zero for $p \rightarrow 0$ but does not vanish when $ p \rightarrow 1$ and it is given by $2/n$.
It is important to stress that the quantum discord $D(\rho_{12}(0))$ is nonzero except in the particular case $ p \rightarrow 0$. To show that the quantum discord does not disappear after the interaction of the system with the dissipative channel, we note that the density matrix $\rho_{12}(t))$ belongs to the class of the so-called circulant states [@Bylicka]. Thus, one can use the vanishing quantum discord criteria, discussed in [@Bylicka], to check that the state $\rho_{12}(t)$ has vanishing quantum discord if and only if $ p \rightarrow 0$. This implies that even when entanglement suddenly disappears in a finite time, quantum discord does not vanish. This agrees with the commonly accepted fact that the quantum discord is more robust than entanglement to sudden death under a dissipative channel.
Concluding remarks
===================
In this paper, we have obtained explicit expressions of quantum discord for symmetric and antisymmetric superpositions of multipartite coherent states. These states cover the generalized ${\rm GHZ}$ and generalized Werner states. In particular, the balanced antisymmetric superpositions ($m$ odd) interpolate continuously between generalized ${\rm GHZ}$ and generalized Werner states. The key point in determining the bipartite quantum correlations is based on the splitting of the entire system in two qubit subsystems. Two inequivalent splitting schemes were discussed. The first one leads to a pure bipartite density and the quantum discord coincides with the entanglement of formation. The second consists in constructing bipartite systems by a trace procedure keeping only the modes in which we are interested. In this way, mixed states are obtained and the corresponding quantum discord was explicitly derived. This derivation requires an optimization over all the measurement needed to extract the amount of quantum correlation which is general difficult to perform. To avoid such a difficulty, we used the purification method together with the Koashi-Winter relation which are advantageous in simplifying the minimization process of the conditional entropy. In the last part of the paper, we discussed the robustness of the quantum discord present in multipartite coherent states in comparison with the entanglement. We have shown that in sending the system through a dephasing channel the entanglement can be lost. This is not the case of quantum discord which behaves more robust against dissipative channels. It must be noticed that the results obtained here can be extended easily to many other classes of coherent states even those associated with higher symmetries which are labeled by several variables. Finally, we stress that the bipartite correlation does not capture genuine multipartite correlations. It follows that it is interesting to investigate the measure of genuine multipartite quantum discord for arbitrary multipartite non orthogonal states in the spirit of the results recently obtained in [@Ma].
[99]{}
R. Horodecki, P. Horodecki, M. Horodecki and K. Horodecki, Rev. Mod. Phys. [**81**]{}(2009) 865.
O. Gühne and G. Tóth, Phys. Rep. [**474**]{} (2009) 1.
K. Modi, A. Brodutch, H. Cable, T. Paterek and V. Vedral, [Quantum discord and other measures of quantum correlation]{}, [arXiv:1112.6238]{}.
M.A. Nielsen and I.L. Chuang, [*Quantum Computation and Quantum Information*]{} (Cambridge Univ. Press, Cambridge, 2000).
V. Vedral, Rev. Mod. Phys. [**74**]{} (2002) 197.
P. Rungta, V. Buzek, C.M. Caves, M. Hillery and G.J. Milburn, Phys. Rev. A [**64**]{} (2001) 042315.
C.H. Bennett, D.P. DiVincenzo, J. Smolin and W.K. Wootters, Phys. Rev. [**A 54**]{} (1996) 3824.
W.K. Wootters, Phys. Rev. Lett. [**80**]{} (1998) 2245.
V. Coffman, J. Kundu and W.K. Wootters, Phys. Rev. A [**61**]{} (2000) 052306.
E. Knill and R. Laflamme, Phys. Rev. Lett. [**81**]{} (1998) 5672.
S.L. Braunstein, C.M. Caves, R. Jozsa, N. Linden, S. Popescu and R. Schack, Phys. Rev. Lett. [**83**]{} (1999) 1054.
C.H. Bennett, D.P. DiVincenzo, C.A. Fuchs, T. Mor, E. Rains, P.W. Shor, J.A. Smolin and W.K. Wootters, Phys. Rev. A [**59**]{} (1999) 1070.
D.A. Meyer, Phys. Rev. Lett. [**85**]{} (2000) 2014.
E. Biham, G. Brassard, D. Kenigsberg and T. Mor, Theor. Comput. Sci. 320 (2004) 15.
A. Datta, S.T. Flammia and C.M. Caves, Phys. Rev. A [**72**]{}, 042316 (2005); A. Datta and G. Vidal, [*ibid*]{}. [**75**]{}(2007) 042310; A. Datta, A. Shaji and C.M. Caves, Phys. Rev. Lett. [**100**]{} (2008) 050502.
L. Henderson and V. Vedral, J. Phys. A [**34**]{}(2001) 6899; V. Vedral, Phys. Rev. Lett. [**90**]{} (2003) 050401; J. Maziero, L. C. Celéri, R.M. Serra and V. Vedral, Phys. Rev A [**80**]{} (2009) 044102.
H. Ollivier and W.H. Zurek, Phys. Rev. Lett. [**88**]{} (2001) 017901.
S. Luo, Phys. Rev. A **77** (2008) 042303; Phys. Rev. A **77** (2008) 022301.
M. Ali, A.R.P. Rau and G. Alber, Phys. Rev. A **81** (2010) 042105.
P. Giorda and M. G. A. Paris, Phys. Rev. Lett. [**105**]{} (2010) 020503.
B. Dakic, Y.O. Lipp, X. Ma, M. Ringbauer, S. Kropatschek, S. Barz, T. Paterek, V. Vedral, A. Zeilinger, C. Brukner and P. Walther, Quantum discord as optimal resource for quantum communication, [arXiv:1203.1629]{} (2012).
B.C Sanders, J. Phys. A: Math. Theor. [**45**]{} (2012) 244002.
B.C. Sanders, Phys. Rev. A 45 (1992) 6811.
X. Wang, B. C. Sanders and S. H. Pan, J. Phys. A 33 (2000) 7451.
B.C. Sanders, Phys. Rev. A [**46**]{} (1992) 2966.
I. Jex, P. T" orm" a and S. Stenholm, J. Mod. Opt. [**42**]{} (1995) 1377.
S.-B. Zheng, Quant. Semiclass. Opt. B: J. European Opt. Soc. B [**10**]{} (1998) 691 .
X. Wang and B.C. Sanders, Phys. Rev. A [**65**]{} (2001) 012303.
H. Jeong and N. B. An, Phys. Rev. A [**74**]{} (2006) 022104 .
H.-M. Li, H.-C. Yuan and H.-Y. Fan, Int. J. Theor. Phys. [**48**]{} (2009) 2849.
P. P. Munhoz, F. L. Semião and Vidiello, Phys. Lett. A [**372**]{} (2008) 3580.
W.-F. Wang, X.-Y. Sun and X.-B. Luo, Chin. Phys. Lett. [**25**]{}(2008) 839.
E.M. Becerra-Castro, W.B. Cardoso, A.T. Avelar and B. Baseia, J. Phys. B: At. Mol. Opt. Phys. [**41**]{} (2008) 085505.
G.J. Milburn, Phys. Rev. A [**33**]{} (1985) 674.
G.J. Milburn and C.A. Holmes, Phys. Rev. Lett. [**56**]{} (1986) 2237.
B. Yurke and D. Stoler, Phys. Rev. Lett. [**57**]{} (1986) 13.
B. Yurke and D. Stoler, Phys. Rev. A [**35**]{} (1987) 4846.
V. Bu$\tilde{z}$ek and P. L. Knight, Progress in Optics [**34**]{} (1995) 1.
M. Brune, E. Hagley, J. Dreyer, X. Maitre, A. Maali, C. Wunderlich, J. M. Raimond and S. Haroche, Phys. Rev. Lett. [**77**]{} (1996) 4887.
A. Perelomov, [*Generalized Coherent States and Their Applications*]{}, Springer, Berlin, 1986.
M. Daoud and M. Kibler, [*Bosonic and k-fermionic coherent states for a class of polynomial Weyl-Heisenberg algebras*]{}, [arXiv:1110.4799]{}, To appear in J. Phys. A (2012).
W. Dür, G. Vidal and J.I. Cirac, Phys. Rev. A [**62**]{} (2000) 062314.
A. Ourjoumtsev, H. Jeong, R. Tualle-Brouri and P. Grangier, Nature [**448**]{} (2007) 784.
T. Gerrits, S. Glancy, T.S. Clement, B. Calkins, A.E. Lita, A.J. Miller, A. L. Migdall, S.W. Nam, R.P. Mirin and E. Knill, Phys. Rev. A [**82**]{} (2010) 031802(R).
S. Hamieh, R. Kobes and H. Zaraket, Phys. Rev. A [**70**]{} (2004) 052325.
F. Galve, G. Giorgi and R. Zambrini, EPL [**96**]{} (2011) 40005
K. Modi, A. Brodutch, H. Cable, T. Paterek and V. Vedral, Quantum discord and other measures of quantum correlation, [arXiv:1112.6238]{}.
M. Koachi and A. Winter, Phys. Rev. A [**69**]{} (2004) 022309.
M. Shi, W. Yang, F. Jiang and J. Du, J. Phys. A: Math. Theor. [**44**]{} (2011) 415304.
F.F. Fanchini, T. Werlang, C.A. Brasil, L.G.E. Arruda and A.O. Caldeira, Phys. Rev. A. [**81**]{} (2010) 052107.
B. Li, Z-X Wang and S-M Fei, Phys. Rev. A [**83**]{} (2011) 022321.
F.F. Fanchini, M.F. Cornelio, M.C. de Oliveira and A.O. Caldeira, Phys. Rev. A [**84**]{} (2011) 012313.
T. Yu and J.H. Eberly, Phys. Rev. Lett. [**97**]{} (2006) 140403 .
M.P. Almeida, F. de Melo, M. Hor-Meyll, A. Salles, S.P. Walborn, P.H. Souto Ribeiro and L. Davidovich, Science [**316**]{} (2007) 579.
W.K. Wootters, Phys. Rev. Lett. [**80**]{} (1998) 2245; W.K. Wootters, Quant. Inf. Comp. [**1**]{} (2001) 27.
S. Hill and W.K. Wootters, Phys. Rev. Lett. [**78**]{} (1997) 5022.
B. Bylicka, D. Chruściński, Circulant states with vanishing quantum discord, [arXiv:1104.1804]{}.
Z-H. Ma and Z-H. Chen, Witness for a measure of genuine multipartite quantum discord for arbitrary $N$ partite quantum state, [ArXiv:1108.4323]{}
[^1]: email: [m$_{-}[email protected]]{}
[^2]: email: [[email protected]]{}
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We have found that the minimum energy configuration of $N=395$ charges confined in a disk and interacting via the Coulomb potential, reported by Cerkaski et al. in Ref. [@Cerkaski15] is not a global minimum of the total electrostatic energy. We have identified a large number of configurations with lower energy, where defects are present close to the center of the disk; thus, the formation of a hexagonal core and valence circular rings for the centered configurations, predicted by the model of Ref. [@Cerkaski15], is not supported by numerical evidence and the configurations obtained with this model cannot be used as a guide for the numerical calculations, as claimed by the authors.'
author:
- Paolo Amore
title: 'Comment to “Thomson rings in a disk”'
---
In a recent paper, ref. [@Cerkaski15], Cerkaski et al. studied the problem of a finite number of equal charges, interacting via the Coulomb potential and confined inside a disk. This problem has been previously studied by several authors in a series of papers, refs. [@Berezin85; @Wille85; @Nurmela98; @Oymak00; @Oymak01; @Worley06; @Moore07], and it can be regarded as a generalization of the well–known Thomson problem [@Thomson] (finding the configurations of minimum energy of $N$ equal charges on the surface of a sphere). Despite the apparent simplicity, both problems provide a serious computational challenge, of increasing difficulty with $N$: in particular, the number of local minima of the total electrostatic energy grows very fast with $N$ (for the case of the Thomson problem see for example the discussion in Ref. [@Calef15]). As a result, the identification of the global minimum of a system of $N$ charges typically requires extensive numerical calculations: in the absence of a formal criterium to establish whether a given configuration of equilibrium is a global minimum, one has to repeat the numerical calculations several times, keeping $N$ fixed, and regard the configuration with lowest energy among those obtained as a probable candidate for a global minimum.
For the case of the disk, Erkoc and Oymak [@Oymak00; @Oymak01] have observed the tendency, for systems with modest number of charges ($N \leq 109$), to accomodate the charges on concentric rings, empirically deducing the rules for the distribution of charges on the disk (incidentally, most of the energies reported by these authors in Tables [@Oymak01] do not correspond to global minima). The analysis performed by Cerkaski et al. in Ref. [@Cerkaski15] is a refinement of the work of Erkoc and Oymak [@Oymak00; @Oymak01], and it relies on the hypothesis that charges arrange on concentric rings, for configurations of minimum energy. In this way, the original problem is reduced to the much simpler problem of calculating the electrostatic energy due to $p$ rings, each carrying an appropriate number of charges, uniformely distributed over the ring; the equilibrium configuration in this case is obtained by solving a system of two equations (their eqs. (16) and (17)).
To test their model the authors have performed numerical (molecular dynamics, MD for short) calculations for systems up to $N=400$ charges. Based on this analysis they conclude that their approach allows one “to determine with high accuracy the equilibrium configurations of a few hundred charged particles”. In particular, for $N \gtrsim 200$ their approach “predicts the formation of the hexagonal core and valence circular rings for the centered configurations”, with “an increasing sequence of rings, starting from the center, matching the regular hexagonal pattern”. For the case of $N=395$, discussed at length in Ref. [@Cerkaski15] the authors observe the formation of an hexagonal structure with rings $\left\{1,6,12,18,24\right\}$. Fig. 2b of Ref. [@Cerkaski15], that reports a comparison between the model and the numerical MD calculations for $N=395$ charges, displays an excellent agreement between the two, with only a small mismatch just outside the hexagonal structure (the green region in the figure is used to highlight the hexagonal structure). The energy reported by the authors for this configuration, which is expected to be a global minimum of the total energy, is $\mathcal{E}_{MD} = 110665.1$, compared to the energy $\mathcal{E}_{avg} = 110667.6$, obtained with their model, with an error of just $2 \times 10^{-3} \%$.
With the purpose of veryfying the results of Ref. [@Cerkaski15] we have carried out extensive numerical calculations, in particular for the case of $395$ charges. The approach that we have implemented allows one to generate configurations with the desired number of charges on the border: in this way we have verified that the lower energies occurr when $N_p=147$ charges are disposed on the border of the disk, in agreement with Ref. [@Cerkaski15]. We have thus generated $3001$ configurations with $N_p=147$, starting from initial configurations where the internal charges are randomly distributed, and we have found that $824$ of them have energy lower that the value reported in Ref. [@Cerkaski15], $\mathcal{E} < \mathcal{E}_{MD} = 110665.1$. The lowest energy among those that we have calculated (possibly a global minimum) is $\mathcal{E}_{\rm MIN} = 110664.44$. The histogram in Fig. \[Fig\_1\] illustrates these points. Interestingly, we have also found that even the configuration with largest energy has an energy slightly lower that the value predicted by the model of Ref. [@Cerkaski15], $\mathcal{E} = 110667.576 < \mathcal{E}_{avg}$ (see Fig. \[Fig\_1\]).
In Fig. \[Fig\_2\] we display the configuration with the lowest energy among those calculated (an hexagonal grid is also plotted, to facilitate the identification of a centered hexagonal structure); note that the color of the vertices, representing the charges, depends on the number of nearest neighbors. Studying this figure, we observe the presence of defects very close to the center of the disk, and of a single, slightly deformed, hexagonal cell (the yellow region), centered at the origin, in sharp contrast with the numerical and theoretical observations of Ref. [@Cerkaski15]. Additionally, we have also found that similar behaviors are also observed for the configurations with slightly larger energy.
We summarize our main findings:
- the occurrence of defects, even very close to the center of the disk, may help to lower the total energy, while disrupting the hexagonal structure;
- all the configurations with $N_p=147$ that we have calculated have energy lower than the theoretical value obtained in Ref. [@Cerkaski15] and about $27 \%$ of them has energy lower than the numerical MD value of Ref. [@Cerkaski15];
- the use of the model of Ref. [@Cerkaski15] as a guide for the numerical MD calculation (which is also claimed to cut the CPU times by a factor $10^3$), as suggested by the authors, is definitively unjustified. Using this approach, we may expect that the solutions not only will correspond to local minima of the energy, but they will also be strongly biased (i.e. the more symmetric structures could be favored);
- the number of configurations grows very fast with $N$, thus requiring an efficient numerical approach: our program allows to generate configurations with a desired number of charges on the border. We have found that in this way the performance is drastically improved;
In light of this findings, the validity of the model of Ref. [@Cerkaski15] must be questioned, particularly for $N \gtrsim 200$; it also appear clear the inadeguacy of the numerical calculation of Ref. [@Cerkaski15], which has failed to identify a very large number of configurations with energy lower than the one reported by the authors (about $27 \%$ of the configurations that we have calculated have lower energy than the one of Ref. [@Cerkaski15]!). We are not sure whether this problem has been triggered by using the output of the model as a guide for the numerical calculation or if the program used by the authors produced a small number of configurations with $N_p=147$.
![(color online) Histogram for the energies of the configurations with $395$ charges with $N_p=147$.[]{data-label="Fig_1"}](histogram.pdf){width="7cm"}
![(color online) Numerical solution corresponding to a configuration of $395$ charges with energy $\mathcal{E} = 110664.44$[]{data-label="Fig_2"}](fig1.pdf){width="8cm"}
Acknowledgments {#acknowledgments .unnumbered}
===============
This research was supported by the Sistema Nacional de Investigadores (México).
[99]{} Cerkaski, M., R. G. Nazmitdinov, and A. Puente, Physical Review E 91.3 (2015): 032312 A.A. Berezin, Nature [**315**]{}, 104 (1985) L.T. Wille and J. Vennik, J. Phys. A: Math. Gen. [**18**]{}, L1113 (1985) K.J. Nurmela, J. Phys. A: Math. Gen. [**31**]{}, 1035 (1998) S. Erkoc and H. Oymak, Phys. Rev.[**E**]{} 62, R3075-R3076 (2000) S. Erkoc and H. Oymak, Physics Letters A 290.1 (2001): 28-34 A. Worley, arXiv:physics/0609231 (2006) Mughal, A. and M. A. Moore, Physical Review E 76.1 (2007): 011606. J.J. Thomson, Philos. Mag. [**7**]{}, 237 (1904) M. Calef, W. Griffiths and A. Schultz, J. Stat. Phys. [**160**]{}, 239-253 (2015)
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
For a polytope $P$ a simplex ${\Sigma}$ with vertex set ${{\ensuremath{\mathcal{V}}}\xspace}({\Sigma})$ is called a *special simplex* if every facet of $P$ contains all but exactly one vertex of ${\Sigma}$.
For such polytopes $P$ with face complex ${{\ensuremath{\mathcal{F}}}\xspace}(P)$ containing a special simplex the subcomplex ${{\ensuremath{\mathcal{F}}}\xspace}(P) {\backslash}{{\ensuremath{\mathcal{V}}}\xspace}({\Sigma})$ of all faces not containing vertices of ${\Sigma}$ is the boundary of a polytope $Q$ — the *basis polytope* of $P$. If additionally the dimension of the affine basis space of ${{\ensuremath{\mathcal{F}}}\xspace}(P) {\backslash}{{\ensuremath{\mathcal{V}}}\xspace}({\Sigma})$ equals $\dim(Q)$, we call $P$ *meek*; otherwise we call $P$ *wild*.
We give a full combinatorial classification and techniques for geometric construction of the class of meek polytopes with special simplices. We show that every wild polytope $P'$ with special simplex can be constructed out of a particular meek one $P$ by intersecting $P$ with particular hyperplanes. It is non–trivial to find all these hyperplanes for an arbitrary basis polytope; we give an exact description for 2–basis polytopes. Furthermore we show that the $f$–vector of each wild polytope with special simplex is componentwise bounded above by the $f$–vector of a particular meek one which can be computed explicitly. Finally, we discuss the $n$–cube as a non–trivial example of a wild polytope with special simplex and prove that its basis polytope is the zonotope given by the Minkowski sum of the $(n-1)$–cube and vector $(1,\ldots,1)$.
Polytopes with special simplex have applications on Ehrhart theory, toric rings and were just used by Francisco Santos to construct a counter–example disproving the Hirsch conjecture.
author:
- Timo de Wolff
bibliography:
- './../References.bib'
title: Polytopes with special simplices
---
Introduction {#SecIntroduction}
============
Studying the combinatorics of polytopes or of classes of polytopes is a main topic in polytope theory with many applications (as a general reference see, e.g., [@Gruenb1; @Zieg1]). In this paper we study *polytopes with a special simplex* which have been introduced by C. Athanasiadis in his work on Ehrhart series, $h$-vectors of polytopes and Gorenstein toric rings [@Atha1].\
For a polytope $P$, let ${{\ensuremath{\mathcal{V}}}\xspace}(P)$ denote the vertex set of $P$, ${{\ensuremath{\mathcal{F}}}\xspace}(P)$ denote the face complex and ${{\ensuremath{\mathcal{F}}}\xspace}(\partial P)$ denote the boundary complex of $P$ (see e.g. [@Zieg1 p. 129]). Similarly, let ${{\ensuremath{\mathcal{V}}}\xspace}({{\ensuremath{\mathcal{F}}}\xspace}(P))$ denote the vertex set of any polytopal complex ${{\ensuremath{\mathcal{F}}}\xspace}(P)$. For every set ${\sigma}\subseteq {{\ensuremath{\mathcal{V}}}\xspace}({{\ensuremath{\mathcal{F}}}\xspace}(P))$ consisting of vertices of a polyhedral complex ${{\ensuremath{\mathcal{F}}}\xspace}(P)$, we denote the maximal subcomplex of ${{\ensuremath{\mathcal{F}}}\xspace}(P)$ which does not contain vertices in ${\sigma}$ as ${{\ensuremath{\mathcal{F}}}\xspace}(P) {\backslash}{\sigma}$.
Let $P$ be an $n$–polytope in ${{\ensuremath{\mathbb{R}}}\xspace}^q$ and let ${\Sigma}$ be a simplex spanned by $m+1$ vertices of $P$. Then we call ${\Sigma}$ a *special simplex* in $P$, if every facet of $P$ contains exactly $m$ vertices of ${\Sigma}$ (i.e. every facet contains all but one vertex of ${\Sigma}$). We define $$\begin{aligned}
\operatorname{\mathcal{S}}_{(n,m)} & := & \{P \ | \ P \text{ is a polytope with some special simplex } {\Sigma}, \dim(P) = n, \dim({\Sigma}) = m\}\end{aligned}$$
\[DefSpezSimpl\]
In [@Atha1] Athanasiadis proves using the results of Stanley ([@Stan1]) as well as Reiner and Welker ([@ReiWel1]) that for a compressed $n$–polytope with special $m$–simplex the Ehrhart series can be written as $$\begin{aligned}
\sum_{r \geq 0} i(P,r)t^r & = & \frac{h(t)}{(1-t)^{n-1}}\end{aligned}$$ with $h(t)$ being the $h$–polynomial of the boundary complex of a simplicial polytope in dimension $n - m$. Therefore, $h(t)$ is in particular symmetric and unimodal (i.e.: $h_0 \leq \ldots \leq h_{\lfloor n / 2 \rfloor}$).
The connection to toric rings shown by Athanasiadis, Ohsugi and Hibi in [@Atha1], [@OhHi2] and [@OhHi3] is that a compressed polytope $P$ contains a special simplex if and only if the toric ring $K[P]$ of $P$ is *Gorenstein*.
Independently from Ohsugi and Hibi, Bruns and Römer generalized Athanasiadis’ theorem to $h$–vectors of Gorenstein polytopes (cf. [@BruRoe1]). See also e.g. [@ConHosTho1] and [@BatNill1].
In [@Santos1] Santos disproves the Hirsch conjecture by constructing a counter example in dimension 43. The key part for the construction of this counter example is his proof for the existence of a 5–dimensional polytope with special 1–simplex[^1] which has a length of 6. The length is here defined as the number of edges one has to pass to get from one vertex of the special simplex to the other one.\
It is known that *Birkhoff polytopes* and *order polytopes* over graded posets come always with a special simplex. One can find some other examples in low dimensions quite easily; furthermore a connection between the *reverse lexicographic triangulation* of such polytopes and *simplicial joins* can be established. However, so far there has not been a systematic classification of this class of polytopes.\
In this article we give a full classification of polytopes with special simplex in the following way:
Firstly, we define *basis polytopes*. Note that we call two polytopal (resp. simplicial / polyhedral) complexes (and therefore in particular: two polytopes) *isomorphic* if their face lattices are isomorphic.
Let $P$ be an $n$–polytope in ${{\ensuremath{\mathbb{R}}}\xspace}^n$ with face complex ${{\ensuremath{\mathcal{F}}}\xspace}(P)$ and special $m$–simplex ${\Sigma}$. We define the *basis polytope* $Q$ as the image of $P$ under the projection $\pi: {{\ensuremath{\mathbb{R}}}\xspace}^n {\rightarrow}{{\ensuremath{\mathbb{R}}}\xspace}^{n-m}$ of the linear subspace, which is parallel to the affine basis space of ${\Sigma}$, to the origin. \[DefBasispolytop\]
We use the notation $$\begin{aligned}
\operatorname{\mathcal{S}}_{(n,m)}(Q) & := & {\left}\{P \ \middle| \ \begin{array}{l}P \text{ is polytope with some special simplex } {\Sigma}, \dim(P) = n, \\
\dim({\Sigma}) = m, Q \text{ is basis polytope of } P
\end{array}{\right}\}\end{aligned}$$
In Proposition \[SatzAequivSpezSimp\] we deduce amongst other things by using results of Athanasiadis ([@Atha1]) as well as Reiner and Welker ([@ReiWel1]) that $$\begin{aligned}
{{\ensuremath{\mathcal{F}}}\xspace}(\partial Q) & = & {{\ensuremath{\mathcal{F}}}\xspace}(P) {\backslash}{{\ensuremath{\mathcal{V}}}\xspace}({\Sigma})\end{aligned}$$ with ${{\ensuremath{\mathcal{F}}}\xspace}(P) {\backslash}{{\ensuremath{\mathcal{V}}}\xspace}({\Sigma})$ being the subcomplex of the faces of $P$ which does not contain any vertex of ${{\ensuremath{\mathcal{V}}}\xspace}({\Sigma})$. Note that it is not obvious that ${{\ensuremath{\mathcal{F}}}\xspace}(P) {\backslash}{{\ensuremath{\mathcal{V}}}\xspace}({\Sigma})$ is the boundary complex of a polytope.\
Basis polytopes and the upper fact induce a natural distinction between *meek* and *wild* polytopes with special simplex:
Let $P$ be an $n$–polytope with special simplex ${\Sigma}$ and basis polytope $Q$. Let ${{\ensuremath{\mathcal{A}}}\xspace}$ be the affine basis space of the polytopal complex ${{\ensuremath{\mathcal{F}}}\xspace}(P) {\backslash}{{\ensuremath{\mathcal{V}}}\xspace}({\Sigma})$. We call $P$
1. *meek*, if and only if $\dim Q = \dim {{\ensuremath{\mathcal{A}}}\xspace}$.
2. *wild*, if and only if $\dim Q < \dim {{\ensuremath{\mathcal{A}}}\xspace}$ and the combinatorial structure of $P$ is different from the meek polytope $\operatorname{conv }({{\ensuremath{\mathcal{V}}}\xspace}(Q) \cup {{\ensuremath{\mathcal{V}}}\xspace}({\Sigma}))$[^2]. We use the notation $$\begin{aligned}
\operatorname{\mathcal{M}}_{(n,m)}(Q) & := & {\left}\{P \ \middle| \ \begin{array}{l} P \text{ is meek polytope with some special simplex } {\Sigma}, \dim(P) = n, \\
\dim({\Sigma}) = m, Q \text{ is basis polytope of } P
\end{array}{\right}\} \\
\operatorname{\mathcal{W}}_{(n,m)}(Q) & := & {\left}\{P \ \middle| \ \begin{array}{l} P \text{ is wild polytope with some special simplex } {\Sigma}, \dim(P) = n, \\
\dim({\Sigma}) = m, Q \text{ is basis polytope of } P
\end{array}{\right}\}\end{aligned}$$
\[DefZahmWild\]
With these tools we firstly get the following theorem which classifies meek polytopes with special simplex:
Let $P$ be a meek $n$–polytope with special $m$–simplex ${\Sigma}$. Then $P$ has an $(n - m)$–dimensional basis polytope $Q$ (given by Definition \[DefBasispolytop\]) such that $$\begin{aligned}
&&
\begin{array}{cccccccc}
\exists \, i \in {{\ensuremath{\mathbb{N}}}\xspace}, j \in {{\ensuremath{\mathbb{N}}}\xspace}_{>0} : & i + j & = & m & \text{ and } & P & = & {\operatorname{pyr }_{i}}{\left}({\Sigma}_j \oplus Q{\right}).
\end{array}\end{aligned}$$ \[ThmKlassSpezSimpRough\]
In this context “${\operatorname{pyr }_{i}}$” denotes the $i$–th *pyramid* and “${\Sigma}_j \oplus Q$” the *direct sum* of a $j$–simplex ${\Sigma}_j$ and $Q$. Recall that the direct sum of two polytopes $P_1 \in {{\ensuremath{\mathbb{R}}}\xspace}^n,P_2 \in {{\ensuremath{\mathbb{R}}}\xspace}^m$ is the polytope given by $$\begin{aligned}
P_1 \oplus P_2 & := & {\left}\{{\lambda}_1 \binom{v}{0} + {\lambda}_2 \binom{0}{w} \in {{\ensuremath{\mathbb{R}}}\xspace}^{n+m} : v \in P_1; w \in P_2; {\lambda}_1,{\lambda}_2 \in {{\ensuremath{\mathbb{R}}}\xspace}_{\geq 0}; {\lambda}_1 + {\lambda}_2 \leq 1; {\right}\}.\end{aligned}$$ Its boundary complex coincides with the polytopal join (cf. Definition \[DefSimplVerein\]) of the boundary complex of the basis polytope and the boundary complex of a $j$–simplex.\
With this notation Theorem \[ThmKlassSpezSimpRough\] proves that for a fixed basis polytope $Q$ and fixed $n,m$ there are (up to labeling) exactly $m+1$ non–isomorphic meek polytopes with special simplex, i.e.: $$\begin{aligned}
\# \operatorname{\mathcal{M}}_{(n,m)}(Q) & = & m+1 \end{aligned}$$ and in particular for every $n, m$ with $m \in {{\ensuremath{\mathbb{N}}}\xspace}_{>0}, n \in {{\ensuremath{\mathbb{N}}}\xspace}_{\geq m}$ there exists an $n$–polytope with special $m$–simplex. For arbitrary $n$ or $Q$ there exist infinitely many polytopes with special simplex if $n - m \geq 3$. Since the $f$–vectors of pyramids and direct sums are known the theorem yields the combinatorial structure of the class $\operatorname{\mathcal{M}}_{(n,m)}(Q)$ and we can therefore decide whether a polytope is meek and contains a special simplex just by knowing its $f$–vector. The theorem explains furthermore how polytopes in $\operatorname{\mathcal{M}}_{(n,m)}(Q)$ can be constructed in a combinatorial *and* in a geometric way and hence classifies $\operatorname{\mathcal{M}}_{(n,m)}(Q)$ completely.\
For $\operatorname{\mathcal{W}}_{(n,m)}(Q)$ we are unable to decide wether a polytope belongs to this class in a pure combinatorial way just by investigating its $f$–vector since the combinatorial structure of particular wild polytopes depends crucially on their basis polytopes (cf. Section \[SubSecWildePolytope\]).
We are able to give a classification in the following sense: We prove that every wild polytope with basis polytope $Q$ can be constructed by intersecting ${\Sigma}\oplus Q$ with particular (via the vertices combinatorially defined) hyperplanes satisfying conditions which we present in Theorem \[ThmWildPolyCharac\]. Therefore we present a full classification of all polytopes with special simplex. Theorem \[ThmWildPolyCharac\] is constructive if one knows all particular hyperplanes satisfying the conditions of the Theorem for a specific basis polytope. In general it seems to be a hard task to find these hyperplanes; we present an explicit description for all polytopes with special simplex and a basis polytope of dimension 2 (Corollary \[CorWildPolyBasisPolyDim2\]). Theorem \[ThmWildPolyCharac\] yields furthermore an upper bound for the $f$–vector of wild polytopes with special simplex (Corollary \[CorBoundaries\]):
Let $P \in \operatorname{\mathcal{W}}_{(n,k)}(Q)$ with special simplex ${\Sigma}$ and $n \geq 3$ (for $n < 2$ we have $\operatorname{\mathcal{W}}_{(n,k)}(Q) = 0$). Then we have for its $f$–vector $f_P$ in particular: $$\begin{aligned}
f_{P}^{(0)} & = & f_{({\Sigma}\oplus Q)}^{(0)} \\
f_{P}^{(i)} & < & f_{({\Sigma}\oplus Q)}^{(i)} \quad \text{ for } i \in \{n-2,n-1\} \\
f_{P}^{(i)} & \leq & f_{({\Sigma}\oplus Q)}^{(i)} \quad \text{ for } i \in \{1,\ldots,n-3\}.\end{aligned}$$ Here $f_{({\Sigma}\oplus Q)}$ denotes the $f$–vector of the direct sum which is explicitly computable.
Finally we discuss the 4–cube as an example for a wild polytope with special simplex which is non–trivial since its basis polytope has dimension 3 and thus is not covered by Corollary \[CorWildPolyBasisPolyDim2\]. We prove in particular that the basis polytope of the $n$–cube is the zonotope given by the $(n-1)$ cube and the vector $(1,\ldots,1)$ (Proposition \[PropBasispolyCube\]).
Acknowledgements {#acknowledgements .unnumbered}
----------------
I would like to thank Hartwig Bosse, Christian Haase, Cordian Riener, Raman Sanyal, Reinhard Steffens and Theresa Szczepanski for their helpful comments. My very special thanks go to Thorsten Theobald who introduced me to the topic, supervised the master thesis (Diplomarbeit) the first part of this article is originally based on and strongly supported me in all intents and purposes from the first to the last moment.
Preliminaries and examples {#SecGrundlagen}
==========================
Preliminaries
-------------
We assume that the reader is familiar with standard geometric structures e.g.: polyhedral / simplicial complexes or $f$–vectors (cf. e.g. [@Zieg1]). In this section we introduce the further structures which are needed to achieve our results.\
For our efforts we need a particular triangulation on polytopal complexes — the *reverse lexicographic triangulation* (cf. eg. [@Lee1], [@Atha1 p. 165]):
Let ${{\ensuremath{\mathcal{F}}}\xspace}$ be a polytopal complex with vertexset ${{\ensuremath{\mathcal{V}}}\xspace}({{\ensuremath{\mathcal{F}}}\xspace}) :=\{v_1,\ldots,v_p\}$; let $\tau := (v_1,\ldots,v_p)$ be a total ordering on ${{\ensuremath{\mathcal{V}}}\xspace}({{\ensuremath{\mathcal{F}}}\xspace})$. Then we define the *reverse lexicographic triangulation* (short: RLT) by: $$\begin{aligned}
{\triangle}_{\tau}(\emptyset) & := & \emptyset \textrm{ and } \\
{\triangle}_{\tau}({{\ensuremath{\mathcal{F}}}\xspace}) & := & {\triangle}_{\tau}({{\ensuremath{\mathcal{F}}}\xspace}{\backslash}v_p) \cup \bigcup_{F \in M_{v_p}} {\left}\{\operatorname{conv }(\{v_p\} \cup G) \ | \ G \in {\triangle}_{\tau}({{\ensuremath{\mathcal{F}}}\xspace}(F)) \cup \{\emptyset\} {\right}\}, \end{aligned}$$ with $M_{v_p}$ containing exactly the facets not containing $v_p$ of all maximal faces of ${{\ensuremath{\mathcal{F}}}\xspace}$ containing $v_p$. ${\triangle}_{\tau}({{\ensuremath{\mathcal{F}}}\xspace}{\backslash}v_p)$ and ${\triangle}_{\tau}({{\ensuremath{\mathcal{F}}}\xspace}(F))$ are defined with respect to subsets of vertices of polytopal complexes and the ordering on the particular subset which is induced by $\tau$. \[DefRueckLexTri\]
Note that there are additional definitions which are equivalent to Definition \[DefRueckLexTri\] (cf. e.g. [@Stan1 p. 333] and [@JoZi1 p. 1253 et seq.]; in the latter the ordering of the vertices is transposed). These triangulations coincide with the successive pulling of vertices into the negative halfspaces of the supporting hyperplanes containing the particular vertices. (cf. [@Gruenb1 p. 78 et seq.]).\
Now we introduce *polytopal joins* and the *quotient polytope*. Polytopal joins are needed for Lemmata from Reiner / Welker and Athanasiadis which we want to use and give furthermore another nice description of direct sums of polytopes. The quotient polytope is needed since basis polytopes are quotient polytopes and it allows us to present a lemma of Reiner and Welker which is essential to prove that two particular conditions are equivalent to the existence of a special simplex in a given polytope (Proposition \[SatzAequivSpezSimp\]).\
For polytopal and simplicial joins cf. [@Atha1 p. 167] and [@ReiWel1 p. 259, Definition 3.7] (observe that in all references only simplicial joins are used.):
Let ${\triangle}_1, {\triangle}_2$ be two polytopal complexes with disjoint vertex sets and affine basis spaces with trivial intersection in the interior. Then we define the *polytopal join* ${\triangle}_1 * {\triangle}_2$ of ${\triangle}_1, {\triangle}_2$ as the polytopal complex containing the following facets $F$: $$\begin{aligned}
F & = & \operatorname{conv }\{F_1,F_2\} \textrm{ with } \forall i \in \{1,2\}: F_i \textrm{ is a maximal face of } {\triangle}_i. \end{aligned}$$ The former faces of ${\triangle}_1 * {\triangle}_2$ are induced by the intersections of the facets (cf. [@Tza1 p. 1193]).
If ${\triangle}_1, {\triangle}_2$ are simplicial complexes, we will call ${\triangle}_1 * {\triangle}_2$ a *simplicial join*. \[DefSimplVerein\]
Note that for polytopal complexes ${\triangle}_1, {\triangle}_2$ we have $\dim({\triangle}_1 * {\triangle}_2) = \dim({\triangle}_1) + \dim({\triangle}_2) + 1$. If one of the two polytopal complexes ${\triangle}_1, {\triangle}_2$ (w.l.o.g.: ${\triangle}_1$) of a polytopal join contains only the empty set, we have $\emptyset * {\triangle}_2 = {\triangle}_2$. This statement is in accord with the one about the dimension since $\dim(\emptyset) = -1$.
Let ${\triangle}_1,{\triangle}_2$ and ${\triangle}_3$ be polytopal complexes whose affine basis spaces share no common non trivial subspace pairwise. Then the polytopal join is associative, i.e.: $$\begin{aligned}
({\triangle}_1 * {\triangle}_2) * {\triangle}_3 & = & {\triangle}_1 * ({\triangle}_2 * {\triangle}_3).\end{aligned}$$
Observe that we have for two polytopes $P \in {{\ensuremath{\mathbb{R}}}\xspace}^n,Q \in {{\ensuremath{\mathbb{R}}}\xspace}^m$ that $$\begin{aligned}
{{\ensuremath{\mathcal{F}}}\xspace}(\partial (P \oplus Q)) & = & {{\ensuremath{\mathcal{F}}}\xspace}(\partial P) * {{\ensuremath{\mathcal{F}}}\xspace}(\partial Q)\end{aligned}$$ This is easy to see: Let $F_1 \in {{\ensuremath{\mathcal{F}}}\xspace}(P), F_2 \in {{\ensuremath{\mathcal{F}}}\xspace}(Q)$ facets of $P$ and $Q$ and let $A_P,A_Q$ denote the affine basis spaces of $P$ resp. $Q$. Then $F_1 \cap F_2 = \emptyset$ and $F_1 \not\parallel F_2$ since $A_P$ and $A_Q$ do online intersect in one point which is located in the interior of $P$ and $Q$. Hence $\operatorname{conv }({{\ensuremath{\mathcal{V}}}\xspace}(F_1) \cup {{\ensuremath{\mathcal{V}}}\xspace}(F_2))$ is contained in a hyperplane $H$ of ${{\ensuremath{\mathbb{R}}}\xspace}^{n+m}$. Since $H$ supports $F_1$ and $F_2$ all vertices ${{\ensuremath{\mathcal{V}}}\xspace}(P {\backslash}F_1)$ and ${{\ensuremath{\mathcal{V}}}\xspace}(Q {\backslash}F_2)$ are contained in $H^+$. Thus, $H$ is a supporting hyperplane of $P \oplus Q$. The inverse way follows directly from the definition of the direct sum.
Let $P \subseteq {{\ensuremath{\mathbb{R}}}\xspace}^q$ be an $n$–polytope and $V$ an arbitrary subspace of ${{\ensuremath{\mathbb{R}}}\xspace}^q$. Then the *quotient polytope* $P / V \subseteq {{\ensuremath{\mathbb{R}}}\xspace}^q / V$ is defined by $$\begin{aligned}
P / V & := & \{p + V \ | \ p \in P\}.\end{aligned}$$ Obviously $P / V$ can be identified with the image $\pi(P)$ of $P$ under a linear embedding $\pi : {{\ensuremath{\mathbb{R}}}\xspace}^q {\rightarrow}{{\ensuremath{\mathbb{R}}}\xspace}^{q - \dim V}$ with kernel $V$ (cf. [@Atha1 p. 167], [@ReiWel1 p. 260, Definition 3.10]). \[DefQuotPolytop\]
Notice that the basis polytope from Definition \[DefBasispolytop\] hence is a particular quotient polytope.\
With this definition we are able to formulate the following essential lemma (cf. [@Atha1 ibid.], cf. also [@ReiWel1 p. 261, Proposition 3.12]):
Let $P$ be a $n$–polytope in ${{\ensuremath{\mathbb{R}}}\xspace}^q$ with a triangulation ${\triangle}$ isomorphic to ${\Sigma}* {\triangle}'$ with ${\triangle}'$ denoting a (not exactly specified) simplicial complex and ${\Sigma}$ a simplex which is not completely contained in the boundary of $P$. Let $V$ be the linear subspace of ${{\ensuremath{\mathbb{R}}}\xspace}^q$ being parallel to the affine span of ${\Sigma}$. Then the boundary complex of the quotient polytope $P / V \subseteq {{\ensuremath{\mathbb{R}}}\xspace}^q / V$ is isomorphic to ${{\ensuremath{\mathcal{F}}}\xspace}(P) {\backslash}{{\ensuremath{\mathcal{V}}}\xspace}({\Sigma})$ and has a triangulation isomorphic to ${\sigma}* {\triangle}'$ with ${\sigma}$ being an interior point of $P / V$. \[Lem2ReiWel1\]
The upper lemma was originally given by Reiner and Welker in [@ReiWel1]. Later it was cited and modified by Athanasiadis in [@Atha1] into the upper version which we want to use here.
Examples
--------
Before we start to prove our results we recall some of the known examples of polytopes with special simplices.
Be aware that special simplices of a polytope $P$ may never completely lie in the boundary, since in this case they would be contained in one particular facet of $P$ in contradiction to the Definition \[DefSpezSimpl\].
There are quite few combinatorially different polytopes with special simplices in low dimensions. Here we present two of them:
(380,190)(0,0) (0,0)[![Left: the four special 1–simplices (blue) in a 3–dimensional cube; right: the special 1– (blue) and 2–simplex (green) in a 3–dimensional bipyramid with triangular base (both lattice models).[]{data-label="AbbSpezielleSimplizesLowDim"}](./SpezSimpBsp2.eps "fig:") ]{}
\[ExaSpezSimp1\]
The set of all doubly stochastic $n \times n$ matrices (i. e.: matrices with positive entries such that all row– and column sums are 1) is a polytope in ${{\ensuremath{\mathbb{R}}}\xspace}^{n \times n}$ — the *Birkhoff polytope* $P_n$.
Birkhoff polytopes are $0 / 1$–polytopes; they have $n!$ vertices (the permutation matrices of $S_n$), $n^2$ facets (in each case consisting of the set of all matrices with $x_{ij} = 0$) and dimension $(n-1)^2$. A complete classification is given by (cf. [@Zieg1 p. 20]): $$\begin{aligned}
P_n & = & {\left}\{X \in {{\ensuremath{\mathbb{R}}}\xspace}^{n \times n} \ \middle| \ \forall i,j \in \{1,\ldots,n\}: x_{ij} \geq 0 {\wedge}\sum_{k = 1}^n x_{ik} = \sum_{k = 1}^n x_{kj} = 1 {\right}\}.\end{aligned}$$
Let $P_n$ be the $n$–th Birkhoff polytope. Let $v_1,\ldots,v_n$ be the $n \times n$ permutation matrices, which correspond to the elements of the cyclic subgroup of the symmetric group generated by the permutation $(1 \ 2 \ \cdots \ n)$ (or $n$ arbitrary permutation matrices with pairwise disjoint support). Then $v_1,\ldots,v_n$ are the vertices of a special simplex in $P_n$, because every facet of $P_n$ is defined by an equation of the form $x_{ij} = 0$ in ${{\ensuremath{\mathbb{R}}}\xspace}^{n \times n}$ and does not contain exactly one of the vertices $v_1,\ldots,v_n$. This is obvious since the $v_k$ have the form: $${\left}(\begin{array}{cccccc}
0 & 1 & 0 & \cdots & 0 & 0 \\
0 & 0 & 1 & \ddots & 0 & 0 \\
\vdots & \vdots & \ddots & \ddots & \ddots & \vdots \\
0 & 0 & 0 & \ddots & 1 & 0 \\
0 & 0 & 0 & \cdots & 0 & 1 \\
1 & 0 & 0 & \cdots & 0 & 0 \\
\end{array}{\right}),
{\left}(\begin{array}{cccccc}
0 & 0 & 1 & \cdots & 0 & 0 \\
\vdots & \vdots & \ddots & \ddots & \ddots & \vdots \\
0 & 0 & 0 & \ddots & 1 & 0 \\
0 & 0 & 0 & \cdots & 0 & 1 \\
1 & 0 & 0 & \cdots & 0 & 0 \\
0 & 1 & 0 & \cdots & 0 & 0 \\
\end{array}{\right}),
\ldots,$$ i.e. in every row the 1 runs through every column successively (cf. [@Atha1 p. 165]). \[Exa1Atha1\]
Let $\Omega := \{x_1,\ldots,x_n\}$ be a poset with respect to the relation $\leq_{\Omega}$. Then the set of all maps $f : \Omega {\rightarrow}{{\ensuremath{\mathbb{R}}}\xspace}$ forms an $n$–dimensional ${{\ensuremath{\mathbb{R}}}\xspace}$–vectorspace (we identify every $f$ with its image vector). Using this fact the *order polytope* ${{\ensuremath{\mathcal{O}}}\xspace}(\Omega)$ is defined as the set of all $f(\Omega) \subset {{\ensuremath{\mathbb{R}}}\xspace}^n$ which satisfy (cf. e.g. [@Stan2]):
1. $\forall x \in \Omega : 0 \leq f(x) \leq 1,$
2. $\forall x,y \in \Omega : {\left}(\neg \exists z: z \neq x {\wedge}z \neq y {\wedge}x \leq_{\Omega} z \leq_{\Omega} y{\right}) {\Rightarrow}f(x) \leq f(y).$
The order polytope ${{\ensuremath{\mathcal{O}}}\xspace}(\Omega)$ has dimension $n$. $f \in {{\ensuremath{\mathcal{O}}}\xspace}(\Omega)$ is contained in a specific facet of the order polytope if and only if one of the following condition holds:
1. $\exists x \in \Omega: f(x) = 0 {\wedge}x \textrm{ is minimal with resp. to } \leq_{\Omega}.$
2. $\exists x \in \Omega: f(x) = 1 {\wedge}x \textrm{ is maximal with resp. to } \leq_{\Omega}.$
3. $\exists x,y \in \Omega: x \neq y {\wedge}f(x) = f(y) {\wedge}y \textrm{ covers } x \textrm{ with resp. to } \leq_{\Omega}$
(recall that in posets $(\Omega,\leq_{\Omega})$ $y$ *covers* $x$ if and only if there is no $z \in \Omega {\backslash}\{x,y\}$ with $x \leq_{\Omega} z \leq_{\Omega} y$).
An *(order) ideal* or *filter* in $\Omega$ is a subset $I \subseteq \Omega$ satisfying the following condition: $$\begin{aligned}
\forall i,j \in \Omega: {\left}(i \leq_{\Omega} j {\wedge}i \in I{\right}) & {\Rightarrow}& j \in I.\end{aligned}$$ The vertices of ${{\ensuremath{\mathcal{O}}}\xspace}(\Omega)$ are the images of the filters of $\Omega$ under the characteristic function (with the filters interpreted as vectors with respect to the relation $\leq_{\Omega^0}$) (cf. [@Stan2 p. 10 et seq.]).
Let $\Omega:= \{x_1,\ldots,x_n\}$ be a graded poset with respect to the relation $\leq_{\Omega}$ with rank $m-1$ ($1 \leq m \leq n$; i.e.: the maximal chain consists of $m-1$ elements). Let $P := {{\ensuremath{\mathcal{O}}}\xspace}(\Omega)$ be the order polytope of $\Omega$ in ${{\ensuremath{\mathbb{R}}}\xspace}^n$. Let $v_i$ be the characteristic vector of the filter consisting of all elements in $\Omega$ with rank greater than $m - i$ for all $i \in \{1,\ldots,m\}$ ($v_i$ is therefore a vertex of $P$). The upper equations defining the facets of an order polytope yield that $v_1,\ldots,v_m$ are the vertices of a special simplex in $P$ (cf. [@Atha1 p. 165]).
To substantiate the intuition of order polytopes let $\Omega := \{x_1,x_2,x_3\}$ and look at the three order polytopes ${{\ensuremath{\mathcal{O}}}\xspace}_{(1,1,1)}, {{\ensuremath{\mathcal{O}}}\xspace}_{(2,1)}$ and ${{\ensuremath{\mathcal{O}}}\xspace}_{(1,2)}$ given by the following three order structures: $$\begin{aligned}
\begin{xy}
\xymatrix{
x_3 \\
x_2 \ar@{-}[u] \\
x_1 \ar@{-}[u]
}
\end{xy}
\quad
\begin{xy}
\xymatrix{
& x_3 & \\
x_1 \ar@{-}[ru] & & x_2 \ar@{-}[lu]
}
\end{xy}
\quad
\begin{xy}
\xymatrix{
x_2 & & x_3 \\
& x_1 \ar@{-}[lu] \ar@{-}[ru] &
}
\end{xy}\end{aligned}$$ Then one can show easily that ${{\ensuremath{\mathcal{O}}}\xspace}_{(1,1,1)}$ is a simplex and its special simplex is the polytope itself; ${{\ensuremath{\mathcal{O}}}\xspace}_{(2,1)}$ and ${{\ensuremath{\mathcal{O}}}\xspace}_{(1,2)}$ are both 3–dimensional polytopes in form of a pyramid with square base each containing a special 2–simplex:
(400,200)(0,0) (0,0)[![The order polytopes ${{\ensuremath{\mathcal{O}}}\xspace}_{(2,1)}$ and ${{\ensuremath{\mathcal{O}}}\xspace}_{(1,2)}$ with their special simplices (green) (lattice models).[]{data-label="AbbOrdnungspolytopeSpezSimp"}](./Orderpoly2.eps "fig:")]{} (67,89) (13,149) (110,163) (146,145) (130,100) (128,105)[(-3,2)[35]{}]{} (267,89) (213,149) (310,163) (346,145) (306,100)
\[Exa2Atha1\]
Raman Sanyal advised me to the class of weak Hannar polytopes: A polytope $P$ is called *weakly Hannar* if $P = \operatorname{conv }({{\ensuremath{\mathcal{V}}}\xspace}(F) \cup {{\ensuremath{\mathcal{V}}}\xspace}(-F))$ for every facet $F$ of $P$. Obviously every weak Hannar polytope is centrally symmetric (cf. [@Hansen1], [@SanWerZieg1]).
In a weak Hannar polytope every pair of points $(x,-x)$ forms a special 1–simplex. This is easy to see since for every facet $F$ $x$ is contained in $F$ or $-F$ by definition. If $x \in F$ we have $-x \in -F$ due to the fact that weak Hannar polytopes are centrally symmetric and for the same reason there is no facet with $x,-x \in F$. Hence $\{x,-x\}$ is a special simplex. \[ExaWeaklyHannar\]
Equivalence conditions for polytopes with special simplices {#AbsExaUndAequivSatz}
===========================================================
Although the original Definition \[DefSpezSimpl\] of special simplices is easy to understand it is not very easy to work with. Hence in Proposition \[SatzAequivSpezSimp\] we introduce two conditions which are together equivalent to the existence of a special simplex. This Proposition \[SatzAequivSpezSimp\] is the foundation to establish our classification of meek polytopes with special simplex. To prove it we need the following lemma of Athanasiadis (cf. [@Atha1 p. 167 et seq.]):
Let $P$ be an $n$–dimensional polytope and let $\tau := (v_p,\ldots,v_1)$ be an ordering of the vertices of $P$ such that ${{\ensuremath{\mathcal{V}}}\xspace}({\Sigma}) := \{v_1,\ldots,v_k\}$ is the vertex set of a special simplex ${\Sigma}$ in $P$. Let ${\triangle}$ be the simplicial complex on the set $\{v_{k+1},\ldots,v_p\}$ induced by the reverse lexicographic triangulation ${\triangle}_{\tau'}({{\ensuremath{\mathcal{F}}}\xspace}(P) {\backslash}{{\ensuremath{\mathcal{V}}}\xspace}({\Sigma}))$ with respect to $\tau' := (v_{k+1},\ldots,v_p)$. Then the following assertions hold:
1. The reverse lexicographic triangulation ${\triangle}_{\tau}(P)$ of $P$ with respect to $\tau$ is isomorphic to the simplicial join ${\Sigma}* {\triangle}$.
2. ${\triangle}$ is isomorphic to the boundary complex of a simplicial polytope of dimension $n - k + 1$. \[Lem2Atha1\]
We can now present our equivalence proposition. The two conditions we mentioned above are firstly that the subcomplex of the original polytope $P$, which is left after eliminating all faces containing vertices of the special simplex, is the boundary complex of a *polytope* $Q$ (the basis polytope) and secondly that *every* hyperplane containing the special simplex divides $Q$. We express this formally:
Let $P$ be an $n$–polytope with ${{\ensuremath{\mathcal{V}}}\xspace}(P) := \{v_0,\ldots,v_p\}$. Then $P$ contains a special $m$–simplex ${\Sigma}$ (w.l.o.g.: ${{\ensuremath{\mathcal{V}}}\xspace}({\Sigma}) = \{v_0,\ldots,v_m\}$ with $1 \leq m \leq n-1$) if and only if the following two conditions hold:
1. For every hyperplane $H \subset {{\ensuremath{\mathbb{R}}}\xspace}^n$ with ${\Sigma}\subset H$ holds: $$\begin{aligned}
\exists i,j \in \{m+1,\ldots,p\}: v_i \in H^+ & {\wedge}& v_j \in H^-.
\end{aligned}$$
2. There is a basis polytope $Q \subset {{\ensuremath{\mathbb{R}}}\xspace}^{n-m}$ (recall Definition \[DefBasispolytop\]) such that $$\begin{aligned}
{{\ensuremath{\mathcal{F}}}\xspace}(P) {\backslash}{{\ensuremath{\mathcal{V}}}\xspace}({\Sigma}) & = & {{\ensuremath{\mathcal{F}}}\xspace}(\partial Q).
\end{aligned}$$
\[SatzAequivSpezSimp\]
Assume w.l.o.g. that $v_0$ is the origin and therefore each of the observed ${\Sigma}$ containing affine hyperplanes is a hyperplane of ${{\ensuremath{\mathbb{R}}}\xspace}^n$.\
Let ${\Sigma}$ be a special simplex in $P$. Assume there is a hyperplane $H$ containing ${\Sigma}$ with $v_i \in H^+$ for all $i \in \{m+1,\ldots,p\}$. Then we have in particular $P \subset H^+$. Since ${\Sigma}\subset P$ and ${\Sigma}\subset H$ this implies that $H$ is a bounding hyperplane of $P$ and ${\Sigma}\subset \partial P$. But then there exists a facet containing ${\Sigma}$ completely in contradiction to Definition \[DefSpezSimpl\] of special simplices. This implies condition (a).
We additionally know by Lemma \[Lem2Atha1\] that the reverse lexicographic triangulation ${\triangle}_{\tau}(P)$ with respect to $\tau := (v_p,\ldots,v_0)$ of $P$ is isomorphic to the simplicial join ${\Sigma}* {\triangle}'$ (with ${\triangle}' := {\triangle}_{\tau'}({{\ensuremath{\mathcal{F}}}\xspace}(P) {\backslash}{{\ensuremath{\mathcal{V}}}\xspace}({\Sigma}))$). Therefore the conditions of Lemma \[Lem2ReiWel1\] are fullfilled which in particular implies that ${{\ensuremath{\mathcal{F}}}\xspace}(P) {\backslash}{{\ensuremath{\mathcal{V}}}\xspace}({\Sigma})$ is isomorphic to the boundary of the quotient polytope $P / V \subset {{\ensuremath{\mathbb{R}}}\xspace}^{n-m}$ with $V := {\textrm{span}}\{v_0,\ldots,v_m\}$. This yields condition (b).\
Let us now assume that $P$ is a polytope satisfying conditions (a) and (b). Since boundary complexes of polytopes are pure we know by (b) that every maximal face $S$ of ${{\ensuremath{\mathcal{F}}}\xspace}(P) {\backslash}{{\ensuremath{\mathcal{V}}}\xspace}({\Sigma})$ has dimension $n - m - 1$. Every facet $F$ of $P$ with $F {\backslash}{{\ensuremath{\mathcal{V}}}\xspace}({\Sigma}) = S$ has dimension $n - 1$ by definition of $P$. Since eliminating a vertex from a facet (resp. more general: a face) lowers the dimension at most by 1, every facet of $P$ needs to contain $m$ affine independent vertices out of the set ${{\ensuremath{\mathcal{V}}}\xspace}({\Sigma})$.\
But, condition (a) yields that no facet may contain all $m+1$ vertices from ${{\ensuremath{\mathcal{V}}}\xspace}({\Sigma})$, since otherwise for the appropriate bounding hyperplane $H$ all vertices of ${{\ensuremath{\mathcal{V}}}\xspace}(P) {\backslash}{{\ensuremath{\mathcal{V}}}\xspace}({\Sigma})$ would be located in $H^+$. Therefore, the vertices of ${\Sigma}$ are affine independent, i.e. ${\Sigma}$ is a simplex and every facet of $P$ contains exactly $m$ vertices of ${\Sigma}$. This is exactly the definition of special simplex.
If $P$ is a simplex itself it makes sense to say that $P$ has a special simplex (which is $P$ itself) as well by Definition \[DefSpezSimpl\] as by the upper proposition since condition (a) and (b) are satisfied trivially. For the rest of this article we will not mention this special case anymore (the reader might just keep it in mind).
Note the following subtle fact in Proposition \[SatzAequivSpezSimp\]: Although the polytopal complex ${{\ensuremath{\mathcal{F}}}\xspace}(P) {\backslash}{{\ensuremath{\mathcal{V}}}\xspace}({\Sigma})$ falls together with the boundary of an $(n-m)$–polytope $Q$, this does **not** in general imply that the affine basis space of this complex has the same dimension as $Q$. The dimension of the complex might be higher; the upper conclusion just applies to the combinatorial structure of ${{\ensuremath{\mathcal{F}}}\xspace}(P) {\backslash}{{\ensuremath{\mathcal{V}}}\xspace}({\Sigma})$ and ${{\ensuremath{\mathcal{F}}}\xspace}(\partial Q)$. \[RemPolymayproofild\]
We give an example to make this abstract remark clearer.
By Example \[ExaSpezSimp1\], Fig. \[AbbSpezielleSimplizesLowDim\] a 3–cube contains a special 1–simplex. By eliminating the two vertices belonging to the simplex, one can see that the remaining complex is not contained in ${{\ensuremath{\mathbb{R}}}\xspace}^2$, although it is the boundary complex of a 6–gon. \[ExaWuerfelistwild\]
This observation motivates the distinction between *meek* and *wild* polytopes with special simplices introduced in Definition \[DefZahmWild\] in the introduction.\
Clearly a polytope $P$ with special simplex ${\Sigma}$ and basis polytope $Q$ is meek if and only if the intersection of the affine basis spaces of ${\Sigma}$ and $Q$ is a single point. Furthermore, all polytopes $P$ with special simplex whose basis polytope $Q$ is a simplex are meek.
Pyramids and direct sums {#AbsPyrundMuscheln}
========================
Before we use Proposition \[SatzAequivSpezSimp\] to show how to construct polytopes out of $\operatorname{\mathcal{M}}_{(n,m)}$ (with $n,m$ arbitrary) we first deduce two corollaries to show that $\operatorname{\mathcal{M}}_{(*,1)}$ consists of bipyramids and to classify the low dimensional cases $\operatorname{\mathcal{M}}_{(2,1)}, \operatorname{\mathcal{M}}_{(3,2)}$ and $\operatorname{\mathcal{M}}_{(3,1)}$. Afterwards we show that constructing the pyramid of a meek $n$–polytope with special $m$–simplex yields a meek $(n+1)$–polytope with special $(m+1)$–simplex. Unfortunately one can not construct all polytopes in $\operatorname{\mathcal{M}}_{(n,m)}$ this way. Therefore we have to look on direct sums additionally. Both together deliver the whole class $\operatorname{\mathcal{M}}_{(n,m)}$ which we show in the next section. Notice that all polytopes with special simplex we look at until Section \[SubSecWildePolytope\] are **meek**.
Let $P$ be a meek $n$–polytope with special simplex ${\Sigma}$. If $\dim({\Sigma}) = 1$ then $P$ is a bipyramid, i.e. there is a $(n-1)$–basis polytope $Q$ with $f$–vector $(1,f_{q_0},\ldots,f_{q_{n-2}},1)$ such that we have for the $f$–vector $f_P := (1,f_{p_0},\ldots,f_{p_{n-1}},1)$ of $P$: $$\begin{aligned}
f_{p_0} \, = \, f_{q_0} + 2, \ f_{p_{n-1}} \, = \, 2 \cdot f_{q_{n-2}}, \ \forall i \in \{1,\ldots,n-2\}: f_{p_i} \, = \, f_{q_i} + 2 \cdot f_{q_{i-1}}. \end{aligned}$$ In other words: $$\begin{aligned}
\operatorname{\mathcal{M}}_{(n,1)} & = & \{\operatorname{bipyr }Q \ | \ Q \text{ is basis polytope, } \dim(Q) = n-1\} \quad \text{ for all } n \in {{\ensuremath{\mathbb{N}}}\xspace}.\end{aligned}$$ \[KorSpezSimp1\]
Let $Q$ with $\dim(Q) = n-1$ be chosen arbitrary, ${{\ensuremath{\mathcal{V}}}\xspace}({\Sigma}) = \{v_0,v_1\}$. Let $H_Q$ be the hyperplane containing $Q$. To achieve condition (a) in Proposition \[SatzAequivSpezSimp\] for every hyperplane containing ${{\ensuremath{\mathcal{V}}}\xspace}({\Sigma})$, one vertex has to be in $H_Q^+$ and one in $H_Q^-$ and the vertices may not be connected (besides $Q$) by an edge. This corresponds exactly to a bipyramid with basis polytope $Q$ since ${{\ensuremath{\mathcal{F}}}\xspace}(P) {\backslash}{{\ensuremath{\mathcal{V}}}\xspace}({\Sigma})$ lies in $H_Q$ because $P$ is meek.
We may now classify all meek polytopes with special simplices in low dimensions.
In ${{\ensuremath{\mathbb{R}}}\xspace}^2$ exactly all convex quadrangles have a special 1–simplex. In ${{\ensuremath{\mathbb{R}}}\xspace}^3$ exactly the polytopes with $f$–vector $(1,5,9,6,1)$ or $(1,5,8,5,1)$ have a special 2–simplex and all arbitrary bipyramids are meek and have a special 1–simplex, i.e.: $$\begin{aligned}
\operatorname{\mathcal{M}}_{(2,1)} & = & \{P \ | \ f_P = (1,4,4,1)\}, \\
\operatorname{\mathcal{M}}_{(3,2)} & = & \{P \ | \ f_P = (1,5,9,6,1) \vee f_P = (1,5,8,5,1)\}, \\
\operatorname{\mathcal{M}}_{(3,1)} & = & \{P \ | \ f_P = (1,k+2,3 \cdot k,2 \cdot k,1), k \in {{\ensuremath{\mathbb{N}}}\xspace}_{\geq 3}\}.\end{aligned}$$ \[KorSpezSimp2\]
The assertions on $\operatorname{\mathcal{M}}_{(2,1)}$ and $\operatorname{\mathcal{M}}_{(3,1)}$ may be deduced from the previous corollary by a simple calculation.
For $\operatorname{\mathcal{M}}_{(3,2)}$ the conditions in Proposition \[SatzAequivSpezSimp\] yield that there is exactly one hyperplane containing ${\Sigma}$ and ${{\ensuremath{\mathcal{F}}}\xspace}(\partial Q) = {{\ensuremath{\mathcal{F}}}\xspace}(P) {\backslash}{{\ensuremath{\mathcal{V}}}\xspace}({\Sigma})$ consists of merely two vertices since $\dim(Q) = 1$. Thus, the intersection of $Q$ and ${\Sigma}$ is either in the interior of ${\Sigma}$ or on a 1–face of ${\Sigma}$. These are the two cases mentioned above. The intersection may not be a vertex since this vertex would no longer be part of ${{\ensuremath{\mathcal{V}}}\xspace}(P)$ and the intersection may not be empty since ${\Sigma}$ would not be a special simplex otherwise.
Note the following further fact Proposition \[SatzAequivSpezSimp\] yields: For all $n \in {{\ensuremath{\mathbb{N}}}\xspace}, m \in {{\ensuremath{\mathbb{N}}}\xspace}_{<n}$ there exists a canonical embedding $\rho: \operatorname{\mathcal{M}}_{(n,m)} {\rightarrow}\operatorname{\mathcal{M}}_{(n+1,m+1)}$ which is induced by a pyramid construction. This leads to the first way to construct a meek polytope with special simplex.
Let be $P$ a meek $n$–polytope (${{\ensuremath{\mathcal{V}}}\xspace}(P) := \{v_0,\ldots,v_p\}$) with special $m$–simplex. Then its pyramid $\operatorname{pyr }P \subset {{\ensuremath{\mathbb{R}}}\xspace}^{n+1}$ has a special $m+1$–simplex and is meek. \[KorSpezSimp3\]
Let ${{\ensuremath{\mathcal{V}}}\xspace}({\Sigma}) := \{v_0,\ldots,v_n\}$ and let $w$ be the vertex which is added as the apex of the pyramid of $P$, i.e.: $$\begin{aligned}
\text{pyr } P & = & {\left}(\operatorname{conv }{{\ensuremath{\mathcal{V}}}\xspace}(P) \cup \{w\}{\right}).\end{aligned}$$ Then ${\Sigma}' := \operatorname{conv }{\left}({{\ensuremath{\mathcal{V}}}\xspace}({\Sigma}) \cup \{w\}{\right})$ is a special simplex in pyr $P$: since if $H'$ is a hyperplane in ${{\ensuremath{\mathbb{R}}}\xspace}^{n+1}$ with ${\Sigma}' \subset H'$ then there exists one and only one hyperplane $H$ of ${{\ensuremath{\mathbb{R}}}\xspace}^n$ with $H = H' \cap {{\ensuremath{\mathbb{R}}}\xspace}^n$ and ${\Sigma}\subset H$. Since ${\Sigma}$ is a special simplex of $P$, condition (a) of Proposition \[SatzAequivSpezSimp\] yields: $$\begin{aligned}
\exists i,j \in \{m+1,\ldots,p\}: v_i \in H^+ {\wedge}v_j \in H^-\end{aligned}$$ and thus, due to the fact that $\{v_{m+1},\ldots,v_p\} \subset {{\ensuremath{\mathbb{R}}}\xspace}^n$, by construction of $H'$: $$\begin{aligned}
\exists i,j \in \{m+1,\ldots,p\}: v_i \in H'^+ {\wedge}v_j \in H'^-.\end{aligned}$$ Therefore condition (a) of Proposition \[SatzAequivSpezSimp\] holds for pyr $P$ and ${\Sigma}'$. Condition (b) is trivial. Hence ${\Sigma}'$ is special $m+1$–simplex of pyr $P$.
The two corollaries \[KorSpezSimp1\] and \[KorSpezSimp3\] put together provide the simple first method to construct an $n$–dimensional polytope with special $m$–simplex:
Let $Q$ be an arbitrary polytope with dimension $n - m$. Then $$\begin{aligned}
{\operatorname{pyr }_{m-1}} (\operatorname{bipyr }Q) & := & \underbrace{\operatorname{pyr }\operatorname{pyr }\cdots \operatorname{pyr }}_{m-1 \text{ times}} \operatorname{bipyr }Q\end{aligned}$$ is a meek $n$–polytope with special $m$–simplex. \[KorSpezSimp4\]
We will always assume from now on w.l.o.g. that the apex of the pyramid $\operatorname{pyr }P$ has the coordinates $(0,\ldots,0,1)^T$.\
Unfortunately, in the class of meek polytopes with special simplices one cannot construct $\operatorname{\mathcal{M}}$ in total just by constructing pyramids of bipyramids of arbitrary polytopes. The reason is that in an $n$–pyramid with special $m$–simplex ${\Sigma}$ there has to be an $(m-1)$–face $F$ of ${\Sigma}$ which is contained in the base of the pyramid but not in the boundary of the base of the pyramid. Thus, $F$ may not be an element of the face lattice of the pyramid (note that in low dimensions this may only appear with special 2–simplices since vertices need to be in the face lattice). But there are examples of polytopes with special simplex ${\Sigma}$ which have every element of the face lattice of ${\Sigma}$ in ${{\ensuremath{\mathcal{F}}}\xspace}(P)$: for example the bipyramid of a 2–simplex (this is the case $(1,5,9,6,1)$ in Corollary \[KorSpezSimp2\] which was already introduced in Example \[ExaSpezSimp1\]; the 2–simplex is the interesting one here) or the Birkhoff polytope $P_3$ of all doubly stochastic $3 \times 3$ matrices — a 4–dimensional polytope with special 2–simplex whose face lattice is completely contained in the face complex of $P_3$.
Hence there has to be another method to create polytopes with special simplex which avoids the problem mentioned above. It turns out that this method is the direct sum:
The direct sum ${\Sigma}\oplus Q$ of an $k$–simplex ${\Sigma}$ and an $m$–polytope $Q$ is a meek $(m+k)$–polytope and always contains ${\Sigma}$ as a special $k$–simplex. Furthermore $$\begin{aligned}
\forall F \in {{\ensuremath{\mathcal{F}}}\xspace}(\partial {\Sigma}) : F \in {{\ensuremath{\mathcal{F}}}\xspace}({\Sigma}\oplus Q),\end{aligned}$$ holds, i.e. all elements of the boundary complex of ${\Sigma}$ are contained in the face complex of the direct sum ${\Sigma}\oplus Q$. \[SadefShell\]
Recall that ${{\ensuremath{\mathcal{F}}}\xspace}(\partial ({\Sigma}\oplus Q)) = {{\ensuremath{\mathcal{F}}}\xspace}(\partial {\Sigma}) * {{\ensuremath{\mathcal{F}}}\xspace}(\partial Q)$. By definition of the polytopal join every face of ${{\ensuremath{\mathcal{F}}}\xspace}(\partial {\Sigma}) * {{\ensuremath{\mathcal{F}}}\xspace}(\partial Q)$ is the convex hull of a facet $F_1$ of ${{\ensuremath{\mathcal{F}}}\xspace}(\partial Q)$ and a facet $F_2$ of ${{\ensuremath{\mathcal{F}}}\xspace}(\partial {\Sigma})$. Since the affine basis spaces of $Q$ and ${\Sigma}$ intersect in only one point located in the interior of both complexes all vertices of both facets have to be contained in $\operatorname{conv }({{\ensuremath{\mathcal{V}}}\xspace}(F_1) \cup {{\ensuremath{\mathcal{V}}}\xspace}(F_2))$. Thus, ${\Sigma}\oplus Q$ contains a special simplex since ${\Sigma}$ is simplex and hence every facet of ${\Sigma}$ contains all but one vertex of ${\Sigma}$.
By definition of the direct sum it follows that ${{\ensuremath{\mathcal{F}}}\xspace}(P) {\backslash}{{\ensuremath{\mathcal{V}}}\xspace}({\Sigma})$ is contained in the affine basis space of $Q$ and therefore that ${\Sigma}\oplus Q$ is meek.
By definition of the polytopal join the convex hull of every pair of a facet of $Q$ and a facet of ${\Sigma}$ is a facet of ${\Sigma}\oplus Q$. This implies in particular that every facet and hence every face of ${\Sigma}$ is contained in the boundary of ${\Sigma}\oplus Q$.
Observe that every pyramid of a polytope $P$ with special $m$–simplex ${\Sigma}_m$ and basis polytope $Q$ may easily be transformed into ${\Sigma}_m \oplus Q$ geometricly. One just has to push ${\Sigma}_m$ out of the pyramids base into the new dimension emerged during pyramid construction in the opposite direction of the apex of the pyramid.
Observe, on the other way around, that one may transform ${\Sigma}_k \oplus Q$ into ${\operatorname{pyr }_{j}}({\Sigma}_{k-j} \oplus Q)$ by intersecting ${\Sigma}_k \oplus Q$ with $j$ hyperplanes such that every hyperplane is parallel to a facet $F$ of ${\Sigma}_k$, contains $Q$ and is oriented such that $F$ is in the negative halfspace. Notice that a hyperplane is determined that way since, if $\dim({\Sigma}_k \oplus Q) = n$, then it has to contain the intersection point of $Q$ and ${\Sigma}_k$, the basis space of $Q$ has dimension $n - k$ and the facet, it has to be parallel to, has dimension $k - 1$.\
The next step is to explain the combinatorical structure (i.e. here: the $f$–vector) of ${\Sigma}_k \oplus Q$ for an arbitrary polytope $Q$. But this is easy due to ${{\ensuremath{\mathcal{F}}}\xspace}(\partial ({\Sigma}_k \oplus Q)) = {{\ensuremath{\mathcal{F}}}\xspace}(\partial Q) * {{\ensuremath{\mathcal{F}}}\xspace}(\partial {\Sigma}_k)$. This means pictorially that (for every $j \in 1 \leq i + \dim(Q)$) the $j$–faces of the face lattice of ${\Sigma}_k \oplus Q$ are all the combinatorially combinable ones from the lattices ${{\ensuremath{\mathcal{F}}}\xspace}(\partial {\Sigma}_k)$ and ${{\ensuremath{\mathcal{F}}}\xspace}(\partial Q)$. This is given more specificly by the following proposition:
Let $Q$ be a $q$–polytope with $f$–vector $f_Q := (1,f_{Q_0},f_{Q_1},\ldots,f_{Q_{q-1}},1)$. Then the $f$–vector $f_{{\Sigma}_k \oplus Q}$ of ${\Sigma}_k \oplus Q$ is given by $$\begin{aligned}
f_{({\Sigma}_k \oplus Q)_j} & = & \sum_{r,l : r + l = j - 1} f_{Q_r} \cdot \binom{i+1}{l+1} \quad \text{ for all } j \in \{0,\ldots,q+i\}.\end{aligned}$$ \[SatzKombStrukShell\]
By the definition of direct sums we know that the intersection of the affine basis spaces of $Q$ and ${\Sigma}_k$ is a single point which is not part of the boundary of ${\Sigma}_k$. Furthermore, we know that all elements of ${{\ensuremath{\mathcal{F}}}\xspace}(\partial {\Sigma}_k)$ are in ${{\ensuremath{\mathcal{F}}}\xspace}({\Sigma}_k \oplus Q)$. Since ${\Sigma}_k$ is a special simplex, condition (b) of Proposition \[SatzAequivSpezSimp\] implies that all elements of ${{\ensuremath{\mathcal{F}}}\xspace}(\partial Q)$ are in ${{\ensuremath{\mathcal{F}}}\xspace}({\Sigma}_k \oplus Q)$ either. Thus, every pair of an $r$–face of ${{\ensuremath{\mathcal{F}}}\xspace}(\partial {\Sigma}_k)$ and an $l$–face of ${{\ensuremath{\mathcal{F}}}\xspace}(\partial Q)$ forms a $r + l + 1$–face of ${{\ensuremath{\mathcal{F}}}\xspace}({\Sigma}_k \oplus Q)$. Since the number of $l$–faces in an $k$–simplex is $\binom{k+1}{l+1}$ the assertion follows.
Full classification of meek polytopes with special simplices {#AbsThmKlassSpezSimp}
============================================================
After discussing the relation between direct sums and pyramids (of bipyramids) of polytopes we finally need to check if there exist additional meek polytopes with special simplices which are not obtained by these two constructions. The following main theorem (which was already given roughly as Theorem \[ThmKlassSpezSimpRough\] in the introduction) yields that this is not the case and therefore gives a full classification of $\operatorname{\mathcal{M}}_{(n,m)}$ since the $f$–vectors of pyramids and direct sums are known:
Let $P$ be an $n$–polytope. Then $P$ is meek and contains a special $m$–simplex ${\Sigma}$ if and only if $P$ has an $(n - m)$–basis polytope $Q$ (cf. Definition \[DefBasispolytop\]) such that $$\begin{aligned}
&&
\begin{array}{cccccccc}
\exists \, i \in {{\ensuremath{\mathbb{N}}}\xspace}, j \in {{\ensuremath{\mathbb{N}}}\xspace}_{>0} : & i + j & = & m & \text{ and } & P & = & {\operatorname{pyr }_{i}}{\left}({\Sigma}_j \oplus Q{\right});
\end{array}\end{aligned}$$ with ${\Sigma}_j$ denoting a $j$–simplex. In other words: $$\begin{aligned}
&& \forall n, m \in {{\ensuremath{\mathbb{N}}}\xspace}: \operatorname{\mathcal{M}}_{(n,m)} =\\
&& \qquad \{P \ | \ \exists \, i, j \in {{\ensuremath{\mathbb{N}}}\xspace}, \exists \, Q : i + j = m {\wedge}\dim(Q) = n - m {\wedge}P = {\operatorname{pyr }_{i}}{\left}({\Sigma}_j \oplus Q{\right})\}.\end{aligned}$$ \[ThmKlassSpezSimp\]
Let $P$ be an arbitrary $n$–polytope with special $m$–simplex ${\Sigma}$. According to condition (b) in Proposition \[SatzAequivSpezSimp\] we know that $P$ has an $(n-m)$–basis polytope $Q$ and all elements in ${{\ensuremath{\mathcal{F}}}\xspace}(Q)$ are elements in ${{\ensuremath{\mathcal{F}}}\xspace}(P)$. ${\Sigma}\cap Q$ is a single point which is contained in the interior of $Q$ since $P$ is meek and hence each facet of $P$ is the convex hull of a facet of $Q$ and a facet of ${\Sigma}$ lying in the boundary complex of $P$. We know further that we are in the situation $$\begin{aligned}
P & = & {\Sigma}\oplus Q\end{aligned}$$ if and only if all elements of ${{\ensuremath{\mathcal{F}}}\xspace}(\partial {\Sigma})$ are elements of ${{\ensuremath{\mathcal{F}}}\xspace}(P)$. If an $i$–face $F$ (with ${{\ensuremath{\mathcal{V}}}\xspace}(F) := \{v_{F_1},\ldots,v_{F_r}\}$) of ${\Sigma}$ is not an element of ${{\ensuremath{\mathcal{F}}}\xspace}(P)$, then we know due to the construction of a pyramid (and the resulting affine dependence of ${{\ensuremath{\mathcal{V}}}\xspace}(F)$ and ${{\ensuremath{\mathcal{V}}}\xspace}(Q)$) that all $(i+1)$–, …, $(m-1)$–faces (thus $i$–, $(i+1)$–,…, $(m-1)$–simplices) containing ${{\ensuremath{\mathcal{V}}}\xspace}(F)$ are not elements of ${{\ensuremath{\mathcal{F}}}\xspace}(P)$ either. Hence we can realize this situation by $$\begin{aligned}
P & = & {\operatorname{pyr }_{i}}{\left}({\Sigma}_{m-i} \oplus Q){\right}) \label{EquMainThm1}.\end{aligned}$$ Now we show that no further cases are possible: Let $A_Q$ and $A_{\Sigma}$ denote the affine basis spaces of $Q$ and ${\Sigma}$. Moving the vertices of ${{\ensuremath{\mathcal{V}}}\xspace}(Q)$ in $A_Q$ changes either nothing of ${{\ensuremath{\mathcal{F}}}\xspace}(P)$ or it also changes ${{\ensuremath{\mathcal{F}}}\xspace}(Q)$ which is not allowed. Moving the vertices of ${\Sigma}$ in $A_{\Sigma}$ combinatorically only effects the position of the point where ${\Sigma}$ intersects $Q$ and thus which faces of ${\Sigma}$ are not contained in ${{\ensuremath{\mathcal{F}}}\xspace}(P)$. Assume there is a further $i'$–face $G$ of ${\Sigma}$ (with ${{\ensuremath{\mathcal{V}}}\xspace}(G) := \{v_{G_1},\ldots,v_{G_s}\}$) next to $F$ not being an element of ${{\ensuremath{\mathcal{F}}}\xspace}(P)$ and there is no $i$ such that (\[EquMainThm1\]) holds. But then the $i''$–face $F \cap G$ is also not contained in ${{\ensuremath{\mathcal{F}}}\xspace}(P)$. If $i'' \leq 1$ this is a contradiction, since ${{\ensuremath{\mathcal{V}}}\xspace}({\Sigma})$ has to be contained in ${{\ensuremath{\mathcal{F}}}\xspace}(P)$ and otherwise we are directly in the situation $$\begin{aligned}
P & = & {\operatorname{pyr }_{i''}}{\left}({\Sigma}_{m-i''} \oplus Q){\right})\end{aligned}$$ in contradiction to our assumption since in this case all $(i''+ 1)$–, …, $(m-1)$–faces of ${\Sigma}$ containing ${{\ensuremath{\mathcal{V}}}\xspace}(G) \cap {{\ensuremath{\mathcal{V}}}\xspace}(F)$ may not be element of ${{\ensuremath{\mathcal{F}}}\xspace}(P)$ again.
Wild polytopes with special simplices {#SubSecWildePolytope}
=====================================
In Remark \[RemPolymayproofild\] and Example \[ExaWuerfelistwild\] we have demonstrated (by looking at the 3–cube) the necessity of the distinction between meek and wild polytopes with special simplex as we have defined it in Definition \[DefZahmWild\]. After classifying the meek polytopes and demonstrating how to construct them geometricly in the former sections one would like to do the same for wild polytopes.
Unfortunately it is not possible to give an analogue complete classification of the wild polytopes with special simplex in general since the question how many exist for a specific basis polytope and how their $f$–vectors look like cannot be answered without additional information about the basis polytope itself.
What we can do, is to give a full classification in the sense that every polytope in ${{\ensuremath{\mathcal{S}}}\xspace}_{(n,k)}(Q)$ can be constructed by intersecting ${\Sigma}_k \oplus Q$ with certain hypersurfaces which we can determine. This Theorem \[ThmWildPolyCharac\] will furthermore yield upper bounds for the $f$–vector of wild polytopes with special simplex.\
Firstly, we show that there is a bijection between the vertex sets of a polytope in ${{\ensuremath{\mathcal{S}}}\xspace}_{(n,k)}(Q)$ and ${\Sigma}_k \oplus Q$:
Let $P$ be an $n$–polytope with special $k$–simplex ${\Sigma}$, basis polytope $Q$. Let $A_{\Sigma}$ and $A_{{{\ensuremath{\mathcal{F}}}\xspace}(P) {\backslash}{{\ensuremath{\mathcal{V}}}\xspace}({\Sigma})}$ denote the affine basis spaces of ${{\ensuremath{\mathcal{F}}}\xspace}(P) {\backslash}{{\ensuremath{\mathcal{V}}}\xspace}({\Sigma})$ resp. ${\Sigma}$ such that w.l.o.g. $0 \in A_{{{\ensuremath{\mathcal{F}}}\xspace}(P) {\backslash}{{\ensuremath{\mathcal{V}}}\xspace}({\Sigma})} \cap {\Sigma}$. Let $\psi: A_{\Sigma}{\rightarrow}A_{\Sigma}$ denote the linear map $v \mapsto w + (1 + {\varepsilon}) \cdot (v-w)$ with ${\varepsilon}> 0$ with $w$ denoting the center point of the normal fan of ${\Sigma}$. Then $\psi$ and the projection $\pi: {{\ensuremath{\mathbb{R}}}\xspace}^n {\rightarrow}{{\ensuremath{\mathbb{R}}}\xspace}^n / A_{\Sigma}$ induce a bijection $\phi: {{\ensuremath{\mathcal{V}}}\xspace}(P) {\rightarrow}{{\ensuremath{\mathcal{V}}}\xspace}({\Sigma}\oplus Q)$. We call $\phi$ the *vertex projection* of $P$. \[LemVertexProjection\]
We have $\# {{\ensuremath{\mathcal{V}}}\xspace}(P) = \# {{\ensuremath{\mathcal{V}}}\xspace}({\Sigma}\oplus Q)$ since ${{\ensuremath{\mathcal{F}}}\xspace}(P) {\backslash}{{\ensuremath{\mathcal{V}}}\xspace}({\Sigma}) = {{\ensuremath{\mathcal{F}}}\xspace}(\partial Q)$ due to Proposition \[SatzAequivSpezSimp\]. If $P$ is meek the lemma is trivial since in this case we have $A_{{{\ensuremath{\mathcal{F}}}\xspace}(P) {\backslash}{{\ensuremath{\mathcal{V}}}\xspace}({\Sigma})} \cap {\Sigma}= \{0\}$ and $\pi_{|{{\ensuremath{\mathcal{V}}}\xspace}(P)|} = {\textrm{id}}$. $\psi$ ensures that $Q \cap {\Sigma}$ intersects in the interior of ${\Sigma}$ since it blows up ${\Sigma}$ slightly. Thus, we obtain ${\Sigma}_k \oplus Q$ due to Theorem \[ThmKlassSpezSimp\] and we have $\phi(v) = v$ for $v \in {{\ensuremath{\mathcal{V}}}\xspace}(P) {\backslash}{{\ensuremath{\mathcal{V}}}\xspace}({\Sigma})$ and $\phi(v) = \psi(v)$ for $v \in {{\ensuremath{\mathcal{V}}}\xspace}({\Sigma})$.
If $P$ is wild Proposition \[SatzAequivSpezSimp\] tells us that ${{\ensuremath{\mathcal{F}}}\xspace}(P) {\backslash}{{\ensuremath{\mathcal{V}}}\xspace}({\Sigma}) = {{\ensuremath{\mathcal{F}}}\xspace}(\partial Q)$ and Definition \[DefBasispolytop\] and Lemma \[Lem2ReiWel1\] yield $\pi({{\ensuremath{\mathcal{F}}}\xspace}(P) {\backslash}{{\ensuremath{\mathcal{V}}}\xspace}({\Sigma})) = {{\ensuremath{\mathcal{F}}}\xspace}(\partial Q)$. Since we know that due to Theorem \[ThmKlassSpezSimp\] there is one unique meek polytope with special simplex determined by $n$,$k$, $Q$, and the position of ${\Sigma}\cap A_{{{\ensuremath{\mathcal{F}}}\xspace}(P) {\backslash}{{\ensuremath{\mathcal{V}}}\xspace}({\Sigma})}$ in ${\Sigma}$, $\pi_{|{{\ensuremath{\mathcal{V}}}\xspace}(P) {\backslash}{{\ensuremath{\mathcal{V}}}\xspace}({\Sigma})}$ induces a bijection between ${{\ensuremath{\mathcal{V}}}\xspace}(P)$ and the vertex set of a meek polytope in ${{\ensuremath{\mathcal{M}}}\xspace}_{(n,k)}(Q)$. From this point one goes on as above.
Now we are attempting to prove our main theorem which will be a generalization of Theorem \[ThmKlassSpezSimp\]. Before we can do that we need to introduce two definitions:
Let $L(P)$ be the face lattice of an $n$–polytope $P$ in ${{\ensuremath{\mathbb{R}}}\xspace}^n$ and $Q$ an $k$–polytope such that ${{\ensuremath{\mathcal{F}}}\xspace}(\partial Q) \subseteq {{\ensuremath{\mathcal{F}}}\xspace}(\partial P)$ (and thus $L(Q) \subseteq L(P)$). Let $H$ be a hyperlane with $H \cap Q \neq \emptyset$ which is combinatorially defined by the vertices of $P$.
1. We call $H$ $Q$–*separating* if it is possible to deform $P$ into an $n$–polytope $P'$ such that there is a bijection between ${{\ensuremath{\mathcal{V}}}\xspace}(P)$ and ${{\ensuremath{\mathcal{V}}}\xspace}(P')$, ${{\ensuremath{\mathcal{F}}}\xspace}(\partial Q) \subseteq {{\ensuremath{\mathcal{F}}}\xspace}(\partial P')$ and the interior of every face of $Q$ is completely contained in $H$ or in $H^+$ or in $H^-$. We call $P'$ an $H$–*realization* of $P$ (resp. $L(P)$)
2. If there is an $H$–realization $P'$ with $H^- \cap Q = \emptyset$, we call $H$ $Q$–*supportable*.
3. If we have $L(P) = L(P')$ we say that $H$ separates (resp. supports) $Q$ *trivially*.
4. Let ${{\ensuremath{\mathcal{H}}}\xspace}$ be a set of hyperplanes. We call ${{\ensuremath{\mathcal{H}}}\xspace}$ *simultaneously* $Q$–*separating / –supportable* if there is one deformation $P'$ of $P$ such that every $H \in {{\ensuremath{\mathcal{H}}}\xspace}$ is $Q$–separating / –supportable with $H$–realization $P'$.
\[DefQSeperable\]
Observe that $Q$–separatability / –supportability is a pure combinatorial property of $P$ which does not depend on the geometry of $P$. Observe furthermore that this definition in particular implies that for an $Q$–separating hyperplane it is possible to arrange $P$ in a way such that the intersection of $H$ and an arbitrary face $F$ of $Q$ is a face of $Q$ again. If the separation is non–trivial, $H$ has to intersect $Q$ along certain connected ridges of $Q$. Finally observe that the cases $\dim(Q) \in \{1,2\}$ are easy: For $\dim(Q) = 1$ every $H$ separates $Q$ trivially. For $\dim(Q) = 2$ exactly every hyperplane intersecting $Q$ in a pair of vertices is $Q$–separating since $Q$ is an $n$–gon.\
1. Let $P = Q$ such that $P$ is the 3–polytope from figure \[AbbExaSeperation\]. The orange hyperplane in the left picture separates the polytope since every face is completely contained in the hyperplane or in its positive / negative halfspace. The lilac hyperplane in the right picture does not separate the polytope since there are three facets which are divided into the positive and the negative halfspace along the blue lines and it is impossible to move vertices in or beneath or beyond the hyperplane without changing $L(P)$. Both hyperplanes are not $Q$–supportable since this is obviously never possible if $\dim(P) = \dim(Q)$.
(400,200)(0,0) (0,0)[![A polytope with a separating and a non–separating hyperplane.[]{data-label="AbbExaSeperation"}](./Seperationb.eps "fig:")]{}
2. Let $Q$ be the 3–zonotope given in the left picture of figure \[Abb4CubeBasispoly\] and $P$ be the 4–polytope $\operatorname{pyr }Q$ with Schlegel–diagram given in the middle picture of figure \[Abb4CubeBasispoly\] (both as lattice models; the green point is the apex of the pyramid; for additional information about Schlegel–diagrams see cf. [@Zieg1 p. 132 et. seq.]). Then the set ${{\ensuremath{\mathcal{H}}}\xspace}$ of four hyperplanes (3–dimensional) which are each given by the apex and three of the turquoise vertices of $Q$ which are intersecting in the six orange 2–planes in the right picture of figure \[Abb4CubeBasispoly\] is simultaneously $Q$–separating as the right picture shows. It is even simultaneously $Q$–supportable as we will see at the end of this section.
(570,160)(0,0) (0,0)[![A 3–polytope $Q$, its pyramid $P := \operatorname{pyr }Q$ and a set of simultaneously $Q$–supportable hyperplanes.[]{data-label="Abb4CubeBasispoly"}](./4CubeBasis.eps "fig:")]{}
\[ExaQSeperable\]
Let $P$ be polytope with special simplex ${\Sigma}$, basis polytope $Q$ and vertex projection $\phi$. Let $H$ be an arbitrary hyperplane. We define the $H$–*corresponding* hyperplane ${\widehat}H$ for ${\Sigma}\oplus Q$ ($= \operatorname{conv }(\phi({{\ensuremath{\mathcal{V}}}\xspace}(P)))$) in the following way: If we have ${{\ensuremath{\mathcal{F}}}\xspace}(P) {\backslash}{{\ensuremath{\mathcal{V}}}\xspace}({\Sigma}) \subset H$ then we define ${\widehat}H$ by $Q \subset {\widehat}H$ and $$\begin{aligned}
v \in {\widehat}H & {\Leftrightarrow}& v \in H \text{ for } v \in {{\ensuremath{\mathcal{V}}}\xspace}({\Sigma}).\end{aligned}$$ If ${{\ensuremath{\mathcal{F}}}\xspace}(P) {\backslash}{{\ensuremath{\mathcal{V}}}\xspace}({\Sigma}) \not\subset H$ we define ${\widehat}H$ as the affine hull of elements of ${{\ensuremath{\mathcal{V}}}\xspace}({\Sigma}\oplus Q)$ such that by $$\begin{aligned}
\phi(v) \in {\widehat}H & {\Leftrightarrow}& v \in H \text{ for } v \in {{\ensuremath{\mathcal{V}}}\xspace}({\Sigma}) \\
\phi(w) \in {\widehat}H \cap {\widehat}H^- & {\Leftrightarrow}& w \in H \text{ for } w \in {{\ensuremath{\mathcal{V}}}\xspace}(P) {\backslash}{{\ensuremath{\mathcal{V}}}\xspace}({\Sigma}).\end{aligned}$$ \[DefCorrespHyperplane\]
That means the corresponding hyperplane intersects all those vertices $\phi(v)$ of $Q$ whose preimage $v$ is as well contained in a face of $Q$ in $H$ as in a face of $Q$ contained in $H^+$.\
Observe that for every hyperplane $H$ there always exists a corresponding hyperplane since $\phi$ maps ${{\ensuremath{\mathcal{F}}}\xspace}(P) {\backslash}{{\ensuremath{\mathcal{V}}}\xspace}({\Sigma}) \cap H$ to a connected part of ${{\ensuremath{\mathcal{F}}}\xspace}(\partial Q)$ and all $v \in {{\ensuremath{\mathcal{V}}}\xspace}({\Sigma})$ are contained in the interior of the simplex spaned up by $\{\phi(v): v \in {{\ensuremath{\mathcal{V}}}\xspace}({\Sigma})\}$.
The green supporting hyperplane of the wild polytope on the left picture has a corresponding hyperplane in the direct sum of the special one simplex and the 8–gon which is given by the violet vertices. The blue part of the 8–gon is the part of $Q$ which is separated by the corresponding hyperperlane and thus is contained in its negative halfspace.
(360,190)(0,0) (0,0)[![A supporting hyperplane of a wild polytope and its corresponding hyperplane in the direct sum. The blue part of the 8–gon is separated by the corresponding hyperplane which intersects the direct sum in the pink vertices.[]{data-label="AbbExaCorrespHyperplane"}](./CorrespHyperplaneBsp2.eps "fig:")]{}
An $n$–polytope $P$ contains a special $k$–simplex ${\Sigma}$ with basis polytope $Q$ (i.e.: $P \in \operatorname{\mathcal{S}}_{(n,k)}(Q)$) if and only if the following conditions hold: Let ${{\ensuremath{\mathcal{H}}}\xspace}$ denote the set of bounding hyperplanes of facets of $P$ and ${\widehat}{{\ensuremath{\mathcal{H}}}\xspace}$ the set of corresponding hyperplanes in ${\Sigma}\oplus Q$:
1. ${\widehat}{{\ensuremath{\mathcal{H}}}\xspace}$ is simultaneously $Q$–supportable in ${\Sigma}\oplus Q$.
2. Every ${\widehat}H \in {\widehat}{{\ensuremath{\mathcal{H}}}\xspace}$ is a bounding hyperplane of ${\Sigma}\oplus Q$, contains $Q$ or there exist a facet $F$ of ${\Sigma}\oplus Q$ such that $\operatorname{int}(F)$ is completely contained in ${\widehat}H^-$.
3. For every ${\widehat}H \in {\widehat}{{\ensuremath{\mathcal{H}}}\xspace}$ there exists exactly one $w_0 \in {{\ensuremath{\mathcal{V}}}\xspace}({\Sigma}) \cap {\widehat}H^+$.
Furthermore $P$ is meek if all corresponding hyperperlanes ${\widehat}H$ separate $Q$ trivially. \[ThmWildPolyCharac\]
First we show that conditions (a) — (c) are satisfied if $P$ is a polytope with special simplex. Due to vertex projection we have a bijection between ${{\ensuremath{\mathcal{V}}}\xspace}(P)$ and ${{\ensuremath{\mathcal{V}}}\xspace}({\Sigma}\oplus Q)$. Proposition \[SatzAequivSpezSimp\] yields ${{\ensuremath{\mathcal{F}}}\xspace}(P) {\backslash}{{\ensuremath{\mathcal{V}}}\xspace}({\Sigma}) = {{\ensuremath{\mathcal{F}}}\xspace}(\partial Q)$. This implies that the corresponding hyperplanes of all bounding hyperplanes of $P$ have to be simultaneously $Q$–supportable since otherwise it would be impossible to move the vertices of ${\widehat}H^-$ into ${\widehat}H$ (simultaneously for every ${\widehat}H \in {\widehat}{{\ensuremath{\mathcal{H}}}\xspace}$) without changing the combinatorical structure of ${{\ensuremath{\mathcal{F}}}\xspace}(P) {\backslash}{{\ensuremath{\mathcal{V}}}\xspace}({\Sigma})$; this is (a).
Let $H \in {{\ensuremath{\mathcal{H}}}\xspace}$. Let $v_1,\ldots,v_j \in {{\ensuremath{\mathcal{V}}}\xspace}(P) {\backslash}{{\ensuremath{\mathcal{V}}}\xspace}({\Sigma})$, $w_1,\ldots,w_k \in {{\ensuremath{\mathcal{V}}}\xspace}({\Sigma})$ denote the vertices of $P$ which are contained in $H$ and $M := \{\phi(v_1),\ldots,\phi(v_j)\} \cap {\widehat}H^- \subset {{\ensuremath{\mathcal{V}}}\xspace}(Q)$ denote the vertices of $Q$ which are contained in ${\widehat}H^-$. Notice that $\{\phi(w_1),\ldots,\phi(w_k)\} \subset {\widehat}H$ or $\{\phi(w_1),\ldots,\phi(w_k)\} \subset {\widehat}H^-$. If $M = \emptyset$, i.e. if ${\widehat}H$ separates $Q$ trivially, and ${{\ensuremath{\mathcal{F}}}\xspace}(P) {\backslash}{{\ensuremath{\mathcal{V}}}\xspace}({\Sigma}) \not\subset H$ we have $\{\phi(w_1),\ldots,\phi(w_k)\} \subset {\widehat}H$ and ${\widehat}H$ is a bounding hyperplane. If $M = \emptyset$ and ${{\ensuremath{\mathcal{F}}}\xspace}(P) {\backslash}{{\ensuremath{\mathcal{V}}}\xspace}({\Sigma}) \subset H$, we have $\{\phi(w_1),\ldots,\phi(w_k)\} \subset {\widehat}H^-$. Then $\phi(w_1),\ldots,\phi(w_k)$ forms a facet of ${\Sigma}\oplus Q$ with every facet of $Q$. These are contained in ${\widehat}H^-$. If $M \neq \emptyset$, then $\phi(v_1),\ldots,\phi(v_j)$ form at least two facets of $Q$ since ${\widehat}H$ is $Q$–supportable and during the deformation from ${\Sigma}\oplus Q$ to $P$ there are only vertices in ${\widehat}H^-$ moved into ${\widehat}H$ due to Definition \[DefCorrespHyperplane\]. These facets form a facet of ${\Sigma}\oplus Q$ together with $\phi(w_1),\ldots,\phi(w_k)$; this is (b).
Let ${{\ensuremath{\mathcal{V}}}\xspace}({\Sigma}) = \{w_0,\ldots,w_k\}$. Since ${\Sigma}$ is special simplex in $P$, we have w.l.o.g. $w_1,\ldots,w_k \in H$ and $w_0 \in H^+$. Since ${\Sigma}\subset \operatorname{conv }(\{\phi(w_0),\ldots,\phi(w_k)\})$ we have $\phi(w_1),\ldots,\phi(w_k) \in {\widehat}H \cup {\widehat}H^-$ and $\phi(w_0) \in {\widehat}H^+$ and (c) holds.\
Now we move the vertices of a polytope $P$ with special $k$–simplex ${\Sigma}$ and basis polytope $Q$ into a set of hyperplanes ${{\ensuremath{\mathcal{H}}}\xspace}$ such that for every hyperplane $H \in {{\ensuremath{\mathcal{H}}}\xspace}$ (a) — (c) holds. We show that this yields a polytope $P'$ with special $k$–simplex and basis polytope $Q$ again.
Since ${{\ensuremath{\mathcal{H}}}\xspace}$ is simultaneously $Q$–supportable due to (a) all vertices may be moved into the hyperplanes $H \in {{\ensuremath{\mathcal{H}}}\xspace}$ without changing the combinatorial structure of ${{\ensuremath{\mathcal{F}}}\xspace}(P) {\backslash}{{\ensuremath{\mathcal{V}}}\xspace}({\Sigma})$. Hence $P'$ has a basis polytope $Q$ again.
Condition (b) yields that $H \cap P'$ is $(n-1)$–dimensional again since $H^-$ contains at least one facet of $P$, the vertices of ${\Sigma}$ remain a simplex since $P'$ is an $n$–polytope and no vertex of ${\Sigma}$ may be contained in the affine span of a face of $Q$ since otherwise the structure of ${{\ensuremath{\mathcal{F}}}\xspace}(P) {\backslash}{{\ensuremath{\mathcal{V}}}\xspace}({\Sigma})$ was changed.
For every $H \in {{\ensuremath{\mathcal{H}}}\xspace}$ there is exactly one $w_0 \in H^+$ due to (c). Thus $\{w_1,\ldots,w_k\} := {{\ensuremath{\mathcal{V}}}\xspace}({\Sigma}) {\backslash}\{w_0\} \subset H \cup H^-$ and will be moved into $H$ and therefore $P'$ has a special $k$–simplex again.\
It remains to prove the condition for meekness of $P$. Let $P$ be as before with $P = ({\Sigma}\oplus Q) \cap \bigcap_{H \in {{\ensuremath{\mathcal{H}}}\xspace}} H$ such that (a) — (c) are satisfied and furthermore all $H$ separate $Q$ trivially. We may assume w.l.o.g. that all $H \in {{\ensuremath{\mathcal{H}}}\xspace}$ are no bounding hyperplanes of $P$. Then $Q \subset \bigcap_{H \in {{\ensuremath{\mathcal{H}}}\xspace}} H$ and thus ${{\ensuremath{\mathcal{F}}}\xspace}(P) {\backslash}{\Sigma}$ is contained in the affine basis space of $Q$. Hence $P$ is meek. Notice that the intersection with hyperplanes of this type corresponds to the transformation of direct sums into pyramids via intersections which we already mentioned after Proposition \[SadefShell\].
Note that the Theorem does not state that it is possible to move all vertices of ${\Sigma}\oplus Q$ in the corresponding hyperplanes of $P$ *successively* and get a polytope in ${{\ensuremath{\mathcal{S}}}\xspace}_{(n,k)}(Q)$ after moving the vertices of one or some particular ${\widehat}H^-$. It states that one gets a polytope in ${{\ensuremath{\mathcal{S}}}\xspace}_{(n,k)}(Q)$ if one moves the vertices of ${\Sigma}\oplus Q$ to all corresponding hyperplanes *simultaneously* such that the hyperplanes satisfies the conditions (a) — (c).
Instead of moving vertices of ${\Sigma}\oplus Q$ into certain corresponding hyperplanes $P$, the whole process might of course also be interpreted as simply intersecting ${\Sigma}\oplus Q$ with the corresponding hyperplanes. This is maybe the easier way to understand the connection between $P$ and ${\Sigma}\oplus Q$ but it has the disadvantage that it is not a continuous deformation anymore.\
Thus, the reason why it is hard to classify all wild polytopes with special $k$–simplex and basis polytope $Q$ is that it is difficult to figure out the sets of simultaneously $Q$–supportable hyperplanes satisfying the upper conditions if $Q$ is sufficiently complicated. The situation gets even worse if one is not only interested in the $f$–vector of the specific wild polytope but also in its face lattice. The following example shows that — in contrast to the meek case —the face lattice of a wild polytope with certain $f$–vector is not unique:
Both polytopes on the pictures below have the $f$–vector $(1,10,22,14,1)$ and both are wild polytopes with special 1–simplex and the 8–gon as basis polytope. Anyway, the polytopes have different face lattices and thus are non–isomorphic resp. have a different geometric realization.
(400,200)(0,0) (0,0)[![Two non–isomorphic, combinatorial equivalent wild polytopes with special 1–simplex and identical basis polytope.[]{data-label="AbbExaWildPoly"}](./WildPolyBspb.eps "fig:")]{}
Indeed we are unable to give specific formulas for possible $f$–vectors of wild polytopes with special simplex without further information about the basis polytope. But we are able to find an upper bound for the $f$–vector:
Let $P \in \operatorname{\mathcal{W}}_{(n,k)}(Q)$ with special simplex ${\Sigma}$ and $P' \in \operatorname{\mathcal{M}}_{(n,k)}(Q)$ the polytope one obtains by applying the vertex projection on the vertices not belonging to ${\Sigma}$. I.e.: $P' = \operatorname{conv }(\phi({{\ensuremath{\mathcal{V}}}\xspace}(P) {\backslash}{{\ensuremath{\mathcal{V}}}\xspace}({\Sigma})) \cup {{\ensuremath{\mathcal{V}}}\xspace}({\Sigma}))$. Then we have for the $f$–vectors $f_P$ and $f_{P'}$: $$\begin{aligned}
f_{P_0} & = & f_{P'_0} \\
f_{P_i} & < & f_{P'_i} \quad \text{ for } i \in \{n-1,n\} \\
f_{P_i} & \leq & f_{P'_i} \quad \text{ for } i \in \{1,\ldots,n-2\}.\end{aligned}$$ Observe that $f_{P'}$ is componentwisely bounded by $f_{{\Sigma}\oplus Q}$ and $f_{{\Sigma}\oplus Q}$ can always be computed explicitly. \[CorBoundaries\]
$f_{P_0} = f_{P'_0}$ follows directly from the definition of $\phi$. Theorem \[ThmWildPolyCharac\] shows that the $P$ can be achieved from $P'$ by intersecting $P'$ with additional hyperplanes such that each intersection combines facets $G_1,\ldots,G_r$ of $P'$ to one large facet of $P$.
Notice furthermore that all polynomials in $\operatorname{\mathcal{S}}_{(n,k)}(Q)$ have the same reverse lexicographic triangulation with vertexorder $\tau := (v_p,\ldots,v_0)$ and ${{\ensuremath{\mathcal{V}}}\xspace}({\Sigma}) = (v_0,\ldots,v_k)$ since Athanasiadis’ Lemma \[Lem2Atha1\] tells us that for every polytope $P \in \operatorname{\mathcal{S}}_{(n,k)}(Q)$ the reverse lexicographic triangulation ${\triangle}_{\tau}(P)$ is given by: $$\begin{aligned}
{\triangle}_{\tau}(P) & = & {\triangle}_{\tau}(Q) * {\Sigma}\end{aligned}$$
As we have already mentioned after Definition \[DefQSeperable\], $Q$–separability / –supportability becomes rather trivial if $\dim(Q) = 2$. Thus, we want to handle this case separately:
All Polytopes $P \in {{\ensuremath{\mathcal{S}}}\xspace}_{n,n-2}(Q)$ can be constructed out of ${\Sigma}_{n-2} \oplus Q$ by intersection with hyperplanes $H \in {{\ensuremath{\mathcal{H}}}\xspace}$ defined on ${{\ensuremath{\mathcal{V}}}\xspace}({\Sigma}_{n-2} \oplus Q)$ such that for every $H \in {{\ensuremath{\mathcal{H}}}\xspace}$:
1. There exists exactly one $w_0 \in H^+ \cap {{\ensuremath{\mathcal{V}}}\xspace}({\Sigma}_{n-2})$, $H$ contains exactly two $v_0,v_1 \in {{\ensuremath{\mathcal{V}}}\xspace}(Q)$ and all $w_i \in {{\ensuremath{\mathcal{V}}}\xspace}({\Sigma}) {\backslash}\{w_0\}$ or it contains $Q$.
2. Every vertex $v \in {{\ensuremath{\mathcal{V}}}\xspace}(Q)$ is contained in at most $n-2$ negative halfspaces of $H \in {{\ensuremath{\mathcal{H}}}\xspace}$ or $Q$–containing hyperplanes.
\[CorWildPolyBasisPolyDim2\]
We have to show that the two upper conditions fall together with the three conditions of Theorem \[ThmWildPolyCharac\] for the case $\dim(Q) = 2$. First we show that the conditions of the corollary imply the three conditions of the theorem. Observe that every single hyperplane defined in (a) is $Q$–supportable since $Q$ is an $n$–gon and thus simplicial and only consists of vertices and facets. Hence every vertex in $H^-$ can be moved into $H$ without changing the face lattice of ${{\ensuremath{\mathcal{F}}}\xspace}(\partial Q)$. Condition (b) ensures that $v$ is still contained in $n$ facets after intersecting with all hyperplanes. In ${\Sigma}_{n-2} \oplus Q$ every vertex $v$ is contained in $2(n-1)$ facets: $v$ is contained in 2 edges of $Q$, every edge forms $n-1$ facets of the direct sum induced by the $n-1$ subsets of ${{\ensuremath{\mathcal{V}}}\xspace}({\Sigma}_{n-2})$ of cardinality $n-2$. Every intersections transforms two of these hyperplanes into a single big one. Thus, $v$ may be contained in at most $n-2$ negative halfspaces since otherwise $v$ would lie only in $n-1$ hyperplanes at all and thus be no vertex of $P$ anymore. This is condition (a) of Theorem \[ThmWildPolyCharac\]. Conditions (b) and (c) are trivial.\
The inverse way is easy since (b) follows from $Q$–supportability (with the argument from above), $w_0 \in H^+$ is trivial and every hyperplane may only contain $2$ or all vertices of $Q$ since it is an $n$–gon. If $H$ contains only $2$ vertices of ${{\ensuremath{\mathcal{V}}}\xspace}(Q)$ we obviously need $n-2$ more to define a hyperplane and hence ${\left}({{\ensuremath{\mathcal{V}}}\xspace}({\Sigma}) {\backslash}\{w_0\}{\right}) \subset H$
Although one does not get *one* combinatorial description of all wild polytopes with special simplex we want to close the article with showing that in principle one can get very explicit results for certain classes of wild polytopes with special simplex. We show that by investigating the $n$–cubes $\square_n$ which always come with a special 1–simplex due to the fact that they are weakly Hannar (cf. Example \[ExaWeaklyHannar\]).
Since we have seen that the wild polytopes with special simplex and 2–dimensional basis polytopes are rather easy, we want to start with discussing the 4–cube $\square_4$ as a non–trivial example since it is a wild polytope with special 1–simplex and 3–dimensional basis polytope. The Schlegel–diagram of the 4–cube looks the following way (cf. eg. [@Zieg1 p. 136]):
(150,150)(0,0) (0,0)[![The Schlegel–diagram of a 4–cube.[]{data-label="Abb4CubeSchlegel"}](./4CubeSchlegel.eps "fig:")]{}
The two green vertices form a special 1–simplex ${\Sigma}$. Each turquoise vertex is connected to one vertex of ${\Sigma}$; each pink vertex is not connected to a vertex of ${\Sigma}$. Observe that there are seven other special 1–simplices due to the symmetry of the 4–cube.
The complex ${{\ensuremath{\mathcal{F}}}\xspace}(\square_4) {\backslash}{{\ensuremath{\mathcal{V}}}\xspace}({\Sigma})$ consists of all faces of $\square_4$ not containing a green vertex. This is obviously a 2–dimensional polytopal complex since it isomorphic to the boundary complex of the 3–dimensional basis polytope. It consists of twelve 4–gons which are the orange planes in the following figure. We splitted it up in two pictures for clarity reasons. In the left picture we have three 4–gons in the outer 3–cube and three 4–gons connecting the outer with the inner cube. In the right picture we have three 4–gons in the inner cube and again three 4–gons connecting the outer with the inner cube.
(300,150)(0,0) (0,0)[![The boundary complex of the basis polytope in the 4–cube.[]{data-label="Abb4CubeBasisComplex"}](./4CubeBasisComplex.eps "fig:")]{}
If one realizes this polytopal complex as the boundary complex of a 3–polytope, the basis polytope of $\square_4$, it is easy to see that one gets a polytope which is combinatorially equivalent with the one on the left picture of figure \[Abb4CubeBasispoly\] (cf. p. ). We denote this polytope as $Q$. Thus, $\square_4$ is realized by intersecting ${\Sigma}_1 \oplus Q$ with a couple of hyperplanes. Observe that ${\Sigma}_1 \oplus Q$ is a bipyramid such that both of the pyramids have a Schlegel–diagram as in the middle picture of figure \[Abb4CubeBasispoly\]. In fact, $\square_4$ is realized by intersecting each pyramid of the bipyramid ${\Sigma}_2 \oplus Q$ with the four 3–spaces given in the right picture of figure \[Abb4CubeBasispoly\] (respectively the other four 3–spaces defined by each 3–set of the other four turquoise vertices). To see that fill in all missing faces in the Schlegel–diagram of $\square_4$ to transform it into ${\Sigma}_1 \oplus Q$ and observe that one has locally the situation on the left picture of the following figure which one has to transform in the right one to obtain one facet of $\square_4$:
(300,100)(0,0) (0,0)[![Transformation of facets of a direct sum into a facet of the 4–cube.[]{data-label="Abb4CubeFacet"}](./4CubeFacet.eps "fig:")]{}
This is exactly what happens in the right picture of figure \[Abb4CubeBasispoly\] and proves at the same time that the set of hyperplanes in the picture are simultaneously $Q$–separating as we have claimed it in Example \[ExaQSeperable\].
$\square_4$ together with the $n$–cube in lower dimensions already provide the structure of the basis polytopes of $\square_n$:
The basis polytope $Q_n$ of $\square_n$ is the zonotope given by the Minkowski sum of $\square_{(n-1)}$ and the vector $(1,\ldots,1) \in {{\ensuremath{\mathbb{R}}}\xspace}^{n-1}$. Thus, it has the $f$–vector which is given by $$\begin{aligned}
f_{(Q_n)}^{(0)} & = & f_{(\square_n)}^{(0)} + f_{(Q_{(n-1)}}^{(0)} \text{ and } \\
f_{(Q_n)}^{(i)} & = & f_{(\square_n)}^{(i)} + f_{(Q_{(n-1)}}^{(i)} + f_{(Q_{(n-1)}}^{(i-1)} \text{ for all } i \in \{1,\ldots,n-1\}\end{aligned}$$ \[PropBasispolyCube\]
Let w.l.o.g. all $\square_n$ be centrally symmetric with vertices in $\{-1,1\}^n$. For every $\square_n$ we call the set of all faces which contain the point $(-1,\ldots,-1) \in {{\ensuremath{\mathbb{R}}}\xspace}^n$ the *lower* and the set of those which contain the point $(1,\ldots,1)$ the *upper half* of $\square_n$. Observe that the intersection of the closure of the lower and the upper half is the basis polytope $Q_n$ since $\square_n$ is weakly Hannar and hence $(-1,\ldots,-1),(1,\ldots,1)$ form a special 1–simplex in $\square_n$.
Let w.l.o.g. $(-1,\ldots,-1,1),(1,\ldots,1,-1)$ be the vertices of the special simplex ${\Sigma}_n$ we want to look at in $\square_n$. We investigate ${{\ensuremath{\mathcal{F}}}\xspace}(\square_n) {\backslash}{{\ensuremath{\mathcal{V}}}\xspace}({\Sigma}_n)$: $\square_n$ is the prism over $\square_n$. Hence on the bottom ${{\ensuremath{\mathcal{F}}}\xspace}(\square_n) {\backslash}{{\ensuremath{\mathcal{V}}}\xspace}({\Sigma}_n)$ consists of the closure of the lower half of $\square_{n-1}$, because $(1,\ldots,1,-1)$ is also one vertex of a special simplex ${\Sigma}_{n-1}$ in $\square_{n-1}$. The same argument holds for the top with the upper half of $\square_{n-1}$. Between top and bottom ${{\ensuremath{\mathcal{F}}}\xspace}(\square_n) {\backslash}{{\ensuremath{\mathcal{V}}}\xspace}({\Sigma}_n)$ contains all the faces $F \times (0,\ldots,0,2)$ where $F$ denotes the faces which are in the closure of the lower half of $\square_{n-1}$ on the bottom and $F$ shifted by $(0,\ldots,2)$ is contained in the closure of the upper half of $\square_{n-1}$ shifted by $(0,\ldots,2)$ in the top. These are exactly the faces belonging to $Q_{n-1}$.
This already gives the $f$–vector. Now we investigate the zonotope $\square_{n-1} + (1,\ldots,1)$: The Minkowski sum leaves the lower half of $\square_{n-1}$ invariant, turns every face $F$ of the intersection of the closure of the lower and the upper half (which is $Q_{n-1}$) into $F \times (1,\ldots,1)$ and shifts the upper half of $\square_{n-1}$ along $(1,\ldots,1)$ such that every face $F$ of the upper half, belonging to $Q_{n-1}$, is transfered to $F$ shifted by $(1,\ldots,1)$. Hence the face lattices of $\square_{n-1} + (1,\ldots,1)$ and $Q_n$ coincide.
Timo de Wolff\
Institute of Mathematics, Goethe–University Frankfurt am Main, Germany\
email: `wolff(at)math.uni-frankfurt.de`\
<http://www.uni-frankfurt.de/fb/fb12/mathematik/dm/personen/dewolff/>
All figures were constructed with the open source 3D creator *Blender*. See: [www.blender.org](www.blender.org)
[^1]: He calls a polytope with special 1–simplex a *spindle*.
[^2]: Background for this condition is that it might be possible to shift some vertices of $Q$ out of the affine basis space without changing the combinatorial structure of $\operatorname{conv }({{\ensuremath{\mathcal{V}}}\xspace}(Q) \cup {{\ensuremath{\mathcal{V}}}\xspace}({\Sigma}))$. Since we are only interested in combinatorial classification and geometric construction we want to omit these special cases.
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'H[ø]{}gni C. Kamban'
- 'Thomas G. Pedersen'
bibliography:
- 'litt.bib'
title: 'Interlayer excitons in van der Waals heterostructures: Binding energy, Stark shift, and field-induced dissociation'
---
Naturally occurring layered materials held together by van-der-Waals-like forces have been intensely studied in recent years. Peeling graphite layer-by-layer and forming graphene [@Novoselov2004], a task thought impossible, turned researchers on to two-dimensional materials. Realizing that these layers may be used as building blocks of artificially layered materials, so-called van der Waals heterostructures, provides an endless array of possible combinations [@Wang2012; @Geim2013]. The extraordinary electronic and optical properties of TMDs [@Mak2010; @Splendiani2010; @Ramasubramaniam2012; @Tongay2013; @Qiu2013; @Gutierrez2013; @Mouri2013; @Scheuschner2014] have made them one of the most interesting classes of building blocks for vdWHs. Potential electro-optical applications of TMDs include photodetectors [@Wang2015; @Lopez2013; @Yin2012], light-emitting diodes [@Ross2014; @Pospischil2014; @Withers2015], and solar cells [@Lopez2014; @Bernardi2013; @Pospischil2014]. It is well known that the optical properties of TMD monolayers are dominated by excitons [@Wang2012; @Geim2013; @Ramasubramaniam2012; @Qiu2013; @Trolle2015], as such two-dimensional excitons can have giant binding energies [@Ramasubramaniam2012; @Olsen2016; @Qiu2013; @Chernikov2014; @Hanbicki2015]. Strongly bound excitons, in turn, make generation of photocurrents difficult, as excitons must first be dissociated into free electrons and holes. This is one of the challenges facing the use of monolayer TMDs in efficient photocurrent devices. A further complication is the fast recombination rates of excitons in these monolayers [@Palummo2015; @Poellmann2015; @Wang2016]. Without inducing dissociation in some way, excitons will typically recombine before they are dissociated. Applying an in-plane electric field to the excitons, however, enhances generation of photocurrents for two reasons: (i) the electric field counteracts recombination by pulling electrons and holes in opposite directions, and (ii) the electric field assists dissociation of excitons [@Pedersen2016ExcitonIonization; @Haastrup2016; @Massicotte2018; @Kamban2019].
When two TMD monolayers are brought together with a type-II band alignment, the conduction band minimum and the valence band maximum reside in two different layers. Electrons and holes in the structure will therefore prefer to reside in separate layers, provided that the loss in exciton binding energy is smaller than the energy gained by band offsets. These spatially indirect electron-hole pairs can still form bound states with large binding energies and they are referred to as interlayer excitons [@Hong2014; @Chen2016; @Miller2017; @Kunstmann2018; @Merkl2019]. Experiments have shown that photoexcited intralayer excitons created in one of the sheets undergo ultrafast tunneling into interlayer excitons [@Merkl2019; @Hong2014]. For instance, in MoS$_2$/WS$_2$ heterostructures, it was found that the hole transferred from the MoS$_2$ layer to the WS$_2$ layer within $50$ fs after optical excitation [@Hong2014], and similar time scales were reported for WSe$_2$/WS$_2$ [@Merkl2019]. After tunneling, the interlayer excitons have a long lifetime due to the small overlap of electron and hole wave functions [@Fogler2014; @Rivera2015; @Rivera2016; @Miller2017; @Nagler2017; @Jin2018]. Thus, when a weak in-plane electric field is present, the photoexcited intralayer excitons will tunnel into interlayer excitons with sufficiently long lifetimes for them to be dissociated by the electric field. We therefore expect to see significantly larger photocurrents from this type of structure compared to TMD monolayers. This is corroborated by the much larger dissociation rates found for interlayer excitons. As an example, we find that interlayer excitons in freely suspended MoS$_2$/WS$_2$ have a dissociation rate of $\Gamma\approx 1.7\times 10^{4}\, \mathrm{s}^{-1}$ for a field strength of $10\,\mathrm{V/\mu m}$, whereas for excitons in monolayer MoS$_2$ it is only $\Gamma\approx 5.3\times 10^{-38}\, \mathrm{s}^{-1}$.
The paper is organized as follows. In \[sec:II\], we set up the Wannier equation and analyse the interaction between electron-hole pairs in bilayer vdWHs. This is done by solving the multilayer Poisson equation, and subsequently showing that the full solution is excellently approximated by an analytical bilayer Keldysh potential. In \[sec:III\], we then turn to computing binding energies as well as Stark shifts and dissociation rates for interlayer excitons. Here, we also compare the numerically exact results to an analytical weak-field approximation, derived by weak-field asymptotic theory. Finally, \[sec:IV\] concludes upon the results. The solution to the multilayer Poisson equation is presented in \[app:poisson\].
Interlayer excitons {#sec:II}
===================
![Interlayer exciton in a bilayer van der Waals heterostructure with zero external field (a), a weak in-plane field polarizing the exciton (b), and a strong in-plane field dissociating the exciton (c). []{data-label="fig:TMDs"}](Current_3_layer_excitonsvg.pdf){width="1\columnwidth"}
![Sketch of the layered geometry used to describe the interaction between an electron at position $z_e$ and a hole at $z_h$. []{data-label="fig:sketch"}](Dielectric_Sketch.pdf){width="0.5\columnwidth"}
A bilayer vdWH supports two distinct types of excitons. The electron and hole may either be localized within the same layer, or they may reside in different layers. These two cases are referred to as intra- and interlayer excitons, respectively. The vdWHs considered in the present paper are bilayers with type-II band alignment. We name the structures TMD1/TMD2 so that the conduction band minimum and valence band maximum reside in the first and second layer, respectively. For the structures considered, the energy won from band offsets by the electron and hole residing in the first and second layer, respectively, is larger than the loss in exciton binding energy. The many-body excitonic ground state is therefore an interlayer exciton. Intralayer excitons in either layer are thus excited states, and will therefore tunnel into the ground state. Direct photoexcitation of interlayer excitons by resonance photons is not as likely as for intralayer excitons due to the weak overlap of the electron and hole wave functions. However, resonantly excited intralayer excitons in either layer quickly transition into interlayer excitons [@Merkl2019] that, therefore, become very important for many properties of vdWHs. To describe excitonic effects from first principles, one must turn to the many-body Bethe-Salpeter equation [@Salpeter1951; @Onida2002]. Solving it is a computationally demanding task even for simple structures. Fortunately, under well-defined approximations, the many-body problem can be simplified to the Wannier equation [@Wannier1937; @Lederman1976], essentially reducing it to a Schrödinger-type problem with a hydrogenic Hamiltonian. The Wannier model has indeed been shown to yield a sufficiently accurate description of many excitonic properties [@Latini2015; @Cudazzo2010; @Cudazzo2011; @Pedersen2016ExcitonStark; @Massicotte2018]. In terms of the relative coordinate $\boldsymbol{r} = \boldsymbol{r}_e - \boldsymbol{r}_h$ of the electron-hole pair, it reads (atomic units are used throughout) $$\begin{aligned}
{\mathopen{}\mathclose\bgroup\originalleft}[-\frac{1}{2\mu}\nabla^2 +V{\mathopen{}\mathclose\bgroup\originalleft}(\boldmath{r}{\aftergroup\egroup\originalright}) + \pmb{\mathcal{E}}\cdot \boldsymbol{r}{\aftergroup\egroup\originalright}]\psi{\mathopen{}\mathclose\bgroup\originalleft}(\boldsymbol{r}{\aftergroup\egroup\originalright})=E\psi{\mathopen{}\mathclose\bgroup\originalleft}(\boldsymbol{r}{\aftergroup\egroup\originalright})\thinspace,\label{eq:Wannier}\end{aligned}$$ where $\mu = m_em_h/{\mathopen{}\mathclose\bgroup\originalleft}(m_e+m_h{\aftergroup\egroup\originalright})$ is the reduced exciton mass, $V$ is the screened Coulomb interaction between the electron and hole, and $\pmb{\mathcal{E}}$ is the electric field. The electric field is taken to point along the $x$-axis throughout the paper, i.e. $\pmb{\mathcal{E}} = \mathcal{E}\boldsymbol{e}_x$. As the valence band maximum and conduction band minimum at the $\mathrm{K}$ point are primarily composed of the $d$ orbitals of the metal atoms [@Wang2017; @Kormanyos2015], electrons and holes will, to a good approximation, reside in the middle of their respective layers. We are therefore able to freeze their out-of-plane motion, which effectively makes solving \[eq:Wannier\] a two dimensional problem. shows an illustration of an interlayer exciton in a bilayer vdWH subjected to three different field strengths: zero (a), weak (b), and strong (c). When an electric field is present, the electron and hole will be pulled in opposite directions. For weak electric fields, the probability of dissociating the exciton is low. It will therefore become polarized, but most likely recombine rather than dissociate. In strong electric fields, however, field induced dissociation becomes likely, and dissociation rates may become extremely large as the field strength increases.
To describe the interaction $V$ between the electron-hole pair, we fix the electron and hole to the middle of their respective sheets. The vertical separation between the electron and hole $d= {\mathopen{}\mathclose\bgroup\originalleft}|z_e-z_h{\aftergroup\egroup\originalright}|$ is therefore $d_{\mathrm{intra}}=0$ and $d_{\mathrm{inter}}={\mathopen{}\mathclose\bgroup\originalleft}(d_1+d_2{\aftergroup\egroup\originalright})/2$ for intra- and interlayer excitons, respectively, where $d_1$ and $d_2$ are the thicknesses of the two TMD sheets (see \[fig:TMDs\]). The van der Waals heterostructure is then modeled as the four layer system in \[fig:sketch\]. Here, the dielectric function $\varepsilon{\mathopen{}\mathclose\bgroup\originalleft}(z{\aftergroup\egroup\originalright})$ is taken to be piecewise constant given by $\varepsilon_a$, $\varepsilon_1$, $\varepsilon_2$, and $\varepsilon_b$ in the superstrate, upper TMD sheet, lower TMD sheet, and substrate, respectively. By Fourier decomposing $V$ and solving the multilayer Poisson equation for the Fourier components (see \[app:poisson\]), one may obtain $V$ expressed as an integral in momentum space $$\begin{aligned}
V{\mathopen{}\mathclose\bgroup\originalleft}(r{\aftergroup\egroup\originalright})=-\int_0^\infty \frac{e^{-dq}J_0{\mathopen{}\mathclose\bgroup\originalleft}(qr{\aftergroup\egroup\originalright})}{\varepsilon_{\mathrm{eff}}{\mathopen{}\mathclose\bgroup\originalleft}(q{\aftergroup\egroup\originalright})}dq\thinspace,\label{eq:fullpotinarticle}\end{aligned}$$ where $J_0$ is a Bessel function. The effective dielectric function $\varepsilon_{\mathrm{eff}}$ is, of course, different for intra- and interlayer excitons. For $q\to 0$, both cases tend to the average dielectric constant of the surrounding media ${\mathopen{}\mathclose\bgroup\originalleft}(\varepsilon_a+\varepsilon_b{\aftergroup\egroup\originalright})/2$, as expected. For $q\to\infty$, however, the dielectric function describing intralayer excitons tends to the dielectric constant of the layer to which they are confined, i.e. to $\varepsilon_1$ and $\varepsilon_2$ for intralayer excitons in the first and second layer, respectively. On the other hand, the function for interlayer excitons tends to the average dielectric constant of the two layers ${\mathopen{}\mathclose\bgroup\originalleft}(\varepsilon_1+\varepsilon_2{\aftergroup\egroup\originalright})/2$. The complete interlayer dielectric function is obtained in \[app:poisson\] and is given by \[eq:dielectric\]. also explains how to obtain the intralayer function. The full potential may readily be obtained for real $r$ by using standard numerical integration techniques in \[eq:fullpotinarticle\]. However, in the present paper, we seek to calculate dissociation rates by using exterior complex scaling (ECS) [@Simon1979; @McCurdy1991; @Rescigno2000; @McCurdy2004]. This implies rotating the radial coordinate into the complex plane outside a radius $R$ by an angle $\phi$, i.e. $$\begin{aligned}
r \to \begin{cases}
r \quad &\mathrm{for }\,\, r<R\\
R + {\mathopen{}\mathclose\bgroup\originalleft}(r-R{\aftergroup\egroup\originalright})e^{i\phi} \quad &\mathrm{for }\,\, r>R\thinspace.
\end{cases}\label{eq:ECStrans}\end{aligned}$$ It is a simple task to show that the integrand in \[eq:fullpotinarticle\] will become an exponentially increasing oscillating function when $r> d/\sin\phi+R$ by using the integral representation for the Bessel function [@Abramowitz1972]. This makes \[eq:fullpotinarticle\] extremely difficult to handle while using ECS. Fortunately, the numerical solution is very accurately approximated by a bilayer Keldysh (BLK) potential. The Keldysh potential has been used extensively to describe excitons in monolayer TMDs. The monolayer Keldysh (MLK) interaction is given by [@Keldysh1979; @Trolle2017] $$\begin{aligned}
V_{\mathrm{MLK}}{\mathopen{}\mathclose\bgroup\originalleft}(r{\aftergroup\egroup\originalright}) = -\frac{\pi}{2r_0}{\mathopen{}\mathclose\bgroup\originalleft}[H_0{\mathopen{}\mathclose\bgroup\originalleft}(\frac{\kappa r}{r_0}{\aftergroup\egroup\originalright})-Y_0{\mathopen{}\mathclose\bgroup\originalleft}(\frac{\kappa r}{r_0}{\aftergroup\egroup\originalright}){\aftergroup\egroup\originalright}]\thinspace,\label{eq:monokeld}\end{aligned}$$ where $H_0$ is the zeroth order Struve function [@Abramowitz1972], $Y_0$ is the zeroth order Bessel function of the second kind [@Abramowitz1972], $\kappa = {\mathopen{}\mathclose\bgroup\originalleft}(\varepsilon_a+\varepsilon_b{\aftergroup\egroup\originalright})/2$ is the average dielectric constant of the surrounding media, and $r_0$ is the screening length proportional to the polarizability of the sheet [@Cudazzo2011]. This potential diverges logarithmically at the origin. On the other hand, the potential describing interlayer excitons is finite at the origin due to the vertical separation between the electron-hole pair.
![ (a) Exciton Coulomb potential for freely suspended WS$_2$/WSe$_2$. The full and approximate potentials are shown as the solid lines and circles, respectively. The black line and symbols show the interlayer exciton potential while the red and green lines correspond to the potential of intralayer excitons in the WS$_2$ and WSe$_2$ halves of the bilayer, respectively. (b) Binding energies for interlayer excitons in the six structures obtained with the full potential vs. those obtained by the BLK potential. []{data-label="fig:potential"}](Potential_sub_SR.pdf){width="1\columnwidth"}
To obtain the bilayer Keldysh potential, we substitute $r \to \sqrt{r^2 + d^2}$ and $r_0 \to r_0^{{\mathopen{}\mathclose\bgroup\originalleft}(1{\aftergroup\egroup\originalright})}+r_0^{{\mathopen{}\mathclose\bgroup\originalleft}(2{\aftergroup\egroup\originalright})}$ into \[eq:monokeld\] which yields $$\begin{aligned}
V_{\mathrm{BLK}}{\mathopen{}\mathclose\bgroup\originalleft}(r{\aftergroup\egroup\originalright}) = -\frac{\pi}{2{\mathopen{}\mathclose\bgroup\originalleft}(r_0^{{\mathopen{}\mathclose\bgroup\originalleft}(1{\aftergroup\egroup\originalright})}+r_0^{{\mathopen{}\mathclose\bgroup\originalleft}(2{\aftergroup\egroup\originalright})}{\aftergroup\egroup\originalright})}{\mathopen{}\mathclose\bgroup\originalleft}[H_0{\mathopen{}\mathclose\bgroup\originalleft}(\frac{\kappa \sqrt{r^2 + d^2}}{r_0^{{\mathopen{}\mathclose\bgroup\originalleft}(1{\aftergroup\egroup\originalright})}+r_0^{{\mathopen{}\mathclose\bgroup\originalleft}(2{\aftergroup\egroup\originalright})}}{\aftergroup\egroup\originalright})-Y_0{\mathopen{}\mathclose\bgroup\originalleft}(\frac{\kappa \sqrt{r^2 + d^2}}{r_0^{{\mathopen{}\mathclose\bgroup\originalleft}(1{\aftergroup\egroup\originalright})}+r_0^{{\mathopen{}\mathclose\bgroup\originalleft}(2{\aftergroup\egroup\originalright})}}{\aftergroup\egroup\originalright}){\aftergroup\egroup\originalright}]\thinspace.\label{eq:BK}\end{aligned}$$ Here, $r_0^{{\mathopen{}\mathclose\bgroup\originalleft}(1{\aftergroup\egroup\originalright})}$ and $r_0^{{\mathopen{}\mathclose\bgroup\originalleft}(2{\aftergroup\egroup\originalright})}$ are the screening lengths of the first and second monolayer, respectively. The first substitution accounts for the possible vertical separation between electrons and holes in bilayer structures. The second substitution accounts for the increased thickness of the structure, as the total screening length is proportional to it [@Keldysh1979; @Trolle2017]. It should be noted that making the first substitution without the second leads to a potential that is far too strongly binding [@Donck2018]. Note, further, that to obtain the interaction for intralayer excitons in a bilayer structure, we use the BLK potential with $d=0$. The screening lengths used in the present paper are *ab initio* values from Ref. [@Olsen2016].
[width=1]{}
[c cc c cc c cc c cc c cc c cc]{}\
& && && &&&&&&\
\
& & & & & & & & & & &\
$\kappa$ &${\mathopen{}\mathclose\bgroup\originalleft}|E_0{\aftergroup\egroup\originalright}|$ & $\Gamma_0$ & &${\mathopen{}\mathclose\bgroup\originalleft}|E_0{\aftergroup\egroup\originalright}|$ &$\Gamma_0$ & &${\mathopen{}\mathclose\bgroup\originalleft}|E_0{\aftergroup\egroup\originalright}|$ &$\Gamma_0$ & &${\mathopen{}\mathclose\bgroup\originalleft}|E_0{\aftergroup\egroup\originalright}|$ &$\Gamma_0$ & &${\mathopen{}\mathclose\bgroup\originalleft}|E_0{\aftergroup\egroup\originalright}|$ &$\Gamma_0$ & &${\mathopen{}\mathclose\bgroup\originalleft}|E_0{\aftergroup\egroup\originalright}|$ &$\Gamma_0$\
\[0.1cm\]\
$1$ & $269$ & $1.328\times 10^{-23}$ & & $280$ & $3.781\times 10^{-20}$ & & $266$ & $5.616\times 10^{-21}$ & & $256$ & $1.653\times 10^{-22}$ & & $260$ & $2.056\times 10^{-20}$ & & $260$ & $1.265\times 10^{-18}$\
$2$ & $183$ & $1.089\times 10^{-14}$ & & $186$ & $7.200\times 10^{-13}$ & & $178$ & $2.570\times 10^{-13}$ & & $173$ & $3.925\times 10^{-14}$ & & $173$ & $5.068\times 10^{-13}$ & & $170$ & $4.371\times 10^{-12}$\
$3$ & $139$ & $1.027\times 10^{-11}$ & & $139$ & $1.864\times 10^{-10}$ & & $133$ & $9.040\times 10^{-11}$ & & $131$ & $2.445\times 10^{-11}$ & & $129$ & $1.444\times 10^{-10}$ & & $125$ & $6.316\times 10^{-10}$\
$4$ & $111$ & $3.037\times 10^{-10}$ & & $109$ & $2.830\times 10^{-9}$ & & $105$ & $1.609\times 10^{-9}$ & & $104$ & $5.849\times 10^{-10}$ & & $101$ & $2.305\times 10^{-9}$ & & $97$ & $7.091\times 10^{-9}$\
$5$ & $92$ & $2.226\times 10^{-9}$ & & $89$ & $1.372\times 10^{-8}$ & & $86$ & $8.611\times 10^{-9}$ & & $85$ & $3.759\times 10^{-9}$ & & $83$ & $1.153\times 10^{-8}$ & & $78$ & $2.847\times 10^{-8}$\
\[tbl:excitonbinding\]
In \[fig:potential\], we compare the bilayer Keldysh potential to the full potential obtained by solving the multilayer Poisson equation. Panel (a) shows the potential for the representative WS$_2$/WSe$_2$ case. Evidently, a good agreement is found. Both the intra- and interlayer potentials (exact and approximate) behave as $-1/\kappa r$ for $r\gg d$, as the vertical separation becomes negligible in this region. For small $r$, on the other hand, the intralayer potential diverges logarithmically, while the interlayer potential behaves as $a+br^2$, where $a$ and $b$ are constants. The quadratic form is readily understood by expanding the Bessel function in \[eq:fullpotinarticle\] to second order, i.e. $$\begin{aligned}
J_0{\mathopen{}\mathclose\bgroup\originalleft}(qr{\aftergroup\egroup\originalright}) = 1-{\mathopen{}\mathclose\bgroup\originalleft}(\frac{qr}{2}{\aftergroup\egroup\originalright})^2 + O{\mathopen{}\mathclose\bgroup\originalleft}(r^4{\aftergroup\egroup\originalright})\thinspace.\end{aligned}$$ Integrating using the second order expansion shows that the integral diverges for $d=0$. For $d>0$, on the other hand, the exponential function leads to converging integrals even for $r=0$. The quadratic behaviour has inspired the use of a harmonic oscillator model, obtained by expanding a potential similar to \[eq:BK\] to second order, to analyse interlayer excitons [@Berman2017]. It should be noted, however, that the binding energies predicted by a second order expansion of \[eq:BK\] are in poor agreement with those obtained using the full potential. In contrast, the binding energies obtained with the BLK potential are in good agreement with the full potential results, as seen in panel (b) of \[fig:potential\]. In fact, they never deviate more than $6\%$ for the cases considered. The binding energies are, furthermore, in excellent agreement with those found in literature. As an example, the binding energy of $1$s interlayer excitons in WS$_2$/WSe$_2$ on a diamond substrate ($\varepsilon_a = 1$ and $\varepsilon_b = 5$) was measured recently to be $126\pm7$ meV [@Merkl2019], and our model yields $125$ meV. Furthermore, Ovesen *et al.* [@Ovesen2019] found the binding energy of interlayer excitons in MoSe$_2$/WSe$_2$ in free space to be $246$ meV using a model similar to our full potential, where we find $260$ meV with the BLK model. It should be mentioned that the model used in the present paper predicts slightly lower free-space binding energies than some of the *ab initio* methods [@Meckbach2018; @Gillen2018; @Torun2018; @Deilmann2018]. Interlayer exciton binding energies for the six TMD bilayer combinations with type-II band alignment [@Gong2013; @Komsa2013] are summarized in \[tbl:excitonbinding\] for various dielectric environments. The effective masses used are obtained from Ref. [@Kormanyos2015], and the TMD widths have been taken as half the vertical lattice constants found in Ref. [@He2014].
Field induced dissociation {#sec:III}
==========================
![Stark shift (left axis) and dissociation rate (right axis) of interlayer excitons in six different bilayer van der Waals heterostructures in various dielectric environments $\kappa$. []{data-label="fig:dissociation"}](Interlayer_exciton_dissociation.pdf){width="1\columnwidth"}
When the exciton is subjected to an electrostatic field, it may be dissociated. This is realized in the Wannier model by the energy eigenvalue obtaining a non-vanishing imaginary part [@Pedersen2016ExcitonIonization; @Pedersen2016Stark; @Haastrup2016; @Massicotte2018]. The field induced dissociation rate is then given by $\Gamma = -2 \,\mathrm{Im}\, E$, where the imaginary part can be obtained efficiently by utilizing the ECS technique [@Simon1979; @McCurdy1991; @Rescigno2000; @McCurdy2004]. As mentioned briefly in the previous section, this technique consists of rotating the radial coordinate into the complex plane outside a radius $R$ (see \[eq:ECStrans\]). The partitioning of the radial domain is efficiently dealt with by resolving the radial part of the eigenstate in a finite element basis consisting of Legendre polynomials, and the angular part in a cosine basis. We have previously used the same numerical procedure to calculate dissociation rates for excitons in various monolayer TMDs [@Kamban2019], and we refer the interested reader to that paper for the technical details of the method. The field induced dissociation rates and Stark shifts for interlayer excitons in the six van der Waals heterostructures are shown in \[fig:dissociation\]. As is evident, the structures support excitons that behave very similarly in electrostatic fields. The exciton Stark shifts can be seen to vary approximately as $\mathcal{E}^2$ for weak electric fields, in accordance with perturbation theory $E \approx E_0 - \frac{1}{2}\alpha\mathcal{E}^2$, where $E_0$ is the unperturbed energy and $\alpha$ the in-plane polarizability. Calculating the polarizability using the Dalgarno-Lewis equation [@Dalgarno1955] and a finite element basis [@Kamban2019] reveals that the interlayer exciton polarizabilities are significantly larger than their monolayer counterparts. For example, freely suspended MoS$_2$/WSe$_2$ supports interlayer excitons with $\alpha_{\mathrm{MoS}_2/\mathrm{WSe}_2}^{{\mathopen{}\mathclose\bgroup\originalleft}(\kappa=1{\aftergroup\egroup\originalright})} = 41\times 10^{-18} \mathrm{eV}{\mathopen{}\mathclose\bgroup\originalleft}(\mathrm{m/V}{\aftergroup\egroup\originalright})^2$, whereas monolayer MoS$_2$ and WSe$_2$ support exciton polarizabilities of around $\alpha_{\mathrm{MoS}_2}^{{\mathopen{}\mathclose\bgroup\originalleft}(\kappa=1{\aftergroup\egroup\originalright})} = 4.6\times 10^{-18} \,\mathrm{eV}{\mathopen{}\mathclose\bgroup\originalleft}(\mathrm{m/V}{\aftergroup\egroup\originalright})^2$ and $\alpha_{\mathrm{WSe}_2}^{{\mathopen{}\mathclose\bgroup\originalleft}(\kappa=1{\aftergroup\egroup\originalright})} = 6.3\times 10^{-18}\, \mathrm{eV}{\mathopen{}\mathclose\bgroup\originalleft}(\mathrm{m/V}{\aftergroup\egroup\originalright})^2$, respectively [@Pedersen2016ExcitonStark]. Encapsulating the materials in hBN ($\kappa = 4.9$ [@Latini2015]) increases the polarizabilities to $\alpha_{\mathrm{MoS}_2/\mathrm{WSe}_2}^{{\mathopen{}\mathclose\bgroup\originalleft}(\kappa=4.9{\aftergroup\egroup\originalright})} = 116\times 10^{-18}\, \mathrm{eV}{\mathopen{}\mathclose\bgroup\originalleft}(\mathrm{m/V}{\aftergroup\egroup\originalright})^2$ whereas $\alpha_{\mathrm{MoS}_2}^{{\mathopen{}\mathclose\bgroup\originalleft}(\kappa=4.9{\aftergroup\egroup\originalright})} = 14.2\times 10^{-18}\, \mathrm{eV}{\mathopen{}\mathclose\bgroup\originalleft}(\mathrm{m/V}{\aftergroup\egroup\originalright})^2$ and $\alpha_{\mathrm{WSe}_2}^{{\mathopen{}\mathclose\bgroup\originalleft}(\kappa=4.9{\aftergroup\egroup\originalright})} = 20.8\times 10^{-18}\, \mathrm{eV}{\mathopen{}\mathclose\bgroup\originalleft}(\mathrm{m/V}{\aftergroup\egroup\originalright})^2$ [@Pedersen2016ExcitonStark]. The reason that these large polarizabilities are observed for interlayer excitons in bilayer vdWHs is the increased screening and the vertical separation of the electron and hole. Both reduce the binding energy and they are therefore much easier to polarize.
![Stark shift (left axis) and dissociation rate (right axis) for interlayer excitons in a WS$_2$/WSe$_2$ bilayer that is either (i) freely suspended, (ii) placed on an hBN substrate, or (iii) encapsulated by hBN. []{data-label="fig:dissWSe2WS2"}](Dissociation_rate_WSe2-WS2.pdf){width="0.5\columnwidth"}
Turning to the dissociation rates, it is clear that the encapsulating media significantly alter how quickly interlayer excitons are dissociated. The same behaviour was observed for their monolayer counterparts [@Kamban2019], and one therefore has a large degree of freedom in device design. For extremely weak fields, the dissociation rates are very low, but they grow rapidly with an increasing field strength. As an example, the dissociation rate of interlayer excitons in freely suspended MoS$_2$/WS$_2$ is around $\Gamma\approx 1.7\times 10^{4}\, \mathrm{s}^{-1}$ already at $\mathcal{E} = 10\,\mathrm{V/\mu m}$, and only $\Gamma\approx 5.3\times 10^{-38}\, \mathrm{s}^{-1}$ and $\Gamma\approx 2.7\times 10^{-33}\, \mathrm{s}^{-1}$ for monolayer MoS$_2$ and WS$_2$, respectively [@Kamban2019]. It should be noted that dissociation rates of intralayer excitons are only important if they are comparable to the rate at which intralayer excitons tunnel over to interlayer excitons. In a recent experiment on MoS$_2$/WS$_2$ structures, the holes of photoexcited excitons in the MoS$_2$ layer of this structure were observed to tunnel into the WS$_2$ layer within $50\, \mathrm{fs}$ [@Hong2014]. Comparing to the dissociation rates of intralayer excitons in the top and bottom layer of MoS$_2$/WS$_2$ for $\mathcal{E}=10\,\mathrm{V/\mu m}$, we find $\Gamma\approx 2.1\times 10^{-3}\, \mathrm{s}^{-1}$ and $\Gamma\approx 2.9\times 10^{4}\, \mathrm{s}^{-1}$, respectively. The large difference between these two rates can be traced back to the reduced masses, which are $\mu = 0.2513$ and $\mu = 0.1543$, respectively. The time it takes to dissociate these excitons with the given field may be approximated as $\tau = 1/\Gamma \approx 476\, \mathrm{s}$ and $\tau \approx 3.5\times 10^{-5}\, \mathrm{s}$, respectively. The intralayer excitons have therefore clearly transitioned to interlayer excitons before they are dissociated by the field. Interlayer tunneling rates are likely affected by material parameters as well as the surrounding media. Assuming, however, similar time scales as observed for MoS$_2$/WS$_2$ $\Gamma_{\mathrm{Tunnel}}\approx 10^{13}\, \mathrm{s}^{-1}$, interlayer dissociation rates are the limiting factor in field induced exciton dissociation for weak to moderate fields. For the largest fields in \[fig:dissociation\], the competition between tunneling and dissociation will be important for an accurate description. Due to the risk of dielectric break-down, such large fields are best avoided in devices, however. The high interlayer dissociation rates suggest that using carefully chosen bilayer TMDs in photocurrent devices is much more attractive than their monolayer counterparts. Moreover, proper encapsulation will further improve device performance. As hBN is a very common material used to encapsulate samples, we show the interlayer dissociation rates for WS$_2$/WSe$_2$ that is either (i) freely suspended, (ii) placed on an hBN substrate, or (iii) encapsulated by hBN in \[fig:dissWSe2WS2\]. Evidently, hBN surroundings increase the dissociation rates by several orders of magnitude, and for weak fields in particular.
![Dissociation rates (a) and binding energies (b) for interlayer excitons in six vdWHs as functions of environment screening $\kappa$. In panel (a), the dots and solid lines represent the numerically exact and analytical approximation \[eq:wfexp\], respectively. The inset shows the approximate and exact dissociation rates for interlayer excitons in WS$_2$/WSe$_2$ as a function of field strength. []{data-label="fig:dissvskappa"}](Dissociation_vs_kappa_SR.pdf){width="1\columnwidth"}
Recently an analytical weak-field approximation was obtained for exciton dissociation rates in monolayer TMDs [@Kamban2019]. The derivation was made using weak-field asymptotic theory [@Tolstikhin2011], and the fact that the MLK potential has a simple asymptotic form. As mentioned in the previous section, the BLK potential has exactly the same asymptotic behaviour. The weak-field approximation for interlayer excitons therefore has exactly the same form, albeit with a different field-independent front factor. We arrive at $$\begin{aligned}
\Gamma \approx \Gamma_0\mathcal{E}^{1/2-2\sqrt{\mu}/{\mathopen{}\mathclose\bgroup\originalleft}(\kappa k{\aftergroup\egroup\originalright})}\exp{\mathopen{}\mathclose\bgroup\originalleft}(-\frac{2\sqrt{\mu}k^3}{3\mathcal{E}}{\aftergroup\egroup\originalright})\thinspace,\label{eq:wfexp}\end{aligned}$$ where $k = \sqrt{2{\mathopen{}\mathclose\bgroup\originalleft}|E_0{\aftergroup\egroup\originalright}|}$ and $\Gamma_0 = \Gamma_0{\mathopen{}\mathclose\bgroup\originalleft}(E_0,\mu,r_0^{{\mathopen{}\mathclose\bgroup\originalleft}(1{\aftergroup\egroup\originalright})}+r_0^{{\mathopen{}\mathclose\bgroup\originalleft}(2{\aftergroup\egroup\originalright})},d,\kappa{\aftergroup\egroup\originalright})$ is a field-independent material constant. The parameters needed to use \[eq:wfexp\] are presented in \[tbl:excitonbinding\], where $\Gamma_0$ has been computed by the integral procedure in [@Kamban2019]. Panel (a) of \[fig:dissvskappa\] shows the interlayer dissociation rates as functions of environment screening $\kappa$ for a field of $5\,\mathrm{V/\mu m}$ computed with the numerical procedure (dots) and the weak-field approximation (solid lines). Evidently a good agreement is found for such a low field. Nevertheless, as the inset shows, the weak-field approximation quickly begins to overestimate the dissociation rate for larger fields. This was also found for monolayer TMDs [@Kamban2019]. It is clear that WS$_2$/WSe$_2$ supports excitons with the largest dissociation rates at $\mathcal{E}=5\,\mathrm{V/\mu m}$, making it an interesting choice in device design. Note that the fully numerical procedure breaks down if the dissociation rate becomes extremely small, and, hence, we are unable to use it to obtain dissociation rates for MoS$_2$/MoSe$_2$ and MoS$_2$/WS$_2$ in surroundings with (very) low screening for this field strength. It is therefore advantageous to have a formula such as \[eq:wfexp\] when the fields become sufficiently weak.
In panel (b) of \[fig:dissvskappa\], we show the binding energies for the same excitons as functions of $\kappa$. They clearly follow similar trends as the dissociation rates do, suggesting that binding energy has a significant impact on dissociation rates. This is to be expected, as strongly bound excitons are harder to pull apart. It should, however, be noted that the binding energy does not uniquely determine the dissociation rate. As an example, the interlayer excitons in MoS$_2$/MoSe$_2$ and MoS$_2$/WS$_2$ have identical binding energies for surrounding media with $\kappa\approx 2.8$. Nevertheless, the dissociation rates are clearly different. In fact, several crossings are observed in the binding energies as a function of $\kappa$ whereas only one is found in the dissociation rates at $5\,\mathrm{V/\mu m}$. The origin is the different reduced exciton masses. For no structure does $\Gamma_0$ nor $\sqrt{\mu}/k$ contain any crossing for $\kappa\in{\mathopen{}\mathclose\bgroup\originalleft}[1;5{\aftergroup\egroup\originalright}]$. However, $\sqrt{\mu}k^3$ crosses for MoS$_2$/WSe$_2$ and MoSe$_2$/WSe$_2$. This complicated behaviour suggests that, to obtain accurate estimates of interlayer exciton dissociation rates, one must turn to either the weak-field approximation or make a full numerical calculation.
Conclusion {#sec:IV}
==========
In the present paper, we have studied binding energies, Stark shifts, and dissociation rates of interlayer excitons in van der Waals heterostructures (vdWHs). The structures analysed are the six bilayer vdWHs with type-II band alignment arising from combinations of MoS$_2$, MoSe$_2$, WS$_2$, and WSe$_2$. The bilayer excitons are described using an analytical bilayer Keldysh potential, which we have verified for accuracy by comparing to the full solution of the multilayer Poisson equation. Exciton binding energies, Stark shifts, and dissociation rates can therefore readily be calculated using the analytical potential. We find interlayer exciton binding energies ranging from $256$ to $280\,\mathrm{meV}$ for freely suspended structures and from $78$ to $92\,\mathrm{meV}$ for heavily screened structures, making them stable at room temperature. Furthermore, both the polarizabilities and dissociation rates found for these excitons are much larger than their monolayer counterparts. For example, interlayer excitons in freely suspended MoS$_2$/WS$_2$ are found to dissociate at a rate of $\Gamma\approx 1.7\times 10^{4}\, \mathrm{s}^{-1}$ in a field strength of $10\,\mathrm{V/\mu m}$ whereas monolayer MoS$_2$ and WS$_2$ have $\Gamma\approx 5.3\times 10^{-38}\, \mathrm{s}^{-1}$ and $\Gamma\approx 2.7\times 10^{-33}\, \mathrm{s}^{-1}$, respectively.
For moderate field strengths, intralayer exciton dissociation rates are significantly lower than the rate at which such excitons tunnel into interlayer excitons. For this reason, interlayer exciton dissociation rates are the limiting factor in generation of photocurrents at weak to moderate fields. Since optically excited excitons in one of the layers tunnel to long-lived interlayer excitons on ultrafast timescales, bilayer vdWHs with favourable band offsets may potentially serve as building blocks in efficient photocurrent devices. Finally, the numerically exact dissociation rates are compared to an analytical weak-field dissociation formula obtained from weak-field asymptotic theory. A good agreement is found in the weak-field limit, and \[eq:wfexp\] therefore serves as a useful formula to quickly estimate field induced dissociation rates of excitons in bilayer vdWHs for weak electric fields.
Multilayer Poisson equation {#app:poisson}
===========================
We want to find the interaction between the two particles in the system represented by \[fig:sketch\]. For charges at $z$ and $z'$ with in-plane separation $\boldsymbol{r}$, the interaction can be Fourier decomposed as $$\begin{aligned}
-V{\mathopen{}\mathclose\bgroup\originalleft}(r,z,z'{\aftergroup\egroup\originalright}) = \frac{1}{4\pi^2}\int \varphi{\mathopen{}\mathclose\bgroup\originalleft}(z,z';q{\aftergroup\egroup\originalright})e^{i\boldsymbol{q}\cdot \boldsymbol{r}}d^2q\thinspace,\end{aligned}$$ where the Fourier components satisfy the Poisson equation $$\begin{aligned}
4\pi\delta{\mathopen{}\mathclose\bgroup\originalleft}(z-z'{\aftergroup\egroup\originalright}) = {\mathopen{}\mathclose\bgroup\originalleft}[q^2\varepsilon{\mathopen{}\mathclose\bgroup\originalleft}(z;q{\aftergroup\egroup\originalright})-\frac{\partial}{\partial z}\varepsilon{\mathopen{}\mathclose\bgroup\originalleft}(z;q{\aftergroup\egroup\originalright})\frac{\partial}{\partial z}{\aftergroup\egroup\originalright}]\varphi{\mathopen{}\mathclose\bgroup\originalleft}(z,z';q{\aftergroup\egroup\originalright})\thinspace.\end{aligned}$$ We take the dielectric function $\varepsilon$ to be piecewise constant $$\begin{aligned}
\varepsilon{\mathopen{}\mathclose\bgroup\originalleft}(z;q{\aftergroup\egroup\originalright}) = \begin{cases}
\varepsilon_a\thinspace, \quad z>d_1\\
\varepsilon_1\thinspace, \quad 0<z<d_1\\
\varepsilon_2\thinspace, \quad -d_2<z<0\\
\varepsilon_b\thinspace, \quad z<-d_2\thinspace,
\end{cases}\end{aligned}$$ where $d_1$ and $d_2$ are the widths of the first and second layer, respectively. For $z'$ confined to the second layer, the solution can be sought on the form $$\begin{aligned}
\varphi{\mathopen{}\mathclose\bgroup\originalleft}(z,z';q{\aftergroup\egroup\originalright}) = \frac{2\pi}{q}\begin{cases}
A_1e^{-qz}\\
A_2e^{-qz} + B_2e^{qz}\\
A_3e^{-qz} + B_3e^{qz} +\varepsilon_2^{-1}e^{-q{\mathopen{}\mathclose\bgroup\originalleft}|z-z'{\aftergroup\egroup\originalright}|}\\
B_4e^{qz}\thinspace,
\end{cases}\end{aligned}$$ in the respective regions. The Fourier components can then be found analytically by solving the system of equations that arises from the boundary conditions. To describe charges confined to distinct layers, we fix the electron and hole to $z=d_1/2$ and $z'=-d_2/2$, respectively. This leads to the interlayer exciton potential $$\begin{aligned}
\varphi{\mathopen{}\mathclose\bgroup\originalleft}(d_1/2,-d_2/2;q{\aftergroup\egroup\originalright}) = \frac{\varphi_0{\mathopen{}\mathclose\bgroup\originalleft}(d_1/2,-d_2/2;q{\aftergroup\egroup\originalright})}{\varepsilon_{\mathrm{eff}}{\mathopen{}\mathclose\bgroup\originalleft}(q{\aftergroup\egroup\originalright})}\thinspace,\end{aligned}$$ where $$\begin{aligned}
\varphi_0{\mathopen{}\mathclose\bgroup\originalleft}(z,z';q{\aftergroup\egroup\originalright}) = \frac{2\pi}{q}e^{-{\mathopen{}\mathclose\bgroup\originalleft}|z-z'{\aftergroup\egroup\originalright}|q}\end{aligned}$$ is the bare interaction. The effective dielectric function is given by $$\begin{aligned}
\varepsilon_{\mathrm{eff}}{\mathopen{}\mathclose\bgroup\originalleft}(q{\aftergroup\egroup\originalright}) = \frac{A{\mathopen{}\mathclose\bgroup\originalleft}(q{\aftergroup\egroup\originalright})}{B{\mathopen{}\mathclose\bgroup\originalleft}(q{\aftergroup\egroup\originalright})}\label{eq:dielectric}.\end{aligned}$$ with $$\begin{aligned}
A{\mathopen{}\mathclose\bgroup\originalleft}(q{\aftergroup\egroup\originalright}) = {\mathopen{}\mathclose\bgroup\originalleft}(1-e^{-2d_1q}{\aftergroup\egroup\originalright}){\mathopen{}\mathclose\bgroup\originalleft}(\varepsilon_1^2\gamma^+ + \varepsilon_a\varepsilon_2\gamma^-{\aftergroup\egroup\originalright}) + {\mathopen{}\mathclose\bgroup\originalleft}(1+e^{-2d_1q}{\aftergroup\egroup\originalright}){\mathopen{}\mathclose\bgroup\originalleft}(\varepsilon_a\varepsilon_1\gamma^+ + \varepsilon_1\varepsilon_2\gamma^-{\aftergroup\egroup\originalright})\thinspace,\end{aligned}$$ $$\begin{aligned}
B{\mathopen{}\mathclose\bgroup\originalleft}(q{\aftergroup\egroup\originalright}) = 2{\mathopen{}\mathclose\bgroup\originalleft}[{\mathopen{}\mathclose\bgroup\originalleft}(1-e^{-d_1q}{\aftergroup\egroup\originalright})\varepsilon_a + {\mathopen{}\mathclose\bgroup\originalleft}(1+e^{-d_1q}{\aftergroup\egroup\originalright})\varepsilon_1{\aftergroup\egroup\originalright}] \times {\mathopen{}\mathclose\bgroup\originalleft}[{\mathopen{}\mathclose\bgroup\originalleft}(1+e^{-d_2q}{\aftergroup\egroup\originalright})\varepsilon_2+{\mathopen{}\mathclose\bgroup\originalleft}(1-e^{-d_2q}{\aftergroup\egroup\originalright})\varepsilon_b{\aftergroup\egroup\originalright}]\thinspace,\end{aligned}$$ where $$\begin{aligned}
\gamma^\pm = {\mathopen{}\mathclose\bgroup\originalleft}(1\pm e^{-2d_2q}{\aftergroup\egroup\originalright})\varepsilon_2 + {\mathopen{}\mathclose\bgroup\originalleft}(1\mp e^{-2d_2q}{\aftergroup\egroup\originalright})\varepsilon_b\thinspace.\end{aligned}$$ This effective dielectric function tends to ${\mathopen{}\mathclose\bgroup\originalleft}(\varepsilon_a+\varepsilon_b{\aftergroup\egroup\originalright})/2$ for $q\to 0$ and to ${\mathopen{}\mathclose\bgroup\originalleft}(\varepsilon_1+\varepsilon_2{\aftergroup\egroup\originalright})/2$ for $q\to \infty$, as expected. The dielectric function describing charges confined to the same layer may be obtained in a similar manner by placing both charges at $z=-d_2/2$. The interaction in real space can then be obtained as the inverse Fourier transform $$\begin{aligned}
-V{\mathopen{}\mathclose\bgroup\originalleft}(r{\aftergroup\egroup\originalright})=\int_0^\infty \frac{e^{-{\mathopen{}\mathclose\bgroup\originalleft}|z-z'{\aftergroup\egroup\originalright}|q}J_0{\mathopen{}\mathclose\bgroup\originalleft}(qr{\aftergroup\egroup\originalright})}{\varepsilon_{\mathrm{eff}}{\mathopen{}\mathclose\bgroup\originalleft}(q{\aftergroup\egroup\originalright})}dq\thinspace.\label{eq:fullpot}\end{aligned}$$ For the dielectric constants, we have used the static in-plane dielectric constants calculated from first principles in Ref. [@Laturia2018].
Acknowledgements {#acknowledgements .unnumbered}
================
The authors gratefully acknowledge financial support from the Center for Nanostructured Graphene (CNG), which is sponsored by the Danish National Research Foundation, Project No. DNRF103. Additionally, T.G.P. is supported by the QUSCOPE Center, sponsored by the Villum Foundation.
Author contributions statement {#author-contributions-statement .unnumbered}
==============================
H.C.K. initiated the work, performed all calculations, and prepared the initial manuscript. T.G.P. assisted in solving the Poisson equation. H.C.K. and T.G.P. analysed the results and prepared the final manuscript.
Additional information {#additional-information .unnumbered}
======================
**Competing interests** The authors declare no competing financial interest.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In the context of black-hole accretion disks, we estimate the effects of the plasma environment on the atomic parameters associated with K-vacancy states in He- and Li-like oxygen ions. Multiconfiguration Dirac–Fock and Breit–Pauli computations have been carried out assuming a time-averaged Debye–Hückel potential for both the electron–nucleus and electron–electron interactions. Plasma effects on the ionization ad excitation thresholds, transition wavelengths, and radiative emission rates are reported for and .'
author:
- |
J. Deprince,$^1$ P. Palmeri,$^1$ P. Quinet,$^{1,2}$ S. Fritzsche,$^{3,4}$ M. Bautista,$^5$\
C. Mendoza,$^5$ T. R. Kallman,$^6$ and J. García$^7$\
title: 'Plasma-Environment Effects on the Atomic Structure and K Lines of He- and Li-like Oxygen Ions'
---
Introduction {#intro}
============
Accurate atomic data to characterize the absorption and emission of high-energy photons in dense plasmas are relevant in a wide range of astrophysical applications. An interesting case is the modeling of the X-rays produced and reprocessed in the region near a compact object such as a black hole or neutron star. In these systems the gas falling into the strong gravitational potential of the compact object spirals in forming an accretion disk from which copious X-ray radiation is produced [see, for instance, @Krolik1999; @Done2007]. The observed spectrum shows atomic features, both in emission and absorption, from different cosmic abundant species such as iron and oxygen ions, the modeling of which provides strong insights into the composition, temperature, and degree of ionization of the plasma . Furthermore, for black holes, their angular momentum can be inferred from the modeling of the distortion of the Fe K emission complex by relativistic effects [e.g., @Reynolds2013; @Garcia2014].
In the present report, we focus on oxygen ions for which the atomic data associated to K-vacancy states have already been computed [@Garcia2005] and incorporated in the collisional–radiative XSTAR code developed by and . However, none of these atomic parameters takes into account plasma embedding effects considered essential by in the interpretation of the X-ray spectra observed by the current satellite-borne telescopes ([*Chandra, XMM–Newton, Suzaku*]{}). In accretion disks around black holes, it turns out that spectral models require uncommonly high iron abundances: several times the solar value [@Garcia2016]. Recent magnetohydrodynamics (MHD) simulations by @Sch2013 of accreting black holes (of 10 $M_{\odot}$ with an accretion rate of 10%) predict disk plasma conditions characterized by temperatures of $10^5{-}10^7$ K and electron densities of $10^{18}{-}10^{21}$ cm$^{-3}$. Such a high-density plasma environment can have significant effects on the atomic processes giving rise to line emission and on the ionization balance of the species present in the plasma. We present preliminary results concerning the plasma embedding effects on the structure and radiative parameters of the K-vacancy states of and .
Theoretical Methods
===================
Atomic Structure Calculations {#methods}
-----------------------------
Two independent theoretical methods have been used to model the atomic structures and compute the radiative data for the He- and Li-like O ions, namely the fully relativistic multiconfiguration Dirac–Fock (MCDF) method, using a combination of the GRASP92 [@Paripa1996] and the RATIP [@Fritzsche2012] implementations, and AUTOSTRUCTURE based on the multiconfiguration Breit–Pauli relativistic framework [AUTOS hereafter, @Badnell1997; @Badnell2011].
In the MCDF approach, all the atomic state functions (ASFs) $\Psi(\gamma, P, J, M_J)$ are expanded in linear combinations of configuration state functions (CSFs) $\Phi(\alpha_i, P, J, M_J)$ of the same parity and total angular momentum $$\Psi(\gamma, P, J, M_J) = \sum_i c_i\,\Phi(\alpha_i, P, J, M_J)\,.$$ The CSFs are in turn antisymmetrized products of a common set of orthonormal monoelectronic spin-orbitals of the form $$\varphi_{n\kappa m}(r,\theta,\phi) = \frac{1}{r}\begin{pmatrix}
P_{n\kappa}(r)\,\chi_{\kappa m}(\theta, \phi)\\
iQ_{n\kappa}(r)\,\chi_{-\kappa m}(\theta, \phi)
\end{pmatrix}\,,$$ where $P_{n\kappa}(r)$ and $Q_{n\kappa}(r)$ are, respectively, the large and small components of the radial wave function, and the angular functions $\chi_{\kappa m}(\theta, \phi)$ are the spinor spherical harmonics. The $\alpha_i$ represent all the one-electron and intermediate quantum numbers needed to completely define the CSF, while $\gamma$ is usually chosen as the $\alpha_i$ corresponding to the CSF with the largest weight $|c_i|^2$. The $\kappa$ quantum number is given by $\kappa = \pm\,(j+1/2)$ where $j$ is the electron total angular momentum quantum number.
In AUTOS, the ASF $\Psi(\gamma,L,S,P,J,M_J)$ is expressed in intermediate coupling and is the eigenvector of the Breit–Pauli Hamiltonian, where the Breit interaction is treated as a perturbation. The electron orbital takes the simpler form $$\varphi_{nlm}(r,\theta,\phi) = \frac{1}{r}P_{n\ell}(r)\,\chi_{\ell m}(\theta, \phi)$$ where $\ell$ is the electron orbital angular momentum and $\chi_{\ell m}$ the spherical harmonic. The radial $P_{n\ell}(r)$ orbitals are obtained in a Thomas–Fermi–Dirac–Amaldi model potential $$V(r)=\frac{Z_{\rm eff}(\lambda_{n\ell},r)}{r}$$ where $\lambda_{n\ell}$ are orbital scaling parameters.
For both and , the ASFs were built by considering all the single and double excitations from the the spectroscopic configurations with $n=2$ and $n=3$ orbitals. Two different optimizations were implemented depending on the desired atomic parameters. For ionization potentials, the computations were carried out by optimizing only on the ground level of each ion, whereas for all the other atomic parameters of interest the calculations were carried out by optimizing a weighted trace of the Hamiltonian using level weights proportional to $2J+1$ or 1.0.
Plasma Screening Effects
------------------------
In these two theoretical methods the effects of the plasma screening on the atomic properties were modeled with a time-averaged Debye-Hückel (DH) potential for both the electron–nucleus and electron–electron interactions. This potential in atomic units (a.u.) takes the form $$V^{\text{DH}}(r,\lambda) = -\sum_{i=1}^N \frac{Ze^{-\lambda r_i}}{r_i} + \sum_{j<i}^N \frac{e^{-\lambda r_{ij}}}{r_{ij}}\ ,$$ where $N$ is the number of bound electrons, $r_i$ is the distance of the $i$th electron from the nucleus, and $r_{ij}$ is the distance between the $i$ and $j$ electrons. The plasma screening parameter $\lambda$ is the inverse of the Debye shielding length ($\lambda_{\text{De}}$), which can be expressed in atomic units in terms of the plasma electron density $n_e$ and temperature $T_e$ as $$\lambda = \frac{1}{\lambda_{\text{De}}} = \left(\frac{4\pi n_e}{k T_e}\right)^{1/2}\ .$$
For a black-hole accretion disk the MHD simulations of @Sch2013 suggest screening parameter values of $0<\lambda \lesssim 0.1$ a.u. ($\lambda = 0$ corresponding to the isolated case). A typical accretion disk can be characterized by a weakly coupled plasma since its corresponding coupling parameter $$\Gamma = \frac{e^2}{4\pi \varepsilon_0 d k T_e}$$ for particles separated by a mean distance $d= \left(3/(4\pi n_e)\right)^{1/3}$ [@Piel2010], which represents the ratio of the Coulomb potential interaction energy between two particles to the thermal energy, is markedly lower than unity ($0.0003 < \Gamma < 0.3$). For weakly coupled plasmas ($\Gamma \ll 1$), the screening effects can be approximated with a Debye–Hückel model potential.
$^\text{a}$Semiempirical calculation [@Edlen1979]
$^\text{b}$[*Ab initio*]{} calculation [@Drake1988]
$^\text{a}$Electron beam ion trap measurements [@Schmidt2004]
$^\text{b}$Spectroscopic tables [@Moore1998]
$^\text{c}$Spectroscopic measurements [@Engstrom1995]
Results and Discussion
======================
The ionization potentials ($E_0$), K-threshold energies ($E_K$), radiative wavelengths, and $A$-values for and obtained with MCDF/RATIP and AUTOS adopting plasma screening parameters of $\lambda = 0$ and $\lambda = 0.1$ a.u. are listed in Tables \[IP-plasma\]–\[A-plasma\].
For the isolated ion case ($\lambda = 0$), the $E_0$ obtained with MCDF/RATIP and AUTOS are both in very good agreement (better than 0.2%) with the NIST standards. A similar agreement is found between the MCDF/RATIP and AUTOS K thresholds $E_K$ and wavelengths, but for the $E_K$ of the matching deteriorates somewhat to just under 1%. $A$-values are within 10% and in fair accord with @Garcia2005, who used the pseudo-relativistic Hartree–Fock method [@Cowan1981], except for the ${\rm 1s2s2p\ ^2P_J - 1s^22s\ ^2S_{1/2}}$ transitions in . This may be due to strong mixing, which makes $A$-values sensitive to configuration interaction, or orbital definition (e.g., relaxation effects).
The threshold plasma effects can be appreciated in Tables \[IP-plasma\]–\[K-threshold-plasma\], where the ionization potential $E_0$ and K threshold $E_K$ are lowered in and by about 16 eV and 19 eV, respectively, with a screening parameter of $\lambda = 0.1$ a.u. Both methods coincide precisely on this finding. Moreover, the lowering of the IPs and K thresholds determined with MCDF/RATIP is found to vary linearly with the plasma screening parameter $\lambda$, as shown in Fig. \[fig:figure1\] for the former.
![Plasma effects on the ionization potential (IP) of and computed by MCDF/RATIP.[]{data-label="fig:figure1"}](shift-IP-O6-O7.eps){width="80.00000%"}
On the other hand, wavelengths and $A$-values are only weakly affected by plasma screening effects as can be seen in Tables \[wavelength-plasma\]–\[A-plasma\]: the K-line wavelengths are redshifted by less than 20 mÅ while the $A$-values are reduced by less than 0.5%. However, in contrast to the linear threshold variation with screening parameter, the respective wavelength and $A$-value variations, although small, are not linear as shown for the latter in Fig. \[fig:figure3\].
![Plasma effects on a K-line radiative transition probability of computed by MCDF/RATIP.[]{data-label="fig:figure3"}](A-O6.eps){width="80.00000%"}
Final Remarks
=============
A similar study of the plasma environment effects on the atomic structure and radiative parameters of K-vacancy states in Li-like and He-like have been carried out by @Deprince2017. It was found therein that, in similar plasma conditions ($\lambda = 0.1$ a.u.), thresholds were lowered by ${\sim}65{-}68$ eV, which is a much larger effect than the one we are reporting here for Li- and He-like oxygen ions. They also found that the K lines in these two Fe ions were also redshifted but by even smaller amounts (${\sim}0.1$ mÅ) than those for and . Furthermore, the Fe $A$-value variations were of comparable magnitude but some were positive while others negative.
JD is a Research Fellow of the Belgian Fund for Research Training in Industry and Agriculture (FRIA). PP and PQ are, respectively, Research Associate and Research Director of the Belgian Fund for Scientific Research (FRS-FNRS). Financial support from these organizations, as well as from the NASA Astrophysics Research and Analysis Program, grant 80NSSC17K0345, are gratefully acknowledged.
natexlab\#1[\#1]{}url \#1[`#1`]{}urlprefix\[2\]\[\][[\#2](#2)]{}
Badnell, N. R.1997, J. Phys. B – At. Mol. Opt., 30, 1
Badnell, N. R. 2011, Comput. Phys. Commun., 182, 1528
Bautista, M. A., & Kallman, T. R. 2001, ApJS, 134, 139
Cowan, R. D. 1981, The Theory of Atomic Structures and Spectra (Berkeley, CA: University of California Press)
Deprince, J., Fritzsche, S., Kallman, T. R., Palmeri, P., & Quinet, P. 2017, AIP Conf. Proc., 1811, 040002
Done, C., Gierlinski, M., & Kubota A., A&A Rev., 15, 1
Drake, G. W. F. 1988, Can. J. Phys., 66, 586
Edlén, B. 1979, Phys. Scripta, 19, 255
Engström, L., & Litzén U. 1995, J. Phys. B. – At. Mol. Opt., 28, 2565
Fritzsche, S. 2012, Comput. Phys. Commun., 183, 1523
García, J., Mendoza, C., Bautista, M. A., et al. 2005, ApJS, 158, 68
García, J., & Kallman, T. R. 2010, ApJ, 718, 695
García, J., Dauser, T., Lohfink, A., et al. 2014, ApJ, 782, 76
García, J., Fabian, A. C., Kallman, T. R., et al. 2016, MNRAS, 462, 751
Kallman, T. R., & Bautista, M. A. 2001, ApJS, 133, 221
Krolik, J. H. 1999, Nature, 398, 678
Moore, C. E. 1998, Tables of Spectra of Hydrogen, Carbon, Nitrogen, and Oxygen Atoms and Ions, edited by J. W. Gallagher (Boca Raton, FL: CRC Press)
Parpia, F. A., Froese Fischer, C., & Grant, I. P. 1996, Comput. Phys. Commun., 94, 249 Piel, A. 2010, Plasma Physics: An Introduction to Laboratory, Space, and Fusion Plasmas (Berlin-Heidelberg, Germany: Springer-Verlag)
Reynolds, C. S. 2013, Class. Quantum Grav., 30, 244004
Ross, R. R., & Fabian, A. C. 2005, MNRAS, 358, 211
Schmidt, M., Beiersdorfer, P., Chen, H. 2004, ApJ, 604, 562
Schnittman, J. D., Krolik, J. H., & Noble S. C. 2013, ApJ, 769, 156
Smith, R. K., & Brickhouse, N. S. 2014, Adv. At. Mol. Opt. Phy., 63, 271
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Making use of a simplified model for protein folding, it can be shown that conformations which are particularly stable when their energy is minimized with respect to amino acid sequence (in the sense that they display a large energy gap to the lowest structrally dissimilar conformation), aside from leading to fast folding, are highly designable (in the sense that many sequences target onto it in the folding process). These results are quite general, do not depend on the particular simmetries displayed by the compact conformation chosen as native, and can be obtained making use of a large class of contact energies. On the other hand, the design of sequences which fold onto native conformations displaying secondary structures and eventually tertiary simmetries, is a difficult task requiring a delicate tuning of the contact energies.'
address:
- '$^1$Dipartimento di Fisica, Università di Milano, Via Celoria 16, I-20133 Milano, Italy.'
- '$^2$INFN, Sezione di Milano, Via Celoria 16, I-20133 Milano, Italy.'
- '$^3$The Niels Bohr Institute, University of Copenhagen, 2100 Copenhagen, Denmark.'
author:
- 'R. A. Broglia$^{1,2,3}$, G. Tiana$^{3}$ and H. E. Roman$^{1,2}$'
title: 'Protein Foldability and Designability: General Physics and Pretty Chemistry'
---
Pacs numbers: 87.15.Da, 61.43.-j, 64.60.Cn, 64.60.Kw
A basic property of “wild type” proteins is that they fold on short call. A second one is that they can accomodate numerous mutations that are neutral with respect to structure changes. Another is that they display, in the native conformation, secondary structures and tertiary symmetries, features that define their biological specificity [@creighton].
Single domain notional proteins must display a large energy gap $\delta$ between the native state and the bulk of misfolded conformations that are structurally dissimilar to the native state (calculated making use of a Monte Carlo formalism), in keeping with the fact that fast folding is tantamount to a value of the “order parameter” $\xi=\delta/{\bar U}\gg1$, where ${\bar U}$ is the average value of the contact energies among the amino acids. In cases where ${\bar U}=0$, the order parameter to be used is $\xi=\delta/\sigma$, where $\sigma$ is the dispersion of the interaction energies [@goldstein92; @ssk94; @sg92; @s94; @goldstein95; @gas96; @bryngelson; @s97]. The designed sequence which in the native conformation fulfills $\xi\gg1$, allows for billions of mutations leading to sequences which have the native conformation as their non-degenerate ground state with energies lying inside the gap and, more importantly, which fold to the native state in a time of the order of that associated with the folding of the designed sequence[@desig_1].
To design a sequence which folds fast and which is highly resilient to point mutations is easily achived making use of a large class of contact energies among the amino acids, and does not require the native conformation to display “wild–type” secondary or tertiary structures [@s94; @ueda; @go; @dill; @sklonik; @covell; @godzik; @socci; @klimov; @Shakh96; @grosberg].
It is not difficult either to derive contact energies which select, as the most designable conformations, those displaying “wild-type” secondary structures [@wingreen]. What is very difficult, requiring a detailed tuning of the contact energies, is to achive that the associated sequences fold, let alone whether they do it fast or slow.
To illustrate the above points we report, in what follows, results obtained making use of a lattice model [@goldstein92; @ssk94; @sg92; @s94; @goldstein95; @gas96; @bryngelson; @s97], where the differences between the amino acids are manifested in pairwise interaction energy of variable magnitude and sign, depending on the identity of the interacting amino acid. The configurational energy is $$E = {1\over2}\sum_{i,j}^N U_{m(i),m'(j)}~
\Delta(\vert {\vec r}_i-{\vec r}_j\vert),
\label{hami}$$ $\{ {\vec r}\}$ being the set of coordinates of all of the monomers describing a chain conformation and $U_{m,m'}$ is the effective interaction energy between monomers $m$ and $m'$. The quantity $\Delta(\vert {\vec r}_i-{\vec r}_j\vert)$ is a contact function. It is equal to one if sites $i$ and $j$ are at unit distance (lattice neighbours) not connected by a covalent bond, and zero otherwise. In addition, it is assumed that on-site repulsive forces prevent two amino acids from occupying the same site simultaneously, so that $\Delta(0)=\infty$.
We start considering the case of proteins displaying a large value of the parameter $\xi$. A particular realization of this situation is provided by a 20-letter amino acid chain, composed of 36 monomers and designed to fold into the native conformation shown in Fig. 1a. The quantities $U_{m,m'}$ used here correspond to the contact energies obtained from a statistical analysis of real proteins data and were taken from Table 6 of ref.[@mj] (“disordered” M-J contact energies (MJ)$_d$). In this case, the average value of the contact energies is ${\bar U}=0$, the corresponding standard deviation $\sigma\cong 0.3$. The 36-mer chain denoted as S$_{36}$ and designed by minimizing, for fixed amino acid composition, the energy of the native conformation with respect to the amino acids sequence is shown in Fig.1b . In the units we use ($R T_{\rm room}=0.6$ kcal/mol), the energy of S$_{36}$ in its native conformation is $E_{\rm nat}=-16.5$ and the value of the gap $\delta=2.5$, yielding a value for the order parameter $\xi=8.3$. Making use of Monte Carlo (MC) simulations it is found that at the temperature $T=0.28$, optimal from the point of view of allowing for the accumulation of statistically representative samples of the different simulations, and at the same time leading to a consistent population of the native conformation ($\approx 10\%$), the sequence S$_{36}$ folds into the native conformation in $8\cdot 10^5$ MC steps.
The characterization of the role played by the different amino acids in the folding process of S$_{36}$ have been carried out in ref.[@my] (cf. also [@aggreg]), by using mutations. It was found that the 36 sites of the native conformation (see Fig.1a) can be classified as “hot” (red bead, numbered 6, 27, and 30), “warm” (yellow beads numbered 3, 5, 11, 14, 16 and 28), and “cold” (the rest of the beads, coloured green) sites. On average, mutations on the 27 “cold” sites yield sequences that still fold to the native structure (neutral mutations), although the folding time is somewhat longer than for S$_{36}$. Sequences obtained from mutations on the six warm sites fold, as a rule, to a unique conformation, sometimes different but in any case very similar to the native one. Mutations on the three hot sites lead, in general, to complete misfolding (denaturation) of the protein. These results were found not to depend on the possible structures displayed by the native conformations, nor on any three-dimensional properties of the native conformation, aside from the general requirement of being compact. In fact, making use of the (MJ)$_d$ contact energies, any compact structure is an equally good native. Similar results as those discussed above were obtained making use of a random force with the same average and standard deviation of the (MJ)$_d$ contact energies, as well as with the Go model [@go].
Because of the large value of $\delta$, and the many possible sites available to introduce neutral mutations, there are of the order of $10^9-10^{10}$ sequences S$_{36}'$ obtained from S$_{36}$ through single- and multiple- amino acid concentration conserving mutations (swapping of amino acids) which have an energy, when calculated in the native conformation, lying within the gap $\delta$. The sequences S$_{36}'$ fold into the native conformation shown in Fig. 1a, in times of the order, although somewhat longer, than that associated with the folding of S$_{36}$ [@desig_1]. These results indicate that the native structure, which, as mentioned above, can be any compact structure, is highly designable.
A revealing example of the difficulties met in trying to design fast folding, highly designable proteins with “wild type”-like secondary structures is provided by the results of ref. [@wingreen] (cf. also [@kardar]), where a complete enumeration and energy calculation of all compact configurations of 2 letters chains containing 27 monomers was presented. The designability of each compact structure was measured by the number of sequences S$_{27}$ that can design the structure, that is, sequences that posses the structure as their non-degenerate ground state. It was found that highly designable conformations posses “protein-like” secondary structures and even tertiary symmetries, and are thermodynamically more stable than other conformations.
In keeping with the fact that the basic test a notional protein should pass to qualify as such is to fold to the native structure in a “short time” (typically of the order of $10^5-10^6$ MC steps), we have calculated the dynamics of a number of the sequences designed by Li [*et al.*]{} [@wingreen]. We have found that neither the sequences associated with the poorly-designable (cf. Fig.2(c)) nor with the highly-designable (cf. Fig.2(a)) conformations fold. Such sequences are thus unlikely candidates for coding functional proteins.
The reason for these results is to be found on the fact that all the structures of the two letter chain display a gap $\delta$ which is negative. In fact elongated, poorly designable and thus structurally dissimilar configurations exist (cf. Fig.2(b)), sampled by the Monte Carlo simulations which, for a given amino acid composition have an energy which is lower than that associated with the original, highly designable configuration. Defining the energy gap $\delta$ as the energy difference between the native and the lowest structural dissimilar [*compact*]{} configuration, leads to a positive value of $\delta$ which is of the order of the average contact energies, and consequently to an “order parameter” $\xi$ of the order of 1.
At the “microscopic” level we have found that all the configurations of the two letter 27mer chains display a very large number of “hot” sites. In fact, essentially half of all sites are “hot” sites (cf. Fig.2). In other words, introducing single point mutations in the most designable structures of ref.[@wingreen] one finds that one, out of two, leads to protein denaturation (i.e., in the present case, makes unstable the native structure), a behaviour not observed in “wild-type” single domain proteins. In fact, the strategy of multi-domain proteins, where the folding of each domain is controlled by a small number of strongly interacting amino acids, is used here to keep in place secondary structures on a single domain protein.
A possible way to design a viable notional protein making use of the 3D, 27-monomer chains of ref. [@wingreen] could be to modify the contact energies used in this paper, eventually introducing an angle dependence (cf. e.g. ref.[@lund1]) so as to selectively change the value of $\delta/{\bar U}$. That is, to force the ratio $(\delta/{\bar U})_{\rm non-desig}$ associated with the non-designable structures to remain at the present value (of the order of unity), while boosting the ratio $(\delta/{\bar U})_{\rm desig}$ characterizing the designable structures to become of the order of 10.
In keeping with this discussion we report results of MC simulations carried out for the twenty letter 27mer making use of contact energies taken from Table 5 of ref.[@mj] (“ordered” Miyazawa-Jerningan contact energies (MJ)$_o$). Also of a U-matrix containing 210 matrix elements of the form $U_{ij}=h_{ij}+\eta$, where $h_{ij}$ can be $-1$, $0$ or $1$ according to the hydrophobicity or polarity of the ith and jth residues, while $\eta$ is a random number taken from a Gaussian distribution centered around zero and with standard deviation $0.34$ [@maksim]. In these cases it is found that not all compact conformations can act as native configurations. In fact, one has to carefully choose the ratio between local and non-local contacts to obtain natives in which the designed sequence displays a large value of $\xi$. Examples are shown in Fig.3, where the corresponding conformations are associated with values of $\xi$ equal to $6.9$ (Fig.3(a)), $3.8$ (Fig.3(b)) and $0.2$ (Fig.3(c)). Although these conformations do not display secondary structures typical of “wild–type” proteins as discussed in ref. [@wingreen], they provide examples of the possible links which, appropriately chosen contact energies may create between selected 3D conformations and high entropy regions of the sequence space (large gap $\delta$), that is between structure and foldability.
One can conclude that stability, designability and fast folding are basic [*physical*]{} properties characterizing notional proteins, properties which depend only on the large number of degrees of freedom displayed by these system in sequence space, while the presence of secondary structures and tertiary symmetries are “pretty” expressions of detailed [*chemistry*]{}.
Financial support by NATO under grant CRG 940231 is gratefully acknowledged. The help of the late Dr. N. D’Alessandro in modelling advice is gratefully ackowledged.
[99]{}
T. Creighton, [*Proteins, Structure and Molecular Properties*]{}, W. H. Freeman, New York (1992)
R. Goldstein, Z.A. Luthey-Schulten and P. Wolynes, Proc. Natl. Acad. Sci. USA [**89**]{}, 4918 (1992)
A. Sali, E.I. Shakhnovich, and M. Karplus, J. Mol. Biol. [**235**]{}, 1614 (1994)
E. Shakhnovich and A. Gutin, Proc.Natl. Acad. Sci. USA [**90**]{}, 7195 (1993)
E.I. Shakhnovich, Phys. Rev. Lett. [**72**]{}, 3907 (1994)
S. Govindarajan and R. Goldstein, Biopolymers [**36**]{}, 43 (1995)
A. Gutin, V. Abkevich, and E. Shakhnovich, Phys. Rev. Lett. [**77**]{}, 5433 (1996)
J. Bryngelson, J.N. Onuchic, N.D. Socci, and P. Wolynes, Proteins: Struct. Funct. and Genetics [**21**]{}, 167 (1995)
E.I. Shakhnovich, Curr. Opin. Struct. Biol. [**7**]{}, 29 (1997)
R.A. Broglia, G. Tiana, H.E. Roman, E. Vigezzi and E. Shakhnovich (to be published)
Y. Ueda, H. Taketomi and N. Go, Int. Journal Peptide Protein Res. [**7**]{}, 445 (1975)
N. Go and H. Abe, Biopolymers [**20**]{}, 991 (1981)
K. Lau and K. Dill, Macromolecules [**22**]{}, 3986 (1989)
J. Sklonik, A. Kolinski and R. Yaris, Comments Mol. Cell Bioph. [**6**]{}, 223 (1990)
D. Covell and R. Jernigan, Biochemistry [**29**]{}, 3287 (1990)
A. Godzik, A. Kolinksi and J. Sklonik, J. Comput. Chem. [**14**]{}, 1194 (1993)
N. Socci, W. Bialek and J. Onuchic, Phys. Rev. [**E 49**]{}, 3440
D. Klimov and D. Thirumalai, Phys. Rev. Lett. [**76**]{}, 4070 (1996)
E.I. Shakhnovich, V. Abkevich and O. Ptitsyn, [*Nature*]{} [**379**]{}, 96 (1996)
V. Pande, A. Grosberg, D. Rokshar and T. Tanaka, Curr. Opin. Struct. Biol. [**8**]{}, 68 (1998)
H. Li, R. Helling, C. Tang and N. Wingreen, Science [**273**]{}, 666 (1996)
S. Miyazawa and R. Jernigan, Macromolecules [**18**]{}, 534 (1985)
G. Tiana, R. A. Broglia, H. E. Roman, E. Vigezzi and E. Shakhnovich, J. Chem. Phys. [**108**]{}, 757 (1998)
R. A. Broglia, G. Tiana, S. Pasquali, H. E. Roman and E. Vigezzi, Proc. Natl. Acad. Sci. USA [**95**]{}, 12930 (1998)
M. Kardar, Science [**273**]{}, 610 (1996)
A. Irback, C. Peterson, F. Potthast and O. Sommelius, J. Chem. Phys. [**107**]{}, 273 (1997)
M. Sgorobogatiy and G. Tiana, Phys. Rev. [**E 58**]{}, 3572 (1998)
W. Humphrey, A. Dalke and K. Schulten, [*J. Mol. Graphics*]{} [**14**]{}, 33 (1996)
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
Micron-scale swimmers move in the realm of negligible inertia, dominated by viscous drag forces. In this paper, we formulate the leading-order dynamics of a slender multi-link ($N$-link) microswimmer assuming small-amplitude undulations about its straight configuration. The energy-optimal stroke to achieve a given prescribed displacement in a given time period is obtained as the largest eigenvalue solution of a constrained optimal control problem. Remarkably, the optimal stroke is an ellipse lying within a two-dimensional plane in the ($N$-1)-dimensional space of joint angles, where $N$ can be arbitrarily large. For large $N$, the optimal stroke is a traveling wave of bending, modulo edge effects.
If the number of shape variables is small, we can consider the same problem when the prescribed displacement in one time period is large, and not attainable with small variations of the joint angles. The fully nonlinear optimal control problem is solved numerically for the cases $N$=3 (Purcell’s three-link swimmer) and $N$=5 showing that, as the prescribed displacement becomes small, the optimal solutions obtained using the small-amplitude assumption are recovered. We also show that, when the prescribed displacements become large, the picture is different. For $N$=3 we recover the non-convex planar loops already known from previous studies. For $N$=5 we obtain non-planar loops, raising the question of characterizing the geometry of complex high-dimensional loops.
author:
- |
Fran[ç]{}ois Alouges, Antonio DeSimone, Laetitia Giraldi,\
Yizhar Or, and Oren Wiezel
title: 'Energy-optimal strokes for multi-link microswimmers: Purcell’s loops and Taylor’s waves reconciled'
---
Introduction
============
The analysis of biological and bio-inspired swimming at microscopic scales has attracted considerable attention in the recent literature, starting from the seminal work by Taylor [@Taylor51], Lighthill [@Lighthill75], and Purcell [@Purcell77]. One of the reasons is that swimming of unicellular organisms is at the root of many fundamental processes in biology: reproduction through the swimming of sperm cells is just one example [@GrayHancock55; @Drezner1981; @Gaffney2011; @Gaffney2017]. Moreover, biology inspires the design of bio-mimetic artificial devices that may have important applications in medicine as drug delivery, diagnostic or therapeutic devices (for example: smart endoscopic capsules), see e.g. [@Dreyfus2005; @micromachines2014; @alouges2015soft]. The large size of the recent literature makes it impossible to provide an exhaustive survey of the state of the art. Specialized reviews have appeared in recent years, such as [@Gaffney2011; @LaugaPowers09; @StokerReview; @GoldsteinReview], and several monographs are available [@Lighthill75; @childress.book; @Bonnard_book]. The interested reader may find in these works and in the references cited therein several hundred papers to explore the subject.
At the scale of a single cell, viscous forces dominate inertia in fluid flows, which are then governed by the (steady) Stokes equations [@BrennerHappel65]. These arise from the Navier-Stokes equation, in the (formal) limit of zero Reynolds and Womersley numbers. The linearity of the Stokes system has the consequence that net propulsion through periodic shape actuation is only possible through histories of shape changes that are not reciprocal in time, a fact popularized as “the scallop theorem” by Purcell [@Purcell77].
Non-reciprocal periodic shape changes can arise in a number of ways: thanks to a non-trivial topology of the space of shapes, through non-trivial loops in the space of shapes, thanks to the propagation of travelling waves of deformation. Corresponding biological examples are, respectively, the rotary motion of helical bacterial flagellar bundles, the different shapes of cilia in the power and in the recovery part of one stroke, the beating of a eukaryotic flagellum causing the propagation of bending waves along the length of the flagellum. In fact, bending waves propagating along cilia/flagella and shape modulation during one stroke are used not only for propulsion by micro-organisms, but also for transport inside organs in humans and other higher organisms [@blake_1971; @Satir1990; @Cicuta2017]. The second example is the one with more connections with the study of minimal artificial swimmers (i.e., swimmers with only two internal degrees of freedom to control shape such as the three-link swimmer of Purcell [@Purcell77; @becker2003self], the three-sphere swimmer of Najafi and Golestanian [@najafi2004simple], and others.) The third example is possibly the most thoroughly exploited paradigm in the fabrication of micro-swimmer prototypes, often through the action of an external magnetic field on a flexible filament [@Dreyfus2005; @alouges2015soft; @Fischer2009; @Sitti2018].
Patterns of optimal actuation have been investigated independently, for each of these three examples, with a variety of analytical and numerical methods. For the case of flagellar and ciliary propulsion, these include [@Lighthill75; @Pironneau74; @berman2013undulatory; @Lauga2014; @HosoiOptimalBiflagellated] leading to recognizing, for example, helical shapes as the optimal ones for filaments in three dimensions, and smoothed saw-tooth traveling waves as the optimal wave forms for the planar beating of a one-dimensional flagellum or for a planar sheet. In the limit of small amplitudes, the latter reduces to Taylor’s traveling sine waves [@Taylor51]. For the study of optimal strokes for minimal artificial swimmers the reader is referred, e.g., to [@becker2003self; @TamHosoi2007; @alouges2013optimally; @alouges2008; @alougesM3AS; @oren_cdc2016; @Giuliani2018].
More recently, it has been recognized that, in the presence of external forces or torques, the scallop paradigm has to be reconsidered. This is a consequence of the fact that, in this case, the governing equations can no longer be cast as an affine control systems without drift. Similar remarks apply to the presence of parts of the swimmer body whose shape is not directly controlled and it instead emerges from the balance between elastic restoring forces and viscous resistances. Since here we will not pursue this issue any further, we refer the reader to some of the relevant literature [@Hosoi2link] and [@spagnolie2010optimal; @passov2012dynamics; @Montino; @cicconofri2016motion; @Gaffney2018].
In spite of all the recent progress, several questions remain open for investigation as testified by the growing pace at which research articles on low Reynolds number swimmers are being published. New aspects of the very same fundamental swimming problem are thus continuously emerging. Rather than adding one more, we are motivated here by the quest for unifying perspectives over this vast literature. For example: what is the connection, if any, between optimal actuation by traveling waves inspired by Taylor’s swimming sheet [@Taylor51] and optimal actuation by closed loops in the space of shapes shown in Tam and Hosoi’s gaits [@TamHosoi2007] for Purcell’s swimmer and further discussed in [@RazAvron]? The two main paradigms for producing non-reciprocal shape changes have remained mostly disconnected in the recent literature, confined to two independent streams of research.
Motivated by this question, we focus on one specific example, namely, a planar swimmer consisting of $N$ equal links of fixed length $L$ representing the $N$-dimensional generalization of Purcell’s famous three-link swimmer, and use geometric control theory to identify some unifying principles. In particular, we consider the problem of prescribing an admissible displacement in one shape cycle, and look for the gait that minimizes a suitable cost functional giving a measure of the expended energy during that cycle.
The problem of optimal control is non-linear in the shape parameters, and determining (even numerically) the optimal gait explicitly becomes very hard as soon as the number of shape parameters becomes large. Under the assumption of small-amplitude joint angles, by considering the approximation at leading order in the shape parameters, we obtain an affine control system which can be analyzed in full detail. In previous studies, asymptotic analysis in this small-angle regime has been used for Purcell’s and other swimmers where the number $N$ of shape degrees of freedom is small [@giraldi2015optimal; @wiezel2016optimization; @alouges2016parking], but never for large $N$. We find that optimal gaits are always two-dimensional elliptical loops, independent of $N$. These gaits bridge Purcell’s loops for the two-dimensional shape space associated with $N$=3, to gaits that, modulo edge effects, can be identified with Taylor’s traveling waves of bending for large $N$. Interestingly, this analysis can be done without any explicit reference to a concrete model for the interaction between the swimmer and the surrounding fluid, which can be modeled with the full detail of Stokes hydrodynamics, or with any of the simplified models to treat the viscous drag for a slender swimmer. The result only depends on structural properties and symmetries of the governing equations, which in turn reflect the geometric symmetry of the physical problems. In fact, in this regime of small-amplitude joint angles, the perturbations from the rectilinear geometry of the reference configuration are small, and a slender one-dimensional swimmer with homogenous geometric and mechanical properties that interacts with a homogeneous surrounding medium is a system which is essentially invariant under shifts along the body axis. This is exactly true for an infinite or a periodic system and approximately true, modulo edge effects, for a system of finite length. The relevance of traveling waves as optimal gaits is therefore naturally suggested by the geometric symmetry of the system and, in fact, it emerges naturally from the symmetries of the governing equations.
Our result can be seen as a unifying principle in the sense that many different concrete problems, where different geometries of micro-swimmers obtained as $N$ is varied, and different models for the fluid-structure interactions are employed, lead to different governing equations, but their behavior can be interpreted on the basis of the same single principle, thanks to the fact that the governing equations all share the same abstract structure and symmetries. To reinforce this point even further, we note that a similar result has been obtained in the different but related context of crawling of one-dimensional objects on solid surfaces or within a solid matrix [@Agostinelli], where it is shown that peristaltic waves used by earthworms for locomotion can be interpreted as optimal gaits when only small deformations are allowed. This is true provided, of course, that both mechanical properties of the crawler and frictional interactions with the surroundings are invariant with respect to shifts along the body axis.
To remove the assumption of small-amplitude joint angles and analyze the case in which large displacements are prescribed (and large joint angle variations are allowed) we can only resort to numerics, and our current capabilities limit the size of the system we can handle (in this paper, $N$=5). This is nevertheless sufficient to reveal that optimal gaits of large amplitude have interesting and nontrivial geometry. They may differ both qualitatively and quantitatively from the planar elliptical loops representing their small amplitude limits. In particular, we find non-convex planar loops for $N$=3, as it was already well-known from the analysis of optimal gaits for Purcell’s swimmer, but also non-planar closed space curves for $N$=5. However, all these nontrivial gaits duly converge to planar elliptical loops when the size of the prescribed displacement in one cycle becomes small.
Seen from the perspective of our results on the small-amplitude angle regime, these results are not surprising. In particular, when the restricted setting of small perturbations from the rectilinear geometry is abandoned, and large shape changes are considered, then invariance under shifts along the body axis is lost and traveling waves are no longer a natural basis for the study of the properties of optimal gaits. In addition, the optimal loops are no longer planar: the study of their geometric properties emerges in this way as a new and completely open field for future investigations that will require the development of new theoretical and numerical tools.
The rest of the paper is organized as follows. We describe the mathematical setting of the optimal control problem for the $N$-link swimmer in Section \[sec:dynamics\]. In Section \[sec:small\_amplitude\] we consider the regime of small-amplitude joint angles, hence of small deviations from the rectilinear shape, derive the general result that optimal loops are planar ellipses, and we solve numerically some specific examples which, for large $N$, show the emergence of traveling waves as optimal gaits. In Section \[sec:large\_amplitude\] we move to the regime of large amplitude joint angles, hence of large deviations from the rectilinear shape, and discuss some numerical results obtained via direct numerical optimization using the software [@bonnans2012Bocop]: we exhibit non-planar complex loops for the case $N$=5 and show that they converge to planar ellipses as the size of the prescribed displacement in one cycle becomes small.
Dynamics of the $N$-link swimmer {#sec:dynamics}
================================
We develop our analysis at leading-order of the dynamics of the $N$-link neutrally buoyant swimmer using only the assumptions following from the linearity of Stokes equations, their invariance properties, and the symmetries of the swimmer. These assumptions are valid for several classic models such as Purcell’s swimmer [@Purcell77], the $N$-link swimmer in [@AlougesDeSimone13], and Taylor’s swimming sheet [@Taylor51]. They also hold irrespectively of whether the forces exerted by the fluid surrounding the swimmer are calculated by solving explicitly the Stokes equations, or they are evaluated by using any of the several approximate methods that have been used for slender swimmers (resistive force theory [@GrayHancock55], slender body theory [@cox1970] etc.).
The swimmer is planar and composed of $N$ identical segments connected by rotational joints, see Fig \[Fig:N-link\_notation\]. We assume in the following that $N$ is odd, and denote by $(x,y)$ the position of the midpoint of the central link and $\theta$ its orientation (the angle that the central link makes with the horizontal axis), see Fig \[Fig:N-link\_notation\]. The case of even $N$ simply requires a different choice of coordinates. The swimmer is immersed in an unbounded domain occupied by a homogeneous viscous fluid governed by Stokes equations and is driven by shape changes, i.e., the joint angles ${\boldsymbol\upphi}=(\phi_1,\phi_2,\ldots\phi_{N-1})^T$ are given functions of time $t$. In particular, we will be concerned with $T-$periodic shape changes ${\boldsymbol\upphi}(t)$, which we call strokes, satisfying ${\boldsymbol\upphi}(t) = {\boldsymbol\upphi}(T+t), \quad \forall t>0$. Our analysis aims at resolving the leading-order terms of the swimmer’s dynamics, and is restricted to up to second order $O({\varepsilon}^2)$ terms in the shape parameters $\phi_i(t)=O({\varepsilon})$, which are assumed to be small, i.e. ${\varepsilon}\ll 1$.
(-5,0) – (5,0) ; (-4,2.3) node [$\bullet$]{} – ++(-55:2) coordinate (A) node [$\bullet$]{} ; (A) – ++(-55:1) ; (A) ++(-55:0.5)\[->\] arc(-55:-30:0.5); (A) ++(-55:0.7) node\[right\][$\phi_1$]{};
\(A) – ++(-30:2) coordinate (B) node [$\bullet$]{} ; (B) – ++(-30:1) ; (B) ++(-30:0.5)\[->\] arc(-30:20:0.5); (B) ++(-5:0.5) node\[right\][$\phi_2$]{};
\(B) – ++(20:2) coordinate (C) node [$\bullet$]{} ; (C) – ++(20:1) ; (C) ++(20:0.5)\[->\] arc(20:45:0.5); (C) ++(40:0.7) node\[right\][$\phi_3$]{};
\(C) – ++(45:2) coordinate (D) node [$\bullet$]{} ; (D) – ++(45:1) ; (D) ++(45:0.5)\[->\] arc(45:10:0.5); (D) ++(35:0.7) node\[right\][$\phi_4$]{};
\(D) – ++(10:2) coordinate (D) node [$\bullet$]{} ;
(B)++(20:1) coordinate (X) node [$\bullet$]{} ; (X)++(1.2,0) arc(0:20:1.2); (X)++(1.2,0.2) node\[right\][$\theta$]{} ; (X)++(0,0.2) node\[above\][$(x,y)$]{};
(-5,0) – (5,0) ; (-4,2.3) node [$\bullet$]{} – ++(-55:2) coordinate (A) node [$\bullet$]{} ; (A) – ++(-55:1) ; (A) ++(-55:0.5)\[->\] arc(-55:-30:0.5); (A) ++(-55:0.7) node\[right\][$\phi_1$]{};
\(A) – ++(-30:2) coordinate (B) node [$\bullet$]{} ; (B) – ++(-30:1) ; (B) ++(-30:0.5)\[->\] arc(-30:20:0.5); (B) ++(-5:0.5) node\[right\][$\phi_2$]{};
\(B) – ++(20:2) coordinate (C) node [$\bullet$]{} ; (C) – ++(20:1) ; (C) ++(20:0.5)\[->\] arc(20:45:0.5); (C) ++(40:0.7) node\[right\][$\phi_3$]{};
\(C) – ++(45:2) coordinate (D) node [$\bullet$]{} ; (D) – ++(45:1) ; (D) ++(45:0.5)\[->\] arc(45:10:0.5); (D) ++(35:0.7) node\[right\][$\phi_4$]{};
\(D) – ++(10:2) coordinate (D) node [$\bullet$]{} ;
(B)++(20:1) coordinate (X) node [$\bullet$]{} ; (X)++(1.2,0) arc(0:20:1.2); (X)++(1.2,0.2) node\[right\][$\theta$]{} ; (X)++(0,0.2) node\[above\][$(x,y)$]{};
(-6,-1.5) – (6,-1.5) ; (5.5,-1.5) node\[above\][$x$-axis]{}; (-5,-3) – (5,-3) ; (-4,-5.3) node [$\bullet$]{} – ++(55:2) coordinate (A) node [$\bullet$]{} ; (A) – ++(55:1) ; (A) ++(55:0.5)\[->\] arc(55:30:0.5); (A) ++(55:1) node\[right\][$-\phi_1$]{};
\(A) – ++(30:2) coordinate (B) node [$\bullet$]{} ; (B) – ++(30:1) ; (B) ++(30:0.5)\[->\] arc(30:-20:0.5); (B) ++(5:0.5) node\[right\][$-\phi_2$]{};
\(B) – ++(-20:2) coordinate (C) node [$\bullet$]{} ; (C) – ++(-20:1) ; (C) ++(-20:0.5)\[->\] arc(-20:-45:0.5); (C) ++(-45:0.6) node\[right\][$-\phi_3$]{};
\(C) – ++(-45:2) coordinate (D) node [$\bullet$]{} ; (D) – ++(-45:1) ; (D) ++(-45:0.5)\[->\] arc(-45:-10:0.5); (D) ++(-39:0.7) node\[right\][$-\phi_4$]{};
\(D) – ++(-10:2) coordinate (D) node [$\bullet$]{} ;
(B)++(-20:1) coordinate (X) node [$\bullet$]{} ; (X)++(1.2,0) arc(0:-20:1.2); (X)++(1.2,-0.2) node\[right\][$-\theta$]{} ; (X)++(0,-0.8) node\[above\][$(x,-y)$]{};
The dynamics of the swimmer’s planar motion is governed by $$\label{eq.dyn}
\left\{
\begin{array}{l}
\dot{x} = {\boldsymbol{f}}(\theta,{\boldsymbol\upphi})\cdot \dot{{\boldsymbol\upphi}}\,,\\
\dot{y} = {\boldsymbol{g}}(\theta,{\boldsymbol\upphi})\cdot \dot{{\boldsymbol\upphi}}\,,\\
\dot{\theta} = {\boldsymbol{h}}({\boldsymbol\upphi})\cdot \dot{{\boldsymbol\upphi}}\,.\\
\end{array}
\right.$$ These equations come from the balance of viscous force and torque, which are linear in $\dot{x},\dot{y},\dot{\theta}$ and $\dot{{\boldsymbol\upphi}}$. Details of the derivation in the case of resistive force theory [@GrayHancock55] are given in [@wiezel2016optimization; @AlougesDeSimone13; @giraldi2013controllability]. The special structure of , namely the fact that functions ${\boldsymbol{f}}$, ${\boldsymbol{g}}$, and ${\boldsymbol{h}}$ are independent of $x$ and $y$, and that the last one is also independent of $\theta$, are consequences of the translational and rotational invariance of the problem.
Due to the geometric structure of the swimmer, the functions ${\boldsymbol{f}},{\boldsymbol{g}}$ and ${\boldsymbol{h}}$ that define the dynamics satisfy further relations that are deduced from symmetries of the system. Among them, the symmetry with respect to the $x-$axis, depicted in Fig. \[Fig:symmetry\_figs\], transforms the swimmer parametrized by ${\boldsymbol\upphi}$ at position $(x,y,\theta)$ to the one parametrized by $-{\boldsymbol\upphi}$, at position $(x,-y,-\theta)$. The invariance of the dynamics under such a transformation leads to $$\left\{
\begin{array}{l}
{\boldsymbol{f}}(\theta,{\boldsymbol\upphi})\cdot \dot{{\boldsymbol\upphi}} = {\boldsymbol{f}}(-\theta,-{\boldsymbol\upphi})\cdot (-\dot{{\boldsymbol\upphi}})\,,\\
{\boldsymbol{g}}(\theta,{\boldsymbol\upphi})\cdot \dot{{\boldsymbol\upphi}} = -{\boldsymbol{g}}(-\theta,-{\boldsymbol\upphi})\cdot (-\dot{{\boldsymbol\upphi}})\,,\\
{\boldsymbol{h}}({\boldsymbol\upphi})\cdot \dot{{\boldsymbol\upphi}} = -{\boldsymbol{h}}(-{\boldsymbol\upphi})\cdot (-\dot{{\boldsymbol\upphi}})\,.
\end{array}
\right.
\label{symmx}$$
Dynamics at leading order {#subsec:leading_order_dynamics}
-------------------------
For $\theta$ and ${\boldsymbol\upphi}$ of order ${\varepsilon}$, we expand the dynamics to second order in $\theta$, ${\boldsymbol\upphi}$ and $\dot{{\boldsymbol\upphi}}$ by expanding ${\boldsymbol{f}}$ to first order as $${\boldsymbol{f}}(\theta,{\boldsymbol\upphi}) \sim {\mathbf{F}}_0 + {\mathbf{F}}_\theta \theta + {\mathbf{F}}_{{\boldsymbol\upphi}} {\boldsymbol\upphi}+O({\varepsilon}^2)
\label{expansion}$$ where ${\mathbf{F}}_0={\boldsymbol{f}}(0,0)\in \mathbb{R}^{N-1}$, ${\mathbf{F}}_\theta=\frac{\partial {\boldsymbol{f}}}{\partial \theta}(0,0)\in \mathbb{R}^{N-1}$ and ${\mathbf{F}}_\phi=\frac{\partial {\boldsymbol{f}}}{\partial {\boldsymbol\upphi}}(0,0)\in \mathbb{R}^{N-1,N-1}$. We also expand similarly ${\boldsymbol{g}}$ and ${\boldsymbol{h}}$. Using the symmetry relations , we get for ${\boldsymbol{f}}$, ${\boldsymbol{g}}$ and ${\boldsymbol{h}}$ respectively $$\begin{aligned}
\left({\mathbf{F}}_0 + {\mathbf{F}}_\theta \theta + {\mathbf{F}}_{{\boldsymbol\upphi}} {\boldsymbol\upphi}\right) \cdot \dot{{\boldsymbol\upphi}} &=& -\left({\mathbf{F}}_0 - {\mathbf{F}}_\theta \theta - {\mathbf{F}}_{{\boldsymbol\upphi}} {\boldsymbol\upphi}\right) \cdot \dot{{\boldsymbol\upphi}}\\
\left({\mathbf{G}}_0 + {\mathbf{G}}_\theta \theta + {\mathbf{G}}_{{\boldsymbol\upphi}} {\boldsymbol\upphi}\right) \cdot \dot{{\boldsymbol\upphi}} &=& \left({\mathbf{G}}_0 - {\mathbf{G}}_\theta \theta - {\mathbf{G}}_{{\boldsymbol\upphi}} {\boldsymbol\upphi}\right) \cdot \dot{{\boldsymbol\upphi}}\\
\left({\mathbf{H}}_0 + {\mathbf{H}}_{{\boldsymbol\upphi}} {\boldsymbol\upphi}\right) \cdot \dot{{\boldsymbol\upphi}} &=& \left({\mathbf{H}}_0 - {\mathbf{H}}_{{\boldsymbol\upphi}} {\boldsymbol\upphi}\right) \cdot \dot{{\boldsymbol\upphi}},\end{aligned}$$ from which we deduce, ${\boldsymbol\upphi}$ and $\dot{{\boldsymbol\upphi}}$ being arbitrary, that ${\mathbf{F}}_0={\mathbf{G}}_\theta = 0$ and ${\mathbf{G}}_{{\boldsymbol\upphi}}={\mathbf{H}}_{{\boldsymbol\upphi}}=0$. The original system therefore reduces to $$\label{eq.dyn2}
\left\{
\begin{array}{l}
\dot{x} = \left({\mathbf{F}}_\theta \theta + {\mathbf{F}}_{{\boldsymbol\upphi}} {\boldsymbol\upphi}\right)\cdot \dot{{\boldsymbol\upphi}}+O(\varepsilon^3)\,,\\
\dot{y} = {\mathbf{G}}_0\cdot \dot{{\boldsymbol\upphi}}+O(\varepsilon^3)\,,\\
\dot{\theta} = {\mathbf{H}}_0\cdot \dot{{\boldsymbol\upphi}}+O(\varepsilon^3)\,.\\
\end{array}
\right.$$
Integrating in time we obtain $$\theta(t) = \int_0^t \dot{\theta} \,dt = \int_0^t{\mathbf{H}}_0\cdot \dot{{\boldsymbol\upphi}}\,dt ={\mathbf{H}}_0\cdot ({\boldsymbol\upphi}(t)-{\boldsymbol\upphi}(0))+O({\varepsilon}^3)
\label{eqtheta}$$ and $$y(t) = \int_0^t \dot{y} \,dt = \int_0^t{\mathbf{G}}_0\cdot \dot{{\boldsymbol\upphi}}\,dt =
{\mathbf{G}}_0 \cdot ({\boldsymbol\upphi}(t)-{\boldsymbol\upphi}(0))+O({\varepsilon}^3) \,.
$$
We deduce that, up to second order, the swimmer experiences no global rotation or transverse translation after one complete stroke. Indeed, ${\boldsymbol\upphi}$ being $T$-periodic, we have $$\Delta \theta = \theta(T) - \theta(0) = O(\varepsilon^3)\,,\quad\quad \Delta y = y(T) - y(0) = O(\varepsilon^3)\,.\label{eq:delta_th_y}$$
Moreover, using the expression for $\theta$ in in the first equation of and integrating over one period leads to $$\Delta x = x(T)-x(0) = \int_0^T \left({\mathbf{H}}_0^T \otimes {\mathbf{F}}_{\theta} + {\mathbf{F}}_{{\boldsymbol\upphi}} \right){\boldsymbol\upphi}\cdot \dot{{\boldsymbol\upphi}}\,dt + O(\varepsilon^3) \,.$$ Calling ${\mathbf{F}}^{\theta}_{{\boldsymbol\upphi}}$ the matrix ${\mathbf{H}}_0^T \otimes {\mathbf{F}}_{\theta} + {\mathbf{F}}_{{\boldsymbol\upphi}}$, integrating by parts the last expression, and using again the $T$-periodicity of ${\boldsymbol\upphi}$ leads to $$\int_0^T {\mathbf{F}}^{\theta}_{{\boldsymbol\upphi}} {\boldsymbol\upphi}\cdot \dot{{\boldsymbol\upphi}}\,dt = - \int_0^T {\mathbf{F}}^{\theta}_{{\boldsymbol\upphi}} \dot{{\boldsymbol\upphi}} \cdot {\boldsymbol\upphi}\,dt = - \int_0^T {{\mathbf{F}}^{\theta}_{{\boldsymbol\upphi}}}^T {\boldsymbol\upphi}\cdot \dot{{\boldsymbol\upphi}}\,dt \,.$$ Finally we get the net longitudinal displacement in one cycle as $$\begin{aligned}
\Delta x
&= \int_0^T \frac12 ({\mathbf{F}}^{\theta}_{{\boldsymbol\upphi}}-{{\mathbf{F}}^{\theta}_{{\boldsymbol\upphi}}}^T) {\boldsymbol\upphi}\cdot \dot{{\boldsymbol\upphi}}\,dt+O({\varepsilon}^3) \label{eq:deltax}\,.\end{aligned}$$
Our analysis at order two in the joint-angle amplitudes proves that the net lateral displacement $\Delta y$ and the net rotation both vanish at order two, as given in Eq. , while $\Delta x$ may be non-zero at order two, as shown in Eq. . The fundamental physical explanation of this result lies in the fact that the swimmer has symmetry about its straightened configurations, and only time-periodic strokes consisting of small-amplitude deviations about this configuration are analyzed. These are zero-mean trajectories, which, due to the swimmer’s symmetry, cancel out all displacements at leading order, except for the X displacement which is along the swimmer’s axis of symmetry. This fact has also been previously explained in some related works . This is also related to the fact the Lie-bracket vector field for each pair of the joint angle inputs, evaluated at the zero (straight) configuration, gives only $x-$displacement while all other motions only appear in higher order Lie brackets, see [@probprog96].
Other symmetries may be also used to obtain further information about the remaining coefficients of the system. For instance, using the rotational invariance of the system one can show that $$\left\{
\begin{array}{l}
{\boldsymbol{f}}(\theta,{\boldsymbol\upphi}) = \cos(\theta){\boldsymbol{f}}(0,{\boldsymbol\upphi}) -\sin(\theta) {\boldsymbol{g}}(0,{\boldsymbol\upphi})\,,\\
{\boldsymbol{g}}(\theta,{\boldsymbol\upphi}) = \sin(\theta){\boldsymbol{f}}(0,{\boldsymbol\upphi}) +\cos(\theta) {\boldsymbol{g}}(0,{\boldsymbol\upphi})\,,
\end{array}
\right.$$ from which one easily obtains ${\mathbf{F}}_\theta = -{\mathbf{G}}_0$ (and ${\mathbf{G}}_\theta = {\mathbf{F}}_0 = 0$). Similarly, using the symmetry of the system with respect to the $y$-axis, it can be shown that both ${\mathbf{F}}_\theta$ and ${\mathbf{G}}_0$ are ‘even’ vectors (in the sense that ${\mathbf{F}}_{\theta,i}={\mathbf{F}}_{\theta,N-i}$ and similarly for ${\mathbf{G}}_0$) while ${\mathbf{H}}_0$ is an ‘odd’ vector (for which ${\mathbf{H}}_{0,i}=-{\mathbf{H}}_{0,N-i}$.)
Power expended
--------------
We now derive a leading-order expression for the mechanical energy expended by the swimmer during one cycle. The instantaneous power (i.e. rate of mechanical work) is given by the scalar product between force and velocity densities integrated over its surface (see e.g. [@Lighthill75]). Moreover, due to the linearity of Stokes equations, both the forces and the velocities acting on the swimmer depend linearly on ${\dot{\boldsymbol\upphi}}$. Thus, the instantaneous power density expended by the swimmer is a quadratic form in ${\dot{\boldsymbol\upphi}}$, with coefficients depending on ${\boldsymbol\upphi}$ (see [@wiezel2016optimization] for concrete expressions using resistive force theory). We may therefore write the total energy $E$ expended by the swimmer during the stroke as $$\label{eq:energy_full}
E({\boldsymbol\upphi}) = \int_0^T < {\mathbf{P}}({\boldsymbol\upphi}) {\dot{\boldsymbol\upphi}},{\dot{\boldsymbol\upphi}}>\,dt \,,$$ where ${\mathbf{P}}({\boldsymbol\upphi})\in \mathcal{M}_{N,N}(\mathbb{R})$ is a symmetric and positive definite matrix, and the bracket $<\cdot,\cdot>$ stands for the scalar product in $\mathbb{R}^N$. Expanding this expression gives the $O({\varepsilon}^2)$ leading-order term of the energy as $$\label{eq:energy}
E_0({\boldsymbol\upphi}) = \int_0^T < {\mathbf{P}}_0 {\dot{\boldsymbol\upphi}},{\dot{\boldsymbol\upphi}}>\,dt \,,$$ where ${\mathbf{P}}_0 := {\mathbf{P}}({\boldsymbol\upphi}=0)$ is symmetric and positive definite.
Optimal strokes of small amplitude {#sec:small_amplitude}
==================================
In this section, we study energy-optimal strokes while focusing on the case in which the joint angles remain small (small amplitude approximation), so that only small perturbations of the rectilinear geometry are allowed.
Analysis of the optimal control problem
---------------------------------------
We begin by reviewing the well-known criterion of optimal energy efficiency due to Lighthill, and discuss its relation with the energy-optimal strokes studied here. Lighthill’s efficiency of swimming is defined as the ratio of $E_{drag} / E_{swim}$, where $E_{drag}$ is the energy needed by an external actor to pull the swimmer during a time $T$ at an average speed $\Delta x/T$, while $E_{swim}$ is the energy expended by the swimmer during a stroke to propel itself at the same average speed. As the drag force is proportional to the velocity, $E_{drag}$ is proportional to $(\Delta x)^2/T$ whereas $E_{swim}$ behaves like $\Delta x$ as and indicate ($\Delta x$ and $E$ being quadratic in $({\boldsymbol\upphi},\dot{{\boldsymbol\upphi}})$ are of the same order $O(\varepsilon^2)$). Thus, Lighthill’s efficiency is proportional to $\Delta x$, and maximizing it amounts to increase $\Delta x$, and thus the amplitude of the stroke as much as possible, beyond the range of small angles. This implies that optimal strokes that maximize Lighthill’s efficiency typically involve large-angle trajectories [@TamHosoi2007; @wiezel2016optimization]. On the other hand, if the amplitude of the stroke is constrained, by e.g. the maximum allowed amplitude of the angles, the displacement will be constrained as well, and it is expected that this constraint will be saturated when attempting to maximize Lighthill’s efficiency, making the resulting optimal strokes of limited general interest. We therefore turn our attention to the problem of maximizing the efficiency (i.e. minimizing the energy) for a displacement achieved during one stroke which is bounded, or more simply fixed. Optimal strokes are thus defined as $T$-periodic strokes ${\boldsymbol\upphi}(t)$ that expend the minimal energy $E$, among all strokes that achieve a given displacement $\Delta x$ in a given time period $T$.
We showed in the previous section that, at leading order, the $y$ and $\theta$ displacements are negligible with respect to the $x$ displacement. Optimal strokes that provide a (longitudinal) displacement $\Delta x$ are thus sought as solutions to the constrained minimization problem $$\label{eq:problem}
\min_{{\boldsymbol\upphi}\in \mathcal{D}} E_0({\boldsymbol\upphi})\,,$$ where, using , we have $$\mathcal{D}:=\left\{{\boldsymbol\upphi}\in (\mathcal{C}^1(0,T))^{N-1} | {\boldsymbol\upphi}(0) = {\boldsymbol\upphi}(T)\,\textrm{and}\,\frac12\int_0^T\!\! ({\mathbf{F}}^{\theta}_{{\boldsymbol\upphi}}-{{\mathbf{F}}^{\theta}_{{\boldsymbol\upphi}}}^T) {\boldsymbol\upphi}\cdot \dot{{\boldsymbol\upphi}}dt = \Delta x\right\}\!.$$
In the following, we show that the solution of this optimization problem describes a planar ellipse in shape space, which is $(N-1)$-dimensional. We also give a method for its computation. Note that discrepancies between the solutions of the optimal control problem defined above under the leading-order approximation and under the exact nonlinear dynamics in and become smaller and smaller by decreasing the displacement $\Delta x$ and making the stroke amplitude ${\varepsilon}$ of order $O(\sqrt{\Delta x})$.\
In order to find the Euler-Lagrange first-order necessary condition associated with the constrained optimization problem (\[eq:problem\]), we set $$\tilde{\mathcal{E}}_0({\boldsymbol\upphi})= \mathcal{E}_0({\boldsymbol\upphi})+\lambda K_0({\boldsymbol\upphi})$$ where $\displaystyle K_0({\boldsymbol\upphi}) = \frac12 \int_0^T ({\mathbf{F}}^{\theta}_{{\boldsymbol\upphi}}-{{\mathbf{F}}^{\theta}_{{\boldsymbol\upphi}}}^T) {\boldsymbol\upphi}\cdot \dot{{\boldsymbol\upphi}}\,dt$ and $\lambda \in \mathbb{R}$ is the Lagrange multiplier associated with the constraint $K_0({\boldsymbol\upphi}) = \Delta x$. Writing that the first order functional derivative of $\tilde{\mathcal{E}}_0$ vanishes at the optimal stroke ${\boldsymbol\upphi}^*$ $$\delta \tilde{\mathcal{E}}_0({\boldsymbol\upphi}^*)=0$$ amounts to writing $$\frac{d}{dt}\frac{\partial L}{\partial \dot{{\boldsymbol\upphi}}}({\boldsymbol\upphi}^*,\dot{{\boldsymbol\upphi}}^*) -\frac{\partial L}{\partial {\boldsymbol\upphi}}({\boldsymbol\upphi}^*,\dot{{\boldsymbol\upphi}}^*)=0
\label{EL}$$ where the Lagrangian $L$ is such that $$\tilde{\mathcal{E}}_0({\boldsymbol\upphi}) = \int_0^T L({\boldsymbol\upphi},\dot{{\boldsymbol\upphi}})\,dt\,.$$ Here, we have $$L({\boldsymbol\upphi},\dot{{\boldsymbol\upphi}}) = <{\mathbf{P}}_0\dot{{\boldsymbol\upphi}}, \dot{{\boldsymbol\upphi}}> +\frac{\lambda}{2} ({\mathbf{F}}^{\theta}_{{\boldsymbol\upphi}}-{{\mathbf{F}}^{\theta}_{{\boldsymbol\upphi}}}^T) {\boldsymbol\upphi}\cdot \dot{{\boldsymbol\upphi}}$$ and therefore becomes $$\label{eq:optimal_stroke}
{\mathbf{P}}_0\ddot{{\boldsymbol\upphi}}^* = \frac{\lambda}{2} ({{\mathbf{F}}^{\theta}_{{\boldsymbol\upphi}}}^T-{\mathbf{F}}^{\theta}_{{\boldsymbol\upphi}}) \dot{{\boldsymbol\upphi}}^* \,.$$
We denote by ${\mathbf{M}}$ the skew symmetric matrix $$\label{eq:matrix_M}
{\mathbf{M}}= \frac12{\mathbf{P}}_0^{-\frac12}({{\mathbf{F}}^{\theta}_{{\boldsymbol\upphi}}}^T-{\mathbf{F}}^{\theta}_{{\boldsymbol\upphi}}){\mathbf{P}}_0^{-\frac12}\,,$$ and decompose the equation along the eigen-elements of ${\mathbf{M}}$. Eigenvectors of skew symmetric matrices go by pairs, associated with conjugate and purely imaginary eigenvalues. We therefore set $({\mathbf{v}}^\pm_j)_{1\leq j\leq N'}$ ($N'=\lfloor \frac{N}{2} \rfloor$) the (complex and orthonormal) eigenvectors associated with the purely imaginary eigenvalue $\pm i\mu_j$ with $\mu_j\geq 0$: $${\mathbf{M}}{\mathbf{v}}^\pm_j = \pm i\mu_j {\mathbf{v}}^\pm_j\,.$$ Projecting ${\mathbf{P}}_0^\frac12{\boldsymbol\upphi}^*$ on the $({\mathbf{v}}^\pm_j)_{1\leq j\leq N'}$ as $${\mathbf{P}}_0^\frac12{\boldsymbol\upphi}^*(t) = \sum_{j=1}^{N'} \psi^\pm_j(t){\mathbf{v}}^\pm_j,$$ we deduce from (\[eq:optimal\_stroke\]) that $$\ddot{\psi}^\pm_j = \pm i \lambda \mu_j \dot{\psi}^\pm_j \,,$$ or $\psi^\pm_j(t) = \frac{\alpha_j^\pm}{\pm i \lambda \mu_j} \exp(\pm i \lambda \mu_j t)+ C^\pm_j$ where $C^\pm_j$ is a constant that we may take equal to 0. The solution of is thus expressed as
$$\label{eq:phi_star}
{\boldsymbol\upphi}^*(t) = \sum_{j=1}^{N'} \frac{ \alpha^\pm_j\, \text{e}^{\pm i \lambda \mu_j t}}{\pm i \lambda \,\mu_j} {\mathbf{P}}_0^{-1/2} {\mathbf{v}}^\pm_j,\quad \text{for $t \in [0,T]$}\,.$$
Since we focus on periodic strokes, ${\boldsymbol\upphi}(0)={\boldsymbol\upphi}(T)$, we must have $$\label{eq:periodic_stroke}
\lambda \mu_j T = 2 \pi k_j\quad 1\leq j \leq N', k_j \in \mathbb{N}.$$ By plugging the solution into , we find $$\label{eq:energy_0}
E_0({\boldsymbol\upphi}^*) = T\,\sum_{j=1}^{N'} (|\alpha^+_j|^2+|\alpha^-_j|^2)\,,$$ while the $x$-displacement is given by $$\label{eq:displacement_0}
\Delta x = T\,\sum_{j=1}^{N'} \frac{(|\alpha^+_j|^2+|\alpha^-_j|^2)}{\lambda} = \frac{E_0({\boldsymbol\upphi}^*)}{\lambda}\,.$$
Since $\Delta x$ is fixed in the optimization problem, minimizing the energy requires choosing $\lambda$ as small as possible. In view of , this is achieved if $\lambda=\pm\frac{2\pi}{\mu_{\scriptscriptstyle M}T}$, where $\mu_{\scriptscriptstyle M} = \max\left\{\mu_j\,,1\leq j \leq N' \right\}$, the direction of the translation depending on the sign of $\lambda$, and $\psi^\pm_j=0$ if $\mu_j\ne \mu_{\scriptscriptstyle M}$. Assuming $\lambda>0$ ($\lambda<0$ is handled similarly), we deduce that $\lambda = 2\pi/{T \mu_{\scriptscriptstyle M}}$ and the solution has only two modes corresponding to ${\mathbf{v}}^\pm_{\scriptscriptstyle M}$ $$\label{eq:phi_star_final}
{\boldsymbol\upphi}^*(t) = \frac{\alpha^+_{\scriptscriptstyle M}\, \text{e}^{ 2i\pi t}}{2 i \pi} {\mathbf{P}}_0^{-1/2} {\mathbf{v}}^+_{\scriptscriptstyle M} - \frac{\alpha^-_{\scriptscriptstyle M}\, \text{e}^{- 2i\pi t}}{2 i \pi} {\mathbf{P}}_0^{-1/2} {\mathbf{v}}^-_{\scriptscriptstyle M},\quad \text{for $t \in [0,T]$}\,,$$ and is therefore an ellipse drawn in the plane $({\mathbf{P}}_0^{-1/2} {\mathbf{v}}^+_{\scriptscriptstyle M},{\mathbf{P}}_0^{-1/2} {\mathbf{v}}^-_{\scriptscriptstyle M})$. Furthermore, noticing that ${\boldsymbol\upphi}^*$ is real and ${\mathbf{v}}_{\scriptscriptstyle M}^+ = \overline{{\mathbf{v}}_{\scriptscriptstyle M}^-}$, we must have $\alpha_{\scriptscriptstyle M}^+ = \overline{\alpha_{\scriptscriptstyle M}^-}$. The net-displacement achieved by the gait will be $\Delta x = \frac{2T}{\lambda}|\alpha^+_{\scriptscriptstyle M}|^2$. Therefore, the optimal gait amplitude scales as $\sqrt{\Delta x}$, as expected from .
Let us point out the important observation that the optimal gait is always an ellipse lying within a two-dimensional plane regardless of the number of links $N$, which makes the dimension of the shape space arbitrarily large. In addition, the optimal gait in can be written equivalently as sinusoidal inputs: $$\label{eq:sin}
\phi_k^*(t) = a_k \sin(\omega t + p_k), \quad k=1,\ldots,N\!-\!1$$ where $\omega = 2 \pi /T$. The amplitudes $a_k$ in are of order $O({\varepsilon})$, and scale as $\sqrt{\Delta x}$. Finally, an important property of the optimal solution ${\boldsymbol\upphi}^*(t)$ is that it satisfies $$P(t)=< {\mathbf{P}}_0 {\dot{\boldsymbol\upphi}}(t),{\dot{\boldsymbol\upphi}}(t)>=2|\alpha_{\scriptscriptstyle M}^+|^2=const\,,
\label{eq:constant_power}$$ where $P(t)$ is the mechanical power generated by the swimmer. Equation implies that the optimal gait generates a constant power over the entire cycle. A similar result for the three-sphere swimmer was proved in [@alouges2008] and this agrees with a fundamental observation made in [@becker2003self], which states that for any given trajectory in shape space, the time-parametrization associated with constant power is the one that minimizes the total energy expenditure.
Numerical results: from Purcell’s loops to Taylor’s waves {#Sec:Numerical}
---------------------------------------------------------
In this section, we present numerical results for the problem of computing optimal gaits for slender swimmers by using the derivation described above and leading to the Euler-Lagrange equations . In order to derive a simple and concrete formulation of the dynamics, we use here the local drag approximation of resistive force theory [@GrayHancock55; @cox1970] for slender links. Our computations are made using . A similar approach based on the the Euler-Lagrange equations was used in [@alouges2008] for the three-sphere swimmer where, however, nonlocal hydrodynamic forces were fully resolved by solving the Stokes system for the surrounding fluid.
Our numerical code is based on the computation of ${\boldsymbol\upphi}^*(t)$ from formula . The first steps do not rely on the small amplitude approximation, and they will be used in later sections as well.
First, we derive the dynamics of the swimmer in using resistive force theory [@cox1970; @GrayHancock55]. This theory states that the viscous drag force ${\boldsymbol{f}}_i$ and torque $m_i$ on the $i^{th}$ slender link of length $l$ under planar motion are proportional to its linear and angular velocities, respectively. Thus, one can write the expression for the drag force and torque exerted on each link: $$\label{eq.rft}
\begin{array}{c}
{\boldsymbol{f}}_i=-c_t l({\mathbf{v}}_i\cdot{\mathbf{t}}_i){\mathbf{t}}_i-c_n l({\mathbf{v}}_i\cdot{\mathbf{n}}_i){\mathbf{n}}_i \\[12pt]
m_i=-\dfrac{1}{12} c_n l^3 \omega_i,
\end{array}$$ where ${\mathbf{v}}_i$, $\omega_i$ are the linear and angular velocities of the $i^{th}$ link, and ${\mathbf{t}}_i,\; {\mathbf{n}}_i$ are unit vectors in its axial and normal directions. The resistance coefficients in are $c_n\!=\!2c_t\!=4\pi\eta/\log(l/a)$ where $\eta$ is the dynamic viscosity of the fluid and $a\ll l$ is the radius of the slender links’ cross-section. Using , the swimmer’s dynamic equations can be derived from force and torque balance, see for instance [@wiezel2016optimization] for Purcell’s swimmer and [@AlougesDeSimone13] for the general $N$-link swimmer.
Next, we exploit the assumption of small amplitude approximation and we calculate the matrices ${\mathbf{F}}^{\theta}_{{\boldsymbol\upphi}}$ and ${\mathbf{P}}_0$ in , associated with the leading-order expressions for the dynamics and the mechanical energy expenditure. Then it only remains to compute the eigenvalues and eigenvectors of the matrix ${\mathbf{M}}$ in in order to assemble the expression of the optimal gait ${\boldsymbol\upphi}^*(t)$ in . This computation is done numerically in .
Purcell’s swimmer and $N$-link swimmers for small $N$ {#sec:Purcell}
-----------------------------------------------------
We use the small amplitude approximation and solve equation numerically to compute optimal strokes for Purcell’s three link swimmer and for other cases with small $N$. Purcell’s three-link swimmer is the minimal model of a linked microswimmer. We used the formula in order to compute the energy-optimal gait, which is shown in Fig. \[fig:optimal\_gait\_3link\] as a trajectory in the plane of joint angles $(\phi_1,\phi_2)$ for $\Delta x = 0.1l$. Snapshots of the swimmer’s configuration at quarter-period times are also illustrated on the plot, and we recover the key ingredient that enables the propulsion of this system, namely a phase shift between the two actuators.
The case of $N=5$ link swimmer is also computed and solutions are shown in Fig. \[Fig:5linksnap\]. Four snapshots of the swimmer obtained again at quarter-period times are presented corresponding to a sequence of shapes along the optimal stroke.
Both cases ($N=3$ and $N=5$) will be discussed further and compared with those obtained without assuming that the prescribed displacements are small and joint angles have small amplitude in Section \[sec:large\_amplitude\].
Optimal gaits for swimmers with many links
------------------------------------------
We now show results of optimal gaits for swimmers with $N=11$ and $N=101$, obtained by solving the eigenvalue problem . Figure \[fig:snapshot\_101&11\] shows a single snapshot of both swimmers under the optimal gait for a prescribed displacement $\Delta x= 0.1l$. The figure indicates that the optimal gaits look like a travelling wave. In order to test this observation quantitatively, we rewrite the optimal gaits using the sinusoidal representation , and compute the joint amplitudes $a_k$ and phase difference between consecutive joints $$\Delta p_k=p_{k+1}-p_k \mbox{ mod }[360^\circ]\,,
\label{eq:deltapk}$$ where $p_k$ are given by . The results of amplitudes and phase differences across the joints for both swimmers are plotted in Figures \[fig:amplitude101&11\] and \[fig:phase101&11\], respectively. Remarkably, the amplitudes $a_k$ for both swimmers display a nearly identical and slightly non-uniform symmetric distribution along the swimmer’s body, with a peak at the center (and up to $14\% $ decrease towards the ends), see Fig \[fig:amplitude101&11\]. The phase difference between joints is very close to a uniform value of $120^\circ$, indicating a travelling wave with wavelength of three links. This is further confirmed when computing the optimal gait for increasing numbers of $N$ links for $\Delta x=0.01l$. Figures \[fig:mean\_normalized\_amp\_vs\_n\] and \[fig:meanphase\_vs\_n\] plot the gait’s amplitude and phase difference, respectively, averaged across all swimmer’s joints. Note that for a swimmer with discrete links, a waveform with length of integer number of $m$ links corresponds to a phase difference of $\Delta \varphi=2 \pi /m$. Taking $m=1$ or $m=2$ results in time-reversible motion, hence the smallest possible integer is $m=3$. That is, the energy-optimal gait has the shortest possible wavelength. This result is in agreement with Taylor’s observation in [@Taylor51].
In hindsight, the emergence of traveling waves as optimal solutions for the control problem of a slender homogeneous swimmer whose shape is assumed to remain close to the rectilinear one, and swimming in a homogeneous fluid is not surprising. Consider for a moment the case in which the swimmer is of infinite length, but possesses a periodic shape (with period $Nl$). In particular, we look for strokes that are periodic in both space (with period $Nl$) and time (with period $T$). This system is translationally invariant with respect to shifts along the body axis. As a consequence, the initial problem , the optimal control problem , the governing operators, and the resulting solutions possess the same symmetry. In particular, this entails that the matrices ${\mathbf{P}}_0$ and ${\mathbf{F}}^{\theta}_{{\boldsymbol\upphi}}-{{\mathbf{F}}^{\theta}_{{\boldsymbol\upphi}}}^T$ (and therefore ${\mathbf{M}}$) are circulant matrices, i.e. matrices ${\mathbf{X}}$ that are constant along their diagonals and whose entries $X_{ij}$ only depend on $i-j$ mod $[N-1]$. Such matrices share the same eigenvectors that are given by ${\mathbf{v}}^\pm_k=\frac{1}{\sqrt{N-1}}(1,\omega^\pm_k,(\omega^\pm_k)^2,\cdots,(\omega^\pm_k)^{N-2})^T$ with $\omega^\pm_k=e^{\pm 2ik\pi/{(N-1)}}$ for $k=1,\cdots,(N-1)/2$. Using shows that the solution of the optimal control problem is therefore a traveling sinusoidal wave. In the finite length case, the same is true modulo edge-effects, as it is apparent from Figure \[Fig:11&101links\]. Notice that a similar result has been obtained in the different but related context of crawling of one-dimensional objects on solid surfaces or within a solid matrix [@Agostinelli], where the optimality of actuation strategies based on peristaltic waves is discussed.
[0.29]{} ![Optimal strokes for swimmers with $N=11$ (red) and $N=101$ (blue and blow-up) for $\Delta x=0.01 l$, where $l$ is the length of a single link. (a) A single snapshot of both swimmers during the optimal stroke. (b) Joint angle’s amplitudes $a_k$ vs. (scaled) joint number $k$. (c) Phase differences $\Delta p_k$, defined by , vs. (scaled) joint number $k$.[]{data-label="Fig:11&101links"}](snapshot_101_11 "fig:"){width="\textwidth"}
[0.33]{} ![Optimal strokes for swimmers with $N=11$ (red) and $N=101$ (blue and blow-up) for $\Delta x=0.01 l$, where $l$ is the length of a single link. (a) A single snapshot of both swimmers during the optimal stroke. (b) Joint angle’s amplitudes $a_k$ vs. (scaled) joint number $k$. (c) Phase differences $\Delta p_k$, defined by , vs. (scaled) joint number $k$.[]{data-label="Fig:11&101links"}](amplitude101_11 "fig:"){width="\textwidth"}
[0.33]{} ![Optimal strokes for swimmers with $N=11$ (red) and $N=101$ (blue and blow-up) for $\Delta x=0.01 l$, where $l$ is the length of a single link. (a) A single snapshot of both swimmers during the optimal stroke. (b) Joint angle’s amplitudes $a_k$ vs. (scaled) joint number $k$. (c) Phase differences $\Delta p_k$, defined by , vs. (scaled) joint number $k$.[]{data-label="Fig:11&101links"}](phase101_11 "fig:"){width="\textwidth"}
Finally, we fix the total length $L=Nl$ of a swimmer, compute the energy-optimal gait for moving a given displacement of $\Delta x = 0.01L$, and obtain the energy $E^*$ along this gait. Figure \[Fig:constant\_L\] shows a log-log plot of $E^*$ as a function of the links number $N$. Remarkably, it can be seen that for large $N$, the energy $E^*$ decays to zero as $1/N$. Analyzing this seemingly counter-intuitive behavior for large $N$ more closely (see Appendix) reveals that the optimal energy indeed scales as $E^* \sim \frac{\Delta x L^2}{N}$ for large $N$. This suggests that the a more suitable performance measure would be the scaled optimal energy defined as $Q=\frac{N E^*}{\Delta x L^2}$. This quantity, (multiplied by $\Delta x$ for better graphical visibility), is also plotted as a function of $N$ as the dashed line in Figure \[Fig:constant\_L\], which indicates that it converges to a finite nonzero value at the limit of large $N$.
[0.45]{} ![Optimal strokes for multi-link swimmers with large $N$ for $\Delta x=0.01 l$, where $l$ is the length of a single link. (a) Joint angle’s amplitude $a_k$ averaged across all joints, as a function of $N$. (b) Phase difference $\Delta p_k$ averaged across all joints, as a function of $N$. []{data-label="Fig:function_of_N"}](mean_normalized_amp_vs_n "fig:"){width="\textwidth"}
[0.45]{} ![Optimal strokes for multi-link swimmers with large $N$ for $\Delta x=0.01 l$, where $l$ is the length of a single link. (a) Joint angle’s amplitude $a_k$ averaged across all joints, as a function of $N$. (b) Phase difference $\Delta p_k$ averaged across all joints, as a function of $N$. []{data-label="Fig:function_of_N"}](meanphase_vs_n "fig:"){width="\textwidth"}
![Log-log plots of the optimal energy $E^*$ (solid line) and scaled energy $Q$ (dashed line) for $\Delta x=0.01L$ as a function of $N$. Here $L=Nl$ is the total length of the swimmer and the energy is normalized by $c_t L^3 / T$.[]{data-label="Fig:constant_L"}](energy_Q_vs_n){width="50.00000%"}
Optimal strokes of large amplitude {#sec:large_amplitude}
==================================
In this section we remove the assumption of small angle amplitudes, and consider the problem of determining energetically-optimal strokes when the prescribed displacement in one cycle is no longer small. At the present state of our knowledge, this can only be done numerically.
We compute numerically the optimal gaits by utilizing the toolbox of optimal control, which uses direct optimization methods by discretizing times and states [@bonnans2012Bocop]. applies numerical integration of the full nonlinear dynamic equations and the energy formula , without assuming small-amplitude strokes. We solve the problem for large prescribed displacements. In addition, we consider the behavior of the optimal solutions when the prescribed displacements become smaller and smaller, in order to obtain an independent check of our analysis of Section \[sec:small\_amplitude\], which is based on leading-order approximation when imposed displacements and angle amplitudes are small. The discretized nonlinear optimization problem is solved by the solver [@wachter2006implementation] with [@amestoy2001fully], while the derivatives are computed by sparse automatic differentiation with [@bell2012cppad]. In the numerical experiments, we used a midpoint (implicit 1st order) discretization with 100 time steps.
Optimal gaits for Purcell’s three-link swimmer {#sec:Purcell_large_amplitude}
----------------------------------------------
The optimal gaits computed within the framework of the small angle amplitudes of Section \[sec:Purcell\] are shown in Fig. \[fig:optimal\_gait\_3link\] as a trajectory in the plane of joint angles $(\phi_1,\phi_2)$ for $\Delta x = 0.1l$. Snapshots of the swimmer’s configuration at quarter-period intervals are also illustrated on the plot.
For comparison, we computed energy-optimal gaits using for different displacements $\Delta x$, and the resulting trajectories scaled by $\sqrt{\Delta x}$ are shown in Fig. \[fig:Bocop\_compare\_3link\]. It can be seen from the plot that for small displacements $\Delta x$ the optimal gait obtained by converges (after scaling) to the one obtained by the formula . Note that we did not constrain the initial conditions of the swimmer in this computation with , nor required any symmetry relations of the gait. The only constraint is on zero net rotation and net displacement of magnitude $\Delta x$, without a specified direction. The agreement between the two methods of computation confirms the validity of our small-amplitude analysis. When the displacement $\Delta x$ is increased further, the energy-optimal gaits obtained with begin to deviate significantly from the small-amplitude one. For $\Delta x = 0.26$, the optimal gait (dashed) coincides with the gait obtained by Tam and Hosoi [@TamHosoi2007] that maximizes Lighthill’s efficiency. This was calculated by considering the first two terms in a Fourier series expansion of the two joint angles. Notably, the long axis of this ellipse-like-shaped gait is “skewed” with respect to the long axis of the exact ellipse representing our small-amplitude energy-optimal gait. The reason for this fundamental difference is the fact that optimization of Lighthill’s efficiency as in [@TamHosoi2007] does not involve a constraint on the travelled distance. Therefore, it produces large displacements through large amplitude strokes, as discussed at the beginning of Section \[sec:small\_amplitude\].
When $\Delta x$ is further increased to an upper limit of $\Delta x = 0.306$, the optimal gait (dash-dotted) deforms into the famous peanut-shaped loop obtained in [@TamHosoi2007] as the maximal-displacement gait. Note that for large displacements, we had to add constraints on symmetries of the gait and initial conditions, otherwise began to search for different gaits which are not “simple loops”.
We now further analyze the motion of Purcell’s swimmer using leading-order terms as studied in [@wiezel2016optimization]. The sinusoidal gait in for the two joint angles can be rewritten as $$\label{eq.gait_phase}
\phi_1(t)={\varepsilon}\sin(t+\varphi/2) \, \;\; \phi_2(t)={\varepsilon}\sin(t-\varphi/2).$$ That is, it has a stroke amplitude of ${\varepsilon}$ and phase difference of $\varphi$. Using resistive force theory and the leading order expansion in ${\varepsilon}$, as explained in [@wiezel2016optimization], the leading-order expressions for displacement $\Delta x$ and energy $E$ in a cycle under the gait are obtained as: $$\label{eq.leadXE}
\Delta x=\frac{5 \pi}{81}{\varepsilon}^2 l \sin(\varphi) , \;\; E=\frac{2\pi c_t}{81}{\varepsilon}^2 (11\cos(\varphi)+16) \; .$$ Therefore, for sinusoidal gaits of the form , our constrained optimization of minimizing the energy to cover a given displacement reduces to minimizing $E / \Delta x$. This gives an optimization problem for a scalar function of $ \varphi$. Using elementary calculus, the minimum of this function is obtained at the optimal phase difference of $\varphi^*=\cos^{-1}(-11/16)=133.43^\circ$, and the resulting optimal gait is exactly identical to the one obtained by using the eigenvalue formulation in , which is shown in Fig. \[fig:optimal\_gait\_3link\].
[0.35]{} ![Optimal strokes for Purcell’s swimmer, plotted as loops in the plane of joint angles $(\phi_1,\phi_2)$. (a) Optimal stroke obtained by our small-amplitude analysis, with snapshots of the swimmer’s configurations. (b) Comparison with optimal strokes computed using for different displacements $\Delta x$, angles $\phi_i$ are scaled by $\sqrt{\Delta x}$.[]{data-label="Fig:3link_optimal_strokes"}](optimal_gait_3link "fig:"){width="\textwidth"}
[0.4]{} ![Optimal strokes for Purcell’s swimmer, plotted as loops in the plane of joint angles $(\phi_1,\phi_2)$. (a) Optimal stroke obtained by our small-amplitude analysis, with snapshots of the swimmer’s configurations. (b) Comparison with optimal strokes computed using for different displacements $\Delta x$, angles $\phi_i$ are scaled by $\sqrt{\Delta x}$.[]{data-label="Fig:3link_optimal_strokes"}](Bocop_compare_3link "fig:"){width="\textwidth"}
Optimal gaits for 5-link swimmer
--------------------------------
We now show a comparison between our analytical formulation of optimal gaits in and numerical optimization using for the five-link swimmer, whose space of joint angles is four-dimensional. Snapshots at quarter-period times of the swimmer’s configuration for the analytical optimal gait corresponding to $\Delta x=0.1l$ are shown in Figure \[Fig:5linksnap\] and compared to the numerical solutions obtained for $\Delta x=l$. According to our small-amplitude analysis, the optimal gait is planar (it lies in the two-dimensional linear subspace $S$ spanned by eigenvectors associated with the pair of imaginary eigenvalues of ${\mathbf{M}}$ corresponding to the maximal magnitude $\mu_{\scriptscriptstyle M}$.) In order to compare further our analytical optimal gaits with the results of computations, we first plot in Figure \[fig:Bocop\_compare\_5link\] the projections onto the planar subspace $S$ of the four-dimensional optimal gaits. These optimal gaits were numerically computed by for different values of $\Delta x$ and then scaled by $\sqrt{\Delta x}$ for comparison with the analytical optimal gaits. In order to test the theoretical prediction that optimal gaits should lie within the two-dimensional subspace $S$, we computed the maximal Euclidean distance $d$ of each optimal gait in $\mathbb{R}^4$ from the plane $S$. Figure \[fig:distance\_from\_plane\] shows a log-log plot of this distance $d$ as a function of $\Delta x$. It can be seen that for small displacements this distance decays to zero as $(\Delta x)^{3/2}$, indicating that the optimal gaits obtained numerically using are indeed converging to planar loops lying within $S$, in agreement with the prediction of our asymptotic analysis.
[0.24]{} ![Snapshots of five-link swimmer during the optimal stroke. (a)-(d): analytical solutions for $\Delta x=0.1l$; (e)-(h): numerical solutions for $\Delta x=l$.[]{data-label="Fig:5linksnap"}](5linksnap1 "fig:"){width="\textwidth"}
[0.24]{} ![Snapshots of five-link swimmer during the optimal stroke. (a)-(d): analytical solutions for $\Delta x=0.1l$; (e)-(h): numerical solutions for $\Delta x=l$.[]{data-label="Fig:5linksnap"}](5linksnap2 "fig:"){width="\textwidth"}
[0.24]{} ![Snapshots of five-link swimmer during the optimal stroke. (a)-(d): analytical solutions for $\Delta x=0.1l$; (e)-(h): numerical solutions for $\Delta x=l$.[]{data-label="Fig:5linksnap"}](5linksnap3 "fig:"){width="\textwidth"}
[0.24]{} ![Snapshots of five-link swimmer during the optimal stroke. (a)-(d): analytical solutions for $\Delta x=0.1l$; (e)-(h): numerical solutions for $\Delta x=l$.[]{data-label="Fig:5linksnap"}](5linksnap4 "fig:"){width="\textwidth"}
[0.24]{} ![Snapshots of five-link swimmer during the optimal stroke. (a)-(d): analytical solutions for $\Delta x=0.1l$; (e)-(h): numerical solutions for $\Delta x=l$.[]{data-label="Fig:5linksnap"}](5linksnap_large_amp1 "fig:"){width="\textwidth"}
[0.24]{} ![Snapshots of five-link swimmer during the optimal stroke. (a)-(d): analytical solutions for $\Delta x=0.1l$; (e)-(h): numerical solutions for $\Delta x=l$.[]{data-label="Fig:5linksnap"}](5linksnap_large_amp2 "fig:"){width="\textwidth"}
[0.24]{} ![Snapshots of five-link swimmer during the optimal stroke. (a)-(d): analytical solutions for $\Delta x=0.1l$; (e)-(h): numerical solutions for $\Delta x=l$.[]{data-label="Fig:5linksnap"}](5linksnap_large_amp3 "fig:"){width="\textwidth"}
[0.24]{} ![Snapshots of five-link swimmer during the optimal stroke. (a)-(d): analytical solutions for $\Delta x=0.1l$; (e)-(h): numerical solutions for $\Delta x=l$.[]{data-label="Fig:5linksnap"}](5linksnap_large_amp4 "fig:"){width="\textwidth"}
[0.45]{} ![Comparison of the small-amplitude optimal stroke of the five-link swimmer with computations for different displacements $\Delta x$. (a) Projection of the strokes onto the plane $S$, angles $\phi_i$ are scaled by $\sqrt{\Delta x}$ ($V1,V2 \in \mathbb{R}^4$ are two orthonormal basis vectors spanning the plane $S$). (b) Log-log plot of the maximal distance $d$ of optimal strokes obtained with from the plane $S$, as a function of displacement $\Delta x$.[]{data-label="Fig:5link_optimal_strokes"}](projection_on_plane "fig:"){width="\textwidth"}
[0.45]{} ![Comparison of the small-amplitude optimal stroke of the five-link swimmer with computations for different displacements $\Delta x$. (a) Projection of the strokes onto the plane $S$, angles $\phi_i$ are scaled by $\sqrt{\Delta x}$ ($V1,V2 \in \mathbb{R}^4$ are two orthonormal basis vectors spanning the plane $S$). (b) Log-log plot of the maximal distance $d$ of optimal strokes obtained with from the plane $S$, as a function of displacement $\Delta x$.[]{data-label="Fig:5link_optimal_strokes"}](distance_from_plane "fig:"){width="\textwidth"}
Moving to the discussion of the optimal gaits of large amplitude, the fact that we are considering a system whose shape space is four-dimensional enables us to address and explore the deviations from planarity of the optimal loops. The computed optimal strokes are shown in the bottom panel of Fig. \[Fig:5linksnap\] and in Fig. \[Fig:5link\_optimal\_strokes\_projected\], where we plot several different projections, by plotting the joint evolution of several pairs of shape variables. While, in the regime of small displacements, we do recover the elliptical loops predicted by our theory, the shape of the strokes for large displacements shows large discrepancies, and the geometry of the strokes is difficult to understand because of their high-dimensionality. One notable feature of Fig. \[Fig:5linksnap\] is that, in the large angle amplitude regime, optimal strokes consist of undulating shapes (wave-forms) such that the distance between successive peaks is of the order of the whole swimmer length, see Fig. \[fig:5linksnap\_large\_amp4\], as it is in fact observed in biological systems such as sperm cells [@Gaffney2011; @Gaffney2017].
[0.31]{} ![Optimal strokes of the five-link swimmer obtained with computations for different displacements $\Delta x$. (a) $\phi_1$ versus $\phi_2$. (b) $\phi_2$ versus $\phi_3$. (c) $\phi_1$ versus $\phi_4$.[]{data-label="Fig:5link_optimal_strokes_projected"}](5linkphi1phi2 "fig:"){width="100.00000%"}
[0.31]{} ![Optimal strokes of the five-link swimmer obtained with computations for different displacements $\Delta x$. (a) $\phi_1$ versus $\phi_2$. (b) $\phi_2$ versus $\phi_3$. (c) $\phi_1$ versus $\phi_4$.[]{data-label="Fig:5link_optimal_strokes_projected"}](5linkphi2phi3 "fig:"){width="\textwidth"}
[0.31]{} ![Optimal strokes of the five-link swimmer obtained with computations for different displacements $\Delta x$. (a) $\phi_1$ versus $\phi_2$. (b) $\phi_2$ versus $\phi_3$. (c) $\phi_1$ versus $\phi_4$.[]{data-label="Fig:5link_optimal_strokes_projected"}](5linkphi1phi4 "fig:"){width="\textwidth"}
In order to give a concrete and visual representation of the emergence of non-planarity when angles and displacements become large, we derive from the computed four-dimensional loops a dimensionally-reduced representation in three dimensions, using the approach [@isomap; @ArroyoDeSimone12]. Figure \[fig:isomap\] shows clearly that the loops are non-planar, and that interesting features of their two-dimensional projections such as cusps and crossings are really an outcome of the projection of higher dimensional, non self-intersecting curves. This interesting behavior is entirely new with respect to the results currently available in the literature for the $N=3$ case of Purcell’s swimmer whose shape space is two-dimensional, and whose optimal strokes are necessarily planar curves. In addition, we see clearly that, as the imposed displacement becomes small, the loops converge to the planar ellipses of the small amplitude approximation of Section \[sec:small\_amplitude\].
![Three-dimensional representation of the optimal loops for the 5-link-swimmer as a function of the amplitude of the prescribed displacement $\Delta x$, in normalized coordinates $(\pi_1,\pi_2,\pi_3)$. The 3D picture is produced using the approach and the optimal strokes for different prescribed displacements are plotted in black. It is therefore a three dimensional non-linear projection of the origianl angles in 4D. As the displacement goes to 0, the strokes converge to a planar ellipse, as the one obtained in the small displacement limit. In turn, when the prescribed displacement increases, the strokes are non-planar and their geometry is complex.[]{data-label="fig:isomap"}](5link3D){width="\textwidth"}
These results open the new and unexplored question of characterizing the geometry of optimal loops for swimmers with shape space of large dimensions, when the amplitude of the joint angles is allowed to become large. Conducting numerical computation with for a number of links $N>5$ is currently beyond its limitations of memory allocation. There is however no such limitations for the solution of the small amplitude version of the optimal control problem presented here.
Conclusion
==========
In this work, we have studied optimal periodic strokes of multi-link micro-swimmers that achieve a given prescribed displacement in a given time with minimum energy. Exploiting the linearity of Stokes flows and geometric symmetries of the swimmer, the optimization has been formulated as a constrained variational problem, where leading-order expansion of the dynamics leads to the optimal solution of an eigenvalue problem. Remarkably, it is proven that energy-optimal strokes for $N$-link swimmers reduce to ellipses lying in a two-dimensional subspace of the space of shapes.
For large $N$, the optimal strokes become traveling waves with the shortest possible wavelength of three links, in agreement with the observation made in Taylor’s classic work [@Taylor51]. A noticeable difference from [@Taylor51] due to the finite length of the swimmer is the slightly non-uniform distribution of energy-optimal wave amplitudes, which decay symmetrically from swimmer’s center towards its ends.
Numerical optimization for the fully nonlinear problem, in which the amplitude of the prescribed displacements and the excursions of the joint angles are allowed to be large, is obtained using the optimization toolbox for multi-link swimmers with three and five links. When the prescribed displacements are small, the numerical results show excellent agreement with those obtained using the small amplitude approximation. When the imposed displacements are large, the picture changes significantly. For Purcell’s 3-link swimmer, we obtain non-convex closed curves similar to the ones previously obtained by Tam and Hosoi for maximizing displacement or Lighthill’s efficiency. For the 5-link swimmer, we obtain non-planar loops of complex and intriguing geometry. A more precise characterization of the properties of these highly-dimensional closed loops seems an interesting open problem. For both $N=3$ and $N=5$, all these complex shapes converge to the planar ellipses of the small amplitude approximation when the prescribed displacements become small.
Possible directions for future extension of this research are comparison with measured strokes of biological swimming microorganisms as in [@berman2013undulatory; @Rossi2017Kinematics], generalization of the model to include elasticity as in [@Montino; @spagnolie2010optimal; @passov2012dynamics; @cicconofri2016motion], and the study of optimal control of magnetically-actuated microswimmers [@alouges2015soft].
Acknowledgements {#acknowledgements .unnumbered}
================
OW wishes to thank the Technion-France Association for their support in his research internship visit to Paris in 2015, in which this work has been initiated. We gratefully acknowledge support by the Israel Science Foundation (OW and YO, grant no. 567/14) and by the European Research Council (ADS, grant no. 340685-MicroMotility). We all wish to warmly thank Dr. Pierre Martinon for his help and technical support with the software.
Appendix - scaling laws for large N {#appendix---scaling-laws-for-large-n .unnumbered}
===================================
In order to study scaling of the displacement and energy for large $N$, we now consider a swimmer with $N$ links of fixed lengths $l=1$, under the gait of travelling wave: $$\label{eq.wave}
\phi_k(t)={\varepsilon}\sin(\omega t + k \Delta \varphi) \mbox{, for } k=1 \ldots N \mbox{, where } \omega = 2 \pi / T.$$ We consider gaits with fixed amplitude ${\varepsilon}$ and phase difference of $\Delta \varphi=2 \pi /3$. Using our formulation based on resistive-force theory , we numerically calculate the displacement $\Delta x$ and energy expenditure $E$ along a cycle, as a function of $N$. The results are shown on log-log scale plots in Figures \[fig:apndx\_x\_n\_const\_l\] and \[fig:apndx\_E\_n\_const\_l\]. It can be seen that for large $N$, the displacement $\Delta x$ converges to a constant. This is analogous to the “infinite sheet” limit of Taylor’s net swimming speed [@Taylor51]. In addition, $\Delta x$ is linear in $l$ and scales quadratically with the stroke amplitude ${\varepsilon}$ as in [@Taylor51]. The energy expenditure $E$, however, grows linearly with $N$. This is because for large $N$, the interaction between links becomes negligible and each link contributes an equal amount of viscous dissipation. In addition, computation for fixed $N$ and varying links’ length $l$ reveals that the energy expenditure scales cubically, as $l^3$ (plot not shown). This is explained by the following observation. The mechanical power dissipation of each link scales as $P_i=f_i v_i$ where $f_i$ is the viscous drag force and $v_i$ is the link’s linear velocity. The viscous force $f_i$ scales as $f_i=R v_i$ where $R$ is a viscous resistance coefficient (cf. [@BrennerHappel65; @cox1970; @GrayHancock55]). The link’s velocity scales as $v_i \sim {\varepsilon}l /T$, while the resistance coefficient $R$ scales linearly with the link’s length $l$. Since total energy dissipation scales as $N P_i T$, we deduce the following scaling laws for large $N$: $$\label{eq.scaling}
\Delta x \sim {\varepsilon}^2 l, \;\;\; E \sim {\varepsilon}^2 N l^3 /T .$$ Consider now the case where the required displacement $\Delta x$ as well as the *total length* of the swimmer $L=Nl$, are held fixed while $N$ is varied. That is, the links’ length scales as $l=L/N$. The scaling laws in will now change to $$\label{eq.scalingL}
\Delta x \sim {\varepsilon}^2 \frac{L}{N}, \;\;\; E \sim {\varepsilon}^2 \frac{L^3}{N^2 T} .$$ Fixing the displacement $\Delta x$, it is deduced from that for large $N$, the amplitude ${\varepsilon}$ increases as $\sqrt{\frac{N\Delta x}{L}}$. Moreover, the energy scales as $E \sim \frac{\Delta x L^2}{NT}$. This scaling relation explains the decay rate of $E^*$ in Figure \[Fig:constant\_L\].
[0.45]{} ![Log-log plots of (a) displacement $\Delta x$ and (b) total energy E as a function of links number $N$, under the traveling-wave input for a chain of $N$ links with equal lengths $l=1$.[]{data-label="Fig:function_of_N_const_l"}](apndx_x_n_const_l "fig:"){width="\textwidth"}
[0.45]{} ![Log-log plots of (a) displacement $\Delta x$ and (b) total energy E as a function of links number $N$, under the traveling-wave input for a chain of $N$ links with equal lengths $l=1$.[]{data-label="Fig:function_of_N_const_l"}](apndx_E_n_const_l "fig:"){width="\textwidth"}
[10]{}
G. I. Taylor, “Analysis of the swimming of microscopic organisms,” [ *Proceedings of the Royal Society of London A: Mathematical, Physical and Engineering Sciences*]{}, vol. 209, pp. 447–461, 1951.
J. Lighthill, “Mathematical biofluiddynamic,” [*Society for Industrial and Applied Mathematics, Philadelphia, PA.*]{}, 1975.
E. M. Purcell, “Life at low [R]{}eynolds number,” [*Am. J. Phys*]{}, vol. 45, no. 1, pp. 3–11, 1977.
J. Gray and J. Hancock, “The propulsion of sea-urchin spermatozoa,” [*J. Exp. Biol.*]{}, vol. 32, no. 4, pp. 802–814, 1955.
R. D. Dresdner and D. F. Katz, “Relationships of mammalian sperm motility and morphology to hydrodynamic aspects of cell function,” [*Biology of Reproduction*]{}, vol. 25, no. 5, pp. 920–930, 1981.
E. Gaffney, H. Gadêlha, D. Smith, J. Blake, and J. Kirkman-Brown, “Mammalian sperm motility: Observation and theory,” [*Annual Review of Fluid Mechanics*]{}, vol. 43, no. 1, pp. 501–528, 2011.
K. Ishimoto, H. Gadêlha, E. Gaffney, D. Smith, and J. Kirkman-Brown, “Coarse-graining the fluid flow around a human sperm,” [*Phys. Rev. Lett.*]{}, vol. 118, no. 12, pp. 124501, 2017.
R. Dreyfus, J. Baudry, M. L. Roper, M. Fermigier, H. A. Stone, and J. Bibette, “Microscopic artificial swimmers,” [*Nature*]{}, vol. 437, no. 7060, pp. 862–865, 2005.
J. Feng and S. Cho, “Mini and micro propulsion for medical swimmers,” [ *Micromachines*]{}, vol. 5, pp. 97–113, 2014.
F. Alouges, A. DeSimone, L. Giraldi, and M. Zoppello, “Can magnetic multilayers propel artificial microswimmers mimicking sperm cells?,” [ *Soft Robotics*]{}, vol. 2, pp. 117–128, 2015.
E. Lauga and T. R. Powers, “The hydrodynamics of swimming microorganisms,” [*Reports on Progress in Physics*]{}, vol. 72, no. 9, p. 096601, 2009.
J. Guasto, R. Rusconi, and R. Stoker, “Fluid mechanics of planktonic microorganisms,” [*Annu. Rev. Fluid Mech.*]{}, vol. 44, pp. 373–400, 2012.
R. E. Goldstein, “Green algae as model organisms for biological fluid dynamics,” [*Annu. Rev. Fluid Mech.*]{}, vol. 47, pp. 343–375, 2015.
S. Childress, [*Mechanics of Swimming and Flying*]{}. Cambridge, U.K.: Cambridge University Press, 1981.
B. Bonnard, M. Chyba, and J. Rouot, [*Geometric and Numerical Optimal Control*]{}. Springer Brief in Mathematics, 2018.
H. Brenner and J. Happel, [*Low [R]{}eynolds number hydrodynamics: with special applications to particulate media.*]{} Springer, 1965.
J. R. Blake, “A spherical envelope approach to ciliary propulsion,” [ *Journal of Fluid Mechanics*]{}, vol. 46, no. 1, p. 199?208, 1971.
P. Satir, and M. A. Sleigh, “The physiology of cilia and muco-ciliary interactions,” [*Annual Review of Physiology*]{}, vol. 52, no. 1, pp. 137–155, 1990. PMID: 2184754.
L. Feriani, M. Juenet, C. Fowler, N. Bruot, M. Chioccioli, S. Holland, C. Bryant, and P. Cicuta, “Assessing the collective dynamics of motile cilia in cultures of human airway cells,” [*Biophys. J.*]{}, vol. 113, pp. 109–119, 2017.
L. E. Becker, S. A. Koehler, and H. A. Stone, “On self-propulsion of micro-machines at low [R]{}eynolds number: [P]{}urcell’s three-link swimmer,” [*Journal of Fluid Mechanics*]{}, vol. 490, pp. 15–35, 2003.
A. Najafi and R. Golestanian, “Simple swimmer at low [R]{}eynolds number: [T]{}hree linked spheres,” [*Physical Review E*]{}, vol. 69, no. 6, p. 062901, 2004.
H. Ghosch and P. Fischer, “Controlled propulsion of artificial magnetic nanostructured propellers,” [*Nano Letters*]{}, vol. 9, pp. 2243–2245, 2009.
W. Hu, G. Lum, M. Mastrangeli, and M. Sitti, “Small-scale soft-bodied robot with multimodal locomotion,” [*Nature*]{}, vol. 554, pp. 81–85, 2018.
O. Pironneau and D. F. Katz, “Optimal swimming of flagellated micro-organisms,” [*Journal of Fluid Mechanics*]{}, vol. 66, pp. 391–415, 1974.
R. S. Berman, O. Kenneth, J. Sznitman, and A. M. Leshansky, “Undulatory locomotion of finite filaments: lessons from [C]{}aenorhabditis elegans,” [ *New Journal of Physics*]{}, vol. 15, no. 7, p. 075022, 2013.
F. Montenegro-Johnson and E. Lauga, “Optimal swimming of a sheet,” [ *Physical Review E*]{}, vol. 89, p. 060701(R), 2014.
D. Tam and A. E. Hosoi, “Optimal feeding and swimming gaits of biflagellated organisms,” [*Proc. Nat. Acad. Sciences USA*]{}, vol. 108, pp. 1001–1006, 2011.
D. Tam and A. E. Hosoi, “Optimal stroke patterns for [P]{}urcell’s three-link swimmer,” [*Physical Review Letters*]{}, vol. 98, p. 068105, 2007.
F. Alouges, A. DeSimone, L. Heltai, A. Lefebvre, and B. Merlet, “Optimally swimming [S]{}tokesian robots,” [*Discrete and Continuous Dynamical Systems B*]{}, vol. 18, pp. 1189–1215, 2013.
F. Alouges, A. DeSimone, and A. Lefebvre, “Optimal strokes for low [R]{}eynolds number swimmers: an example,” [*Journal of Nonlinear Science*]{}, vol. 18, pp. 277–302, 2008.
F. Alouges, A. DeSimone, and L. Heltai, “Numerical strategies for stroke optimisation of axisymmetric micro-swimmers,” [*Mathematical Models and Methods in Applied Sciences*]{}, vol. 21, pp. 361–387, 2011.
O. Wiezel and Y. Or, “Using optimal control to obtain maximum displacement gait for (p)urcell’s three-link swimmer,” in [*2016 IEEE 55th Conference on Decision and Control (CDC)*]{}, pp. 4463–4468, Dec 2016.
N. Giuliani, N. Heltai, and A. DeSimone, “Predicting and optimizing micro-swimmer performance from the hydrodynamics of its components: the relevance of interactions,” [*Soft Robotics*]{}, 2018.
L. Burton, R. Hatton, D. Smith, H. Choset, and A. Hosoi, “Two-link swimming using buoyant orientation,” [*Physics of Fluids*]{}, vol. 22, p. 091703, 2010.
S. E. Spagnolie and E. Lauga, “The optimal elastic flagellum,” [*Physics of Fluids*]{}, vol. 22, no. 3, p. 455, 2010.
E. Passov (Gutman) and Y. Or, “Dynamics of [P]{}urcell’s three-link microswimmer with a passive elastic tail,” [*The European Physical Journal E*]{}, vol. 35, no. 8, pp. 1–9, 2012.
A. Montino and A. DeSimone, “Three-sphere low-[R]{}eynolds-number swimmer with a passive elastic arm,” [*The European Physical Journal E*]{}, vol. 38, no. 42, pp. 1–10, 2015.
G. Cicconofri and A. DeSimone, “Motion planning and motility maps for flagellar microswimmers,” [*The European Physical Journal E*]{}, vol. 39, no. 7, p. 72, 2016.
K. Ishimoto and E. Gaffney, “An elastohydrodynamical simulation study of filament and spermatozoan swimming driven by internal couples,” [*IMA J. Appl. Math.*]{}, vol. 83, pp. 655–679, 2018.
O. Raz and J. Avron, “Comment on ‘[O]{}ptimal stroke patterns for [P]{}urcell’s three-link swimmer’,” [*Phys. Rev. Lett.*]{}, vol. 100, p. 029801, 2008.
L. Giraldi, P. Martinon, and M. Zoppello, “Optimal design of [P]{}urcell’s three-link swimmer,” [*Physical Review E*]{}, vol. 91, no. 2, p. 023012, 2015.
O. Wiezel and Y. Or, “Optimization and small-amplitude analysis of [P]{}urcell’s three-link microswimmer model,” in [*Proc. R. Soc. A*]{}, vol. 472, p. 20160425, The Royal Society, 2016.
F. Alouges and G. Di Fratta, “Parking 3-sphere swimmer. [I]{}. [E]{}nergy minimizing strokes,” Disc. & Cont. Dyn. Syst. - B, vol. 23, no. 4, p. 1797–1817, 2018.
D. Agostinelli, F. Alouges, and A. DeSimone, “Peristaltic waves as optimal gaits in metameric bio-inspired robots,” [*Frontiers in Robotics and AI*]{}, vol. 5, p. 99, 2018.
F. Bonnans, P. Martinon, and V. Gr[é]{}lard, [*Bocop - A collection of examples*]{}. PhD thesis, INRIA, 2012.
F. Alouges, A. DeSimone, L. Giraldi, and M. Zoppello, “Self-propulsion of slender micro-swimmers by curvature control: $n$-link swimmers,” [ *International Journal of Non-Linear Mechanics*]{}, vol. 56, pp. 132–141, 2013.
R. G. Cox, “The motion of long slender bodies in a viscous fluid part 1. [G]{}eneral theory,” [*Journal of Fluid Mechanics*]{}, vol. 44, no. 4, p. 791?810, 1970.
L. Giraldi, P. Martinon, and M. Zoppello, “Controllability and optimal strokes for [N]{}-link microswimmer,” in [*52nd IEEE Conference on Decision and Control (CDC)*]{}, pp. 3870–3875, Dec 2013.
J. E. Avron and O. Raz, “A geometric theory of swimming: [P]{}urcell’s swimmer and its symmetrized cousin,” [*New Journal of Physics*]{}, vol. 10, no. 6, p. 063016, 2008.
E. Gutman and Y. Or, “Symmetries and gaits for [P]{}urcell’s three-link microswimmer model,” [*IEEE Transaction on Robotics*]{}, vol. 32, no. 1, pp. 53–69, 2016.
J. Koiller, K. Ehlers, and R. Montgomery, “Problems and progress in microswimming,” [*Journal of Nonlinear Science*]{}, vol. 6, pp. 507–541, Nov 1996.
A. W[ä]{}chter and L. T. Biegler, “On the implementation of an interior-point filter line-search algorithm for large-scale nonlinear programming,” [ *Mathematical Programming*]{}, vol. 106, no. 1, pp. 25–57, 2006.
P. R. Amestoy, I. S. Duff, J.-Y. L’Excellent, and J. Koster, “A fully asynchronous multifrontal solver using distributed dynamic scheduling,” [ *SIAM Journal on Matrix Analysis and Applications*]{}, vol. 23, no. 1, pp. 15–41, 2001.
B. M. Bell, “Cppad: a package for c++ algorithmic differentiation,” [ *Computational Infrastructure for Operations Research. http://www.coin-or.org/CppAD*]{}.
J. Tenenbaum, V. Silva, and J. Langford, “A global geometric framework for non- linear dimensionality reduction,” [*Science*]{}, vol. 290, pp. 2319–2323, 2000.
M. Arroyo, D. Milan, L. Heltai, and A. DeSimone, “Reverse engineering the euglenoid movement,” [*Proc. Nat. Acad. Sciences USA*]{}, vol. 109, pp. 17874–17879, 2012.
M. Rossi, G. Cicconofri, A. Beran, G. Noselli, and A. De[S]{}imone, “Kinematics of flagellar swimming in [E]{}uglena gracilis: [H]{}elical trajectories and flagellar shapes,” [*Proc. Nat. Acad. Sciences USA*]{}, vol. 114, no. 50, pp. 13085–13090, 2017.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We present the STARS library, a grid of tidal disruption event (TDE) simulations interpolated to provide the mass fallback rate ($dM/dt$) to the black hole for a main-sequence star of any stellar mass, stellar age, and impact parameter. We use a 1D stellar evolution code to construct stars with accurate stellar structures and chemical abundances, then perform tidal disruption simulations in a 3D adaptive-mesh hydrodynamics code with a Helmholtz equation of state, in unprecedented resolution: from 131 to 524 cells across the diameter of the star. The interpolated library of fallback rates is publicly available[^1] and one can query the library for any stellar mass, stellar age, and impact parameter. We provide new fitting formulae for important disruption quantities ($\beta_{\rm crit}, \Delta M, \dot M_{\rm peak}, t_{\rm peak}, n_\infty$) as a function of stellar mass, stellar age, and impact parameter. Each of these quantities vary significantly with stellar mass and stellar age, but we are able to reduce all of our simulations to a single relationship that depends only on stellar structure, characterized by a single parameter $\rho_c/\bar\rho$, and impact parameter $\beta$. We also find that in general, more centrally concentrated stars have steeper $dM/dt$ rise slopes and shallower decay slopes. For the same $\Delta M$, the shape of the $dM/dt$ varies significantly with stellar mass, promising the potential determination of stellar properties from the TDE light curve alone. The shape of the fallback-rate curves depends strongly on stellar structure and to a certain extent stellar mass, meaning that fitting TDEs using this library offers a better opportunity to determine the nature of the disrupted star and the black hole.'
author:
- 'Jamie A.P. Law-Smith'
- 'David A. Coulter'
- James Guillochon
- Brenna Mockler
- 'Enrico Ramirez-Ruiz'
bibliography:
- 'export-bibtex.bib'
title: 'Stellar TDEs with Abundances and Realistic Structures (STARS): Library of Fallback Rates'
---
Introduction {#sec:introduction}
============
Near the region of influence of a galactic massive black hole (BH), a star scattered onto an orbit that brings it close to the BH can be tidally disrupted by the BH’s gravitational field—this is a tidal disruption event (TDE). The stellar material that remains bound to the BH produces a flare that is the signature of this event. TDEs were first explored theoretically with pioneering studies by @1975Natur.254..295H, , @1988Natur.333..523R, and @1989ApJ...346L..13E.
Several dozen such flares have been observed at the centers of other galaxies [for reviews of observations, see e.g. @2015JHEAp...7..148K; @2017ApJ...838..149A; @2017ApJ...842...29H; @2020arXiv200101409V], with observations now regularly capturing both the rise [e.g., @2019ApJ...883..111H] and decay [e.g., @2018MNRAS.480.5689H; @2019ApJ...878...82V] of the transient in great detail, and even signatures of an accretion disk [e.g., @2019ApJ...880..120H; @2020arXiv200309427H]. Fitting theoretical models to observed TDEs allows one to extract the properties of the disruption: BH mass, BH spin, stellar mass, stellar age, impact parameter, and radiative efficiency. The first attempt to systematically extract BH masses from TDEs [@2019ApJ...872..151M] was remarkably successful, obtaining errors of order that of the $M$–$\sigma$ relationship. However, determination of other parameters is made difficult by degeneracies between stellar properties and BH properties. One needs better theoretical models of TDEs in order to extract more accurate information from observed events.
A combination of detailed theoretical modeling and high resolution observations can turn TDEs into unique tools to probe several astrophysical questions: (1) the BH mass function and in particular the possible existence of a cutoff in the BH mass function at low masses ($M_{\rm BH} \lesssim 10^5 M_\sun$), (2) the BH spin distribution, (3) the radiative efficiency of BH accretion and other questions of accretion physics, (4) the stellar populations (stellar masses and ages) in galactic centers, as the stars at the centers of distant galaxies are exposed through their disruption and accretion, (5) the dynamics operating in galactic centers; e.g., which mechanisms (two-body, resonant relaxation, secular effects, etc.) dominate how stars and BHs interact.
At the order-of-magnitude level, tidal disruption occurs when a star crosses the tidal disruption radius $$r_{\rm t}=\left(\frac{M_{\rm BH}}{M_\star}\right)^{1/3} R_\star \propto \bar\rho_\star^{-1/3},
\label{eq:rt}$$ at which point a star’s self-gravity is smaller than the tidal acceleration across its radius. The pericenter passage time of the star is approximately equal to the star’s dynamical time, $t_{\rm p} \sim r_{\rm p} / v_{\rm p} \sim \sqrt{R_\star^3/GM_\star} = t_{\rm dyn, \star}$. The star is spun-up to a large fraction of its breakup angular velocity, and a quadrupole tidal distortion develops across its surface. These two effects are what tidally disrupt a star, and (again, at the order-of-magnitude level) they occur over the star’s dynamical timescale, so the star does not have time to react hydrodynamically. The impact parameter of the encounter $$\beta \equiv \frac{r_{\rm t}}{r_{\rm p}}$$ is the ratio of the tidal radius to the pericenter distance; it is an order-of-magnitude measure of the “strength” of the tidal interaction. The critical impact parameter $\beta_{\rm crit}$ is defined as the smallest impact parameter of full disruption (i.e., where the entire star’s mass is disrupted), below which are partial disruptions and above which are “post-critical” encounters.
The mass fallback rate $dM/dt \equiv \dot M_{\rm fb}$ of debris to pericenter is a central quantity of interest as it appears to track the luminosity evolution of observed TDEs closely [e.g., @2012Natur.485..217G; @2014ApJ...783...23G; @2019ApJ...872..151M]. For non-relativistic disruptions, one can scale mass fallback rate and time of return to pericenter with BH mass as follows: $$\begin{aligned}
\dot M_{\rm peak} &\propto M_{\rm BH}^{-1/2}\ M_\star^{2}\ R_\star^{-3/2} \\
t_{\rm peak} &\propto M_{\rm BH}^{1/2}\ M_\star^{-1}\ R_\star^{3/2}
\label{eq:mdotpeaktpeakscalings}\end{aligned}$$ where the $\dot M_\text{peak} \propto M_\star^{2}$ scaling results when we assume that a constant fraction of the star’s mass is lost to the BH in the disruption. Note that these formulae are for the peak quantities of the mass fallback rate, but can be applied to scale the entire $dM/dt$ curve.
Because the tidal radius depends inversely on the average density of the star (Eq. \[eq:rt\]), there is a maximum BH mass for disruption outside the innermost-bound spherical orbit for different types of stars. See Figure 1 of @2017ApJ...841..132L for the phase space in $M_{\rm object}$ and $M_{\rm BH}$ of stellar objects (from planets to evolved stars) that are expected to produce bright tidal disruption flares. We expect that most observed TDEs will be from MS stars, and that the mass function of TDEs is relatively flat for $M_\star \lesssim M_\sun$ [@2016MNRAS.461..371K] (see more detailed discussion in Section \[sec:conclusion\]). Thus, it is important to have a library of MS star simulations with which to fit observed events. This work is aimed to be the definitive library of the MS star parameter space. One can extend this library to include relativistic encounters, but these are rarer, and account for a small fraction of MS star disruptions (see discussion in Section \[sec:conclusion\]).
Our own Galactic Center’s nuclear star cluster is host to young, massive stars, giant-branch stars, and an old population of main sequence stars . Stars in the Galactic Center also exhibit a range of metallicities, from metal-rich to metal-poor [@2017MNRAS.464..194F]. In nuclear clusters outside our galaxy, there is evidence for a diversity of stellar ages and types [@2010ApJ...714..713S] and a wide range of star formation histories [@2014MNRAS.441.3570G]. Additionally, TDEs appear to be observed preferentially in post-starburst galaxies [@2014ApJ...793...38A; @2016ApJ...818L..21F; @2017ApJ...850...22L; @2018ApJ...853...39G]. Thus, we expect some TDEs to be sourced by more massive stars and also by stars of varying ages and compositions. Observations of TDEs suggest that many disruptees are non-ZAMS stars, as is expected—spectra of TDEs show metal lines that evolve with time [e.g., @2019ApJ...887..218L]. However, the theoretical modelling of TDEs thus far has largely ignored the stellar evolution aspect of the problem.
A brief (incomplete) summary of recent theoretical work, where we highlight features relevant to or differing in significant ways from this work, is below. Before discussing the simulation work, we mention the so-called “frozen-in” approximation, studied in detail by @2009MNRAS.392..332L, in which the star arrives intact to pericenter. One can then integrate across the star in equal-orbital-energy slices and determine the mass fallback rate to the BH as a function of time analytically. In this framework, only the structure of the star matters. This approximation can only handle full disruptions (a single impact parameter). @2016MNRAS.458..127K investigated abundance anomalies using this framework and 1D stellar evolution models from MESA. @2018ApJ...857..109G, following the above two works, developed an analytic framework to study the composition as a function of time following disruption, using MESA stars. In a recent significant extension to @2009MNRAS.392..332L, @2019ApJ...883L..17C developed an analytic framework to determine the asymptotic power-law slope of the fallback rate, finding $\propto t^{-5/3}$ for full disruptions and $\propto t^{-9/4}$ for partial disruptions, in general (but not detailed) agreement with simulations.
Despite the success and usefulness of the “frozen-in” approximation in calculating TDE properties, we have learned from hydrodynamical simulations that the star arrives at pericenter significantly distorted and spinning [for a detailed discussion see @2019MNRAS.485L.146S]. @1982ApJ...263..377N performed the first 3D tidal disruption simulations, of a $\gamma=5/3$ polytropic stellar structure. @1993ApJ...418..163K [@1993ApJ...418..181K] were the first to study stars with different stellar structures ($\gamma=(5/3, 1.5, 4/3)$) numerically. @2009ApJ...697L..77R studied the tidal disruption and initial disk formation for $\gamma=5/3$ and $\gamma=1.4$ polytropes in 3D hydrodynamics. @2013ApJ...767...25G performed the first systematic parameter-space study of the effect of both stellar structure and impact parameter, exploring $\gamma=4/3$ and $\gamma=5/3$ polytropes and a wide range of impact parameters. The effect of impact parameter made it clear that tidal disruptions are a 3-body nonlinear hydrodynamical problem that needs to be studied by simulations, at least to a certain extent. @1993ApJ...410L..83L were the first to explore relativistic tidal disruptions, on a Schwarzchild metric. @1997ApJ...479..164D studied the disruption of a $\gamma=5/3$ polytrope by a rotating BH on a Kerr metric. @2012ApJ...749..117H simulated the disruption of a white dwarf by a spinning BH. @2014PhRvD..90f4020C and @2017MNRAS.469.4483T compared relativistic simulations to Newtonian simulations in detail, finding good agreement for non-relativistic encounters. @2019MNRAS.487.4790G performed a grid of general-relativistic simulations for a $\gamma=5/3$ polytrope, exploring higher impact parameters and spinning BH’s, providing new fitting formulae for the relativistic regime, and again finding good agreement for non-relativistic encounters. See also @2019GReGr..51...30S for a review of the status of TDEs in general relativity. @2008ApJ...679.1385R [@2008CoPhC.179..184R; @2009ApJ...695..404R] studied the tidal disruption of white dwarfs in detail for the first time, while @2012ApJ...757..134M were the first to study the tidal disruption of giant stars, whose highly segregated density profiles did not allow them to be fully disrupted. @2017ApJ...841..132L performed a case-study simulating realistic stellar structures and compositions, with a hydrogen-envelope helium-core white dwarf. @2019ApJ...872..163G studied the stellar spin dependence of fallback rates for a $\gamma=5/3$ polytrope. @2019MNRAS.487..981G ran moving-mesh simulations of a ZAMS 1$M_\sun$ star with MESA stellar structure. Their results were consistent with the $\gamma=4/3$ result from @2013ApJ...767...25G. @2019ApJ...882L..26G ran simulations with MESA stellar structures, for three stellar masses and ages, at one impact parameter $\beta=3$, and argued that the inferred BH mass from fitting TDEs to polytropic hydrodynamical simulations can be incorrect at the order-of-magnitude level. @2019ApJ...882L..25L performed simulations with MESA stellar structures and a Helmholtz EOS, tracking chemical abundances for 49 elements in the 3D hydrodynamical simulations for the first time. They found significant differences with the polytropic results of @2013ApJ...767...25G and also significant differences with the analytic predictions of the fallback-rate composition of @2018ApJ...857..109G. @2020arXiv200103501R [@2020arXiv200103502R; @2020arXiv200103503R; @2020arXiv200103504R] recently posted results of a parameter-space study in a general-relativistic framework, for a range of stellar masses, impact parameters, and BH masses, and at a single stellar age.
A number of theoretical studies have focused on the disk formation process [e.g., @2014ApJ...783...23G; @2015ApJ...804...85S; @2015ApJ...806..164P; @2015ApJ...809..166G; @2016MNRAS.461.3760H; @2016MNRAS.455.2253B; @2020MNRAS.492..686L; @2020MNRAS.495.1374B]. At present, the emission mechanism(s) responsible for the Optical/UV, X-ray, and Radio components of TDE flares is an open question—the main candidates are (1) rapid disk circularization and accretion emission or (2) stream self-intersections and disk-formation-process emission—, but it is clear that the luminosity evolution of TDEs (the “light curve”) tracks the mass fallback rate to the BH ($dM/dt$, the main output of the simulations in this work) very closely (see references above). This means that the fallback rate, which, as we show in this paper, is nearly solely determined by stellar structure, is a useful output.
In this paper, we study the disruption of main-sequence (MS) stars in a grid of stellar mass, stellar age, and impact parameter. Previously published systematic studies have used polytropic stellar structures, where $P=K \rho^\gamma = K \rho^{(n+1)/n}$. This work is a parameter space study using realistic internal stellar structures, chemical abundances, and equations of state (EOS). We significantly expand upon the stellar structure study of @2013ApJ...767...25G by considering a wide range of density profiles derived from more accurate stellar models (as compared to polytropes): we study 14 distinct stellar structures, corresponding to different stellar masses and ages. The EOS is incorporated via the Helmholtz EOS, consistent with the MESA EOS of the stellar models. We find (see Appendix) that the EOS contribution to the pressure support is small, and we argue that one can predict many of the properties of tidal disruption from stellar structure and impact parameter alone. In order to reduce our simulations into one relationship for various tidal disruption quantities, we parameterize the stellar structure by the single parameter $\rho_c/\bar\rho$, the ratio of the star’s central density to its average density. The simulations presented in this work have a full Helmholtz EOS tracking 49 elements; however, we study the chemical abundance of the fallback debris in a followup paper.
This paper is organized as follows: Section \[sec:methods\] describes our methods, Section \[sec:results\] describes our results, Section \[sec:conclusion\] concludes, and our Appendix describes the interpolated `STARS_library` package, as well as several other issues.
Methods {#sec:methods}
=======
We use the 1D stellar evolution code MESA [@2011ApJS..192....3P] to run a grid of models from 0.1$M_\sun$ to 10$M_\sun$, from pre-MS to zero-age main-sequence (ZAMS) to terminal-age main-sequence (TAMS). We define TAMS as a central hydrogen mass fraction of $10^{-3}$. We use the `mesa_49.net` nuclear network, including 49 elements. See also @2018ApJ...857..109G and @2019ApJ...882L..25L for details on the MESA setup. Table \[tab:MESA\_params\] in the Appendix lists relevant or non-standard parameters for the MESA simulations. Stars with $M_\star \lesssim 0.8 M_\sun$ have a MS lifetime longer than the age of the universe; thus, for these stars, the oldest model we use in our tidal disruption calculations is at 10 Gyr, rather than TAMS.
We then map the 1D stellar density profiles and chemical abundances into the 3D adaptive-mesh refinement (AMR) hydrodynamics code FLASH [@2000ApJS..131..273F]. We use the Helmholtz EOS and an extended Helmholtz table[^2] that spans $10^{-12} \leq \rho\ {\rm [g/cm^3]} \leq 10^{15}$ and $10^3 \leq T\ {\rm [K]} \leq 10^{13}$. This Helmholtz EOS is the backbone of the EOS module in MESA, so our mapping is self-consistent. In FLASH, we use the sinks module to integrate trajectories and to track the position of the BH relative to the star, as in @2017ApJ...834L..19G. This is also an update from @2013ApJ...767...25G. In the multipole gravity solver, we use a maximum angular number of the multipole expansion of $l_m=20$. Our 3D box is 1000$R_\star$ on a side and we set the background density to $10^{-11}$ g/cm$^3$. This is such that the mass of the box is typically $\ll 0.1$% of the stellar mass. We use the following hydrodynamics parameters: the interpolation order is 3rd order, the slope limiter is the “hybrid” one, and we use the hybrid Riemann solver. We have no magnetic fields ($B=0$). As in @2013ApJ...767...25G, in calculating the binding energy of each cell in the simulation with respect to the BH and stellar debris, we shift the center of mass of the star to be on a parabolic orbit (this is equivalent to centering the $dM/de$ distribution at $e=0$). @2013ApJ...767...25G show that, because of the magnitude of this shift, it only affects the $dM/dt$ for $t>100$ yrs. For the vast majority of events, the star is expected to approach on a parabolic or nearly-parabolic orbit [@2018ApJ...855..129H]. See also @2009ApJ...705..844G, @2011ApJ...732...74G, @2013ApJ...767...25G, and @2019ApJ...882L..25L for details on the FLASH setup. Table \[tab:FLASH\_params\] in the Appendix lists several additional FLASH parameters, including the background grid values.
The simulations in this paper are run with $1.5\times10^8$ maximum cells. This is a higher maximum cell count than in @2019ApJ...882L..25L. We choose the maximum initial refinement based on the central concentration of the stellar density profile. We use 131 cells across the diameter of the star for less centrally concentrated stars ($\rho_c/\bar\rho \lesssim 150$) and 524 cells across the diameter of the star for more centrally concentrated stars ($\rho_c/\bar\rho \gtrsim 150$). In a few test simulations, results are nearly identical if we use 524 cells across the diameter of the star for the less centrally concentrated stars as well. For comparison, the simulations in @2013ApJ...767...25G had $\approx$50 cells across the initial diameter, so our simulations have a factor of 2.6X to 10.5X higher initial resolution. The moving-mesh simulations of @2019MNRAS.487..981G have a maximum of $2.4\times10^5$ cells, thus initially $\approx$60 cells across the diameter of the star. The simulations of @2020arXiv200103501R [@2020arXiv200103502R; @2020arXiv200103503R; @2020arXiv200103504R] initially have $\approx$50 cells across the diameter of the star and, in an important difference, the final box size is $17\times9\times14 R_\star$.
[l l l l l l l]{} 0.1 $M_\sun$ & 0.1214 $R_\sun$ & 0 Gyr & 5.5 & 213 s & 0.5–1.2$^\ast$ & 25–10\
& 0.1215 $R_\sun$ & 10 Gyr & 5.5 & 213 s & 0.5–1.2$^\ast$ & 25–10\
0.3 $M_\sun$ & 0.2814 $R_\sun$ & 0 Gyr & 5.8 & 434 s & 0.6, 0.7, 0.8, 0.9, 1.0, 2.0 & 33, 28, 25, 22, 20, 10\
& 0.2989 $R_\sun$ & 10 Gyr & 5.8 & 475 s & 0.6, 0.7, 0.8, 0.9, 1.0, 2.1 & 35, 30, 26, 23, 21, 10\
0.5 $M_\sun$ & 0.4452 $R_\sun$ & 0 Gyr & 11 & 669 s & 0.6, 0.8, 1.0, 1.15, 1.4, 2.6 & 44, 33, 26, 23, 19, 10\
& 0.4564 $R_\sun$ & 10 Gyr & 12 & 715 s & 0.6, 0.8, 1.0, 1.2, 1.4, 2.8 & 46, 35, 28, 23, 20, 10\
0.7 $M_\sun$ & 0.6485 $R_\sun$ & 0 Gyr & 23 & 994 s & 0.75, 1.0, 1.25, 1.5, 3.4 & 46, 34, 28, 23, 10\
& 0.6793 $R_\sun$ & 10 Gyr & 36 & 1065 s & 0.8, 1.0, 1.15, 1.3, 1.5, 1.7, 3.6 & 45, 36, 31, 28, 24, 21, 10\
1.0 $M_\sun$ & 0.9012 $R_\sun$ & 0 Gyr & 42 & 1362 s & 1.0, 1.25, 1.5, 1.75, 2.0, 4.2 & 42, 34, 28, 24, 21, 10\
& 1.0455 $R_\sun$ & 4.8 Gyr & 138 & 1702 s & 1.0, 1.5, 2.0, 2.5, 3.0, 3.5, 4.9 & 49, 33, 24, 20, 16, 14, 10\
& 1.2872 $R_\sun$ & 8.4 Gyr & 756 & 2325 s & 1.0, 1.5, 2.0, 3.0, 4.0, 4.5, 5.0, 6.0 & 60, 40, 30, 20, 15, 13, 12, 10\
1.5 $M_\sun$ & 1.6275 $R_\sun$ & 0 Gyr & 128 & 2699 s & 1.0, 1.5, 2.0, 2.75, 6.7 & 67, 51, 34, 24, 10\
& 2.0805 $R_\sun$ & 2 Gyr & 1697 & 3901 s & 2.0, 4.0, 6.0, 8.6 & 43, 21, 14, 10\
3.0 $M_\sun$ & 1.8896 $R_\sun$ & 0 Gyr & 73 & 2387 s & 0.6–4.0$^\ast$ & 103–15\
& 3.3192 $R_\sun$ & 0.3 Gyr & 1198 & 5558 s & 1.5, 2.0, 3.0, 4.0, 4.5, 5.0, 7.0, 10.8 & 72, 54, 36, 27, 24, 22, 16, 10\
10 $M_\sun$ & 3.6870 $R_\sun$ & 0 Gyr & 38 & 3564 s & 1.0–4.2$^\dagger$ & 80–19\
& 8.4232 $R_\sun$ & 0.02 Gyr& 1292 & 12307 s& 1.5–10.8$^\dagger$ & 123–17\
Table \[tab:grid\] lists the parameter space in stellar mass, stellar age, and impact parameter studied in this work. For each star, we run a range of impact parameters $\beta$ (the ratio of the tidal radius to the pericenter distance) from grazing partial disruptions to post-critical disruptions. The lowest $\beta$ for a given star corresponds to 1% to 10% mass lost (unbound) from the star in the encounter. We run one very-post-critical (post-full-disruption) $\beta$ for each star such that $r_{\rm p}=10r_{\rm g}$ for this highest $\beta$ encounter, where $r_{\rm g}\equiv GM/c^2$ in this paper. This is for the purpose of interpolating our grid of $dM/dt$ curves in the maximum applicable range. We also plan to extend this library with relativistic simulations (applicable to the small fraction of very relativistic encounters) in future work. For comparison, the most relativistic encounter in @2013ApJ...767...25G was the $\beta=4$ for the $\gamma=4/3$ star, which was $r_{\rm p}=11.8 r_{\rm g}$.
Table \[tab:grid\] also lists several other quantities, such as the ratio of the star’s central density to average density $\rho_c/\bar\rho$ (a parameterization of the central concentration of the star that we use extensively later on), the dynamical time of the star, defined in this paper as $t_{\rm dyn} \equiv \sqrt{R_\star^3 / GM_\star}$, and the ratio of pericenter distance to gravitational radius, $r_{\rm p}/r_{\rm g}$, corresponding to each $\beta$. In order to extend the range of the interpolated fallback rate library (see Appendix), we include a few results scaled from simulations where the stellar structures are nearly identical. We scale $\gamma=5/3$ results for the 0.1$M_\sun$ stars, $\gamma=4/3$ results for the ZAMS 3$M_\sun$ star, our ZAMS 1$M_\sun$ results for the ZAMS 10$M_\sun$ star, and our TAMS 3$M_\sun$ results for the TAMS 10$M_\sun$ star. Note that we do perform a 10$M_\sun$, $\beta=1.5$ simulation in Section \[sec:stellar\_structure\_vs\_EOS\].
We run at a single BH mass of $M_{\rm BH} = 10^6 M_\sun$. Our simulations are directly applicable to any non-relativistic encounter with a different BH mass because of the scaling of disruption quantities with $M_{\rm BH}$ in Eq. (\[eq:mdotpeaktpeakscalings\]). See Figure \[fig:mbh\_beta\_phase\_space\] in the Appendix for the range of applicability of our simulations. This grid of simulations applies to tidal disruptions from most stellar masses, ages, impact parameters, and BH masses. For $r_{\rm p}>10 r_{\rm g}$, the difference in $dM/dt$’s between Newtonian and relativistic simulations is $\lesssim10\%$ (see more detailed discussion in Section \[sec:conclusion\]).
At the last timestep before debris begins to leave the computational domain, we calculate the specific binding energy of every cell in the simulation relative to the BH and the star (if any self-bound mass remains), including only material bound to the BH and excluding material bound to the star or unbound from the BH. From this spread in specific binding energy distribution, or $dM/de$, we calculate the mass return rate to the BH (more accurately, to pericenter) as a function of time, using Kepler’s third law: $$\frac{dM}{dt}=\frac{dM}{de}\frac{de}{dt}=\frac{dM}{de}\frac{1}{3} (2 \pi G M_{\rm BH})^{2/3} t^{-5/3}.$$
We run the simulations to a maximum time of 100$t_{\rm dyn}$ and the star reaches pericenter at approximately 20$t_{\rm dyn}$ into the simulation. For moderate $\beta$’s, the debris remains inside the box for 100$t_{\rm dyn}$, but for high $\beta$’s the debris begins to leave the box earlier—for the most extreme $\beta$’s we study this is at approximately 50$t_{\rm dyn}$. So $dM/de$ distributions are calculated at 30–80$t_{\rm dyn}$ after pericenter. @2013ApJ...767...25G studied fallback rates at up to 550$t_{\rm dyn}$ after pericenter, showing that the resulting shape is identical to those calculated at earlier times as we do. By not having a large enough box to follow the entire debris for $t>100 t_{\rm dyn}$, we do not sacrifice any accuracy in the final result, but instead gain precision as a smaller box allows for higher resolution. Note that however, unlike in @2020arXiv200103501R [@2020arXiv200103502R; @2020arXiv200103503R; @2020arXiv200103504R], where debris leaves the computational domain after a few dynamical timescales, the debris in our simulations remains in the box for many dynamical timescales after pericenter.
As an example of our analysis method, Figure \[fig:dmde\] shows the specific binding energy distribution and resulting mass return rate to the BH for a $0.3M_\sun$ ZAMS star in a $\beta=0.9$ encounter with a $10^6 M_\sun$ BH. $dM/de$ is plotted in units of $M_\star/\Delta e$, where $\Delta e= G M_\star^{2/3} M_{\rm BH}^{1/3} / R_\star$, an order-of-magnitude estimate of the range in fluid binding energies [@2013MNRAS.435.1809S]. The $x$-axis is normalized similarly. The hydrodynamical grid data from the simulation is binned and then fit with a B-spline. The $dM/dt$ is extended by finding the average slope of the last 10–20% of the $dM/dt$. The extended section is not visible on this plot as it is for $t\gtrsim10^4$ s. The result for a $\gamma=5/3$ polytrope, for this $\beta$, scaled to the same mass and radius as this star, is also shown. It is expected to match quite closely as the stellar structure of a $0.3 M_\sun$ ZAMS star is well approximated by a $\gamma=5/3$ polytrope (see Figure \[fig:MESA\_profiles\_vs\_polytropes\_norm\] in the Appendix).
The small differences between the polytropic and MESA initial condition simulations are likely due to differences in resolution, numerical method, and smoothing algorithm between the two works. The B-splines applied to smooth the $dM/de$ distributions in this work have been examined and calibrated in detail for the $dM/de$ result from each simulation, and so the resulting $dM/dt$ curves in this work have higher fidelity to the raw hydrodynamical grid data.
Results {#sec:results}
=======
Stellar structure
-----------------
Figure \[fig:structure\] shows the ratio of central density to average density $\rho_c/\bar\rho$ for MS stars, interpolated based on a finely-spaced MESA grid (more finely spaced than for our TDE simulations; Table \[tab:grid\]). This shows the range of stellar structures on the main sequence. We perform TDE simulations in FLASH for structures at the extremes and interpolate the resulting $dM/dt$’s in stellar mass and stellar age in between these extremes. For some regions one can use polytropic stellar structures rather than MESA initial conditions (e.g., for $M_\star \lesssim 0.3 M_\sun$).
The top panels show $\rho_c/\bar\rho$ in the space of stellar age vs. stellar mass. Here one sees that lower-mass stars ($\lesssim 0.8 M_\sun$), whose main-sequence lifetimes are $\gtrsim$ the age of the universe, evolve slowly and have roughly constant stellar structures over 10 Gyr. More massive stars ($\gtrsim 0.8 M_\sun$) have shorter lifetimes and evolve through different stellar structures more rapidly. We expect a wide variety of stellar masses and stellar ages to source observed TDEs; see Section \[sec:conclusion\] for discussion.
The bottom panels show normalized density profiles colored by $\rho_c/\bar\rho$, both individually for the stars we simulate in FLASH and interpolated on a more finely-spaced grid of MESA stars. One can see that $\rho_c/\bar\rho$ maps nearly 1-1 onto the stellar density profile, at least for the main sequence. This makes it a very good single-value parameter to describe the stellar structure, and we use it to reduce our simulations into a single relationship and to provide fitting formulae for each TDE quantity.
For comparison, for giant stars, $\rho_c/\bar\rho \gtrsim 10^6$. In @2012ApJ...757..134M, due to numerical limitations, the authors use $\rho_{\rm core}/\bar\rho \approx 2\times10^3$ for the hydrodynamical simulations. They are unable to fully disrupt the giant stars, due to the highly segregated density profile (much more centrally concentrated than that of MS stars despite the artificial value of $\rho_{\rm core}/\bar\rho$) and extended envelope structure.
Figure \[fig:MESA\_profiles\_vs\_polytropes\_norm\] in the Appendix shows density profiles for each star we simulate as compared to $\gamma=5/3$ and $\gamma=4/3$ stellar structures. Also see Figure 1 in @2019ApJ...882L..25L for the stellar structure evolution of a $1M_\sun$ and $3M_\sun$ star over their MS lifetimes (also compared to polytropic stellar profiles). Only a few stars correspond closely to polytropic stellar structures; for all others, MESA initial conditions are significantly more accurate.
Qualitative tidal disruption results
------------------------------------
As an example of one of our FLASH simulations, Figure \[fig:vol\_rend\] shows a volume rendering of the disruption of a $1M_\sun$ ZAMS star at $t-t_{\rm p} \approx 3 t_{\rm dyn}$ in a $\beta=1$ encounter with a $10^6 M_\sun$ BH (where $t_{\rm p}$ is the pericenter time and $t_{\rm dyn}$ is the dynamical time of the star). Color corresponds to density. One can see the stratified structure following the disruption of a star with an accurate internal stellar structure. This encounter is a partial disruption in which a core survives. Note that this snapshot is zoomed in on the star in order to highlight the density structure of the debris a few dynamical times after pericenter, but that our computational domain is 1000$R_\star$ on a side (roughly 100 times the size of this volume rendering), and the debris eventually expands to fill this.
As an example of the range of vulnerability to tidal disruption of our grid, Figure \[fig:butterfly\] shows 2D density slices of the disruption of several different stars at different impact parameters, all at $\approx 2 t_{\rm dyn}$ after pericenter. The axes are arranged such that $\beta$ increases from left to right, and central concentration ($\rho_c/\bar\rho$) increases from top to bottom. The white contours correspond to absolute values of density (1, $10^{-1}$, and $10^{-2}$ g/cm$^3$), illustrating the different stellar structures and also the amount the different layers of the star are spun up. Increasing $\beta$ both increasingly distorts the star and spins it up. The more centrally concentrated stars have “layers” that are more differentiated. This leads to the outer layers being torqued more than the inner layers, and the core remaining sequestered and undisturbed at higher $\beta$’s for more centrally concentrated stars. Increasing central concentration allows the star to survive higher-$\beta$ encounters. For example, a ZAMS 0.3$M_\sun$ star has a larger fraction of its mass at larger radii, and is thus fully disrupted by a $\beta=1$ encounter, whereas a TAMS 1$M_\sun$ star has only a small fraction of its mass at larger radii from its sequestered core, and is thus relatively undisturbed by a $\beta=1$ encounter.
Mass lost
---------
Figure \[fig:deltaM\_vs\_beta\] shows fractional mass lost from the star $\Delta M/M_\star$ as a function of impact parameter $\beta$. The $x$-axis in the top left panel is the raw $\beta$ and the $x$-axis in the top right panel is normalized to the critical $\beta$ for full disruption for each star. Generally, more massive stars and stars further along in their MS evolution are more centrally concentrated, and thus must be disrupted deeper relative to their nominal “tidal radii” in order to lose the same amount of fractional mass. One can see that the mass-loss prescriptions for $\gamma=5/3$ and $\gamma=4/3$ polytropes are inadequate to describe the more centrally concentrated stars, which have critical impact parameters of $\beta_{\rm crit}>2$.
The bottom panels have a scaled $x$-axis, $$x = \exp \left[ \left( \beta/\beta_{\rm crit} \right)^\alpha -1 \right],\ \ \ \alpha = (\rho_c/\bar\rho)^{-1/3},$$ constructed in order to reduce all of the simulations into a single relationship. This formula accounts for the dependence of $\Delta M$ vs. $\beta$ on stellar structure. The fact that we are able to express all of the simulations in a single relationship implies that $\Delta M$ depends only on $\rho_c/\bar\rho$ and $\beta$. Analytic and B-spline fits for this relation, allowing one to obtain the mass lost for any stellar mass, stellar age, and impact parameter, are provided in the Appendix. The simulations in this work all reduce to a single relation for other disruption quantities as well (see below in this section).
Critical impact parameter
-------------------------
We did not sample finely enough in $\beta$ to find the exact critical impact parameter for full disruption for each star, but we estimate this within the bounds associated with our spacing in $\beta$ and the mass lost as a function of $\beta$ for the star. Figure \[fig:beta\_crit\_vs\_Mstar\] shows the critical $\beta$ as a function of $\rho_c/\bar\rho$. For nearly all stars, the critical $\beta$ is well-fit by a simple relation, $$\beta_{\rm crit} \approx 0.52 \left(\frac{\rho_c}{\bar\rho} \right)^{1/3},\ \ \ \rho_c/\bar\rho \lesssim 500.
\label{eq:beta_crit}$$ At the highest central concentrations ($\rho_c/\bar\rho \gtrsim 500$), the critical $\beta$ is higher than predicted with this relation, and is instead better fit by a steeper relation, $$\beta_{\rm crit} \approx 0.39 \left(\frac{\rho_c}{\bar\rho} \right)^{1/2.3},\ \ \ \rho_c/\bar\rho \gtrsim 500.
\label{eq:beta_crit2}$$ This allows one to predict the critical impact parameter for full disruption for any main-sequence star. All one needs is $\rho_c/\bar\rho$ for that star, obtained from, e.g., a MESA model directly or pre-computed grids such as the MIST models [@2016ApJ...823..102C]. So the ability to fully disrupt a star is a simple function of this ratio of densities.
Note that for 2 points at the highest central concentrations, the quoted $\beta_{\rm crit}$ is obtained through extrapolation, as the maximum $\beta$ simulated for that star did not fully disrupt it. The lower limits obtained from our simulations are also shown. We note also that the resolution (in term of maximum number of cells in the simulation) required to precisely determine $\beta_{\rm crit}$ becomes significantly higher for these most centrally concentrated stars. At the end of the simulation when the debris has expanded, and thus the linear resolution has decreased from the initial maximum level of refinement (in order to resolve the same number of cells in the simulation box), the highest density “core” of the star—for the stars with $\rho_c/\bar\rho \gtrsim 10^3$—becomes smaller than a grid cell and so is not resolved. However, the aim of this study is not to precisely determine the critical impact parameter for these most centrally concentrated stars, but rather to determine the mass fallback rate to the BH following their disruption.
Table \[tab:beta\_crit\] lists the critical impact parameter $\beta_{\rm crit}$ for full disruption for all stars studied. Errors represent the grid spacing and so are overestimates. We also list the corresponding physical pericenter distance in gravitational radii $r_{\rm p}/r_{\rm g}$. For stars that do not evolve significantly in structure over 10 Gyr, their $\beta_{\rm crit}$’s remain the same or very similar. For stars that evolve significantly over $<10$ Gyr, the $\beta_{\rm crit}$ increases dramatically over a star’s lifetime. For example, for a 1$M_\sun$ star, the $\beta_{\rm crit}$ increases from $\approx$1.8 to $\approx$7.0 over its main-sequence lifetime, a result of its $\rho_c/\bar\rho$ increasing from 42 to 756. This corresponds to the critical $r_{\rm p}/r_{\rm g}$ for full disruption decreasing by a factor of 2.7X, from 23.4 to 8.6. Thus, a TAMS 1$M_\sun$ star must approach 2.7X closer the BH in order to be fully disrupted compared to its ZAMS self. Similar trends are seen for the 1.5$M_\sun$ and 3$M_\sun$ stars, where the critical $r_{\rm p}/r_{\rm g}$ decreases by a factor $>$2X from ZAMS to TAMS. Thus, for higher-mass stars ($M_\star \gtrsim 0.8 M_\sun$), at the same pericenter distance, the ZAMS star is roughly 2X as vulnerable to tidal disruption and associated mass loss as the TAMS star. For lower-mass stars ($M_\star \lesssim 0.8 M_\sun$), the critical $r_{\rm p}/r_{\rm g}$ is roughly constant over 10 Gyr. See the Appendix for the dependence of $\beta_{\rm crit}$ and the associated $r_{\rm p}/r_{\rm g}$ with stellar mass.
[l l l l l]{} 0.3 $M_\sun$ & 0.2814 $R_\sun$ & 0 Gyr & $0.9 \pm 0.1$ & 22.0\
& 0.2989 $R_\sun$ & 10 Gyr & $0.9 \pm 0.1$ & 23.4\
0.5 $M_\sun$ & 0.4452 $R_\sun$ & 0 Gyr & $1.1 \pm 0.1$ & 24.0\
& 0.4564 $R_\sun$ & 10 Gyr & $1.1 \pm 0.1$ & 24.7\
0.7 $M_\sun$ & 0.6485 $R_\sun$ & 0 Gyr & $1.5 \pm 0.2$ & 23.2\
& 0.6793 $R_\sun$ & 10 Gyr & $1.6 \pm 0.1$ & 22.1\
1.0 $M_\sun$ & 0.9012 $R_\sun$ & 0 Gyr & $1.8 \pm 0.1$ & 23.4\
& 1.0455 $R_\sun$ & 4.8 Gyr & $2.7 \pm 0.2$ & 18.1\
& 1.2872 $R_\sun$ & 8.4 Gyr & $7.0^\dagger$ & 8.6\
1.5 $M_\sun$ & 1.6275 $R_\sun$ & 0 Gyr & $2.85 \pm 0.2$ & 23.5\
& 2.0805 $R_\sun$ & 2 Gyr & $10^\dagger$ & 8.6\
3.0 $M_\sun$ & 1.8896 $R_\sun$ & 0 Gyr & $2.0^\ast \pm 0.25$ & 30.8\
& 3.3192 $R_\sun$ & 0.3 Gyr & $8.5 \pm 1.5$ & 12.7\
As $\rho_c/\bar\rho$ increases, the star’s expansion in response to mass loss decreases. As a result, the star becomes less vulnerable to tidal dissipation and thus the critical impact parameter for full disruption monotonically increases with increasing $\rho_c/\bar\rho$. For example, note the transition in critical impact parameter from $\beta_{\rm crit} = 0.9 < 1$ for the 0.3$M_\sun$ star to $\beta_{\rm crit} = 1.1 > 1$ for the 0.5$M_\sun$ star. The 0.3$M_\sun$ star corresponds to a $\gamma=5/3$ polytrope (see Figure \[fig:MESA\_profiles\_vs\_polytropes\_norm\]) and the 0.5$M_\sun$ star has an intermediate structure in between a $\gamma=5/3$ and $\gamma=4/3$ polytrope. In the most extreme cases, for example for giant stars [@2012ApJ...757..134M] with a sequestered core-envelope structure, the remaining envelope contracts and thus its disruption requires increasingly higher $\beta$’s for the same amount of mass loss (in fact, these stars are unable to be fully disrupted by massive BHs).
Mass fallback rates
-------------------
### All encounters
Figure \[fig:mdots\_allgrouped\] shows mass fallback rates $dM/dt$ for all of our simulations, grouped by star. Each panel is a particular stellar mass and age, and shows all impact parameters $\beta$ for that object. One can perhaps pick out trends in the shape of the $dM/dt$ curves that we will explore in more detail below. Higher $\beta$ generally corresponds to shifting the $dM/dt$ curve upwards and to the left, giving a shorter peak timescale and higher peak fallback rate. After the critical $\beta$ for full disruption, the $dM/dt$ curve flattens, particularly near the peak. The rise and decay slopes also vary with stellar mass, stellar age, and impact parameter.
The top panels of Figure \[fig:mdots\_allone\] show $dM/dt$’s for a single stellar mass, 1$M_\sun$, for all $\beta$’s and stellar ages, in order to demonstrate the effect of stellar age. Note that the $\beta$’s are not the same for each stellar age (see Table \[tab:grid\] or Figure \[fig:mdots\_allgrouped\]). The left panel shows the raw curves and the right panel is normalized to the peak fallback rate ($\dot M_{\rm peak}$) and the associated peak fallback time ($t_{\rm peak}$) in order to focus on the rise and decay slopes. In the left panel, the curves largely overlap, demonstrating the degeneracy between stellar structure and $\beta$. This is why there is a degeneracy in the `MOSFiT` TDE fitting between stellar mass and impact parameter (and so efficiency) [@2019ApJ...872..151M]. This is similar to the shifting of $dM/dt$’s from $\gamma=5/3$ and $\gamma=4/3$ polytropes to lie on top of each other demonstrated in @2013ApJ...767...25G. In the right panel, one sees that the rise and decay slopes also largely overlap for a single stellar mass at multiple ages.
Thus, it is more challenging to identify the age of the star by the light curve alone, due to the degeneracy introduced by $\beta$. However, stars of different age have significantly different compositions, and thus the stellar debris will have different chemical abundances. This is a promising avenue to distinguish stellar age, and indeed also stellar mass, if one can tie the composition of the fallback debris to observed spectral features. See @2019ApJ...882L..25L for a first study of the chemical abundance of the debris.
The middle and bottom panels of Figure \[fig:mdots\_allone\] show $dM/dt$’s for all of our simulations. The middle panels are colored by stellar mass, with multiple stellar ages and $\beta$’s for each mass. From the middle left panel, evidently, more massive stars produce $dM/dt$ curves with higher normalizations than less massive stars, but the peak timescales are similar if BH mass is constant. This in principle would offer a way to determine stellar mass in an observed TDE, but there is a degeneracy between stellar mass and efficiency in the current fitting of TDE observations [e.g., @2019ApJ...872..151M]. One can have more efficient conversion of matter to radiation with a smaller total stellar debris mass, or a less efficient conversion of matter to radiation with a larger total stellar debris mass.
From the middle right panel, one sees that, generally, the rise slope is shallower with increasing stellar mass and (though this is more difficult to see) the decay slope is steeper with increasing stellar mass. The rise slope is a stronger trend, whereas the decay slope appears to mainly be a function of the fact that for more massive stars, a higher fraction of encounters studied in this work are partial disruptions. Indeed, with lower $\beta$ encounters being more likely, it is more likely to have a steeper decay slope for a more massive star. We also tabulate the asymptotic power law indices $n_\infty$ later on in this section.
The bottom panels are colored by $\rho_c/\bar\rho$. In the bottom right panel, one can see that more centrally concentrated stars have shallower rise slopes and (though it is difficult to see because of the density of lines) steeper decay slopes. The fact that more centrally concentrated stars have slower rises can be understood by thinking about (1) the density profile of the star and (2) the hydrodynamics of disruption. First, stars with higher values of $\rho_c/\bar\rho$ have relatively more extended, lower density envelopes. Just considering the undisturbed spherically-symmetric profile (see Figure \[fig:MESA\_profiles\_vs\_polytropes\_norm\]), the material that will be most bound to the BH has a slower transition in density, leading to a shallower rise slope. Stars with lower values of $\rho_c/\bar\rho$, on the other hand, have more uniform density profiles, and their outer layers exhibit a sharper transition in density, leading to a steeper rise slope. Second, there is a hydrodynamical effect that, for a more centrally concentrated star, the outer layers are more vulnerable to tidal disruption and this material is thus further stretched out post-disruption (compare the density contours in Figure \[fig:butterfly\]). This results in a greater difference in rise slopes than might be predicted from the undisturbed density profiles alone. This can be seen more clearly in Figure \[fig:mdots\_full\], which shows only full disruptions and has fewer profiles plotted. The fact that more centrally concentrated stars have faster decays relates to the survivability of the core. While full disruptions decay closer to $\propto t^{-5/3}$, the presence of a surviving core changes the binding energies of the material bound to the BH [@2013ApJ...767...25G]. This results in a steeper decay slope when there is a surviving core [@2019ApJ...883L..17C discuss this effect in more detail]. The transition between partial disruptions and full disruptions occurs at higher $\beta$’s with increasing $\rho_c/\bar\rho$; thus, more centrally concentrated stars are more likely to exhibit steeper decays.
The structure of the star is imprinted on the shape and slope of the fallback curve. It is thus easier to determine the stellar structure of the disrupted star in an observed TDE than more degenerate properties such as stellar mass and stellar age. This also hints that in principle, there may be a mapping from a single structural parameter such as $\rho_c/\bar\rho$, in combination with $\beta$, to the shape of the $dM/dt$. One can imagine a mapping in which $M_\star$ gives the normalization of the $dM/dt$ while $\rho_c/\bar\rho$ gives the “tilt” corresponding to the rise and decay slopes.
### Full disruptions
Figure \[fig:mdots\_full\] shows only critical (full) disruptions for all stellar masses and ages. These encounters thus all have the same fractional mass lost $\Delta M/M_\star \approx 1$. Here the trends are more evident than in Figure \[fig:mdots\_allone\], though the decay slopes are relatively similar, as all full disruptions give $n_\infty \approx -5/3$ (see below in the section). The top panels show a single stellar mass $M_\star=1M_\sun$ in order to highlight the stellar age dependence. Here the actual amount of mass unbound from the star $\Delta M$ and thus the mass eventually fed to the BH, $\approx \Delta M/2$, is identical between the stars. The rise slopes are shallower with increasing stellar age.
The middle panels are colored by stellar mass. In the left panel one primarily sees the normalization to $dM/dt$ given by $M_\star$. With increasing $M_\star$, the $dM/dt$ curve is shifted upwards and slightly to to the left. In the right panel, one sees that the rise slopes are shallower with increasing stellar mass. There is a weak trend towards steeper decay slopes with increasing stellar mass.
The bottom panels are colored by $\rho_c/\bar\rho$. Here one sees perhaps most clearly the trends discussed above, as we are coloring by the key physical parameter. It appears that, for the same $\Delta M/M_\star$, $\rho_c/\bar\rho$ nearly uniquely determines the rise slope of $dM/dt$ (this is also true for the decay slope, but is more clear when we study $n_\infty$ vs. $\beta$ below). Increasing central concentration leads to shallower rise slopes, and, for full disruptions, similar decay slopes. See discussion above for the physical intuition behind this effect.
\
\
\
### At fixed mass lost
Figure \[fig:mdots\_fixed\_deltam\] shows a comparison of $dM/dt$ curves at fixed mass lost $\Delta M$. This allows one to compare fallback rates for the same amount of total material supplied to the BH. We compare at $\Delta M = (0.1, 0.29, 0.5, 0.7, 1.0)M_\sun$; half of this unbound material returns to the BH. Of course, some values of $\Delta M$ are inaccessible to some stellar masses. To do this, we construct a very finely interpolated grid (see Appendix) of impact parameters for each star, and select the $\beta$ for which the $dM/dt$ integrates to the given $\Delta M$.
The top four panels show stars at 0 Gyr (ZAMS) and the bottom four panels show stars at 10 Gyr (and 8.4 Gyr for the 1$M_\sun$ star). If the efficiency of conversion of matter to light is relatively constant, then the total energy radiated in a TDE is determined by $\Delta M$. Thus, this figure shows $dM/dt$’s from a stellar population at a single age, and (assuming constant efficiency) for a given radiated energy.
At fixed $\Delta M$, the shape of the $dM/dt$’s vary significantly with stellar mass. The peak fallback rate, the peak timescale, and the rise and decay slopes all vary significantly. Most strikingly, for both stellar age populations, the decay slopes are steeper with increasing stellar mass. This is because for the same $\Delta M$, the encounter is more of a partial disruption for higher stellar masses. Higher stellar masses also correspond to higher $\dot M_{\rm peak}$ values (by up to a factor of $\approx$5) at fixed $\Delta M$. The differences in the shape of the $dM/dt$’s are not particularly more or less significant for different values of $\Delta M$. These differences suggest that fitting TDEs with this library of fallback rates will allow for better determination of stellar properties, and perhaps that the nature of the disrupted star can be determined from the light curve alone.
\
Derived quantities
------------------
### Peak timescale and fallback rate
Next we discuss quantities derived from the $\dot M_{\rm fb}$ results. The top panels of Figure \[fig:mdotpeaktpeak\] show the peak mass fallback rate $\dot M_{\rm peak}$ and the associated peak time $t_{\rm peak}$ as a function of impact parameter $\beta$. The $t_{\rm peak}$’s are plotted with a linear $y$-axis and so may appear to vary significantly, but nearly all $t_{\rm peak}$ values are $\approx$20–40 days. This is because the mass-radius relationship of main sequence stars nearly cancels out the $M_\star$ and $R_\star$ dependence ($M_\star^{-1} R_\star^{3/2}$) in Equation (\[eq:mdotpeaktpeakscalings\]). The general trend is that $t_{\rm peak}$ decreases sharply with $\beta$ until full disruption and then increases more slowly post critical-$\beta$.
The $\dot M_{\rm peak}$ values show a marked $M_\star$ dependence, and span several orders of magnitude. We can understand this because of the $M_\star$ and $R_\star$ dependence ($M_\star^{2} R_\star^{-3/2}$) in Equation (\[eq:mdotpeaktpeakscalings\]) has a higher power on $M_\star$ than in the $t_{\rm peak}$ scaling. The general trend is that $\dot M_{\rm peak}$ increases sharply with $\beta$ until full disruption and then decreases more slowly post critical-$\beta$. Trends for both $t_{\rm peak}$ and $\dot M_{\rm peak}$ are qualitatively similar to the @2013ApJ...767...25G simulations, but our study spans a wider range due to the many different stellar structures. Focusing on the 1$M_\sun$ star at 3 ages for comparison, at the same $\beta$, the more centrally concentrated (older) star peaks later and has a lower peak fallback rate.
In the bottom panels of Figure \[fig:mdotpeaktpeak\], the $x$-axis is scaled with the structural parameter $\alpha = (\rho_c/\bar\rho)^{-1/3}$ in the same way as in Figure \[fig:deltaM\_vs\_beta\]. Here $\dot M_{\rm peak}$ is also normalized by the stellar mass. Note that unlike for $\Delta M/M_\star$ in Figure \[fig:deltaM\_vs\_beta\], the $y$-axes of these two plots are not normalized to be unitless, and so have some spread (note that $t_{\rm peak}$ appears to have more spread, but it is on a linear scale). Despite this, nearly all of the simulations collapse into one coherent relationship. Fitting formulae for $t_{\rm peak}$ and $\dot M_{\rm peak}$ are provided in the Appendix, such that one can obtain the $t_{\rm peak}$ and $\dot M_{\rm peak}$ for any stellar mass, stellar age, and impact parameter.
\
\
Figure \[fig:mdotpeak\_times\_tpeak\] shows the quantity $\dot M_{\rm peak} \times t_{\rm peak}/M_\star$ vs. $\beta$ scaled with structural parameter as above. To order-of-magnitude, $\dot M_{\rm peak} \times t_{\rm peak}/M_\star \approx \Delta M_{\rm bound}/M_\star \approx \Delta M/2M_\star$, so this relation is similar to the bottom left panel of Figure \[fig:deltaM\_vs\_beta\]. This is a good internal consistency check. Note that the integrated $dM/dt$ is only approximately equal to the product of its peak values (e.g., for full disruptions $\Delta M/2M_\star = 0.5$, while $\dot M_{\rm peak} \times t_{\rm peak}/M_\star$ asymptotes to $\approx$0.25).
### Power law indices
The top panels of Figure \[fig:indices\] show the instantaneous power law index $n(t)$ of the mass fallback rate as a function of time. The top left panel shows the parameter space occupied by tidal disruptions of MS stars for a $10^6 M_\sun$ BH in absolute units. The $x$-axis in the top right panel is normalized by the peak time, and we also compare to the results of @2013ApJ...767...25G for $\gamma=5/3$ and $\gamma=4/3$ polytropes. The regions are comparable, but this work covers more parameter space, as expected by the larger range of stellar structures studied. The different range of $\beta$’s sampled may also account for some differences.
The bottom panels of Figure \[fig:indices\] show the asymptotic decay power law index $n_\infty$ for individual stars as a function of $\beta$, and as a function of the scaled $x$-axis incorporating stellar structure familiar from earlier figures. Partial disruptions generally have $n_\infty \approx -2.2$ [$\approx 9/4$; @2019ApJ...883L..17C], while full and post-critical disruptions generally have $n_\infty \approx -5/3$, though the $\beta$-dependence is more nuanced than this. For example, $n_\infty$ peaks near the critical $\beta$ for full disruption, then falls slightly for post-critical disruptions. There is also a small region at small $\beta$ where there is a sharp transition from higher $n_\infty$ to $n_\infty \approx -2.2$—this was better captured in @2013ApJ...767...25G, who sampled $\beta$ more finely in this region.
In the bottom left panel, one can see that the more centrally concentrated stars (which correspond roughly to the more massive stars) have $n_\infty \approx -2.2$ for progressively higher $\beta$’s. Aside from fitting individual TDEs, one can use this information with a statistical population of TDEs: given that the TDE rate depends inversely on $\beta$ to some power (see Section \[sec:conclusion\] for more detailed discussion), one can use $n_\infty$ as a probe of the stellar structure and more indirectly the stellar mass and age. More specifically, the lower-mass stars spend only a small portion of $\beta$ parameter space at $n_\infty \approx -2.2$, while the higher-mass stars spend much more of $\beta$ parameter space here, especially when weighted inversely by $\beta$ to some power. If the stellar mass function of TDEs is roughly flat for $M_\star \lesssim M_\sun$ [@2016MNRAS.461..371K], and especially if mass segregation in galactic centers is important (see Section \[sec:conclusion\]) then if an $n_\infty \approx -2.2$ is observed, it is more likely due to the partial disruption of a more massive star. Less massive stars are more likely to be full disruptions, and thus if $n_\infty \approx -5/3$ is observed, it is more likely due to the full disruption of a lower-mass star.
In the bottom right panel, after rescaling the $x$-axis with $\rho_c/\bar\rho$, the simulations reduce into a single relationship describing $n_\infty$ for any star and impact parameter. We provide a B-spline fit to this relation in the Appendix, so that one can obtain the $n_\infty$ for any stellar mass, stellar age, and impact parameter.
Conclusion {#sec:conclusion}
==========
Comparison to other work
------------------------
We compare with @2013ApJ...767...25G throughout, but perhaps the most salient point is that, as we stress in this paper, every tidal disruption quantity that we have calculated depends primarily on the stellar structure, parameterized by $\rho_c/\bar\rho$, and impact parameter $\beta$. Thus, the study of @2013ApJ...767...25G is only directly applicable to stars with exactly $\gamma=5/3$ or $\gamma=4/3$ stellar structures, which do not represent the range of stellar masses and stellar ages involved in TDEs. This work offers a significant improvement upon the @2013ApJ...767...25G grid of simulations, and applies to any main-sequence stellar mass and stellar age. The fitting formulae (see Appendix) we find for key disruption quantities can be used to predict the disruption quantities for any star, provided one knows the $\rho_c/\bar\rho$. Additionally, the `STARS_library` githup repository (see Appendix) allows one to retrieve the $dM/dt$ for any stellar mass and age.
Our results are consistent with the hydrodynamical simulations of @2019MNRAS.487..981G, @2019MNRAS.487.4790G in the non-relativistic limit, and @2019ApJ...882L..26G for the impact parameters and stars they consider. The asymptotic decay power law indices we find are broadly consistent with the analytic predictions of @2019ApJ...883L..17C ($n_\infty=-9/4$ for partial disruptions, $n_\infty=-5/3$ for full disruptions), though our results indicate a more nuanced $\beta$ dependence (see Figure \[fig:indices\] and associated discussion). We note that we find $n_\infty \approx -2.2$ for partial disruptions, slightly larger than the $n_\infty=-2.25$ found by @2019ApJ...883L..17C.
Our results are broadly consistent with those of @2020arXiv200103501R [@2020arXiv200103502R; @2020arXiv200103503R; @2020arXiv200103504R], though we note significant differences, even in the non-relativistic limit. Rather than comparing $dM/dt$’s in detail, we note that the critical impact parameters we find for full disruption differ significantly. See Figure \[fig:beta\_crit\_vs\_Mstar2\] in the Appendix for a comparison. Where our two results disagree, e.g., the $\beta_{\rm crit}=0.9$ result for the 0.3 $M_\sun$ star (which has an identical profile to a $\gamma=5/3$ polytrope), this result has been independently found by other authors for polytropic stellar models, and the non-relativistic regime (for this particular encounter $r_{\rm p} = 22 r_{\rm g}$) is one in which our grid-based adaptive-mesh FLASH framework has also been independently verified with two other numerical methods (SPH and moving-mesh) and resolutions by other authors—the simulations of @2013ApJ...767...25G, , @2019MNRAS.487..981G, @2019ApJ...882L..26G, and [@2019MNRAS.487.4790G] agree with our result. We also note that @2020arXiv200103501R [@2020arXiv200103502R; @2020arXiv200103503R; @2020arXiv200103504R] studied only a single stellar age at the middle of the main sequence lifetime for each star, and at 13-14 Gyr for $M_\star < 1 M_\sun$.
Caveats/extensions
------------------
Our calculations are Newtonian, and thus do not capture the (minority of) encounters in which relativistic effects significantly alter the $dM/dt$. @2014PhRvD..90f4020C, @2017MNRAS.469.4483T, and @2019MNRAS.487.4790G [for a review of TDEs in GR see also @2019GReGr..51...30S] studied differences between the fallback rates from relativistic versus Newtonian tidal disruption simulations in detail, and found that for $r_{\rm p}>10 r_{\rm g}$, the difference is $\lesssim10\%$. In general, the error in a non-relativistic tidal disruption simulation goes as $\approx r_{\rm g}/r_{\rm p}$. For very relativistic encounters, the decay power law index remains approximately the same, and the most important differences are that the peak time is shifted to the right and the peak fallback rate is shifted down compared to the Newtonian case. The fraction of disruptions that require relativistic hydrodynamics in order to accurately model the mass fallback rate is $\approx r_{\rm g}/r_{\rm t}$. For a 1$M_\sun$, 1$R_\sun$ star, this is $\approx$5% for a $10^6 M_\sun$ BH and $\approx$20% for a $10^7 M_\sun$ BH. See Figure \[fig:mbh\_beta\_phase\_space\] in the Appendix for the range of applicability of Newtonian hydrodynamics simulations. Note that while relativistic effects may cause significant apsidal precession, the effect on the mass fallback rate to the BH is relatively small—and it is this quantity that tracks the luminosity evolution of TDEs so closely. While the library presented in this paper applies to the majority of TDEs, we do plan to extend our setup to include relativistic encounters in future work.
Other extensions include studying orbits with different eccentricities, incorporating stellar spin [rather than performing new simulations with spinning stars, we could scale our $dM/dt$ results with a parameterized stellar spin dependence based on the results of @2019ApJ...872..163G], studying magnetic fields [we already have the framework for this, first studied in @2017ApJ...834L..19G], adding more stellar masses and ages to the grid, sampling more finely in $\beta$ (though this will not make a significant difference because of our $dM/dt$ interpolation), and various extensions and upgraded functionality of the `STARS_library` github repository. Finally, we plan to use the interpolated fallback-rate library as the new backbone for the publicly available TDE fitting software `MOSFiT` [@2018ApJS..236....6G; @2019ApJ...872..151M].
We do not perform an analysis of TDE rates as a function of stellar mass and stellar age, but below is a brief discussion. In terms of broad demographics, @2016MNRAS.461..371K calculated that for $M_{\rm BH} \lesssim 10^7 M_\sun$, the typical TDE is due to a $0.3 M_\sun$ star, but that the mass function is relatively flat for $M_\star \lesssim M_\sun$. This study only considered effects due to the initial mass function (IMF) and did not include any dynamical interactions between stars, such as mass segregation in galactic centers, which segregates more massive stars closer to the BH and ejects less massive stars [e.g., @2004ApJ...613.1143B; @2016ApJ...819...70M]—this would cause the stellar mass function for TDEs to peak at higher masses than suggested by a convolution of the IMF and the luminosity function of TDEs [@2012ApJ...760..103D; @2013ApJ...777..133M]. See also Figure 15 in @2012ApJ...757..134M for the fraction of stars at different evolutionary stages contributing to TDE flares. For $M_{\rm BH} \lesssim 10^8 M_\sun$, MS stars make up $>$85%. In more detail, @2012ApJ...757..134M find that the tidal disruption rate scales with the tidal radius of a given star as $\dot n \propto r_{\rm t}^\alpha$, where $\alpha \approx 0.2$–$0.5$. Ignoring other considerations, the physical cross-section increases with stellar age and leads to higher rates of tidal disruption for older MS stars. However, the $\Delta M$ lost at a given $\beta$ decreases as the star becomes more centrally concentrated, leading to fainter flares.
The dynamical mechanisms operating in the relevant galactic center and the associated most likely age for a star to be disrupted are also important. The two-body relaxation time, which is $\approx$2 Gyr for a $10^6 M_\sun$ BH [@2013ApJ...764...52B] (the MS lifetime of a 1.5$M_\sun$ star), gives a characteristic limit for the approximate mass and age of a star upon disruption. However, there are many other mechanisms that disrupt stars earlier. Stars can interact with one another coherently inside the sphere of influence of the BH, in contrast to two-body relaxation, leading to rapid angular momentum evolution [@1996NewA....1..149R]. For example, this occurs for eccentric nuclear disks [@2018ApJ...853..141M], galaxy mergers, or SMBH binaries [@2015MNRAS.451.1341L]. Finally, the fact that TDEs appear to be observed preferentially in post-starburst galaxies [@2014ApJ...793...38A; @2016ApJ...818L..21F; @2017ApJ...850...22L; @2018ApJ...853...39G] means that many TDEs are sourced by unique stellar populations.
Summary
-------
We summarize the main points of this paper below.
1. We present a grid of tidal disruption simulations of stars with accurate stellar structures and chemical abundances, using MESA models as initial conditions to FLASH simulations with a Helmholtz EOS.
2. We interpolated this grid in 3D (stellar mass, stellar age, and impact parameter) to provide the `STARS_library` github repository (see Section \[sec:stars\_lib\]), where one can request the $dM/dt$ for any stellar mass, stellar age, and impact parameter.
3. The quantities $\beta_{\rm crit}, \Delta M, t_{\rm peak}, \dot M_{\rm peak}$, and $n_\infty$ vary significantly with stellar mass and stellar age, but can be reduced to a single relationship that depends only on stellar structure, parameterized by $\rho_c/\bar\rho$, and impact parameter $\beta$ (see Figures \[fig:deltaM\_vs\_beta\], \[fig:beta\_crit\_vs\_Mstar\], \[fig:mdotpeaktpeak\], and \[fig:indices\]). We provide fitting formulae for these quantities applicable to any MS star (see Figure \[fig:fitting\_formulae\] in Appendix).
4. For the same $\Delta M$, the shape of the $dM/dt$ curve varies significantly with stellar mass, promising the potential determination of stellar properties from the TDE light curve alone (see Figure \[fig:mdots\_fixed\_deltam\]).
5. The critical impact parameter for full disruption increases with increasing central concentration, and scales approximately as $(\rho_c/\bar\rho)^{1/3}$ for $\rho_c/\bar\rho \lesssim 500$ and $(\rho_c/\bar\rho)^{1/2.3}$ for $\rho_c/\bar\rho \gtrsim 500$ (see Figure \[fig:beta\_crit\_vs\_Mstar\]).
6. In general, more centrally concentrated stars have steeper $dM/dt$ rise slopes and shallower decay slopes (see Figures \[fig:mdots\_allone\] and \[fig:mdots\_full\]).
7. We show that the shape of $dM/dt$ depends only on the stellar density profile, and has little dependence on the internal EOS of the star (see Section \[sec:stellar\_structure\_vs\_EOS\] where we compare a 1$M_\sun$ and 10$M_\sun$ star with nearly identical density profiles), thus extending the range of applicability of our interpolated library and fitting formulae to any MS star.
We hope the community makes use of the `STARS_library` github repository (see Appendix), and we look forward to incorporating this library as the new backbone of the `MOSFiT` fitting software for TDEs, or any other TDE fitting framework, allowing for more accurate determinations of both the nature of the disrupted star and the BH.
We thank Jieun Choi, Josiah Schwab, Dongwook Lee, Kevin McKinnon, Katie Auchettl, Tiara Hung, Jane Dai, Emily Martin, and Nick Leaf for intellectual contributions. We acknowledge use of the [*lux*]{} supercomputer at UCSC, funded by NSF MRI grant AST 1828315, and the HPC facility at the University of Copenhagen, funded by a grant from VILLUM FONDEN (project number 16599). We thank the Heising-Simons Foundation, the Danish National Research Foundation (DNRF132), and the NSF (AST-1911206 and AST-1852393) for support. D.C. acknowledges support from the NSF GRFP under grant DGE1339067. We thank NVIDIA for helping with visualizations and volume renderings of the simulations—these were performed with the GPU-enabled NVIDIA IndeX plug-in for ParaView.
Interpolated fallback-rate library {#sec:stars_lib}
==================================
The STARS library of interpolated fallback rates, as well as up-to-date instructions for loading and using the library, are available at <https://github.com/jamielaw-smith/STARS_library>. One can create custom interpolated grids and one can also query for any stellar mass, stellar age, and impact parameter.
The basic interpolation between $dM/dt$’s is the same as in the `MOSFiT` software [@2018ApJS..236....6G; @2019ApJ...872..151M]. We first interpolate in $\beta$ for a given stellar mass and age. We then interpolate each of these $\beta$’s in stellar mass, for a given fractional main-sequence stellar age (e.g., 0.3$M_\sun$ ZAMS to 0.5$M_\sun$ ZAMS). We then interpolate in stellar age for a given stellar mass. In the query/retrieval mode, in order to retrieve a $dM/dt$ corresponding to a specific stellar mass, stellar age, and impact parameter that does not already exist in the interpolated library, we perform a quick series of interpolations on the nearest neighbor points in 3D space.
Figure \[fig:STARS\_lib\] shows the STARS library interpolated $dM/dt$’s for a small grid in stellar mass, stellar age, and impact parameter, for a single BH mass $M_{\rm BH}=10^6M_\sun$. This grid has 10 interpolated $\beta$ points, 11 interpolated mass points, and 5 interpolated age points, with spacing in $\beta$ and stellar mass logarithmic, and spacing in stellar age linear.
Figure \[fig:STARS\_lib\_allinone\] shows all of the STARS library interpolated $dM/dt$’s in the same plot, for a single BH mass $M_{\rm BH}=10^6M_\sun$. The left panel is in absolute units, in order to emphasize the several orders of magnitude in $\dot M$ and time covered by this grid even for a single BH mass—roughly 6 orders of magnitude in fallback rate and 4 orders of magnitude in time (we extend the $dM/dt$’s to later times than shown in this plot). The right panel is normalized to peak time and peak fallback rate, in order to emphasize the range of rise and decay slopes exhibited by the library. See Section \[sec:results\] for more discussion of how these slopes depend on stellar structure. In general, more centrally concentrated stars have steeper rise slopes and shallower decay slopes. The more rounded behavior near peak is from post-critical-$\beta$ disruptions; see e.g. the high-$\beta$ behavior for a given star in Figure \[fig:mdots\_allgrouped\]. We also overplot three $dM/dt$’s in gray: a ZAMS 0.3$M_\sun$ star with $\beta=0.6$, a ZAMS 1$M_\sun$ star with $\beta=4.2$, and a ZAMS 3$M_\sun$ star with $\beta=2.0$.
Fits to disruption quantities
=============================
Figure \[fig:fitting\_formulae\] shows fitted B-splines and analytic formulae for the mass lost from the star ($\Delta M/M_\star$), time of peak mass fallback rate ($t_{\rm peak}$), peak mass fallback rate ($\dot M_{\rm peak}$), and asymptotic decay power law index ($n_\infty$). These fitting relations allow one to obtain the $\Delta M/M_\star$, $t_{\rm peak}$, $\dot M_{\rm peak}$, and $n_\infty$ for any stellar mass, stellar age, and impact parameter. One must first obtain the $\rho_c/\bar\rho$ value for this star, e.g. from Table \[tab:grid\], a MESA model directly, or a pre-packaged grid of MESA models such as the MIST models [@2016ApJ...823..102C].
The simple analytic fits are below. We were unable to fit the $n_\infty$ data with a simple formula, so the only option for this quantity is the B-spline. $$\begin{aligned}
\Delta M/M_\star &=& 1.055 \tanh[(x+0.04)^{17}]-0.015\\
t_{\rm peak}\ {\rm [day]} &=& 2/x^{12} + 17x + 8\\
\dot M_{\rm peak}/(M_\star/M_\sun)\ {\rm [M_\sun/yr]} &=& 0.9 \sin(15x-1.5)/x^6 -1.5x+4.3\\
n_\infty &=& {\rm (unable\ to\ find\ simple\ analytic\ fit;\ see\ B\ spline\ below)}\\
x&=&\exp[(\beta/\beta_{\rm crit})^\alpha -1],\ \ \ \alpha=(\rho_c/\bar\rho)^{-1/3}\\
\beta_{\rm crit} &=&
\begin{cases}
0.52 (\rho_c/\bar\rho)^{1/3},\ \rho_c/\bar\rho \lesssim 500\\
0.39 (\rho_c/\bar\rho)^{1/2.3},\ \rho_c/\bar\rho \gtrsim 500 \label{eq:beta_crit_appendix}
\end{cases}\end{aligned}$$ The range for $\Delta M/M_\star$ is $x \in [0.8,1.0]$ and for $t_{\rm peak}$ and $\dot M_{\rm peak}$ is $x \in [0.8,1.7]$.
Below we provide python code to read in and evaluate the B-spline fits to the disruption quantities $\Delta M/M_\star$, $t_{\rm peak}$, $\dot M_{\rm peak}$, and $n_\infty$. This code will reproduce the blue lines in Figure \[fig:fitting\_formulae\]. Note that the order of all of the splines is 3. The knots and coefficients have been rounded to 3 decimals, which gives precision indistinguishable from the original fitted splines.
[-6pt ` import numpy as np from scipy.interpolate import splev # deltam/mstar knots = [0.804, 0.804, 0.804, 0.804, 0.861, 0.906, 0.962, 1.007, 1.007, 1.007, 1.007]; coeffs = [0.04, 0.05, -0.004, 0.383, 0.836, 0.974, 1.001, 0.0, 0.0, 0.0, 0.0] # tpeak knots = [0.804, 0.804, 0.804, 0.804, 1.082, 1.638, 1.638, 1.638, 1.638]; coeffs = [41.644, 25.813, 27.055, 36.797, 34.184, 0.0, 0.0, 0.0, 0.0] # mdotpeak/mstar knots = [0.804, 0.804, 0.804, 0.804, 0.897, 0.943, 1.082, 1.638, 1.638, 1.638, 1.638]; coeffs = [0.151, 0.222, 2.663, 3.459, 1.719, 2.042, 2.022, 0.0, 0.0, 0.0, 0.0] # ninf knots = [0.804, 0.804, 0.804, 0.804, 0.897, 0.943, 0.99, 1.036, 1.082, 1.175, 1.268, 1.314, 1.499, 1.638, 1.638, 1.638, 1.638]; coeffs = [-2.089, -2.216, -2.185, -2.232, -1.326, -1.631, -1.653, -1.659, -1.683, -1.698, -1.664, -1.667, -1.674, 0.0, 0.0, 0.0, 0.0] # evaluate B-splines x = np.linspace(min(knots), max(knots), 100); y = splev(x, [knots, coeffs, 3]) ` ]{}
Note that in order to use the B-splines to retrieve values for a specific star, one needs to use the scaled $x$-axis, and thus input the $\rho_c/\bar\rho$ and $\beta_{\rm crit}$ values for the star. $\rho_c/\bar\rho$ can be obtained as above and $\beta_{\rm crit}$ can be obtained from either Table \[tab:beta\_crit\] or our simple formula (Eq. \[eq:beta\_crit\_appendix\]). Then one plugs these into $$x=\exp[(\beta/\beta_{\rm crit})^\alpha -1],\ \ \ \alpha=(\rho_c/\bar\rho)^{-1/3}.$$ This $x$ is the value to plug into the B-spline in order to retrieve the desired $y$-axis value.
MESA profiles vs. polytropes
============================
Figure \[fig:MESA\_profiles\_vs\_polytropes\_norm\] shows MESA density profiles vs. $\gamma=5/3$ and $\gamma=4/3$ polytropes, normalized to central density and stellar radius. One can see where a polytropic stellar model is sufficient and where we need to use a MESA stellar model as initial conditions for the FLASH simulations. For $M_\star=0.1, 0.3M_\sun$, the profiles are nearly identical to $\gamma=5/3$ polytropic stellar structures. For $M_\star=0.5, 0.7 M_\sun$ the profiles are in between $\gamma=5/3$ $\gamma=4/3$ polytropic stellar structures, and could in principle be simulated with a polytropic simulation using a $\gamma$ that has been matched to this star. For $M_\star \gtrsim 1M_\sun$, as the star evolves off the ZAMS, the profile before more centrally concentrated than $\gamma=4/3$ and thus cannot be simulated self-consistently with a polytropic stellar structure, as these are unstable for $\gamma \lesssim 4/3$. One requires a Helmholtz EOS in order to provide hydrodynamic support for these non-ZAMS stars. Interestingly, the ZAMS $M_\star \gtrsim 1M_\sun$ stars are not uniformly well represented by $\gamma=4/3$ polytropes. The only star shown here that is quite well approximated by a $\gamma=4/3$ polytrope is the 3$M_\sun$ ZAMS star.
One can also see that $\rho_c/\bar\rho$ does not map 1-1 exactly to the stellar density profile, though it is a very good approximation. Consider the 10Gyr 0.7$M_\sun$ star ($\rho_c/\bar\rho=36$) and the ZAMS 10$M_\sun$ star ($\rho_c/\bar\rho=38$). The 10$M_\sun$ star has a higher value of $\rho_c/\bar\rho$ but its density profile is less centrally concentrated than that of the 0.7$M_\sun$ star.
Initial profiles after relaxation
=================================
Figure \[fig:initial\_profiles\] shows the stellar density profiles after $5 t_{\rm dyn}$ of relaxation onto the hydrodynamical grid in FLASH, compared to initial profiles from MESA. To be clear, these two profiles match exactly at the start ($t=0$) of the FLASH simulation, but here they are shown after $5 t_{\rm dyn}$. Our setup has been tested such that the density profile does not change significantly after $\approx 100 t_{\rm dyn}$ of relaxation onto the hydrodynamical grid (without a BH present) in @2009ApJ...705..844G.
As mentioned in Section \[sec:methods\], the most centrally concentrated stars, for which $\rho_c/\bar\rho \gtrsim 150$, have a higher initial resolution in FLASH of 512 cells across the diameter of the star. Stars with $\rho_c/\bar\rho \lesssim 150$ are initially resolved by 131 cells across their diameters.
Critical impact parameter and pericenter distance vs. stellar mass
==================================================================
Figure \[fig:beta\_crit\_vs\_Mstar2\] shows critical $\beta$ and critical pericenter distance over gravitational radius ($r_{\rm p}/r_{\rm g}$) as a function of stellar mass. See also Figure \[fig:beta\_crit\_vs\_Mstar\] for critical $\beta$ as a function of $\rho_c/\bar\rho$, in which there is a simple power-law relationship. We find large ranges in critical $\beta$ and critical $r_{\rm p}/r_{\rm g}$, especially with stellar age for $M_\star \gtrsim 0.8 M_\sun$. At a fixed stellar mass, $\beta_{\rm crit}$ can vary by a factor of $\approx$4. We also compare to the results from @2020arXiv200103501R [@2020arXiv200103502R; @2020arXiv200103503R; @2020arXiv200103504R]. To avoid repetition, see Section \[sec:methods\] and Section \[sec:conclusion\] for discussion of the differences between these works. We note that these authors studied only a single stellar age and that the difference in critical $\beta$’s occurs in the non-relativistic regime where our results have been verified by several other authors. For physical pericenter distance, our average from 0.3 to 3$M_\sun$ is $r_{\rm p}/r_{\rm g} \approx 20.4$. @2020arXiv200103501R [@2020arXiv200103502R; @2020arXiv200103503R; @2020arXiv200103504R] get $r_{\rm p}/r_{\rm g} \sim 26.9$, averaging from 0.15$M_\sun$ to 3$M_\sun$, $\approx 35\%$ higher than our value.
Range of applicability of nonrelativistic TDE hydrodynamics simulations
=======================================================================
Figure \[fig:mbh\_beta\_phase\_space\] shows regions where $r_{\rm p}>10 r_{\rm g}$ for a few stars (0.1$M_\sun$, 1$M_\sun$, 10$M_\sun$). This is where we expect nonrelativistic tidal disruption simulations to have $\lesssim$10% error. For more detailed discussion of relativistic effects, see Section \[sec:conclusion\]. Note that the BH masses of the host galaxies of observed TDEs, found independently through new observations and the $M$–$\sigma$ relationship [@2017MNRAS.471.1694W; @2019MNRAS.487.4136W], lie in the range $3\times10^5 M_\sun \leq M_{\rm BH} \leq 2 \times 10^7 M_\sun$ and peak at $10^6 M_\sun$. The stellar masses obtained through fitting TDEs [@2019ApJ...872..151M] are almost all $M_\star \lesssim 1 M_\sun$.
Test of stellar structure vs. EOS {#sec:stellar_structure_vs_EOS}
=================================
In this paper we claim that stellar structure is the sole determinant of several tidal disruption quantities. To make this claim more robust we must first disentangle the effects of the equation of state. As a test of the effects of stellar structure vs. the EOS, we run simulations of two stars that have very similar density profiles but different energy support (radiative vs. convective). We use a ZAMS 1$M_\sun$ star (36 Myr) and a close-to-ZAMS 10$M_\sun$ star (4 Myr). The 10$M_\sun$ star is at central hydrogen fraction of $X=0.65$, compared the ZAMS value of $X=0.71$. The 1$M_\sun$ and 10$M_\sun$ star have very similar (but not identical) normalized density profiles, and $\rho_c/\bar\rho=42$ for the 1$M_\sun$ star and $\rho_c/\bar\rho=45$ for the 10$M_\sun$ star. We chose to have the profiles match nearly exactly, rather than have the $\rho_c/\bar\rho$ values match. We ran one simulation at $\beta=1.5$ for each, a “solid” partial disruption.
Figure \[fig:stellar\_structure\_vs\_EOS\_density\_profile\] shows a comparison of the density profiles for these two stars. They are very different in absolute scales, but normalized to $\rho_c$ and $R_\star$, the profiles are nearly indistinguishable.
Figure \[fig:stellar\_structure\_vs\_EOS\_dmdt\] shows the $dM/dt$’s, absolute and scaled with $M_\star$ and $R_\star$. The 1$M_\sun$ star has $\Delta M/M=0.581$ and the 10$M_\sun$ star has $\Delta M/M=0.583$. The $dM/dt$’s are very similar, which supports the argument advanced in this paper that the shape of the $dM/dt$ curve is determined by the stellar density profile, and allows us to provide the fitting formulae as a function of only $\rho_c/\bar\rho$ and $\beta$ for disruption quantities for any stellar mass and age within our simulation grid range. In fact, if, from this initial study of stellar structure vs. EOS, we can conclude that these fitting formulae are robust for any star, then we can extend the scalings to stars with masses outside our grid range, as their different EOS’s will have only a small effect on the resulting disruption quantities.
The differences in shape between these two $dM/dt$’s could be due partially to the slightly different density profiles (note the small “notch” in the profile for the 10$M_\sun$ star in Figure \[fig:stellar\_structure\_vs\_EOS\_density\_profile\]) and values of $\rho_c/\bar\rho$ of these two stars. However—and this is the motivation for this test—it could be due to differences in the EOS of the two stars. In particular, the change in slope at $t\approx100$ days may be due to different transitions between $\gamma=5/3$ and $\gamma=4/3$ pressure support in the $\rho$–$T$ plane for these two stars [e.g. see Figure 7 in @2017ApJ...845..173M].
MESA and FLASH parameters
=========================
In Table \[tab:MESA\_params\] and Table \[tab:FLASH\_params\] we list a few relevant parameters for the MESA and FLASH simulations. See Section \[sec:methods\] for explanations. MESA inlists are available upon request. We also turn on overshooting (not shown in the table) for $M_\star > 3M_\sun$ using the same parameter choices as in the MIST models [@2016ApJ...823..102C].
[l l]{} `create_pre_main_sequence_model` & `.true.`\
`new_net_name` & `mesa_49.net`\
`new_rates_preference` & `2 ! jina`\
`kappa_file_prefix` & `a09`\
`initial_zfracs` & `6 ! AGSS09`\
`kappa_lowT_prefix` & `lowT_fa05_a09p`\
`kappa_CO_prefix` & `a09_co`\
`initial_z` & `0.0142d0`\
`initial_y` & `0.2703d0`\
`Lnuc_div_L_zams_limit` & `0.999d0`\
`mixing_length_alpha` & `2`\
`delta_lg_XH_cntr_hard_limit` & `0.00432d0`\
`do_element_diffusion` & `.true.`\
`xa_central_lower_limit_species(1)` & `h1`\
`xa_central_lower_limit(1)` & `0.001d0`\
[l l]{} `xmax` & `1.e3`\
`tmax` & `1.e2`\
`mpole_lmax` & `20`\
`eos_tolerance` & `1.e-6`\
`smalle` & `1.e7`\
`smallt` & `1.e3`\
`smlrho` & `1.e-12`\
`smallp` & `1.e0`\
`sim_rhoAmbient` & `1.e-11`\
`sim_pAmbient` & `1.e1`\
`sim_tAmbient` & `1.e3`\
`order` & `3`\
`slopeLimiter` & `hybrid`\
`charLimiting` & `.true.`\
`use_3dFullCTU` & `.true.`\
`use_hybridOrder` & `.true.`\
`RiemannSolver` & `HYBRID`\
`shockDetect` & `.true.`\
[^1]: <https://github.com/jamielaw-smith/STARS_library>
[^2]: Available at <http://cococubed.asu.edu/code_pages/eos.shtml>.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We study the inverse Robin problem for the Schrödinger equation in a half-space. The potential is assumed to be compactly supported. We first solve the direct problem for dimensions two and three. We then show that the Robin-to-Robin map uniquely determines the potential $q$.'
address: 'Department of Mathematics and Statistics, University of Helsinki. P.O. Box 68 (Gustaf Hällströminkatu 2b) FI-00014'
author:
- Lassi Päivärinta and Miren Zubeldia
title: 'The inverse Robin boundary value problem in a half-space'
---
Introduction
============
In this paper we consider a Robin boundary problem for wave propagation modelled by the Schrödinger equation $$\label{schro}
(\Delta + q(x))u(x)=0$$ in the half-space ${\mathbb{R}}^{3}_{+} = \{(x_1, x_2, x_3) \in {\mathbb{R}}^3 : x_3 > 0 \}$ and in the half-plane ${\mathbb{R}}^2_{+}= \{ (x_1, x_2)\in {\mathbb{R}}^2 : x_2 > 0 \}$, with the impedance boundary condition $$\label{impedance}
\frac{\partial u}{\partial x_d} + \theta u = f \quad \quad \text{over} \quad \quad x_d=0,$$ for $d=2,3$ and $\theta\in \mathbb{C}$, subject to suitable radiation conditions in each case. This problem arises in the propagation of electromagnetic and acoustic waves.
We assume $\Im q(x) \geq 0$ in ${\mathbb{R}^{d}_{+}}$, $d=2,3$ and $\Im q(x) >0$ in an open and bounded domain $\Omega \subset {\mathbb{R}^{d}_{+}}$. In addition, it is required that $q(x)$ differs from the constant $k^{2}$ only in a bounded domain that contains $\Omega$, where the wave number $k$ is positive and fixed. It is customary to write the impedance parameter $\theta$ as $\theta = ik\beta$ where $\beta$ is a given constant, called an acoustic admittance.
For the applications, $\beta$ would have to be taken as a complex-valued function in order to model the physical properties of the boundaries considered. Positive values of the real part $\Re \beta >0$ are associated with energy absorbing boundaries and those problems have been widely studied by Chandler-Wilde [@cw0] and Chandler-Wilde and Peplow [@cwp0]. If $\Re \beta =0$, the boundary is said to be passive [@Ka] or non-absorbing. Energy-absorbing boundaries allow the propagation of evanescent waves having an exponential decay from a source point at the boundary and in any direction. In the critical situation when $\Re \beta =0$ and $\Im \beta <0$, induced surface waves propagate being guided by the boundary and their exponential decay is strictly reduced to the cross-sectional direction (similar to a Rayleigh wave). So this is the critical case that maximizes the energy that propagates at boundary level, which is of interest for many applications in acoustics. See [@DMN2], [@DMN1], [@DMN] and the references given there.
On the one hand, we study the case when $\Re \beta =0$ and $\Im \beta < 0$, so that $\theta > 0$. These hypotheses on the impedance coefficients imply the appearance of surface waves guided by the domains infinite boundary. The Helmholtz equation $\Delta u + k^2 u = 0$ with this boundary condition is studied by Durán, Muga, Nédélec [@DMN2], [@DMN]. Looking for outgoing solutions, the authors firstly observe that the speed of an outgoing surface wave is different from the speed of the volume waves. Thus, a usual Sommerfeld radiation condition does not describe this phenomenon. In order to exhibit the correct radiation condition, they compute the associated Green’s function and analyze its far field.
In addition, we assume either $\beta = 0$ (rigid boundary) or $\Re \beta >0$ (energy-absorbing boundary). This means that either $\theta = 0$ or $\theta \in\mathbb{C}\backslash \{0\}$ such that $\Im \theta > 0$. In applications in outdoor sound propagation, $\beta$ lies only within a restricted region of the physically feasible half-plane $\Re \beta >0$. One of the basic theoretical problems arising from a study of outdoor sound propagation is the problem of predicting the far field behavior of the sound field emitted from a monofrequency monopole point source located above a homogeneous impedance plane [@cs1], [@cs2], [@hf], [@khn], [@r], [@Th]. In mathematical terms, the solution for this problem is the Green function for Helmholtz equation in a half-space with impedance boundary condition. See for example [@cw3], [@cw4], [@cw1], [@Ha], [@Ra] among others. Thus a precise evaluation of the Green’s function is of great importance. Accurate and efficient calculation of propagation from a line source for the widest range of source and receiver positions as well as a generalized asymptotic expansion for the Green’s function in the far field is obtained by Chandler-Wilde and Hothershall [@cw1], [@cw2]. This problem is two-dimensional. Propagation from a point source, i.e., three-dimensional case, has been considered by Thomasson [@Th] among others and has been studied in more detail by Chandler-Wilde [@cw].
It turns out that in the papers mentioned above, the wave number $k$ is a positive constant. The complex wave propagation phenomena involving infinite half-planes, or perturbation of them, having impedance boundary conditions is studied by Durán, Hein and Nédélec [@DHN]. Many problems that appear in sciences and engineering motivated their study. See for example [@cwp0], [@ht], [@ma], [@sa]. Desiring to treat these problems, they study a classical two-dimensional model of time-harmonic waves, based on the Helmholtz equation. As the Helmholtz equation permits to describe a wide range of different wave propagation phenomena when the oscillations act in the linear range, the understanding of these phenomena for a relatively simple model allows to study them later for more complex and specific cases.
The first goal of this paper is to study the direct problem for the Schrödinger equation (\[schro\]) with the impedance boundary condition (\[impedance\]) for $d=2,3$. We assume that $\theta \geq 0$ or $\Im \theta > 0$. We base on the works [@DMN2], [@DMN1]-[@DMN], [@Th], [@cw1]-[@cw2] mentioned above. We construct the unique solution of the Robin boundary problem $$\label{1}
\left\{ \begin{array}{ll}
(\Delta + q(x))u= 0 & \text{in ${\mathbb{R}^{d}_{+}}$}\\
\\
\frac{\partial u}{\partial x_{d}} + \theta u = f & \text{over $\{x_{d} = 0\}$}\\
\\
u \quad \text{satisfies the corresponding radiation condition}.
\end{array} \right.$$ for $d=2,3$ in terms of the Green’s function $G_{\theta}$ associated to the Helmholtz equation $\Delta u + k^2 u = 0$ with impedance boundary condition (\[impedance\]).
Observe that when $\theta = 0$, (\[1\]) becomes into a Neumann problem. Thus proving existence and uniqueness of solution of the problem (\[1\]) solves the direct Neumann boundary problem for the Schrödinger equation in half-space. As far as we know, this does not appear in the literature.
The radiation condition that we refer in the problem (\[1\]) is established from the one that the corresponding Green’s function satisfies. We see that in the case of a non-absorbing boundary, the established radiation condition is somewhat different from the usual Sommerfeld’s one, due to the appearance of surface waves. In the rigid or energy-absorbing boundary, we have the usual one. See Appendix (Section \[appendix\]) below.
We assume that the boundary data $f$ is in the weighted Sobolev space $H^{-1/2, -\delta}({\mathbb{R}}^{d-1})$ with $\delta > 1/2$ defined by $$H^{-1/2, -\delta}({\mathbb{R}}^{d-1}) = \left\{ u: \frac{u}{(1+|x|^{2})^{\delta/2}} \in H^{-1/2}({\mathbb{R}}^{d-1}) \right\}.$$ Here we use the definition of the Sobolev space in terms of the Fourier transfrom $$\label{sobolev}
H^{s}({\mathbb{R}}^d) = \left\{ f\in L^2({\mathbb{R}}^d): \int_{{\mathbb{R}}^d} |\hat{f}(\xi)|^2 (1+|\xi|^2)^s \, d\xi < \infty \right\}, \quad \quad s\in {\mathbb{R}}.$$ Thus it follows that $$\label{inclusion}
H^{1/2, -\delta}({\mathbb{R}}^{d-1}) \subset H^{-1/2, -\delta}({\mathbb{R}}^{d-1}).$$
We show that the problem (\[1\]) has a unique solution $u$ belonging to the weighted Sobolev space $H^{1, -\delta}({\mathbb{R}^{d}_{+}})$ with $\delta > 1/2$. Thus using the trace theorem that characterizes the existence of the trace operators $$\begin{aligned}
& \gamma_0 : H^1({\mathbb{R}^{d}_{+}}) \to H^{1/2}({\mathbb{R}}^{d-1}), \, \quad \quad \quad \, \gamma_0 v = v|_{x_d = 0}\notag\\
& \gamma_1 : H^1({\mathbb{R}^{d}_{+}}) \to H^{-1/2}({\mathbb{R}}^{d-1}), \quad \quad \quad \gamma_1 v= \frac{\partial v}{\partial x_d}\Big{|}_{x_d=0}\notag\end{aligned}$$ and the definition of the weighted Sobolev spaces as above, the trace of $u$ and its normal derivative can be defined and it holds that $$\begin{aligned}
u|_{x_d=0} &\in H^{1/2, -\delta}({\mathbb{R}}^{d-1}), \notag\\
\frac{\partial u}{\partial x_d}\Big{|}_{x_d =0} &\in H^{-1/2, -\delta}({\mathbb{R}}^{d-1}).\notag\end{aligned}$$ Hence, we define the set of the Cauchy data for solutions of the Schrödinger equation in half-space as follows, $$\label{cauchy}
C_{q}:= \left\{ \left(u|_{x_{d}=0}, \frac{\partial u}{\partial x_{d}}\Big{|}_{x_{d}=0}\right) : u\in H^{1,-\delta}({\mathbb{R}^{d}_{+}}) \, \, \, \delta>\frac{1}{2}, \, \, \,(\Delta + q(x))u = 0 \, \, \text{in} \, \, {\mathbb{R}^{d}_{+}}\right\}.$$
We then introduce the Dirichlet-to-Neumann map $\Lambda_{q}$, the Neumann-to-Dirichlet map $N_{q}$ and the Robin-to-Robin map $R_{\theta_{1}, \theta_{2}}^{q}$ associated with $(\Delta + q(x))$ on ${\mathbb{R}^{d}_{+}}$ as follows,
$$\label{dnmap}
\Lambda_{q}: \left\{ \begin{array}{lcr}
H^{1/2, -\delta}({\mathbb{R}}^{d-1}) & \longrightarrow & H^{-1/2, -\delta}({\mathbb{R}}^{d-1}) \\
u|_{x_{d}=0} & \longmapsto & \frac{\partial u}{\partial x_{d}}\Big{|}_{x_{d}=0},
\end{array} \right.$$
$$\label{ndmap}
N_{q}: \left\{ \begin{array}{lcr}
R(\Lambda_{q}) \subset H^{-1/2, -\delta}({\mathbb{R}}^{d-1}) & \longrightarrow & H^{1/2, -\delta}({\mathbb{R}}^{d-1}) \\
\frac{\partial u}{\partial x_{d}}\Big{|}_{x_{d}=0} & \longmapsto & u|_{x_{d}=0},
\end{array} \right.$$
where $R(\Lambda_q)$ denotes the range of $\Lambda_q$,
$$\label{rrmap}
{R_{\theta_{1}, \theta_{2}}^{q}}: \left\{ \begin{array}{lcr}
H^{-1/2, -\delta}({\mathbb{R}}^{d-1}) & \longrightarrow & H^{-1/2, -\delta}({\mathbb{R}}^{d-1}) \\
\frac{\partial u}{\partial x_{d}} + \theta_{1}u\Big{|}_{x_{d}=0} & \longmapsto & \frac{\partial u}{\partial x_{d}} + \theta_{2}u\Big{|}_{x_{d}=0},
\end{array} \right.$$
where $\theta_{1}\neq \theta_{2}$.
The inverse Robin boundary value problem is to determine $q(x)$ in the upper half-space from knowledge of the Robin-to-Robin map $R_{\theta_1, \theta_2}^q$. As far as we know, this problem is not studied for general $\theta_1$, $\theta_2$ in the half-space. The case when $\theta_1 =0$ and $\theta_2 = \infty$ is considered by Cheney, Lassas and Uhlmann [@clu] in dimension $d=3$. The inverse scattering problem for the acoustic equation with both $\theta = \cot \alpha < 0$ (active boundary) and $\theta= \cot \alpha >0$ (passive boundary) and for exponentially decaying potentials $q$ with $\Im \theta > 0$, has been considered by Karamyan in [@Ka1], [@Ka] and [@Ka2], respectively. The impedance boundary problem in open bounded domain for the particular case of $\theta_1 =-\tan\alpha$, $\theta_2=\cot\alpha$ with $\alpha\in {\mathbb{R}}$ has been studied by Isaev and Novikov [@in1], [@in2]. More precisely, the authors give stability estimates for determination of potential [@in1] and reconstruction of a potential [@in2] from the Robin-to-Robin map.
General Robin-to-Robin map for bounded Lipschitz domains $\Omega \subset {\mathbb{R}}^d$, $d\geq 2$ is considered by Gesztesy and Mitrea [@gm1], [@gm2], where the direct Robin boundary problem for the Schrödinger operator $\Delta + q$ for not necessarily real-valued potentials $q$ satisfying $q\in L^\infty(\Omega)$ is solved. More concretely, the authors study the following generalized Robin boundary value problem $$\label{bounded}
\left\{ \begin{array}{ll}
(\Delta + q(x))u= 0 & \text{in $\Omega$}, \quad u\in H^1(\Omega)\\
\\
\frac{\partial u}{\partial \nu} + \theta u = f & \text{on $\partial \Omega$}
\end{array} \right.$$ for every $f\in H^{-1/2}(\partial\Omega)$, where $\nu$ denotes the outward pointing normal unit vector to $\partial\Omega$. Here $\theta$ corresponds to a more general boundary operator which in particular can be the operator of multiplication by $\theta\in{\mathbb{R}}$. Note that in the bounded domain case, one can work with the standard Sobolev space $$H^s(\Omega)= \{ u\in \mathcal{D}'(\Omega) : u= U|_{\Omega} \quad \text{for some} \quad U\in H^{s}({\mathbb{R}}^d)\},$$ where $\mathcal{D}'(\Omega)$ denotes the usual set of distributions on $\Omega \subset {\mathbb{R}}^d$ and $H^s({\mathbb{R}}^d)$ is as in (\[sobolev\]). By the unique solvability of the problem (\[bounded\]), one can introduce the Dirichlet-to-Neumann map $\Lambda_{q,\Omega}$, the Neumann-to-Dirichlet map $N_{q,\Omega}$ and the Robin-to-Robin map $R_{\theta_1,\theta_2,\Omega}^q$ associated with $\Delta + q$ on $\Omega$ as follows,
$$\label{dnomega}
\Lambda_{q,\Omega}: \left\{ \begin{array}{lcr}
H^{1/2}(\partial\Omega) & \longrightarrow & H^{-1/2}(\Omega) \\
u|_{\partial\Omega} & \longmapsto & \frac{\partial u}{\partial \nu}\big{|}_{\partial\Omega},
\end{array} \right.$$
$$\label{ndomega}
N_{q,\Omega}: \left\{ \begin{array}{lcr}
H^{-1/2}(\Omega) & \longrightarrow & H^{1/2}(\Omega) \\
\frac{\partial u}{\partial \nu}\big{|}_{\partial\Omega} & \longmapsto & u|_{\partial\Omega},
\end{array} \right.$$
$$\label{rromega}
R_{\theta_1, \theta_2, \Omega}^q: \left\{ \begin{array}{lcr}
H^{-1/2}(\Omega) & \longrightarrow & H^{-1/2}(\Omega) \\
\frac{\partial u}{\partial \nu} + \theta_{1}u\big{|}_{\partial\Omega} & \longmapsto & \frac{\partial u}{\partial \nu} + \theta_{2}u\big{|}_{\partial\Omega},
\end{array} \right.$$
where $\theta_{1}\neq \theta_{2}$.
The question of whether the knowledge of the Robin-to-Robin map $R_{\theta_1, \theta_2, \Omega}^q$ determines $q(x)$ in an open bounded domain $\Omega \subset {\mathbb{R}}^d$, $d\geq 2$ is also of our interest.
The main result of this manuscript can be formulated as follows.
\[main\] Let $d=2, 3$. We assume that $\theta_i \geq 0$ or $\Im \theta_i >0$, $i=1,2$ such that $\theta_1 \neq \theta_2$, $\Im q_i(x) \geq 0$ in ${\mathbb{R}^{d}_{+}}$ and $\Im q_i(x) > 0$ in an open and bounded domain $\Omega \subset {\mathbb{R}^{d}_{+}}$. Suppose the set $B$ containing the supports of $q_{1}-k^{2}$ and $q_{2}-k^{2}$ is strictly contained in the upper half-space ${\mathbb{R}^{d}_{+}}$ and $\Omega \subset B$. Furthermore, let $q_1, q_2 \in L^\infty (B)$. If $R_{\theta_{1}, \theta_{2}}^{q_{1}} = R_{\theta_{1}, \theta_{2}}^{q_{2}}$, then $q_{1} = q_{2}$.
The theorem is also true for bounded domains. In fact, we can also show the following result.
\[domain\] Let $d\geq 2$ and $\theta_i \in {\mathbb{R}}$, $i=1,2$ with $\theta_1 \neq \theta_2$. Let $\Omega \subset {\mathbb{R}}^d$ be a open bounded domain with smooth boundary. Let $q_i\in L^{\infty}(\Omega)$, $i=1,2$. If $R_{\theta_1, \theta_2, \Omega}^{q_1} = R_{\theta_1,\theta_2,\Omega}^{q_2}$, then $q_1 = q_2$.
The main idea of the proof is to reduce the Robin-to-Robin problem to a more studied Dirichlet-to-Neumann problem. We first prove that if $R_{ \theta_1, \theta_2}^{q_1}f = R_{\theta_1, \theta_2}^{q_2}f$ for a given data $f \in H^{-1/2, -\delta}({\mathbb{R}}^{d-1})$, then $\Lambda_{q_1}g = \Lambda_{q_2}g$ for every $g\in H^{1/2,-\delta}({\mathbb{R}}^{d-1})$. The same reasoning applies to the bounded domain case. Indeed, the same argument works for showing that the fact that $R_{\theta_1, \theta_2, \Omega}^{q_1} = R_{\theta_1, \theta_2, \Omega}^{q_2}$ implies that $\Lambda_{q_1,\Omega} = \Lambda_{q_2, \Omega}$. Thus our problem reduces to determine $q(x)$ from the knowledge of the Dirichlet-to-Neumann maps $\Lambda_q$ and $\Lambda_{q,\Omega}$.
When $d=3$, it is known [@clu] that knowledge of $\Lambda_{q}$ uniquely determines $q(x)$ in the half-space geometry. As far as we know, the problem in the half-space is not studied in the two dimensional case. However, same method as in [@clu] allows us to extend the result to the half-plane. See Section \[clu2d\] below. In the bounded domain case, it is known that knowledge of $\Lambda_{q, \Omega}$ uniquely determines the potential $q(x)$. For $d\geq 3$, see the seminal work of Sylvester and Uhlmann [@su]. The two dimensional case was open until the paper of Bukhgeim [@Bu] in 2008 for piecewise $W^{1,p}$ potentials with $p>2$ and later improved to $W^{\varepsilon, p}$ potentials with $\varepsilon >0$ by Blåsten [@Bl1]. The final result is given by Imanuvilov and Yamamoto [@iy] for $q\in L^\infty$ and we use this in the sequel.
The paper is organized as follows. In Section \[1.1\] we study the direct impedance boundary problem in half-space ${\mathbb{R}^{d}_{+}}$. Existence and uniqueness of solution of the problem (\[1\]) for $d=2,3$ is proved. Section \[prooftheorem\] is devoted to the proof of Theorem \[main\]. We first give a detailed proof of how to reduce the Robin-to-Robin problem to the Dirichlet-to-Neumann one in ${\mathbb{R}^{d}_{+}}$ for a general dimension $d$. Same argument works for bounded domain case. See Remark \[remarkbounded\] below. We then prove that $\Lambda_q$ uniquely determines $q$ in the half-plane. Finally, in Section \[appendix\] (Appendix) we include a small review of the papers [@DMN2], [@DMN1]-[@DMN], [@cw] and [@cw1]-[@cw2] pointing out the main properties of the Green’s function associated to the Helmholtz equation in ${\mathbb{R}}^3_{+}$ and ${\mathbb{R}}^2_{+}$, respectively.
Acknowledgements {#acknowledgements .unnumbered}
================
The authors would like to express their gratitude to Katya Krupchyk for helpful and useful comments on the paper.
The direct problem {#1.1}
==================
In this section we study the unique solvability of the Robin boundary problem (\[1\]) for $d=2,3$ and for the impedance parameter $\theta$ such that $\theta >0$ (non-absorbing boundary), $\theta = 0$ (rigid boundary) and $\Im \theta >0$ (energy-absorbing boundary).
The proof relies very much on the unperturbed Robin Green’s functions given in Appendix (Section \[appendix\]). Since there are some differences in their asymptotics and the corresponding radiation condition depending on the boundary that we consider, we split the section into two parts.
We first study the non-absorbing case $\theta >0$ in detail. Afterwards, since the rigid $\theta=0$ and energy absorbing case $\Im \theta > 0$ follows by the same method, we only focus on the differences respect to the first case. Finally, we give the existence and uniqueness result for both cases.
Non-absorbing boundary ($\theta >0$) {#uni3}
------------------------------------
Let $d=2,3$. Let us consider the Robin boundary problem $$\label{12}
\left\{ \begin{array}{ll}
(\Delta + q(x))u= 0 & \text{in ${\mathbb{R}^{d}_{+}}$}\\
\\
\frac{\partial u}{\partial x_{d}} + \theta u = f & \text{over $\{x_{d} = 0\}$},
\end{array} \right.$$ where $u$ also satisfies the radiation condition given by $$\label{RC3}
\left\{ \begin{array}{ll}
\left| \frac{\partial u}{\partial r} - iku\right| < \frac{C}{r^{\left(2\alpha + \frac{1}{2} \right)}} & \text{in ${\mathbb{R}}^3_{+}(\alpha_{+}) = \{x_{3} > r^{\alpha}\}$}\\
\\
\left|\frac{\partial u}{\partial r} - i \sqrt{k^{2} + \theta^{2}}u \right| < \frac{C}{r^{\left(\frac{3}{2} -\alpha \right)}} & \text{in ${\mathbb{R}}^3_{+}(\alpha_{-}) = \{0 \leq x_{3} < r^{\alpha} \}$},
\end{array} \right.$$ when $r \to + \infty$, for any $\alpha \in \left(\frac{1}{4}, \frac{1}{2} \right)$ and $$\label{RC2}
\left\{ \begin{array}{ll}
\left| \frac{\partial u}{\partial r} - iku\right| < \frac{C}{r^{1+\alpha}} & \text{in ${\mathbb{R}}^2_{+}(\alpha_{+}) = \{x_{2} > r^{\alpha}\}$}\\
\\
\left|\frac{\partial u}{\partial r} - i \sqrt{k^{2} + \theta^{2}}u \right| < \frac{C}{r^{1+\alpha}} & \text{in ${\mathbb{R}}^2_{+}(\alpha_{-}) = \{0 \leq x_{2} < r^{\alpha} \}$},
\end{array} \right.$$ when $r \to + \infty$, for any $\alpha \in \left(0,\frac{1}{2} \right)$.
Our first goal is to construct the unique solution of the problem (\[12\]) using the Green’s function given by Durán, Muga and Nédélec [@DMN2], [@DMN1]-[@DMN]. See Appendix \[sectiongreen\] below.
Let $G_{\theta}(x,y)$ denote the unperturbed Robin Green’s function satisfying the boundary problem $$\label{G1}
\left\{ \begin{array}{ll}
(\Delta + k^{2})G_{\theta}(x, y)= -\delta(x-y) & \text{in ${\mathbb{R}^{d}_{+}}$}\\
\\
\frac{\partial G_{\theta}(x,y)}{\partial x_{d}} + \theta G_{\theta}(x,y) = 0 & \text{over $\{x_{d} = 0\}$}\\
\\
G_{\theta} \, \text{satisfies the radiation condition (\ref{RC3}) or (\ref{RC2})},
\end{array} \right.$$ for $x, y\in {\mathbb{R}^{d}_{+}}$.
We construct the solution to the problem (\[12\]) as the solution to the following integral equation $$\label{+}
u(x) = \frac{1}{2 C_\pi C'_\pi} \int_{y_{d}=0} G_{\theta}(x,y)f(y)\, d\sigma(y) + \int_{y_{d} > 0} G_{\theta}(x,y) \left( q(y) - k^{2} \right)u(y)\, dy,$$ where $C_\pi = \frac{1}{\sqrt{8\pi}}, \, C'_\pi= \frac{1}{4\pi}$ if $d=2$ and $C_\pi = \frac{1}{4\pi},\, C'_\pi = \frac{1}{2\pi}$ when $d=3$.
We first need to show that $u$ has the desired properties. It clearly satisfies the corresponding radiation condition. To see that the Robin boundary condition $\frac{\partial u}{\partial x_d} + \theta u = f$ holds at $x_d = 0$, observe that the entire contribution to $u$ comes from the first term on the right-hand side of (\[+\]) because of the Robin boundary condition that satisfies the Green’s function if $x_d = 0$ and $y_d >0$. Thus we need to check that when $y_d = 0$, we obtain $$\label{poisson0}
\frac{\partial G_\theta(x,y)}{\partial x_d} + \theta G_\theta(x,y) = 2 C_\pi C'_\pi\delta(x'-y') \quad \quad \text{over} \quad \{x_d = 0\},$$ where $x=(x', x_d)$, $y=(y', y_d)$ and $C_\pi, \, C'_\pi$ are as above. For this purpose, from (\[fouriergreen\]) and (\[spatialrobin\]), observe that on the surface $y_d=0$ yields $$\notag
G_\theta(x,y) = 2 C_\pi C'_\pi\int_{{\mathbb{R}}^{d-1}}\frac{1}{\theta -\sqrt{\xi^2 - k^2}}e^{-\sqrt{\xi^2 - k^2}x_d}e^{-i\xi\cdot (x'-y')} \, d\xi$$ and $$\notag
\frac{\partial G_\theta(x,y)}{\partial x_d} = -2C_\pi C'_\pi\int_{{\mathbb{R}}^{d-1}}\frac{\sqrt{\xi^2 - k^2}}{\theta -\sqrt{\xi^2 - k^2}} e^{-\sqrt{\xi^2 - k^2}x_d}e^{-i\xi\cdot (x'-y')} \, d\xi.$$ Hence, evaluating at $x_d = 0$, (\[poisson0\]) follows.
We point out that when dealing with Robin Green’s function $G_\theta$, by (\[fouriergreen\]) and (\[spatialrobin\]) the solution $u$ can be also defined as the solution of the integral equation $$\label{+'}
u(x) = - \frac{1}{2\theta C_\pi C'_\pi} \int_{y_{d}=0} \frac{ \partial G_{\theta}(x,y)}{\partial y_d}f(y)\, d\sigma(y) + \int_{y_{d} > 0} G_{\theta}(x,y) \left( q(y) - k^{2} \right)u(y)\, dy,$$ being $C_\pi$ and $C'_\pi$ as above and $\theta \neq 0$. It is easy to check that when $y_d = 0$, yields $$\label{poisson}
\frac{\partial}{\partial x_d}\frac{\partial G_\theta(x,y)}{\partial y_d} + \theta\frac{\partial G_\theta(x,y)}{\partial y_d} = -2\theta C_\pi C'_\pi\delta(x'-y') \quad \quad \text{over} \quad \{x_d = 0\}.$$ This follows by showing that on the surface $y_d=0$, we have $$\notag
\frac{\partial G_\theta(x,y)}{\partial y_d} = -C_\pi C'_\pi\int_{{\mathbb{R}}^{d-1}}\frac{2\theta}{\theta -\sqrt{\xi^2 - k^2}}e^{-\sqrt{\xi^2 - k^2}x_d}e^{-i\xi\cdot (x'-y')} \, d\xi$$ and $$\notag
\frac{\partial}{\partial x_d} \frac{\partial G_\theta(x,y)}{\partial y_d} = C_\pi C'_\pi\int_{{\mathbb{R}}^{d-1}}\frac{2\theta\sqrt{\xi^2 - k^2}}{\theta -\sqrt{\xi^2 - k^2}} e^{-\sqrt{\xi^2 - k^2}x_d}e^{-i\xi\cdot (x'-y')} \, d\xi,$$ which evaluating at $x_d = 0$ implies (\[poisson\]).
In order to prove existence and uniqueness of solution of (\[12\]), we first need to show that equation (\[+\]) is uniquely solvable in a suitable function space. Indeed, we prove that (\[+\]) has a unique solution in $H^{1,-\delta}({\mathbb{R}^{d}_{+}})$ with $\delta>\frac{1}{2}$ by applying the Fredholm alternative. For simplicity of notation, we write $V(y) = q(y) - k^{2}$ and let $G_{\theta}V$ stand for the operator given by $$G_{\theta}V u(x)= \int_{y_{d} > 0} G_{\theta}(x,y) V(y) u(y) \, dy.$$
\[pro1\] Let $d=2,3$. If $V$ is a bounded function of compact support, the operator $G_{\theta}V$ is compact in $H^{1,-\delta}({\mathbb{R}^{d}_{+}})$ for any $\delta>\frac{1}{2}$.
From the assumption that $V$ has compact support in the lower half-space, we write $G_{\theta}V$ as $G_{\theta}\chi V$, where $\chi$ is the function that is one on the support of $V$ and zero everywhere else. Let $r_1, r_2 >0$ such that $supp \, V \subset \Omega \subset \{ r_1 \leq |y| \leq r_2\}$.
As in Agmon [@A], we first note that multiplication by $V$ is a compact operator mapping $H^{1,-\delta}({\mathbb{R}^{d}_{+}})$ into $L^{2,\delta}({\mathbb{R}^{d}_{+}})$ and then show that the operator $G_{\theta}\chi$ is a bounded operator mapping $L^{2,\delta}({\mathbb{R}^{d}_{+}})$ into $H^{1,-\delta}({\mathbb{R}^{d}_{+}})$. Hence the product operator $G_{\theta}\chi V = G_{\theta}V$ is a compact operator on $H^{1,-\delta}({\mathbb{R}^{d}_{+}})$.
Multiplication by $V$ is a compact operator from $H^{1,-\delta}$ to $L^{2,\delta}$ by the Sobolev embedding theorem.
To show that $G_{\theta}\chi$ is bounded from $L^{2,\delta}$ into $H^{1,-\delta}$, we use a functional analysis approach. Let $$k(x,y) = \frac{G_\theta(x,y)\chi(y)}{(1+|x|^2)^{\delta/2}(1+|y|^2)^{\delta/2}}$$ denote a function $k:{\mathbb{R}^{d}_{+}}\times {\mathbb{R}^{d}_{+}}\to \mathbb{C}$ and we define the associated integral operator $$Kf(x)=\int_{{\mathbb{R}^{d}_{+}}} k(x,y)f(y)dy.$$ We need to prove that $K: L^2 \to H^1$ is bounded.
We first show that for $\delta > 1/2$ $$\label{goal21}
\int_{{\mathbb{R}^{d}_{+}}}\int_{{\mathbb{R}^{d}_{+}}} |k(x,y)|^2 \, dy \, dx < +\infty.$$ This implies that $K:L^2 \to L^2$ is a Hilbert Schmidt operator and hence, compact and bounded. Secondly, we prove that the integral kernel $$h_i(x,y)= \frac{\partial}{\partial x_i}k(x,y)\quad \quad \quad i=1,\ldots, d$$ defines, as well, a bounded integral operator $H_i$ from $L^2$ to $L^2$. This together with shows that $K: L^2 \to H^1$ is bounded.
To show note that by (\[grb20\]), (\[grb21\]), (\[grb\]), the Robin Green’s function holds $$|G_\theta(x,y)|\leq \frac{C}{|x-y|^{\frac{d-1}{2}}}.$$ Let $\varphi \in C_0^\infty({\mathbb{R}^{d}_{+}})$ be such that $\varphi(x)=1$ for $|x|\leq r_2 + 1$ and $\varphi(x)=0$ for $|x|\geq 2r_2 + 1$. Thus for any $\delta > 0$ and $d=2,3$, we get $$\begin{aligned}
\notag
\int_{{\mathbb{R}^{d}_{+}}} \int_{{\mathbb{R}^{d}_{+}}} |\varphi(x)k(x,y)|^2 \, dx\, dy & \leq C \int_{\Omega} \frac{dy}{(1+|y|^2)^\delta} \int_{\tilde{\Omega}\cap \{|x-y|\leq 1\}} \frac{dx}{|x-y|^{d-1}}\notag\\
& + \int_{\Omega} \frac{dy}{(1+|y|^2)^\delta} \int_{\tilde{\Omega}\cap \{|x-y|\geq 1\}} \frac{dx}{(1+|x|^2)^\delta} < \infty,\notag\end{aligned}$$ where $\tilde{\Omega}$ is a bounded domain containing the support of $\varphi$. In addition, since $1-\varphi(x) = 0$ when $|x|\leq r_2 +1$ and in the support of $V$, using that $|x-y|\geq |x|-r_2$, we have $$\begin{aligned}
\notag
\int_{{\mathbb{R}^{d}_{+}}} \int_{{\mathbb{R}^{d}_{+}}} |(1-\varphi(x))k(x,y)|^2 \, dx\, dy & \leq C \int_\Omega \frac{dy}{(1+|y|^2)^\delta}\int_{|x|\geq r_2 +1} \frac{dx}{|x|^{2\delta}|x-y|^{d-1}}\\
& \leq C\int_{|x|\geq r_2 +1} \frac{dx}{||x|-r_2|^{2\delta +d-1}} < \infty,\notag \end{aligned}$$ for $2\delta - 1 >0$ i.e. $\delta > 1/2$.
To prove the claim for $H_i$ note that $$\notag
|h_i(x,y)| \leq \frac{C}{(1+|y|^2)^{\delta/2}}\left(\frac{1}{(1+|x|^2)^{\frac{\delta}{2}}|x-y|^{\frac{d+1}{2}}} + \frac{1}{(1+|x|^2)^{\frac{\delta+1}{2}}|x-y|^{\frac{d-1}{2}}} \right).$$ The second term of the above inequality is square integrable for all $\delta >0$ and defines thus a Hilbert Schmidt operator, as above. Similarly, $(1-\varphi(x))h_i(x,y)$ is square integrable for any $\delta >0$. The proof is completed by applying the Schur test to the first term of $\varphi(x)h_i(x,y)$, i.e. checking that $$\notag
\sup_{|x|\leq 2r_2+1} \frac{1}{(1+|x|^2)^{\delta/2}} \int_{|y|\leq r_2} \frac{dy}{(1+|y|^2)^{\delta/2}|x-y|^{\frac{d+1}{2}}} < \infty$$ and $$\notag
\sup_{|y|\leq r_2} \frac{1}{(1+|y|^2)^{\delta/2}} \int_{|x|\leq 2r_2+1} \frac{dx}{(1+|x|^2)^{\delta/2}|x-y|^{\frac{d+1}{2}}} < \infty.$$ These inequalities are true for all $\delta >0$ and $d=2,3$ since $$\begin{aligned}
\notag
\sup_{|x|\leq R} \int_{|y|\leq R'} \frac{dy}{|x-y|^{\frac{d+1}{2}}} \leq \sup_{|x|\leq R} \int_{|z|\leq R+R'} \frac{dz}{|z|^{\frac{d+1}{2}}} < \infty\end{aligned}$$ for any $R, R' >0$ and $d>1$.
Now we are ready to prove the uniqueness result for the integral equation (\[+\]).
\[++\] Let $d=2,3$, $\Im q(x) \geq 0$ in ${\mathbb{R}^{d}_{+}}$ and $\Im q(x) >0$ in an open and bounded domain $\Omega \subset {\mathbb{R}^{d}_{+}}$. Suppose that $V$ is a bounded function with compact support. Then equation (\[+\]) has a unique solution $u$ in $H^{1,-\delta}({\mathbb{R}^{d}_{+}})$.
By Proposition \[pro1\], the Fredholm alternative guarantees that (\[+\]) has a unique solution provided that the corresponding homogeneous equation has only the zero solution. In what follows, $B$ stands for a compact set in ${\mathbb{R}^{d}_{+}}$ that contains the support of $V$.
Observe that a solution of the homogeneous equation $$u(x) = \int_{y_{d}> 0} G_{\theta}(x,y) V(y) u(y)\, dy$$ is also a solution of the equation $$\label{2.4}
(\Delta + q(x)) u(x) = 0 \quad \text{in} \quad {\mathbb{R}^{d}_{+}},$$ with boundary condition $$\label{robin}
\frac{\partial u}{ \partial x_{d}} + \theta u = 0 \quad \text{over} \quad \{x_{d}=0\}.$$
To show that such a $u$ must be identically zero, we use an energy identity. We first introduce some notation and see some properties related to the radiation condition. Let ${S_{R}^{+}}$ denote the surface of the half-sphere of radius $R$ contained in the half-space ${\mathbb{R}^{d}_{+}}$. Let ${S_{R}^{1}}$ be the part of ${S_{R}^{+}}$ contained in the domain ${\mathbb{R}^{d}_{+}(\alpha_{+})}$. Let ${S_{R}^{2}}$ be the complementary part, i.e. the part of ${S_{R}^{+}}$ contained in the domain ${\mathbb{R}^{d}_{+}(\alpha_{-})}$. Note that the radiation conditions (\[RC3\]) and (\[RC2\]) imply that $$\label{radisphere}
\left\{ \begin{array}{ll}
\lim_{R \to +\infty} \int_{{S_{R}^{1}}} \left\vert \frac{\partial u}{\partial r} - iku \right\vert^{2}\, dS = 0 \\
\\
\lim_{R \to +\infty} \int_{{S_{R}^{2}}} \left\vert \frac{\partial u}{\partial r} - i\sqrt{\theta^2 + k^2}u \right\vert^{2}\, dS = 0. \\
\end{array} \right.$$
The procedure for obtaining the energy identity is to multiply (\[2.4\]) by the complex conjugate $\bar{u}$ and integrate over the domain within ${S_{R}^{+}}$ and the plane $x_d =0$. We call this domain $B_R^{+}$ and we choose $R$ such that $B \subset {B_{R}^{+}}$. After an application of the divergence theorem, we get $$\label{a2}
\int_{B_R^{+}} \left( |\nabla u|^{2} - q(x) |u|^{2} \right) \, dx = \int_{{S_{R}^{+}}} \frac{\partial u }{\partial r} \bar{u}\, dS - \int_{\{x_d =0\}\cap {B_{R}^{+}}}\frac{\partial u }{\partial x_d} \bar{u}\, d\sigma(x).$$ We now take the imaginary part of the above equality. By the Robin boundary condition (\[robin\]), since $\theta \in {\mathbb{R}}$, it follows that the integral over $\{x_d = 0\}$ does not contribute and we obtain $$\begin{aligned}
-\int_{{B_{R}^{+}}} \Im q(x) \vert u \vert^2 \, dx &= \Im \int_{{S_{R}^{+}}} \frac{\partial u}{\partial r}\bar{u}\, dS \label{sr}\\
& = \Im \int_{{S_{R}^{1}}} \left[\frac{\partial u}{\partial r} - iku \right]\bar{u}\, dS + k\int_{{S_{R}^{1}}} \vert u \vert^2 \, dS\notag\\
& + \Im \int_{{S_{R}^{2}}} \left[\frac{\partial u}{\partial r} - i \sqrt{k^2 + \theta^2}u \right]\bar{u}\, dS + \sqrt{\theta^2 + k^2}\int_{{S_{R}^{2}}} \vert u \vert^2 \, dS.\notag\end{aligned}$$
Observe that on the one hand, writting $q(x) = V(x) + k^2$, since $k\in {\mathbb{R}}$ and $supp \, V \subset B$, the left-hand side of the above identity can be given by $$\label{v}
-\Im\int_{B} V(x)|u|^2\, dx,$$ which is finite as $V$ is a bounded potential and $u\in L^2(B)$. In addition, by the assumption that $\Im q(x) >0$, we get that (\[v\]) is negative. On the other hand, letting $R\to \infty$, we obtain $$\label{lim}
\lim_{R \to \infty} -\int_{{B_{R}^{+}}} \Im q(x)|u|^2 \, dx = -\int_{{\mathbb{R}^{d}_{+}}} \Im q(x) |u|^2\, dx.$$
To deal with the right-hand side of (\[sr\]), we first note that its boundedness implies that in particular $\int_{S_R^i} |u|^2 \, dS < \infty$, $i=1,2$. Thus by the radiation condition (\[radisphere\]) we get $$\label{li11}
\lim_{R \to \infty} \left[\Im \int_{{S_{R}^{1}}} \left(\frac{\partial u}{\partial r} - iku \right)\bar{u}\, dS +\Im \int_{{S_{R}^{2}}} \left(\frac{\partial u}{\partial r} - i \sqrt{k^2 + \theta^2}u \right)\bar{u} \right]\, dS = 0.$$ Hence, according to the above remark related to the fact that (\[v\]) is negative and finite, by using the positivity of $k$, $\sqrt{\theta^2 + k^2}$, it follows that $$0 < \lim_{R \to +\infty} \left[k \int_{{S_{R}^{1}}} \vert u \vert^2\, dS + \sqrt{\theta^2 + k^2}\int_{{S_{R}^{2}}} \vert u \vert^2 \, dS \right] \leq 0.$$ This together with (\[li11\]) implies that the right-hand side of (\[sr\]) tends to zero as $R \to \infty$. As a consequence, combining this with (\[lim\]) gives $$\int_{{\mathbb{R}^{d}_{+}}} \Im q(x) |u|^2 \, dx= 0.$$
Finally, since $\Im q(x) > 0$ in $\Omega \subset {\mathbb{R}^{d}_{+}}$, it may be concluded that $u$ must be zero there, and the unique continuation gives $u=0$ in ${\mathbb{R}^{d}_{+}}$, which is our claim.
Rigid and absorbing boundary ($\theta = 0$ and $\Im \theta >0$) {#uni2}
---------------------------------------------------------------
Let us consider the problem (\[12\]) in half-space ${\mathbb{R}^{d}_{+}}$, with the radiation conditions (\[RC3\]) and (\[RC2\]) replaced by the usual Sommerfeld’s one, $$\label{S}
\frac{\partial u}{\partial r} -i ku = o(r^{-\frac{(d-1)}{2}}),$$ uniformly in $\alpha$ as $r=|x| \to \infty$ with $0 < \alpha < \pi$, where here $(r, \alpha)$ denotes the plane polar coordinates of $x\in {\mathbb{R}^{d}_{+}}$.
The same approach as above applies to these cases. By using (\[grbcw0\]) and (\[grbcw1\]), the proof of Propositions \[pro1\] and \[++\] runs as before, the only difference being in the uniqueness result when $\Im \theta >0$.
In order to prove that the integral equation (\[+\]) for $\Im \theta >0$ has a unique solution, one needs to work a bit more. After multiplying equation $\Delta u + q(x)u = 0$ by $\bar{u}$ and integrating over the domain $B_{R}^+$, since in this case $\theta \notin {\mathbb{R}}$, when we take the imaginary part of the identity (\[a2\]), we obtain $$\begin{aligned}
\label{tita2}
-\Im \int_{B_{R}^+}& q(x)|u|^2 \, dx = \Im \int_{{S_{R}^{+}}} \frac{\partial u}{\partial r}\bar{u} \, dS + \Im\theta \int_{\{x_d =0\}\cap {B_{R}^{+}}} |u|^2 \, d\sigma(x)\\
& = \Im \left[ \int_{{S_{R}^{+}}} \left(\frac{\partial u}{\partial r} -iku \right)\bar{u}\, dS + ik\int_{{S_{R}^{+}}} |u|^2\, dS + \theta \int_{\{x_d =0\}\cap {B_{R}^{+}}} |u|^2 \, d\sigma(x)\right].\notag\end{aligned}$$
We deal with the left hand-side of the above identity as in the $\theta >0$ case. From the fact that $u \in L^2_{loc}({\mathbb{R}}^2_{+})$ and $\Im q > 0$ in $B \subset B_R^+$, we have that the right hand-side of (\[tita2\]) is also bounded. In particular, $\int_{{S_{R}^{+}}} |u|^2 \, dS < \infty$. Thus by the radiation condition (\[S\]), it follows that $$\lim_{R \to \infty} \Im \int_{{S_{R}^{+}}} \left(\frac{\partial u}{\partial r} -iku \right)\bar{u}\, dS = 0.$$ In addition, since for $R$ large enough, the left-hand side of (\[tita2\]) is negative and the right-hand side is positive, we have $$\lim_{R \to \infty} \left[ k\int_{{S_{R}^{+}}} |u|^2 \, dS + \Im\theta \int_{\{x_d =0\}\cap{B_{R}^{+}}} |u|^2 \, d\sigma(x) \right] = 0.$$
The rest of the proof runs as before.
Note that this argument breaks down in the case that $\Im k < 0$ or/and $\Im \theta < 0$. This is the reason that we are not able to show the case when $\Re\theta < 0$ which is studied in [@DHN]. Our argument would apply for $k\in \mathbb{C}$ with $\Im k > 0$ and $\theta \in \mathbb{C}$ with $\Im \theta >0$. The rest of the cases need a different reasoning.
Existence and uniqueness {#exun}
------------------------
We can now formulate our main result of this section.
Let $d=2,3$. We assume that $\theta\geq 0$ or $\Im \theta >0$, $\Im q(x)\geq 0$ in ${\mathbb{R}^{d}_{+}}$ and $\Im q(x) >0$ in a bounded domain $\Omega \subset {\mathbb{R}^{d}_{+}}$. Let $f$ be a function in $H^{-1/2, -\delta}({\mathbb{R}}^{d-1})$, $\delta >\frac{1}{2}$. Then the problem (\[1\]) has a unique solution $u \in H^{1, -\delta}({\mathbb{R}^{d}_{+}})$ .
- - *Existence:* We will show that the solution $u$ of the integral equation (\[+\]) satisfies problem (\[1\]).
By construction, since $G_{\theta}$ is solution of the problem (\[G1\]), u satisfies equation (\[2.4\]), the boundary condition $$\frac{\partial u}{\partial x_{d}} + \theta u = f \quad \text{over} \quad \{ x_{d} = 0\}$$ and the radiation conditions (\[RC3\]), (\[RC2\]) or (\[S\]).
- *Uniqueness:* The uniqueness is guaranteed by showing that $u\equiv 0$ if $f=0$. This is done in the proof of Proposition \[++\].
Proof of the main result {#prooftheorem}
========================
This section deals with the proof of Theorem \[main\].
It will be divided into two steps. We first prove that knowledge of the Robin-to-Robin map implies knowledge of the Dirichlet-to-Neumann map. This argument works for both two and three dimensions, and any $\theta \in \mathbb{C}$. We next concern with determining the potential $q(x)$ in the half-space from the Dirichlet-to-Neumann map. In the three dimensional case, this is done by Cheney, Lassas and Uhlmann [@clu]. Thus we leave it to the reader. We will focus on $d=2$, giving a detailed proof for this case.
From Robin-to-Robin to Dirichlet-to-Neumann {#fromrrtodn}
-------------------------------------------
Let $R_{\theta_1, \theta_2}^{q_i}$, $\Lambda_{q_i}$, $N_{q_i}$ denote the Robin-to-Robin map, the Dirichlet-to-Neumann map and the Neumann-to-Dirichlet map associated with $\Delta + q_i$ on ${\mathbb{R}^{d}_{+}}$, defined by (\[dnmap\]), (\[ndmap\]) and (\[rrmap\]), respectively.
Our main goal here is to show the following result. Unless otherwise stated, we work under the assumptions of Theorem \[main\].
\[rn3\] Let $f\in H^{-1/2, -\delta}({\mathbb{R}}^{d-1})$. If $R_{\theta_1, \theta_2}^{q_1}f = R_{\theta_1, \theta_2}^{q_2}f$, then $\Lambda_{q_1}(u_1|_{x_d=0}) = \Lambda_{q_2}(u_2|_{x_d=0})$, where $u_i\in H^{-1,-\delta}({\mathbb{R}^{d}_{+}})$, $i=1,2$ is the unique solution of the problem $$\label{rob1}
\left\{ \begin{array}{ll}
(\Delta + q_i(x))u_i= 0 & \text{in ${\mathbb{R}^{d}_{+}}$}\\
\frac{\partial u_i}{\partial x_{d}} + \theta_1 u_i = f & \text{over $\{x_{d} = 0\}$}\\
u_i \quad \text{satisfies the corresponding radiation condition}.
\end{array} \right.$$ As a consequence, $\Lambda_{q_1}g = \Lambda_{q_2}g$ for any $g\in H^{1/2,-\delta}({\mathbb{R}}^{d-1})$.
To this end, we first give the following result that relates the mappings $R_{\theta_1, \theta_2}^q$ and $\Lambda_q$ and is fundamental for our approach.
\[lema2\] Let $\theta, \theta' \in \mathbb{C}$ with $\theta \neq \theta'$ and $S = R_{\theta, \theta'}^{q} - I:H^{-1/2,-\delta}({\mathbb{R}}^{d-1})\to H^{-1/2,-\delta}({\mathbb{R}}^{d-1})$. Then $$\label{sq}
S(\Lambda_{q} + \theta)= (\theta - \theta')I .$$ Moreover,
- $\Lambda_{q} + \theta$ is injective.
- $\Lambda_{q} + \theta$ has a dense range.
- $(\Lambda_{q} +\theta)S=(\theta - \theta')I$.
- $S$ is one to one with $S^{-1}= (\theta - \theta')(\Lambda_q + \theta)$. Hence, $\Lambda_q + \theta$ is invertible.
Let $f\in H^{-1/2,-\delta}({\mathbb{R}}^{d-1})$. We denote $u_f = u|_{x_d =0}$ and $\frac{\partial u_f}{\partial x_d}= \frac{\partial u}{\partial x_d}\Big|_{x_d=0}$, where $u\in H^{-1,-\delta}({\mathbb{R}^{d}_{+}})$ is the unique solution of the Robin problem $$(\Delta + q)u = 0 \quad \text{in} \quad {\mathbb{R}^{d}_{+}}\, , \quad \quad \frac{\partial u}{\partial x_d} + \theta u = f \quad \text{over} \quad \{x_d = 0\}.$$
An easy computation shows that $$\begin{aligned}
S(\Lambda_{q} + \theta) u_f & = R_{\theta, \theta'}^{q} (\Lambda_{q}u_f + \theta u_f) - \left(\frac{\partial u_f}{\partial x_d} + \theta u_f \right)\notag\\
& = \left(\frac{\partial u_f}{\partial x_d} + \theta' u_f \right) - \left(\frac{\partial u_f}{\partial x_d} + \theta u_f\right)\notag\\
& = (\theta' - \theta)u_f,\notag\end{aligned}$$ which gives (\[sq\]).
To deal with the injectivity of the operator $\Lambda_{q} + \theta$, let $g\in H^{1/2,-\delta}({\mathbb{R}}^{d-1})$ and consider the Dirichlet problem $$\label{dir01}
\left\{ \begin{array}{ll}
(\Delta + q(x))u= 0 & \text{in ${\mathbb{R}^{d}_{+}}$}\\
u_{f} = g \\
u \quad \text{satisfies the corresponding radiation condition},
\end{array} \right.$$ so that $\Lambda_q g = \Lambda_q u_f$. Now observe that if $(\Lambda_{q} +\theta)u_f = 0,$ then $S(\Lambda_{q} + \theta)u_f = 0$ and by (\[sq\]) we have $(\theta' - \theta) u_f = 0$. Hence, since $\theta \neq \theta'$, it follows that $u_f=0$, which is our claim.
We next prove that if $v\in H^{1/2, -\delta}({\mathbb{R}}^{d-1})$ is such that $\langle (\Lambda_{q} +\theta)u, v \rangle=0$ $\forall u \in H^{1/2, -\delta}({\mathbb{R}}^{d-1})$, then $v=0$. Observe that for any $u\in H^{1/2, -\delta}({\mathbb{R}}^{d-1})$, yields $$\begin{aligned}
\notag
\langle (\Lambda_{q} +\theta)u, v \rangle=0 & \Leftrightarrow \langle u, (\Lambda_{q} +\overline{\theta})v \rangle=0.\end{aligned}$$ Thus $(\Lambda_{q} +\overline{\theta}) v = 0$ and by (i) we obtain $v=0$ and (ii) follows.
In order to get (iii), using $f=\frac{\partial u_f}{\partial x_d} + \theta u_f$, we check that $$\begin{aligned}
(\Lambda_{q} + \theta) Sf & = (\Lambda_q + \theta)(R_{\theta,\theta'}-I)\left(\frac{\partial u_f}{\partial x_d}+\theta u_f \right)\notag\\
& = (\Lambda_q + \theta)(\theta'-\theta)u_f\notag\\
& = (\theta' -\theta)f\notag\end{aligned}$$ Then by property (ii) we extend the operator to $H^{-1/2,-\delta}({\mathbb{R}}^{d-1})$.
Finally, by (\[sq\]) and (iii) it follows that S is surjective and injective respectively, thus $S$ is invertible with $S^{-1} = (\Lambda_q + \theta)$ and the proof is completed.
Now we are ready to prove the main result of this section.
[**[Proof of Proposition \[rn3\]]{}**]{}
Our proof starts with the observation that $$\begin{aligned}
& R_{\theta_{1}, \theta_{2}}^{q_{1}} \left( \Lambda_{q_{1}} + \theta_{1} \right) = \Lambda_{q_{1}} + \theta_{2},\label{r1} \\
& R_{\theta_{1}, \theta_{2}}^{q_{2}} \left( \Lambda_{q_{2}} + \theta_{1} \right) = \Lambda_{q_{2}} + \theta_{2}\label{r2}.\end{aligned}$$
By the assumption $R_{\theta_{1}, \theta_{2}}^{q_{1}} = R_{\theta_{1}, \theta_{2}}^{q_{2}}$ and the fact that $R_{\theta_{1}, \theta_{2}}^{q_{i}}$ is a linear operator, subtracting (\[r1\]) and (\[r2\]) we obtain $$\label{r3}
R_{\theta_{1}, \theta_{2}} \left(\Lambda_{q_{1}} - \Lambda_{q_{2}}\right) = \Lambda_{q_{1}} - \Lambda_{q_{2}}.$$ Here and subsequently, $R_{\theta_{1}, \theta_{2}}$ stands for $R_{\theta_{1}, \theta_{2}}^{q_{i}}$, $i=1,2$.
Let us denote $T= R_{\theta_{1}, \theta_{2}} - I$ and $u_{1_{f}}=u_1|_{x_d=0}$, $u_{2_{f}}=u_{2}|_{x_d=0}$. Thus from (\[r3\]) we have $$T(\Lambda_{q_{1}}u_{1_{f}} - \Lambda_{q_{2}}u_{2_{f}}) = 0$$ and by Lemma \[lema2\] using that $T$ is injective, we obtain $$\label{dn11}
\Lambda_{q_{1}}u_{1_{f}} - \Lambda_{q_{2}}u_{2_{f}} = 0,$$ which proves the first part of the proposition.
Let $g\in H^{1/2,-\delta}({\mathbb{R}}^{d-1})$. The proof is completed by showing that $\Lambda_{q_1}g = \Lambda_{q_2}g$. For that, we consider the Dirichlet problem $$\label{dir1}
\left\{ \begin{array}{ll}
(\Delta + q_i(x))u_i= 0 & \text{in ${\mathbb{R}^{d}_{+}}$}\\
u_{i_f} = g \\
u_i \quad \text{satisfies the corresponding radiation condition},
\end{array} \right.$$ where $u_i$ is the unique solution of the Robin problem (\[rob1\]) and $u_{i_f}= u_i|_{x_d = 0}$, $i=1,2$. Observe that this can be seen as the particular case of (\[rob1\]) with $\theta=\infty$. Thus for each $g\in H^{1/2,-\delta}({\mathbb{R}}^{d-1})$ it is guaranteed the existence of the unique solution of the problem (\[dir1\]) and the associated Dirichlet-to-Neumann map $\Lambda_{q_i}$ is given by $$\Lambda_{q_i} g = \Lambda_{q_i}u_{i_f}.$$ This together with (\[dn11\]) gives our claim and the proof is complete.
\[remarkbounded\] The same reasoning applies to the sufficiently smooth bounded domain case. In the same manner, using the maps defined by (\[dnomega\]), (\[ndomega\]) and (\[rromega\]), we can show that for a given $f\in H^{-1/2}(\partial\Omega)$ if $R_{\theta_1, \theta_2, \Omega}^{q_1}f = R_{\theta_1, \theta_2, \Omega}^{q_2}f$, then $\Lambda_{q_1, \Omega}g = \Lambda_{q_2, \Omega}g$, for all $g\in H^{1/2}(\partial\Omega)$ which proves Theorem \[domain\].
It is worth pointing out that in a bounded domain case there is an alternative approach to prove that $\Lambda_{q,\Omega} + \theta$ is one to one. In fact, it can be shown that the operator $\Lambda_{q,\Omega} + \theta$ is Fredholm with index zero. For this purpose, on the one hand it needs to be checked that $\Lambda_{q,\Omega}N_{q, \Omega} = I = N_{q, \Omega}\Lambda_{q, \Omega}$. This is done by Gesztesy and Mitrea, see Remark 3.8 [@gm1] or Remark 3.6 [@gm2]. On the other hand, the compact embedding $H^{1/2}(\partial\Omega) \hookrightarrow H^{-1/2}(\partial\Omega)$ for sufficiently smooth bounded domains $\Omega \subset {\mathbb{R}}^d$ is used.
This argument breaks down in the half-space case. In particular, the embedding $H^{1/2,-\delta}({\mathbb{R}}^{d-1}) \longrightarrow H^{-1/2,-\delta}({\mathbb{R}}^{d-1})$ is not compact. However, it can be proved that the Dirichlet-to-Neumann map and the Neumann-to-Dirichlet map are inverse operators in our setting.
\[key\] Let $\delta > 1/2$. For any $f\in H^{1/2, -\delta}({\mathbb{R}}^{d-1})$, it follows that $$\label{dnd}
\Lambda_{q} N_{q} f = f = N_{q} \Lambda_{q} f.$$
Let $f\in H^{1/2, -\delta}({\mathbb{R}}^{d-1})$. We first show that $$\label{dnda}
N_{q} \Lambda_{q} f = f.$$
For a given data $f\in H^{1/2, -\delta}({\mathbb{R}}^{d-1})$, we consider the following Dirichlet problem: $$\label{dirichlet}
\left\{ \begin{array}{ll}
(\Delta + q) u= 0 & \text{in ${\mathbb{R}^{d}_{+}}$}\\
\\
u = f & \text{over $\{x_{d} = 0\}$}\\
\\
\lim_{R \to +\infty} \int_{|x|=R} \left\vert \frac{\partial u}{\partial r} - iku \right\vert^{2} \, dS = 0 .
\end{array} \right.$$
As mentioned above, this is just the particular case of our Robin problem (\[1\]) when $\theta = \infty$. Thus we know that there exists a unique solution $u_{f} \in H^{1, -\delta}({\mathbb{R}}^{d-1})$ of (\[dirichlet\]) and we can define the associated Cauchy-data set as in (\[cauchy\]) such that $C_{q} \subset H^{1/2, -\delta}({\mathbb{R}}^{d-1}) \times H^{-1/2, -\delta}({\mathbb{R}}^{d-1})$, as well as the maps $\Lambda_{q}$, $N_{q}$ as in (\[dnmap\]), (\[ndmap\]), respectively.
Let $f_{u}=u_{f}|_{x_d =0}$ and $g_{u} = \Lambda_{q} u_{f} = \frac{\partial u_{f}}{\partial x_{d}}|_{x_{d}=0}$. Observe that by definition, $$N_{q} g_{u} = f_{u}, \quad \quad \quad \Lambda_{q} f_{u} = g_{u}.$$ Hence, $$\begin{aligned}
N_{q}\Lambda_{q} f = N_{q} \Lambda_{q} f_{u} = N_{q} g_{u} = f_{u} = f,\end{aligned}$$ which gives (\[dnda\]).
The same argument works for showing that $$\Lambda_{q} N_{q} f = f,$$ with the only difference that in this case we consider the associated Neumann problem $$\label{neumann}
\left\{ \begin{array}{ll}
(\Delta + q) u= 0 & \text{in ${\mathbb{R}^{d}_{+}}$}\\
\\
\frac{\partial u}{\partial x_{d}} = f & \text{over $\{x_{d} = 0\}$}\\
\\
\lim_{R \to +\infty} \int_{|x|=R} \left\vert \frac{\partial u}{\partial r} - iku \right\vert^{2} \, dS = 0 .
\end{array} \right.$$ which is the particular case of the Robin problem when $\theta = 0$ and we take $f_{u} = \frac{\partial u_{f}}{\partial x_{d}}|_{x_{d}=0}$, $g_{u} = N_{q}f_u = u_{f}|_{x_d =0}$.
Although this property of the mappings is not necessary for our proof, we find it interesting and it can be useful for some other problems.
Uniqueness for a wave propagation inverse problem in a half-plane {#clu2d}
-----------------------------------------------------------------
The aim of this section is to determine $q(x)$ in the upper half-plane from knowledge of the Dirichlet-to-Neumann map $\Lambda_{q}$. For this purpose, we follow the same arguments as in [@clu], adapting them to the two dimensional case.
The main idea is to first prove that if $\Lambda_{q_1}=\Lambda_{q_2}$ on some open subset of the boundary $x_2 = 0$, then the following orthogonality relation for the potentials $q_1$ and $q_2$ holds $$\label{8}
\int_B (q_1 - q_2)v_1 v_2 \, dx = 0.$$ Here $B$ is an open set containing the supports of $q_1-k^2$ and $q_2-k^2$, while $v_1$ and $v_2$ are solutions of the Dirichlet problem
$$\label{di2d}
\left\{ \begin{array}{ll}
(\Delta + q_i)v_i= 0 & \text{in ${\mathbb{R}}^2_{+}$}\\
\\
v_i|_{x_2 =0} = g & \text{over $\{ x_{2} = 0\}$}\\
\\
v_i \, \, \text{satisfies the outgoing radiation condition},
\end{array} \right.$$
for $i=1,2$ and $g\in H^{1/2,-\delta}({\mathbb{R}}^1)$.
In order to get (\[8\]), we will use the associated Dirichlet Green’s function, which can be defined by the method of images. We use a tilde to denote the image point, so that $\tilde{y}$ is the point obtained reflecting $y$ across the $y_2 = 0$ plane. First we recall that the free-plane outgoing Green’s function corresponding to the medium parameter in the lower half-plane is $$G(x,y) = \frac{i}{4}H_{0}^{(1)}(k|x-y|),$$ where $H_{0}^{(1)}$ denotes the zeroth order Hankel function of the first kind. Then the Green’s function of the homogeneous half-plane is defined by the method of images as $$G^{D}(x, y) = G(x,y) - G(x,\tilde{y}).$$
This function satisfies the problem $$\left\{ \begin{array}{ll}
(\Delta + k^{2})G^{D}(x, y)= -\delta(x-y) & \text{in ${\mathbb{R}}^2_{+}$}\\
\\
G^{D}(x,y) = 0 & \text{over $\{x_{2} = 0\}$}\\
\\
\lim_{|x| \to \infty} |x|\left(\frac{\partial}{\partial |x|} -ik \right) G^{D}(x,y) = 0.
\end{array} \right.$$
Thus we can define the perturbed free-plane Green’s function associated to the problem (\[di2d\]) as the unique solution of the Lippman-Schwinger equation $$\label{dir}
G^{D}_q(x,y) = G^{D}(x,y) + \int G^{D}(x,z)(q(z)-k²)G^{D}_q(z,y)\, dz.$$
Next we compute the kernel of the Dirichlet-to-Neumann map $\Lambda_{q}$ using this Green’s function. More concretely, it can be proved that the kernel of $\Lambda_q$ is $$-\frac{\partial}{\partial x_2}\frac{\partial}{\partial y_2} G^{D}_q(x,y)\Big{|}_{x_2 =0, y_2 =0}.$$ See Proposition 2.1 of [@clu] for more details. Therefore, we say that $\Lambda_{q_1} = \Lambda_{q_2}$ on some open subset $\Gamma$ of the boundary $x_2 =0$ if the kernels of the operators coincide on $\Gamma \times \Gamma$, i.e. $$\frac{\partial}{\partial x_2}\frac{\partial}{\partial y_2}G_{q_1}^{D}(x,y)\Big{|}_{x_2=0, y_2=0} = \frac{\partial}{\partial x_2}\frac{\partial}{y_2}G_{q_2}^{D}(x,y)\Big{|}_{x_2=0, y_2=0}$$ for $x$ and $y$ in $\Gamma$.
We point out that one can also prove that the boundary value problem (\[di2d\]) has a unique solution by using the Dirichlet Green’s function $G^D_q$ given above, as done by Cheney and Isaacson [@ci] for the three-dimensional case.
Now we are ready to state the main result of this section.
Suppose the set $B$ containing the supports of $q_1 - k^2$ and $q_2 - k^2$ are strictly contained in the upper half-plane. Let $q_1, q_2 \in L^\infty(B)$. If $\Lambda_{q_1} = \Lambda_{q_2}$ on some open subset $\Gamma$ of the boundary $x_2 = 0$, then $q_1 = q_2$.
The proof will be divided into two steps. Following the same method as in [@clu], it can be proved that under the hypotheses of the theorem, the following orthogonality relation holds $$\label{8'}
\int_{B} (q_1 - q_2)v_1 v_2 \, dx =0$$ for $v_1, v_2 \in V$, where $V= \{ v \in H^2(B) : (\Delta + q)v = 0\}$.
We only give the main ideas of the proof. For more details we refer the reader to [@clu]. The proof involves a series of five lemmas. The first two show that knowledge of the Dirichlet-to-Neumann map suffices to determine the Dirichlet Green’s function outside the perturbation of $q$. The third and fourth lemmas stablish a version of the Green’s theorem identity that is often used for uniqueness arguments. The last lemma shows that the linear combinations of the Dirichlet Green’s function can be used to approximate the Bukhgeim solutions.
In the three dimensional case, Cheney, Lassas and Uhlmann [@clu] use the Sylvester-Uhlmann solutions to conclude from (\[8’\]) that $q_1=q_2$. In our case, in the two dimensional case, we make use of Imanuvilov-Yamamoto’s solutions [@iy]. We include here a brief outline of these solutions.
Let $i=\sqrt{-1}$, $x=(x_1,x_2)$, $x_1, x_2 \in {\mathbb{R}}$, $z=x_1 + ix_2$, $\bar{z}$ denote the complex conjugate of $z\in\mathbb{C}$. We set $\partial_z=\frac{1}{2}(\partial_{x_1}-i\partial_{x_2})$, $\partial_{\bar{z}}=\frac{1}{2}(\partial_{x_1}+i\partial_{x_2})$ and we introduce the operators $$\partial_{\bar{z}}^{-1}g = -\frac{1}{\pi}\int_{B} \frac{g(\xi_1,\xi_2)}{\zeta-z}\, d\xi_1 \, d\xi_2, \quad \quad \partial_{z}^{-1}g = -\frac{1}{\pi}\int_{B}\frac{g(\xi_1,\xi_2)}{\bar{\zeta}-\bar{z}} \, d\xi_1 \, d\xi_2,$$ where $\zeta=\xi_1 + i\xi_2$. Consider a holomorphic function $\Phi(x,y)=\left(z-(y_1 + iy_2) \right)^2$ with $y=y_1 +iy_2$ and we introduce two operators $$\tilde{R}_{\tau}g = \frac{1}{2}e^{\tau(\bar{\Phi}-\Phi)}\partial_{z}^{-1}\left(ge^{\tau(\Phi-\bar{\Phi})} \right), \quad \quad R_\tau g = \frac{1}{2}e^{\tau(\Phi-\bar{\Phi})}\partial_{\bar{z}}^{-1}
\left(ge^{\tau(\bar{\Phi}-\Phi)} \right).$$ Set $U_0 = 1$, $U_1 = \tilde{R}_\tau \left(\frac{1}{2}\left(\partial_{\bar{z}}^{-1}q_1 -\beta_1 \right) \right)$, $U_j = \tilde{R}_\tau\left(\frac{1}{2}\partial_{\bar{z}}^{-1}(q_1 U_{j-1}) \right)$ for all $j\geq 2$, where $\beta_1$ is a fixed constant. We then construct a solution to the Schrödinger equation with $q_1$ in the form $$\label{Aiy}
v_1 = \sum_{j=0}^\infty e^{\tau\Phi}(-1)^j U_j.$$ Similarly, we also construct the complex geometric optics solution for the Schrödinger equation with the potential $q_2$ $$\label{Biy}
v_2 = \sum_{j=0}^\infty e^{-\tau\bar{\Phi}} (-1)^j V_j,$$ where $V_0 =1$, $V_1=R_{-\tau}(\partial_z^{-1}q_2 - \beta_2)$, $V_j = R_{-\tau}\left(\partial_z^{-1}(q_2 V_{j-1}) \right)$ and $\beta_2$ is a specific constant.
After plugging the solutions (\[Aiy\]) and (\[Biy\]) into the orthogonality relation (\[8’\]), by the stationary phase argument, it can be showed that $$\label{Ciy}
0 = \int_B (q_1 - q_2) v_1 v_2 \, dx = \int_B (q_1 - q_2)e^{\tau(\bar{\Phi} - \Phi)}\, dx + o\left(\frac{1}{\tau} \right) \quad \quad \text{as} \quad \tau \to +\infty.$$ Finally, from (\[Ciy\]) by using some measure theory and the stationary phase method, it can be concluded that $q_1 - q_2 = 0$.
Appendix: Unperturbed Robin Green’s function {#appendix}
============================================
Let $d=2, 3$. In this section we present the Green’s function satisfying the boundary value problem $$\label{ro}
\left\{ \begin{array}{ll}
(\Delta + k^{2})G_{\theta}(x,y) = -\delta(x-y) & \text{in ${\mathbb{R}}^d_{+}$}\\
\\
\frac{\partial G_{\theta}(x,y)}{\partial x_{d}} + \theta G_{\theta}(x,y) = 0 & \text{over $\{x_{d} = 0\}$},
\end{array} \right.$$ where $y\in {\mathbb{R}^{d}_{+}}$ is a fixed source point and $x\in{\mathbb{R}^{d}_{+}}$ is the receiver point.
A useful method for constructing the Robin Green’s function for a general $\theta$ is taking a Fourier transform in the horizontal directions $x_1$, $x_{d-1}$ and reducing the PDE of the problem (\[ro\]) to an ordinary differential equation. Thus the spectral Green’s function can be given by $$\label{fouriergreen}
\widehat{G_\theta}(\xi,k,x_d,y_d)= C_\pi\left(\frac{\theta +\sqrt{\xi^2 - k^2}}{\theta-\sqrt{\xi^2 -k^2}}\frac{e^{-\sqrt{\xi^2 -k^2}(x_d + y_d)}}{\sqrt{\xi^2 - k^2}} - \frac{e^{-\sqrt{\xi^2 -k^2}|x_d -y_d|}}{\sqrt{\xi^2 - k^2}} \right)$$ where $C_\pi= \frac{1}{\sqrt{8\pi}}$ when $d=2$, $C_\pi = \frac{1}{4\pi}$ if $d=3$. Using the inverse Fourier transform, the spatial Green’s function is represented as $$\label{spatialrobin}
G_{\theta}(x,y)=C'_\pi\int_{{\mathbb{R}}^{d-1}} \widehat{G_{\theta}}(\xi,k,x_d,y_d)e^{-i\xi\cdot(x'-y')} \, d\xi,$$ for $C'_\pi = \frac{1}{4\pi}$ in the two dimensional case and $C'_\pi=\frac{1}{2\pi}$ when $d=3$. See [@DMN2] and [@DMN1] for more details.
We consider the cases when $\theta >0$, the non-absorbing boundary case, studied by Durán, Muga and Nédélec [@DMN2] ($d=2$), [@DMN1]-[@DMN] (d=3) and also the rigid or energy-absorbing boundary case, $\theta =0$ or $\Im \theta > 0$, given by Thomasson [@Th], Chandler-Wilde [@cw] ($d=3$) and Chandler-Wilde, Hothersall [@cw1]-[@cw2] ($d=2$). We include here a brief summary of these works pointing out the main properties of $G_{\theta}$ that we need for solving the direct problem. More precisley, we focus on the asymptotic expansion of $G_\theta$.
Non-absorbing boundary ($\theta > 0$) {#sectiongreen}
-------------------------------------
The Green’s function is given by (\[spatialrobin\]). In order to get its asymptotics, some analysis techniques like the stationary phase and the calculus of residues are used.
In the two dimensional case, it has been showed [@DMN2] that $$G_{\theta}(x,y) = G_{\theta}^1 (x,y) + G_{\theta}^2(x,y),$$ where $$\label{grb20}
G_{\theta}^1(x,y) = \left\{ \begin{array}{ll}
\left(\frac{\theta - ik \sin\gamma}{\theta + ik\sin\gamma}e^{2ik\sin\gamma y_2} -1 \right)\frac{e^{i(kr +\frac{\pi}{4})}}{\sqrt{8\pi kr}} + o\left(r^{-\frac{3}{2}} \right), & \text{when $x_2 - y_2 >0$,}\\
\\
o\left(r^{-1} \right) & \text{when $x_2 - y_2 \leq 0$}
\end{array} \right.$$ and $$\label{grb21}
G_{\theta}^2(x,y) = \frac{-i\theta}{\sqrt{\theta^2 + k^2}} e^{-\theta(y_2 + x_2)}e^{i\sqrt{\theta^2 + k^2}|y_1 - x_1|} + o\left(r^{-1}\right).$$ Here $(r, \gamma)$ are the polar coordinates for the spatial variables. Then, the general radiation condition for $r$ large and $0 < \alpha < \frac{1}{2}$ is the following: $$\label{RCapp}
\left\{ \begin{array}{ll}
\left| \frac{\partial G_{\theta}}{\partial r} - ik G_{\theta}\right| < cr^{-(1-\alpha)} & \text{in ${\mathbb{R}}^2_{+}(\alpha_{+}) = \{ (x_1, x_2)\in {\mathbb{R}}^2_{+} : x_{2} > cr^{\alpha}\}$}\\
\\
\left|\frac{\partial G_{\theta}}{\partial r} - i \sqrt{k^{2} + \theta^{2}}G_{\theta} \right| < cr^{-(1-\alpha)} & \text{in ${\mathbb{R}}^2_{+}(\alpha_{-}) = \{(x_1, x_2)\in {\mathbb{R}}^2_{+} : x_{2} < cr^\alpha \}$}.
\end{array} \right.$$
Regarding to the three dimensional case ([@DMN1], [@DMN]), the following estimation is obtained. For $0 < \alpha <1$, $$\label{grb}
G_{\theta}(x,y) = \left\{ \begin{array}{ll}
\left(\frac{\theta - ik \cos\gamma}{\theta + ik\cos\gamma}e^{2ik\cos\gamma y_3} -1 \right)\frac{e^{ikr}}{4\pi r} + O\left(r^{-(2\alpha +1/2)} \right), & \text{when $r\cos\gamma > r^\alpha$,}\\
\\
\frac{\theta e^{-\theta(x_3 + y_3)}}{i\sqrt{2\pi}(\theta^2 + k^2)^{1/4}}\frac{e^{i\left(\sqrt{\theta^2 + k^2}\rho - \pi/4 \right)}}{\sqrt{\rho}} + O\left(\rho^{-3/2} \right) & \text{when $r\cos\gamma < r^\alpha$},
\end{array} \right.$$ as $r\to \infty$ and $\rho \to +\infty$, respectively. Here $(r,\gamma,\varphi)$ are the spherical coordinates in the spatial variables and $\rho$ stands for the horizontal radial variable defined as the radius of the projection over the horizontal plane, that is, $$\rho = r\sin\gamma = \sqrt{(x_1 - y_1)^2 + (x_2 - y_2)^2} = \vert x' - y' \vert,$$ where $x'=(x_1,x_2)$ and $y'=(y_1,y_2)$.\
As a consequence, it may be concluded that the Green’s function satisfies the following general radiation condition $$\label{RCapp}
\left\{ \begin{array}{ll}
\left| \frac{\partial G_{\theta}}{\partial r} - ik G_{\theta}\right| < \frac{C}{r^{\left(2\alpha + \frac{1}{2} \right)}} & \text{in ${\mathbb{R}}^3_{+}(\alpha_{+}) = \{x_{3} > r^{\alpha}\}$}\\
\\
\left|\frac{\partial G_{\theta}}{\partial r} - i \sqrt{k^{2} + \theta^{2}}G_{\theta} \right| < \frac{C}{r^{\left(\frac{3}{2} -\alpha \right)}} & \text{in ${\mathbb{R}}^3_{+}(\alpha_{-}) = \{ 0 \leq x_{3} < r^\alpha \}$},
\end{array} \right.$$ when $r \to + \infty$ and for any $\alpha \in \left(\frac{1}{4}, \frac{1}{2} \right)$.
Rigid or energy-absorbing boundary ($\theta= 0$ or $\Im \theta > 0$) {#cw}
--------------------------------------------------------------------
In the case of a rigid boundary, the solution of the problem (\[ro\]) is easily found by the method of images to be $$\label{grbcw0}
G_{0}(x,y) = \left\{ \begin{array}{ll}
-\frac{e^{ikR}}{4\pi R} - \frac{e^{ikR'}}{4\pi R'}, & \text{when $d=3$,}\\
\\
-\frac{i}{4}H_{0}^{(1)}(kR) - \frac{i}{4}H_{0}^{(1)}(kR') & \text{when $d=2$},
\end{array} \right.$$ where $R=|x-y|$ and $R'=|x-y'|$, as $y'$ is the image position with $(y_1, y_2, -y_3)$ in the three-dimensional case and $(y_1,-y_2)$ in the two-dimensional one. In addition, we denote $(r,\gamma)$ the polar coordinates of $x$ and $\rho=kR'$. Then it follows that for $d=2,3$, $G_{0}$ satisfies the usual Sommerfeld radiaton condition $$\label{usuradi}
\frac{\partial G_{0}}{\partial r} - ik G_{0} = o(r^{-\frac{(d-1)}{2}}), \quad \quad \text{as} \quad \quad r\to \infty.$$
In the general case $\Im \theta > 0$, $G_{\theta}$ is expressed as the sum of $G_{0}$ and a correction term $P_{\theta}$, i.e., $$\label{new}
G_{\theta}(x,y) = G_0(x,y) + P_{\theta}(x,y).$$ The representation for $P_{\theta}$ and its first derivatives are derived in the form of Laplace-type integral. This allows to obtain asymptotic expansions of $P_{\theta}$ in the far fiel (as $\rho \to \infty$) by using the modified saddle point method of Ott. $P_{\theta}$ is approximated by $P_{\theta,N}$ for $N=0,1, \ldots$, and the following bound on the error of this approximation is obtained [@Th], [@cw] [@cw1], [@cw2] for $N=0,1,\ldots$ and for all $\rho_{0} >0$: $$\label{error}
|P_{\theta} - P_{\theta,N}| \leq \left\{ \begin{array}{ll}
\frac{C}{\rho^{N+1}}, & \text{when $d=3$,}\\
\\
\frac{C}{\rho^{N+\frac{3}{2}}} & \text{when $d=2$},
\end{array} \right.$$ when $\rho \geq \rho_{0}$, where $C$ is a positive constant depending on $N$ and $\rho_0$. Hence, $$P_{\theta} \to 0 \quad \quad \text{as} \quad \quad \rho \to \infty,$$ uniformly on $\theta$. As a consequence, $$\label{grbcw1}
G_{\theta} \to G_0 \quad \quad \text{as} \quad \quad \rho \to \infty,$$ and it may be concluded that $G_{\theta}$ also satisfies the radiation condition (\[usuradi\]).
[99]{}
M. Abramowitz, I.A. Stegun, *Handbook of Mathematical Functions with Formulas, Graphs, and Mathematical Tables*, Dover, New York (1965).
S. Agmon, *Spectral properties of Schrödinger operators and scattering theory*, Ann. Scuola Norm. Sup. Pisa Cl. Sci. 4 (1975), 151-218.
E. Blåsten, *Stability and uniqueness for the inverse problem of the Schrödinger equation in $2D$ with potentials in $W^{\varepsilon,p}$*, <arXiv:1106.0632> (2011).
A.L. Bukhgeim, *Recovering a potential from Cauchy data in the two-dimensional cas*, J. Inv. Ill-Posed Probles, 16 (2008), 19-33.
S.N. Chandler-Wilde, *Ground effects in environmental sound propagation*, Ph.D. Thesis, University of Bradford, U.K. (1988).
S.N. Chandler-Wilde, *The impedance boundary value problem for the Helmholtz equation in a half-plane*, Math. Methods Appl. Sci. 20, 813-840 (1997).
S.N. Chandler-Wilde, D.C. Hothersall, *Sound propagation above an inhomogeneous impedance plane*, J. Sound Vibration 98 (1985) 475-491.
S.N. Chandler-Wilde, D.C. Hothersall, *The boundary integral equation method in outdoor sound propagation*, Proceedings of the Institute of Acoustics 9 (1987) 37-44.
S.N. Chandler-Wilde, D.C. Hothersall, *Efficient calculation of the Green function for acoustic propagation above a homogeneous impedance plane*, J. Sound Vibration 180 (1995), no. 5, 705-724.
S.N. Chandler-Wilde, D.C. Hothersall, *A uniformly valid far field asymptotic expansion of the Green function for two-dimensional propagation above a homogeneous impedance plane*, J. Sound Vibration 182 (1995), no. 5, 665-675.
S.N. Chandler-Wilde, A.T. Peplow, *A boundary integral equation formulation for the Helmholtz equation in a locally perturbed half-plane*, Z. Angew. Math. Mech. 85 (2005), no. 2, 79-88.
M. Cheney, M. Lassas, G. Uhlmann, *Uniqueness for a wave propagation inverse problem in a half-space*, Inverse Problems 14 (1998) 679-684.
M. Cheney, D. Isaacson, *Inverse problems for a perturbed dissipative half-space*, Inverse Problems 11 (1995) 865-888.
C.F. Chien, W.W. Soroka, *Sound propagation along an impedance plane*, J. Sound Vribation 43 (1975) 9-20.
C.F. Chien, W.W. Soroka, *A note on calculation of sound propagation along an impedance surface*, J. Sound Vibration 69 (1980) 340-343.
M. Durán, I. Muga, J.C. Nédélec, *The Helmholtz equation with impedance in a half-plane*, C. R. Acad. Sci. Paris. Ser. I 340 (2005) 483-488.
M. Durán, I. Muga, J.C. Nédélec, *The Helmholtz equation with impedance in a half-space*, C. R. Acad. Sci. Paris. Ser. I 341 (2005) 561-566.
M. Durán, I. Muga, J.C. Nédélec, *The Helmholtz equation in a locally perturbed half-space with non-absorbing boundary*. Arch. Rational Mech. Anal. 191 (2009) 143-172.
M. Durán, R. Hein, J.C. Nédélec, *Computing numerically the Green’s function of the half-plane Helmholtz operator with impedance boundary conditions*, Numer. Math. 107 (2007) 295-314.
F. Gesztesy, M. Mitrea, *Generalized Robin boundary conditions, Robin-to-Dirichlet maps, and Krein-type resolvent formulas for Schrödinger operators on bounded Lipschitz domains*, Perspectives in partial differential equations, harmonic analysis and applications, 105-173, Proc. Sympos. Pure Math., 79, Amer. Math. Soc., Providence, RI, 2008.
F. Gesztesy, M. Mitrea, *Robin-to-Robin maps and Krein-type resolvent formulas for Schrödinger operators on bounded Lipschitz domains*, Modern analysis and applications. The Mark Krein Centenary Conference. Vol. 2: Differential operators and mechanics, 81-113, Oper. Theory Adv. Appl., 191, Birkhäuser Verlag, Basel, 2009.
D. Habault, *Sound propagation above an inhomogeneous plane: boundary integral equation methods*, J. Sound Vibration 100 (1985) 55-67.
D. Habault, P.J.T. Filippi, *Ground effect analysis: surface wave and layer potential representations*, J. Sound Vibration 79 (1981) 529-550.
L. Hwang, E. Tuck, *On the oscillations of harbours of arbitrary shape*, J. Fluid. Mech. 42 (1970) 447-464.
O. Imanuvilov, M. Yamamoto, *Inverse boundary value problem for Schrödinger equation in two dimensions*, <arXiv:1208.3775> (2012).
M.I. Isaev, R.G. Novikov, *Stability estimates for determination of potential from the impedance boundary map*, <arXiv:1112.3728> (2012).
M.I. Isaev, R.G. Novikov, *Reconstruction of a potential from the impedance boundary map*, <arXiv:1204.0076> (2012).
G. Karamyan, *The inverse scattering problem for the acoustic equation in a half-space*, Inverse Problems 18 (2002) 1673-1686.
G. Karamyan, *Inverse scattering in a half-space with passive boundary*, Commun. Partial Diff. Equ. 28 (9-10), 1627-1641 (2003).
G. Karamyan, *The inverse scattering problem with impedance boundary in a half-space*, Inverse Problems 20 (2004) 1485-1495.
T. Kawai, T. Hidaka, T. Nakajima, *Sound propagation above an impedance boundary*, J. Sound Vibration 83 (1982) 125-138.
C. L. Mader, *Numerical Modeling of Detonation*, University of California Press, Berkeley (1979).
H. Ott, *Die Sattelpunktsmethode in der Umgebung eines Pols*, Annalen der Physik 43 (1943) 393-403.
K.B. Rasmussen, *The effect of terrain profile on sound propagation outdoors*, Report Number 111, Danish Acoustic Institute (1982).
A.D. Rawlins, *The field of a spherical wave reflected from a plane absorbent surface expressed in terms of an infinite series of Legrende polynomials* J. Sound Vibration 89 (1983) 359-363.
K. Sakoda, *Optical Properties of Photonic crystals*, Springer Series in Optical Sciences 80. Springer, Berlin (2001).
G. Schwarz, *Hodge decomposition - A method for solving boundary value problems*, Springer 1995.
J. Sylvester, G. Uhlmann, *A global uniqueness theorem for an inverse boundary value problem*, Ann. of Math., 125 (1987), 153-169.
S.-I. Thomasson, *A powerful asymptotic solution for sound propagation above an impedance boundary*, Acustica 45 (1980) 122-125.
I.N. Vekua, *Generalized Analytic Functions*, Oxford, Pergamon Press, 1962.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'A weighted game or a threshold function in general admits different weighted representations even if the sum of non-negative weights is fixed to one. Here we study bounds for the diameter of the corresponding weight polytope. It turns out that the diameter can be upper bounded in terms of the maximum weight and the quota or threshold. We apply those results to approximation results between power distributions, given by power indices, and weights.'
address: 'Sascha Kurz, University of Bayreuth, 95440 Bayreuth, Germany'
author:
- Sascha Kurz
title: Bounds for the diameter of the weight polytope
---
Introduction {#sec_introduction}
============
Consider a stock corporation whose shares are hold by three major stockholders owning 35%, 34%, and 17%, respectively. The remaining 14% are widely spread. Assuming that decisions are made by a simple majority rule, all three major stockholders have equal influence on the company’s decisions, while the private shareholders have no say. To be more precise, any two major stockholders can adopt a proposal, while the private shareholders together with an arbitrary major stockholder need further affirmation. Such decision environments can be captured by means of weighted voting games. Formally, a weighted (voting) game consists of a set of players or voters $N=\{1,\dots,n\}$, a vector of non-negative weights $w=(w_1,\dots,w_n)$, and a positive quota $q$. A proposal is accepted if and only if the weight sum of its supporters meets or exceeds the quota.
Committees that decide between two alternatives have received wide attention. Von Neumann and Morgenstern introduced the notion of simple games, which is a super class of weighted games, in [@von1953theory]. Examples of decision-making bodies that can be modeled as weighted games are the US Electoral College, the Council of the European Union, the UN Security Council, the International Monetary Fund or the Governing Council of the European Central Bank. Many applications seek to evaluate players’ influence or power in simple or weighted games, see, e.g., [@leech1987ownership]. The initial example illustrates that shares or weights can be a poor proxy for the distribution of power. Using the taxicab metric, i.e., the $\Vert\cdot\Vert_1$-distance, the corresponding distance between shares and relative power is $\left|0.35-\tfrac{1}{3}\right|+\left|0.34-\tfrac{1}{3}\right|+\left|0.17-\tfrac{1}{3}\right|+
\left|0.14-0\right|\approx$32.67%. If the weights add up to one, then we speak of relative or normalized weights. The insight that the power distribution differs from relative weights, triggered the invention of so-called power indices like the Shapley-Shubik index [@shapley1954method], the Penrose-Banzhaf index [@banzhaf1964weighted], or the nucleolus [@schmeidler1969nucleolus]. Due to the combinatorial nature of most of those power indices, qualitative assessments are technically demanding and large numbers of involved parties cause computational challenges [@chalkiadakis2011computational].
One reason for the difference between relative weights and power is that a weighted game permits different representations. If there are two normalized representations whose weight vectors are at large distance then at least one of the relative weight vectors also has a large distance to the power distribution. So, here we study bounds for the diameter of the weight polytope, i.e., bounds for the maximal distance between two normalized vectors of the same weighted game. We will study those bounds in terms of the number of players, the relative quota, and the maximum relative weight in a given representation of the game.
Each weighted game, also called threshold function in threshold logic, admits a representation with integer weights. Bounds for the necessary magnitude of integer weighted are studied in the literature, see e.g. [@babai2010weights] and the references therein.
The remaining part of the paper is structured as follows. In Section \[sec\_weight\_polytope\] we give the necessary definitions for simple games, weighted games and the weight polytope. Worst case lower bounds on the diameter of the weight polytope are given in Section \[sec\_lower\_bounds\] and upper bounds are given in Section \[sec\_upper\_bounds\]. Applications to approximation results for power indices are given in Section \[sec\_applications\] before we draw a brief conclusion in Section \[sec\_conclusion\]. Some lengthy or more technical proofs are moved to an appendix.
The weight polytope of a weighted game {#sec_weight_polytope}
======================================
For a positive integer $n$ let $N=\{1,\dots, n\}$ be the set of players. A *simple game* is a mapping $v\colon 2^N\to\{0,1\}$ from the subsets of $N$ to binary outcomes satisfying $v(\emptyset)=0$, $v(N)=1$, and $v(S)\le v(T)$ for all $\emptyset\subseteq S\subseteq T\subseteq N$. The interpretation in the context of binary voting systems is as follows. A subset $S\subseteq N$, also called coalition, is considered as the set of players that are in favor of a proposal, i.e., which vote yes. If $v(S)=1$ we call coalition $S$ winning and losing otherwise. By $\mathcal{W}(v)$ we denote the set of winning coalitions and by $\mathcal{L}(v)$ we denote the set of losing coalitions of $v$. If coalition $S$ is winning but each proper subset is losing, then we call $S$ minimal winning. Similarly, if $S$ is losing but each proper superset of $S$ is winning, then we call $S$ maximal losing. By $\mathcal{W}^m(v)$ we denote the set of minimal winning and by $\mathcal{L}^m(v)$ we denote the set of maximal losing coalitions. $v(S)$ encodes the group decision, i.e., $v(S)=1$ if the proposal is accepted and $v(S)=0$ otherwise. So, these assumptions for a simple game are quite natural for a voting system with binary options in the input and output domain. The dual $v^d$ of a simple game $v$ is defined via $v^d(S)=v(N)-v(N\backslash S)=1-v(N\backslash S)$ for all $S\subseteq N$ and is a simple game itself. If $v(S)=v(S\cup\{i\})$ for all $S\subseteq N$, then we call player $i$ a null player. Player $i$ is a passer if $v(\{i\})=1$. Two players $i$ and $j$ are equivalent if $v(S\cup\{i\})=v(S\cup\{j\})$ for all $S\subseteq N\backslash\{i,j\}$.
A simple game $v$ is called *weighted* if there exist weights $w\in\mathbb{R}_{\ge 0}^n$ and a quota $q\in\mathbb{R}_{>0}$ such that $v(S)=1$ if and only if $w(S):=\sum_{i\in S} w_i\ge q$. From the conditions of a simple game we conclude $0<q\le w(N)$. If $w(N)=1$ we speak of normalized or relative weights, where $0<q\le 1$. We denote the respective game by $v=[q;w]$ and refer to the pair $(q;w)$ as a weighted representation, i.e., we can have $[q;w]=[q';w']$ but $(q;w)\neq (q';w')$. The example from the introduction can, e.g., be represented by $(51\%;35\%,34\%,17\%,14\%)$, $\left(\tfrac{1}{2};\tfrac{1}{3},\tfrac{1}{3},\tfrac{1}{3},0\right)$, or $(6;4,3,3,1)$, where the fourth player mimics the private shareholders.
\[lemma\_dual\_weights\] If $(q;w)$ is a normalized representation of a weighted game $v$, then $(1-q+\varepsilon;w)$ is a normalized representation of the dual game $v^d$ for each $0<\varepsilon< \min\{q-w(S)\mid S\in\mathcal{L}(v)\}$.
For each losing coalition $S$ of $v^d$ the coalition $N\backslash S$ is winning in $v$, so that $w(N\backslash S)=1-w(S)\ge q$ and $w(S)\le 1-q<1-q+\varepsilon$. Now let $S$ be a winning coalition of $v^d$, so that $N\backslash S$ is losing in $v$ and $\varepsilon<q-w(N\backslash S)=q-1+w(S)$, which is equivalent to $w(S)>1-q+\varepsilon$. Since $\emptyset$ is a losing coalition in $v$ we have $\varepsilon<q-w(\emptyset)=q$, so that $1-q+\varepsilon<1$.
Note that $\min\{q-w(S)\mid S\in\mathcal{L}(v)\}>0$.
Given a weighted game $v$, we call a weight vector $w\in\mathbb{R}_{\ge 0}$ feasible if there exists a quota $q\in\mathbb{R}_{>0}$ satisfying $v=[q;w]$. Obviously, such a quota exists iff the largest weight of a losing coalition is strictly smaller than the smallest weight of a winning coalition. Thus, c.f. [@kaniovski2015representation Lemma 3.2], the set of feasible normalized weight vectors is given by $$\begin{aligned}
&&\left\{ w\in\mathbb{R}_{\ge 0}^n\mid w(N)=1, v(S)>v(T)\quad\forall S\in\mathcal{W}(v), T\in\mathcal{L}(v)\right\}\\
&=& \left\{ w\in\mathbb{R}_{\ge 0}^n \mid w(N)=1, v(S)>v(T)\quad\forall S\in\mathcal{W}^m(v), T\in\mathcal{L}^m(v)\right\}.\end{aligned}$$ Note that these sets only depend on the game $v$ and are non-empty for weighted games. Due to the involved strict inequalities we have to consider their closure in order to obtain polytopes.
\[def\_weighted\_polytope\] For a weighted game $v$ we define the weight polytope of $v$ by $$\mathsf{W}(v)= \left\{ w\in\mathbb{R}_{\ge 0}^n \mid w(N)=1, v(S)\ge v(T)\quad\forall S\in\mathcal{W}^m(v), T\in\mathcal{L}^m(v)\right\}$$ and call $$\operatorname{diam}(\mathsf{W}(v))=\max\left\{\Vert w-w'\Vert_1 \mid w,w'\in \mathsf{W}(v)\right\}$$ its diameter, where $\Vert x\Vert_1:=\sum_i \left| x_i\right|$.
As an example we consider the weighted game $v=[2;1,1,1]$. For $w\in\mathsf{W}(v)$ the conditions $w(S)\ge w(T)$ for all $S\in\mathcal{W}^m(v)$ and all $T\in\mathcal{L}^m(v)$ read $w_1+w_2\ge w_3$, $w_1+w_3\ge w_2$, and $w_2+w_3\ge w_1$. The normalization $w(N)=1$ can be used to eliminated $w_3$ via $w_3=1-w_1-w_2$. Finally, respecting $w\in\mathbf{R}_{\ge 0}^3$ gives $$\mathsf{W}(v)=\left\{ \left(w_1,w_2,1-w_1-w_2\right)\mid 0\le w_1\le \frac{1}{2}, 0\le w_2\le \frac{1}{2}, w_1+w_2\ge \frac{1}{2}\right\}.$$ Since $w:=\left(\tfrac{1}{2},\tfrac{1}{2},0\right)\in\mathsf{W}(v)$ and $w':=\left(\tfrac{1}{2},0,\tfrac{1}{2}\right)\in\mathsf{W}(v)$, we have $$\operatorname{diam}(\mathsf{W}(v))\ge \Vert w-w'\Vert_1=1.$$ Indeed, it can be shown that $\Vert \tilde{w}-\hat{w}\Vert_1\le 1$ for all $\tilde{w},\hat{w}\in \mathsf{W}(v)$, so that $\operatorname{diam}(\mathsf{W}(v))=1$ in our example.
For a simple game $v$ the set $\mathsf{W}(v)$ is non-empty iff $v$ is a so-called roughly weighted game, which is a relaxation of a weighted game. While also for a weighted game $v$ not any element in $\mathsf{W}(v)$ can be completed by a suitable quota $q\in(0,1]$ to a normalized representation $(q;w)$, Definition \[def\_weighted\_polytope\] makes sense nevertheless since $\dim(\mathsf{W}(v))=n-1$, see e.g. [@kaniovski2015representation Lemma 3.4], i.e., the weight polytope is full-dimensional. More concretely, for each weighted game $v$ and each $\varepsilon\in\mathbb{R}_{>0}$ there are $w,w'\in \mathsf{W}(v)$ and $q,q'\in(0,1]$ such that $v=[q;w]=[q';w']$ and $$\operatorname{diam}(\mathsf{W}(v))-\varepsilon\le \Vert w-w'\Vert_1 \le \operatorname{diam}(\mathsf{W}(v)).$$ Given the indicated linear programming formulation, $\operatorname{diam}(\mathsf{W}(v))$ can be computed in polynomial time (in terms of the number of minimal winning and maximal losing coalitions). The same is true if we replace $\Vert\cdot\Vert_1$ by the maximum norm $\Vert x\Vert_\infty=\max\{x_i \mid 1\le i\le n\}$ for $x\in\mathbb{R}^n$. We denote the corresponding diameter by $\operatorname{diam}^\infty(\mathsf{W}(v))$. For an arbitrary $p$-norm $\Vert x\Vert_p:=\left(\sum_i x_i^p\right)^{1/p}$ with $1<p<\infty$, we can obtain lower and upper bounds via $\Vert x\Vert_\infty\le\Vert x\Vert_p\le\Vert x\Vert_1$, so that we restrict ourselves to the corresponding two distance functions. The bound $\Vert x\Vert_{\infty}\le\Vert x\Vert_1$ can be slightly improved in our context.
\[lemma\_improved\_relation\_infty\_1\] For $w,w'\in\mathbb{R}^n_{\ge 0}$ with $\Vert w\Vert_1=\Vert w'\Vert_1=1$, we have $\Vert w-w'\Vert_{\infty}\le \frac{1}{2}\Vert w-w'\Vert_1$.
With $S:=\{1\le i\le n\mid w_i\le w'_i\}$ and $A:=\sum_{i\in S} \left(w'_i-w_i\right)$, $B:=\sum_{i\in N\backslash S} \left(w_i-w'_i\right)$, where $N=\{1,\dots,n\}$, we have $A-B=0$ since $\Vert w\Vert_1=\Vert w'\Vert_1$ and $w,w'\in\mathbb{R}_{\ge 0}^n$. Thus, $\Vert w-w'\Vert_1=2A$ and $\Vert w-w'\Vert_{\infty}\le \max\{A,B\}=A$.
What can be said about $\operatorname{diam}(\mathsf{W}(v))$ and $\operatorname{diam}^\infty(\mathsf{W}(v))$ in general without solving the specific linear programs? Obviously, we have $\operatorname{diam}(\mathsf{W}(v))\le 2$ and $\operatorname{diam}^\infty(\mathsf{W}(v))\le 1$. These bounds are asymptotically attained for $n\ge 2$ and $v=[n;(1,\dots,1)]$, i.e., for any $0<\varepsilon<\frac{1}{n}$ we can set $w=(1-(n-1)\cdot\varepsilon,\dots,\varepsilon)$, $w'=(\varepsilon,\dots,\varepsilon,1-(n-1)\cdot \varepsilon)$, $q=q'=1-\varepsilon$ so that $v=[q;w]=[q';w']$, $\Vert w-w'\Vert_1=2\cdot(1-n\varepsilon)$, and $\Vert w-w'\Vert_\infty=1-2\varepsilon$. In other words, $(1,0,\dots,0),(0,\dots,0,1)\in \mathsf{W}([n;1,\dots,1])$ attain the desired distances. For the weighted game $v$ with $n=1$ players we have $\operatorname{diam}(\mathsf{W}(v))=\operatorname{diam}^\infty(\mathsf{W}(v))=0$ since $\mathsf{W}(v)=\{(1)\}$.
In order to obtain tighter bounds for the diameter of the weight polytope we need more information besides the number of players. Given an exemplary normalized representation $(q;w)$, we study *key parameters* like the relative quota $q\in(0,1]$, i.e., the quota of a normalized representation, or the maximum relative weight $\Delta(w):=\Vert w\Vert_\infty\in(0,1]$, where we write $\Delta$ whenever $w$ is clear from the context. Besides this, also more sophisticated invariants of weight vectors have been studied in applications. The so-called *Laakso-Taagepera index* a.k.a. *Herfindahl-Hirschman index*, c.f. [@laakso1981proportional], is used in Industrial Organization to measure the concentration of firms in a market, see, e.g., [@curry1983industrial], and given by $$L(w)
=\left(\sum\limits_{i=1}^{n}w_i\right)^2 / \sum\limits_{i=1}^{n} w_i^2.$$ for $w\in\mathbb{R}_{\ge 0}^n$ with $w\neq 0$. In general we have $1\le L(w)\le n$. If the weight vector $w$ is normalized, then the formula simplifies to $L(w)=1/\sum_{i=1}^n w_i^2$. Under the name effective number of parties the index is widely used in political science to measure party fragmentation, see, e.g., [@laakso1979effective]. However, we observe the following relations between the maximum relative weight $\Delta=\Delta(w)$ and the Laakso-Taagepera index $L(w)$:
\[lemma\_relation\_maximum\_laakso\_taagepera\] For $w\in\mathbb{R}_{\ge 0}^n$ with $\Vert w\Vert_1=1$, we have $$\frac{1}{\Delta}\le
\frac{1}{\Delta\left(1-\alpha(1-\alpha)\Delta\right)}\le L(w)\le
\frac{1}{\Delta^2+\frac{(1-\Delta)^2}{n-1}}\le\frac{1}{\Delta^2}$$ for $n\ge 2$, where $\alpha:=\frac{1}{\Delta}-\left\lfloor\frac{1}{\Delta}\right\rfloor\in[0,1)$. If $n=1$, then $\Delta=L(w)=1$.
Optimize $\sum\limits_{i=1}^n w_i^2$ with respect to the constraints $w\in\mathbb{R}^n$, $\Vert w\Vert_1=1$, and $\Delta(w)=\Delta$, see the appendix for the technical details.
So, any lower or upper bound involving $L(w)$ can be replaced by a bound involving $\Delta$ instead. Since $\Delta$ has nicer analytical properties and requires less information on $w$, we stick to $\Delta$ in the following. We remark that there are similar inequalities for other indices measuring market concentration. Upper bounds on $\operatorname{diam}(\mathsf{W}(v))$, in terms of $n$, $q$, and $\Delta$, will be given in Section \[sec\_upper\_bounds\] and worst case lower bounds for $\operatorname{diam}(\mathsf{W}(v))$ and $\operatorname{diam}^\infty(\mathsf{W}(v))$ will be given in Section \[sec\_lower\_bounds\].
Worst case lower bounds for the diameter of the weight polytope {#sec_lower_bounds}
===============================================================
For integers $1\le k\le s$ and $t\ge 0$ we denote by $v_{k,s,t}$ the weighted game with $s$ players of weight one, $t$ players of weight zero, and a quota of $k$, i.e., $v_{k,s,t}=[k;1,\dots,1,0,\dots 0]$. Players $1,\dots,s$ are pairwise equivalent as well as players $s+1,\dots s+t$, which are null players. If $k=1$, then each player $1\le i\le s$ is a passer. First we study lower bounds for the diameter of those weighted games.
\[lemma\_diameter\] For integers $1\le k< s$ and $t\ge 0$ we have $$\operatorname{diam}(\mathsf{W}(v_{k,s,t}))\ge \max\left\{\tfrac{1}{10k},\tfrac{1}{10(s-k)}\right\}
\quad\text{and}\quad
\operatorname{diam}^\infty(\mathsf{W}(v_{k,s,t}))\ge \tfrac{1}{s}.$$
Let $S=\{1,\dots s\}$ and $T=\{s+1,\dots,s+t\}$. We start with the lower bound for $\operatorname{diam}(\mathsf{W}(v_{k,s,t}))$. If $s$ is even, then we set $S_1=\{1,\dots,s/2\}$, $S_0=\emptyset$, and $S_{-1}=\{s/2+1,\dots,s\}$. If $s$ is odd, then we set $S_{1}=\{1,\dots,(s-1)/2\}$, $S_0=\{(s+1)/2\}$, and $S_{-1}=\{(s+3)/2,\dots,s\}$. Let $0\le \gamma\le\frac{1}{s}$ be a parameter that we specify latter depending on further case differentiations. With this, we set $w_i=\frac{1}{s}+\gamma$ for all $i\in S_1$, $w_i=\frac{1}{s}$ for all $i\in S_0$, $w_i=\frac{1}{s}-\gamma$ for all $i\in S_{-1}$, $w_i=\bar{w}_i=0$ for all $i\in T$, and $\bar{w_i}=w_{s+1-i}$ for all $i\in S$. It is easily verified that $w\in\mathbb{R}_{\ge 0}^{s+t}$ and $\Vert w\Vert_1=1$. In order to conclude $w\in \mathsf{W}(v_{k,s,t})$ it suffices to check $w(U)+w(T)=w(U)\le w(V)$ for all $U,V\subseteq S$ with $|U|=k-1$ and $|V|=k$. Since $\bar{w}$ is a permutation of $w$, $w\in \mathsf{W}(v_{k,s,t})$ implies $\bar{w}\in \mathsf{W}(v_{k,s,t})$, so that $$\operatorname{diam}(\mathsf{W}(v_{k,s,t}))\ge \Vert w-\bar{w}\Vert_1=2\gamma\cdot |S_1|=2\gamma\cdot\left\lfloor\frac{s}{2}\right\rfloor
\ge \frac{\gamma s}{2},$$ where we have used $s\ge 2$ for the last inequality.
If $k\le \frac{s+1}{2}$ we set $\gamma=\frac{1}{s(2k-1)}\le \frac{1}{s}$. For $U,V\subseteq S$ with $|U|=k-1$ and $|V|=k$ we have $w(U)\le (k-1)\cdot \left(\frac{1}{s}+\gamma\right)$ and $w(V)\ge k\cdot\left(\frac{1}{s}-\gamma\right)$ so that $w(U)\le w(V)$ and $\operatorname{diam}(\mathsf{W}(v_{k,s,t}))\ge\tfrac{1}{4k}\ge \tfrac{1}{10(s-k)}$.
If $k\ge \frac{s+2}{2}$ we set $\gamma=\frac{1}{s(2s+3-2k)}\le\frac{1}{s}$. For $U,V\subseteq S$ with $|U|=k-1$ and $|V|=k$ we have $$w(U)\le \frac{s}{2}\cdot \left(\frac{1}{s}+\gamma\right) +\frac{1}{s}+ \left(k-1-\frac{s}{2}-1\right)\cdot \left(\frac{1}{s}-\gamma\right)$$ and $$w(V)\ge \frac{s}{2}\cdot \left(\frac{1}{s}-\gamma\right) +\frac{1}{s}+ \left(k-\frac{s}{2}-1\right)\cdot \left(\frac{1}{s}+\gamma\right)$$ so that $w(U)\le w(V)$ and $$\operatorname{diam}(\mathsf{W}(v_{k,s,t}))\ge \frac{\gamma s}{2}\ge \frac{1}{2(2(s-k)+3)}\overset{s-k\ge 1}{\ge} \frac{1}{10(s-k)}\ge \frac{1}{10k}.$$
Next we consider the lower bound for $\operatorname{diam}^\infty(\mathsf{W}(v_{k,s,t}))$. We set $\gamma=\frac{1}{2s}$, $w_1=\bar{w}_2=\tfrac{1}{s}+\gamma$, $w_2=\bar{w}_1=\tfrac{1}{s}-\gamma$, $w_i=\bar{w}_i=\frac{1}{s}$ for all $3\le i\le s$, and $w_i=\bar{w}_i=0$ for all $i\in T$. It is easily verified that $w\in\mathbb{R}_{\ge 0}^{s+t}$ and $\Vert w\Vert_1=1$. In order to conclude $w\in \mathsf{W}(v_{k,s,t})$ it suffices to check $w(U)+w(T)=w(U)\le w(V)$ for all $U,V\subseteq S$ with $|U|=k-1$ and $|V|=k$. The latter follows from $w(U)\le \frac{k-1}{s}+\gamma$ and $w(V)\ge \frac{k}{s}-\gamma$. Since $\bar{w}$ is a permutation of $w$, we also have $\bar{w}\in \mathsf{W}(v_{k,s,t})$, so that $$\operatorname{diam}^\infty(\mathsf{W}(v_{k,s,t}))\ge \Vert w-\bar{w}\Vert_{\infty}=2\gamma=\frac{1}{s}.$$
For the excluded cases $k=s$ we have:
\[lemma\_diameter2\] For integers $s\ge 1$ and $t\ge 0$ with $t+s\ge 2$ we have $$\operatorname{diam}(\mathsf{W}(v_{s,s,t}))\ge \frac{2}{3}
\quad\text{and}\quad
\operatorname{diam}^\infty(\mathsf{W}(v_{s,s,t}))\ge \frac{1}{3}.$$
Let $0<\varepsilon<\tfrac{1}{s}$ be arbitrary. If $s\ge 2$ we choose $w_1=\bar{w}_s=1-(s-1)\varepsilon$, $w_i=\bar{w}_{s+1-i}=\varepsilon$ for all $2\le i\le s$, and $w_i=\bar{w}_i=0$ for all $s+1\le i\le s+t$. We can easily check $w,\bar{w}\in \mathsf{W}(v_{s,s,t})$. Since $\Vert w-\bar{w}\Vert_1=2\cdot (1-s\varepsilon)$ and $\Vert w-\bar{w}\Vert_{\infty}=1-s\varepsilon$ we have $\operatorname{diam}(\mathsf{W}(v_{s,s,t}))\ge \frac{2}{3}$ and $\operatorname{diam}^\infty(\mathsf{W}(v_{s,s,t}))\ge \frac{1}{3}$ using $\varepsilon<\tfrac{2}{3s}$.
If $s=1$ then we consider $w=(1,0,0,\dots, 0)\in \mathsf{W}(v_{1,1,t})$ and $\bar{w}=(\tfrac{2}{3},\tfrac{1}{3},0,\dots, 0)\in \mathsf{W}(v_{1,1,t})$. Thus, $\operatorname{diam}(\mathsf{W}(v_{1,1,t}))\ge \Vert w-\bar{w}\Vert_1=\tfrac{2}{3}$ and $\operatorname{diam}^\infty(\mathsf{W}(v_{1,1,t}))\ge \Vert w-\bar{w}\Vert_{\infty}=\tfrac{1}{3}$.
Next we show that for a given relative quota $q\in(0,1]$ or a given maximum relative weight $\Delta\in(0,1]$ we can construct a weighted game $v$, for any suitably large number of players, with matching representation such that $\operatorname{diam}(\mathsf{W}(v))$ is lower bounded by a positive constant independent of $q$ or $\Delta$. Actually, we construct two representations of the same weighted game and give a lower bound for the distance between the two normalized weight vectors.
\[lemma\_lb\_approximation\_q\] For each $q\in(0,1]$ there exists a weighted game $v=[q;w]=[q;\bar{w}]$ with $n\ge 2$ players, where $w,\bar{w}\in\mathbb{R}^n_{\ge 0}$, and $\Vert w\Vert_1=\Vert \bar{w}\Vert_1=1$, such that $\Vert w-\bar{w}\Vert_{\infty}\ge \frac{1}{3}$ and $\Vert w-\bar{w}\Vert_{1}\ge \frac{2}{3}$.
We give general constructions for different ranges of $q$:
- $\frac{2}{3}<q\le 1$: $w=\left(\frac{2}{3},\frac{1}{3},0,\dots,0\right)$, $\bar{w}=\left(\frac{1}{3},\frac{2}{3},0,\dots,0\right)$;
- $\frac{1}{3}<q\le \frac{2}{3}$: $w=\left(\frac{2}{3},\frac{1}{3},0,\dots,0\right)$, $\bar{w}=\left(1,0,\dots,0\right)$;
- $0<q\le \frac{1}{3}$: $w=\left(\frac{2}{3},\frac{1}{3},0,\dots,0\right)$, $\bar{w}=\left(\frac{1}{3},\frac{2}{3},0,\dots,0\right)$.
\[lemma\_lb\_approximation\_delta\] Let $\Delta\in(0,1]$ and $n\ge \frac{1}{\Delta}+1$. There exist $w,\bar{w}\in\mathbb{R}_{\ge 0}^n$, $q,\bar{q}\in(0,1]$ with $\Vert w\Vert_1=\Vert \bar{w}\Vert_1=1$, $\Delta(w)=\Delta$, $[q;w]=[\bar{q},\bar{w}]$, and $\tfrac{1}{2}\cdot\Vert w-\bar{w}\Vert_1\ge \Vert w-\bar{w}\Vert_\infty \ge \tfrac{1}{7}$.
We set $s=\left\lfloor\tfrac{1}{\Delta}\right\rfloor\ge 1$ and $t=n-s\ge 1$, since $n\ge \frac{1}{\Delta}+1\ge s+1$. For $w=(\Delta,\dots,\Delta,1-s\Delta,0,\dots,0)\in\mathbb{R}_{\ge 0}^n$, with $s$ entries being equal to $\Delta$, we have $\Delta(w)=\Delta$ and $[q;w]=v_{s,s,t}$ for $0<q=s\Delta\le 1$. Due to Lemma \[lemma\_diameter2\] we have $\operatorname{diam}^\infty(\mathsf{W}(v_{s,s,t}))\ge \frac{1}{3}$, so that the triangle inequality implies the existence of a vector $w'\in \mathsf{W}(v_{s,s,t})$ with $\Vert w-w'\Vert_\infty\ge \frac{1}{6}$. If $w'$ is on the boundary of $\mathsf{W}(v_{s,s,t})$ we slightly perturb $w'$ to $\bar{w}$ in the interior of $\mathsf{W}(v_{s,s,t})$ and complete it to a representation $(\bar{q},\bar{w})$ with $\bar{q}\in(0,1]$, $[q;w]=[\bar{q},\bar{w}]$, and $\Vert w-\bar{w}\Vert_\infty\ge \frac{1}{7}$. The inequality $\tfrac{1}{2}\cdot \Vert w-\bar{w}\Vert_1\ge \vert w-\bar{w}\Vert_\infty$ follows from Lemma \[lemma\_improved\_relation\_infty\_1\].
By a tailored construction we can obtain a slightly more general result:
\[lemma\_lb\_approximation\_delta\_1\] For each $\Delta\in(0,1)$ there exists a weighted game $v=[q;w]=[q;\bar{w}]$ with $n\ge \frac{4}{3\Delta}+6$ players, where $q\in(0,1)$, $w,\bar{w}\in\mathbb{R}^n_{\ge 0}$, $\Delta(w)=\Delta(\bar{w})=\Delta$, and $\Vert w\Vert_1=\Vert \bar{w}\Vert_1=1$, such that $\Vert w-\bar{w}\Vert_{1}\ge \frac{2}{3}$ and $\Vert w-\bar{w}\Vert_{\infty}\ge\Delta/2$.
If $\Delta\ge \frac{2}{3}$, we can consider a weighted game with two passers and $n-2$ null players. One representation is given by $q=1-\Delta$ and $w=(\Delta,1-\Delta,0,\dots,0)$. Of course we can swap the weights of the first two players and obtain a second representation given by quota $q$ an weight vector $\bar{w}=(1-\Delta,\Delta,0,\dots,0)$. With this, we compute $\Vert w-\bar{w}\Vert_{1}=2\cdot (2\Delta-1)\ge \frac{2}{3}$ and $\Vert w-\bar{w}\Vert_{\infty}=2\Delta-1\ge \Delta/2$.
If $0<\Delta<\frac{2}{3}$, we define an integer $a:=\left\lfloor\frac{2}{3\Delta}\right\rfloor\ge 1$ and consider a weighted game with $2a$ passers and $n-2a$ null players. One representation is given by $q=\Delta/2$, $w_{2i-1}=\Delta$, $w_{2i}=\Delta/2$ for $1\le i\le a$, $w_{2a+1}=w_{2a+3}=w_{2a+5}=\frac{1}{3}-\frac{a\Delta}{2}\ge 0$, $w_{2a+2}=w_{2a+4}=w_{2a+6}=0$, and $w_i=0$ for all $2a+7\le i\le n$. By assumption we have $n\ge \tfrac{4}{3\Delta}+6\ge 2a+6$ and the first $2a$ players are obviously passers. By checking $0\le \frac{1}{3}-\frac{a\Delta}{2}<\frac{\Delta}{2}$ we conclude that the remaining players are null players and have a non-negative weight. By construction, the weights of the $n$ players sum up to one. Changing the weights of player $2i-1$ and player $2i$ for $1\le i\le a$ does not change the game so that we obtain a second representation with quota $q$ and weights $\bar{w}_{2i}=\Delta$, $\bar{w}_{2i-1}=\Delta/2$ for $1\le i\le a$, $\bar{w}_{2a+2}=\bar{w}_{2a+4}=\bar{w}_{2a+6}=\frac{1}{3}-\frac{a\Delta}{2}\ge 0$, $w_{2a+1}=w_{2a+3}=w_{2a+4}=\bar{w}_{2a+1}=\bar{w}_{2a+2}=\bar{w}_{2a+3}=0$, and $\bar{w}_i=0$ for all $2a+7\le i\le n$. With this, we have $\Vert w-\bar{w}\Vert_{1}=a\Delta +2-3a\Delta=2(1-a\Delta)\ge\frac{2}{3}$ and $\Vert w-\bar{w}\Vert_{\infty}=\Delta/2$.
For each $w,\bar{w}\in\mathbb{R}_{\ge 0}^n$ with $\Delta(w)=\Delta(\bar{w})$, we obviously have $\Vert w-\bar{w}\Vert_\infty\le \Delta(w)$. So, a constant lower bound for the $\Vert\cdot\Vert_\infty$-distance can only exist if we slightly weaken the assumptions as done in Lemma \[lemma\_lb\_approximation\_delta\].
In some applications only weighted games with a quota of at least one half are considered, which clashes with some of our constructions in the proofs of the previous lemmas. However, by considering the dual of a given weighted game we can turn a quota below one half to a quota above one half, see Lemma \[lemma\_dual\_weights\]. So, instead of small quotas we get large quotas.
So, either knowing the relative quota or the maximum relative weight is not sufficient in order to deduce a non-constant upper bound on the diameter of the weight polytope for a suitably large number of players. However, as we will see in the next section, knowing the relative quota and the maximum relative weight is indeed sufficient for such an upper bound, see Theorem \[thm\_weighted\_representation\]. Our next aim is to show that this upper bound is tight up to a constant.
\[lemma\_lb\_diam\_representation\_polytop\] For each $0<q<1$, $0<\Delta\le 1$, and each integer $n\ge \tfrac{1}{\Delta}+2$ there exist weight vectors $w,\bar{w}\in\mathbb{R}^n_{\ge 0}$ with $\Vert w\Vert_1=\Vert\bar{w}\Vert_1=1$, $\Delta(w)=\Delta$ and a quota $0<\bar{q}\le 1$ with $[q;w]=[\bar{q};\bar{w}]$ such that $$\Vert w-\bar{w} \Vert_1\ge \frac{1}{200}\cdot \min\left\{2, \frac{4\Delta}{\min\{q,1-q\}}\right\}.$$ Under the same assumptions there exist weight vectors $w,\bar{w}\in\mathbb{R}^n_{\ge 0}$ with $\Vert w\Vert_1=\Vert\bar{w}\Vert_1=1$, $\Delta(w)=\Delta$ and a quota $0<\bar{q}\le 1$ with $[q;w]=[\bar{q};\bar{w}]$ such that $\Vert w-\bar{w} \Vert_\infty\ge \frac{\Delta}{5}$.
We set $a=\left\lfloor\tfrac{1}{\Delta}\right\rfloor\ge 1$ and choose the unique integer $b$ with $b\Delta<q$ and $(b+1)\Delta\ge q$. With this we set $k=b+1\ge 1$ and $w=(\Delta,\dots,\Delta,1-a\Delta,0,\dots,0)$, where $0\le 1-a\Delta<\Delta$, so that $w\in\mathbb{R}_{\ge 0}^n$ and $\Vert w\Vert_1=1$. If $b\Delta+(1-a\Delta)<q$ we set $s=a$ and $s=a+1$ otherwise, so that $[q;w]=v_{k,s,n-s}$. Note that $n-s\ge 1$.
If $k=s$, then Lemma \[lemma\_diameter2\] gives $\operatorname{diam}(\mathsf{W}(v_{s,s,t}))\ge \frac{2}{3}$, so that the triangle inequality implies the existence of a vector $w'\in \mathsf{W}(v_{s,s,t})$ with $\Vert w-w'\Vert_1\ge \frac{1}{3}$. If $k<s$, then Lemma \[lemma\_diameter\] gives $\operatorname{diam}(\mathsf{W}(v_{k,s,t}))\ge \max\left\{\frac{1}{10k},\frac{1}{10(s-k)}\right\}$, so that the triangle inequality implies the existence of a vector $w'\in \mathsf{W}(v_{k,s,t})$ with $$\Vert w-w'\Vert_1\ge \max\left\{\frac{1}{20k},\frac{1}{20(s-k)}\right\}=\frac{1}{20s}\cdot\frac{1}{\min\left\{\frac{k}{s},\frac{s-k}{s}\right\}}.$$ In the following we make several case distinctions for the subcase $k<s$.
If $k=1$ or $s-k=1$, then $\Vert w-w'\Vert_1\ge\frac{1}{20}$. In the following we assume $k\ge 2$ and $s-k\ge 2$. By construction we have $\tfrac{k}{2}\le (k-1)\Delta<q$, $k\Delta\ge q$, and $(s-1)\Delta\le 1$, so that $k<\frac{2q}{\Delta}$, $\tfrac{s-k}{2}\Delta\le (s-1)\Delta-k\Delta\le 1-q$ and $s-k\le \frac{2(1-q)}{\Delta}$.
If $k\le s-k$, i.e., $2k\le s$, then $q\le \tfrac{1}{2}$ and $$\Vert w-w'\Vert_1\ge \frac{1}{20s}\cdot\frac{1}{\min\left\{\frac{k}{s},\frac{s-k}{s}\right\}}
=\frac{1}{20k}\ge \frac{1}{40}\cdot \frac{\Delta}{q}=\frac{1}{40}\cdot \frac{\Delta}{\min\{q,1-q\}}.$$ If $k> s-k$, i.e., $2k>s$, then $q> \tfrac{1}{2}$ and $$\Vert w-w'\Vert_1\ge \frac{1}{20s}\cdot\frac{1}{\min\left\{\frac{k}{s},\frac{s-k}{s}\right\}}
=\frac{1}{20(s-k)}\ge \frac{1}{40}\cdot \frac{\Delta}{1-q}=\frac{1}{40}\cdot \frac{\Delta}{\min\{q,1-q\}}.$$ Thus, $$\Vert w-w' \Vert_1\ge \frac{1}{160}\cdot \min\left\{2, \frac{4\Delta}{\min\{q,1-q\}}\right\}$$ in all cases. If $w'$ is on the boundary of $\mathsf{W}(v_{k,s,n-s})$, then we slightly perturb $w'$ to $\bar{w}$ in the interior of $\mathsf{W}(v_{k,s,n-s})$ and choose a quota $\bar{q}\in(0,1]$ such that $[\bar{q};\bar{w}]=v_{k,s,n-s}$. This gives the statement for the $\Vert\cdot\Vert_1$-distance, if the pertubation is small enough to be covered by our decrease of the factor $\tfrac{1}{160}$ to $\tfrac{1}{200}$.
For the $\Vert\cdot\Vert_\infty$-distance we choose $w$ with $[q;w]=v_{k,s,n-s}$ as above. If $k=s$, then Lemma \[lemma\_diameter2\] gives $\operatorname{diam}^\infty(\mathsf{W}(v_{s,s,t}))\ge \frac{1}{3}$, so that the triangle inequality implies the existence of a vector $w'\in \mathsf{W}(v_{s,s,t})$ with $\Vert w-w'\Vert_\infty\ge \frac{1}{6}$. If $k<s$, then Lemma \[lemma\_diameter\] gives $\operatorname{diam}^\infty(\mathsf{W}(v_{k,s,t}))\ge \frac{1}{s}$, so that the triangle inequality implies the existence of a vector $w'\in \mathsf{W}(v_{k,s,t})$ with $\Vert w-w'\Vert_\infty\ge \tfrac{1}{2s}$. For $s=1$ this gives $\Vert w-w'\Vert_\infty\ge \tfrac{1}{2}$. For $s\ge 2$ we have $s\le\tfrac{2}{\Delta}$ so that $\Vert w-w'\Vert_\infty\ge \tfrac{\Delta}{4}$. Since $\Delta\le 1$ we have $\Vert w-w'\Vert_\infty\ge \tfrac{\Delta}{4}$ in all cases, so that the stated result follows possibly by a perturbation.
Upper bounds for the diameter of the weight polytope {#sec_upper_bounds}
====================================================
Before we start to upper bound $\operatorname{diam}(\mathsf{W}(v))$ in terms of $\Delta$ and $q$, we provide a slightly more general result.
\[lemma\_qdelta\_bound\_winning\] Let $w\in\mathbb{R}^n_{\ge 0}$ with $\Vert w\Vert_1=1$ for an integer $n\in\mathbb{N}_{>0}$ and $0<q<1$. For each $x\in\mathbb{R}^n_{\ge 0}$ with $\Vert x\Vert_1=1$ and $x(S)=\sum_{s\in S}x_s\ge q$ for every winning coalition $S$ of $[q;w]$, we have $$\Vert w-x\Vert_1 \le \frac{2\Delta}{\min\{q+\Delta,1-q\}}\le\frac{2\Delta}{\min\{q,1-q\}},$$ where $\Delta=\Delta(w)$.
Consider a winning coalition $T$ such that $x(T)$ is minimal and invoke $x(T)\ge q$, see the appendix for the technical details.
From Lemma \[lemma\_qdelta\_bound\_winning\] we can directly conclude:
Let $w,\bar{w}\in\mathbb{R}^n_{\ge 0}$ with $\Vert w\Vert_1=\Vert \bar{w}\Vert_1=1$ for an integer $n\in\mathbb{N}_{>0}$ and $0<q, \bar{q}<1$. If $[q;w]=[\bar{q};\bar{w}]$, then we have $$\Vert w-\bar{w}\Vert_1 \le \max\left\{ \frac{2\Delta(w)}{\min\{q,1-q\}},
\frac{2\Delta(\bar{w})}{\min\{\bar{q},1-\bar{q}\}}\right\}
\le\frac{2\Delta(w)}{\min\{q,1-q\}}+\frac{2\Delta(\bar{w})}{\min\{\bar{q},1-\bar{q}\}}.$$
Unfortunately, this does not allow us to derive an upper bound of $\Vert w-\bar{w}\Vert_1$ which only depends on $q$ and $\Delta(w)$. However, we can obtain the following analog of Lemma \[lemma\_qdelta\_bound\_winning\] for losing instead of winning coalitions.
\[lemma\_qdelta\_bound\_losing\] Let $w\in\mathbb{R}^n_{\ge 0}$ with $\Vert w\Vert_1=1$, $\Delta=\Delta(w)$, and $0<q<1$. For each $x\in\mathbb{R}^n_{\ge 0}$ with $\Vert x\Vert_1=1$ and $x(S)=\sum_{s\in S}x_s\le q$ for every losing coalition $S$ of $[q;w]$, we have $$\Vert w-x\Vert_1 \le\frac{4\Delta}{\min\{q,1-q\}}.$$ Moreover, if $q>\Delta$, then $
\Vert w-x\Vert_1 \le \frac{2\Delta}{\min\{q-\Delta,1-q+\Delta\}}\le\frac{2\Delta}{\min\{q-\Delta,1-q\}}
$.
Consider a losing coalition $T$ such that $x(T)$ is maximal and invoke $x(T)\le q$. Technical details are provided in the appendix.
\[thm\_weighted\_representation\] Let $w,\bar{w}\in\mathbb{R}^n_{\ge 0}$ with $\Vert w\Vert_1=\Vert \bar{w}\Vert_1=1$, $\Delta=\Delta(w)$, and $0<q, \bar{q}<1$. If $[q;w]=[\bar{q};\bar{w}]$, then we have $$\Vert w-\bar{w}\Vert_1\le \min\left\{2,\frac{4\Delta}{\min\{q,1-q\}}\right\}\le\frac{4\Delta}{\min\{q,1-q\}},$$ i.e., $\operatorname{diam}(\mathsf{W}([q;w]))\le \frac{4\Delta(w)}{\min\{q,1-q\}}$. Moreover, if $q>\Delta$, then we have $$\Vert w-\bar{w}\Vert_1\le \frac{2\Delta}{\min\{q-\Delta,1-q\}}.$$
In Section \[sec\_weight\_polytope\] we have observed $\Vert w-\bar{w}\Vert_1\le 2$. If $\bar{q}\ge q$, then $\bar{w}(S)\ge \bar{q}\ge q$ for every winning coalition $S$ of $[q;w]$. Here, we can apply Lemma \[lemma\_qdelta\_bound\_winning\]. Otherwise we have $\bar{w}(T)<\bar{q}<q$ for every losing coalition $T$ of $[q;w]$ and Lemma \[lemma\_qdelta\_bound\_losing\] applies.
As an example we consider the normalized weight vector $w=\tfrac{1}{120}\cdot(15,14,\dots,1)$ and the quota $\tfrac{3}{5}$. Let $(\bar{q};\bar{w})$ be another normalized representation of the weighted game $[q;w]$, then the first bound gives $\Vert w-\bar{w}\Vert_1\le \tfrac{5}{4}$. Since $\Delta=\tfrac{1}{8}>q$, also the second bound applies yielding $\Vert w-\bar{w}\Vert_1\le \tfrac{5}{8}$. We remark that for this specific example the diameter $\operatorname{diam}(\mathsf{W}([q;w]))$ is much smaller than $\tfrac{5}{8}$.
Applications {#sec_applications}
============
A power index $\varphi$ is a mapping from the set of weighted games on $n$ players into $\mathbb{R}_{\ge 0}^n$. We call $\varphi$ efficient if $\Vert\varphi(v)\Vert_1=1$ for all weighted games $v$. The difference $\Vert w-\varphi([q;w])\Vert_1$ between relative weights and the corresponding power distribution is studied in the literature, see e.g. [@dubey1979mathematical; @kurz2014nucleolus; @neyman1982renewal]. Lemma \[lemma\_qdelta\_bound\_winning\] is a generalization of [@kurz2014nucleolus Lemma 1]: if $\varphi$ is the nucleolus, see e.g. [@schmeidler1969nucleolus], and $0<q<1$ then $$\label{ie_ub_nucleolus}
\Vert w-\varphi([q;w])\Vert_1 \le\frac{2\Delta(w)}{\min\{q,1-q\}}$$ for all $w\in\mathbb{R}_{\ge 0}^n$ with $\Vert w\Vert_1=1$. From Theorem \[thm\_weighted\_representation\] we directly conclude:
\[cor\_ub\_general\] Let $w\in\mathbb{R}^n_{\ge 0}$ with $\Vert w\Vert_1=1$ and $0<q<1$. If an efficient power index $\varphi$ permits the existence of a quota $q'\in(0,1)$ such that $[q';\varphi([q;w])]=[q;w]$, i.e., the power vector of the given weighted game can be completed to a representation of the same game, then $$\Vert w-\varphi([q;w])\Vert_1
\le \frac{4\Delta(w)}{\min\{q,1-q\}}.$$
*Representation compatibility* of $\varphi$ for $[q;w]$ is automatically satisfied for the modified nucleolus (modiclus) [@sudholter1996modified], minimum sum representation index [@freixas2014minimum] or one of the power indices based on averaged representations [@kaniovski2015average] for all weighted games and for the Penrose-Banzhaf index for all spherically separable simple games [@houy2014geometry]. The theorem also applies to the bargaining model for weighted games analyzed in [@market_value_model], cf. [@prop_payoffs]. It is unknown whether there exists a constant $c\in\mathbb{R}_{>0}$ such that $$\label{ie_ub_ssi}
\Vert w-\operatorname{SSI}([q;w])\Vert_1
\le \frac{c\Delta(w)}{\min\{q,1-q\}}.$$ holds for the Shapley-Shubik index $\operatorname{SSI}$ and all $w\in\mathbb{R}^n_{\ge 0}$ with $\Vert w\Vert_1=1$ and $0<q<1$. For the Penrose-Banzhaf index such a constant $c$ can not exist, see [@kurz2018note Proposition 2].
For the other direction we have:
\[lemma\_general\_lower\_distance\_bound\] Let $n\in\mathbb{N}_{>0}$, $q,\bar{q}\in(0,1]$, $w,\bar{w}\in\mathbb{R}^n_{\ge 0}$ with $\left\Vert w\right\Vert_1=
\left\Vert \bar{w}\right\Vert_1=1$ and $[q;w]=[\bar{q};\bar{w}]$, $\Vert\cdot\Vert$ be an arbitrary norm on $\mathbb{R}^n$ and $\varphi$ be a mapping from the set of weighted games (on $n$ players) into $\mathbb{R}_{\ge 0}^n$, then we have $$\max\left\{\left\Vert w-\varphi\left(\left[q;w\right]\right)\right\Vert,
\left\Vert \bar{w}-\varphi\left(\left[\bar{q};\bar{w}\right]\right)\right\Vert
\right\}\ge \frac{\left\Vert w-\bar{w}\right\Vert}{2}.$$
Using the triangle inequality yields $\left\Vert w-\varphi\left(\left[q;w\right]\right)\right\Vert+
\left\Vert \bar{w}-\varphi\left(\left[\bar{q};\bar{w}\right]\right)\right\Vert\ge
\left\Vert w-\bar{w}\right\Vert$ from which we can conclude the stated inequality.
\[prop\_impossible\] Let $\varphi$ be a mapping from the set of weighted games (on $n$ players) into $\mathbb{R}_{\ge 0}^n$.
- For each $q\in(0,1]$ and each integer $n\ge 2$ there exists a weighted game $[q;w]$, where $w\in\mathbb{R}^n_{\ge 0}$ and $\Vert w\Vert_1=1$, such that $\Vert w-\varphi([q;w])\Vert_{1}\ge \frac{1}{3}$ and $\Vert w-\varphi([q;w])\Vert_{\infty}\ge \frac{1}{6}$.
- For each $\Delta\in(0,1)$ and each integer $n\ge \frac{4}{3\Delta}+6$ there exists a weighted game $[q;w]$, where $q\in(0,1]$, $w\in\mathbb{R}^n_{\ge 0}$, $\Vert w\Vert_1=1$, and $\Delta(w)=\Delta$, such that $\Vert w-\varphi([q;w])\Vert_{1}\ge \frac{1}{3}$, and $\Vert w-\varphi([q;w])\Vert_{\infty}\ge\Delta/4$.
Combine Lemma \[lemma\_general\_lower\_distance\_bound\] with lemmas \[lemma\_lb\_approximation\_q\] and \[lemma\_lb\_approximation\_delta\_1\].
\[prop\_impossible2\] Let $\varphi$ be a mapping from the set of weighted games (on $n$ players) into $\mathbb{R}_{\ge 0}^n$. For each $q\in(0,1)$, $\Delta\in(0,1]$, there exist $w,\bar{w}\in\mathbb{R}_{\ge 0}^n$, $\bar{q}\in(0,1]$ with $\Vert w\Vert_1=\Vert\bar{w}\Vert_1=1$, $\Delta(w)=\Delta$, $[q;w]=[\bar{q};\bar{w}]$, and $$\Vert \bar{w}-\varphi([\bar{q};\bar{w}])\Vert_1\ge \frac{1}{200}\cdot \min\left\{2, \frac{4\Delta}{\min\{q,1-q\}}\right\}.$$
We construct $w$ as in the proof of Lemma \[lemma\_lb\_diam\_representation\_polytop\] and choose integers $k$, $s$, and $t$ such that $[q;w]=v_{k,s,t}$. In the proof of Lemma \[lemma\_lb\_diam\_representation\_polytop\] we have actually verified $$\operatorname{diam}(\mathsf{W}([q;w]))\ge \frac{1}{80}\cdot \min\left\{2, \frac{4\Delta}{\min\{q,1-q\}}\right\}=:\Lambda.$$ Now choose $w',w''\in \mathsf{W}([q;w])$ with $\Vert w'-w''\Vert_1\ge \Lambda$. By the triangle inequality we have either $\Vert w'-\varphi([q;w])\Vert_1\ge \Lambda/2$ or $\Vert w''-\varphi([q;w])\Vert_1\ge \Lambda/2$. By choosing $\bar{w}$ as $w'$ or $w''$ and eventually moving it into the interior of $\mathsf{W}([q;w])$ we obtain the stated result.
So, upper bounds for the $\Vert\cdot\Vert_1$-distance between normalized weights and a power distribution, as in Inequality (\[ie\_ub\_nucleolus\] or Inequality (\[ie\_ub\_ssi\]) are tight up to the constant $c$ if only the normalized quota and the normalized maximum weight are taken into account.
Conclusion {#sec_conclusion}
==========
In this paper we have introduced the concept of the diameter of the weight polytope of a weighted game. This number measures how diverse two different normalized weight vectors, representing the same given game, can be. In Theorem \[thm\_weighted\_representation\] we have shown that $$\operatorname{diam}(\mathsf{W}([q;w]))\le \min\left\{2,\frac{4\Delta}{\min\{q,1-q\}}\right\}\le\frac{4\Delta}{\min\{q,1-q\}},$$ for any $q\in(0,1)$ and any $w\in\mathbb{R}_{\ge 0}^n$ with $\Vert w\Vert_1=1$. Lemma \[lemma\_lb\_diam\_representation\_polytop\] certifies that this upper bound is in general, i.e., in the worst case, tight up to a constant. (This paper traded smaller constants for easier proofs.) The super-exponential growth of the number of weighted games (see [@zuev1989asymptotics]) indicates that this is not the case for the majority of weighted games. Thus, it would be interesting to determine other parameters of a representation of a weighted game that permit tight upper bounds on the diameter of the corresponding weight polytope. Another possible line for future research is to consider games with a priori unions, spatial games, or games with restricted communication.
As shown in Section \[sec\_applications\], there are connections to approximations of power indices by weight vectors. Proposition \[prop\_impossible2\] gives a partial explanation for the conditions of the main theorem of [@neyman1982renewal] on a limit result for the Shapley-Shubik index. Moreover, for a general power index it shows that upper bounds for the $\Vert\cdot\Vert_1$-distance between normalized weights and a power distribution, taking only the normalized quota and the normalized maximum weight into account, as in Corollary \[cor\_ub\_general\], would be tight up to a constant.
Acknowledgment {#acknowledgment .unnumbered}
==============
The author would like to thank the anonymous referees of a previous submission for their very helpful remarks and suggestions.
Delayed proofs {#sec_evacuated_proofs}
==============
(Lemma \[lemma\_relation\_maximum\_laakso\_taagepera\])\
For $n=1$, we have $w_1=1$, $\Delta(w)=1$, $\alpha=0$, and $L(w)=1$, so that we assume $n\ge 2$ in the remaining part of the proof. For $w_i\ge w_j$ consider $a:=\frac{w_i+w_j}{2}$ and $x:=w_i-a$, so that $w_i=a+x$ and $w_j=a-x$. With this we have $w_i^2+w_j^2=2a^2+2x^2$ and $(w_i+y)^2+(w_j-y)^2=2a^2+2(x+y)^2$. Let us assume that $w^\star$ minimizes $\sum_{i=1}^n w_i^2$ under the conditions $w\in\mathbb{R}_{\ge 0}$, $\Vert w\Vert_1=1$, and $\Delta(w)=\Delta$. (Since the target function is continuous and the feasible set is compact and non-empty, a global minimum indeed exists.) W.l.o.g. we assume $w_1^\star=\Delta$. If there are indices $2\le i,j\le n$ with $w_i^\star>w_j^\star$, i.e., $x>0$ in the above parameterization, then we may choose $y=-x$. Setting $w_i':=w_i^\star+y=a=\frac{w_i^\star+w_j^\star}{2}$, $w_j':=w_j^\star-y=a
=\frac{w_i^\star+w_j^\star}{2}$, and $w_h':=w_h^\star$ for all $1\le h\le n$ with $h\notin\{i,j\}$, we have $w'\in \mathbb{R}_{\ge 0}^n$, $\Vert w'\Vert_1=1$, $\Delta(w')=\Delta$, and $\sum_{h=1}^n \left(w_h'\right)^2=\sum_{h=1}^n \left(w_h^\star\right)^2\,-\,x^2$. Since this contradicts the minimality of $w^\star$, we have $w_i^\star=w_j^\star$ for all $2\le i,j\le n$, so that we conclude $w_i^\star=\frac{1-\Delta}{n-1}$ for all $2\le i\le n$ from $1=\Vert w^\star\Vert_1=\sum\limits_{h=1}^n w_h^\star$. Thus, $L(w)\le 1/\left(\Delta^2+\frac{(1-\Delta)^2}{n-1}\right)$, which is tight. Since $\Delta\le 1$ and $n\ge 2$, we have $1/\left(\Delta^2+\frac{(1-\Delta)^2}{n-1}\right)\le \frac{1}{\Delta^2}$, which is tight if and only if $\Delta=1$, i.e., $n-1$ of the weights have to be equal to zero.
Now, let us assume that $w$ maximizes $\sum_{i=1}^n w_i^2$ under the conditions $w\in\mathbb{R}_{\ge 0}$, $\Vert w\Vert_1=1$, and $\Delta(w)=\Delta$. (Due to the same reason a global maximum indeed exists.) Due to $1=\Vert w\Vert_1\le n\Delta$ we have $0<\Delta\le 1/n$, where $\Delta=1/n$ implies $w_i=\Delta$ for all $1\le i\le n$. In that case we have $L(w)=n$ and $\alpha=0$, so that the stated lower bounds for $L(w)$ are valid. In the remaining cases we assume $\Delta>1/n$. If there would exist two indices $1\le i,j\le n$ with $w_i\ge w_j$, $w_i<\Delta$, and $w_j>0$, we may strictly increase the target function by moving weight from $w_j$ to $w_i$ (this corresponds to choosing $y>0$), by an amount small enough to still satisfy the constraints $w_i\le \Delta$ and $w_j\ge 0$. Since $\Delta>0$, we can set $a:=\lfloor 1/\Delta\rfloor\ge 0$ with $a\le n-1$ due to $\Delta>1/n$. Thus, for a maximum solution, we have exactly $a$ weights that are equal to $\Delta$, one weight that is equal to $1-a\Delta\ge 0$ (which may indeed be equal to zero), and $n-a-1$ weights that are equal to zero. With this and $a\Delta=1-\alpha\Delta$ we have $
\sum_{i=1}^n w_i^2=a\Delta^2 (1-a\Delta)^2=\Delta-\alpha\Delta^2+\alpha^2\Delta^2=\Delta(1-\alpha\Delta+\alpha^2\Delta)
=\Delta\left(1-\alpha(1-\alpha)\Delta\right)\le\Delta
$. Here, the latter inequality is tight if and only if $\alpha=0$, i.e., $1/\Delta\in\mathbb{N}$.
(Lemma \[lemma\_qdelta\_bound\_winning\])\
We set $N=\{1,\dots,n\}$, $w(U)=\sum_{u\in U} w_u$ and $x(U)=\sum_{u\in U} x_u$ for each $U\subseteq N$. Let $S^+=\{i\in N \mid x_i>w_i\}$ and $S^-=\{i\in N\mid x_i\le w_i\}$, i.e., $S^+$ and $S^-$ partition the set $N$ of players. We have $w(S^+)<1$ since $w(S^+)<x(S^+)\le x(N)=1$, so that $w(S^-)>0$. Define $0\le \delta\le 1$ by $x(S^-)=(1-\delta)w(S^-)$. We have $$x(S^+)=1-x(S^-)=w(S^+)+w(S^-)-(1-\delta)w(S^-)=w(S^+)+\delta w(S^-)$$ and $$\label{eq_dist_delta}
\Vert w-x\Vert_1
=\left(x(S^+)-w(S^+)\right)+\left(w(S^-)-x(S^-)\right)=2\delta w(S^-).$$ Generate a set $T$ by starting at $T=\emptyset$ and successively add a remaining player $i$ in $N\backslash T$ with minimal $x_i/w_i$, where all players $j$ with $w_j=0$ are the worst ones. Stop if $w(T)\ge q$. By construction $T$ is a winning coalition of $[q;w]$ with $w(T)<q+\Delta$, since the generating process did not stop earlier and $w_j\le \Delta(w)$ for all $j\in N$.
If $w(S^-)\ge q$, we have $T\subseteq S^-$ and $x(T)/w(T)\le x(S^-)/w(S^-)=1-\delta$. Multiplying by $w(T)$ and using $w(T)<q+\Delta$ yields $$x(T)\le (1-\delta)w(T)<(1-\delta)(q+\Delta)= (1-\delta)q +(1-\delta)\Delta.$$ Since $x(T)\ge q$, as $T$ is a winning coalition, we conclude $\delta<\Delta/(q+\Delta)$. Using this and $w(S^-)<1$ in Equation (\[eq\_dist\_delta\]) yields $$\Vert w-x\Vert_1<\frac{2\Delta}{q+\Delta}<\frac{2\Delta}{q}.$$
If $w(S^-)< q$, we have $S^-\subseteq T$, $x(T)=x(S^-)+x(T\backslash S^-)$, $w(T\backslash S^-)>0$, and $w(S^+)>0$. Since $T\backslash S^-\subseteq S^+$, $x(T\backslash S^-)/w(T\backslash S^-)\le x(S^+)/w(S^+)$, so that $$\begin{aligned}
x(T)&=&x(S^-)+x(T\backslash S^-)\le (1-\delta)w(S^-)+\frac{x(S^+)}{w(S^+)}\cdot\left(w(T)-w(S^-)\right)\\
&\le& (1-\delta)w(S^-)+\frac{x(S^+)}{w(S^+)}\cdot\left(q+\Delta-w(S^-)\right)\\
&=& q+\frac{x(S^+)\Delta-(1-q)\delta w(S^-)}{w(S^+)}\\
&\le & q+\frac{\Delta-(1-q)\delta w(S^-)}{w(S^+)}.
\end{aligned}$$ Since $x(T)\ge q$, we conclude $(1-q)\delta w(S^-)\le \Delta$, so that $\Vert w-x\Vert_1\le \frac{2\Delta}{1-q}$.
(Lemma \[lemma\_qdelta\_bound\_losing\])\
If $q\le 2\Delta$, then $\frac{4\Delta}{\min\{q,1-q\}}\ge \frac{4\Delta}{q}\ge 2\ge \Vert x-w\Vert_1$, so that we can assume $q>\Delta$.
Using the notation from the proof of Lemma \[lemma\_qdelta\_bound\_winning\], we have $x(S^+)=w(S^+)+\delta w(S^-)$ and $\Vert w-x\Vert_1=2\delta w(S^-)$.
Generate $T$ by starting at $T=\emptyset$ and successively add a remaining player $i$ in $N\backslash T$ with maximal $x_i/w_i$, where all players $j$ with $w_j=0$ are taken in the first rounds, as long as $w(T)+w_i< q$. By construction $T$ is a losing coalition of $[q;w]$ with $q-\Delta\le w(T)<q$, since the generating process did not stop earlier. If $w(S^+)\ge q$, we have $T\subseteq S^+$ and $x(T)/w(T)\ge x(S^+)/w(S^+)=1+\frac{\delta w(S^-)}{w(S^+)}\ge 1+\delta w(S^-)$. Multiplying by $w(T)$ and using $w(T)\ge q-\Delta$ yields $$x(T)\ge \left(1+\delta w(S^-)\right)w(T)\ge\left(1+\delta w(S^-)\right)(q-\Delta)= (q-\Delta) +\delta w(S^-)(q-\Delta).$$ Since $x(T)\le q$, as $T$ is a losing coalition, we conclude $\delta w(S^-)\le\Delta/(q-\Delta)$, so that $\Vert w-x\Vert_1<\frac{2\Delta}{q-\Delta}$.
If $w(S^+)< q$, we have $S^+\subseteq T$, $x(T)=x(S^+)+x(T\backslash S^+)$, $w(T\backslash S^+)>0$, and $w(S^-)>0$. Since $T\backslash S^+\subseteq S^-$, $x(T\backslash S^+)/w(T\backslash S^+)\ge x(S^-)/w(S^-)$, so that $$\begin{aligned}
x(T)&=&x(S^+)+x(T\backslash S^+)\ge w(S^+)+\delta w(S^-) +\frac{x(S^-)}{w(S^-)}\cdot\left(w(T)-w(S^+)\right)\\
&\ge& w(S^+)+\delta w(S^-) +(1-\delta)\cdot\left(q-\Delta-w(S^+)\right)\\
&=& \delta w(S^-)+q-\Delta-\delta q+\delta \Delta+\delta w(S^+)
= q-\Delta+\delta(1-q+\Delta).
\end{aligned}$$ Since $x(T)\le q$, $\delta\le \frac{\Delta}{1-q+\Delta}$, so that $\Vert w-x\Vert_1\le \frac{2\Delta}{1-q+\Delta}$ due to $w(S^-)\le 1$.
So, for $q>\Delta$ we have $\Vert w-x\Vert_1 \le \frac{2\Delta}{\min\{q-\Delta,1-q+\Delta\}}\le \frac{2\Delta}{\min\{q-\Delta,1-q\}}$. In order to show $\Vert w-x\Vert_1\le \frac{4\Delta}{\min\{q,1-q\}}$ it remains to consider the case $q\le 1-q$. For $q>2\Delta$, see the start of the proof, we have $\Vert w-x\Vert_1\le \frac{2\Delta}{\min\{q-\Delta,1-q\}}\le \frac{2\Delta}{q-\Delta}\le
\frac{4\Delta}{q}\le \frac{4\Delta}{\min\{q,1-q\}}$.
[10]{}
L. Babai, K. A. Hansen, V. V. Podolskii, and X. Sun. Weights of exact threshold functions. In [*International Symposium on Mathematical Foundations of Computer Science*]{}, pages 66–77. Springer, 2010.
J. F. Banzhaf III. Weighted voting doesn’t work: a mathematical analysis. , 19:317, 1964.
G. Chalkiadakis, E. Elkind, and M. Wooldridge. Computational aspects of cooperative game theory. , 5(6):1–168, 2011.
B. Curry and K. D. George. Industrial concentration: a survey. , 31(3):203–255, 1983.
P. Dubey and L. S. Shapley. Mathematical properties of the [B]{}anzhaf power index. , 4(2):99–131, 1979.
J. Freixas and S. Kaniovski. The minimum sum representation as an index of voting power. , 233(3):739–748, 2014.
N. Houy and W. S. Zwicker. The geometry of voting power: weighted voting and hyper-ellipsoids. , 84:7–16, 2014.
S. Kaniovski and S. Kurz. The average representation –a cornucopia of power indices? , 32(2):169–181, 2015.
S. Kaniovski and S. Kurz. Representation-compatible power indices. , 264(1):235–265, 2018.
S. Kurz. A note on limit results for the [P]{}enrose-[B]{}anzhaf index. , 2018.
S. Kurz, S. Napel, and A. Nohn. The nucleolus of large majority games. , 123(2):139–143, 2014.
M. Laakso and R. Taagapera. Effective number of parties: A measure with application to [W]{}estern [E]{}urope. , 12(1):3–27, 1979.
M. Laakso and R. Taagepera. Proportional representation and effective number of parties in [F]{}inland. In [*Power, voting, and voting power*]{}, pages 107–120. Springer, 1982.
D. Leech. Ownership concentration and the theory of the firm: a simple-game-theoretic approach. , pages 225–240, 1987.
M. Montero. Proportional payoffs in legislative bargaining with weighted voting: a characterization. , 12(3):325–346, 2017.
A. Neyman. Renewal theory for sampling without replacement. , pages 464–481, 1982.
G. Owen, I. Lindner, S. L. Feld, B. Grofman, and L. Ray. A simple market value bargaining model for weighted voting games: characterization and limit theorems. , 35(1):111–128, 2006.
D. Schmeidler. The nucleolus of a characteristic function game. , 17(6):1163–1170, 1969.
L. S. Shapley and M. Shubik. A method for evaluating the distribution of power in a committee system. , 48(3):787–792, 1954.
P. Sudh[ö]{}lter. The modified nucleolus as canonical representation of weighted majority games. , 21(3):734–756, 1996.
J. von Neumann and O. Morgenstern. . Princeton University Press, 3rd edition, 1953.
Y. A. Zuev. Asymptotics of the logarithm of the number of threshold functions of the algebra of logic. , 39(3):512–513, 1989.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'It is believed that curvature singularities are a creation of general relativity and hence, in the absence of a quantum gravity, models of nonsingular black holes have received significant attention. We study the shadow (apparent shape), an optical appearance because of its strong gravitational field, cast by a nonsingular black hole which is characterized by three parameters, i.e., mass ($M$), spin ($a$), and a deviation parameter ($k$). The nonsingular black hole under consideration, is a generalization of the Kerr black hole [that]{} can be recognized asymptotically ($r>>k, k>0$) explicitly as the Kerr-Newman black hole, and in the limit $k \rightarrow 0$ as the Kerr black hole. It turns out that the shadow of a nonsingular black hole is a dark zone covered by [a]{} deformed circle. Interestingly, it is seen that the shadow of a black hole is affected due to the parameter $k$. Indeed, for a given $a$, the size of a shadow reduces as the parameter $k$ increases and the shadow becomes more distorted as we increase the value of the parameter $k$ when compared with the analogous Kerr black hole shadow. We also investigate, in detail, how the ergoregion of a black hole is changed due to the deviation parameter $k$.'
author:
- Muhammed Amir
- 'Sushant G. Ghosh'
title: Shapes of rotating nonsingular black hole shadows
---
Introduction
============
The determination of black hole spin is an open issue in astrophysics that has received significant attention using various methods [@Tanaka:1995en; @Fabian:1995qz; @Genzel:2003as; @Connars:1980; @Asada:2003nf; @Tanaka:1996ht; @Finn:2000sy]. It turns out that [the astronomical]{} observation of a black hole shadow (apparent shape) may significantly help us to resolve this important open issue. It is widely believed that the black hole shadow may contain information about the nature of a black hole, and it can help us to determine the black hole spin [@Takahashi:2004xh]. It is [shown]{} that the shape of a rotating black hole is deformed in an optically thin emitting medium due to the presence of spin [@Bardeen:1973gb; @Chandrasekhar:1992; @Falcke:1999pj], which for the Schwarzschild black hole or a nonrotating black hole is circular [@Bozza:2009yw].
It is also believed that a supermassive black holes exist in the center of a galaxy. Falcke [@Falcke:1999pj] determined the formation of images of supermassive black holes to demonstrate that a strong gravitational field bends the trajectories of photons released by [a]{} bright object and an observer can see a dark zone around a black hole. In 2002, Holz and Wheeler [@Holz:2002uf] detected the retro-images or retro-MACHOS of the sun by a Schwarzschild black hole. In [@Takahashi:2004xh], the author discussed the qualitative and quantitative analyses of the shape and position of the rotating black hole shadow on an optically thin accretion disk and its dependence on the angular momentum.
A black hole casts a shadow as an optical appearance because of its strong gravitational field, which for the Schwarzschild black hole is investigated in the pioneer study of Synge [@Synge:1966] and Luminet [@Luminet:1979] who also suggested a formula to calculate the angular radius of the shadow [@Synge:1966]. It is shown that a shadow of the Schwarzschild black hole is a perfect circle [@Bozza:2009yw] through the gravitational lensing either in vacuum [@Virbhadra:2008ws] or in plasma [@Morozova:2013dxb; @Bisnovatyi-Kogan:2015dxa]. Bardeen [@Bardeen:1973gb] studied the appearance of a shadow cast by the Kerr black hole (see also [@Chandrasekhar:1992]). It can be seen for the Kerr black hole that the shadow is no longer circular; it has a deformed or distorted shape in the direction of rotation [@Bozza:2007gt; @Chandrasekhar:1992; @Falcke:1999pj; @Zakharov:2005ek; @Nedkova:2013msa; @Bambi:2008jg]. The shadow of the Kerr black hole or a Kerr naked singularity by constructing observables has been discussed by Hioki and Maeda [@Hioki:2009na]. The shadows of several black holes and naked singularities, have been rigorously studied in the last few years, e.g., for the Kerr-Newman black hole [@Takahashi:2005hy], Einstein-Maxwell-Dilaton-Axion black hole [@Wei:2013kza], Kerr-Taub-NUT black hole [@Abdujabbarov:2012bn], rotating braneworld black hole [@Amarilla:2011fx], Kaluza-Klein rotating dilaton black hole [@Amarilla:2013sj], rotating non-Kerr black hole [@Atamurotov:2013sca], Kerr-Newman-NUT black holes with cosmological constant [@Grenzebach:2014fha], and the Kerr black hole with scalar hair [@Cunha:2015yba]. A coordinate-independent characterization of a black hole shadow is discussed in [@Abdujabbarov:2015xqa]. The study of shadows has also been extended for the five-dimensional rotating Myers-Perry black hole [@Papnoi:2014aaa]. Recently, significant attention has been devoted to studying various [nonsingular]{} models [@Bambi:2014nta; @Stuchlik:2014qja; @Ghosh:2014mea; @Schee:2015nua; @Amir:2015pja; @Ghosh:2015pra] including its shadow [@Li:2013jra; @Tinchev:2015apf; @Abdujabbarov:2016hnw].
In the absence of a well-defined quantum gravity, models of nonsingular black holes have received significant attention [@Ansoldi:2008jw; @Bambi:2013ufa; @Toshmatov:2014nya; @Ghosh:2014hea]. In contrast to classical black holes where singularities are unavoidable, the nonsingular black holes [also]{} have horizons, but there is no singularity and their metrics as well as their curvature invariants are well behaved everywhere. Bardeen [@Bardeen] proposed the first nonsingular black hole which is a solution of the Einstein equations coupled to an electromagnetic field, yielding an alteration of the Reissner-Nordstr[ö]{}m metric. However, the physical source associated with a Bardeen solution was obtained much later by Ay[ó]{}n-Beato and Garc[í]{}a [@ABG99]. Later, the exact self-consistent solutions for the nonsingular black hole for the dynamics of gravity coupled to nonlinear electrodynamics was obtained [@AGB; @AGB1; @AGB2; @regular; @regular1; @regular2; @regular3; @Hayward; @Culetu:2014lca; @lbev; @Balart:2014cga; @Xiang]. Subsequently, there has been an intense activity in the investigation of nonsingular black holes [@regular; @regular1; @regular2; @regular3; @Hayward; @Culetu:2014lca; @lbev; @Balart:2014cga; @Xiang]. Recently, Ghosh [@Ghosh:2014pba] employed a probability distribution inspired mass function $m(r)$ to replace the Kerr black hole mass $ M $ to obtain a rotating nonsingular black hole. It has an additional deviation parameter $k$ which is identified asymptotically ($r>>k$) exactly as the Kerr-Newman black hole and as the Kerr black hole when $k=0$. Thus, the rotating nonsingular black hole is a generalization of the Kerr black hole. The subject of this paper is to study the size and apparent shape of this rotating nonsingular black hole, and its changes due to the parameter $k$, by analyzing in detail the unstable circular orbits. Thus, we have studied the effect of the parameter $k$ on the shape of a shadow of the Kerr black hole. As in case of the Kerr black hole, we use two observables$-$the radius $R_s$ and the distortion parameter $\delta_s$, characterizing the apparent shape. We found that these observables and, hence, the shape of the shadows are affected by the parameter $k$. Interestingly, the size of the shadow, decreases with parameter $k$ for a given value of $a$ in rotating nonsingular black hole, thereby suggesting a smaller shadow than in the Kerr black hole. Thus, the larger value of the deviation parameter $k$ leads to a decrease in the size of the shadow, and thus $R_s$ decreases with an increase in $k$. The distortion of the shadow is characterized by the observable $\delta_s$ which for the case of a rotating nonsingular black hole increases monotonically with the parameter $k$. Thus, the black hole becomes more distorted or deformed with the increase in parameter $k$ for a given value of $a$.
The paper is organized as follows: In Sec. \[spacetime\], we briefly review the rotating nonsingular black hole and the study of the ergoregion for this black hole. In Sec. \[geodesics\], we obtain the geodesics of the photon and also discuss the effective potential. In Sec. \[shadow\], we obtain the apparent shapes of the rotating nonsingular black hole and calculate the observables, and in Sec. \[EE rate\], we study the energy emission rate of the black hole. Finally, we conclude in Sec. \[conclusion\].
Rotating nonsingular black holes {#spacetime}
================================
The metric of the rotating nonsingular black holes was obtained in Ref. [@Ghosh:2014pba], which in Boyer-Lindquist coordinates ($t$, $r$, $\theta$, $\phi$), with gravitational units $G=c=1$, reads $$\begin{aligned}
\label{metric}
ds^2 &=& - \left( 1- \frac{2Mr e^{-k/r}}{\Sigma} \right) dt^2 + \frac{\Sigma}{\Delta}dr^2 + \Sigma d \theta^2 \nonumber\\
&-& \frac{4aMr e^{-k/r}}{\Sigma } \sin^2 \theta dt d\phi \nonumber \\
& + & \left[r^2+ a^2 + \frac{2M r a^2 e^{-k/r}}{\Sigma} \sin^2 \theta \right] \sin^2 \theta d\phi^2,\end{aligned}$$ where $\Sigma$ and $\Delta$ are given by $$\begin{aligned}
\Sigma = r^2 + a^2 \cos^2\theta \;\;\; \text{and}\;\;\; \Delta=r^2 + a^2 - 2 M r e^{-k/r}.\end{aligned}$$ The parameters $M$, $a$, and $k$ ($k>0$), respectively, represent the mass, spin, and deviation parameter. The metric (\[metric\]) reduces to the Kerr solution [@Kerr:1963ud] if the parameter $k=0$ and to the Schwarzschild solution for both $k=a=0$ [@schw]. The detailed analysis of the horizon structure and the energy conditions of the metric (\[metric\]) are also discussed [@Ghosh:2014pba]. It is noted that the metric (\[metric\]) is singular at $\Delta=0$ and $\Sigma=0$. However, $\Delta=0$ is a coordinate singularity. It is seen that the metric (\[metric\]), for $r>>k$, can be identified as Kerr-Newman [@Newman:1965my], i.e., for $r>>k$, one gets $$\begin{aligned}
g_{tt}&=&1-\frac{2Mr-q^2}{\Sigma}+\mathcal{O}(k^2/r^2),\nonumber \\
\Delta &=& r^2+a^2-2Mr+q^2+\mathcal{O}(k^2/r^2), \nonumber\end{aligned}$$ where the charge $q$ and mass $M$ are related to the deviation parameter $k$ via $q^2=2Mk$.
The astrophysical black hole candidates are expected to be the Kerr black holes, but the actual nature of these astrophysical objects has still to be verified [@Bambi:2011mj]. The nonsingular metric can be seen as the prototype of a large class of non-Kerr black hole metrics, in which the components of the metric tensor have the same expressions as that of the Kerr black hole with $M$ replaced by a mass function $ m(r) $ that reduces to $ M $ as $r \rightarrow 0$ [@Bambi:2013sha]. Further, non-Kerr black holes or nonsingular ones may indeed look like Kerr black holes with different spin; e.g., the nonsingular ones with some nonzero values of the deviation parameter and spin parameter that are higher (lower) than the value of the spin parameter (say $a^*$) look like Kerr black holes with spin parameter higher (lower) than $a^*$. These black holes have an additional deviation parameter. The standard procedure to test the Kerr black hole with the observational x-ray data in Cygnus X-1 is developed in Refs. [@Gou:2011nq; @Gou:2013dna]. Thus, to test the nonsingular metric with the x-ray data of the black hole candidate in Cygnus X-1 would be to repeat the algorithm of Refs. [@Gou:2011nq; @Gou:2013dna] with the Kerr background replaced by the nonsingular black holes.
The procedure has been applied to test the Bardeen metric with the x-ray data of the black hole candidate in Cygnus X-1 with the Kerr black hole background replaced by the Bardeen [@Bambi:2014nta]. The spin parameter and the inclination angle can be determined by observing the apparent shape of the shadow, which is distorted mainly by these two parameters, defining some observables characterizing the apparent shape (its radius and distortion parameter). It has been shown that the spin parameter and the inclination angle of a Kerr black hole can be determined by the observation [@Hioki:2009na]. One can assume that the observations may allow non-Kerr black holes or the nonsingular black holes considered as real astrophysical black hole candidates; hence, the technique [@Hioki:2009na] can be used to determine how the spin parameter may change due to the additional parameter $k$.
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- -- -- --
{width="0.245\linewidth"} {width="0.245\linewidth"} {width="0.245\linewidth"} {width="0.245\linewidth"}
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- -- -- --
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- -- -- --
{width="0.245\linewidth"} {width="0.245\linewidth"} {width="0.245\linewidth"} {width="0.245\linewidth"}
{width="0.245\linewidth"} {width="0.245\linewidth"} {width="0.245\linewidth"} {width="0.245\linewidth"}
{width="0.245\linewidth"} {width="0.245\linewidth"} {width="0.245\linewidth"} {width="0.245\linewidth"}
{width="0.245\linewidth"} {width="0.245\linewidth"} {width="0.245\linewidth"} {width="0.245\linewidth"}
{width="0.245\linewidth"} {width="0.245\linewidth"} {width="0.245\linewidth"} {width="0.245\linewidth"}
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- -- -- --
Ergoregion
----------
The nonsingular rotating black hole metric (\[metric\]) or any other rotating nonsingular black hole, e.g., the rotating Bardeen black hole [@Bambi:2008jg], can be thought of as a non-Kerr black hole metric with $M$ replaced by $m(r)$, and it reduces to the standard Kerr black hole metric as $r \rightarrow \infty.$ Hence, the event horizon of the black hole is a null surface determined by $g^{rr}=\Delta=0$, and the zeros of $$\label{hor}
r^2 + a^2 - 2 M r e^{-k/r} =0,$$ give the black hole horizons. The numerical solution of Eq. (\[hor\]) reveals that, for a suitable choice of parameters, it admits two roots, $r^{-}$ and $r^{+}$, which correspond to the Cauchy and event horizons, respectively [@Ghosh:2014pba]. Furthermore, it is worthwhile to understand that one can find values of parameters such that $r^{+}=r^{-}$, which means one gets an extremal black hole, and $r^{+}>r^{-}$ describes a nonextremal black hole. The horizons exist only for a certain range of the parameters [@Ghosh:2014pba]. On the other hand, at the static limit surface, the timelike Killing vector $\xi^{a}=(\frac{\partial}{\partial t})^{a}$ becomes null, i.e., $$\xi^{a}\xi_{a}=g_{tt}=0,$$ or $$\label{sls}
r^2 + a^2 \cos^2 \theta - 2 M r e^{-k/r} =0.$$ By solving Eq. (\[sls\]) numerically, it turns out that the metric (\[metric\]) admits two positive roots (cf. [@Ghosh:2014pba] for details). The region between $r^{+}_{EH} < r < r^{+}_{SLS}$ is called the ergoregion. In an ergoregion the gravitational forces change the movement of an object, and an object in this region can no longer remain stationary. In this region, an object can escape the gravitational forces of the black hole if the speed of this object is greater than the escape velocity. One of the important properties of an ergoregion is that, in this region, a timelike Killing vector $\partial_{t}$ becomes spacelike or vice versa. It is known that an ergoregion plays a significant role in the energy extraction process or Penrose process of a black hole [@Penrose:1971uk].
The plots of ergoregions of the Kerr black hole ($k=0$) can be seen from Fig. \[ergo2\] for different values of spin $a$, showing that the ergoregion area increases with $a$. We have shown the effect of parameter $k$ on the shape of an ergoregion in Fig. \[ergo\]. It can be observed from these figures that the ergoregion is easily affected by the parameter $k$. It can be seen from Fig. \[ergo\] that one can find a critical value of $k$ ($k^c$) where the two horizons almost coincide (near extremal black hole). It can be observed for $k>k^c$ that the horizons [vanish]{} (cf., Fig. \[ergo\]). One can easily observe from Fig. \[ergo\] that there is an increase in the ergoregion area with the parameter $k$ for a given value of spin $a$. It turns out that the critical values of parameter $k^{c}=0.5565, 0.4758, 0.3876, 0.2938, 0.1981$ correspond, respectively, for $a=0.4, 0.5, 0.6, 0.7, 0.8$.
Geodesics in nonsingular black hole {#geodesics}
===================================
In this section, we present the calculation necessary for the shape of the black hole shadow, which demands the study of geodesics motion around the nonsingular black hole (\[metric\]). The rotating nonsingular spacetime is characterized by three constants of the motion which are energy $E$, component of angular momentum $L_{z}$ and the Carter constant $\mathcal{K}$. To derive the geodesic equations, first we start with the Lagrangian $$\label{lagrangian}
\mathcal{L}=\frac{1}{2}g_{\mu \nu}\dot{x}^{\mu}\dot{x}^{\nu},$$ where an overdot denotes partial derivative with respect to the affine parameter $\sigma$. The constants of motion $E$ and $L_{z}$ are given by [@Bardeen:1972fi] $$\label{E}
-E=\frac{\partial \mathcal{L}}{\partial \dot{t}}=g_{tt}\dot{t}+g_{t\phi}\dot{\phi},$$ and $$\label{L}
L_{z}=\frac{\partial \mathcal{L}}{\partial \dot{\phi}}=g_{\phi \phi}\dot{\phi}+g_{t\phi}\dot{t},$$ with $E$ and $L_z$, respectively, the energy and angular momentum. One can easily obtain the following geodesic equations by solving Eqs. (\[E\]) and (\[L\]), simultaneously: $$\label{u^t}
\Sigma \frac{d t}{d \sigma} = -a \left(aE \sin^2 \theta - L_{z}\right) + \frac{(r^2 + a^2) \left[(r^2+a^2)E-aL_{z}\right]}{r^2 + a^2- 2 M r e^{-k/r}},$$ $$\label{u^Phi}
\Sigma \frac{d \phi}{d \sigma} = -\left(aE - L_{z}\csc^2 \theta \right) + \frac{a \left[(r^2+a^2)E-aL_{z}\right]}{r^2 + a^2- 2 M r e^{-k/r}}.$$ The remaining geodesic equations for the rotating nonsingular black hole with metric tensor $g^{\mu \nu}$ can be determined by using the Hamilton-Jacobi equation, which has the following form, $$\label{hje}
\frac{\partial S}{\partial \sigma} = -\frac{1}{2} g^{\mu\nu} \frac{\partial S}{\partial x^{\mu}} \frac{\partial S}{\partial x^{\nu}},$$ where $S$ is the Jacobi action. The Hamilton-Jacobi Eq. (\[hje\]) has a solution of the following form, $$\label{hja}
S = \frac{1}{2} m_0^2 \sigma -Et + L_{z} \phi + S_{r}(r) + S_{\theta}(\theta),$$ where $ m_0 $ corresponds to the rest mass of the particle, and $ S_{r} $ and $ S_{\theta} $ are functions of $ r $ and $ \theta $, respectively. Furthermore, one can obtain the geodesic equations by using Eqs. (\[hje\]) and (\[hja\]) and separating out the coefficients of $r$ and $\theta$ equal to the Carter constant ($\pm \mathcal{K}$), $$\label{u^r}
\Sigma \frac{d r}{d \sigma} = \pm \sqrt{\mathcal{R}},$$ $$\label{u^theta}
\Sigma \frac{d \theta}{d \sigma} = \pm \sqrt{\Theta},$$ where $$\begin{aligned}
\label{R}
\mathcal{R} &=& \left[(r^2+a^2)E-aL_{z}\right]^2 \nonumber \\
&-&\left(r^2 + a^2- 2 M r e^{-k/r}\right) \left[m_0^2 r^2 +\mathcal{K}+ (L_{z}-a E)^2 \right], \nonumber \\ \end{aligned}$$ and $$\begin{aligned}
\label{Th}
\Theta=\mathcal{K} +\cos^2 \theta \left(a^2E^2-L_{z}^2\csc^2 \theta \right),\end{aligned}$$ where, without loss of generality one can take $m_{0}=1$ for particles, and $m_{0}=0$ for photons. We are interested in photon trajectories; therefore, the Eq. (\[R\]) transforms into $$\begin{aligned}
\label{R1}
\mathcal{R}&=& \left[(r^2+a^2)E-aL_{z}\right]^2 \nonumber \\
&-&\left(r^{2}+a^{2}-2Mr e^{-k/r}\right) \left[\mathcal{K}+(L_{z}-a E)^2\right].\end{aligned}$$ Now, we introduce dimensionless quantities such that $\chi=L_{z}/E$ and $\zeta=\mathcal{K}/E^2$, which are constant along the geodesics; therefore, in terms of these quantities Eq. (\[R1\]) looks like $$\begin{aligned}
\label{R2}
\mathcal{R}&=& E^2\left[r^2+a^2-a\chi \right]^2 \nonumber \\
&-& E^2 \left(r^{2}+a^{2}-2Mr e^{-k/r}\right) \left[\zeta +(\chi -a )^2\right].\nonumber \\\end{aligned}$$ Furthermore, to discuss the radial motion of the particle, the effective potential is a very important tool which can be calculated by using the equation [@Wei:2013kza] $$\begin{aligned}
\left(\Sigma \frac{d r}{d \sigma}\right)^2 + V_{eff}=0,\end{aligned}$$ hence, the expression of effective potential ($V_{eff}=-\mathcal{R}$) reads $$\begin{aligned}
\label{effpot}
V_{eff} &=& -E^2\left[r^2+a^2-a \chi \right]^2 \nonumber \\
&+& E^2 (r^2+a^2-2Mr e^{-k/r})\left[\zeta +(\chi -a)^2 \right].\nonumber \\ \end{aligned}$$ Indeed, the effective potential not only depends on parameter $k$ but also on $\chi$ and $\zeta$. Next, we are interested in circular orbits of the photons to find out $\chi$ and $\zeta$, which satisfy [@Bardeen:1972fi] $$\label{condition}
V_{eff} = 0 \;\;\;\;\; \text{and} \;\;\;\;\; \frac{d V_{eff}}{dr}=0.$$ By using Eqs. (\[effpot\]) and (\[condition\]), we can easily get the parameters $\chi$ and $\zeta$ in the following form: $$\begin{aligned}
\label{xi}
\chi &=& \frac{(r^2+a^2)r^2 e^{k/r}+M \left[(k-3 r) r^2+a^2 (k+r)\right]}{a M (k+r)-a r^2e^{k/r}},
\nonumber \\ \end{aligned}$$ $$\begin{aligned}
\label{eta}
\zeta &=& -\frac{4 a^2Mr^4 (k-r) e^{k/r} +r^4(k M-3 Mr + r^2 e^{k/r})^2}{a^2 (k M+Mr-r^2 e^{k/r})^2}.
\nonumber \\ \end{aligned}$$ If one substitutes $k=0$, we get $$\begin{aligned}
\label{xi0}
\chi &=& \frac{(r^2+a^2)r^2 +M r(-3 r^2 +a^2 )}{a r(M-r)},\end{aligned}$$ $$\begin{aligned}
\label{eta0}
\zeta &=& \frac{4 a^2Mr^5 - r^5(r-3 M)^2}{a^2 r^2(M-r)^2},\end{aligned}$$ which are exactly the same as the Kerr black hole [@Hioki:2009na]. Similarly, in the limit $k \rightarrow 0$, all quantities in this section correspond to the Kerr black hole [@Hioki:2009na]. The $V_{eff}$ dependence of the parameter $k$ with the particular choice of $\chi$ and $\zeta$ can be seen in Fig. \[eff\], which shows that there are various bounds of the effective potential according to the parameter $k$.
--------------------------------------------------------------------------------------------- --
{width="0.5\linewidth"} {width="0.5\linewidth"}
--------------------------------------------------------------------------------------------- --
Nonsingular black hole shadow {#shadow}
=============================
Next, we study the shadow of the rotating nonsingular black hole. We consider that the photons released from a bright object are coming towards the black hole, which is situated between an observer and a bright object. There should be three possible trajectories of the photon geodesics around the black hole: (i) falling into the black hole (capture orbits), (ii) scattered away from the black hole to infinity (scatter orbits), and (iii) critical geodesics which separate the first two sets (unstable orbits). The observer which is far away from the black hole can see only the scattered photons while those photons which fall into the black hole form a dark region. This dark region is known as the black hole shadow. Furthermore, to study about this dark region or black hole shadow we are going to define the new coordinates ($\alpha,\beta$) [@Bardeen:1973gb], which are known as celestial coordinates [(cf. Fig. \[celestial\])]{} defined by
![\[celestial\] Plot showing the geometrical representation of the celestial coordinates for a distant observer.](fig1.eps){width="0.85\linewidth"}
![\[a=0\] Plot showing the shapes of shadow cast by a nonrotating black hole ($a=0$) for different values of $k$.](figure4a.eps)
![\[k-obs\] Plot showing the behavior of the radius of the black hole shadow for different values of $k$.](figure4b.eps)
---------------------------------------------------------------------------------------------------------------------------------------------- -- --
{width="0.33\linewidth"} {width="0.33\linewidth"} {width="0.33\linewidth"}
{width="0.33\linewidth"} {width="0.33\linewidth"} {width="0.33\linewidth"}
---------------------------------------------------------------------------------------------------------------------------------------------- -- --
$$\label{alpha}
\alpha = \lim_{r_{0} \rightarrow \infty} \left(-r_{0}^2 \sin \theta_{0} \frac{d \phi}{dr} \right),$$
$$\label{beta}
\beta = \lim_{r_{0} \rightarrow \infty} \left(r_{0}^2 \frac{d \theta}{dr} \right),$$
where $r_{0}$ is the distance from the black hole to the observer and $\theta_{0}$ represents the inclination angle between the rotation axis of the black hole and the direction to an observer. By using Eqs. (\[u\^t\]), (\[u\^Phi\]), (\[u\^r\]), and (\[u\^theta\]), we can easily calculate $d\phi /dr$ and $d \theta /dr$ when inserting these values in Eqs. (\[alpha\]) and (\[beta\]), where the celestial coordinates read [@Bardeen:1973gb; @Chandrasekhar:1992] $$\label{alpha1}
\alpha = -\chi \csc \theta_{0},$$ $$\label{beta1}
\beta = \pm \sqrt{\zeta +a^2 \cos^2 \theta_{0} -\chi^2 \cot^2 \theta_{0}}.$$ where $\theta_{0}$ is the inclination angle between the rotation axis and the direction to an observer. If an observer is situated in the equatorial plane of the black hole, in this case the inclination angle is $\theta_{0}=\pi /2$. Hence, the Eqs. (\[alpha1\]) and (\[beta1\]) transform into $$\label{alpha2}
\alpha = -\chi,$$ $$\label{beta2}
\beta = \pm \sqrt{\zeta}.$$ Next, to visualize the shapes of the black hole shadow, we need to plot $\beta$ vs $\alpha$, where coordinates $\alpha$ and $\beta$ satisfy the following relation: $$\label{alpha&beta}
\alpha^2 + \beta^2 = \chi^2 +\zeta.$$ Using Eqs. (\[xi\]), (\[eta\]), and (\[alpha&beta\]), we can obtain $$\begin{aligned}
\label{alpha&beta1}
\alpha^2 + \beta^2 &=& \frac{1}{(k M+ Mr- r^2 e^{k/r})^2}\Big[2 r^2 \nonumber \\
& \times& \left(r^4 e^{2 k/r}+M^2(k^2 -2 k r-3 r^2)\right) \nonumber \\
&+&a^2 \left(r^2 e^{k/r}+M(k +r) \right)^2\Big].\end{aligned}$$ Now we can plot the shadow of the rotating nonsingular black hole (\[metric\]) by using Eqs. (\[alpha2\]) and (\[beta2\]). It is clear that the shapes or the contours of a black hole shadow depend on its spin, inclination angle and parameter $k$. We shall plot $\alpha$ vs $\beta$ to display the contour of the shadow of the nonsingular black hole for various values of $a$ and $k$ at different inclination angles.
---------------------------------------------------------------------------------------------------------------------------------------------- -- --
{width="0.33\linewidth"} {width="0.33\linewidth"} {width="0.33\linewidth"}
{width="0.33\linewidth"} {width="0.33\linewidth"} {width="0.33\linewidth"}
---------------------------------------------------------------------------------------------------------------------------------------------- -- --
Nonrotating case
----------------
We first consider the nonrotating case, $a=0$, in which case the metric (\[metric\]) reduces to one obtained in [@Culetu:2014lca]. In this case, the parameters $\alpha$ and $\beta$ satisfy $$\label{nonrot}
\alpha^2 + \beta^2 = \frac{2 r^2 (k^2-2 k r-3 r^2+ r^{4} e^{2 k/r})}{(r+ k- r^{2} e^{k/r})^2},$$ where $r$ is the radius of circular orbits. The black hole shadow or plots of Eqs. (\[alpha2\]) and (\[beta2\]) are presented in Fig. \[a=0\]. They are perfect circles. It can be [observed]{} from the plots that the parameter $k$ reduces the radius of the shadow (cf. Fig. \[k-obs\]). Consider the case when $k=0$, and Eq. (\[nonrot\]) transforms into $$\label{nonrot0}
\alpha^2 + \beta^2 = \frac{2 r^2( r^2-3)}{(1- r)^2}.$$ Equation (\[nonrot0\]) is exactly the same as the one obtained for the Schwarzschild black hole [@Synge:1966].
![\[observable\] Plot showing the schematic representation of the observable.](fig2.eps){width="0.8\linewidth"}
{width="0.48\linewidth"} {width="0.48\linewidth"}
Rotating case
-------------
It is shown that for a rotating black hole the shadow has an [atypical]{} deformation due to spin parameters instead of being a perfect circle. Here we wish to further analyze the effect of parameter $k$ on the apparent shapes of the rotating nonsingular black hole. Furthermore, if $k=0$, Eq. (\[alpha&beta1\]) reduces to $$\begin{aligned}
\label{alpha&beta01}
\alpha^2 + \beta^2 &=& \frac{2r^2(r^2 -3 M^2)+a^2 (r+M)^2}{(r-M)^2}.\end{aligned}$$ Now, consider the case when the spin is nonzero, i.e., $a\neq0$. We have plotted the shapes of the shadow in Figs. \[theta\] and \[k\]. In Fig. \[theta\], we show the contours of the black hole shadow for various values of deviation parameter $k$ with different inclination angles $\theta_{0}$. Indeed, the shapes of the shadow are dependent on the inclination angle of the observer. Furthermore, Fig. \[k\] shows the effect of parameter $k$ on the shadow of a black hole. It is clear from Fig. \[k\] that when $a$ and $k$ take critical values, the shape of the shadow is more deformed.
Furthermore, to calculate the actual size and distortion of the black hole shadow, let us define the observable that characterizes the black hole shadow, namely, $R_{s}$ and $\delta_{s}$, where the parameters $R_{s}$ and $\delta_{s}$ correspond to the actual size of the shadow and distortion in the shape of the shadow, respectively. The radius of the shadow can be calculated by considering a circle passing through the three points at the top, bottom, and rightmost corresponding to ($\alpha_{t}$,$\beta_{t}$), ($\alpha_{b}$,$\beta_{b}$), and ($\alpha_{r}$,$0$), respectively [@Hioki:2009na]. The definitions of these observables are given by [@Hioki:2009na] $$\label{Rs}
R_{s}=\frac{(\alpha_{t}-\alpha_{r})^2+\beta_{t}^2}{2(\alpha_{t}-\alpha_{r})},$$ and $$\label{dels}
\delta_{s}=\frac{(\tilde{\alpha_{p}}-\alpha_{p})}{R_{s}},$$ where $(\tilde{\alpha_{p}},0)$ and $(\alpha_{p},0)$ are the points where the reference circle and the contour of the shadow cut the horizontal axis at the opposite side of $(\alpha_{r},0)$ (cf., Fig. \[observable\]). The plots of the observable can be seen from Fig. \[obs\] which shows that when parameter $k$ increases, the radius of the shadow decreases and the distortion increases.
Energy emission rate {#EE rate}
====================
Next, we are interested in calculating the rate of energy emission by the rotating nonsingular black hole. First of all, we know that the limiting constant value of the absorption cross section for a spherically symmetric black hole is given by $$\sigma_{lim} \approx \pi R_{s}^2.$$ The black hole shadow is responsible for the high-energy absorption cross section for a distant observer. The mentioned limiting constant value is derived in terms of geodesics and can be analyzed for the wave theories. For a black hole endowed with a photon sphere, the limiting constant value is the same as the geometrical cross section of this photon sphere. Now, by using this limiting constant value, we can compute the energy emission rate [@Wei:2013kza] by $$\frac{d^2 E(\omega)}{d\omega dt}= \frac{2 \pi^2 R_{s}^2}{e^{\omega/ T}-1}\omega ^3,$$ where $$\begin{aligned}
T &=& \frac{1}{4 \pi r_{+}^2}\frac{r_{+}^2 (r_{+}-k)-a^2 (r_{+}+k)}{(r_{+}^2+a^2)}.\end{aligned}$$ This generalizes the Hawking temperature of the Kerr black hole, and $r_{+}$ represents the event horizon. If $k=0$, then one obtains the Hawking temperature, $$\begin{aligned}
T &=& \frac{1}{4 \pi r_{+}}\frac{r_{+}^2-a^2}{r_{+}^2+a^2},\end{aligned}$$ for the Kerr black hole [@Good:2014uja]. The plots of the energy emission rate ($d^2 E(\omega)/d\omega dt$) versus the frequency ($\omega$) can be seen from Fig. \[ee\] for $a=0$ and $a=0.9$ by varying the parameter $k$. It is clear that, for small values of $k$, the energy emission rate is high and if it takes a maximum value, then the energy emission rate is low.
{width="0.48\linewidth"} {width="0.48\linewidth"}
Conclusion
==========
The investigation of the shadows of different types of black holes has been an important subject of research because that the observations of a black hole in the center of a galaxy may be determined [@Falcke:1999pj]. The shadow of a Kerr black hole is distorted mainly by its spin parameter and the inclination angle. It is a perfect circle for the Schwarzschild black hole.
In this paper, we make a qualitative analysis of the shapes of the black hole shadow cast by a rotating nonsingular black hole which has an additional deviation parameter $k$ and also perform a [detailed]{} study of its ergoregion. The dependence of black hole shadows and ergoregions on spin $a$ and parameter $k$ are explicitly discussed. With the help of a study of null geodesics around the rotating nonsingular black hole, we construct the parameters $\zeta$ and $\chi$ by using the circular orbit conditions and discuss the behavior of the effective potential. We consider the black hole shadows for both nonrotating and rotating cases to bring out the effect of spin $a$ and parameter $k$. The shape of the shadows for the nonrotating black hole are perfect circles, and the size decreases with increasing value of $k$. In the next paragraph, we studied the shape of shadows for the rotating black hole. In this case, the shape is not a perfect circle, but it is deformed due to the presence of spin $a$ and parameter $k$. One can observe that the area of the ergoregion increases with an increase either in parameter $k$ or spin $a$. The observables $R_{s}$ and $\delta_{s}$ are defined to characterize the shape of the black hole shadow. We demonstrate the influence of the parameter $k$ on observables $R_{s}$ and $\delta_{s}$. It shows that the value of $R_s$ decreases and $\delta_s$ increases with the parameter $k$.
The results obtained are similar to the Kerr black hole shadows [@Hioki:2009na], but the size of the shadow decreases due to the deviation parameter $k$. The results obtained here may provide insight to into the qualitative features of a nonsingular black hole. We have done a detailed analysis of a rotating nonsingular black hole from the viewpoint of the ergoregion and shadow that contains the Kerr black hole analysis as the special case when the deviation parameter $k=0$.
M.A. acknowledges the University Grant Commission, India, for financial support through the Maulana Azad National Fellowship For Minority Students scheme (Grant No. F1-17.1/2012-13/MANF-2012-13-MUS-RAJ-8679). S.G.G. would like to thank SERB-DST for Research Project Grant No. SB/S2/HEP-008/2014. We also thank IUCAA for hospitality while a part of this work was done and ICTP for Grant No. OEA-NET-76.
[00]{}
Y. Tanaka [*et al.*]{}, Nature (London) [**375**]{}, 659 (1995).
A.C. Fabian, K. Nandra, C.S. Reynolds, W.N. Brandt, C. Otani, Y. Tanaka, H. Inoue, and K. Iwasawa, Mon. Not. R. Astron. Soc. [**277**]{}, L11 (1995).
R. Genzel, R. Schodel, T. Ott, A. Eckart, T. Alexander, F. Lacombe, D. Rouan, and B. Aschenbach, Nature (London) [**425**]{}, 934 (2003).
P.A. Connors, R.F. Stark, and T. Piran, Astrophys. J. [**235**]{}, 224 (1980).
H. Asada, M. Kasai, and T. Yamamoto, Phys. Rev. D [**67**]{}, 043006 (2003).
T. Tanaka, Y. Mino, M. Sasaki, and M. Shibata, Phys. Rev. D [**54**]{}, 3762 (1996).
L.S. Finn and K. S. Thorne, Phys. Rev. D [**62**]{}, 124021 (2000).
R. Takahashi, Astrophys. J. [**611**]{}, 996 (2004); R. Takahashi, Publ. Astron. Soc. Jpn. [**57**]{}, 273 (2005).
J.M. Bardeen, in [*Black holes, in Proceeding of the Les Houches Summer School, Session*]{} 215239, edited by C. De Witt and B.S. De Witt and B.S. De Witt (Gordon and Breach, New York, 1973).
S. Chandrasekhar, *The Mathematical Theory of Black Holes* (Oxford University Press, New York, 1992).
H. Falcke, F. Melia and E. Agol, Astrophys. J. [**528**]{}, L13 (2000).
V. Bozza, Gen. Relativ. Gravit. [**42**]{}, 2269 (2010).
D.E. Holz and J.A. Wheeler, Astrophys. J. [**578**]{}, 330 (2002).
J.L. Synge, Mon. Not. R. Astron. Soc. [**131**]{}, 463 (1966).
J.P. Luminet Astron. Astrophys. [**75**]{}, 228 (1979).
K.S. Virbhadra, Phys. Rev. D [**79**]{}, 083004 (2009).
G.S. Bisnovatyi-Kogan and O.Y. Tsupko, Plasma Phys. Rep. [**41**]{}, 562 (2015).
V.S. Morozova, B.J. Ahmedov, and A.A. Tursunov, Astrophys. Space Sci. [**346**]{}, 513 (2013).
P.G. Nedkova, V.K. Tinchev, and S.S. Yazadjiev, Phys. Rev. D [**88**]{}, 124019 (2013).
A.F. Zakharov, F. De Paolis, G. Ingrosso, and A.A. Nucita, New Astron. [**10**]{}, 479 (2005); A.F. Zakharov, F. De Paolis, G. Ingrosso, and A.A. Nucita, Astron. Astrophys. [**442**]{}, 795 (2005); F. De Paolis, G. Ingrosso, A.A. Nucita, A. Qadir and A.F. Zakharov, Gen. Relativ. Gravit. [**43**]{}, 977 (2011).
C. Bambi and K. Freese, Phys. Rev. D [**79**]{}, 043002 (2009).
V. Bozza and G. Scarpetta, Phys. Rev. D [**76**]{}, 083008 (2007).
K. Hioki and K. i. Maeda, Phys. Rev. D [**80**]{}, 024042 (2009).
R. Takahashi, Publ. Astron. Soc. Jap. [**57**]{}, 273 (2005).
S. W. Wei and Y. X. Liu, J. Cosmol. Astropart. Phys. 11 (2013) 063.
A. Abdujabbarov, F. Atamurotov, Y. Kucukakca, B. Ahmedov, and U. Camci, Astrophys. Space Sci. [**344**]{}, 429 (2013).
L. Amarilla and E. F. Eiroa, Phys. Rev. D [**85**]{}, 064019 (2012).
L. Amarilla and E. F. Eiroa, Phys. Rev. D [**87**]{}, 044057 (2013).
F. Atamurotov, A. Abdujabbarov, and B. Ahmedov, Phys. Rev. D [**88**]{}, 064004 (2013).
A. Grenzebach, V. Perlick, and C. L[ä]{}mmerzahl, Phys. Rev. D [**89**]{}, 124004 (2014).
P.V. P. Cunha, C.A. R. Herdeiro, E. Radu, and H.F. Runarsson, Phys. Rev. Lett. [**115**]{}, 211102 (2015).
A.A. Abdujabbarov, L. Rezzolla, and B. J. Ahmedov, Mon. Not. R. Astron. Soc. [**454**]{}, 2423 (2015).
U. Papnoi, F. Atamurotov, S. G. Ghosh, and B. Ahmedov, Phys. Rev. D [**90**]{}, 024073 (2014).
Z. Stuchlík and J. Schee, Int. J. Mod. Phys. D [**24**]{}, 1550020 (2015).
C. Bambi, Phys. Lett. B [**730**]{}, 59 (2014).
S.G. Ghosh, P. Sheoran, and M. Amir, Phys. Rev. D [**90**]{}(10), 103006 (2014).
J. Schee and Z. Stuchlik, J. Cosmol. Astropart. Phys. 06 (2015) 048.
M. Amir and S.G. Ghosh, J. High Energy Phys. 07 (2015) 015.
S. G. Ghosh and M. Amir, Eur. Phys. J. C [**75**]{}, 553 (2015).
Z. Li and C. Bambi, J. Cosmol. Astropart. Phys. 01 (2014) 041.
V.K. Tinchev, Chin. J. Phys. [**53**]{}, 110113 (2015).
A. Abdujabbarov, M. Amir, B. Ahmedov, and S.G. Ghosh, Phys. Rev. D [**93**]{}, 104004 (2016).
S. Ansoldi, arXiv:0802.0330.
C. Bambi and L. Modesto, Phys. Lett. B [**721**]{}, 329 (2013).
B. Toshmatov, B. Ahmedov, A. Abdujabbarov, and Z. Stuchlik, Phys. Rev. D [**89**]{}, 104017 (2014).
S.G. Ghosh and S.D. Maharaj, Eur. Phys. J. C [**75**]{}, 7 (2015).
J.M. Bardeen, in *Conference Proceedings of GR5, Tbilisi*, USSR, 1968, p. 174.
E. Ay[ó]{}n-Beato, A. Garc[í]{}a, Phys. Lett. B [**493**]{}, 149 (2000).
E. Ay[ó]{}n-Beato, A. Garc[í]{}a, Phys. Rev. Lett. [80]{}, 5056 (1998).
E. Ay[ó]{}n-Beato and A. Garc[í]{}a, Gen. Rel. Grav. [**31**]{}, 629 (1999).
E. Ay[ó]{}n-Beato and A. Garc[í]{}a, Gen. Relativ. Gravit. [**37**]{}, 635 (2005).
I. Dymnikova, Gen. Relativ. Gravit. [**24**]{}, 235 (1992).
I. Dymnikova, Classical Quantum Gravity [**21**]{}, 4417 (2004).
K.A. Bronnikov, Phys. Rev. D [**63**]{}, 044005 (2001).
S. Shankaranarayanan and N. Dadhich, Int. J. Mod. Phys. D [**13**]{}, 1095 (2004).
S.A. Hayward, Phys. Rev. Lett. [**96**]{}, 031103 (2006).
H. Culetu, Int. J. Theor. Phys. [**54**]{}, 2855 (2015).
L. Balart and E.C. Vagenas, Phys. Lett. B [**730**]{}, 14 (2014).
L. Balart and E.C. Vagenas, Phys. Rev. D [**90**]{}(12), 124045 (2014).
L. Xiang, Y. Ling, and Y.G. Shen, Int. J. Mod. Phys. D [**22**]{}, 1342016 (2013).
S.G. Ghosh, Eur. Phys. J. C [**75**]{}, 532 (2015).
R.P. Kerr, Phys. Rev. Lett. [**11**]{}, 237 (1963).
K. Schwarzschild, Uber das Gravitationsfeld eines Manpunktes nach der Einsteinschen Theorie, Sitzber. Deutsch. Akad. Wiss. Berlin, Kl. Math. Phys. Tech., 189, (1916).
E.T. Newman, R. Couch, K. Chinnapared, A. Exton, A. Prakash, and R. Torrence, J. Math. Phys. [**6**]{}, 918 (1965).
C. Bambi, Mod. Phys. Lett. A [**26**]{}, 2453 (2011); Astronomy Review [**8**]{}, 4 (2013).
C. Bambi, J. Cosmol. Astropart. Phys. 08 (2013) 055.
L. Gou [*et al.*]{}, Astrophys. J. [**742**]{}, 85 (2011).
L. Gou [*et al.*]{}, Astrophys. J. [**790**]{}, 29 (2014).
R. Penrose and R.M. Floyd, Nature (London) [**229**]{}, 177 (1971).
J.M. Bardeen, W.H. Press and S.A. Teukolsky, Astrophys. J. [**178**]{}, 347 (1972).
M.R.R. Good and Y.C. Ong, Phys. Rev. D [**91**]{}, 044031 (2015).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In the context of Standard Model (SM) extensions, the seesaw mechanism provides the most natural explanation for the smallness of neutrino masses. In this work we consider the most economical type-I seesaw realization in which two right-handed neutrinos are added to the SM field content. For the sake of predictability, we impose the maximum number of texture zeros in the lepton Yukawa and mass matrices. All possible patterns are analyzed in the light of the most recent neutrino oscillation data, and predictions for leptonic CP violation are presented. We conclude that, in the charged-lepton mass basis, eight different texture combinations are compatible with neutrino data at $1\sigma$, all of them for an inverted-hierarchical neutrino mass spectrum. Four of these cases predict a CP-violating Dirac phase close to $3\pi/2$, which is around the current best-fit value from global analysis of neutrino oscillation data. If one further reduces the number of free parameters by considering three equal elements in the Dirac neutrino Yukawa coupling matrix, several texture combinations are still compatible with data but only at $3\sigma$. For all viable textures, the baryon asymmetry of the Universe is computed in the context of thermal leptogenesis, assuming (mildly) hierarchical heavy Majorana neutrino masses $M_{1,2}$. It is shown that the flavored regime is ruled out, while the unflavored one requires $M_{1} \sim 10^{14}$ GeV.'
author:
- 'D. M. Barreiros'
- 'R. G. Felipe'
- 'F. R. Joaquim'
title: 'The minimal type-I seesaw model with maximally-restricted texture zeros'
---
Introduction {#Intro}
============
The discovery of neutrino oscillations provided a solid evidence for physics beyond the Standard Model (SM), by confirming the existence of neutrino masses and mixing. From the theory viewpoint, the most straightforward and elegant way of accounting for them consists of adding right-handed (RH) neutrinos to the SM field content. If heavy enough, these states can mediate neutrino masses at the classical level through the well-known seesaw mechanism [@Minkowski:1977sc]. Besides supplying an explanation for small neutrino masses, the addition of RH neutrinos to the SM allows for the leptogenesis mechanism [@Fukugita:1986hr] to work through the out-of-equilibrium decays of the heavy neutrinos in the early Universe (for reviews see e.g. [@Buchmuller:2004nz; @Davidson:2008bu; @Chun:2017spz; @Fong:2013wr]). This offers an answer for another SM puzzle: the baryon asymmetry of the Universe (BAU).
Although, in principle, the number of RH neutrinos is arbitrary, at least two are necessary to explain the present neutrino oscillation data, namely, three nonzero neutrino mixing angles and two mass-squared differences. Interestingly, at least two RH neutrinos are also required for leptogenesis to be realized. Therefore, the two RH neutrino seesaw model (2RHNSM) is not only a minimal model for neutrino masses, but also for the generation of the BAU in the context of leptogenesis. Still, even in this scenario, the number of parameters describing the neutrino Lagrangian at high energies is larger than the number of low-energy observables currently (or potentially) measured by experiments. One way of increasing predictability is to consider texture zeros in the lepton Yukawa and mass matrices, which can be motivated, for instance, by imposing U(1) Abelian flavor symmetries [@Grimus:2004hf; @Cebola:2015dwa]. In general, texture zeros imply predictions not only for low-energy neutrino parameters but also for the BAU, since leptogenesis is sensitive to the couplings which control neutrino masses and mixing. Therefore, a complete study of all possible texture zeros in the light of most recent neutrino data is welcome. In particular, since neutrino experiments are starting to deliver some information regarding leptonic CP violation [@Branco:2011zb], predictions for low-energy CP phases are of utmost importance. At the same time, a connection with leptogenesis can also be established in this framework [@Branco:2011zb; @Hagedorn:2017wjy]. These questions have already been partially covered in the literature. For instance, the compatibility of texture-zero hypothesis in the 2RHNSM with neutrino data has been studied in Refs. [@Frampton:2002qc; @Ibarra:2003up; @Harigaya:2012bw; @Rink:2016vvl; @Shimizu:2017fgu] and, in the context of leptogenesis, in Refs. [@GonzalezFelipe:2003fi; @Joaquim:2005zv; @Abada:2006ea; @Zhang:2015tea; @Siyeon:2016wro; @Geib:2017bsw; @Achelashvili:2017nqp; @Shimizu:2017vwi].
In this work, we revisit the 2RHNSM in maximally restricted texture-zero scenarios, i.e. when the maximum number of texture zeros is imposed in the lepton Yukawa and mass matrices. Moreover, we consider cases in which equality relations among the Dirac neutrino Yukawa couplings exist. For textures that reproduce the observed neutrino mass and mixing patterns, we present the predictions for low-energy CP violation, neutrinoless double beta decay and the BAU. Special attention will be paid to the treatment of leptogenesis in the 2RHNSM. Contrary to what is usually done in the literature, where only the decay of the lightest heavy neutrino is considered, we include decays of both heavy neutrinos in our analysis. Moreover, flavor effects which arise from the fact that lepton interactions become out of equilibrium at different temperatures are taken into account.
This paper is organized as follows. In Section \[sec1\] we set the basics of the 2RHNSM, by describing the model and identifying the number of parameters at high and low energies. Afterwards, in Section \[sec2\], the maximally-restricted texture zero matrices are identified, and their compatibility with neutrino data is analyzed. Furthermore, the predictions for Dirac and Majorana CP phases are shown, together with those for the effective neutrino mass parameter relevant for neutrinoless double beta decays. We also consider cases with three equal elements in the Dirac neutrino Yukawa coupling matrix in Section \[sec3a\]. We then compute the BAU in the thermal leptogenesis framework in Section \[sec3\], and determine under which conditions its value is compatible with the observed one. Our conclusions are drawn in Section \[sec4\].
The two right-handed neutrino seesaw model {#sec1}
==========================================
Considering only Yukawa and mass terms, the lepton Lagrangian density for the SM extended with RH neutrino fields $\nu_{R}$ is $\mathcal{L}=\mathcal{L}_{\ell}+\mathcal{L}_\nu$ with $$\begin{aligned}
\mathcal{L}_\nu&=-\overline{\ell_{L}}{\mathbf{Y}^\nu}\tilde{\Phi}\nu_{R} -\frac{1}{2}\overline{(\nu_{R})^c}{\mathbf{M}_R}\nu_{R} + \text{H.c.}\,, \label{LtypeI}\\
\mathcal{L}_\ell&=-\overline{\ell_{L}}{\mathbf{Y}^\ell}\Phi\,e_{R} + \text{H.c.}\,.\label{Lcl}\end{aligned}$$ Here, $\ell_{L}$ and $\Phi$ are the SM lepton and Higgs doublets, respectively, $\tilde{\Phi}=i\sigma_2 \Phi^\ast$, and $e_R$ denote the RH charged-lepton fields. The Dirac neutrino Yukawa couplings and RH neutrino mass matrices are described by ${\mathbf{Y}^\nu}$ and ${\mathbf{M}_R}$. For $N$ RH neutrinos, ${\mathbf{Y}^\nu}$ and ${\mathbf{M}_R}$ are $3\times N$ and $N\times N$ general complex matrices, being ${\mathbf{M}_R}$ symmetric. After integrating out the $\nu_R$’s, the effective Majorana neutrino mass matrix ${\mathbf{M}^\nu}$, obtained upon electroweak symmetry breaking, is given by the seesaw formula [@Minkowski:1977sc] $$\begin{aligned}
{\mathbf{M}^\nu}=-v^2{\mathbf{Y}^\nu}{\mathbf{M}_R}^{-1}{{\mathbf{Y}^\nu}}^T\,,
\label{Mnuseesaw}\end{aligned}$$ which is valid for ${\mathbf{M}_R}\gg v$, where $v=174\,{\rm GeV}$ is the vacuum expectation value of the neutral component of $\Phi$. This (symmetric) matrix is diagonalized by a unitary matrix $\mathbf{U_\nu}$ as $$\begin{aligned}
{\mathbf{U}_\nu}^T {\mathbf{M}^\nu}{\mathbf{U}_\nu}=\text{diag}(m_1,m_2,m_3)\equiv \mathbf{d}_m\,,
\label{Mnudiag}\end{aligned}$$ where $m_i$ are the (real and positive) effective neutrino masses. Considering that ${\mathbf{U}_\ell}$ rotates the left-handed (LH) charged-lepton fields to their diagonal mass basis, lepton mixing in charged currents is encoded in the so-called Pontecorvo-Maki-Nakagawa-Sakata (PMNS) unitary matrix ${\mathbf{U}}$ given by $$\begin{aligned}
{\mathbf{U}}={\mathbf{U}_\ell}^\dag {\mathbf{U}_\nu}\,.
\label{UPMNS}\end{aligned}$$ Throughout this work we will use the standard parametrization [@Patrignani:2016xqp]
$$\begin{gathered}
\mathbf{U}=\begin{pmatrix}
c_{12}c_{13}&s_{12}c_{13}&s_{13}e^{-i\delta}\\
-s_{12}c_{23}-c_{12}s_{23}s_{13}e^{i\delta}&c_{12}c_{23}-s_{12}s_{23}s_{13}e^{i\delta}&s_{23}c_{13}\\
s_{12}s_{23}-c_{12}c_{23}s_{13}e^{i\delta}&-c_{12}s_{23}-s_{12}c_{23}s_{13}e^{i\delta}&c_{23}c_{13}
\end{pmatrix}\!\!\begin{pmatrix}
1&0&0\\
0&e^{i\alpha_{21}/2}&0\\
0&0&e^{i\alpha_{31}/2}\\
\end{pmatrix}\,,
\label{Uparam}\end{gathered}$$
where $c_{ij}\equiv \cos\theta_{ij}$, $s_{ij}\equiv \sin\theta_{ij}$ and $\theta_{ij}\,(i < j=1,2,3)$ are the three lepton mixing angles. The phases $\delta$ and $\alpha_{21,31}$ are Dirac and Majorana-type CP-violating phases, respectively.
The present values for $\theta_{ij}$, $\delta$ and $\Delta m^2_{ij}=m_i^2-m_j^2$, extracted from global analyses of all neutrino oscillation data [@deSalas:2017kay; @Esteban:2016qun; @Capozzi:2016rtj], are given in Table \[datatable\] for both normally-ordered (NO) and inverted-ordered (IO) neutrino mass spectra defined as: $$\begin{aligned}
{\text {NO}:}& m_1 < m_2 < m_3\;\, ({\Delta m^2_{31}}>0)\,,\\
{\text {IO}:}& m_3 < m_1 < m_2\;\, ({\Delta m^2_{31}}<0)\,.
\label{NOIO}\end{aligned}$$ Notice that although neutrino mixing angles and mass-squared differences are known with very good precision, the experimental sensitivity to the value of $\delta$ is still limited, and the statistical significance of the presented ranges for that parameter is low.
**Parameter** Best Fit $\pm1\sigma$ $3\sigma$ range
---------------------------------------------------------------------- ------------------------ -----------------------
$\theta_{12}\;(^{\circ})$ $34.5_{-1.0}^{+1.1}$ $31.5\rightarrow38.0$
\[0.15cm\] $\theta_{23}\;(^{\circ})$ \[NO\] $41.0\pm1.1$ $38.3\rightarrow52.8$
$\theta_{23}\;(^{\circ})$ \[IO\] $50.5\pm1.0$ $38.5\rightarrow53.0$
\[0.15cm\] $\theta_{13}\;(^{\circ})$ \[NO\] $8.44_{-0.15}^{+0.18}$ $7.9\rightarrow8.9$
$\theta_{13}\;(^{\circ})$ \[IO\] $8.41_{-0.17}^{+0.16}$ $7.9\rightarrow8.9$
\[0.15cm\] $\delta\;(^{\circ})$ \[NO\] $252_{-36}^{+56}$ $0\rightarrow360$
$0\rightarrow31$
$142\rightarrow360$
\[0.15cm\] $\Delta m_{21}^2\;(\times 10^{-5}\;\text{eV}^2)$ $7.56\pm0.19$ $7.05\rightarrow8.14$
\[0.15cm\] $|\Delta m_{31}^2|\;(\times 10^{-3}\;\text{eV}^2)$ \[NO\] $2.55\pm0.04$ $2.43\rightarrow2.67$
$|\Delta m_{31}^2|\;(\times 10^{-3}\;\text{eV}^2)$ \[IO\] $2.49\pm0.04$ $2.37\rightarrow2.61$
: Neutrino oscillation parameters obtained from the global analysis of Ref. [@deSalas:2017kay] (see also Refs. [@Esteban:2016qun] and [@Capozzi:2016rtj]).[]{data-label="datatable"}
Let us now consider the simplest type-I seesaw model which can account for the data presented in Table \[datatable\], i.e. the 2RHNSM. In this case, ${\mathbf{Y}^\nu}$ and ${\mathbf{M}_R}$ are $3\times 2$ and $2 \times 2$ matrices, respectively. In the mass-eigenstate basis of $\nu_R$, the free parameters in the Lagrangian (\[LtypeI\]) are the two RH neutrino masses $M_{1,2}$, and the twelve real parameters of ${\mathbf{Y}^\nu}$. By rotating the LH charged-lepton fields, one is able to eliminate three parameters from ${\mathbf{Y}^\nu}$, leaving a total of eleven. Since for the 2RHNSM the effective neutrino mass matrix ${\mathbf{M}^\nu}$ given in Eq. (\[Mnuseesaw\]) is rank two, $m_1=0\,(m_3=0)$ for NO (IO).[^1] Moreover, the diagonal phase matrix in Eq. (\[Uparam\]) must be replaced by $\text{diag}(1,e^{i\alpha/2},1)$ since, in the presence of a massless neutrino, only one Majorana phase is physical. Thus, in the 2RHNSM, the low-energy neutrino sector is described by seven parameters (two masses, three mixing angles and two CP-violating phases), to be compared with the eleven at high energies.
One convenient way of parameterizing ${\mathbf{Y}^\nu}$ relies on the so-called Casas-Ibarra parametrization [@Casas:2001sr]. In the basis where both ${\mathbf{M}_R}$ and ${\mathbf{Y}^\ell}$ are diagonal, $$\begin{gathered}
\mathbf{Y}^\nu=v^{-1}{\mathbf{U}}^*\,\mathbf{d}_m^{1/2}\,\mathbf{R}\,\mathbf{d}_M^{1/2}\,,
\label{CasasandIbarra}\end{gathered}$$ with $\mathbf{d}_M=\text{diag}(M_1,M_2)$. The matrix $\mathbf{R}$ is a $3\times 2$ complex orthogonal matrix which can be parametrized by a single complex angle $z$ in the following way $$\begin{gathered}
\mathbf{R}_{\text{NH}}=\begin{pmatrix}
0&0\\
\cos z&-\sin z\\
\xi \sin z&\xi \cos z
\end{pmatrix}
\;,\;
\mathbf{R}_{\text{IH}}=\begin{pmatrix}
\cos z &-\sin z \\
\xi \sin z &\xi \cos z\\
0&0
\end{pmatrix}\; ,
\label{RmatrixIO}\end{gathered}$$ with $\xi=\pm 1$. Notice that, in the case of a non-diagonal ${\mathbf{M}_R}$, the right-hand side of Eq. (\[CasasandIbarra\]) must be multiplied on the right by ${\mathbf{U}_R}^\dag$, being ${\mathbf{U}_R}$ the unitary matrix which diagonalizes ${\mathbf{M}_R}$ as ${\mathbf{U}_R}^T {\mathbf{M}_R}{\mathbf{U}_R}=\mathbf{d}_M$.
Clearly, even in the simplest minimal type-I seesaw model, there are more free independent parameters at high energies than at low energies. In order to reduce the degree of arbitrariness of the 2RHNSM, in the next section we will introduce maximally-restricted texture zeros and study their phenomenological implications.
Maximally-restricted texture zeros {#sec2}
==================================
In this section we will study the implications of imposing texture zeros in ${\mathbf{Y}^\ell}$, ${\mathbf{Y}^\nu}$ and ${\mathbf{M}_R}$. Our guiding principle is to consider the maximum number of zeros such that the charged-lepton masses and neutrino data can be accommodated. In the former case, this corresponds to having six zeros in ${\mathbf{Y}^\ell}$, which guarantees three non-degenerate masses. There are six textures of this type related among each other by permutations of rows and/or columns applied to ${\mathbf{Y}^\ell}_{\rm diag}={\rm diag}(y_e,y_\mu,y_\tau)$, where $y_{e,\mu,\tau}=m_{e,\mu,\tau}/v$. Textures for ${\mathbf{Y}^\nu}$ with three or more zeros lead to vanishing mixing angles and/or two massless neutrinos, being therefore excluded experimentally. In principle, with two texture zeros in ${\mathbf{Y}^\nu}$, all neutrino data could be reproduced. There are fifteen different types of $3\times 2$ matrices with two vanishing entries. Some of them are automatically excluded by present neutrino data, namely,
- [Textures with two zeros placed in the same line $j$ of ${\mathbf{Y}^\nu}$ are excluded since these lead to the case in which the two RH neutrino fields are decoupled from the lepton flavor $j$. Therefore, all elements in line (and column) $j$ of the Majorana neutrino mass matrix ${\mathbf{M}^\nu}$ vanish, implying the existence of two vanishing mixing angles $\theta_{ij}$, which is excluded by the data. In practice, this corresponds to the situation in which one neutrino flavor state coincides with its mass eigenstate.]{}
- [If both zeros are placed in lines $(i,j)$ of the same column in ${\mathbf{Y}^\nu}$, then lines (and columns) $(i,j)$ of ${\mathbf{M}^\nu}$ are linearly dependent. Thus, at least one mixing angle $\theta_{ij}$ is zero, leading to the unrealistic case in which one flavor eigenstate is a superposition of only two of the three mass eigenstates.]{}
We therefore conclude that the maximally-allowed number of texture zeros in ${\mathbf{Y}^\nu}$ is two. The ${\mathbf{Y}^\nu}$ textures to be analyzed are of the type: $$\begin{gathered}
\begin{matrix}
{\rm T}_1:\;\begin{pmatrix}
0&\times\\
\times&0\\
\times&\times
\end{pmatrix},&
{\rm T_2}: \;\begin{pmatrix}
0&\times\\
\times&\times\\
\times&0
\end{pmatrix},&
{\rm T_3}:\; \begin{pmatrix}
\times&\times\\
0&\times\\
\times&0
\end{pmatrix},\\
&&&\\
{\rm T_4}:\; \begin{pmatrix}
\times&0\\
0&\times\\
\times&\times
\end{pmatrix},&
{\rm T_5}: \;\begin{pmatrix}
\times&0\\
\times&\times\\
0&\times
\end{pmatrix},&
{\rm T_6}:\; \begin{pmatrix}
\times&\times\\
\times&0\\
0&\times
\end{pmatrix},
\end{matrix}
\label{Tstructures}\end{gathered}$$ where the symbol $\times$ denotes a generic non-vanishing entry.
[K[1.5cm]{}|K[1.5cm]{}|c|K[1.2cm]{}|K[1.2cm]{}]{} ${\mathbf{Y}^\nu}$&${\mathbf{M}_R}$&${\mathbf{M}^\nu}$&**NH**&**IH**\
\[-0.15cm\][T$_1$, T$_2$]{}&\[-0.15cm\][R$_2$]{}&&\[-0.15cm\]&\[-0.25cm\]\
\[-0.15cm\][T$_4$, T$_5$]{}&\[-0.15cm\][R$_3$]{}&&\
T$_1$, T$_4$&R$_1$&B: $\begin{pmatrix}
\times &0&\times\\
\cdot&\times&\times\\
\cdot&\cdot&\times
\end{pmatrix}$&&(${1\sigma}$)\
T$_2$, T$_5$&R$_1$&C: $\begin{pmatrix}
\times &\times&0\\
\cdot&\times&\times\\
\cdot&\cdot&\times
\end{pmatrix}$&&(${1\sigma}$)\
\[-0.15cm\][T$_3$, T$_4$]{}&\[-0.15cm\][R$_2$]{}&&\[-0.15cm\]&\[-0.25cm\][(${1\sigma}$)]{}\
\[-0.15cm\][T$_1$, T$_6$]{}&\[-0.15cm\][R$_3$]{}&&\
T$_3$, T$_6$&R$_1$&E: $\begin{pmatrix}
\times &\times&\times\\
\cdot&\times&0\\
\cdot&\cdot&\times
\end{pmatrix}$&&\
\[-0.15cm\][T$_5$, T$_6$]{}&\[-0.15cm\][R$_2$]{}&&\[-0.15cm\]&\[-0.25cm\][(${3\sigma}$)]{}\
\[-0.15cm\][T$_2$, T$_3$]{}&\[-0.15cm\][R$_3$]{}&&\
As for ${\mathbf{M}_R}$, with more that two texture zeros, at least one of the RH neutrinos is massless. On the other hand, with one texture zero, there are three different patterns for ${\mathbf{M}_R}$: $$\begin{aligned}
{\rm R}_1:\begin{pmatrix}
\times&0\\
\cdot&\times
\end{pmatrix}\;,\;
{\rm R_2}:\begin{pmatrix}
0&\times\\
\cdot&\times
\end{pmatrix}\;,\;
{\rm R_3}:\begin{pmatrix}
\times&\times\\
\cdot&0
\end{pmatrix}\,,
\label{Rstructures}\end{aligned}$$ with the dot $(\cdot)$ indicating the symmetric nature of the matrix. Combining them with the ${\mathbf{Y}^\nu}$ textures (\[Tstructures\]) through the seesaw formula (\[Mnuseesaw\]), one obtains the textures for ${\mathbf{M}^\nu}$ given in the third column of Table \[tabR1R2R3\]. All cases A-F feature the presence of one texture zero in ${\mathbf{M}^\nu}$. Notice that sets of $({\mathbf{Y}^\nu},{\mathbf{M}_R})$ textures related by simultaneous permutations of the columns in ${\mathbf{Y}^\nu}$, and lines and columns in ${\mathbf{M}_R}$, lead to the same ${\mathbf{M}^\nu}$ due to invariance of Eq. (\[Mnuseesaw\]) under $\nu_R$ rotations. Moreover, when ${\mathbf{M}_R}$ is diagonal (texture R$_1$), ${\mathbf{M}^\nu}$ is the same for ${\mathbf{Y}^\nu}$ textures related by a column permutation. For instance, the sets $(\mathrm{T}_1,\mathrm{R}_1)$ and $(\mathrm{T}_4,\mathrm{R}_1)$ lead to the same low-energy predictions since $\mathrm{T}_1$ and $\mathrm{T}_4$ are related by column permutation.
The condition ${\mathbf{M}^\nu}_{\alpha\beta}=0$ imposes relations among the neutrino parameters. In particular, from Eq. (\[Mnudiag\]) it is straightforward to conclude that [@Xing:2003ic; @Gautam:2015kya] $$\begin{aligned}
\text{NH}:&\dfrac{m_2}{m_3}&\!\!=-\dfrac{{\mathbf{U}}^*_{\alpha 3}{\mathbf{U}}^*_{\beta 3}}{{\mathbf{U}}^*_{\alpha 2}{\mathbf{U}}^*_{\beta 2}}\;,
\label{NHrelation}\\
\text{IH}:&\dfrac{m_1}{m_2}&\!\!=-\dfrac{{\mathbf{U}}^*_{\alpha 2}{\mathbf{U}}^*_{\beta 2}}{{\mathbf{U}}^*_{\alpha 1}{\mathbf{U}}^*_{\beta 1}}\;.
\label{IHrelation}\end{aligned}$$ Taking into account that neutrino masses $m_i$ are real and positive, $m_2^2={\Delta m^2_{21}}$ ($m_2^2={\Delta m^2_{21}}+|{\Delta m^2_{31}}|$) and $m_3^2={\Delta m^2_{31}}$ ($m_1^2=|{\Delta m^2_{31}}|$) for NH (IH). Thus, we have $$\begin{aligned}
\text{NH}:\;\;\;\;\;\;\;r_\nu&=&\left|\dfrac{{\mathbf{U}}^*_{\alpha 3}{\mathbf{U}}^*_{\beta 3}}{{\mathbf{U}}^*_{\alpha 2}{\mathbf{U}}^*_{\beta 2}}\right|^2,
\label{NHrelationb}\\
\text{IH}:\dfrac{1}{1+r_\nu}&=&\left|\dfrac{{\mathbf{U}}^*_{\alpha 2}{\mathbf{U}}^*_{\beta 2}}{{\mathbf{U}}^*_{\alpha 1}{\mathbf{U}}^*_{\beta 1}}\right|^2,\;\;\;r_\nu\equiv \dfrac{{\Delta m^2_{21}}}{|{\Delta m^2_{31}}|}\,.
\label{IHrelationb}\end{aligned}$$
Given the parametrization in Eq. (\[Uparam\]), and the experimentally-allowed ranges for the mixing angles presented in Table \[datatable\], one can test which textures lead to viable values of $r_\nu$ using the above relations. From all cases, the simplest one to be analyzed is texture A, for which $r_\nu$ is simply given by $$\begin{aligned}
{\rm NH:} \,\,r_\nu=\frac{t_{13}^4}{s_{12}^4}\simeq 0.005\;,\;
{\rm IH:}\,\, r_\nu=\frac{1}{t_{12}^4}-1\simeq 3.5\,.\end{aligned}$$ These numerical estimates, obtained using the best-fit values given in Table \[datatable\], indicate that texture A is disfavored by data, independently of the value of $\delta$.
By varying the mixing angles in their experimentally $1\sigma$ and $3\sigma$ allowed regions,[^2] we plot $r_\nu$ as a function of $\delta$ in Figs. \[NHmassratios\] and \[IHmassratios\] for NH and IH, respectively, using Eqs. (\[NHrelationb\]) and (\[IHrelationb\]) together with Eq. (\[Uparam\]). In light (dark) blue we show the $r_\nu$ regions obtained when all mixing angles vary in their $3\sigma$ ($1\sigma$) experimental ranges. The horizontal pink bands (red line) indicate the $3\sigma$ experimental range (best-fit value) for $r_\nu$. From these results, we conclude that all textures with one zero in ${\mathbf{M}^\nu}$ are incompatible with neutrino data at more than $3\sigma$ level for NH. In the context of the 2RHNSM with texture zeros in ${\mathbf{Y}^\nu}$ and ${\mathbf{M}_R}$, this means that all combinations shown in Table \[tabR1R2R3\] are excluded for that type of neutrino mass spectrum. For IH (Fig. \[IHmassratios\]) and specific ranges of $\delta$, one obtains values for $r_\nu$ compatible with the data at $1\sigma$ for textures B, C and D, and only at $3\sigma$ for texture F. Therefore, all combinations of textures for ${\mathbf{Y}^\nu}$ and ${\mathbf{M}_R}$ leading to textures B, C, D and F for ${\mathbf{M}^\nu}$ are viable. Notice that only textures B and C predict $r_\nu$ values in its $1\sigma$ range, for $\delta$ around its best-fit value.
[ll]{} ${\mathbf{M}^\nu}\;$ &$\hspace{2cm}$**CP-violating phases**\
& $c_\delta=2\dfrac{[s^4_{12}(1+r_\nu)-c^4_{12}]s^2_{23}s^2_{13}+r_\nu c^2_{23}s_{12}^2c^2_{12}}{[s^2_{12}(1+r_\nu)+c^2_{12}]\sin(2\theta_{12})\sin(2\theta_{23})s_{13}}$
------------------------------------------------------------------------
------------------------------------------------------------------------
\
& $c_\alpha=\dfrac{(2+r_\nu)c^2_{23}s^2_{12}c_{12}^2-[s^4_{12}(1+r_\nu)+c^4_{12}]s^2_{23}s^2_{13}}{2\sqrt{1+r_\nu}(c^2_{23}+s^2_{23}s^2_{13})s^2_{12}c^2_{12}}$
------------------------------------------------------------------------
------------------------------------------------------------------------
\
& $c_\delta=-2\dfrac{[s^4_{12}(1+r_\nu)-c^4_{12}]c^2_{23}s^2_{13}+r_\nu s^2_{23}s_{12}^2c^2_{12}}{[s^2_{12}(1+r_\nu)+c^2_{12}]\sin(2\theta_{12})\sin(2\theta_{23})s_{13}}$
------------------------------------------------------------------------
------------------------------------------------------------------------
\
&$c_\alpha=\dfrac{(2+r_\nu)s^2_{23}s^2_{12}c_{12}^2-[s^4_{12}(1+r_\nu)+c^4_{12}]c^2_{23}s^2_{13}}{2\sqrt{1+r_\nu}(s^2_{23}+c^2_{23}s^2_{13})s^2_{12}c^2_{12}}$
------------------------------------------------------------------------
------------------------------------------------------------------------
\
& $c_\delta=2\dfrac{(c^2_{12}\sqrt{1+r_\nu}-s^2_{12})c^2_{23}+(s^2_{12}\sqrt{1+r_\nu}-c^2_{12})s^2_{23}s^2_{13}}{(\sqrt{1+r_\nu}+1)\sin(2\theta_{12})\sin(2\theta_{23})s_{13}}$
------------------------------------------------------------------------
------------------------------------------------------------------------
\
&$c_\alpha\simeq-\dfrac{3+\cos(4\theta_{12})-16s_{13}^2t_{23}^2}{2\sin^2(2\theta_{12})}$
------------------------------------------------------------------------
------------------------------------------------------------------------
\
& $c_\delta=2\dfrac{(s^2_{12}-c^2_{12}\sqrt{1+r_\nu})s^2_{23}+(c^2_{12}-s^2_{12}\sqrt{1+r_\nu})c^2_{23}s^2_{13}}{(\sqrt{1+r_\nu}+1)\sin(2\theta_{12})\sin(2\theta_{23})s_{13}}$
------------------------------------------------------------------------
------------------------------------------------------------------------
\
&$c_\alpha\simeq-\dfrac{3t_{23}^2+t_{23}^2\cos(4\theta_{12})+16s_{13}^2}{2t_{23}^2\sin^2(2\theta_{12})}$
------------------------------------------------------------------------
------------------------------------------------------------------------
\
\[CPtable\]
Having identified the compatible textures, we now obtain expressions for $\delta$ in terms of the mixing angles and $r_\nu$ using Eq. (\[IHrelationb\]). By imposing that the right-hand side of Eq. (\[IHrelation\]) is real, we can obtain analytical expressions for the Majorana phase $\alpha$ as a function of $\theta_{ij}$, $r_\nu$ and $\delta$. In Table \[CPtable\], we present the results for $c_\delta\equiv\cos\delta$ and $c_\alpha\equiv\cos\alpha$ for textures B, C, D and F when the neutrino mass spectrum is of IH type. It is worth mentioning that, although of different nature, $\delta$ and $\alpha$ are not independent phases in our case. This is due to the presence of zeros in the effective neutrino mass matrix. Taking $\theta_{ij}$, ${\Delta m^2_{21}}$ and ${\Delta m^2_{31}}$ in their $3\sigma$ $(1\sigma)$ experimental ranges, we show in Fig. \[deltaalpha\] the light (dark) blue allowed regions in the $(\alpha,\delta)$ parameter space for textures B, C, D and F of ${\mathbf{M}^\nu}$. We conclude that, for textures B and C, values of $\delta \simeq 3 \pi/2$ close to the best-fit value are allowed (cf. Table \[datatable\]). For such values of $\delta$, $\alpha \simeq 1.9 \pi\,(0.08\pi)$ is predicted for texture B (C). In fact, for these textures $$\begin{aligned}
{\rm B:}\;&c_\delta \simeq \frac{r_\nu \sin(2\theta_{12})}{4s_{13}t_{23}}-\frac{s_{13}t_{23}}{\tan(2\theta_{12})}\,,\\
{\rm C:}\;&c_\delta \simeq -\frac{r_\nu t_{23} \sin(2\theta_{12})}{4s_{13}}+\frac{s_{13}}{t_{23}\tan(2\theta_{12})},\end{aligned}$$ from which we see that $|c_\delta|\ll 1$, implying $\delta \simeq \pm \pi/2$. Instead, for textures D and F $$\begin{aligned}
{\rm D:}\;&c_\delta \simeq \frac{1}{2s_{13}t_{23}\tan(2\theta_{12})}\,,\\
{\rm F:}\;&c_\delta \simeq -\frac{t_{23}}{2s_{13}\tan(2\theta_{12})}\,,\end{aligned}$$ one obtains $|c_\delta| \sim \mathcal{O}(1)$ meaning that $\delta$ is far from $\pm \pi/2$. Therefore, as anticipated above, only textures B and C lead to $\delta$ values within the $1\sigma$ range of Table \[datatable\]. For textures D and F, the obtained values for $\delta$ are out of the $1\sigma$ range, but still within the $3\sigma$ one. Presently, attempts to probe the Majorana nature of neutrinos are mainly based on neutrinoless double beta decay ($0\nu\beta\beta$) experiments. The observation of $0\nu\beta\beta$ decay would also provide a measurement of the neutrino mass scale, since the rate of this process is related to the square of the neutrino mass. A relevant quantity for $0\nu\beta\beta$ decay is the effective mass $m_{\beta\beta}$, which, for an IH neutrino mass spectrum, is given by $$\begin{aligned}
m_{\beta\beta}&=\left|\sum_{i=1}^{3}m_i {\mathbf{U}}_{1i}^2\right|\nonumber\\
&=c_{13}^2|{\Delta m^2_{31}}|^{1/2}\,\left|c_{12}^2+(1+r_\nu)^{1/2}\,s_{12}^2e^{i\alpha} \right|.\end{aligned}$$ Given that $\alpha$ is a function of $\theta_{ij}$, $\delta$ and $r_\nu$, in Fig. \[mbbdelta\], we show the allowed regions in the $(m_{\beta\beta},\delta)$-plane, taking into account the experimental ranges for the neutrino parameters (the color codes are the same used in previous figures). The results are presented for textures B, C, D and F, where one can see that the value of $m_{\beta\beta}$ is around 50 meV (15 meV) for textures B and C (D and F). These values are compatible with all constraints coming from $0\nu\beta\beta$ decay and cosmological experiments [@Capozzi:2017ipn] for IH, but lie out of the sensitivity range of leading experiments like EXO-200 [@Albert:2017qto], KamLAND-Zen [@KamLAND-Zen:2016pfg], GERDA [@Agostini:2017dxu] and CUORE-0 [@Alduino:2016vtd]. Nevertheless, next-generation experiments will be able to test the IH spectrum (for a general discussion about future prospects and sensitivities of $0\nu\beta\beta$ decay experiments see e.g. Ref. [@Ostrovskiy:2016uyx]).
In the above analysis, we have studied the cases with one texture zero in ${\mathbf{M}_R}$. Notice, however, that the maximally-allowed number of zeros in this matrix is actually two, leading to a single possible texture $$\begin{aligned}
{\rm R_4:\,}\begin{pmatrix}
0&\times\\
\cdot&0
\end{pmatrix},
\label{R4text}\end{aligned}$$ which is characterized by a spectrum with two degenerate RH neutrinos. Combining through the seesaw formula (\[Mnuseesaw\]) the matrix R$_4$ with all ${\mathbf{Y}^\nu}$ textures presented in Eq. , one obtains the textures for ${\mathbf{M}^\nu}$ given in the third column of Table \[tabR4\]. One can see that in all cases ${\mathbf{M}^\nu}$ contains two zeros, which have been tested individually above.[^3] Moreover, additional relations among the elements of ${\mathbf{M}^\nu}$ (see fourth column of Table \[tabR4\]) arise due to the specific form of ${\mathbf{M}_R}$, which contains a single parameter. For NH, all cases with ${\rm R}_4$ are excluded, since all textures with one zero in ${\mathbf{M}^\nu}$ were already shown to be incompatible with data (see Table \[tabR1R2R3\]). For IH, combinations leading to textures ${\rm A}_1$ and ${\rm A}_2$ for ${\mathbf{M}^\nu}$ are excluded due to the condition ${\mathbf{M}^\nu}_{11}=0$ (see Table \[tabR1R2R3\]).
[K[1.2cm]{}|K[0.7cm]{}|c|K[2.1cm]{}|K[0.7cm]{}|K[0.7cm]{}]{} ${\mathbf{Y}^\nu}$&${\mathbf{M}_R}$&${\mathbf{M}^\nu}$&**Relation in** ${\mathbf{M}^\nu}$&**NH**&**IH**\
T$_1$, T$_4$&\[-1.5cm\][R$_4$]{}&A$_1$: $\begin{pmatrix}
0 &\times&\times\\
\cdot&0&\times\\
\cdot&\cdot&\times
\end{pmatrix}$&$\dfrac{{\mathbf{M}^\nu}_{33}}{2{\mathbf{M}^\nu}_{23}}=\dfrac{{\mathbf{M}^\nu}_{13}}{{\mathbf{M}^\nu}_{12}}$&&\
T$_2$, T$_5$&&A$_2$: $\begin{pmatrix}
0 &\times&\times\\
\cdot&\times&\times\\
\cdot&\cdot&0
\end{pmatrix}$&$\dfrac{{\mathbf{M}^\nu}_{22}}{2{\mathbf{M}^\nu}_{23}}=\dfrac{{\mathbf{M}^\nu}_{12}}{{\mathbf{M}^\nu}_{13}}$&&\
T$_3$, T$_6$&&D$_1$ : $\begin{pmatrix}
\times &\times&\times\\
\cdot&0&\times\\
\cdot&\cdot&0
\end{pmatrix}$&$\dfrac{{\mathbf{M}^\nu}_{11}}{2{\mathbf{M}^\nu}_{12}}=\dfrac{{\mathbf{M}^\nu}_{13}}{{\mathbf{M}^\nu}_{23}}$&&\
![Predictions for the low-energy phases $\delta$ and $\alpha$ for textures B, C, D and F, using the $3\sigma$ (light blue) and $1\sigma$ (dark blue) ranges given in Table \[datatable\] for the mixing angles and neutrino mass-squared differences. The black dot corresponds to the predictions obtained with the best-fit values of $~\theta_{ij}$, ${\Delta m^2_{21}}$ and $|{\Delta m^2_{31}}|$.[]{data-label="deltaalpha"}](fig3.pdf "fig:")\
![Predictions for $\delta$ and $m_{\beta\beta}$ for textures B, C, D and F, using the $3\sigma$ (light blue) and $1\sigma$ (dark blue) ranges given in Table \[datatable\] for the mixing angles and neutrino mass-squared differences. The black dot corresponds to the predictions obtained with the best-fit values of $~\theta_{ij}$, ${\Delta m^2_{21}}$ and $|{\Delta m^2_{31}}|$.[]{data-label="mbbdelta"}](fig4.pdf "fig:")\
As for texture D$_1$, although the conditions ${\mathbf{M}^\nu}_{22}=0$ and ${\mathbf{M}^\nu}_{33}=0$ are individually compatible with the data at $3\sigma$, they cannot be simultaneously verified, as one can see in Fig. \[IHmassratios\], comparing the results for textures D and F. Indeed, from these plots one concludes that there is no overlap between the regions allowed by the data for the same values of $\delta$. This seems to contradict previous results obtained in the literature which state that textures with ${\mathbf{M}^\nu}_{22}={\mathbf{M}^\nu}_{33}=0$ are compatible with the data (see e.g. Ref [@Cebola:2015dwa]). Notice, however, that the results in those references were obtained for a general neutrino spectrum with $m_{1,2,3}\neq 0$. One can understand why texture D$_1$ in our case ($m_3=0$) is not valid by inspecting the relations between neutrino masses and ${\mathbf{U}}$ when the conditions ${\mathbf{M}^\nu}_{22}={\mathbf{M}^\nu}_{33}=0$ are imposed, namely [@Fritzsch:2011qv], $$\begin{aligned}
\frac{m_3}{m_1}=\left|\frac{{\mathbf{U}}_{22}^2{\mathbf{U}}_{31}^2-{\mathbf{U}}_{21}^2{\mathbf{U}}_{32}^2}{{\mathbf{U}}_{23}^2{\mathbf{U}}_{32}^2-{\mathbf{U}}_{22}^2{\mathbf{U}}_{33}^2}\right|\,,\\
\frac{m_3}{m_2}=\left|\frac{{\mathbf{U}}_{22}^2{\mathbf{U}}_{31}^2-{\mathbf{U}}_{21}^2{\mathbf{U}}_{32}^2}{{\mathbf{U}}_{21}^2{\mathbf{U}}_{33}^2-{\mathbf{U}}_{23}^2{\mathbf{U}}_{31}^2}\right|\,.\end{aligned}$$ Therefore, if $m_3=0$ the condition $$\begin{aligned}
\left|{\mathbf{U}}_{22}^2{\mathbf{U}}_{31}^2-{\mathbf{U}}_{21}^2{\mathbf{U}}_{32}^2\right|=0\end{aligned}$$ must be verified for texture D$_1$. The above relation can be approximately written as $$\begin{aligned}
c_\delta\simeq \dfrac{2\cos(2\theta_{12})\cos(2\theta_{23})\pm\sqrt{2}\sqrt{\cos(4\theta_{12})+\cos(4\theta_{23})}}{4\sin(2\theta_{12})\sin(2\theta_{23})s_{13}}\,,\end{aligned}$$ which, taking into account the current mixing angle data, always leads to a complex $c_\delta$. In conclusion, we have analyzed all possible textures with six zeros in ${\mathbf{Y}^\ell}$, two zeros in ${\mathbf{Y}^\nu}$, and one or two zeros in ${\mathbf{M}_R}$. The compatibility of all textures is summarized in the last two columns of Tables \[tabR1R2R3\] and \[tabR4\], for NH and IH. We remark that no restriction has been imposed in the non-zero elements of those matrices.
The results presented above are valid in the basis where ${\mathbf{Y}^\ell}={\rm diag}(y_e,y_\mu,y_\tau)\equiv {\mathbf{Y}^\ell}_{\rm diag}$ so that the charged lepton mass matrix is $\mathbf{M}^\ell={\rm diag}(m_e,m_\mu,m_\tau)$. One may wonder whether these conclusions hold for any other ${\mathbf{Y}^\ell}$ with six zeros and only three nonzero elements. First, it is straightforward to see that any two nonzero elements in the same line/column lead to a massless charged lepton. This leaves us with six viable textures for ${\mathbf{Y}^\ell}$ with six zeros, which can be obtained from ${\mathbf{Y}^\ell}_{\rm diag}$ by applying permutations of lines and/or columns: $$\begin{gathered}
\begin{matrix}
{\rm L}_1:\;\begin{pmatrix}
\times&0 &0\\
0&\times &0\\
0&0 &\times
\end{pmatrix},&
{\rm L_2}: \;\begin{pmatrix}
0&\times &0\\
\times&0 &0\\
0&0 &\times
\end{pmatrix},&
{\rm L_3}:\; \begin{pmatrix}
0&0 &\times\\
0&\times &0\\
\times&0 &0
\end{pmatrix},\\
&&&\\
{\rm L}_4:\;\begin{pmatrix}
\times&0 &0\\
0&0 &\times\\
0&\times &0
\end{pmatrix},&
{\rm L_5}: \;\begin{pmatrix}
0&0 &\times\\
\times&0 &0\\
0&\times &0
\end{pmatrix},&
{\rm L_6}:\; \begin{pmatrix}
0&\times &0\\
0&0 &\times\\
\times&0 &0
\end{pmatrix}.
\end{matrix}
\label{Ylstructures}\end{gathered}$$ Obviously, if only column permutations (rotation of RH charged-lepton fields) are performed, then the results for a specific set of ${\mathbf{Y}^\nu}$ and ${\mathbf{M}_R}$ textures remain unchanged. However, if a permutation of the lines $i$ and $j$ in ${\mathbf{Y}^\ell}$ is involved (rotation of LH charged-lepton fields by the permutation matrix $\mathbf{P}_{ij}$), then the same line permutation has to be performed in ${\mathbf{Y}^\nu}$. At the effective level, this corresponds to permuting the lines and columns $i$ and $j$ in the effective neutrino mass matrix ${\mathbf{M}^\nu}$. Under these rotations, textures T$_1$-T$_6$ of ${\mathbf{Y}^\nu}$ and, consequently, A-F of ${\mathbf{M}^\nu}$, are transformed among themselves. Thus, even if a given texture pair (${\mathbf{Y}^\nu}$,${\mathbf{M}_R}$) is not compatible with data in the ${\mathbf{Y}^\ell}_{\rm diag}$ basis, this may not be the case in another ${\mathbf{Y}^\ell}$ basis obtained from a line permutation $\mathbf{P}_{ij}$.
To check the viability of a given set of textures $({\mathbf{Y}^\ell},{\mathbf{Y}^\nu},{\mathbf{M}_R};{\mathbf{M}^\nu})=({\rm L}_i,{\rm T}_i,{\rm R}_i;\text{A-F})$ one has to identify the permutation $\mathbf{P}_{ij}$ which brings $\text{L}_i$ to ${\mathbf{Y}^\ell}_{\rm diag}$, and find the transformed ${\mathbf{M}^\nu}$ texture. For instance, consider the case $({\mathbf{Y}^\ell}_{\rm diag},{\rm T}_3,{\rm R}_1;{\rm E})$, shown in Table \[tabR1R2R3\] to be incompatible with data. Under $\mathbf{P}_{13}$, ${\mathbf{Y}^\ell}_{\rm diag}$ is transformed into ${\rm L}_3$, while texture E becomes texture B, which is compatible with data at $1\sigma$. Therefore, although the set $({\mathbf{Y}^\ell}_{\rm diag},{\rm T}_3,{\rm R}_1;{\rm E})$ is not viable, the set $({\rm L}_3,{\rm T}_3,{\rm R}_1;{\rm E})$ is, since it corresponds to $({\mathbf{Y}^\ell}_{\rm diag},{\rm T}_4,{\rm R}_1;{\rm B})$ under $\mathbf{P}_{13}$. In Table \[permut\] we summarize the transformation properties of each ${\mathbf{M}^\nu}$ texture under line permutations $\mathbf{P}_{ij}$, identifying in each case the compatibility with data taking into account the results obtained for ${\mathbf{Y}^\ell}_{\rm diag}$ given in Table \[tabR1R2R3\]. Notice that when ${\mathbf{M}_R}$ is of type $\text{R}_4$, the results presented in Table \[tabR4\] are valid for any ${\mathbf{Y}^\ell}$ texture of type $\text{L}_i$. This is due to the fact that, under any permutation of lines and/or columns in ${\mathbf{Y}^\ell}$, textures A$_{1,2}$ and D$_1$ (which are all excluded by data) transform among themselves.
[K[1.0cm]{}l|K[1cm]{}l|K[1cm]{}l|K[1cm]{}l]{} &&&\
A&&D&($1\sigma$)&F&($3\sigma$)&A&\
B&($1\sigma$)&B&($1\sigma$)&E&&C&($1\sigma$)\
C&($1\sigma$)&E&&C&($1\sigma$)&B&($1\sigma$)\
D&($1\sigma$)&A&&D&($1\sigma$)&F&($3\sigma$)\
E&&C&($1\sigma$)&B&($1\sigma$)&E&\
F&($3\sigma$)&F&($3\sigma$)&A&&D&($1\sigma$)\
Imposing relations among the elements of ${\mathbf{Y}^\nu}$ {#sec3a}
-----------------------------------------------------------
We now intend to further restrict the two texture zero patterns analyzed above by imposing equality relations among the elements of ${\mathbf{Y}^\nu}$. The first obvious choice would be to consider all elements in ${\mathbf{Y}^\nu}$ to be equal. However, one can show that the eigenvector associated to $m_3=0$ is always $v_3=(\mp 1,-1,1)/\sqrt{3}$, leading to $s_{13}=\pm 1/\sqrt{3}$, which is excluded by the data. Thus, we move to the analysis of textures with two zeros in ${\mathbf{Y}^\nu}$ and three equal elements. Each case will be denoted by the labels of ${\mathbf{Y}^\nu}$, ${\mathbf{M}_R}$ and corresponding ${\mathbf{M}^\nu}$ (see first column of Table \[Tabrel\]), and indexes of the ${\mathbf{Y}^\nu}$ equal elements (see second column of Table \[Tabrel\]). For instance, the cases with ${\mathbf{Y}^\nu}_{21}={\mathbf{Y}^\nu}_{31}={\mathbf{Y}^\nu}_{12}$ are denoted by $(21,31,12)$. Due to the highly constrained form of the involved matrices, extra relations among the elements of ${\mathbf{M}^\nu}$ arise. These are shown in the third column of Table \[Tabrel\] for all possible combinations. Compatibility with neutrino data is determined by checking whether those relations are verified taking the allowed ranges for the neutrino parameters given in Table \[datatable\]. Also notice that the heavy Majorana neutrino masses and the elements of ${\mathbf{M}^\nu}$ are related. In particular, defining the ratio
[K[2.4cm]{}|K[1.6cm]{}|K[3.6cm]{}|K[3.8cm]{}|K[0.8cm]{}|c|c]{} (${\mathbf{Y}^\nu}$, ${\mathbf{M}_R}$, ${\mathbf{M}^\nu}$)&**Equal elements in** ${\mathbf{Y}^\nu}$& **Relations in** ${\mathbf{M}^\nu}$&$r_N\equiv M_2/M_1$&**IH**&
-------------------------------------------------------------------
**Low energy predictions**
$(\theta_{12},\theta_{23},\theta_{13})^\circ$
$({\Delta m^2_{31}},{\Delta m^2_{21}})\,\times10^{-3}\text{eV}^2$
$(\delta,\alpha)^\circ\;,\;m_{\beta\beta}\,(\text{meV})$
-------------------------------------------------------------------
&$r_N$\
&$(21,31,12)$& ${\mathbf{M}^\nu}_{22}={\mathbf{M}^\nu}_{23}$ &$r_N=\left|\dfrac{{\mathbf{M}^\nu}_{22}}{{\mathbf{M}^\nu}_{11}}\right|$&&&1.91
------------------------------------------------------------------------
------------------------------------------------------------------------
\
&$(21,31,32)$&$\dfrac{{\mathbf{M}^\nu}_{11}({\mathbf{M}^\nu}_{33}-{\mathbf{M}^\nu}_{22})}{({\mathbf{M}^\nu}_{13})^2}=1$&$r_N=\left|\dfrac{{\mathbf{M}^\nu}_{11}{\mathbf{M}^\nu}_{22}}{({\mathbf{M}^\nu}_{13})^2}\right|$&&&12.00
------------------------------------------------------------------------
------------------------------------------------------------------------
\
&$(21,12,32)$& ${\mathbf{M}^\nu}_{11}={\mathbf{M}^\nu}_{13}$ &$r_N=\left|\dfrac{{\mathbf{M}^\nu}_{22}}{{\mathbf{M}^\nu}_{11}}\right|$&&&$-$
------------------------------------------------------------------------
------------------------------------------------------------------------
\
&$(31,12,32)$&$\dfrac{{\mathbf{M}^\nu}_{22}({\mathbf{M}^\nu}_{33}-{\mathbf{M}^\nu}_{11})}{({\mathbf{M}^\nu}_{23})^2}=1$&$r_N=\left|\dfrac{({\mathbf{M}^\nu}_{23})^2}{{\mathbf{M}^\nu}_{11}{\mathbf{M}^\nu}_{22}}\right|$&&&$-$
------------------------------------------------------------------------
------------------------------------------------------------------------
\
&$(21,31,12)$& ${\mathbf{M}^\nu}_{23}={\mathbf{M}^\nu}_{33}$, &$r_N=\left|\dfrac{{\mathbf{M}^\nu}_{33}}{{\mathbf{M}^\nu}_{11}}\right|$&&&1.91
------------------------------------------------------------------------
------------------------------------------------------------------------
\
&$(21,31,22)$&$\dfrac{{\mathbf{M}^\nu}_{11}({\mathbf{M}^\nu}_{22}-{\mathbf{M}^\nu}_{33})}{({\mathbf{M}^\nu}_{12})^2}=1$&$r_N=\left|\dfrac{{\mathbf{M}^\nu}_{11}{\mathbf{M}^\nu}_{33}}{({\mathbf{M}^\nu}_{12})^2}\right|$&&&12.00
------------------------------------------------------------------------
------------------------------------------------------------------------
\
&$(21,12,22)$& ${\mathbf{M}^\nu}_{11}={\mathbf{M}^\nu}_{12}$, &$r_N=\left|\dfrac{({\mathbf{M}^\nu}_{23})^2}{{\mathbf{M}^\nu}_{11}{\mathbf{M}^\nu}_{33}}\right|$&&&$-$
------------------------------------------------------------------------
------------------------------------------------------------------------
\
&$(31,12,22)$&$\dfrac{{\mathbf{M}^\nu}_{33}({\mathbf{M}^\nu}_{22}-{\mathbf{M}^\nu}_{11})}{({\mathbf{M}^\nu}_{23})^2}=1$&$r_N=\left|\dfrac{{\mathbf{M}^\nu}_{33}}{{\mathbf{M}^\nu}_{11}}\right|$&&&$-$
------------------------------------------------------------------------
------------------------------------------------------------------------
\
&$(11,31,12)$&&$\dfrac{r_N-1}{r_N-\sqrt{r_N}-1}=\left|\dfrac{{\mathbf{M}^\nu}_{33}}{{\mathbf{M}^\nu}_{13}}\right|$&&&$-$
------------------------------------------------------------------------
------------------------------------------------------------------------
\
&$(11,31,22)$&&$\dfrac{\sqrt{r_N}}{r_N-1}=\left|\dfrac{{\mathbf{M}^\nu}_{23}}{{\mathbf{M}^\nu}_{33}}\right|$&&&$-$
------------------------------------------------------------------------
------------------------------------------------------------------------
\
&$(11,12,22)$&&$\dfrac{\sqrt{r_N}}{r_N-1}=\left|\dfrac{({\mathbf{M}^\nu}_{23})^2}{{\mathbf{M}^\nu}_{12}{\mathbf{M}^\nu}_{33}}\right|$&&&$-$
------------------------------------------------------------------------
------------------------------------------------------------------------
\
&$(31,12,22)$&&$\dfrac{\sqrt{r_N}}{r_N-1}=\left|\dfrac{{\mathbf{M}^\nu}_{23}}{{\mathbf{M}^\nu}_{33}}\right|$&&&$-$
------------------------------------------------------------------------
------------------------------------------------------------------------
\
&$(11,21,12)$&&$\dfrac{r_N-1}{r_N-\sqrt{r_N}-1}=\left|\dfrac{{\mathbf{M}^\nu}_{22}}{{\mathbf{M}^\nu}_{12}}\right|$&&&$-$
------------------------------------------------------------------------
------------------------------------------------------------------------
\
&$(11,21,32)$&&$\dfrac{\sqrt{r_N}}{r_N-1}=\left|\dfrac{{\mathbf{M}^\nu}_{23}}{{\mathbf{M}^\nu}_{22}}\right|$&&&$-$
------------------------------------------------------------------------
------------------------------------------------------------------------
\
&$(11,12,32)$&&$\dfrac{\sqrt{r_N}}{r_N-1}=\left|\dfrac{({\mathbf{M}^\nu}_{23})^2}{{\mathbf{M}^\nu}_{13}{\mathbf{M}^\nu}_{22}}\right|$&&&$-$
------------------------------------------------------------------------
------------------------------------------------------------------------
\
&$(21,12,32)$&&$\dfrac{\sqrt{r_N}}{r_N-1}=\left|\dfrac{{\mathbf{M}^\nu}_{23}}{{\mathbf{M}^\nu}_{22}}\right|$&&&$-$
------------------------------------------------------------------------
------------------------------------------------------------------------
\
&$(21,31,12)$&&$\dfrac{\sqrt{r_N}}{r_N-1}=\left|\dfrac{{\mathbf{M}^\nu}_{12}}{{\mathbf{M}^\nu}_{11}}\right|$&&&1.46
------------------------------------------------------------------------
------------------------------------------------------------------------
\
&$(21,31,32)$&&$\dfrac{\sqrt{r_N}}{r_N-1}=\left|\dfrac{({\mathbf{M}^\nu}_{12})^2}{{\mathbf{M}^\nu}_{11}{\mathbf{M}^\nu}_{23}}\right|$&&&1.08
------------------------------------------------------------------------
------------------------------------------------------------------------
\
&$(21,12,32)$&&$\dfrac{\sqrt{r_N}}{r_N-1}=\left|\dfrac{{\mathbf{M}^\nu}_{12}}{{\mathbf{M}^\nu}_{11}}\right|$&&&$-$
------------------------------------------------------------------------
------------------------------------------------------------------------
\
&$(31,12,32)$&&$\dfrac{r_N-1}{r_N-\sqrt{r_N}-1}=\left|\dfrac{{\mathbf{M}^\nu}_{11}}{{\mathbf{M}^\nu}_{13}}\right|$&&&$-$
------------------------------------------------------------------------
------------------------------------------------------------------------
\
&$(21,31,12)$&&$\dfrac{\sqrt{r_N}}{r_N-1}=\left|\dfrac{{\mathbf{M}^\nu}_{13}}{{\mathbf{M}^\nu}_{11}}\right|$&&&1.46
------------------------------------------------------------------------
------------------------------------------------------------------------
\
&$(21,31,22)$&&$\dfrac{\sqrt{r_N}}{r_N-1}=\left|\dfrac{({\mathbf{M}^\nu}_{13})^2}{{\mathbf{M}^\nu}_{11}{\mathbf{M}^\nu}_{23}}\right|$&&&1.08
------------------------------------------------------------------------
------------------------------------------------------------------------
\
&$(21,12,22)$&&$\dfrac{r_N-1}{r_N-\sqrt{r_N}-1}=\left|\dfrac{{\mathbf{M}^\nu}_{11}}{{\mathbf{M}^\nu}_{12}}\right|$&&&$-$
------------------------------------------------------------------------
------------------------------------------------------------------------
\
&$(31,12,22)$&&$\dfrac{\sqrt{r_N}}{r_N-1}=\left|\dfrac{{\mathbf{M}^\nu}_{13}}{{\mathbf{M}^\nu}_{11}}\right|$&&&$-$
------------------------------------------------------------------------
------------------------------------------------------------------------
\
$$\begin{aligned}
r_N=\frac{M_2}{M_1}\,,
\label{rNdef}\end{aligned}$$
where $M_{2,1}$ are the eigenvalues of ${\mathbf{M}_R}$, we obtain the relations shown in the fourth column of Table \[Tabrel\]. Our analysis shows that only eight combinations are compatible with neutrino data at the $3\sigma$ level (see fifth and sixth columns of Table \[Tabrel\]). The low-energy predictions for the neutrino parameters correspond to the case in which the data is best fitted. It is possible to show analytically that, for all compatible sets of matrices, $\theta_{23}=\pi/4$, which is confirmed by the numerical result. It is worth mentioning that any texture combination obtained from those presented in Table \[Tabrel\] by permuting the columns of ${\mathbf{Y}^\nu}$ will remain valid. For instance, the first case shown in Table \[Tabrel\] becomes (T$_4$,R$_1$,B) with equal elements (11,22,32), leading to the same predictions. Therefore, there are actually sixteen different cases compatible with the data. As mentioned above, the equality among elements of ${\mathbf{Y}^\nu}$ fixes the value of $r_N$, which is indicated in the last column of Table \[Tabrel\]. From inspection of the same table, one can also conclude that none of the texture configurations is compatible with the data at $1\sigma$.
As in the analysis presented in the previous section, the results obtained with equal ${\mathbf{Y}^\nu}$ elements correspond to ${\mathbf{Y}^\ell}={\mathbf{Y}^\ell}_{\rm diag}$. For a different ${\mathbf{Y}^\ell}$ texture related to ${\mathbf{Y}^\ell}_{\rm diag}$ by permutations of lines (and columns), the ${\mathbf{Y}^\nu}$ textures transform among themselves, and the equal elements of ${\mathbf{Y}^\nu}$ change position. Thus, combinations which are incompatible with data (see Table \[Tabrel\]) in the charged-lepton mass basis may become compatible for a non-diagonal ${\mathbf{Y}^\ell}$, related to ${\mathbf{Y}^\ell}_{\rm diag}$ by permutation of lines. In Table \[tabrelper\] we summarize the transformation properties of each combination $({\rm T}_i,\text{R}_i,\text{A-F})$ with equal ${\mathbf{Y}^\nu}$ elements under line permutations $\mathbf{P}_{ij}$ (and up to possible column permutation). In each case, we identify the compatibility with data taking into account the results given in Table \[Tabrel\] for ${\mathbf{Y}^\ell}={\mathbf{Y}^\ell}_{\rm diag}$.
[K[1.5cm]{}K[0.35cm]{}|K[1.5cm]{}K[0.35cm]{}|K[1.5cm]{}K[0.35cm]{}|K[1.5cm]{}K[0.35cm]{}]{} &&&
------------------------------------------------------------------------
------------------------------------------------------------------------
\
&&&
------------------------------------------------------------------------
------------------------------------------------------------------------
\
(21,31,12)&&(21,12,32)&&&&(21,31,12)&\
(21,31,32)&&(31,12,32)&& & &(21,31,22)&\
(21,12,32)&&(21,31,12)&& & &(31,12,22)&\
(31,12,32)&&(21,31,32)&& & &(21,12,22)&\
&&&
------------------------------------------------------------------------
------------------------------------------------------------------------
\
(21,31,12)&&&&(31,12,22)&&(21,31,12)&\
(21,31,22)&&&&(21,12,22)&& (21,31,32)&\
(21,12,22)&&&&(21,31,22)&&(31,12,32)&\
(31,12,22)&&&&(21,31,12)&& (21,12,32)&\
&&&
------------------------------------------------------------------------
------------------------------------------------------------------------
\
(11,31,12)&&&&(31,12,32)&&(11,21,12)&\
(11,31,22)&& &&(21,12,32)&& (11,21,32)&\
(11,12,22)&& &&(21,31,32)&& (11,12,32)&\
(31,12,22)&&&&(21,31,12)&& (21,12,32)&\
&&&
------------------------------------------------------------------------
------------------------------------------------------------------------
\
(11,21,12)&&(21,12,22)&&&&(11,31,12)&\
(11,21,32)&&(31,12,22)&&& &(11,31,22)&\
(11,12,32)&& (21,31,22)&&& &(11,12,22)&\
(21,12,32)&&(21,31,12)&&& &(31,12,22)&\
&&&
------------------------------------------------------------------------
------------------------------------------------------------------------
\
(21,31,12)&&&&(31,12,22)&&(21,31,12)&\
(21,31,32)&& &&(11,12,22)&&(21,31,22)&\
(21,12,32)&& &&(11,31,22)&& (31,12,22)&\
(31,12,32)&& &&(11,31,12)&& (21,12,22)&\
&&&
------------------------------------------------------------------------
------------------------------------------------------------------------
\
(21,31,12)&&(21,12,32)&&&&(21,31,12)&\
(21,31,22)&& (11,12,32)&&& &(21,31,32)&\
(21,12,22)&&(11,21,12)&&& &(31,12,32)&\
(31,12,22)&&(11,21,32)&&& &(21,12,32)&\
Leptogenesis in the 2RHNSM with texture zeros {#sec3}
=============================================
In the previous sections, several mass matrix patterns were found to be compatible with current neutrino oscillation data at $1\sigma$ and $3\sigma$ C.L., in the framework of the minimal type-I seesaw model with maximally restricted texture zeros. Here, we further analyze these patterns requiring their compatibility with successful leptogenesis [@Fukugita:1986hr]. We recall that the baryon asymmetry of the Universe is parametrized through the baryon-to-photon ratio $$\begin{gathered}
\eta_B\equiv\dfrac{n_B-n_{\bar{B}}}{n_\gamma}\; ,
\label{baryontophotonratio}\end{gathered}$$ where $n_B$, $n_{\bar{B}}$ and $n_\gamma$ are the number densities of baryons, anti-baryons and photons, respectively. From cosmic microwave background (CMB) measurements provided by the Planck collaboration [@Ade:2015xua], the present value of $\eta_B$ is $$\begin{gathered}
\eta_B^0=(6.11\pm0.04)\times 10^{-10}\,.
\label{presentetab}\end{gathered}$$
In a minimal type-I seesaw context with two right-handed neutrinos, the leptogenesis mechanism may proceed via the out-of-equilibrium decays of the heavy neutrinos $N_1$ and $N_2$ in the early Universe. The generated lepton asymmetry in such decays is partially converted into a baryon asymmetry by ($B+L$)-violating sphaleron processes, leading to [@Antusch:2011nz] $$\begin{gathered}
\eta_B=a_\text{sph}\,\dfrac{N_{B-L}}{N_\gamma^\text{rec}}\simeq 9.58\times 10^{-3}\, N_{B-L}\; ,
\label{finaletab}\end{gathered}$$ where $a_\text{sph} \equiv B/(B-L)=28/79$ is the conversion factor, $N_{B-L}$ is the final asymmetry calculated in a comoving volume, and $N_{\gamma}^\text{rec}$ is the number of photons in the same volume ($N_{\gamma}^\text{rec}\simeq 37.01$) at the recombination temperature.
Flavored and unflavored CP asymmetries {#subsecCP}
--------------------------------------
An important ingredient in the generation of the BAU is the CP asymmetry produced in the decays of the heavy neutrinos into the lepton flavors $\alpha=e,\mu,\tau$. Working in the mass eigenbasis of the heavy neutrinos $N_i$ and the charged leptons $\ell_\alpha$, the CP asymmetries $\epsilon_i^\alpha$ may be computed as [@Covi:1996wh] $$\begin{gathered}
\epsilon_i^\alpha=\dfrac{\Gamma(N_i\rightarrow\Phi\ell_\alpha)-\Gamma(N_i\rightarrow\Phi^\dagger\bar{\ell}_\alpha)}{\sum_\beta[\Gamma(N_i\rightarrow\Phi\ell_\beta)+\Gamma(N_i\rightarrow\Phi^\dagger\bar{\ell}_\beta)]},
\label{generalCP}\end{gathered}$$ where $\Gamma(N_i\rightarrow\Phi\ell_\alpha)\equiv \Gamma_i^\alpha$ and $\Gamma(N_i\rightarrow\Phi^\dagger\bar{\ell}_\alpha)\equiv \overline{\Gamma}_i^\alpha$ are the $N_i$ decay rates into leptons and antileptons, respectively. At tree level, $$\begin{gathered}
\Gamma_i^\alpha=\overline{\Gamma}_i^\alpha=M_i\dfrac{|\mathbf{Y}^\nu_{\alpha i}|^2}{16 \pi},
\label{flvdecayrate}\end{gathered}$$ with the sum in the denominator of running over the three lepton flavors. The leading non-zero contributions to the asymmetry $\epsilon_i^\alpha$ arise from interference of the tree-level process with its one-loop corrections. For the two RH neutrino case, the result is [@Branco:2011zb] $$\begin{aligned}
\epsilon_i^\alpha =&\frac{1}{8\pi}\frac{1}{\mathbf{H}_{ii}^\nu}\{\text{Im}[\mathbf{Y}_{\alpha i}^{\nu *}\mathbf{H}_ {ij}^\nu\mathbf{Y}_{\alpha j}^\nu ][f(x_j)+g(x_j)]+\nonumber\\ &\text{Im}[\mathbf{Y}_{\alpha i}^{\nu *}\mathbf{H}_ {ji}^\nu\mathbf{Y}_{\alpha j}^{\nu}]g'(x_j)\},
\label{flavouredcp}\end{aligned}$$ where $j\neq i=1,2$, $x_j=M_j^2/M_i^2$ and $\mathbf{H}^\nu=\mathbf{Y}^{\nu\dagger} \mathbf{Y}^\nu$. The loop functions $f(x)$, $g(x)$ and $g'(x)$ correspond to the one-loop vertex and self-energy corrections, given by $$\begin{aligned}
\label{loopfunctions1}
f(x)&=\sqrt{x}\left[1-(1-x)\ln\left(1+\frac{1}{x}\right)\right],\\
g(x)&=\sqrt{x}g'(x)=-\frac{\sqrt{x}}{(x-1)}\,.\end{aligned}$$ Summing over the lepton flavors in Eq. , the unflavored CP asymmetry is recovered, $$\begin{gathered}
\epsilon_i=\frac{1}{8\pi}\frac{1}{\mathbf{H}_{ii}^\nu}\text{Im}[(\mathbf{H}_{ij}^\nu)^2][f(x_j)+g(x_j)].
\label{unflavouredcp}\end{gathered}$$
In our study, two temperature regimes will be of interest [@Barbieri:1999ma; @Abada:2006fw; @Nardi:2006fx; @Abada:2006ea]. For temperatures above $10^{12}$ GeV in the early Universe, the charged-lepton Yukawa interactions are out of equilibrium. Hence, for this temperature range, the three lepton flavors are indistinguishable (unflavored regime), and the lepton asymmetry may be represented rigorously by a single flavor eigenstate. In this case, the relevant CP asymmetry for leptogenesis is given by Eq. . In the temperature interval $10^{9}\lesssim T \lesssim 10^{12}$ GeV, the $\tau$ Yukawa interactions enter thermal equilibrium and processes involving leptons are able to distinguish between two different flavors: the $\tau$ and a coherent superposition of $e$ and $\mu$ (two-flavored regime). The corresponding CP asymmetries, $\epsilon_i^\tau$ and $\epsilon_i^\gamma\equiv \epsilon_i^e+\epsilon_i^\mu$, are then obtained from Eq. . The CP asymmetries given in Eq. depend on the Yukawa coupling matrix ${\mathbf{Y}^\nu}$, which can be written in terms of the Casas-Ibarra parametrization presented in Eq. . This allows to rewrite the asymmetry in a more convenient form for leptogenesis analysis,
$$\begin{gathered}
\epsilon_i^\alpha=-\dfrac{1}{8\pi v^2}\dfrac{M_j}{\sum_k m_k |\mathbf{R}_{ki}|^2}\sum_{k,k'}\sqrt{m_k}m_{k'}\{\sqrt{m_{k'}}\, \text{Im}[\mathbf{U}_{\alpha k}^*\mathbf{U}_{\alpha k'}\mathbf{R}_{ki}\mathbf{R}_{k'i}][f(x_j)+g(x_j)]\nonumber\\
+\sum_{k''}\sqrt{m_{k''}}\,\text{Im}[\mathbf{U}_{\alpha k}^*\mathbf{U}_{\alpha k''}\mathbf{R}_{ki}\mathbf{R}_{k'i}^*\mathbf{R}_{k' j}\mathbf{R}_{k'' j}^*]g'(x_j)\}\, ,
\label{flavouredcpCI}
\end{gathered}$$
where the orthogonal matrix $\mathbf{R}$ is parametrized by a single complex parameter $z$, as shown in Eq. . For an inverted hierarchical neutrino mass spectrum,[^4] the flavored asymmetries generated by $N_1$ and $N_2$ decays are written in terms of $z$ as $$\begin{aligned}
\label{flavouredCPtanz1}
\epsilon_1^{\alpha}=&-\dfrac{M_2}{8\pi v^2}\dfrac{ A_1^\alpha\left[f(x_2) + g(x_2)\right]+B_1^\alpha g'(x_2)}{m_1|c_z|^2+m_2|s_z|^2}\, ,\\
\epsilon_2^{\alpha}=&-\dfrac{M_1}{8\pi v^2}\dfrac{ A_2^\alpha\left[f(x_1) + g(x_1)\right]+B_2^\alpha g'(x_1)}{m_1|s_z|^2+m_2|c_z|^2}\, ,
\label{flavouredCPtanz2}\end{aligned}$$ where $c_z\equiv\cos z$, $s_z\equiv \sin z$ and $$\begin{aligned}
\label{Afactor}
A_1^\alpha=&(m_2^2|\mathbf{U}_{\alpha 2}|^2-m_1^2|\mathbf{U}_{\alpha 1}|^2)\, \text{Im}[s^2_z]+\xi\sqrt{m_1 m_2}\nonumber\\
&\lbrace(m_2-m_1)\text{Im}[\textbf{U}_{\alpha 1}^*\textbf{U}_{\alpha 2}]\text{Re}[c_z s_z]+\nonumber\\
&+(m_2+m_1)\text{Re}[\mathbf{U}^*_{\alpha 1}\mathbf{U}_{\alpha 2}]\text{Im}[c_z s_z]\rbrace\, ,\\
\nonumber\\
B_1^\alpha=&m_1m_2\,(|\mathbf{U}_{\alpha 2}|^2-|\mathbf{U}_{\alpha 1}|^2)\, \text{Im}[c^2_z\, (s^2_z)^*]+\xi\sqrt{m_1 m_2}\nonumber\\
&\lbrace(\,|c_z|^2+|s_z|^2\,)(m_2-m_1)\text{Im}[\textbf{U}_{\alpha 1}^*\textbf{U}_{\alpha 2}]\text{Re}[c_z\, s_z^*]+\nonumber\\
&+(\,|c_z|^2-|s_z|^2\,)(m_2+m_1)\text{Re}[\mathbf{U}^*_{\alpha 1}\mathbf{U}_{\alpha 2}]\text{Im}[c_z\, s_z^*]\rbrace.\nonumber\\
\label{Bfactor}\end{aligned}$$ The factors $A_2^\alpha$ and $B_2^\alpha$ are obtained replacing $s_z\rightarrow c_z$, $c_z\rightarrow s_z$ and $\xi\rightarrow-\xi$ in Eqs. and , respectively. These factors have the following properties $$\begin{gathered}
\sum_{\alpha}A_1^\alpha={\Delta m^2_{21}}\text{Im}[s_z^2], \quad \sum_{\alpha}A_2^\alpha={\Delta m^2_{21}}\text{Im}[c_z^2]\, ,\nonumber\\
\sum_{\alpha}B_i^\alpha=0\, .\end{gathered}$$ Using these relations, the unflavored CP asymmetries are easily obtained, $$\begin{gathered}
\label{unflavouredcpCIz1}
\epsilon_1=-\dfrac{M_2}{8\pi v^2}\dfrac{{\Delta m^2_{21}}\text{Im}[s_z^2]}{m_1\,|c_z|^2 + m_2\, |s_z|^2}\left[f(x_2)+g(x_2)\right],\\
\epsilon_2=-\dfrac{M_1}{8\pi v^2}\dfrac{{\Delta m^2_{21}}\text{Im}[c_z^2]}{m_1\,|s_z|^2 + m_2\, |c_z|^2}\left[f(x_1)+g(x_1)\right].
\label{unflavouredcpCIz2}\end{gathered}$$
The presence of a texture zero in ${\mathbf{Y}^\nu}$ allows for the determination of $z$ in terms of low-energy parameters and $M_{1,2}$, as one may see from Eq. . For instance, in the basis where the charged-lepton and RH neutrino mass matrices are diagonal, the condition ${\mathbf{Y}^\nu}_{11}=0$ implies, for IH, $$\begin{gathered}
\sqrt{m_1}\,{\mathbf{U}}_{11}^*c_z+\xi\sqrt{m_2}\,{\mathbf{U}}_{12}^*s_z=0\; ,\end{gathered}$$ leading to $$\begin{gathered}
\tan z=-\xi\sqrt{\dfrac{m_1}{m_2}}\dfrac{{\mathbf{U}}_{11}^*}{{\mathbf{U}}_{12}^*}\; .\end{gathered}$$ In Table \[tabletan\], we present the expressions for $\tan z$ according to the position of the texture zero in ${\mathbf{Y}^\nu}$ and considering the matrix forms R$_{1,2,3}$ for ${\mathbf{M}_R}$. From this table it is straightforward to see that requiring the presence of two simultaneous zeros in ${\mathbf{Y}^\nu}$ leads to relations among the mixing angles, neutrino masses and the low-energy phases, as expected from Eq. .
[K[1cm]{}|K[5cm]{}|K[5cm]{}]{} ${\mathbf{M}_R}$&$\tan z$ for ${\mathbf{Y}^\nu}_{\alpha 1}=0$&$\tan z$ for ${\mathbf{Y}^\nu}_{\alpha 2}=0$\
R$_1$&$-\xi\sqrt{\dfrac{m_1}{m_2}}\dfrac{\mathbf{U}_{\alpha 1}^*}{\mathbf{U}_{\alpha 2}^*}$&$\xi\sqrt{\dfrac{m_2}{m_1}}\dfrac{\mathbf{U}_{\alpha 2}^*}{\mathbf{U}_{\alpha 1}^*}$
------------------------------------------------------------------------
------------------------------------------------------------------------
\
R$_2$&$i$&$\dfrac{- i\, \sqrt{m_1}M_1\mathbf{U}^*_{\alpha 1}+\,\xi\sqrt{m_2}M_2\mathbf{U}^*_{\alpha 2}}{\sqrt{m_1}M_2\mathbf{U}^*_{\alpha 1}+\,i\,\xi\sqrt{m_2}M_1\mathbf{U}^*_{\alpha 2}}$
------------------------------------------------------------------------
------------------------------------------------------------------------
\
R$_3$&$\dfrac{i\, \sqrt{m_1}M_1\mathbf{U}^*_{\alpha 1}+\,\xi\sqrt{m_2}M_2\mathbf{U}^*_{\alpha 2}}{\sqrt{m_1}M_2\mathbf{U}^*_{\alpha 1}-\,i\,\xi\sqrt{m_2}M_1\mathbf{U}^*_{\alpha 2}}$& $i$
------------------------------------------------------------------------
------------------------------------------------------------------------
\
Replacing in Eqs. and the expressions for $\tan z$ given in Table \[tabletan\], and using the low-energy relations of Table \[CPtable\], we obtain predictions for the flavored CP asymmetries $\epsilon_i^\tau$ and $\epsilon_i^\gamma$, for each of the valid texture-zero cases identified in the Section \[sec2\].
![Flavored CP asymmetries $|\epsilon_{1,2}^\gamma|$ and $|\epsilon_{1,2}^\tau|$ as functions of the low-energy CP-violating phases $\alpha$ and $\delta$, for the texture-zero case ${\mathbf{Y}^\nu}_{11}=0$ and R$_1$. The gray-scale contour regions show the maximum values of $|\epsilon_i^\alpha|$, taking $\theta_{ij}$, ${\Delta m^2_{21}}$ and $|{\Delta m^2_{31}}|$ in the $3\sigma$ experimental range (see Table \[datatable\]) and for $10^9\lesssim M_{1,2}\lesssim 10^{12}$ GeV with $M_2\gtrsim 3M_1$. The colored contour lines are the results obtained for the minimum value of $\chi^2$. In the plot, the triangles and squares correspond to the ($\alpha$,$\delta$) pairs fixed by the conditions ${\mathbf{Y}^\nu}_{11}={\mathbf{Y}^\nu}_{22}=0$ and ${\mathbf{Y}^\nu}_{11}={\mathbf{Y}^\nu}_{32}=0$, respectively (cf. textures B and C in Table \[tabR1R2R3\]).[]{data-label="cpasymflav"}](fig5.pdf "fig:")\
It turns out that, even if one considers a single texture zero in ${\mathbf{Y}^\nu}$, the CP asymmetries are highly suppressed in the flavored regime. As illustration, in Fig. \[cpasymflav\] we show the asymmetries $|\epsilon_i^\gamma|$ and $|\epsilon_i^\tau|$, $i=1,2$, for the case R$_1$ and ${\mathbf{Y}^\nu}_{11}=0$ on the plane ($\alpha$,$\delta$) of the low-energy CP-violating phases. The maximum value for the CP asymmetries (gray scale) is presented for the $3\sigma$ range of the mixing angles and the neutrino mass-squared differences. Notice that we have imposed $M_2\gtrsim 3M_1$ to ensure a nonresonant regime, and $10^9\lesssim M_{1,2}\lesssim 10^{12}$ GeV since $\mu$ and $e$ interactions are in equilibrium. In the same plot, the $|\epsilon_i^\alpha|$ values calculated for the minimum of $\chi^2$ (varying the mixing angles and mass-squared differences) are presented as colored lines. The points marked by triangles and squares correspond to ($\alpha$,$\delta$) fixed by the two-zero conditions ${\mathbf{Y}^\nu}_{11}={\mathbf{Y}^\nu}_{22}=0$ and ${\mathbf{Y}^\nu}_{11}={\mathbf{Y}^\nu}_{32}=0$, respectively, i.e. textures B and C for ${\mathbf{M}^\nu}$ (see Table \[tabR1R2R3\]). We may also see that for the whole $\delta$ and $\alpha$ ranges, the obtained CP asymmetries are highly suppressed being the maximum values below $10^{-6}$. Moreover, $|\epsilon_i^\alpha|\lesssim 10^{-7}$ for ($\alpha$,$\delta$) fixed by textures B and C. Thus, for the case with ${\mathbf{Y}^\nu}_{11}=0$ and $R_1$, the CP asymmetries are too small to ensure efficient leptogenesis. One can show that all other combinations of textures with zeros in ${\mathbf{Y}^\nu}$ and ${\mathbf{M}_R}$ allowed by neutrino data yield similar results.
We conclude that thermal leptogenesis in the flavored regime with $10^9\lesssim T\lesssim 10^{12}$ GeV cannot successfully reproduce the observed baryon asymmetry given in Eq. . This conclusion will be corroborated in the next section when the final baryon asymmetry is computed.
Let us consider now the unflavored regime. In this case, the CP asymmetries are enhanced. For each of the valid two-zero textures, the CP asymmetries $\epsilon_1$ and $\epsilon_2$ given in Eqs. and are computed using the expressions of Tables \[tabletan\] and \[CPtable\]. In Fig. \[cpasymunflav\], we present $|\epsilon_1|$ (blue contour regions) and $|\epsilon_2|$ (gray-scale contour lines) in the $(r_N,M_1)$ plane, for the low-energy neutrino parameters that best fit the 2RHNSM with ${\mathbf{Y}^\nu}$ and ${\mathbf{M}_R}$ textures (T,R). We only show the results for the six combinations (T$_{1,5}$,R$_1$), (T$_{3,4}$,R$_2$), and (T$_{1,6}$,R$_3$), that lead to $\eta_B>0$. From the same plot we see that the maximum values for $|\epsilon_i|$ can now reach $10^{-4}$, which is two orders of magnitude higher than the ones in the flavored regime (cf. Fig \[cpasymflav\]). Furthermore, as the ratio $r_N$ increases, the CP asymmetry $|\epsilon_2|$ gets slightly suppressed with respect to $|\epsilon_1|$.
![Unflavored CP asymmetries $|\epsilon_i|$, $i=1,2$ on the plane ($r_N$, $M_1$), $r_N= M_2/M_1$, for the low-energy neutrino parameters that best fit the texture pairs (T,R). The blue contour regions (gray-scale contour lines) show $|\epsilon_1|$ ($|\epsilon_2|$).[]{data-label="cpasymunflav"}](fig6.pdf "fig:")\
Baryon asymmetry production
---------------------------
In the calculation of the final lepton asymmetry we will consider the contributions of both $N_1$ and $N_2$. In the flavored and unflavored regimes, the leptonic CP asymmetries generated in the $N_i$ decays are most likely to be washed out by the out-of-equilibrium inverse decays and scattering processes in which the heavy neutrinos participate. In general, a measure of the washout strength is given by the so-called decay parameter $K_i$, which for a lepton flavor channel $\alpha$ reads $$\begin{gathered}
K_i^\alpha=\dfrac{\tilde{m}_i^\alpha}{m_*},
\label{flvdecayparam}\end{gathered}$$ where $\tilde{m}_i^\alpha$ is the flavored effective neutrino mass, $$\begin{gathered}
\tilde{m}_i^\alpha=\dfrac{
v^2|{\mathbf{Y}^\nu}_{\alpha i}|^2}{M_i},\end{gathered}$$ and $m_*\simeq 1.09\times 10^{-3}$ eV is the equilibrium neutrino mass. Summing over flavors in Eq. , one obtains the total decay parameter, $$\begin{gathered}
K_i=\sum_\alpha K_i^\alpha=\frac{\tilde{m}_i}{m_*},\end{gathered}$$ with $$\begin{gathered}
\tilde{m}_i=\sum_\alpha \tilde{m}_i^\alpha=\frac{v^2\,\mathbf{H}^\nu_{ii}}{M_i}.\end{gathered}$$ The relation between $\tilde{m}_i$ and $m_*$ gives a measure of thermal equilibrium for the decays, namely, if $\tilde{m}_i\gg m_*$ ($\tilde{m}_i\ll m_*$) the asymmetry is strongly (weakly) washed out by inverse decays.
The fraction of surviving lepton asymmetry can be expressed in terms of efficiency factors $\kappa\in[0,1]$, which are obtained by solving the relevant Boltzmann equations. In our study, we will use instead the simple and accurate analytical approximations for $\kappa_i^\alpha(K_i^\alpha)$ and $\kappa_i(K_i)$ from Refs. [@Antusch:2011nz] and [@Buchmuller:2004nz], respectively. The imposed hierarchy $M_2\gtrsim 3 M_1$ implies $N_{N_1}(T\sim M_2)\simeq N_{N_2}(T\sim M_1)\simeq 0$, so that the computation of the final asymmetry may be split into the $N_1$ and $N_2$ leptogenesis phases. Furthermore, we consider a strong-coupling $N_1$ scenario, where part of the lepton asymmetry generated by $N_2$ decays is projected onto a flavor-direction protected against the washout from $N_1$ interactions [@Antusch:2011nz].
The final ($B-L$)-asymmetry for the flavored temperature regime can be written as [@Antusch:2011nz] $$\begin{gathered}
N_{B-L}=N_{\Delta_{\gamma_1}}+N_{\Delta_{\gamma_1^\perp}}+N_{\Delta_{\tau}},
\label{finalasym}\end{gathered}$$ where the $\Delta_\alpha\equiv B/3-L_\alpha$ number densities in each flavor state read $$\begin{aligned}
N_{\Delta_{\gamma_1\;}}&\simeq -P_{\gamma_2\gamma_1}\,\epsilon_2^\gamma\,\kappa_2^\gamma\, e^{-\frac{3\pi}{8}K_1^\gamma}-\epsilon_1^\gamma\,\kappa_1^\gamma,\\
N_{\Delta_{\tau\;\;\;}}&\simeq -\epsilon_2^\tau\,\kappa_2^\tau\, e^{-\frac{3\pi}{8}K_1^\tau}-\epsilon_1^\tau\,\kappa_1^\tau,\\
N_{\Delta_{\gamma_1^\perp}}&\simeq -\,(1-P_{\gamma_2\gamma_1})\,\epsilon_2^\gamma\,\kappa_2^\gamma,\end{aligned}$$ in which $\gamma_1$ and $\gamma_1^\perp$ are the parallel and orthogonal flavor components to the interaction channels of $N_1$, respectively. Here, $\kappa_i^\alpha$ are the efficiency factors defined in [@Antusch:2011nz], and $P_{\gamma_2\gamma_1}$ is the probability of flavor $\gamma_2$, generated in the $N_2$ decay, to be transformed into $\gamma_1$ under the $N_1$ decay process, $$\begin{aligned}
P_{\gamma_2\gamma_1}=\dfrac{\left|\sum_{\alpha} \mathbf{Y}^{\nu *}_{\alpha 1} {\mathbf{Y}^\nu}_{\alpha 2}\right|^2}{\left(\,\sum_{\alpha} |{\mathbf{Y}^\nu}_{\alpha 1}|^2\right) \left(\, \sum_{\alpha} |{\mathbf{Y}^\nu}_{\alpha 2}|^2 \right)},\end{aligned}$$ where $\alpha=e, \mu$.
In the unflavored regime, the lepton flavors are indistinguishable in the primordial plasma and the final $(B-L)$-asymmetry reads [@Blanchet:2011xq] $$\begin{gathered}
\label{unflavouredbaryonasym}
N_{B-L}\simeq-\epsilon_1\kappa_1-\left(1-P_{21}+P_{21}e^{-\frac{3\pi K_1}{8}}\right)\epsilon_2\kappa_2\; ,\end{gathered}$$ with $\kappa_i$ being defined in [@Buchmuller:2004nz]. Here, $P_{21}$ is the probability of the lepton asymmetry produced in $N_2$ leptogenesis being projected onto the flavor direction of the asymmetry due to $N_1$ interactions, $$\begin{aligned}
P_{21}=\dfrac{\left|\mathbf{H}^\nu_{12}\right|^2}{\mathbf{H}^\nu_{11} \mathbf{H}^\nu_{22}}.\end{aligned}$$ After computing the densities $N_{B-L}$, for both flavored and unflavored regimes, using Eqs. and , the final baryon-to-photon ratio $\eta_B$ is obtained from Eq. .
![Baryon-to-photon ratio $\eta_B$ as a function of the low-energy CP-violating phases $\alpha$ and $\delta$ in the flavored regime, for the texture-zero case ${\mathbf{Y}^\nu}_{11}=0$ and R$_1$. The gray-scale contour regions show the maximum value of $\eta_B$, taking $\theta_{ij}$, ${\Delta m^2_{21}}$ and $|{\Delta m^2_{31}}|$ in the $3\sigma$ experimental range (see Table \[datatable\]) and for $10^9\lesssim M_{1,2}\lesssim 10^{12}$ GeV with $M_2\gtrsim 3M_1$. The colored contour lines are the results obtained for the minimum value of $\chi^2$. In the plot, the triangles and squares correspond to the ($\alpha$,$\delta$) pairs fixed by the conditions ${\mathbf{Y}^\nu}_{11}={\mathbf{Y}^\nu}_{22}=0$ and ${\mathbf{Y}^\nu}_{11}={\mathbf{Y}^\nu}_{32}=0$, respectively (cf. textures B and C in Table \[tabR1R2R3\]).[]{data-label="etabflav"}](fig7.pdf "fig:")\
In Fig. \[etabflav\], we present $\eta_B$ computed for the illustrative case of ${\mathbf{Y}^\nu}_{11}=0$ with R$_1$, for which the flavored CP asymmetries were already analyzed in Section \[subsecCP\]. In that figure, the gray-scale contour regions correspond to the maximum of $\eta_B$ in the $3\sigma$ experimental range of the mixing angles and the neutrino mass-squared differences, taking $10^9\lesssim M_{1,2}\lesssim10^{12}$ GeV. As expected from the small values of $|\epsilon_i^\alpha|$ (see Fig. \[cpasymflav\]), the final baryon asymmetry is suppressed in the whole allowed parameter region. Indeed, the final $\eta_B$ lies between one to two orders of magnitude below the observed value $\eta_B^0$. Moreover, for the ${\mathbf{M}^\nu}$ textures B and C, marked in the figure by a triangle and a square, respectively, $\eta_B\lesssim10^{-12}$ is verified. For all the other combinations of textures T and R that are compatible with neutrino oscillation data, similar results were obtained for the flavored regime, corroborating the fact that thermal leptogenesis in the two-flavor case is not viable.
For the unflavored regime, sufficiently large (and positive) values for $\eta_B$ are obtained for six of the twelve pairs (T,R) of textures compatible with neutrino data (see Table \[tabR1R2R3\]). This is shown in Fig. \[etabunflav\], where we present the predicted $\eta_B$ (gray-scale contour regions) as a function of $M_1$ and the mass ratio $r_N$, considering the low-energy neutrino data that best fit the six textures. In fact, for all these cases, the observed baryon-to-photon ratio $\eta_B^0$ (red contour line in Fig. \[etabunflav\]) is achieved for $M_1\sim 10^{14}$ GeV, being $\kappa_i\sim \mathcal{O}(10^{-3})$ (strong washout regime). Hence, one concludes that the texture combinations (T$_{1,5}$,R$_1$), (T$_{3,4}$,R$_2$), and (T$_{1,6}$,R$_3$) lead to successful thermal leptogenesis in the unflavored regime.
![The baryon-to-photon ratio $\eta_B$ on the plane ($r_N$, $M_1$), $r_N= M_2/M_1$, for the unflavored regime and taking the low-energy neutrino parameters that best fit the texture pairs (T,R). The gray-scale contour regions represent the final value of $\eta_B$, while the red contour line corresponds to the observed value $\eta_B^0$ given in Eq. .[]{data-label="etabunflav"}](fig8.pdf "fig:")\
One may wonder whether the above conclusion remains valid, if one considers the more restricted cases discussed in Section \[sec3a\], in which three elements of ${\mathbf{Y}^\nu}$ are equal. We will only consider the cases that were proved to be compatible with neutrino data and, additionally, verify the condition $r_N\gtrsim3$, for which our leptogenesis assumptions hold. From Table \[Tabrel\] and Fig. \[etabunflav\], one can see that only the cases (T$_1$,R$_1$,B) with ${\mathbf{Y}^\nu}_{21}={\mathbf{Y}^\nu}_{31}={\mathbf{Y}^\nu}_{32}$ and (T$_5$,R$_1$,C) with ${\mathbf{Y}^\nu}_{21}={\mathbf{Y}^\nu}_{22}={\mathbf{Y}^\nu}_{32}$ meet those requirements ($r_N\sim 12$) and, simultaneously, yield $\eta_B>0$. In Fig. \[etabcorr\], we present the $\eta_B$ region allowed by the $3\sigma$ experimental interval for the low-energy neutrino parameters (blue region) as a function of the mass $M_1$. Here we also show the results obtained when the contribution of the second neutrino $N_2$ is not taken into account for leptogenesis (gray region). One concludes that for temperatures below $10^{14}$ GeV the effect of the second neutrino $N_2$ is negligible, while for higher temperatures the $N_2$ contribution tends to lower $\eta_B$. The value of $\eta_B^0$ (red horizontal line) is achieved for masses $M_1\sim 10^{14}$ GeV.
\
Conclusions {#sec4}
===========
In this paper, we have revisited the 2RHNSM considering maximally restricted texture-zero patterns for the lepton Yukawa and mass matrices. Our results are summarized in Table \[tabR1R2R3\]. We conclude that textures B, C and D for the effective neutrino mass matrix ${\mathbf{M}^\nu}$ are compatible with current neutrino data (mixing angles and mass-squared differences) at $1\sigma$, while texture F is compatible at $3\sigma$. In all cases, only an inverted hierarchical neutrino mass spectrum is allowed. A remarkable prediction of textures B and C is that one of the viable solutions for the low-energy CP-violating Dirac phase is $\delta\sim 3\pi/2$, which is very close to the best-fit value obtained from the combined fit of neutrino oscillation data.
Aiming at reducing the number of free parameters in the model, we have also explored scenarios in which additional relations (equality) among the Dirac neutrino Yukawa couplings are imposed. The cases with the maximum number of equal elements in ${\mathbf{Y}^\nu}$ which are compatible with neutrino data are presented in Table \[Tabrel\]. As can be seen from the table, compatibility is only verified at the $3\sigma$ confidence level.
For the phenomenologically viable textures, we have studied their implications for the BAU in the framework of type-I seesaw thermal leptogenesis. We paid special attention to the treatment of leptogenesis in the 2RHNSM. Contrary to what is customary in the literature, where only the decay of the lightest heavy neutrino is considered, we included the decays of both heavy neutrinos in our analysis. Moreover, flavor effects that arise from the fact that lepton interactions exit thermal equilibrium at different temperatures in the early Universe were taken into account. We considered two temperature regimes for leptogenesis: the two-flavored regime ($10^9\lesssim T\lesssim 10^{12}$ GeV) and the unflavored regime ($T\gtrsim10^{12}$ GeV). Within our assumptions ($M_2\gtrsim 3M_1$), we showed that the CP asymmetries in the flavored regime are too small to generate the required lepton asymmetry for successful leptogenesis. On the other hand, for the unflavored case, the CP asymmetries are enhanced, and the observed baryon-to-photon ratio is achieved in the 2RHNSM for the texture combinations (T$_{1,5}$,R$_1$), (T$_{3,4}$,R$_2$), and (T$_{1,6}$,R$_3$), for $M_1\sim 10^{14}$ GeV. Furthermore, the cases (T$_1$,R$_1$) and (T$_5$,R$_1$), with three equal elements in ${\mathbf{Y}^\nu}$ in the positions (21,31,32) and (21,22,32), respectively, were shown to be also compatible with the present value of the baryon asymmetry for the same leptogenesis temperature $T \sim 10^{14}$ GeV.
The nature of the flavor structure of the fermion sector in the standard model and theories beyond it remains puzzling. A common approach to address this problem is to assume certain constraints on the coupling and/or mass matrices in order to reduce the number of free parameters. The lepton textures considered in this work were taken as the simplest and most economical patterns that can be implemented in the framework of the 2RHNSM. We have shown that the maximally constrained 2RHNSM is compatible with current neutrino oscillation data and can also explain the matter-antimatter asymmetry in the Universe via the leptogenesis mechanism. This conclusion holds for several mass matrix textures with the maximal number of allowed zeros and, in a more restricted set, having equal elements in the Dirac Yukawa coupling matrix. It would be interesting to see if such predictive textures could arise from a flavor symmetry principle. This is a subject that certainly deserves to be further explored [@inprep].\
[**Acknowledgements:**]{} This work was supported by Fundaç[ã]{}o para a Ci[ê]{}ncia e a Tecnologia (FCT, Portugal) through the project CFTP-FCT Unit 777 (UID/FIS/00777/2013).
[99]{}
P. Minkowski, Phys. Lett. B [**67**]{} 421 (1977); M. Gell-Mann, P. Ramond and R. Slansky, in [*Supergravity*]{}, eds. P. Van Nieuwenhuizen and D. Freedman (North-Holland, Amsterdam, 1979),p. 315; T. Yanagida, in [*Proceedings of the Workshop on the Unified Theory and the Baryon Number in the Universe*]{}, eds. O. Sawada and A. Sugamoto (KEK, Tsukuba, 1979), p. 95; S.L. Glashow, in [*Quarks and Leptons*]{}, eds. M. Lévy et al., (Plenum, 1980, New-York), p. 707; R.N. Mohapatra and G. Senjanović, Phys. Rev. Lett. [**44**]{}, 912 (1980). M. Fukugita and T. Yanagida, Phys. Lett. B [**174**]{}, 45 (1986). W. Buchmuller, P. Di Bari and M. Plumacher, Annals Phys. [**315**]{}, 305 (2005) \[hep-ph/0401240\]. S. Davidson, E. Nardi and Y. Nir, Phys. Rept. [**466**]{}, 105 (2008) \[arXiv:0802.2962 \[hep-ph\]\]. C. S. Fong, E. Nardi and A. Riotto, Adv. High Energy Phys. [**2012**]{}, 158303 (2012) \[arXiv:1301.3062 \[hep-ph\]\]. E. J. Chun [*et al.*]{}, arXiv:1711.02865 \[hep-ph\]. W. Grimus, A. S. Joshipura, L. Lavoura and M. Tanimoto, Eur. Phys. J. C [**36**]{}, 227 (2004) \[hep-ph/0405016\]. L. M. Cebola, D. Emmanuel-Costa and R. G. Felipe, Phys. Rev. D [**92**]{}, no. 2, 025005 (2015) \[arXiv:1504.06594 \[hep-ph\]\]. G. C. Branco, R. G. Felipe and F. R. Joaquim, Rev. Mod. Phys. [**84**]{}, 515 (2012) \[arXiv:1111.5332 \[hep-ph\]\]. C. Hagedorn, R. N. Mohapatra, E. Molinaro, C. C. Nishi and S. T. Petcov, arXiv:1711.02866 \[hep-ph\]. P. H. Frampton, S. L. Glashow and T. Yanagida, Phys. Lett. B [**548**]{}, 119 (2002) \[hep-ph/0208157\]. A. Ibarra and G. G. Ross, Phys. Lett. B [**591**]{}, 285 (2004) \[hep-ph/0312138\]. K. Harigaya, M. Ibe and T. T. Yanagida, Phys. Rev. D [**86**]{}, 013002 (2012) \[arXiv:1205.2198 \[hep-ph\]\]. T. Rink and K. Schmitz, JHEP [**1703**]{}, 158 (2017) \[arXiv:1611.05857 \[hep-ph\]\]. Y. Shimizu, K. Takagi and M. Tanimoto, arXiv:1709.02136 \[hep-ph\].
R. González Felipe, F. R. Joaquim and B. M. Nobre, Phys. Rev. D [**70**]{}, 085009 (2004) \[hep-ph/0311029\]. F. R. Joaquim, Nucl. Phys. Proc. Suppl. [**145**]{}, 276 (2005) \[hep-ph/0501221\]. A. Abada, S. Davidson, A. Ibarra, F.-X. Josse-Michaux, M. Losada and A. Riotto, JHEP [**0609**]{}, 010 (2006) \[hep-ph/0605281\]. J. Zhang and S. Zhou, JHEP [**1509**]{}, 065 (2015) \[arXiv:1505.04858 \[hep-ph\]\]. K. Siyeon, J. Korean Phys. Soc. [**69**]{}, no. 11, 1638 (2016) \[arXiv:1611.04572 \[hep-ph\]\]. T. Geib and S. F. King, arXiv:1709.07425 \[hep-ph\]. A. Achelashvili and Z. Tavartkiladze, arXiv:1710.10955 \[hep-ph\]. Y. Shimizu, K. Takagi and M. Tanimoto, arXiv:1711.03863 \[hep-ph\]. C. Patrignani [*et al.*]{} \[Particle Data Group\], Chin. Phys. C [**40**]{}, no. 10, 100001 (2016). P. F. de Salas, D. V. Forero, C. A. Ternes, M. Tortola and J. W. F. Valle, arXiv:1708.01186 \[hep-ph\]. I. Esteban, M. C. Gonzalez-Garcia, M. Maltoni, I. Martinez-Soler and T. Schwetz, JHEP [**1701**]{}, 087 (2017) \[arXiv:1611.01514 \[hep-ph\]\]. F. Capozzi, E. Lisi, A. Marrone, D. Montanino and A. Palazzo, Nucl. Phys. B [**908**]{}, 218 (2016) \[arXiv:1601.07777 \[hep-ph\]\].
J. A. Casas and A. Ibarra, Nucl. Phys. B [**618**]{}, 171 (2001) \[hep-ph/0103065\]. Z. z. Xing, Phys. Rev. D [**69**]{}, 013006 (2004) \[hep-ph/0307007\]. R. R. Gautam, M. Singh and M. Gupta, Phys. Rev. D [**92**]{}, no. 1, 013006 (2015) \[arXiv:1506.04868 \[hep-ph\]\]. F. Capozzi, E. Di Valentino, E. Lisi, A. Marrone, A. Melchiorri and A. Palazzo, Phys. Rev. D [**95**]{}, no. 9, 096014 (2017) \[arXiv:1703.04471 \[hep-ph\]\]. J. B. Albert [*et al.*]{} \[EXO-200 Collaboration\], arXiv:1710.07670 \[hep-ex\]. A. Gando [*et al.*]{} \[KamLAND-Zen Collaboration\], Phys. Rev. Lett. [**117**]{}, no. 8, 082503 (2016), Addendum: \[Phys. Rev. Lett. [**117**]{}, no. 10, 109903 (2016)\] \[arXiv:1605.02889 \[hep-ex\]\]. M. Agostini [*et al.*]{} \[GERDA Collaboration\], arXiv:1710.07776 \[nucl-ex\]. C. Alduino [*et al.*]{} \[CUORE Collaboration\], Eur. Phys. J. C [**77**]{}, no. 1, 13 (2017) \[arXiv:1609.01666 \[nucl-ex\]\]. I. Ostrovskiy and K. O’Sullivan, Mod. Phys. Lett. A [**31**]{}, no. 18, 1630017 (2016) Erratum: \[Mod. Phys. Lett. A [**31**]{}, no. 23, 1692004 (2016)\] \[arXiv:1605.00631 \[hep-ex\]\]. Z. z. Xing, Phys. Lett. B [**539**]{}, 85 (2002) \[hep-ph/0205032\]. S. Dev, S. Kumar, S. Verma and S. Gupta, Phys. Rev. D [**76**]{}, 013002 (2007) \[hep-ph/0612102\]. H. Fritzsch, Z. z. Xing and S. Zhou, JHEP [**1109**]{}, 083 (2011) \[arXiv:1108.4534 \[hep-ph\]\]. P. O. Ludl, S. Morisi and E. Peinado, Nucl. Phys. B [**857**]{}, 411 (2012) \[arXiv:1109.3393 \[hep-ph\]\]. D. Meloni and G. Blankenburg, Nucl. Phys. B [**867**]{}, 749 (2013) \[arXiv:1204.2706 \[hep-ph\]\]. W. Grimus and P. O. Ludl, J. Phys. G [**40**]{}, 055003 (2013) \[arXiv:1208.4515 \[hep-ph\]\]. S. Dev, R. R. Gautam, L. Singh and M. Gupta, Phys. Rev. D [**90**]{}, no. 1, 013021 (2014) \[arXiv:1405.0566 \[hep-ph\]\]. T. Kitabayashi and M. Yasuè, Phys. Rev. D [**93**]{}, no. 5, 053012 (2016) \[arXiv:1512.00913 \[hep-ph\]\].
P. A. R. Ade [*et al.*]{} \[Planck Collaboration\], Astron. Astrophys. [**594**]{}, A13 (2016) \[arXiv:1502.01589 \[astro-ph.CO\]\]. S. Antusch, P. Di Bari, D. A. Jones and S. F. King, Phys. Rev. D [**86**]{}, 023516 (2012) \[arXiv:1107.6002 \[hep-ph\]\]. L. Covi, E. Roulet and F. Vissani, Phys. Lett. B [**384**]{}, 169 (1996) \[hep-ph/9605319\]. R. Barbieri, P. Creminelli, A. Strumia and N. Tetradis, Nucl. Phys. B [**575**]{}, 61 (2000) \[hep-ph/9911315\]. A. Abada, S. Davidson, F. X. Josse-Michaux, M. Losada and A. Riotto, JCAP [**0604**]{}, 004 (2006) \[hep-ph/0601083\]. E. Nardi, Y. Nir, E. Roulet and J. Racker, JHEP [**0601**]{}, 164 (2006) \[hep-ph/0601084\].
S. Blanchet, P. Di Bari, D. A. Jones and L. Marzola, JCAP [**1301**]{}, 041 (2013) \[arXiv:1112.4528 \[hep-ph\]\]. D. M. Barreiros, R. G. Felipe and F. R. Joaquim, in preparation.
[^1]: From now on we will denote these two cases by normal (NH) and inverted hierarchy (IH), respectively.
[^2]: We will perform our analysis considering a diagonal charged-lepton Yukawa matrix ${\mathbf{Y}^\ell}_{\rm diag}$. In the end of this section, we will comment on how the results change when the remaining five ${\mathbf{Y}^\ell}$ textures with six zeros are considered.
[^3]: Analyses of ${\mathbf{M}^\nu}$ with two texture zeros have been presented in Refs. [@Cebola:2015dwa; @Xing:2002ap; @Dev:2006qe; @Ludl:2011vv; @Fritzsch:2011qv; @Meloni:2012sx; @Grimus:2012zm; @Dev:2014dla; @Kitabayashi:2015jdj] for the general case $m_{1,2,3}\neq 0$.
[^4]: Hereafter, we consider only the IH case since, as shown in Section \[sec2\], this is the only type of spectrum compatible with low-energy neutrino data.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
It is well-known (cf. K.-Pudlák[@KP-jsl]) that a polynomial time algorithm finding tautologies hard for a propositional proof system $P$ exists iff $P$ is not optimal. Such an algorithm takes $1^{(k)}$ and outputs a tautology $\tau_k$ of size at least $k$ such that $P$ is not p-bounded on the set of all $\tau_k$’s.
We consider two more general search problems involving finding a hard formula, [**Cert**]{} and [**Find**]{}, motivated by two hypothetical situations: that one can prove that ${{\mbox{NP}}}\neq co{{\mbox{NP}}}$ and that no optimal proof system exists. In [**Cert**]{} one is asked to find a witness that a given non-deterministic circuit with $k$ inputs does not define $TAUT \cap {{\{0,1\}^k}}$. In [**Find**]{}, given $1^{(k)}$ and a tautology $\alpha$ of size at most $k^{c_0}$, one should output a size $k$ tautology $\beta$ that has no size $k^{c_1}$ $P$-proof from substitution instances of $\alpha$.
We shall prove, assuming the existence of an exponentially hard one-way permutation, that [**Cert**]{} cannot be solved by a time $2^{O(k)}$ algorithm. Using a stronger hypothesis about the proof complexity of Nisan-Wigderson generator we show that both problems [**Cert**]{} and [**Find**]{} are actually only partially defined for infinitely many $k$ (i.e. there are inputs corresponding to $k$ for which the problem has no solution). The results are based on interpreting the Nisan-Wigderson generator as a proof system.
author:
- Jan Krajíček
date: |
Faculty of Mathematics and Physics\
Charles University in Prague
---
A [**propositional proof system**]{} in the sense of Cook and Reckhow [@CR] is a polynomial time relation $P(x,y)$ such that for a binary string $\tau$: $$\tau \in \mbox{TAUT}\ \ \mbox{iff }\ \ \exists \pi \in {{\{0,1\}^*}}P(\tau, \pi)$$ where TAUT is the set of propositional tautologies (in DeMorgan language for the definiteness). Any string $\pi$ for which $P(\tau, \pi)$ holds is called a [**$P$-proof**]{} of $\tau$. A proof system (tacitly propositional from now on) is [**p-bounded**]{} iff there exists a constant $c \geq 1$ such that the above holds even with the requirement that $|\pi| \le |\tau|^c$. Cook and Reckhow [@CR] noted that a p-bounded proof system exists iff ${{\mbox{NP}}}= \mbox{co}{{\mbox{NP}}}$. Hence proving that no p-bounded proof system exists would imply ${{\mbox{NP}}}\neq \mbox{co}{{\mbox{NP}}}$ and thus also ${{\mbox{P}}}\neq {{\mbox{NP}}}$. This fact elevated the investigation of lengths of proofs into a fundamental topic of mathematical logic approach to computational complexity.
Strong lower bounds were proved for a variety of proof systems and several different methods for this purpose were invented. Examples of proof systems that appear to be outside of the scope of current methods are the so called [**Frege systems**]{}: the usual text-book propositional calculi based on a finite number of axioms schemes and inference rules (only quadratic lower bounds are known for them, cf.[@Kra-speed]). This apparent failure could cause an uninformed reader to dismiss the whole area of proof complexity. However, although we may not be near proving that ${{\mbox{NP}}}\neq \mbox{co}{{\mbox{NP}}}$, the lower bounds for weaker proof systems proved so far do have consequences interesting in their own right. For example, a single lower bound for a proof system $P$ implies time lower bounds for a class $Alg(P)$ of SAT algorithms associated with $P$ and all commonly used SAT algorithms belong to such a class for some $P$ for which we have an exponential lower bound (cf.[@Kra-sat] and references given there). Another type of consequences can be found in bounded arithmetic: a lengths-of-proofs lower bound for $P$ often implies the unprovability of a true $\Pi^0_1$ sentence in a first-order theory $T_P$ associated with $P$. These unprovability arguments do not use Gödel’s theorem and the $\Pi^0_1$ sentences involved have typically a clear combinatorial meaning. And last but not least, any super-polynomial lower bound for $P$ also implies that ${{\mbox{P}}}\neq {{\mbox{NP}}}$ is consistent with $T_P$. We shall not survey these proof complexity topics in detail here and instead refer the reader to expositions in [@kniha; @Kra-methods; @Kra-ecm; @Pud-survey] or in [@k2 Chpt.27].
In this paper we are interested in the question how hard it is - to be measured in terms of computational complexity here - to find tautologies hard (i.e. requiring long proofs) for a given proof system. Proposing plausible candidates for tautologies hard for Frege systems and for stronger proof systems turned out to be a quite delicate issue. The lack of a variety of good candidates is one of principal obstacles for proving lower bounds for strong systems. Of course, in principle one would be happy to accept a suggestion for such a hard tautology from a friendly oracle. However, the experience with known lower bound proofs shows that it is essential to have explicit formulas with a transparent combinatorial meaning.
In particular, all first super-polynomial lower bounds for proof systems for which we have any such bounds were proved for some sequence of tautologies $\{\tau_k\}_k$ of size $|\tau_k| \geq k$ and constructible in polynomial time (or even log space) from $1^{(k)}$. This type of sequences of hard tautologies has been considered in [@KP-jsl] and [@kniha Chpt.14] and it exists for a proof system $P$ iff $P$ is not [**optimal**]{}, i.e. there exists a proof system $Q$ that has a super-polynomial speed-up over $P$ (w.r.t. lengths of proofs) on an infinite set of formulas. It is consistent with the present knowledge, and indeed most researchers seem to conjecture that, that no optimal proof system exist and hence that for each $P$ a p-time constructible sequence of hard formulas exist. However, deriving the existence of hard formulas from the assumption of non-optimality is not very illuminating: it is a basic proof complexity result that $P$ cannot admit polynomial size proofs of formulas expressing the soundness of a proof system $Q$ (these formulas are log space constructible) if $Q$ has a super-polynomial speed-up over $P$ on an infinite set of formulas (cf.[@kniha Chpt.14]). Hence, in a sense, deriving the existence of a polynomial time sequence of hard tautologies from the non-optimality assumption amounts just to restating the assumption in a different terminology. We refer the reader to Beyersdorff-Sadowski[@BeySad] for further information and up-to-date references.
We shall consider in this paper two more general search problems in which the task includes a requirement to find a hard tautology. The two problems model in their ways two hypothetical situations: a situation when one can prove ${{\mbox{NP}}}\neq co{{\mbox{NP}}}$ (i.e. super-polynomial lower bounds for all proof systems) and a situation when one can prove that no optimal proof system exists by having a uniform method how to construct from a given proof system a stronger one. These two tasks, [**Cert**]{} and [**Find**]{}, will be defined in Section \[tasks\] (in Section \[pairs\] we add one more search task [**Pair**]{} involving disjoint pairs of ${{\mbox{NP}}}$ sets).
We will prove (using the hypothesis of the existence of a hard one-way permutation) that [**Cert**]{} cannot be solved by exponential time algorithms and (using a stronger hypothesis about the proof complexity of the Nisan-Wigderson generator) that both [**Cert**]{} and [**Find**]{} actually cannot be solved at all on infinitely many input lengths. Our primary motivation for this research is to understand what kind of consequences do various - both proven and conjectural - statements about the proof complexity of the Nisan-Wigderson generator have.
The paper is organized as follows. After the motivation and the definition of the search tasks [**Cert**]{} and [**Find**]{} in Section \[tasks\] we review some complexity theory in Section \[1\] and some proof complexity in Section \[2\]. The hardness results are proved in Sections \[4\] and \[6\], respectively (after a proof complexity interlude in Section \[5\]). The paper is concluded by Section \[pairs\] considering a related search task for disjoint pairs of sets and a few remarks in Section \[rema\].
We do assume only basic complexity theory and proof complexity (e.g. the well-known relation between reflection principles and simulations). But the reader may still benefit from understanding a wider proof complexity context. In particular, [@k2 Chpt.27] overviews some fundamental problems of proof complexity and [@k2 Chpts.29 and 30] survey[^1] the theory proof complexity generators (and list relevant literature).
The search tasks [**Cert**]{} and [**Find**]{} {#tasks}
==============================================
We are going to consider two search tasks asking us to find formulas with certain properties (and in Section \[pairs\] we add one more). Both are more complex than the mere task to construct hard tautologies for a given proof system that was discussed in the introduction. To motivate them we shall describe two thought situations in proof complexity; the search tasks are then abstract (and simplified) versions of those.
First assume that you can prove (i.e. ZFC can) that ${{\mbox{NP}}}\ne co{{\mbox{NP}}}$ and thus, in particular, super-polynomial lower bounds for all proof systems. For a proof system $P$ and a constant $c \geq 1$ denote by ${{\mbox{LB}_P(c)}}$ the statement $$\forall 1^{(k)} \exists \tau \
[|\tau|\geq k \wedge \tau \in TAUT \wedge
\forall \pi (|\pi|\le |\tau|^c) \neg P(\tau, \pi)]$$ formalizing a polynomial lower bound for $P$ with degree $c$.
It is easy to see that for any decent proof system (see Section \[5\] for a formal definition of decency), as long as we can prove some specific polynomial lower bound we can also prove its soundness. The decency assumption allows to extend a proof of a falsifiable formula $\varphi$ to a proof of $0$ and further to a proof of any $\tau$, all in polynomial time.
But by a simple application of Gödel’s theorem ZFC is not able to prove the soundness of all proof systems. This suggests that we should be proving lower bounds conditioned upon the assumption that $P$ is indeed a Cook-Reckhow proof system. If $P$ were not complete we do not need to bother with lower bounds for it, so the interesting clause of the Cook-Reckhow definition is the soundness and we are lead to implications: $$Ref_P \rightarrow {{\mbox{LB}_P(c)}}$$ where $Ref_P$ is a universal sentence (in the language $L_{PV}$ of Section \[2\]) formalizing that any formula with a $P$-proof must be a tautology.
Now we simplify the situation a bit more. Let $D(x,y)$ be a circuit in $k$ variables $x = (x_1,\dots, x_k)$ and $\ell = k^c$ variables $y = (y_1, \dots, y_{\ell})$ which we interpret as the provability relation of a proof system restricted to formulas of size $k$ and proofs of size at most $\ell$. This motivates the following [**search task**]{} $\mbox{{\bf Cert}}(c)$ defined for any constant $c \geq 1$:
- input: $1^{(k)}$ and a size $k^{c^2}$ circuit $D(x,y)$ in $k$ variables $x = (x_1,\dots, x_k)$ and $\ell = k^c$ variables $y = (y_1, \dots, y_{\ell})$
- required output: either a size $k$ falsifiable formula $\varphi$ such that $D(\varphi, y)$ is satisfiable or a size $k$ tautology $\tau$ such that $D(\tau,y)$ is unsatisfiable.
The output of $Cert(c)$ thus certifies that $D$ is not a non-deterministic circuit (with input $x$ and non-deterministic variables $y$) that accepts $TAUT \cap {{\{0,1\}^k}}$. In other words, the output either witnesses that $D$ does not correspond to a sound proof system or that the proof system admits a lower bound.
The provability relation of a proof system restricted to size $k$ formulas and size $\ell$ proofs can be computed by circuits of size $(k+\ell)^{O(1)}$. In the formulation of the problem we have represented the $O(1)$ constant by $c$ as well. In addition [**Cert**]{} ignores the uniformity of such circuits corresponding to a particular proof system (they can be constructed in log space from $1^{(k)}, 1^{(\ell)}$). This is in line with the prevailing approach in complexity theory to reduce uniform problems to non-uniform finite combinatorial problems. Finally note that in our simplification we are taking the reflection principle just for the proof lengths corresponding to the lower bound we should witness; this make sense due to the non-uniformity.
The second search task we shall consider is motivated by another thought experiment. Assume that you can prove that no optimal proof system exists and, in fact, that you have a uniform construction that from a proof system $P$ produces a stronger proof system $Q(P)$ (i.e. not simulable by $P$). For definiteness, assume that there is one oracle polynomial time machine that for all $P$ defines $Q(P)$ when having the oracle for $P$. Then we expect to be able to prove $$Ref_P\ \rightarrow\ Ref_{Q(P)}$$ and, most importantly, that it is stronger $$Ref_P\ \rightarrow\
\forall 1^{(k)} \forall \pi (|\pi|\le k^c) \neg P({{{\sf ref}_{Q(P)}^k}}, \pi)]$$ where ${{{\sf ref}_{Q(P)}^k}}$ is a size $k^{O(1)}$ tautology formalizing the soundness of $Q(P)$ w.r.t. all proof of size at most $k$ (we assume for simplicity that a proof is always at least as long as the formula it proves so one parameter suffices). See a similar formula in (\[m0\]) in the proof of Lemma \[5.3\].
Any decent proof system can simulate $Q(P)$ if it can use ${{{\sf ref}_{Q(P)}^k}}$ as extra axioms (see Section \[5\]). In the following problem $\alpha$ represents a bit more[^2] generally any extra axiom.
Let $P$ be a proof system and $c_1 \geq c_0 \geq 1$ be constants. Consider the following promise [**computational task**]{} $\mbox{{\bf Find}}(P,c_1,c_0)$:
- input: $1^{(k)}$ and a tautology $\alpha$ such that $|\alpha| \le k^{c_0}$
- required output: any size $k$ tautology $\beta$ that has no proof in proof system $P + \alpha$, $P$ augmented by $\alpha$ as an extra axiom scheme[^3], of size less than $k^{c_1}$.
The requirement that the size of $\beta$ is exactly $k$ is just for a technical convenience; we could allow any interval $[k^{\Omega(1)}, k^{O(1)}]$ instead.
Computational complexity preliminaries {#1}
======================================
Let $n \rightarrow m = m(n)$ be an injective function such that $m(n) > n$ and let $f : {{\{0,1\}^*}}\rightarrow {{\{0,1\}^*}}$ be a Boolean function. The [**Nisan-Wigderson generator**]{} $NW_{A,f} : {{\{0,1\}^n}}\rightarrow {{\{0,1\}^m}}$ is defined using the notion of a design. A $(d,\ell)$-design on $[n]$ is a set system $A = \{J_i\subseteq [n]\}_{i\in [m]}$ on $[n] = \{1, \dots, n\}$ such that:
- $|J_i| = \ell$, for all $i$,
- $|J_i \cap J_j| \le d$, for all $i \neq j$.
The $i$-th bit of $NW_{A,f}(x)$ is computed by $f_{\ell} : {{\{0,1\}^{\ell}}}\rightarrow \{0,1\}$ from the $\ell$-bit string $x(J_i) := x_{j_1}\dots x_{j_{\ell}}$, where $$J_i\ =\ \{j_1< \dots < j_{\ell}\}$$ and $f_{\ell}$ is the restriction of $f$ to ${{\{0,1\}^{\ell}}}$. In the future the parameter $\ell$ will be determined by $n$ and we shall denote the restriction $f$ as well. Nisan and Wigderson[@NW] showed that there are such designs for a wide range of parameters $n,m,d,\ell$ and that one can construct them uniformly and feasibly. In particular, we can fix the parameters as follows: $$\label{e1}
\ell := n^{1/3}\ \ \mbox{ and }\ \ m := 2^{n^{\delta}}\ \
\mbox{ and }\ \ \ d := \log(m)\ ,$$ where $1/3 \geq \delta > 0$ is arbitrary. We shall thus assume that fixing $n$ and $\delta$ fixes the other parameters and also some set system $A_n$ constructed from $1^{(n)}, 1^{(m)}$ in time $m^{O(1)}$ and with parameters meeting the requirements. In fact, we need that $$\label{e2}
\mbox{{\em $J_i$ is computable from $i$ and $1^{(n)}$ in polynomial time.}}$$ The design from [@NW L.2.5] has this property.
In our construction the function $f$ will be ${{{{\mbox{NP}}}\cap \mbox{co}{{\mbox{NP}}}}}$. By this we mean that it is the characteristic function of a language in ${{{{\mbox{NP}}}\cap \mbox{co}{{\mbox{NP}}}}}$. Hence $f$ is defined by two ${{\mbox{NP}}}$ predicates $$\label{e3}
\exists y (|y| \le |u|^{c} \wedge F_0(u,y))\ \ \mbox{ and }
\ \
\exists y (|y| \le |u|^{c} \wedge F_1(u,y))$$ with $F_0$ and $F_1$ polynomial-time relations and $c$ a constant such that $$\label{i0}
f(u) = a\ \mbox{ iff }\ \exists y (|y| \le |u|^{c} \wedge F_a(u,y))$$ for $a = 0, 1$. Any string $y$ witnessing the existential quantifier will be called a witness for $f(u)$.
We shall use results from [@Kra-nwg] and those do assume that $f$ has unique witnesses, meaning that for each $u$ there is exactly one witness for $f(u)$. A natural source of ${{{{\mbox{NP}}}\cap \mbox{co}{{\mbox{NP}}}}}$ functions with unique witnesses are hard bits of one-way permutations. That is, for a polynomial time (and intended to be one-way) permutation $h : {{\{0,1\}^*}}\rightarrow {{\{0,1\}^*}}$ we have $$f(u)\ :=\ B(h^{(-1)}(u))$$ where $B(v)$ is a hard bit predicate for $h$.
The hardness of one-way permutations is measured as follows. A polynomial time permutation $h$ is defined to be [**$\epsilon(\ell)$ one way with security parameter $t(\ell)$**]{} iff for all $\ell$ and any circuit $D$ with $\ell$ inputs and of size at most $t(\ell)$ it holds: $${{\mbox Prob}}_{v \in \{0,1\}^{\ell}}[ D(h(v)) = v] \ \le \ \epsilon(\ell)\ .$$ Using the Goldreich-Levin theorem we may assume that such a permutation $h$ has a hard bit function $B(v)$. The details can be found in Goldreich [@Gol].
Our construction needs to assume that $f$ is hard in the sense of Nisan and Wigderson [@NW]. They define $f$ to be $(\epsilon(\ell), S(\ell))$-hard if for every $\ell$ and every circuit $C$ with $\ell$ inputs and of size at most $S(\ell)$ it holds: $${{\mbox Prob}}_{u \in \{0,1\}^{\ell}}[C(u) = f(u) ]\ < \
1/2 + \epsilon/2\ .$$ They define then the [**hardness of $f$**]{}, denoted [**$H_f(\ell)$**]{}, to be the maximal $S$ such that the function is $(1/S, S)$-hard. This simplification makes sense when $\epsilon$ has the rate about $m^{-O(1)}$ as in Nisan and Wigderson [@NW].
In the proof complexity situations studied in [@Kra-nwg] the parameter $S$ plays the main role, with $\epsilon$ being primarily of the rate $\ell^{- O(1)}$. This corresponds to the fact that in applications of the original Nisan-Wigderson generators $m$ is usually exponentially large but for various purposes of proof complexity (especially lengths-of-proofs lower bounds) the best choice would be at the opposite end: $m=n+1$. This lead in [@Kra-nwg] to keeping $\epsilon$ and $S$ separate and using the notion of the approximating hardness (defined there) in place of $H_f(\ell)$. In this paper, however, we shall use only those results from [@Kra-nwg] where $m$ is exponentially large as in (\[e1\]) and thus using the measure $H_f(\ell)$ suffices here.
A one-way permutation $h$ with a hard bit $B$ is [**exponentially hard**]{} iff it is $2^{-{\ell}^{\Omega(1)}}$ one-way with security parameter $2^{{\ell}^{\Omega(1)}}$. The hardness $H_f(\ell)$ of $f$ is then $2^{{\ell}^{\Omega(1)}}$ as well. Details of these constructions can be found in Goldreich [@Gol].
We will use in Sections \[4\] and \[pairs\] the hypothesis that an exponentially hard one-way permutation exists instead of the presumably weaker assumption that an ${{{{\mbox{NP}}}\cap \mbox{co}{{\mbox{NP}}}}}$ function $f$ with unique witnesses and with exponential hardness $H_f$ exists. The only reason is that the former hypothesis is more familiar than the latter one.
Proof complexity preliminaries {#2}
==============================
Although the formulation of the search tasks [**Cert**]{} and [**Find**]{} may not suggests so explicitly this investigation resulted from a research program in proof complexity about the so called proof complexity generators and we shall use some ideas from this theory.
We shall start with a proof complexity conjecture of Razborov[@Raz03 Conjecture 2]. Take an arbitrary string $b \in {{\{0,1\}^m}}$ that is outside of the range $Rng({{{NW}_{A_n,f}}})$ of ${{{NW}_{A_n,f}}}$. The statement $b \notin Rng({{{NW}_{A_n,f}}})$ is a $\mbox{co}{{\mbox{NP}}}$ property of $b$ and can be expressed by a propositional formula $\tau({{{NW}_{A_n,f}}})_b$ in the sense that $$\tau({{{NW}_{A_n,f}}})_b \in \mbox{TAUT }\ \mbox{ iff }\
b \notin Rng({{{NW}_{A_n,f}}})\ .$$ The construction of the propositional translation of the $\mbox{co}{{\mbox{NP}}}$ statement is analogous to the usual proof of the NP-completeness of SAT. The details can be found in any of [@Coo75; @kniha; @Pud-survey; @k2]). Note that the size of the formulas is polynomial in $m$. [**Razborov’s conjecture**]{} says that these tautologies are hard for Extended Frege system EF for ${{{NW}_{A_n,f}}}$ defined as above, with $m = 2^{n^{\Omega(1)}}$ and based on an ${{{{\mbox{NP}}}\cap \mbox{co}{{\mbox{NP}}}}}$ function $f$ that is hard on average for ${{\mbox{P}}}/\mbox{poly}$. Pich [@Pich] proved the conjecture for all proof systems admitting feasible interpolation in place of EF.
In [@Kra-nwg] we have considered a generalization of the conjecture. We shall recall only one part of that generalization dealing with exponentially large $m$; in the other parts $m = n+1$ and they use the notion of approximating hardness of a function mentioned in the previous section.
\[2.1\]
Assume $f$ is an ${{\mbox{NP}}}\cap co{{\mbox{NP}}}$ function with unique witnesses that has an exponential Nisan-Wigderson hardness $H_f(\ell) = 2^{\ell^{\Omega(1)}}$.
Then there is $\delta > 0$ such that for $m(n) = 2^{n^{\delta}}$ and for any infinite ${{\mbox{NP}}}$ set $R$ that has infinitely many elements whose length equals to $m(n)$ for $n\geq 1$ it holds: $$Rng({{{NW}_{A_n,f}}}) \cap R \ \neq \ \emptyset \ .$$
Let us observe that Conjecture \[2.1\] has a proof complexity corollary including Razborov’s conjecture.
\[2.2\]
Let $P$ be any proof system. Assume that Conjecture \[2.1\] holds and that the Nisan-Wigderson hardness $H_f(\ell)$ of an ${{{{\mbox{NP}}}\cap \mbox{co}{{\mbox{NP}}}}}$ function $f$ with unique witnesses is $2^{\ell^{\Omega(1)}}$.
Then there exists $\delta > 0$ such that for all $c \geq 1$, the size of $P$-proofs of formulas $\tau({{{NW}_{A_n,f}}})_b$ for all large enough $n$ and all $b \notin Rng({{{NW}_{A_n,f}}})$ of size $|b| = m(n)$ is bigger than $|\tau({{{NW}_{A_n,f}}})_b|^c$.
Note that the set $R$ of all $b$ of lengths $m(n)$ for $n \geq 1$ for which $\tau({{{NW}_{A_n,f}}})_b$ has a $P$-proof of size at most $|\tau({{{NW}_{A_n,f}}})_b|^c$ is in ${{\mbox{NP}}}$.
Now we recall (a part of) the consistency result from [@Kra-nwg] concerning Conjecture \[2.1\]. Its technical heart is a lower bound on complexity of functions solving a certain search task associated with ${{{NW}_{A_n,f}}}$ and that would, in principle, suffice for our purposes here. Using the consistency result itself, however, seems to decrease the number of technicalities one otherwise needs to discuss.
We first recapitulate a few basic definitions. Cook [@Coo75] has defined a theory PV whose language $L_{PV}$ has a name for every polynomial-time algorithm obtained from a few basic algorithms by the composition and by the limited recursion on notation, following Cobham’s [@Cob64] characterization of polynomial time. The details of the definition of $L_{PV}$ can be found in [@Coo75; @kniha] but are not important here. In fact, neither is the theory PV itself as we shall work with the true universal first-order theory of ${\mathbf N}$ in the language $L_{PV}$. We shall denote this theory ${\mbox{T}_{PV}}$, as in [@Kra-nwg]. Note that ${\mbox{T}_{PV}}$ contains formulas expressing the soundness of all proof systems.
Let $f$ be an ${{\mbox{NP}}}\cap \mbox{co}{{\mbox{NP}}}$ function defined as in (\[i0\]). Let us abbreviate by $G(w,z,x,y)$ the open $L_{PV}$ formula $$(z_x=0 \wedge F_0(w(J_x), y)) \vee (z_x=1 \wedge F_1(w(J_x), y))$$ where $J_x$ is from the set system $A_n$ (polynomial time definable from $1^{(n)}$ and $x$) and $F_0$, $F_1$ are from (\[e3\]).
We do not have a symbol in $L_{PV}$ for the function on ${{\{0,1\}^*}}$ computed for $n \geq 1$ on ${{\{0,1\}^n}}$ by ${{{NW}_{A_n,f}}}$ as it is not a polynomial time function, and the function has to be defined. One possible formalization of the statement ${{{NW}_{A_n,f}}}(w)= z$ for $|w|=n$ and $|z|=m$ is then $$\forall x \in [m] \exists y (|y| \le \ell^c)\ G(w,z,x, y)$$ with $c$ from (\[e3\]). Now we are ready to state the result from [@Kra-nwg] we shall need.
\[3.1\]
Assume $f$ is an ${{\mbox{NP}}}\cap co{{\mbox{NP}}}$ function with unique witnesses having the Nisan-Wigderson hardness $H_f(\ell)$ at least $2^{\ell^{\Omega(1)}}$.
Then there is $\delta > 0$ such that for any ${{\mbox{NP}}}$ set $R$ that has infinitely many elements whose length equals to $m(n)$ for $n\geq 1$ and defined by $L_{PV}$ formula $\exists v (|v| \le |z|^d) R_0(z,v)$, with $R_0$ open, theory ${\mbox{T}_{PV}}$ does not prove the universal closure of the formula $$A\ \rightarrow\ B$$ where $A$ is the formula with variables $v, w, z, n, m, \ell$ $$n = |w| \wedge m = |z| \wedge m = 2^{n^{\delta}} \wedge \ell = n^{1/3}
\wedge |v| \le m^d \wedge R_0(z,v)$$ and $B$ is the formula $$\exists x \in [m] \forall y (|y| \le \ell^c)
\ \neg G(w, z, x, y)\ .$$
This statement is in [@Kra-nwg] derived from a bit finer model-theoretic result.
The hardness of task [**Cert**]{} {#4}
=================================
The argument we shall use to derive the hardness of [**Cert**]{} applies to a more general situation which we describe now.
By an ${{{({{{{\mbox{NP}}}\cap \mbox{co}{{\mbox{NP}}}}})/\mbox{poly}}}}$ [**algorithm**]{} we shall mean two polynomial time predicates $F_0(x,y,z)$ and $F_1(x,y,z)$ and a constant $c \geq 1$ similarly as in (\[e3\]) but now with an extra argument $z$ for the non-uniform advice, and a sequence of advice strings $\{w_k\}_k$ such that $|w_k| \le k^c$ (w.l.o.g. we use constant $c$ also in the bound to the length of advice strings). We shall assume that $$\label{h1}
\forall x, z (|z|\le |x|^c)\
[\exists y (|y| \le |x|^c) F_0(x,y,z)] \oplus [\exists y (|y| \le |x|^c) F_1(x,y,z)]$$ is valid where $\oplus$ is the exclusive disjunction. Thus an ${{({{{{\mbox{NP}}}\cap \mbox{co}{{\mbox{NP}}}}})/\mbox{poly}}}$ algorithm is an ${{{{\mbox{NP}}}\cap \mbox{co}{{\mbox{NP}}}}}$ set of pairs $(x, z)$ of appropriate lengths augmented by a sequence of advice strings substituted for $z$. In our situation it is more natural to talk about algorithms than sets as we shall be looking for “errors they make”. We shall denote such an algorithm $({\cal F}, \{w_k\}_k)$ where ${\cal F}$ is the triple $(F_0, F_1, c)$ from (\[h1\]).
For $L$ a language let us denote by $L_k$ the truth table of the characteristic function of $L$ on ${{\{0,1\}^k}}$. If $L \in {{{{\mbox{NE}}}\cap \mbox{co}{{\mbox{NE}}}}}$ then the set $R^L$ of such strings $\{L_k\ |\ k \geq 1\}$ is in ${{\mbox{NP}}}$ and can be defined by an $L_{PV}$ formula as $$z \in R^L\ \mbox{ iff }\
\exists v (|v| \le |z|^d) R^L_0(z,v)$$ with $R^L_0$ an open formula. Any $v$ witnessing the existential quantifier for $z$ will be called a witness for $z \in R^L$. Note that $TAUT \in {{{{\mbox{NE}}}\cap \mbox{co}{{\mbox{NE}}}}}$.
For a language $L \in {{{{\mbox{NE}}}\cap \mbox{co}{{\mbox{NE}}}}}$ and a triple $\cal F$ as in (\[h1\]) define the [**search task $\mbox{{\bf Err}}(L, {\cal F})$**]{} as follows:
- input: $1^{(k)}$, string $L_k$ and a witness $v$ for $L_k \in R^L$, and a string $w$ such that $|w| \le k^c$
- required output: a string $x \in {{\{0,1\}^k}}$ such that $\cal F$ using $w$ as an advice string makes an error on $x$: $$\forall y (|y|\le |x|^c) \
[(x \in L_k \wedge \neg F_1(x,y,w)) \vee
(x \notin L_k \wedge \neg F_0(x,y,w))]$$
\[4.1\] Assume that an exponentially hard one-way permutation exists. Let $L$ be a language such that $L \in {{{{\mbox{NE}}}\cap \mbox{co}{{\mbox{NE}}}}}$.
Then there exists a triple ${\cal F}$ as in (\[h1\]) such that no deterministic polynomial time algorithm can solve $Err(L,{\cal F})$ on all inputs for all sufficiently large lengths $k$.
Assume that language $L$ satisfies the hypothesis of the theorem and let $\cal F$ be any triple as in (\[h1\]). Assume that $\cal A$ is a deterministic polynomial time algorithm that attempts to solve $Err(L, {\cal F})$ on all inputs for all $k \geq k_0$, for some $k_0 \geq 1$.
We are going to define a universal $L_{PV}$ sentence $${\Psi_{L,{\cal F}, {\cal A}, k_0}}$$ that is true iff $\cal A$ solves $Err(L, {\cal F})$ for all inputs for all $k \geq k_0$.
The sentence ${\Psi_{L,{\cal F}, {\cal A}, k_0}}$ is the universal closure of: $$C\ \rightarrow \ D$$ where $C$ is the formula $$|z|=2^k \wedge |v|\le |z|^d \wedge R^L_0(z,v)
\wedge k \geq k_0 \wedge |w| \le k^c \wedge
x = {\cal A}(1^{(k)},z,v,w)$$ with $\cal A$ represented by an $L_{PV}$ function symbol, and $D$ is the formula $$|x| = k \ \wedge \
\forall y (|y|\le |x|^c)
[(x \in z \wedge \neg F_1(x,y,w)) \vee
(x \notin z \wedge \neg F_0(x,y,w))]\ .$$ The following should be obvious:
[**Claim 1:**]{}[ *Algorithm $\cal A$ solves $Err(L, {\cal F})$ for all inputs for all $k \geq k_0$ iff the sentence ${\Psi_{L,{\cal F}, {\cal A}, k_0}}$ is true.* ]{}
We are going now to define a specific ${{({{{{\mbox{NP}}}\cap \mbox{co}{{\mbox{NP}}}}})/\mbox{poly}}}$ triple ${\cal F}$ as in (\[h1\]) such that ${\Psi_{L,{\cal F}, {\cal A}, k_0}}$ will be false for all $L \in {{{{\mbox{NE}}}\cap \mbox{co}{{\mbox{NE}}}}}$, all polynomial time algorithms $\cal A$ and all $k_0 \geq 1$.
Let $h$ be an exponentially hard one-way permutation with a hard bit $B$. Hence the function $f$ from (\[i0\]) has the exponential hardness $H_f(\ell) = 2^{- \ell^{\Omega(1)}}$. The existence of such $h$ and $B$ is guaranteed by the hypothesis of the theorem.
Take $\delta > 0$ provided by Theorem \[3.1\] and put $k:= n^\delta$. Using ${{{NW}_{A_n,f}}}: {{\{0,1\}^n}}\rightarrow {{\{0,1\}^m}}= \{0,1\}^{2^k}$ define an ${{({{{{\mbox{NP}}}\cap \mbox{co}{{\mbox{NP}}}}})/\mbox{poly}}}$ triple ${\cal F} = (F_0, F_1, c)$ as n (\[h1\]) as follows: put $c := \delta^{-1}$ and for $x \in {{\{0,1\}^k}}$, $y \in {{\{0,1\}^n}}= \{0,1\}^{k^c}$, $w$ of size $|z| = n$ and $a = 0,1$ define: $$\label{h2}
F_a(x,y,w)\ :=\
[h(y) = w(J_x) \wedge B(y) = a]\ .$$ In other words, on input $x$ the algorithm computes the $x$-th bit of ${{{NW}_{A_n,f}}}(w)$.
[**Claim 2:**]{} [ *For no $L \in {{{{\mbox{NE}}}\cap \mbox{co}{{\mbox{NE}}}}}$, no polynomial time algorithm $\cal A$ and no $k_0 \geq 1$ is the sentence ${\Psi_{L,{\cal F}, {\cal A}, k_0}}$ true.* ]{}
To see this note that by substituting term ${\cal A}(1^{(k)},z,v,w)$ for $x$ in $C \rightarrow D$ and quantifying it existentially $\exists x (x \in [m])$ allows to deduce from ${\Psi_{L,{\cal F}, {\cal A}, k_0}}$ the universal closure of $A \rightarrow B$ from Theorem \[3.1\]. Hence, by that theorem, ${\Psi_{L,{\cal F}, {\cal A}, k_0}}$ cannot be true.
Claims 1 and 2 imply the theorem.
We remark that the argument can be actually extended to rule out a larger class of algorithms $\cal A$: the so called Student - Teacher interactive computations of [@KPT; @KPS] (see also [@kniha]).
Let $({\cal F}, \{w_k\}_k)$ be as above. Define circuits $D_k(x,y)$ to be (some canonical) circuits with $k$ inputs $x$ and $k^c$ inputs $y$ that outputs $1$ iff $F_1(x,y,w_k)$ holds. We can choose $c \geq 1$ large enough so that $D_k$ has size at most $k^{c^2}$.
Given $1^{(k)}$, a time $2^{O(k)}$ algorithm can compute the string ${{\mbox{TAUT}_k}}$ (as well as the witness for ${{\mbox{TAUT}_k}}\in R^{TAUT}$ required in the general formulation of the theorem). Such an algorithm is then polynomial in the size of ${{\mbox{TAUT}_k}}$. If $\tau$ is a solution to $\mbox{{\bf Cert}}(c)$ on input $D_k$ then either $\tau \in {{\mbox{TAUT}_k}}$ and $\forall y (|y| \le k^c) \neg F_1(\tau, y, w_k)$ or $\tau \notin {{\mbox{TAUT}_k}}$ and $D_k(\tau, y)$ is satisfiable, in which case ${\mbox{T}_{PV}}$ implies that $\forall y (|y| \le k^c) \neg F_0(\tau, y, w_k)$. In other words, an algorithm solving $\mbox{{\bf Cert}}(c)$ solves $\mbox{{\bf Err}}(TAUT, {\cal F})$ too. This yields the following statement as a corollary to Theorem \[4.1\].
Assume that an exponentially hard one-way permutation exists. Then there is $c \geq 1$ such that no deterministic time $2^{O(k)}$ algorithm can solve $\mbox{{\bf Cert}}(c)$ on all lengths $k \geq 1$.
Proof systems with advice and with extra axioms {#5}
===============================================
The task [**Find**]{} was formulated using the provability in a proof system and in this section we develop a technical tool allowing us to move from ${{({{{{\mbox{NP}}}\cap \mbox{co}{{\mbox{NP}}}}})/\mbox{poly}}}$ algorithms to proof systems. We shall recall first the notion of a [**proof system with advice**]{} as introduced by Cook-K.[@CooKra Def.6.1]. It is defined as the ordinary Cook-Reckhow proof system (cf.the introduction) except that the binary relation $P(x,y)$ is decidable in polynomial time using an advice string that depends only on the length of $x$ (the formula). We say that the advice is polynomial iff its length is $|x|^{O(1)}$. This concept has some interesting properties; for example, in the classes of these proof systems - with varying bounds on the size of advice strings - there exists an optimal one. We refer the reader to [@CooKra Sec.6] and to subsequent [@BKM1; @BKM2; @BeyMul] for further information.
Our aim in this section is to link proof systems with polynomial advice with proof systems with extra axioms, as defined below. A sequence of formulas $\{\alpha_k\}_k$ will be called [**p-bounded**]{} iff $|\alpha_k| \le k^{O(1)}$ for all $k$.
\[5.1\] Let $P(x,y)$ be an ordinary Cook-Reckhow proof system.
1. For a tautology $\alpha$ the proof system $P+\alpha$ is defined as follows:
a string $\pi$ is a $(P+\alpha)$-proof of formula $\tau$ iff $\pi$ is a $P$-proof of a disjunction of the form $$\bigvee_i \neg \alpha'_i\ \vee \ \tau$$ where $\alpha'_i$ are arbitrary substitution instances of $\alpha$ obtained by substituting constants and variables for variables.
2. For a p-bounded sequence $\{\alpha_k\}_k$ of tautologies define a string $\pi$ to be an $(P+ \{\alpha_k\}_k)$-proof of formula $\tau$ iff it is a $(P+\alpha_k)$-proof of $\tau$ for $k = |\tau|$.
We allow only substitutions of constants and variables in instances $\alpha_i'$ in part 1 as that makes sense for all proof systems (e.g. we do not have to discuss various limitations on depth for constant depth Frege systems) and it suffices here. Systems $(P+ \{\alpha_k\}_k)$ are not meant to genuinely formalize the informal notion of proof systems with extra axioms; such systems should not pose restrictions on which extra axioms can be used in proofs of which formulas. We use them here only as a technical vehicle allowing us to move from proof systems with advice to ordinary proof systems.
Note that while $P+\alpha$ is a Cook-Reckhow proof system, $P+ \{\alpha_k\}_k$ is generally not. The following lemma is obvious as we can use the sequence $\{\alpha_k\}_k$ as advice strings to recognize $(P+ \{\alpha_k\}_k)$ - proofs.
\[5.2\] Let $P$ be a Cook-Reckhow proof system. For every p-bounded sequence $\{\alpha_k\}_k$ of tautologies $P+ \{\alpha_k\}_k$ is a proof system with polynomial advice in the sense of Cook-K.[@CooKra].
In Section \[tasks\] we used informally the notion of a decent proof system, meaning a proof system that can perform efficiently a few simple manipulations with proofs. We shall use the formalization of this notion from [@Kra-sat Sec.2].
In the following ${{\sf sat}_k}(u,x,v)$ are formulas for $k \geq 1$ and suitable $r = k^{O(1)}$ with $u = (u_1,\dots, u_k)$, $x = (x_1,\dots, x_k)$ and $v = (v_1, \dots, v_r)$ such that for all $a, \varphi \in {{\{0,1\}^k}}$ it holds that:
- [*${{\sf sat}_k}(a,\varphi, v) \in \mbox{TAUT}$ iff $a$ is a truth assignment satisfying formula $\varphi$.*]{}
The extra variables $v$ are used to compute the truth value, as in the ${{\mbox{NP}}}$-completeness of SAT.
A proof system (ordinary or with advice) $P$ is [**decent**]{} iff the following tasks can be performed by polynomial time algorithms[^4]:
1. From a $P$-proof $\pi$ of formula $\psi(x)$ and a truth assignment $a$ to variables $x$ construct a $P$-proof of $\psi(a)$.
2. Given a true sentence $\psi$ (i.e. no variables) construct its $P$-proof.
3. Given $P$-proofs $\pi_1$ of $\psi$ and $\pi_2$ of $\psi \rightarrow \eta$ construct a proof of $\eta$.
4. Given a formula $\varphi(u_1,\dots, u_n)$ and a $P$-proof of formula ${{\sf sat}_k}(u, \varphi, v)$ with variables $u, v$ construct a $P$-proof of $\varphi$.
Conditions D1-3 are easy to verify for many of the usual proof systems (e.g. Frege systems mentioned in the introduction or resolution). The algorithm for condition D4 is defined by induction on the number of connectives in $\varphi$, cf.[@kniha Chpt.9].
\[5.3\] Let $Q$ be a proof system with polynomial advice and $P$ a decent Cook-Reckhow proof system. Then for every constant $c \geq 1$ there exists a p-bounded sequence $\{\alpha_k\}_k$ of tautologies and $d \geq 1$ such that:
Any tautology $\tau$ having a $Q$-proof of size $\le |\tau|^c$ has an $(P+ \{\alpha_k\}_k)$-proof of size $\le |\tau|^d$.
Assume that $Q$ uses polynomial size advice string $w_k$ for formulas of size $k$. For $k \geq 1$ denote by ${{{\sf prov}_Q^{k^c}}}(x,y,t,s)$ a propositional formula such that:
- ${{{\sf prov}_Q^{k^c}}}$ has $k$ atoms $x=(x_1,\dots,x_k)$ for bits of a formula, $k^c$ atoms $y = (y_1,\dots,y_{k^c})$ for bits of a $Q$-proof, $k^c$ atoms $t = (t_1, \dots, t_{k^c})$ for bits of an advice and $k^{O(1)}$ atoms $s = (s_1, \dots, s_{k^{O(1)}})$ for bits of the computation of the truth value of $Q(x,y)$ with advice $t$,
- For size $k$ formula $\varphi$ and a $k^c$ size strings $\pi$ and $w$: ${{{\sf prov}_Q^{k^c}}}(\varphi, \pi, w, s) \in \mbox{SAT}$ iff $Q(\varphi,\pi)$ with advice $w$ is true.
Then take for $\alpha_k$ the formula with variables $x,y,z,s,v$ $$\label{m0}
{{{\sf prov}_Q^{k^c}}}(x,y,t/w_k,s)\ \rightarrow\ {{\sf sat}_k}(x,z,v)$$ where $w_k$ is the string used by $Q$ for size $k$ formulas. The formula expresses the soundness of $Q$ and hence it is a tautology. Its total size is $k^{O(c)}$.
Let $\varphi$ be a size $k$ formula with variables among $z=(z_1,\dots,z_k)$ and having a size $\le k^c$ $Q$-proof $\pi$. Let $e$ be bits of the computation of $Q(\varphi,\pi)$ with advice $w_k$.
Take the following substitution instance of $\alpha_k$: $${{{\sf prov}_Q^{k^c}}}(\varphi, \pi, w_k, e) \ \rightarrow\ {{\sf sat}_k}(\varphi, z, v)$$
[**Claim:**]{}[ *For some $d \geq 1$ depending only on $c \geq 1$ and $P$ the formula $\varphi$ has a $(P+\alpha_k)$-proof of size $\le k^d$.*]{}
We shall use the decency of $P$. By the choice of $\pi$ and $e$ the sentence ${{{\sf prov}_Q^{k^c}}}(\varphi, \pi, w_k, e)$ is true and hence has, by D2, a size $k^{O(c)}$ $P$-proof. Then, using D3, use modus ponens to derive in size $k^{O(c)}$ the formula $${{\sf sat}_k}(\varphi, z, v)\ .$$ D4 then allows to derive in $P$ the formula $\varphi$, in size $k^{O(1)}$. The total size of the $P$-proof is $k^{O(c)}$.
Note that we have not used the decency condition D1 explicitly; it’s role is replaced here by the definition of the system $P + \alpha_k$ which takes as axioms all substitution instances of $\alpha_k$.
Formulas $\alpha_k$ depend not only on $k$ and $w_k$ but also on the bound $k^c$ to the length of $y$. This is the reason why we cannot simply say that $P + \{\alpha_k\}_k$ simulates $Q$ (it simulates only[^5] $Q$-proofs of size at most $k^c$).
The partial definability of tasks [**Cert**]{} and [**Find**]{} {#6}
===============================================================
We will need the following notion. For a complexity class $\mathcal X$ and a language $L$ define the property that [**$L$ is infinitely often in ${\mathcal X}$**]{}, denoted $L \in_{i.o.} {\mathcal X}$, iff there exists $L'\in {\mathcal X}$ such that $$L \cap {{\{0,1\}^k}}\ =\ L' \cap {{\{0,1\}^k}}$$ for infinitely many lengths $k$. Recall the definition of the class ${{({{{{\mbox{NP}}}\cap \mbox{co}{{\mbox{NP}}}}})/\mbox{poly}}}$ at the beginning of Section \[4\].
The following consequence of Conjecture \[2.1\] was noted at the end of [@k2 Sec.30.2] and it uses an idea linking the output/input ratio of proof complexity generators with the unprovability of circuit lower bounds due to Razborov [@Raz95], quite similar to the reasoning in Razborov-Rudich [@RR].
\[2.3\]
Assume that Conjecture \[2.1\] holds and that an exponentially hard one-way permutation exists. Then for every $L \in {{{{\mbox{NE}}}\cap \mbox{co}{{\mbox{NE}}}}}$: $$L \in_{i.o.} {{({{{{\mbox{NP}}}\cap \mbox{co}{{\mbox{NP}}}}})/\mbox{poly}}}\ .$$ In particular, $TAUT \in_{i.o.} {{\mbox{NP}}}/poly$.
Take $\delta > 0$ from Conjecture \[2.1\]. Put $k := n^{\delta}$ and think of a string $b \in {{\{0,1\}^m}}$ as of the truth-table of the characteristic function of the language $L$ on inputs of length $k$; denote it $L_k$ as in Section \[4\].
For any language $L$ in ${{{{\mbox{NE}}}\cap \mbox{co}{{\mbox{NE}}}}}$ the set of strings $\{L_k\ |\ k \geq 1\}$ is in ${{\mbox{NP}}}$: the ${{\mbox{NP}}}$ witness can collect all $2^k$ ${{\mbox{NE}}}$ witnesses for each variable setting - this will have size $2^{O(k)}$ - and check their validity.
In particular, if some $L \in {{{{\mbox{NE}}}\cap \mbox{co}{{\mbox{NE}}}}}$ would determine the truth-tables $L_k$ for $k = n^{\delta}$ such that all but finitely many lie outside the range of ${{{NW}_{A_n,f}}}$ we would get a contradiction with Conjecture \[2.1\]. Hence we get:
[**Claim:**]{} [*For infinitely many $n$, for $k = n^{\delta}$ and $m = 2^k$: $$L_k \in {{\{0,1\}^m}}\cap Rng({{{NW}_{A_n,f}}})\ .$$*]{} For $L_k \in Rng({{{NW}_{A_n,f}}})$ let $a \in {{\{0,1\}^n}}$ be such that ${{{NW}_{A_n,f}}}(a) = L_k$. Then computing $L$ on $i \in {{\{0,1\}^k}}$ amounts to computing $f$ on $a(J_i)$. But by the requirement (\[e2\]) posed on $A_n$ the set $a(J_i)$ can be computed from $i$ and $a$ (taken as the advice string) in time polynomial in $n$ and $f$ is an ${{\mbox{NP}}}\cap \mbox{co}{{\mbox{NP}}}$ function.
This lemma has an immediate consequence for problem [**Cert**]{}.
Assume that Conjecture \[2.1\] holds and that an exponentially hard one-way permutation exists.
Then for some $c \geq 1$ the task $\mbox{{\bf Cert}}(c)$ is only partially defined for infinitely many lengths $k \geq 1$, i.e. there are inputs corresponding to $k$ for which the problem has no solution.
We shall derive the same consequence for the task [**Find**]{} using the results of Section \[5\]. For a triple $\cal F$ as in (\[h1\]) define a proof system with polynomial advice $Q(x,y)$ by:
- either $0 < |w|\le |x|^c \wedge F_1(x,y,w)$,
- or $w$ is the empty word and $y$ is a Frege proof of $x$,
thinking of $w$ as the advice. Note that condition (\[h1\]) does not guarantee that $F_1(x,y,w)$ is a proof system for some $w$ but the second clause of the definition of $Q(x,y)$ allows us to fall back on a Frege system by taking for the advice the empty word.
Now let $\{w_k\}_k$ such that $|w_k| \le k^c$ be a sequence of advice words defining a proof system with advice $Q$ (there exists at least one such: the sequence of empty strings). Let $P$ be any decent Cook-Reckhow proof system and let formulas $\alpha_k$ and constant $d \geq 1$ be those provided by Lemma \[5.3\] for $c$ from $\cal F$. Assume $c_0 \geq 1$ is such that $|\alpha_k|\le k^{c_0}$ and assume also w.l.o.g. that $d \geq c_0$.
Consider the task $Find(P,c_0,d)$. A solution for input $1^{(k)}$ and $\alpha_k$ is a size $k$ tautology $\beta$ and by the choice of $c_0, d$ it must be that $$\forall y (|y| \le k^c)\ \neg F_1(\beta, y, w_k)\ .$$ But Conjecture \[2.1\] implies analogously as above that for the triple $\cal F$ coming from ${{{NW}_{A_n,f}}}$ there will be infinitely many lengths $k\geq 1$ and strings $w_k$ for which $\alpha_k$ is a tautology but no such $\beta$ exists. Hence we have the following statement.
Assume that Conjecture \[2.1\] holds and that an exponentially hard one-way permutation exists.
Then for all decent Cook-Reckhow proof systems $P$ there are constants $c_1 \geq c_0 \geq 1$ such that the task $\mbox{{\bf Find}}(P, c_0, c_1)$ has no solution for infinitely many lengths $k \geq 1$.
Disjoint ${{\mbox{NP}}}$ pairs {#pairs}
==============================
Let $(U,V)$ and $(A,B)$ be two pairs of disjoint subsets of ${{\{0,1\}^*}}$. A [**reduction**]{} of $(A,B)$ to $(U,V)$ is a function $f : {{\{0,1\}^*}}\rightarrow {{\{0,1\}^*}}$ such that for all $u$: $$u \in A \rightarrow f(u) \in U\ \wedge\
u \in B \rightarrow f(u) \in V\ .$$ It is (non-uniform) [**p-reduction**]{} if $f$ is (non-uniform) p-time.
Disjoint ${{\mbox{NP}}}$ pairs appear quite naturally in several places of proof complexity. Most notably, the reflection principle for a proof system just asserts that two ${{\mbox{NP}}}$ sets (that of formulas with bounded size proofs and of falsifiable formulas) are disjoint. A particularly elegant form of this observation was found by Razborov [@Raz-pairs] in the notion of the canonical pair of a proof system. Shadowing the relation of the provability of reflection principles to simulations, a similar one exists between the provability of disjointness of such pairs and simulations. We refer the reader to [@Pud-survey] for more background.
Given two pairs of disjoint sets $(U,V)$ and $(A,B)$ and a constant $c \geq 1$ consider the [**search task**]{} $\mbox{{\bf Pair}}_{U,V}^{A,B}(c)$:
- input: $1^{(k)}$ and a circuit $C$ with $k$ inputs, several outputs and of size at most $k^c$
- required output: a string $u \in {{\{0,1\}^k}}$ such that $$u \in A \wedge f(u) \notin U\ \mbox{ or }\
u \in B \wedge f(u) \notin V\ .$$
In other words, the output string $u$ certifies that circuit $C$ is not a reduction of $(A, B)$ to $(U,V)$ on ${{\{0,1\}^k}}$.
Take a triple $\cal F$ as in (\[h1\]) and define $U$ and $V$ to be the sets of pairs $(x,z)$ such that $|z|\le |x|^c$ and $\exists y (|y|\le |x|^c)F_0(x,y,z)$ or $\exists y (|y|\le |x|^c)F_1(x,y,z)$, respectively.
For a disjoint pair $A, B$ of sets such that $A \in {{{{\mbox{NE}}}\cap \mbox{co}{{\mbox{NE}}}}}$ take for language $L$ on ${{\{0,1\}^k}}$ simply $A$. For $w$ of size $\le k^c$ consider circuit $C_w$ that takes size $k$ input $x$ and outputs the pair $(x,w)$; note that $|C_w| \le k^{c+1}$ for $k >> 0$. Then a solution to $\mbox{{\bf Pair}}_{U,V}^{A,B}(c+1)$ for input $1^{(k)}$ and $C_w$ is also a solution to $Err(L,{\cal F})$. Hence Theorem \[4.1\] implies the following statement.
Assume that an exponentially hard one-way permutation exists.
Then there are two disjoint ${{\mbox{NP}}}$ sets $U, V$ and $c \geq 1$ such that for any pair $A, B$ of disjoint sets such that $A \in {{{{\mbox{NE}}}\cap \mbox{co}{{\mbox{NE}}}}}$ the task $\mbox{{\bf Pair}}_{U,V}^{A,B}(c)$ is not solvable by a deterministic time $2^{O(k)}$ algorithm.
The reader familiar with the canonical pairs of proof systems mentioned earlier may note that these sets are in ${{\mbox{E}}}\subseteq {{{{\mbox{NE}}}\cap \mbox{co}{{\mbox{NE}}}}}$ and thus the theorem applies to them.
Concluding remarks {#rema}
==================
The role of Conjecture \[2.1\] is rather ambivalent: it implies that ${{\mbox{NP}}}\neq co{{\mbox{NP}}}$ (Lemma \[2.2\]) but also that $TAUT \in _{i.o.} {{\mbox{NP}}}/poly$ (Lemma \[2.3\]). This is caused by the dual role of the Nisan-Wigderson generator; it is a source of hard tautologies but also a strong proof system. The reader should consider, before dismissing Conjecture \[2.1\] because of Lemma \[2.3\], how little contemporary complexity theory understands about the power of non-uniform advice.
It would be interesting to have a variety of candidate combinatorial constructions $Q(P)$ of a proof system stronger than $P$, as discussed in Section \[tasks\]. At present only the construction of $iP$, the implicit $P$, from [@Kra-implicit] applies to all proof systems and it is consistent with the present knowledge that it indeed yields stronger proof systems. An indirect plausible construction of $Q(P)$ may use the relation between proof systems and first-order theories: take theory $T_P$ corresponding to $P$, extend $T_P$ by $Con(T_P)$ (or in some other Gödelian fashion) getting $S$, and then take for $Q(P)$ the proof system simulating $S$ (cf.[@KP-jsl; @kniha] about $T_P$ etc.). But it is hardly combinatorially transparent.
The referee pointed out that problems $Cert$ and $Find$ seem to be very close to each other and asked whether they could be, in fact, reduced to each other. One can reduce $Find(P, c_0, c_1)$ to $Cert(c)$ for suitable $c$: for $k \geq 1$ and a tautology $\alpha$ of size at most $k^{c_0}$ take for $D$ a circuit (canonically constructed) computing the provability relation for $P + \alpha$ on formulas of size $k$ and proofs of size at most $k^{c_1}$. The size of $D$ is polynomial in $k^{c_1}$ and so $D$ is a valid input to $Cert(c)$ for some $c_1 \le c = O(c_1)$. A solution to $Cert(c)$ cannot certify that $D$ is unsound and so it has to be a size $k$ formula certifying the lower bound $k^c \geq k^{c_1}$ for $D$ and hence the $k^{c_1}$ lower bound for $P + \alpha$ too. I do not know whether $Cert$ can be also reduced to $Find$ but I suspect it cannot be: valid inputs to $Cert$ are allowed either to fail to be a sound proof system or to fail to have short proofs while inputs to $Find$ cannot be unsound.
I thank the anonymous referee for comments that lead to a better presentation of the results.
[44]{}
, Nondeterministic Instance Complexity and Proof Systems with Advice, in: Proc. 3rd International Conference on Language and Automata Theory and Applications (LATA), Springer-Verlag, LNCS 5457, (2009), pp.164-175.
, Proof Systems that Take Advice, [*Information and Computation*]{}, Vol. [**209 (3)**]{}, (2011), pp.320-332.
, Does Advice Help to Prove Propositional Tautologies?, in: Proc. 12th International Conference on Theory and Applications of Satisfiability Testing (SAT), Springer-Verlag, LNCS 5584, (2009), pp.65-72.
, Characterizing the Existence of Optimal Proof Systems and Complete Sets for Promise Classes. in: A.E.Frid, A.Morozov, A.Rybalchenko, K.W.Wagner (Eds.), [*Computer Science - Theory and Applications*]{}, Fourth International Computer Science Symposium in Russia, CSR 2009, Novosibirsk, Russia, (August 18-23, 2009). Lecture Notes in Computer Science [**5675**]{}, Springer. (2009), pp.47-58.
, The intrinsic computational difficulty of functions, in : [*Proc. Logic, Methodology and Philosophy of Science*]{}, ed. Y. Bar-Hillel, North-Holland, (1965), pp. 24-30.
, Feasibly constructive proofs and the propositional calculus, in: [*Proc. 7$^{\mbox{th}}$ Annual ACM Symp. on Theory of Computing*]{}, (1975), pp. 83-97. ACM Press.
, Consequences of the Provability of $NP \subseteq P/poly$, [*J. of Symbolic Logic*]{}, [**72(4)**]{}, (2007), pp. 1353-1371.
, The relative efficiency of propositional proof systems, [*J. Symbolic Logic*]{},[**44(1)**]{}, (1979), pp.36-50.
, [*Foundations of cryptography*]{}, Vol.1, Cambridge University Press, (2001).
, Speed-up for Propositional Frege Systems via Generalizations of Proofs, [*Commentationes Mathematicae Universitas Carolinae*]{}, [**30(1)**]{}, (1989), pp.137-140.
, [*Bounded arithmetic, propositional logic, and complexity theory*]{}, Encyclopedia of Mathematics and Its Applications, Vol. [**60**]{}, Cambridge University Press, (1995).
, On methods for proving lower bounds in propositional logic, in: [*Logic and Scientific Methods*]{} Eds. M. L. Dalla Chiara et al., (Vol. 1 of Proc. of the Tenth International Congress of Logic, Methodology and Philosophy of Science, Florence (August 19-25, 1995)), Synthese Library, Vol.259, Kluwer Academic Publ., Dordrecht, (1997), pp.69-83.
, Implicit proofs, [*J. of Symbolic Logic*]{}, [**69(2)**]{}, (2004), pp.387-397.
, Proof complexity, in: Laptev, A. (ed.), European congress of mathematics (ECM), Stockholm, Sweden, June 27–July 2, 2004. Zurich: European Mathematical Society, (2005), pp.221-231.
, [*Forcing with random variables and proof complexity*]{}, London Mathematical Society Lecture Notes Series, Vol. [**382**]{}, Cambridge University Press, (2011).
, On the proof complexity of the Nisan-Wigderson generator based on a hard ${{{{\mbox{NP}}}\cap \mbox{co}{{\mbox{NP}}}}}$ function, [*J. Mathematical Logic*]{}, Vol.[**11(1)**]{}, (2011), pp.11-27.
, A note on SAT algorithms and proof complexity, [*Information Processing Letters*]{}, [**112**]{}, (2012), pp.490-493.
, Propositional proof systems, the consistency of first order theories and the complexity of computations, [*J. Symbolic Logic*]{}, [**54(3)**]{}, (1989), pp.1063-1079.
, Interactive Computations of Optimal Solutions, in: B. Rovan (ed.): [*Mathematical Foundations of Computer Science*]{} (B. Bystrica, August ’90), Lecture Notes in Computer Science 452, Springer-Verlag, (1990), pp. 48-60.
, Bounded arithmetic and the polynomial hierarchy, [*Annals of Pure and Applied Logic*]{}, [**52**]{}, (1991), pp.143–153.
, Hardness vs. randomness, [*J. Comput. System Sci.*]{}, Vol.[**49**]{}, (1994), pp.149–167.
, Nisan-Wigderson generators in proof systems with forms of interpolation, [*Mathematical Logic Quarterly*]{}, [**57(3)**]{}, (2011), pp.379-383.
, The lengths of proofs, in: Handbook of Proof Theory, S.R. Buss ed., Elsevier, (1998), pp.547-637.
, Unprovability of lower bounds on the circuit size in certain fragments of bounded arithmetic, [*Izvestiya of the R.A.N.*]{}, [**59(1)**]{}, (1995), pp.201-224.
, On Provably Disjoint NP-pairs, unpublished manuscript, (1994).
, Pseudorandom generators hard for $k$-DNF resolution and polynomial calculus resolution, preprint, (May’03).
, Natural proofs, [*J. of Computer and Systems Science*]{}, [**55(1)**]{}, (1997), pp.24-35.
[^1]: One can read these chapters independently of the rest of the book.
[^2]: Really just a bit: for decent $P$ adding $\alpha$ is equivalent to adding the reflection principles for $P+\alpha$.
[^3]: The proof system $P + \alpha$ will be defined in Section \[5\].
[^4]: Polynomially bounded functions would suffice for us here but such a weakening would not put more of the usual proof systems into the class and so we just stick with the definition from [@Kra-sat].
[^5]: $P$ augmented by all such formulas $\alpha_k$, constructed for all $k$, $w_k$ and all $c \geq 1$, does simulate $Q$ as by the proof of the claim $d = O(c)$.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
We have carried out a deep (t=30000s) x-ray search of the eastern portion of the L1495 cloud centered on the well known weak line T Tauri star (WTTS) V410 Tau using the ROSAT PSPC. This deep exposure enabled a search for candidate pre-main sequence (PMS) objects in this cloud to a limit $\sim 20$ times more sensitive than that typical of the fields examined with the [*Einstein*]{} searches. Despite assertions that the PMS population in Taurus-Auriga is nearly completely known, this x-ray survey revealed 8 new PMS objects in a region 50$\:$ in diameter, as compared to a previously known stellar population of 12 objects, including deeply embedded IRAS sources.
Spectroscopic and photometric observations enable us to place these objects in the HR Diagram. The newly discovered objects are predominantly stars of spectral type M0 and later, and a large fraction (6/8) appear to be surrounded by circumstellar accretion disks as judged by their infrared excess and H$\alpha$ emission. We combined the data for these x-ray discovered objects with extant and new data for the previously identified PMS stars in this region to examine the history of star formation and the frequency distribution of stellar masses in this cloud.
If the “post ROSAT” population is either complete or representative, we conclude (1) that star formation in L1495 East took place $\sim1\times 10^6$ yrs ago and that the spread in ages is small; (2) the frequency distribution of masses, N(M), in this apparently coeval group appears to peak near $\log M = -0.5$ (using masses derived from the recently published PMS tracks of D’Antona & Mazzitelli (1993) and Swenson et al.(1993)) and to decline toward lower masses. The derived N(log M) for L1495E compares well with the IMF derived from studies of stars in the solar neighborhood, a result which suggests that the Taurus-Auriga clouds are currently producing stars whose mass spectrum approximates the time/space averaged IMF for the solar neighborhood.
author:
- 'Karen M. Strom and Stephen E. Strom'
title: A Multiwavelength Study of Star Formation in the L1495E Cloud in Taurus
---
Introduction
============
X-ray searches in regions of recent star formation have proven efficacious not only in recovering previously known solar-type PMS objects (the classical T Tauri stars or CTTS), but in locating a class of objects of which few were previously identified, the weak line T Tauri stars (WTTS) which lack the emission phenomena (H$\alpha$ emission; metallic line emission; optical excess (veiling) and infrared excess emission) which characterize CTTS. We report in this contribution the results of a deep search for x-ray sources in the vicinity of an actively star forming cloud within the Taurus-Auriga cloud complex: L1495 E(ast). Our ROSAT PSPC image is the result of a total integration time of $\sim 33000$s and is centered on V410 Tau. We had originally hoped to use these data to search for periodic variations in x-ray flux from V410 Tau. However the exposure did not span a large enough fraction of the 1.88 day rotation period to provide a definite result. The long effective integration time did permit a search for sources with x-ray luminosities as faint as $3.0\times10^{28}\,$ ergs/s (unreddened), $\sim 20\:$ times deeper than any previous x-ray exposure in this region. As a result we located a number of additional candidate PMS stars (8 in number, compared to a previously identified PMS population of 12 objects in the same area). The goals of this study are (1) to locate optical counterparts for the newly identified sources; (2) to locate these objects in the HR Diagram using red spectra and optical and near infrared photometry; (3) to determine masses, ages and disk frequency for these objects; and (4) to construct an IMF for this population.
Cloud Environment and Previous Studies of the Stellar Population
================================================================
Large scale studies of the L1495 molecular cloud have been published by Duvert, Cernicharo & Baudry (1986) and Fukui et al. (1992). These maps were made with radio telescopes of 2.5m and 4m diameter and therefore of relatively low spatial resolution ($\theta \sim 2\farcm 5$). However both maps show that the greatest concentration of molecular material is centered on the region containing V410 Tau, V892 Tau, DD Tau and CZ Tau, the region upon which our ROSAT PSPC image is centered. Unfortunately, the region has not yet been studied either at high spatial resolution or in molecules sensitive to high densities (e.g. NH$_3$ ). Most molecular studies have been concentrated in the less dense western part of this cloud. In view of the youth of this region and the degree of star forming activity, it would be of great interest to pursue such studies.
Recent observations with the Nagoya University radio telescope, centered on known PMS objects show that the velocity dispersion as measured by line widths in this part of L1495 is larger than that seen in any other part of the Taurus-Auriga clouds, $\geq$3 km/s. The velocity dispersion drops considerably (to 1-1.5 km/s) in the outer parts of the cloud. In the western star forming center of L1495 (centered on V773 Tau) the velocity dispersion is $\sim 2$ km/s. Duvert et al. (1986) attributed the measured velocity dispersion to multiple components with a cloud collision possibly underway. However, while some of the line profiles appear to have weak additional components, the line profiles at cloud center appear to be smooth gaussians with FWHM of $\geq3$ km/s. The mass of L1495E, within the region covered by the ROSAT PSPC image, lies between 180 $M_\odot$ as determined from the Nagoya C$^{18}$O map and 230 $M_\odot$ as determined from the $^{13}$CO map (Ohnishi & Mizuno 1993)
Within a circle of 1$\,$ radius surrounding V410 Tau there are 19 known PMS objects; 12 of these lie within a circle of 25$\,$ radius centered in the ROSAT field. There is one Herbig Ae/Be star, V892 Tau (Elias 1), centrally located in the cloud; also located in the central region are 5 WTTS, 3 CTTS and 3 heavily embedded Class I sources (Lada 1987), one of which is known to drive a jet seen in H$\alpha$ and \[\] (Strom et al.1986, Goodrich 1993). In the outer region, 30 $<$ r $<$ 55, there are 4 CTTS and 3 WTTS. While M type stars are known among this population, they are all [*early*]{} M stars (spectral types M2 or earlier). This region of the Taurus clouds thus exhibits the spectral type distribution characteristic of the rest of the Taurus clouds which show an apparent deficiency of late M stars, spectral types later than M2 (Cohen & Kuhi 1979), and therefore an apparent deficiency of stars with masses $M <
0.2 M_\odot$.
Observations
============
The x-ray observation of the L1495E cloud was obtained in pointing mode with the ROSAT telescope using the position sensitive proportional counter (PSPC) , between $4^h 09^m$ on 1991 Mar. 4 UT and $23^h 25^m$ on 1991 Aug. 2 UT. The total observation time over this period was 32986s. A description of the satellite, the telescope and the detector are given by Trümper (1983) and Pfeffermann et al. (1986). Because a portion of the data included in the standard analysis (SASS, version 5-6) contained observations taken during a period when the satellite was rolling (as indicated by the “housekeeping” data) the observation was rereduced excluding data taken during this period; the effective observation time is 25757s. The data were rereduced, excluding the portion affected by satellite roll, and analyzed using the PROS package (versions 2.1 and 2.2) under IRAF.
The optical photometry at R and I was obtained on the night of 1992 Dec 11 UT with a Tektronix 2048x2048 CCD mounted in a universal CCD dewar on the Kitt Peak National Observatory 0.9m telescope by Patrick Hartigan. The plate scale for the camera/telescope combination used for this observation was 0775/px. A series of exposure times ranging from 10s through 300s was used for observations at I in order to obtain unsaturated images of the bright stars. A single exposure at R of 30s duration was made. Only 2 stars suffered from saturated images in this exposure. Since the field was not crowded, the photometry was reduced using the DIGIPHOT package in IRAF using the new Landolt standard fields (Landolt 1992) and the PHOTCAL routine (Davis 1993).
The near infrared photometry was obtained during 4 different runs due to the convolved effects of the small size of infrared arrays and the terrible weather in Arizona during the winter of 1993. The largest batch of near infrared observations was taken by Ronald Probst on the night of 1993 Jan 12 UT using the SQIID photometer on the 1.3m telescope at KPNO. Simultaneous images in J, H, and K were obtained using 256 x 256 PtSi arrays over a field of dimension $5 \farcm 4$ x $5 \farcm 4$; the size of each pixel is 136. Each image is the sum of 2 “dithered” images, offset by $\sim20\arcsec \:$ with an integration time of 180s apiece. One additional field was obtained by Lynne Hillenbrand and the authors on 1992 Dec 27 UT, again with SQIID on the 1.3m telescope at KPNO. The field centered on V892 Tau (Elias 1) was observed on 1992 Jan 27 UT with this same instrumentation by the authors. These data were all reduced using the SQIID package under IRAF. A few additional single channel photometer measurements collected for other programs at San Pedro Martir are also included.
One additional field, as well as 2 previously observed fields, were observed by Michael Meyer and Patricia Knesek on 1993 March 11 UT at the 2.4m Hiltner telescope at the MDM observatory on Kitt Peak using the NicMass camera, equipped with a 256 x 256 HgCdTe NICMOS array having a field of dimension $4 \farcm 25$ x $4 \farcm 25$, and a plate scale of $0 \farcs 81$/px. A series of four 3s integrations in each filter was followed by a series of four 10s integrations. These data were also reduced using the SQIID package under IRAF.
Spectra were obtained with HYDRA, the fiber optic coupled spectrograph on the 4m telescope at KPNO. We used the red fiber cable in conjunction with the BL181 grating (316 l/mm) set at a central wavelength of 7600Å, and a Tektronix 2048x2048 CCD to obtain spectra with a usable range from $5500\AA \,$ to $9200\AA$; the effective resolution was $\sim$3 pixels or 6Å. The spectra were then reduced using the DOHYDRA package within IRAF. Further analysis was done using the ONEDSPEC package. Spectral standards were established using the main sequence proper motion members of Praesepe (Jones & Stauffer 1991) for spectral types between F6 and M4. These standards were supplemented by the spectra of Kirkpatrick, Henry & McCarthy (1991) for the late M stars.
Identification of Optical Counterparts to ROSAT Sources
=======================================================
Our deep ROSAT PSPC exposure centered on V410 Tau allows us to assess the completeness of the known pre-main sequence stellar population in this nearby active star formation region, of which only the periphery was observed with the [*Einstein*]{} satellite. The pre-ROSAT surveys do not provide the basis for evaluating the number of very low mass stars ($M \leq 0.3 M_\odot$) in this region, leaving the question of the population of the low mass end of the IMF unaddressed. Some information is provided by extant H$\alpha$ surveys and the LkCa survey. This part of the Taurus clouds was covered by the Herbig, Vrba & Rydgren (1986) survey for stars exhibiting Ca II H & K emission and three such stars (LkCa 4, 5, & 7) were found within the field of this ROSAT observation. These stars exhibit H$\alpha$ emission and strong $\lambda 6708\AA$ absorption (Strom et al. 1989a, Walter et al. 1988) and are clearly pre-main sequence objects. However their H$\alpha$ emission equivalent widths are too small to allow detection by H$\alpha$ objective prism surveys. Unfortunately, both the H$\alpha$ and emission line surveys reach only the brighter stars in the Taurus clouds (R $\leq$ 13), and are incomplete for fainter stars, usually locating only those faint stars with unusually strong H$\alpha$ emission.
Our deep ROSAT PSPC image of the V410 Tau region should enable a deep search for faint pre-main sequence stars based on the elevated x-ray luminosity characteristic of PMS stars. Relative to main sequence stars of the same spectral type, the x-ray luminosities of PMS stars may be $10^3$ times greater (Feigelson 1986, Strom et al. 1990). Both CTTS and WTTS are strong x-ray emitters, with no significant difference in the x-ray properties of the two groups (Feigelson & Kriss, Strom et al. 1990, Gauvin & Strom 1992). Our observation should allow us to detect lightly obscured stars at the distance of the Taurus clouds (160 pc) having x-ray luminosities a factor of 20 lower than those found with the [*Einstein*]{} observations.
Unfortunately the sensitivity of a ROSAT PSPC observation is not uniform across the field. The point spread function of the ROSAT PSPC is a strong nonlinear function of the distance from field center, varying from FWHM $\sim25\arcsec \,$ at field center to $1\arcmin \,$ at $\sim 30\arcmin \:$ from field center to $4\arcmin \;$ at the edge of the $2\deg \,$ field; therefore the sensitivity decreases rapidly outside the central 25$\,$ of the field. All but 3 (CK Tau 1, PSC04154+2823, PSC 04158+2805) of the 12 previously identified pre-main sequence stars (Herbig & Bell 1988) within the central field of radius $25\arcmin \,$ were easily detected. There is a faint extension on the x-ray image of Hubble 4 that may be attributable to CK Tau 1. In the outer annulus $(25\arcmin \leq r \leq
55\arcmin)\;$ 6 more previously identified members (RY Tau, BP Tau, HDE 283572, FO Tau, LkCa 4, LkCa 7) of the Taurus cloud population are found. One of the previously known PMS stars, FO Tau, is not detected in this exposure. It is located at a distance $51\farcm 8 \;$ from the field center. While other PMS stars were detected at a distance from field center almost this great, (notably RY Tau and HDE 283572 at $r = 46\farcm2$), the much lower luminosity of FO Tau apparently precludes its detection. In addition to the previously known members of the PMS population, eight additional x-ray detections ($> 5 \sigma$) appear in the central region of our exposure, which is centered on the densest part of the dark cloud. The positions of the previously known PMS stars are offset from the x-ray positions derived from the standard analysis by $\Delta \alpha =
-3\farcs8 \pm 1\farcs0$ and $\Delta \delta = 8\farcs0 \pm 2\farcs5$. Using these offsets as a template, we were able to identify optical counterparts for each of the new x-ray sources on our deep I band CCD frame. Within the positional error given by the local x-ray point spread function ($\Delta\theta \leq 10\arcsec$), 6 sources (V410 x-ray 1, 2, 3, 4, 6, & 7) appear to have unique optical counterparts while only two sources (V410 x-ray 5 & 8) within the central region of the field had more than one optical object as a possible identification. The object farthest from field center (V410 x-ray 8, r = 185) is located near the cloud periphery, hence complicating the identification owing to increasing contamination by background field stars. In this case we examined 6 possible identifications. Although most of these 6 candidates are much too far from the x-ray position, we felt that the possibility that the brightest star of the 6, x-ray 8a, was the true x-ray source should be explored. The positional information for the sources detected in our ROSAT observation, as well as for the other PMS objects in the field, are presented in Table 1. The table gives the object name, the optical position, the x-ray position, the difference between the x-ray and optical positions in the sense x-ray minus optical, and the radial distance from field center. A finding chart is given in Figure 1 (Plate XX).
X-ray Properties of the Sources
-------------------------------
For each detected x-ray photon, the time of arrival, position and energy are recorded. This allows us to evaluate the variability of the source and to fit its spectral energy distribution with models which take into account both extinction from the intervening interstellar medium and the source temperature. Assuming that the x-ray emission from each of these sources arises in a stellar corona, we model the observed spectrum with a Raymond-Smith (Raymond & Smith 1977) thermal plasma, assuming solar abundances. We then solved for the temperature of the plasma and the column density of Hydrogen atoms along the line of sight assuming the Morrison-McCammon (Morrison & McCammon 1983) interstellar x-ray absorption model. For the brighter x-ray sources, the signal to noise was sufficiently high to note strong systematic behavior in the fit residuals. Where appropriate, these sources were then refit with a two temperature model (Schrijver, Lemen & Mewe 1989) where the line of sight extinction was held constant but two components of different temperature were included in the fit. These models provided substantially better fits (the $\chi ^2$ values were improved by factors of 5 to 35) to the x-ray spectra of the brighter sources. Single component models were used for the fainter sources. The results of these fits for all stars are given in Table 2. Examples of the fits obtained for two high signal to noise sources (V410 Tau and Hubble 4) with different line of sight extinctions are shown in Figure 2. Also shown are the $\chi^2$ plots in the temperature-$\log N_H$ plane. These plots demonstrate the sensitivity of the fits to line of sight extinction, even when the count rate is high. Using these fits, we are then able to calculate extinction corrected x-ray luminosities (0.2-2.4 KeV) for these sources. When line of sight extinction is sufficiently high that few low energy photons are detected, the error in the temperature determination increases, sometimes substantially. The two temperature model serves only as an approximation to the more likely situation of a continuous temperature distribution. Therefore we feel that the presentation of formal error bars for the temperature and emission measure values would not be physically meaningful. Higher spectral resolution is necessary in order to make more physically meaningful measurements of the temperature distribution found in the region where the x-ray emission is formed. For sources with too few counts to yield a reliable fit, a model with $\log
N_H = 21$ and $T = 1$ KeV was fit in order to allow an estimate of the x-ray luminosity.
We can compare the distribution of x-ray luminosities of the newly discovered x-ray objects with those of the previously identified PMS population of L1495E. The distributions of the 2 populations are shown in Figure 3. It is obvious that, while the [*range*]{} in x-ray luminosity for the 2 groups is similar, the [*distributions*]{} are vastly different. The median x-ray luminosity for the previously identified population is $\log L_x =$ 30.26; the median x-ray luminosity for the newly identified population is $\log L_x =$ 29.00. Therefore, by substantially increasing our sensitivity to low luminosity sources, we have demonstrated that [*PMS*]{} stars may have x-ray luminosities at least as low as those found for Pleiades [*main sequence*]{} stars of solar type (Stauffer et al. 1993).
While it was hoped that this observation would cover a full rotation period of V410 Tau, scheduling constraints led to the coverage of only $\sim\onehalf $ of the period of this star. V410 Tau shows clear evidence of high amplitude variation. We were unable to comment definitively on the relationship between the x-ray light curve and that seen in the optical. However we are able to extract variability data for each of our detected objects on shorter timescales because the arrival time of each detected photon is recorded as part of the observation. The Kolgomorov-Smirnov and Cramer-von Mises tests for constancy were performed on the time sorted event files for the brighter sources in this field. Nine of the 12 sources bright enough to yield reliable results demonstrate clear variability. These results are also tabulated in Table 2 where we list the probability of variability found for the sources examined as well as the ratio of the maximum count rate to the minimum count rate seen in an 860s bin. The large number of x-ray variables is not surprising in light of the extreme variability already noted for some of the PMS sources in the [*Einstein*]{} observations (Montmerle et al. 1983, Feigelson & DeCampli 1981). None of the 8 newly discovered x-ray sources in L1495E was sufficiently bright to enable reliable variability indications.
At least one, and possibly 2, x-ray flares occurred on PMS stars within the field during the observation period (Figure 4). DD Tau clearly flared during the second half of the observation period following a long period of very low x-ray flux. Unfortunately, owing to a gap in the observation, we are unable to determine the rise time of the flare. The emission at the peak of the flare was a factor of 15 higher than that seen prior to the flare. The e-folding time for the decay of the flare is $\sim$1 hour; however the star remained a factor of 2 above the flux level of its previous quiesent stage more than 5 hours after the flare began, at the end of the observation. It is possible that a less extreme event was seen on HDE 283572. Here the rise time is $\sim 1.4$ hours and the decay time is $\sim 6$ hours. The peak flux is $\sim 6$ times the pre-event flux. Another interpretation of the HDE 283572 observation would be that we are seeing variability associated with the rotation of an active region on the surface of the star. The rotation period of this star is $\sim 37$ hours (1.55d) (Walter et al. 1988); therefore our observation covered 0.65 periods. Lacking more data, we are unable to determine which interpretation is more likely.
The Optical and Near Infrared Photometry and Spectroscopy for the Candidate Objects
===================================================================================
We have obtained new Cousins R and I band CCD observations of the central region of the ROSAT field, as well as new near infrared (J, H, and K) observations of the x-ray sources in the same region. These data are presented in Table 3. When multiple observations have been made, the epoch is also given.
We were also able to obtain classification spectra in the red region for most of the possible x-ray source identifications (see Figure 1) with the HYDRA fiber optic fed spectrograph. Because of restrictions imposed by the placement of the optical fibers in the focal plane of the telescope, when a choice had to be made between two or more objects, the more likely identification (the brightest star nearest the x-ray position) was chosen. For this reason, no spectra were obtained of V410 x-ray 5b, V410 x-ray 8b and c. Although fibers were placed on V410 x-ray 2 and 4, no usable spectra were obtained owing to the faintness of the stars. No spectrum was obtained for DD Tau on this run. The spectral type for this star was taken from Hartigan, Strom & Strom (1994). The spectral type for CZ Tau is a compromise since there is a definite change of spectral type with wavelength of at least 2 spectral subclasses, with later spectral types being found at longer wavelengths. This does not appear to be due to a variable veiling contribution, but is perhaps due to contamination of the spectrum with light from the subarcsecond companion found by Leinert et al. (1992).
Observations of the Praesepe main sequence stars were used as spectral standards with which to classify our spectra of the x-ray sources in L1495E. The spectral types obtained are also given in Table 3; the spectral types should be good to 1 spectral subclass earlier that M2 and $\,$ a spectral subclass for stars later than M2. While the resolution of the spectra are too low to enable measurement of the strength of the Li $\lambda 6708$[Å]{} line, it was possible to measure the equivalent width of the H$\alpha$ emission line, thus providing another criterion for including the star as a candidate member of the PMS population. These values are presented in Table 3 as well.
Anonymous Stars
---------------
Because there were 97 fibers available for placement upon stars or sky positions when using the HYDRA spectrograph, we chose to place some of these fibers upon anonymous stars that fell within the area of the L1495E cloud. We obtained usable spectra for 5 of these stars, optical photometry for these 5, and, coincidentally, near infrared photometry for 4 of the stars. This data is also given in Table 3.
Other anonymous stars appeared in our infrared imaging frames. Data for these stars is also given in the Appendix.
Stellar Properties
==================
On the basis of the data tabulated in Table 3, we can assess the likelihood that we have properly identified the stellar x-ray sources and compare the distribution of the spectral types of the newly identified stars with that of the previously identified population. We must first evaluate the extinction suffered by each star. The spectral type and observed R-I color give a color excess for each star, E(R-I), which, when combined with a standard reddening law, yields the extinction. These values are also tabulated in Table 3.
Among the candidate objects for identification with the x-ray sources, only the spectrum for V410 x-ray 8a clearly disqualifies it as a member of the cloud population. The spectral type of V410 x-ray 8a is M0-2 III. This spectral type combined with the available photometry indicates substantial extinction. Since the object lies near the cloud edge, it is likely that we are viewing the star through a substantial path length of the L1495 cloud; the luminosity classification of this spectrum clearly identifies the star as a background giant. The positional discrepancy with the x-ray source location is also much too large to be acceptable. Therefore the x-ray source is most likely associated with the H$\alpha$ emission object x-ray 8e, although we have no spectrum for x-ray 8d which lies marginally closer to the x-ray position. The spectrum of V410 x-ray 8e indicates that this star is heavily veiled with a spectral type near M4 or possibly later. Due to the veiling and the approximate spectral type, the reddening is poorly determined. The H$\alpha$ emission is very broad. The spectra of all of the other x-ray candidate objects exhibit H$\alpha$ in emission and M type spectra.
Among the anonymous stars for which we have obtained spectra, there are 4 M type stars. One of these stars, V410 anon 13, clearly shows significant extinction, indicating that it cannot be a foreground star. This star has an H$\alpha$ emission equivalent width of 22[Å]{} indicating that it is most likely a member of the cloud population of PMS stars. The other 3 M stars show little or no extinction ($A_V \leq 0.10$) and little, if any, H$\alpha$ emission ($W(H\alpha) \leq 2\AA$); therefore they are likely to be foreground M dwarfs. The fifth star is an A2 star, showing no H$\alpha$ emission, and a large extinction, 10.45 mag. at V. The strength of the triplet at 7774Å is consistent with the spectral type found from the H$\alpha$ line profile and a luminosity class of V-III. The H$\alpha$ profile indicates a main sequence star. At a galactic latitude of $-$15.5 and a galactic longitude of 168.7, only 0.1 A III stars /$\sq\deg$ are expected within the volume formed by the cloud size and the distance to the star were it a giant. This star is thus most likely a main sequence A star at or close to the cloud distance.
We can now examine the distribution of spectral types for the previously identified cloud population and compare it with that found for the newly identified population. This is shown in Figure 5. Clearly the [*newly discovered cloud members include a much larger number of later type stars*]{}, indicating that previous studies greatly undersampled the lowest mass PMS stars. We can also examine the distribution of reddening values within the 2 groups as shown in Figure 6. The reddening values in the previously identified population are heavily weighted toward low values with a median at 1.25 mag. The median value for the newly identified population is 3.80 mag. indicating that a major reason for the absence of these stars from previous surveys is the relatively large extinction associated with these intrinsically faint objects.
The Infrared Color-Color Diagram
--------------------------------
In Figure 7 we present the near infrared color-color diagram for the stars in Table 3. The solid lines show the locus of the main sequence and of the giant branch. The dashed lines show the bounds of the region within which reddened main sequence stars may fall. Stars lying to the right of this region show definite evidence for excess emission believed to arise from circumstellar accretion disks. Stars lying between these lines may still possess accretion disks. However it is impossible to distinguish such stars from reddened main sequence stars or reddened WTTS in this diagram. Also shown on the lower and left axes of this figure are the marginal distributions of all of the data for these sources. This allows us to also indicate the locations for objects too red to have measurements at all 3 wavelengths and allows quick evalution of the distribution of points in each color (Tufte 1983).
We may also obtain a measure of the extinction suffered by each star from this diagram. This is useful, both as a consistency check with the optical and x-ray data, and to provide extinction estimates for objects for which we have no other extinction measure. Weak line T Tauri stars successfully deredden back to the main sequence in this diagram, using the extinction derived from the optical region photometry. However CTTS define a different sequence when the reddening is derived from their optical colors in the red region and high signal to noise spectra are used to assign spectral types (Strom, Strom & Merrill 1993). Those stars which we are able to clearly assign to one of these 2 classes we deredden to the appropriate sequence. We derive extinctions for the remaining stars, assuming that all of these stars are WTTSs. This procedure overestimates the extinctions for those stars which are actually CTTS owing to the fact that such stars have intrinsic excess emission arising in circumstellar accretion disks. For most stars, the extinctions determined by each of these methods are in reasonable agreement. Extinctions derived in this way are used only for the 4 objects in Table 3 lacking optical photometry and/or spectral types. The derived values are given in Table 4.
We can see from this diagram that most of the candidates for identification with the x-ray sources are clearly PMS stars exhibiting the excess infrared emission characteristic of circumstellar accretion disks. The three apparently most heavily reddened objects are two of the x-ray discovered sources and IRAS PSC04154+2823. The only candidate for identification with V410 x-ray 8 which clearly exhibits disk emission is x-ray 8e; it is the object which also has the smallest positional discrepancy and which exhibits broad H$\alpha \;$ emission. The infrared colors of V410 anon 13 also are in agreement with assignment to cloud membership based on its observed infrared excess. The other reddened serendipitous source, V410 anon 9, is also a possible cloud member. This star has a spectral type of A2 and an inferred extinction of 10.45 magnitudes at V. Kim (1990) detected this star in his near infrared survey and placed it in his definite PMS group from its location in the infrared color-color diagram. While our colors are little different from his ($\Delta (J-H) = -0.16$; $\Delta (H-K) = 0.03$), we feel that this star can deredden naturally to the appropriate colors for its spectral type, given the errors in the photometry, although excess infrared emission cannot be ruled out. There is no reflection nebula associated with this star and there is no emission at H$\alpha $. Therefore it is not certain whether this star is associated with L1495E.
The HR Diagram
--------------
One of the major goals of the study of a young star formation region is to place the stars in the HR Diagram to estimate the masses of the stars and their approximate ages in order to determine the rate of star formation versus time in the region. With mass estimates for the stars, we can calculate the star formation efficiency of the molecular cloud and compare the local IMF to that observed in the field. The data in Table 3 enable us to locate the population of L1495E in the HR Diagram.
From our knowledge of the components of the spectral energy distributions of T Tauri stars (Bertout, Basri, and Bouvier 1988, Hartigan et al.1992), we know that the maximum contribution of the stellar photosphere to the spectral energy distribution is made at wavelengths near $1\mu$m. Flux variations due to rotational modulation are also small at these wavelengths (Vrba et al. 1986, Bouvier & Bertout 1989). Therefore we will choose to use the flux at I or J as our measure of the stellar photospheric flux. When data at J are available, J is preferred. We must still estimate the amount of extinction for each star, although we have minimized the importance of extinction corrections by working this far to the red. The spectral type and R$-$I color yield a color excess for each star, E(R$-$I) = 1.4 $A_J$, which, when combined with a standard reddening law, becomes an extinction at the chosen wavelength, I or J ($A_J$ = 0.46 $\times
A_I$ = 0.28 $\times A_V$). Using the bolometric corrections of Bessell & Brett (1988), we find that the bolometric correction at I varies rapidly as a function of spectral type beyond M0, while the variation at J is slow and monotonic over the entire spectral range of interest. Therefore we will use the J magnitude, whenever possible, (corrected for extinction, distance (m-M = 5.88) and bolometric correction) as our measure of the stellar luminosity. Using the temperature-spectral type calibration of Hartigan et al. (1994) compiled from the work of Bessel & Brett (1988) and Schmidt-Kaler (1982), we derive an effective temperature for each star for which we have a spectrum. The errors in placement of stars in the HR Diagram vary with spectral type. They can be estimated by using the values given below for two spectral types. $$K5 \pm 1~\mbox{subtype} \left\{ \begin{array}{rlrlr}
+0.019 & \mbox{in $\log T_{eff}$} & -0.03 & \mbox{in $R_c - I_c$ \& } &
Observed \\
-0.020 & & +0.11 & \mbox{E($R_c - I_c$)} \\
& & & \\
-0.02 & \mbox{in $A_J$} & -0.05 & \mbox{in $BC_J$} & Derived \\
+0.08 & & +0.01 & \\
& & & \\
+0.01 & \mbox{in $\log L$ } & & & Result \\
+0.03 & & &
\end{array}
\right.$$
$$M3 \pm 1~\mbox{subtype} \left\{ \begin{array}{rlrlr}
+0.019 & \mbox{in $\log T_{eff}$} & -0.21 & \mbox{in $R_c - I_c$ \& } &
Observed \\
-0.027 & & +0.26 & \mbox{E($R_c - I_c$)} \\
& & & \\
-0.15 & \mbox{in $A_J$} & -0.08 & \mbox{in $BC_J$} & Derived \\
+0.19 & & +0.08 & \\
& & & \\
+0.03 & \mbox{in $\log L$ } & & & Result \\
-0.04 & & &
\end{array}
\right.$$
However, the sometimes large light variations of PMS stars prove to be the largest source of random error in the placement of these stars in the HR diagram (Hartigan et al. 1994). This can be seen immediately in Table 3 where where our 2 observations of V410 x-ray 1 differ by more than a magnitude at J and of V410 x-ray 3 differ by more than 0.5 magnitude. There are multiple sources for the variability exhibited by PMS stars. Rotational periods have been determined for many PMS stars by observation of the spot modulation of their light curves. The amplitude of this variation may be as small as a few hundreths of a magnitude or as large as 1 magnitude (at V) (AA Tau, Vrba et al. 1988). These stars are also known to have an irregular component to their variability with very large amplitudes sometimes exceeding several magnitudes (Herbig and Bell 1988). These irregular fluctuations are presumed to be due to variations in the accretion rate through the disks of these stars (Kenyon et al. 1994). In the extreme case, an FU Ori event may cause a rise of more than 5 magnitudes over a relatively short period of time (weeks) with a much longer decay time (years) (Hartmann, Kenyon and Hartigan 1993). Hartigan et al. (1994) found that the error in placment of PMS stars in the HR diagram was dominated by the variability of these stars. They estimated that variability was responsible for $\sigma \simeq 0.05$ in $\log L$ although there is a long tail at higher amplitudes to the distribution. When all of the possible sources of error are combined in quadature, the result is a “typical” $1\sigma$ uncertainty of 0.08 in $\log L$.
The data placing each of our stars in the HR Diagram is given in Table 4 and they are plotted in Figure 5 with the $10^8$ yr isochrone from the D’Antona & Mazzitelli(1993) evolution tracks. The objects V410 x-ray 2, 4, PSC04154+2823 and PSC04158+2805 are placed in the HR Diagram using the (uncertain) extinctions estimated from their near infrared colors, a bolometric correction and effective temperature assuming they are middle M stars similar to the other objects found in this sample. For V410 anon 3 & 17, for which there was no near infrared photometry, we used I as the standard wavelength. The final luminosities for the 5 objects known to be binaries are then corrected for the light contributed by any secondary star discovered by high spatial resolution observations (see section 5.3). These final values are given in the column L(J$_c$).
It is obvious from Figure 8 that the stellar population of L1495E is very young. However, it is difficult to evaluate exactly how young and how coeval this population is because different sets of “modern” evolutionary tracks (D’Antona & Mazzitelli 1993, Swenson et al. 1993) show sufficient differences to make interpretation difficult. Also, the isochrones are in intervals of $\log$ Age; therefore the stars are evolving very rapidly in the upper part of the HR diagram and more slowly as they age. In Figure 9 we show the data for these stars plotted on a) the D’Antona & Mazzitelli tracks using the Alexander opacities and the Canuto and Mazzitelli convection formulation; and b) the Swenson et al. tracks. It is clear that, for stars this young ($\leq 10^6$ yrs) the tracks differ sufficiently to influence our estimate of coevality as well as the “absolute” age. On the D’Antona & Mazzitelli tracks, for stars with masses less than 0.5 $M_\odot$, the population appears approximately coeval with an age of $t \sim 5 \times 10^5\,$yrs. Using a program written by Patrick Hartigan which interpolates between the isochrones to assign an age to each star, we find that 20 of the 22 low mass PMS stars, excluding those viewed by scattered light (see later in this section, and the anonymous stars, have ages $\leq 7.1 \times 10^5$ years. On the Swenson et al.tracks, for the same mass range, the apparent age increases as the mass decreases and the mean age for the L1495E population is $10^6\,$yrs. Given the uncertainties in the model isochrones and the fact that the models are still sensitive to initial conditions at these early ages, this group can be considered coeval to within $\leq 1$Myr. The two apparently older stars merit discussion. We have only one complete set of colors for FQ Tau. However, at the time of the Leinert et al. (1992) speckle observations, FQ Tau was more than 0.5 mag. brighter at K than when we observed the star. If this brightness change is also reflected at J, this star is then $< 1\sigma$ above the mean age for the group. The remaining star lies 3.75$\sigma$ above the mean age of the group; this is not unreasonable for a sample this size.
One possible problem apparent from both sets of tracks is the gradually increasing age found for stars with masses $\geq 1.5 M_{\odot}$. While RY Tau and HDE 283572 lie off the face of the cloud, V892 Tau (Elias 1) is heavily embedded in the cloud core and illuminates a reflection nebula. It is difficult to believe, given the apparent velocity dispersion of L1495E of $\sim 3.0$ km/s, or 3.0 pc$/10^6$ yr,that this star is over $3 \times 10^6$ yrs old as the tracks imply. A star formed in the center of the cloud would have moved more than $200\arcmin$ away from the cloud center in $3 \times 10^6$ yrs. The motion of this star would have to be aligned within 10$\,$ of the line of sight and be moving away from us in order to remain located in cloud center and still exhibit an extinction of 9 mag. The only alternative (other than suspecting a systematic error in the tracks) is that the age of this star is correct and that star formation actually began $3 \times 10 ^6\,$yr ago. In this case the timescale for “building” a star via infall and accretion, before it appears on these tracks, would have to be much longer for low mass stars.
From examination of Figures 8 and 9 and Table 4, several objects deserve special mention. There are two members of the previously identified PMS population that fall well below the main body of the distribution. The images of these objects, CK Tau 1 and PSC 04158+2805, appear definitely extended on our deep R and I frames, the PSC object appearing triangular and the object near Hubble 4 appearing flamelike (Figure 10). We suspect that we are not seeing these objects directly but instead via light scattered through a circumstellar envelope or from nearby molecular cloud material. Recent submillimeter observations of low luminosity embedded sources in the Taurus region (Barsony & Kenyon 1992) indicate very large extinctions for these sources, (50 $< A_V <$ 600 ), sufficient to render them undetectable even at $2\mu m$. However these objects are nevertheless visible in the near infrared and sometimes at even shorter wavelengths. This paradox can be resolved if the density distribution surrounding these objects is highly asymmetric. Models of sources such as these using asymmetric density distributions have been developed recently by Kenyon, Calvet & Hartmann (1993) and Kenyon et al.(1993). These authors find it necessary to introduce a bipolar hole in the density distribution in order to reproduce both the observed spectral energy distributions and the extended images of such sources. Both the spectral energy distributions and direct images of the objects shown in Figure 10 are very similar to those represented by these models. The (R$-$I) colors of these objects indicate 5-10 magnitudes of extinction. However, the models of Kenyon et al. (1993), while only calculated for the J,H, & K bands, suggest that the actual extinction that should be applied to the observed magnitudes could be a factor of 3 or more higher than that inferred from the observed near infrared colors. The discrepancy should be even greater for extinctions determined from the R-I color. The visible and near infrared light is due to scattering and not transmission, and therefore the anomalous location of these stars in the HR diagram can easily be accounted for by an underestimate of the extinction.
Our spectral type for CK Tau 1 is uncertain as the spectrum appears heavily veiled and contains strong emission lines of H, \[ \], \[ \], \[ \] among others. This contributes a major source of error in the reddening estimate as we are probably not measuring photospheric colors. Our previously published images of CK Tau 1 in H$\alpha $ (Strom et al. 1986) indicated a faint bipolar jet emanating from this star. The recent narrow band imaging by Goodrich (1993) confirms this result, showing a bipolar jet clearly in the \[S II\] images.
Among the x-ray discovered sources in this cloud, it is clear that V410 x-ray 1, 2, 3, 4, 5a, 6, 7, and 8e are member of the PMS population of L1495E. In most cases they are the only possible optical object which could be identified with the x-ray source. The near infrared colors of these objects indicate excess infrared emission. When we have spectra for these objects, H$\alpha $ appears in emission ($W(H\alpha ) \geq 8\AA $). As discussed previously, x-ray 8a is a background M giant seen through the edge of the cloud. We can make no statement about the possible association of V410 x-ray 8b, 8c, 8d, 8f and 5b since we have no spectra for those objects.
Of the serendipitously observed stars, V410 anon 3, 12 & 17 show little or no extinction or H$\alpha $ emission. Their derived luminosities (at the distance of the Taurus clouds) place them above the main sequence. However they are more likely [*main*]{} sequence stars in front of the cloud. The galactic model (Bahcall and Soneria 1980) predicts $\sim$ 10 M stars between us and the Taurus clouds projected on the central region of the cloud ($r \leq 25
\arcmin$). The other two serendipitously observed stars may well be members of the cloud population. V410 anon 13 has an H$\alpha $ emission equivalent width of $22\AA \:$ and an inferred extinction of 2.75 magnitudes at V. This star is clearly a strong candidate for inclusion in this young group. The lack of detection in the ROSAT image is probably due to its low luminosity, below that of any of the x-ray detected objects. The star V410 anon 9, discussed previously, appears to be a normal A2 star with no H$\alpha$ emission. When dereddened and placed at the distance of the Taurus clouds, the star lies just above the main sequence, not far below V892 Tau. It is possible that this star lies behind the cloud, but it cannot lie far behind because the probability of finding an A giant this far above the galactic plane is small.
Binaries in L1495E
------------------
While we have approximately doubled the previously known population within the central region of L1495E, all with M stars, it is still possible that even more low mass stars may be hidden as secondaries in binary systems. Fortunately there has been much recent high spatial resolution work which allows us to evaluate this possibility (Ghez 1993, Simon et al.1992, Leinert et al. 1992, Simon 1992, Ghez, Neugebauer & Matthews 1992).
Of the previously identified PMS sample (19 stars), 7 are known to be binaries (V410 Tau, DD Tau, CZ Tau, V892 Tau, FO Tau, FQ Tau, and LkCa 7). Eight stars were found to be single at the separations, flux ratios (and sometimes position angles) to which the measurements were sensitive. Four stars (CK Tau 1, PSC04154+2823, PSC04158+2805 and LkCa 5) were not examined. V892 Tau was discovered to be a binary by Thornley et al. (1989) using near infrared imaging, The secondary was later confirmed by Skinner et al. (1993) in their VLA observations of this region. They detected both V892 Tau and its close companion to the NE as well as CK Tau 1 and Hubble 4 (Skinner 1993).
While a few B and A stars in young stellar groups have been identified as possible x-ray sources (Caillault & Zoonematkermani 1989, Strom et al. 1990, Pravdo & Angelini 1993), there is always the possibility that the x-ray emission is attributible to a low mass companion (Schmitt et al. 1985). In this case, it is reasonable to attribute the x-ray luminosity of V892 Tau to the low mass secondary of the system. The derived x-ray luminosity is consistent with the inferred luminosity of the secondary if we assume that the luminosity ratio of the two components of V892 Tau is close to the ratio of their infrared luminosities, the star lies on the mean isochrone for this group, and we use the bolometric correction inferred from this placement.
Flux ratios are in general available at only one wavelength for these binaries, usually K (2.2$\mu$m). For 5 of the 7 known binaries, the flux ratios at K lie between 1.1 and 2.2. The other 2 stars (V892 Tau and V410 Tau) have relatively much fainter companions. If we assume that this ratio represents the true luminosity ratio of the components, the extinction for both components is the same, and that they lie on the $10^6$ yr isochrone, we can estimate the masses of these stars. Therefore these 7 companions can also be added to the low mass ($M < 0.3 M_{\odot}$) population of the cloud.
The Cloud IMF
-------------
Among the basic problems to be addressed by studies of star formation are (1) to derive the initial mass function (IMF) for isolated star formation regions and compare with that derived from study of stars in the solar neighborhood; and (2) the local efficiency of star formation in the star forming cores. Implicit in the determination of the IMF is the question of whether stars of mass lower than that in which hydrogen burning can be sustained (brown dwarfs) are plentiful and may therefore contribute to solving the “missing mass” problem.
It is now possible to estimate the star formation efficiency of the L1495E cloud. The mass of the cloud has been estimated at between 180 and 230 $M_\odot$ (Ohnishi & Mizuno 1993). The total stellar mass in the central region of the cloud depends slightly on the set of tracks used to make the mass estimate. If the Swenson et al.(1993) tracks are used, the total stellar mass is estimated at $\sim 13
M_\odot$; if the D’Antona & Mazzitelli (1993) tracks are used, the total stellar mass is estimated at $\sim 10.5 M_\odot$. This leads to star formation efficiencies between 5% and 7%. If the stars in the outer region of the cloud are included, the star formation efficiencies increase to between 8% and 9.3%. Current constraints on the formation of bound clusters (Lada, Margulis & Dearborn 1984) require that at least 30% of the cloud mass must be converted to stars. Unless a large number of low mass members of this group remain hidden within the cloud, at luminosities below the limit reached by this ROSAT exposure, it is unlikely that this group will remain bound.
One of the first studies to examine the distribution of spectral types in nearby star forming regions was carried out by Cohen & Kuhi (1979). While a small number of stars later than spectral type M0 were found in Taurus, those authors assumed that the flattening of the mass spectrum at low masses apparent in their data, was a selection effect due to a magnitude cutoff. Since that work, there have been several surveys aimed at increasing the known PMS population of the Taurus clouds: the proper motion surveys of Jones & Herbig (1979), Hartmann et al. (1991), and Gomez et al. (1992); the H,K emission survey of Herbig et al. (1986); the x-ray sources from [*Einstein*]{} pointed observations (Walter et al. 1988); and the deep H$\alpha$ emission survey of Briceño et al. (1993). The recent proper motion surveys, (Hartmann et al. 1991, Gomez et al.1992) have found a small number (7) of additional members of the Taurus PMS population, usually late type stars (SpT later than M3) with relatively weak H$\alpha$ emission. Most recently, Briceño et al. (1993) have used very deep (V $\approx 18$) objective prism H$\alpha$ emission surveys to search for as yet undetected pre-main sequence stars in the Taurus clouds. They identified 12 new T Tauri stars, 5 of which were found in L1544, a region previously unsurveyed in H$\alpha$ and containing only one identified PMS star. Most of the newly discovered PMS stars are classical or strong emission T Tauri stars (CTTS). The PMS evolutionary tracks used until this point (Cohen & Kuhi 1979, Strom et al. 1989b) have identified the peak of the N(SpT) distribution in the Taurus clouds (at K7-M0) with stars of $\sim 1 M_\odot$ with the number of stars decreasing with mass until few stars are found below 0.4 $M_\odot$. This result is in marked disagreement with the IMF derived for the solar neighborhood. The IMF found for the local solar neighborhood (Scalo 1986, Kroupa, Tout & Gilmore 1990, Tinney, Mould & Reid 1992) can be well fit by a log normal function peaking at $\log M =
-0.5$ and then either decreasing as the mass decreases or possibly staying constant for all lower masses. The PMS stars identified in the above surveys have spectral types with either M or continuum + emission characteristics. Their R magnitudes range from 14.2 to 18.5. Even though most of these newly discovered stars are M stars, the apparent deficiency of low mass stars remained.
One possibility for hiding additional low mass stars is as secondaries in binary and multiple systems. Recent studies (Ghez 1993, Simon et al. 1992, Leinert et al. 1992) have suggested that most stars form in multiple systems. This possibility must also be evaluated in order to fully account for the IMF in star forming regions.
It is also important to note that the masses for these stars obtained from the “modern” tracks are quite different from those obtained in the past. Here stars that were recently considered as typical 1 $M_{\odot}$ PMS stars are now found to be 0.3-0.5$M_{\odot}$ stars. While this may seem a small change, these typical TTS will not be G stars when they arrive on the main sequence but early M stars. This change may have a significant effect on the analyses of disk stability and binary period distributions (Ghez 1993, Leinert et al. 1992) since derived ratios of disk mass to stellar mass will increase and the total system masses for binaries will decrease.
As a consequence of (1) the newly discovered low mass members of the L1495E cloud population (x-ray detections, 8 stars; anonymous stars, 1 star; low mass companions, 7 known), and (2) masses derived from new tracks, the apparent deficit in the population of the low mass end of the IMF in this part of the Taurus clouds will be decreased considerably (see below).
The combination of newly observed and already known PMS stars in L1495E enables a modern discussion of the IMF. Since we have a deep survey only in the central region of the ROSAT field, we will use only the objects (known and newly discovered) that fall within that area. We begin with the 12 previously known members of the cloud population. To this we add 8 x-ray discovered PMS stars, 1 certain PMS star from the group of anonymous optically visible stars projected upon the cloud, 1 possible addition to the intermediate mass star population associated with the cloud, and, finally, the 5 known secondary stars to the cloud members within the central region that have been observed at high spatial resolution. This gives a total of 27 probable cloud members within a circle of radius 25$,$ centered on V410 Tau.
In order to construct an IMF for the PMS population of this cloud, we must infer a mass for each star from its position in the HR Diagram relative to the calculated evolution tracks. Here we must deal with the problem that the mass deduced for each star differs depending on the tracks we choose to use. This is particularly true at the very low mass end of the range in which we are interested, where the two sets of tracks are most divergent. We construct an IMF for each set of tracks in order to better compare the derived shape of the cloud IMF. The results are presented in figure 11. The results from both sets of tracks yield IMFs that peak at $\log M = -$0.5, identical to the result found by Scalo(1986) from solar neighborhood stars. There are too few stars of masses greater than $\sim1 M_{\odot}$ to allow us to draw any conclusions about the higher mass end of the IMF in L1495E, and the behavior of the derived IMF for masses below the peak is dependent upon the tracks chosen. While the IMF drops toward lower masses, the drop is much more severe for the D’Antona & Mazzitelli tracks. If the Swenson et al.tracks are used, 10 stars are found in the 2 bins of lowest mass ($ -1.2 \leq \log M \leq$ -0.8), while if the D’Antona & Mazzitelli tracks are chosen, only 2 stars are found in these bins. For this reason, and because (1) less than half of the cloud members have been searched for companions of subarcsecond separation; and (2) there is no knowledge of binary systems with separation below the sensitivity of current instrumentation, we are unable to make a definitive statement about the shape of the cloud IMF at these low masses. However, there is [*no evidence for a rise toward lower masses*]{} if the PMS population we have identified is either complete or representative.
We have attempted to estimate the number of possible additional members of the cloud population from our near infrared images. In addition to deriving the JHK magnitudes for the candidate x-ray sources, we also obtained near infrared magnitudes for all other objects appearing within these frames. Most of these objects, in the frames away from the cloud edge, are quite faint. Consequently, the photometric errors are larger and their colors are less definitive than we would like for the purpose of identifying possible cloud members. However, there are clearly some stars (8) whose presence in the central area of the cloud and whose colors (indicating the presence of disk emission) suggest that they are members of the cloud population. One of these stars, not visible on our long I image, just 40$\;$ south-southeast of V410 Tau, is brighter than $K = 10$ and extremely red. Some of the stars in the region near V892 Tau are so red that no J magnitudes can be obtained, however they cluster about the tight group of stars found in this region, suggesting membership. A table giving the approximate coordinates of these stars and their photometry can be found in the Appendix.
We must emphasize again the need for very deep x-ray searches of star formation regions in order to locate the low mass members of the PMS population. The lack of other comparably deep observations of star forming regions with ROSAT while the PSPC was operative is extremely unfortunate.
Other nearby star formation regions have not shown a deficiency of M stars comparable to that seen in Taurus-Auriga. Hughes and Hartigan (1992) compare the spectral type distribution in the Chamaeleon II cloud with that seen in Taurus. They find a higher fraction of stars with late spectral types in Cha II than are seen in Taurus. Krautter & Kelemen (1987) show the distribution of spectral types for the Lupus star formation region, at a comparable distance, which has only 2 stars earlier than K5 while 47 stars later than M0 are found. The new PMS evolution tracks can explain the difference in the spectral type distribution between the Taurus and the Lupus clouds. The mean age of the stars in the Lupus clouds is considerably older than that of the stars in the Taurus clouds (Hughes & Hartigan, private communication). Therefore, using the “modern” tracks, as shown in Figure 9, the mean spectral type of the population will be considerably later in Lupus than in Taurus for the same distribution of stellar masses since these tracks take a considerable turn to lower effective temperature after $\sim 1 \times
10^6$ yrs.
Correlations of X-ray Properties with the Stellar Parameters
============================================================
Because the x-ray emission from PMS stars appears to be consistent with coronal emission models, the strength of the x-ray emission should be correlated with the strength of the dynamo generating the stellar magnetic field. In turn, we expect that a measure of the rotation of the star will be the proper surrogate for the dynamo strength; and therefore that stellar rotation should correlate with the x-ray emission. However, one complication in this chain of logic is the fact that these very young stars are completely convective, and the currently favored dynamo models operate at the interface between the convective and the radiative zone, driven by differential rotation. However there is a distributed dynamo model (Rosner 1980) that works throughout the convection zone.
In this section we will evaluate the simplest correlations that we can make, the correlation of the x-ray luminosity with the stellar luminosity and with the rotational velocity. In these correlations, we will omit the data for V892 Tau since it is likely that the x-ray emission is due to the fainter component of the system for which we are unable to establish a reliable luminosity estimate since we have no spectral type.
Our x-ray data consist solely of measured values with no upper limits, therefore we have no need to use survival statistics methods in our analysis. The x-ray luminosity and the stellar luminosity are very strongly correlated, with a formal probability of $<$0.001 that they are uncorrelated. The slope of the relationship between $\log L_x \;$ and $\log L_\star \;$ found from a least squares solution is $1.80 \pm 0.01$. However, since the values of $\log L_x \;$ are instantaneous snapshots of highly variable sources, the accuracy of the slope determination is no doubt worse than that quoted. This relationship is shown in Figure 12. The WTTS, CTTS and stars that are no longer fully convective are indicated. While there is considerable overlap between the CTTS and WTTS, the WTTS are clearly systematically more luminous than the CTTS. While it is true that some of the WTTS are found at earlier spectral types than are the CTTS and therefore at higher stellar luminosities, it also appears to be true that the WTTS are more luminous at x-ray wavelengths than the CTTS when compared at constant stellar luminosity. DD Tau is plotted twice in this figure, at its pre-flare and flare locations. The pre-flare location falls just below the center of the band, in the midst of a group of other points, while the flare point lies well above the majority of the points. It is possible that the other points lying well above the major part of the distribution represent stars that were not in their quiescent state at the time of the observation. It is also interesting to note that the x-ray luminosities of the least luminous of these PMS stars have x-ray luminosities comparable to those of the young near main sequence K and M stars of the Pleiades (Stauffer et al. 1993). In fact, because our detection limit is lower than that of the Pleiades observations, some are even less luminous than the upper limits found for the undetected Pleiades members. This must be taken into consideration when testing dynamo models and angular momentum evolution scenarios.
We can compare this distribution with that found for the PMS members of the Cha I cloud (Feigelson et al. 1993). These authors find a correlation with a slope of $1.00 \pm 0.15$ for the relationship between the stellar luminosity and the x-ray luminosity for the Cha I stars. For the stellar luminosities they used the values found by Gauvin & Strom (1992) adjusted for the closer distance that they chose to use. Since in this relationship the distance enters in the same manner on both axes, it does not matter which distance is chosen. However, the method by which the stellar luminosity is calculated has now been improved. In order to remove any systematic effects introduced in this way, we have recalculated the stellar luminosities for the Cha I stars using the J magnitude as our measure of the stellar photosphere as discussed in section 6.2 and using the same data compiled in Gauvin and Strom (1992). The x-ray luminosities are taken from Feigelson et al. (1993). In Figure 13 we have plotted the data for Cha I with the relationship derived for the stars in L1495E and shown in Figure 12. We can clearly see that the Cha I CTTS, fainter WTTS and radiative track stars exhibit the same slope to this relationship as found for the stars in L1495E, however they have systematically lower x-ray luminosities at constant stellar luminosity. The stars deviating from this relationship are the most luminous stars in the cloud (excluding the 2 A stars not considered here). Most of these stars are no longer fully convective; the most luminous WTTS lies on the boundary of the fully convective region. The only real anomalies are the 2 CTTS (VW Cha and WW Cha). One of these, VW Cha, is a binary with a separation of 17$\;$(Reipurth and Zinnecker 1993, Schwartz 1977), both components of which must lie within the x-ray beam. Reexamination of the spectra of these stars (Appenzeller, Jankovics & Krautter 1983) shows that these stars would better be classified as continuum + emission stars leading us to overestimate the stellar luminosity for 2 reasons, 1) the flux at J is contaminated by disk emission, and 2) the assigned early K spectral types are most likely too early leading us to use a bolometric correction too small. There is no simple way to make a correction to the luminosities assigned to these stars, short of obtaining high resolution spectra. Therefore we merely indicate that the stellar luminosities for these stars are upper limits. The relative absence of WTTS in this region is no doubt due to the fact that the stars with adequate photometric and spectroscopic data are those which were discovered by H$\alpha$ emission surveys. Many more cloud members have been discovered in the x-ray emission surveys and in more sensitive H$\alpha$ emission surveys (Hartigan 1993), but as yet lack the information necessary to place them in the HR Diagram.
An interesting aspect of this diagram is that the x-ray luminosity of stars which are now on their radiative tracks appears to be decreasing faster with the age of the star than do the x-ray luminosities of stars still fully convective. This could reflect the fact that the surface magnetic field strength is relatively weaker for these stars and therefore that the dynamo generation mechanism is not as efficient as that in the fully convective stars. A possible correlated effect has been seen by Edwards et al. (1993) in the angular momentum statistics of these stars. While the circumstellar disk appears able to regulate the rotation period of the fully convective stars quite efficiently through the linking of the stellar magnetic field with the disk, this process appears to be less efficient in stars that have radiative cores.
Another way to look at the correlation of rotation with coronal heating is to use the relationship first proposed by Bouvier (1990) between x-ray surface flux and rotational period of the star. We are fortunate to have 6 rotational periods determined for the stars in L1495E (Rydgren et al. 1984, Bouvier et al. 1986, Vrba et al. 1986, Walter et al. 1987, Vrba et al. 1988, Bouvier et al. 1993), and also v$\sin i$ data (Hartmann et al. 1986, Hartmann & Stauffer 1989) for this group. There are many fewer rotational periods determined in Cha I (Bouvier & Bertout 1988). In order to make use of a larger database, we will translate the rotation periods into projected equatorial velocities through knowledge of the stellar radius. In Figure 14a we show the relationship between $\log F_x$ and rotational velocity. Those velocities that are derived from rotation periods are shown by filled circles while those known only from measurement of stellar line widths are marked with open circles. In Figure 14b the same relationship is shown for the Cha I cloud stars; the v$\sin i$ values are taken from Franchini et al. (1988) and Walter (1992). The two stars marked with limiting values in both rotational velocity and surface flux are VS Cha and WW Cha, mentioned earlier. These stars are very heavily veiled and therefore the luminosity and thus the radii are probably over estimated, leading to an underestimate in x-ray surface flux. Franchini et al. (1988) note that the rotational velocities that they have determined would be over estimated if a substantial veiling flux is present. In Figure 14a the data for the Pleiades stars is shown (Stauffer et al. 1993). The Pleiades data can be described as filling a wide band in log $F_x$ at rotational velocities $>20$ km/s. Below that velocity much lower values of log $F_x$ can be found. It is difficult to look for a functional relationship in this data since values of v$\sin i$ give only lower limits to $V_{eq}$. However, it is clear that the slowly rotating Pleiades stars exhibit much lower x-ray surface fluxes (Stauffer et al. 1993). The L1495E x-ray data is consistent with the Pleiades data. However, because of the small number of stars in this group having measured v$\sin i$’s, only 1 slow rotator, no independent assessment of this behavior can be made for PMS stars from this data. Therefore we cannot separate age and rotation effects.
Because the derivation of other stellar parameters, in particular the stellar mass, are dependent upon the set of stellar evolutionary tracks used, we prefer to wait until the new tracks have been further evaluated until attempting further correlations with derived stellar parameters. It will also be useful to have a larger body of x-ray data derived from the ROSAT pointed observations for stellar populations of a range of ages. We will then be able to separate the effects of age and rotation for stars in a small mass range from first appearance to the main sequence.
Conclusions
===========
The L1495E cloud is a region of very active star formation. Our multiwavelength deep survey of this region has considerably increased our knowledge of the PMS population within the cloud. The ROSAT PSPC observation allowed us to identify 8 new low mass ($0.08 M_{\odot} \leq M \leq 0.6 M_{\odot}$) members of the cloud population. Serendipitous observations of otherwise undistinguished stars has added 1 (or possibly 2) more members. Nearly all of the new members of the cloud population are of spectral type M3 and later and thus represent a heretofore unrepresented fraction of the cloud population. This population appears to be nearly coeval; that is the spread in the HR Diagram is small at any given mass. The mean age lies between $5 \times
10^5$ and $1 \times 10^6$ yr depending upon the adopted tracks.
These observations combined with the use of “modern” evolutionary tracks allow us to address an outstanding problem concerning the PMS population in the Taurus molecular clouds, the apparent deficiency of stars with masses below $\sim 0.40
M_{\odot}$. While the addition of the newly discovered cloud members and the identification of the low mass companions of other cloud members helps us to alleviate this deficiency, the use of the new evolutionary tracks is the key factor in understanding the mass distribution in the L1495E population. The masses deduced for stars of $1 \times 10^6$ yrs and younger are considerably lower than those previously derived (Strom et al. 1989b). This allows us to understand the large difference in the effective temperature distribution, as reflected in spectral types, between the PMS population in Lupus and that in Taurus as merely a reflection of different ages of the two groups.
The near infrared observations of the newly recognized cloud members show that 6 of 8 exhibit colors characteristic of disk emission. With the spectra obtained with HYDRA, we find that, with the possible exception of the serendipitously observed A2 star, the new cloud members are low mass stars ($M < 0.6 M_\odot$) and are typically deeply embedded in the cloud. The images of 2 of the low luminosity IRAS sources are extended. These images, in combination with their spectral energy distributions, and the models of Kenyon et al. (1993), demonstrate that these objects are still surrounded by dense core material, even though one of the objects is actually found at the edge of the main cloud.
Although there are only 35 known members of this stellar aggregate, 27 of which lie within the completely sampled region, we show that the observed IMF for this group is consistent with the IMF found by Scalo (1986) for the field population, with a peak near $\log M
= -$0.5 and a falloff toward lower masses. There is no evidence for a rise as mass decreases.
The x-ray luminosity of the L1495E stars is highly correlated with the stellar luminosity with a slope to the relationship of 1.80 $\pm$ 0.01. This can be seen in the observations of the Cha I stars as well, although the higher mass stars in this group have much lower x-ray luminosities than their counterparts in L1495E, perhaps due to the greater age of the Cha I association ($\sim 2
\times
10^6$ yr on the Alexander tracks). When the x-ray surface flux of the Pleiades sample is examined as a function of the rotational velocity of the stars, it is seen that no systematic variation is found for velocities greater than 20 km/s. However, below that velocity the range of observed x-ray surface flux increases dramatically with stars of low projected rotational velocities showing the lowest surface fluxes. When the rotational velocity drops below 20 km/s, the energy source powering the x-ray emission in these stars is substantially diminished. The x-ray data for the L1495E stars is consistent with the Pleiades data. However there are too few stars in the sample having measured rotational velocities to enable us to separate age and rotation effects.
The authors would like to thank Patrick Hartigan for obtaining the R and I band CCD frames, and Ronald Probst, Michael Meyer, Patricia Knezak and Lynne Hillenbrand for each obtaining part of the near infrared frames. Without their help this paper would have greatly suffered. The authors would also like to acknowledge support from the National Science Foundation (SES), the NASA Planetary Science and Origins of Solar Systems programs, the NASA ADP program and, of course, the ROSAT observations support (KMS).
Appendix
========
In this section we give the data on the anonymous stars that were serendipitously observed with HYDRA and with SQIID. In Table 5 are given the positions and photometric data for the optically visible stars which are seen projected on the face of the cloud. Five of these stars are discussed in the text.
In table 6 are given approximate positions and near infrared photometric data for stars appearing in our SQIID frames whose colors indicate that they might be cloud members.
Appenzeller, I., Jankovics, I. & Krautter, J. 1983, , 53, 291 Bahcall, J. N. & Soneira, R. M. 1980, , 44, 73. Barsony, M. & Kenyon, S. J. 1992, , 384, L53 Beichman, C. A., Boulanger, F. & Moshir, M. 1992, , 386, 248 Bertout, C., Basri, G., & Bouvier, J. 1988, , 330, 350 Bouvier, J., 1990, , 99, 946 Bouvier, J. & Bertout, C. 1989, , 211, 99 Bouvier, J., Bertout, C., Benz, W., & Mayor, M. 1986, , 165, 110 Bouvier, J., Cabrit, S., Fernández, M., Martín, E. L., & Matthews, J. M. 1993, , 272, 176 Bessell, M. S. & Brett, J. M. 1988, , 100, 1134 Briceño, C., Calvet, N., Gomez, M., Hartmann, L., Kenyon, S., & Whitney, B., 1993, , 105, 686 Caillault, J.-P. & Zoonematkermani, S. 1989, , 338, L57 Cohen, M. & Kuhi, L. V. 1979, , 41, 743 D’Antona, F. & Mazzitelli, I. 1993, , (in press) Davis, L. 1993 in Astronomical Data Analysis Software and Systems II, ed. R. J. Hanish, R. J. V. Brissenden & J. Barnes (ASP. Conf. Ser.), 52. Duvert, G., Cernicharo, J. & Baudry, A. 1986, , 164, 349 Edwards, S. et al. 1993, , 106, 372 Feigelson, E. D.1986 in Protostars and Molecular Clouds, ed. T. Montmerle & C. Bertout(Saclay:Centre d’Etudes Nucleaire de Saclay), 123 Feigelson, E. D., Casanova, S., Montmerle, T. & Guibert, J. 1993, , (in press) Feigelson, E. D. & DeCampli, W. M. 1981, , 243, L89 Feigelson, E. D. & Kriss, G. A. 1989, , 338, 262 Franchini, M., Magazzú, A., & Stalio. R. 1988, , 189, 132; errata in 1988, . 197, 354 Fukui. Y., Mizuno, A., Nagahama, T., Imaoka, K. & Ogawa, H. 1991, Mem. S. A. It., 62, 801 Gauvin, L. S. & Strom, K. M. 1992, , 385, 217 Ghez, A. 1993, Ph.D. Thesis, Calif. Inst. of Technology Ghez, A., Neugebauer, G., & Matthews, K. 1992, in IAU Colloq. 135, Complementary Approaches to Double and Multiple Star Research, ed. H. A. McAlister & W. I. Hartkopf (ASP. Conf. Ser.), 21 Gomez, M., Jones, B. F., Hartmann, L. W., Kenyon, S. J., Stauffer, J. R., Hewett, R., & Reid, N., 1992, , 104, 762 Goodrich, R.W. 1993, , 86, 499 Hartigan, P. 1993, , 105, 1511 Hartigan, P., Kenyon, S. J., Hartmann, L. W., Strom, S. E., Edwards, S., Welty, A. D., & Stauffer, J. 1992, , 382, 617 Hartigan, P., Strom, K. M. & Strom, S. E. 1994, (in press) Hartmann, L. W., Hewett, R., Stahler, S. & Mathieu, R. D. 1986, , 309, 275 Hartmann, L. W., Jones, B. F., Stauffer, J. R., & Kenyon, S. J., 1991, , 101, 1050 Hartmann, L. W., Kenyon, S. J., & Hartigan, P. in Protostars and Planets III, edited by E. H. Levy and J. I. Lunine (University of Arizona Press, Tucson) Hartmann, L. W., Soderblom, D. R. & Stauffer, J. R. 1987, , 93, 907 Hartmann, L. W. & Stauffer, J. R. 1989, , 97, 873 Herbig, G. H. & Bell, K. R. 1988, Lick Obs. Bull. Ser., No. 1111 Herbig, G. H., Vrba, F. J., & Rydgren, A. E. 1986, , 91, 575 Hughes, J. & Hartigan, P. 1992, , 104, 680 Jones, B. F. & Herbig, G. H. 1979, , 84, 1872 Jones, B. F. & Stauffer, J. R. 1991, , 102, 1080 Kenyon, S. J., Calvet, N. & Hartmann, L. 1993, , (in press) Kenyon, S. J., Whitney, B. A., Gomez. M. & Hartmann, L. 1993, , (in press) Kim, Ch. Y. 1990, in Memoirs of the Faculty of Science, Kyoto University, Series of Physics, Astrophysics, Geophysics and Chemistry; Vol. XXXVIII, No. 1, Article 1. Kirkpatrick, J. D., Henry, T. J. & McCarthy, D. W., Jr. 1991, , 77,417 Kroupa, P., Tout, C. A. & Gilmore, G. 1990, . 244, 76 Krautter, J. & Kelemen, J. 1987, Mitt.Astron.Ges. 70, 397 Lada, C. J. 1987, in IAU Symposium 115, Star Forming Regions, ed. M. Peimbert & J. Jugaku, (Dordrecht: Reidel), p 1 Lada, C. J., Margulis, M. & Dearborn, D. 1984, , 285, 141 Landolt, A. U., 1992, , 104, 340 Leinert, Ch. et al. 1992, in IAU Colloq. 135, Complementary Approaches to Double and Multiple Star Research, ed. H. A. McAlister & W. I. Hartkopf (ASP. Conf. Ser.), 21 Marcy, G. W. 1980, , 85, 230 Montmerle, T., Koch-Miramond, L., Falgerone, E., & Grindlay, J. E. 1983, , 269, 182 Morrison, R. & McCammon, D. 1983, , 270, 119 Ohnishi, T. & Mizuno, A. 1993, private communication. Pravdo, S. H. & Angelini, L. 1993, , 407, 232 Pfefferman, E. et al. 1986, SPIE, 773, 519 Raymond, J. C. & Smith, B. W. 1977, , 35, 419 Reipurth, B. & Zinnecker, H. 1993, , (in press) Rosner, R. 1980, in Cool Stars, Stellar Systems, and the Sun, ed. A. K. Dupree (SAO Rep., No. 389), 79 Rydgren, A. E., Zak, D. S., Vrba, F. J., Chugainov, P. F., & Zajtseva, G. V. 1984, , 89, 1015 Scalo, J. 1986, Fund. Cosmic Phys., 11,1 Schmidt-Kaler, T. H. 1982, “Physical Parameters of Stars”, Landolt-Bornstein New Series, Vol. 2b, Astronomy and Astrophysics - Stars and Star Clusters, Eds. K. Shaifers and H. H. Voigt (New York: Springer-Verlag) Schmitt, J. H. M. M., Golub, L., Harnden, F. R. Jr., Maxon, C. W., Rosner, R. and Vaiana, G. S. 1985, , 290, 307 Schrijver, C. J., Lemen, J. R. & Mewe, R. 1989, , 341, 484 Schwartz, R. D. 1977, , 35. 161 Simon, M. 1992, in IAU Colloq. 135, Complementary Approaches to Double and Multiple Star Research, ed. H. A. McAlister & W. I. Hartkopf (ASP. Conf. Ser.), 21 Simon, M., Chen, W. P., Howell, R. R., Benson, J. A. & Slowik, D. 1992, , 384, 212 Skinner, S. L. 1993, , 408, 660 Skinner, S. L., Brown, A., Stewart, R. T. 1993, , 87, 217 Stauffer, J. R., Caillault, J.-P., Gagné, M., Prosser, C. F., & Hartmann, L. W. 1993, , (submitted) Strom, K. M., Strom, S. E., Edwards, S., Cabrit, S., & Skrutskie, M. F., 1989b, , 97, 1451 Strom, K. M., Strom, S. E. & Merrill, K. M. 1993, , 412, 233 Strom, K. M., Strom, S. E., Morgan, J., & Wolff, S. 1986, , 62, 39 Strom, K. M. et al. 1990, , 362, 168 Strom, K. M., Wilkin, F. P., Strom, S. E. & Seaman, R. L. 1989a, , 98, 1444 Swenson, F. J., Faulkner, J., Rogers, F. J. & Iglesias, C. A. 1993, , (in press) Thornley, M., Woodward, C. E., Pipher, J. L., Forrest, W. J., Sellgren, K., & Shure, M. A. 1989, , 21, 1085 Tinney, C. G., Mould, J. R. & Reid, I. N. 1992. , 396, 173 Trümper, J. 1983, Adv. Space Res. 2, 241 Tufte, E. R. 1983, The Visual Display of Quantitative Information, Graphics Press, Cheshire, CT. Vrba, F. J., Rydgren A. E., Chugainov, P. F., Shakovskaya, N. I., & Weaver, W. B. 1988, , 97, 483 Vrba, F. J., Rydgren, A. E., Chugainov, P. F., Shakovskaya, N. I., & Zak, D. S. 1986, , 306, 199 Walter, F. M. 1992, , 104, 758 Walter, F. M., Brown, A., Mathieu, R. D., Myers, P. C., & Vrba, F.J. 1988, , 96, 297
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The nuclear tunneling crossover temperature ($T_c$) of hydrogen transfer reactions in supported molecular-switch architectures can lie close to room temperature. This calls for the inclusion of nuclear quantum effects (NQE) in the calculation of reaction rates even at high temperatures. However, standard computations of NQE relying on parametrized dimensionality-reduced models, quickly become inadequate in these environments. In this letter, we study the paradigmatic molecular switch based on porphycene molecules adsorbed on metallic surfaces with full-dimensional calculations that combine density-functional theory for the electrons with the semi-classical ring-polymer instanton approximation for the nuclei. We show that the double intramolecular hydrogen transfer (DHT) rate can be enhanced by orders of magnitude due to surface fluctuations in the deep tunneling regime. We also explain the origin of an Arrhenius temperature-dependence of the rate below $T_c$ and why this dependence differs at different surfaces. We propose a simple model to rationalize the temperature dependence of DHT rates spanning diverse fcc \[110\] surfaces.'
author:
- Yair Litman
- Mariana Rossi
title: |
Multidimensional Hydrogen Tunneling in Supported Molecular Switches:\
The Role of Surface Interactions
---
Nuclear tunneling is an inherently quantum-mechanical process that can strongly impact the properties of matter in a wide variety of situations, ranging from biological enzymes to organic-based technologies [@LayfieldHammes-Schiffer2013; @JiangShuai2015; @KaapLeeuw2016; @Koch_JACS_2017]. In complex environments, it has been recognized that signatures of tunneling on rate processes are often not well captured by textbook theories [@Meisner_Angew_2016; @Topaler_JCP_1994] and in particular, in hydrogen transfer reactions, the small mass of hydrogen makes tunneling pronounced [@HYDROGEN_TRANSFER]. Still, a theoretical description of nuclear tunneling that goes beyond a simple one-dimensional approximation and considers anharmonic coupling between many degrees of freedom in larger-scale systems remains a challenge [@Fang_NatCom_2020; @Rommel_JPCB_2012; @Mills_CPL_1997]. To build up a systematic understanding of multidimensional rate processes in the deep tunneling regime and the impact of the environment on hydrogen dynamics, a fully *ab initio* treatment of simple, yet non-trivial, reactions in well controlled conditions is desired.
![(a) Top view of the local minima of porphycene on Cu(110). Concerted and stepwise DHT mechanisms are represented by blue and orange arrows, respectively. Lateral view of the *cis* (b) and *trans* (c) conformers, of porphycene on Cu(110) and Ag(110).[]{data-label="fig:Porphycene"}](fig1.png){width="0.70\columnwidth"}
Tetrapyrrole macrocycles, like porphyrin, naphtalocianine, and porphycene, show a remarkable diversity of functional properties [@Auwarter_NatChem_2015; @Kobke_NatNano_2020; @Seufert_NatChem_2011; @Shubina_JACS_2007; @Grill_Rev_2008], and were proposed, among others, as molecular switches [@Zhang_ChemSocRev_2015; @Waluk_ChemRev_2017; @Qiu_PRL_2004]. The intramolecular hydrogen transfer reaction that occurs in the inner cage of these molecules, known as tautomerization, can be triggered remotely by different external stimuli [@Liljeroth_2007; @Kugel_NanoLett_2017; @Kumagai_PRL_2013; @Bockmann_NanoLett_2016; @Ladenthin_NAT_2016; @Mangel_PCCP_2020]. Moreover, because the reaction takes place without a pronounced conformational change, these molecules can be incorporated in nanoscale devices. Within this area, controlling the dynamical properties of these molecules is central to advance rational design. Molecular diffusion and rotations have been more often addressed [@Eichberger_NanoLett_2008; @Marbach_ChemComm_2014; @Shea_JCP_2014; @Buchner_JPCC_2011; @Sonnleitner_PRL_2011], while key aspects of the hydrogen transfer reaction mechanism and its temperature dependence remain poorly understood. This is especially due to the challenges in the description of multidimensional quantum dynamics that cannot be captured by perturbative treatments of anharmonic couplings [@Warshel_1982_JPC; @HabershonMano_Rev_2013; @Wei_IntRev_2019; @LitmanRossi2019; @Meisner_Angew_2016].
In this Letter, we study the effect of tunneling on the double intramolecular hydrogen transfer (DHT) of two representative systems, namely porphycene on Cu(110) and Ag(110) surfaces. In these reactions, the tunneling crossover temperature, which represents the temperature below which tunneling becomes greater than classical hopping over the barrier ($T_\text{c}=\hbar \omega_b /2\pi k_\text{B}$, where $\omega_b$ is the imaginary frequency of the unstable mode at the transition state geometry [@Gillan1987]) lies close to room temperature. Specifically, we i) identify the multi-dimensional DHT mechanisms at different temperatures, ii) clarify their temperature dependence, and iii) elucidate the role of surface fluctuations in the deep tunneling regime. The results we obtain are able to explain puzzling experimental measurements [@Kumagai_PRL_2013; @Koch_JACS_2017] that showed an unexpected temperature dependence of the DHT. In addition, they show that instead of acting only as a passive observer of the reaction, in certain situations the surface takes a prominent role in the tunneling event.
We here employ a combination of density-functional theory (DFT) for the electronic degrees of freedom with the ring polymer instanton (RPI) approximation [@JOR_JCP_2009; @Arni_thesis] for the nuclear degrees of freedom. RPI can be viewed as an extension of Eyring [@Eyring] transition state theory (TST) which includes tunnelling and captures anharmonic contributions along the reaction pathway. It is a semi-classical method that uses discretized closed Feynman path (CFP) integrals to evaluate tunneling rates in the deep tunneling regime. This approximation finds the dominant stationary pathways in the CFP that connect reactants and products, the instanton pathways. Then, a series of steepest descent approximations to evaluate the flux-side correlation function leads to the instanton approximation for thermal reaction rates $k_\text{inst}$ [@Jor_review_2018]. The rate can be expressed as $$\label{eq:instanton}
\begin{split}
k_\text{inst}(\beta) &= A_\text{inst}(\beta) e^{-S[\mathbf{x}_\text{inst}(\beta)]/\hbar},
\end{split}$$ where $A_\text{inst}$ accounts for the harmonic fluctuations around the instanton pathway $\mathbf{x}_\text{inst}$, $S$ is the Euclidean action, and $\beta=1/k_\text{B}T$ with $k_\text{B}$ the Boltzmann constant and $T$ the temperature. In the discretized CFP space, $\mathbf{x}_\text{inst}$ is a first order saddle point. Albeit approximate, this method shows the best tradeoff in situations where the quantum exponential wall would prevent a full dimensional evaluation of the exact tunneling rate [@Richardson_PCCP_2017; @Topaler_JCP_1994].
Within the Born-Oppenheimer (BO) approximation, evaluating Eq. \[eq:instanton\] requires the BO energies and forces. We here employ the most accurate level of theory affordable for the system size and number of system replicas required. We employ DFT with the Perdew-Burke-Ernzerhof (PBE) exchange-correlation functional [@PBE] including the Tkatchenko-Scheffler [@TS] dispersion correction modified to treat physisorption on surfaces [@TSsurf]. We thereby ensure a good description of the anisotropic electron density redistribution upon porphycene adsorption [@Kumagai_JCP_2018; @Jingtai_JCPL_2019]. At this level of theory, we do not expect reaction barriers to be quantitatively accurate [@Barone_CPL_1994], but expect to capture qualitative trends and the correct physics. In the supporting information (SI), we report selected geometries calculated with a range-separated hybrid functional for comparison. These calculations were enabled by the combination of the FHI-aims [@FHI-AIMS] all-electron code and the i-PI [@i-pi; @i-pi2] universal force engine. Details and convergence tests are provided in the SI.
{width="90.00000%"}
In porphycene, the inner-cage hydrogen atoms can adopt different configurations, giving rise to two stable tautomers, coined *cis* and *trans*. Each tautomer is two-fold degenerate and their structures on Cu(110) and Ag(110) are shown in Fig. \[fig:Porphycene\]. On these surfaces, the [*cis*]{} tautomer is more stable than the [*trans*]{} tautomer by 172 and 20 meV, respectively. Indeed, on the same surfaces, experiments observe almost exclusively the *cis* tautomer [@Ladenthin_NAT_2016; @Koch_JACS_2017; @Kumagai_NatCom_2014]. The DHT of the *cis*$\rightarrow$*cis* reaction can happen through two possible mechanisms. The concerted mechanism, where the hydrogen atoms are transferred together, without the existence of a stable intermediate and passing through a second-order saddle point, and the stepwise mechanism, where the hydrogen atoms are transferred sequentially and the reaction involves a *trans* intermediate.
We have recently studied the DHT of porphycene in the gas phase [@LitmanRossi2019], and found that there is a competition between a concerted and a stepwise DHT at different temperatures. However, new effects arise due to the interaction between the molecule and the surface. In order to rationalize them, we divide the influence of the surface on the physisorbed molecule into static and dynamical effects. The static surface effects refer to the change in the potential energy landscape upon adsorption, and can affect both static and dynamical properties of the adsorbed molecule. For example, while the global minimum in the gas phase is the *trans* tautomer in a flat conformation, on Cu(110) and Ag(110) the relative tautomer stability is reversed and the molecule is buckled, as shown in Fig. \[fig:Porphycene\]b and \[fig:Porphycene\]c. This distortion changes the DHT energy barriers and consequently the hydrogen dynamics. Dynamical surface effects, on the other hand, refer to the impact that the motion of the surface atoms have on the molecular properties. As it will be shown, they can significantly influence the dynamics of the DHT.
On Cu(110), we calculate $T_c = 264$ K for the DHT and show the instanton tunneling rates of the *cis*$\rightarrow$*cis* reaction in Fig. \[fig:Cu110Rates\]a. We show rates between 75 and 150 K because they lie well within the deep tunneling regime and we wish to compare with experimental results in the same temperature range [@Kumagai_PRL_2013], shown in the inset. In this case, we find a very good agreement with experiment throughout the six-orders of magnitude variation of the rate. The calculated effective activation energy $E_\text{A}$ is 190 meV. This compares well with the value of 168 $\pm$ 12 meV reported by experiments. Both the calculated rates and the measured ones show an Arrhenius-like temperature dependence even though tunneling is dominant at this temperature range. We proceed to explain the origin of this dependence.
The calculated DHT rates contain contributions almost exclusively from the stepwise mechanism, because at these temperatures it is several orders of magnitude faster than the concerted alternative. We propose that the *trans* intermediate was not observed because its predicted residence time is $\approx$ 0.1 ns, which lies beyond the time resolution of STM experiments ($\approx 100$ $\mu \text{s}$). The instanton trajectory provides an intuitive view of the reactive process by showing the main “instantaneous" tunneling configuration that the delocalized nuclei adopt. The trajectory is visualized in Fig. S1 at several temperatures. As the temperature is decreased, the reaction takes shorter pathways and crosses regions of higher PES energy. Moreover, we observe a considerable contribution from heavier atoms like C and N and, interestingly, even Cu atoms to the tunneling mechanism (see Tab. S11). These are manifestations of the multidimensional nature of the tunnelling process and prove that reducing the problem dimensionality is inappropriate [@Tuckerman_PRL_2001; @Chenfang_2019_JPCL; @Fang_NatCom_2020].
After these considerations, the observed $E_\text{A}$ can be understood as follows. At low enough temperatures, when only the vibrational ground states (VGS) are populated and a further decrease of the temperature does not affect the vibrational populations, the *trans*$\rightarrow$*cis* reaction will proceed from the VGS of the reactant and will be constant with temperature. As a consequence and because of detailed balance, the inverse reaction *cis*$\rightarrow$*trans*, which is the rate-controlling step of the stepwise mechanism, must show $E_\text{A}$ equal to the difference between the VGS energies of reactant and product, that we call $\Delta E$ (see Fig. \[fig:Cu110Rates\]c). The height and width of the barrier impact the absolute value of the rate, but they do not affect the Arrhenius slope in the low temperature limit of an asymmetric reaction. Indeed, a harmonic estimation of $\Delta E$ is 172 meV, which is very close to the calculated $E_A=190$ meV from Fig. \[fig:Cu110Rates\]a.
Surface $T$ (K) $\kappa_{\text{tun}}$ KIE$_{\text{inst}}$ KIE$_{\text{TST}}$ SFE
--------- --------- ----------------------- --------------------- -------------------- -----
Cu(110) 100 1.0 21 89 34
Cu(110) 85 1.2 32 197 106
Cu(110) 75 2.4 66 397 464
Ag(110) 75 14.9 - - 4
: Tunneling enhancement factor ($\kappa_{\text{tun}}$), and kinetic isotopic effects (KIE) and surface fluctuations enhancement (SFE). See definitions and discussion in the text.[]{data-label="tab:KIE"}
We further analyze how tunneling manifests itself in these reaction rates in Table \[tab:KIE\]. A standard procedure to estimate the impact of tunneling is to compare $k_{\text{inst}}$ with the rate predicted by the Eyring TST ($k_\text{TST}$), since the latter neglects tunneling but includes ZPE. The tunnelling enhancement factor is $\kappa_\text{tun} = k_\text{inst}/k_\text{TST}$ [@Beyer_PCL_2016], which is reported in Tab. \[tab:KIE\]. Surprisingly, because these factors are close to 1, they would seem to indicate that tunnelling plays a minor role. To understand this observation, we computed the kinetic isotope effect (KIE), defined as $k^\text{H}/k^\text{D}$, where $k^\text{D}$ was obtained from calculations where the inner-cage hydrogen atoms were replaced by deuterium. If tunnelling would be a minor effect, the only difference in these rates should be ZPE, and since ZPE is captured by both $k_{\text{inst}}$ and $k_\text{TST}$, the KIE of both should be similar. However, as shown in Tab. \[tab:KIE\], these numbers are different. TST overestimates the KIE in this particular case, for reasons outlined in the SI, Section VII. We thus conclude that $\kappa_{\text{tun}} \approx 1$ because of the following. On Cu(110) the $\Delta E$ (including harmonic ZPE) between reactants and products, which is a good estimate for $E_A$ in $k_\text{inst}$ as discussed above, happens to be similar to the energy difference between the ZPE corrected reactant and transition state, which defines $E_A$ for $k_\text{TST}$ (see Table \[tab:Surfaces\]). Additionally, because close to $T_c$ both rates are comparable, if $E_A$ are similar, the prefactors also must be. This observation explains why TST fared reasonably well in these systems in the past, even without including the relevant physics of tunneling [@Jingtai_JCPL_2019]. Thus, we propose that KIE$_\text{inst}$/KIE$_\text{TST}$ can be an alternative measure of tunneling contributions to hydrogen transfer reactions.
The investigation of the dynamical surface effects on the DHT required RPI calculations where we fixed the surface atoms at the reactant position. The rates obtained as a result of this constrained optimization lack all contributions from fluctuations of the surface degrees-of-freedom. We call the ratio between the rates with and without those constraints the “surface fluctuations enhancement" (SFE). The SFE for $k_\text{inst}$ on Cu(110) are reported in Table \[tab:KIE\] and can adopt surprisingly large values. The results show that the dynamical surface effects act on the opposite direction of the static ones, increasing the tunneling rates up to two orders of magnitude. Interestingly, the SFE become larger at lower temperatures because the contribution of heavy atoms to tunneling increases with decreasing temperature.
We then compare the Cu(110) with the Ag(110) substrate, a surface with a weaker static interaction. On Ag(110), porphycene is less buckled upon adsorption (see Fig. \[fig:Porphycene\]b) and the *trans* conformer lies only 20 meV above *cis*, but $T_c$ is also 264 K. Accordingly, it was observed in experiments that the DHT rates are substantially faster on Ag(110) than on Cu(110) [@Koch_JACS_2017]. Unlike Cu(110), the measured rates show two distinct regimes [@Koch_JACS_2017], reproduced in the inset of Fig. \[fig:Cu110Rates\]b. Above $\sim 10$ K there is an Arrhenius behaviour, while below $\sim 10$ K the rate shows almost no temperature dependence. In Fig. \[fig:Cu110Rates\]b we show the calculated $k_\text{inst}$ for the stepwise and the concerted mechanisms of the *cis*$\to$*cis* DHT of porphycene at Ag(110) (see calculation details in the SI). We obtain an $E_\text{A}$ of 45 meV for the stepwise mechanism, which compares reasonably well with the experimental $E_A$, which we estimate to be 12 $\pm$ 3 meV. The calculated harmonic value of $\Delta E$ is again close to $E_\text{A}$ for this reaction. At 10 K the $\mathbf{x}_\text{inst}$ for the concerted mechanism starts at the reactant minimum, indicating that tunneling takes place from the reactant VGS. As such, the rates for the concerted mechanism (which is symmetric) do not change with temperature below this point and it becomes dominant below 8.5 K. Hence, we can explain the two regimes observed in experiment by the change in the DHT mechanism. The lack of quantitative agreement between the calculated and measured, transition temperature and rate of the concerted mechanism comes most likely from the potential energy surface used in the calculations, but some dependence of the measured rates on the STM tip cannot be fully discarded. Finally, as shown in Table \[tab:KIE\], the SFE are smaller here than they were for Cu(110), accounting for a factor 4 increase of the rate at 75 K. This is consistent with the weaker adsorbate-surface interaction strength.
Building up on these considerations, the dependence of these DHT rates can be schematically understood as shown in Fig. \[fig:Cu110Rates\]d and predicted for other metallic fcc \[110\] surfaces. One needs to compute the ZPE-corrected energy difference between the *cis* and *trans* tautomers $\Delta E$, the ZPE-corrected energy barrier for the stepwise $E^*_\text{step}$ and the $T_c$ of the stepwise reaction. At high temperatures the reaction behaves classically and proceeds by hopping over the lower barrier, which is normally the stepwise one, yielding a slope of $\approx E^*_\text{step}$, labeled 1 in Fig. \[fig:Cu110Rates\]d. Considerably below $T_c$, the low temperature limit of the stepwise tunneling reaction is achieved and a slope of $\approx \Delta E$ should be observed (labeled 2). Finally, below a transition temperature $T_t$, the concerted mechanism becomes dominant and the rate becomes independent on temperature (labeled 3).
Surface $\Delta E$ (meV) $T_c$ (K) $E^*_\text{step}$ (meV) $T_\text{t}$ (K)
--------- ------------------ ----------- ------------------------- ------------------
Ag(110) 20 264 61 3 - 9
Au(110) 59 264 85 8 - 22
Cu(110) 172 264 222 15 - 41
Ni(110) 170 297 347 13 - 35
Pd(110) 148 293 326 12 - 34
: Calculated energies ($\Delta E$, $E^*_\text{step}$), crossover temperatures ($T_c$) and estimated transition temperatures ($T_\text{t}$) for the DHT of porphycene at several fcc 110 surfaces. See definitions in Fig. \[fig:Cu110Rates\] and text. The $T_\text{t}$ interval is given by considering calculated or experimental references for the Ag(110) case (see SI). The surface reconstruction of Au(110) [@Moritz_SurSci_1979] was ignored for the sake of comparison. \[tab:Surfaces\]
Using Ag(110) as a reference and building a 1D potential model for which it is necessary to calculate the barrier for the concerted mechanism $E^*_\text{con}$ (see model in the SI), $T_\text{t}$ can be estimated for other surfaces. In Table \[tab:Surfaces\] the calculated values for different fcc \[110\] surfaces are reported, together with the corresponding estimation of $T_\text{t}$. All $T_c$ values are similar and close to 300 K, showing the importance of tunneling at considerably high temperatures. While the estimated $T_t$ represent temperatures that can be achieved in different experiments especially for the stronger interacting surfaces, the resulting rates in Cu(110), Ni(110), and Pd(110) would be smaller than 10$^{-10}$ Hz, which lies far beyond the STM detection limit.
To conclude, we have shown how surface interactions can impact tunneling within a prototype molecular switch based on porphycene molecules adsorbed on metallic surfaces. This study was able to show the inadequacy of dimensionality-reduction schemes on these problems, the counter-intuitive origin of different temperature-dependencies of the rates in the deep-tunneling regime, and the effects of surface interaction on the intramolecular hydrogen tunneling. Even though full-dimensional calculations are required to get quantitative results and understand the underlying processes, we could propose a simple estimator to predict the DHT temperature dependence on different metallic surfaces.
The methodology we presented can be straightforwardly applied to other molecules on surfaces where the calculation of internal hydrogen transfer rates are sought. Limitations in the RPI approximation may arise when several local minima of the adsorbate with similar energies are present [@Kumagai_JCP_2018].
The well-defined system addressed in this work allowed us to disentangle and quantify static and dynamic effects of the environment (in this case the metallic surface) on quantum hydrogen dynamics. We showed that dynamical effects of the environment can promote hydrogen tunneling. Such a quantification is normally not straightforward in condensed phase or biological systems. In this sense, this work shows how single-crystal substrates can be an ideal playground where cutting-edge theory and experiment can meet to provide a deeper understanding of quantum dynamics in fluctuating environments. These findings will help to address hydrogen dynamics in biology [@KlinmanKohen2013] and in functional materials [@Tayi_NatChem_2015], as well as guide the design and interpretation of future experiments.
The authors thank Stuart Althorpe, Aaron Kelly and Matthias Koch for fruitful discussions, and thank Takashi Kumagai and Jeremy Richardson for numerous discusssions and a careful assessment of the manuscript. The authors acknowledge financing from the Max Planck Society, and computer time from the Max Planck Computing and Data Facility (MPCDF) and the Swiss National Supercomputing Centre (CSCS) under project ID s883.
[55]{}ifxundefined \[1\][ ifx[\#1]{} ]{}ifnum \[1\][ \#1firstoftwo secondoftwo ]{}ifx \[1\][ \#1firstoftwo secondoftwo ]{}““\#1””@noop \[0\][secondoftwo]{}sanitize@url \[0\][‘\
12‘\$12 ‘&12‘\#12‘12‘\_12‘%12]{}@startlink\[1\]@endlink\[0\]@bib@innerbibempty @noop [****, ()]{} @noop [****, ()]{} [****, ()](\doibase
10.1103/PhysRevB.93.140206) @noop [****, ()]{} [****, ()](\doibase 10.1002/anie.201511028) [****, ()](\doibase 10.1063/1.468244) @noop [**]{} (, ) @noop [****, ()]{} [****, ()](\doibase 10.1021/jp308526t) [****, ()](\doibase https://doi.org/10.1016/S0009-2614(97)00886-5) [****, ()](\doibase 10.1038/nchem.2159) [****, ()](\doibase 10.1038/s41565-019-0594-8) [****, ()](\doibase 10.1038/nchem.956) @noop [****, ()]{} [****, ()](\doibase 10.1088/0953-8984/20/05/053001) [****, ()](\doibase 10.1039/C4CS00377B) [****, ()](\doibase 10.1021/acs.chemrev.6b00328) [****, ()](\doibase 10.1103/PhysRevLett.93.196806) [****, ()](\doibase 10.1126/science.1144366) [****, ()](https://doi.org/10.1021/acs.nanolett.7b02419) [****, ()](https://link.aps.org/doi/10.1103/PhysRevLett.111.246101) @noop [****, ()]{} [****, ()](\doibase
10.1038/nchem.2552) [****, ()](\doibase
10.1039/C9CP06868F) @noop [****, ()]{} [****, ()](\doibase 10.1039/C4CC01744G) [****, ()](\doibase 10.1063/1.4890661) @noop [****, ()]{} [****, ()](\doibase
10.1103/PhysRevLett.107.186103) [****, ()](\doibase 10.1021/j100209a016) @noop [****, ()]{} [****, ()](\doibase
10.1080/0144235X.2019.1558623) @noop [****, ()]{} @noop [****, ()]{} [****, ()](\doibase 10.1063/1.3267318) **, @noop [Ph.D. thesis]{}, () [****, ()](\doibase 10.1021/cr60056a006) [****, ()](\doibase 10.1080/0144235X.2018.1472353) [****, ()](\doibase 10.1039/C6CP07808G) [****, ()](\doibase 10.1103/PhysRevLett.77.3865) [****, ()](\doibase 10.1103/PhysRevLett.102.073005) [****, ()](\doibase
10.1103/PhysRevLett.108.146103) @noop [****, ()]{} @noop [****, ()]{} [****, ()](\doibase https://doi.org/10.1016/0009-2614(94)01238-5) [****, ()](\doibase http://dx.doi.org/10.1016/j.cpc.2009.06.022) [****, ()](\doibase http://dx.doi.org/10.1016/j.cpc.2013.10.027) @noop [****, ()]{} [****, ()](\doibase 10.1038/nchem.1804) @noop [****, ()]{} @noop [****, ()]{} [****, ()](\doibase 10.1021/acs.jpclett.6b02115) @noop [****, ()]{} @noop [****, ()]{} [****, ()](\doibase
10.1038/nchem.2206)
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'SLAM based techniques are often adopted for solving the navigation problem for the drones in GPS denied environment. Despite the widespread success of these approaches, they have not yet been fully exploited for automation in a warehouse system due to expensive sensors and setup requirements. This paper focuses on the use of low-cost monocular camera-equipped drones for performing warehouse management tasks like inventory scanning and position update. The methods introduced are at par with the existing state of warehouse environment present today, that is, the existence of a grid network for the ground vehicles, hence eliminating any additional infrastructure requirement for drone deployment. As we lack scale information, that in itself forbids us to use any 3D techniques, we focus more towards optimizing standard image processing algorithms like the thick line detection and further developing it into a fast and robust grid localization framework. In this paper, we show different line detection algorithms, their significance in grid localization and their limitations. We further extend our proposed implementation towards a real-time navigation stack for an actual warehouse inspection case scenario. Our line detection method using skeletonization and centroid strategy works considerably even with varying light conditions, line thicknesses, colours, orientations and partial occlusions. A simple yet effective Kalman Filter has been used for smoothening the $\rho$ and $\theta$ outputs of the two different line detection methods for better drone control while grid following. A generic strategy that handles the navigation of the drone on a grid for completion of the allotted task is also developed. Based on the simulation and real-life experiments, the final developments on the drone localization and navigation in a structured environment are discussed.'
author:
-
bibliography:
- 'IEEEexample.bib'
title: '**Grid-based Localization Stack for Inspection Drones towards Automation of Large Scale Warehouse Systems** '
---
INTRODUCTION
============
Use of drones has boomed in the past few years especially towards the industrial usage that ranges from delivery to fault checks. All these applications have a standard fundamental requirement i.e., smooth navigation of the robot from an initial location to the final location, for which the robot needs to know it’s local/global position in real-time. Indoor localization introduces an another layer of difficulty as the presence of GPS signal is generally weak or unavailable for the drones to localize in a global frame. This type of scenario is commonly found in industrial and warehouse environments that more or less functions in the indoor conditions. Figure 1 shows a scenario from a large warehouse system updating towards its automation. For any mobile robot working in an application, the primary task it must be excelling at is navigation. There are various navigation strategies used for mobile robots depending upon environment viz. indoor, outdoor and depending upon information viz. map-based, map-less. Also, just designing an algorithm is not sufficient. For navigation, various sensors are assembled onboard the mobile robot that senses the surrounding environment and acts as a feedback. This feedback ensures that the robot follows the commands given by the algorithm as accurately as possible. Today, various sensors like LiDAR, Camera, Radar, Sonar, etc., are explicitly designed for onboard use on mobile robots. However, researchers now are focusing more towards camera-based navigation as it is substantially cost-effective compared to the rest (Radar, LiDAR are still expensive) and provides detailed information about the environment which may not be available even after using a combination of the other sensors. UAVs perform active manoeuvres in their application environment which limits their size and further reducing their payload carrying capacity. Industrial applications that demand significant UAV flight time should also be light and cost-efficient. Thus, efforts are made to reduce the number of onboard sensors. For all these requirements, a smart usage of the monocular camera can serve the overall purpose.
A navigation stack will only make sense if the drone is capable of localizing in an environment. For the presented case, we need to localize using the lines and nodes existing in the environment for which we need to detect the thick guidelines first. One of the most widely used line detection algorithms is Hough-line-based detection, but it has many limitations that are expected to create problems in a real-world scenario. Much of the efficiency of the Hough transform [@Hough_1962] relies upon the quality of input data i.e., the edges have to be detected well for the Hough transform to be efficient. Using the vanilla hough transform on the noisy images is considerably prone to the haywire detections, and generally, to tackle this, a denoising stage is prefixed. A rule-based denoising framework leads to the occurrence of variance by threshold configurations, whereas, a learning-based denoising framework leads to an increase in computational load. Hence, we propose a pipeline for robust line detection that overcomes these problems by reducing the artifacts to a state easily detected by hough line algorithm with less severe errors.
A standard thick line detection algorithm generally supports the following steps:
1. Denoising
2. Line Segmentation / Thresholding
3. Edge Detection
4. Hough Line Transform
5. Combining multiple results
The pipeline we propose in this paper, as shown in Figure 2, starts with subscribing to the down-facing camera images from the drone platform. A thick-line detection algorithm was implemented as a part of machine vision which holds significant value for the optimal performance of localization. The paper is divided into following sections. Literature review and relevant studies in this direction are discussed in Section II. Sections III states all the assumptions and conventions considered while the development of the framework. Section IV gives a walk-through of different methods applied with a brief discussion. Section V describes the control and localization implementation for the drone using the output from our thick-line detection algorithm. Any offset of drone from the closest safe point of hovering (i.e. intersection of two guidelines) is reduced with multiple PID loops acting on the lines detected within the image. Thus, discretizing the drone’s location to nodes co-ordinates. Finally, section VI discusses a navigation strategy that we suggest for the drone to follow in the given structured environment.
RELATED WORK
============
Large warehouses across the country lose millions of dollar every year due to lost inventory. Manual scanning of inventory is both time-consuming and prone to error. This problem of real-time warehouse inspection system has been approached by different entities. Kiva Systems, now owned by Amazon, has implemented an autonomous inventory system in their warehouses that track both the location and contents of every bin in the warehouse using AGVs. Few of the problems that this system face include a limited height of the shelf, a requirement of a manual process for the inventory scanning and low-speed manoeuvre of the ground vehicles. Another option available in the industry involves deploying an infrastructure of permanent shelves with built-in RFID readers such that the shelf can recognize the item being stored in it. While this system gives a real-time scan of the inventory, the permanent shelf doesn’t allow the ground robots to automate the inventory movement.
MIT’s RFly [@rfly] used relays which during the flight picked RFID responses from various sites along the path of the drone, and handled these spatial measurements as an antenna array. Later, by applying the antenna array equations to these measurements, they can localized RFIDs. This method is based on previous work on Synthetic Aperture Radar (SAR) systems. Another organization, Infinium Robotics, used a miniature ground vehicle mounted with a visual tag for the drone to follow between the shelf gaps. This type of method increases the cost per drone because of the requirement of an additional ground robot to act as a guide.
For all such drones developed towards warehouse automation, localization stands as the most critical module. Hence, we considered it important to review the present state of localization methods for structured environments. The current state-of-the-art algorithms in this direction are majorly the existing algorithms in simultaneous localization and mapping (SLAM) for autonomous robots using different sensor setups. For example, LSD-SLAM [@lsdstereo] and ORB-SLAM2 [@orb] used stereo camera, DSO [@dso] and VINS-Mono [@vins] used monocular camera with inertial sensor, the work presented by Hess et. al. [@2dlidar] and Bosse et. al. [@bosse] used LiDARs, etc. Although these methods work quite well even in unstructured environments, sensors like stereo camera and LiDAR systems are, in general, bulky to be used on-board UAVs in a constrained environment with a considerable cost of deployment for large-scale application. Moreover, these systems have high computational power and memory need that renders them unsuitable for on-board use of a micro-aerial vehicle (MAV). Our method, therefore, utilizes the presence of structured environment, like the use of grid lines available in typical warehouses for the movement of ground robots, to build a robust and computationally efficient solution that allows the drone to localize itself on a warehouse map with a sufficient accuracy. A previous grid-based localization system [@manash] used the properties of particularly structured grid-lines to localize a MAV inside each sub-cell of the grid. However, it required a minimum of three vertical and horizontal equally spaced grid lines for sufficiently accurate localization. Our method removes this extra constraint of having multiple lines by employing a combination of elegant and simpler strategies with discrete localization on nodes.
ASSUMPTIONS AND CONVENTIONS
===========================
Conventions
-----------
The following conventions are used in this paper:
1. The drone’s coordinate system: X-axis and Y-axis are along the roll and pitch axis of the drone in hovering mode respectively, the origin being the centre of the drone.
2. Vertical line: The straight line parallel to the shelf.
3. Horizontal line: The straight lines perpendicular to the shelf.
4. $\theta$ : Angle made by the perpendicular to the vertical line from origin with the X-axis in drone’s co-ordinate system
5. $\rho$ : The length of the perpendicular to the vertical line from origin with the X-axis in drone’s Co-ordinate system. Figure 3 shows a sample snapshot containing drone with its co-ordinate axes, the vertical line, the parameters $rho$ and $theta$.
6. The pitch angle of the drone is along the horizontal line and roll angle along the vertical line.
7. -10$\degree$ $\leq$ roll $\leq$ 10$\degree$, though we only used to give roll in range of -1$\degree$ and 1$\degree$.
8. -10$\degree$ $\leq$ pitch $\leq$ 10$\degree$, though we only used to give pitch in range of -1$\degree$ and 1$\degree$.
9. -10$\degree$ $\leq$ yaw-rate $\leq$ 10, though we only used to give yaw-rate in range of -1$\degree$ and 1.
Assumptions
-----------
A warehouse is a well-structured environment. The boxes or cargo placed in a warehouse shelf may have different sizes i.e., there may not be a specific constant horizontal distance between any two adjacent boxes. To take such situations into account and to ensure that no box is left unscanned, some acceptable conditions have been assumed. The lines and nodes are assumed to be next to the cargo shelf where drone needs to stop and scan for the QR codes and report the location corresponding to the nodes. A typical warehouse setup following the stated assumptions is shown in Figure 4. The horizontal line should end before some distance ($D_Y$) from the shelf to avoid any contact of the drone with the boxes. $D_Y$, $D_X$ (distance between any two adjacent nodes) and $\sigma$ can be related by a simple equation given by $$D_X = 2\times D_Y \times\tan( \frac{\sigma}{2} )$$ where
$\sigma$ $\gets$ Field of view of the mounted camera. Equation (1) can be easily realized from Figure 2.
The vertical length between the start of any shelf and the first node must be less than or equal to $\frac{D_x}{2}$.
For a controlled environment like a warehouse, constant pressure with negligible wind is assumed to minimize the errors in height estimation using a barometer, good lighting condition is expected for crisp grid detection and QR code detection. Any moving obstacle like forklifts, overhead chains, etc which can hinder the path of the drone is assumed to be absent or away from the drone’s route.
VISION SYSTEM
=============
For the presented work, vision algorithms hold a crucial position for the drone navigation as it allows the drone to localize within the grid structure. In such a scenario, using a simple method consisting of canny and hough transform as line detection method might seem promising, but is equally risky to use citing its naivety under extreme conditions. Hence, it is necessary to improve certain sections of the grid detection algorithm focusing towards robustness while also keeping the time complexity similar to the simple method as above. Following sub-sections discuss some of the improvements, we propose.
Line Thresholding
-----------------
Image input is converted to HSV colour space for thresholding as it provides a better thresholding accuracy compared to RGB colour space. A Support Vector Machine model [@svm] was used to train the thresholding of different colour segments and used region growing algorithm on top of it to identify the line bands for varying brightness and intensities. We trained our model to detect yellow colour under varying conditions and lighting. Results were tested on blurred and distorted video and our model was robust enough to quickly threshold the line with minimal loss in accuracy. Figure 5 shows the effectiveness of colour segmentation using the above method.
Line Tracking
-------------
### Centroid Method
The thresholded image was broken into $W\times H$ slices which are dynamically identified based on the lines coverage area in the image, e.g., increasing the altitude of drone reduces the thickness of line thus greater slices are needed to detect the line features properly. For a very large area of lines in the image, the slices can be reduced, while for a small area of lines more slices are needed to cover all the curves and turn in the lines. Each slice was processed in parallel to find the centroid of each slice using OpenCV and contour detection. Moments of each contour was calculated and the centroid was returned for each slice of the image. Figure 6 shows the centroid extraction result on a sample image.\
Moment of any object in 2D space is given by $$M_{pq} = \int_{-\infty}^{\infty}\int_{-\infty}^{\infty}x^py^qf(x,y)\mathrm{d}x\mathrm{d}y$$ For a grayscale image with pixel intensity I(x,y) , raw moment is given by $$M_{ij} = \sum_x\sum_yx^iy^jI(x,y)$$
Centroid of the object then can be calculated as $$\begin{aligned}
\bar{x} = \frac{M_{10}}{M_{00}}\\
\bar{y} = \frac{M_{01}}{M_{00}}\end{aligned}$$ Image slices were then concatenated to form an image with the centroid points. These centroid points were used for RANSAC [@ransac] model fitting to find the best fitting line for the given set of points.
$images\gets divide(image,N_{height},N_{width})$ $contours\gets contour(img)$ $contour_{Max} \gets Max_{area}(contour)$ $M \gets moment(contour_{Max})$ $Centre_X = M_{10}/M_{00} $ $Centre_Y = M_{01}/M_{00}$ **push** $Centre_X, Centre_Y$ $image \gets concatenate(images)$ **return** $image$
### Skeletonization
The centroid method can get computationally expensive for an on-board processor fitted on drones like Raspberry Pi; an alternative method can be used to attain respectable performance compared to the centroid method.
The morphological skeleton of a shape is a thin version of the form that is formed at the centre of the shape, equidistant from the edges. It usually emphasizes the geometric and topological features of the structure, such as shape, topology, length, direction, and width.
The thresholded image was converted to binary and image was eroded to obtain the skeleton view of the image. Thinning algorithm by Zhang-Suen [@thinning] was used to thin the contours to a few pixels wide producing a single straight line for the hough line to work properly. Hough transform was then used to find the lines from the reduced skeletonized points. Figure 7 shows one such sample of application of this method with comparison with canny edge detection. To further reduce the error, mean clustering was used to cluster lines to an accuracy of $5$ degrees. Two ranges of $\theta$ were considered for classifying as a vertical or horizontal line.
$$Line = \left\{
\begin{array}{@{}ll@{}}
vertical, & \text{if}\ 0\degree \leq \theta < 30\degree \\
vertical, & \text{if}\ 150\degree < \theta < 180\degree \\
horizontal, & \text{otherwise} \\
\end{array}\right.$$
Lines are sorted into vertical and horizontal topics using means clustering with a specific threshold and are clustered into a group of two lines. Figure 8 shows the threshold range and permitted perturbation for the line clustering.
If the skeletonization method is used for line detection, there may be an offset in the line detected because of the nature of skeletonization algorithm. This can be tackled by fusing the data from centroid as well as skeletonization, if the hardware permits, to gain more accurate line detections.
CONTROL AND LOCALISATION
========================
The centroid or skeletonization algorithm returns a set of lines that sometimes are found to be fluctuating by a small quantity, which then affects the stability of the drone while following the line. Therefore, a simple Kalman filter was applied on the output line equations (parameters in the Hessian form) to minimize the effect of jitters and random false lines produced by the cluttered noise in the image.
A clustering algorithm was also implemented in case of skeletonization which further normalizes any erroneous data produced due to the branching of the skeleton of the line i.e., multiple perpendicular lines to the reference line occurring in an image. Assuming resolution of the line to be in the range of $5\degree$, lines were clustered in different pairs with $5\degree$ difference between them. Figure 9 shows the detection of two nodes after applying this method to a sample image. This helps to provide yaw feedback to the drone with respect to the line and can correct any other orientation errors induced in the system.
A closed-loop PID controller was used on the roll, pitch and yaw rate of the drone with the lines pose as feedback to control the $x$ and $y$ position, and yaw of the drone with respect to the grid lines. The design of PID controller that was implemented for our purpose is shown in Figure 10. The velocity output of each of the control loop was clipped to a range $[-0.1, 0.1]$ to avoid the possibility of drone deviating by unexpectedly high value and the grid line getting out of the field-of-view of the down-facing camera. PID equation of each subsystem is given as follows:
$$PID = K_P e(t) + K_I\int e(t) + K_D\frac{de(t)}{dt}$$
where:
$K_P$: Proportional Constant
$K_I$: Integral Constant
$K_D$: Derivative Constant
$w$: Image Width
$h$: Image Height
$\Delta x$ = $\frac{w}{2}$ - $\rho_{vertical}$
$\Delta y$ = $\frac{h}{2}$ - $\rho_{horizontal}$
$\Delta \theta$ = 0 - $\theta_{horizontal}$
integral = $integral_{previous}$ + error
derr = error - $error_{previous}$
dt : execution time
Even though the roll and pitch angles were clipped to a small value, the instantaneous orientation of the drone created an offset in the position of the line in an image as shown in Figure 11. This offset was observed to be proportional to the altitude (H) and tangent (trigonometric function) of the roll or pitch angle of the drone for vertical line and horizontal line respectively. To compensate for this offset for accurate estimation of the position of lines, we implemented an offset correction step before sending the line error signal to the controller. The correction step is formulated as follow:
$$\rho_{vertical}^{Corrected} = \rho_{vertical} - H\times\tan(roll)$$
$$\rho_{horizontal}^{Corrected} = \rho_{horizontal} - H\times\tan(pitch)$$
Finally, the PID control loops were tuned independently one by one first and then together using the previously estimated parameters as the starting point. Figure 14 shows the experimental setup along with the drone used for tuning the parameters.
STRATEGY
========
While it is important to understand the environment and localize oneself, it is of equal importance to formulate a strategy for accomplishing the set goal. Our navigation stack helps the drone to comply with this simple notion. Like any other robotic decision-making process, we defined the following key modules on which our system works.
a\. Perception
b\. Control and Localization
c\. Strategy
Sections IV and V appertain to the first two modules. Depending on the environment, the strategy employed by the drone for its functioning can be modified. For example, if the grid structure is available, grid nodes are used as intermediate targets. In our implementation, these intermediate targets were identified and tracked by the strategy module depending upon the global target or task to be accomplished, like detection and localization of the goods using the bar-codes and QR codes.
Turn detection
--------------
Apart from nodes and vertical lines, curved or sharp turns can also be present in the environment which can be mistaken as a node if not dealt with properly by the algorithm. To avoid such a situation, an L-detection algorithm was implemented which differentiated an L from a regular node. The ratio of the number of pixels on the left to right side of the skeletonized line was considered, and a threshold ratio was set which, when exceeded, referred to a turn in the path.
Multiple Line Handling
----------------------
The case of multiple nodes being detected in a given position is handled by tracking the goal node. Once the lines and their $\rho$ and $\theta$ values are detected, lines are divided into two different topics, vertical and horizontal. Vertical line published is used for the traversal of the drone. An imaginary horizontal line is assumed in front of the horizontal lines which helps the drone steer forwards. Multiple horizontal lines may cause problems when the height is changing as $\rho$ and $\theta$ value of each line might change significantly thus drone’s returning to the same line becomes a difficult task. To tackle the problem, we have a constantly updating line coordinate which keeps track of current drone hover line parameters and updates the specific lines at different heights so that the drone doesn’t lose the tracking in case multiple horizontal lines are visible. The strategy commands the controller to hover the drone over a specific goal node. We also included some exceptional cases that are bound to occur, e.g., the starting point/take-off strategy and endpoint/charge station detection for actual deployment scenario.
a\. {width="1in" height="1in"} {width="1in" height="1in"} {width="1in" height="1in"} {width="1in" height="1in"} {width="1in" height="1in"} {width="1in" height="1in"}
b\. {width="1in" height="1in"} {width="1in" height="1in"} {width="1in" height="1in"} {width="1in" height="1in"} {width="1in" height="1in"} {width="1in" height="1in"}
c\. {width="1in" height="1in"} {width="1in" height="1in"} {width="1in" height="1in"} {width="1in" height="1in"} {width="1in" height="1in"} {width="1in" height="1in"}
d\. {width="1in" height="1in"} {width="1in" height="1in"} {width="1in" height="1in"} {width="1in" height="1in"} {width="1in" height="1in"} {width="1in" height="1in"}
e\. {width="1in" height="1in"} {width="1in" height="1in"} {width="1in" height="1in"} {width="1in" height="1in"} {width="1in" height="1in"} {width="1in" height="1in"}
EXPERIMENTAL SETUP AND EVALUATION
=================================
All the demonstrated methods and strategies were implemented on a Parrot AR Drone 2 with variants like off-board processing on remote compute unit and on-board processing with Raspberry Pi 3. The remote compute unit was equipped with Intel Core i7 7th Gen @2.7GHz x 4 processor with 32 GB on-board RAM. For the other variant, Raspberry Pi 3 was attached to the bottom of AR Drone which led to a decrease in flight time as it drew power from the drone’s battery and also added up to the drone’s weight. This was compensated using a Lithium-ion battery that provided more charge density than the standard one. We implemented our pipeline on these variants and evaluated the performance. Before the actual hardware deployment, all the necessary tests were made on a Gazebo simulation as shown in Figure 13. AR drone hosts two cameras, from which we targeted the down-facing camera for localization and the front-facing camera for detecting QR-Code/Barcode. We chose Robot Operating System (ROS) for retrieving drone’s on-board sensors data and also for sending appropriate command actions to it. ROS package ardrone\_autonomy was used for this purpose. Image processing on drone-sent camera data was done with the help of the OpenCV library. Finally, all the data handling and processing codes were written in optimized multi-thread C++ code to match the industrial standards. Table I shows On-board (fps) Vs Off-board (fps) for Centroid and Skeletonization Methods.
Evaluation Metric and Results
-----------------------------
Better performance of the grid-localization relates to better grid-line detection by the machine vision algorithm. Hence, we provide an evaluation, particularly for the line detection algorithms. Any line in a plane can be represented by Hessian normal form of the equation of line in parameters $r$ (normal distance of the line from the origin) and $\alpha$ (angle made by the normal vector to the line with one of the axis). We measured the performance of different methods by calculating the absolute errors in $r$ and $\alpha$ of the generated line with respect to the ground truth line. We collected about 50 images containing different views of the grid structure both from the simulation world and real world experimental setup each. The ground truth for each of these images was drawn manually as perceived by the human mindset and was later converted into hessian normal form. Some of the images and the result of the application of discussed methods in the previous sections are shown in Figure 12. Table II shows the performance of methods on an average over 50 images. Results show Naive approach suffer due to thick lines. This causes it to create multiple lines and thus reducing the accuracy of the approach. Moreover, multiple horizontal lines are averaged out as one due to clustering which further causes error, especially in real test cases. Centroid approach works by dividing the line into smaller parts and taking average over it which reduces the errors significantly and provides very accurate $\rho$ and $\theta$ values. Skeletonization, on other hand, works by reducing the lines to single pixel thickness. This induces errors in $\theta$ of the final line but comparatively fewer errors in $\rho$ compared to centroid approach. Combined approach sees a balanced accuracy of both the approaches separately as it normalizes some of the drawbacks from each approach.
**Method** **On-board** (fps) **Off-board** (fps)
----------------- -------------------- ---------------------
Centroid 4.23 15.10
Skeletonization 8.93 20.0
: Performance evaluation of our method on different hardware.[]{data-label="table_example"}
----------------- ----------------- ----------------------------- ----------------- -------------------- -- --
$\Delta r$ (px) $\Delta \alpha$ ($\degree$) $\Delta r$ (px) $\Delta \alpha$ ()
Naive approach 56 3 78 12
Centroid 10 3 18 5
Skeletonization 5 4 15 10
Combined 6 3 16 7
----------------- ----------------- ----------------------------- ----------------- -------------------- -- --
: Performance evaluation of different method on a dataset of images of size 1920$\times$1080.
CONCLUSIONS
===========
The proposed pipeline demonstrates promising result in both simulation world and real world scenario. Both of our line detection algorithms get away the limitations of pure hough line transform. Depending upon the hardware available, we show the capability of the method to outperform classical techniques. Our clustering algorithm on rho, theta of multiple lines effectively gives the closest node to which the intermediate target should be set to. At any point in time, the offset from this set-position is reduced iteratively by multiple PID loops. This simple yet effective localization method gives a competitive performance regarding speed as compared to the application of heavy SLAM algorithms on same computing power. Though we cannot expect comparable accuracy as we discretized the locations of the drone to nodes co-ordinates, the algorithms achieve decent accuracy in terms of node coordinates as the application area doesn’t require per position localization in a well-defined environment. The proposed navigation strategy makes the drone capable of planning a sub-optimal route to attain global target/setpoint. We finally end our conclusion with the restatement that monocular camera-based vision techniques when designed appropriately can perform significantly well as compared to price one needs to pay for using full SLAM-based solutions.
ACKNOWLEDGMENT {#acknowledgment .unnumbered}
==============
We would like to thank the institute authorities at Indian Institute of Technology Kharagpur (session 2017-18) and IIT Gymkhana body to avail us the funds as a support for the presented work. Also, a special thanks to members of Aerial Robotics Kharagpur group for providing us their working space and insights time to time for the betterment of the project.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'I review recent developments in the application of effective field theory to nuclear physics. Emphasis is placed on precision two-body calculations and efforts to formulate the nuclear shell model in terms of an effective field theory.'
address: |
Department of Physics, University of Washington,\
Seattle, WA 98915\
and\
Jefferson Lab., 12000 Jefferson Avenue, Newport News,\
Virginia 23606.
author:
- 'Martin J. Savage'
title: |
Effective Field Theory in\
Nuclear Physics[^1]
---
Introduction {#introduction .unnumbered}
============
A question that I have been asked many times is “[*Why use Effective Field Theory in Nuclear Physics*]{}”? The simple and somewhat glib answer to this question is that the only other option that one has to using an effective field theory (EFT) is to use the “[*Theory of Everything*]{}” (TOE), string theory or some derivative thereof. All other descriptions [**must**]{} be incomplete at some level and when precise predictions are compared with precise measurements, differences will become obvious. It is a daunting prospect for us (maybe only me) to use the TOE to compute low-energy hadronic processes, and in fact, it is quite silly to even consider such calculations. After all, we know that processes in QED can be computed to high precision which agree with experimental observations, without knowing anything about physics at the Planck scale, $M_{\rm pl}$.
The renormalizability of QED assures that ultra-violet divergences, arising from our lack of understanding of physics at short-distances, can be explicitly removed by a few constants (electric charge and fermion mass), allowing observables to be related to each other to arbitrary precision. In contrast, EFT’s are non-renormalizable but are still predictive when a systematic power counting in small expansion parameters can be established. Relations between observables at a given precision will involve a finite number of constants that are not dictated by the symmetries of the EFT alone. For instance, the standard model is a renormalizable field theory but processes at energies much less than the scale of electroweak symmetry breaking can be described by an non-renormalizable EFT of reduced symmetry, $SU(3)_c\otimes SU(2)_L\otimes U_Y(1)
\rightarrow SU(3)_c\otimes U_{\rm em}(1)$, where weak interactions are incorporated by higher-dimension four-fermi operators. Even with quantum effects, the theory provides a systematic expansion of observables in terms of $q^2/ M_W^2$ and $m_f^2/M_W^2$, where $m_f$ is a fermion mass, $M_W$ is the mass of the weak gauge bosons, and $q$ is the external momentum.
If one were interested in calculating the cross section for $np\rightarrow
d\gamma$–radiative neutron capture by a proton to form a deuteron–directly from QCD, then a lattice calculation is the only technique available. The lattice calculation will provide an unambiguous cross section in terms of the quark masses and $\Lambda_{\rm QCD}$, or equivalently in terms of other hadronic observables. A cartoon of a contribution to $np\rightarrow d\gamma$ in terms of perturbative quarks and gluons is shown in Figure \[fig1\].
=2.5in
Sources of quarks and gluons that have non-zero overlap with the proton, neutron, deuteron and a source for the photon would used to generate the amplitude for $np\rightarrow d\gamma$. It is clear that a significant amount of work goes into forming the hadronic states themselves, let alone computing the interaction terms. In addition, as the deuteron has such a small binding energy and hence is quite extended compared to the nucleon, considerable effort will be required to generate the deuteron itself. Unfortunately, at this point in time the lattice community is not even close to being able to perform this multi-hadron calculation. Indeed, the deuteron itself remains to be generated in lattice calculations. [^2] If all nuclear lengths scales were of order the chiral symmetry breaking scale $\Lambda_\chi$ then (very) naively lattice computations of nuclear observables would not be that much harder than computations in the single-nucleon sector. However, there are several low-energy length scales that play important roles in nuclear physics. Firstly $\Lambda_\chi$, below which a hadronic description makes sense, and higher-dimension operators are induced that describe contributions from scales above $\Lambda_\chi$. Secondly, the scale of the repulsive part of the nucleon-nucleon interaction, which is conventionally modeled by the exchange of vector mesons (far from mass-shell) and numerically is of order $\Lambda_\chi$. Thirdly, the mass of the pion, which is much less than $\Lambda_\chi$ due to its special status as a pseudo-Goldstone boson. Finally, nuclear binding energies which are much smaller than one would naively guess.
If one is interested in this process at energies much less than $\Lambda_\chi$, or the mass of the $\rho$-meson, but comparable to the mass of the pion, then it should be sufficient to use an EFT with only nucleons, pions and photons as dynamical degrees of freedom. All contributions from higher mass scales will be encapsulated in the infinite number of higher dimension operators that arise in the momentum and chiral expansions. Some diagrams that will contribute to $np\rightarrow d\gamma$ are shown in Figure \[fig2\].
=3.2in
Unlike most EFT’s that one encounters, higher dimension operators (dim-6) involving the nucleon field play a central role[@Weinberg]-[@KSW]. The fact that the deuteron is barely bound, with a binding energy of $B=2.2~{\rm MeV}$, requires a fine-tuning between pion exchange and short-distance physics. Naively, one would expect a binding energy set by $f_\pi$, the pion decay constant, much larger than one finds in nature. Therefore, the class of diagrams shown in Figure \[fig2\](b) is not expected to be suppressed compared to those in Figure \[fig2\](a). In addition, contributions from diagrams shown in Figure \[fig2\](c) must be included. These arise from an insertion of operators that are gauge invariant by themselves, and are not related in any way to operators describing nucleon-nucleon scattering. They arise from short-distance physics and have a scale typically set by $\Lambda_\chi$.
Continuing our descent in energy, if one is interested in this process at energies much less than the pion mass, $m_\pi$, then it should be sufficient to use an EFT with only nucleons and photons as dynamical degrees of freedom. All contributions from mass scales greater than $m_\pi$ will be encapsulated in the infinite number of higher dimension operators that arise in the momentum expansion (chiral symmetry is explicitly broken, leaving isospin symmetry as the only relic of the flavor symmetries). Some diagrams that will contribute to $np\rightarrow d\gamma$ are shown in Figure \[fig3\].
=2.5in
It is important to realize that nucleon-nucleon scattering described by this EFT uniquely reproduces Effective Range Theory (ERT)[@ERtheory]. However, for all other observables, such as those involving electroweak gauge fields, ERT (e.g. [@ERtheory; @Noyes]) is seen to be an uncontrolled approximation to EFT (e.g. [@CRSa]).
In the following sections I attempt to indicate the status of EFT descriptions in the various energy regimes. Firstly, I will discuss low-energy $|{\bf p}|\ll m_\pi$ processes involving two and three nucleons, focusing on the recent high precision calculations that have been performed. Secondly, the issues, results and the present roadblocks to successfully describing the intermediate energy regime $|{\bf p}|{\raisebox{-0.7ex}{$\stackrel{\textstyle >}{\sim}$ }}m_\pi$ are presented. Finally, efforts to translate the understanding gained in EFT developments to many-nucleon systems are described. Such translation is necessary in order to achieve the ultimate goal of having a perturbative theory of nuclei that faithfully reproduces QCD.
$np\rightarrow d\gamma$ at Low Energies {#nprightarrow-dgamma-at-low-energies .unnumbered}
=======================================
During the past year there has been considerable focus placed on the radiative capture process $np\rightarrow d\gamma$. Firstly, it was pointed out[@BNTTa] that the uncertainty in the cross section for $np\rightarrow d\gamma$ contributed significantly to the uncertainties in the predictions of Big-Bang-Nucleosynthesis (BBN) of light element abundances. This resulted from the lack of data in the energy region important for BBN from either $np\rightarrow d\gamma$ or $\gamma d\rightarrow np$. Further, available potential model calculations of these processes[@ENDF] are undocumented and error estimates are absent. Tools have recently been developed that allow for a $\sim 1\%$ calculation of the cross section with EFT[@CSa; @Ra]. This was facilitated in part by realizing that it is advantageous to get not only the location of the deuteron pole correct, but also the normalization of the deuteron s-state component. This had long been implemented in the methods of [@Parka; @Parkeft], and implicit in the construction of Weinberg[@Weinberg; @Bira], but was only recently implemented in the dimensionally regulated EFT[@PRSa]. Finally, there are experimental efforts to measure the small isoscalar $E2_S$ and $M1_S$ amplitudes contributing to $np\rightarrow d\gamma$ using polarized neutrons on polarized protons[@nppolexpt]. At the second workshop on [Effective Field Theory in Nuclear Physics]{} held at the [*Institute for Nuclear Physics*]{} at the [*University of Washington*]{} in 1999, Mannque Rho challenged the participants to compute the $E2_S$ and $M1_S$ amplitudes with the group whose predictions are verified experimentally winning a bottle of exceptional wine. With such a wonderful prize at stake, many workshop participants redirected their efforts to this project. This is now known as the [*Rho-challenge*]{}.
$np\rightarrow d\gamma$ for Big-Bang Nucleosynthesis [^3] {#nprightarrow-dgamma-for-big-bang-nucleosynthesis .unnumbered}
---------------------------------------------------------
As existing potential model calculations of $np\rightarrow d\gamma$ are undocumented and error estimates unavailable, a $5\%$ uncertainty was assigned to the cross section as input into BBN codes[@SKMa]. It would be somewhat dismal if, after several decades of investigations into nuclear physics, the cross section for this process was uncertain at the $5\%$ level. However, EFT calculations have demonstrated that the actual uncertainty is much less than $5\%$. EFT is a well defined method of calculation and estimates of the uncertainty in a given calculation can be made by considering the magnitude of higher order terms that have been omitted. The expression for the cross section for $np\rightarrow d\gamma$ valid at the $\sim 3\%$ level (with nonrelativistic kinematics) is $$\begin{aligned}
\sigma & = & {4\pi\alpha \left(\gamma^2+|{\bf P}|^2\right)^3
\over \gamma^3 M_N^4 |{\bf P}|}
\left[\ |\tilde{X}_{M1}|^{2}
\ +\ |\tilde{X}_{E1}|^{2}\
\right]
\ \ \ .
\label{unpol}\end{aligned}$$ The isovector $M1_V$ and $E1_V$ amplitudes are $$\begin{aligned}
|\tilde{X}_{M1}|^2 & = &
{\kappa_1^2 \gamma^4 \left( {1\over a_1}-\gamma\right)^2 \over
\left({1\over a_1^2} +|{\bf P}|^2\right) \left(\gamma^2 + |{\bf P}|^2 \right)^2
}
\left[Z_d
- r_0 { \left( {\gamma\over a_1}+|{\bf P}|^2\right) |{\bf P}|^2 \over
\left({1\over a_1^2} +|{\bf P}|^2\right) \left( {1\over a_1}-\gamma\right)}
- {L_{np}\over\kappa_1}{M_N\over 2\pi}
{ \gamma^2 + |{\bf P}|^2 \over {1\over a_1}-\gamma}
\right]
\nonumber\\
|\tilde{X}_{E1}|^2 & = &
{|{\bf P}|^2 M_N^2 \gamma^4 \over \left(\gamma^2+|{\bf P}|^2\right)^4}
\left[ Z_d \ +\ {M_N\gamma\over 6\pi}
\left({\gamma^2\over 3}+|{\bf P}|^2\right)\overline{C}^{(P)}
\right]
\ \ \ ,
\label{eq:EMamps}\end{aligned}$$ where $\kappa_1$ is the isovector nucleon magnetic moment, $a_1=-23.714\pm 0.013~{\rm fm}$ is the scattering length in the ${{}^1\kern-.14em S_0}$ channel, $r_0=2.73\pm 0.03~{\rm fm}$ is the effective range in the ${{}^1\kern-.14em S_0}$ channel, $\gamma=\sqrt{M_N B}$ is the deuteron binding momentum, $\overline{C}^{(P)}$ is a number derivable from the nucleon-nucleon p-wave amplitudes, $Z_d~=~1/(1~-~\gamma~\rho_d)$ is the residue of the ${{}^3\kern-.14em S_1}$ nucleon-nucleon amplitude at the deuteron pole, and $\rho_d$ is the effective range in the ${{}^3\kern-.14em S_1}$ channel. There is also a contribution from an operator that is not related by gauge invariance to nucleon-nucleon scattering, $L_{np}$. This is the coefficient of a local operator involving four-nucleon fields and a magnetic photon. For incident neutrons with speed $|v|=2200~{\rm m/s}$ the cross section for capture by protons at rest is measured to be $\sigma^{\rm expt}_{\rm cold} = 334.2\pm 0.5~{\rm mb}$[@CWCa]. The value of $L_{np}$ is fixed by requiring that the expressions in eqs. (\[unpol\]) and (\[eq:EMamps\]) reproduce $\sigma^{\rm expt}_{\rm cold}$. The amplitudes in eq. (\[eq:EMamps\]) have been computed to one higher order by Rupak[@Ra], including relativistic effects and the appearance of an E1 counterterm, providing a $\sim 1\%$ calculation of $np\rightarrow d\gamma$. Once $L_{np}$ has been fixed, the photo-dissociation cross section $\gamma d\rightarrow n p$ can be determined, and is shown in Figure \[fig4\].
=2.0in
One finds that this relatively simple analytic expression reproduces the data very well, once the counterterm $L_{np}$ has been determined at a given energy. An idea of the convergence of the effective field theory calculation can be obtained from the numerical results of Rupak[@Ra], $$\begin{aligned}
\sigma (2~{\rm MeV}) & = &
0.0218 \left( 1 + 0.6389 + 0.0135 - 0.0053 - 0.0001 + ...\right)~{\rm fm^2}
\nonumber\\
\sigma (20~{\rm keV}) & = &
0.1917 \left( 1 + 0.1076 + 0.0001 + ...\right)~{\rm fm^2}
\ \ \ ,
\label{eq:conv}\end{aligned}$$ which are both seen to converge rapidly.
$E$ (MeV) total EFT $\sigma$ (mb)[@Ra] ENDF $\sigma$ (mb)[@ENDF]
----------------------- ------------------------------ ---------------------------
$1.264\times 10^{-8}$ $334.2^{\ (*)}$ 332.0
$5.0\times 10^{-4}$ 1.668(0) 1.660
$1.0\times 10^{-3}$ 1.172(0) 1.193
$5.0\times 10^{-3}$ 0.4982(0) 0.496
$1.0\times 10^{-2}$ 0.3324(0) 0.324
$5.0\times 10^{-2}$ 0.1081(0) 0.108
0.100 0.06352(0) 0.0633
0.500 0.0341(1) 0.0345
1.00 0.0349(4) 0.0342
: $\sigma (np\rightarrow d\gamma)$ as a function of the nucleon center-of-mass energy, $E$. The asterisk denotes an input.
\[table1\]
The cross section at various energies computed with EFT by Rupak[@Ra], along with those from the on-line nuclear data center[@ENDF] are shown in Table \[table1\]. As expected the EFT calculation agrees with the numerical values from [@ENDF] at the $\sim 1\%$ level.
Isoscalar M1 and E2 Amplitudes in $np\rightarrow d\gamma$ {#isoscalar-m1-and-e2-amplitudes-in-nprightarrow-dgamma .unnumbered}
---------------------------------------------------------
As mentioned earlier, the [*Rho Challenge*]{} focused on the $M1_S$ and $E2_S$ isoscalar amplitudes that contribute to $np\rightarrow d\gamma$, so that predictions can be compared with the imminent measurement of polarization observables[@nppolexpt]. Two works were completed soon after the challenge was issued, one by Kubodera, Park, Min, and Rho[@PKMRnp], and one by Chen, Rupak and myself[@CRSb].
There are a couple of angular distributions that can be measured in $\vec{n}+\vec{p}\rightarrow d\gamma$, but if in addition, the polarization of the $\gamma$ can be measured one finds that there is a different cross section for production of right-handed versus left-handed circularly polarized photons. Defining the asymmetry $A^\gamma (\theta)$ to be the ratio of the difference to the sum of these cross sections, $$\begin{aligned}
A^\gamma_{\eta_{\rm n}} (\theta) & = &
\eta_{\rm n} \left[\
\left( P_\gamma (M1)\ +\ P_\gamma (E2) \right)\ \cos\theta
\ +\ P_\gamma (E1)\ \sin^2\theta
\right]
\ \ \ ,
\label{eq:gamas}\end{aligned}$$ where the $P_\gamma (\Pi L)$ are combinations of the $M1_V$, $M1_S$, $E1_V$ and $E2_S$ amplitudes, and $\eta_{\rm n}$ is the neutron polarization vector.
The amplitudes and polarizations are computed in two very different ways. In [@PKMRnp] EFT wavefunctions are developed with a coordinate-space cut-off, which are then used to determine matrix elements of the various electric and magnetic multipole operators. Pions appear as dynamical degrees of freedom and determine the long-range part of the nucleon-nucleon interaction. In addition, counterterms are included via short-distance interactions (e.g. “delta-shell” and others) so that the magnetic and quadrupole moment of the deuteron are recovered. A very different construction is used in [@CRSb]. The EFT without pions is used and divergences are dimensionally regulated. As in [@PKMRnp], the four-nucleon-one photon counterterms are chosen to recover the deuteron magnetic and quadrupole moments, to give $P_\gamma (M1)\ =\ -7.1\times 10^{-4}$, $P_\gamma (E2) \ = \ -3.5\times 10^{-4}$ and a total of $P_\gamma =-1.06\times 10^{-3}$ in the forward direction, approximately $2/3$ of the experimentally determined value of[@Bazh] $P_\gamma^{\rm expt} =-(1.5\pm 0.3) \times 10^{-3}$. Given the large uncertainty in the calculation of the $M1_S$ amplitude, and the uncertainty of the measurement, the two are not inconsistent. $P_\gamma (M1)= -7.1\times 10^{-4}$ agrees with the results of Burichenko and Kriplovich[@BKa] of $P_\gamma (M1)= -7.0\times 10^{-4}$ from a Reid soft-core calculation, but is somewhat less than their zero-range calculation of $P_\gamma (M1)= -9.2\times 10^{-4}$. However, given the large uncertainty in the $M1_S$ amplitude of [@CRSb], both values are consistent. $P_\gamma (E2) = -3.5\times 10^{-4}$ calculated in [@CRSb] agrees well with that computed in [@PKMRnp], and therefore these observables do not distinguish between the two EFT methods.
Weak Interactions of the Deuteron {#weak-interactions-of-the-deuteron .unnumbered}
=================================
Weak interaction processes involving the deuteron are central to current research efforts in nuclear physics. In addition to the accelerator based programs to elucidate the flavor structure of the nucleon, such as the SAMPLE experiments[@SAMPLE] at Bates, the interactions between neutrinos and the deuteron form the core of our efforts to learn about the neutrino and look beyond the standard model of electroweak interactions. Both in production, e.g. $pp\rightarrow d e^+\nu_e$, and in detection at SNO (Sudbury Neutrino Observatory), e.g. $\nu_\mu d\rightarrow \nu_\mu np$, charged and neutral current weak interaction matrix elements between the deuteron and continuum states are required.
Much effort over the past few decades has been put into calculating the production mechanism $pp\rightarrow d e^+\nu_e$, both from standard non-relativistic quantum mechanics[@PotBah], and from sophisticated potential model techniques[@PotSch]. Recently, EFT has been applied to this process by Kong and Ravndal[@KRpp] and by Park, Kubodera, Min and Rho[@PKMRpp] giving elegant expressions and numerical values of the weak capture cross section that are consistent with previous estimates[^4]. The cross section depends somewhat on the value of a four-nucleon-one-weak-gauge-boson interaction, with coefficient $L_{1,A}$, as defined in [@BCa].
The detection reactions, $\nu d\rightarrow n p\nu$, $\nu_e d\rightarrow e^- pp$ and $\overline{\nu}_e d\rightarrow e^+ nn$ had been looked at by two groups, Ying, Haxton and Henley (YHH)[@YHH] and Kubodera and Nozawa (KN)[@KN], using sophisticated potential models. The two sets of calculations differ at the $5\%$ level, due to different treatments of meson-exchange currents (MEC). Recently, Butler and Chen[@BCa] have determined the break-up cross sections with EFT. Only $L_{1,A}$ needs to be fixed in order to perform a $\sim 1\%$ calculation. The YYH and KN numerical results can be recovered with different choices of $L_{1,A}$, as shown in Figure \[fig5\][^5],
=2.0in
confirming that the difference between the two potential-model calculations is short-distance in origin. Therefore, to predict the break-up cross section with a precision of better than $\sim 5\%$, one has two options. Firstly, compute the $\beta$-decay of tritium, and use this to determine the counterterm $L_{1,A}$ in the EFT, or equivalently the MEC’s in the potential models[^6]. Secondly, one can perform an experiment to measure one of the break-up cross sections to high accuracy, and thereby extract $L_{1,A}$, or the MEC’s. Such an experiment is currently under consideration[@Avig].
=2.0in
An important input into determining if neutrinos are changing flavor as they move out of the sun and to the earth is the ratio of charged current to neutral current cross sections. Figure \[fig6\] shows that this ratio, unlike the individual cross sections, is relatively insensitive to the counterterm $L_{1,A}$.
Low-Energy Three-Body Processes {#low-energy-three-body-processes .unnumbered}
===============================
Significant progress has been made in understanding three-body systems with EFT[@threebod; @triton]. A couple of years ago, Bedaque, Hammer and van Kolck[^7] showed very clearly that low-energy $Nd $ scattering in the $J={3\over 2}^+$ channel could be described by EFT using contact interactions alone. One of the more impressive results was the calculation of $a_{3\over 2}$, the scattering length in the quartet s-wave channel. Bedaque and van Kolck calculated the first three terms to be $a_{3\over 2} \ =\ 5.01\ +\ 1.0\ +\ 0.32\ + ...~{\rm fm}$ where the ellipses denote higher order contributions, that are estimated to be $\pm 0.1~{\rm fm}$. The calculated $a_{3\over 2}=6.32\pm 0.1~{\rm fm}$ agrees very well[^8] with the experimental $a_{3\over 2}^{\rm expt}=6.35\pm 0.02~{\rm fm}$[@a32expt]. The first term had been computed in 1957 by Skornyakov and Ter Martirosian[@SMa] while the second term was computed in 1991 by Efimov[@Efi]. The third term had not been computed before and was determined unambiguously from the nucleon-nucleon scattering amplitude [@threebod]. The results of extending this analysis to non-zero energy[@threebod] can be seen in Figure \[fig7\] and are found to agree well with data.
=2.5in
Scattering in higher partial waves has been examined by Bedaque, Gabbiani and Grießhammer[@BGGa] in the theory with only contact interactions between nucleons. A comparison between the EFT calculations[^9], sophisticated potential model calculations and data for one partial wave is shown in Figure \[fig8\].
=3.3in
-0.6in
Computations in the spin-doublet channel required much more development as local three-nucleon operators can contribute to the scattering amplitude[@triton]. The scale-dependence of the counterterm (more commonly known as the three-body force) is quite different from those that are familiar from perturbative field theory. Bedaque, Hammer and van Kolck showed that a single momentum independent counterterm could absorb all cut-off dependence from the leading operators resummed by the integral equation that describe three-body systems. The observed periodicity as a function of scale indicates that if a given calculation is performed with a given value of the cut-off, it is possible that the three-body “force” vanishes, while for a different value of the cut-off the three-body “force” may dominate. Further, the appearance of only one three-body counterterm required to render the scattering amplitude scale independent also naturally explains the [*Phillips line*]{} found in potential-models.
As a last comment on the three-body work, the techniques that have been developed in the nuclear physics setting are being applied to atomic systems, most notably scattering lengths and recombination rates in Bose condensates[@Bose].
Issues at Higher Energies {#issues-at-higher-energies .unnumbered}
=========================
The situation regarding the applicability of EFT’s at higher momentum, near or above the pion mass, is much less clear. In Weinberg’s scheme[@Weinberg], pion exchange and local four-nucleon operators contribute to nucleon-nucleon scattering at the same order in the expansion parameter. A great calculational and conceptual simplification would arise if the exchange of pions between nucleons could be treated in perturbation theory in all partial waves, as suggested by Kaplan, Wise and myself (KSW)[@KSW]. Unfortunately, one of the more disappointing results found during the past year is that the EFT with perturbative pions[@KSW] appears not to be converging [@FMSa] (also, see earlier work by Cohen and Hansen[@cohen]). Fleming, Mehen and Stewart [@FMSa] performed an analytic calculation of the NNLO amplitude for nucleon-nucleon scattering in the theory with pions and KSW power-counting[@KSW]. They found large non-analytic contributions that appear to destroy the convergence of the series. In contrast, several computations were performed with Weinberg’s power-counting [@Weinberg] for the nucleon-nucleon potential which appears to give converging amplitudes. However, the formal inconsistency of Weinbergs power-counting remains. Amplitudes are not renormalization (cut-off) independent at any order in Weinberg’s expansion, however, the cut-off dependence is found to be numerically small when renormalized at a typical strong interaction scale. Therefore, at this point in time there is no formally consistent, converging, perturbative EFT to describe nuclear interactions for momenta of order or higher than the mass of the pion.
To give you an idea of what has been attempted at these higher energies with both Weinberg and KSW power-counting, let me show you the results obtained for $\gamma d\rightarrow \gamma d$, deuteron Compton scattering. Three-orders in Weinberg’s counting have been completed for $\gamma d\rightarrow \gamma d$ at higher energies[@BMPKa]. The angular distribution of scattering photons at $E_\gamma=69~{\rm MeV}$ is shown in Figure \[fig9\][^10].
=2.5in =2.0in
One can see from Figure \[fig9\] that the expansion appears to be converging nicely to the experimental values. Similarly, $\gamma d\rightarrow \gamma d$ has been computed to two orders in KSW power-counting[@CGSSa], the results of which can be seen in Figure \[fig10\].
=5.0in
Very good agreement between data and the parameter free-prediction at NLO is found at $E_\gamma=49~{\rm MeV}$. The agreement is somewhat worse at $E_\gamma=69~{\rm MeV}$, and does not appear to be approaching the data in the same way that the calculation with Weinberg’s counting appears to. It is clear that higher order calculations must be performed once the perturbative pions versus non-perturbative pions issue is understood, and further, it is clear that more precise data is required at low-energies. It is worth mentioning at this point that neither counting schemes, nor any theoretical calculation that exists at present comes close to reproducing the recent data at $E_\gamma=95~{\rm MeV}$[@sask].
There is much work still to be done in this area.
On the Road to Nuclei {#on-the-road-to-nuclei .unnumbered}
=====================
In parallel to the efforts that I have described in the two- and three-body sectors, Haxton and collaborators[@wick] have been developing techniques to apply the ideas underpinning EFT to the nuclear shell model (efforts are ongoing by others[@others] but I will not discuss their work in this talk).
Before I discuss this work I wish to show you the results of a relatively simple but demonstrative calculation by Phillips[@PhilT] (see also [@PCa]). To show how the ideas of EFT can be translated into a potential-model mode of thinking, Phillips compared the deuteron quadrupole form factor, presented as $T_{20}$, computed with the Nijmegen93 potential[@Nij] with that generated by an effective potential, $V^L_{\rm eff} (r)$ (where $L$ denotes the orbital angular momentum state) and local quadrupole moment counterterm. The effective potential consists of one-pion exchange at long distances and a square-well at short-distances, $V^L_{\rm eff} (r)=V^{\rm OPE} (r)$ for $r>R$, and $V^L_{\rm eff} (r)=V_{0,L}$ for $r<R$. The values of the $V_{0,L}$ are chosen to reproduce the deuteron binding energy, and low-energy nucleon-nucleon scattering for each choice of $R$. Therefore, the long-distance behavior of the “true” potential and effective potential are identical. Further, the tail of the deuteron wave-function produced by the “true” potential and effective potential are identical. As one is brutalizing the nucleon-nucleon interaction at short-distances while preserving nucleon-nucleon scattering, it is expected that predictions for other observables, such as electromagnetic form factors, will deviate significantly from nature when probing distance scale comparable to or less than $R$, for reasonable values of $R$. In addition one expects to find that the static moments differ somewhat from nature. It was shown in [@CRSa; @KSWem] that such short-distance modifications can be compensated by the inclusion of gauge invariant local operators. In the case of the deuteron quadrupole moment it is necessary to introduce a four-nucleon-one-quadrupole-photon operator, that is in no way related to the operators determining nucleon-nucleon scattering. These operators are induced at the chiral symmetry breaking scale, and must be included in any consistent calculation, and in fact, their omission is responsible for the discrepancy between all sophisticated potential model calculations of the deuteron quadrupole moment[@QuadPot] and its experimental value. Phillips choose a value of this operator to reproduce the observed deuteron quadrupole moment for each value of $R$, and then predicted $T_{20}^{\rm eff}$ in the effective theory, the results of which are shown in Figure \[fig11\][^11].
=2.5in =3.5in
It is clear from Figure \[fig11\] that by fixing parameters in the effective theory to reproduce low-energy observables that, even with a nucleon-nucleon potential that has been brutalized at short-distances, one can essentially recover the “true” form factor over a quite impressive range of momentum transfers. This provides a very clear demonstration that $T_{20}$ is determined largely by the tail of the deuteron wavefunction, the OPE tail of the nucleon-nucleon interaction and the deuteron quadrupole moment. For potential model calculations to accurately determine $T_{20}$, they must first recover the deuteron quadrupole moment, and to do so they must include a four-nucleon quadrupole operator[@CRSa; @KSWem].
A very similar exploration is ongoing by Haxton and collaborators [@wick]. They are attempting to construct a model-space-independent shell model, and are presently focusing on the deuteron to optimize their techniques. The underpinnings of EFT are basis independent, and as such should be able to be implemented in both a plane-wave basis (PWB) or a harmonic oscillator basis (HOB). For a continuum process, obviously the PWB is preferable, but for a bound state it seems reasonable that a bound state basis, such as the HOB, will be optimal. In an HOB basis it is natural to integrate out the levels of the HO step by step to allow for an easier calculation in a reduced model-space. As each HO level is removed, the hamiltonian and coefficients of gauge invariant operators are redefined to preserve observables. This is a discrete analog of the renormalization group (RG) implemented in the PWB.
=4.0in
-2in
In Figure \[fig12\][^12] the magnetic form factor of the deuteron $F_{M1} (q)$ is shown versus momentum transfer, where the $n=50$ calculation provides the “true” calculation. The other curves correspond to the same calculation, but with the insertion of the bare $M1$ operator in each reduced model space. The solid line, however, is not just the $n=50$ calculation but also the calculation from ALL reduced model-spaces when the renormalized $M1$ operator is inserted and NOT the bare operator. Clearly, a discrete RG can be implemented in an HOB.
One of the advantages of constructing a discrete RG for the nuclear shell model would be to greatly reduce the computer time required to compute matrix elements in a nucleus with $A\gg 2$. Presently, efforts are being made to reduce the shell-model space for the deuteron down from $n=140$, which reproduces the deuteron binding energy perfectly (by construction), down to $n=20$ or $30$ and faithfully reproduce all deuteron observables. Part of the effort is to include the short-range part of the nucleon-nucleon interaction by local operators. If successful, this program will allow for high precision computations of nuclear properties, with greatly increased speed.
Discussion {#discussion .unnumbered}
==========
I have tried to give you an overview of a number of important developments of the last year or so. At low-energies, high precision calculations, $\sim 1\%$ have been performed in the two-body sector, and progress is being made toward calculations of similar precision in the three-body sector.
At somewhat higher energies, a formally consistent and converging EFT describing nucleon-nucleon interactions is yet to be uncovered. Weinberg’s power-counting is formally ill-defined, yet gives numerical results that appear to be converging. In contrast, KSW power-counting is formally well-defined, yet appears not to be converging! I suspect, as do others, that some sort of union between the two power-countings may in fact be the correct one, but this is merely speculation.
Some of the more interesting developments this year were made in the implementation of EFT ideas in nuclear many-body calculations. There are indications that the EFT techniques may provide a means to compute properties of nuclei presently beyond reach. However, more work is required before any conclusions can be drawn.
G. P. Lepage, Summary talk presented at the INT workshop on [*Effective Field Theory in Nuclear Physics II*]{}, edited by P.F. Bedaque, M.J. Savage, R. Seki and U. van Kolck, ISBN-981-02-4181-X.
S. Weinberg, [*Phys. Lett.*]{} [**B251**]{}, 288 (1990); [*Nucl. Phys.*]{} [**B363**]{}, 3 (1991); [*Phys. Lett.*]{} [**B295**]{}, 114 (1992).
C. Ordonez and U. van Kolck, [*Phys. Lett.*]{} [**B291**]{}, 459 (1992); C. Ordonez, L. Ray and U. van Kolck, [*Phys. Rev. Lett.*]{} [**72**]{}, 1982 (1994); [*Phys. Rev.*]{} [**C53**]{}, 2086 (1996); U. van Kolck, [*Phys. Rev.*]{} [**C49**]{}, 2932 (1994).
T.S. Park, D.P. Min and M. Rho, [*Phys. Rev. Lett.*]{} [**74**]{}, 4153 (1995); [*Nucl. Phys.*]{} [**A596**]{}, 515 (1996).
D.B. Kaplan, M.J. Savage and M.B. Wise, [*Nucl. Phys.*]{} [**B478**]{}, 629 (1996).
D.B. Kaplan, M.J. Savage and M.B. Wise, [*Phys. Lett.*]{} [**B424**]{}, 390 (1998); [*Nucl. Phys.*]{} [**B534**]{}, 329 (1998).
H. A. Bethe, [*Phys. Rev.*]{} [**76**]{}, 38 (1949); H. A. Bethe and C. Longmire, [*Phys. Rev.*]{} [**77**]{}, 647 (1950).
H. P. Noyes, [*Nucl. Phys.*]{} [**74**]{}, 508 (1965).
J.-W. Chen, G. Rupak and M. J. Savage, [*Nucl. Phys.* ]{} [**A653**]{} 386 (1999).
S. Burles, K. M. Nollet, J. W. Truran and M. S. Turner, [*Phys. Rev. Lett.*]{} [**82**]{}, 4176 (1999).
ENDF online database at the NNDC Online Data Service, [http://www.nndc.bnl.gov]{}.
D. R. Phillips, G. Rupak, M. J. Savage [*Phys. Lett.*]{} [**B473**]{}, 209 (2000).
J.-W. Chen and M. J. Savage, [*Phys. Rev.*]{} [**C60**]{}, 065205 (1999).
G. Rupak, [nucl-th/9911018]{}.
T.-S. Park, D.-P. Min and M. Rho, [*Phys. Rev. Lett.*]{} [**74**]{}, 4153 (1995); [*Nucl. Phys.* ]{} [**A596**]{} 515 (1996).
T.-S. Park, K. Kubodera, D.-P. Min, and M. Rho, [*Astrophys. Jour.*]{} [**507**]{}, 443 (1998); [*Phys. Rev.*]{} [**C58**]{}, R637 (1998).
C. Ordonez and U. van Kolck, [*Phys. Lett.*]{} [**B291**]{}, 459 (1992); C. Ordonez, L. Ray and U. van Kolck, [*Phys. Rev. Lett.*]{} [**72**]{}, 1982 (1994); [*Phys. Rev.* ]{} [**C53**]{}, 2086 (1996); U. van Kolck, [*Phys. Rev.* ]{} [**C49**]{}, 2932 (1994).
T. M. Muller, [*Private Communication*]{}.
M. S. Smith, L. H. Kawano and R. A. Malaney, [*Astrophys. J. Suppl. Ser.*]{} [**85**]{} 219 (1993).
A.E. Cox, S.A.R. Wynchank and C.H. Collie, [*Nucl. Phys.*]{} [**74**]{}, 497 (1965).
T.-S. Park, K. Kubodera, D.-P. Min, and M. Rho, [*Phys. Lett.*]{} [**B472**]{}, 232 (2000).
J.-W Chen, G. Rupak and M. J. Savage, [*Phys. Lett.*]{} [**B464**]{}, 1 (1999).
A. N. Bazhenov [*et al*]{}., [*Phys. Lett.*]{} [**B289**]{}, 17 (1992).
A. P. Burichenko and I. B. Khriplovich, [*Nucl. Phys.*]{} [**A515**]{}, 139 (1990).
D. T. Spayde [*et al.*]{} (SAMPLE Collaboration) [*Phys. Rev. Lett.*]{} [**84**]{}, 1106 (2000).
J. N. Bahcall and R. M. May, [*Ap. J*]{} [**55**]{}, 501 (1969).
R. Schiavilla [*et al*]{}., [*Phys. Rev.*]{} [**C58**]{} 1263 (1998).
X. Kong and F. Ravndal, [nucl-th/0004038]{}; [*Nucl. Phys.*]{} [**A665**]{} 137 (2000); [*Nucl. Phys.*]{} [**A656**]{} 421 (1999); [*Phys. Lett.*]{} [**B470**]{}, 1 (1999).
T.- S. Park, K. Kubodera, D.- P Min and M. Rho, [*Nucl. Phys.*]{} [**A646**]{}, 83 (1999).
A. N. Ivanov, H. Oberhummer, N. I. Troitskaya, and M. Faber [nucl-th/9910021]{} and references therein.
M. N. Butler and J.-W Chen, [nucl-th/9905059]{}.
S. Ying, W. C. Haxton and E. M. Henley, [*Phys. Rev.*]{} [**C45**]{}, 1982 (1992); [*Phys. Rev.*]{} [**D40**]{}, 3211 (1989).
K. Kubodera and S. Nozawa, [*Int. J. Mod. Phys.*]{} [**E3**]{}, 101 (1994); Y. Kohyama and K. Kubodera, USC(NT)-report-92-1, unpublished.
F. Avignone, private communication.
P. F. Bedaque and H. W. Grießhammer, [*Nucl. Phys.*]{} [**A671**]{}, 357 (2000); P.F. Bedaque, H.W. Hammer and U. van Kolck, [*Phys. Rev. Lett.*]{} [**82**]{}, 463 (1999); [*Nucl. Phys.*]{} [**A646**]{}, 444 (1999); [*Phys. Rev.*]{} [**C58**]{}, R641 (1998). P.F. Bedaque and U. van Kolck, [*Phys. Lett.*]{} [**B428**]{}, 221 (1998).
P.F. Bedaque, H.W. Hammer and U. van Kolck, [nucl-th/9906032]{}.
F. Gabbiani, P. F. Bedaque and H. W. Grießhammer, [nucl-th/9911034]{}.
J. L. Friar, D. Huber, H. Witala and G. L. Payne, [*Acta Phys. Polon.*]{} [**B31**]{}, 749 (2000).
W. Dilg, L. Koester and W. Nistler, [*Phys. Lett.*]{} [**B36**]{}, 208 (1971).
G. V. Skornyakov and K. A. Ter-Martirosian, [*Sov. Phys. JETP*]{} [**4**]{}, 648 (1957).
V. Efimov, [*Phys. Rev.*]{} [**C47**]{}, 1876 (1993).
A. Kievsky, S. Rosati, W. Tornow and M. Viviani, [*Nucl. Phys.*]{} [**A607**]{}, 402 (1996).
E. Huttel, W. Arnold, H. Baumgart, H. Berg and G. Clausnitzer, [*Nucl. Phys.*]{} [**A406**]{}, 443 (1983); P. A. Schmelzbach, W. Grubler, R. E. White, V. Konig, R. Risler and P. Marmier, [*Nucl. Phys.*]{} [**A197**]{}, 273 (1972).
W. Tornow and H. Witala, “Proton-Deuteron Phase-Shift Analysis above the Deuteron Breakup Threshold”, Technical Report TUNL XXXVI (1996-97).
P. F. Bedaque, E. Braaten and H. W. Hammer, [cond-mat/0002365]{}. P. F. Bedaque, H. W. Hammer, and U. van Kolck, [*Nucl. Phys.*]{} [**A646**]{}, 444 (1999).
S. Fleming, T. Mehen and I. W. Stewart, [nucl-th/9911001]{}; [*Phys. Rev.*]{} [**C61**]{}, 044005 (2000).
T. D. Cohen and J. M. Hansen, [nucl-th/9908049]{}; [*Phys. Rev.*]{} [**C59**]{}, 3047 (1999); [*Phys. Rev.*]{} [**C59**]{}, 13 (1999); T. D. Cohen, [nucl-th/9904052]{}
S. R. Beane, M. Malheiro, D. R. Phillips and U. van Kolck, [*Nucl. Phys.*]{} [**A656**]{}, 367 (1999); private communication.
M. A. Lucas, Ph. D. thesis, University of Illinois at Urbana-Champaign (1994)
J.-W. Chen, H. W. Grießhammer, M. J. Savage and R. P. Springer, [*Nucl. Phys*]{} [**A644**]{}, 245 (1998).
D. L. Hornidge [*et al.*]{}, [*Phys. Rev. Lett.*]{} [**84**]{}, 2334 (2000).
W. C. Haxton and C. L. Song, [nucl-th/9907097]{}; [nucl-th/9906082]{}.
E. Epelbaoum, W. Glockle and U.-G. Meissner, [*Nucl. Phys.*]{} [**A671**]{}, 295 (2000); [*Few Body Syst. Suppl.*]{} [**10**]{}, 479 (1999); [*Phys. Lett.*]{} [**B439**]{}, 1 (1998); E. Epelbaoum, W. Glockle, A. Kruger and U.-G. Meissner, [*Nucl. Phys.*]{} [**A645**]{}, 413 (1999); M. Lutz, [9906028]{}; H. W. Hammer and R. J. Furnstahl, [nucl-th/0004043]{}.
D. R. Phillips, [nucl-th/0004060]{}.
D. R. Phillips and T. D. Cohen, [*Nucl. Phys.*]{} [**A668**]{}, 45 (2000).
V. G. J. Stoks, R. A. M. Klomp, C. P. F. Terheggen and J. J. de Swart, [*Phys. Rev.*]{} [**C49**]{}, 2950 (1994).
D. B. Kaplan, M. J. Savage and M. B. Wise, [*Phys. Rev.*]{} [**C59**]{}, 617 (1999).
R. Machleidt, [nucl-th/0006014]{}; R. B. Wiringa, V.G.J. Stoks, R. Schiavilla [*Phys. Rev.*]{} [**C51**]{}, 38 (1995).
[^1]: Talk presented at the [*7th Conference on the Intersections of Particle and Nuclear Physics*]{}, Quebec City, Canada, May 22-28, 2000. NT@UW-00-018.
[^2]: During the [*Effective Field Theory*]{} workshop to be held at the [*Institute for Nuclear Physics*]{} at the [*University of Washington*]{} during the summer of 2000, efforts will be made to estimate the computer resources necessary to determine the deuteron binding energy from lattice QCD [@Lepage].
[^3]: I thank Gautam Rupak for allowing me to present his results in this section
[^4]: There has also been recent work in [@Ivan] that I have so far failed to comprehend.
[^5]: I thank Malcolm Butler and Jiunn-Wei Chen for allowing me to reproduce their figures
[^6]: This method of fixing the MEC’s has already been implemented for $pp\rightarrow d e^+\nu_e$[@PotSch].
[^7]: I thank Paulo Bedaque, Hans-Werner Hammer and Bira van Kolck for allowing me to reproduce their figure.
[^8]: Subsequent “second generation” potential-model calculations agree with this result[@FHWPa].
[^9]: I thank Paulo Bedaque, Fabrizio Gabbiani and Harald Grießhammer for allowing me to reproduce their figure.
[^10]: I thank Daniel Phillips, Silas Beane and Bira van Kolck for allowing me to reproduce their figure.
[^11]: I thank Daniel Phillips for allowing me to reproduce his figure.
[^12]: I thank Wick Haxton for allowing me to reproduce his figure.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Let $T$ be a consistent o-minimal theory extending the theory of densely ordered groups and let $T''$ be a consistent theory. Then there is a complete theory $T^*$ extending $T$ such that $T$ is an open core of $T^*$, but every model of $T^*$ interprets a model of $T''$. If $T''$ is NIP, $T^*$ can be chosen to be NIP as well. From this we deduce the existence of an NIP expansion of the real field that has no distal expansion.'
address: |
Department of Mathematics\
University of Illinois at Urbana-Champaign\
1409 West Green Street\
Urbana, IL 61801
author:
- Philipp Hieronymi
- Travis Nell
- Erik Walsberg
bibliography:
- 'HW-Bib.bib'
date:
-
-
title: 'Wild theories with o-minimal open core'
---
[^1]
Introduction
============
Let ${\ensuremath{\mathcal{R}}}$ be an expansion of a dense linear order $(R,<)$ without endpoints. The **open core of ${\ensuremath{\mathcal{R}}}$**, denoted by ${\ensuremath{\mathcal{R}}}^{\circ}$, is the structure $(R,(U))$, where $U$ ranges over all open sets of all arities definable in ${\ensuremath{\mathcal{R}}}$. Miller and Speissegger introduced this notion of an open core for expansions of $({\mathbb{R}},<)$ in [@MS99], and established sufficient conditions on ${\ensuremath{\mathcal{R}}}$ such that its open core is o-minimal. Here we want to answer the following question:
*Is there any restriction on what kind of structures can be interpreted in an expansion of $(R,<)$ with o-minimal open core?*
This question, although formulated slightly differently, was already asked by Dolich, Miller and Steinhorn in a preprint version of [@DMS-Indepedent]. Our answer is negative. To give a precise statement of our result, we need to recall the notion of an open core of a theory as introduced in [@DMS1]. Let $T^*$ be a theory extending the theory of dense linear orders without endpoints in a language ${\ensuremath{\mathcal{L}}}^*\supseteq \{ < \}$, and let $T$ be another theory in a language ${\ensuremath{\mathcal{L}}}$. We say that **$T$ is an open core of $T^*$** if for every ${\ensuremath{\mathcal{N}}}\models T^*$ there is ${\ensuremath{\mathcal{M}}} \models T$ such that ${\ensuremath{\mathcal{N}}}^{\circ}$ is interdefinable with ${\ensuremath{\mathcal{M}}}$.
Let $T$ be a consistent o-minimal theory extending the theory of densely ordered groups and let $T'$ be a consistent theory. Then there is a complete theory $T^*$ extending $T$ such that
1. $T^*$ interprets a model of $T'$,
2. $T$ is an open core of $T^*$,
3. $T^*$ is NIP if $T'$ is NIP,
4. $T^*$ is strongly dependent if $T'$ is strongly dependent.
Statements (3) and (4) of Theorem A indicate that we can choose $T^*$ in such a way that not only the open core of $T^*$ is o-minimal, but also $T^*$ remains tame in the sense of Shelah’s combinatorial tameness notions. For definitions of NIP and strong dependence, we refer the reader to Simon [@Simon-Book].
We will deduce the following analogue for o-minimal expansions of the ordered real additive group from the proof of Theorem A.
Let ${\ensuremath{\mathcal{R}}}$ be an o-minimal expansion of $({\mathbb{R}},<,+)$ in a language ${\ensuremath{\mathcal{L}}}$ and let $T'$ be a consistent theory such that $|{\ensuremath{\mathcal{L}}}|<|{\mathbb{R}}|$ and $|T'|\leq |{\mathbb{R}}|$. Then there exists an expansion ${\ensuremath{\mathcal{S}}}$ of ${\ensuremath{\mathcal{R}}}$ such that
1. ${\ensuremath{\mathcal{S}}}$ interprets a model of $T'$,
2. the open core of ${\ensuremath{\mathcal{S}}}$ is interdefinable with ${\ensuremath{\mathcal{R}}}$,
3. ${\ensuremath{\mathcal{S}}}$ is NIP if $T'$ is NIP,
4. ${\ensuremath{\mathcal{S}}}$ is strongly dependent if $T'$ is strongly dependent.
We will deduce from work in [@DMS1] that an expansion of $({\mathbb{R}},<,+)$ has o-minimal open core if and only if it does not define a discrete linear order. Therefore in Theorem B the statement *${\ensuremath{\mathcal{S}}}$ interprets a model of $T'$* cannot be replaced by the statement *${\ensuremath{\mathcal{S}}}$ defines a model of $T'$*.
The outline of the proof of the above results is as follows. For simplicity, let ${\ensuremath{\mathcal{R}}}$ be $({\mathbb{R}},<,+)$ and let $T'$ be a consistent theory in a countable language ${\ensuremath{\mathcal{L}}}'$ with an infinite model. Take a dense basis $P$ of ${\mathbb{R}}$ as a ${\mathbb{Q}}$-vector space. By [@DMS-Indepedent 2.25] the open core of the structure $({\mathbb{R}},<,+,P)$ is ${\ensuremath{\mathcal{R}}}$. We further expand $({\mathbb{R}},<,+,P)$ by a binary predicate $E$ such that $E$ is an equivalence relation on $P$, has countably many equivalence classes and each equivalence class of $E$ is dense in $P$. Now take a countable model $M$ of $T'$ and expand $({\mathbb{R}},<,+,P,E)$ to an expansion ${\ensuremath{\mathcal{S}}}$ such that the quotient $P/E$ becomes an ${\ensuremath{\mathcal{L}}}'$-structure that is isomorphic to $M$. Since each equivalence class of $E$ is dense in $P$ and hence in ${\mathbb{R}}$, we can define this *fusion* ${\ensuremath{\mathcal{S}}}$ of $({\mathbb{R}},<,+,P,E)$ and $M$ in a way that the open core of the resulting structure ${\ensuremath{\mathcal{S}}}$ is still ${\ensuremath{\mathcal{R}}}$. Indeed we use ideas and techniques from [@DMS-Indepedent] to prove a quantifier-elimination result for ${\ensuremath{\mathcal{S}}}$ analogous to the one of $({\mathbb{R}},<,+,P)$ (see [@DMS-Indepedent 2.9]), and from that deduce that the open core of ${\ensuremath{\mathcal{S}}}$ is ${\ensuremath{\mathcal{R}}}$.
In the special case that ${\ensuremath{\mathcal{L}}}'$ is empty and $T'$ is the theory of infinite sets, the construction we outlined above gives the following extension of the results from [@DMS-Indepedent].
Let $T$ be a complete o-minimal theory extending the theory of densely ordered groups in a language ${\ensuremath{\mathcal{L}}}$, and let ${\ensuremath{\mathcal{L}}}_e$ be the language ${\ensuremath{\mathcal{L}}}$ augmented by a unary predicate $P$ and a binary predicate $E$. Let $T_{e,\infty}$ be the ${\ensuremath{\mathcal{L}}}_e$-theory containing $T$ and axiom schemata expressing the following statements:
- $P$ is dense and ${\operatorname{dcl}}_T$-independent,
- $E \subseteq P^2$ is an equivalence relation on $P$,
- each equivalence class of $E$ is dense in $P$,
- $E$ has infinitely many equivalence classes.
Then $T_{e,\infty}$ is complete, and $T$ is an open core of $T_{e,\infty}$.
Theorem B should be compared to Friedman and Miller [@FM-Sparse Theorem A]. Among other things, the latter result implies the existence of an expansion of the real field that defines a model of first-order arithmetic, but every subset of ${\mathbb{R}}$ definable in this expansion is a finite union of an open set and finitely many discrete sets. Therefore both our result and [@FM-Sparse] describe situations in which topological tameness exists without model-theoretic tameness.
In general our results rule out that the property of having an o-minimal open core has any consequences in terms of model-theoretic tameness of the whole structure. At first glance this might look like a disappointing result. However, we do not share this viewpoint. We regard our results as further evidence that in model-theoretically wild situations geometric tameness can often prevail. In some of those situations the open core of a structure or theory seems to be the right tool that can capture precisely this tameness, making certain phenomena trackable by model-theoretic analysis.
Theorem B(3) has a few interesting corollaries about NIP expansions of $({\mathbb{R}},<,+)$. First of all, it states that for every NIP theory $T'$ of cardinality at most continuum there is an NIP expansion of $({\mathbb{R}},<,+)$ that interprets a model of $T'$. Therefore the model theory of NIP expansions of $({\mathbb{R}},<,+)$ is in general as complicated as the model theory of arbitrary NIP theories. We use this observation to deduce a new result about the distality of NIP expansions of $({\mathbb{R}},<,+)$. The notion of distality was introduced by Simon in [@Simon-Distal] to single out those NIP theories and structures that can be considered purely unstable. While every o-minimal expansions of $({\mathbb{R}},<,+)$ is distal, there are several natural examples of non-distal NIP expansions of $({\mathbb{R}},<,+)$ (see [@HN-Distal]). However, by Chernikov and Starchenko [@CS-RegDis] even just having a distal expansion guarantees certain desirable combinatorial properties of definable sets (the strong Erdös-Hajnal property). Therefore it is interesting to know whether or not all NIP expansions of $({\mathbb{R}},<,+)$ have a distal expansion. Although we do not know it, we expect all examples of non-distal NIP expansions of $({\mathbb{R}},<,+)$ produced in [@HN-Distal] to have distal expansions. So far the only known NIP theory without an distal expansion is the theory of algebraically closed fields of characteristic $p$ by [@CS-RegDis Proposition 6.2]. Combining this with Theorem B, we almost immediately obtain the following.
There is an NIP expansion of $({\mathbb{R}},<,+)$ that does not have a distal expansion.
This is also the first example of an NIP expansion of any densely ordered set that does not have a distal expansion.
While in general for every countable NIP theory there is an expansion of $({\mathbb{R}},<,+)$ that interprets a model of this theory, there is a natural class of expansions of $({\mathbb{R}},<,+)$ in which models of certain NIP theories in countable languages cannot be interpreted. A set $X\subseteq {\mathbb{R}}$ is somewhere dense and co-dense if there is an open interval $I$ such that $X\cap I$ is dense and co-dense in $I$. We say an expansion of $({\mathbb{R}},<)$ is **noiseless** if it does not define a somewhere dense and co-dense subset of ${\mathbb{R}}$.[^2] The expansion ${\ensuremath{\mathcal{S}}}$ we produce for Theorem B is not noiseless. It is therefore natural to ask whether in Theorem B we can require ${\ensuremath{\mathcal{S}}}$ to be noiseless. The answer to this question is negative.
Let ${\ensuremath{\mathcal{R}}}$ be a noiseless NIP expansion of $({\mathbb{R}},<,+,1)$. Then ${\ensuremath{\mathcal{R}}}$ has definable choice, that is: for $A\subseteq {\mathbb{R}}^{m}\times {\mathbb{R}}^n$ $\emptyset$-definable in ${\ensuremath{\mathcal{R}}}$ there is an $\emptyset$-definable function $f: \pi(A) \to {\mathbb{R}}^n$ such that
1. ${\operatorname{gr}}(f)\subseteq A$,
2. $f(a) = f(b)$ whenever $a,b\in \pi(A)$ and $A_a=A_b$,
where $\pi : {\mathbb{R}}^{m+n} \to {\mathbb{R}}^m$ is the projection onto the first $m$ coordinates.
It follows from Theorem E that if a noiseless NIP expansion ${\ensuremath{\mathcal{R}}}$ of $({\mathbb{R}},<,+,1)$ interprets a structure ${\ensuremath{\mathcal{M}}}$, then ${\ensuremath{\mathcal{R}}}$ defines an isomorphic copy of ${\ensuremath{\mathcal{M}}}$. We will prove Theorem E in greater generality. In particular, Theorem E not only holds for noiseless NIP expansions, but also for noiseless NTP$_2$ expansions (for a definition of NTP$_2$ see [@Simon-Book]).
We now show that Theorem B fails when we require ${\ensuremath{\mathcal{S}}}$ to be noiseless. Let $p$ be a prime and $\mathbb{F}_p$ be the field with $p$ elements. By Shelah and Simon [@shelahsimon Theorem 2.1] if ${\ensuremath{\mathcal{V}}} = (V, +, \ldots)$ is an infinite $\mathbb{F}_p$-vector space and $\prec$ is a linear order on $V$, then $({\ensuremath{\mathcal{V}}}, \prec)$ has IP. Suppose now that ${\ensuremath{\mathcal{M}}} = (M, <, \ldots)$ is an expansion of an infinite linear order $(M,<)$ and that ${\ensuremath{\mathcal{V}}}$ is an ${\ensuremath{\mathcal{M}}}$-definable infinite $\mathbb{F}_{p}$-vector space with underlying set $V \subseteq M^k$. The lexicographic order on $M^k$ induced by $<$ is linear and induces a linear order on $V$. It follows that ${\ensuremath{\mathcal{M}}}$ has IP. Thus no NIP expansion of a linear order defines an infinite $\mathbb{F}_p$-vector space. By Theorem E no noiseless NIP expansion of $({\mathbb{R}},<,+,1)$ interprets an infinite vector space over a finite field.
Open questions {#open-questions .unnumbered}
--------------
We end the introduction with a few open questions.
**1.** We work here in the context of ordered structures and o-minimal open core. It is likely that our techniques can be used to extend our results to various other settings. In particular, by using the technology from Berenstein and Vassiliev [@BV-Selecta] rather than from [@DMS-Indepedent] one should be able produce analogues of Theorem A and B for other geometric structures such as the field of $p$-adic numbers.
**2.** Similar questions can be asked about NIP expansions of $({\mathbb{N}},<)$. Since every such expansion has definable Skolem functions, we again have some limitations on what kind of theories can be interpreted in such a structure. Can we say anything more? For example: can an NIP expansion of $({\mathbb{N}},<)$ interpret an infinite field? Is there an NIP expansion of $({\mathbb{N}},<)$ that does not admit a distal expansion?
**3.** Is there a noiseless NIP expansion of $({\mathbb{R}},<,+)$ that does not admit a distal expansion? Is every infinite field interpretable in a noiseless NIP expansion isomorphic to $({\mathbb{R}},+,\cdot)$ or $(\mathbb{C}, +, \cdot)$?
The previous question is even open for d-minimal NIP expansions, a subclass of the class of noiseless NIP expansions (see [@Miller-tame] for a definition of d-minimality). It follows from Fornasiero [@F-dmingroup Theorem 4.13] that any uncountable field interpretable in a d-minimal expansion is isomorphic to $({\mathbb{R}},+,\cdot)$ or $(\mathbb{C},+,\cdot)$. Thus in this setting it suffices to show that no d-minimal NIP expansion interprets a countable field. It is not difficult to show that any countable set definable in a d-minimal expansion admits a definable order with order type $\omega$. Thus, if the above question about the interpretability of infinite fields in NIP expansions of $({\mathbb{N}},<)$ has a negative answer, then any infinite field interpretable in a d-minimal NIP expansion of $({\mathbb{R}},<,+)$ is isomorphic to $({\mathbb{R}},+,\cdot)$ or $(\mathbb{C},+,\cdot)$.
Is every noiseless NIP expansion of $({\mathbb{R}},<,+)$ d-minimal? We doubt that this statement is true, but it seems difficult to produce a counterexample.
Acknowledgments {#acknowledgments .unnumbered}
---------------
The authors thank Antongiulio Fornasiero and Chris Miller for helpful conversations around the topic of this paper.
Notation {#notation .unnumbered}
--------
We will use $m,n$ for natural numbers and $\kappa$ for a cardinal. Let $X,Y$ be sets. We denote the cardinality of $X$ by $|X|$. For a function $f: X\to Y$, we denote the graph of $f$ by ${\operatorname{gr}}(f)$. If $Z\subseteq X \times Y$ and $x\in X$, then $Z_x$ denotes the set $\{ y\in Y \ : \ (x,y) \in Z\}$. If $a=(a_1,\dots,a_n)$, we sometimes write $Xa$ for $X\cup \{a_1,\dots,a_n\}$, and $XY$ for $X\cup Y$.Let ${\ensuremath{\mathcal{L}}}$ be a language and $T$ an ${\ensuremath{\mathcal{L}}}$-theory. Let $M \models T$ and $A\subseteq M$. In this situation, ${\ensuremath{\mathcal{L}}}$-definable always means ${\ensuremath{\mathcal{L}}}$-definable with parameters. If we want to be precise about the parameters we write ${\ensuremath{\mathcal{L}}}$-$A$-definable to indicate ${\ensuremath{\mathcal{L}}}$-definability with parameters from $A$. Let $b\in M^n$. Then we write ${\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}}(b|A)$ for the ${\ensuremath{\mathcal{L}}}$-type of $b$ over $A$. Moreover, ${\operatorname{dcl}}_T(A)$ denotes the definable closure of $A$ in $M$. Whenever $T$ is o-minimal, ${\operatorname{dcl}}_T$ is a pregeometry.
The fusion {#Section:Fusion}
==========
Let $T$ be a consistent o-minimal theory extending the theory of densely ordered groups with a distinguished positive element, and let ${\ensuremath{\mathcal{L}}}$ be its language. Let ${\ensuremath{\mathcal{L}}}'$ be a relational language disjoint from ${\ensuremath{\mathcal{L}}}$, and let $T'$ be a consistent ${\ensuremath{\mathcal{L}}}'$-theory. In this section we will construct a language ${\ensuremath{\mathcal{L}}}^* \supseteq {\ensuremath{\mathcal{L}}}$ and a complete ${\ensuremath{\mathcal{L}}}^*$-theory $T^*$ extending $T$ such that $T$ is an open core of $T^*$ and $T^*$ interprets $T'$. In Section \[Section:NIP\] we show that $T^*$ is NIP whenever $T$ is, and in Section \[Section:STDEP\] we prove that strong dependence of $T'$ implies strong dependence of $T^*$.
By replacing $T$ by a completion of $T$ and $T'$ by a completion of $T'$, we can directly reduce to the case that both $T$ and $T'$ are complete. So from now, we assume that $T$ and $T'$ are complete.
Let ${\ensuremath{\mathcal{L}}}_e$ be ${\ensuremath{\mathcal{L}}}$ expanded by a unary predicate $P$ and a binary predicate $E$ such that neither $P$ nor $E$ are not in ${\ensuremath{\mathcal{L}}}'$. Let $T_{e}$ be the extension of $T$ by axiom schemata expressing the following statements:
- $P$ is dense and ${\operatorname{dcl}}_T$-independent,
- $E \subseteq P^2$ is an equivalence relation on $P$,
- each equivalence class of $E$ is dense in $P$.
Let ${\ensuremath{\mathcal{L}}}^* = {\ensuremath{\mathcal{L}}}_e \cup {\ensuremath{\mathcal{L}}}'$. For a given ${\ensuremath{\mathcal{L}}}'$-formula $\theta$ we define a ${\ensuremath{\mathcal{L}}}^*$-formula $\theta_e$ recursively as follow: $$\begin{aligned}
&\hbox{if $\theta$ is $x=y$, then define $\theta_e$ as $Exy$,}\\
&\hbox{if $\theta$ is $Rx_1\dots x_n$ where $R$ is an $n$-ary predicate in ${\ensuremath{\mathcal{L}}}'$, then define $\theta_e$ as $Rx_1\dots x_n$,}\\
&\hbox{if $\theta$ is $\neg\theta'$, then define $\theta_e$ as $\neg \theta'_e$,} \\
&\hbox{if $\theta$ is $\theta'\wedge \theta''$, then define $\theta_e$ as $\theta'_e\wedge \theta''_e$,} \\ \displaybreak
&\hbox{if $\theta$ is $\theta'\vee \theta''$, then define $\theta_e$ as $\theta'_e\vee \theta''_e$,} \\
&\hbox{if $\theta$ is $\exists x \theta'$, then define $\theta_e$ as $\exists x (Px \wedge \theta'_e),$}\\
&\hbox{if $\theta$ is $\forall x \theta'$, then define $\theta_e$ as $\forall x (Px \rightarrow \theta'_e).$}\end{aligned}$$
Let $T^*$ be the extension of $T_e$ by the following axiom schemata:
- $R\subseteq P^n$ and $$\forall x_1 \forall y_1\dots \forall x_n \forall y_n \ \left( \bigwedge_{i=1}^n Ex_iy_i \right) \rightarrow \Big(Rx_1\dots x_n \leftrightarrow Ry_1\dots y_n\Big)$$ for every $R \in {\ensuremath{\mathcal{L}}}'$ with $\operatorname{ar}(R)=n$,
- $\varphi_e$ for every $\varphi \in T'$.
We now fix some further notation. Given a model ${\ensuremath{\mathcal{M}}}$ of $T^*$, we will denote the underlying model of $T$ by $M$, the interpretation of $P$ and $E$ by $P_{{\ensuremath{\mathcal{M}}}}$ and $E_{\mathcal M}$. For $b\in P_{{\ensuremath{\mathcal{M}}}}^n$ and $A\subseteq P_{{\ensuremath{\mathcal{M}}}}$ we denote by ${\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}'}(b|A)$ the set of all ${\ensuremath{\mathcal{L}}}^*$-formulas of the form $\varphi_e(x,a)$ for some ${\ensuremath{\mathcal{L}}}'$-formula $\varphi(x,y)$ such that $a \in A^m$ and ${\ensuremath{\mathcal{M}}} \models \varphi_e(b,a)$.
A standard induction on ${\ensuremath{\mathcal{L}}}'$-formulas together with Axiom (T4) gives the following.
\[lem:sametype\] Let ${\ensuremath{\mathcal{M}}}\models T^*$, $a,b \in P_{{\ensuremath{\mathcal{M}}}}$ and $A\subseteq P_{{\ensuremath{\mathcal{M}}}}$ . If $(a,b) \in E_{{\ensuremath{\mathcal{M}}}}$, then ${\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}'}(a|A)={\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}'}(b|A)$.
We now show that given a model of $T$ with enough ${\operatorname{dcl}}_T$-independent elements, this model can be expanded to a model of $T^*$. This result will be used to show consistency of $T^*$.
\[lem:expand\] Let $M\models T$ and let $(A_b)_{b\in B}$ be a family of dense subsets of $M$ such that
- $\bigcup_{b\in B} A_b$ is ${\operatorname{dcl}}_T$-independent,
- $A_b \cap A_{b'} = \emptyset$ whenever $b\neq b'$,
- there is a model of $T'$ with the same cardinality as $B$.
Then $M$ can be expanded to a model of $T^*$.
Let ${\ensuremath{\mathcal{N}}}$ be a model of $T'$ with the same cardinality as $B$. Without loss of generality, we can assume that $B$ is the universe of ${\ensuremath{\mathcal{N}}}$. We now expand $M$ to an ${\ensuremath{\mathcal{L}}}^*$-structure ${\ensuremath{\mathcal{M}}}$. We interpret the relation symbol $P$ as $P_{\mathcal M}:= \bigcup_{b\in B} A_b$. For $a,a'\in P_{{\ensuremath{\mathcal{M}}}}$ we say $a E_{{\ensuremath{\mathcal{M}}}} a'$ if and only if there is $b\in B$ such that $a,a'\in A_b$. It is clear that $E_{{\ensuremath{\mathcal{M}}}}$ is an equivalence relation on $P_{{\ensuremath{\mathcal{M}}}}$ and that every equivalence class of $E_{{\ensuremath{\mathcal{M}}}}$ is dense in $M$. Thus $(M,P_{\mathcal M},E_{{\ensuremath{\mathcal{M}}}})$ is an ${\ensuremath{\mathcal{L}}}_e$-structure that models $T_e$. It is left to interpret the elements of ${\ensuremath{\mathcal{L}}}'$. Let $R$ be an $n$-ary relation symbol in ${\ensuremath{\mathcal{L}}}'$. We define its interpretation $R_{{\ensuremath{\mathcal{M}}}}$ by $$\{ (a_1,\dots,a_n) \in P_{{\ensuremath{\mathcal{M}}}}^n \ : \ \exists b_1,\dots,b_n \in B \ \bigwedge_{i=1}^n a_i \in A_{b_i} \wedge {\ensuremath{\mathcal{N}}} \models R(b_1,\dots b_n)\}.$$ Let ${\ensuremath{\mathcal{M}}} := (M,P_{{\ensuremath{\mathcal{M}}}},E_{{\ensuremath{\mathcal{M}}}}, \big(R_{{\ensuremath{\mathcal{M}}}}\big)_{R\in {\ensuremath{\mathcal{L}}}'})$. It is clear from the definition of $E_{{\ensuremath{\mathcal{M}}}}$ and $R_{{\ensuremath{\mathcal{M}}}}$ that ${\ensuremath{\mathcal{M}}}$ satisfies (T4). By a straightforward induction on formulas we see that for every ${\ensuremath{\mathcal{L}}}'$-formula $\varphi(x)$ and for every $a_1,\dots,a_n \in P_{{\ensuremath{\mathcal{M}}}}$ and $b_1,\dots,b_n\in B$ with $a_i \in A_{b_i}$ we have $${\ensuremath{\mathcal{M}}} \models \varphi_e(a_1,\dots,a_n) \hbox{ if and only if } {\ensuremath{\mathcal{N}}} \models \varphi(b_1,\dots,b_n).$$ Thus ${\ensuremath{\mathcal{M}}}$ satisfies (T5), and therefore ${\ensuremath{\mathcal{M}}}\models T^*$.
\[prop:consistent\] The theory $T^*$ is consistent.
By [@DMS-Indepedent 1.11] there is a model $M$ of $T$ and a family $(A_b)_{b\in B}$ of dense subsets of $M$ such that the family $(A_b)_{b \in B}$ satisfies the assumptions of Lemma \[lem:expand\]. The statement of the proposition then follows from Lemma \[lem:expand\].
\[prop:interpret\] Every model of $T^*$ interprets a model of $T'$.
Let ${\ensuremath{\mathcal{M}}} :=(M,P_{{\ensuremath{\mathcal{M}}}},E_{{\ensuremath{\mathcal{M}}}}, \big(R_{{\ensuremath{\mathcal{M}}}}\big)_{R\in {\ensuremath{\mathcal{L}}}'})\models T^*$. Let $N$ be the set of equivalence classes of $E_{{\ensuremath{\mathcal{M}}}}$. For an $r$-ary relation symbol $R$, let $$R_{{\ensuremath{\mathcal{N}}}} := \{ ([a_1]_{E_{{\ensuremath{\mathcal{M}}}}},\dots,[a_n]_{E_{{\ensuremath{\mathcal{M}}}}}) \in N^n \ : \ (a_1,\dots,a_n) \in R_{{\ensuremath{\mathcal{M}}}}\}.$$ Note that $R_{{\ensuremath{\mathcal{N}}}}$ is well-defined by Lemma \[lem:sametype\]. Let ${\ensuremath{\mathcal{N}}} = (N, \big(R_{{\ensuremath{\mathcal{N}}}}\big)_{R\in {\ensuremath{\mathcal{L}}}'})$. Since ${\ensuremath{\mathcal{N}}}$ is interpretable in ${\ensuremath{\mathcal{M}}}$, it is only left to show that ${\ensuremath{\mathcal{N}}} \models T'$. Using a straightforward induction on ${\ensuremath{\mathcal{L}}}'$-formulas and Axiom (T4) the reader can check that for every ${\ensuremath{\mathcal{L}}}'$-formula $\varphi(x)$ and for every $a_1,\dots,a_n \in P_{{\ensuremath{\mathcal{M}}}}$ $${\ensuremath{\mathcal{M}}} \models \varphi_e(a_1,\dots,a_n) \hbox{ if and only if } {\ensuremath{\mathcal{N}}} \models \varphi([a_1]_{E_{{\ensuremath{\mathcal{M}}}}},\dots,[a_n]_{E_{{\ensuremath{\mathcal{M}}}}}).$$ Thus ${\ensuremath{\mathcal{N}}} \models T'$, since ${\ensuremath{\mathcal{M}}}$ satisfies (T5).
Proposition \[prop:interpret\] shows that $T^*$ satisfies condition (1) of Theorem A. In the rest of this section we will show that $T^*$ also satisfies condition (2). In order to do so we have to carefully analyse the definable sets in models of $T^*$.
Back-and-forth system
---------------------
To better understand definable sets and types in models of $T^*$, we follow the general strategy of the proofs of [@DMS-Indepedent 2.8] and van den Dries [@densepairs Theorem 2.5] by constructing a back-and-forth system between models of $T^*$. Let $\kappa$ be a cardinal larger than $|T^*|$. Let ${\ensuremath{\mathcal{M}}}_1$ and ${\ensuremath{\mathcal{M}}}_2$ be two $\kappa$-saturated models of $T^*$. Let ${\ensuremath{\mathcal{I}}}$ be the set of all partial ${\ensuremath{\mathcal{L}}}$-isomorphisms $\iota : X \to Y$ between ${\ensuremath{\mathcal{M}}}_1$ and ${\ensuremath{\mathcal{M}}}_2$ such that there are
- finite $A\subseteq P_{{\ensuremath{\mathcal{M}}}_1}$ and $A'\subseteq P_{{\ensuremath{\mathcal{M}}}_2}$,
- finite $Z\subseteq {\ensuremath{\mathcal{M}}}_1$ and $Z'\subseteq {\ensuremath{\mathcal{M}}}_2$
with
- $\iota(A)=A'$ and $\iota(Z)=Z'$.
- $Z$ and $Z'$ are ${\operatorname{dcl}}$-independent over $P_{{\ensuremath{\mathcal{M}}}_1}$ and $P_{{\ensuremath{\mathcal{M}}}_2}$ respectively,
- $X={\operatorname{dcl}}_T(AZ)$ and $Y={\operatorname{dcl}}_T(A'Z')$,
- for all $a_1,\dots, a_n \in A$, $${\ensuremath{\mathcal{M}}}_1 \models \varphi_e(a_1,\dots,a_n) \hbox{ if and only if } {\ensuremath{\mathcal{M}}}_2 \models \varphi_e(\iota(a_1),\dots,\iota(a_n)).$$
In the following we will show that ${\ensuremath{\mathcal{I}}}$ is back-and-forth system of partial ${\ensuremath{\mathcal{L}}}^*$-iso-morphisms.
Let $\iota : X \to Y \in {\ensuremath{\mathcal{I}}}$ and let $A,Z \subseteq {\ensuremath{\mathcal{M}}}_1$ and $A',Z'\subseteq {\ensuremath{\mathcal{M}}}_2$ be such that $A,A,Z,Z'$ satisfy conditions (i)-(iv) above. Then $\iota$ is a partial ${\ensuremath{\mathcal{L}}}^*$-isomorphism and $X \cap P_{{\ensuremath{\mathcal{M}}}_1} = A$.
We first show that $X \cap P_{{\ensuremath{\mathcal{M}}}_1} = A$. Suppose there is $z \in (X\cap P_{{\ensuremath{\mathcal{M}}}_1})\setminus A$. Since $P_{{\ensuremath{\mathcal{M}}}_1}$ is ${\operatorname{dcl}}_T$-independent and $A\subseteq P_{{\ensuremath{\mathcal{M}}}_1}$, we have that $z \notin {\operatorname{dcl}}_T(A)$. Thus $z \in {\operatorname{dcl}}_T(AZ)\setminus {\operatorname{dcl}}_T(A)$. Since ${\operatorname{dcl}}_T$ is a pregeometry and $z\in P_{{\ensuremath{\mathcal{M}}}_1}$, this contradicts the ${\operatorname{dcl}}_T$-independence of $Z$ over $P_{{\ensuremath{\mathcal{M}}}_1}$. Similarly we can show that $Y\cap P_{{\ensuremath{\mathcal{M}}}_2}=A'$. Since $\iota(A)=A'$, it follows that $\iota(X\cap P_{{\ensuremath{\mathcal{M}}}_1})=Y\cap P_{{\ensuremath{\mathcal{M}}}_2}$. Since $(x=y)_e$ is $Exy$, we can easily deduce from (iv) that $\iota$ is an ${\ensuremath{\mathcal{L}}}_e$-isomorphism. Applying (iv) once more, we see that $\iota$ is also an ${\ensuremath{\mathcal{L}}}^*$-isomorphism.
The set ${\ensuremath{\mathcal{I}}}$ is a back-and-forth system.
Let $\iota : X \to Y \in {\ensuremath{\mathcal{I}}}$ and $b \in {\ensuremath{\mathcal{M}}}_1$. By symmetry it is enough to show that if $b\notin X$, then we can find $\iota'\in {\ensuremath{\mathcal{I}}}$ extending $\iota$ such that $b$ is in the domain of $\iota'$. From now on, assume that $b\notin X$.
**Case I:** $b \in P_{{\ensuremath{\mathcal{M}}}_1}$. Let $p$ be the collection of all ${\ensuremath{\mathcal{L}}}'$-formulas $\varphi(x,\iota(a))$ such that $a\in A^n$ and ${\ensuremath{\mathcal{M}}}_1 \models \varphi_e(b,a)$. By saturation of ${\ensuremath{\mathcal{M}}}_2$ and since $\iota \in {\ensuremath{\mathcal{I}}}$, there is $b' \in P_{{\ensuremath{\mathcal{M}}}_2}$ such that ${\ensuremath{\mathcal{M}}}_2 \models \varphi_e(b',\iota(a))$ for every $\varphi(x,\iota(a))\in p$. By density of the equivalence classes of $E$, we can take an element $b'' \in P_{{\ensuremath{\mathcal{M}}}_2}$ such that $(b',b'') \in E_{{\ensuremath{\mathcal{M}}}_2}$ and the cuts realized by $b$ in $X$ and by $b''$ in $Y$ correspond via $\iota$. Thus $\iota$ extends to an ${\ensuremath{\mathcal{L}}}$-isomorphism $\iota' : {\operatorname{dcl}}_T(ZAb) \to {\operatorname{dcl}}_T(ZA'b'')$ with $\iota'(b)=b''$. Since $(b',b'') \in E_{{\ensuremath{\mathcal{M}}}_2}$, we have by Lemma \[lem:sametype\] that ${\ensuremath{\mathcal{M}}}_2 \models \varphi_e(b'',\iota(a))$ for every $\varphi(x,\iota(a))\in p$. It is clear from our choice of $b''$ that $\iota' \in {\ensuremath{\mathcal{I}}}$.
**Case II:** $b \in {\operatorname{dcl}}_T(ZAP_{{\ensuremath{\mathcal{M}}}_1})$. Let $a_1,\dots,a_m \in P_{{\ensuremath{\mathcal{M}}}_1}$ be such that $b\in {\operatorname{dcl}}_T(ZAa_1\dots a_m)$. By applying Case I $m$ times, we can find an element $\iota' \in {\ensuremath{\mathcal{I}}}$ extending $\iota$ such that $a_1,\dots,a_m$ are in the domain of $\iota'$. Since the domain of $\iota'$ contains ${\operatorname{dcl}}_T(ZAa_1\dots a_m)$, it also contains $b$.
**Case III:** $b \notin {\operatorname{dcl}}_T(AP_{{\ensuremath{\mathcal{M}}}_1})$. By [@DMS-Indepedent 2.1] and saturation of ${\ensuremath{\mathcal{M}}}_2$, there exists an element $b'\in {\ensuremath{\mathcal{M}}}_2\setminus {\operatorname{dcl}}_T(A'P_{{\ensuremath{\mathcal{M}}}_2})$ such that the cuts realized by $b$ over $X$ and $b'$ over $Y$ correspond via $\iota$. Therefore we can find an ${\ensuremath{\mathcal{L}}}$-isomorphism $\iota' : {\operatorname{dcl}}_T(AZb) \to {\operatorname{dcl}}_T(A'Z'b')$ extending $\iota$ and mapping $b$ to $b'$. It is easy to check that $\iota' \in {\ensuremath{\mathcal{I}}}$.
Completeness and quantifier-reduction
-------------------------------------
We now use the back-and-forth system ${\ensuremath{\mathcal{I}}}$ to deduce certain desirable properties of $T^*$. In particular, we show completeness of $T^*$ and a quantifier-reduction result.
The theory $T^*$ is complete.
By Proposition \[prop:consistent\], $T^*$ is consistent. In the previous section we constructed a back-and-forth system between any two $\kappa$-saturated models of $T^*$. This implies that two such models are elementary equivalent. Completeness of $T^*$ follows.
We call an ${\ensuremath{\mathcal{L}}}^*$-formula $\chi(y)$ **special** if it is of the form $$\exists x \ Px \wedge \psi_e(x) \wedge \varphi(x,y),$$ where $\psi$ is an ${\ensuremath{\mathcal{L}}}'$-formula and $\varphi$ is an ${\ensuremath{\mathcal{L}}}$-formula.
We now establish that $T^*$ has quantifier-elimination up to boolean combinations of special formulas (compare this result and its proof to [@DMS-Indepedent 2.9]) and [@densepairs Theorem 1]).
Each ${\ensuremath{\mathcal{L}}}^*$-formula is $T^*$-equivalent to a boolean combination of special formulas.
Let ${\ensuremath{\mathcal{M}}}$ be a $\kappa$-saturated model of $T^*$. Let ${\ensuremath{\mathcal{M}}}_1 := {\ensuremath{\mathcal{M}}}_2 :={\ensuremath{\mathcal{M}}}$ and let ${\ensuremath{\mathcal{I}}}$ be the back-and-forth system between ${\ensuremath{\mathcal{M}}}_1$ and ${\ensuremath{\mathcal{M}}}_2$ constructed in the previous section. Let $a=(a_1,\dots,a_n),b=(b_1,\dots,b_n) \in M^n$ be such that $a$ and $b$ satisfy the same special formulas. To establish the theorem it suffices to show that ${\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}^*}(a)={\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}^*}(b)$. To prove the latter statement, it is enough to find $\iota\in I$ that maps $a$ to $b$. By permuting the coordinates we can assume there is $r \in \{0,\dots,n\}$ such that $a_1,\dots,a_r$ are ${\operatorname{dcl}}_T$-independent over $P_{{\ensuremath{\mathcal{M}}}}$ and $a_{r+1},\dots,a_n \in {\operatorname{dcl}}_T(a_1\dots a_rP_{{\ensuremath{\mathcal{M}}}})$. Since $a$ and $b$ satisfy the same special formulas, the reader can easily verify that $b_1,\dots,b_r$ are ${\operatorname{dcl}}_T$-independent over $P_{{\ensuremath{\mathcal{M}}}}$. Let $m\in {\mathbb{N}}$ and $g=(g_1,\dots,g_m) \in P_{{\ensuremath{\mathcal{M}}}}^m$ be such that $a_{r+1},\dots,a_n \in {\operatorname{dcl}}_T(a_1\dots a_rg)$. For $i=r+1,\dots,n$, let $f_i : M^{r+m} \to M$ be an ${\ensuremath{\mathcal{L}}}$-$\emptyset$-definable function such that $f_i(a_1,\dots,a_r,g) = a_i$. We will now find $h=(h_1,\dots,h_m) \in P_{{\ensuremath{\mathcal{M}}}}^m$ such that
- ${\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}'}(h)={\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}'}(g)$,
- $f_i(b_1,\dots,b_r,h) = b_i$ for each $i=r+1,\dots,n$.
If we have such $h$, we can find an ${\ensuremath{\mathcal{L}}}$-isomorphism $\iota : {\operatorname{dcl}}(a_1\dots a_rg) \to {\operatorname{dcl}}(b_1\dots b_rh)$ such that $\iota(g)=h$ and $\iota(a_i) = b_i$ for each $i=1,\dots,r$. Since $h$ satisfies (1) and each of the sets $\{a_1\dots a_r\}$ and $\{b_1\dots b_r\}$ is ${\operatorname{dcl}}_T$-independent over $P_{{\ensuremath{\mathcal{M}}}}$, it is easy to check that $\iota \in {\ensuremath{\mathcal{I}}}$. Because $h$ also satisfies (2), we get that $\iota(a_i)=b_i$ for $i=r+1,\dots,n$. Thus $\iota$ is the desired element of ${\ensuremath{\mathcal{I}}}$.
We now prove the existence of an $h \in P_{{\ensuremath{\mathcal{M}}}}^m$ satisfying (1) and (2). Observe that there is an ${\ensuremath{\mathcal{L}}}$-formula $\psi(x,y)$ such that an element $h \in M^m$ satisfies (2) if and only if ${\ensuremath{\mathcal{M}}}\models \psi(h,b)$. By saturation, in order to find $h$ satisfying (1) and (2), it is enough to find for every ${\ensuremath{\mathcal{L}}}'$-formula $\varphi(x)$ with ${\ensuremath{\mathcal{M}}}\models \varphi_e(g)$ an $h\in P_{{\ensuremath{\mathcal{M}}}}^m$ such that ${\ensuremath{\mathcal{M}}} \models \varphi_e(h)\wedge \psi(h,b)$. So let $\varphi(x)$ be an ${\ensuremath{\mathcal{L}}}'$-formula with ${\ensuremath{\mathcal{M}}}\models \varphi_e(g)$. Consider the special formula $\chi(y)$ given by $$\exists x \ Px \wedge \varphi_e(x) \wedge \psi(x,y).$$ Since ${\ensuremath{\mathcal{M}}}\models \chi(a)$ and $a$ and $b$ satisfy the same special formulas, we get that ${\ensuremath{\mathcal{M}}}\models \chi(b)$. Thus there exists $h\in P_{{\ensuremath{\mathcal{M}}}}^m$ such that ${\ensuremath{\mathcal{M}}} \models \varphi_e(h)\wedge \psi(h,b)$.
Types
-----
In order to show statements (2)-(4) of Theorem A we need better control over the ${\ensuremath{\mathcal{L}}}^*$-types in models of $T^*$. We establish the necessary results in this section. Throughout let ${\ensuremath{\mathcal{M}}}$ be a $\kappa$-saturated model of $T^*$. We first introduce the following notation: for $C \subseteq {\ensuremath{\mathcal{M}}}$ and $n\in {\mathbb{N}}$ we denote by $D_n(C)$ the set $$\{ z \in {\ensuremath{\mathcal{M}}}^n \ : \ z \hbox{ is ${\operatorname{dcl}}_T$-independent from } CP_{{\ensuremath{\mathcal{M}}}}\}.$$
\[prop:type\] Let $a \in P_{{\ensuremath{\mathcal{M}}}}^k$, $z \in D_l(\emptyset)$, $b \in P_{{\ensuremath{\mathcal{M}}}}^m$ and $y\in D_n(z)$. Then ${\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}^*}(by|az)$ is implied by the conjunction of
- ${\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}}(by|az)$,
- “$b\in P_{{\ensuremath{\mathcal{M}}}}^m$” and ${\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}'}(b|a)$,
- “$y \in D_n(z)$”.
Set ${\ensuremath{\mathcal{M}}}_1 := {\ensuremath{\mathcal{M}}}_2 :={\ensuremath{\mathcal{M}}}$ and let ${\ensuremath{\mathcal{I}}}$ be the back-and-forth system between ${\ensuremath{\mathcal{M}}}_1$ and ${\ensuremath{\mathcal{M}}}_2$ constructed in the previous section. Let $b_1,b_2 \in P_{\mathcal M}^m$ and $y_1,y_2 \in D_n(z)$ be such that ${\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}}(b_1y_1|az) = {\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}}(b_2y_2|az)$ and ${\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}'}(b_1 | a)= {\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}'}(b_2 | a)$. In order to show that ${\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}^*}(b_1y_1 | az) = {\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}^*}(b_2y_2 | az)$, we only need to find $\iota\in {\ensuremath{\mathcal{I}}}$ such that $\iota(b_1y_1)=b_2y_2$ and the coordinates of $a$ and $z$ are in the domain of $\iota$. It is immediate that the identity on ${\operatorname{dcl}}_T(az)$ is in ${\ensuremath{\mathcal{I}}}$. Since ${\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}}(b_1y_1|az)={\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}}(b_2y_2|az)$, there is a partial ${\ensuremath{\mathcal{L}}}$-isomorphism from ${\operatorname{dcl}}_T(azb_1y_1)$ to ${\operatorname{dcl}}_T(azb_2y_2)$ mapping $b_1y_1$ to $b_2y_2$. Because ${\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}'}(b_1 | a)= {\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}'}(b_2 | a)$ and $y_1,y_2 \in D_n(z)$, it is immediate that $\iota \in {\ensuremath{\mathcal{I}}}$.
We immediately obtain the following three corollaries from Proposition \[prop:type\].
\[lem:typeI\] Let $C\subseteq {\ensuremath{\mathcal{M}}}$ be finite and $y \in D_n(C)$. Then ${\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}^*}(y | C)$ is implied by ${\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}}(y | C)$ in conjunction with “$y \in D_n(C)$”.
\[lem:induced\] Let $a \in P_{{\ensuremath{\mathcal{M}}}}^k$, $z \in D_l(\emptyset)$ and $b \in P_{{\ensuremath{\mathcal{M}}}}^n$. Then ${\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}^*}(b|az)$ is implied by ${\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}}(b|az)$, “$b \in P_{{\ensuremath{\mathcal{M}}}}^n$” and ${\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}'}(b|a)$.
Let $Z \subseteq P_{{\ensuremath{\mathcal{M}}}}^n$ be ${\ensuremath{\mathcal{L}}}^*$-definable. Then there is an ${\ensuremath{\mathcal{L}}}$-definable set $Y \subseteq {\ensuremath{\mathcal{M}}}^n$ and an ${\ensuremath{\mathcal{L}}}'$-formula $\varphi(x)$ such that $$Z = Y \cap \{ a \in P_{{\ensuremath{\mathcal{M}}}}^n \ : \ {\ensuremath{\mathcal{M}}} \models \varphi_e(a)\}.$$
Combining Proposition \[prop:type\] with a result of Boxall and Hieronymi [@BoxallH], we are now able to deduce statement (2) of Theorem A.
\[thm:opencoreTstar\] The theory $T$ is an open core of $T^*$.
We will use [@BoxallH Corollary 3.1] to show that every ${\ensuremath{\mathcal{L}}}^*$-definable open set in ${\ensuremath{\mathcal{M}}}$ is also ${\ensuremath{\mathcal{L}}}$-definable. Let $X\subseteq {\ensuremath{\mathcal{M}}}^n$ be open and ${\ensuremath{\mathcal{L}}}^*$-definable over some finite parameter set $C$. We will now apply [@BoxallH Corollary 3.1], using $D_n(C)$ as $D_{S_1\dots S_n}$. Therefore it is left to check that conditions (1)-(3) of [@BoxallH Corollary 3.1] hold for $D_n(C)$. These three conditions are
1. $D_n(C)$ is dense in ${\ensuremath{\mathcal{M}}}$,
2. for every $y\in D_n(C)$ and every open set $U\subseteq {\ensuremath{\mathcal{M}}}^n$, if ${\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}}(y|C)$ is realized in $U$, then ${\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}}(y|C)$ is realized in $U\cap D_n(C)$,
3. for every $y \in D_n(z)$, ${\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}^*}(y|C)$ is implied by ${\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}}(y | C)$ in conjunction with “$y \in D_n(C)$”.
Condition (1) follows easily from saturation of ${\ensuremath{\mathcal{M}}}$ and [@DMS-Indepedent 2.1]. Using o-minimality of $T$, it is easy to deduce Condition (2) from Condition (1). Finally, Condition (3) holds by Corollary \[lem:typeI\].
Completions of $T_e$
--------------------
Using results from the previous sections we will now give a characterizations of all complete ${\ensuremath{\mathcal{L}}}_e$-theories containing $T_e$.
Let $T_{e,\infty}$ be the ${\ensuremath{\mathcal{L}}}_e$-theory consisting of $T_e$ and an axiom schema expressing the following statement:
- $E$ has infinitely many equivalence classes.
Similarly, for every $n\in {\mathbb{N}}_{>0}$ define $T_{e,n}$ to be the ${\ensuremath{\mathcal{L}}}_e$-theory consisting of $T_e$ and a sentence stating that $E$ has exactly $n$ equivalence classes.
\[thm:tecompletion\] Let $p \in {\mathbb{N}}_{>0}\cup \{\infty\}$. The theory $T_{e,p}$ is complete.
We first consider the case that $p=\infty$. Let ${\ensuremath{\mathcal{L}}}'$ be empty and $T'$ be the (complete) ${\ensuremath{\mathcal{L}}}'$-theory of infinite sets. Let ${\ensuremath{\mathcal{L}}}^*$ and ${\ensuremath{\mathcal{T}}}^*$ be constructed as above. Since ${\ensuremath{\mathcal{L}}}'=\emptyset$, we have that ${\ensuremath{\mathcal{L}}}^* = {\ensuremath{\mathcal{L}}}_e$. Since $T^*$ is complete, it is enough to show that every model of $T_{e,\infty}$ is also a model of $T^*$. Let ${\ensuremath{\mathcal{M}}} \models T_{e,\infty}$. Since ${\ensuremath{\mathcal{L}}}'=\emptyset$, we immediately get that ${\ensuremath{\mathcal{M}}}$ satisfies (T4). It is left to show that ${\ensuremath{\mathcal{M}}}$ satisfies (T5). Let $\varphi \in T'$. Since $T'$ is the theory of infinite sets, there is $n\in {\mathbb{N}}$ such that $\varphi$ is the following formula $$\exists x_1 \dots \exists x_n \bigwedge_{1\leq i<j \leq n} x_i \neq x_j.$$ It is easy to check that $\varphi_e$ is the ${\ensuremath{\mathcal{L}}}_e$-formula $$\exists x_1 \dots \exists x_n \bigwedge_{1\leq i<j \leq n} Px_i \wedge \neg Ex_ix_j.$$ Since ${\ensuremath{\mathcal{M}}}$ satisfies (T6), we get that ${\ensuremath{\mathcal{M}}} \models \varphi_e$. Thus ${\ensuremath{\mathcal{M}}}$ satisfies (T5).The proof of the case $p\in {\mathbb{N}}_{>0}$ can be done similarly by replacing the ${\ensuremath{\mathcal{L}}}'$-theory of infinite sets by the ${\ensuremath{\mathcal{L}}}'$-theory of a set with exactly $p$ elements.
From Theorem \[thm:tecompletion\] we can directly deduce the following characterization of completions of $T_e$.
Let $\tilde{T}$ be a complete ${\ensuremath{\mathcal{L}}}_e$-theory such that $T_e \subseteq \tilde{T}$. Then there is $p \in {\mathbb{N}}_{>0} \cup \{\infty\}$ such that $T_{e,p}\models \tilde{T}$.
We obtain the following corollary as an immediate consequence of Theorem \[thm:opencoreTstar\] and the proof of Theorem \[thm:tecompletion\].
Let $p \in {\mathbb{N}}_{>0}\cup \{\infty\}$. The theory $T$ is an open core of $T_{e,p}$.
Preservation of NIP {#Section:NIP}
===================
Let $T$ be a complete o-minimal extension of the theory of densely ordered groups with a distinguished positive element, and let ${\ensuremath{\mathcal{L}}}$ be its language. As before, let ${\ensuremath{\mathcal{L}}}'$ be a relational language disjoint from ${\ensuremath{\mathcal{L}}}$, and let $T'$ be a complete ${\ensuremath{\mathcal{L}}}'$-theory. Furthermore, let $T^*$ be the ${\ensuremath{\mathcal{L}}}^*$-theory constructed in the previous section. We will now show that $T^*$ is NIP if $T'$ is NIP. As we will see, this can be deduced rather directly from Corollaries \[lem:typeI\] and \[lem:induced\] and the following result of G[ü]{}naydin and Hieronymi [@GH-Dependent].
\[gh:fact\][@GH-Dependent Proposition 2.4] Let ${\ensuremath{\mathcal{L}}}_0$ be a first-order language and let ${\ensuremath{\mathcal{L}}}_1$ be a language containing ${\ensuremath{\mathcal{L}}}_0$ and a unary predicate symbol $U$ not in ${\ensuremath{\mathcal{L}}}_0$. Let $T_0$ be a complete ${\ensuremath{\mathcal{L}}}_0$-theory and let $T_1$ be a complete ${\ensuremath{\mathcal{L}}}_1$-theory extending $T_0$. Let $\mathbb M$ be a monster model of $T_1$. Suppose that
- ${\operatorname{dcl}}_{T_0}$ is a pregeometry,
- for every ${\ensuremath{\mathcal{L}}}_1$-formula $\varphi(x,y)$, indiscernible sequence $(g_i)_{i\in\omega}$ from $U_{\mathbb M}^p$ and $b\in\mathbb M^q$, the set $\{ i\in\omega \ : \ \mathbb{M} \models
\varphi(g_i,b)\}$ is either finite or co-finite (in $\omega$),
- for every formula $\varphi(x,y)$, indiscernible sequence $(a_i)_{i\in\omega}$ from $\mathbb M$ and $b\in \mathbb M^q$ with $a_i\notin{\operatorname{dcl}}_{{\ensuremath{\mathcal{L}}}_0}(U_{\mathbb M}b)$ for every $i\in\omega$, the set $\{ i\in\omega \ : \ \mathbb{M} \models
\varphi({a_i},b)\}$ is either finite or co-finite (in $\omega$).
Then $T_1$ is NIP.
If $T'$ is NIP, so is $T^*$.
We apply Fact \[gh:fact\] with $T_0:=T$ and $T_1:= T^*$. Since $T$ is o-minimal, ${\operatorname{dcl}}_T$ is a pregeometry.For (ii), let $\varphi(x,y)$ be an ${\ensuremath{\mathcal{L}}}^*$-formula, $(g_i)_{i\in\omega}$ an indiscernible sequence from $P_{\mathbb M}^p$ and $b\in\mathbb M^q$. Without loss of generality, we can assume that there are $b_1\in P_{\mathbb M}^{q_1}$ and $b_2 \in \mathbb M^{q_2}$ such that $b_2$ is ${\operatorname{dcl}}_T$-independent over $P_{\mathbb M}$ and $b=(b_1,b_2)$. By Corollary \[lem:induced\] there is an ${\ensuremath{\mathcal{L}}}$-formula $\psi(x,u,v)$ and an ${\ensuremath{\mathcal{L}}}'$-formula $\theta(x,u)$ such that for all $a \in P_{\mathbb M}^p$ $$\label{eq:NIP}
\tag{$\ast$} \mathbb{M} \models \varphi(a,b) \leftrightarrow \big(\psi(a,b_1,b_2) \wedge \theta_e(a,b_1)\big).$$ Since both $T$ and $T'$ are NIP, it follows immediately from that $\{ i\in\omega \ : \ \mathbb{M} \models \varphi(g_i,b)\}$ is either finite or co-finite.For (iii), let $\varphi(x,y)$ be an ${\ensuremath{\mathcal{L}}}^*$-formula, $(a_i)_{i\in\omega}$ an indiscernible sequence from $\mathbb M$ and $b\in \mathbb M^q$ with $a_i\notin{\operatorname{dcl}}_{T}(P_{\mathbb M}b)$ for every $i\in\omega$. By Corollary \[lem:typeI\] there is an ${\ensuremath{\mathcal{L}}}$-formula $\psi(x,b)$ such that for all $a \in \mathbb{M} \setminus {\operatorname{dcl}}_{T}(P_{\mathbb M}b)$ $$\mathbb{M} \models \varphi(a,b) \leftrightarrow \psi(a,b).$$ Since $T$ is NIP, $\{ i\in\omega \ : \ \mathbb{M} \models \varphi({a_i},b)\}$ is either finite or co-finite (in $\omega$).
We can now give a proof of Theorem D that there is an NIP expansion of $T$ without a distal expansion.
Fix a prime $p$. Let $T'$ be $ACF_p$. Since $T'$ is stable, $T'$ is NIP. Suppose $T^*$ has a distal expansion $\tilde{T}$. Then $\tilde{T}^{eq}$ is distal by [@Simon-Book Remark after Definition 9.17]. However, by Proposition \[prop:interpret\] every model of $\tilde{T}^{eq}$ defines a model of $T'$. By [@CS-RegDis Proposition 6.2] $\tilde{T}^{eq}$ cannot be distal. A contradiction.
Preservation of Strong Dependence {#Section:STDEP}
=================================
In this section, we will show that $T^*$ (as constructed in Section \[Section:Fusion\]) is strongly dependent if $T'$ is. We essentially follow the proof of Berenstein, Dolich and Onshuus [@BDO Theorem 2.11].
Let ${\ensuremath{\mathcal{L}}}_0$ be a first-order language containing $<$ and let ${\ensuremath{\mathcal{L}}}_1$ be a language containing ${\ensuremath{\mathcal{L}}}_0$ and a unary predicate symbol $U$ not in ${\ensuremath{\mathcal{L}}}_0$. Let $T_0$ be a complete ${\ensuremath{\mathcal{L}}}_0$-theory extending the theory of linear ordered sets such that ${\operatorname{dcl}}_{T_0}$ is a pregeometry. Let $T_1$ be a complete ${\ensuremath{\mathcal{L}}}_1$-theory extending $T_0$, and let $\mathbb M$ be a monster model of $T_1$. If $X,Y$ are subsets of $\mathbb M$, we say $X$ is **$U$-independent** over $Y$ if $X\setminus U_{\mathbb M}$ is ${\operatorname{dcl}}_{T_0}$-independent over $U_{\mathbb M}Y$. If $Y=\emptyset$, we simply say that $X$ is $U$-independent. We say an indiscernible sequence $(a_i)_{i \in I}$ of tuples of elements of $\mathbb M$ is **$U$-independent** if each $a_i$ is $U$-independent.
\[lem:indi\] Let $\kappa$ be an infinite cardinal and let $(a_i)_{i \in I} $ be an indiscernible sequence of tuples of elements of $\mathbb M$ of length $\kappa$. Then there is an $U$-independent indiscernible sequence $J=(b_i)_{i \in I}$ of tuples of elements of $\mathbb M$ of length $\kappa$ such that for every $j<\kappa$ there is an ${\ensuremath{\mathcal{L}}}_0$-$\emptyset$-definable function $f : \mathbb M^n \to \mathbb M$, and $j_1,\ldots, j_n<\kappa$ such that for every $i \in I$ $$a_{i,j} = f(b_{i,j_1},\dots,b_{i,j_n}).$$
We inductively construct a sequence $(b_i)_{i\in I}$ from the sequence $(a_i)_{i \in I}$ by removing $U$-dependencies. Let $\alpha< \kappa$ be minimal such that there is $i\in I$ such that $\{a_{i,j}:j\leq \alpha\}$ is not $U$-independent. By minimality of $\alpha$ there are $j_1<\ldots < j_m < \alpha$ and an ${\ensuremath{\mathcal{L}}}_0$-$\emptyset$-definable $f:M^{m+\ell+1}\to M$ such that $$\label{eq:indi}\tag{$*$}
\exists u_{i,0},\ldots, u_{i,\ell}\in U_{\mathbb M} \ f(a_{i,j_1},\ldots,a_{i,j_m},u_{i,0},\ldots,u_{i,\ell}) = a_{i,\alpha}.$$ By indiscernibility of $(a_i)_{i\in I}$, holds for every $i\in I$. For each $i\in I$, define a set $$S_i = \{ (u_{0},\ldots,u_{\ell}) \in U^{l+1} \ : \ f(a_{i,j_1},\ldots,a_{i,j_m},u_{0},\ldots,u_{\ell}) = a_{i,\alpha}\}.$$ By indiscernibility of $(a_i)_{i\in I}$, we have that $S_i$ is finite for some $i$ if and and only $S_{i}$ is finite for every $i\in I$. We first consider the case that $S_i$ is finite for every $i \in I$. Then for each $i\in I$ we may choose $u_i = (u_{i,0},\ldots u_{i,\ell})$ to be the lexicographically least member of $S_i$. Let $b_i$ be the tuple where $a_{i,\alpha}$ is replaced by $u_i$. As $a_{i,\alpha}$ and $u_i$ are interdefinable over $\{a_{i,j_1},\ldots, a_{i,j_m}\}$, $(b_i)_{i\in I}$ is indiscernible. Furthermore, the set $\{b_{i,1},\dots,b_{i,\alpha+\ell}\}$ is $U$-independent for each $i\in I$.Now suppose that $S_i$ is infinite. Consider the collection of formulas in variables $(x_{i,j})_{i \in I}$ for $j<\kappa$ stating:
1. $x_{i,j}=a_{i,j}$ for $j<\alpha$
2. $f(a_{i,j_1},\ldots a_{i,j_m},x_{i,\alpha},\ldots, x_{i,\alpha+\ell})=a_{i,\alpha}$
3. $x_{i,\alpha},\ldots, x_{i,\alpha+\ell} \in U$
4. $x_{i,\alpha+\ell+j} = a_{i,\alpha+j}$ for $j \ge 1$
5. The sequence $(x_i)_{i \in I}$ is indiscernible.
As $S_i$ is infinite, it can be shown by a standard argument using Ramsey’s theorem that this collection is finitely satisfiable. Therefore, by saturation there is a realization $(b_i)_{i \in I}$ of this collection. By construction, we have for every $i\in I$ that $\{b_{i,1},\dots, b_{i,\alpha + \ell}\}$ is $U$-independent and $a_{i,\alpha} = f(b_{i,j_1},\ldots b_{i,j_m},b_{i,\alpha},\ldots, b_{i,\alpha+\ell})$. Inductively continuing, we arrange the sequence $(b_i)_{i\in I}$ as desired.
We will use Lemma \[lem:indi\] to show a criterion for strong dependence for $T_1$. Before we do so, we recall the definition of strong dependence. If $I$ is a linear order, we denote its completion by ${\operatorname{compl}}(I)$. If $I$ is a linear order, $c=(c_1,\dots,c_n)\in {\operatorname{compl}}(I)^n$ and $i,i' \in I$, we write $i \sim_c i'$ if $$\bigwedge_{j=1}^n \big((i < c_j \leftrightarrow i' < c_j) \wedge (i = c_j \leftrightarrow i' = c_j)\big)$$ Note that $\sim_{c}$ defines an equivalence relation $\sim_{c}$ on $I$.
A theory $\tilde{T}$ in a language $\tilde{{\ensuremath{\mathcal{L}}}}$ is **strongly dependent** if for every $\mathcal{M}\models \tilde{T}$, every $b\in \mathcal M^m$ and every indiscernible sequence $(a_i)_{i \in I}$, there is $n\in {\mathbb{N}}$ and $c\in {\operatorname{compl}}(I)^n$ such that $i\sim_c j \Rightarrow {\operatorname{tp}}_{\tilde{{\ensuremath{\mathcal{L}}}}}(a_i | b) = {\operatorname{tp}}_{\tilde{{\ensuremath{\mathcal{L}}}}}(a_j|b)$.
For more details and other equivalent definitions of strong dependence, we refer the reader to [@Simon-Book Chapter 4].
\[lem:stdep\] The following are equivalent:
- For every $b\in \mathbb M$, and every $U$-independent indiscernible sequence $(a_i)_{i \in I}$, if $b$ is $U$-independent over $\{a_i : i \in I\}$, then there is $n\in {\mathbb{N}}$ and $c\in {\operatorname{compl}}(I)^n$ such that $i\sim_c j \Rightarrow {\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}_1}(a_i | b) = {\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}_1}(a_j|b)$.
- For every $b\in \mathbb M$, and indiscernible sequence $(a_i)_{i \in I}$, if $b$ is $U$-independent over $\{a_i : i \in I\})$, then there is $n\in {\mathbb{N}}$ and $c\in {\operatorname{compl}}(I)^n$ such that $i\sim_c j \Rightarrow {\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}_1}(a_i | b) = {\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}_1}(a_j|b)$.
- $T_1$ is strongly dependent.
It is clear that (iii)$\Rightarrow$(ii)$\Rightarrow$(i). Observe that (i)$\Rightarrow $(ii) follows easily from Lemma \[lem:indi\]. So we only need to show that (ii) implies (iii). Let $b \in \mathbb M^m$ and $(a_i)_{i \in I}$ be an indiscernible sequence of possibly infinite tuples from $\mathbb M$. It is enough to consider the case $m=1$ (see for example [@Simon-Book Proposition 4.26]). Suppose $b\in {\operatorname{dcl}}_{T_0}(U_{\mathbb M}\{a_i : i \in I\})$. Then there are $g \in U_{\mathbb M}^l$, $e=(e_1,\dots, e_k)\in I^k$ and an ${\ensuremath{\mathcal{L}}}_0$-$\emptyset$-definable function $f$ such that $$\label{eq:stdep}\tag{$\dagger$}
b=f(g,a_{e_1},\dots, a_{e_k}).$$ Without loss of generality assume that $e_1 < \dots < e_k$. Set $a_e=(a_{e_1},\dots, a_{e_k})$, $e_0=-\infty$ and $e_{k+1}=+\infty$. Let $t \in \{0,\dots,k\}$. Now observe that $(a_ea_j)_{j \in (e_{t},e_{t+1})\cap I}$ is an indiscernible sequence. By (ii) there is $d_t \in ({\operatorname{compl}}(I)\cap (e_{t},e_{t+1}))^{n_t}$ such that for all $i,j \in (e_{t},e_{t+1})$ we have $i \sim_{d_t} j \Rightarrow {\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}_1}(a_{e}a_i|g) = {\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}_1}(a_{e}a_j|g)$. By we get that for all such $i,j$ $$i \sim_{d_t} j \Rightarrow {\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}_1}(a_i|b) = {\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}_1}(a_j|b).$$ Set $c:=(d_0e_1d_1\dots e_kd_{k+1})$. It can be checked easily that this is the desired $c\in {\operatorname{compl}}(I)^n$.
Let us now recall the setting of Section \[Section:Fusion\]. Let $T$ be a complete o-minimal extension of the theory of densely ordered groups with a distinguished positive element, and let ${\ensuremath{\mathcal{L}}}$ be its language. As before, let ${\ensuremath{\mathcal{L}}}'$ be a language disjoint from ${\ensuremath{\mathcal{L}}}$, and let $T'$ be a complete ${\ensuremath{\mathcal{L}}}'$-theory. Furthermore, let $T^*$ be the ${\ensuremath{\mathcal{L}}}^*$-theory constructed in Section \[Section:Fusion\].
If $T'$ is strongly dependent, so is $T^*$.
We now apply Lemma \[lem:stdep\] with $T_0:=T$, $T_1:=T^*$ and $U:=P$. As before, note that ${\operatorname{dcl}}_T$ is a pregeometry, since $T$ is o-minimal.Let $b\in \mathbb M$ and $(a_i)_{i \in I}$ be an $P$-independent sequence such that $b$ is $P$-independent over $\{a_i : i \in I\}$. Since each $a_i$ is $P$-independent, we have (after possibly changing the order of entries of the $a_i$’s) that for each $i\in I$ there are tuples $u_i,v_i$ of elements of $\mathbb M$ such that for each $i \in i$
- $a_i = u_iv_i$,
- $u_i$ is a tuple of elements in $P_{\mathbb M}$,
- $v_i$ is ${\operatorname{dcl}}_T$-independent over $P_{\mathbb M}$.
Since $b$ is $P$-independent over $\{a_i : i \in I\}$, we get that either $b\in P_{\mathbb M}$ or $b\notin {\operatorname{dcl}}_T(\{a_i : i \in I\}P_{\mathbb{M}})$. We consider the two different cases.
Let $b\in P_{\mathbb M}$. By Proposition \[prop:type\] the type ${\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}^*}(u_iv_i|b)$ is determined by
- ${\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}}(u_iv_i|b)$,
- the statement “$u_i$ is a tuple of elements of $P_{\mathbb{M}}$” and ${\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}'}(u_i|b)$,
- the statement “$v_i$ is ${\operatorname{dcl}}_T$-independent over $P_{\mathbb M}$”.
Since both $T$ and $T'$ are strongly dependent, we can find $c\in {\operatorname{compl}}(I)^n$ such that for every $i,j \in I$ $$i\sim_c j \Rightarrow \Big( {\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}}(u_iv_i|b)={\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}}(u_jv_j|b) \hbox{ and } {\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}'}(u_i|b)={\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}'}(u_j|b) \Big).$$ Thus for every $i,j \in I$ with $i\sim_c j$ we get ${\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}^*}(a_i|b)={\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}^*}(a_j|b)$.
Now suppose that $b\notin {\operatorname{dcl}}_T(\{a_i : i \in I\}P_{\mathbb{M}})$. In particular, $b \notin {\operatorname{dcl}}_T(P_{\mathbb{M}})$. Since ${\operatorname{dcl}}_T$ is a pregeometry, $v_i$ is ${\operatorname{dcl}}_T$-independent over $P_{\mathbb{M}}b$ for each $i\in I$. By Proposition \[prop:type\], for each $i\in I$ the type ${\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}^*}(u_iv_i|b)$ is determined by
- ${\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}}(u_iv_i|b)$,
- the statement “$u_i$ is a tuple of elements of $P_{\mathbb{M}}$” and ${\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}'}(u_i)$,
- the statement “$v_i$ is ${\operatorname{dcl}}_T$-independent over $P_{\mathbb M}b$”.
As before using strong dependence of $T$ and $T'$, we can find $c\in {\operatorname{compl}}(I)^n$ such that for every $i,j \in I$ $$i\sim_c j \Rightarrow \Big( {\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}}(u_iv_i|b)={\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}}(u_jv_j|b) \hbox{ and } {\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}'}(u_i)={\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}'}(u_j) \Big).$$ Thus for every $i,j \in I$ with $i\sim_c j$ we get ${\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}^*}(a_i|b)={\operatorname{tp}}_{{\ensuremath{\mathcal{L}}}^*}(a_j|b)$.
This completes the proof of Theorem A. In the next section we will deduce Theorem B from Theorem A.
It is worth pointing out in this section on strong dependence that by Dolich and Goodrick [@DG Corollary 2.4] every strongly dependent expansion of the real field has o-minimal open core. In contrast to this restriction, our Theorem B(4) shows that there is a large variety of such expansions of the real field.
Proof of Theorem B
==================
The purpose of this section is twofold. We first deduce Theorem B from our proof of Theorem A. Then we show that in Theorem B the statement “${\ensuremath{\mathcal{S}}}$ interprets a model of $T'$” cannot be replaced by the statement “${\ensuremath{\mathcal{S}}}$ defines a model of $T'$”.
Let ${\ensuremath{\mathcal{R}}}=({\mathbb{R}},<,+,\dots)$ be an o-minimal expansion of the real ordered additive group in a language ${\ensuremath{\mathcal{L}}}$ and let $T'$ be a theory such that $|{\ensuremath{\mathcal{L}}}|<|{\mathbb{R}}|$ and $|T'|\leq |{\mathbb{R}}|$. Let $T^*$ be the theory as constructed in Section \[Section:Fusion\]. Since $T^*$ satisfies the statements (1)-(4) of Theorem A, it is only left to show that ${\ensuremath{\mathcal{R}}}$ can be expanded to a model of $T^*$. Since $|{\ensuremath{\mathcal{L}}}|<|{\mathbb{R}}|$, we can find a ${\operatorname{dcl}}_T$-basis of cardinality at least $|T'|$. Since ${\operatorname{dcl}}_T(\emptyset)$ is dense in ${\mathbb{R}}$, we are able to choose this basis such that it is dense in ${\mathbb{R}}$. Now apply Lemma \[lem:expand\].
\[prop:ocdlo\] Let ${\ensuremath{\mathcal{S}}}$ be an expansion of $({\mathbb{R}},<,+)$. The following are equivalent
1. ${\ensuremath{\mathcal{S}}}$ defines an infinite discrete linear order.
2. ${\ensuremath{\mathcal{S}}}$ defines an order with order type $\omega$.
3. The open core of ${\ensuremath{\mathcal{S}}}$ is not o-minimal.
We show that (1) implies (2). Suppose ${\ensuremath{\mathcal{S}}}$ defines an infinite discrete linear order $(D,\prec)$. Fix $d \in D$. Either $D_{ \prec d}$ or $D_{\succ d}$ is infinite. After replacing $\prec$ with the reverse order if necessary, we may suppose that $D_{ \succ d}$ is infinite. After replacing $(D,\prec)$ with $(D_{\succeq d }, \prec)$ if necessary we suppose that $(D, \prec)$ has a minimal element. Let $E \subseteq D$ be the set of $e$ such that $D_{\prec e}$ is finite. Recall that a subset of ${\mathbb{R}}^n$ is finite if and only if it is closed, bounded and discrete. It follows that $E$ is definable. Note that $E_{\prec e}$ is finite for all $e \in D$. Then $(E, \prec)$ is a discrete linear order with minimal element and finite initial segments. Thus it has order type $\omega$.We now show that (2) implies (3). Suppose that $(D, \prec)$ is a definable order with order type $\omega$ and $D \subseteq {\mathbb{R}}^n$. First suppose that there is no coordinate projection $\pi: {\mathbb{R}}^n \to {\mathbb{R}}$ such that $\pi(D)$ is somewhere dense. Since $D$ is infinite, there is a coordinate projection $\rho : {\mathbb{R}}^n \to {\mathbb{R}}$ such that $\rho(D)$ is infinite. Then $\rho(D)$ is an infinite, nowhere dense, subset of ${\mathbb{R}}$. Thus the open core of ${\ensuremath{\mathcal{S}}}$ is not o-minimal. Now let $\pi: {\mathbb{R}}^n \to {\mathbb{R}}$ be a coordinate projection such that $\pi(D)$ is somewhere dense. Let $a,b\in {\mathbb{R}}$ such that $(a,b)$ is an interval in in which $\pi(D)$ is dense. We now reduce to the case when $D$ is a dense subset of an open interval. Note that $D' = \{ e \in D : a < \pi(e) < b \}$ is an infinite, and hence $\prec$-cofinal, subset of $D$. It follows that $(D',\prec)$ has order type $\omega$. After replacing $D$ with $D'$ if necessary we suppose that $\pi(D)$ is a subset of $(a,b)$. We put an order $\prec_\pi$ on $\pi(D)$ by declaring $x \prec_\pi y$ if there is a $e \in D$ such that $\pi(e) = x$ and $\pi(e') \neq y$ for all $e' \prec e$. It is easy to see that $(\pi(D), \prec_\pi)$ has order type $\omega$. After replacing $(D,\prec)$ with $(\pi(D), \prec_\pi)$ we suppose that $D$ is a dense subset of $(a,b)$. We declare $$Y := \{ x \in D \ : \ \forall z \in D (z\prec x) \rightarrow (z < x) \}.$$ That is, $Y$ is the set of $e \in D$ such that $e$ the $<$-maximal element of $D_{\preceq e}$. By density of $D$ in $(a,b)$, it is easy to see that $Y$ is infinite and that $(Y, <)$ is order-isomorphic to $({\mathbb{N}}, <)$. Thus $Y$ is an infinite discrete definable subset of ${\mathbb{R}}$. Hence the closure of $Y$ does not have interior, but infinitely many connected components. Therefore ${\ensuremath{\mathcal{S}}}$ does not have o-minimal open core.
Since (2) trivially implies (1), it is enough to show that (3) implies (2). Suppose that the open core of ${\ensuremath{\mathcal{S}}}$ is not o-minimal. By [@DMS1 2.14 (2)] there is an infinite discrete subset $D\subseteq {\mathbb{R}}$ definable in ${\ensuremath{\mathcal{S}}}$. First consider the case that $D\cap [-a,a]$ is a finite set for every $a \in {\mathbb{R}}_{>0}$. Then either $((-D) \cap [0,\infty),<)$ or $(D\cap[0,\infty),<)$ has order type $\omega$. From now on we can assume that there is $a\in {\mathbb{R}}_{>0}$ such that the cardinality of $D\cap [-a,a]$ is infinite. Thus without loss of generality we can assume that $D$ is bounded. For $\varepsilon \in {\mathbb{R}}_{>0}$ set $$D_{\varepsilon} := \{ d \in D \ : \ (d-\varepsilon,d+\varepsilon) \cap D = \{d\}\}.$$ Since $D$ is bounded, each $D_{\varepsilon}$ is finite. Moreover, since $D$ is discrete and infinite, there is a function $f: D\to {\mathbb{R}}_{>0}$ definable in ${\ensuremath{\mathcal{S}}}$ mapping $d\in D$ to the supremum of all $\varepsilon \in {\mathbb{R}}_{>0}$ with $d\in D_{\varepsilon}$. We now define the following order on $D$: let $d_1,d_2\in D$, we set $d_1 \prec d_2$ whenever one of the following conditions holds:
- $f(d_1) > f(d_2)$,
- $f(d_1)=f(d_2)$ and $d_1 < d_2$.
It can be checked easily that $(D,\prec)$ has order type $\omega$.
Let $T'$ be the theory of an infinite discrete order. By Theorem B there exists an expansion of $({\mathbb{R}},<,+)$ that has o-minimal open core and interprets a model of $T'$. However, by Proposition \[prop:ocdlo\] there is no expansion of $({\mathbb{R}},<,+)$ that has o-minimal open core and defines a model of $T'$. Therefore in Theorem B the statement “${\ensuremath{\mathcal{S}}}$ interprets a model of $T'$” cannot be replace by the statement “${\ensuremath{\mathcal{S}}}$ defines a model of $T'$”.
Noiseless NIP expansions of $({\mathbb{R}},<,+)$
================================================
Recall that an expansion of $({\mathbb{R}},<)$ is noiseless if it does not define a somewhere dense and co-dense subset of ${\mathbb{R}}$. In this section we show that every noiseless NIP expansion of $({\mathbb{R}},<,+,1)$ has definable choice and hence eliminates imaginaries. This statement will be established for the slightly larger class of noiseless expansions of $({\mathbb{R}},<,+,1)$ that do not define a Cantor set. A **Cantor set** is a non-empty compact subset of ${\mathbb{R}}$ that neither has interior nor isolated points. By [@HW-Monadic Theorem B] every NTP$_2$ (and hence every NIP) expansion of $({\mathbb{R}},<,+)$ does not define a Cantor set.
Fix a noiseless expansion ${\ensuremath{\mathcal{R}}}$ of $({\mathbb{R}},<,+,1)$ that does not define a Cantor set. Throughout this section, *definable* will mean *definable in ${\ensuremath{\mathcal{R}}}$*. For a subset $X\subseteq {\mathbb{R}}^n$, we denote the (topological) closure of $X$ by ${\operatorname{Cl}}(X)$ and the interior of $X$ by ${\operatorname{Int}}(X)$.
\[lem:isolatedpoint\] Let $X\subseteq {\mathbb{R}}$ be a non-empty definable set with empty interior. Then $X$ contains an isolated point.
Since ${\ensuremath{\mathcal{R}}}$ is noiseless, the closure ${\operatorname{Cl}}(X)$ of $X$ has empty interior. Because ${\ensuremath{\mathcal{R}}}$ does not define a Cantor set, ${\operatorname{Cl}}(X)$ has an isolated point. It follows directly that $X$ has an isolated point.
Therefore in an expansion of $({\mathbb{R}},<,+)$ that does not define a Cantor set, every definable subset of ${\mathbb{R}}$ contains a locally closed point. For expansions of the real field, the existence of definable Skolem functions in expansions satisfying the latter condition was shown in [@F-tameopen Lemma 9.1].
\[lem:nowheredense\] Let $C\subseteq {\mathbb{R}}^{n+1}$ be $\emptyset$-definable such that $C_x$ has empty interior for every $x\in {\mathbb{R}}^n$. Then there is an $\emptyset$-definable function $f: \pi(C) \to {\mathbb{R}}$ such that ${\operatorname{gr}}(f) \subseteq C$, where $\pi : {\mathbb{R}}^{n+1} \to {\mathbb{R}}^n$ is the projection onto the first $n$ coordinates.
By Lemma \[lem:isolatedpoint\] we have that for all $x\in \pi(C)$ the set $C_x$ has an isolated point whenever $C_x$ is non-empty. Let $g : \pi(C) \to {\mathbb{R}}$ map $x\in \pi(C)$ to $$\sup \{r \in {\mathbb{R}}_{>0} \ : \ \exists y \in C_x \ (y-r,y+r) \cap C_x = \{y\} \}$$ if such supremum exists, and to $1$ otherwise. Define $$D:= \{ (x,y) \in C \ : \ (y-\tfrac{g(x)}{2},y+\tfrac{g(x)}{2}) \cap C_x = \{y\}\}.$$ It is easy to check that $D_x$ is non-empty if and if $C_x$ is non-empty. For each $x\in \pi(D)$ and $y_1,y_2 \in D_x$, we have $|y_1-y_2| \geq \tfrac{g(x)}{2}$. Therefore the set $D_x$ is closed and discrete for each $x \in \pi(D)$. Let $f: \pi(C) \to {\mathbb{R}}$ be the function defined by $$x\mapsto \left\{
\begin{array}{ll}
\min D_x \cap [0,\infty), & \hbox{if $D_x \cap [0,\infty)$ is non-empty} \\
\max D_x \cap (-\infty,0), & \hbox{otherwise.}
\end{array}
\right.$$ Observe that $f$ is well-defined, because $D_x$ is closed and discrete. From the definition of $f$ we obtain directly that ${\operatorname{gr}}(f)\subseteq C$.
\[prop:skolem\] Let $A\subseteq {\mathbb{R}}^{m}\times {\mathbb{R}}^n$ be $\emptyset$-definable. Then there is an $\emptyset$-definable function $f: \pi(A) \to {\mathbb{R}}^n$ such that
1. ${\operatorname{gr}}(f)\subseteq A$,
2. $f(a) = f(b)$ whenever $a,b\in \pi(A)$ and $A_a=A_b$,
where $\pi : {\mathbb{R}}^{m+n} \to {\mathbb{R}}^m$ is the projection onto the first $m$ coordinates.
Using induction it is easy to reduce to the case that $n=1$. We can split $A$ into $B, C\subseteq {\mathbb{R}}^{m+n}$ such that $A=B\cup C$ and $$B := \{(x,y) \in A \ : \ y \in {\operatorname{Int}}(A_x)\}, \quad C := \{(x,y) \in A \ : \ y \in A_x \setminus {\operatorname{Int}}(A_x)\}.$$ Observe that $C_x$ has empty interior for each $x\in \pi(C)$. Thus by Lemma \[lem:nowheredense\] there is a definable function $f_1 : \pi(C) \to {\mathbb{R}}$ such that ${\operatorname{gr}}(f_1) \subseteq C$. Now define a subset $D\subseteq {\mathbb{R}}^{m+1}$ such that $(x,y)\in D$ whenever one the following conditions holds:
- $y$ is a midpoint of a connected component of $B_x$,
- $y = 1 + \sup ({\mathbb{R}}\setminus B_x)$ and ${\mathbb{R}}\setminus B_x$ is bounded from above,
- $y=-1 + \inf ({\mathbb{R}}\setminus B_x)$ and ${\mathbb{R}}\setminus B_x$ is bounded from below,
- $y= 0$ and $B_x = {\mathbb{R}}$.
It is easy to see that $D$ is definable, $D\subseteq B$ and $\pi(B)=\pi(D)$. Moreover, $D_x$ has empty interior for each $x\in \pi(D)$. By Lemma \[lem:nowheredense\] there is a definable function $f_2: \pi(D) \to {\mathbb{R}}$ such that ${\operatorname{gr}}(f_2) \subseteq D$. We now define $f: \pi(A) \to {\mathbb{R}}$ by $$x \mapsto \left\{
\begin{array}{ll}
f_1(x), & \hbox{if $B_x=\emptyset$;} \\
f_2(x), & \hbox{otherwise.}
\end{array}
\right.$$ It follows directly that ${\operatorname{gr}}(f)\subseteq A$. Furthermore, the reader can easily check that for $a\in \pi(A)$ the value of $f(a)$ only depends on $A_a$ and not on $a$. Therefore condition (2) holds for $f$ as well.
Theorem E follows immediately from Proposition \[prop:skolem\]. Note that Theorem E fails for NIP expansions of $({\mathbb{R}},<,+,1)$ in general. For example, the structure $({\mathbb{R}},<,+,1,{\mathbb{Q}})$ is NIP (see for example [@GH-Dependent Corollary 3.2]), does not have definable Skolem functions ([@DMS1 5.4]) and does not eliminate imaginaries ([@DMS1 5.5]).
[^1]: The first author was partially supported by NSF grant DMS-1654725.
[^2]: The name *noiseless* was suggested by Chris Miller. Being noiseless is equivalent to the statement that every definable subset of ${\mathbb{R}}$ either has interior or is nowhere dense. The latter condition has also been called *i-minimality* by Fornasiero [@F-tameopen].
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In this paper, a double-pivot simplex method is proposed. Two upper bounds of iteration numbers are derived. Applying one of the bounds to some special linear programming (LP) problems, such as LP with a totally unimodular matrix and Markov Decision Problem (MDP) with a fixed discount rate, indicates that the double-pivot simplex method solves these problems in a strongly polynomial time. A variant of Klee-Minty cube is used to show that the estimated bounds of the iteration numbers are very tight. Numerical test on three variants of Klee-Minty cubes is performed for the problems with sizes as big as $200$ constraints and $400$ variables. Dantzig’s simplex method cannot handle Klee-Minty cube problem with $200$ constraints because it needs about $2^{200} \approx 10^{60}$ iterations. But the proposed algorithm performs extremely good for all three variants.'
author:
- 'Yaguang Yang[^1]'
title: 'A double-pivot simplex algorithm and its upper bounds of the iteration numbers'
---
[**Keywords:**]{} Double-pivot algorithm, simplex method, linear programming.
Introduction
============
Since Dantzig invented the simplex method in 1940s [@dantzig49], its complexity has been a topic attracted many researchers. Since simplex methods search the optimizer among vertices which are defined by the linear constraints, the iterate moves from one vertex to the next vertex along an edge of the polytope. Therefore, the diameter of a polytope, defined as the shortest path or the least number of edges between any two vertices of the polytope, is the smallest number of iterations that the best simplex algorithm can possibly achieve. Hirsch in 1957 [@dantzig63] conjectured that the diameter of the polytope is $m-n$ for the polytope $P=\{ \x\in \R^n: \A\x \le \b \}$ where $\A \in \Z^{m \times n}$ and $m>n$. This conjecture was disapproved by Santos [@santos12] after $50$ years worth of efforts of many experts. Now, some experts, for example Santos [@santos12a], believe that the diameter of the convex polytope can be bounded by a polynomial of $(m-n)n$. This new conjecture of the bound for the diameter of the convex polytope is far away from the best-known quasi-polynomial upper bounds which are due to Kalai and Kleitman [@kalai92], Todd [@todd14], and Sukegawa [@sukegawa17]. In a recent effort [@yang18], this author showed that for a given polytope, the diameter is bounded by $\mathcal{O} \left( \frac{n^3 \Delta}{\det(\A^*)} \right)$, where $\Delta$ is the largest absolute value among all $(n-1) \times (n-1)$ sub-determinants of $\A$ and ${\det(\A^*)}$ is the smallest absolute value among all nonzero $n \times n$ sub-determinants of $\A$.
Finding the diameter of convex polytopes provides only a surmised lowest iteration number for which an optimal pivot rule may achieve. Finding actually such a pivot rule (the way to choose the next neighbor vertex) is also a difficult problem. Researchers proposed many pivot rules with the hope that they may achieve the number of iterations (in the worst case) bounded by a polynomial (see [@tz93] and references therein). However, since Klee and Minty [@km72] constructed a cube and showed that Dantzig’s rule needs an exponential number of iterations in the worst case to solve the Klee and Minty cube problem, people have showed similar results for almost every popular pivot rule [@friedmann11; @gs79; @jeroslow73; @psz08]. It is now believed that finding a pivot rule that will solve all linear programming problems in the worst case in polynomial time is a very difficult problem [@smale99].
Existing pivot rules consider one of many merit criteria to select an entering variable, such as the most negative index in the reduced cost vector (Dantzig’s rule), the best improvement rule, Bland’s least index pivoting rule, the steepest edge simplex rule, Zadeh’s rule, among others [@tz93]. Each merit criterion has its own appealing feature. However, existing simplex algorithms cannot use multiple merits at the same time because each of these algorithms updates only one variable at a time. In a slightly different view, a merit criterion may be a good choice in most scenarios but may be a poor choice in some spacial case, such as the Klee-Minty cube. Therefore, randomized pivot rules [@ghz98; @kp06] that randomly select an entering variable from the set of possible entering variables that will improve the objective function have been proposed and proved to be able to find an optimizer in a polynomial time on average [@kp06]. This shows that using a combination of merits in the selection of pivot can be beneficial.
In this paper, we consider a novel simplex algorithm for linear programming problem. This algorithm is different from all existing simplex algorithms in that it updates two variables at one iteration. This strategy looks two pivots ahead instead of focus only on the next step. We believe that this strategy is better than all existing pivot rules because it looks longer term benefit instead of a short-sighted one-step achievement. Since the proposed algorithm updates two variables at a time, it can use multiple merits in the selection of pivots at the same time in a deterministic way which is different from the randomized rules. We wish that these features give us some hope to find some strong polynomial algorithms to solve linear programming problems. We may extend the proposed algorithm to select more than two entering variables, but there is a trade-off between reducing iteration numbers and reducing the cost of each iteration.
In this paper, we use small letters with bold font for vectors and capital letters with bold font for matrices. The remainder of the paper is organized as follows. Section 2 describes the proposed algorithm. Section 3 analyzes the iteration numbers of the algorithm. Section 4 provides the information on the numerical test for three variants of Klee-Minty cubes. The concluding remarks are in Section 5.
The proposed algorithm
======================
We consider the primal linear programming problem in the standard form: $$\begin{aligned}
\begin{array}{cl}
\min & \c^{\T}\x, \\
\mbox{\rm subject to}
& \A\x=\b, \hspace{0.1in} \x \ge \0,
\end{array}
\label{LP}\end{aligned}$$ where $\A \in {\R}^{m \times n}$, $\b \in {\R}^{m}$, $\c \in {\R}^{n}$ are given, and $\x \in {\R}^n$ is the vector to be optimized. Associated with the linear programming is the dual programming that is also presented in the standard form: $$\begin{aligned}
\begin{array}{cl}
\max & \b^{\T}\y, \\
\mbox{\rm subject to} &
\A^{\T}\y +\s=\c, \hspace{0.1in} \s \ge \0,
\end{array}
\label{DP}\end{aligned}$$ where dual variable vector $\y \in {\R}^{m}$, and dual slack vector $\s \in {\R}^{n}$.
A feasible solution of the linear program satisfies the conditions of $\A \x =\b$ and $\x \ge \0$. We will denote by $B \subset \{ 1, 2, \ldots, n \}$ the index set with cardinality $| B | = m$ and $N = \{ 1, 2, \ldots, n \} \setminus B$ with cardinality $| N |=n-m$ the complementary set of $B$ such that matrix $\A$ and vector $\x$ can be partitioned as $\A=[ \A_B, \A_N]$ and $\x =[\x_B^{\Tr}, \x_N^{\Tr}]^{\Tr}$, moreover the columns of $\A_B$ are linearly independent and $\A_B \x_B = \b$, hence $\x_N=\0$. We call this $\x=[\x_B,\0]$ as the basic feasible solution. Similarly, we can partition $\c$ and $\s$ as follows: $$\c = \left[ \begin{array}{c}
\c_B \\ \c_N
\end{array} \right], \hspace{0.2in}
\s = \left[ \begin{array}{c}
\s_B \\ \s_N
\end{array} \right].$$ We denote by $\mathcal{B}$ the set of all bases $B$. In the discussion below, we make the following assumptions:
- $rank(\A) = m$.
- The primal problem (\[LP\]) has an optimal solution.
- Initial basic feasible solution $\x^0$ is given and is not an optimizer.
- All basic feasible solutions are bounded above and below, more specifically, for all $i \in B \subset \mathcal{B}$, $ \delta \le x_i \le \gamma$.
The first three assumptions are standard. The last assumption implies that the primal problem (\[LP\]) is not degenerate. We also denote by $\x^*= (\x_{B^*}, \x_{N^*})$ the optimal basic solution of (\[LP\]) with $\x_{B^*}=\A_{B^*}^{-1} \b \ge \0$ and $\x_{N^*}=\0$, by $(\y^*, \s^*)$ the optimal basic solution of the dual problem (\[DP\]) with $\y^*=\A_{B^*}^{-\Tr} \c_{B^*}$, $\s_{B^*}=\0$, and $\s_{N^*}=\c_{N^*}-\A_{N^*}^{\Tr}\A_{B^*}^{-\Tr}\c_{B^*}$, by $z^* = \c^{\Tr} \x^* = \b^{\T}\y^*$ the optimal value. Using $B-N$ partition, we can write the primal problem as $$\begin{aligned}
\begin{array}{cl}
\min & \c_B^{\T}\x_B + \c_N^{\T}\x_N, \\
\mbox{\rm subject to}
& \A_B\x_B + \A_N\x_N=\b,
\hspace{0.1in} \x_B \ge \0, \hspace{0.1in} \x_N \ge \0.
\end{array}
\label{LP1}\end{aligned}$$ Since $\A_B$ is non-singular, we can rewrite (\[LP1\]) as $$\begin{aligned}
\begin{array}{cl}
\min & \c_B^{\T}\A_B^{-1} \b
+ (\c_N - \A_N^{\T} \A_B^{-\Tr} \c_B)^{\Tr} \x_N, \\
\mbox{\rm subject to}
& \x_B = \A_B^{-1} \b - \A_B^{-1}\A_N\x_N,
\hspace{0.1in} \x_B \ge \0, \hspace{0.1in} \x_N \ge \0.
\end{array}
\label{LP2}\end{aligned}$$ Let superscript $k$ represent the $k$th iteration, the matrices and vectors in the $k$th iteration are then denoted by $\A_{B^k}$, $\A_{N^k}$, $\c_{B^k}$, $\c_{N^k}$, $\x_{B^k}$, $\x_{N^k}$, $\s_{B^k}$, and $\s_{N^k}$, where $\x^k=[\x_{B^k},\x_{N^k}]$ is the basic feasible solution of (\[LP\]) with $\x_{B^k} > \0$ and $\x_{N^k}= \0$. It is worthwhile to note that the partition of $(B^k,N^k)$ keeps updating and it is different from the partition $(B^*,N^*)$ before an optimizer is found. Let $$\bar{\c}_{N^k}^{\Tr} =( \c_{N^k} - \A_{N^k}^{\Tr}
\A_{B^k}^{-\Tr} \c_{B^k} )^{\Tr}
\label{cNbar}$$ be the reduced cost vector. Clearly, if $\bar{\c}_{N^k} \ge \0$, an optimizer is found; if $\bar{c}_{j^k} <0$ for some $j^k \in N^k$, then the entering variable $x_{\jmath^k}$ in the next vertex is chosen from the set of $\{ j^k ~|~ \bar{c}_{j^k} <0 \}$ because by increasing $x_{\jmath^k}$, the objective function $\c^{\Tr} \x=\c_{B^k}^{\T}\x_{B^k}+\bar{c}_{\jmath^k}x_{\jmath^k}$ will be reduced. Many different ways have been developed for the selection of the entering variable $x_{\jmath^k}$ under the constraint: $$\jmath^k \in \{ j^k ~|~ \bar{c}_{j^k} <0 \}.$$ Once the entering variable is selected, existing pivot rules determine the leaving variable using the following method: Denote $\bar{\b} = \A_{B^k}^{-1} \b$ and $\bar{\a}_{\jmath^k} = \A_{B^k}^{-1} \A_{\jmath^k}$, $\jmath^k \in N^k$ corresponding to the entering variable, the leaving variable $x_{\imath}$, ${\imath} \in B^k$, is determined by the following condition. $$\begin{aligned}
x_{\imath} = \min_{i \in \{ 1,\ldots,m \}}
\bar{b}_i / \bar{a}_{\jmath^k,i},
\hspace{0.1in} \mbox{subject to}
\hspace{0.1in} \bar{a}_{\jmath^k,i}>0.
\label{outVar}\end{aligned}$$ The corresponding step-size is given by $$\begin{aligned}
\min_{i \in \{ 1,\ldots,m \}}
\bar{b}_i / \bar{a}_{\jmath^k,i},
\hspace{0.1in} \mbox{subject to}
\hspace{0.1in} \bar{a}_{\jmath^k,i}>0 .
\label{stepsize}\end{aligned}$$
As we pointed out above, our strategy is to select, in a deterministic way, two entering variables from the set of non-basic variables that will reduce the objective function. According to some extensive computational experience, for example [@ps14], Dantzig’s rule is the most efficient on average among all popular pivot rules (even though Dantzig’s rule needs exponentially many pivots to find the optimal solution for Klee-Minty cubes in the worst case), therefore, we select the first entering variable $x_{\jmath_1^k}$ using Dantzig’s rule: $${\jmath_1^k} :=\{ {\jmath_1^k} ~|~ \bar{c}_{\jmath_1^k}
= \min_{j^k \in N^k} \bar{\c}_{j^k} \}.
\label{enteringVar}$$ Kitahara and Mizuno [@km13] showed that the number of iterations in existing pivot rules is significantly affected by the minimum values of all the positive elements of primal basic feasible solutions. Carefully studying Klee-Minty cube and its variants [@greenberg97; @km11; @ibrahima13] indicates that the other entering variable should be determined by taking the variable among all $j^k \in N^k$ with $\bar{c}_{j^k}<0$ such that a particular $\jmath_2^k$ will maximize the step-size, i.e., $$x_{\jmath_2^k} = \max_{\bar{\c}_{j^k}<0}
\Big\{
\min_{i \in \{ 1,\ldots,m \}} \bar{b}_i / \bar{a}_{j^k,i},
\hspace{0.1in} \mbox{subject to}
\hspace{0.1in} \bar{a}_{j^k,i}>0
\Big\}.
\label{iniVar2}$$ This strategy will be justified again in the proof of Theorem \[thm2\] and in the discussion of Remark \[longestStep\]. If ${\jmath_2^k}={\jmath_1^k}$ (which means that the most negative rule will generate the longest step), then we take the second entering variable $x_{\jmath_2^k}$ which has the second largest step-size.
Now we discuss how to choose the leaving variables. To make our notation simple, we drop the iteration index $k$ if it does not cause confusion. Let $\bar{\A}_{(\jmath_1,\jmath_2)} =
\A_B^{-1} \A_{(\jmath_1,\jmath_2)}$ where $\A_{(\jmath_1,\jmath_2)}$ is composed of the $\jmath_1$ and $\jmath_2$ columns of $\A_N$, and $\bar{\c}_{(\jmath_1,\jmath_2)} < \0$ be the two corresponding elements in $\bar{\c}_N$. For two entering indices $(\jmath_1,\jmath_2) \in N^k$ such that that $\x_{(\jmath_1,\jmath_2)}^{\Tr} =[x_{\jmath_1},x_{\jmath_2}] \ge \0$, we need $$\x^{k+1}=\x_{B^k}=\A_{B^k}^{-1}\b
-\A_{B^k}^{-1}\A_{N^k}\x_{N^k}
= \bar{\b} - \bar{\A}_{(\jmath_1,\jmath_2)}
\x_{(\jmath_1,\jmath_2)} \ge \0.
\label{updatedXB}$$ Therefore, the problem of finding a good new vertex is reduced to minimize the following linear programming problem. $$\begin{aligned}
\begin{array}{cl}
\min & \bar{\c}_{(\jmath_1,\jmath_2)}^{\T}
\x_{(\jmath_1,\jmath_2)}, \\
\mbox{\rm subject to}
& \bar{\A}_{(\jmath_1,\jmath_2)}
\x_{(\jmath_1,\jmath_2)} \le \bar{\b},
\hspace{0.1in} \x_{(\jmath_1,\jmath_2)} \ge \0.
\end{array}
\label{twoDim}\end{aligned}$$ Here the third merit criterion is introduced, which is to determine the values of the two entering variables to minimize the objective function under the constraints of (\[twoDim\]). These constraints make sure that the updated $\x_B \ge \0$ and the leaving variables are zeros. As this problem has only two variables, the solution is slightly more complicate than the selection of the single leaving variable in existing pivot rules, but is still simple and straightforward. Divide $\bar{\A}_{(\jmath_1,\jmath_2)}$ into two parts: $\bar{\A}_1$ has the rows with at least one positive element, and $\bar{\A}_2 \le \0$ has the rows with all elements smaller than or equal to zero . Partition $\bar{\b}=\A_B^{-1} \b$ into the corresponding $\bar{\b}_1$ and $\bar{\b}_2$. Since elements in $\bar{\A}_2$ are smaller than to equal to zero, in view of (\[updatedXB\]) or (\[twoDim\]), introducing positive entering variables will keep the corresponding elements in $\x_B$ to be positive. For $\bar{\A}_1$, in view of (\[updatedXB\]) or (\[twoDim\]), introducing positive variables may change the sign of $\x_B$. If the number of rows in $\bar{\A}_1$ is greater than or equal to $2$, for any two independent rows $(i_1,i_2)$ of $\bar{\A}_1$, solving $$\bar{\A}_1(i_1,i_2) \left[ \begin{array}{c}
x_{\jmath_1} \\ x_{\jmath_2} \end{array} \right]
= \bar{\b}_1(i_1,i_2)
\label{subEq}$$ will give a possible vertex in the convex polygon defined in (\[twoDim\]). Therefore $\x_2 := [x_{\jmath_1}, x_{\jmath_2}] \ge \0$ is a feasible vertex of the polygon if $\bar{\A}_1 \x_2 \le \bar{\b}_1$ holds. Otherwise, it is not feasible and will not be considered further. Two special feasible vertices should also be considered, i.e., $\x_2 := [x_{\jmath_1}, 0]$ and $\x_2 := [0, x_{\jmath_2}]$ which correspond to the most negative rule and the longest step-size rule respectivily. For all feasible vertices of the convex polygon defined in (\[twoDim\]), we select the vertex that minimizes the objective function of (\[twoDim\]). The corresponding row indices $(\imath_1,\imath_2)$ that form the selected vertex determine the leaving variables. If the number of rows in $\A_1$ is exact one, the longest step pivot rule is used.
The proposed algorithm is therefore as follows:
[ ]{}\
Data: Matrix $\A$, vectors $\b$ and $\c$. [ ]{}\
initial basic feasible solution $\x^0$, and its related partitions $\x_{B^0}$, $\x_{N^0}$, $\A_{B^0}$, $\A_{N^0}$, $\c_{B^0}$, $\c_{N^0}$, $(\A_{B^0})^{-1}$, and $\bar{\c}_{N^0}=\c_{N^0} - \c_{B^0} (\A_{B^0})^{-1} \A_{N^0}$.
- While $\min (\bar{\c}_{N^k}) < 0$
- If at least two elements of $\bar{\c}_{N^k}$ are smaller than zero
- The first entering variable $x_{\jmath_1^k}$ is determined by Dantzig’s rule. For all negative elements of $\bar{\c}_{N^k}$ other than the most negative elements $\bar{\c}_{\jmath_1^k}$, determine the $x_{\jmath_2^k}$ such that the second entering variable will take the longest step. Two special vertices, $[x_{\jmath_1^k}, 0]$ and $[0, x_{\jmath_2^k}]$ are obtained.
- Divide $\bar{\A}_({\jmath_1,\jmath_2})$ into two parts: $\bar{\A}_1$ whose row has positive elements and $\bar{\A}_2 \le \0$. Partition $\A_{B^k}^{-1} \b$ into the corresponding $\bar{\b}_1$ and $\bar{\b}_2$.
- If the number of rows of $\bar{\A}_1$ is greater than or equal to $2$
- Compute all vertices in two dimensional plane by solving (\[subEq\]).
- Determine all feasible vertices which satisfy $\x_2=[x_{\jmath_1},x_{\jmath_2}]^{\Tr} \ge \0$ and $\bar{\A}_1 \x_2 \le \bar{\b}_1$.
- Find a pair of entering variables among all feasible vertices $\x_2$ (including the two special vertices) that minimizes the objective $[\bar{c}_{\jmath_1}, \bar{c}_{\jmath_2}] \x_2$.
- Update base $\A_{B^k}$ and $\c_{B^k}$, non-base $\A_{N^k}$ and $\c_{N^k}$. Compute $\A_{B^k}^{-1}$ and $\bar{\c}_{N^k} = \c_{N^k} - \c_{B^k} \A_{B^k}^{-1} \A_{N^k}$.
- Else if there is only one row in $\bar{\A}_1$
- The longest step rule is applied.
- Update base $\A_{B^k}$ and $\c_{B^k}$, non-base $\A_{N^k}$ and $\c_{N^k}$. Compute $\A_{B^k}^{-1}$ and $\bar{\c}_{N^k} = \c_{N^k} - \c_{B^k} \A_{B^k}^{-1} \A_{N^k}$.
- end (if)
- Else if only one element of $(\bar{\c}_{N^k})$ is negative,
- Dantzig’s rule (which is also the longest rule) is applied.
- Update base $\A_{B^k}$ and $\c_{B^k}$, non-base $\A_{N^k}$ and $\c_{N^k}$. Compute $\A_{B^k}^{-1}$ and $\bar{\c}_{N^k} = \c_{N^k} - \c_{B^k} \A_{B^k}^{-1} \A_{N^k}$.
- end (if)
- $k \Leftarrow k+1$.
- end (while)
\[mainAlg\]
We can modify the algorithm by selecting two entering variables using the indices corresponding to the [**two**]{} most negative elements in $\bar{\c}_{N^k}$. In the numerical test section, we will show that this is a not a good strategy. \[mostNeg\]
Analysis of the algorithm
=========================
In this section, we provide two upper bounds of the iteration numbers for the proposed algorithm using the strategy developed in [@km11; @km13; @ye11].
Let $r$ be a real number and $\lceil r \rceil$ be the smallest integer bigger than $r$. Let $\gamma_P^*$ be the maximum value of all elements of $\x^*$ and $$\gamma_D = \max_{k} \{ \gamma_D^k \} =
\max_{k} \Big\{ \max_{j^k \in N^k}
\{ -\bar{c}_{j^k} ~|~ \bar{c}_{j^k} < 0 \} \Big\}.
\label{gammaD}$$ Let $(B^k,N^k)$ and $(B^*,N^*)$ be the partitions of base and non-base variables at iteration $k$ and at the end of the program when the optimization is achieved. Let $\x^*$ be partitioned using $(B^k,N^k)$ but not $(B^*,N^*)$, i.e., $$\x^*= \left[ \begin{array}{c}
\x_{B^k}^* \\ \x_{N^k}^* \end{array} \right].$$ The first lemma is derived using exactly the same argument but stating a slightly improved result of [@km11; @km13].
(Kitahara and Mizuno) Let $\x^*$ be partitioned using $(B^k,N^k)$ and $z^*$ be the optimal value of (\[LP\]), we have $$\c^{\Tr} \x^k - z^*
\le \gamma_D^k \| \x_{N^k}^* \|_1.
\label{1stEst}$$ \[firstL\]
Since $\x^*$ is partitioned using $(B^k,N^k)$, we have $(\x_{B^k}^* , \x_{N^k}^*) \ge \0$, and $$\begin{aligned}
\A_{B^k}\x_{B^k}^* +\A_{N^k}\x_{N^k}^*= \b. \nonumber\end{aligned}$$ This gives $$\begin{aligned}
\x_{B^k}^*
= \A_{B^k}^{-1} \b -\A_{B^k}^{-1} \A_{N^k}\x_{N^k}^*.
\nonumber \end{aligned}$$ Therefore, we have $$\begin{aligned}
\c^{\Tr} \x^* & = &
\c_{B^k}^{\Tr}\x_{B^k}^* + {\c}_{N^k}^{\Tr}\x_{N^k}^*
\nonumber \\
& = & \c_{B^k}^{\Tr} \A_{B^k}^{-1} \b
-{\c}_{B^k}^{\Tr}\A_{B^k}^{-1}\A_{N^k}\x_{N^k}^*
+{\c}_{N^k}^{\Tr}\x_{N^k}^*
\nonumber \\
& = & \c_{B^k}^{\Tr} \A_{B^k}^{-1} \b
+ ({\c}_{N^k}^{\Tr}-{\c}_{B^k}^{\Tr}\A_{B^k}^{-1}\A_{N^k})
\x_{N^k}^*.\end{aligned}$$ Using this relation and (\[cNbar\]), we have $$\begin{aligned}
z^* & = & \c^{\Tr} \x^* \nonumber \\
& = & \c_{B^k}^{\Tr} \A_{B^k}^{-1} \b
+ \bar{\c}_{N^k}^{\Tr}\x_{N^k}^* \nonumber \\
& \ge & \c^{\Tr} \x^k - \gamma_D^k \| \x_{N^k}^* \|_1.
\label{est1}\end{aligned}$$ This finishes the proof. [<1.5em -1.5em plus0em minus0.5em height0.75em width0.5em depth0.25em]{}
If $B^* \ne B^k$, i.e., $\x^k$ is not an optimizer, from (\[est1\]), it must have $\bar{\c}_{N^k}^{\Tr}\x_{N^k}^* <0$. Therefore, there are $j^k \in N^k$ such that $$\bar{c}_{j^k} <0 \hspace{0.1in} \mbox{and}
\hspace{0.1in} x_{j^k}^*>0.
\label{NkBstar}$$ This means that for $j^k \in N^k \cap B^*$, $x_{j^k}$ should be the entering variable. The problem is that one does not know $B^*$ before an optimizer is found. \[totalD\]
We may also partition $\x^k$ using $(B^*,N^*)$ as $$\x^k= \left[ \begin{array}{c}
\x_{B^*}^k \\ \x_{N^*}^k \end{array} \right].$$ This gives $$\begin{aligned}
\A_{B^*}\x_{B^*}^k +\A_{N^*}\x_{N^*}^k= \b, \nonumber\end{aligned}$$ and $$\begin{aligned}
\x_{B^*}^k
= \A_{B^*}^{-1} \b -\A_{B^*}^{-1} \A_{N^*}\x_{N^*}^k.
\nonumber \end{aligned}$$ Similar to the derivation of (\[1stEst\]), we have $$\begin{aligned}
\c^{\Tr} \x^k = \c^{\Tr} \x^* +
({\c}_{N^*}- {\c}_{B^*}\A_{B^*}^{-1} \A_{N^*})^{\Tr} \x_{N^*}^k
= z^* + \bar{\c}_{N^*}^{\Tr}\x_{N^*}^k.
\nonumber \end{aligned}$$ If $x_{j}^k > 0$, we have $x_{j}^k \in B^k$. For $j \in {N^*} \cap B^k$, since $\bar{\c}_{N^*} \ge \0$, we have $$\begin{aligned}
z^* & \ge \c^{\Tr} \x^k -
\max \{ x_j^k ~|~ j \in N^* \cap B^k \}
\| \bar{\c}_{N^*} \|_1.\end{aligned}$$
If $N^* \ne N^k$, i.e., $\x^k$ is not an optimizer, it must have $\bar{\c}_{N^*}^{\Tr}\x_{N^*}^k >0$. Therefore, there are $j^* \in N^*$ such that $$\bar{c}_{j^*} >0 \hspace{0.1in} \mbox{and}
\hspace{0.1in} x_{j^*}^k>0.
\label{NkBstar2}$$ This means that for $j^* \in N^* \cap B^k$, $0 < x_{j^*}^k \in B^k$ should be the leaving variable. The problem is that one does not know $N^*$ before an optimizer is found. \[totalDiff\]
Let $\gamma_{\ell}=\min_{k} x_{\jmath_2^k}$, where $x_{\jmath_2^k}$ is defined in (\[iniVar2\]), i.e., $x_{\jmath_2^k}$ is the longest step among all possible entering variables with $\bar{c}_{j^k} <0$ and $j^k \in N^k$; and define $$\delta_D =\min_{k} \delta_D^k
= \min_{k} \Big\{ \min \{ -\bar{c}_{j^k} ~| ~j^k \in N^k
\hspace{0.1in}\mbox{and} \hspace{0.1in} \bar{c}_{j^k}<0 \}
\Big\}.$$ Considering Algorithm \[mainAlg\], our next lemma is an improvement of the one in [@km13].
Let $\x^k$ and $\x^{k+1}$ be the $k$th and $(k+1)$th iterates generated by Algorithm \[mainAlg\]. If $\x^k$ is not optimal and $\x^k \neq \x^{k+1}$, then, we have $$\c^{\Tr} \x^k - \c^{\Tr} \x^{k+1} \ge \delta_D \gamma_{\ell}.
\label{myIneq}$$ \[diffObj\]
Since $\x^k \neq \x^{k+1}$, from the derivation of (\[twoDim\]), the difference of the objective functions between $k$th and $(k+1)$th iterations is actually the solution of (\[twoDim\]), which is smaller than the special case when only one entering variable $x_{\jmath_2^k}$, which would generate the longest step among $\bar{c}_{j^k}<0$ for all $j^k \in N^k$, is selected. Let $\bar{\x}_{(\jmath_1^k,\jmath_2^k)}$ be the optimal solution of (\[twoDim\]) at iteration $k$. Therefore $$\begin{aligned}
\c^{\Tr} \x^k - \c^{\Tr} \x^{k+1}
& = & - \bar{\c}_{(\jmath_1^k,\jmath_2^k)}^{\T}
\bar{\x}_{(\jmath_1^k,\jmath_2^k)}
\nonumber \\
& \ge & -\bar{c}_{\jmath_2^k}\bar{x}_{\jmath_2^k}
\nonumber \\
& \ge & \delta_D \gamma_{\ell}.
\label{myIneq1}\end{aligned}$$ This finishes the proof. [<1.5em -1.5em plus0em minus0.5em height0.75em width0.5em depth0.25em]{}
Lemma \[diffObj\] says that for Algorithm \[mainAlg\], the objective value decreases in every iteration at least by a constant $\delta_D \gamma_{\ell}$.\[constantD\]
From Lemmas \[firstL\], \[diffObj\] and Remark \[constantD\], it is easy to show that the following upper bound of iteration numbers of Algorithm \[mainAlg\] holds.
Suppose that we generate a sequence of basic feasible solutions by Algorithm \[mainAlg\] from an initial iterate $\x^0$. Then, the number of total iterations is bounded above by $$\Bigl\lceil
\frac{\c^{\Tr} \x^0 - z^*}{\delta_D \gamma_{\ell}}
\Bigr\rceil \le
\Bigl\lceil
\frac{\gamma_D^0\| \x^* \|_1 }
{\delta_D \gamma_{\ell}}
\Bigr\rceil
\label{bound}$$ \[mainThe\]
The bound of the left side is obvious. By the defintion of $\gamma_D^k$, we have $$\bar{\c}_{N^k}^{\Tr}\x_{N^k}^* \ge
-{\gamma_D^k \| \x_{N^k}^* \|_1 }.$$ Therefore, for initial step, the last inequality of (\[est1\]) can be replaced by $$\c^{\Tr} \x^0 - z^* \le \gamma_D^0 \| \x^* \|_1 .$$ Since every iteration will reduce the objective function at least a constant $\delta_D \gamma_{\ell}$, we need at most $$\Bigl\lceil
\frac{\gamma_D^0 \| \x^* \|_1 }
{\delta_D \gamma_{\ell}}
\Bigr\rceil$$ iterations to find the optimal solution. [<1.5em -1.5em plus0em minus0.5em height0.75em width0.5em depth0.25em]{}
The upper bound given in Theorem \[mainThe\] is smaller than the one in [@km13] because (a) $\gamma_{\ell}$ is the smallest value in all longest steps among all iterates while the corresponding number in [@km13] is the smallest value in all nonzero components among all iterates $\x^k$, (b) $\| \x^* \|_1 $ depends only on the optimal solution of $\x^*$, and (c) $\gamma_D^0$ depends only on the $\c$.
Now, we present a upper bound in terms of only $\delta$ and $\gamma$.
Assume that the $k$th iterate generated by Algorithm \[mainAlg\] is not an optimizer. Let $$t = m \frac{\gamma}{\delta}
\log\left( m \frac{\gamma}{\delta} \right)
\label{tIter}$$ then there is a $\bar{j} \in B^k$, a corresponding $x_{\bar{j}}^k>0$, after at most another $\lceil t \rceil$ iterations, $x_{\bar{j}}^{k+t}$ becomes zero and stay zero since then. \[thm2\]
In view of (\[est1\]) in Lemma \[firstL\], since $\x^*$ has at most $m$ nonzero elements and $\bar{\c}_{N^k}^{\Tr}\x_{N^k}^* \ge -\gamma_D^k (m \gamma)$, we have $${\c}^{\Tr} \x^k -z^* \le m \gamma \gamma_D^k.$$ Using this inequality, together with (\[myIneq1\]) in Lemma \[diffObj\] and (\[enteringVar\]), we have $$\begin{aligned}
\c^{\Tr} \x^k - \c^{\Tr} \x^{k+1}
& = & - \bar{\c}_{(\jmath_1^k,\jmath_2^k)}^{\T}
\bar{\x}_{(\jmath_1^k,\jmath_2^k)}
\nonumber \\
& \ge & -\bar{c}_{\jmath_1^k}\bar{x}_{\jmath_1^k}
\nonumber \\
& \ge & \gamma_D^k \delta
\nonumber \\
& \ge & \frac{\delta}{m\gamma}
\left( {\c}^{\Tr} \x^k -z^* \right).
\nonumber $$ This shows $$\begin{aligned}
\c^{\Tr} \x^k - z^* - (\c^{\Tr} \x^{k+1} -z^*)
\ge \frac{\delta}{m\gamma}
\left( {\c}^{\Tr} \x^k -z^* \right)
\nonumber $$ or equivalently $$\begin{aligned}
\c^{\Tr} \x^{k+1} -z^*
\le \left( 1 - \frac{\delta}{m\gamma} \right)
\left( {\c}^{\Tr} \x^k -z^* \right).
\nonumber $$ Therefore, for any integer $t>0$, we have $$\begin{aligned}
\frac{\c^{\Tr} \x^{k+t} -z^*} {\c^{\Tr} \x^k -z^*}
\le \left( 1 - \frac{\delta}{m\gamma} \right)^t.
\label{myIneq2}\end{aligned}$$ Since $| B^k |=m$ and $$\c^{\Tr} \x^k -z^* = \x^{k^{\Tr}} \s^*
= \sum_{j \in B^k} x_j^k s_j^*,$$ there must have a $\bar{j} \in B^k$ such that $$x_{\bar{j}}^k s_{\bar{j}}^* \ge \frac{1}{m} (\c^{\Tr} \x^k -z^* ).$$ Using Assumption 4, $\gamma \ge x_{\bar{j}}^k > 0$, we have $$\begin{aligned}
s_{\bar{j}}^* \ge \frac{1}{mx_{\bar{j}}^k } (\c^{\Tr} \x^k -z^* )
\ge \frac{1}{m \gamma } (\c^{\Tr} \x^k -z^* ).
\label{myIneq3}\end{aligned}$$ Moreover, for any integer $t >0$, we have $$\c^{\Tr} \x^{k+t} -z^* = \s^{*^{\Tr}}\x^{k+t}
\ge x_{\bar{j}}^{k+t} s_{\bar{j}}^*.$$ this gives $$\begin{aligned}
x_{\bar{j}}^{k+t} \le \frac{\c^{\Tr} \x^{k+t} -z^* }{s_{\bar{j}}^*}.
\label{myIneq4}\end{aligned}$$ Substituting (\[myIneq3\]) and (\[myIneq2\]) into (\[myIneq4\]) gives $$\begin{aligned}
x_{\bar{j}}^{k+t}
\le m \gamma \frac{\c^{\Tr} \x^{k+t} -z^* }
{\c^{\Tr} \x^k -z^* }
\le m \gamma \left( 1 - \frac{\delta}{m\gamma} \right)^t.
\label{myIneq5}\end{aligned}$$ Substituting (\[tIter\]) into (\[myIneq5\]) and using the identity $x^{\log_b y}= y^{\log_b x}$ and the inequality $\log(1-x) \le -x$ for all $x \le 1$, we have $$\begin{aligned}
x_{\bar{j}}^{k+t} & \le & m \gamma \left( 1 - \frac{\delta}{m\gamma} \right)^{m \frac{\gamma}{\delta}
\log\left( m \frac{\gamma}{\delta} \right)}
\nonumber \\
& = & m \gamma \left[ \left( 1 - \frac{\delta}{m\gamma}
\right)^{\log\left( m \frac{\gamma}{\delta} \right)}
\right]^{m \frac{\gamma}{\delta}}
\nonumber \\
& = & m \gamma \left[
\left( m \frac{\gamma}{\delta} \right)^{\log
\left( 1 - \frac{\delta}{m\gamma} \right)}
\right]^{m \frac{\gamma}{\delta}}
\nonumber \\
& \le & m \gamma \left[ \left(
m \frac{\gamma}{\delta} \right)^{- \frac{\delta}{m\gamma}}
\right]^{m \frac{\gamma}{\delta}} \le \delta.
\label{myIneq6}\end{aligned}$$ Therefore, after at most $\lceil t \rceil$ iterations, $x_{\bar{j}}^{k+t}< \delta$ holds. In view of Assumption 4, we conclude that $x_{\bar{j}}^{k+t}$ is not a basic variable of $B^{k+t}$ and (\[myIneq2\]) asserts that it will not be a basic variable thereafter. [<1.5em -1.5em plus0em minus0.5em height0.75em width0.5em depth0.25em]{}
The scenario described in the theorem can occur at most one time for each optimal non-basic variable and since there are $n-m$ non-basic optimal variables, we have the following theorem.
For the double-pivot algorithm \[mainAlg\], it needs at most $(n-m) \Bigl\lceil
m \frac{\gamma}{\delta}
\log\left( m \frac{\gamma}{\delta} \right)
\Bigr\rceil$ iterations to find the optimal solution of (\[LP\]).
The way of selecting $x_{\bar{j}}^{k}$ below (\[myIneq2\]) implies that one should consider the entering variable that takes the longest step because this variable is likely an optimal non-basic variable. \[longestStep\]
Similar to the argument in [@km13a], we can apply the above theorem to some special linear programming problems, such as LP with a totally unimodular matrix and Markov Decision Problem with a fixed discount rate, and show that the double-pivot algorithm solves these special LP problems in a strongly polynomial time.
The tightness of the bounds in the two theorems can be seen from the following problem provided in [@km11]: $$\begin{aligned}
\begin{array}{cl}
\min & -\sum_{i=1}^m x_i \\
\mbox{subject to} &
x_1 + x_{m+1} = 1, \\
& 2 \sum_{i=1}^{k-1} x_i +x_k + x_{m+k} = 2^k-1
\hspace{0.1in} k=2, \ldots, m,
\\
& x_i \ge 0 \hspace{0.1in} i=1, \ldots, 2m.
\end{array}
\label{3rdProblem}\end{aligned}$$ Assuming that the initial point is taken as $\x^0 =[0,\ldots,0,1,\ldots,1]$ (there are $m$ zeros and $n-m$ ones) and Dantzig’s rule is used, for this problem, Kitathra and Mizuno showed [@km11] that the bound of Theorem \[thm2\] is reduced to $\lceil (2m \log 2)2^m \rceil$, while the actual iteration number is $2^m-1$. The estimated bound is reasonably tight. We show that the bound of Theorem \[mainThe\] is much tighter than the one of Theorem \[thm2\]. For this problem, it is easy to see that the first $m$ variables of the optimal solution are $[x_1^*,\ldots,x_m^*]=[ 0, \ldots, 0,2^m-1]$ with optimal objective function $-(2^m-1)$ and the objective function at initial $\x^0$ is zero. Therefore, we have $\c^{\Tr}\x^0-z^*=2^m-1$. Since every component of any basic feasible solution is a positive integer (because every basic matrix is lower triangle and it is easy to see the claim), this shows that $\delta_D = 1$ because either $x_1=1$ or $x_{m+1}=1$. In the first iteration, noticing that $B^0=\{ m+1, m+2,\ldots, 2m \}$ and the entering variable $\bar{x}_{\jmath_2^k}=x_m=2^m-1$. In view of Theorem \[mainThe\], it needs only one iteration to find the optimal solution and $\gamma_{\ell}=2^m-1$. This claim is verified in the numerical test in the next section for several variants of Klee-Minty cube.
Numerical test
==============
Klee-Minty cube and its variants have been used to prove that several popular simplex algorithms need exponential number of iterations in the worst case to find an optimizer. In this section, three variants of Klee-Minty cube [@greenberg97; @ibrahima13; @km11] are used to test the proposed algorithm.
The first variant of Klee-Minty cube is given in [@greenberg97]: $$\begin{aligned}
\begin{array}{cl}
\min & -\sum_{i=1}^m 2^{m-i} x_i \\
\mbox{subject to} &
\left[
\begin{array}{cccccc}
1 & 0 & 0 & \ldots & 0 & 0 \\
2^2 & 1 & 0 & \ldots & 0 & 0 \\
2^3 & 2^2 & 1 & \ldots & 0 & 0 \\
\vdots & \vdots & \vdots & \ddots & 0 & 0 \\
2^{m-1} & 2^{m-2} & 2^{m-3} & \ldots & 1 & 0 \\
2^m & 2^{m-1} & 2^{m-2} & \ldots & 2^2 & 1 \\
\end{array}
\right]
\left[ \begin{array}{c}
x_1 \\ x_2 \\ \vdots \\ \vdots \\ x_{m-1} \\ x_m
\end{array} \right]
\le
\left[ \begin{array}{c}
5 \\ 25 \\ \vdots \\ \vdots \\ 5^{m-1} \\ 5^m
\end{array} \right]
\\
& x_i \ge 0 \hspace{0.1in} i=1, \ldots, m.
\end{array}
\label{1stProblem}\end{aligned}$$ The optimizer is $[ 0, \ldots, 0,5^m ]$ with optimal objective function $-5^m$.
The second variant of Klee-Minty cube is given in [@ibrahima13]: $$\begin{aligned}
\begin{array}{cl}
\min & -\sum_{i=1}^m 10^{m-i} x_i \\
\mbox{subject to} &
2 \sum_{j=1}^{i-1} 10^{i-j} x_j +x_i \le 100^{i-1}
\hspace{0.1in} i=1, \ldots, m, \\
& x_i \ge 0 \hspace{0.1in} i=1, \ldots, m.
\end{array}
\label{2ndProblem}\end{aligned}$$ The optimizer is $[ 0, \ldots, 0,10^{2(m-1)}]$ with optimal objective function $-10^{2(m-1)}$.
The third variant of Klee-Minty cube is given in [@km11] (its standard form was discussed in the previous section): $$\begin{aligned}
\begin{array}{cl}
\min & -\sum_{i=1}^m x_i \\
\mbox{subject to} &
x_1 \le 1, \\
& 2 \sum_{i=1}^{k-1} x_i +x_k \le 2^k-1
\hspace{0.1in} k=2, \ldots, m,
\\
& x_i \ge 0 \hspace{0.1in} i=1, \ldots, m.
\end{array}
\label{3rdProblem}\end{aligned}$$ The optimizer is $[ 0, \ldots, 0,2^m-1]$ with optimal objective funtion $-(2^m-1)$. The test results are summarized in Table \[table1\]. All initial points are selected as $[0, \ldots, 0]^{\Tr}$. For the first variant of Klee-Minty cube [@greenberg97], using the most two negative elements of $\bar{c}_{N^k}$ to choose the entering variables (described in Remark \[mostNeg\]) is better than the strategy of Dantzig’s rule which uses the most negative element of $\bar{c}_{N^k}$ to choose the entering variable. The pivot rule with the most two negative elements uses half of the iterations of Dantzig’s rule but the iteration numbers still increase exponentially fast. When the size $m \ge 18$, the program freezes because iteration numbers are very big and the computational time is very long.
Algorithm \[mainAlg\] is much more impressive. For all problems in three variants, only one iteration is needed to find the optimal solution, except for the problem with dimension $m=200$ in variant 2 [@ibrahima13] because Matlab R2016a on computer Dell Inspiron 3847 cannot store the big value (bigger than 10E+310) in vector $\b$. This shows that the estimated bound of Theorem \[mainThe\] is reachable.
We also compared the tests result with the one in [@ibrahima13] which uses randomized pivot method. For $m=100$, the randomized pivot method uses more than $1000$ iterations to find the solution for a variant of Klee-Minty cube on average of $200$ runs; for $m=200$, the randomized pivot method uses more than $5000$ iterations to find the solution on average of $200$ runs. Using Algorithm \[mainAlg\], it takes one iteration for these problems. The proposed double-pivot algorithm is much more efficient than the randomized algorithm for these Klee-Minty cube problems. This result justifies a moderate computational cost increase in each iteration.
[|c|c|c|c|c|c|]{} & & &\
& [Dantzig]{} & Remark \[mostNeg\] & Alg. 2.1 & Alg. 2.1 & Alg. 2.1\
2 & 3 & 2 & 1 & 1 & 1\
3 & 7 & 4 & 1 & 1 & 1\
4 & 15 & 8 & 1 & 1 & 1\
5 & 31 & 16 & 1 & 1 & 1\
6 & 63 & 32 & 1 & 1 & 1\
7 & 127 & 64 & 1 & 1 & 1\
8 & 255 & 128 & 1 & 1 & 1\
9 & 511 & 256 & 1 & 1 & 1\
10 & 1023 & 512 & 1 & 1 & 1\
11 & $2^{11}-1$ & 1024 & 1 & 1 & 1\
12 & $2^{12}-1$ & $2^{11}$ & 1 & 1 & 1\
13 & $2^{13}-1$ & $2^{12}$ & 1 & 1 & 1\
14 & $2^{14}-1$ & $2^{13}$ & 1 & 1 & 1\
15 & $2^{15}-1$ & $2^{14}$ & 1 & 1 & 1\
16 & $2^{16}-1$ & $2^{15}$ & 1 & 1 & 1\
17 & - & $2^{16}$ & 1 & 1 & 1\
18 & - & - & 1 & 1 & 1\
19 & - & - & 1 & 1 & 1\
20 & - & - & 1 & 1 & 1\
21 & - & - & 1 & 1 & 1\
22 & - & - & 1 & 1 & 1\
23 & - & - & 1 & 1 & 1\
24 & - & - & 1 & 1 & 1\
25 & - & - & 1 & 1 & 1\
26 & - & - & 1 & 1 & 1\
27 & - & - & 1 & 1 & 1\
28 & - & - & 1 & 1 & 1\
29 & - & - & 1 & 1 & 1\
30 & - & - & 1 & 1 & 1\
100 & - & - & 1 & 1 & 1\
200 & - & - & 1 & - & 1\
\[table1\]
Conclusion
==========
In this paper, a double-pivot simplex method is proposed. Two upper-bounds of the iteration numbers for the proposed algorithm are derived. The first bound is very tight and can be reached. The second bound, when it is applied to some special linear programming problems, such as LP with a totally unimodular matrix and Markov Decision Problem with a fixed discount rate, shows that the proposed algorithm will find the optimal solution in a strongly polynomial time. The numerical test for Klee-Minty cube problems shows very promising result. It is hoped that the double-pivot strategy may lead to some strongly polynomial algorithms for general linear programming problems.
The Matlab codes used for the tests of the Klee-Minty variants in [@greenberg97; @ibrahima13; @km11] are available up request to the author.
[99]{}
N. Bonifas, M.D. Summa, F. Eisenbrand, N. Hahnle, and M. Niemeier, (2014), On sub-determinants and the diameter of polyhedra. Discrete and Computational Geometry, 52, 102-115.
G.B. Dantzig, (1949), Programming in a linear structure, Econometrica 17, 73–74.
G.B. Dantzig, (1963), Linear programming and extensions, Princeton University Press, Princeton, 1963.
O. Friedmann, (2011), A subexponential lower bound for Zadeh’s pivoting rule for solving linear programs and games., In: IPCO, pp. 192–206.
B. Gartner, M. Henk, and G.M. Ziegler (1998), Randomized simplex algorithms on Klee-Minty cubes Combinatorica, 18(3), 49-372.
D. Goldfarb and W.Y. Sit, (1979), Worst case behavior of the steepest edge simplex method, Discrete Applied Mathematics, 1, 277-285.
H. J. Greenberg, (1997), Klee-Minty Polytope Shows Exponential Time Complexity of Simplex Method, University of Colorado at Denver, http://www.cudenver.edu/ hgreenbe.
F. Ihrahima, (2013), Degeneracy and geometry in the simplex method, Stanford University report, available from the Internet, https://pdfs.semanticscholar.org/0b28/52b085df3288d0ddcc28c5e511082fd03fef.pdf.
R.G., Jeroslow, (1973), The simplex algorithm with the pivot rule of maximizing criterion improvement, Discrete Mathematics, 4, 367-377.
G. Kalai and D. Kleitman, (1992), A quasi-polynomial bound for the diameter of graphs of polyhedra, Bulletin of the American Mathematical Society, 26, 315-316.
J.A. Kelner and D.A. Spielman, (2006), A randomized polynomial-time simplex algorithm for linear programming, Proceedings of the thirty-eighth annual ACM symposium on Theory of Computing, 51-60.
T. Kitahara and S. Mizuno, (2011), Klee-Minty’s LP and Upper Bounds for Dantzig’s Simplex Method, Operations Research Letters, 39(2), 88-91.
T. Kitahara and S. Mizuno, (2013), A bound for the number of different basic solutions generated by the simplex method, Mathematical Programming, 137, 579-586.
T. Kitahara and S. Mizuno, (2013) An upper bound for the number of different solutions generated by the primal simplex method with any selection rule of entering variables, Asia-Pacific Journal of operational research, 30, 1340012, \[10 pages\].
V. Klee and G.J. Minty, (1972), How good is the simplex algorithm? In O. Shisha, editor, Inequalities, III, 159-175. Academic Press, New York, NY.
Paparrizos K., Samaras N., Zissopoulos D. (2008) Linear Programming: Klee-Minty Examples. In: Floudas C., Pardalos P. (eds) Encyclopedia of Optimization. Springer, Boston, MA
N. Ploskas and N. Samaras, (2014), Pivoting rules for the revised simplex algorithm, Yugoslav Journal of Operations Research, 24(3), 321-332.
F. Santos, (2012), A counterexample to the Hirsch conjecture. Annals of Mathematics, 176, 383-412.
F. Santos, (2012), The Hirsch conjecture has been disproved: An interview with Francisco Santos. EMS Newsletter December 2012.
S. Smale, (1999). Mathematical problems for the next century. In Arnold, V. I.; Atiyah, M.; Lax, P.; Mazur, B. Mathematics: frontiers and perspectives. American Mathematical Society, 271–294.
N. Sukegawa, (2017), Improving bounds on the diameter of a polyhedron in high dimensions, Discrete Mathematics, 340, 2134-2142.
T. Terlaky and S. Zhang, (1993), Pivot rules for linear programming: A survey on recent theoretical developments, Annals of Operations Research, Vol. 46 (1), 203–233.
M. J. Todd, (2014), An improved Kalai–Kleitman bound for the diameter of a polyhedron, SIAM Journal on Discrete Mathematics, 26(2), 1944-1947.
Y. Yang, (2018), On the diameter of polytopes, arXiv:1809.06780v1.
Y. Ye, (2011), The simplex and policy-iteration methods are strongly polynomial for the Markov decision problem with a fixed discount rate, Mathematics of Operations Research, 36(4), 593-603.
[^1]: US NRC, Office of Research, 11555 Rockville Pike, Rockville, 20850. Email: [email protected].
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We develop a non-perturbative “gauge Mean Field Theory” (gMFT) method to study a general effective spin-$1/2$ model for magnetism in rare earth pyrochlores. gMFT is based on a novel exact slave-particle formulation, and matches both the perturbative regime near the classical spin ice limit and the semiclassical approximation far from it. We show that the full phase diagram contains two exotic phases: a quantum spin liquid and a coulombic ferromagnet, both of which support deconfined spinon excitations and emergent quantum electrodynamics. Phenomenological properties of these phases are discussed.'
author:
- Lucile Savary
- Leon Balents
bibliography:
- 'pyro\_theory.bib'
title: 'Coulombic Quantum Liquids in Spin-1/2 Pyrochlores'
---
Amongst the celebrated exotic phases of matter, of particular recent interest are the Quantum Spin Liquids (QSLs) [@balents2010spin]. Behind [*seemingly*]{} innocuous defining properties –strong spin correlations, the absence of static magnetic moments, and unbroken crystalline symmetry–, QSLs display the consequences of [*extreme*]{} quantum entanglement. These include emergent gauge fields and fractional excitations, which take these states beyond the usual “mean field” paradigm of phases of matter. Not only are these phases challenging to predict and describe, they have also proven very hard to find in the laboratory, rendering their search and discovery even more tantalizing.
A consensual place to look for QSLs is among frustrated magnets [@balents2010spin]. Frustration allows the spins to avoid phases where they are either ordered or frozen, with relatively small fluctuations and correlations between them. Recent experiments have given compelling evidence of a QSL state in certain two-dimensional organic materials [@kanoda11:_mott_physic_organ_conduc_trian_lattic], but both microscopic and fully consistent phenomenological theories are lacking. By contrast, [*classical*]{} spin liquids have been conclusively seen and microscopically understood in the spin ice pyrochlores [@gingras11:_spin_ice]. This raises the possibility, suggested experimentally [@jointpaper] and theoretically [@tb2ti2o7-gingras-canals], of QSLs in those rare earth pyrochlores in which spins are non-classical, supported by recent results on [[Yb$_2$Ti$_2$O$_7$]{}]{} [@jointpaper]. However, for any material, only detailed, quantitative theory predicting the [ *type(s)*]{} and properties of QSLs that appear [*and*]{} matching experiments can take the physics to the next level.
We take up this challenge here for quantum rare earth pyrochlores. Our analysis confirms that a “$U(1)$” QSL phase exists in the phase diagram (Fig. \[fig:MF-phasediag\]) of a spectrum of real materials, and is furthermore supplemented by another exotic phase, a Coulombic ferromagnet, which contains spinons, but displays non-zero magnetization. We also study the confinement transitions out of these Coulomb phases, which are analogous to “Higgs” transitions [@fradkin-shenker]. Finally, we discuss experimental signatures of the $U(1)$ QSL, and of the $U(1)$ Coulomb ferromagnet.
![Gauge mean field phase diagram obtained for $J_{\pm\pm}=0$ and $J_{zz}>0$. “QSL”, “CFM”, “FM” and “AFM” denote the $U(1)$ Quantum Spin Liquid, Coulomb Ferromagnet, standard ferromagnet, and standard antiferromagnet, respectively. Phase boundaries with/without white lines indicate continuous/discontinuous transitions in gMFT. Note that the diagram is symmetric in $J_{z\pm}\rightarrow-J_{z\pm}$. []{data-label="fig:MF-phasediag"}](fig1_gMFT-phasediag.pdf){width="3.3in"}
The most general nearest-neighbor symmetry-allowed exchange Hamiltonian for spin-$1/2$ spins (real or effective) on the pyrochlore lattice is $$\begin{aligned}
\label{eq:spinham}
H & = & \sum_{\langle ij\rangle} \Big[ J_{zz} \mathsf{S}_i^z \mathsf{S}_j^z - J_{\pm}
(\mathsf{S}_i^+ \mathsf{S}_j^- + \mathsf{S}_i^- \mathsf{S}_j^+) \nonumber \\
&& +\, J_{\pm\pm} \left[\gamma_{ij} \mathsf{S}_i^+ \mathsf{S}_j^+ + \gamma_{ij}^*
\mathsf{S}_i^-\mathsf{S}_j^-\right] \nonumber \\
&& +\, J_{z\pm}\left[ \mathsf{S}_i^z (\zeta_{ij} \mathsf{S}_j^+ + \zeta^*_{ij} \mathsf{S}_j^-) +
{i\leftrightarrow j}\right]\Big],\end{aligned}$$ where $\gamma$ is a $4\times4$ complex unimodular matrix, and $\zeta=-\gamma^*$. The explicit expression of $\gamma$ and of the local bases whose components are used in Eq. are given in the Supplementary Material. The first term (we assume in this paper $J_{zz}>0$), taken alone, gives the highly frustrated classical nearest-neighbor spin ice model, which exhibits an extensive ground state degeneracy of “two-in-two-out” states.
In fact, this model has been studied theoretically in the special case $J_{z\pm}=J_{\pm\pm}=0$, where it reduces to an “XXZ” model with global XY spin-rotation symmetry [@hermele]. There, it was shown that for $J_\pm
\ll J_{zz}$, it is perturbatively equivalent, order by order, to a lattice $U(1)$ gauge theory, with gauge fields that describe the spin configurations [*constrained*]{} to the spin ice manifold of ground states. This gauge theory was furthermore argued to exhibit a so-called “Coulomb phase”, which corresponds to a $U(1)$ QSL phase. Subsequent numerical simulations [@banerjee; @shannon-sikora-2011] verified this prediction. This Coulombic QSL is not only magnetically disordered, but also supports several exotic excitations: spinons (called magnetic monopoles in the spin ice literature), dual “electric monopoles”, and an emergent photon. This understanding, however, was limited to the perturbative regime $J_\pm \ll J_{zz}$ and considered only the XXZ case. Here we develop a [*non-perturbative*]{} method to analyze the full Hamiltonian in Eq. .
Non-perturbative theories of QSLs based on “slave particles” have been developed and used extensively in $SU(2)$ invariant $S=1/2$ Heisenberg and Hubbard models [@wen2004quantum]. Generally these approaches work by embedding the Hilbert space on each site in some larger “spinon” one, with a microscopic gauge symmetry which acts to project back to the physical space. QSL phases are found when, in a mean field sense, this microscopic gauge symmetry is incompletely broken in the ground state. Here, we follow the spirit but not the letter of these approaches, by introducing redundant degrees of freedom not for each spin but for each tetrahedron of the pyrochlore lattice. This new slave particle representation is, like the aforementioned standard ones, formally exact, but additionally naturally describes the Coulombic QSL found before in the perturbative analysis, when that limit is taken. It also has the added advantage that, unlike in standard approaches, the gauge fields appear explicitly in the slave particle Hamiltonian, rendering the analogy to lattice gauge theory more direct and transparent.
By dint of the theory developed in Refs. , we define our slave particles on the centers of the “up” and “down” tetrahedra of the pyrochlore lattice, which comprise two FCC sublattices (I/II, with $\eta_\mathbf{r}=\pm1$) of sites, denoted with boldface characters $\mathbf{r}$, of a dual diamond lattice. The sites of the original pyrochlore lattice are bonds of the dual lattice. The perturbative analysis of Ref. identified the low energy states of $H$ as the spin ice ones, supplemented by spinons corresponding to defect tetrahedra. As mentioned above, this inspires us to enlarge the Hilbert space and define “spinon” slave operators, which in turn can be seen as particles in a fluctuating vacuum (the two-in-two-out manifold dear to the spin ice community). We consider $\mathcal{H}_{big}=\mathcal{H}_{spin}\otimes\mathcal{H}_{Q}$, where $\mathcal{H}_{spin}=\bigotimes_N\mathcal{H}_{1/2}$ is the Hilbert space of Eq. and $\mathcal{H}_{Q}$ is the Hilbert space of a field $Q_\mathbf{r}\in\mathbb{Z}$. $Q_\mathbf{r}$ is defined on all the sites of the dual diamond lattice and, at this stage, is free and unphysical. We further define the real and compact operator $\varphi_\mathbf{r}$ to be the canonically conjugate variable to $Q_\mathbf{r}$, $[\varphi_\mathbf{r},Q_\mathbf{r}]=i$. In $\mathcal{H}_{Q}$, the bosonic operators $\Phi_\mathbf{r}^\dagger=e^{i\varphi_\mathbf{r}}$ and $\Phi_\mathbf{r}=e^{-i\varphi_\mathbf{r}}$ thus act as raising and lowering operators, respectively, for $Q_\mathbf{r}$. Note that, by construction, $|\Phi_\mathbf{r}|=1$. We now take the restriction of $\mathcal{H}_{big}$ to the subspace $\mathcal{H}$, in which $$\label{eq:gauss}
Q_\mathbf{r} = \eta_\mathbf{r}
\sum_\mu {\sf
s}^z_{\mathbf{r},\mathbf{r}+\eta_\mathbf{r}\mathbf{e}_\mu},$$ where the $\mathbf{e}_\mu$’s are the four nearest-neighbor vectors of an $\eta_\mathbf{r}=1$ (I) diamond sublattice site. This constraint can be viewed as analogous to Gauss’ law, where now $Q_\mathbf{r}$ counts the number of spinons. The restriction of $Q_\mathbf{r}$, $\Phi_\mathbf{r}$ and $\Phi_\mathbf{r}^\dagger$ to $\mathcal{H}$ exactly reproduces all matrix elements of the original $\mathcal{H}_{spin}$, with the replacements $$\label{eq:1}
\mathsf{S}_{\mathbf{r},\mathbf{r}+\mathbf{e}_\mu}^+ =
\Phi_{\mathbf{r}}^\dagger\,
\mathsf{s}_{\mathbf{r},\mathbf{r}+\mathbf{e}_\mu}^+
\Phi_{\mathbf{r}+\mathbf{e}_\mu}, \qquad
\mathsf{S}_{\mathbf{r},\mathbf{r}+\mathbf{e}_\mu}^z =
\mathsf{s}_{\mathbf{r},\mathbf{r}+\mathbf{e}_\mu}^z.$$ Here $\mathbf{r}\in\mbox{I}$, and $\mathsf{s}^\pm_{\mathbf{rr}'}, \mathsf{s}^z_{\mathbf{rr}'}$ act within the $\mathcal{H}_{spin}$ subspace of $\mathcal{H}_{big}$. Note especially that, by itself, $\mathsf{s}^\pm_{\mathbf{rr}'} \neq
\mathsf{S}^\pm_{\mathbf{rr}'}$ is not the physical spin, and does not remain within $\mathcal{H}$.
In this paper we focus on the case where $J_{\pm\pm}=0$ (which otherwise introduces additional complications to be dealt with in a separate publication), and the Hamiltonian then becomes
$$\begin{aligned}
\label{eq:ham-spinons}
H & = & \sum_{\mathbf{r} \in {\rm I,II}} \frac{J_{zz}}{2} Q_\mathbf{r}^2 - J_\pm \left\{ \sum_{\mathbf{r}\in
{\rm I}} \sum_{\mu,\nu\neq\mu} \Phi_{\mathbf{r}+\mathbf{e}_\mu}^\dagger
\Phi_{\mathbf{r}+\mathbf{e}_\nu}^{\vphantom\dagger} {\sf
s}^-_{\mathbf{r},\mathbf{r}+\mathbf{e}_\mu} {\sf
s}^+_{\mathbf{r},\mathbf{r}+\mathbf{e}_\nu} + \sum_{\mathbf{r}\in
{\rm II}} \sum_{\mu,\nu\neq\mu} \Phi_{\mathbf{r}-\mathbf{e}_\mu}^\dagger
\Phi_{\mathbf{r}-\mathbf{e}_\nu}^{\vphantom\dagger} {\sf s}^+_{\mathbf{r},\mathbf{r}-\mathbf{e}_\mu} {\sf s}^-_{\mathbf{r},\mathbf{r}-\mathbf{e}_\nu}
\right\} \\
&& - J_{z\pm} \left\{ \sum_{r\in {\rm I}} \sum_{\mu,\nu\neq\mu} \left(
\gamma^*_{\mu\nu} \Phi_\mathbf{r}^\dagger \Phi_{\mathbf{r}+\mathbf{e}_\nu}^{\vphantom\dagger}
{\sf s}^z_{\mathbf{r},\mathbf{r}+\mathbf{e}_\mu} {\sf s}^+_{\mathbf{r},\mathbf{r}+\mathbf{e}_\nu} + {\rm h.c.}\right)
+ \sum_{\mathbf{r}\in {\rm II}} \sum_{\mu,\nu\neq\mu} \left(
\gamma^*_{\mu\nu} \Phi_{\mathbf{r}-\mathbf{e}_\nu}^\dagger \Phi_{r}^{\vphantom\dagger}
{\sf s}^z_{\mathbf{r},\mathbf{r}-\mathbf{e}_\mu} {\sf s}^+_{\mathbf{r},\mathbf{r}-\mathbf{e}_\nu} + {\rm h.c.}\right) \right\}+\mbox{const.}.
\nonumber \end{aligned}$$
The integer-valued constraint in Eq. commutes with $H$ and thereby ensures that Eq. is a $U(1)$ gauge theory. Explicitly, it is invariant under the transformations $$\label{eq:gauge-sym}
\begin{cases}
\Phi_\mathbf{r}\rightarrow\Phi_\mathbf{r}\,e^{-i\chi_\mathbf{r}}\\
{\sf s}^\pm_{\mathbf{rr}'}\rightarrow {\sf s}_{\mathbf{rr}'}^\pm e^{\pm i(\chi_{\mathbf{r}'}-\chi_{\mathbf{r}\vphantom{'}})}
\end{cases},$$ with arbitrary $\chi_\mathbf{r}$. This invariance, and the Gauss’ law in Eq. can be made formally identical to that in lattice electrodynamics by writing $\mathsf{s}^z_{\mathbf{rr}'}=E_{\mathbf{rr}'}$ and $\mathsf{s}^\pm_{\mathbf{rr}'}= e^{\pm iA_{\mathbf{rr}'}}$, where $E$ and $A$ are lattice electric and magnetic fields [@hermele]. This clarifies that $\mathsf{s}^\pm_{\mathbf{rr}'}$ is to be regarded as an element of the $U(1)$ gauge group. However, the notation is unnecessary and we use it only when conceptually valuable.
Eq. can be viewed as spinons hopping in the background of fluctuating gauge fields, and thereby lends itself to the application of standard mean field theory methods for lattice gauge models [@PhysRevD.10.2445], which we call gauge Mean Field Theory (gMFT). Upon performing gMFT, we will get a Hamiltonian for spinons hopping in a fixed background. Specifically, we perform the replacement: $$\begin{aligned}
\label{eq:MFTreplace}
&&\Phi^\dagger\Phi\,\mathsf{s}\,\mathsf{s}\rightarrow\\
&&\qquad\Phi^\dagger\Phi\langle\mathsf{s}\rangle\langle\mathsf{s}\rangle+\langle\Phi^\dagger\Phi\rangle\mathsf{s}\langle\mathsf{s}\rangle+\langle\Phi^\dagger\Phi\rangle\langle\mathsf{s}\rangle\mathsf{s}-2\langle\Phi^\dagger\Phi\rangle\langle\mathsf{s}\rangle\langle\mathsf{s}\rangle,\nonumber\end{aligned}$$ and thereby split the Hamiltonian in a spinon part $H_\Phi^{{\rm MF}}$, and a gauge part $H_g^{{\rm MF}}$. Note that unlike conventional Curie-Weiss mean field theory, which entirely neglects any quantum entanglement, gMFT, while suppressing some fluctuations, still allows high correlations and entanglement.
The gMFT order parameters are closely analogous to those in $U(1)$ Higgs theory [@PhysRevD.10.2445; @PhysRevD.28.360]. A non-zero expectation value $\langle \mathsf{s}^\pm\rangle \neq 0$ implies the phase of $\mathsf{s}^\pm$ is relatively well-defined, i.e. there are small fluctuations of the vector potential $A$. The converse case, $\langle
\mathsf{s}^\pm\rangle =0$ would indicate confinement, but does not occur here. A non-zero scalar expectation value, $\langle \Phi\rangle \neq
0$, analogous to a Higgs phase, indicates spinon condensation and generation of a mass for the gauge field, and a conventional, non-exotic state. Combined with $\langle \mathsf{s}^\pm\rangle \neq 0$, it also implies “XY” magnetic order. Conversely, $\langle \Phi\rangle =0$ indicates the spinons have a gap, and is characteristic of the Coulomb phase. The remaining gMFT order parameter, $\mathsf{s}^z$, is gauge invariant, and thus indicates only the presence ($\langle
\mathsf{s}^z\rangle\neq 0$) or absence ($\langle \mathsf{s}^z\rangle=
0$) of “Ising” magnetic order, i.e. time-reversal symmetry breaking. Combining this together, the phases in gMFT are summarized in Table \[tab:phase-class\].
$\langle\Phi\rangle\quad$ $\langle \mathsf{s}^z \rangle\quad$ $\langle \mathsf{s}^\pm\rangle\quad$ phase$\quad$
--------------------------- ------------------------------------- -------------------------------------- --------------
$0$ $0$ $\neq0$ QSL
$0$ $\neq0$ $\neq0$ CFM
$\neq0$ $\neq0$ $\neq0$ FM
$\neq0$ $0$ $\neq0$ AFM
: Order parameters and phases in gMFT.
\[tab:phase-class\]
We emphasize that despite the fact that $\langle \Phi\rangle$ does not appear explicitly in the decoupling in Eq. , the gMFT does generally allow for Higgs phases where $\Phi$ is indeed condensed. As we will show below, the Higgs phase appears in a manner similar to Bose-Einstein condensation in an ideal Bose gas.
We now use the following Ansatz, valid when $J_\pm>0$ (which we assume hereafter), $$\langle\mathsf{s}_\mu^z\rangle=\frac{1}{2}\sin\theta\,\varepsilon_\mu,\qquad\langle\mathsf{s}^-_\mu\rangle=\frac{1}{2}\cos\theta,
\label{eq:ansatz}$$ where $\mu=0,..,3$ and $\varepsilon=(1,1,-1,-1)$, which assumes translational invariance and fully polarized “spins” $\vec{\mathsf{s}}$, in accord with Eq. , and is compatible with FM polarization along the (global) $x$ axis ($\langle\mathsf{s}^+_\mu\rangle=\langle\mathsf{s}^-_\mu\rangle$). Note that Eq. shows that the gMFT allows fluctuations of $E$ and $A$, so long as $\theta\neq\pi/2$ and $\theta\neq0$, respectively. Defining the dot product, and through it the vector notation, $\vec{\mathsf{u}}\cdot\vec{\mathsf{v}}=\mathsf{u}^z\mathsf{v}^z+\frac{1}{2}\left(\mathsf{u}^+\mathsf{v}^-+\mathsf{u}^-\mathsf{v}^+\right)$, we find $$H^{\rm MF}_{\sf s} =- \sum_{\mathbf{r}\in {\rm I}} \sum_\mu \vec{{\sf h}}_{{\rm eff},\mu}(\mathbf{r})\cdot \vec{{\sf s}}_{\mathbf{r},\mathbf{r}+\mathbf{e}_\mu},$$ where ${\sf h}_{{\rm eff},\mu}^z = 4\, \varepsilon_\mu J_{z\pm}
I_1\cos\theta$ and ${\sf h}_{{\rm eff},\mu}^- = 4 J_{z\pm}
I_1\sin\theta +2J_\pm I_2\cos\theta$, and we have defined $I_1=\varepsilon_\mu\langle\Phi_\mathbf{r}^\dagger\Phi^{\vphantom\dagger}_{\mathbf{r}+\mathbf{e}_\mu}\rangle$ (no summation implied) and $I_2=\sum_{\nu\neq\mu}\langle\Phi_\mathbf{r}^\dagger\Phi^{\vphantom\dagger}_{\mathbf{r}+\mathbf{e}_\mu-\mathbf{e}_\nu}\rangle$ ($\mu$ is fixed). These quantities turn out to be independent of the diamond bond $\mu$. To treat the spinons, we relax the $|\Phi_\mathbf{r}|=1$ constraint to a global one by introducing a Lagrange multiplier $\lambda$ via the term $\lambda\sum_\mathbf{r}\left(|\Phi_\mathbf{r}|^2-1\right)$ in a path integral formulation, with free integration over $\Phi$ and $\Phi^*$. The spinon Lagrangian is $$\mathcal{L}^{{\rm MF}}_\Phi=\frac{1}{N_{{\rm u.c.}}}\sum_{\mathbf{k}}\int_{\omega_n}\Phi_{\mathbf{k},\omega_n}^*\cdot G^{-1}_{\mathbf{k},\omega_n}\cdot\Phi^{\vphantom*}_{\mathbf{k},\omega_n},$$ where $N_{{\rm u.c.}}$ is the number of unit cells, $\left[G_{\mathbf{k},\omega_n}\right]_{ab}=\langle\Phi_b^*\Phi_a\rangle$, and we find the equal time Green’s function to be $$\label{eq:green-fn}
G_{\mathbf{k},\tau=0}=\frac{1}{2}\sqrt{\frac{J_{zz}}{2}}
\begin{pmatrix}
Z_\mathbf{k}^+ & -\frac{M_\mathbf{k}}{|M_\mathbf{k}|}Z^-_\mathbf{k}\\
-\frac{M_\mathbf{k}^*}{|M_\mathbf{k}|}Z^-_\mathbf{k} & Z_\mathbf{k}^+
\end{pmatrix},$$ where $M_\mathbf{k}=\sum_\mu\varepsilon_\mu e^{i\mathbf{k}\cdot
\mathbf{e}_\mu}$, $Z_\mathbf{k}^\pm\left(\theta,\lambda\right)=\frac{1}{z_\mathbf{k}^+}\pm\frac{1}{z_\mathbf{k}^-}$, $z_\mathbf{k}^\pm\left(\theta,\lambda\right)=\sqrt{\lambda-\ell_\mathbf{k}^\pm(\theta)}$, $\ell^\pm_\mathbf{k}(\theta)=\frac{1}{2}J_\pm\cos^2\theta
L_\mathbf{k}\mp\left|\frac{1}{2}J_{z\pm}\sin2\theta
M_\mathbf{k}\right|$, $L_\mathbf{k}=\sum_{\mu,\nu<\mu}\cos\left[\mathbf{k}\cdot(\mathbf{e}_\mu-\mathbf{e}_\nu)\right]$. A couple of remarks are in order: (i) $\lambda>\max_\mathbf{k}\ell_\mathbf{k}^-$ (ii) the spinon dispersion relations are $\omega^\pm_\mathbf{k}(\theta,\lambda)=\sqrt{2J_{zz}}\,z^\pm_\mathbf{k}(\theta,\lambda)$.
The gMFT consistency conditions on $\theta$ and $\lambda$ (for fixed $J_\pm,J_{z\pm}$) arise from requiring Eqs. and $\langle \mathsf{s}^\kappa_\mu \rangle
= \mathsf{h}^\kappa_\mu/(2|\vec{\mathsf{h}}_\mu|)$, and from the normalization condition on $|\Phi|^2$, and can be written $$\begin{cases}\tan\theta=\dfrac{2J_{z\pm}I_1(\theta,\lambda)}{2J_{z\pm}I_1(\theta,\lambda)\tan\theta+J_\pm I_2(\theta,\lambda)}\\
I_3\left(\theta,\lambda\right)=1\end{cases},
\label{eq:consistency}$$ where $I_3\left(\theta,\lambda\right)=\langle\Phi_\mathbf{r}^\dagger\Phi_\mathbf{r}^{\vphantom{\dagger}}\rangle$. The explicit expressions for the $I_i=N_{{\rm
u.c.}}^{-1}\sum_\mathbf{k}\mathcal{I}^i_\mathbf{k}$, needed to solve Eqs. , are readily derived from Eq. , and are given in the Supplementary Material, Eq. . Since Eqs. may allow several distinct solutions, we must choose the solution of Eq. with the lowest energy. In the mean field approximation, the ground state energy can be calculated by taking the expectation value of the Hamiltonian. We find, per unit cell, $\epsilon_{\rm GS}=\epsilon_{av}+\epsilon_{kin}$, with $$\begin{aligned}
\label{eq:energy-av}
\epsilon_{av}&=&-2I_2(\theta,\lambda)\cos^2\theta
J_\pm-4I_1(\theta,\lambda)\sin2\theta\, J_{z\pm}\\
\epsilon_{kin}&=&\frac{1}{2}\int_\mathbf{k}\left[\omega^+_\mathbf{k}(\theta,\lambda)+\omega^-_\mathbf{k}(\theta,\lambda)\right],
\label{eq:energy-kin}\end{aligned}$$ where $\omega^\pm=\sqrt{2J_{zz}}\,z^\pm$. Here $\epsilon_{kin}$ measures the “kinetic” energy associated with the spinon modes, while $\epsilon_{av}$ represents the “background” energy in which the latter evolve.
We now discuss how the different phases are obtained from the solutions of the gMFT equations. Condensed and uncondensed phases are distinguished by the value of $\lambda$. As in the theory of superfluidity, condensation is synonymous with off-diagonal long-range order, i.e. $\langle \Phi_\mathbf{r}\rangle^* \langle
\Phi_{\mathbf{r}'}\rangle \equiv
\lim_{|\mathbf{r}-\mathbf{r}'|\rightarrow\infty}
\langle\Phi_\mathbf{r}^\dagger\Phi_{\mathbf{r}'}^{\vphantom{\dagger}}\rangle
\neq 0$. This expectation value $\langle\Phi_{\mathbf{r}'}^{\vphantom{\dagger}}\Phi_\mathbf{r}^\dagger\rangle=N_{{\rm
u.c.}}^{-1}\sum_\mathbf{k}G_\mathbf{k}e^{i\mathbf{k}\cdot(\mathbf{r}'-\mathbf{r})}$ is non-zero in the long-distance limit if and only if the usual conversion of the sum to an integral [*fails*]{}, i.e. if there exists one $\mathbf{k}_0$ such that $G_{\mathbf{k}_0}=O(N_{{\rm u.c.}})$. Like the chemical potential in Bose-Einstein condensation, $\lambda$ in a Higgs phase differs from its minimum allowed value by a sub-extensive part. $\lambda=\lambda_{{\rm
min}}(\theta)+\frac{\delta^2}{N_{{\rm u.c.}}^2}$ therefore defines condensation, where $\delta$ is of order $O(1)$, and $\lambda_{{\rm
min}}(\theta)=\max_\mathbf{k}\ell_\mathbf{k}^-(\theta)$. If instead $\lambda-\lambda_{{\rm min}}(\theta)$ is $O(1)$, one has a phase with deconfined gapped spinons and a Coulombic gauge structure. As we already mentioned above, these classes of phases can be further subdivided into “polarized” (i.e. with magnetization along the [*local*]{} $z$ axis) and “unpolarized” when $\theta\neq0$ and $\theta=0$, respectively.
The phase diagram resulting from the gMFT solution (see Supplementary material) is shown in Fig. \[fig:MF-phasediag\]. It contains two “exotic” phases in which spinons are deconfined and uncondensed, indicated as QSL and CFM. The QSL state, with $\theta=0$, is completely absent magnetic order, and is the phase studied in Refs. . Its low energy physics mimics quantum electrodynamics, and thereby contains a photonic excitation (gapless and linear near the origin) and gapped fractional monopole excitations (spinon and “electric” monopole) that interact via Coulomb interactions. In the present formalism, the photon is only obtained once quadratic fluctuations around the gMFT solution are considered, but is a universal feature of the exotic phases. The CFM, or “Coulombic Ferromagnet” phase, is a new phase of matter that can be seen as a polarized version of the $U(1)$ QSL. Despite being magnetic, its [ *elementary*]{} magnetic [*excitations*]{} are spinons rather than spin waves, and it also supports a gapless photon mode. Indeed, in gMFT the transition from the QSL to CFM is second order, and consists of a continuous rise of magnetization from zero. For larger $J_{z\pm},J_\pm$, one obtains Higgs phases, which are conventional states of matter without exotic excitations and are continuously connected to the usual magnetically ordered states described by Curie-Weiss MFT. Interestingly, we find the exotic CFM state is considerably more stable than the “pure” QSL, occupying a much more substantial portion of the phase diagram.
How do we recognize a Coulomb phase in experiment? A generic sign of fractionalization is a two-particle continuum in inelastic neutron scattering, two spinons being excited by one neutron [@ross2009; @jointpaper]. In addition the photon can be detected directly by inelastic neutron scattering, as a linearly dispersing transverse mode. It is, however, more challenging to observe than the usual acoustic spin wave, because its scattering intensity becomes small ($\propto\omega$) at low energy (see Supp. Mat.), in contrast to the spin wave for which the intensity diverges ($\sim 1/\omega$) in the same limit. Interestingly, the pinch points in the static structure factor present for classical spin ice are absent for the quantum Coulomb phase [@shannon-sikora-2011], so this is not a useful measurement. Perhaps the most striking signature of the Coulomb phase is likely to be thermodynamic. Like the phonons, the photons contribute as $B\, T^3$ to the specific heat at low temperatures, but their speed is $v_{photon}\sim J\ll c$, the speed of sound. Crudely estimating $J\sim 2$ K appropriate for [[Yb$_2$Ti$_2$O$_7$]{}]{}, we obtain a coefficient $B_{photon} \approx 10^3$mJ/mole-K$^4$, approximately 1000 times larger (!) than the phonon contribution $B_{phonon} \approx 0.5$ mJ/mole-K$^4$ measured for the isostructural material Y$_2$Ti$_2$O$_7$ [@PhysRevB.79.224111].
With a phase diagram and a new phase of matter in hand, we take heart at discovering yet more new exciting facts in the pyrochore lattice. Future studies should address the more frustrated case $J_\pm<0$, phase transitions in applied field, and the influence of defects.
We thank Peter Holdsworth, Bruce Gaulin and Kate Ross for discussions. This work was supported by the DOE through BES grant DE-FG02-08ER46524.
[**SUPPLEMENTARY MATERIAL**]{}
Geometry
========
Local bases
-----------
The local cubic bases in which the Hamiltonian Eq. is expressed are the following $(\mathbf{\hat{a}}_i,\mathbf{\hat{b}}_i,\mathbf{\hat{e}}_i)$ bases $$\left\{\begin{array}{l}
\mathbf{\hat{e}}_0=(1,1,1)/\sqrt{3}\\
\mathbf{\hat{e}}_1=(1,-1,-1)/\sqrt{3}\\
\mathbf{\hat{e}}_2=(-1,1,-1)/\sqrt{3}\\
\mathbf{\hat{e}}_3=(-1,-1,1)/\sqrt{3},
\end{array}\right.,
\quad
\left\{\begin{array}{l}
\mathbf{\hat{a}}_0=(-2,1,1)/\sqrt{6}\\
\mathbf{\hat{a}}_1=(-2,-1,-1)/\sqrt{6}\\
\mathbf{\hat{a}}_2=(2,1,-1)/\sqrt{6}\\
\mathbf{\hat{a}}_3=(2,-1,1)/\sqrt{6}
\end{array}\right.,$$ $\mathbf{\hat{b}}_i=\mathbf{\hat{e}}_i\times\mathbf{\hat{a}}_i$, such that spin $\mathbf{S}_i$ on sublattice $i$ is $\mathbf{S}_i=\mathsf{S}^+_i(\mathbf{\hat{a}}_i-i\mathbf{\hat{b}}_i)/2+\mathsf{S}^-_i(\mathbf{\hat{a}}_i+i\mathbf{\hat{b}}_i)/2+\mathsf{S}^z_i\mathbf{\hat{e}}_i$.
The $4\times4$ matrix $\gamma$ introduced in Eq. is $$\gamma=\begin{pmatrix}
0 & 1 & w & w^2\\
1 & 0 & w^2 & w\\
w & w^2 & 0 & 1\\
w^2 & w & 1 & 0
\end{pmatrix},$$ where $w=e^{2\pi i/3}$ is a third root of unity.
Lattice vectors
---------------
The four nearest-neighbor vectors of a I-sublattice diamond site (sublattice I corresponds to “up” tetrahedra) are $\mathbf{e}_\mu=\frac{a\sqrt{3}}{4}\mathbf{\hat{e}}_\mu$, where $a$ is the usual FCC lattice spacing. The four pyrochlore sites of the “up” tetrahedron centered at the origin are located at $\mathbf{e}_\mu/2$, $\mu=0,..,3$.
The FCC primitive lattice vectors are $\mathbf{A}_i=\mathbf{e}_0-\mathbf{e}_i$, $i=1,..,3$, while the reciprocal lattice basis vectors are defined as usual by $\mathbf{B}_1=2\pi\frac{\mathbf{A}_2\times\mathbf{A}_3}{v_{{\rm u.c.}}}$ and its cyclic permutations, where $v_{{\rm u.c.}}=\mathbf{A}_1\cdot(\mathbf{A}_2\times\mathbf{A}_3)$ is the volume of the (real space) unit cell. If the $q_i$’s are defined as $$\label{eq:qi}
\mathbf{k}=\sum_{i=1}^3 q_i\,\mathbf{B}_i,$$ the first Brillouin zone can be considered the “cube” with unit sides described by $-1/2<q_i<1/2$ (note that the $q_i$’s are dimensionless).
Brillouin Zone Sums
===================
In the main text, we defined $$\begin{aligned}
I_1&=&\varepsilon_\mu\langle\Phi_\mathbf{r}^\dagger\Phi_{\mathbf{r}+\mathbf{e}_\mu}\rangle\quad\mbox{(no summation implied)}, \\
I_2&=&\sum_{\nu\neq\mu}\langle\Phi_\mathbf{r}^\dagger\Phi_{\mathbf{r}+\mathbf{e}_\mu-\mathbf{e}_\nu}\rangle\quad\mbox{($\mu$ is fixed)},\\
I_3&=&\langle\Phi_\mathbf{r}^\dagger\Phi_\mathbf{r}^{\vphantom{\dagger}}\rangle.\end{aligned}$$ For brevity, we define the $\mathcal{I}^i_\mathbf{k}$’s through $$\label{eq:formal-sums}
I_i=\frac{1}{N_{{\rm u.c.}}}\sum_{\mathbf{k}\in{\rm BZ}}\mathcal{I}^i_\mathbf{k}=\frac{1}{N_{{\rm u.c.}}}\sum_{\mathbf{q}\in{\rm cube}}\mathcal{I}^i_\mathbf{q}$$ ($\mathbf{q}$ is defined in Eq. ), whose explicit expressions are $$\begin{aligned}
&&\mathcal{I}^1_\mathbf{q}\left(\theta,\lambda\right)=\nonumber\\
&&\quad\frac{1}{2}\sqrt{\frac{J_{zz}}{2}}\frac{Z^-_\mathbf{q}}{|M_\mathbf{q}|}\left(-1-\cos2\pi q_1+\cos2\pi q_2+\cos2\pi q_3\right)\nonumber,\\
&&\mathcal{I}^2_\mathbf{q}\left(\theta,\lambda\right)=\frac{1}{2}\sqrt{\frac{J_{zz}}{2}}Z^+_\mathbf{q}\left(\cos2\pi q_1+\cos2\pi q_2+\cos2\pi q_3\right),\nonumber\\
&&\mathcal{I}^3_\mathbf{q}\left(\theta,\lambda\right)=\frac{1}{2}\sqrt{\frac{J_{zz}}{2}}Z^+_\mathbf{q},
\label{eq:integrals}\end{aligned}$$ where we used the Fourier transform convention $\Phi_\mathbf{r}=\frac{1}{N_{{\rm u.c.}}}\sum_\mathbf{k}\Phi_\mathbf{k}\,e^{i\mathbf{k}\cdot\mathbf{r}}$, and in turn $$\langle\Phi^\dagger_\mathbf{r}\Phi_{\mathbf{r}'}\rangle=\frac{1}{N_{u.c.}}\sum_\mathbf{k}[G_\mathbf{k}]_{21}e^{i\mathbf{k}\cdot(\mathbf{r}'-\mathbf{r})}\quad\mbox{for}\; \mathbf{r}\in\mbox{I},\mathbf{r}'\in\mbox{II},$$ with $G_{\mathbf{k}}$ defined in Eq. .
These Brillouin zone sums need to be evaluated. While in the Coulomb phases, they can be directly identified with the corresponding integrals, the situation is more complex within the condensed phases. We address this issue below.
Brillouin zone sums in the condensed phases
===========================================
The condensed region occurs if $I_3(\lambda_{{\rm min}})<1$. In this case, we must allow for the subextensive part of $\lambda$. Let $$\label{eq:59}
\lambda= \lambda_{\rm min} + \frac{\delta^2}{N_{u.c.}^2},$$ where $N_{u.c.}$ is the number of unit cells and $\delta>0$. Then the sums Eq. cannot merely be replaced by the corresponding Brillouin zone integrals, but must be written as follows: $$\label{eq:57}
\frac{1}{N_{u.c.}} \sum_\mathbf{k} f(\mathbf{k}) = \frac{f(\mathbf{k}_{\rm min})}{N_{u.c.}} + \frac{1}{N_{u.c.}} \sum_{\mathbf{k}\neq\mathbf{k}_{\rm min}} f(\mathbf{k})$$ and $$\begin{aligned}
&&\frac{1}{N_{u.c.}} \sum_{\mathbf{k}\neq\mathbf{k}_{\rm min}} f(\mathbf{k}) = \frac{V_{u.c.}}{V} \sum_{\mathbf{k}\neq\mathbf{k}_{\rm min}} f(\mathbf{k})\\
&&\quad = \frac{1}{V_{BZ}}\sum_{\mathbf{k}\neq\mathbf{k}_{\rm min}} \left(\frac{2\pi}{L}\right)^d f(\mathbf{k}) \rightarrow \int \! \frac{d^d k}{V_{BZ}} f(\mathbf{k})=\int_\mathbf{k} f(\mathbf{k}),\nonumber\end{aligned}$$ where $V_{BZ}$ is the volume of the Brillouin zone. In general, we denote this decomposion of an integral as $I= I_{\rm min}
+ I'$, where $I'$ is the continuous integral part evaluated at $\lambda=\lambda_{\rm min}$, and $I_{\rm min}$ is the first term in the right hand side of Eq. . In $I_{\rm min}$, we need only retain the part which is non-vanishing as $N_{u.c.}\rightarrow \infty$.
The value $\lambda_{{\rm min}}$ is determined by the condition that $\min_\mathbf{q} z_\mathbf{q}^- =
0$, which corresponds to $\lambda_{{\rm min}} = \max_\mathbf{q}\ell_\mathbf{q}^-(\theta)=\max_\mathbf{q} (\frac{1}{2}\cos^2\theta J_\pm L_\mathbf{q}+ |\frac{1}{2}J_{z\pm}\sin2\theta M_\mathbf{q}|)$. Here, this maximum $\max_\mathbf{q}\ell_\mathbf{q}^-$ always occurs for the (dimensionless) wavevector of the form $\mathbf{q}_{\rm min}=(q_1,q_2,q_3)=(0,q,q)$ ($\mathbf{q}_{\rm min}$ corresponds to $\mathbf{k}_0$ in the main text). Defining $$y=\cos 2\pi q,$$ $\max_\mathbf{q}\ell_\mathbf{q}^-$ is realized for the wavevector $(0,q,q)$ corresponding to $$\label{eq:56}
y_{{\rm min}} =
\begin{cases}
1 - \frac{1}{8}\left(\frac{J_2[J_{z\pm},\theta]}{J_1[J_\pm,\theta]}\right)^2 & \mbox{for } J_2[J_{z\pm},\theta]< 4 J_1[J_\pm,\theta] \\
-1 & \mbox{for }
J_2[J_{z\pm},\theta]>4 J_1[J_\pm,\theta]
\end{cases},$$ where we defined $$\begin{aligned}
\label{eq:alpha}
J_1[J_{\pm},\theta]&=&\frac{1}{2}J_\pm\cos^2\theta, \\
J_2[J_{z\pm},\theta]&=&\frac{1}{2}J_{z\pm}\sin2\theta,
\label{eq:beta}\end{aligned}$$ i.e. $\ell_\mathbf{k}^\pm(\theta)=J_1[J_\pm,\theta] L_\mathbf{k}\mp |J_2[J_{z\pm},\theta] M_\mathbf{k}|$, and assumed $J_2[J_{z\pm},\theta] >0$. Then we find $$\begin{aligned}
\label{eq:lambdamin}
&& \lambda_{{\rm min}} =\\
&&\;\; \begin{cases}
6 J_1[J_{\pm},\theta] + \frac{J_2[J_{z\pm},\theta]^2}{2J_1[J_{\pm},\theta]} & \mbox{for }
J_2[J_{z\pm},\theta]< 4 J_1[J_{\pm},\theta] \\
4 J_2[J_{z\pm},\theta]-2J_1[J_{\pm},\theta] & \mbox{for } J_2[J_{z\pm},\theta]> 4 J_1[J_{\pm},\theta]
\end{cases}.\nonumber\end{aligned}$$
There is one subtlety. When $\theta=0$, $z^+_\mathbf{k}=z^-_\mathbf{k}$, so that there will be a degeneracy of the minimum energy state. This changes the condensed contributions. The two cases can actually be treated together, provided we formulate everything in terms of a new variable $\rho$, the “condensed density”, rather than $\delta$, where $$\begin{cases}
\rho(\theta>0)=\frac{1}{2}\sqrt{\frac{J_{zz}}{2}} \frac{1}{\delta} \\
\rho(\theta=0) =\sqrt{\frac{J_{zz}}{2}} \frac{1}{\delta}
\end{cases},$$ i.e. the condensed density is larger in the case $\theta=0$ by a factor of $2$. Then, from Eqs. and , we find $$\begin{aligned}
\label{eq:60}
\rho &=& 1 - I'_3,\\
\label{eq:62}
I_1 &=& \frac{\rho}{\sqrt{2}} \sqrt{1-y_{\rm min}} + I'_1,\\
I_2 &=& \rho(1+2y_{\rm min}) + I'_2.
\label{eq:63}\end{aligned}$$ When $\theta=0$, the minimum is reached at $\mathbf{q}_{\rm min}=\mathbf{0}$, i.e. $y_{\rm min}=1$. Eq. then reduces to $I_1'(\theta=0)=0$ since $I_1(\theta=0)$ vanishes identically (which reflects the fact that the two FCC sublattices are decoupled), and Eq. reduces to $I_2 = 3\rho+ I'_2$.
Ground State Energy Calculation
===============================
In some cases, we find multiple solutions of the mean-field equations. The physical solution is the one with the lowest ground state energy. To find it, we must evaluate the energy in the mean field approximation. There are different ways to address this. Here we show how to calculate the ground state energy “directly”, i.e. by taking the expectation value of the Hamiltonian. We arrive at Eqs. and given in the main text.
We write the Hamiltonian in Eq. as $$\label{eq:47}
H = \sum_{\mathbf{r} \in {\rm I},{\rm II}} \frac{J_{zz}}{2} Q_\mathbf{r}^2 + \mathcal{H}[\Phi,{\sf s}].$$ We can treat the gMFT as a variational calculation, with the trial wavefunction being the ground state of the gMFT Hamiltonian. The expectation value of the second term is straightforward, but that of the first is not. We need to write the Hamiltonian generalization of the approximation of “softening” the $|\Phi|=1$ constraint in the path integral.
To do this, we write $\Phi_\mathbf{r} = x_\mathbf{r} + i y_\mathbf{r}$, and introduce canonical momenta $p_{x,\mathbf{r}}, p_{y,\mathbf{r}}$, such that $[x_\mathbf{r}, p_{x,\mathbf{r}}]=i$ etc. Then if we write $\Pi_\mathbf{r} = p_{x,\mathbf{r}}+i p_{y,\mathbf{r}}$, the equivalent approximation in Hamiltonian form is $$\label{eq:48}
H \rightarrow \sum_{\mathbf{r} \in {\rm I},{\rm II}} \left\{ \frac{J_{zz}}{2} \Pi_\mathbf{r}^\dagger \Pi_\mathbf{r}^{\vphantom\dagger} +
\lambda (\Phi_\mathbf{r}^\dagger \Phi_\mathbf{r}^{\vphantom\dagger}-1) \right\} +
\mathcal{H}[\Phi,{\sf s}],$$ since $$\Pi^\dagger\Pi\sim-\frac{1}{|\Phi|}\frac{\partial}{\partial|\Phi|}\left(|\Phi|\frac{\partial}{\partial|\Phi|}\right)+\frac{1}{|\Phi|^2}Q^2\sim Q^2,$$ because $|\Phi|=1$, as enforced by $\lambda (\Phi_\mathbf{r}^\dagger \Phi_\mathbf{r}^{\vphantom\dagger}-1)$ with $\lambda>0$. One can check that, writing the path integral for the Hamiltonian Eq. and integrating out $\Pi_\mathbf{r}$, one obtains the action used previously. Note that we can see that there should be four harmonic oscillators per unit cell, arising from $p_{x,\mathbf{r}},p_{y,\mathbf{r}}$ for the two sites $\mathbf{r}$ in the diamond basis. This implies in turn that each of the two spinon branches $\omega^\pm_\mathbf{k}$ is doubly degenerate.
Returning to the energy calculation, we simply can take the expectation value of Eq. . Note that, because we choose $\langle \Phi_\mathbf{r}^\dagger\Phi_\mathbf{r}\rangle = 1$, the expectation value of the second term in the brackets (with the $\lambda$ coefficient) is zero. Hence the energy is $$\label{eq:49}
E_{\rm GS} = \sum_{\mathbf{r} \in {\rm I},{\rm II}} \frac{J_{zz}}{2} \left\langle
\Pi_\mathbf{r}^\dagger \Pi_\mathbf{r}^{\vphantom\dagger} \right\rangle +
\left\langle\mathcal{H}[\Phi,{\sf s}] \right\rangle = E_{kin}+ E_{av}.$$ The second term is $$\begin{aligned}
\label{eq:50}
E_{av} & = & \left\langle\mathcal{H}[\Phi,{\sf s}] \right\rangle \\
& = & N_{u.c.}\left[ - 2 I_2 \cos^2\theta J_\pm - 4 I_1
\sin2\theta J_{z\pm}\right], \nonumber\end{aligned}$$ i.e. half of the constant term obtained from the replacement Eq. (up to a minus sign). The “kinetic energy” $E_{kin}$ requires more thought. We can calculate it by the path integral. One obtains $$\label{eq:51}
\left\langle
\Pi_\mathbf{r}^\dagger \Pi_\mathbf{r}^{\vphantom\dagger} \right\rangle =
\frac{1}{J_{zz}}\left[ 2\delta(\tau=0) - \frac{\langle \partial_\tau
\Phi^\dagger_\mathbf{r} \partial_\tau \Phi_\mathbf{r}^{\vphantom\dagger}\rangle}{J_{zz}}\right].$$ Each of the two terms in the square brackets is formally divergent, but together they give a finite answer. This can be seen by writing it all in Fourier space: $$\begin{aligned}
\label{eq:52}
E_{kin} & = & \frac{J_{zz}}{2} N_{u.c.} \sum_{\mathbf{r}\in{\rm I},{\rm II}} \left\langle
\Pi_\mathbf{r}^\dagger \Pi_\mathbf{r}^{\vphantom\dagger} \right\rangle \\
& = & \frac{2N_{u.c.}}{2J_{zz}} \int \! \frac{d\omega_n}{2\pi} \int_\mathbf{k} \left[
2J_{zz} - \omega_n^2 \left[G_{\mathbf{k},\omega_n}\right]_{11}\right]. \nonumber \end{aligned}$$ Now we can rewrite the Green’s function as $$\begin{aligned}
\label{eq:53}
\left[G_{\mathbf{k},\omega_n}\right]_{11} & = & \frac{\Omega_n^2 + \frac{1}{2}(z_+^2 +
z_-^2)}{(\Omega_n^2 + z_+^2)(\Omega_n^2+z_-^2)},\end{aligned}$$ where $\Omega_n = \omega_n/\sqrt{2J_{zz}}$, and $z_\pm=z^\pm_\mathbf{k}$. Changing variables from $\omega_n$ to $\Omega_n$, we have $$\begin{aligned}
\label{eq:54}
E_{kin} & = & 2 N_{u.c.} \sqrt{2J_{zz}}\int_\mathbf{k} \int \! \frac{d\Omega_n}{2\pi}
\frac{\Omega_n^2 (z_+^2+z_-^2)/2 + z_+^2 z_-^2}{(\Omega_n^2 +
z_+^2)(\Omega_n^2+z_-^2)} \nonumber \\
&=& \frac{N_{u.c.}}{2} \int_\mathbf{k} [\omega^+_\mathbf{k} + \omega^-_\mathbf{k}].\end{aligned}$$
We note that this formulation makes it clear that this form of the energy is variational. Specifically, even if $\theta$ is [*not*]{} chosen to satisfy the mean field condition, this should give an upper bound to the ground state energy. This means that this form of the energy should be minimized when $\theta$ equals its true value, which is the minimum energy mean field solution. So this form could be used, if desired, to [*search*]{} for the ground state by just minimizing this energy as a function of $\theta$. Note that we do need to choose $\lambda$ as a function of $\theta$ to enforce the $|\Phi|=1$ condition for this to be valid, though.
Phase Boundaries
================
Most generally, the phase transitions of Figure \[fig:MF-phasediag\] are found numerically as the loci of points where the lowest-energy solutions to the gMFT equations have different $(\theta,\lambda)$ characteristics on either side of the transition line. It turns out that some of these lines coincide with curves which can be found more directly. Here we outline which transitions have a direct physical meaning, and how we find them.
We define, and shall use extensively, the dimensionless exchange constants $$\label{eq:dimensionless}
\tilde{J}_i=\frac{J_i}{J_{zz}}.$$ Finally, the problem being symmetric in $J_{z\pm}\rightarrow-J_{z\pm}$, we will always assume $J_{z\pm}\geq0$. In the latter case, all the values of $\theta$ of interest are contained in the $[0,\pi/2[$ interval, so that $J_2$, defined in Eq. , is always positive. Hence, $|J_2|=J_2$.
QSL - AFM
---------
In both the QSL and AFM, $\theta=0$. This considerably simplifies the problem. Using Eqs. and , we find that the maximum allowed $J_{\pm}$ [*within the QSL*]{} is such that $\lambda=3J_{\pm}$, from which $$\label{eq:QSL-AFM}
\left(\frac{J_{\pm}}{J_{zz}}\right)_{{\rm max}}^{\rm QSL}=\frac{1}{2}\left(\int_\mathbf{k}\frac{1}{\sqrt{3-\frac{1}{2}L_\mathbf{k}}}\right)^2$$ follows. The actual QSL-AFM transition happens when $\frac{J_{\pm}}{J_{zz}}=\left(\frac{J_{\pm}}{J_{zz}}\right)_{{\rm max}}^{\rm QSL}$.
QSL - CFM
---------
We find that the QSL-CFM transition is continuous in $\theta$. We can thus find the transition line as follows. We expand the consistency equations Eqs. to first order in $\theta$, around $\theta=0$, $$\begin{aligned}
I_1(\theta,\lambda)&=&\theta \left|J_{z\pm}\right|\hat{I}_1(\lambda)+O(\theta^2)\\
I_2(\theta,\lambda)&=&I_2(0,\lambda)+O(\theta^2)\\
I_3(\theta,\lambda)&=&I_3(0,\lambda)+O(\theta^2)\end{aligned}$$ where $\hat{I}_1(\lambda)=\frac{1}{2}\sqrt{\frac{J_{zz}}{2}}\left(\frac{Z_\mathbf{k}^+(\theta=0,\lambda)}{2}\right)^{3}$ (note that $\hat{I}_1$ has the dimension of $1/J$), so that Eqs. become $$\begin{cases}
J_\pm I_2(0,\lambda)=2J_{z\pm}^2\hat{I}_1(\lambda)\\
I_3(0,\lambda)=1
\end{cases},$$ where neither $I_2(0,\lambda)$, $I_3(0,\lambda)$ nor $\hat{I}_1(\lambda)$ depend on $J_{z\pm}$. For every value of $J_\pm$, $\lambda$ is the solution of $I_3(0,\lambda)=1$, so that we find $J_{z\pm}/J_{zz}$ at the transition to be $$\left(\frac{J_{z\pm}}{J_{zz}}\right)_{crit}^{\rm QSL-CFM}=\sqrt{\frac{\tilde{J}_\pm I_2(0,\lambda)}{2J_{zz}\hat{I}_1(\lambda)}},$$ with $\lambda(J_\pm)$ first determined by solving $I_3(0,\lambda)=1$ at fixed $J_\pm$.
CFM - FM
--------
The CFM-FM transition coincides with the first appearance of a condensed solution (at $J_\pm$ fixed for increasing $J_{z\pm}$) to the gMFT equations Eqs. . To find this transition line, we use the generic form of $\ell_\mathbf{k}^\pm=J_1 L_\mathbf{k}\mp|J_2 M_\mathbf{k}|$. In other words, we treat $J_1$ and $J_2$ defined in Eqs. and as parameters. We also define $r=J_2/J_1=2\frac{J_{z\pm}}{J_\pm}\tan\theta$, so that $$\ell^\pm_\mathbf{k}=J_1 L_\mathbf{k}\mp|J_2 M_\mathbf{k}|=J_1\left(L_\mathbf{k}\mp r |M_\mathbf{k}|\right).$$
The system being at the onset of a [*possible*]{} condensate, we know the possible values of $\lambda$ at this onset (see above section, Eq. ), $$\begin{cases}
\lambda=2J_1\left(2r-1\right)\\
\mbox{or}\;\lambda=J_1\left(6+\frac{r^2}{2}\right)
\end{cases},$$ so that, from $I_3(\theta,\lambda)=1$ (with these values of $\lambda$), we get either $$\label{eq:alpha-r1}
J_1\left(r\right)=\frac{J_{zz}}{8}\left[\int d^3k\sum_{\nu=\pm1}\frac{1}{\sqrt{r(4+\nu |M_\mathbf{k}|)-2-L_\mathbf{k}}}\right]^2,$$ or $$\label{eq:alpha-r2}
J_1\left(r\right)=\frac{J_{zz}}{8}\left[\int d^3k\sum_{\nu=\pm1}\frac{1}{\sqrt{6-L_\mathbf{k}+r\left(\frac{1}{2}r+\nu |M_\mathbf{k}|\right)}}\right]^2,$$ i.e. $J_1$ as a function of $r$. Using the above equations, we can rewrite the consistency equation Eq. in a form suitable for determining $r$ as an unknown. For example, in the first case Eq. , the transition line is best found as a function of $J_\pm$, so we fix $J_\pm$, and use the simple relations (obtained from Eqs. and ) $$\begin{aligned}
\theta&=&\arccos\sqrt{\frac{2J_1}{J_\pm}}\\
J_{z\pm}&=&\frac{1}{2}r\frac{1}{\tan\theta}J_\pm.\end{aligned}$$ Using these relations in Eq. and Eq. , we obtain an equation which can be solved numerically for $r$ as a function of $J_\pm$. Then, by substituting back, we get $J_{z\pm}$ and hence the phase boundary in the $J_{\pm}$–$J_{z\pm}$ plane. In the second case Eq. , we fix $\theta$ (as opposed to $J_\pm$), use the relations $$\begin{aligned}
J_\pm&=&\frac{2J_1}{\cos^2\theta}\\
J_{z\pm}&=&\frac{1}{2}r\frac{1}{\tan\theta}J_\pm,\end{aligned}$$ and proceed similarly.
CFM - AFM and AFM - FM
----------------------
The transitions are here discontinuous in $\theta$, and we find the transitions indirectly, except for a small portion of the CFM-AFM transition line, which coincides with the limiting boundary described above, with $\lambda=6J_1+\frac{J_2^2}{2J_1}$.
Benchmarking gMFT
=================
Classical approach
------------------
In Figure \[fig:classical-phasediag\], we compare the phase diagram of the gMFT with that expected by treating the spins in Eq. classically. The large $J_\pm/J_{z\pm}$ region is found to be antiferromagnetic, while the large $J_{z\pm}/J_{\pm}$ is ferromagnetic.
The transition line is found analytically. To achieve this, we use the fact that, both in the FM and AFM phases, the classical spins take a specific and simple form.
In the FM phase, $$\begin{cases}
\vec{S}_0^{\rm FM}=(x\mathbf{\hat{a}}_0+y\mathbf{\hat{b}}_0+z\mathbf{\hat{e}}_0)/2\\
\vec{S}_1^{\rm FM}=(-x\mathbf{\hat{a}}_1-y\mathbf{\hat{b}}_1-z\mathbf{\hat{e}}_1)/2\\
\vec{S}_2^{\rm FM}=(-x\mathbf{\hat{a}}_2-y\mathbf{\hat{b}}_2-z\mathbf{\hat{e}}_2)/2\\
\vec{S}_3^{\rm FM}=(x\mathbf{\hat{a}}_3+y\mathbf{\hat{b}}_3+z\mathbf{\hat{e}}_3)/2
\end{cases},$$ where $x,y,z$ are free parameters with $x^2+y^2+z^2 = 1$. In polar coordinates, we write $x=\sin\Theta\cos\phi, y = \sin\Theta\sin\phi,
z=\cos\Theta$, and the energy Eq. per unit cell becomes $$\begin{aligned}
\epsilon^{\rm FM}_{class}&=&\frac{1}{8}\Big[-2\cos^2\Theta J_{zz}+4\sin^2\Theta J_\pm\\
&&\qquad\left.+4\sin2\Theta\left(\cos\phi+\sqrt{3}\sin\phi\right)J_{z\pm}\right].\nonumber\end{aligned}$$ The energy is then found to be minimized for $\phi=\pi/3$ and $\Theta=\arctan\left[\frac{1+2\tilde{J}_\pm-\sqrt{1+4\tilde{J}_\pm(\tilde{J}_\pm+1)+64\tilde{J}_{z\pm}^2}}{8\tilde{J}_{z\pm}}\right]$, which yields $\epsilon^{\rm FM}_{class}=\dfrac{J_{zz}}{8}\left[-1+2\tilde{J}_\pm-\sqrt{(1+2\tilde{J}_\pm)^2+64\tilde{J}_{z\pm}^2}\right]$, where $\tilde{J}_i=J_i/J_{zz}$, Eq. .
In the AFM phase, the solution takes the form $$\vec{S}_\mu^{\rm FM}=\frac{1}{2}\left(X\mathbf{\hat{a}}_\mu+Y\mathbf{\hat{b}}_\mu\right),$$ with arbitrary $X,Y$ such that $X^2+Y^2=1$. This gives the energy per unit cell $$\epsilon^{\rm AFM}_{class}=-\frac{3J_\pm}{2}.$$ Equating $\epsilon^{\rm FM}_{class}=\epsilon^{\rm AFM}_{class}$, we readily find that the transition line takes the form $$\left(\frac{J_{z\pm}}{J_{zz}}\right)_{crit}^{class}=\sqrt{\frac{\tilde{J}_\pm(6\tilde{J}_\pm-1)}{2}}.$$ The comparison between the classical and the gMFT diagrams shows that the two approaches agree in the semiclassical limit of the gMFT, i.e. for $J_\pm/J_{zz},J_{z\pm}/J_{zz}\gg1$.
{width=".8\textwidth"}
Perturbative approach
---------------------
In the small $J_{z\pm}/J_{zz}$ and $J_{\pm}/J_{zz}$ limit, the perturbative approach developed in Refs. and applies. There, the effective Hamiltonian is $$\begin{aligned}
H_{\rm eff}^{\rm pert}&=&-K\sum_{\{i,j,k,l,m,n\}=\hexagon}\left(\mathsf{S}_i^+\mathsf{S}_j^-\mathsf{S}_k^+\mathsf{S}_l^-\mathsf{S}_m^+\mathsf{S}_n^-+\mbox{h.c.}\right)\nonumber \\
&& \;-J_{(3)}\sum_{\langle\langle\langle i,j\rangle\rangle\rangle}\mathsf{S}_i^z\mathsf{S}_j^z,\end{aligned}$$ with $K=\dfrac{12J_\pm^3}{J_{zz}^2}$ and $J_{(3)}=\dfrac{3J_{z\pm}^2}{J_{zz}}$, and $\hexagon$ represents the (flat) hexagons of the [*pyrochlore*]{} lattice. While the first term induces the spin liquid physics, the second, taken alone has the six ferromagnetic ground states described in Ref. , with polarization along the $\langle100\rangle$ directions. The actual phase diagram is therefore expected to differ from that of the gMFT in the small $J_{z\pm}/J_{zz}$ region (but agree when $J_{z\pm}/J_{zz}=0$). Dimensionally we find that the transition between the FM and the CFM in this region occurs along $$\left(\frac{J_{z\pm}}{J_{zz}}\right)_{crit}^{\rm PT}\sim\left(\frac{J_{\pm}}{J_{zz}}\right)^{3/2},$$ in the region where perturbation theory applies. A [*sketch*]{} of the expected gMFT diagram which includes this perturbative limit is sketched in Figure \[fig:modified-gMFT-diag\].
![Modified gMFT diagram, which takes into account the known perturbative limit $J_{z\pm}/J_{zz}\ll1$. Note that the FM-CFM transition (white dashed line) in the latter region is a [*sketch*]{}.[]{data-label="fig:modified-gMFT-diag"}](fig3_gMFTmod-phasediag.pdf){width="3.3in"}
Physical Properties
===================
Staggered magnetization
-----------------------
Here we calculate the “staggered” magnetization within the QSL and AFM phases, i.e. the [*local*]{} magnetization of each spin. This quantity will allow us to distinguish between the QSL (uncondensed) and AFM (condensed) since the reasons for overall zero magnetization in both phases are different. Indeed, in the AFM, the local the spin expectation values are non-zero but compensate, while in the QSL, the (local) XY symmetry remains unbroken, and $\langle \vec{\mathsf{S}}_i\rangle=\vec{0}$ everywhere.
We calculate the staggered magnetization along the local $x$ axes. To do so, we compute the correlation function $\langle\mathsf{S}_i^+\mathsf{S}_j^-\rangle$, where sites $i$ and $j$ are taken far apart from one another, where they can be considered independent. $$\begin{aligned}
\langle\mathsf{S}_i^+\mathsf{S}_j^-\rangle&=&\langle\Phi_{\mathbf{r}}^\dagger\,\mathsf{s}^+_{\mathbf{r},\mathbf{r}+\mathbf{e}_\mu}\Phi_{\mathbf{r}+\mathbf{e}_\mu}\Phi_{\mathbf{r}'+\mathbf{e}_\nu}^\dagger\mathsf{s}^-_{\mathbf{r}',\mathbf{r}'+\mathbf{e}_\nu}\Phi_{\mathbf{r}'}\rangle\\
&=&\langle\mathsf{s}_\mu^+\rangle\langle\mathsf{s}_\nu^-\rangle\langle\Phi_{\mathbf{r}}^\dagger\Phi_{\mathbf{r}'}\rangle\langle\Phi_{\mathbf{r}+\mathbf{e}_\mu}\Phi_{\mathbf{r}'+\mathbf{e}_\nu}^\dagger\rangle,\end{aligned}$$ where $\mathbf{r}$ and $\mathbf{r}'$ are on diamond sublattice I and, of course, $i\in\langle\mathbf{r},\mathbf{r}+\mathbf{e}_\mu\rangle$ and $j\in\langle\mathbf{r}',\mathbf{r}'+\mathbf{e}_\nu\rangle$, and where we made use of Wick’s theorem. From Eq. , we find $$\begin{aligned}
\langle\Phi_{\mathbf{r}}^\dagger\Phi_{\mathbf{r}'}\rangle&=&\frac{1}{2N_{u.c.}}\sqrt{\frac{J_{zz}}{2}}\sum_\mathbf{k}Z_{\mathbf{k}}^+(0,\lambda)e^{i\mathbf{k}\cdot(\mathbf{r}'-\mathbf{r})}.\end{aligned}$$ In the QSL, the sum can be replaced by its corresponding integral, $$\frac{1}{N_{u.c.}}\sum_\mathbf{k}Z_{\mathbf{k}}^+(0,\lambda)e^{i\mathbf{k}\cdot(\mathbf{r}'-\mathbf{r})}\rightarrow\int_\mathbf{k}Z_{\mathbf{k}}^+(0,\lambda)e^{i\mathbf{k}\cdot(\mathbf{r}'-\mathbf{r})}=0,$$ because $|\mathbf{r}'-\mathbf{r}|\rightarrow\infty$, and $z^-_\mathbf{k}$ never reaches $0$ within the integration domain. We thereby recover $\langle\mathsf{S}^x\rangle=0$ in the QSL. In the AFM, however, the “minimum term” survives via the subextensive part of $\lambda$, Eq. , so that $$\begin{aligned}
\langle\Phi_{\mathbf{r}}^\dagger\Phi_{\mathbf{r}'}\rangle&=&\sqrt{\frac{J_{zz}}{2}}\frac{1}{\delta},\end{aligned}$$ since $\mathbf{k}_{\rm min}=\mathbf{0}$ when $\theta=0$. Using Eq. , we find $$\begin{aligned}
\sqrt{\frac{J_{zz}}{2}}\frac{1}{\delta}&=&1-\sqrt{\frac{J_{zz}}{2}}\left[\int_\mathbf{k}\frac{1}{\sqrt{\lambda_{\rm min}-\frac{1}{2}J_\pm L_\mathbf{k}}}\right]\\
&=&1-\sqrt{\frac{J_{zz}}{J_\pm}}\sqrt{\left(\frac{J_\pm}{J_{zz}}\right)_{\rm max}^{\rm QSL}},\end{aligned}$$ as defined in Eq. . Similarly, $\langle\Phi_{\mathbf{r}+\mathbf{e}_\mu}\Phi_{\mathbf{r}'+\mathbf{e}_\nu}^\dagger\rangle=\sqrt{\frac{J_{zz}}{2}}\frac{1}{\delta}$ in the AFM, so that $$\label{eq:stagmag}
\langle\mathsf{S}^x\rangle=\sqrt{\langle\mathsf{S}_i^+\mathsf{S}_j^-\rangle}=\frac{1}{2}\left(1-\sqrt{\frac{\tilde{J}_\pm^c}{\tilde{J}_\pm}}\right),$$ in the AFM phase for $J_{z\pm}=0$, where we have defined $\tilde{J}_\pm=J_\pm/J_{zz}$ and $\tilde{J}_\pm^c=\left(J_\pm/J_{zz}\right)_{\rm max}^{\rm QSL}$. $\langle\mathsf{S}^x\rangle$ at $J_{z\pm}=0$ is plotted versus $J_\pm/J_{zz}$ in Figure \[fig:staggered-mag\].
![Staggered magnetization $\langle\mathsf{S}^x\rangle$ as a function of $J_\pm/J_{zz}$, Eq. , for $J_{z\pm}=0$.[]{data-label="fig:staggered-mag"}](fig4_staggered-mag.pdf){width="3.3in"}
Structure factor
----------------
The excitations above the uncondensed ground states of Eq. comprise the photon, spinons and “electric” monopoles [@hermele; @jointpaper]. In general, the spinons and monopoles are gapped excitations, and so do not contribute to the low-energy part of the structure factor. Below some threshold, only the photon survives, and gives rise to a universal form of the scattering. Here we sketch the general argument which shows that it appears in the inelastic structure factor as a linearly dispersing mode at arbitrarily low energy, with an unusually small spectral weight and polarization dependence which distinguishes it from the usual spin waves. We will be deliberately vague here about lattice details, orientation of local spin axes, etc, as these affect only $O(1)$ dependencies of the amplitudes and not the important scaling properties. A much more complete and microscopic study of the structure factor in the QSL and CFM phases is planned for a future publication.
Rather than derive it microscopically, we rely on the universality of the low energy description of the emergent photon. We expect that it is described by the usual electromagnetic action, which may be written in the continuum for small momentum and low energy. We write this in Euclidean time in terms of the vector potential, and choose the gauge in which the scalar potential vanishes. The result is familiar: $$\label{eq:2}
\mathcal{S}_{\rm QED} = \int_{\tau {\bf r}} \Big[ \frac{c_1}{2} (\partial_\tau
{\bf A})^2 + \frac{c_2}{2} ({\boldsymbol \nabla}\times {\bf A})^2 \Big].$$ Here $c_1$ and $c_2$ are phenomenological parameters related to the effective dielectric constant and magnetic permeability for the effective electrodynamics. Importantly, the photon velocity is given by $v=\sqrt{c_2/c_1}$. We recall the relation between the formalism of Eq. and that of lattice electrodynamics, given in the main text, $$\begin{cases}
\mathsf{S}^z_{\mathbf{rr}'}=\mathsf{s}^z_{\mathbf{rr}'}=E_{\mathbf{rr}'}\\
\mathsf{S}^\pm_{\mathbf{rr}'}=\Phi_\mathbf{r}^\dagger\,\mathsf{s}^\pm_{\mathbf{rr}'}\Phi_{\mathbf{r}'}= \Phi_\mathbf{r}^\dagger\,e^{\pm iA_{\mathbf{rr}'}}\Phi_{\mathbf{r}'}
\end{cases},
\label{eq:spins-ELD}$$ where $\mathbf{r}$ is on diamond sublattice I. Because the transverse components of the spins, $\mathsf{S}^\pm$, involve the spinon fields, we expect that they contribute only above the two-spinon threshold. For low energies, it suffices to consider $\mathsf{S}^z$ only. We take the continuum limit by defining the electric field to be “oriented” along the diamond lattice bond $\mbox{I}\rightarrow\mbox{II}$ on which the spin lies and centered midway along it, i.e. $${\bf E}_{\mathbf{r}+\mathbf{e}_\mu/2}=E_{\mathbf{r},\mathbf{r}+\mathbf{e}_\mu}\mathbf{\hat{e}}_\mu,$$ Similarly, recall that the magnetic moment due to $\mathsf{S}^z_{\mathbf{r},\mathbf{r}+\mathbf{e}_\mu}$ is along the $\mathbf{\hat{e}}_\mu$ direction in real space, $$\label{eq:3}
\mathbf{S}_{\mathbf{r}+{\mathbf{e}_\mu}/2} = {\mathsf
S}^z_{\mathbf{r},\mathbf{r}+\mathbf{e}_\mu} {\bf\hat e}_\mu = {\bf
E}_{\mathbf{r}+\mathbf{e}_\mu/2} .$$ So we see that, in this low energy subspace where excited spinons may be neglected, the spin operator is precisely the electric field operator. Then, using the usual relation $\mathbf{E}(\mathbf{r})=-\boldsymbol{\nabla}_\mathbf{r}V(\mathbf{r})-\partial_\tau
\mathbf{A}(\mathbf{r})$ for $V=0$ (gauge choice), $$\label{eq:4}
{\mathbf S}_{\mathbf{k},\omega_n} = i\omega_n {\mathbf A}_{\mathbf{k},\omega_n},$$ where $\omega_n$ is the bosonic Matsubara frequency. Thus the Matsubara correlation function of the spins is $$\label{eq:6}
\langle S^i_{-\mathbf{k},-\omega_n}S^j_{\mathbf{k},\omega_n}\rangle=\omega_n^2
\langle
A^i_{-\mathbf{k},-\omega_n} A^j_{\mathbf{k},\omega_n}\rangle,$$ Calculation of the gauge field propagator is a textbook exercise: $$\label{eq:5}
\left\langle A^i_{-\mathbf{k}-\omega} A^j_{\mathbf{k}\omega}\right\rangle =
c_1^{-1}\frac{\omega_n^2 \delta_{ij}+v^2 k_i k_j }{\omega_n^2 + v^2
\mathbf{k}^2}.$$ Inserting this into Eq. and analytically continuing $i\omega_n \rightarrow \omega+i0^+$, we arrive finally at the inelastic structure factor $$\begin{aligned}
\label{eq:7}
{\mathcal F}_{\mathbf{k},\omega} &=& -\left.{\rm Im}\left[
\langle S^i_{-\mathbf{k},-\omega_n}S^j_{\mathbf{k},\omega_n}\rangle\right]\right|_{i\omega_n
\rightarrow \omega+i0^+}
\\
&\sim& \frac{\pi }{c_1} \big[ \delta_{ij} -
\frac{k_i k_j}{\mathbf{k}^2}\big] \; \omega \,\delta( \omega-v |\mathbf{k}|).\nonumber\end{aligned}$$ We see that the photon appears as a sharp peak in the structure factor, with a weight proportional to its frequency $\omega$. This indicates, as mentioned in the text, a strong suppression of the weight at low energy, especially relative to the familiar case of a spin wave, for which the weight [*diverges*]{} like $1/\omega$ in the same limit.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Rough sets are efficient for data pre-processing in data mining. As a generalization of the linear independence in vector spaces, matroids provide well-established platforms for greedy algorithms. In this paper, we apply rough sets to matroids and study the contraction of the dual of the corresponding matroid. First, for an equivalence relation on a universe, a matroidal structure of the rough set is established through the lower approximation operator. Second, the dual of the matroid and its properties such as independent sets, bases and rank function are investigated. Finally, the relationships between the contraction of the dual matroid to the complement of a single point set and the contraction of the dual matroid to the complement of the equivalence class of this point are studied. **Keywords.** rough set; approximation operator; matroid; contraction'
author:
- 'Jingqian Wang, William Zhu [^1]'
title: Rough sets and matroidal contraction
---
Introduction {#section0}
============
The theory of rough sets was proposed by Pawlak [@Pawlak91Rough; @Pawlak82Rough] in 1982 as a tool to conceptualize, organize and analyze various types of data in data mining. This theory is especially useful for dealing with uncertain and vague knowledge in information systems. Using the concepts of lower and upper approximations in rough set theory, knowledge hidden in information systems may be unraveled and expressed in the form of decision rules. The basic operators in rough set theory are approximation operators. Many examples of applications of the rough set theory to process control, economics, medical diagnosis, biochemistry, environmental science, biology, chemistry psychology, conflict analysis and other fields can be found in [@Liu06TheTransitive; @LiuWang06Semantic; @HungSankarSkowron11Rough; @QinPeiDu05TheRelationship; @WangLiu00TheInconsistency; @WuZhang06Rough; @YaoChen05Subsystem; @ZhongYaoOhshima03peculiarity; @ZhuWang06Covering].
Matroids [@Lai01Matroid; @Li07Some; @Mao06TheRelation; @WangZhu11Matroidal; @ZhuWang11Matroidal] are structures that generalize linear independence in vector spaces. They borrow extensively linear algebra and graph theory, and have a variety of applications in combinatorial optimization, integer programming, secret communication, and so on. Recently, the combination of rough sets and matroids has been studied in [@LiLiu12Matroidal; @WangZhu11Matroidal; @WangZhuZhu10Abstract; @ZhuWang11Matroidal; @ZhuWang11Rough].
In this paper, a matroidal structure of rough sets is constructed, and then the contraction of the dual of the matroid is studied. First, a matroid is induced by the lower approximation operator in rough sets. Second, the dual of the matroid is investigated. Finally, the relationships between the contraction of the dual matroid to the complement of a single point set and the contraction of the dual matroid to the complement of the equivalence class of this point are studied.
The rest of this paper is organized as follows. Section \[section1\] reviews some fundamental concepts about rough sets and matroids. In Section \[section2\], a matroid is induced by the lower approximation operator in rough sets, and the relationships between the contraction of the dual matroid to the complement of a single point set and the contraction of the dual matroid to the complement of the equivalence class of this point are studied. Finally, this paper is concluded and further work is indicated in Section \[section3\].
Basic definitions {#section1}
=================
This section recalls some fundamental definitions related to rough sets and matroids.
(Approximation space [@Yao98Relational; @Yao98Constructive]) \[definition1\] Let $U$ be a nonempty and finite set called universe and $R$ an equivalence relation on $U$. The ordered pair $(U,R)$ is called a Pawlak’s approximation space.
In rough sets, a pair of approximation operators are used to describe an object. In the following definition, a widely used pair of approximation operators are introduced.
(Approximation operator [@Yao98Relational; @Yao98Constructive]) \[definition2\] Let $R$ be an equivalence relation on $U$. A pair of approximation operators $R_{*}$, $R^{*}:2^{U}\rightarrow 2^{U}$, are defined as follows: for all $X\subseteq U$,\
$R_{*}(X)=\{x\in U:RN(x)\subseteq X\}$.\
$R^{*}(X)=\{x\in U:RN(x)\bigcap X\neq \emptyset\}$.\
Where $RN(x)=\{y\in U:xRy\}$ is called the equivalence class of $x$, and $RN(x)$ is a member of the partition generated by $R$ on $U$. They are called the lower and upper approximation operators with respect to $R$, respectively.
Let $\emptyset$ be the empty set, $X^{c}$ the complement of $X$ in $U$, from the definition of approximation operators, we have the following conclusions about them.
( [@Yao98Relational; @Yao98Constructive]) \[proposition12\] The properties of the Pawlak’s rough sets:\
(1L) $ R_{*}(U)= U $ (1H) $ R^{*}(U)= U $\
(2L) $ R_{*}(\phi)= \phi$ (2H) $ R^{*}(\phi)= \phi$\
(3L) $ R_{*}(X)\subseteq X $ (3H) $X \subseteq R^{*}(X) $\
(4L) $R_{*}(X \bigcap Y )= R_{*}(X) \bigcap R_{*}(Y )$\
(4H) $R^{*}(X \bigcup Y )= R^{*}(X) \bigcup R^{*}(Y )$\
(5L) $R_{*}(R_{*}(X))= R_{*}(X) $ (5H) $R^{*}(R^{*}(X))= R^{*}(X) $\
(6L) $X \subseteq Y \Rightarrow R_{*}(X) \subseteq R_{*}(Y )$ (6H) $X \subseteq Y \Rightarrow R^{*}(X) \subseteq R^{*}(Y )$\
(7L) $R_{*}((R_{*}(X))^{c})= (R_{*}(X))^{c}$ (7H) $R^{*}((R^{*}(X))^{c})= (R^{*}(X))^{c}$\
(8LH) $R_{*}(X^{c})= (R^{*}(X))^{c} $\
(9LH) $R_{*}(X) \subseteq R^{*}(X)$
Matroids are algebraic structures that generalize linear independence in vector spaces. They have a variety of applications in integer programming, combinatorial optimization, algorithm design, and so on. In the following definition, one of the most valuable definitions of matroids is presented in terms of independent sets.
(Matroid [@Lai01Matroid]) \[definition3\] A matroid is an ordered pair $M=(U,\textbf{I})$ where $U$(the ground set) is a finite set, and $\textbf{I}$ (the independent sets) a family of subsets of $U$ with the following properties:\
(I1) $\emptyset \in \textbf{I}$.\
(I2) If $I\in \textbf{I}$, and $I^{'}\subseteq I$, then $I^{'}\in \textbf{I}$.\
(I3) If $I_{1}, I_{2}\in \textbf{I}$, and $|I_{1}|< |I_{2}|$, then there exists $e\in I_{2}-I_{1}$, such that $I_{1}\bigcup \{e\}\in \textbf{I}$, where $|I|$ denotes the cardinality of $I$.
\[example1\] Let $U=\{a,b,c,d\}$, $\textbf{I}=\{\emptyset,\{a\},\{b\},\{c\},\{a,b\},\{a,c\},\{b,$ $c\}\}$. Then $(U,\textbf{I})$ is a matroid.
If a subset of the ground set is not an independent set of a matroid, then it is called a dependent set of the matroid. Based on the dependent set, we introduce the circuit of a matroid. For this purpose, two denotations are presented.
([@Lai01Matroid]) \[definition4\] Let $U$ be a nonempty and finite set and $\textbf{A}\subseteq 2^{U}$ a family of subsets of $U$. One can denote:\
$Max(\textbf{A})=\{X\in \textbf{A}:\forall Y\in \textbf{A},X\subseteq Y\Rightarrow X=Y\}$.\
$Min(\textbf{A})=\{X\in \textbf{A}:\forall Y\in \textbf{A},Y\subseteq X\Rightarrow X=Y\}$.
The dependent set of a matroid generalizes the linear dependence in vector spaces and the cycle in graphs. Any circuit of a matroid is a minimal dependent set.
(Circuit [@Lai01Matroid]) \[definition5\] Let $M=(U,\textbf{I})$ be a matroid. A minimal dependent set in $M$ is called a circuit of $M$, and we denote the family of all circuits of $M$ by $\textbf{C}(M)$, i.e., $\textbf{C}(M)=Min(\textbf{I}^{c})$, where $\textbf{I}^{c}=2^{U}-\textbf{I}$.
(Continued from Example \[example1\]) $\textbf{C}(M)=\{\{d\},\{a,b,c\}\}.$
A base of a matroid is a maximal independent set.
(Base [@Lai01Matroid]) \[definition6\] Let $M=(U,\textbf{I})$ be a matroid. A maximal independent set in $M$ is called a base of $M$, and we denote the family of all bases of $M$ by $\textbf{B}(M)$, i.e., $\textbf{B}(M)=Max(\textbf{I})$.
(Continued from Example \[example1\]) $\textbf{B}(M)=\{\{a,b\},\{a,$ $c\},\{b,c\}\}$.
The following proposition shows that a matroid can be determined by bases.
(Base axiom [@Lai01Matroid]) \[proposition0\] Let $U$ be a finite set and $\textbf{B}$ a family of subsets of $U$. Then there exists $M=(U,\textbf{I})$ such that $\textbf{B}=\textbf{B}(M)$ iff $\textbf{B}$ satisfies the following two conditions:\
(B1) $\textbf{B}\neq \emptyset$.\
(B2) If $B_{1}$, $B_{2}\in \textbf{B}$ and $x\in B_{1}-B_{2}$, then there exists an element $y\in B_{2}-B_{1}$ such that $(B_{1}-\{x\})\bigcup \{y\}\in \textbf{B}$.
The dimension of a vector space and the rank of a matrix are useful concepts in linear algebra. The rank function of a matroid is a generalization of these two concepts.
(Rank function [@Lai01Matroid]) \[definition7\] Let $M=(U,\textbf{I})$ be a matroid. The rank function $r_{M}$ of $M$ is defined as $r_{M}(X)=max\{|I|:I\subseteq X,I\in \textbf{I}\}$ for all $X\subseteq U$. $r_{M}(X)$ is called the rank of $X$ in $M$.
(Continued from Example \[example1\]) Suppose $X_{1}=\{a\}$, $X_{2}=\{a,b,c\}$, $X_{3}=\{a,b,d\}$. Then $r_{M}(X_{1})=1$, $r_{M}(X_{2})=2$, $r_{M}(X_{3})=2 $.
The following proposition presents the properties of the rank function of a matroid.
([@Lai01Matroid]) \[proposition1\] Let $M=(U,\textbf{I})$ be a matroid and $r_{M}$ the rank function of $M$. Then $r_{M}$ satisfies the following conditions:\
(R1) If $X\in 2^{U}$, then $0\leq r_{M}(X)\leq |X|$.\
(R2) If $X\subseteq Y\subseteq U$, then $r_{M}(X)\leq r_{M}(Y)$.\
(R3) If $X, Y\subseteq U$, then $r_{M}(X)+ r_{M}(Y)\geq r_{M}(X\bigcup Y)+r_{M}(X\bigcap Y)$.
Given a matroid, we can generate a new matroid through the following proposition.
([@Lai01Matroid]) \[proposition2\] Let $M=(U,\textbf{I})$ be a matroid and $\textbf{B}^{*}=\{U-B:B\in \textbf{B}(M)\}$. Then $\textbf{B}^{*}$ is the family of bases of a matroid on $U$.
The new matroid in the above proposition, whose ground set is $U$ and whose set of bases is $\textbf{B}^{*}$, is called the dual of $M$ and denoted by $M^{*}$. Generally, the independent sets, the circuits, the bases and the rank function of $M^{*}$ are called the coindependent sets, the cocircuits, the cobases and the corank function of $M$.
Rough sets and matroidal contraction {#section2}
====================================
In this section, we firstly establish a matroidal structure of rough sets by the lower approximation operator. The dual of the matroid and its properties such as independent sets, bases and rank function are provided. Finally, the relationships between the contraction of the dual matroid to the complement of a single point set and the contraction of the dual matroid to the complement of the equivalence class of this point are studied.\
In the following proposition, for an equivalence relation on a universe, we establish a family of subsets through the lower approximation operator, and prove it satisfies the independent set axiom of a matroid. In other words, it determines a matroid.
\[proposition3\] Let $R$ be an equivalence relation on $U$. Then $\textbf{I(}R)=\{X\subseteq U:R_{*}(X)=\emptyset \}$ satisfies (I1), (I2) and (I3) of Definition \[definition3\].
First, according to Proposition \[proposition12\], $R_{*}(\emptyset)=\emptyset $. Then $ \emptyset \in \textbf{I}(R)$.\
Second, let $I\in \textbf{I}(R)$, $I^{'}\subseteq I$. Since $I\in \textbf{I}(R)$, so $R_{*}(I)=\emptyset$. According to Proposition \[proposition12\], $R_{*}(I^{'})\subseteq R_{*}(I)=\emptyset$. Therefore, $R_{*}(I^{'})=\emptyset $, i.e., $I^{'}\in \textbf{I}(R)$.\
Third, let the partition generated by $R$ on $U$ be $U/R=\{P_{1},P_{2},\cdots,P_{m}\}$. Let $I_{1}$, $I_{2}\in \textbf{I}(R)$ and $|I_{1}|<|I_{2}|$. Since $I_{1}=I_{1}\bigcap U$ and $I_{2}=I_{2}\bigcap U$, so $I_{1}=I_{1}\bigcap (\bigcup \limits_{i=1}^{m}P_{i})=\bigcup \limits_{i=1}^{m}(I_{1}\bigcap P_{i})$ and $I_{2}=I_{2}\bigcap (\bigcup \limits_{i=1}^{m}P_{i})=\bigcup \limits_{i=1}^{m}(I_{2}\bigcap P_{i})$. Since $I_{1}$, $I_{2}\in \textbf{I}(R)$, so $R_{*}(I_{1})=\emptyset$ and $R_{*}(I_{2})=\emptyset$. Thus, for all $1\leq i\leq m$, $(I_{1}\bigcap P_{i})\subset P_{i}$ and $(I_{2}\bigcap P_{i})\subset P_{i}$. Since $|I_{1}|<|I_{2}|$, so $|\bigcup \limits_{i=1}^{m}(I_{1}\bigcap P_{i})|<|\bigcup \limits_{i=1}^{m}(I_{2}\bigcap P_{i})|$, i.e., $\sum\limits_{i=1}^{m}|I_{1}\bigcap P_{i}|< \sum\limits_{i=1}^{m}|I_{2}\bigcap P_{i}|$. Therefore, there exists $1\leq i\leq m$ such that $|I_{1}\bigcap P_{i}|<|I_{2}\bigcap P_{i}|$ (In fact, if for all $1\leq i\leq m$ such that $|I_{1}\bigcap P_{i}|\geq |I_{2}\bigcap P_{i}|$, then $\sum\limits_{i=1}^{m}|I_{1}\bigcap P_{i}|\geq \sum\limits_{i=1}^{m}|I_{2}\bigcap P_{i}|$, i.e., $|I_{1}|\geq |I_{2}|$. It is contradictory with $|I_{1}|<|I_{2}|$). Thus, $|I_{1}\bigcap P_{i}|<|I_{2}\bigcap P_{i}|<|P_{i}|$, and there exists $e\in (I_{2}\bigcap P_{i})-(I_{1}\bigcap P_{i})\subseteq I_{2}-I_{1}$ such that $(I_{1}\bigcap P_{i})\bigcup \{e\}\subset P_{i}$, i.e., $R_{*}(I_{1}\bigcup \{e\})=\emptyset$. Hence $I_{1}\bigcup \{e\}\in \textbf{I}(R)$. This completes the proof.
In conclusion, $\textbf{I}(R)$ satisfies (I1), (I2) and (I3) of Definition \[definition3\]. Therefore, there exists a matroid on $U$ such that $\textbf{I}(R)$ is the family of its independent sets, and the matroid is denoted by $M(R)=(U,\textbf{I}(R))$. The family of bases of $M(R)$ denoted by $\textbf{B}(R)$.
\[example2\] Let $U=\{a,b,c,d,e\}$, $R$ an equivalence relation on $U$ and the partition generated by $R$ on $U$ be $U/R=\{\{a,b\},\{c,d,e\}\}$. According to Proposition \[proposition3\], $\textbf{I}(R)$ $=\{\emptyset,\{a\},\{b\},\{c\},\{d\},\{e\},\{a,c\},\{a,d\},\{a,$ $e\},\{b,c\},\{b,d\},\{b,e\},\{c,d\},\{c,e\},\{d,e\},\{a,c,d\},\{a,c,e\},\{a,d,e\},$ $\{b,c,d\},\{b,c,e\},\{b,d,e\}\}.$ Therefore, the matroid induced by the lower approximation operator is $M(R)=(U,\textbf{I}(R))$.
The following two corollaries represent the independent sets and the bases of the matroid induced by the lower approximation operator.
\[corollary1\] Let $R$ be an equivalence relation on $U$ and $M(R)$ the induced matroid. Then $\textbf{I}(R)=\{X\subseteq U:R_{*}(X)=\emptyset\}=\{X\subseteq U:\forall x\in U,RN(x)$ $\nsubseteq X \}.$
\[corollary2\] Let $R$ be an equivalence relation on $U$, $M(R)$ the induced matroid. Then $\textbf{B}(R)=Max(\textbf{I}(R))=\{X\subseteq U:\forall x\in U,|RN(x)$ $\bigcap X|=|RN(x)|-1\}.$
(Continued from Example \[example2\]) \[example3\] $\textbf{B}(R)=\{\{a,c,d\},\{a,c,e\},$ $\{a,d,e\},\{b,c,d\},\{b,c,e\},\{b,d,e\}\}$.
The dual of the matroid and its properties such as independent sets, bases and rank function are investigated in Proposition \[proposition4\], Corollary \[corollary3\] and Corollary \[corollary4\].
\[proposition4\] Let $R$ be an equivalence relation on $U$, $M(R)$ the induced matroid and $\textbf{B}^{*}(R)=\{U-B:B\in \textbf{B}(R)\}$$=\{X\subseteq $ $U:\forall x\in U,|RN(x)\bigcap X|=1\}$. Then $\textbf{B}^{*}(R)$ satisfies the base axiom.
According to Proposition \[proposition2\], it is straightforward.
(Continued from Example \[example2\]) \[example4\] $\textbf{B}^{*}(R)=\{\{a,c\},\{a,d\},\{a,e\},$ $\{b,c\},\{b,d\},\{b,e\}\}$.
According to Proposition \[proposition4\], $\textbf{B}^{*}(R)$ is the family of bases of a matroid, and the matroid denoted by $M^{*}(R)$. In fact, we know $M^{*}(R)$ is the dual matroid of $M(R)$, and we denote the family of the independent sets and the rank function of $M^{*}(R)$ as $\textbf{I}^{*}(R)$ and $r_{M^{*}(R)}$, respectively.
\[corollary3\] Let $R$ be an equivalence relation on $U$ and $M^{*}(R)$ the dual of the induced matroid $M(R)$. Then $\textbf{I}^{*}(R)=Min(\textbf{B}^{*}(R))=\{X\subseteq $ $U:\forall x\in U,|RN(x)\bigcap$ $ X|\leq1\}.$
We denote the dual matroid as $M^{*}(R)=(U,I^{*}(R))$.
(Continued from Example \[example2\]) $\textbf{I}^{*}(R)=\{\emptyset,\{a\},\{b\},\{c\},\{d\},$ $\{e\},\{a,c\},\{a,d\},\{a,e\},\{b,c\},\{b,d\},\{b,e\}\}$. Therefore the dual of the matroid $M(R)$ is $M^{*}(R)=(U,I^{*}(R))$.
\[corollary4\] Let $R$ be an equivalence relation on $U$ and $M^{*}(R)$ the dual of the induced matroid $M(R)$. For all $x\in U$ and $X\subseteq U$, $r_{M^{*}(R)}(X)=|\{RN(x):RN(x)\bigcap X\neq \emptyset\}|.$
(Continued from Example \[example2\]) Suppose $X_{1}=\{a,c\}$ and $X_{2}=\{c,d,e\}$. Then $r_{M^{*}(R)}(X_{1})=|\{\{a,b\},\{c,d,e\}\}|=2$ and $r_{M^{*}(R)}(X_{2})=|\{\{c,d,e\}\}|=1$.
The following proposition shows a relationship between the ranks of two subsets of a universe.
([@Lai01Matroid]) \[proposition5\] Let $U$ be a finite set and $r:2^{U}\rightarrow Z$ a function satisfying conditions (R2) and (R3) of Proposition \[proposition1\]. If $X,Y\subseteq U$ such that for all $y\in Y-X$, $r(X)=r(X\bigcup \{y\})$, then $r(X)=r(X\bigcup Y)$.
The following two definitions show two special matroids.
(Restriction [@Lai01Matroid]) \[definition8\] Let $M=(U,\textbf{I})$ be a matroid and $X\subseteq U$. Then $M|X=(X,\textbf{I}_{X})$ is a matroid called the restriction of $M$ to $X$, where $\textbf{I}_{X}=\{I\subseteq X:I\in \textbf{I}\}$.
(Contraction [@Lai01Matroid]) \[definition9\] Let $M=(U,\textbf{I})$ be a matroid, $T\subseteq U$ and $B_{T}$ be a base of $M|T$, i.e., $B_{T}\in \textbf{B}(M|T)$. Then $M/T=(U-T,\textbf{I}^{'})$ is a matroid called the contraction of $M$ to $U-T$, where $\textbf{I}^{'}=\{I\subseteq U-T:I\bigcup B_{T}\in \textbf{I}\}$. (The definition of $M/T$ has no relationship with the selection of $B_{T}\in \textbf{B}(M|T)$)
The following proposition shows a relationship of ranks between a matroid and the restriction of the matroid to a subset.
([@Lai01Matroid]) \[proposition6\] Let $M=(U,\textbf{I})$ be a matroid and $T\subseteq U$. For all $X\subseteq U-T$, $r_{M/T}(X)=r_{M}(X\bigcup T)-r_{M}(T)$.
In the following four propositions, we investigate the relationships between some characteristics of the contraction of the dual matroid to the complement of a single point set and to the complement of the equivalence class of this point, respectively, such as independent sets, bases, rank functions and circuits.
\[proposition7\] Let $R$ be an equivalence relation on $U$. For all $x\in U$, $\textbf{I}(M^{*}(R)/\{x\})= \textbf{I}(M^{*}(R)/RN(x))$.
According to Proposition \[proposition4\] and Corollary \[corollary3\], we know $\{x\}\in \textbf{B}(M^{*}(R)|\{x\})$ and $\{x\}\in \textbf{B}(M^{*}(R)|RN(x))$. Thus $\textbf{I}(M^{*}(R)/\{x\})=\{I\subseteq U-\{x\}:I\bigcup \{x\}\in \textbf{I}^{*}(R)\}$, $\textbf{I}(M^{*}(R)/RN(x))=$ $\{I\subseteq U-RN(x):I\bigcup \{x\}\in \textbf{I}^{*}(R)\}$. For all $Y\subseteq RN(x)-\{x\}$ and $Y\neq \emptyset$, $Y\bigcup \{x\}\notin \textbf{I}^{*}(R)$. Thus $\textbf{I}(M^{*}(R)/\{x\})=\{I\subseteq U-\{x\}:I\bigcup \{x\}\in \textbf{I}^{*}(R)\}=\{I\subseteq U-RN(x):I\bigcup \{x\}\in $ $\textbf{I}^{*}(R)\}$. Hence $\textbf{I}(M^{*}(R)/\{x\})=\textbf{I}(M^{*}(R)/RN(x))$. This completes the proof.
\[proposition9\] Let $R$ be an equivalence relation on $U$. For all $x\in U$, $\textbf{B}(M^{*}(R)/\{x\})=\textbf{B}(M^{*}(R)/RN(x))$.
According to Definition \[definition6\], $\textbf{B}(M^{*}(R)/\{x\})=Max(\textbf{I}(M^{*}(R)/$ $\{x\}))$, and $\textbf{B}(M^{*}(R)/RN(x))=Max(\textbf{I}(M^{*}(R)/RN(x)))$. According to Proposition \[proposition7\], $\textbf{I}(M^{*}(R)/\{x\})=\textbf{I}(M^{*}(R)/RN(x))$. Thus $\textbf{B}(M^{*}(R)/$ $\{x\})=\textbf{B}(M^{*}(R)/RN(x))$. This completes the proof.
\[proposition10\] Let $R$ be an equivalence relation on $U$. For all $x\in U$ and $X\subseteq U-RN(x)$, $r_{M^{*}(R)/\{x\}}(X)=r_{M^{*}(R)/RN(x)}(X)$.
For all $X\subseteq U-RN(x)$, $X\subseteq U-\{x\}$. According to Proposition \[proposition6\], $r_{M^{*}(R)/\{x\}}(X)=r_{M^{*}(R)}(X\bigcup \{x\})-r_{M^{*}(R)}(\{x\})$ and $r_{M^{*}(R)/RN(x)}$\
$(X)=r_{M^{*}(R)}(X\bigcup RN(x))-r_{M^{*}(R)}(RN(x))$. According to Corollary \[corollary4\], $r_{M^{*}(R)}(\{x\})=r_{M^{*}(R)}(RN(x))=1$. So, we only need to proof $r_{M^{*}(R)}(X\bigcup \{x\})=r_{M^{*}(R)}(X\bigcup RN(x))$. For all $y\in (X\bigcup RN(x))-(X\bigcup \{x\})=RN(x)-\{x\}$, $r_{M^{*}(R)}(X\bigcup \{x\})=r_{M^{*}(R)}(X\bigcup \{x,y\})$. According to Proposition \[proposition5\], $r_{M^{*}(R)}(X\bigcup \{x\})=r_{M^{*}(R)}((X\bigcup \{x\})\bigcup (X$ $\bigcup RN(x)))= r_{M^{*}(R)}(X\bigcup RN(x))$. This completes the proof.
\[proposition8\] Let $R$ be an equivalence relation on $U$. For all $x\in U$, $\textbf{C}(M^{*}(R)/RN(x))\subseteq \textbf{C}(M^{*}(R)/\{x\})$.
According to Proposition \[proposition7\], $\textbf{I}(M^{*}(R)/\{x\})=\textbf{I}(M^{*}(R)/RN(x))$. According to Definition \[definition5\], $\textbf{C}(M^{*}(R)/\{x\})=Min((\textbf{I}(M^{*}(R)/\{x\}))^{c})$, where $(\textbf{I}(M^{*}(R)/\{x\}))^{c}=2^{U-\{x\}}-\textbf{I}(M^{*}(R)/\{x\})$, and $\textbf{C}(M^{*}(R)/$ $RN(x))=Min((\textbf{I}(M^{*}(R)/RN(x)))^{c})$, where $(\textbf{I}(M^{*}(R)$ $/RN(x)))^{c}=2^{U-RN(x)}-\textbf{I}(M^{*}(R)/RN(x))$. Therefore, $Min((\textbf{I}(M^{*}(R)/RN(x)))^{c})$ $\subseteq Min((\textbf{I}(M^{*}(R)/\{x\}))^{c})$, i.e., $\textbf{C}(M^{*}(R)/RN(x))\subseteq \textbf{C}(M^{*}(R)/\{x\})$. This completes the proof.
In the following proposition, we study when the contraction of the dual matroid to the complement of a single point set and the contraction of the dual matroid to the complement of the equivalence class of this point have the same circuits.
\[proposition11\] Let $R$ be an equivalence relation on $U$. For all $x\in U$, $\textbf{C}(M^{*}(R)/\{x\}|(U-RN(x)))=\textbf{C}(M^{*}(R)/RN(x))$.
According to Proposition \[proposition7\], $\textbf{I}(M^{*}(R)/\{x\})=\textbf{I}(M^{*}(R)/RN(x))$ and $\textbf{I}(M^{*}(R)/\{x\})=$ $\textbf{I}(M^{*}(R)/\{x\}|(U-RN(x)))$. According to Definition \[definition5\], $\textbf{C}(M^{*}(R)/\{x\}|(U-RN(x)))=Min((\textbf{I}(M^{*}(R)/\{x\}|(U-RN(x))))^{c})$, where $(\textbf{I}(M^{*}(R)/\{x\}|(U-RN(x))))^{c}=2^{U-RN(x)}-\textbf{I}(M^{*}($ $R)/\{x\}|(U-RN(x)))=2^{U-RN(x)}-\textbf{I}(M^{*}($ $R)/\{x\})$, and $\textbf{C}(M^{*}(R)/$ $RN(x))=Min((\textbf{I}(M^{*}(R)/RN(x)))^{c})$, where $(\textbf{I}(M^{*}(R)/RN(x)))^{c}=2^{U-RN(x)}-\textbf{I}(M^{*}(R)/RN(x))$. Hence, $\textbf{C}(M^{*}($ $R)/\{x\}|(U-RN(x)))=\textbf{C}(M^{*}(R)/RN(x))$. This completes the proof.
As we know, if a single point set is the equivalence class of this point, then the contraction of the dual matroid to the complement of the single point set and the contraction of the dual matroid to the complement of the equivalence class of this point have the same independent sets, circuits, bases and rank functions.
Conclusions {#section3}
===========
\[S:Conclusions\] In this paper, we establish a matroidal structure of rough sets by the lower approximation operator. We provide the dual of the matroid and its properties such as independent sets, bases and rank function. Moreover, we study the relationships between the contraction of the dual matroid to the complement of a single point set and the contraction of the dual matroid to the complement of the equivalence class of this point. We will do more works in combining rough sets and matroids.
Acknowledgments
===============
This work is supported in part by the National Natural Science Foundation of China under Grant No. 61170128, the Natural Science Foundation of Fujian Province, China, under Grant Nos. 2011J01374 and 2012J01294, and the Science and Technology Key Project of Fujian Prov-ince, China, under Grant No. 2012H0043.
[10]{} Lai, H.: Matroid theory. Higher Education Press, Beijing (2001)
Li, X., Liu, S.: Matroidal approaches to rough set theory via closure operators. International Journal of Approximate Reasoning **53** (2012) 513–527
Li, Y.: Some researches on fuzzy matroids. PhD thesis, Shaanxi Normal University (in Chinese) (2007)
Liu, G.: The transitive closures of matrices over distributive lattices. In: Granular Computing. (2006) 63–66
Liu, Q., Wang, J.: Semantic analysis of rough logical formulas based on granular computing. In: Granular Computing. (2006) 393–396
Mao, H.: The relation between matroid and concept lattice. Advance in Mathematics **35** (2006) 361–365
Nguyen, H.S., Pal, S.K., Skowron, A.: Rough sets and fuzzy sets in natural computing. Theoretical Computer Science **412** (2011) 5816–5819
Pawlak, Z.: Rough sets: theoretical aspects of reasoning about data. Kluwer Academic Publishers, Boston (1991)
Pawlak, Z.: Rough sets. International Journal of Computer and Information Sciences **11** (1982) 341–356
Qin, K., Pei, Z., Du, W.: The relationship among several knowledge reduction approaches. In: Fuzzy Systems and Knowledge Discovery. Volume 3613 of LNCS. (2005) 1232–1241
Wang, G., Liu, F.: The inconsistency in rough set based rule generation. In: Rough Sets and Current Trends in Computing. Volume 2005 of LNCS. (2000) 370–377
Wang, S., Zhu, W.: Matroidal structure of covering-based rough sets through the upper approximation number. International Journal of Granular Computing, Rough Sets and Intelligent Systems **2** (2011) 141–148
Wang, S., Zhu, W., Zhu, P.: Abstract interdependency in rough sets. Journal of Nanjing University (in Chinese) **46** (2010) 507–510
Wu, W., Zhang, W.: Rough set approximations vs. measurable spaces(in chinese). In: Granular Computing. (2006) 329–332
Yao, Y., Chen, Y.: Subsystem based generalizations of rough set approximations. In: Lecture Notes in Computer Science. Volume 3488. (2005) 210–218
Yao, Y.: Relational interpretations of neighborhood operators and rough set approximation operators. Information Sciences **111** (1998) 239–259
Yao, Y.: Constructive and algebraic methods of theory of rough sets. Information Sciences **109** (1998) 21–47
Zhong, N., Yao, Y., Ohshima, M.: Peculiarity oriented multidatabase mining. IEEE Transactions on Knowledge and Data Engineering **15** (2003) 952–960
Zhu, W., Wang, F.: Covering based granular computing for conflict analysis. In: Intelligence and Security Informatics. Volume 3975 of LNCS. (2006) 566–571
Zhu, W., Wang, S.: Matroidal approaches to generalized rough sets based on relations. International Journal of Machine Learning and Cybernetics **2** (2011) 273–279
Zhu, W., Wang, S.: Rough matroid. In: Granular Computing. (2011) 817–822
[^1]: Corresponding author. E-mail: [email protected] (William Zhu)
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In the present paper, we present some numerical methods for computing approximate solutions to some large differential linear matrix equations. In the first part of this work, we deal with differential generalized Sylvester matrix equations with full rank right-hand sides using a global Galerkin and a norm-minimization approaches. In the second part, we consider large differential Lyapunov matrix equations with low rank right-hand sides and use the extended global Arnoldi process to produce low rank approximate solutions. We give some theoretical results and present some numerical experiments.'
author:
- 'M. Hached [^1]'
- 'K. Jbilou[^2]'
date:
title: Numerical methods for differential linear matrix equations via Krylov subspace methods
---
[: 65F10]{}.\
[: Sylvester equation, Lyapunov equation, Global Arnoldi, matrix Krylov subspace, preconditioning.]{}
Introduction
============
In this work, we are interested in the numerical solution of two differential linear matrix equations. First, we consider the linear matrix differential equation with a full right-hand side $$\label{linmatrix}
\left\{
\begin{array}{l}
\dot{X}(t) = \displaystyle \sum_{i=1}^q A_i\, X(t) \, B_i+C,\\
X(t_0)=X_0,\; \; t \in [t_0, \, T_f],
\end{array}
\right.$$ where $A_i \in \mathbb{R}^{n \times n}$, $B_i \in \mathbb{R}^{p
\times p}$, $i=1,\ldots,q$, $C\, {\rm and}\, X \in \mathbb{R}^{n
\times p}$, and we assume that the right hand term $C$ is full rank. Differential Sylvester and Lyapunov matrix equations are particular cases of (\[linmatrix\]).\
The second differential matrix equation that will be considered in this paper, is the well known differential Lyapunov matrix equation with a low rank right hand side $$\label{lyap1}
\left\{
\begin{array}{l}
\dot X(t)=A\,X(t)+X(t)\,A^T+BB^T;\; (DLE) \\
X(t_0)=X_0,\; \; t \in [t_0, \, T_f],
\end{array}
\right.$$
where the matrix $ A \in \mathbb{R}^{n \times n} $ is assumed to be large, sparse and nonsingular and $ B \in
\mathbb{R}^{n \times p} $ is a full rank matrix, with $ p \ll n $. The initial condition $X_0={\widetilde Z}_0 {\widetilde Z}_0 ^T$ is assumed to be a given symmetric and positive low-rank matrix.\
The differential linear matrix equations and play an important role in many areas such as control, filter design theory, model reduction problems, differential equations and robust control problems [@abou03; @corless].\
Notice that the two linear differential matrix equations above can be reformulated as $$\label{lin1}
{\dot x}(t) = {\cal M} x(t)+ b,$$ where $x(t)=vec(X(t))$, the matrix ${\cal M}$ is given by ${\cal M} = \displaystyle \sum_{1}^q (B_i^T \otimes A_i)$ for the problem and ${\cal M} = I \otimes A +A\otimes I$ for , while the right hand side $b$ is given by $b=vev(C)$ for and $b=vev(BB^T)$ for respectively, and $vec(C)$ is the long vector obtained by stacking the columns of the matrix $C$. So we can use classical solvers for computing solutions of .\
The exact solution of is given by $$\label{lin2}
x(t) = e^{t {\cal M}} x_0 + \displaystyle \int_{t_0}^t e^{(t-\tau){\cal M}} b \, d \tau,$$ which can also be expressed as $$\label{lin3}
x(t)=t \psi_1(t {\cal M}) (b+{\cal M}x_0)+x_0,$$ where $$\label{lin4}
\psi_1(z)=\displaystyle \frac{e^z-1}{z}.$$ However, in the cases for which the matrix ${\cal M}$ is very large, this approach would not be appropriate.\
The rest of the paper is organized as follows. In Section 2, we recall the definitions of the Kronecker and the $\diamond$ products with some of their properties that will be of use in this work. In Section 3, we give a numerical method for solving the problem by using projections onto matrix Krylov subspaces, based on a Global-Galerkin orthogonality condition. In Section 4, we will be interested in the numerical solution of the Lyapunov differential matrix equation . The approximate solutions will be obtained via projection onto matrix Krylov subspaces using the extended global Arnoldi algorithm. The last section is devoted to some numerical experiments.
Preliminaries
=============
Definitions
-----------
We begin by recalling some notations that will be used in the sequel. We define the inner product $$\left<Y,Z\right>_F = tr(Y^TZ),$$ where $ tr(Y^TZ)$ denotes the trace of the matrix $Y^TZ$ such that $Y,Z \in {\hbox{ I\hskip -2pt R}}^{n\times p}$. The associated norm is the Frobenius norm denoted by $\|Z\|_F=\sqrt{\left<Z,Z\right>_F}$.\
The matrix product $A \otimes B = [a_{i,j}B]$ denotes the well known Kronecker product of the matrices $A$ and $B$ which satisfies the following properties:
1. $(A \otimes B)(C \otimes D) = (AC \otimes BD)$.
2. $(A\otimes B)^T = A^T \otimes B^T $.
3. $(A\otimes B)^{-1} = A^{-1} \otimes B^{-1} $, if $A$ and $B$ are nonsingular square matrices.
We also use the matrix product $\diamond$ defined in [@bouyouli] as follows.
Let ${\cal Z} = [Z_1,...,Z_m]$ and ${\cal W} = [W_1,..., W_l]$ be matrices of dimension $n \times mp$ and $n\times lp$ respectively, where $Z_i$ and $W_j$ $(i = 1,..., m$ $j = 1,...,l)$ are ${\hbox{ I\hskip -2pt R}}^{n\times p}$. Then the ${\hbox{ I\hskip -2pt R}}^{m\times l}$ matrix ${\cal Z}^T \diamond {\cal W}$ is defined as: $${\cal Z}^T \diamond {\cal W}= [\left< Z_i, W_j \right>]_{1\le i \le m; 1 \le j \le l}$$
A block matrix ${\cal W} = [W_1,..., W_l]$ is said to be F-orthonormal if $$\left< W_i,W_j \right>= \delta_{i,j} = \left\{\begin{array}{ccccc}
0 & if & i\neq j \\
1 & if & i= j
\end{array}\right. \; i,j=1,...,l.$$ which is equivalent to $${\cal W}^T \diamond {\cal W}=I_l.$$ The following proposition gives some properties satisfied by the above product.
\[propd\][@bouyouli] Let $A,\ B,\ C \in {\hbox{ I\hskip -2pt R}}^{n\times ps}$, $D \in {\hbox{ I\hskip -2pt R}}^{n\times n}$, $L \in {\hbox{ I\hskip -2pt R}}^{p\times p}$ and $α \in {\hbox{ I\hskip -2pt R}}$. Then we have,
1. $(A + B)^T \diamond C = A^T \diamond C + B^T \diamond C$.
2. $A^T \diamond (B + C) = A^T \diamond B + A^T \diamond C$.
3. $(\alpha A)^T \diamond C = \alpha(A^T \diamond C)$.
4. $(A^T \diamond B)^T = B^T \diamond A$.
5. $(DA)^T \diamond B = A^T \diamond (D^T B)$.
6. $A^T \diamond (B(L \otimes I_s)) = (A^T \diamond B)L.$
7. $\| A^T \diamond B \|_F \leq \| A \|_F \| B \|_F$.
Let $A$ and $V$ be $n \times n$ and $n \times p$ matrices, respectively, then the matrix (also called the global) Krylov subspace $\mathbb{K}_m(A,V)$ associated to the pair $(A,V)$ is the subspace of ${\mathbb{R}}^{n \times p}$ generated by $V, AV,\ldots,A^{m-1}V$, i.e., $$\mathbb{K}_m(A,V)=span\{V,AV,\ldots,A^{m-1}V\}.$$
In the next proposition, we recall the global QR (gQR) factorisation of an $n \times mp$ matrix $Z$. The algorithm of such a matrix factorisation is given in [@bouyouli].
[@bouyouli] Let $\mathcal{Z} = \left [\begin{array}{llll} Z_{1}, & Z_{2}, &\ldots, &Z_{m} \end{array}\right ]$ be an $n \times mp$ matrix with $Z_i \in {\hbox{ I\hskip -2pt R}}^{n \times p}$, $i=1,\ldots,m$. Then, the matrix $\mathcal{Z}$ can be factored as $${\mathcal Z} = {\mathcal Q}\,(R \otimes I_p),$$ where ${\mathcal Q}=[Q_1,\ldots,Q_m]$ is an $n \times mp$ F-orthonormal matrix satisfying ${\mathcal Q}^T \diamond {\mathcal Q}=I_m$ and $R$ is an upper triangular matrix of dimension $m \times m$.
The following proposition will be useful later.
\[prop4\][@jbiloumessaoudi] Let ${\cal V}_m=[V_1,\cdots,V_m]$, be an $n \times mp$ F-orthonormal matrix. Let $Z=[z_{i,j}]$ and $G=[g_{i,j}]$ be matrices of sizes ${m\times r}$ and ${mp\times q}$ respectively, where $r$ and $q$ are any integers. Then we have $$\|{\cal V}_m(Z\otimes I_p)\|_F= \|Z\|_F$$ and $$\|{\cal V}_m\,G\|_F\le \|G\|_F.$$
Global-based Krylov subspace methods for the problem
=====================================================
The Global-Galerkin Krylov subspace method for linear matrix differential equations {#galerkin1}
-----------------------------------------------------------------------------------
In this section, we consider the differential linear matrix equation and will present an iterative projection method to get numerical approximate solutions. Let ${\cal A}$ be the linear matrix operator defined as follows $$\begin{aligned}
\mathcal{A}:\;\mathbb{R}^{n \times p} &\longrightarrow&
\;\mathbb{R}^{n \times p}\\
X \; &\longrightarrow& \;\displaystyle \sum_{i=1}^q A_i\, X \,
B_i.\end{aligned}$$ Notice that the transpose of the operator $\mathcal{A}$ with respect to the inner product $\langle .\,,\, . \rangle_F$ is defined as the application mapping $X \in\mathbb{R}^{n \times p}$ $\mathcal{A}^T(X) =\displaystyle \sum_{i=1}^q A_i^T\, X \, B_i^T.$\
Let $V$ be any $n \times p$ matrix then we define the matrix Krylov subspace associated to the pair $({\mathcal{A}},V)$ and an integer $m$ defined by $${\cal
K}_{m}({\mathcal{A}},V)=span\{V,{\mathcal{A}}(V),\ldots,{\mathcal{A}}^{m-1}(V)\},$$ Where ${\mathcal{A}}^i(V)$ is defined recursively as ${\mathcal{A}}^i(V)={\mathcal{A}}({\mathcal{A}}^{i-1} (V))$. Notice that the matrix Krylov subspace ${\cal
K}_{m}({\mathcal{A}},V)$ is a subspace of $\mathbb{R}^{n \times
p}$, which means that if a matrix $Y$ is in ${\cal
K}_{m}({\mathcal{A}},V)$, then we have $Y=\displaystyle \sum_{i=1}^m \alpha_i {\cal A}^{i-1}(V)$ where $\alpha_i \in \mathbb{R}$, $i=1,\ldots,m$. Next, we remind the modified global Arnoldi algorithm [@jms] that allows us to construct an F-orthonormal basis $V_1,V_2,\ldots,V_m$ of the matrix Krylov subspace ${\cal K}_{m}({\mathcal{A}},V)$, i.e. $$\langle
V_i,V_j\rangle_F=\delta_{i,j},\quad\hbox{for}\quad
i,j=1,\cdots,k,$$ where $\delta_{i,j}$ denotes the classical Kronecker symbol. The algorithm is described as follows.
- [Set]{} $V_1=V/\|V\|_F$.
- [For]{} $j=1,\ldots, k$
1. ${\tilde
V}={\mathcal{A}}( V_j)$,
2. [for]{} $i=1,\ldots, j$.
1. $h_{i,j}=\langle V_i,{\tilde V}\rangle_F$
2. ${\tilde V}={\tilde
V}-h_{i,j}V_i,$
3. [EndFor]{}
3. $h_{j+1,j}=\parallel {\tilde V}
\parallel_F,$
4. $V_{j+1}={\tilde V}/h_{j+1,j}.$
- [EndFor]{}.
The matrix ${\widetilde H}_m$ denotes the $(m+1) \times m$ upper Hessenberg matrix whose nonzero entries $h_{i,j}$ are defined by Algorithm 1 and $H_m$ is the $m \times m$ matrix obtained from ${\widetilde H}_m$ by deleting its last row. The $n \times mp$ block matrix ${\cal V}_m=[V_1,V_2,\ldots,V_m]$ is F-orthonormal which means that the matrices $V_1,\ldots,V_m$ are orthonormal with respect to the scalar product $\langle \,.,.\, \rangle_F$ which is equivalent to $$\label{v1}
{\cal V}_{m}^T \diamond {\cal V}_{m} = I_p.$$ For the extended global Arnoldi algorithm, we have the following relations $$\label{Arnoldi1}
[{\mathcal{A}}(V_1),\ldots,{\mathcal{A}}(V_m)]={\cal V}_m(H_m
\otimes I_p) + E_{m+1},$$ where $E_{m+1}=h_{m+1,m}\,[0_{n \times p},\ldots,0_{n
\times p},V_{m+1}]$, and $$\label{Arnoldi2}
[{\mathcal{A}}(V_1),\ldots,{\mathcal{A}}(V_m)]={\cal
V}_{m+1}({\widetilde H}_m \otimes I_p).$$ Starting from an initial guess $X_0(t) \in \mathbb{R}^{n \times p}$ and the corresponding residual $\mathcal{R}_0(t)={\dot X}_0(t)-{\mathcal{A}}(X_0(t))-C$, at step $m$, we define the approximation $X_m$ as follows $$\label{approx1}
X_m(t)=X_0(t)+Z_m(t)\;\; {\rm with}\;\; Z_m(t) \in {\cal
K}_m({\mathcal{A}},C)$$ and $$\label{ortho1}
\mathcal{R}_m(t)={\dot X}_m(t)-{\mathcal{A}}(X_m(t))-C\, \perp_F \, {\cal
K}_{m}({\mathcal{A}},C).$$ The Galerkin condition is equivalent to $$\label{ortho2}
{\cal V}_m^T \diamond \mathcal{R}_m(t) = 0.$$ The condition can be written as $$\label{approx2}
X_m(t) = X_0(t)+\displaystyle \sum_{i=1}^m y_m^{(i)}(t) V_i =X_0(t)+ {\cal V}_m (y_m(t) \otimes I_p),$$ where $y_m(t)$ is a vector of $\mathbb{R}^m$ and $y_m^{(i)}(t)$ is the $i$-th component of $y_m(t)$. Therefore, the residual $\mathcal{R}_m(t)$ can be expressed as $$\begin{aligned}
\label{res1}
\mathcal{R}_m(t) &=& {\dot X}_m(t)-{\cal A}(X_m(t))-C\\
& = & {\dot X}_0(t)-{\cal A}(X_0(t))+{\cal V}_m (\dot{y}_m(t) \otimes I_p)-{\cal A}({\cal V}_m (y_m(t) \otimes I_p))-C\\
& = & {\cal V}_m (\dot{y}_m(t) \otimes I_p)- {\cal A}( \displaystyle \sum_{i=1}^p y_m^{(i)}(t) V_i) +{\cal R}_0(t)\\
& = & {\cal V}_m (\dot{y}_m(t) \otimes I_p)- ( \displaystyle \sum_{i=1}^p y_m^{(i)}(t) {\cal A}(V_i))+{\cal R}_0(t)\\
& =& {\cal V}_m (\dot{y}_m(t) \otimes I_p)- [{\cal A}(V_1), {\cal A}(V_2),\ldots,{\cal A}(V_m)] (y_m(t) \otimes I_p))+{\cal R}_0(t).\end{aligned}$$ Using the relation , it follows that $$\mathcal{R}_m(t)= {\cal V}_m (\dot{y}_m(t) \otimes I_p)- {\cal V}_m (H_m \otimes I_p) (y_m(t) \otimes I_p))-E_{m+1}(y_m(t) \otimes I_p)+{\cal R}_0(t).$$ On the other hand, $E_{m+1}$ can be expressed as $E_{m+1} =h_{m+1,m}V_{m+1} [0,0,\ldots,I]$ which can be written as $E_{m+1} =h_{m+1,m}V_{m+1} (\widetilde E \otimes I_p)$. Then we get a new expression of the residual given by $$\label{res2}
\mathcal{R}_m(t) = {\cal V}_m (\dot{y}_m(t) \otimes I_p)- {\cal V}_m (H_m y_m(t) \otimes I_p)-h_{m+1,m}V_{m+1} ({\widetilde E}y_m(t) \otimes I_p)+{\cal R}_0(t).$$ Using the properties of the $\diamond $ product given in Proposition \[propd\] and the fact that ${\cal V}_m^T \diamond V_{m+1}=0$, the F-orthogonality condition reduces to the low dimensional linear differential system of equations $$\label{sol11}
{\dot y}_m(t)=H_my_m(t)+c_m(t),$$ where $c_m =- {\cal V}_m^T \diamond {\cal R}_0(t)\in \mathbb{R}^m$.\
The solution of the ODE is given by $$\label{sol12}
y_m(t) = e^{t\, H_m}y_m(0) +\displaystyle \int_0^t e^{(t-\tau)H_m}\, c_m(\tau) d\tau.$$ A simple way to compute approximate solutions $x_n \approx x(t_n)$ is to use Euler’s method defined as follows $$\label{euler}
x_{n+1}=e^{hH_m}x_n+h \psi_1(h H_m)c(t_n),$$ where $t_n=nh$ and $h$ is a stepsize and the function $\psi_1$ is defined by .
In the next algorithm, we summarize the main steps of the global-Galerkin
1. Choose a tolerance $\varepsilon$ and a maximum number of Arnoldi iteration $m_{max}$.
2. Compute $\beta=||{\cal R}_0(t)||_F$, and $V_1= {\cal R}_0(t)/\beta$.
3. [For]{} $m=1$ ... $m_{max}$
1. Construct the $F$-orthonormal basis $V_1,V_2,\ldots, V_m$ by Algorithm \[alg1\].
2. Determine $y_m$ as solution of the problem the problems .
3. Compute the residual norm $||{\cal R}_m(t)||_F$.
4. If $||{\cal R}_m(t)||_F<\varepsilon$ stop, else, Goto (a).
4. [EndFor]{}.
5. Compute the approximation $X_m=X_0+{\cal V}_m ( y_m \otimes I_p)$.
Global projection methods for large differential Lyapunov equations with low-rank right-hand sides
==================================================================================================
In this section, we consider the following large scale differential Lyapunov equation . Differential Lyapunov equations play a fundamental role in many topics such as control, model reduction problems, differential equations and robust control problems [@abou03; @corless]. We notice that, as the problem is large and square, we cannot apply the methods developed in Section 3.\
The expression of the exact solution is given by $$\label{solexacte}
X(t)=e^{(t-t_0)A}X_0e^{(t-t_0)A^T}+\int_{t_0}^t e^{(t-\tau)A}\, B B^Te^{(t-\tau)A^T}\, d\tau.$$ In this section, we consider low-rank approximate solutions to the exact solution $X$ using the global or the extended global Arnoldi process [@druskin; @heyouni; @heyouni1; @simoncini].
Projecting and using the extended global Arnoldi process
--------------------------------------------------------
We will consider extended global Krylov subspaces associated to the pair $(A,B)$ and defined as follows $$\label{kry1}
\mathbb{K}_m(A,B)=span(A^{-m},\ldots,A^{-1}B,B,AB,\ldots,A^{m-1}B).$$ Notice that $$\mathbb{K}_m(A,B) = {\cal K}_{m}(A,B)\, + \, {\cal K}_m(A^{-1},A^{-1}B),$$ where ${\cal K}_m(A,B)$ is the global Krylov subspace associated to the pair $(A,B)$. To compute an F-orthonormal basis of $ \mathbb{K}_m(A,B) $, we can use the extended global Arnoldi algorithm defined as follows [@heyouni]
1. Compute the global QR decomposition: $[B, A^{-1}B]= V_1 (R \otimes I_p)$
2. [For]{} $j=1,\ldots,m$
1. Set $V_j^{(1)}$: the first $p$ columns of $V_j$ and $V_j^{(2)}$: the second $p$ columns of $V_j$,
2. Set ${\mathbb V}_j= [{\mathbb V}_{j-1}, V_j ]$ and $U=[AV_j^{(1)}, A^{-1}V_j^{(2)} ]$,
3. F-orthogonalize $U$ w.r. to ${\mathbb V}_j$ to get $V_{j+1}$, i.e.
4. [For]{} $i=1,2,\ldots,j$
1. $H_{i,j} =V_i^T \diamond U$,
2. $U=U-V_i (H_{i,j} \otimes I_p)$
5. [EndFor]{}
3. Compute the QR decomposition $U=V_{j+1} (H_{j+1,j} \otimes I_p)$
4. [EndFor]{}
If the upper triangular $2 \times 2$ matrices $H_{j+1,j}$ ($j=1,\ldots,m$) are full rank, then Algorithm \[extalgo\] computes an F-orthonormal basis of the global extended Krylov subspace $ \mathbb{K}_m(A,B)$, the obtained $n \times 2mp$ matrix $\mathbb{V}_m=[V_1,\ldots,V_m]$ is F-orthonormal $$\mathbb{V}_m^T \diamond \mathbb{V}_m =I_{2p}.$$ Let ${\mathbb T}_m= \mathbb{V}_m^T \diamond (A\mathbb{V}_m)=[T_{i,j]}]$ with $T_{i,j}=V_i^T \diamond (AV_j) \in \mathbb{R}^{2 \times 2}$, $i,j=1,\ldots,m$. Then it can be shown that ${\mathbb T}_m$ is a $2m \times 2m$ upper block Hessenberg matrix whose elements can be obtained from the matrix-coefficients $H_{i,j}$ computed by the extended global Arnoldi algorithm. Let ${\widetilde {\mathbb T}}_m = \mathbb{V}_{m+1}^T \diamond (A \mathbb{V}_m)$, then ${\mathbb T}_m$ can be obtained from ${\widetilde {\mathbb T}}_m$ by deleting the last 2 rows of ${\widetilde {\mathbb T}}_m$.\
We have the following algebraic relations [@heyouni]. $$\begin{aligned}
\label{rel2}
A\mathbb{V}_m &=& \mathbb{V}_{m+1} ({\widetilde {\mathbb T}}_m \otimes I_p)\\ & = & \mathbb{V}_{m} ({{\mathbb T}}_m
\otimes I_p) + V_{m+1} T_{m+1,m}(E_m^T \otimes I_p),
\end{aligned}$$ where $E_m^T=[0,0,\ldots,I_2]$ the matrix of the last two rows of the identity matrix $I_{2m}$.\
Let $X_m(t)$ be the desired low-rank approximate solution given as $$\label{approxL}
X_m(t) = {\mathbb V}_m (Y_m(t) \otimes I_p) {\mathbb V}_m^T,\; t\in [t_0,\,T_f],$$ where $Y_m(t) \in \mathbb{R}^{2m \times 2m}$, satisfies the Petrov-Galerkin orthogonality condition $$\label{galerkin}
{\cal V}_m^T R_m(t) {\cal V}_m =0,\; t \in [t_0,\; T_f],$$ where ${\cal R}_m(t)$ is the residual $ {\cal R}_m(t) = \displaystyle {\dot X}_m(t)-A\,X_m(t)-X_m(t)\,A^T- BB^T $. Then, from and , we obtain the low dimensional differential Lyapunov equation $$\label{lowlyap}
\displaystyle {\dot Y}_m(t)- \mathbb{T}_m\,Y_m(t)-Y_m(t)\,\mathbb{T}_m^T - B_mB_m^T=0, \; t\in [t_0,\,T_f],$$ with $ { B}_m= {\cal V}_m^T \diamond B $. Notice that $$\label{r}
[B,A^{-1}B]=V_1(R \otimes I_p),\; {\rm and}\; B_m=r_{1,1}e_1^{(2m)},$$ with $R=[r_{i,j}]$, $1\le i,j \le 2$ and $e_1^{(2m)}$ is the first vector of the canonical basis of $ \mathbb{R}^{2m}$.\
The low-dimensional differential Lyapunov equation will be solved by using some classical linear differential equations solvers.\
In order to limit the computational effort, we give an upper of the norm of the residual that will allow to stop the iterations without explicitly forming $X_m(t)$ which will be given only at the end of the process.
Let $\mathcal{R}_m(t)$ be the residual obtained at step $m$, then we have $$\label{normres}
\Vert \mathcal{R}_m(t) \Vert \le \sqrt{2}\, \Vert T_{m+1,m} E_m^T Y_m(t) \Vert_F,\; t\in [t_0,\,T_f].$$
Using and , the residual $\mathcal{R}_m(t)=\displaystyle {\dot X}_m(t)-A\,X_m(t)-X_m(t)\,A^T- BB^T $ is expressed as $$\begin{aligned}
{\mathcal R}_m(t) &= &{\mathbb V}_m ({\dot Y}_m(t) \otimes I_p) {\mathbb V}_m^T-\left[ \mathbb{V}_{m} ({{\mathbb T}}_m
\otimes I_p) + V_{m+1} (T_{m+1,m}E_m^T \otimes I_p) \right ] (Y_m(t) \otimes I_p) {\mathbb V}_m^T \\
&- &{\mathbb V}_m (Y_m(t) \otimes I_p) \left[ \mathbb{V}_{m} ({{\mathbb T}}_m
\otimes I_p) + V_{m+1} (T_{m+1,m}E_m^T \otimes I_p) \right]^T-BB^T.\end{aligned}$$ Therefore, using the fact that $Y_m$ is solution of the low dimensional differential problem , the residual can be expressed as follows $$\label{res3}
{\mathcal R}_m(t) = \mathbb{V}_{m+1}\, \left (
\left[
\begin{array}{cc}
0& Y_m(t)E_mT_{m+1,m}^T\\
T_{m+1,m} E_m^T Y_m(t) & 0
\end{array}
\right] \otimes I_p \right)\, {\mathbb V}_{m+1}^T.$$ Therefore, applying Proposition \[prop4\], we get for any $t \in [t_0,\; T_f]$ the following upper bound $$\label{resm}
\Vert {\mathcal R}_m(t) \Vert_F^2 \le 2\, \Vert T_{m+1,m} E_m^T Y_m(t) \Vert_F^2.$$
Solving the low dimensional differential Lyapunov equation
----------------------------------------------------------
We have now to solve the low dimensional differential Lyapunov equation by some integration method such as the well known Backward Differentiation Formula (BDF). We notice that BDF is especially used for the solution of stiff differential equations.\
At each time $t_k$, let $Y_{m,k}$ denote the approximation of $Y_m(t_k)$, where $Y_m$ is a solution of (\[lowlyap\]). Then, the new approximation $Y_{m,k+1}$ of $Y_m(t_{k+1})$ obtained at step $k+1$ by ↨$\l$-step BDF is defined by the implicit relation $$\label{bdf}
Y_{m,k+1} = \displaystyle \sum_{i=0}^{l-1} \alpha_i Y_{m,k-i} +h_k \beta {\mathcal F}(Y_{m,k+1}),$$ where $h_k=t_{k+1}-t_k$ is the step size, $\alpha_i$ and $\beta_i$ are the coefficients of the BDF method as listed in Table \[tab1\] and ${\mathcal F}(X)$ is given by $${\mathcal F}(Y)= {\mathbb T}_m\,Y+Y\,{\mathbb T}_m^T+\,B_m\,B_m^T.$$
$l$ $\beta$ $\alpha_0$ $\alpha_1$ $\alpha_2$
----- --------- ------------ ------------ ------------
1 1 1
2 2/3 4/3 -1/3
3 6/11 18/11 -9/11 2/11
: Coefficients of the $l$-step BDF method with $l \le 3$.[]{data-label="tab1"}
The approximate $Y_{m,k+1}$ solves the following matrix equation $$-Y_{m,k+1} +h_k\beta ({\mathbb T}_m Y_{m,k+1} + Y_{m,k+1} {\mathbb T}_m^T+ B_m B_m^T) + \displaystyle \sum_{i=0}^{l-1} \alpha_i Y_{m,k-i} = 0,$$ which can be written as the following algebraic Lyapunov matrix equation
$$\label{lyapbdf}
{\cal T}_m\, Y_{m,k+1} + \,Y_{m,k+1} {\cal T}_m^T+ {\cal B}_{m,k}\, {\cal B}_{m,k}^T =0.$$
We assume that at each time $t_k$, the approximation $Y_{m,k}$ is factorised as a low rank product $Y_{m,k}\approx Z_{m,k} {Z_{m,k}}^T$, where $Z_{m,k} \in \mathbb{R}^{n \times m_k}$, with $m_k \ll n$. In that case, the coefficient matrices appearing in are given by $${\cal T}_m= h_k\beta {\mathbb T}_m -\displaystyle \frac{1}{2}I \; \mbox{and} \; {\cal B}_{m,k+1}=[\sqrt{h_k\beta} B_m, \sqrt{\alpha_0}Z_{m,k},\ldots,\sqrt{\alpha_{l-1}} Z_{m,k+1-l}].$$ The Lyapunov matrix equation can be solved by applying direct methods based on Schur decomposition such as the Bartels-Stewart algorithm [@bartels; @gnv].\
In the following algorithm, we summarise the main steps of the extended global Arnoldi method for solving the differential Lyapunov matrix equation .
1. Inputs: $A$, $B$ a maximum number of extended Arnoldi iteration $m_max$ and a tolerance $tol$.
2. Apply the extended global Arnoldi Algorithm to the pair $(A,B)$ to get an F-orthonormal matrix $\mathbb{V}_m=[V_1,\ldots,V_m]$ and the upper block Hessenberg matrix $\mathbb{T}_m$.
3. Solve the low dimensional problem by the BDF method.
4. If $\mathcal{R}_m(t) < tol$ stop and compute the obtained approximate solution.
Using the approximation of the exponential of a matrix
------------------------------------------------------
In this subsection, we will see how to use the expression to get low rank approximate solutions. It is known [@higham; @horn] that for any square matrix $A$, we have the Cauchy’s integral representation $$\label{cauchy1}
f(A)=\displaystyle \frac{1}{2\pi i}\int_{\Gamma}^{} f(\lambda) (\lambda I -A)^{-1}d \lambda,$$ where $f$ is an analytic function on and inside a closed contour $\Gamma \subset \mathbb{C}$ that encloses the spectrum $\sigma(A)$. A very important topic consists in approximating see [@benzi; @gallopoulos; @hoc; @saad2]. $$\label{cauchy2}
f(A)B= \displaystyle \frac{1}{2\pi i}\int_{\Gamma}^{} f(\lambda) (\lambda I -A)^{-1}Bd \lambda.$$ On the other hand, using the global Arnoldi algorithm, we can show [@jms] that $$(\lambda I -A)^{-1}B \approx {\cal V}_m \left ( (\lambda I-{\cal H}_m)^{-1} \beta e_1 \otimes I_p)\, \right),$$ where ${\cal V}_m$ is the F-orthonormal matrix obtained from the global Arnoldi process applied to the pair $(A,V)$ and ${\cal H}_m ={\cal V}_m^T \diamond (A{\cal V}_m)$. Then $$\label{cauchy3}
(\lambda I -A)^{-1}B \approx \beta {\cal V}_m \left ( (\lambda I-{\cal H}_m)^{-1} \otimes I_p)\, \tilde E_1 \right),$$ where $\tilde E_1= e_1 \otimes I_p$ and $\beta = \Vert B \Vert_F$. Therefore, if the contour $\Gamma$ contains also the spectrum of ${\cal H}_m$, (which is the case for example if we choose the countour of field of values of the matrix $A$) we get $$\label{eqq0}
f(A)B \approx \beta {\cal V}_m \displaystyle \frac{1}{2\pi i}\int_{\Gamma}^{} f(\lambda) \left ( (\lambda I-{\cal H}_m)^{-1} \otimes I_p)\, \tilde E_1 \right) d \lambda,$$ which can be written as $$\label{eqq1}
f(A)B \approx \beta {\cal V}_m \displaystyle \frac{1}{2\pi i}\int_{\Gamma}^{} f(\lambda) \left ( (\lambda I-{\cal H}_m)^{-1} \otimes I_p)\, \tilde E_1 \right) d \lambda= \beta {\cal V}_m (f({\cal H}_m) \otimes I_p)\, \tilde E_1.$$ Using the fact that $\tilde E_1= e_1 \otimes I_p$, we get $$\label{eqq2}
f(A)B \approx \beta {\cal V}_m (f({\cal H}_m)e_1 \otimes I_p).$$ Notice that using some Kronecker product relations, the expression can also be written $$\label{eqq3}
f(A)B \approx \displaystyle \frac{\beta {\cal V}_m }{2\pi i}\int_{\Gamma}^{} f(\lambda) \left (\lambda I-\left [{\cal H}_m \otimes I_p \right ]^{-1} \right ) \tilde E_1 d \lambda,$$ an then $$\label{eqq4}
f(A)B \approx \beta {\cal V}_m f({\cal H}_m \otimes I_p) \tilde E_1.$$ The two expressions on the right hand sides in and are the same. Applying these results to the function $f(x)=e^x$, we get the approximation to the exponential appearing in the expression of the exact solution $$\label{cauchy5}
e^{(t-\tau)A} B \approx \beta {\cal V}_m (e^{(t-\tau){\cal H}_m} e_1 \otimes I_p).$$ Assuming that $X_0=0$, we consider approximations $X_m(t)$ to the solution as follows $$\label{approx11}
X_m(t)= \int_{t_0}^t Z_m(\tau) {Z_m(\tau)}^T\, d \tau,$$ where $$\label{approx12}
Z_m(\tau) = \beta {\cal V}_m (e^{(t-\tau){\cal H}_m} e_1 \otimes I_p).$$ Hence, from and , we get $$\label{approx3}
X_m(t)={\cal V}_m (G_m(t) \otimes I_p) {\cal V}_m^T,$$ where $$\label{approx4}
G_m(t) = \int_{t_0}^t {\widetilde G}_m(\tau) {\widetilde G}_m(\tau)^T d\tau,$$ with ${\widetilde G}_m(\tau)= \beta e^{(t-\tau){\cal H}_m} e_1$. So, to compute the approximation $X_m(t)$, we need to compute the integral which will be done by using a quadrature formula.\
The next theorem states that the matrix function $G_m$ is solution of a low dimensional differential Lyapunov equation.
The function $G_m$ defined by the relation satisfies the following differential Lyapunov equation, $$\label{low1}
{\dot G}_m(t)={\cal H}_m G_m(t) + G_m(t) {\cal H}_m^T + \beta^2 e_1 e_1^T.$$
The proof can easily be obtained by deriving the expression .
Next, we give a result that allows us the computation of the norm of the residual.
\[tres\] Let $ X_m(t) = {\cal V}_m(G_m(t) \otimes I_p){\cal V}_m^T $ be the approximation obtained at step $ m $ by the global Arnoldi method. Then the residual $ {\cal R}_m(t) $ satisfies $$\label{result2}
\parallel {\cal R}_m(t) \parallel_F \le \vert h_{m+1,m} \vert \Vert {\bar G}_m(t) \parallel_2,$$ where $ {\bar G}_m(t) $ is the last row of $ G_m(t) $.
Let $X_m(t)$ be the approximate solution given by . Then we have $$\label{pertu}
\displaystyle {\dot X}_m(t)=A\,X_m(t)+X_m(t)\,A^T+{\cal L}_m.$$ where ${\cal L}_m =BB^T- L_m-L_m^T$ with $L_m(t)=h_{m+1,m} E_m^T (G_m(t) \otimes I) {\cal V}_m^T$.\
The error $\mathcal{E}_m(t)=X(t)-X_m(t)$ satisfies the following equation $$\label{pertu2}
\displaystyle {\dot {\mathcal E}}_m(t)=A\ {\mathcal E}_m(t)+{\mathcal E}_m(t)A^T-\mathcal{R}_m(t),$$ and then $$\label{error3}
\mathcal{E}_m(t)=e^{(t-t_0)A}\mathcal{E}_{m,0}e^{(t-t_0)A^T}+\int_{t_0}^t e^{(t-\tau)A}\mathcal{R}_m(\tau)e^{(t-\tau)A^T}d\tau,\; t \in [t_0,\, T_f].$$ where $\mathcal{E}_{m,0}=\mathcal{E}_m(t_0)$.
The proof of is obtained by using the expression of the approximate solution $X_m(t)$ and the relation . The expression of the error is easily derived by extracting the initial problem from the expression of the residual ${\cal R}_m(t)$.
\[Theoerr2\] Assume that $X(t_0)=X_m(t_0)$, then we have the following upper bound $$\label{upperbound}
\parallel \mathcal{E}_m(t) \parallel \le \displaystyle \vert h_{m+1,m} \vert \, \parallel \bar G_m \parallel_{\infty} \frac{e^{2(t-t_0)\mu_2(A)}-1}{2 \mu_2(A)},\; \forall t \in [t_0,\, T_f],\\
$$ where $\mu_2(A)=\displaystyle \frac{1}{2} \lambda_{max}(A+A^T)$ is the 2-logarithmic norm and $\parallel \bar G_m \parallel_{\infty} =\displaystyle \max_{\tau \in [t_0,\, t]} \parallel \bar G_m(\tau) \parallel$ where where $ {\bar G}_m(t) $ is the last row of $ G_m(t) $.
Using the expression of $\mathcal{E}_m(t)$ and the fact that $\parallel e^{tA} \parallel \le e^{\mu_2(A)t}$, we get $$\begin{aligned}
\parallel \mathcal{E}_m(t) \parallel & \le & \parallel h_{m+1,m} \bar G_m \parallel_{\infty} \displaystyle \int_{t_0}^t e^{2(t-\tau) \mu_2(A)} d\tau\\
& \le & \vert h_{m+1,m} \vert \, \parallel \bar G_m \parallel_{\infty} e^{2t\mu_2(A)} \displaystyle \int_{t_0}^t e^{-2\tau \mu_2(A)} d\tau\\
& = & \displaystyle \vert h_{m+1,m} \vert\, \parallel \bar G_m \parallel_{\infty} \frac{e^{2(t-t_0)\mu_2(A)}-1}{2 \mu_2(A)},\end{aligned}$$ which gives the desired result.
Notice that instead of using the global Arnoldi, we can also use the extended global Arnoldi to obtain approximation to $f(A)B$. In this case we have $$\label{expext1}
e^{(t-\tau)A}B \approx (e^{(t-\tau) {\mathbb T}_m} \otimes I_p){\widetilde B}_m,$$ where ${\widetilde B}_m=r_{1,1}e^{(2m)} \otimes I_p$ given by . Then $$\label{expext2}
e^{(t-\tau)A}B \approx r_{1,1} (e^{(t-\tau) {\mathbb T}_m} e^{(2m)} \otimes I_p).$$
Therefore, all the relations stated for the global Arnoldi are still valid for the extended block Arnoldi with some variations. From the numerical point of view, the extended global Arnoldi methodis faster than global Arnoldi.
Conclusion
==========
We presented in the present paper different new approaches for computing approximate solutions to large scale differential differential matrix equations. These approaches are based on projection onto matrix Krylov subspaces using the globlal and the extended global Arnoldi algorithms. For problems with full rank right hand sides, the problem reduces to the computation of solutions of differential linear systems of equations by classical methods. In the second part of this work, we considered a differential Lyapunov matrix equation with a decomposed low rank hand sides. The initial problem was projected onto matrix Krylov subspaces to get low dimensional differential Lyapunov equation that is solved by the classical BDF methods. Numerical experiments will be provided to show that both methods are promising for large-scale problems
[99]{} , [*Matrix Riccati Equations in Control and Sytems Theory*]{}, in Systems & Control Foundations & Applications, Birkhauser, (2003).
, [*Algorithm 432: Solution of the matrix equation AX+XB=C*]{}, Circ. Syst. Signal Proc., 13 (1972), 820–826. , [*Decay rates and O(n) algorithms for approximating functions of sparse matrices*]{}, Electr. Trans. Numer. Anal., 28 (2007), pp. 16–-39.
, A note on the numerical approximate solutions for generalized Sylvester matrix equations with applications, Applied Mathematics and Computation, 206(2) (2008) 687–694. R. Bouyouli, K. Jbilou, R. Sadaka and H. Sadok, Convergence properties of some block Krylov subspace methods for multiple linear systems. J. Comput. Appl. Math., 196 (2006) 498–511.
, [ Computational Aspects of Linear Control]{}, Kluwer, Dordrecht, 2002.
, [ Linear systems and control - An operator perspective]{}, Pure and Applied Mathematics. Marcel Dekker, New York-Basel, 2003.
, [ Numerical Methods for Linear Control Systems Design and Analysis]{}, Elsevier Academic Press, (2003).
, [ Extended Krylov subspaces: approximation of the matrix square root and related functions]{}, SIAM J. Matrix Anal. Appl., 19(3) (1998), 755–771.
, [Efficient solution of parabolic equations by Krylov approximation methods]{}, SIAM J. Sci. Statist. Comput., 13 (1992), 1236–1264.
, IEEC Trans. Autom. Contr., AC-24(1979), pp. 909-913.
, [ Extended Arnoldi methods for large Sylvester matrix equations]{}, App. Num. Math., 60(11) (2010) 1171–1182.
, [ An extended block Arnoldi algorithm for large-scale solutions of the continuous-time algebraic Riccati equation]{}, Elect. Trans. Num. Anal., 33(2009) 53–62.
, Functions of Matrices, Theory and Computation, SIAM, Philadelphia, 2008.
, On Krylov subspace approximations to the matrix exponential operator, SIAM J. Numer. Anal., 34 (1997), 1911–1925. , Topics in matrix analysis, Cambridge University Press, 1991.
, Appl. Num. Math 31(1999), 49–63.
, A computational method for symmetric Stein matrix equations. In Numerical Linear Algebra in Signals, Systems and Control, 80(2011), 295–311.
, [*Principal component analysis in linear systems: controllability, observability and model reduction*]{}, IEEE Trans. Automatic Contr., AC-26 (1981) 17–32.
, [*Numerical solution of large Lyapunov equations*]{}, in Signal Processing, Scattering, Operator Theory and Numerical Methods. Proceedings of the international symposium MTNS-89, vol. 3, M.A. Kaashoek, J.H. van Schuppen and A.C. Ran, eds., Boston, 1990, Birkhauser, pp. 503–511.
, [*Analysis of some Krylov subspace approximations to the matrix exponential operator*]{}, SIAM J. Numer. Anal., 29 (1992), 209–228.
, [*A new iterative method for solving large-scale Lyapunov matrix equations*]{}, SIAM J. Sci. Comp., 29(3) (2007), 1268–1288.
[^1]: Laboratoire P. Painlevé UMR 8524, UFR de Mathématiques, Université des Sciences et Technologies de Lille, IUT A, Rue de la Recherche, BP 179, 59653 Villeneuve d’Ascq Cedex, France, [email protected]
[^2]: L.M.P.A, Université du Littoral Côte d’Opale, 50 rue F. Buisson BP. 699, F-62228 Calais Cedex, France, [email protected]
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
In the framework of Kontsevich–Zagier periods, we derive integral representations for weight-$k$ automorphic Green’s functions invariant under modular transformations in $\varGamma_0(N)$ ($N\in\mathbb Z_{\geq1} $), provided that there are no cusp forms on the respective Hecke congruence groups with an even integer weight $k\geq4$. These Kontsevich–Zagier integral representations for automorphic Green’s functions give explicit formulae for certain Eichler–Shimura maps connecting Eichler cohomology to Maa[ß]{} cusp forms. We construct integral representations for weight-4 Gross–Zagier renormalized Green’s functions (automorphic self-energy) from limit scenarios of the respective Kontsevich–Zagier integrals. We reduce the weight-4 automorphic self-energy on $ X_0(4)(\mathbb C)=\varGamma_0(4)\backslash\mathfrak H^*$ to an explicit form, which supports an algebraicity conjecture of Gross and Zagier.\
\
*Keywords*: Kontsevich–Zagier periods, automorphic Green’s functions, Ramanujan’s alternative base theory, Jacobi elliptic functions, Eichler–Shimura maps, Gross–Zagier renormalization\
\
*Subject Classification (AMS 2010)*: 11F03, 11F37, 11F67, 11G07, 11M06, 11Y70, 33B15, 33C05, 33C20, 33C75, 33C80, 33E05, 33E20, 33E30
address: 'Program in Applied and Computational Mathematics, Princeton University, Princeton, NJ 08544, USA'
author:
- Yajun Zhou
bibliography:
- 'AGF.bib'
title: |
Kontsevich–Zagier Integrals for Automorphic Green’s Functions.\
I
---
Introduction
============
Background and Motivations\[subsec:background\]
-----------------------------------------------
Let $ \varGamma$ be a congruence subgroup of the modular group $ SL(2,\mathbb Z):=\left\{ \left.\left(\begin{smallmatrix}a&b\\c&d\end{smallmatrix}\right)\right| a,b,c,d\in\mathbb Z;ad-bc=1\right\}$ with projective counterpart $ \overline{\varGamma}=\varGamma/Z(\varGamma)$, where $ Z(\varGamma):=\{\hat\zeta\in\varGamma|\hat \zeta\hat\gamma=\hat\gamma\hat\zeta,\forall\hat\gamma\in\varGamma\}$ is the center of the group $ \varGamma$. For an even number $k>2$, the automorphic Green’s function of weight $k$ on the projective congruence subgroup $ \overline{\varGamma}$ is explicitly given as $$\begin{aligned}
G^{\mathfrak H/\overline {\varGamma}}_{k/2}(z_1,z_2):=-\frac{2}{[\varGamma:\overline{\varGamma}]}\sum_{\hat \gamma\in\varGamma}Q_{\frac{k}{2}-1}
\left( 1+\frac{\vert z_{1} -\hat \gamma z_2\vert ^{2}}{2\operatorname{Im}z_1\operatorname{Im}(\hat\gamma z_2)} \right)=-2\sum_{\hat \gamma\in\overline{\varGamma}}Q_{\frac{k}{2}-1}
\left( 1+\frac{\vert z_{1} -\hat \gamma z_2\vert ^{2}}{2\operatorname{Im}z_1\operatorname{Im}(\hat\gamma z_2)} \right),\label{eq:auto_Green_defn_Q_nu}\end{aligned}$$where $ \hat \gamma z:=\frac{az+b}{cz+d}$ for any transformation $\hat \gamma={\left(\begin{smallmatrix}a&b\\c& d\end{smallmatrix}\right)}$, and $Q_\nu $ is the Legendre function of the second kind defined by the Laplace integral$$\begin{aligned}
\label{eq:Q_nu_Laplace_int}Q_{\nu}(t):=\int_0^\infty\frac{\operatorname{d}u}{(t+\sqrt{t^2-1}\cosh u)^{\nu+1}},\quad t>1,\nu>-1.\end{aligned}$$The function $ G^{\mathfrak H/\overline {\varGamma}}_{k/2}(z_1,z_2)$ is bi-$ \varGamma$-invariant: $ G^{\mathfrak H/\overline {\varGamma}}_{k/2}(z_1,z_2)=G^{\mathfrak H/\overline {\varGamma}}_{k/2}(\hat \gamma z_1,\hat \gamma' z_2),\forall\hat \gamma,\hat\gamma'\in\varGamma$ (thus referred to as “automorphic”), and serves as the propagator of a Maa[ß]{} wave equation on the Riemann surface $ (\mathfrak H\cup\mathbb Q\cup\{i\infty\})\backslash\varGamma$ (hence the name “Green’s function”). Here in Eq. \[eq:auto\_Green\_defn\_Q\_nu\], we have adopted the normalization in [@GrossZagier1985 p. 207], [@GrossZagierI pp. 238–239] and [@GrossZagierII p. 544]; the same function has also been defined in [@Hejhal1983 §6.6] and [@IwaniecGSM53 §5], up to different normalizing constants. In the automorphic Green’s function, the arguments $ z_1$ and $ z_2$ both reside in the upper half-plane $ \mathfrak H=\{z\in\mathbb C|\operatorname{Im}z>0\}$. When the automorphic Green’s function on $ \overline{\varGamma}$ assumes a finite value, its arguments are not equivalent per any transformation in the respective symmetry group: $ z_1\notin \varGamma z_2$. In view of the $ \varGamma$-invariance, the automorphic Green’s function $G^{\mathfrak H/\overline {\varGamma}}_{k/2}(z_1,z_2)$ is effectively defined on the “off-diagonal” points in the Cartesian product of two orbit spaces $(\varGamma\backslash\mathfrak H)\times(\varGamma\backslash\mathfrak H) $. The notation $G^{\mathfrak H/\overline {\varGamma}}_{k/2}$ with superscript $\mathfrak H/\overline {\varGamma} $ (as opposed to $\overline \varGamma\backslash\mathfrak H $) is a matter of historical convention.
The automorphic Green’s function $ G^{\mathfrak H/\overline {\varGamma}}_{k/2}(z_1,z_2),z_1\notin \varGamma z_2$ has intriguing behavior when its arguments are CM points, *i.e.* when $ \operatorname{Re}z_1$, $\operatorname{Re}z_2$, $(\operatorname{Im}z_1)^2$, $(\operatorname{Im}z_2)^2$ are all rational numbers. In the absence of holomorphic cusp forms of even weights $k\geq 4$ on Hecke congruence groups $ \varGamma=\varGamma_0(N):=\left\{ \left.\left(\begin{smallmatrix}a&b\\c&d\end{smallmatrix}\right)\right| a,b,c,d\in\mathbb Z;ad-bc=1;c\equiv0\pmod N\right\}$ of levels $N\in\mathbb Z_{\geq1}$ (*i.e.* $\dim\mathcal S_k(\varGamma)=\dim\mathcal S_k(\varGamma_0(N))=0$), it has been postulated that the values of the automorphic Green’s function at CM points are expressible in terms of the logarithms of certain algebraic numbers (see [@GrossZagierI p. 317] and [@GrossZagierII p. 556]):
[align\*]{}&\
az\^2+bz+c=&a’z’\^2+b’z’+c’=0 a,b,c,a’,b’,c’.
This is known as (the “cusp-form-free version” of) the Gross–Kohnen–Zagier algebraicity conjecture. It has also been speculated that the minimal polynomial for the (conjecturally) algebraic number $ \exp \left[(\operatorname{Im}z\operatorname{Im}z')^{(k-2)/2}G^{\mathfrak H/\overline {\varGamma}}_{k/2}(z,z')\right]$ always has solvable Galois group. For situations where there are cusp forms of even weights $k\geq4$ (*i.e.* $\dim\mathcal S_k(\varGamma)=\dim\mathcal S_k(\varGamma_0(N))>0$), a more technical algebraicity conjecture has been formulated for certain linear combinations of automorphic Green’s functions that are modified by Hecke operators [@GrossZagierI; @GrossZagierII]. In [@GrossZagierI p. 317], Gross and Zagier originally formulated an algebraicity conjecture for $ \dim\mathcal S_k(\varGamma_0(N))\geq0$ with the additional constraint that the CM points $ z$ and $z'$ share the same discriminant, that is, $ b^2-4ac=b'^2-4a'c'$ for the coefficients of their respective minimal polynomials. Such a restricted version of the Gross–Kohnen–Zagier algebraicity conjecture [@GrossZagierII p. 556] is also termed as the Gross–Zagier algebraicity conjecture.
The Gross–Kohnen–Zagier algebraicity conjecture has triggered a series of critical developments in arithmetic algebraic geometry and automorphic function theory during the past few decades, among which the Chow group method and the Borcherds lift approach have produced proofs of the conjecture in various particular scenarios. In the cases where the discriminants for the two CM arguments coincide [@GrossZagierI p. 317, Conjecture (4.4)], S.-W. Zhang has interpreted the automorphic Green’s function on $\varGamma_0(N)$ in terms of CM-Chow-cycles and has demonstrated the truth of the Gross–Zagier algebraicity conjecture for non-degenerate height pairings on CM-cycles [@SWZhang1997]. In his PhD thesis [@MellitThesis], A. Mellit has proposed an extension of Zhang’s method [@SWZhang1997], leading to a confirmation of the statement $\exp( G_2^{\mathfrak H/PSL(2,\mathbb Z)}(z,i)\operatorname{Im}z)\in\overline {\mathbb Q}$ for all the CM points $z$. Building upon the work of R. E. Borcherds [@Borcherds; @Borcherds1999], J. H. Bruiner and T. H. Yang have furnished explicit formulae for certain special CM values of automorphic Green’s functions associated with Hilbert modular groups [@BruinierYang2006]. M. S. Viazovska has adopted the Borcherds lift [@Borcherds] to verify the algebraicity conjecture on the full modular group $ PSL(2,\mathbb Z)$ for a pair of points $ z_1,z_2\in\mathbb Q(i\sqrt{|D|})$ belonging to the same imaginary quadratic field [@Viazovska2011; @Viazovska2012], along with explicit factorization formulae [@Viazovska2012] for the algebraic number in question, up to valuation at ramified primes in the ring of integers for $\mathbb Q(i\sqrt{|D|}) $. These theoretical developments accommodate to generic combinations of weights $k$ and groups $ \varGamma$ (allowing the presence of cusp forms $ \dim\mathcal S_k(\varGamma)\geq0$), but the algebraicity results for automorphic Green’s functions are established only for restrictive types of CM points.
In this series of works, we shall access the values of automorphic Green’s functions at arbitrary CM points via Kontsevich–Zagier periods [@KontsevichZagier] and elliptic function theory. In Part I, we first construct integral representations for automorphic Green’s functions fulfilling the cusp-form-free condition $ \dim\mathcal S_k(\varGamma_0(N))=0$, in the spirit of Kontsevich and Zagier; we then explicitly compute the Kontsevich–Zagier integrals for automorphic Green’s functions in several reducible scenarios.
Notations and Statement of Results
----------------------------------
Before stating the main results of Part I, we fix the notations and terminologies for certain arithmetic functions and give a brief overview of their analytic and algebraic properties.
For $ z\in\mathfrak H$, we write $$\begin{aligned}
\left\{\begin{array}{r@{\;:=\;}l}E_4(z)&\dfrac{45}{\pi^{4}}{ \mathop{ \raisebox
{-1.0\depthofsumsign+1\depthofsumsign}
{\scalebox
{1.0}
{$\displaystyle\sum$} }
}
}\limits_{\substack{m,n\in \mathbb Z\\m^2+n^2\neq0}}\dfrac{1}{(m+nz)^{4}}=1+240{ \mathop{ \raisebox
{-1.0\depthofsumsign+1\depthofsumsign}
{\scalebox
{1.0}
{$\displaystyle\sum$} }
}
}\limits_{n=1}^\infty\dfrac{n^3e^{2\pi inz}}{1-e^{2\pi inz}}\\E_6(z)&\dfrac{945}{2\pi^{6}}{ \mathop{ \raisebox
{-1.0\depthofsumsign+1\depthofsumsign}
{\scalebox
{1.0}
{$\displaystyle\sum$} }
}
}\limits_{\substack{m,n\in \mathbb Z\\m^2+n^2\neq0}}\dfrac{1}{(m+nz)^{6}}=1-504{ \mathop{ \raisebox
{-1.0\depthofsumsign+1\depthofsumsign}
{\scalebox
{1.0}
{$\displaystyle\sum$} }
}
}\limits_{n=1}^\infty\dfrac{n^5e^{2\pi inz}}{1-e^{2\pi inz}}\end{array}\right.\quad \label{eq:E4_E6_defn}\end{aligned}$$for the Eisenstein series, and $$\begin{aligned}
\Delta(z):=\frac{[E_4(z)]^{3}-[E_6(z)]^2}{1728}=e^{2\pi iz}\left[\prod_{n=1}^\infty(1-e^{2\pi inz})\right]^{24}\label{eq:Delta_defn}\end{aligned}$$for the modular discriminant of Weierstra[ß]{}. (Some authors may define $ E_4$, $ E_6$ and $ \Delta$ with other normalizing constants, making each of them differ from our conventions in Eqs. \[eq:E4\_E6\_defn\] and \[eq:Delta\_defn\] by a rational number times a certain integer power of $ \pi$.) Writing $ \eta(z)=e^{\pi iz/12}\prod_{n=1}^\infty(1-e^{2\pi inz}),z\in\mathfrak H$ for the Dedekind eta function, we may recast Eq. \[eq:Delta\_defn\] into $ \Delta(z)=[\eta(z)]^{24}$.
In this work, we set $ \Delta'(z):=\partial \Delta(z)/\partial z$, and use the *ad hoc* notation for the “weight-2 Eisenstein series” as follows:$$\begin{aligned}
E_2^{\vphantom{*}}(z)=\frac{1}{2\pi i}\left[ \frac{\Delta'(z)}{\Delta(z)} -\frac{6i}{\operatorname{Im}z}\right],\quad E_2^*(z)=\frac{1}{2\pi i} \frac{\Delta'(z)}{\Delta(z)}=1-24\sum_{n=1}^\infty\frac{ne^{2\pi inz}}{1-e^{2\pi inz}}.\label{eq:E2_defn}\end{aligned}$$Accordingly, one has the transformation laws under $ SL(2,\mathbb Z) $ [@Schoeneberg p. 68]: $$\begin{aligned}
E_2^*(\hat\gamma z)=(cz+d)^2E^*_2(z)-\frac{6ic}{\pi}(cz+d),\quad \forall\hat\gamma=\begin{pmatrix}a & b \\
c & d \\
\end{pmatrix}\in SL(2,\mathbb Z)\end{aligned}$$ and under $ \varGamma_0(N)$ [@RN3 p. 484]: $$\begin{aligned}
NE^*_2(N(\hat\gamma z))-E_2^*(\hat\gamma z)=(cz+d)^2[NE^*_2(N z)- E_2^*( z)],\quad \forall\hat\gamma=\begin{pmatrix}a & b \\
c & d \\
\end{pmatrix}\in \varGamma_0(N),N\in\mathbb Z_{>0}.\label{eq:E2_star_HeckeN_transf}\end{aligned}$$ We reserve the notation $ E_2$ (*without* an asterisk) for a non-holomorphic function (see [@RLN2 Chap. 11]) so that the functions $ E_2$, $ E_4$ and $ E_6$ follow similar transformation laws [@Schoeneberg p. 67]:$$\begin{aligned}
E_{2}(\hat \gamma z)=(cz+d)^2 E_2(z),\quad E_{4}(\hat \gamma z)=(cz+d)^4 E_4(z),\quad E_{6}(\hat \gamma z)=(cz+d)^6 E_6(z)\label{eq:E2E4E6_mod_transf}\end{aligned}$$for $ z\in\mathfrak H$ and $\hat\gamma={\left(\begin{smallmatrix}a&b\\c& d\end{smallmatrix}\right)}\in SL(2,\mathbb Z) $. (Some authors may switch the nomenclature for $ E_2^{\vphantom{*}}$ and $ E_2^*$, placing priority on holomorphy.) The following Eisenstein series are descendants of $ E_4$ and $ E_6$:$$\begin{aligned}
\left\{\begin{array}{r@{\;=\;}l}
E_{8}(z) & [E_4(z)]^2=1+480{ \mathop{ \raisebox
{-1.0\depthofsumsign+1\depthofsumsign}
{\scalebox
{1.0}
{$\displaystyle\sum$} }
}
}\limits_{n=1}^{\infty}\dfrac{n^{7}e^{2\pi inz}}{1-e^{2\pi inz}},\\[12pt]
E_{10}(z) & E_4(z)E_6(z)=1-264{ \mathop{ \raisebox
{-1.0\depthofsumsign+1\depthofsumsign}
{\scalebox
{1.0}
{$\displaystyle\sum$} }
}
}\limits_{n=1}^\infty\dfrac{n^{9}e^{2\pi inz}}{1-e^{2\pi inz}}, \\[12pt]
E_{14}(z) & [E_4(z)]^{2}E_6(z)=1-24{ \mathop{ \raisebox
{-1.0\depthofsumsign+1\depthofsumsign}
{\scalebox
{1.0}
{$\displaystyle\sum$} }
}
}\limits_{n=1}^\infty\dfrac{n^{13}e^{2\pi inz}}{1-e^{2\pi inz}}. \\
\end{array}\right.
\label{eq:E8E10E14_defn}\end{aligned}$$
We normalize Klein’s $j$-invariant as$$\begin{aligned}
j(z):=\frac{1728[E_4(z)]^{3}}{[E_4(z)]^{3}-[E_6(z)]^2}=\frac{[E_4(z)]^{3}}{\Delta(z)},\end{aligned}$$so that $ j(i)=1728$ and $ j(e^{\pi i/3})=0$. If $N$ is a positive integer and $z\in\mathfrak H$ is a CM point (such that $ [\mathbb Q(z):\mathbb Q]=2$), then each of the following expressions$$\begin{aligned}
j(z),\quad \frac{E_2(Nz)}{E_2(z)},\quad \frac{E_4(Nz)}{E_4(z)},\quad \frac{E_6(Nz)}{E_6(z)},\quad \frac{\Delta(Nz)}{\Delta(z)},\quad \frac{[E_2(z)]^{2}}{E_4(z)},\quad \frac{[E_2(z)]^{3}}{E_6 (z)}\label{eq:fns_alg_val_at_CM_pts}\end{aligned}$$ either is infinity or represents an algebraic number solvable in radical form [@Zagier2008Mod123 p. 86, Proposition 27]. Furthermore, these algebraic numbers generate abelian extensions of the imaginary quadratic field $ \mathbb Q(z)$. The rationale behind the aforementioned assertions on algebraicity and field extensions is sketched in Appendix \[app:algebraicity\].
The main results of this article are summarized in the following two theorems. In what follows, the real and imaginary parts of a point $ z$ (sometimes marked with an additional superscript or subscript) in the upper half-plane $ \mathfrak H$ are abbreviated as $ x\equiv \operatorname{Re}z$ and $ y\equiv \operatorname{Im}z$ (sometimes marked with a corresponding superscript or subscript, as is applicable to $ z$).
\[thm:KZ\_int\_repns\]
1. \[itm:KZ\_b\] For any pair of non-equivalent points $ z$ and $z'$ (such that $ z\notin\varGamma_0(N)z'$) with $ N\in\{2,3,4\}$, the following integral identity holds$$\begin{aligned}
G_2^{\mathfrak H/\overline{\varGamma}_0(N)}(z,z')={}&\frac{4\pi^{2}}{y'}\operatorname{Re}\int_{z'}^{i\infty} \alpha_N(\zeta)[1-\alpha_N(\zeta)][NE_{2}(N\zeta)-E_2(\zeta)]^{2}\varrho^{\mathfrak H/\overline\varGamma_0(N)}_2(\zeta,z)\frac{(\zeta-z')(\zeta-\overline{z'})\operatorname{d}\zeta}{i(N-1)^{2}}\notag\\&-\frac{4\pi^{2}}{y'}\operatorname{Re}\int_{0}^{i\infty} \alpha_N(\zeta)[1-\alpha_N(\zeta)][NE_{2}(N\zeta)-E_2(\zeta)]^{2}\varrho^{\mathfrak H/\overline\varGamma_0(N)}_2(\zeta,z)\frac{\zeta^{2}\operatorname{d}\zeta}{i(N-1)^{2}},\label{eq:G2HeckeN_unified}\end{aligned}$$where $$\begin{aligned}
\alpha_N(z):=
\left\{1+\frac{1}{N^{6/(N-1)}}\left[ \frac{\eta(z)}{\eta(Nz)} \right]^{24/(N-1)}\right\}^{-1}\end{aligned}$$is $ \varGamma_0(N)$-invariant for $ N\in\{2,3,4\}$, and $$\begin{aligned}
\varrho^{\mathfrak H/\overline\varGamma_0(N)}_2(\zeta,z)={}&-(N-1)\frac{y}{2\pi }\frac{\partial}{\partial y}\left\{ \frac{1}{y} \frac{1}{\alpha_{N}(\zeta)-\alpha_{N}(z)}\frac{1}{NE_{2}(Nz)-E_{2}(z)}\right\}.\end{aligned}$$
2. \[itm:KZ\_c\] Suppose that $z\notin\varGamma_0(2)z'$, then we have the following integral formula:$$\begin{aligned}
G_3^{\mathfrak H/\overline{\varGamma}_0(2)}(z,z')={}&\frac{4\pi^{2}}{(y')^{2}}\operatorname{Re}\int_{z'}^{i\infty} \alpha_2(\zeta)[1-\alpha_2(\zeta)][2E_{2}(2\zeta)-E_2(\zeta)]^{3}\varrho^{\mathfrak H/\overline\varGamma_0(2)}_3(\zeta,z)\frac{(\zeta-z')^{2}(\zeta-\overline{z'})^{2}\operatorname{d}\zeta}{i}\notag\\&-\frac{4\pi^{2}}{(y')^{2}}\operatorname{Re}\int_{0}^{i\infty} \alpha_2(\zeta)[1-\alpha_2(\zeta)][2E_{2}(2\zeta)-E_2(\zeta)]^{3}\varrho^{\mathfrak H/\overline\varGamma_0(2)}_3(\zeta,z)\frac{\zeta^{4}\operatorname{d}\zeta}{i},\label{eq:G3Hecke2_KZ_int}\end{aligned}$$where$$\begin{aligned}
\varrho^{\mathfrak H/\overline\varGamma_0(2)}_3(\zeta,z)=\frac{ y^2}{8\pi}\left( \frac{\partial}{\partial y}\frac{1}{y} \right)^2\left\{ \frac{1}{\alpha_{2}(\zeta)-\alpha_2(z)}\frac{1}{[2E_{2}(2z)-E_2(z)]^2} \right\}.\label{eq:rho_3_Hecke2}\end{aligned}$$
3. \[itm:KZ\_a\] The following integral representations hold for weights $k\in\{4,6,8,10,14\}$ and $ j(z)\neq j(z')$:$$\begin{aligned}
G_{k/2}^{\mathfrak H/PSL(2,\mathbb Z)}(z,z')={}&\frac{1728\pi^{2}}{ (y')^{(k-2)/2}}\operatorname{Re}\int_{z'}^{i\infty} \frac{E_{k}(\zeta)}{j(\zeta)}\rho^{\mathfrak H/PSL(2,\mathbb Z)}_{k/2}(\zeta,z)\frac{(\zeta-z')^{(k-2)/2}(\zeta-\overline{z'})^{(k-2)/2}\operatorname{d}\zeta}{i}\notag\\&-\frac{1728\pi^{2}}{ (y')^{(k-2)/2}}\operatorname{Re}\int_{0}^{i\infty} \frac{E_{k}(\zeta)}{j(\zeta)}\rho^{\mathfrak H/PSL(2,\mathbb Z)}_{k/2}(\zeta,z)\frac{\zeta^{k-2}\operatorname{d}\zeta}{i}\label{eqn:GkPSL2Z}\end{aligned}$$ where$$\begin{aligned}
\label{eqn:rho_k_PSL2Z}
\rho^{\mathfrak H/PSL(2,\mathbb Z)}_{k/2}(\zeta,z)={}&\frac{(-1)^{k/2}}{2^{(k-4)/2}}\frac{1}{\left( \frac{k-2}{2} \right)!}\frac{ y^{(k-2)/2}}{864 \pi}\left( \frac{\partial}{\partial y}\frac{1}{y} \right)^{(k-2)/2}\left[ \frac{j(\zeta)j(z)}{j(\zeta)-j(z)}\frac{{E_{6}(z)}}{E_{4}(z)E_{k}(z)}\right].\end{aligned}$$Here in Eq. \[eqn:rho\_k\_PSL2Z\], the Eisenstein series follow the definitions in Eqs. \[eq:E4\_E6\_defn\] and \[eq:E8E10E14\_defn\]; it is also understood that $ j(z)E_6(z)/[E_4(z)E_k(z)]=[E_4(z)]^2 E_6(z)/[\Delta(z)E_k(z)]$ extends to a smooth function in $ z\in\mathfrak H$, so that $ \rho^{\mathfrak H/PSL(2,\mathbb Z)}_{k/2}(\zeta,z)$ is well-behaved whenever $ j(\zeta)\neq j(z)$.
<!-- -->
1. \[itm:GZ\_rn\_a\] We have the following special values of weight-4 Gross–Zagier renormalized Green’s functions (automorphic self-energies):[$$\begin{aligned}
G_2^{\mathfrak H/PSL(2,\mathbb Z)}(i):={}&\lim_{z\to i}\left[ G_2^{\mathfrak H/PSL(2,\mathbb Z)}(z,i) -4\log(2\pi|z-i|)\right]-\frac{2\log|\Delta(i)|}{3}=-4(\log2+\log3);\\G_2^{\mathfrak H/\overline\varGamma_{0}(2)}\left( \frac{i}{\sqrt{2}} \right):={}&\lim_{z\to i/\sqrt{2}}\left[ G_2^{\mathfrak H/\overline\varGamma_{0}(2)}\left(z,\frac{i}{\sqrt{2}}\right) -2\log\left(2\pi\left\vert z-\frac{i}{\sqrt{2}}\right\vert\right)\right]-\frac{\log|\Delta(i/\sqrt{2})|}{3}=-3\log2;\\G_2^{\mathfrak H/\overline\varGamma_{0}(3)}\left( \frac{i}{\sqrt{3}} \right):={}&\lim_{z\to i/\sqrt{3}}\left[ G_2^{\mathfrak H/\overline\varGamma_{0}(3)}\left(z,\frac{i}{\sqrt{3}}\right) -2\log\left(2\pi\left\vert z-\frac{i}{\sqrt{3}}\right\vert\right)\right]-\frac{\log|\Delta(i/\sqrt{3})|}{3}=-2\log\frac{3}{\sqrt[3]{4}};\\G_2^{\mathfrak H/\overline\varGamma_{0}(4)}\left( \frac{i}{\sqrt{4}} \right):={}&\lim_{z\to i/\sqrt{4}}\left[ G_2^{\mathfrak H/\overline\varGamma_{0}(4)}\left(z,\frac{i}{\sqrt{4}}\right) -2\log\left(2\pi\left\vert z-\frac{i}{\sqrt{4}}\right\vert\right)\right]-\frac{\log|\Delta(i/\sqrt{4})|}{3}=-\log2;\end{aligned}$$$$\begin{aligned}
G_2^{\mathfrak H/PSL(2,\mathbb Z)}\left(\frac{1+i\sqrt{3}}{2}\right):={}&\lim_{z\to\frac{1+ i\sqrt{3}}{2}}\left[ G_2^{\mathfrak H/PSL(2,\mathbb Z)}\left(z,\frac{1+i\sqrt{3}}{2}\right) -6\log\left(2\pi\left\vert z-\frac{1+i\sqrt{3}}{2}\right\vert\right)\right]-\log\left\vert \Delta\left( \frac{1+i\sqrt{3}}{2} \right)\right\vert\notag\\={}&-3(2\log2+\log3);\\G_2^{\mathfrak H/\overline\varGamma_{0}(2)}\left(\frac{i-1}{2}\right):={}&\lim_{z\to \frac{i-1}{2}}\left[ G_2^{\mathfrak H/\overline\varGamma_{0}(2)}\left(z,\frac{i-1}{2}\right) -4\log\left(2\pi\left\vert z-\frac{i-1}{2}\right\vert\right)\right]-\frac{2}{3}\log\left\vert \Delta\left( \frac{i-1}{2} \right)\right\vert\notag\\={}&-4\log2;\\G_2^{\mathfrak H/\overline\varGamma_{0}(3)}\left(\frac{3+i\sqrt{3}}{6}\right):={}&\lim_{z\to \frac{3+i\sqrt{3}}{6}}\left[ G_2^{\mathfrak H/\overline\varGamma_{0}(3)}\left(z,\frac{3+i\sqrt{3}}{6}\right) -6\log\left(2\pi\left\vert z-\frac{3+i\sqrt{3}}{6}\right\vert\right)\right]-\log\left\vert \Delta\left( \frac{3+i\sqrt{3}}{6} \right)\right\vert\notag\\={}&-3\log3.\end{aligned}$$]{}
2. \[itm:GZ\_rn\_b\] The weight-4, level-4 automorphic self-energy can be evaluated in closed form:$$\begin{aligned}
G_2^{\mathfrak H/\overline{\varGamma}_0(4)}(z):=\lim_{z'\to z}\left[G_2^{\mathfrak H/\overline{\varGamma}_0(4)}(z,z')-2\log|2\pi(z-z')|\right]-\frac{\log|\Delta(z)|}{3}=-\frac{1}{3}\log\left\vert\frac{\Delta(z)}{\Delta(2z)}\right\vert,\quad \forall z\in\mathfrak H.\label{eq:G2_Hecke4_KZ_rn_statement}\end{aligned}$$ In particular, when $z$ is a CM point, Eq. \[eq:G2\_Hecke4\_KZ\_rn\_statement\] represents the logarithm of an algebraic number, solvable in radical form (see Eq. \[eq:fns\_alg\_val\_at\_CM\_pts\]), and the Galois group $ \operatorname{Gal}(\mathbb Q(z,e^{-6G_2^{\mathfrak H/\overline{\varGamma}_0(4)}(z)})/\mathbb Q(z))$ is abelian.
Plan of the Proof
-----------------
We devote the entire §\[sec:int\_repn\_AGF\] to the proof of Theorem \[thm:KZ\_int\_repns\], namely, the analytic derivation of integral representations for all the automorphic Green’s functions satisfying the cusp-form-free condition $ \dim\mathcal S_k(\varGamma_0(N))=0$. (See Appendix \[app:dimSk\_vanish\] for the reason why the claimed formulae in Theorem \[thm:KZ\_int\_repns\] have exhausted all the cusp-form-free scenarios.) Our analysis in §\[sec:int\_repn\_AGF\] is inspired by and amplified from the following laconic expression for $ G_2^{\mathfrak H/PSL(2,\mathbb Z)}(i,i\sqrt{2})$ mentioned in the thought-provoking survey of Kontsevich and Zagier [@KontsevichZagier §3.4]:$$\begin{aligned}
-\frac{G_2^{\mathfrak H/PSL(2,\mathbb Z)}(i,i\sqrt{2})}{\sqrt{2}}\equiv\frac{20G}{\pi}+1728\pi^2\int_{\sqrt{2}}^\infty\frac{E_4(iy)\Delta(iy)}{[E_6(iy)]^{2}}(y^2-2)\operatorname{d}y.\label{eq:KZ_laconic}\end{aligned}$$Here, $ G=\sum_{\ell=0}^{\infty}{(-1)^{\ell}}{(2\ell+1)^{-2}}$ is Catalan’s constant. In §\[sec:int\_repn\_AGF\], the guiding principle for our confirmation of the integral representations stated in Theorem \[thm:KZ\_int\_repns\] will be a uniqueness theorem (Lemma \[lm:spec\_AGF\]) that characterizes automorphic Green’s functions.
The analytic techniques we employ in §\[sec:int\_repn\_AGF\] may appear familiar to experts in the Eichler–Shimura theory [@Eichler1957; @Shimura1959; @Manin1973]. As seen from the statement of Theorem \[thm:KZ\_int\_repns\], all these integral representations for automorphic Green’s functions (*d’après* Kontsevich–Zagier) can be written in terms of $ \int_{z'}^{i\infty}F_{k}(\zeta,z)(\zeta-z')^{(k-2)/2}(\zeta-\overline{z'})^{(k-2)/2}\operatorname{d}\zeta$, where $ F_{k}(\zeta,z)$ is a weight-$k$ meromorphic cusp form in the variable $\zeta$. If one replaces $ F_{k}(\zeta,z)$ by a holomorphic cusp form $ F_k(\zeta)$, then the Eichler integrals $ \int_{z}^{i\infty}F_{k}(\zeta)(z-\zeta)^{k-2}\operatorname{d}\zeta$ are known to be connected to the harmonic Maa[ß]{} forms [@BringmannGuerzhoyKentOno2013; @MuehlenbruchRaji], via the Eichler–Shimura map. Our analysis in §\[sec:int\_repn\_AGF\] is tailored for meromorphic versions of Eichler integrals, and effectively presents some explicit formulae for Eichler–Shimura maps that send meromorphic versions of Eichler cohomology (integral representations of automorphic Green’s functions in Theorem \[thm:KZ\_int\_repns\]) to Maa[ß]{} cusp forms (definitions of automorphic Green’s functions as infinite series in Eq. \[eq:auto\_Green\_defn\_Q\_nu\]). As we have to cope with the meromorphic versions of Eichler integrals $ \int_{z'}^{i\infty}F_{k}(\zeta,z)(\zeta-z')^{(k-2)/2}(\zeta-\overline{z'})^{(k-2)/2}\operatorname{d}\zeta$ in our proof of Theorem \[thm:KZ\_int\_repns\], some non-trivial residue analysis is required to establish the path independence and modular invariance of the proposed integral representations of automorphic Green’s functions. In particular, the proof of parts \[itm:KZ\_c\] and \[itm:KZ\_a\] in Theorem \[thm:KZ\_int\_repns\] for weights $ k>4$ (see §\[subsec:high\_weight\_KZ\]) draws heavily on the knowledge of elliptic integrals and Ramanujan’s elliptic function theory to alternative bases (see §\[subsec:add\_form\_Legendre\_Ramanujan\]).
The major purpose of Theorem \[thm:KZ\_int\_repns\] is to construct some identities involving *integrals* and *series*, which are relevant to the Gross–Kohnen–Zagier algebraicity conjecture in the cusp-form-free scenarios (boxed equation in §\[subsec:background\]). This provides some concrete examples for an abstract theoretical framework of Eichler–Shimura isomorphisms between the Eichler cohomology groups (“*integrals*”) and the space of Maa[ß]{} cusp forms (“*series*”), which has been laid out by Mellit in [@MellitThesis Chap. 1]. In the current and subsequent instalments for this series of works, we shall treat the integral representations in Theorem \[thm:KZ\_int\_repns\] with explicit computations. This computational approach is motivated by the following observation: the Gross–Kohnen–Zagier algebraicity conjecture for automorphic Green’s functions at CM points amounts to the analytic verifications of certain proposed identities involving only Kontsevich–Zagier periods (absolutely convergent integrals of algebraic functions over algebraic domains) [@KontsevichZagier §1.1]. A “motivic Hodge conjecture” formulated by Kontsevich and Zagier [@KontsevichZagier §1.2 and §4.1] suggests the feasibility to prove any “identity involving periods” using only finitely many steps of permissible algebraic manipulations on integrals.
In §\[sec:analysis\_KZ\_int\_Part1\] of this article, we prove Theorem \[thm:GZ\_rn\], thereby accomplishing a modest goal of evaluating Kontsevich–Zagier integrals for certain weight-4 automorphic self-energies. (For a formulation of the Gross–Zagier renormalization procedure that leads to automorphic self-energies, see [@GrossZagierI Chap. II, §5] or §\[subsec:int\_repn\_G2\_GZ\_rn\] of this article.) Upon appropriate variable substitutions, the integrands we encounter in §\[sec:analysis\_KZ\_int\_Part1\] are certain products of complete elliptic integrals and elementary functions. Such integrals over elliptic integrals are sometimes referred to as *multiple elliptic integrals*, on which various developments have been reported recently [@Bailey2008; @BBGW; @BNSW2011; @Zucker2011; @Wan2012; @Zhou2013Pnu; @Zhou2013Spheres; @RogersWanZucker2013preprint; @Zhou2013Int3Pnu].
To achieve our goal stated in Theorem \[thm:GZ\_rn\], we develop a variety of analytic tools (with combinatorial, modular or geometric flavor) to simplify some Kontsevich–Zagier integrals into closed-form expressions. In §\[subsec:hypergeo\_Pnumu\], we present a glimpse into the kaleidoscope of hypergeometric transformations on Kontsevich–Zagier integrals, which lead to the proof of Theorem \[thm:GZ\_rn\]\[itm:GZ\_rn\_a\] in §\[subsec:int\_repn\_G2\_GZ\_rn\]. We prepare some multiple integral identities in §\[subsec:Ramanujan\_Jacobi\], which are inspired by spherical geometry and modular transformations of elliptic functions. These efforts culminate in the closed-form evaluation of the weight-4, level-4 automorphic self-energy in §\[subsec:G2\_Hecke4\_GZ\_rn\], which proves Theorem \[thm:GZ\_rn\]\[itm:GZ\_rn\_b\]. We note that the analytic expression in Eq. \[eq:G2\_Hecke4\_KZ\_rn\_statement\] lends evidence to a special case of the extended Gross–Zagier algebraicity conjecture [@GrossZagierI p. 317, Conjecture (4.4)].
Acknowledgements {#acknowledgements .unnumbered}
----------------
This work was partly supported by the Applied Mathematics Program within the Department of Energy (DOE) Office of Advanced Scientific Computing Research (ASCR) as part of the Collaboratory on Mathematics for Mesoscopic Modeling of Materials (CM4). The manuscript was completed during the author’s visit to Prof. Weinan E at Princeton University in 2013. The author thanks Dr. Stefan T. Patrikis (Harvard/MIT) for his inspiring lectures on “$L$-Functions and Arithmetic” at Harvard University in 2007, which formed an excellent introduction to the wonderful world of modular forms and elliptic curves. The author is indebted to Prof. Xiaowei Zhuang (Harvard) for her support of research related to elliptic functions in early 2008. The author is grateful to Prof. Bruce C. Berndt (UIUC), Prof. Michael P. Brenner (Harvard), Prof. Benedict H. Gross (Harvard), Prof. M. Lawrence Glasser (Clarkson) and Prof. Shou-Wu Zhang (Princeton) for their encouragements. The author acknowledges Prof. Bruce C. Berndt (UIUC), Dr. Qingtao Chen (ICTP, Trieste), Prof. M. Lawrence Glasser (Clarkson University), Prof. Fei Han (National University of Singapore), and an anonymous referee for their valuable help on improving the readability of this manuscript.
\[sec:int\_repn\_AGF\]Integral Representations of Certain Automorphic Green’s Functions
========================================================================================
In this section, we shall derive integral representations of automorphic Green’s functions on certain Hecke congruence groups, for even weights $ k\geq4$. The main focus of §§\[subsec:KZ\_integrals\] and \[subsec:add\_form\_Legendre\_Ramanujan\] will be the weight-4 cases (covering Theorem \[thm:KZ\_int\_repns\]\[itm:KZ\_b\] and the $ k=4$ part of Theorem \[thm:KZ\_int\_repns\]\[itm:KZ\_a\]), and the higher weight scenarios will be treated in §\[subsec:high\_weight\_KZ\] (proving Theorem \[thm:KZ\_int\_repns\]\[itm:KZ\_c\] and the $ k>4$ part of Theorem \[thm:KZ\_int\_repns\]\[itm:KZ\_a\]).
As we may recall, the Hecke congruence group of level $ N$ is defined as [@Shimura1994 Eq. 1.6.5] $$\begin{aligned}
\varGamma_0(N):=\left\{ \left.\begin{pmatrix}a & b \\
Nc & d \\
\end{pmatrix}\right|a,b,c,d\in\mathbb Z;ad-Nbc=1 \right\},\end{aligned}$$and its projective counterpart is given by $\overline {\varGamma}_0(N)=\varGamma_0(N)/\{\hat I,-\hat I\} $. The Hecke congruence group $ \varGamma_0(N)$ ($ N>1$) is normalized by the Fricke involution $ \widehat w_N:z\mapsto -1/(Nz)$, as the matrix multiplication$$\begin{aligned}
\begin{pmatrix}0&-1/\sqrt{N}\\\sqrt{N}&0 \\
\end{pmatrix}\begin{pmatrix}a&b\\cN&d \\
\end{pmatrix} =\begin{pmatrix}d&-c\\-b N&a \\
\end{pmatrix}\begin{pmatrix}0&-1/\sqrt{N}\\\sqrt{N}&0 \\
\end{pmatrix}\label{eq:Fricke_inv_norm_Hecke}\end{aligned}$$ entails the conjugation relation $\widehat w_N\varGamma_0(N)=\varGamma_0(N)\widehat w_N$. Now that the resolvent kernel of the upper half-plane has $ SL(2,\mathbb R)$-homogeneity at any pair of distinct points $ z_1,z_2\in\mathfrak H$: $$\begin{aligned}
G^{\mathfrak H}_{s}(z_1,z_2):=-2Q_{s-1}
\left( 1+\frac{\vert z_{1} - z_2\vert ^{2}}{2y_1y_2} \right)=G^{\mathfrak H}_{s}(\hat\gamma z_1,\hat \gamma z_2),\quad \forall\hat \gamma\in SL(2,\mathbb R),\label{eq:SL2_homogeneity}\end{aligned}$$one can deduce from Eqs. \[eq:Fricke\_inv\_norm\_Hecke\]–\[eq:SL2\_homogeneity\] the following identity [@GrossZagierI Chap. II, §2, Eq. 2.25]:$$\begin{aligned}
G_s^{\mathfrak H/\overline{\varGamma}_0(N)}(z,z')=G_s^{\mathfrak H/\overline{\varGamma}_0(N)}\left( -\frac{1}{Nz} ,-\frac{1}{Nz'}\right),\quad \operatorname{Re}s>1,
\label{eq:Fricke_inv_HeckeN_GZ}\end{aligned}$$whenever both sides of Eq. \[eq:Fricke\_inv\_HeckeN\_GZ\] are finite.
We will be mainly interested in weight-$k$ automorphic Green’s functions on $ \varGamma_0(N)$ when there are no cusp forms on the same Hecke congruence group with the same weight: $ \dim\mathcal S_k(\varGamma_0(N))=0$. For even weights $ k\geq4$, the solutions to the equation $ \dim\mathcal S_k(\varGamma_0(1))=\dim\mathcal S_k(SL(2,\mathbb Z))=0$ are exhausted by $ k=4,6,8,10,14$ [@DiamondShurman Theorem 3.5.2]. The following dimension formulae are standard: $\dim\mathcal S_4(\varGamma_0(2))=\dim\mathcal S_4(\varGamma_0(3))=\dim\mathcal S_4(\varGamma_0(4))=\dim\mathcal S_6(\varGamma_0(2))=0$ [@Miyake1989J1 Table A]. In fact, there are no other solutions to $\dim\mathcal S_k(\varGamma_0(N))=0 $ for even weights $ k\geq4$ and integers $ N>1$, as one can prove using lower bounds on the dimensions of the spaces of cusp forms [@Martin2005] (see Appendix \[app:dimSk\_vanish\]).
![(Adapted from Fig. 61 in [@FrickeKlein].) (*a*) Fundamental domain $ \protect\mathfrak D$ of the full modular group $ PSL(2,\protect\mathbb Z)$. The modular elliptic curve $ X_0(1)(\mathbb C)=SL(2,\mathbb Z)\backslash\mathfrak H^*$ compactifies $ Y_0(1)(\mathbb C)=SL(2,\mathbb Z)\backslash \mathfrak H$, a quotient space that identifies the corresponding sides of the boundary $\partial \protect\mathfrak D$ along the arrows. (*b*) Tessellation of the upper half-plane $ \protect\mathfrak H$ by successive translations \[generator $\hat T=\protect\left(\protect\begin{smallmatrix}1&1\\0&1\protect\end{smallmatrix}\protect\right)$\] and inversions \[generator $\hat S=\protect\left(\protect\begin{smallmatrix}0&-1\\1&0\protect\end{smallmatrix}\protect\right)$\] of the fundamental domain $\protect \mathfrak D$. Each tile is then subdivided and colored in black or grey according as the pre-image satisfies $ \operatorname{Re}z<0$ or $ \operatorname{Re}z>0$ in the fundamental domain $\protect \mathfrak D$. (*c*)–(*e*) Fundamental domains $ \mathfrak D_N$ of the projective Hecke congruence subgroups $ \overline{\varGamma}_0(N)$ where $ N\in\{2,3,4\}$. (*f*) Overlay of some horizontal translates of the fundamental domains in panels *c*–*e* on top of panel *b*. It is graphically evident that $ [\varGamma_0(1):\varGamma_0(2)]=3$, $ [\varGamma_0(1):\varGamma_0(3)]=4$ and $ [\varGamma_0(1):\varGamma_0(4)]=6$. []{data-label="fig:fundomain"}](funddom.eps){width="15cm"}
Instead of converting the infinite sum in Eq. \[eq:auto\_Green\_defn\_Q\_nu\] directly into integrals, we shall resort to an alternative characterization of automorphic Green’s functions (see [@GrossZagier1985; @SWZhang1997; @MellitThesis; @Viazovska2011]; see also [@Hejhal1983 §6.6] and [@IwaniecGSM53 §5.1] for the same function with different normalizing constants) recapitulated in the next lemma.
\[lm:spec\_AGF\]The function $ G^{\mathfrak H/\overline {\varGamma}}_{k/2}(z_1,z_2)$ (where $\operatorname{Re}k>2 $) can be prescribed as the unique solution to the following set of requirements:
1. \[itm:AGF1\]*(**Symmetry**)* For any modular transformation $\hat \gamma\in\varGamma$ and two points $ z_1,z_2\in\mathfrak H$ in the upper half-plane, one has $ G^{\mathfrak H/\overline {\varGamma}}_{k/2}(z_1,z_2)=G^{\mathfrak H/\overline {\varGamma}}_{k/2}(z_2,z_1)=G^{\mathfrak H/\overline {\varGamma}}_{k/2}(\hat \gamma z_1,z_2)=G^{\mathfrak H/\overline {\varGamma}}_{k/2}( z_1,\hat \gamma z_2)\in\mathbb C\cup\{-\infty\}$. Moreover, if $z_1\notin \varGamma z_2 :=\{\hat\gamma z_2|\hat \gamma\in\varGamma\}$, then $ G^{\mathfrak H/\overline {\varGamma}}_{k/2}(z_1,z_2)\in\mathbb C$.
2. \[itm:AGF2\] *(**Differential Equation**)* The function $ G_{k/2}^{\mathfrak H/\overline {\varGamma}}(z_1,z_2)$ is smooth on the set $\mathfrak H\times\mathfrak H\smallsetminus\{(z,\hat\gamma z)|z\in\mathfrak H,\hat \gamma\in\varGamma\} $,[^1] and we have the following differential equation$$y_{m}^2\left( \frac{\partial^2}{\partial x_{m}^2}+ \frac{\partial^2}{\partial y_{m}^2}\right)G_{k/2}^{\mathfrak H/\overline {\varGamma}}(z_1,z_2)=\frac{k}{2}\left(\frac{k}{2}-1\right)G_{k/2}^{\mathfrak H/\overline {\varGamma}}(z_1,z_2),\quad m=1,2$$when $z_1\notin \varGamma z_2 $.
3. \[itm:AGF3\]*(**Asymptotic Behavior**)* In the limit of $ z_1\to z_2$, one has $ G_{k/2}^{\mathfrak H/\overline {\varGamma}}(z_1,z_2)=2m^{\overline {\varGamma}}_{z_2}\log|z_1-z_2|+O(1)$, where $ m^{\overline {\varGamma}}_{z_2}:=\#\{\hat \gamma\in\overline{\varGamma}|\hat \gamma z_2=z_2\}$ counts the number of inequivalent transformations that leave the point $ z_2$ intact. As $ z_1$ tends to a cusp of the orbit space $ \varGamma\backslash\mathfrak H^*\equiv\varGamma\backslash(\mathfrak H\cup\mathbb Q\cup\{i\infty\})$, the function $ G_{k/2}^{\mathfrak H/\overline {\varGamma}}(z_1,z_2)$ tends to zero.
One can verify directly that Eq. \[eq:auto\_Green\_defn\_Q\_nu\] satisfies all the requirements in \[itm:AGF1\]–\[itm:AGF3\]. Now pick the point $ z_2=z'$, and suppose that we have a function $ F(z),z\in\mathfrak H$ satisfying the following properties:
1. \[itm:AGF1’\][(**Symmetry**)]{} For any $\hat \gamma\in\varGamma$, one has $ F(z)=F(\hat \gamma z)\in\mathbb C\cup\{-\infty\}$. Moreover, if $z\notin \varGamma z' $, then $ F(z)\in\mathbb C$.
2. \[itm:AGF2’\] [(**Differential Equation**)]{} The smooth function $ F(z),z\notin \varGamma z'$ satisifies the following differential equation$$\Delta_z^{\mathfrak H}F(z)\equiv y^2\left( \frac{\partial^2}{\partial x^2}+ \frac{\partial^2}{\partial y^2}\right)F(z)=\frac{k}{2}\left(\frac{k}{2}-1\right)F(z).$$
3. \[itm:AGF3’\][(**Asymptotic Behavior**)]{} As $ z\to z'$, one has $ F(z)=2m^{\overline {\varGamma}}_{z'}\log|z-z'|+O(1)$. As $ z$ tends to a cusp of $ \varGamma\backslash\mathfrak H^*$, the function $ F(z)$ tends to zero.
Then, the expression $ h(z)=F(z)-G_{k/2}^{\mathfrak H/\overline {\varGamma}}(z,z')$ extends to a smooth and bounded automorphic function on $ \varGamma\backslash\mathfrak H$, and so does $ \Delta_z^{\mathfrak H}h(z)$. In other words, we have $ h\in\mathcal D(\varGamma\backslash\mathfrak H)=\{f:\mathfrak H\longrightarrow \mathbb C|f\in C^\infty(\mathfrak H);f(\hat\gamma z)=f(z),\forall\hat\gamma\in\varGamma;\sup_{z\in\mathfrak H}|f(z)|<+\infty;\sup_{z\in\mathfrak H}|\Delta_z^{\mathfrak H}f(z)|<+\infty\}$ [@IwaniecGSM53 §4.1]. Had $h(z)$ been not identically zero, it would serve as an eigenfunction of the (negative semi-definite) operator $ \Delta_z^{\mathfrak H}:\mathcal D(\varGamma\backslash\mathfrak H)\longrightarrow\mathcal D(\varGamma\backslash\mathfrak H)$, with eigenvalue $ k(k-2)/4\in\mathbb C\smallsetminus(-\infty,0]$ — a contradiction.
Here, we recall that the totality of cusps for $ \varGamma\backslash\mathfrak H^*$ is represented by the orbit space $ \varGamma\backslash\mathbb P^1(\mathbb Q)=\varGamma\backslash(\mathbb Q\cup\{i\infty\})$ [@DiamondShurman p. 58]. In particular, the orbit space $ X_{0}(1)(\mathbb C):=SL(2,\mathbb Z)\backslash\mathfrak H^*$ has only one cusp at infinity. For the orbit spaces $ X_0(N)(\mathbb C):=\varGamma_0(N)\backslash\mathfrak H^*$ related to the Hecke congruence groups of levels $N=2$, $3$ and $4$, the following facts are also familiar: $ \varGamma_0(2)\backslash(\mathbb Q\cup\{i\infty\})=\varGamma_0(3)\backslash(\mathbb Q\cup\{i\infty\})=\{0,i\infty\}$, $\varGamma_0(4)\backslash(\mathbb Q\cup\{i\infty\})=\{0,\frac{1}{2},i\infty\} $. The geometric constructions for the modular curves $ Y_0(N)(\mathbb C)=\varGamma_0(N)\backslash\mathfrak H$ ($ N\in\{1,2,3,4\}$) are illustrated in Fig. \[fig:fundomain\]. For a technical description of $ X_0(N)(\mathbb C)=\varGamma_0(N)\backslash\mathfrak H^*$ as a compact Riemann surface, see [@Shimura1994 Chap. 1] or [@DiamondShurman Chap. 2].
\[rmk:Green\_recip\]One can use Eq. \[eq:auto\_Green\_defn\_Q\_nu\] to demonstrate “Green’s reciprocity” $ G_{k/2}^{\mathfrak H/\overline {\varGamma}}(z,z')=G_{k/2}^{\mathfrak H/\overline {\varGamma}}(z',z)$ by transitivity of the group actions. According to the uniqueness argument in the proof of Lemma \[lm:spec\_AGF\], we see that Green’s reciprocity will automatically be honored once criteria \[itm:AGF1’\]–\[itm:AGF3’\] are met. Therefore, it is *not* a requirement that an integral representation for an automorphic Green’s function should display the reciprocal symmetry explicitly.
Kontsevich–Zagier Integrals for Weight-4 Automorphic Green’s Functions on $\overline \varGamma_0(N)$ ($ N\in\{1,2,3,4\}$)\[subsec:KZ\_integrals\]
-------------------------------------------------------------------------------------------------------------------------------------------------
We shall start constructing integral representations for automorphic Green’s functions following a suggestion of Kontsevich and Zagier [@KontsevichZagier], after reading between the lines around their statement of the integral formula for $ G_2^{\mathfrak H/PSL(2,\mathbb Z)}(i,i\sqrt{2})$ (Eq. \[eq:KZ\_laconic\]). We shall refer to such formulae for automorphic Green’s functions as “Kontsevich–Zagier integrals”, both to acknowledge the source of our inspirations and to reckon the fact that, up to a scaling factor that is an integer power of $ \pi$, such integral formulae turn out to be “Kontsevich–Zagier periods” (integrals of algebraic functions over algebraic domains) when the arguments are CM points.
\[prop:G2HeckeNunified\] We have the following integral representations for weight-4 automorphic Green’s functions evaluated at a pair of non-equivalent points $ z$ and $z'$ (such that $ z\notin\varGamma_0(N)z'$) for $ N\in\{2,3,4\}$:$$\begin{aligned}
G_2^{\mathfrak H/\overline{\varGamma}_0(N)}(z,z')={}&\frac{4\pi^{2}}{y'}\operatorname{Re}\int_{z'}^{i\infty} \alpha_N(\zeta)[1-\alpha_N(\zeta)][NE_{2}(N\zeta)-E_2(\zeta)]^{2}\varrho^{\mathfrak H/\overline\varGamma_0(N)}_2(\zeta,z)\frac{(\zeta-z')(\zeta-\overline{z'})\operatorname{d}\zeta}{i(N-1)^{2}}\notag\\&-\frac{4\pi^{2}}{y'}\operatorname{Re}\int_{0}^{i\infty} \alpha_N(\zeta)[1-\alpha_N(\zeta)][NE_{2}(N\zeta)-E_2(\zeta)]^{2}\varrho^{\mathfrak H/\overline\varGamma_0(N)}_2(\zeta,z)\frac{\zeta^{2}\operatorname{d}\zeta}{i(N-1)^{2}},\label{eq:G2HeckeN_unified}\end{aligned}$$where $$\begin{aligned}
\varrho^{\mathfrak H/\overline\varGamma_0(N)}_2(\zeta,z)={}&\frac{\alpha_N(z)[1-\alpha_N(z)]}{[\alpha_N(\zeta)-\alpha_N(z)]^{2}}-\frac{N-1}{12}\frac{N^{2}[E_{2}(N z)]^2-N^{2}E_{4}(Nz)-[E_{2}(z)]^2+E_{4}(z)}{[\alpha_N(\zeta)-\alpha_N(z)][NE_{2}(N z)-E_{2}(z)]^2},\label{eq:rho2HeckeN_unified}\\ \text{with }\frac{1}{\alpha_N(z)}={}&1+\frac{1}{N^{6/(N-1)}}\left[ \frac{\eta(z)}{\eta(Nz)} \right]^{24/(N-1)}=\frac{1}{1- \alpha_N(-1/(Nz))},\quad \lim_{z\to i\infty}\alpha_N(z)=0.\label{eq:alpha_HeckeN_unified}\end{aligned}$$ Here in Eq. \[eq:G2HeckeN\_unified\], the paths of integration can be chosen as arbitrary curves joining the end points in the complex $ \zeta$-plane, so long as the singularities of the integrands are circumvented.
We shall establish Eq. \[eq:G2HeckeN\_unified\] by going through criteria \[itm:AGF1’\]–\[itm:AGF3’\] with respect to the variable $z\in\mathfrak H$.
It is well known that the functions $ \alpha_N(z),N\in\{2,3,4\}$ (Eq. \[eq:alpha\_HeckeN\_unified\]) are modular invariants that induce bijective mappings $ \alpha_N:\varGamma_0(N)\backslash\mathfrak H^*\longrightarrow\mathbb C\cup\{\infty\}$ on the respective Riemann surfaces (see [@ApostolVol2 Theorem 4.9] and [@Maier2009 Table 2]). Thus, whenever the function $ \varrho^{\mathfrak H/\overline\varGamma_0(N)}_2(\zeta,z)$ defined in Eq. \[eq:rho2HeckeN\_unified\] is regular, one can directly check the modular invariance $ \varrho^{\mathfrak H/\overline\varGamma_0(N)}_2(\zeta,\hat \gamma z) =\varrho^{\mathfrak H/\overline\varGamma_0(N)}_2(\zeta,z),\forall\hat \gamma\in\varGamma_0(N)$ where $ N\in\{2,3,4\}$. However, the regularity of an integral representation with respect to the variable $z$ cannot be taken for granted: one needs two additional steps to verify the symmetry criterion \[itm:AGF1’\], as we explain in the next couple of paragraphs.
First, we use residue calculus to demonstrate the path independence of the integral formula in Eq. \[eq:G2HeckeN\_unified\]. Simple computations reveal that[^2] $$\begin{aligned}
\alpha_N'(w):=\frac{\partial\alpha_{N}(w)}{\partial w}={}&\frac{2\pi i}{N-1}\alpha_{N}(w)[1-\alpha_{N}(w)][NE_{2}(Nw)-E_{2}(w)],&& \text{a.e. }w\in\mathfrak H;
\label{eq:alpha_N_deriv_E2}\\
\varrho^{\mathfrak H/\overline\varGamma_0(N)}_2(\zeta,z)={}&\frac{y}{i }\frac{\partial}{\partial y}\left\{ \frac{1}{y} \frac{\alpha_{N}(z)[1-\alpha_N(z)]}{[\alpha_{N}(\zeta)-\alpha_{N}(z)]\alpha_{N}'(z)}\right\},&& \text{a.e. }\zeta,z\in\mathfrak H,\label{eq:rho2HeckeN_unified_deriv_form}\end{aligned}$$so the right-hand side of Eq. \[eq:G2HeckeN\_unified\] may be rewritten as$$\begin{aligned}
\mathcal I_N(z,z')\equiv{}&
\operatorname{Re}\int_{z'}^{i\infty} \frac{[\alpha_N'(\zeta)]^{2}}{\alpha_N(\zeta)[1-\alpha_N(\zeta)]}y\frac{\partial}{\partial y}\left\{ \frac{1}{y} \frac{\alpha_{N}(z)[1-\alpha_N(z)]}{[\alpha_{N}(\zeta)-\alpha_{N}(z)]\alpha_{N}'(z)}\right\}\frac{(\zeta-z')(\zeta-\overline{z'})\operatorname{d}\zeta}{y'}\notag\\&-\operatorname{Re}\int_{0}^{i\infty} \frac{[\alpha_N'(\zeta)]^{2}}{\alpha_N(\zeta)[1-\alpha_N(\zeta)]}y\frac{\partial}{\partial y}\left\{ \frac{1}{y} \frac{\alpha_{N}(z)[1-\alpha_N(z)]}{[\alpha_{N}(\zeta)-\alpha_{N}(z)]\alpha_{N}'(z)}\right\}\frac{ \zeta^2\operatorname{d}\zeta}{y'}.\label{eq:deriv_form_G2Hecke234}\tag{\ref{eq:G2HeckeN_unified}$'$}\end{aligned}$$ Here as usual, a complex-analytic function $ f(z)=f(x+iy)$ is regarded as a bivariate function of $ (x,y)$ during the computations of partial derivatives with respect to $ y=\operatorname{Im}z$. For any two well-behaved analytic functions $ f$ and $g$ that are $ \varGamma$-invariant (*i.e. *$f(\hat\gamma z)=f(z)$, $ g(\hat \gamma z)=g( z)$, $ \forall\hat\gamma\in\varGamma\leq SL(2,\mathbb Z)$), the residues of the following three functions of $ \zeta\in\mathfrak H$: $$\begin{aligned}
\frac{[f'(\zeta)]^{2}\zeta^n}{g(\zeta)}y\frac{\partial}{\partial y}\left\{ \frac{1}{y} \frac{g(z)}{f(\zeta)-f(z)}\frac{1}{f'(z)}\right\} ,\quad n\in\{0,1,2\}\label{eq:G2_three_fns}\end{aligned}$$ at the point $ \zeta=\hat \gamma z$ are equal to $$\begin{aligned}
y\frac{\partial}{\partial y}\left\{ \frac{( \hat \gamma z)^{n}}{y} \frac{1}{\partial( \hat \gamma z)/\partial z}\right\} ,\quad n\in\{0,1,2\}.\end{aligned}$$These three residues are all real numbers:$$\begin{aligned}
-\frac{1}{\operatorname{Im}( \hat \gamma z)},\quad(n=0);\qquad-\frac{\operatorname{Re}( \hat \gamma z)}{\operatorname{Im}( \hat \gamma z)},\quad (n=1);\qquad -\frac{|( \hat \gamma z)|^2}{\operatorname{Im}( \hat \gamma z)},\quad (n=2).\end{aligned}$$ Therefore, the real parts of the contour integrals in Eq. \[eq:deriv\_form\_G2Hecke234\] are always path independent.
Second, we check that the expression for $ \varrho^{\mathfrak H/\overline\varGamma_0(N)}_2(\zeta,z)$ given in Eq. \[eq:rho2HeckeN\_unified\_deriv\_form\] is indeed regular whenever $ z\in\mathfrak H\smallsetminus(\varGamma_0(N)\zeta)$. With the formulae:[^3]$$\begin{aligned}
\left.\begin{array}{r@{\;=\;}l}
\left\{\dfrac{\alpha_{2}(z)[1-\alpha_2(z)]}{\alpha_{2 }'(z)}\right\} ^6& -\dfrac{\alpha_{2}(z)[1-\alpha_2(z)]^{2}}{2^{12} \pi ^6 [\eta (z)]^{24}} \\[10pt]
\left\{\dfrac{\alpha_{3}(z)[1-\alpha_3(z)]}{\alpha_{3 }'(z)}\right\}^6 & -\dfrac{\alpha_{3}(z)[1-\alpha_3(z)]^{3}}{2^6 3^3 \pi ^6[\eta (z)]^{24}} \\[10pt]\left\{\dfrac{\alpha_{4}(z)[1-\alpha_4(z)]}{\alpha_{4 }'(z)}\right\}^6 & -\dfrac{\alpha_{4}(z)[1-\alpha_4(z)]^{4}}{2^{10} \pi ^6[\eta (z)]^{24}} \end{array}\right\}\label{eq:alpha_N_ratio_ids}\end{aligned}$$and the non-vanishing property of the Dedekind eta function $ \eta(z)\neq0,\forall z\in\mathfrak H$, one can confirm that the function $ \varrho^{\mathfrak H/\overline\varGamma_0(N)}_2(\zeta,z)$ is smooth wherever $ \alpha_N(z)\in\mathbb C\smallsetminus\{\alpha_N(\zeta)\}$. It remains to show that $ \varrho^{\mathfrak H/\overline\varGamma_0(N)}_2(\zeta,z)$ is regular when $ \alpha_N(z)\to\infty$ for certain points $ z\in\mathfrak H$. By direct computation on the expressions in Eq. \[eq:rho2HeckeN\_unified\], one can show that (see Remark \[rmk:classical\_invariants\] below) $$\begin{aligned}
\lim_{a_2(z)\to\infty}
\varrho^{\mathfrak H/\overline\varGamma_0(2)}_2(\zeta,z)=-\frac{1}{2},\quad \alpha_2(\zeta)\in\mathbb C;\qquad \lim_{a_3(z)\to\infty}
\varrho^{\mathfrak H/\overline\varGamma_0(3)}_2(\zeta,z)=-\frac{1}{3}, \quad \alpha_3(\zeta)\in\mathbb C,\label{eq:alpha_inf_rho_lim}\end{aligned}$$ and $ \alpha_4(z) $ never diverges for $z\in\mathfrak H$. (Here, we note that $ \alpha_2(z)=\infty$ corresponds to the period-2 elliptic point $ m_{z}^{\overline\varGamma_0(2)}=2$, $ z\in\varGamma_0(2)\frac{i-1}{2}$; while $ \alpha_3(z)=\infty$ corresponds to the period-3 elliptic point $ m_{z}^{\overline\varGamma_0(3)}=3$, $z\in\varGamma_0(3)\frac{3+i\sqrt{3}}{6}$. The orbit space $ X_0(4)(\mathbb C)=\varGamma_0(4)\backslash\mathfrak H^*$ contains no elliptic points: $ m_z^{\overline\varGamma_0(4)}\equiv1,\forall z\in\mathfrak H$.) This completes the verification of criterion \[itm:AGF1’\].
It is relatively straightforward to show that the integral representation in Eq. \[eq:deriv\_form\_G2Hecke234\] satisfies the correct differential equation in the variable $z$, as dictated by criterion \[itm:AGF2’\]. This is because$$\begin{aligned}
\left[ y^2\left( \frac{\partial^{2}}{\partial x^{2}} +\frac{\partial^{2}}{\partial y^{2}}\right)-2\right]\left[y\frac{\partial}{\partial y}\frac{f(z)}{y}\right]=0\label{eq:weight_4_diff_eqn_test}\end{aligned}$$holds for any complex-analytic function $ f(z)=f(x+iy)$.
The verification of criterion \[itm:AGF3’\] naturally breaks down into two subtasks: the logarithmic asymptotics at the diagonal and the vanishing behavior at the cusps.
For the logarithmic asymptotics, we may momentarily assume that $ z'$ is not an elliptic point, so that $ m_{z'}^{\overline\varGamma_0(N)}=1$ and $ \alpha_N(z') $ is finite. As $ z$ approaches $z'$, the first term in Eq. \[eq:rho2HeckeN\_unified\] dominates the contribution to $ \varrho^{\mathfrak H/\overline\varGamma_0(N)}_2(\zeta,z)$, so the integral representation in Eq. \[eq:G2HeckeN\_unified\] goes asymptotically as $$\begin{aligned}
\sim {}&\frac{4\pi^{2}}{y'}\operatorname{Re}\int_{z'}^{i\infty} \alpha_N(\zeta)[1-\alpha_N(\zeta)][NE_{2}(N\zeta)-E_2(\zeta)]^{2}\frac{\alpha_N(z)[1-\alpha_N(z)]}{[\alpha_N(\zeta)-\alpha_N(z)]^{2}}\frac{(\zeta-z')(\zeta-\overline{z'})\operatorname{d}\zeta}{i(N-1)^{2}}\notag\\\sim{}&-\operatorname{Re}\int_{z'}^{2z'} \frac{[\alpha_N'(\zeta)]^{2}}{\alpha_N(\zeta)[1-\alpha_N(\zeta)]}\frac{\alpha_N(z)[1-\alpha_N(z)]}{[\alpha_N'(\zeta)]^{2}(\zeta-z)^2}\frac{(\zeta-z')(\zeta-\overline{z'})\operatorname{d}\zeta}{iy'}=-\operatorname{Re}\int_{z'}^{2z'} \frac{1}{\zeta-z}\frac{(z'-\overline{z'})\operatorname{d}\zeta}{iy'}+O(1)\notag\\={}&2\log|z'-z|+O(1).\end{aligned}$$
Before analyzing the behavior of Eq. \[eq:G2HeckeN\_unified\] near the infinite cusp $ z\to i\infty$, we note that Eqs. \[eq:alpha\_N\_deriv\_E2\] and \[eq:rho2HeckeN\_unified\_deriv\_form\] allow us to compute the following limit:$$\begin{aligned}
\varrho^{\mathfrak H/\overline\varGamma_0(N)}_2(\zeta,z)={}&(1-N)\frac{y}{2\pi }\frac{\partial}{\partial y}\left\{ \frac{1}{y} \frac{1}{[\alpha_{N}(\zeta)-\alpha_{N}(z)][NE_{2}(Nz)-E_{2}(z)]}\right\}\to0,\quad\text{as }z\to i\infty\label{eq:rho2_cusp_limit}\end{aligned}$$for every fixed $ \zeta\in\mathfrak H$.
We then exploit the path independence of the integral representation in Eq. \[eq:deriv\_form\_G2Hecke234\] and recast it into$$\begin{aligned}
\mathcal I_N(z,z')={}&-\operatorname{Re}\int_{0}^{z'} \frac{i[\alpha_N'(\zeta)]^{2}\varrho^{\mathfrak H/\overline\varGamma_0(N)}_2(\zeta,z)}{\alpha_N(\zeta)[1-\alpha_N(\zeta)]}\frac{ \zeta^2\operatorname{d}\zeta}{y'}-\operatorname{Re}\int_{z'}^{i\infty} \frac{i[\alpha_N'(\zeta)]^{2}\varrho^{\mathfrak H/\overline\varGamma_0(N)}_2(\zeta,z)}{\alpha_N(\zeta)[1-\alpha_N(\zeta)]}{\frac{ 2\zeta x'\operatorname{d}\zeta}{y'}}\notag\\&+\operatorname{Re}\int_{z'}^{i\infty} \frac{i[\alpha_N'(\zeta)]^{2}\varrho^{\mathfrak H/\overline\varGamma_0(N)}_2(\zeta,z)}{\alpha_N(\zeta)[1-\alpha_N(\zeta)]}{\frac{ |z'|^{2}\operatorname{d}\zeta}{y'}}.\label{eq:G2HeckeN_unified''}\tag{\ref{eq:G2HeckeN_unified}$''$}\end{aligned}$$We need to treat the three integrals in Eq. \[eq:G2HeckeN\_unified”\] separately in order to conclude that $ \lim_{z\to i\infty} \mathcal I_N(z,z')=0$. Here, by Eq. \[eq:rho2\_cusp\_limit\], we can conclude that the first integral in Eq. \[eq:G2HeckeN\_unified”\] vanishes as $ z$ tends to the infinite cusp:$$\begin{aligned}
\lim_{z\to i\infty}\operatorname{Re}\int_{0}^{z'} \frac{i[\alpha_N'(\zeta)]^{2}\varrho^{\mathfrak H/\overline\varGamma_0(N)}_2(\zeta,z)}{\alpha_N(\zeta)[1-\alpha_N(\zeta)]}\frac{ \zeta^2\operatorname{d}\zeta}{y'}=0.\label{eq:cusp_lim0a}\end{aligned}$$ In Eq. \[eq:cusp\_lim0a\], it is easy to bound $ |\varrho^{\mathfrak H/\overline\varGamma_0(N)}_2(\zeta,z)|$ by a finite constant $ C\in(0,+\infty)$ for $ z\to i\infty$ and bounded $ |\zeta|$, and we have absolute integrability:$$\begin{aligned}
C\int_{0}^{z'} \left\vert\frac{i[\alpha_N'(\zeta)]^{2}}{\alpha_N(\zeta)[1-\alpha_N(\zeta)]}\right\vert\frac{ |\zeta|^2|\operatorname{d}\zeta|}{y'}<+\infty.\end{aligned}$$(Bearing in mind that $ 1-\alpha_N(\zeta)=\alpha_N(-1/(N\zeta))=O(e^{-2\pi i/(N\zeta)})$ and $ \alpha'_N(\zeta)=O(|\zeta|^{-2}e^{-2\pi i/(N\zeta)})$ when $\operatorname{Im}\zeta\to0^+$, one sees that the integrand above is bounded, hence absolutely integrable.) So, one can interchange the limit and the integral sign in Eq. \[eq:cusp\_lim0a\], according to the dominated convergence theorem. Meanwhile, one can also apply the dominated convergence theorem to the second integral in Eq. \[eq:G2HeckeN\_unified”\] after transforming it into an integral from $ 0$ to $z'$:$$\begin{aligned}
-\operatorname{Re}\int_{z'}^{i\infty} \frac{i[\alpha_N'(\zeta)]^{2}\varrho^{\mathfrak H/\overline\varGamma_0(N)}_2(\zeta,z)}{\alpha_N(\zeta)[1-\alpha_N(\zeta)]}{\frac{ 2\zeta x'\operatorname{d}\zeta}{y'}}=\operatorname{Re}\int_{0}^{z'} \frac{i[\alpha_N'(\zeta)]^{2}\varrho^{\mathfrak H/\overline\varGamma_0(N)}_2(\zeta,z)}{\alpha_N(\zeta)[1-\alpha_N(\zeta)]}{\frac{ 2\zeta x'\operatorname{d}\zeta}{y'}}\to0,\quad\text{as }z\to i\infty.\label{eq:cusp_lim0b}\end{aligned}$$Here, the first equality in Eq. \[eq:cusp\_lim0b\] owes to the following vanishing identity:$$\begin{aligned}
\operatorname{Re}\int_{0}^{i\infty} \frac{i[\alpha_N'(\zeta)]^{2}\varrho^{\mathfrak H/\overline\varGamma_0(N)}_2(\zeta,z)\zeta\operatorname{d}\zeta}{\alpha_N(\zeta)[1-\alpha_N(\zeta)]}=0,\quad \forall z\in\mathfrak H.\label{eq:rho2_int0}\end{aligned}$$One can easily verify Eq. \[eq:rho2\_int0\] for boundary points $ z\in\mathfrak H\cap\partial\mathfrak D_N$ (Fig. \[fig:fundomain\]*c*–*e*), that is, for all the points $ z\in\mathfrak H$ satisfying$$\begin{aligned}
|\operatorname{Re}z|=\frac{1}{2},\operatorname{Im}z>\frac{1}{2}\sqrt{\frac{4-N}{N}}\quad\text{or}\quad-\frac{1}{2}<\operatorname{Re}z<0 ,\left|z+\frac{1}{N}\right |=\frac{1}{N}\quad \text{or}\quad 0<\operatorname{Re}z<\frac{1}{2} ,\left|z-\frac{1}{N}\right |=\frac{1}{N},\label{eq:HeckeN_fun_dom_bd}\end{aligned}$$ by choosing the contour of integration along the imaginary axis $ \zeta/i\in(0,+\infty)$. This is because in such scenarios, $ \alpha_N(z) $, $ \alpha_N(\zeta)$, $ i\alpha'_N(\zeta)$ and $ \varrho^{\mathfrak H/\overline\varGamma_0(N)}_2(\zeta,z)$ are all real numbers. As the integral in Eq. \[eq:rho2\_int0\] is annihilated by the differential operator $$\begin{aligned}
-2+ y^2 \left( \frac{\partial^2}{\partial x^2}+\frac{\partial^2}{\partial y^2}\right)=-2-(z-\overline{z})^{2}\frac{\partial}{\partial z}\frac{\partial}{\partial\overline{ z}},\end{aligned}$$and satisfies homogeneous Dirichlet boundary condition on the domain $ \mathfrak D_N$ bounded by the curves specified in Eq. \[eq:HeckeN\_fun\_dom\_bd\], we see that Eq. \[eq:rho2\_int0\] must hold within $ \mathfrak D_N$. Such a vanishing identity then extends to every $z\in\mathfrak H$, because the closure of the aforementioned domain $ \mathfrak H\cap(\mathfrak D_N\cup\partial\mathfrak D_N)$ tessellates the upper half-plane under the actions of $ \varGamma_0(N)$ ($ N\in\{2,3,4\}$). After the confirmation of the vanishing asymptotics for the first two integrals in Eq. \[eq:G2HeckeN\_unified”\], we need to handle the last one with integration by parts:[$$\begin{aligned}
&
\operatorname{Re}\int_{z'}^{i\infty} \frac{i[\alpha_N'(\zeta)]^{2}\varrho^{\mathfrak H/\overline\varGamma_0(N)}_2(\zeta,z)}{\alpha_N(\zeta)[1-\alpha_N(\zeta)]}{\operatorname{d}\zeta}\notag\\={}&(N-1)\frac{y}{2\pi }\frac{\partial}{\partial y}\operatorname{Re}\left\{ \frac{1}{y} \frac{1}{NE_{2}(Nz)-E_{2}(z)}\int_{z'}^{i\infty} \frac{[\alpha_N'(\zeta)]^{2}}{i\alpha_N(\zeta)[1-\alpha_N(\zeta)]}{\frac{\operatorname{d}\zeta}{\alpha_{N}(\zeta)-\alpha_{N}(z)}}\right\}\notag\\={}&y\frac{\partial}{\partial y}\operatorname{Re}\left\{ \frac{1}{y} \frac{1}{NE_{2}(Nz)-E_{2}(z)}\int_{z'}^{i\infty} [NE_{2}(N\zeta)-E_{2}(\zeta)]\operatorname{d}\log[\alpha_{N}(\zeta)-\alpha_{N}(z)]\right\}\notag\\={}&-y\frac{\partial}{\partial y}\operatorname{Re}\left\{ \frac{1}{y} \frac{1}{NE_{2}(Nz)-E_{2}(z)}\int_{z'}^{i\infty} \log[\alpha_{N}(\zeta)-\alpha_{N}(z)]\frac{\partial[NE_{2}(N\zeta)-E_{2}(\zeta)-N+1]}{\partial\zeta}\operatorname{d}\zeta\right\}\notag\\&-y\frac{\partial}{\partial y}\operatorname{Re}\left\{ \frac{1}{y} \frac{NE_{2}(Nz')-E_{2}(z')}{NE_{2}(Nz)-E_{2}(z)}\log[\alpha_{N}(z')-\alpha_{N}(z)]\right\}.\label{eq:cusp_lim0c}\end{aligned}$$]{}As $ y\to+\infty$, one has $ \alpha_N(z)=O(e^{2\pi iz})$, $ NE_{2}(Nz)-E_{2}(z)=N-1+O(e^{2\pi iz})$, so accordingly,$$\begin{aligned}
&\int_{z'}^{i\infty} \log[\alpha_{N}(\zeta)-\alpha_{N}(z)]\frac{\partial[NE_{2}(N\zeta)-E_{2}(\zeta)-N+1]}{\partial\zeta}\operatorname{d}\zeta\notag\\\to{}&\int_{z'}^{i\infty} \log[\alpha_{N}(\zeta)]\frac{\partial[NE_{2}(N\zeta)-E_{2}(\zeta)-N+1]}{\partial\zeta}\operatorname{d}\zeta\in\mathbb C\end{aligned}$$and the contributions to[^4]$$\begin{aligned}
&\frac{1}{NE_{2}(Nz)-E_{2}(z)}\frac{\partial}{\partial z}\int_{z'}^{i\infty} \log[\alpha_{N}(\zeta)-\alpha_{N}(z)]\frac{\partial[NE_{2}(N\zeta)-E_{2}(\zeta)-N+1]}{\partial\zeta}\operatorname{d}\zeta\notag\\={}&\frac{2\pi i[1-\alpha_{N}(z)]}{N-1}\left(\int_{z'}^{z-\frac{i}{3}}+\int_{z-\frac{i}{3}}^{z+\frac{i}{3}}+\int_{z+\frac{i}{3}}^{i\infty}\right) \frac{\alpha_{N}(z)}{\alpha_{N}(z)-\alpha_N(\zeta)}\frac{\partial[NE_{2}(N\zeta)-E_{2}(\zeta)-N+1]}{\partial\zeta}\operatorname{d}\zeta\end{aligned}$$ can be decomposed into three parts:[$$\begin{aligned}
\left|\int_{z'}^{z-\frac{i}{3}}\frac{\alpha_{N}(z)}{\alpha_{N}(z)-\alpha_N(\zeta)}\frac{\partial[NE_{2}(N\zeta)-E_{2}(\zeta)-N+1]}{\partial\zeta}\operatorname{d}\zeta\right|\leq{}&C_1\left\vert\int_{z'}^{z-\frac{i}{3}}|\alpha_{N}(z)|\operatorname{d}\zeta\right\vert=O(|z\alpha_{N}(z)|);\notag\\\left|\int_{z-\frac{i}{3}}^{z+\frac{i}{3}}\frac{\alpha_N(z)}{\alpha_{N}(z)-\alpha_N(\zeta)}\frac{\partial[NE_{2}(N\zeta)-E_{2}(\zeta)-N+1]}{\partial\zeta}\operatorname{d}\zeta\right|\leq{}& C_2\int_{0}^\pi\frac{|e^{2\pi iz+\frac{2\pi i}{3} e^{i\phi}}|}{|1-e^{\frac{2\pi i}{3} e^{i\phi}}|}\operatorname{d}\phi=O(e^{-\pi y});\notag\\\left|\int_{z+\frac{i}{3}}^{i\infty}\frac{\alpha_{N}(z)}{\alpha_{N}(z)-\alpha_N(\zeta)}\frac{\partial[NE_{2}(N\zeta)-E_{2}(\zeta)-N+1]}{\partial\zeta}\operatorname{d}\zeta\right|\leq{}& C_3\left|\int_{z+\frac{i}{3}}^{i\infty}\frac{ \alpha_{N}(z)e^{2\pi i\zeta}}{\alpha_{N}(z)-\alpha_N(\zeta)}\operatorname{d}\zeta\right|=O(|z\alpha_{N}(z)|),\end{aligned}$$]{}where $ C_1$, $ C_2$ and $ C_3$ are constant bounds, deducible from the mean value theorems for integration. As a consequence, one can verify that the expression in Eq. \[eq:cusp\_lim0c\] has order $ O(1/y)$ as $ z\to i\infty$.
From the result $ \lim_{z\to i\infty} \mathcal I_N(z,z')=0$ proved in the last paragraph, we can also deduce the vanishing behavior at another cusp $ \lim_{z\to0} \mathcal I_N(z,z')=0$. This becomes possible once we confirm that $\mathcal I_N(z,z')$ respects the Fricke involution in the same way as $ G_2^{\mathfrak H/\overline{\varGamma}_0(N)}(z,z')$ (Eq. \[eq:Fricke\_inv\_HeckeN\_GZ\]). Indeed, by the symmetry $ \varrho^{\mathfrak H/\overline\varGamma_0(N)}_2(\zeta,z)=\varrho^{\mathfrak H/\overline\varGamma_0(N)}_2(-1/(N\zeta),-1/(Nz))$ along with Eqs. \[eq:G2HeckeN\_unified”\] and \[eq:rho2\_int0\], we can verify that $$\begin{aligned}
\mathcal
I_N(z,z')-\mathcal
I_N\left( -\frac{1}{Nz} ,-\frac{1}{Nz'}\right)=-\operatorname{Re}\int_{0}^{i\infty} \frac{i[\alpha_N'(\zeta)]^{2}\varrho^{\mathfrak H/\overline\varGamma_0(N)}_2(\zeta,z)}{\alpha_N(\zeta)[1-\alpha_N(\zeta)]}{\frac{ 2\zeta x'\operatorname{d}\zeta}{y'}}\equiv0.\label{eq:I_N_Fricke_inv_check}\end{aligned}$$
The orbit space $ \varGamma_0(4)\backslash\mathfrak H^{*}$ has a third cusp at $ z=1/2$. Akin to Eq. \[eq:alpha\_inf\_rho\_lim\], one can compute$$\begin{aligned}
\lim_{z\to1/2}
\varrho^{\mathfrak H/\overline\varGamma_0(4)}_2(\zeta,z)=\lim_{a_4(z)\to\infty}
\varrho^{\mathfrak H/\overline\varGamma_0(4)}_2(\zeta,z)=0, \quad \alpha_4(\zeta)\in\mathbb C.\label{eq:rho2Hecke4cusp_half}\end{aligned}$$As we can always choose the paths of integration for $ \mathcal I_4(z,z')$ (Eq. \[eq:G2HeckeN\_unified”\]) away from the cusp $ z=1/2$, the vanishing limit $ \lim_{z\to1/2}\mathcal I_4(z,z')=0$ follows directly from Eq. \[eq:rho2Hecke4cusp\_half\] and an application of the dominated convergence theorem.
To summarize, we have verified that the integral representation $ \mathcal I_N(z,z')$ ($ N\in\{2,3,4\}$) satisfies all the three criteria \[itm:AGF1’\]–\[itm:AGF3’\], provided that $z'$ is not an elliptic point: $ m_{z'}^{\overline\varGamma_0(N)}=1$. Hence, in these scenarios, $ \mathcal I_N(z,z')$ must be identical to the automorphic Green’s function $G_2^{\mathfrak H/\overline\varGamma_0(N)}(z,z') $, according to Lemma \[lm:spec\_AGF\]. For the isolated elliptic points where $ m_{z'}^{\overline\varGamma_0(N)}>1$, $ N\in\{2,3\}$, one can justify the integral representation in Eq. \[eq:G2HeckeN\_unified\] by continuity in the variable $z'$.
\[rmk:classical\_invariants\]The function $ \alpha_3(z)$ is just the Ramanujan cubic invariant (see [@RN5 Chap. 33, §§2–8] and [@RLN2 Chap. 9]):$$\begin{aligned}
\alpha_{3}(z):=\left( \frac{[\eta(z/3)]^3}{3[\eta(3z)]^3}+1 \right)^{-3}=\left( \frac{[\eta(z)]^{12}}{27[\eta(3z)]^{12}}+1 \right)^{-1},\quad z\in\mathfrak H\smallsetminus\left({\varGamma}_0(3)\frac{3+i\sqrt{3}}{6}\right)\label{eq:alpha3_defn}\end{aligned}$$where the equivalence of the two definitions for $ \alpha_{3}(z)$ in Eq. \[eq:alpha3\_defn\] follows from the cubic modular equation for the Dedekind eta function [@RN3 p. 345, Eq. (iv)], which is also provable by the cubic identity of Borwein–Borwein–Garvan [@BorweinBorweinGarvan]. The function $ 1/\alpha_3(z)$ extends to a holomorphic function on $ \mathfrak H$, with a third-order zero at $ z=(3+i\sqrt{3})/6$, which corresponds to the period-3 elliptic point on $ X_0(3)(\mathbb C)=\varGamma_0(3)\backslash\mathfrak H^*$. Meanwhile, the modular lambda function $ \lambda(z):=16[\eta (z/2)]^8 [\eta (2 z)]^{16}/[\eta (z)]^{24},z\in\mathfrak H $ is related to the modular invariants $\alpha_2 $ and $ \alpha_4$ as follows:$$\begin{aligned}
\alpha_{2}(z)=1-\frac{1}{[1-2\lambda(2z+1)]^2};\quad \alpha_4(z)=\lambda(2z).\label{eq:alpha2_alpha4}\end{aligned}$$One has $ \lambda(i)=1/2$, so $ \alpha_2(z)$ diverges at $z= (i-1)/2$, which corresponds to the period-2 elliptic point on the orbit space $ X_0(2)(\mathbb C)=\varGamma_0(2)\backslash\mathfrak H^*$.
We note that the viability of the integral representations in Eq. \[eq:G2HeckeN\_unified\] draws heavily on the special geometric properties (Fig. \[fig:fundomain\]*c*–*e*) of the compact Riemann surfaces $ \varGamma_0(N)\backslash\mathfrak H^*$ ($ N\in\{2,3,4\}$). While the mapping $ \alpha_N:\varGamma_0(N)\backslash\mathfrak H^*\longrightarrow \mathbb C\cup\{\infty\}$ remains bijective so long as $ (N-1)$ divides $24 $ [@Maier2009 Table 2], the regularity of $ \alpha_N(z)[1-\alpha_N(z)]/{\alpha'_N}(z)$ (see Eq. \[eq:alpha\_N\_ratio\_ids\]) will be lost for $ N>4$. For example, the reader may check that $ \alpha_5'(z)$ vanishes at $ z=(\pm 3+i)/5$, which are the two non-equivalent period-2 elliptic points on $ \varGamma_0(5)\backslash\mathfrak H^*$ (see [@Maier2009 Table 2, Theorem 5.5, and Corollary 5.5.2]). Therefore, for $ N=5$, the integral in Eq. \[eq:G2HeckeN\_unified\] is not regular at $ z=(\pm 3+i)/5$, thus failing criterion \[itm:AGF1’\]. In fact, Eq. \[eq:G2HeckeN\_unified\] ceases to be a valid integral representation for automorphic Green’s functions on the Hecke congruence group $ \overline\varGamma_0(N)$ for all the higher levels $N\geq5$. Nonetheless, integral representations like Eq. \[eq:G2HeckeN\_unified\] do adapt to certain automorphic Green’s functions on higher level Hecke congruence groups with Nebentypus characters. As an example, for a special Dirichlet character $$\chi_5(d)\equiv\left( \frac{d}{5} \right)=\left( \frac{5}{d} \right)=\begin{cases}1, & d\equiv\pm1\pmod5 \\
-1, & d\equiv\pm2\pmod5 \\
0, & d\equiv0\pmod5 \\
\end{cases}$$ we have $$\begin{aligned}
G_2^{\mathfrak H/(\overline \varGamma_0(5),\chi_5)}(z,z'):={}&-2\sum_{\hat \gamma=\left(\begin{smallmatrix}a&b\\c&d\end{smallmatrix}\right)\in\overline{\varGamma}_0(5)}\chi _{5}(d)Q_{1}
\left( 1+\frac{\vert z -\hat \gamma z'\vert ^{2}}{2\operatorname{Im}z\operatorname{Im}(\hat\gamma z')} \right)\notag\\={}&\frac{\pi^{2}}{y'}\operatorname{Re}\int_{z'}^{i\infty} \alpha_5(\zeta)\frac{[\eta(\zeta)]^{5}}{\eta(5\zeta)}[5E_{2}(5\zeta)-E_2(\zeta)]\varrho^{\mathfrak H/(\overline \varGamma_0(5),\chi_5)}_2(\zeta,z)\frac{(\zeta-z')(\zeta-\overline{z'})\operatorname{d}\zeta}{i}\notag\\&-\frac{\pi^{2}}{y'}\operatorname{Re}\int_{0}^{i\infty} \alpha_5(\zeta)\frac{[\eta(\zeta)]^{5}}{\eta(5\zeta)}[5E_{2}(5\zeta)-E_2(\zeta)]\varrho^{\mathfrak H/(\overline \varGamma_0(5),\chi_5)}_2(\zeta,z)\frac{\zeta^{2}\operatorname{d}\zeta}{i},\end{aligned}$$ where the function$$\begin{aligned}
\varrho^{\mathfrak H/(\overline \varGamma_0(5),\chi_5)}_2(\zeta,z):=-\frac{y}{2\pi\ }\frac{\partial}{\partial y}\left\{ \frac{1}{y} \frac{1-\alpha_{5}(z)}{\alpha_{5}(\zeta)-\alpha_{5}(z)}\frac{\eta(5z)}{[\eta(z)]^{5}}\right\}\end{aligned}$$ is regular for $ z\in\mathfrak H\smallsetminus(\varGamma_0(5)\zeta)$.
The projective counterpart for $ SL(2,\mathbb Z)\equiv\varGamma_0(1)$ is the full modular group $ PSL(2,\mathbb Z)=SL(2,\mathbb Z)/\{\hat I,-\hat I\}$. The fundamental domain (Fig. \[fig:fundomain\]*a*)$$\begin{aligned}
\mathfrak D:=\left\{ z\in\mathfrak H\left| |z|>1,-\frac12<\operatorname{Re}z\leq\frac{1}{2} \right. \right\}\cup \left\{ z\in\mathfrak H\left| |z|=1,0\leq \operatorname{Re}z\leq\frac{1}{2} \right. \right\}\label{eq:fun_domain_SL2Z}\end{aligned}$$tessellates the upper half plane $ \mathfrak H$ under actions of the full modular group $ PSL(2,\mathbb Z)$, generated by the translation $ \hat{T}:z\mapsto z+1$ and the inversion $ \hat{S}:z\mapsto-1/z$ (Fig. \[fig:fundomain\]*b*). It is possible to assign two univalent branches to the square root expression $ \sqrt{\smash[b]{(j(z)-1728)/j(z)}}$ so that it induces a smooth bijection between $ (\mathfrak D\cup \hat{S}\mathfrak D)\smallsetminus\{e^{\pi i/3},e^{2\pi i/3}\}$ and $ \mathbb C$. In particular, we shall require that $ \sqrt{\smash[b]{(j(z)-1728)/j(z)}}>0$ for $ z/i>1$, and $ \sqrt{\smash[b]{(j(z)-1728)/j(z)}}<0$ for $ 0<z/i<1$. In this manner, we have $ j(z)=j(-1/z),\forall z\in\mathfrak H$, but $\sqrt{\smash[b]{(j(z)-1728)/j(z)}}=-\sqrt{\smash[b]{(j(-1/z)-1728)/j(-1/z)}},\forall z\in\mathfrak D\smallsetminus\{e^{\pi i/3}\}$.
Define $$\begin{aligned}
\alpha_1(z)=\frac{1-\sqrt{(j(z)-1728)/j(z)}}{2},\quad \forall z\in(\mathfrak D\cup \hat{S}\mathfrak D)\smallsetminus\{e^{\pi i/3},e^{2\pi i/3}\}\label{eq:alpha1_defn}\end{aligned}$$with the aforementioned convention about square roots, then we have $$\begin{aligned}
\alpha_1(z)=1-\alpha_1\left(-\frac1z\right),\quad\lim_{z\to i\infty}\alpha_1(z)=0,\end{aligned}$$in parallel to Eq. \[eq:alpha\_HeckeN\_unified\]. The definition in Eq. \[eq:alpha1\_defn\] can be smoothly extended to all $ z\in\mathfrak H\smallsetminus(SL(2,\mathbb Z)e^{\pi i/3})$, by actions of an index-2 normal subgroup of $ SL(2,\mathbb Z)$ [@Schoeneberg p. 86]. The derivative of $\alpha_1(z)$ can be computed from the differentiation formula for the $ j$-invariant:$$\begin{aligned}
i
\frac{\partial j(z)}{\partial z}=2\pi j(z)\frac{E_{6}(z)}{E_4(z)},\quad \text{a.e. }z\in\mathfrak H.\label{eq:j'z_E4E6}\end{aligned}$$
With the preparations in the last two paragraphs, we can state and prove the Kontsevich–Zagier integral representations for $ G_2^{\mathfrak H/PSL(2,\mathbb Z)}(z,z')$ in the next proposition.
\[prop:G2PSL2Z\]For a pair of non-equivalent points $ z,z'\in\mathfrak H$ such that $ j(z)\neq j(z')$, one has $$\begin{aligned}
G_2^{\mathfrak H/PSL(2,\mathbb Z)}(z,z')={}&\frac{1728\pi^{2}}{y'}\operatorname{Re}\int_{z'}^{i\infty} \frac{E_{4}(\zeta)}{j(\zeta)}\rho^{\mathfrak H/PSL(2,\mathbb Z)}_2(\zeta,z)\frac{(\zeta-z')(\zeta-\overline{z'})\operatorname{d}\zeta}{i}\notag\\&-\frac{1728\pi^{2}}{y'}\operatorname{Re}\int_{0}^{i\infty} \frac{E_{4}(\zeta)}{j(\zeta)}\rho^{\mathfrak H/PSL(2,\mathbb Z)}_2(\zeta,z)\frac{\zeta^{2}\operatorname{d}\zeta}{i},
\label{eq:G2PSL2Z_Eisenstein_form}\end{aligned}$$where $$\begin{aligned}
\rho^{\mathfrak H/PSL(2,\mathbb Z)}_2(\zeta,z)=\frac{[j(\zeta)]^{2}j(z)}{432[j(\zeta)-j(z)]^{2}}-\frac{2j(\zeta)[j(\zeta)+j(z)]}{[j(\zeta)-j(z)]^{2}}+\frac{j(\zeta)j(z)}{2592[j(\zeta)-j(z)]}{\frac{E_6(z)}{[E_4(z)]^2}}\left[ E_{2}(z)-\frac{E_6(z)}{E_4(z)} \right].\label{eq:rho2PSL2Z_Eisenstein}\end{aligned}$$Here in Eq. \[eq:G2PSL2Z\_Eisenstein\_form\], the integration paths can be arbitrary, provided that $ j(\zeta)\neq j(z)$ along the way.
As in Proposition \[prop:G2HeckeNunified\], we shall check criteria \[itm:AGF1’\]–\[itm:AGF3’\] for the variable $z\in\mathfrak H$.
The symmetry property in \[itm:AGF1’\] follows from two simple observations. First, we point out that the function $ \rho^{\mathfrak H/PSL(2,\mathbb Z)}_2(\zeta,z)$ defined in Eq. \[eq:rho2PSL2Z\_Eisenstein\] is regular so long as $ j(\zeta)\neq j(z)$, according to the following computation:$$\begin{aligned}
\lim_{j(z)\to0}
\rho^{\mathfrak H/PSL(2,\mathbb Z)}_2(\zeta,z)=\lim_{E_{4}(z)\to0}
\rho^{\mathfrak H/PSL(2,\mathbb Z)}_2(\zeta,z)=-\frac{4}{3}.\label{eq:alpha_inf_rho_lim_PSL2Z}\end{aligned}$$Second, using Eq. \[eq:j’z\_E4E6\] and Ramanujan’s differential equation for the Eisenstein series [@Ramanujan1916 Eq. 30], one can rewrite Eq. \[eq:rho2PSL2Z\_Eisenstein\] as$$\begin{aligned}
\rho^{\mathfrak H/PSL(2,\mathbb Z)}_2(\zeta,z)=\frac{y}{864 \pi}\frac{\partial}{\partial y}\left\{ \frac{1}{y} \frac{j(\zeta)j(z)}{j(\zeta)-j(z)}\frac{{E_{6}(z)}}{[E_{4}(z)]^2}\right\},
\label{eq:rho2PSL2Z_Eisenstein'}\tag{\ref{eq:rho2PSL2Z_Eisenstein}$'$}\end{aligned}$$ so that one may confirm path independence by a variant of Eq. \[eq:G2\_three\_fns\]. Namely, given that $ f(\hat \gamma z)=f(z)$ and $ M(\hat \gamma z)=(cz+d)^4 M(z)$ for $ \forall\hat\gamma=\left( \begin{smallmatrix}a&b\\c&d\end{smallmatrix}\right)\in SL(2,\mathbb Z)$, the residues of the following three functions of $ \zeta\in\mathfrak H$: $$\begin{aligned}
&M(\zeta)\zeta^ny\frac{\partial}{\partial y}\left\{ \frac{1}{y} \frac{f'(z)}{f(\zeta)-f(z)}\frac{1}{M(z)}\right\}\notag\\ ={}&M(\zeta)\zeta^ny\frac{\partial}{\partial y}\left\{ \frac{1}{y} \frac{f'(z)}{f(\hat\gamma^{-1}\zeta)-f(z)}\frac{1}{M(z)}\right\} ,\quad n\in\{0,1,2\}\label{eq:G2_three_fns'}\tag{\ref{eq:G2_three_fns}$'$}\end{aligned}$$ at the point $ \zeta=\hat \gamma z$ are all real numbers:$$\begin{aligned}
-\frac{1}{\operatorname{Im}( \hat \gamma z)},\quad(n=0);\qquad-\frac{\operatorname{Re}( \hat \gamma z)}{\operatorname{Im}( \hat \gamma z)},\quad (n=1);\qquad -\frac{|( \hat \gamma z)|^2}{\operatorname{Im}( \hat \gamma z)},\quad (n=2).\end{aligned}$$
To verify the differential equation in criterion \[itm:AGF2’\], it would suffice to combine Eq. \[eq:rho2PSL2Z\_Eisenstein’\] with Eq. \[eq:weight\_4\_diff\_eqn\_test\].
For criterion \[itm:AGF3’\], the procedures for logarithmic asymptotics are essentially similar to that of Proposition \[prop:G2HeckeNunified\]. For the verification of cusp behavior, we need a decomposition $ \rho^{\mathfrak H/PSL(2,\mathbb Z)}_2(\zeta,z)=\varrho(\zeta,z)+\varrho(\zeta,-1/z)$ where$$\begin{aligned}
\varrho(\zeta,z)=\frac{y}{i }\frac{\partial}{\partial y}\left\{ \frac{1}{y} \frac{\alpha_{1}(z)[1-\alpha_1(z)]}{[\alpha_{1}(\zeta)-\alpha_{1}(z)]\alpha_{1}'(z)}\right\}=\varrho\left( -\frac{1}{\zeta} ,-\frac{1}{z}\right).\end{aligned}$$If we replace $ \rho^{\mathfrak H/PSL(2,\mathbb Z)}_2(\zeta,z)$ by $ \varrho(\zeta,z)$ in the integrands of Eq. \[eq:G2PSL2Z\_Eisenstein\_form\] and call the result $ \mathcal I_1(z,z')$, then we can justify the limits $ \lim_{z\to0}\mathcal I_1(z,z')=\lim_{z\to i\infty}\mathcal I_1(z,z')=0$ in the same fashion as what we did for $ \mathcal I_N(z,z'),N\in\{2,3,4\}$ in Proposition \[prop:G2HeckeNunified\]. The right-hand side of Eq. \[eq:G2PSL2Z\_Eisenstein\_form\], being equal to $\mathcal I_1(z,z')+\mathcal I_1(-1/z,z')$, should thus vanish as $ z$ goes to the infinite cusp $ i\infty$.
This completes the justification of the integral representation in Eq. \[eq:G2PSL2Z\_Eisenstein\_form\].
Recalling the asymptotic behavior of automorphic Green’s functions [@GrossZagier1985 §5]:$$\begin{aligned}
G_{s}^{\mathfrak H/PSL(2,\mathbb Z)}(z_1,z_2)\sim{}&\frac{2\pi}{1-2s}\frac{y_1^{1-s}y_{2}^{s}}{\zeta(2s)}\sum_{\substack{m,n\in\mathbb Z\\m^2+n^2\neq0}}\frac{1}{|mz_{2}+n|^{2s}},\quad y_1\to+\infty,\operatorname{Re}s>1\label{eq:Gs_asympt}\end{aligned}$$ we now have$$\begin{aligned}
\label{eq:Epstein_s2_int_star}
\sum_{\substack{m,n\in\mathbb Z\\m^2+n^2\neq0}}\frac{ y^2}{|mz+n|^{4}}={}&-\frac{\pi^3}{60}\lim_{y'\to+\infty}G_2^{\mathfrak H/PSL(2,\mathbb Z)}(z,z')y'\notag\\={}&\frac{144\pi^{5}}{5}\operatorname{Re}\int_{0}^{i\infty} \frac{E_{4}(\zeta)}{j(\zeta)}\rho^{\mathfrak H/PSL(2,\mathbb Z)}_2(\zeta,z)\frac{\zeta^{2}\operatorname{d}\zeta}{i},\end{aligned}$$where each one of the three terms is $ SL(2,\mathbb Z)$-invariant with respect to $z$.
For $ z=i$, one has the asymptotic behavior$$\begin{aligned}
G_s^{\mathfrak H/PSL(2,\mathbb Z)}(z,i)\sim\frac{4\pi}{1-2s}\frac{2\zeta(s)L(s,\chi_{-4})}{\zeta(2s) y^{s-1}}:=\frac{4\pi}{1-2s}\frac{2\zeta(s)}{\zeta(2s) y^{s-1}}\sum^\infty_{n=0}\frac{(-1)^n}{(2n+1)^s},\quad y\to+\infty,
\label{eq:Gs_zi_asympt}\end{aligned}$$according to Eq. \[eq:Gs\_asympt\] and an integration of the two-squares theorem [@SteinII p. 298]$$\begin{aligned}
\label{eq:two_squares}\sum_{\substack{m,n\in\mathbb Z\\m^2+n^2\neq0}}y^{s-1}e^{-(m^{2}+n^2)y}=2\sum_{\ell=1}^\infty\frac{y^{s-1}}{\cosh\ell y}\end{aligned}$$over $ y\in(0,+\infty)$. In Eq. \[eq:Gs\_zi\_asympt\], we have used the following notation for $ L$-functions: $$\begin{aligned}
L(s,\chi_D):=\sum_{n=1}^\infty\left( \frac{D}{n} \right)\frac{1}{n^s}\label{eq:L_s_chi_D_defn}\end{aligned}$$where $ \left(\frac{\cdot}{ n}\right)$ is the Jacobi–Kronecker symbol.
In particular, for $ s=2$, the right-hand side of Eq. \[eq:Gs\_zi\_asympt\] involves Catalan’s constant $ G:=\sum_{n=0}^\infty(-1)^n(2n+1)^{-2}=\frac{1}{2}\int_0^\infty y/(\cosh y)\operatorname{d}y$. Noting that Eq. \[eq:rho2PSL2Z\_Eisenstein\] entails $ \rho^{\mathfrak H/PSL(2,\mathbb Z)}_2(\zeta,i)=2j(\zeta)/[j(\zeta)-1728]=2j(\zeta)\Delta(\zeta)/[E_{6}(\zeta)]^{2}$, one may combine Eqs. \[eq:G2PSL2Z\_Eisenstein\_form\] and \[eq:Epstein\_s2\_int\_star\] into the following identity:$$\begin{aligned}
&
G_2^{\mathfrak H/PSL(2,\mathbb Z)}(i,z)=-\frac{40 G}{\pi y}-3456\pi^{2}\operatorname{Re}\int_{z/i}^\infty\frac{E_4(it)\Delta(it)}{[E_6(it)]^{2}}\frac{(t+ix)^2- y^2}{y}\operatorname{d}t,\quad j(z)\neq 1728,\label{eq:G2_PSL2Z_z_i}\end{aligned}$$which incorporates the integral representation of $ G_2^{\mathfrak H/PSL(2,\mathbb Z)}(i,i\sqrt{2})$ (Eq. \[eq:KZ\_laconic\]) mentioned in the survey of Kontsevich–Zagier [@KontsevichZagier] as a special case.
So far, for an arbitrary pair of non-equivalent points $ z,z'\in\mathfrak H$, we have furnished integral representations for weight-4 automorphic Green’s functions on the congruence subgroups $ PSL(2,\mathbb Z)=\overline \varGamma_0(1),$ $\overline \varGamma_0(2),$ $\overline \varGamma_0(3)$ and $\overline \varGamma_0(4)$, in the spirit of Kontsevich and Zagier. This completes the proof of Theorem \[thm:KZ\_int\_repns\]\[itm:KZ\_b\], as well as the $ k=4$ case of Theorem \[thm:KZ\_int\_repns\]\[itm:KZ\_a\]. The Gross–Kohnen–Zagier algebraicity conjecture (boxed equation in §\[subsec:background\]) can be accessed by the analysis of the corresponding integral representations of automorphic Green’s functions at all the CM points. One may wish to carry on Mellit’s algebro-geometric proof of the claim $ \exp(G_2^{\mathfrak H/PSL(2,\mathbb Z)}(z,i)\operatorname{Im}z)\in\overline{\mathbb Q}$ at CM values of $ z$ [@MellitThesis] to the analysis of other integral representations in this subsection (as well as the rest of Theorem \[thm:KZ\_int\_repns\]), in an attempt to verify the Gross–Kohnen–Zagier algebraicity conjecture for all the cusp-form-free scenarios. Whilst it is plausible to pursue such an abstract approach to the Gross–Kohnen–Zagier algebraicity of automorphic Green’s functions in the language of Arakelov intersections and Chow groups for elliptic curves [@SWZhang1997; @MellitThesis] as well as Mellit’s criterion of “geometric representability” [@MellitThesis], it is beyond the scope of this article. In the rest of this paper (see, in particular, §§\[subsec:add\_form\_Legendre\_Ramanujan\], \[subsec:int\_repn\_G2\_GZ\_rn\] and \[subsec:G2\_Hecke4\_GZ\_rn\]), we will only evaluate a special subset of weight-4 automorphic Green’s functions, using explicit and constructive methods.
Addition Formulae and Legendre–Ramanujan Representations for Automorphic Green’s Functions\[subsec:add\_form\_Legendre\_Ramanujan\]
-----------------------------------------------------------------------------------------------------------------------------------
The automorphic Green’s functions on different congruence subgroups are related to each other by some “addition formulae”. Apart from the congruence subgroups $ \varGamma_0(N)$ ($ N\in\{1,2,3,4\}$) studied in §\[subsec:KZ\_integrals\], it is sometimes useful to consider the $ \varLambda$-group and $ \vartheta$-group, so as to facilitate the statements for certain types of “addition formulae”.
Here, the $ \varLambda$-group $\varLambda\equiv \varGamma(2)$ is identical to the principal congruence subgroup of level 2, where [@Shimura1994 Eq. 1.6.1]: $$SL(2,\mathbb Z)\supseteq \varGamma_0(N)\supseteq\varGamma(N):=\left\{\left.\begin{pmatrix}Na+1&Nb\\Nc&Nd+1\end{pmatrix}\right|a,b,c,d\in\mathbb Z,(Na+1)(Nd+1)-N^{2}bc=1\right\}.$$The $ \varLambda$-group characterizes the symmetry of the modular lambda function $ \lambda(z):=\frac{16[\eta (z/2)]^8 [\eta (2 z)]^{16}}{[\eta (z)]^{24}},$ $ z\in\mathfrak H $, as $ \lambda(\hat \gamma z)=\lambda(z)$ for any $ \hat \gamma\in\varLambda$. The canonical isomorphism $ \varGamma_0(4)\longrightarrow \varGamma(2)$ is induced by the duplication map $ \left( \begin{smallmatrix}2&0\\0&1\end{smallmatrix} \right):z\mapsto 2z$ [@Zagier2008Mod123 p. 28]. Therefore, one has (see Eq. \[eq:alpha2\_alpha4\]) $$\begin{aligned}
G_s^{\mathfrak H/\overline{\varGamma}_{0}(4)}(z_1,z_2)=G_{s}^{\mathfrak H/\overline{\varGamma}(2)}(2z_{1},2z_{2}),\quad z_1\in\mathfrak H\smallsetminus(\varGamma_0(4)z_{2}),\operatorname{Re}s>1.\label{eq:Hecke4_add}\end{aligned}$$
The function $ 1-\alpha_2((z-1)/2)=1/[1-2\lambda(z)]^{2}$ (see Eq. \[eq:alpha2\_alpha4\]) is invariant under the transformations $\hat T^2:z\mapsto z+2$ and $\hat S:z\mapsto -1/z$, which are generators of the projective $ \vartheta$-group $ \overline{\varGamma}_{\vartheta}$ [@Schoeneberg p. 85]. The $ \vartheta$-group $ \varGamma_{\vartheta}$ is related to the Hecke congruence group $ \varGamma_0(2)$ by a conjugation [@Schoeneberg p. 86]:$$\begin{aligned}
\begin{pmatrix}1 & -1 \\
1 & 0 \\
\end{pmatrix}\varGamma_0(2)=\varGamma_{\vartheta}\begin{pmatrix}1 & -1 \\
1 & 0 \\
\end{pmatrix}
.\label{eq:theta_Hecke2_conj}\end{aligned}$$Naturally, one can show that$$\begin{aligned}
G_{s}^{\mathfrak H/\overline{\varGamma}_\vartheta}(z,z')=G_s^{\mathfrak H/\overline{\varGamma}_0(2)}\left( -\frac{1}{z-1},-\frac{1}{z'-1} \right)=G_s^{\mathfrak H/\overline{\varGamma}_0(2)}\left( \frac{z-1}{2} ,\frac{z'-1}{2}\right),\label{eq:GsTheta_GsHecke2_reln}\end{aligned}$$by invoking the conjugation relation $ \varGamma_0(2)\cong \varGamma_\vartheta$ (Eq. \[eq:theta\_Hecke2\_conj\]) as well as the Fricke involution on $ \varGamma_0(2)\backslash\mathfrak H^*$ (Eq. \[eq:Fricke\_inv\_HeckeN\_GZ\]).
\[prop:add\_form\_auto\_Green\]We have the following algebraic identities whenever all the summands are finite:$$\begin{aligned}
G_s^{\mathfrak H/PSL(2,\mathbb Z)}(z,z')={}&G_s^{\mathfrak H/\overline{\varGamma}_\vartheta}(z,z')+G_s^{\mathfrak H/\overline{\varGamma}_\vartheta}\left(-\frac{z+1}{z},z'\right)+G_s^{\mathfrak H/\overline{\varGamma}_\vartheta}(z+1,z'),
\label{eq:SL2Z_Theta_add}\\
G_s^{\mathfrak H/PSL(2,\mathbb Z)}(z,z')={}&G_s^{\mathfrak H/\overline{\varGamma}_{0}(3)}(z,z')+G_s^{\mathfrak H/\overline{\varGamma}_{0}(3)}\left( -\frac{1}{z} ,z'\right)+G_s^{\mathfrak H/\overline{\varGamma}_{0}(3)}\left( -\frac{1}{z+1} ,z'\right)+G_s^{\mathfrak H/\overline{\varGamma}_{0}(3)}\left( -\frac{1}{z-1} ,z'\right),\label{eq:Gs_Hecke3_add_G2_PSL}\\G_s^{\mathfrak H/PSL(2,\mathbb Z)}(z,z')={}&G_s^{\mathfrak H/\overline{\varGamma}(2)}(z,z')+G_s^{\mathfrak H/\overline{\varGamma}(2)}\left(-\frac{1}{z},z'\right)+G_s^{\mathfrak H/\overline{\varGamma}(2)}\left(-\frac{z+1}{z},z'\right)+G_s^{\mathfrak H/\overline{\varGamma}(2)}\left(\frac{z\vphantom{1}}{z+1},z'\right)\notag\\& +G_s^{\mathfrak H/\overline{\varGamma}(2)}(z+1,z')+G_s^{\mathfrak H/\overline{\varGamma}(2)}\left(-\frac{1}{z+1},z'\right),\label{eq:SL2Z_Gamma2_add}
\\G_s^{\mathfrak H/\overline{\varGamma}_{0}(2)}(z,z')={}&G_s^{\mathfrak H/\overline{\varGamma}(2)}(z,z')+G_s^{\mathfrak H/\overline{\varGamma}(2)}(z+1,z'),\label{eq:Hecke2_Gamma2_add}\\G_s^{\mathfrak H/\overline{\varGamma}_\vartheta}(z,z')={}&G_s^{\mathfrak H/\overline{\varGamma}(2)}(z,z')+G_s^{\mathfrak H/\overline{\varGamma}(2)}\left(-\frac1z,z'\right).\label{eq:G_s_Theta_add_form}\end{aligned}$$
One simply goes through criteria \[itm:AGF1’\]–\[itm:AGF3’\] for the right-hand side of each proposed formula, with respect to the variable $z $. The details of these routine procedures are left to the readers.
The numbers of terms on both sides of Eqs. \[eq:SL2Z\_Theta\_add\]–\[eq:SL2Z\_Gamma2\_add\] are consistent with the indices of subgroups $ [\varGamma_0(1):\varGamma_0(2)]=3$, $ [\varGamma_0(1):\varGamma_0(3)]=4$ and $ [\varGamma_0(1):\varGamma_0(4)]=6$ (see Fig. \[fig:fundomain\]*f*), while the situations in Eqs. \[eq:Hecke2\_Gamma2\_add\] and \[eq:G\_s\_Theta\_add\_form\] are compatible with $[\varGamma_0(2):\varGamma_0(4)] =[\varGamma_0(1):\varGamma_0(4)]/[\varGamma_0(1):\varGamma_0(2)]=2$. If our scope is not restricted to the groups isomorphic to $ \varGamma_0(N)$ ($N\in\{1,2,3,4\}$), then it is also possible to construct other addition formulae in a similar spirit.
The integral representations for $ G_2^{\mathfrak H/\overline\varGamma_0(N)}(z,z')$ ($ N\in\{1,2,3,4\}$) in Eqs. \[eq:G2HeckeN\_unified\] and \[eq:G2PSL2Z\_Eisenstein\_form\] involve the Eisenstein series. One can recast such integral representations in terms of fractional degree Legendre functions $P_{-1/2} $, $ P_{-1/3}$, $ P_{-1/4}$ and $ P_{-1/6}$. The case of $P_{-1/2} $ is directly related to the complete elliptic integral of the first kind $\mathbf K(\sqrt{\lambda})=\frac{\pi}{2}P_{-1/2}(1-2\lambda)=\int_0^{\pi/2}(1-\lambda\cos^2\theta)^{-1/2}\operatorname{d}\theta,\lambda\in\mathbb C\smallsetminus[1,+\infty)$. By convention, for $ \lambda>1$, one defines $ \mathbf K(\sqrt{\lambda}):=\mathbf K(\sqrt{\lambda-i0^+})=\overline{\mathbf K(\sqrt{\lambda+i0^+})}$. The other three Legendre functions $ P_{-1/3}$, $ P_{-1/4}$ and $ P_{-1/6}$ feature prominently in Ramanujan’s elliptic function theory to alternative bases [@RN5 Chap. 33].
For any complex degree $\nu$, the Legendre function of the first kind $P_\nu $ is defined via the Mehler–Dirichlet integral $$\begin{aligned}
P_\nu(1)=1;\quad P_\nu(\cos\theta):=\frac{2}{\pi}\int_0^\theta\frac{\cos\frac{(2\nu+1)\beta}{2}}{\sqrt{\smash[b]{2(\cos\beta-\cos\theta)}}}\operatorname{d}\beta,\quad \theta\in(0,\pi),\nu\in\mathbb C, \label{eq:Pnu_defn_MD}\end{aligned}$$and admits an analytic continuation as a well-defined function on the slit plane: $ P_{\nu }(\xi),\xi\in\mathbb C\smallsetminus(-\infty,-1]$. In particular, for $ -1<\nu<0$, one can use the Euler integral representation of hypergeometric functions:[^5]$$\begin{aligned}
_2F_1\left( \left.\begin{array}{c}
a,b \\
c \\
\end{array}\right|w \right)=\frac{\Gamma(c)}{\Gamma(b)\Gamma(c-b)}\int_0^1t^{b-1}(1-t)^{c-b-1}(1-tw)^{-a}\operatorname{d}t,\quad \operatorname{Re}c>\operatorname{Re}b>0,-\pi<\arg(1-w)<\pi\label{eq:Euler_int}\end{aligned}$$to define $ P_{\nu}(\xi)={_2}F_1\left( \left.\begin{smallmatrix}-\nu,\nu+1\\1\end{smallmatrix}\right| \smash{\frac{1-\xi}{2}}\right),\xi\in\mathbb C\smallsetminus(-\infty,-1] $. Hereinafter, unless explicitly specified otherwise (such as the sign convention for $ \sqrt{\smash[b]{(j(z)-1728)/j(z)}}$), fractional powers of non-zero complex numbers are taken as $ w^\nu:=e^{\nu\log w}$ where $ \log w:=\log|w|+i\arg w$, $ \log|w|\in\mathbb R$ and $ -\pi<\arg w\leq \pi$.
\[prop:G2\_HeckeN\_Pnu\]For degrees $ \nu\in\{-1/2,-1/3,-1/4,-1/6\}$, the function $$\begin{aligned}
R_{\nu}( \xi):={}&\frac{1-\xi^{2}}{P_{\nu}(\xi)}\frac{\operatorname{d}P_{\nu}(\xi)}{\operatorname{d}\xi}+\frac{1-\xi^{2}}{P_{\nu}(-\xi)}\frac{\operatorname{d}P_{\nu}(-\xi)}{\operatorname{d}\xi}-\frac{\sin(\nu\pi)}{\pi}\left\{\frac{1}{ [P_{\nu}(\xi)]^{2}\operatorname{Im}\frac{iP_{\nu}(-\xi)}{P_{\nu}(\xi)}}-\frac{1}{[P_{\nu}(-\xi)]^{2}\operatorname{Im}\frac{iP_{\nu}(\xi)}{P_{\nu}(-\xi)}}\right\},\label{eq:R_nu_defn}\end{aligned}$$is well-defined for $ \xi\in\mathbb C\smallsetminus((-\infty,-1]\cup[1,+\infty))$, and extends continuously to all $ \xi\in\mathbb C$, smoothly across the branch cut of $ P_\nu(\xi),\xi\in\mathbb C\smallsetminus(-\infty,-1]$:$$\begin{aligned}
R_\nu(1)=-R_\nu(-1):=\lim_{\xi\to1}R_\nu(\xi)=0;\quad R_\nu(\xi)=-R_\nu(-\xi):= R_\nu(\xi+i0^+)\equiv R_\nu(\xi-i0^+),\quad \forall \xi>1.\label{eq:R_nu_defn_ext}\end{aligned}$$For degrees $ \nu\in\{-1/2,-1/3,-1/4,-1/6\}$, and the corresponding levels $ N=4\sin^2(\nu\pi)\in\{1,2,3,4\}$, define $$\begin{aligned}
\varrho_{2,\nu}(\xi|z):=\frac{4\alpha_N(z)[1-\alpha_N(z)]}{[\xi-1+2\alpha_N(z)]^{2}}-\frac{R_{\nu}(1-2\alpha_N(z))}{\xi-1+2\alpha_N(z)},\quad \emph{a.e. }\xi\in\mathbb C, z\in\mathfrak H,\label{eq:rho_2_nu_xi_z_defn}\end{aligned}$$and $ \rho_{2,-1/6}(\xi|z)=\varrho_{2,-1/6}(\xi|z)+\varrho_{2,-1/6}(\xi|-1/z)$, then we have an integral representation [$$\begin{aligned}
G_2^{\mathfrak H/PSL(2,\mathbb Z)}(z,z')={}&\frac{\pi}{\operatorname{Im}\frac{iP_{-1/6}(-\sqrt{(j(z')-1728)/j(z')})}{P_{-1/6}(\sqrt{(j(z')-1728)/j(z')})}}\operatorname{Re}\int_{\sqrt{(j(z')-1728)/{j(z')}}}^1[P_{-1/6}(\xi)]^2\rho_{2,-1/6}(\xi|z)\times\notag\\&\times\left[\frac{iP_{-1/6}(-\xi)}{P_{-1/6}(\xi)}-\vphantom{\overline{\frac{\sqrt{j}}{}}}\frac{iP_{-1/6}(-\sqrt{(j(z')-1728)/j(z')})}{P_{-1/6}(\sqrt{(j(z')-1728)/j(z')})}\right]\times\notag\\&\times\left[\frac{iP_{-1/6}(-\xi)}{P_{-1/6}(\xi)}-\overline{\left(\frac{iP_{-1/6}(-\sqrt{(j(z')-1728)/j(z')})}{P_{-1/6}(\sqrt{(j(z')-1728)/j(z')})}\right)}\right]\operatorname{d}\xi\notag\\&+\frac{\pi}{\operatorname{Im}\frac{iP_{-1/6}(-\sqrt{(j(z')-1728)/j(z')})}{P_{-1/6}(\sqrt{(j(z')-1728)/j(z')})}
}\operatorname{Re}\int_{-1}^1[P_{-1/6}(\xi)]^2\rho_{2,-1/6}(\xi|z)\operatorname{d}\xi,\quad \emph{a.e. }z,z'\in\mathfrak H\label{eq:G2_z_z'_arb}
\intertext{and the following integral representations for $ \nu\in\{-1/2,-1/3,-1/4\}$ and $ N=4\sin^2(\nu\pi)\in\{2,3,4\}$:}G_2^{\mathfrak H/\overline{\varGamma}_0(N)}(z,z')={}&\frac{\pi}{\sqrt{N}\operatorname{Im}\frac{iP_\nu(2\alpha_{N}(z')-1)}{P_\nu(1-2\alpha_{N}(z'))}}\operatorname{Re}\int_{1-2\alpha_{N}(z')}^1[P_\nu(\xi)]^2\varrho_{2,\nu}(\xi|z)\times\notag\\&\times\left[\frac{iP_\nu(-\xi)}{P_\nu(\xi)}-\vphantom{\overline{\frac{1}{}}}\frac{iP_\nu(2\alpha_{N}(z')-1)}{P_\nu(1-2\alpha_{N}(z'))}\right]\left[\frac{iP_\nu(-\xi)}{P_\nu(\xi)}-\overline{\left(\frac{iP_\nu(2\alpha_{N}(z')-1)}{P_\nu(1-2\alpha_{N}(z'))}\right)}\right]\operatorname{d}\xi\notag\\&+\frac{\pi}{\sqrt{N}\operatorname{Im}\frac{iP_{\nu}(2\alpha_{N}(z')-1)}{P_\nu(1-2\alpha_{N}(z'))}
}\operatorname{Re}\int_{-1}^1[P_{\nu}(-\xi)]^2\varrho_{2,\nu}(\xi|z)\operatorname{d}\xi,\quad \emph{a.e. }z,z'\in\mathfrak H.\label{eq:G2Hecke234_Pnu}\end{aligned}$$]{}Here in Eqs. \[eq:G2\_z\_z’\_arb\] and \[eq:G2Hecke234\_Pnu\], the paths of integration can be taken as any curves in the double slit plane $ \xi\in \mathbb C\smallsetminus((-\infty,-1]\cup[1,+\infty))$ that miss the singularities of the integrands.
From Ramanujan’s elliptic function theory to alternative bases, one may read off the following identities (see [@RN5 Chap. 33, Theorem 11.6, Theorem 9.11, Corollary 2.11] and [@RN3 Chap. 17, Entries 13(viii)–(ix)]):$$\begin{aligned}
\left[ P_{-1/6}\left( \sqrt{\frac{j(z)-1728}{j(z)}} \right) \right]^{12}={}&\Delta(z)j(z)=[E_{4}(z)]^{3},&&\operatorname{Im}z>0,|\operatorname{Re}z|<\frac{1}{2},|z+1|>1,|z-1| >1;\label{eq:P_sixth_eta}\\{[}P_\nu(1-2\alpha_N(z))]^2={}&\frac{NE_{2}(Nz)-E_2(z)}{N-1},&&\operatorname{Im}z>0,|\operatorname{Re}z|<\frac{1}{2},\left\vert z+\frac{1}{N} \right\vert>\frac{1}{N},\left\vert z-\frac{1}{N} \right\vert>\frac{1}{N}.\label{eq:P_nu_sqr_E2_diff}\end{aligned}$$Here, in Eq. \[eq:P\_nu\_sqr\_E2\_diff\], the degree $ \nu\in\{-1/2,-1/3,-1/4\}$ corresponds to level $ N=4\sin^2(\nu\pi)\in\{2,3,4\}$. Referring back to Eqs. \[eq:alpha\_N\_deriv\_E2\] and \[eq:alpha\_N\_ratio\_ids\], we see that with $ \nu\in\{-1/2,-1/3,-1/4,-1/6\}$, the relation $ P_\nu(\xi)\neq0$ holds for $ \xi\in\mathbb C\smallsetminus((-\infty,-1]\cup[1,+\infty))$. Thus, Eq. \[eq:R\_nu\_defn\] is well defined.
By taking ratios of a formula (any one among Eq. \[eq:P\_sixth\_eta\] and the three forms of Eq. \[eq:P\_nu\_sqr\_E2\_diff\]) at a pair of points $ z$ and $ -1/(Nz)$, where $ N\in\{1,2,3,4\}$, one can verify the following formulae:$$\begin{aligned}
z=\frac{i P_\nu(2\alpha_N(z)-1)}{\sqrt{N}P_\nu(1-2\alpha_N(z))},&&
\operatorname{Im}z>0,|\operatorname{Re}z|<\frac{1}{2},\left\vert z+\frac{1}{N} \right\vert>\frac{1}{N},\left\vert z-\frac{1}{N} \right\vert>\frac{1}{N}\label{eq:z_Pnu_ratios}\end{aligned}$$ after extracting appropriate roots, and recalling that $E_2(-1/(Nw))=N^2 w^2 E_2(Nw) $, $ \Delta(-1/z)=z^{12}\Delta(z)$. With the Legendre differential equation and special values of the Legendre function, one can verify the following differentiation formula that appeared in Ramanujan’s notebooks [@RN2 p. 88]: $$\begin{aligned}
\frac{\operatorname{d}}{\operatorname{d}\xi}\frac{P_{\nu }(-\xi)}{P_{\nu }(\xi)}=-\frac{2}{\pi}\frac{\sin(\nu\pi)}{(1-\xi^{2})[P_{\nu }(\xi)]^2},\quad \xi\in\mathbb C\smallsetminus((-\infty,-1]\cup[1,+\infty)).\label{eq:P_nu_ratio_deriv}\end{aligned}$$ Using the information above, together with Ramanujan’s differential equation for the Eisenstein series [@Ramanujan1916 Eq. 30], one can verify the following identities:$$\begin{aligned}
R_{-1/6}\left( \sqrt{\frac{j(z)-1728}{j(z)}}
\right)\sqrt{\frac{j(z)-1728}{j(z)}}={}&-\frac{E_{6}(z)}{3[E_{4}(z)]^{2}}\left[ E_2(z)-\frac{ E_6(z)}{E_4(z)} \right],\label{eq:R_sixth_Eisenstein}\\ R_{\nu}(1-2\alpha_N(z))={}&-\frac{N-1}{6}\frac{N^{2}[E_{2}(Nz)]^2-N^{2}E_{4}(N z)-[E_{2}(z)]^2+E_{4}(z)}{[NE_{2}(Nz)-E_{2}(z)]^{2}},\label{eq:Rnu_Eisenstein}\end{aligned}$$ where $\nu\in\{-1/2,-1/3,-1/4\}$ corresponds to $ N=4\sin^2(\nu\pi)\in\{2,3,4\}$. This proves the continuous extensibility of $R_{\nu}$ for $\nu\in\{-1/2,-1/3,-1/4,-1/6\}$ (Eq. \[eq:R\_nu\_defn\_ext\]). Upon variable substitutions according to Eq. \[eq:z\_Pnu\_ratios\], one can show that $ \rho_{2,-1/6}(1-2\alpha_1(\zeta)|z)=\rho_2^{\mathfrak H/PSL(2,\mathbb Z)}(\zeta,z)$ as well as $ \varrho_{2,\nu}(1-2\alpha_N(\zeta)|z)=\varrho_2^{\mathfrak H/\overline\varGamma_0(N)}(\zeta,z)$ for $ \nu\in\{-1/2,-1/3,-1/4\}$ and $N=4\sin^2(\nu\pi)\in\{2,3,4\} $. This finally reveals the analytic connections between Eqs. \[eq:G2\_z\_z’\_arb\] and \[eq:G2PSL2Z\_Eisenstein\_form\] (resp. \[eq:G2Hecke234\_Pnu\] and \[eq:G2HeckeN\_unified\]) .
If $z$ is a CM point, then one can use the explicit formulae in Eqs. \[eq:R\_sixth\_Eisenstein\] and \[eq:Rnu\_Eisenstein\] to verify that the respective values of $ R_\nu$ are solvable algebraic numbers (see Eq. \[eq:fns\_alg\_val\_at\_CM\_pts\] and Appendix \[app:algebraicity\]). Consequently, the related expression $ \varrho_\nu(\xi|z)$ will be an algebraic function of $ \xi$ with algebraic coefficients. Recalling the Euler integral representation for hypergeometric functions from Eq. \[eq:Euler\_int\], we see that Eqs. \[eq:G2\_z\_z’\_arb\] and \[eq:G2Hecke234\_Pnu\] (up to an overall factor that is an integer power of $ \pi$) are integrals of algebraic functions over algebraic domains, if both $ z$ and $z'$ are CM points. This qualifies them as Kontsevich–Zagier periods [@KontsevichZagier §1.1]. As pointed out in the survey of Kontsevich and Zagier [@KontsevichZagier §3.4], the conjectural relation $ \sqrt{y_1y_2}G_2^{\mathfrak H/\overline{\varGamma}_0(N)}(z_1,z_2)\in\log\overline{\mathbb Q}$ (where $ N\in\{1,2,3,4\}$, $ [\mathbb Q(z_1):\mathbb Q]=[\mathbb Q(z_2):\mathbb Q]=2$, $ z_1\notin\varGamma_0(N)z_2$) is a type of “period identity”.
\[rmk:spec\_val\_G2\]Now we showcase a few examples where the “period identity at CM points” $ \sqrt{y_1y_2}$ $G_2^{\mathfrak H/\overline{\varGamma}_0(N)}(z_1,z_2)\in\log\overline{\mathbb Q}$ can be proved by elementary manipulations of integrals.
For $ \nu\in\{-1/6,-1/4,-1/3\}$, the integral representations in Eqs. \[eq:G2\_z\_z’\_arb\] and \[eq:G2Hecke234\_Pnu\] immediately bring us some special values of automorphic Green’s functions:$$\begin{aligned}
G_{2}^{\mathfrak H/PSL(2,\mathbb Z)}\left(\frac{1+i\sqrt{3}}{2},\frac{i}{\sqrt{1}}\right)={}&-\frac{8\pi}{3}\int_{0}^{1}[P_{-1/6}(\xi)]^2\operatorname{d}\xi=-\frac{12}{\sqrt{3}}\log(2+\sqrt{3}),\label{eq:G2_ell3_i_spec_val}\\G_2^{\mathfrak H/\overline{\varGamma}_{0}(2)}\left( \frac{i-1}{2},\frac{i}{\sqrt{2}} \right)={}&-\frac{\pi}{\sqrt{2}}\int_{0}^{1}[P_{-1/4}(\xi)]^2\operatorname{d}\xi=-\frac{4}{\sqrt{2}}\log(1+\sqrt{2}),\label{eq:G2_Hecke2_spec_val}\\G_2^{\mathfrak H/\overline{\varGamma}_{0}(3)}\left( \frac{3+i\sqrt{3}}{6},\frac{i}{\sqrt{3}} \right)={}&-\frac{2\pi}{3\sqrt{3}}\int_{0}^{1}[P_{-1/3}(\xi)]^2\operatorname{d}\xi=-2\log2.\label{eq:G2_Hecke3_spec_val}\end{aligned}$$In all these three cases of $ G_2^{\mathfrak H/\overline\varGamma_0(N)}(z,z')$, we have $ \alpha_N(z)=\infty$ (see Eqs. \[eq:alpha\_inf\_rho\_lim\] and \[eq:alpha\_inf\_rho\_lim\_PSL2Z\]) and $ \alpha_N(z')=1/2$, so it is clear that the values of the respective automorphic Green’s functions are certain algebraic multiples of $$\begin{aligned}
\pi\int_0^1[P_\nu(\xi)]^2\operatorname{d}\xi=\frac{\pi}{2 \nu +1}\left\{1+\frac{\sin (\nu\pi )}{\pi} \left[ \psi ^{(0)}\left(\frac{\nu+2 }{2}\right)-\psi ^{(0)}\left(\frac{\nu+1 }{2}\right )\right]\right\},\quad \nu\in\mathbb C\smallsetminus\{-1/2\}.\label{eq:Pnu_sqr_0to1}\end{aligned}$$Here, the logarithmic derivatives of Euler’s gamma function are the polygamma functions $ \psi^{(m)}(w):=\operatorname{d}^{m+1}\log\Gamma(w)/\operatorname{d}w^{m+1}$ for $ m\in\mathbb Z_{\geq0}$. The integral formula in Eq. \[eq:Pnu\_sqr\_0to1\] (see [@GradshteynRyzhik item 7.113.1]) can be proved by simple applications of the Legendre differential equations. For $ w\in\mathbb Q\cap(0,1)$, one can evaluate $ \psi^{(0)}(w)$ using the explicit formula provided by the digamma theorem of Gau[ß]{} [@HTF1 §1.7.3]. This leads to the logarithmic expressions in Eqs. \[eq:G2\_ell3\_i\_spec\_val\]–\[eq:G2\_Hecke3\_spec\_val\].
Exploiting the addition formulae in Proposition \[prop:add\_form\_auto\_Green\], one can deduce a few special values of $ G_2^{\mathfrak H/\overline\varGamma_0(4)}(z/2,i/\sqrt{4})=G_2^{\mathfrak H/\overline\varGamma(2)}(z,i)=G_2^{\mathfrak H/\overline\varGamma_{\vartheta}}(z,i)/2=G_2^{\mathfrak H/\overline\varGamma_{0}(2)}((z-1)/2,(i-1)/2)/2$ (see Eqs. \[eq:Hecke4\_add\], \[eq:G\_s\_Theta\_add\_form\] and \[eq:GsTheta\_GsHecke2\_reln\]) that are related to Eqs. \[eq:G2\_ell3\_i\_spec\_val\] and \[eq:G2\_Hecke2\_spec\_val\]:[$$\begin{aligned}
-\frac{2}{\sqrt{3}}\log(2+\sqrt{3})={}&\frac{G_2^{\mathfrak H/\overline{\varGamma}_0(2)}(e^{\pi i/3},i)}{2}\notag\\={}&G_2^{\mathfrak H/\overline{\varGamma}(2)}(e^{\pi i/3},i)=G_2^{\mathfrak H/\overline{\varGamma}(2)}(i\sqrt{3},i)=G_2^{\mathfrak H/\overline{\varGamma}(2)}(e^{2\pi i/3},i)=G_2^{\mathfrak H/\overline{\varGamma}(2)}(i/\sqrt{3},i),\label{eq:log2_plus_sqrt3_more}\\
-\frac{2}{\sqrt{2}}\log(1+\sqrt{2})={}&G_2^{\mathfrak H/\overline\varGamma(2)}(-1+i\sqrt{2},i)=G_2^{\mathfrak H/\overline\varGamma(2)}\left(\frac{-1+i\sqrt{2}}{3},i\right)\notag\\={}&G_2^{\mathfrak H/\overline\varGamma(2)}\left(\frac{1+i\sqrt{2}}{3},i\right)=G_2^{\mathfrak H/\overline\varGamma(2)}(1+i\sqrt{2},i).\label{eq:G2Gamma2spec_val}\end{aligned}$$]{}The main idea here is the following four-fold symmetry:$$\begin{aligned}
G_s^{\mathfrak H/\overline\varGamma(2)}(z,i)=G_s^{\mathfrak H/\overline\varGamma(2)}\left( \frac{1+z}{1-z} ,i\right)=G_s^{\mathfrak H/\overline\varGamma(2)}\left( -\frac{1}{z} ,i\right)= G_s^{\mathfrak H/\overline\varGamma(2)}\left( \frac{z-1}{z+1} ,i\right),\label{eq:GsGamma2i_D4}\end{aligned}$$which descends from the identity $G_s^{\mathfrak H/\overline\varGamma(2)}(z,i) =G_s^{\mathfrak H/\overline\varGamma_0(4)}(z/2,i/2)=G_s^{\mathfrak H/\overline\varGamma_0(4)}(-1/(2z),i/2)$ $=G_s^{\mathfrak H/\overline\varGamma(2)}(-1/z,i) $ (see Eq. \[eq:Fricke\_inv\_HeckeN\_GZ\]) along with Eqs. \[eq:GsTheta\_GsHecke2\_reln\] and \[eq:G\_s\_Theta\_add\_form\].
One can derive a special addition formula$$\begin{aligned}
G_s^{\mathfrak H/PSL(2,\mathbb Z)}(z,i)=2[G_s^{\mathfrak H/\overline{\varGamma}(2)}(z,i)+G_s^{\mathfrak H/\overline{\varGamma}(2)}(z+1,i)+G_s^{\mathfrak H/\overline{\varGamma}(2)}(2z+1,i)]\label{eq:GsPSL2Zi_GsGamma2i_add_form}\end{aligned}$$from Eqs. \[eq:SL2Z\_Gamma2\_add\] and \[eq:GsGamma2i\_D4\].
Recalling the definition of the $L$-function $ L(s,\chi_D)$ from Eq. \[eq:L\_s\_chi\_D\_defn\], and that$$\begin{aligned}
G\equiv L(2,\chi_{-4})={}&\sum_{\ell=0}^\infty\frac{(-1)^\ell}{(2\ell+1)^2}=\frac{\psi ^{(1)}\left(\frac{1}{4}\right)-\psi ^{(1)}\left(\frac{3}{4}\right)}{16},\label{eq:L2chi_4_defn_G}\\
L(2,\chi_{-3})={}&\sum_{\ell=0}^\infty\left[ \frac{1}{(3\ell+1)^{2}} -\frac{1}{(3\ell+2)^{2}}\right]=\frac{\psi ^{(1)}\left(\frac{1}{3}\right)-\psi ^{(1)}\left(\frac{2}{3}\right)}{9},\label{eq:L2chi_3_defn}\\L(2,\chi_{-8})=L(2,\chi_{-2})={}&\sum _{n=0}^{\infty } \frac{(-1)^{ n (n-1)/2}}{(2 n+1)^2}=\frac{\psi ^{(1)}\left(\frac{1}{8}\right)+\psi ^{(1)}\left(\frac{3}{8}\right)-\psi ^{(1)}\left(\frac{5}{8}\right)-\psi ^{(1)}\left(\frac{7}{8}\right)}{64},\end{aligned}$$ we may state and prove some integral identities involving fractional degree Legendre functions and these special $L$-values.
\[prop:Pnu\_sec\_L\_weight4\]We have the following identities:[$$\begin{aligned}
L(2,\chi_{-4})={}&-\frac{\pi}{40}\lim_{y'\to+\infty}G_2^{\mathfrak H/PSL(2,\mathbb Z)}(i,z')y'=-\frac{\pi^{2}}{20}\operatorname{Re}\int_{-1}^1\frac{[P_{-1/6}(\xi)]^2}{\xi^{2}}\operatorname{d}\xi\notag\\={}&-\frac{\pi}{16}\lim_{y'\to+\infty}G_2^{\mathfrak H/\overline\varGamma_{\vartheta}}(i,z')y'=\frac{\pi^{2}}{32}\int_{-1}^1[P_{-1/4}(\xi)]^2\operatorname{d}\xi\notag\\={}&-\frac{\pi}{8}\lim_{y'\to+\infty}G_2^{\mathfrak H/\overline\varGamma(2)}(i,z')y'=-\frac{\pi^{2}}{16}\operatorname{Re}\int_{-1}^1\frac{[P_{-1/2}(\xi)]^2}{\xi^2}\operatorname{d}\xi;\label{eq:L2chi_4_Pnu}\\ L(2,\chi_{-3})={}&-\frac{\pi}{45}\lim_{y'\to+\infty}G_{2}^{\mathfrak H/PSL(2,\mathbb Z)}\left( \frac{1+i\sqrt{3}}{2},z'\right)y'=\frac{4\pi^{2}}{135}\int_{-1}^1[P_{-1/6}(\xi)]^2\operatorname{d}\xi\notag\\={}&-\frac{2\pi}{9}\lim_{y'\to+\infty}G_{2}^{\mathfrak H/\overline{\varGamma}_0(3)}\left( \frac{3+i\sqrt{3}}{6},z'\right)y'=\frac{2\pi^2}{81}\int_{-1}^1[P_{-1/3}(\xi)]^2\operatorname{d}\xi\notag\\={}&-\frac{2\pi}{135}\lim_{y'\to+\infty}G_2^{\mathfrak H/PSL(2,\mathbb Z)}(i\sqrt{3},z')y'\notag\\={}&-\frac{56\pi}{135}\lim_{y'\to+\infty}G_2^{\mathfrak H/\overline{\varGamma}_{0}(3)}\left( \frac{i}{\sqrt{3}},z'\right)y'=-\frac{4\pi^{2}}{81}\operatorname{Re}\int_{-1}^1\frac{[P_{-1/3}(\xi)]^2}{\xi^2}\operatorname{d}\xi;\label{eq:L2chi3}\\ L(2,\chi_{-2})={}&-\frac{\pi}{40}\lim_{y'\to+\infty}G_2^{\mathfrak H/PSL(2,\mathbb Z)}(i\sqrt{2},z')y'=-\frac{\pi}{16}\lim_{y'\to+\infty}G_2^{\mathfrak H/\overline\varGamma_{\vartheta}}(1+i\sqrt{2},z')y'\notag\\={}&-\frac{\pi}{16}\lim_{y'\to+\infty}G_2^{\mathfrak H/\overline\varGamma_{0}(2)}\left( \frac{i}{\sqrt{2}},\frac{z'-1}{2} \right)y'=-\frac{\pi^2}{16}\operatorname{Re}\int_{-1}^1\frac{[P_{-1/4}(\xi)]^2}{\xi^2}\operatorname{d}\xi
.\label{eq:L2chi_2_Pnu}\end{aligned}$$]{}Here, all the integration paths are taken as curves in the slit plane $ \mathbb C\smallsetminus(-\infty,-1]$ that circumvent the singularities of the integrands.
It is clear that all these integrals involving Legendre functions can be associated with the claimed asymptotic behavior of automorphic Green’s functions, by Eqs. \[eq:G2\_z\_z’\_arb\] and \[eq:G2Hecke234\_Pnu\]. What remains to be shown are their connections to the special $ L$-values. We shall explain such connections for Eq. \[eq:L2chi3\] in detail and sketch the proof for the rest.
The first equality in Eq. \[eq:L2chi3\] follows from Eq. \[eq:Gs\_asympt\] and an integration of Ramanujan’s formula (see [@BorweinBorweinCubic], [@Berndt1992] or [@NTRamanujan Theorem 3.7.10])$$\begin{aligned}
\sum_{\substack{m,n\in\mathbb Z\\m^2+n^2\neq0}}ye^{-(m^{2}+mn+n^2)y}=6\sum_{\ell=0}^\infty\left[\frac{y}{e^{(3\ell+1)y}-1} -\frac{y}{e^{(3\ell+2)y}-1}\right]\end{aligned}$$over $ y\in(0,+\infty)$. The first two lines in Eq. \[eq:L2chi3\] are related to each other by an addition formula for $ z\in\mathfrak H\smallsetminus(SL(2,\mathbb Z)\frac{1+i\sqrt{3}}{2})$:$$\begin{aligned}
&
G_2^{\mathfrak H/PSL(2,\mathbb Z)}\left(z,\frac{1+i\sqrt{3}}{2}\right)\notag\\={}&G_2^{\mathfrak H/\overline{\varGamma}_{0}(3)}\left( z ,\frac{3+i\sqrt{3}}{6}\right)+G_2^{\mathfrak H/\overline{\varGamma}_{0}(3)}\left( \frac{z}{3} ,\frac{3+i\sqrt{3}}{6}\right)+G_2^{\mathfrak H/\overline{\varGamma}_{0}(3)}\left( \frac{z+1}{3} ,\frac{3+i\sqrt{3}}{6}\right)+G_2^{\mathfrak H/\overline{\varGamma}_{0}(3)}\left( \frac{z-1}{3} ,\frac{3+i\sqrt{3}}{6}\right),\label{eq:G2_Hecke3_add_G2_PSL}\end{aligned}$$ which can be proved by applying the Fricke involution (Eq. \[eq:Fricke\_inv\_HeckeN\_GZ\]) to the last three terms on the right-hand side of Eq. \[eq:Gs\_Hecke3\_add\_G2\_PSL\]. In particular, Eq. \[eq:G2\_Hecke3\_add\_G2\_PSL\] entails the ratio between the asymptotic behavior of two automorphic Green’s functions: $$\begin{aligned}
\lim_{y\to+\infty}G_2^{\mathfrak H/PSL(2,\mathbb Z)}\left(z,\frac{1+i\sqrt{3}}{2}\right)y=(1+3+3+3)\lim_{y\to+\infty}G_2^{\mathfrak H/\overline{\varGamma}_{0}(3)}\left( z ,\frac{3+i\sqrt{3}}{6}\right)y.\label{eq:PSL_Hecke3_asympt_conv}\end{aligned}$$ The third line in Eq. \[eq:L2chi3\] is related to $ L(2,\chi_{-3})$ by Eq. \[eq:Gs\_asympt\] and another generalization of the two-squares theorem (Eq. \[eq:two\_squares\]) due to Ramanujan [@NTRamanujan p. 75, Eq. 3.7.8]:$$\begin{aligned}
\sum_{\substack{m,n\in\mathbb Z\\m^2+n^2\neq0}}ye^{-(3m^{2}+n^2)y}=2\sum_{\ell=0}^\infty y\left( \frac{1}{e^{(3\ell+1)y}-1}- \frac{1}{e^{(3\ell+2)y}-1}\right)+4\sum_{\ell=0}^\infty y\left( \frac{1}{e^{4(3\ell+1)y}-1}- \frac{1}{e^{4(3\ell+2)y}-1}\right).\end{aligned}$$ The last line in Eq. \[eq:L2chi3\] follows from an addition formula$$\begin{aligned}
&
G_2^{\mathfrak H/PSL(2,\mathbb Z)}(z,i\sqrt{3})=G_2^{\mathfrak H/\overline{\varGamma}_{0}(3)}\left( z ,\frac{i}{\sqrt{3}}\right)+G_2^{\mathfrak H/\overline{\varGamma}_{0}(3)}\left( \frac{z}{3} ,\frac{i}{\sqrt{3}}\right)+G_2^{\mathfrak H/\overline{\varGamma}_{0}(3)}\left( \frac{z+1}{3} ,\frac{i}{\sqrt{3}}\right)+G_2^{\mathfrak H/\overline{\varGamma}_{0}(3)}\left( \frac{z-1}{3} ,\frac{i}{\sqrt{3}}\right),\label{eq:G2PSL2Zsqrt3}\end{aligned}$$which can be verified in a similar vein as Eq. \[eq:G2\_Hecke3\_add\_G2\_PSL\].
The readers may fill in the details for Eqs. \[eq:L2chi\_4\_Pnu\] and \[eq:L2chi\_2\_Pnu\] by referring back to the addition formulae in Eqs. \[eq:SL2Z\_Theta\_add\], \[eq:G\_s\_Theta\_add\_form\] and \[eq:GsPSL2Zi\_GsGamma2i\_add\_form\], as well as Ramanujan’s generalization of the two-squares theorem to the sum $|i\sqrt{2}m+n|^2=2m^2+n^2 $ [@NTRamanujan Theorems 3.7.2 and 3.7.3].
The following integral formula is a standard result [@GradshteynRyzhik item 7.112.3]:$$\begin{aligned}
\label{eq:P_nu_sqr}\int_{-1}^1[P_\nu(\xi)]^2\operatorname{d}\xi={}&\frac{2}{2\nu+1}\left[1-\frac{2\sin^2(\nu\pi)}{\pi^{2}}\psi^{(1)}(\nu+1)\right],\quad \nu\in\mathbb C\smallsetminus(\mathbb Z_{<0}\cup\{-1/2\}),\end{aligned}$$and can be proved by Legendre differential equations. The first two lines in Eq. \[eq:L2chi3\] imply the following identity$$\begin{aligned}
5\int_{-1}^1[P_{-1/3}(\xi)]^2\operatorname{d}\xi=6\int_{-1}^1[P_{-1/6}(\xi)]^2\operatorname{d}\xi.\label{eq:5third_6sixth}\end{aligned}$$This can also be verified by spelling out both sides of Eq. \[eq:5third\_6sixth\] with the help of Eq. \[eq:P\_nu\_sqr\], before invoking the duplication and reflection formulae of $ \psi^{(1)}(w)=\operatorname{d}^2\log\Gamma(w)/\operatorname{d}w^2$: $$\begin{aligned}
4\psi^{(1)}(2w)={}&\psi^{(1)}(w)+\psi^{(1)}\left(w+\frac{1}{2}\right)&&\Longrightarrow&&4\psi^{(1)}\left(\frac{1}{3}\right)=\psi^{(1)}\left(\frac{1}{6}\right)+\psi^{(1)}\left(\frac{2}{3}\right),\label{eq:psi1_id1}\\\frac{\pi^{2}}{\sin^2(\pi w)}={}&\psi^{(1)}(w)+\psi^{(1)}(1-w)&&\Longrightarrow&& 4\pi^{2}=\psi^{(1)}\left(\frac{1}{6}\right)+\psi^{(1)}\left(\frac{5}{6}\right)=3\left[ \psi^{(1)}\left(\frac{1}{3}\right) +\psi^{(1)}\left(\frac{2}{3}\right)\right],\label{eq:psi1_id2}\end{aligned}$$which descend from the respective properties of Euler’s gamma function. We will encounter some higher-weight analogs of Eq. \[eq:5third\_6sixth\] later in Proposition \[prop:high\_w\_L\_values\] of §\[subsec:high\_weight\_KZ\].
A generic integral formula for [$$\begin{aligned}
\operatorname{Re}\int_{-1}^1\frac{[P_\nu(\xi)]^2}{\xi^2}\operatorname{d}\xi,\quad -1<\nu<0\end{aligned}$$]{}will be given in Eq. \[eq:PnuPnu\_xx\_int\_unit\_interval\_polygamma\]. The readers are invited to check Eq. \[eq:PnuPnu\_xx\_int\_unit\_interval\_polygamma\] against the special cases $ \nu\in\{-1/2,-1/3,-1/4,-1/6\}$ evaluated in the proposition above.
Kontsevich–Zagier Integrals for Higher Weight Automorphic Green’s Functions\[subsec:high\_weight\_KZ\]
------------------------------------------------------------------------------------------------------
In this subsection, we will treat the five remaining cases where there are no cusp forms on $ \varGamma_0(N)$, namely, $\dim\mathcal S_6(\varGamma_0(2))=\dim\mathcal S_6(\varGamma_0(1))=\dim\mathcal S_8(\varGamma_0(1))=\dim\mathcal S_{10}(\varGamma_0(1))=\dim\mathcal S_{14}(\varGamma_0(1))=0 $.
\[prop:G3Hecke2\]For $z\notin\varGamma_0(2)z'$, we have the following integral representation:$$\begin{aligned}
G_3^{\mathfrak H/\overline{\varGamma}_0(2)}(z,z')={}&\frac{4\pi^{2}}{(y')^{2}}\operatorname{Re}\int_{z'}^{i\infty} \alpha_2(\zeta)[1-\alpha_2(\zeta)][2E_{2}(2\zeta)-E_2(\zeta)]^{3}\varrho^{\mathfrak H/\overline\varGamma_0(2)}_3(\zeta,z)\frac{(\zeta-z')^{2}(\zeta-\overline{z'})^{2}\operatorname{d}\zeta}{i}\notag\\&-\frac{4\pi^{2}}{(y')^{2}}\operatorname{Re}\int_{0}^{i\infty} \alpha_2(\zeta)[1-\alpha_2(\zeta)][2E_{2}(2\zeta)-E_2(\zeta)]^{3}\varrho^{\mathfrak H/\overline\varGamma_0(2)}_3(\zeta,z)\frac{\zeta^{4}\operatorname{d}\zeta}{i},\label{eq:G3Hecke2_KZ_int}\end{aligned}$$where$$\begin{aligned}
\varrho^{\mathfrak H/\overline\varGamma_0(2)}_3(\zeta,z)=\frac{ y^2}{8\pi}\left( \frac{\partial}{\partial y}\frac{1}{y} \right)^2\left\{ \frac{1}{\alpha_{2}(\zeta)-\alpha_2(z)}\frac{1}{[2E_{2}(2z)-E_2(z)]^2} \right\}.\label{eq:rho_3_Hecke2}\end{aligned}$$Here, the invariant $ \alpha_2(z)$ is the same as the one prescribed in Eq. \[eq:alpha\_HeckeN\_unified\] or \[eq:alpha2\_alpha4\], and the paths of integration should avoid the singularities of the integrands.
One can go through criteria \[itm:AGF1’\]–\[itm:AGF3’\] in a similar fashion as what we did in Proposition \[prop:G2HeckeNunified\]. The only non-trivial step is to demonstrate that the integral representation in Eq. \[eq:G3Hecke2\_KZ\_int\] respects the Fricke involution (Eq. \[eq:Fricke\_inv\_HeckeN\_GZ\]). While it is easy to verify $ \varrho^{\mathfrak H/\overline\varGamma_0(2)}_3(\zeta,z)=-\varrho^{\mathfrak H/\overline\varGamma_0(2)}_3(-1/(2\zeta),-1/(2z))$, we still need to show that (see Eq. \[eq:I\_N\_Fricke\_inv\_check\])$$\begin{aligned}
\operatorname{Re}\int_{0}^{i\infty} \alpha_2(\zeta)[1-\alpha_2(\zeta)][2E_{2}(2\zeta)-E_2(\zeta)]^{3}\varrho^{\mathfrak H/\overline\varGamma_0(2)}_3(\zeta,z)\frac{\zeta^{n}\operatorname{d}\zeta}{i}=0,\quad n\in\{1,2,3\}.\label{eq:rho3Hecke2_int0}\end{aligned}$$ For $ n\in\{1,3\}$, the trick used in proving Eq. \[eq:rho2\_int0\] is applicable to Eq. \[eq:rho3Hecke2\_int0\]. It remains to verify Eq. \[eq:rho3Hecke2\_int0\] for the scenario $n=2$. This requires some new techniques, as elaborated in the rest of the proof.
As in the verification of Eq. \[eq:rho2\_int0\], now it would suffice to prove$$\begin{aligned}
\operatorname{Re}\int_{0}^{i\infty} \alpha_2(\zeta)[1-\alpha_2(\zeta)][2E_{2}(2\zeta)-E_2(\zeta)]^{3}\varrho^{\mathfrak H/\overline\varGamma_0(2)}_3(\zeta,z)\frac{\zeta^{2}\operatorname{d}\zeta}{i}=0,\label{eq:G3Hecke2_mid_int0}\end{aligned}$$for all the points $z\in\mathfrak H\cap \partial\mathfrak D_2$ (Fig. \[fig:fundomain\]*c*) satisfying $$|\operatorname{Re}z|=\frac{1}{2},\operatorname{Im}z>\frac{1}{2}\quad \text{or}\quad -\frac{1}{2}<\operatorname{Re}z<0,\left\vert z+\frac{1}{2} \right\vert=\frac{1}{2}\quad\text{or}\quad 0<\operatorname{Re}z<\frac{1}{2},\left\vert z-\frac{1}{2} \right\vert=\frac{1}{2},$$and choosing the contour of integration along the $ \operatorname{Im}\zeta$-axis. To fulfill this task, we shall produce an explicit formula for $$\begin{aligned}
\int_{0}^{i\infty} \frac{\alpha_2(\zeta)[1-\alpha_2(\zeta)][2E_{2}(2\zeta)-E_2(\zeta)]^{3}}{\alpha_{2}(\zeta)-\alpha_2(z)}\frac{\zeta^{2}\operatorname{d}\zeta}{i}\label{eq:alpha2_weight6_int0}\end{aligned}$$by manipulating elliptic integrals.
The following (degree-1) modular transformations for the complete elliptic integrals of the first kind are well known.
1. Imaginary modulus transformation:$$\begin{aligned}
\mathbf K(\sqrt{\lambda})=\frac{1}{\sqrt{1-\lambda}}\mathbf K\left( \sqrt{\frac{\lambda}{\lambda-1}} \right),\quad \lambda\in\mathbb C\smallsetminus[1,+\infty).\label{eq:im_mod}\end{aligned}$$
2. Inverse modulus transformation:$$\begin{aligned}
\mathbf K\left( \frac{1}{\sqrt{\lambda}} \right)=\begin{cases}\sqrt{\lambda}[\mathbf K(\sqrt{\lambda})-i\mathbf K(\sqrt{1-\lambda})], & \operatorname{Im}\lambda>0\text{ or }\lambda<1;\\[6pt]
\sqrt{\lambda}[\mathbf K(\sqrt{\lambda})+i\mathbf K(\sqrt{1-\lambda})], & \operatorname{Im}\lambda<0\text{ or }\lambda>1. \\
\end{cases}\label{eq:inv_mod}\end{aligned}$$
The transformation laws of the complete elliptic integral $ \mathbf K$ (Eqs. \[eq:im\_mod\]–\[eq:inv\_mod\]) and the $ N=4$ case of Eq. \[eq:z\_Pnu\_ratios\] together imply the following functional equations:$$\begin{aligned}
\begin{array}{r@{\;=\;}lr@{\;=\;}lr@{\;=\;}l}
\lambda(z+1)&\dfrac{\lambda(z)}{\lambda(z)-1},\quad & \lambda\left(-\dfrac{1}{z}\right)&1-\lambda(z),\quad &\lambda(z+2)&\lambda(z),\\[8pt]\lambda\left( -\dfrac{1}{z-1} \right)&\dfrac{1}{1-\lambda(z)},& \lambda\left( \dfrac{z\mathstrut}{z+1} \right)&\dfrac{1}{\lambda(z)},& \lambda\left( 1-\dfrac{1}{z} \right)&1-\dfrac{1}{\lambda(z)}.
\end{array}\label{eq:lambda_transf}\end{aligned}$$
From a special case of Eq. \[eq:P\_nu\_sqr\_E2\_diff\], we know that $$\begin{aligned}
{[}P_{-1/4}(1-2\alpha_2(w))]^2={}&2E_{2}(2w)-E_2(w),&&\operatorname{Im}w>0,|\operatorname{Re}w|<\frac{1}{2},\left\vert w+\frac{1}{2} \right\vert>\frac{1}{2},\left|w-\frac{1}{2} \right\vert>\frac{1}{2}.\label{eq:P_quarter_sqr_E2_diff}\end{aligned}$$According to Ramanujan’s base-4 theory, the Legendre function of degree $-1/4 $ satisfies [@RN5 Chap. 33, Eqs. 9.1 and 9.2]: $$\begin{aligned}
P_{-1/4}(\cos\theta)={}&\frac{2}{\pi}\frac{1}{\sqrt{\smash[b]{1+\sin(\theta/2)}}}\mathbf K\left( \sqrt{\frac{2\sin(\theta/2)}{1+\sin(\theta/2)}} \right)=\frac{2\sqrt{2}}{\pi}\frac{1}{\sqrt{\smash[b]{1+\cos(\theta/2)}}}\mathbf K\left( \sqrt{\frac{1-\cos(\theta/2)}{1+\cos(\theta/2)}} \right),\quad 0\leq\theta<\pi.\label{eq:P_quarter_sin4}\end{aligned}$$
Now, it is clear that the integral in Eq. \[eq:G3Hecke2\_mid\_int0\] is equal to $$\begin{aligned}
\frac1{4\pi}\int_{-1}^1\frac{[P_{-1/4}(\xi)P_{-1/4}(-\xi)]^2}{\xi-1+2\alpha_{2}(z)}\operatorname{d}\xi=\frac{8}{\pi^{5}}\int_0^1\frac{[\mathbf K(\sqrt{t})\mathbf K(\sqrt{1-t})]^{2}}{\alpha_{2}(z)-\left(\frac{t}{2-t}\right)^2}\frac{t\operatorname{d}t}{2-t}=\frac{8}{\pi^{5}}\int_0^1\frac{[\mathbf K(\sqrt{t})\mathbf K(\sqrt{1-t})]^{2}}{\alpha_{2}(z)-\left(\frac{1-t}{1+t}\right)^2}\frac{(1-t)\operatorname{d}t}{1+t}.
\label{eq:G3Hecke2_mid_int0'}\tag{\ref{eq:G3Hecke2_mid_int0}$'$}\end{aligned}$$On one hand, we have the following identity by residue calculus:[^6]$$\begin{aligned}
&
\int_{-\infty+i0^+}^{+\infty+i0^+}\frac{[\mathbf K(\sqrt{t})\mathbf K(\sqrt{1-t})]^{2}}{\alpha_{2}(z)-\left(\frac{1-t}{1+t}\right)^2}\frac{(1-t)\operatorname{d}t}{1+t}\notag\\={}&\frac{\pi i}{2}\left[ \frac{1-2\lambda(2z+1)}{1-\lambda(2z+1)} \right]^2\left[ \mathbf K\left( \sqrt{\frac{1}{1-\lambda(2z+1)}+i0^{+}} \right) \mathbf K\left( \sqrt{\frac{\lambda(2z+1)}{\lambda(2z+1)-1}-i0^{+}} \right)\right]^2\notag\\={}
&\frac{\pi ^{5}i}{16}[P_{-1/4}( 1-2\alpha_{2}(z)+i0^{+})P_{-1/4}( 2\alpha_{2}(z)-1-i0^{+} )]^2.\label{eq:P_quarter_4_int_Re}\end{aligned}$$Here in the last step, we have used the identities
$$\begin{aligned}
P_{-1/4}\left( 1-\frac{2}{(1-2\lambda)^2} \right)={}&\frac{1}{\pi}\sqrt{\frac{2(2\lambda-1)}{\lambda-1}}\mathbf K\left( \sqrt{\frac{1}{1-\lambda}} \right)=\frac{1}{\pi}\sqrt{\frac{2(2\lambda-1)}{\lambda}}\mathbf K\left( \sqrt{\frac{1}{\lambda}} \right),\quad\text{a.e. }\lambda\in\mathbb C;\label{eq:P_quarter_K_a}\\P_{-1/4}\left( \frac{2}{(1-2\lambda)^2} -1\right)={}&\begin{cases}\dfrac{2\sqrt{1-2\lambda}}{\pi}\mathbf K(\sqrt{\lambda}), & \operatorname{Re}\lambda<1/2; \\[8pt]
\dfrac{2\sqrt{2\lambda-1}}{\pi}\mathbf K(\sqrt{1-\lambda}), & \operatorname{Re}\lambda>1/2,
\end{cases}\label{eq:P_quarter_K_b}\end{aligned}$$
which follow from the transformation laws of the complete elliptic integrals (Eqs. \[eq:im\_mod\]–\[eq:inv\_mod\]) and analytic continuations of Eq. \[eq:P\_quarter\_sin4\].
On the other hand, by some transformations of the complete elliptic integrals of the first kind, we can establish an identity $$\begin{aligned}
&
\operatorname{Re}\int_{-\infty+i0^+}^{+\infty+i0^+}\frac{[\mathbf K(\sqrt{t})\mathbf K(\sqrt{1-t})]^{2}}{\alpha_{2}(z)-\left(\frac{1-t}{1+t}\right)^2}\frac{(1-t)\operatorname{d}t}{1+t}\notag\\={}&\int_0^1\frac{6[\mathbf K(\sqrt{t})\mathbf K(\sqrt{1-t})]^{2}}{\alpha_{2}(z)-\left(\frac{t}{2-t}\right)^2}\frac{t\operatorname{d}t}{2-t}-\operatorname{Re}\left\{\frac{\pi ^{5}i}{32}[P_{-1/4}( 2\alpha_{2}(z)-1-i0^{+} )]^4\right\}\label{eq:P_quarter_4_int_Re'}\tag{\ref{eq:P_quarter_4_int_Re}$'$}\end{aligned}$$for $ |\operatorname{Re}z|=1/2,\operatorname{Im}z>1/2$. Concretely speaking, we have $$\begin{aligned}
\int_{1+i0^+}^{+\infty+i0^+}\frac{[\mathbf K(\sqrt{t})\mathbf K(\sqrt{1-t})]^{2}}{\alpha_{2}(z)-\left(\frac{1-t}{1+t}\right)^2}\frac{(1-t)\operatorname{d}t}{1+t}\overset{t=\frac{1}{1-s}}{=\!\!=\!\!=\!\!=\!\!=\!\!=}{}&-\int_{0+i0^+}^{1+i0^+}\frac{\left[\mathbf K\left(\sqrt{\frac{1}{1-s}}\right)\mathbf K\left(\sqrt{\frac{s}{s-1}}\right)\right]^{2}}{\alpha_{2}(z)-\left(\frac{s}{2-s}\right)^2}\frac{s\operatorname{d}s}{(2-s)(1-s)^{2}}\notag\\\underset{\text{Eq.~\ref{eq:inv_mod}}}{\overset{\text{Eq.~\ref{eq:im_mod}}}{=\!\!=\!\!=\!\!=\!\!=\!\!=}}{}&-\int_0^1\frac{[\mathbf K(\sqrt{1-s})+i\mathbf K(\sqrt{s})]^{2}[\mathbf K(\sqrt{s})]^2}{\alpha_{2}(z)-\left(\frac{s}{2-s}\right)^2}\frac{s\operatorname{d}s}{2-s};\label{eq:K4_int_T1}\\\int_{-\infty+i0^+}^{0+i0^+}\frac{[\mathbf K(\sqrt{t})\mathbf K(\sqrt{1-t})]^{2}}{\alpha_{2}(z)-\left(\frac{1-t}{1+t}\right)^2}\frac{(1-t)\operatorname{d}t}{1+t}\overset{t=\frac{s}{s-1}}{=\!\!=\!\!=\!\!=\!\!=\!\!=}{}&\int_{0-i0^+}^{1-i0^+}\frac{\left[\mathbf K\left(\sqrt{\frac{1}{1-s}}\right)\mathbf K\left(\sqrt{\frac{s}{s-1}}\right)\right]^{2}}{\alpha_{2}(z)-\left(\frac{1}{1-2s}\right)^2}\frac{\operatorname{d}s}{(1-2s)(1-s)^{2}}\notag\\\underset{\text{Eq.~\ref{eq:inv_mod}}}{\overset{\text{Eq.~\ref{eq:im_mod}}}{=\!\!=\!\!=\!\!=\!\!=\!\!=}}{}&\int_0^1\frac{[\mathbf K(\sqrt{1-s})-i\mathbf K(\sqrt{s})]^{2}[\mathbf K(\sqrt{s})]^2}{\alpha_{2}(z)-\left(\frac{1}{1-2s}\right)^2}\frac{\operatorname{d}s}{1-2s},\label{eq:K4_int_T2}\end{aligned}$$which add up to$$\begin{aligned}
&
\operatorname{Re}\int_{-\infty+i0^+}^{0+i0^+}\frac{[\mathbf K(\sqrt{t})\mathbf K(\sqrt{1-t})]^{2}}{\alpha_{2}(z)-\left(\frac{1-t}{1+t}\right)^2}\frac{(1-t)\operatorname{d}t}{1+t}+\operatorname{Re}\int_{1+i0^+}^{+\infty+i0^+}\frac{[\mathbf K(\sqrt{t})\mathbf K(\sqrt{1-t})]^{2}}{\alpha_{2}(z)-\left(\frac{1-t}{1+t}\right)^2}\frac{(1-t)\operatorname{d}t}{1+t}\notag\\={}&-\int_0^1\frac{[\mathbf K(\sqrt{s})]^4}{\alpha_{2}(z)-\left(\frac{1}{1-2s}\right)^2}\frac{\operatorname{d}s}{1-2s}-\int_0^1\frac{\{[\mathbf K(\sqrt{1-t})]^{2}-[\mathbf K(\sqrt{t})]^{2}\}[\mathbf K(\sqrt{t})]^2}{\alpha_{2}(z)-\left(\frac{t}{2-t}\right)^2}\frac{t\operatorname{d}t}{2-t}\notag\\={}&-\frac{1}{2}\int_0^1\frac{[\mathbf K(\sqrt{s})]^4-[\mathbf K(\sqrt{1-s})]^4}{\alpha_{2}(z)-\left(\frac{1}{1-2s}\right)^2}\frac{\operatorname{d}s}{1-2s}-\int_0^1\frac{\{[\mathbf K(\sqrt{1-t})]^{2}-[\mathbf K(\sqrt{t})]^{2}\}[\mathbf K(\sqrt{t})]^2}{\alpha_{2}(z)-\left(\frac{t}{2-t}\right)^2}\frac{t\operatorname{d}t}{2-t}.\label{eq:K4_over_1_2s_int}\end{aligned}$$We can treat the remaining integral over $s$ in the last line of Eq. \[eq:K4\_over\_1\_2s\_int\] in two ways. By residue calculus, we have $$\begin{aligned}
&
\int_{-\infty+i0^+}^{+\infty+i0^+}\frac{[\mathbf K(\sqrt{s})]^4-[\mathbf K(\sqrt{1-s})]^4}{\alpha_{2}(z)-\left(\frac{1}{1-2s}\right)^2}\frac{\operatorname{d}s}{1-2s}\notag\\={}&\frac{\pi i}{2}\left[ 1-2\lambda\left( -\frac{1}{z}+1 \right) \right]^2\left\{ \left[\mathbf K\left( \sqrt{1-\lambda\left( -\frac{1}{z}+1 \right)} \right) \right]^4-\left[\mathbf K\left( \sqrt{\lambda\left( -\frac{1}{z}+1 \right)} \right) \right]^4\right\}\notag\\={}&\operatorname{Re}\left\{\frac{\pi ^{5}i}{16}[P_{-1/4}( 2\alpha_{2}(z)-1-i0^{+} )]^4\right\}\end{aligned}$$for $ |\operatorname{Re}z|=1/2,\operatorname{Im}z>1/2$. Meanwhile, transformations in the spirit of Eqs. \[eq:K4\_int\_T1\] and \[eq:K4\_int\_T2\] would bring us$$\begin{aligned}
&
\left(
\int_{-\infty+i0^+}^{0+i0^+}+\int_{1+i0^+}^{+\infty+i0^+}\right)\frac{[\mathbf K(\sqrt{s})]^4-[\mathbf K(\sqrt{1-s})]^4}{\alpha_{2}(z)-\left(\frac{1}{1-2s}\right)^2}\frac{\operatorname{d}s}{1-2s}\notag\\={}&\int_0^1\frac{\{2[\mathbf K(\sqrt{t})]^{4}-[\mathbf K(\sqrt{1-t})-i\mathbf K(\sqrt{t})]^{4}-[\mathbf K(\sqrt{1-t})+i\mathbf K(\sqrt{t})]^{4}\}}{\alpha_{2}(z)-\left(\frac{1-t}{1+t}\right)^2}\frac{(1-t)\operatorname{d}t}{1+t}\notag\\={}&\int_0^1\frac{\{12[\mathbf K(\sqrt{1-t})]^{2}-2[\mathbf K(\sqrt{t})]^{2}\}[\mathbf K(\sqrt{t})]^2}{\alpha_{2}(z)-\left(\frac{t}{2-t}\right)^2}\frac{t\operatorname{d}t}{2-t}.\label{eq:K4_diff_comb_int}\end{aligned}$$Thus, we see that the identity claimed in Eq. \[eq:P\_quarter\_4\_int\_Re’\] is a result of Eqs. \[eq:K4\_over\_1\_2s\_int\]–\[eq:K4\_diff\_comb\_int\].
To wrap up, we combine Eqs. \[eq:P\_quarter\_4\_int\_Re\] and \[eq:P\_quarter\_4\_int\_Re’\] into $$\begin{aligned}
\int_{0}^{i\infty} \frac{\alpha_2(\zeta)[1-\alpha_2(\zeta)][2E_{2}(2\zeta)-E_2(\zeta)]^{3}}{\alpha_{2}(\zeta)-\alpha_2(z)}\frac{\zeta^{2}\operatorname{d}\zeta}{i}=-\frac{ \operatorname{Im}z+4 (\operatorname{Im}z)^3}{12}[2E_2(2z)-E_2(z)]^2\label{eq:alpha2_weight6_int0'}\tag{\ref{eq:alpha2_weight6_int0}$'$}\end{aligned}$$where the integration is carried out along the $ \operatorname{Im}\zeta$-axis, and $ |\operatorname{Re}z|=1/2,\operatorname{Im}z>1/2$. By differentiation, we have justified Eq. \[eq:G3Hecke2\_mid\_int0\] for $ |\operatorname{Re}z|=1/2,\operatorname{Im}z>1/2$. The rest of the verification for Eq. \[eq:G3Hecke2\_mid\_int0\] follows readily from the reflection formula $ \varrho^{\mathfrak H/\overline\varGamma_0(2)}_3(\zeta,z)=-\varrho^{\mathfrak H/\overline\varGamma_0(2)}_3(-1/(2\zeta),-1/(2z))$.
\[rmk:spec\_high\_w\_auto\_G\_int\]Due to the subtle restriction in Eq. \[eq:rho3Hecke2\_int0\], the integral representation for $ G_3^{\mathfrak H/\overline{\varGamma}_0(2)}(z,z')$ does not immediately generalize to weight-6 automorphic Green’s functions with levels $ N>2$, or to automorphic Green’s functions with weights higher than 6. Nevertheless, when one of the arguments in the automorphic Green’s functions is a special CM point, say $ \alpha_N(z)=\infty$, it is possible to establish certain integral representations for automorphic Green’s functions by going through criteria \[itm:AGF1’\]–\[itm:AGF3’\] with respect to the other variable $ z'$. Some examples of weights 6 and 10 are given below:[$$\begin{aligned}
&
G_3^{\mathfrak H/PSL(2,\mathbb Z)}\left(\frac{1+i\sqrt{3}}{2},z'\right)\notag\\={}&\notag\frac{4\pi^2}{9\left[ \operatorname{Im}\frac{iP_{-1/6}(-\sqrt{(j(z')-1728)/j(z')})}{P_{-1/6}(\sqrt{(j(z')-1728)/j(z')})} \right]^2}\operatorname{Re}\int^1_{\sqrt{(j(z')-1728)/j(z')}}\xi[P_{-1/6}(\xi)]^4\times\notag\\&\times\left[\tfrac{iP_{-1/6}(-\xi)}{P_{-1/6}(\xi)}-\vphantom{\overline{\tfrac{\sqrt{j}}{}}}\tfrac{iP_{-1/6}(-\sqrt{(j(z')-1728)/j(z')})}{P_{-1/6}(\sqrt{(j(z')-1728)/j(z')})}\right]^2\left[\tfrac{iP_{-1/6}(-\xi)}{P_{-1/6}(\xi)}-\overline{\left(\tfrac{iP_{-1/6}(-\sqrt{(j(z')-1728)/j(z')})}{P_{-1/6}(\sqrt{(j(z')-1728)/j(z')})}\right)}\right]^2\operatorname{d}\xi\notag\\&+\frac{4\pi^2}{9\left[ \operatorname{Im}\frac{iP_{-1/6}(-\sqrt{(j(z')-1728)/j(z')})}{P_{-1/6}(\sqrt{(j(z')-1728)/j(z')})} \right]^2}\int_{-1}^{1}\xi[P_{-1/6}(\xi)]^4\operatorname{d}\xi,\quad \text{a.e. }z'\in\mathfrak H;\label{eq:KZ_G3_ell3_star}\\
&\frac{G_3^{\mathfrak H/\overline{\varGamma}_0(2)}\left(\frac{i-1}{2},z'\right)}{2}=G_3^{\mathfrak H/\overline{\varGamma}(2)}(i,2z'+1)\notag\\={}&\frac{\pi^{2}}{32\left[\operatorname{Im}\frac{iP_{-1/4}(2\alpha_{2}(z')-1)}{P_{-1/4}(1-2\alpha_{2}(z'))}\right]^2}\operatorname{Re}\int^1_{1-2\alpha_{2}(z')}\xi[P_{-1/4}(\xi)]^4\times \notag\\&\times\left[\frac{iP_{-1/4}(-\xi)}{P_{-1/4}(\xi)}-\vphantom{\overline{\tfrac{\sqrt{j}}{}}}\frac{iP_{-1/4}(2\alpha_{2}(z')-1)}{P_{-1/4}(1-2\alpha_{2}(z'))}\right]^2\left[\frac{iP_{-1/4}(-\xi)}{P_{-1/4}(\xi)}-\overline{\left(\frac{iP_{-1/4}(2\alpha_{2}(z')-1)}{P_{-1/4}(1-2\alpha_{2}(z'))}\right)}\right]^2\operatorname{d}\xi\notag\\{}&+\frac{\pi^{2}}{32\left[\operatorname{Im}\frac{iP_{-1/4}(2\alpha_{2}(z')-1)}{P_{-1/4}(1-2\alpha_{2}(z'))}\right]^2}\int_{-1}^1\xi[P_{-1/4}(\xi)]^4\operatorname{d}\xi, \quad\text{a.e. }z'\in\mathfrak H;\label{eq:G3_Gamma2}\\&G_3^{\mathfrak H/\overline{\varGamma}_{0}(3)}\left( \frac{3+i\sqrt{3}}{6},z'\right)\notag\\={}&\frac{\pi^{2}}{54\left[\operatorname{Im}\frac{iP_{-1/3}(2\alpha_{3}(z')-1)}{P_{-1/3}(1-2\alpha_{3}(z'))}\right]^2}\operatorname{Re}\int^1_{1-2\alpha_{3}(z')}\xi[P_{-1/3}(\xi)]^4\times \notag\\&\times\left[\frac{iP_{-1/3}(-\xi)}{P_{-1/3}(\xi)}-\vphantom{\overline{\tfrac{\sqrt{j}}{}}}\frac{iP_{-1/3}(2\alpha_{3}(z')-1)}{P_{-1/3}(1-2\alpha_{3}(z'))}\right]^2\left[\frac{iP_{-1/3}(-\xi)}{P_{-1/3}(\xi)}-\overline{\left(\frac{iP_{-1/3}(2\alpha_{3}(z')-1)}{P_{-1/3}(1-2\alpha_{3}(z'))}\right)}\right]^2\operatorname{d}\xi\notag\\&+\frac{ \pi ^2}{54\left[\operatorname{Im}\frac{iP_{-1/3}(2\alpha_{3}(z')-1)}{P_{-1/3}(1-2\alpha_{3}(z'))}\right]^2}\int_{-1}^1\xi[P_{-1/3}(\xi)]^4\operatorname{d}\xi,\quad\text{a.e. }z'\in\mathfrak H;\label{eq:G3Hecke3_spec}\\&\frac{G_5^{\mathfrak H/\overline{\varGamma}_0(2)}\left(\frac{i-1}{2},z'\right)}{2}=G_5^{\mathfrak H/\overline{\varGamma}(2)}(i,2z'+1)\notag\\={}&-\frac{\pi^{4}}{6144\left[\operatorname{Im}\frac{iP_{-1/4}(2\alpha_{2}(z')-1)}{P_{-1/4}(1-2\alpha_{2}(z'))}\right]^4}\int_{1-2\alpha_{2}(z')}^1\xi(5-6\xi^2)[P_{-1/4}(\xi)]^8\times\notag\\&\times\left[\frac{iP_{-1/4}(-\xi)}{P_{-1/4}(\xi)}-\vphantom{\overline{\tfrac{\sqrt{j}}{}}}\frac{iP_{-1/4}(2\alpha_{2}(z')-1)}{P_{-1/4}(1-2\alpha_{2}(z'))}\right]^4\left[\frac{iP_{-1/4}(-\xi)}{P_{-1/4}(\xi)}-\overline{\left(\frac{iP_{-1/4}(2\alpha_{2}(z')-1)}{P_{-1/4}(1-2\alpha_{2}(z'))}\right)}\right]^4\operatorname{d}\xi\notag\\&-\frac{\pi^{4}}{6144\left[\operatorname{Im}\frac{iP_{-1/4}(2\alpha_{2}(z')-1)}{P_{-1/4}(1-2\alpha_{2}(z'))}\right]^4}\int_{-1}^1\xi(5-6\xi^2)[P_{-1/4}(\xi)]^8\operatorname{d}\xi,\quad\text{a.e. }z'\in\mathfrak H.\label{eq:G5Gamma2_spec}\end{aligned}$$]{}Here, Eq. \[eq:G3\_Gamma2\] reflects the fact that$$\begin{aligned}
\lim_{z\to(i-1)/2}
\varrho^{\mathfrak H/\overline\varGamma_0(2)}_3(\zeta,z)=\frac{\pi}{8}[1-2\alpha_2(\zeta)].\end{aligned}$$The validity of Eq. \[eq:G5Gamma2\_spec\] hinges on a non-trivial vanishing identity $$\begin{aligned}
\int _{-1}^1\xi(5-6\xi^2)[P_{-1/4}(-\xi)]^2[P_{-1/4}(\xi)]^6\operatorname{d}\xi=0,\end{aligned}$$ which can be proved in a similar manner as Eq. \[eq:G3Hecke2\_mid\_int0\].
\[prop:G3457\_PSL2Z\]With the notations for the Eisenstein series as given in Eqs. \[eq:E4\_E6\_defn\] and \[eq:E8E10E14\_defn\], we have the following integral representations for weights $k\in\{6,8,10,14\}$ and $ j(z)\neq j(z')$:$$\begin{aligned}
G_{k/2}^{\mathfrak H/PSL(2,\mathbb Z)}(z,z')={}&\frac{1728\pi^{2}}{(y')^{(k-2)/2}}\operatorname{Re}\int_{z'}^{i\infty} \frac{E_{k}(\zeta)}{j(\zeta)}\rho^{\mathfrak H/PSL(2,\mathbb Z)}_{k/2}(\zeta,z)\frac{(\zeta-z')^{(k-2)/2}(\zeta-\overline{z'})^{(k-2)/2}\operatorname{d}\zeta}{i}\notag\\&-\frac{1728\pi^{2}}{(y')^{(k-2)/2}}\operatorname{Re}\int_{0}^{i\infty} \frac{E_{k}(\zeta)}{j(\zeta)}\rho^{\mathfrak H/PSL(2,\mathbb Z)}_{k/2}(\zeta,z)\frac{\zeta^{k-2}\operatorname{d}\zeta}{i}\label{eq:GkPSL2Z}\end{aligned}$$ where$$\begin{aligned}
\label{eq:rho_k_PSL2Z}
\rho^{\mathfrak H/PSL(2,\mathbb Z)}_{k/2}(\zeta,z)={}&\frac{(-1)^{k/2}}{2^{(k-4)/2}}\frac{1}{\left( \frac{k-2}{2} \right)!}\frac{ y^{(k-2)/2}}{864 \pi}\left( \frac{\partial}{\partial y}\frac{1}{y} \right)^{(k-2)/2}\left[ \frac{j(\zeta)j(z)}{j(\zeta)-j(z)}\frac{{E_{6}(z)}}{E_{4}(z)E_{k}(z)}\right].\end{aligned}$$Here in Eq. \[eq:rho\_k\_PSL2Z\], it is understood that $ j(z)E_6(z)/[E_4(z)E_k(z)]=[E_4(z)]^2 E_6(z)/[\Delta(z)E_k(z)]$ defines a smooth function for all $ z\in\mathfrak H$, so that $ \rho^{\mathfrak H/PSL(2,\mathbb Z)}_{k/2}(\zeta,z)$ is well-behaved when $ j(\zeta)\neq j(z)$.
Contrary to the practice in Proposition \[prop:G2PSL2Z\], we shall go over criteria \[itm:AGF1’\]–\[itm:AGF3’\] with respect to the variable $z'$. The major advantage of this approach is the simplification of the proofs for criteria \[itm:AGF2’\] and \[itm:AGF3’\]. Especially, it is easy to check the compatibility with the cusp behavior: $ \lim_{z'\to i\infty}G_{k/2}^{\mathfrak H/PSL(2,\mathbb Z)}(z,z')=0$.
The remaining challenge resides in the symmetry criterion \[itm:AGF1’\]. In particular, we need to expend some effort to justify that the proposed integral representations in Eq. \[eq:GkPSL2Z\] are invariant under the inversion $ z'\mapsto-1/z'$. This amounts to the verification of the following vanishing identities (see Eq. \[eq:rho3Hecke2\_int0\]):$$\begin{aligned}
\operatorname{Re}\int_{0}^{i\infty} \frac{E_{k}(\zeta)}{j(\zeta)}\rho^{\mathfrak H/PSL(2,\mathbb Z)}_{k/2}(\zeta,z)\frac{\zeta^{n}\operatorname{d}\zeta}{i}=0,\quad n\in[2,k-4]\cap(2\mathbb Z),|\operatorname{Re}z|=\frac{1}{2},\operatorname{Im}z>\frac{\sqrt{3}}{2},\label{eq:rho_k_PSL_int0}\end{aligned}$$ where the paths of integration are along the $ \operatorname{Im}\zeta$-axis. Once Eq. \[eq:rho\_k\_PSL\_int0\] is confirmed, the vanishing identity remains valid on (see Eq. \[eq:fun\_domain\_SL2Z\])$$\begin{aligned}
\mathfrak H\cap\partial(\mathfrak D\cup\hat S\mathfrak D)=\left\{ z\in\mathfrak H\left| |\operatorname{Re}z|=\frac{1}{2},\operatorname{Im}z>\frac{\sqrt{3}}{2} \right. \right\}\cup \left\{ z\in\mathfrak H\left| |\operatorname{Re}z|\leq\frac{1}{2},|z+1|=1,|z-1|=1 \right. \right\}\label{eq:fun_domain_SL2Z_S_bd}\end{aligned}$$ by the inversion $ \hat S:z\mapsto-1/z$. Consequently, Eq. \[eq:rho\_k\_PSL\_int0\] is applicable to the whole fundamental domain $ \mathfrak D$ (by a homogeneous Dirichlet boundary value problem) and to the entire upper half-plane $ \mathfrak H$ (by tessellation under the actions of $ SL(2,\mathbb Z)$).
As in Proposition \[prop:G3Hecke2\], we can prove Eq. \[eq:rho\_k\_PSL\_int0\] by supplying explicit integral formulae in the forms of$$\begin{aligned}
\int_{0}^{i\infty} \frac{E_{k}(\zeta)j(z)}{j(\zeta)-j(z)}\frac{{E_{6}(z)}}{E_{4}(z)E_{k}(z)}\frac{\zeta^{n}\operatorname{d}\zeta}{ i}=p_{k,n} (\operatorname{Im}z),\quad n\in[2,k-4]\cap(2\mathbb Z),|\operatorname{Re}z|=\frac{1}{2},\operatorname{Im}z>\frac{\sqrt{3}}{2},\label{eq:p_k_n_defn}\end{aligned}$$ where the polynomials $p_{k,n} $ are given below:$$\begin{aligned}
\left\{\begin{array}{r@{\;=\;}l}
p_{6,2} (y)& 0, \\[2pt]
p_{8,2}(y)=-p_{8,4}(y) & \dfrac{9y+40 y^3+16 y^5}{80} ,\\[6pt]
p_{10,2}(y)=-p_{10,6}(y) & \dfrac{27 y+84 y^3-112 y^5-64 y^7}{448} ,\\[6pt] p_{10,4}(y) &0,\\[6pt]
p_{14,2}(y)=-p_{14,10}(y) & \dfrac{261 y-5500 y^3-32736 y^5-29568 y^7-14080 y^9-3072 y^{11}}{33792}, \\[6pt]p_{14,4}(y)=-p_{14,8}(y)&\dfrac{9 y + 112 y^3 + 480 y^5 + 768 y^7 + 256 y^9}{2304},\\[6pt] p_{14,6}(y)&0.
\end{array}\right.\label{eq:p_k_n_list}\end{aligned}$$It is clear that Eqs. \[eq:p\_k\_n\_defn\] and \[eq:p\_k\_n\_list\] lead to a verification of Eq. \[eq:rho\_k\_PSL\_int0\].
We now illustrate Eq. \[eq:p\_k\_n\_list\] with a detailed computation for $ p_{8,2}(y)$.
We start by pointing out that the integral representation for $ p_{8,2} y$ can be rewritten using elliptic integrals:$$\begin{aligned}
\int_{0}^{i\infty} \frac{E_{8}(\zeta)j(z)}{j(\zeta)-j(z)}\frac{{E_{6}(z)}}{E_{4}(z)E_{8}(z)}\frac{\zeta^{2}\operatorname{d}\zeta}{ i}=-\frac{64}{\pi^{7}}\frac{j(z)E_{6}(z)}{E_{4}(z)E_{8}(z)}\int_0^1\frac{[\mathbf K(\sqrt{t})]^4[\mathbf K(\sqrt{1-t})]^2}{\frac{256(1-t+t^2)^3}{t^{2}(1-t)^2}-j(z)}\frac{(1-t+t^2)^2\operatorname{d}t}{t(1-t)}\label{eq:p_8_2_ell_int_form}\end{aligned}$$for $ |\operatorname{Re}z|=1/2,\operatorname{Im}z>{\sqrt{3}}/{2}$. To show Eq. \[eq:p\_8\_2\_ell\_int\_form\], we recall the expression of the $ j$-invariant as a rational function of the modular lambda function:$$\begin{aligned}
&&
j(z)={}&\frac{256\{1-\lambda(z)+[\lambda(z)]^2\}^3}{[\lambda(z)]^2[1-\lambda(z)]^2},&&\forall z\in\mathfrak H,&&\end{aligned}$$ along with Ramanujan’s work [@Ramanujan1916] on the relation between the Eisenstein series and the complete elliptic integrals of the first kind $ \mathbf K(\sqrt{\lambda(z)})$: $$\begin{aligned}
&&E_4(z)={}&\left[ \frac{2\mathbf K(\sqrt{\lambda(z)})}{\pi} \right]^{4}\{1-\lambda(z)+[\lambda(z)]^{2}\},&&\operatorname{Im}z>0,|\operatorname{Re}z|<1,\left|z+\frac12\right|>\frac12,\left|z-\frac12\right|>\frac{1}{2},&&\label{eq:E4_Ell_Ramanujan}\\&&E_6(z)={}&\left[ \frac{2\mathbf K(\sqrt{\lambda(z)})}{\pi} \right]^{6}\frac{[\lambda(z)+1][\lambda(z)-2][2\lambda(z)-1]}{2},&&\operatorname{Im}z>0,|\operatorname{Re}z|<1,\left|z+\frac12\right|>\frac12,\left|z-\frac12\right|>\frac{1}{2}.&&\label{eq:E6_Ell_Ramanujan}\end{aligned}$$Furthermore, the variable transformation from $ \zeta$ to $ t=\lambda(\zeta)$ is mediated by the identities:$$\begin{aligned}
z={}&\frac{i\mathbf K(\sqrt{1-\lambda(z)})}{\mathbf K(\sqrt{\lambda(z)})},&& \operatorname{Im}z>0,|\operatorname{Re}z|<1,\left|z+\frac12\right|>\frac12,\left|z-\frac12\right|>\frac{1}{2},\label{eq:lambda_K_ratio_z}\\
\frac{\operatorname{d}}{\operatorname{d}t}\frac{\mathbf K(\sqrt{1-t})}{\mathbf K(\sqrt{t})}={}&-\frac{\pi}{4t(1-t)[\mathbf K(\sqrt{t})]^2},&&t\in\mathbb C\smallsetminus((-\infty,0]\cup[1,+\infty)),
\label{eq:ratio_deriv}\end{aligned}$$which are special cases of Eqs. \[eq:z\_Pnu\_ratios\] and \[eq:P\_nu\_ratio\_deriv\], respectively.
Then, we consider the following integral related to the right-hand side of Eq. \[eq:p\_8\_2\_ell\_int\_form\]:$$\begin{aligned}
-\frac{64}{\pi^{7}}\frac{j(z)E_{6}(z)}{E_{4}(z)E_{8}(z)}\int_{-\infty+i0^+}^{+\infty+i0^+}\frac{[\mathbf K(\sqrt{t})]^4[\mathbf K(\sqrt{1-t})]^2}{\frac{256(1-t+t^2)^3}{t^{2}(1-t)^2}-j(z)}\frac{(1-t+t^2)^2\operatorname{d}t}{t(1-t)}\end{aligned}$$and compute it in two ways. In the first approach, we can close the contour in the upper half $t$-plane and collect residues at the three simple poles therein. Without loss of generality, we may assume that $ \operatorname{Re}z=1/2$, and these three poles are$$\begin{aligned}
t=\lambda(z),\quad t=\lambda\left( \frac{z-1}{z} \right)=1-\frac{1}{\lambda(z)},\quad t=\lambda\left( -\frac{1}{z-1} \right)=\frac{1}{1-\lambda(z)}.\label{eq:3poles_t}\end{aligned}$$By the differentiation formula for the $ j$-invariant (Eq. \[eq:j’z\_E4E6\]), the transformation laws of the Eisenstein series (Eq. \[eq:E2E4E6\_mod\_transf\]) and residue calculus, we have$$\begin{aligned}
&
-\frac{64}{\pi^{7}}\frac{j(z)E_{6}(z)}{E_{4}(z)E_{8}(z)}\int_{-\infty+i0^+}^{+\infty+i0^+}\frac{[\mathbf K(\sqrt{t})]^4[\mathbf K(\sqrt{1-t})]^2}{\frac{256(1-t+t^2)^3}{t^{2}(1-t)^2}-j(z)}\frac{(1-t+t^2)^2\operatorname{d}t}{t(1-t)}\notag\\={}&\frac{1}{i}\left[ z^2+\left(\frac{z-1}{z}\right)^2 z^6+\left(-\frac{1}{z-1}\right)^2 (z-1)^6 \right]\notag\\={}&\frac{9 y + 40 y^3 + 16 y^5}{16}+\frac{21 - 156 y^2 + 48 y^4 - 64 y^6}{64i},\quad\text{for }\operatorname{Im}z=\frac{z-\frac{1}{2}}{i}=y.\label{eq:p_8_2_comp1}\end{aligned}$$In the second approach, we employ the transformation laws of $ \mathbf K$ (Eqs. \[eq:im\_mod\] and \[eq:inv\_mod\]) to deduce (see Eqs. \[eq:K4\_int\_T1\], \[eq:K4\_int\_T2\] and \[eq:K4\_diff\_comb\_int\])$$\begin{aligned}
&-\frac{64}{\pi^{7}}\frac{j(z)E_{6}(z)}{E_{4}(z)E_{8}(z)}\int_{-\infty+i0^+}^{+\infty+i0^+}\frac{[\mathbf K(\sqrt{t})]^4[\mathbf K(\sqrt{1-t})]^2}{\frac{256(1-t+t^2)^3}{t^{2}(1-t)^2}-j(z)}\frac{(1-t+t^2)^2\operatorname{d}t}{t(1-t)}\notag\\{}&+\frac{64}{\pi^{7}}\frac{j(z)E_{6}(z)}{E_{4}(z)E_{8}(z)}\int_0^1\frac{[\mathbf K(\sqrt{t})]^4[\mathbf K(\sqrt{1-t})]^2}{\frac{256(1-t+t^2)^3}{t^{2}(1-t)^2}-j(z)}\frac{(1-t+t^2)^2\operatorname{d}t}{t(1-t)}\notag\\={}&\frac{64}{\pi^{7}}\frac{j(z)E_{6}(z)}{E_{4}(z)E_{8}(z)}\int_{0}^{1}\frac{[\mathbf K(\sqrt{t})]^4[\mathbf K(\sqrt{1-t})-i\mathbf K(\sqrt{t})]^2+[\mathbf K(\sqrt{1-t})-i\mathbf K(\sqrt{t})]^4[\mathbf K(\sqrt{t})]^2}{\frac{256(1-t+t^2)^3}{t^{2}(1-t)^2}-j(z)}\frac{(1-t+t^2)^2\operatorname{d}t}{t(1-t)}\notag\\={}&\frac{64}{\pi^{7}}\frac{j(z)E_{6}(z)}{E_{4}(z)E_{8}(z)}\int_0^1\frac{4i[\mathbf K(\sqrt{1-t})\mathbf K(\sqrt{t})]^3-6i\mathbf K(\sqrt{1-t})[\mathbf K(\sqrt{t})]^{5}-4[\mathbf K(\sqrt{t})]^4[\mathbf K(\sqrt{1-t})]^2}{\frac{256(1-t+t^2)^3}{t^{2}(1-t)^2}-j(z)}\frac{(1-t+t^2)^2\operatorname{d}t}{t(1-t)}.\label{eq:p_8_2_comp2}\end{aligned}$$Comparing Eqs. \[eq:p\_8\_2\_comp1\] and \[eq:p\_8\_2\_comp2\], we have completed the evaluation of $ p_{8,2}(y)$.
Except that one needs to solve $ p_{14,2}(y)$ and $ p_{14,4}(y)$ from two simultaneous equations, the derivations of all the other polynomials in Eq. \[eq:p\_k\_n\_list\] will follow from similar procedures as the computation for $ p_{8,2}(y)$.
At this point, we have completed the verification of all the integral representations of automorphic Green’s functions proposed in Theorem \[thm:KZ\_int\_repns\]. As direct applications of the results in Propositions \[prop:G3Hecke2\] and \[prop:G3457\_PSL2Z\], we list some integral representations of special $ L$-values in the next proposition.
\[prop:high\_w\_L\_values\]We have the following integral formulae for special $ L$-values:[$$\begin{aligned}
\zeta(3)={}&-\frac{\pi^{2}}{42}\lim_{y'\to+\infty}G_{3}^{\mathfrak H/PSL(2,\mathbb Z)}\left( \frac{1+i\sqrt{3}}{2} ,z'\right)(y')^{2}=-\frac{2\pi^4}{189}\int_{-1}^1\xi[P_{-1/6}(\xi)]^4\operatorname{d}\xi\notag\\={}&-\frac{2\pi^{2}}{21}\lim_{y'\to+\infty}G_3^{\mathfrak H/\overline{\varGamma}(2)}(i,z')(y')^2=-\frac{\pi^{4}}{168}\int_{-1}^1 \xi[P_{-1/4}(\xi)]^4\operatorname{d}\xi\notag\\={}&-\frac{2\pi^{2}}{3}\lim_{y'\to+\infty}G_{3}^{\mathfrak H/\overline{\varGamma}_0(3)}\left( \frac{3+i\sqrt{3}}{6},z'\right)(y')^{2}=-\frac{\pi^4}{243}\int_{-1}^1\xi[P_{-1/3}(\xi)]^4\operatorname{d}\xi\notag\\={}&-\frac{4\pi^{2}}{189}\lim_{y'\to+\infty}G_3^{\mathfrak H/PSL(2,\mathbb Z)}(i,z')(y')^2=\frac{4\pi^{4}}{189}\operatorname{Re}\int_{-1}^1\left( 1-\frac{5\xi^2}{18} \right)\frac{[P_{-1/6}(\xi)]^{4}}{\xi^{3}}\operatorname{d}\xi\notag\\={}&-\frac{8\pi^{2}}{567}\lim_{y'\to+\infty}G_3^{\mathfrak H/PSL(2,\mathbb Z)}(i\sqrt{2},z')(y')^{2}=-\frac{2\pi^{2}}{63}\lim_{y'\to+\infty}G_3^{\mathfrak H/\overline\varGamma_{\vartheta}}(1+i\sqrt{2},z')(y')^{2}\notag\\={}&-\frac{2\pi^{2}}{63}\lim_{y'\to+\infty}G_3^{\mathfrak H/\overline\varGamma_{0}(2)}\left( \frac{i}{\sqrt{2}},\frac{z'-1}{2} \right)(y')^{2}=\frac{2\pi^{4}}{63}\operatorname{Re}\int_{-1}^1\frac{(8-3\xi^2)[P_{-1/4}(\xi)]^{4}}{16\xi^{3}}\operatorname{d}\xi;\label{eq:zeta3_P_sixth_P_third}\\L(4,\chi_{-4})={}&-\frac{\pi^{3}}{120}\lim_{y'\to+\infty}G_4^{\mathfrak H/PSL(2,\mathbb Z)}(i,z')(y')^3=-\frac{\pi^6}{120}\operatorname{Re}\int_{-1}^1\left( 1-\frac{19\xi^2}{27} \right)\frac{[P_{-1/6}(\xi)]^{6}}{2\xi^{4}}\operatorname{d}\xi;\\L(4,\chi_{-3})={}&-\frac{4\pi^{3}}{405}\lim_{y'\to+\infty}G_{4}^{\mathfrak H/PSL(2,\mathbb Z)}\left( \frac{1+i\sqrt{3}}{2},z'\right)(y')^{3}\notag\\={}&-\frac{\pi^6}{10935}\int_{-1}^1(7-16\xi^2)[P_{-1/6}(\xi)]^{6}\operatorname{d}\xi;\\\zeta(5)={}&-\frac{2\pi^{4}}{385}\lim_{y'\to+\infty}G_{5}^{\mathfrak H/PSL(2,\mathbb Z)}\left( \frac{1+i\sqrt{3}}{2},z'\right)(y')^{4}=\frac{\pi^{8}}{249480}\int_{-1}^1(51-64\xi^2)\xi[P_{-1/6}(\xi)]^{8}\operatorname{d}\xi\notag\\={}&-\frac{8\pi^{4}}{525}\lim_{y'\to+\infty}G_5^{\mathfrak H/\overline\varGamma(2)}(i,z')(y')^4=\frac{\pi^{8}}{100800}\int_{-1}^1\xi(5-6\xi^2)[P_{-1/4}(\xi)]^8\operatorname{d}\xi\notag\\={}&-\frac{64\pi^{4}}{17325}\lim_{y'\to+\infty}G_5^{\mathfrak H/PSL(2,\mathbb Z)}(i,z')(y')^4\notag\\={}&\frac{64\pi^{8}}{17325}\operatorname{Re}\int_{-1}^1\left( 1-\frac{61\xi^2}{54}\, \, +\frac{5\xi^4}{24} \right)\frac{[P_{-1/6}(\xi)]^{8}}{4\xi^{5}}\operatorname{d}\xi;\label{eq:zeta5_int_P_sixth}\\\zeta(7)={}&-\frac{8\pi^{6}}{8085}\lim_{y'\to+\infty}G_{7}^{\mathfrak H/PSL(2,\mathbb Z)}\left( \frac{1+i\sqrt{3}}{2},z'\right)(y')^{6}\notag\\={}&-\frac{\pi^{12}}{58939650}\int_{-1}^1(168-485\xi^2+320\xi^4)\xi[P_{-1/6}(\xi)]^{12}\operatorname{d}\xi\notag\\={}&-\frac{1024\pi^{6}}{1902285}\lim_{y'\to+\infty}G_7^{\mathfrak H/PSL(2,\mathbb Z)}(i,z')(y')^6\notag\\={}&\frac{1024\pi^{12}}{1902285}\operatorname{Re}\int_{-1}^1\left( 1-\frac{107 \xi^2}{54}+\frac{33871 \xi^4}{29160}-\frac{2077 \xi^6}{11664} \right)\frac{[P_{-1/6}(\xi)]^{12}}{16\xi^{7}}\operatorname{d}\xi.\label{eq:zeta7_int_P_sixth}\end{aligned}$$]{}
These are routine computations based on the asymptotic behavior (Eq. \[eq:Gs\_asympt\]), the addition formulae (Proposition \[prop:add\_form\_auto\_Green\]), and the integral representations (Proposition \[prop:G3Hecke2\], Remark \[rmk:spec\_high\_w\_auto\_G\_int\], Proposition \[prop:G3457\_PSL2Z\]) for higher weight automorphic Green’s functions.
For example, via explicit computations of $ \rho^{\mathfrak H/PSL(2,\mathbb Z)}_{k/2}(\zeta,e^{i\pi/3})$ (Eq. \[eq:rho\_k\_PSL2Z\]) for $ k\in\{6,10,14\}$, one may spell out [$$\begin{aligned}
&G_3^{\mathfrak H/PSL(2,\mathbb Z)}(e^{\pi i/3},z)-\frac{4\pi^2}{9 y^2}\int_{-1}^{1}\xi[P_{-1/6}(\xi)]^4\operatorname{d}\xi\notag\\={}&\frac{4\pi^2}{9 y^2}\operatorname{Re}\int^1_{\sqrt{\frac{j(z)-1728}{j(z)}}}\xi[P_{-1/6}(\xi)]^4\left[\frac{iP_{-1/6}(-\xi)}{P_{-1/6}(\xi)}- z\right]^{2}\left[\frac{iP_{-1/6}(-\xi)}{P_{-1/6}(\xi)}- \overline{z}\right]^{2}\operatorname{d}\xi,\label{eq:GT3_KZ}\\
&G_{5}^{\mathfrak H/PSL(2,\mathbb Z)}(e^{\pi i/3},z)+\frac{\pi^{4}}{1296 y^{4}}\int_{-1}^1(51-64\xi^2)\xi[P_{-1/6}(\xi)]^{8}\operatorname{d}\xi\notag\\={}&-\frac{\pi^{4}}{1296 y^{4}}\operatorname{Re}\int^1_{\sqrt{\frac{j(z)-1728}{j(z)}}}(51-64\xi^{2})\xi[P_{-1/6}(\xi)]^{8}\left[\frac{iP_{-1/6}(-\xi)}{P_{-1/6}(\xi)}- z\right]^{4}\left[\frac{iP_{-1/6}(-\xi)}{P_{-1/6}(\xi)}- \overline{z}\right]^{4}\operatorname{d}\xi;\label{eq:GT5_KZ}\\{}&G_{7}^{\mathfrak H/PSL(2,\mathbb Z)}(e^{\pi i/3},z)-\frac{\pi^{6}}{58320 y^{6}}\int_{-1}^1(168-485\xi^2+320\xi^4)\xi[P_{-1/6}(\xi)]^{12}\operatorname{d}\xi\notag\\={}&\frac{\pi^{6}}{58320 y^{6}}\operatorname{Re}\int^1_{\sqrt{\frac{j(z)-1728}{j(z)}}}(168-485\xi^2+320\xi^4)\xi[P_{-1/6}(\xi)]^{12}\times\notag\\{}&\times\left[\frac{iP_{-1/6}(-\xi)}{P_{-1/6}(\xi)}- z\right]^{6}\left[\frac{iP_{-1/6}(-\xi)}{P_{-1/6}(\xi)}- \overline{z}\right]^{6}\operatorname{d}\xi,\label{eq:GT7_KZ}\end{aligned}$$]{}for $ |z|\geq1,-\frac12<\operatorname{Re}z\leq\frac{1}{2},z\neq \frac{1}{2}+i\frac{\sqrt{3}}{2}$, where all the integrals are taken over straight line segments. Asymptotic analysis of Eqs. \[eq:GT3\_KZ\]–\[eq:GT7\_KZ\] then reveals certain integral representations for $ \zeta(3)$, $ \zeta(5)$ and $ \zeta(7)$ in Eqs. \[eq:zeta3\_P\_sixth\_P\_third\], \[eq:zeta5\_int\_P\_sixth\] and \[eq:zeta7\_int\_P\_sixth\].
We note that those conversion ratios between the asymptotic behavior of certain automorphic Green’s functions can be recovered from their addition formulae, such as $$\begin{aligned}
&\lim_{y\to+\infty}G_3^{\mathfrak H/PSL(2,\mathbb Z)}\left(z,\frac{1+i\sqrt{3}}{2}\right) y^{2}\notag\\={}&(1+3^{2}+3^{2}+3^{2})\lim_{y\to+\infty}G_3^{\mathfrak H/\overline{\varGamma}_{0}(3)}\left( z ,\frac{3+i\sqrt{3}}{6}\right) y^{2},\label{eq:PSL_Hecke3_asympt_conv'}\tag{\ref{eq:PSL_Hecke3_asympt_conv}$'$}\end{aligned}$$which is a weight-6 analog of Eq. \[eq:PSL\_Hecke3\_asympt\_conv\].
It is worth noting that the deep connections between Eichler integrals and special $ L$-values have been discovered and expounded by Shimura [@Shimura1959] and Manin [@Manin1973]. The recent work of Bringmann, Fricke and Kent [@BringmannFrickeKent2014] discussed special $ L$-values associated with Eichler integral representations of harmonic Maa[ß]{} forms.
$$\begin{aligned}
\begin{array}{r@{\;\in\;}l|l|l|l|l}\hline\hline \vphantom{\dfrac{\frac12}{\frac12}}z&\mathfrak Z_{D}& j(z)&e^{-\frac{2 y^2}{3}G_3^{\mathfrak H/PSL(2,\mathbb Z)}(e^{\pi i/3},z)}\!\!\!\!\!&e^{-24 y^4G_5^{\mathfrak H/PSL(2,\mathbb Z)}(e^{\pi i/3},z)}\!\!\!\!\!&e^{-96 y^6G_7^{\mathfrak H/PSL(2,\mathbb Z)}(e^{\pi i/3},z)}\!\!\!\!\! \\\hline \vphantom{\dfrac{\frac\int2}{\frac12}}i &\mathfrak Z_{-4}&2^63^3&(3)^{2}&\left(\dfrac{2^{56}}{3^{27}}\right)^2&\left(\dfrac{3^{45}}{2^{64}}\right)^4 \\[9pt] \dfrac{1+i\sqrt{7}}{2}&\mathfrak Z_{-7}&-3^{3}5^3&\dfrac{5^{3}}{3}&\dfrac{3^{189}}{5^{91}}&\dfrac{5^{133}}{3^{63}}\\[6pt]i\sqrt{2}&\mathfrak Z_{-8}&2^65^3&(5)^{2}&\left( \dfrac{5^{37}}{2^{56}} \right)^2&\left( \dfrac{2^{832}}{5^{347}} \right)^4\\[8pt] \dfrac{1+i\sqrt{11}}{2}&\mathfrak Z_{-11}&-2^{15}&2^{5}&2^{75}&2^{165}\\[5pt]i\sqrt{3}&\mathfrak Z_{-12}&2^43^35^3&\left(\dfrac{3^{3}}{5}\right)^2&\left( \dfrac{5^{277}}{2^{224}3^{243}} \right)^2&\left( \dfrac{3^{1215}5^{67}}{2^{2048}} \right)^4\\[8pt]i\sqrt{4}&\mathfrak Z_{-16}&2^33^311^3&\left(\dfrac{11^{3}}{3^{5}}\right)^2&\left( \dfrac{2^{448}}{3^{27}11^{107}} \right)^2&\left( \dfrac{3^{3447}}{2^{2048}11^{977}} \right)^4\\[8pt]\dfrac{1+i\sqrt{19}}{2}&\mathfrak Z_{-19}&-2^{15}3^3&\dfrac{3^{5}}{2^{3}}&\dfrac{3^{351}}{2^{485}}&\dfrac{2^{28493}}{3^{17883}}\\[8pt]\dfrac{1+i3\sqrt{3}}{2}&\mathfrak Z_{-27}&-2^{15}3^15^3&\dfrac{5^{13}}{2^{11}3^9}&\dfrac{2^{8555}}{3^{2187}5^{2161}}&\dfrac{5^{41533}}{2^{2699}3^{59049}}\\[6pt]i\sqrt{7}& \mathfrak Z_{-28}&3^{3}5^{3}17^3&\left( \dfrac{3^{7}17^{3}}{5^{9}} \right)^2&\left( \dfrac{17^{637}}{3^{1323}5^{203}} \right)^2&\left( \dfrac{3^{15435}5^{17563}}{17^{15953}} \right)^4\\[9pt]\dfrac{1+i\sqrt{43}}{2}&\mathfrak Z_{-43}&-2^{18}3^35^{3}&\dfrac{2^{42}}{3^{19}5^3}&\dfrac{3^{4239}5^{6959}}{2^{22806}}&\dfrac{3^{271773}}{2^{213078}5^{93683}}\\[8pt]\dfrac{1+i\sqrt{67}}{2}&\mathfrak Z_{-67}&-2^{15}3^35^{3}11^3&\dfrac{3^{29}11^{21}}{2^{51}5^{27}}&\dfrac{5^{11999}11^{959}}{2^{17285}3^{8721}}&\dfrac{3^{153837}5^{939013}}{2^{298819}11^{614315}}\\[8pt]\dfrac{1+i\sqrt{163}}{2}&\mathfrak Z_{-163}&-2^{18}3^35^{3}23^329^3&\dfrac{23^{21}29^{69}}{2^{198}3^{139}5^3}&\dfrac{5^{95279}23^{68159}}{2^{94326}3^{131841}29^{46561}}&\dfrac{2^{81641562}29^{1033765}}{3^{7144587}5^{7794803}23^{12653867}}\\[8pt]\hline\hline\end{array}\end{aligned}$$
\[rmk:GKZ\_class\_number\_one\]For some special points $z,z'\in\mathfrak H$ and $ \varGamma=SL(2,\mathbb Z)\equiv\varGamma_0(1)$, the Gross–Kohnen–Zagier algebraicity conjecture (boxed equation in §\[subsec:background\]) can be directly verified by the methods developed in the joint works of Gross–Zagier [@GrossZagierI] and Gross–Kohnen–Zagier [@GrossZagierII]. For example, if one defines the totality of CM points whose minimal polynomial has discriminant $D$ by $ \mathfrak Z_D:=\{z\in\mathfrak H|\exists a,b,c \in\mathbb Z,a>0,\gcd(a,b,c)=1,b^2-4ac=D,az^2+bz+c=0\}$, then one can show that [@ZagierKyushuJ1 p. 50, Theorem II.2] $$\begin{aligned}
\exp\left[\frac{(D_1D_2)^{(k-2)/4}}{2}G_{k/2}^{\mathfrak H/PSL(2,\mathbb Z)}(z_1,z_2)\right]\in\mathbb Q,\quad k=6,10,14\label{eq:Zagier_Q}\end{aligned}$$holds for $ z_1\in\mathfrak Z_{D_1},z_2\in\mathfrak Z_{D_2}$, where the discriminants $ D_1$ and $D_2$ are two distinct members of the finite set $ \{-3$, $-$4, $-$7, $-$8, $-$11, $-$12, $-$16, $-$19, $-$27, $-$28, $-$43, $-$67, $-163\}$. Fixing the point $ z_1=e^{\pi i/3}\in\mathfrak Z_{-3} $, one can tabulate the special values of $G_3^{\mathfrak H/PSL(2,\mathbb Z)}(e^{\pi i/3},z) $, $G_5^{\mathfrak H/PSL(2,\mathbb Z)}(e^{\pi i/3},z) $ and $G_7^{\mathfrak H/PSL(2,\mathbb Z)}(e^{\pi i/3},z) $ that support the statement in Eq. \[eq:Zagier\_Q\] (Table \[tab:Zagier\_Q\_examples\]). The exact factorizations of all the rational numbers in Table \[tab:Zagier\_Q\_examples\] can be justified by the Gross–Kohnen–Zagier theory (*e.g.* [@ZagierKyushuJ1 p. 76, Example II.6]). One may also wish to check these tabulated values numerically[^7] against the integral formulae in Eqs. \[eq:GT3\_KZ\]–\[eq:GT7\_KZ\].
Some Analytic Tools for Kontsevich–Zagier Integrals and Gross–Zagier Renormalization\[sec:analysis\_KZ\_int\_Part1\]
====================================================================================================================
In §§\[subsec:KZ\_integrals\] and \[subsec:high\_weight\_KZ\], we have constructed Kontsevich–Zagier integral representations for automorphic Green’s functions $ G_{k/2}^{\mathfrak H/\overline\varGamma_0(N)}(z,z')$ with even weights $ k\geq4$ satisfying the cusp-form-free condition $ \dim\mathcal S_k(\varGamma_0(N))=0$. We have also computed a few special values of weight-4 automorphic Green’s functions from their respective Kontsevich–Zagier integrals (see Remark \[rmk:spec\_val\_G2\]).
In this section, we prepare a few analytic results that enable us to compute various Kontsevich–Zagier integrals for automorphic Green’s functions, and work out closed-form results for some special scenarios, as announced in Theorem \[thm:GZ\_rn\]. The generic cases will be treated systematically in subsequent work.
In §\[subsec:hypergeo\_Pnumu\], we present some hypergeometric techniques for evaluating certain integrals involving products of Legendre functions. These integrals over Legendre functions will lead to special values of weight-4 “Gross–Zagier renormalized” automorphic Green’s functions in §\[subsec:int\_repn\_G2\_GZ\_rn\]. Here, the Gross–Zagier renormalization [@GrossZagierI Chap. II, §5] is a procedure to subtract logarithmic divergence of automorphic Green’s functions on the “diagonal points”. In §\[subsec:Ramanujan\_Jacobi\], we fulfill two tasks. We first recall some geometric transformations inspired by Ramanujan’s Notebooks, which facilitate the reduction of certain multiple elliptic integrals; we then combine Ramanujan’s transformations with the Jacobi elliptic functions to evaluate some non-trivial multiple elliptic integrals. The integral identities derived in §\[subsec:Ramanujan\_Jacobi\] will assist in the quantitative analysis of weight-4 “Gross–Zagier renormalized” automorphic Green’s functions on $ \overline\varGamma_0(4)$ in §\[subsec:G2\_Hecke4\_GZ\_rn\].
In §§\[subsec:Ramanujan\_Jacobi\]–\[subsec:G2\_Hecke4\_GZ\_rn\], we will draw extensively on modular transformations. In addition to the degree-1 modular transformations (Eqs. \[eq:im\_mod\]–\[eq:lambda\_transf\]), we need Landen’s transformations (sometimes named after Landen and Gau[ß]{} [@ByrdFriedman items 163.02 and 164.02], depending on context) for the complete elliptic integrals of the first kind:$$\begin{aligned}
\mathbf K(\sqrt{1-\lambda})={}&\frac{2}{1+\sqrt{\lambda}}\mathbf K\left( \frac{1-\sqrt{\lambda}}{1+\sqrt{\lambda}} \right),\quad \lambda\in\mathbb C\smallsetminus(-\infty,0];\label{eq:Landen_1}\\\mathbf K(\sqrt{\lambda})={}&\frac{1}{1+\sqrt{\lambda}}\mathbf K\left( \frac{2\sqrt[4]{\lambda}}{1+\sqrt{\lambda}} \right),\quad |\lambda|<1.\label{eq:Landen_2}\end{aligned}$$ With the “$ \lambda$-$ \mathbf K$ relation” (Eq. \[eq:lambda\_K\_ratio\_z\]), one can derive from Landen’s transformations the “duplication formula” and the “dimidiation formula” [@WeberVol3 §135] for the modular lambda function $ \lambda(z)=16[\eta (z/2)]^8 [\eta (2 z)]^{16}/[\eta (z)]^{24}$:$$\begin{aligned}
\label{eq:double_half_lambda}\lambda(2z)=\left[ \frac{1-\sqrt{1-\lambda(z)}}{1+\sqrt{1-\lambda(z)}} \right]^2,\quad \lambda\left( \frac{\vphantom{1}z}{2} \right)=\frac{4\sqrt{\lambda(z)}}{[1+\sqrt{\lambda(z)}]^2},\quad -1<\operatorname{Re}z<1,\left|z+\frac12\right|>\frac12,\left|z-\frac12\right|>\frac{1}{2}.\end{aligned}$$ We shall refer to Eqs. \[eq:Landen\_1\]–\[eq:double\_half\_lambda\] (and descendants thereof) collectively as degree-2 modular transformations.
Apart from hypergeometric, modular and geometric transformations, the calculations in this section are essentially applications of residue calculus to elliptic integrals, the prototype of which is presented in the next paragraph.
Let $ f(\lambda),\lambda\in\mathbb C\smallsetminus[1,+\infty)$ be a complex analytic function in the slit plane with well-defined one-sided limits $ f(\lambda\pm i0^+)$ for $ \lambda>1$, satisfying the bounds $ |f(\lambda)|=O(\log^\nu|1-\lambda|)$ for $\nu\in\mathbb R\smallsetminus\{0\},\lambda\to1$ and $ |f(\lambda)|=O(|\lambda|^{-\nu'})$ for $0<\nu'<1,|\lambda|\to+\infty$. One might use Cauchy’s integral formula to verify that $$\begin{aligned}
f(\lambda)=\frac{1}{2\pi i}\int_0^1\frac{f\big(\frac1\mu-i0^+\big)-f\big(\frac{1}{\mu}+i0^+\big)}{1-\lambda\mu}\frac{\operatorname{d}\mu}{\mu},\quad \lambda\in\mathbb C\smallsetminus[1,+\infty), \label{eq:f_slit_plane_Cauchy_int}\end{aligned}$$where the integration is carried along the open unit interval $ (0,1)$ in the complex $ \mu$-plane. Setting $ f(\lambda)=[\mathbf K(\sqrt{\lambda})]^n,n\in\mathbb Z_{>0}$ in Eq. \[eq:f\_slit\_plane\_Cauchy\_int\] and referring to the inverse modulus transformation (Eq. \[eq:inv\_mod\]), one obtains the integral identity $$\begin{aligned}
{[}\mathbf K(\sqrt{\lambda})]^n={}&\frac{1}{\pi}\operatorname{Im}\int_0^1\frac{\mathbf [\mathbf K(\sqrt{\smash[b]{\vphantom{1}\mu}})+i\mathbf K(\sqrt{\smash[b]{1-\mu}})]^n}{1-\lambda\mu}\mu^{(n-2)/2}\operatorname{d}\mu,\quad 0<\lambda<1,n\in\mathbb Z_{>0}.\label{eq:Kn}\tag{\ref{eq:f_slit_plane_Cauchy_int}-K$^{n}$}\end{aligned}$$ In recent literature [@Wan2012; @Zhou2013Pnu], one may find discussions on the particular cases where $ n=1$ and $2$:$$\begin{aligned}
&&\mathbf K(\sqrt{\lambda})={}&\frac{1}{\pi}\int_0^1\frac{\mathbf K(\sqrt{\smash[b]{1-\mu}})}{1-\lambda\mu}\frac{\operatorname{d}\mu}{\sqrt{\smash[b]{\mu}}},&& \lambda\in\mathbb C\smallsetminus[1,+\infty),&&
\label{eq:K1}\tag{\ref{eq:f_slit_plane_Cauchy_int}-K$^{1}$}\\&&{[}\mathbf K(\sqrt{\lambda})]^2={}&\frac{2}{\pi}\int_0^1\frac{\mathbf K(\sqrt{\smash[b]{\vphantom{1}\mu}})\mathbf K(\sqrt{\smash[b]{1-\mu}})}{1-\lambda\mu}\operatorname{d}\mu,&& \lambda\in\mathbb C\smallsetminus[1,+\infty),&&\label{eq:K2}\tag{\ref{eq:f_slit_plane_Cauchy_int}-K$^{2}$}\end{aligned}$$which were proved by either combinatorial algorithms [@Wan2012 Eqs. 26–27] or geometric transformations [@Zhou2013Pnu Eqs. 31 and 40$^*$].
Hypergeometric Evaluations of Some Definite Integrals Involving Associated Legendre Functions\[subsec:hypergeo\_Pnumu\]
-----------------------------------------------------------------------------------------------------------------------
For $ -1<\nu<0$, we have already defined two real-valued functions $ P_\nu(t),t>1$ and $ Q_\nu(t),t>1$ using Eqs. \[eq:Euler\_int\] and \[eq:Q\_nu\_Laplace\_int\], respectively. In this subsection, we also need the associated Legendre functions $ P^\mu_\nu(t),t>1$ and $ Q^\mu_\nu(t),t>1$ where $ \mu>-1/2$. A convenient way to introduce these associated Legendre functions is to use Hobson’s integral representations [@Hobson1931 p. 270, Eq. 139 and p. 276, Eq. 150]: $$\begin{aligned}
P^{\mu}_\nu(\cosh\psi):={}&\frac{2^{\mu+1}(\sinh\psi)^\mu}{\sqrt{\pi}\Gamma(\frac{1}{2}-\mu)}\int_0^\psi\frac{\cosh\frac{(2\nu+1)u}{2}\operatorname{d}u}{(2\cosh\psi-2\cosh u)^{\mu+\frac{1}{2}}},&& \mu>-\frac{1}{2},\psi>0;\label{eq:Pmunu_defn}\\Q^{\mu}_\nu(\cosh\psi):={}& e^{i\mu\pi}2^\mu\frac{\sqrt{\pi}(\sinh\psi)^\mu}{\Gamma(\frac{1}{2}-\mu)}\int_\psi^\infty\frac{e^{-(\nu+\frac{1}{2})u}\operatorname{d}u}{(2\cosh u-2\cosh \psi)^{\mu+\frac{1}{2}}},&&\mu>-\frac{1}{2},\mu+\nu+1>0,\psi>0,\label{eq:Qmunu_defn}\end{aligned}$$where the integrals are taken along the real axis, so that one always raises a real-valued base (with zero phase) to a positive power of $ \mu+\frac{1}{2}$. The definitions of associated Legendre functions in Eqs. \[eq:Pmunu\_defn\]–\[eq:Qmunu\_defn\] are compatible with the previously defined Legendre functions, in the sense that $ P_\nu^{\vphantom0}(t)=P_\nu^{0}(t),t>1$ and $ Q_\nu^{\vphantom0}(t)=Q_{\nu}^{0}(t),t>1$. Furthermore, these associated Legendre functions satisfy Whipple’s relation [@Hobson1931 p. 245, Eq. 92]:$$\begin{aligned}
Q^{\mu}_\nu(\cosh\psi)={}&e^{i\mu\pi}\sqrt{\frac{\pi}{2}}\frac{\Gamma(\mu+\nu+1)}{\sqrt{\smash[b]{\sinh\psi}}}P^{-\nu-1/2}_{-\mu-1/2}(\coth\psi),\label{eq:Whipple1}\end{aligned}$$whenever $ \psi>0$ and the associated Legendre functions on both sides are defined in the domains specified by Eqs. \[eq:Pmunu\_defn\]–\[eq:Qmunu\_defn\]. Using Whipple’s relation (Eq. \[eq:Whipple1\]), one can readily verify the following formula: $$\begin{aligned}
\int_1^\infty\frac{[P_\nu^{\mu}(\xi)]^2}{\xi^2}\operatorname{d}\xi={}&-\frac{2e^{2i\nu\pi}}{\pi[\Gamma(-\mu-\nu)]^2}\int_1^\infty\frac{\big[Q^{-\nu-1/2}_{-\mu-1/2}(\xi)\big]^2}{\xi^{2}}\operatorname{d}\xi,\label{eq:PnumuPnumu_int_Whipple}\end{aligned}$$provided that both integrals are well-defined. We refer to such an integral identity as “Legendre–Whipple duality”.
In the following proposition, we will present some integral formulae involving associated Legendre functions $ P^\mu_\nu$ and $ Q^\mu_\nu$, with the (usually) tacit assumptions that the order $ \mu$ and the degree $\nu$ are appropriately chosen to meet various requirements.
1. We have the following integral identities:$$\begin{aligned}
\int_1^\infty\frac{[Q_\nu^{\mu}(\xi)]^2}{\xi^2}\operatorname{d}\xi={}&\frac{e^{2i\mu\pi}}{2(1+\nu ) (1+\mu +\nu )}\frac{\mu\pi}{\sin(\mu\pi)} \frac{\Gamma (1+\mu +\nu )}{\Gamma (1-\mu +\nu )}{_3}F_2\left(\left.\begin{array}{c}1,1+\mu,1+\mu+\nu\\2+\nu ,2+\mu+\nu \end{array}\right|1\right)\notag\\={}&\frac{e^{2i\mu\pi}}{2(1+\nu ) (1-\mu +\nu )}\frac{\mu\pi}{\sin(\mu\pi)} \frac{\Gamma (1+\mu +\nu )}{\Gamma (1-\mu +\nu )}{_3}F_2\left(\left.\begin{array}{c}1,1-\mu,1-\mu+\nu\\2+\nu ,2-\mu+\nu \end{array}\right|1\right)\notag\\={}&\frac{e^{2i\mu\pi}}{2}\frac{\mu\pi}{\sin(\mu\pi)} \left[\frac{\Gamma (1+\mu +\nu )}{\Gamma (2+\nu )}\right]^2{_3}F_2\left(\left.\begin{array}{c}1+\mu,1-\mu,1+\nu\\2+\nu ,2+\nu \end{array}\right|1\right).\label{eq:QnumuQnumu_xx_int_3F2}\end{aligned}$$In particular, we have$$\begin{aligned}
\int_1^\infty\frac{[Q_\nu(\xi)]^2}{\xi^2}\operatorname{d}\xi={}&\frac{1}{2(\nu+1)^{2}}{_3}F_2\left(\left.\begin{array}{c}1,1,\nu+1\\\nu +2 ,\nu +2\end{array}\right|1\right)\label{eq:QnuQnu_xx_int_3F2}\end{aligned}$$and $$\begin{aligned}
\int_1^\infty\frac{[P_\nu(\xi)]^2}{\xi^2}\operatorname{d}\xi={}&\frac{(2\nu+1)\tan(\nu\pi)}{\nu\pi}{_3}F_2\left(\left.\begin{array}{c}1,\frac{1}{2}-\nu,-\nu\\[4pt]\frac{3}{2},1-\nu \end{array}\right|1\right)\notag\\={}&-\frac{(2\nu+1)\tan(\nu\pi)}{(\nu+1)\pi}{_3}F_2\left(\left.\begin{array}{c}1,\frac{3}{2}+\nu,1+\nu\\[4pt]\frac{3}{2},2+\nu \end{array}\right|1\right)\notag\\={}&\frac{2(2\nu+1)}{\pi\cos(\nu\pi)}{_3}F_2\left(\left.\begin{array}{c}\frac{1}{2},\frac{3}{2}+\nu,\frac{1}{2}-\nu\\[4pt]\frac{3}{2},\frac{3}{2} \end{array}\right|1\right).\label{eq:PnuPnu_xx_int_3F2}\end{aligned}$$Furthermore, we have the following closed-form evaluation for $\nu\in(-1,-1/2)\cup(-1/2,0) $:$$\begin{aligned}
\int_1^\infty\frac{[P_\nu(\xi)]^2}{\xi^2}\operatorname{d}\xi={}&\frac{2}{\pi}\int_0^{\pi/2}\frac{\sin(2\nu+1)\theta}{\sin\theta\cos(\nu\pi)}\operatorname{d}\theta=\frac{1}{\cos(\nu\pi)}+\frac{\tan(\nu\pi)}{\pi}\left[ \psi ^{(0)}\left(\frac{\nu+2 }{2}\right)-\psi ^{(0)}\left(\frac{\nu+1 }{2}\right )\right].
\label{eq:PnuPnu_xx_int_closed_form}\end{aligned}$$
2. The following integral formula holds:$$\begin{aligned}
&
\int_1^\infty\frac{[P_\nu^{\mu}(\xi)]^2}{\xi^2}\operatorname{d}\xi\notag\\={}&\frac{\psi ^{(0)}(\frac{1}{2}-\nu )-\psi ^{(0)}(-\mu -\nu )-\psi ^{(0)}(1-\mu +\nu )-\gamma_{0}+\dfrac{2(\mu ^{2}-\nu^{2} )}{1-2\nu}{_4} F_3\left(\left.\begin{array}{c}1,1,1-\mu -\nu ,1+\mu -\nu \\[4pt]2,2,\frac{3}{2}-\nu \end{array}\right|1\right)}{\cos(\mu\pi)\Gamma (1-\mu +\nu ) \Gamma (-\mu -\nu )},\label{eq:PnumuPnumu_xx_int_4F3}\end{aligned}$$ where $ \gamma_0=-\psi^{(0)}(1)$ is the Euler–Mascheroni constant. In particular, one has$$\begin{aligned}
&
\int_1^\infty\frac{[P_\nu(\xi)]^2}{\xi^2}\operatorname{d}\xi\notag\\={}&\frac{\sin(\nu\pi)}{\pi}\left[ \frac{2}{\nu }+\pi \cot (\nu\pi)-\psi ^{(0)}\left(\frac{1}{2}-\nu \right)+2 \psi ^{(0)}(\nu )+\gamma _{0}+\frac{2\nu ^2}{1-2\nu}{ _4}F_3\left(\left.\begin{array}{c}1,1,1-\nu ,1-\nu \\[4pt]2,2,\frac{3}{2}-\nu \end{array}\right|1\right)\right]\label{eq:PnuPnu_xx_int_4F3}\end{aligned}$$ and$$\begin{aligned}
\int_1^\infty\frac{[Q_\nu(\xi)]^2}{\xi^2}\operatorname{d}\xi={}&\frac{\pi}{\sin(\nu\pi)}\left[ \psi ^{(0)}(\nu +1)+\gamma_{0}-\frac{\nu(\nu+1)}{2} {_4}F_3\left(\left.\begin{array}{c}1,1,1-\nu ,2+\nu \\[4pt]2,2,2 \end{array}\right|1\right)\right].\label{eq:QnuQnu_xx_int_4F3}\end{aligned}$$Furthermore, the following formula is true for $ \nu\in(-1,0)$:$$\begin{aligned}
\int_1^\infty\frac{[Q_\nu(\xi)]^2}{\xi^2}\operatorname{d}\xi={}&\frac{\pi[\psi ^{(0)}(\nu +1)-\log 2]}{\sin(\nu\pi)}+\frac{\left[\psi ^{(0)}\left(\frac{\nu+2 }{2}\right)+\psi ^{(0)}\left(\frac{1-\nu}{2}\right)\right]^2-\left[\psi ^{(0)}\left(\frac{\nu +1}{2}\right)+\psi ^{(0)}\left(-\frac{\nu }{2}\right)\right]^2}{8}\notag\\&+\frac{\psi ^{(1)}\left(\frac{\nu +1}{2}\right)+\psi ^{(1)}\left(-\frac{\nu }{2}\right)-\psi ^{(1)}\left(\frac{\nu+2 }{2}\right)-\psi ^{(1)}\left(\frac{1-\nu}{2}\right)}{8}.\label{eq:QnuQnu_xx_int_closed_form}\end{aligned}$$
3. The following identity holds for $ -1<\nu<0$:$$\begin{aligned}
\int_1^\infty\frac{P_\nu(\xi)Q_{\nu}(\xi)}{\xi^2}\operatorname{d}\xi={}&\frac{1}{2}\left[ \frac{\Gamma (\nu +1)}{\Gamma (\nu +\frac{3}{2})} \right]^2{_3}F_2\left(\left.\begin{array}{c}\frac{1}{2},\frac{1}{2},\nu+\frac{1}{2}\\[4pt]\nu +\frac{3}{2} ,\nu +\frac{3}{2}\end{array}\right|1\right)=\frac{1}{2}\left[ \psi ^{(0)}\left(\frac{\nu+2 }{2}\right)-\psi ^{(0)}\left(\frac{\nu+1 }{2}\right )\right],\label{eq:PnuQnu_xx_int_closed_form}\end{aligned}$$and the integral formula $$\begin{aligned}
\int_{-1}^1\frac{P_\nu(\xi)P_\nu(-\xi)-[P_\nu(0)]^{2}}{\xi^2}\operatorname{d}\xi=2\{[P_\nu(0)]^{2}-1\},\quad \forall\nu\in\mathbb C\label{eq:minus2_id'}\end{aligned}$$is true so long as the contour of integration lies in the double-slit plane $ \xi\in\mathbb C\smallsetminus((-\infty,-1]\cup[1,+\infty))$.
Additionally, we have an integral formula for $ -1<\nu<0$: [$$\begin{aligned}
\operatorname{Re}\int_{-1}^1\frac{[P_\nu(\xi)]^2}{\xi^2}\operatorname{d}\xi={}&-2\cos^2(\nu\pi)\int_1^\infty\frac{[P_\nu(\xi)]^2}{\xi^2}\operatorname{d}\xi+\frac{4\sin(\nu\pi)\cos(\nu\pi)}{\pi}\int_1^\infty\frac{P_\nu(\xi)Q_{\nu}(\xi)}{\xi^2}\operatorname{d}\xi\notag\\{}&-\frac{4\sin^2(\nu\pi)}{\pi^2}\int_1^\infty\frac{[Q_\nu(\xi)]^2}{\xi^2}\operatorname{d}\xi\notag\\={}&-2\cos(\nu\pi)-\frac{4 \sin ( \nu\pi ) [ \psi ^{(0)}(\nu +1)-\log2]}{\pi }\notag\\&-\frac{\sin ^2( \nu\pi ) }{2 \pi ^2}\left\{\left[\psi ^{(0)}\left(\frac{\nu+2 }{2}\right)+\psi ^{(0)}\left(\frac{1-\nu}{2}\right)\right]^2-\left[\psi ^{(0)}\left(\frac{\nu +1}{2}\right)+\psi ^{(0)}\left(-\frac{\nu }{2}\right)\right]^2\right\}\notag\\&-\frac{\sin ^2( \nu\pi )}{2 \pi ^2} \left[\psi ^{(1)}\left(\frac{\nu +1}{2}\right)+\psi ^{(1)}\left(-\frac{\nu }{2}\right)-\psi ^{(1)}\left(\frac{\nu+2 }{2}\right)-\psi ^{(1)}\left(\frac{1-\nu}{2}\right)\right],\label{eq:PnuPnu_xx_int_unit_interval_polygamma}\end{aligned}$$]{}which generalizes the evaluations in Proposition \[prop:Pnu\_sec\_L\_weight4\].
4. For $ -1<\nu<0$, we have$$\begin{aligned}
&
\int_{0}^1\left\{\frac{[P_{\nu}(\xi)]^2-[P_{\nu}(-\xi)]^2}{\xi^{2}}-\frac{4\sin(\nu\pi)}{\pi \xi}\right\}\operatorname{d}\xi+\operatorname{Re}\int_{-1}^1\frac{[P_\nu(\xi)]^2}{\xi^2}\operatorname{d}\xi\notag\\={}&-2\cos^2(\nu\pi)\int_1^\infty\frac{[P_\nu(\xi)]^2}{\xi^2}\operatorname{d}\xi-\frac{4}{\pi}\int_{0}^1\frac{P_{\nu}(\xi)P_{\nu}(-\xi)-[P_\nu(0)]^{2}}{\xi^{2}}\log \xi\operatorname{d}\xi+\frac{4[P_\nu(0)]^2\sin(\nu\pi)}{\pi}\notag\\={}&\frac{4\sin(\nu\pi) [1-\gamma_{0} - \psi ^{(0)}(\nu +1)-\log 2]}{\pi }-2\cos(\nu\pi).
\label{eq:Pnu_sqr_diff_int_log_over_sqr}\end{aligned}$$For $ -1/2<\nu<0$, we have$$\begin{aligned}
&
\int_1^\infty\left[ P_\nu(\xi)Q_\nu(\xi)-\frac{1}{(2\nu+1)\xi} \right]\operatorname{d}\xi\notag\\={}&\frac{\tan(\nu\pi)}{\pi}\int_1^\infty[Q_\nu(\xi)]^2\operatorname{d}\xi-\frac{\pi}{4}\int_0^1\frac{[P_\nu(\xi)]^2-[P_\nu(-\xi)]^2}{\sin(\nu\pi)\cos(\nu\pi)}\operatorname{d}\xi-\int_{0}^1\frac{P_{\nu}(\xi)P_{\nu}(-\xi)}{\cos(\nu\pi)}\log \xi\operatorname{d}\xi\notag\\={}&\frac{\pi \tan ( \nu\pi )+\psi ^{(0)}\left(\frac{1}{2}-\nu \right)-2 \psi ^{(0)}(\nu )+\psi ^{(0)}\left(\nu +\frac{3}{2}\right)+2 \log 2}{2 (2 \nu +1)}-\frac{1}{\nu (2 \nu +1)}.\label{eq:PQ-1_over_x_int}\end{aligned}$$
<!-- -->
1. We may compute $$\begin{aligned}
&\int_1^\infty\frac{[Q_\nu^{\mu}(\xi)]^2}{\xi^2}\operatorname{d}\xi=\int_0^\infty\frac{[Q_\nu^{\mu}(\sqrt{1+t})]^2}{2(1+t)^{3/2}}\operatorname{d}t\notag\\={}&\frac{e^{2i\mu\pi}}{2\pi i}\frac{\Gamma(1+\mu+\nu)}{\Gamma(1-\mu+\nu)}\int_{c-i\infty}^{c+i\infty}\frac{\Gamma (s)\Gamma (s+\mu)\Gamma(s-\mu)\ \Gamma (\nu +1-s)\Gamma(1-s) }{ 2\Gamma (\nu +1+s)}\operatorname{d}s\label{eq:QmunuQmunu_xx_int_comp}\end{aligned}$$from the following Mellin inversion formula [@Marichev1983 §10.11, Eq. 54(1)] $$\begin{aligned}
{[}Q_\nu^{\mu}(\sqrt{1+t})]^2={}&\frac{\sqrt{\pi}}{2}\frac{e^{2i\mu\pi}}{2\pi i}\frac{\Gamma(1+\mu+\nu)}{\Gamma(1-\mu+\nu)}\int_{c-i\infty}^{c+i\infty}\frac{\Gamma (s)\Gamma (s+\mu)\Gamma(s-\mu)\ \Gamma (\nu +1-s) }{ \Gamma (\nu +1+s)\Gamma (s+\frac{1}{2})}\frac{\operatorname{d}s}{t^s},&&t>0,|\mu|<c<\nu+1\end{aligned}$$and an elementary identity $ \int_0^\infty t^{-s}(1+t)^{-3/2}\operatorname{d}t=2\Gamma(1-s)\Gamma(s+\frac{1}{2})/\sqrt{\pi}$ for $ -1/2<\operatorname{Re}s<1$. Here, by convention, the integrations $\int_{c-i\infty}^{c+i\infty}(\cdots)\operatorname{d}s:= \lim_{T\to +\infty}\int_{c-iT}^{c+iT}(\cdots)\operatorname{d}s
$ are carried out along a vertical line $\operatorname{Re}s = c$.
With the help of Barnes’s second lemma [@Slater Eq. 4.2.2.1], one can reduce the contour integral in Eq. \[eq:QmunuQmunu\_xx\_int\_comp\] into particular values of generalized hypergeometric series $ _3F_2$, as claimed in Eq. \[eq:QnumuQnumu\_xx\_int\_3F2\]. Specializing Eq. \[eq:QnumuQnumu\_xx\_int\_3F2\] to $ Q_\nu^{\vphantom0}=Q_\nu^0$, one obtains Eq. \[eq:QnuQnu\_xx\_int\_3F2\]. Applying the Legendre–Whipple duality in Eq. \[eq:PnumuPnumu\_int\_Whipple\] to Eq. \[eq:QnumuQnumu\_xx\_int\_3F2\], one can verify Eq. \[eq:PnuPnu\_xx\_int\_3F2\].
In the last line of Eq. \[eq:PnuPnu\_xx\_int\_3F2\], one can represent the generalized hypergeometric series $ _3F_2$ as a definite integral involving the hypergeometric series $ _2F_1$ [@GradshteynRyzhik item 7.512.5]:$$\begin{aligned}
&
{_3}F_2\left(\left.\begin{array}{c}\frac{1}{2},\frac{3}{2}+\nu,\frac{1}{2}-\nu\\[4pt]\frac{3}{2},\frac{3}{2} \end{array}\right|1\right)=\frac{1}{2}\int_0^1{_2}F_1\left(\left.\begin{array}{c}\frac{3}{2}+\nu,\frac{1}{2}-\nu\\[4pt]\frac{3}{2} \end{array}\right|u\right)\frac{\operatorname{d}u}{\sqrt{u}}=\frac{1}{2}\int_0^1\frac{\sin \left((2 \nu+1 )\arcsin\sqrt{u} \right)}{(2\nu+1)\sqrt{u(1-u)}}\frac{\operatorname{d}u}{\sqrt{u}},\end{aligned}$$where we have spelt out the hypergeometric series $_2F_1 $ in the integrand using elementary functions [@GradshteynRyzhik item 9.121.30]. After a trigonometric substitution $ u=\sin^2\theta$, we arrive at the first identity in Eq. \[eq:PnuPnu\_xx\_int\_closed\_form\].
We can compute the integral over $ \theta$ via integration by parts and a familiar Fourier expansion [@GradshteynRyzhik item 1.422.2]: $$\begin{aligned}
\int_0^{\pi/2}\frac{\sin(2\nu+1)\theta}{\sin\theta}\operatorname{d}\theta={}&(2\nu+1)\int_0^{\pi/2}\cos(2\nu+1)\theta\log\cot\frac{\theta}{2}\operatorname{d}\theta\notag\\={}&2(2\nu+1)\int_0^{\pi/2}\cos(2\nu+1)\theta\sum_{\ell=0}^\infty\frac{\cos(2\ell+1)\theta}{2\ell+1}\operatorname{d}\theta\notag\\={}&\frac{\sin ( \nu\pi)}{2}\sum_{\ell=0}^{\infty}(-1)^\ell\left(\frac{1 }{ \ell+\nu +1}-\frac{1 }{\ell-\nu}\right),\label{eq:angle_int_psi}\end{aligned}$$before invoking the standard partial fraction expansion for the digamma function $ \psi^{(0)}(z)$. Clearly, the $ \nu=-1/2$ scenario can be recovered in a limit procedure:$$\begin{aligned}
\int_1^\infty\frac{[P_{-1/2}(\xi)]^2}{\xi^2}\operatorname{d}\xi=\lim_{\nu\to-1/2}\int_1^\infty\frac{[P_\nu(\xi)]^2}{\xi^2}\operatorname{d}\xi={}&\frac{4}{\pi^{2}}\int_0^{\pi/2}\frac{\theta\operatorname{d}\theta}{\sin\theta}=\frac{8G}{\pi^{2}}.\end{aligned}$$
2. With the following Mellin inversion formula [@Marichev1983 §10.11, Eq. 28(1)] $$\begin{aligned}
&
[P_{\nu}^{\mu}(\sqrt{1+t})]^2\notag\\= {}&\frac{1}{\sqrt{\pi}\Gamma(1-\mu+\nu)\Gamma(-\mu-\nu)}\frac{1}{2\pi i}\int_{c-i\infty}^{c+i\infty}\frac{\Gamma (s- \mu) \Gamma (\frac{1}{2}-s) \Gamma (\nu +1-s) \Gamma (-\nu-s )}{\Gamma (1-s)\Gamma(1-\mu-s)}\frac{\operatorname{d}s}{t^s},&&\mu<c<\min\{-\nu,\nu+1,1/2\},\label{eq:PnuPnu_Mellin}\end{aligned}$$one can compute $$\begin{aligned}
&
\int_1^\infty\frac{[P_\nu^{\mu}(\xi)]^2}{\xi^2}\operatorname{d}\xi=\int_0^\infty\frac{[P_\nu^{\mu}(\sqrt{1+t})]^2}{2(1+t)^{3/2}}\operatorname{d}t\notag\\={}&\frac{1}{{\pi}\Gamma(1-\mu+\nu)\Gamma(-\mu-\nu)}\frac{1}{2\pi i}\int_{c-i\infty}^{c+i\infty}\frac{ \Gamma (\frac{1}{2}-s) \Gamma (\nu +1-s) \Gamma (-\nu-s )\Gamma (s- \mu)\Gamma (\frac{1}{2}+s)}{\Gamma(1-\mu-s)}\operatorname{d}s\notag\\={}&\frac{1}{{\pi}\Gamma(1-\mu+\nu)\Gamma(-\mu-\nu)}G_{3,3}^{2,3}\left(1\left|
\begin{array}{c}
\frac{1}{2},-\nu ,\nu +1 \\
-\mu ,\frac{1}{2},\mu \\
\end{array}
\right.\hspace{-.5em}\right).\end{aligned}$$By taking the limit $ \xi\to1$ in the following reduction of the Meijer $G$-function $ G^{2,3}_{3,3}$:$$\begin{aligned}
G_{3,3}^{2,3}\left(\xi\left|
\begin{array}{c}
\frac{1}{2},-\nu ,\nu +1 \\
-\mu ,\frac{1}{2},\mu \\
\end{array}
\right.\hspace{-.5em}\right)={}&\frac{2 \pi (2 \nu +1) \sqrt{\xi} }{(2 \mu -1) (2 \mu +1)\cos(\nu\pi)}{_3}F_2\left(\left.\begin{array}{c}1,\frac{1}{2}-\nu ,\nu +\frac{3}{2}\\\frac{3}{2}-\mu ,\mu +\frac{3}{2}\end{array}\right|\xi\right)\notag\\&+\frac{\pi \Gamma (1-\mu +\nu ) \Gamma (-\mu -\nu ) }{\Gamma (1-2 \mu )\xi^{\mu}\cos(\mu\pi)} {_2}F_1\left(\left.\begin{array}{c}-\mu -\nu ,1-\mu +\nu \\1-2 \mu\end{array}\right|\xi\right),\end{aligned}$$one arrives at Eq. \[eq:PnumuPnumu\_xx\_int\_4F3\], after somewhat laborious computations. One can deduce Eq. \[eq:PnuPnu\_xx\_int\_4F3\] as a special case of Eq. \[eq:PnumuPnumu\_xx\_int\_4F3\] with $ P_\nu^{\vphantom0}=P^0_\nu$. By the Legendre–Whipple duality (Eq. \[eq:PnumuPnumu\_int\_Whipple\]), we obtain Eq. \[eq:QnuQnu\_xx\_int\_4F3\] as a particular situation of Eq. \[eq:PnumuPnumu\_xx\_int\_4F3\] for $ P_{-1/2}^{-\nu-1/2}$.
One may also derive Eq. \[eq:QnuQnu\_xx\_int\_4F3\] directly from Eq. \[eq:QnuQnu\_xx\_int\_3F2\], without going through the Meijer $ G$-function and Legendre–Whipple duality. To show this, we start by appealing to Kummer’s $ {_3}F_2$ transformation [@AAR Corollary 3.3.5] $$\begin{aligned}
\label{eq:Kummer3F2}
{_3F_2}\left( \left.\begin{array}{c}
a_{1},a_{2},a_{3} \\[4pt]
b_{1} ,b_{2}\\
\end{array}\right| 1\right)=\frac{\Gamma(b_2)\Gamma(b_1+b_2-a_1-a_2-a_3)}{\Gamma(b_{2}-a_{1})\Gamma(b_{1}+b_{2}-a_{2}-a_{3})}{_3F_2}\left( \left.\begin{array}{c}
a_{1},b_{1}-a_{2},b_{1}-a_{3} \\[4pt]
b_{1} ,b_{1}+b_{2}-a_2-a_3\\
\end{array}\right| 1\right)\end{aligned}$$ with $ a_1=1,a_2=\nu+1,a_3=1,b_1=\nu+2,b_2=\nu+2$. Such a procedure furnishes us with an identity$$\begin{aligned}
{_3}F_2\left(\left.\begin{array}{c}1,1,\nu+1\\\nu +2 ,\nu +2\end{array}\right|1\right)=\frac{\Gamma (\nu +1) \Gamma (\nu +2)}{\Gamma (2 \nu +2)}{_3}F_2\left(\left.\begin{array}{c}\nu+1,\nu+1,\nu+1\\\nu +2 ,2\nu +2\end{array}\right|1\right).\end{aligned}$$For a suitably chosen non-vanishing $ \varepsilon$, we may use Thomae’s second fundamental relation (see [@Thomae1879 Eq. 45] or [@Slater Eq. 4.3.4]) to verify the following identity:$$\begin{aligned}
&{_3}F_2\left(\left.\begin{array}{c}\nu+1,\nu+1,\nu+1+\varepsilon\\\nu +2 ,2\nu +2\end{array}\right|1\right)\notag\\={}&\frac{\Gamma (-\nu ) \Gamma (2 \nu +3) \Gamma (\varepsilon )}{2 \Gamma (\varepsilon+\nu +1)}+\frac{(\nu +1)^2 \Gamma (-\nu -1) \Gamma (2 \nu +2)}{\varepsilon ^2 \Gamma (\varepsilon ) \Gamma (-\varepsilon+\nu +1)}{_3}F_2\left(\left.\begin{array}{c}\varepsilon ,\varepsilon -\nu ,\nu+1 +\varepsilon\ \\1+\varepsilon ,1+\varepsilon\end{array}\right|1\right)\notag\\={}&\left[\frac{\Gamma (-\nu ) \Gamma (2 \nu +3) \Gamma (\varepsilon )}{2 \Gamma (\varepsilon+\nu +1)}+\frac{(\nu +1)^2 \Gamma (-\nu -1) \Gamma (2 \nu +2)}{\varepsilon ^2 \Gamma (\varepsilon ) \Gamma (-\varepsilon+\nu +1)}\right]\notag\\{}&+\frac{(\nu +1)^2 \Gamma (-\nu -1) \Gamma (2 \nu +2)}{\varepsilon ^2 \Gamma (\varepsilon ) \Gamma (-\varepsilon+\nu +1)}\sum_{n=1}^\infty\frac{(\varepsilon)_{n}(\varepsilon-\nu)_{n} (\nu+1 +\varepsilon)_{n}}{(1+\varepsilon)_{n}(1+\varepsilon)_{n}n!}.\label{eq:Thomae2nd_epsilon}\end{aligned}$$In passage to the limit of $ \varepsilon\to0$, we may use Eq. \[eq:Thomae2nd\_epsilon\] to compute$$\begin{aligned}
&{_3}F_2\left(\left.\begin{array}{c}\nu+1,\nu+1,\nu+1\\\nu +2 ,2\nu +2\end{array}\right|1\right)\notag\\={}&-\frac{4^{\nu +1} (\nu +1) \Gamma (-\nu ) \Gamma (\nu +\frac{3}{2}) [\psi ^{(0)}(\nu +1)+\gamma_{0} ]}{\sqrt{\pi }}+\frac{(\nu +1)^2 \Gamma (-\nu -1) \Gamma (2 \nu +2)}{\Gamma (\nu +1)}\sum_{n=1}^\infty\frac{(-\nu)_{n} (\nu+1 )_{n}}{(n!)^2n}\notag\\={}&-\frac{4^{\nu +1} (\nu +1) \Gamma (-\nu ) \Gamma (\nu +\frac{3}{2}) [\psi ^{(0)}(\nu +1)+\gamma_{0} ]}{\sqrt{\pi }}+\frac{(\nu +1)^2 \Gamma (-\nu -1) \Gamma (2 \nu +2)}{\Gamma (\nu +1)}\int_{-1}^1\frac{P_\nu(\xi)-1}{1-\xi}\operatorname{d}\xi.
\label{eq:Thomae2nd_app}\end{aligned}$$ Here, to deduce the integral representation in the last step of Eq. \[eq:Thomae2nd\_app\], we have invoked the hypergeometric form of Legendre functions $ P_{\nu}(\xi)={_2}F_1\left( \left.\begin{smallmatrix}-\nu,\nu+1\\1\end{smallmatrix}\right| \smash{\frac{1-\xi}{2}}\right),$ $\xi\in\mathbb C\smallsetminus(-\infty,-1] $. It can be shown that the infinite series in Eq. \[eq:Thomae2nd\_app\] has termwise agreement with the expansion for $ -\nu(\nu+1){_4}F_3\left(\left.\begin{smallmatrix}1,1,1-\nu,\nu+2\\2 ,2,2\end{smallmatrix}\right|1\right)$, thus Eq. \[eq:QnuQnu\_xx\_int\_4F3\] is recovered.
We shall evaluate the remaining integral in Eq. \[eq:Thomae2nd\_app\] as$$\begin{aligned}
\int_{-1}^1\frac{P_\nu(\xi)-1}{1-\xi}\operatorname{d}\xi=\lim_{\mu\to0^{+}}\int_{-1}^{1-\mu}\frac{P_\nu^{\mu}(\xi)-1}{1-\xi}\operatorname{d}\xi,\end{aligned}$$while adopting Hobson’s definition for $ P^\mu_\nu(\xi),-1<\xi<1$ as follows [@Hobson1931 p. 227, Eq. 55]:$$\begin{aligned}
P^{\mu}_\nu(\xi)=\frac{1}{\Gamma(1-\mu)}\left( \frac{1+\xi}{1-\xi} \right)^{\mu/2}{_2}F_1\left( \left.\begin{array}{c}
-\nu,\nu+1\ \\
1-\mu\ \\
\end{array}\right| \frac{1-\xi}{2}\right),\quad -1<\xi<1.\end{aligned}$$ Bearing in mind the associated Legendre differential equation [@GradshteynRyzhik item 8.700.1] $$\begin{aligned}
\frac{\operatorname{d}}{\operatorname{d}\xi}\left[ (1-\xi^2) \frac{\operatorname{d}P^\mu_\nu(\xi)}{\operatorname{d}\xi}\right]+\left[ \nu(\nu+1)-\frac{\mu^2}{1-\xi^2} \right]P^\mu_\nu(\xi)=0,\label{eq:Pmunu_diff_eqn}\end{aligned}$$and the recursion relation $ (2\nu+1) \xi P^\mu_\nu(\xi)=(\nu-\mu+1)P^\mu_{\nu+1}(\xi)+(\nu+\mu)P^{\mu}_{\nu-1}(\xi)$ [@GradshteynRyzhik item 8.733.2], one may put down[$$\begin{aligned}
&\int_{-1}^{1-\mu}\frac{P_\nu^{\mu}(\xi)-1}{1-\xi}\operatorname{d}\xi=\log\frac{\mu}{2}+\int_{-1}^{1-\mu}\left[ P_\nu^{\mu}(\xi)+\frac{(\nu-\mu+1)P^\mu_{\nu+1}(\xi)+(\nu+\mu)P^{\mu}_{\nu-1}(\xi)}{2\nu+1} \right]\frac{\operatorname{d}\xi}{1-\xi^{2}}\notag\\={}& \log\frac{\mu}{2}+\left.\frac{(1-\xi^2)}{\mu^{2}} \frac{\operatorname{d}}{\operatorname{d}\xi}\left[ P_\nu^{\mu}(\xi)+\frac{(\nu-\mu+1)P^\mu_{\nu+1}(\xi)+(\nu+\mu)P^{\mu}_{\nu-1}(\xi)}{2\nu+1} \right]\right|_{x=1-\mu}\notag\\&+\frac{1}{\mu^{2}}\int_{-1}^{1-\mu}\left[ \nu(\nu+1)P_\nu^{\mu}(\xi)+\frac{(\nu+1)(\nu+2)(\nu-\mu+1)P^\mu_{\nu+1}(\xi)+(\nu-1)\nu(\nu+\mu)P^{\mu}_{\nu-1}(\xi)}{2\nu+1} \right]\operatorname{d}\xi\notag\\={}&\log\frac{\mu}{2}+\left.\frac{(1-\xi^2)}{\mu^{2}} \frac{\operatorname{d}}{\operatorname{d}\xi}\left[ (1+\xi)P_\nu^{\mu}(\xi) \right]\right|_{\xi=1-\mu}+\frac{2^{\mu }\sin(\nu\pi)}{\mu\sin\frac{\mu\pi}{2}} \left[\frac{\Gamma (\frac{1-\nu}{2}) \Gamma (\frac{\nu }{2}+1)}{\Gamma (\frac{-\mu -\nu +1}{2} ) \Gamma (\frac{-\mu +\nu +2}{2})}-\frac{\Gamma (-\frac{\nu }{2}) \Gamma (\frac{\nu +1}{2})}{\Gamma (-\frac{\mu+\nu }{2}) \Gamma (\frac{-\mu +\nu +1}{2} )}\right]\notag\\&-\frac{1}{\mu^{2}}\int_{1-\mu}^{1}\left[ \nu(\nu+1)P_\nu^{\mu}(\xi)+\frac{(\nu+1)(\nu+2)(\nu-\mu+1)P^\mu_{\nu+1}(\xi)+(\nu-1)\nu(\nu+\mu)P^{\mu}_{\nu-1}(\xi)}{2\nu+1} \right]\operatorname{d}\xi\notag\\={}&\frac{2}{\mu}-2\gamma_{0}+\frac{2^{\mu }\sin(\nu\pi)}{\mu\sin\frac{\mu\pi}{2}} \left[\frac{\Gamma (\frac{1-\nu}{2}) \Gamma (\frac{\nu }{2}+1)}{\Gamma (\frac{-\mu -\nu +1}{2} ) \Gamma (\frac{-\mu +\nu +2}{2})}-\frac{\Gamma (-\frac{\nu }{2}) \Gamma (\frac{\nu +1}{2})}{\Gamma (-\frac{\mu+\nu }{2}) \Gamma (\frac{-\mu +\nu +1}{2} )}\right]+O(\mu\log^2\mu).\label{eq:small_mu}\end{aligned}$$]{}Here, we have relied on the standard integral formula for $ \int_{-1}^1 P^\mu_\nu(\xi)\operatorname{d}\xi$ applicable to $ |\operatorname{Re}\mu|<2$ [@GradshteynRyzhik item 7.132.1] and have used the expansion$$\begin{aligned}
P_{\nu}^\mu(\xi)={}& \frac{2^{\mu /2}}{\Gamma (1-\mu )}(1-\xi)^{-\frac{\mu }{2}}-\frac{2^{\frac{\mu }{2}-2} [\mu (1-\mu)+2 \nu (\nu +1)]}{\Gamma (2-\mu )}(1-\xi)^{1-\frac{\mu }{2}}\notag\\&+\frac{2^{\frac{\mu }{2}-5} \left[4 \mu \nu (\nu +1)-\frac{4 (\nu -1) \nu (\nu +1) (\nu +2)}{\mu -2}-(\mu -2) (\mu -1) \mu \right]}{\Gamma (2-\mu )}(1-\xi)^{2-\frac{\mu }{2}}+O((1-\xi)^{3-\frac{\mu }{2}})\end{aligned}$$for the computation of the derivative at $ \xi=1-\mu$ as well as the integration over $ \xi\in[1-\mu,1]$. As we take the limit $ \mu\to 0^+$ in Eq. \[eq:small\_mu\], we can evaluate Eq. \[eq:Thomae2nd\_app\] in closed form, and hence Eq. \[eq:QnuQnu\_xx\_int\_closed\_form\].
3. Using the following Mellin inversion formula [@Marichev1983 §10.11, Eq. 57(1)]: $$\begin{aligned}
&
P_\nu^{-\mu}(\sqrt{1+t})Q_\nu^{\mu}(\sqrt{1+t})\notag\\={}&\frac{e^{i\mu\pi}}{2\sqrt{\pi}}\frac{1}{2\pi i}\int_{c-i\infty}^{c+i\infty}\frac{\Gamma (s)\Gamma(s+\mu)\ \Gamma (\frac{1}{2}-s) \Gamma (\nu +1-s) }{ \Gamma (\nu +1+s)\Gamma (1+\mu-s)}\frac{\operatorname{d}s}{t^s},&&\max\{0,-\mu\}<c<\min\{\nu+1,1/2\}\end{aligned}$$one can verify that $$\begin{aligned}
\int_1^\infty\frac{P_\nu(\xi)Q_{\nu}(\xi)}{\xi^2}\operatorname{d}\xi={}&\int_0^\infty\frac{P_\nu(\sqrt{1+t})Q_{\nu}(\sqrt{1+t})}{2(1+t)^{3/2}}\operatorname{d}t=\frac{1}{2\pi i}\int_{c-i\infty}^{c+i\infty}\frac{[\Gamma (s)]^{2}\Gamma(s+\frac{1}{2}) \Gamma (\frac{1}{2}-s) \Gamma (\nu +1-s) }{ \Gamma (\nu +1+s)}\frac{\operatorname{d}s}{2\pi}.\end{aligned}$$Again, one can treat such a contour integral with Barnes’s second lemma [@Slater Eq. 4.2.2.1], which results in the first equality in Eq. \[eq:PnuQnu\_xx\_int\_closed\_form\].
Instead of directly manipulating the hypergeometric series in Eq. \[eq:PnuQnu\_xx\_int\_closed\_form\], we shall evaluate it by a comparison to Eq. \[eq:PnuPnu\_xx\_int\_closed\_form\]. We may employ the residue theorem to establish a vanishing integral:$$\begin{aligned}
0={}&\int_{-\infty+i0^+}^{+\infty+i0^+}\frac{[P_\nu(\xi)]^2-[P_\nu(0)]^2}{\xi^2}\operatorname{d}\xi\notag\\={}&\int_{-\infty+i0^+}^{-1+i0^+}\frac{[P_\nu(\xi)]^2-[P_\nu(0)]^2}{\xi^2}\operatorname{d}\xi+\mathcal P\int_{-1}^{1}\frac{[P_\nu(\xi)]^2-[P_\nu(0)]^2}{\xi^2}\operatorname{d}\xi\notag\\{}&-2i\sin(\nu\pi)+\int_1^\infty\frac{[P_\nu(\xi)]^2-[P_\nu(0)]^2}{\xi^2}\operatorname{d}\xi.\label{eq:PnuPnu_xx_int_3parts}\end{aligned}$$ For $ \xi<-1$, we use the identity$$\begin{aligned}
P_\nu(\xi\pm i0^{+})=[\cos ( \nu\pi )\pm i \sin (\nu\pi )] P_{\nu }(-\xi)-\frac{2 \sin ( \nu\pi ) Q_{\nu }(-\xi)}{\pi }\label{eq:Pnu_Qnu_rln}\end{aligned}$$ to rewrite the first integral in the last line of Eq. \[eq:PnuPnu\_xx\_int\_3parts\] as $$\begin{aligned}
\int_{-\infty+i0^+}^{-1+i0^+}\frac{[P_\nu(\xi)]^2-[P_\nu(0)]^2}{\xi^2}\operatorname{d}\xi=\int_{1}^\infty\left\{ [\cos ( \nu\pi )+ i \sin (\nu\pi )]P_{\nu }(\xi)-\frac{2 \sin (\nu\pi) Q_{\nu }(\xi)}{\pi } \right\}^2\frac{\operatorname{d}\xi}{\xi^2}-[P_\nu(0)]^2.\end{aligned}$$As we read off the imaginary part of Eq. \[eq:PnuPnu\_xx\_int\_3parts\], we obtain $$\begin{aligned}
2\sin(\nu\pi)\cos(\nu\pi)\int_1^\infty\frac{[P_\nu(\xi)]^2}{\xi^2}\operatorname{d}\xi-\frac{4\sin^2(\nu\pi)}{\pi}\int_1^\infty\frac{P_\nu(\xi)Q_{\nu}(\xi)}{\xi^2}\operatorname{d}\xi-2\sin(\nu\pi)=0,\label{eq:Pnu_Qnu_x_2_0sum}\end{aligned}$$ which allows us to deduce the final evaluation in Eq. \[eq:PnuQnu\_xx\_int\_closed\_form\] from Eq. \[eq:PnuPnu\_xx\_int\_closed\_form\].
Using the residue theorem as well as the relation between $ P_\nu$ and $Q_\nu$ (Eq. \[eq:Pnu\_Qnu\_rln\]), we can verify$$\begin{aligned}
\operatorname{Re}\int_{-1}^1\frac{P_\nu(\xi)P_\nu(-\xi)}{\xi^2}\operatorname{d}\xi={}&-2\cos(\nu\pi)\int_1^\infty\frac{[P_\nu(\xi)]^2}{\xi^2}\operatorname{d}\xi+\frac{4\sin(\nu\pi)}{\pi}\int_{1}^\infty\frac{P_\nu(\xi)Q_\nu(\xi)}{\xi^2}\operatorname{d}\xi=-2\label{eq:minus2_id}\end{aligned}$$for $ -1<\nu<0$, by virtue of the evaluations given in Eqs. \[eq:PnuPnu\_xx\_int\_closed\_form\] and \[eq:PnuQnu\_xx\_int\_closed\_form\]. It is not hard to show that Eq. \[eq:minus2\_id\] entails Eq. \[eq:minus2\_id’\].
As we take the real part of Eq. \[eq:PnuPnu\_xx\_int\_3parts\], we arrive at the first equality in Eq. \[eq:PnuPnu\_xx\_int\_unit\_interval\_polygamma\]. The remaining steps of Eq. \[eq:PnuPnu\_xx\_int\_unit\_interval\_polygamma\] are straightforward computations.
4. Adhering to the convention that $ \xi^s=e^{s\log \xi}$ with $ \log \xi=\log|\xi|+i\arg \xi,-\pi<\arg \xi\leq \pi$, we may extract the imaginary part of a vanishing identity $$\begin{aligned}
0=\int_{-\infty+i0^+}^{+\infty+i0^+}\frac{[P_\nu(\xi)]^2-[P_\nu(-\xi)]^2-2i\sin(\nu\pi)P_\nu(\xi)P_\nu(-\xi)}{(-\xi)^s}\operatorname{d}\xi,\quad 1<s<2,-1<\nu<0
$$as follows:$$\begin{aligned}
0={}&\frac{4}{\pi}\sin(\nu\pi)\sin(s\pi)\left[\cos(\nu\pi)\int_1^\infty\frac{P_\nu(\xi)Q_{\nu}(\xi)}{\xi^s}\operatorname{d}\xi-\frac{\sin(\nu\pi)}{\pi}\int_1^\infty\frac{[Q_\nu(\xi)]^2}{\xi^s}\operatorname{d}\xi\right]\notag\\&+\sin(s\pi)\int_0^1\left\{\frac{[P_\nu(\xi)]^2-[P_\nu(-\xi)]^2}{\xi^{s}}-\frac{4\sin(\nu\pi)}{\pi \xi^{s-1}}\right\}\operatorname{d}\xi+\frac{4\sin(s\pi)\sin(\nu\pi)}{\pi(2-s)}\notag\\&-2\sin(\nu\pi)[1+\cos(s\pi)]\left\{\int_{0}^1\frac{P_{\nu}(\xi)P_{\nu}(-\xi)-[P_\nu(0)]^{2}}{\xi^{s}}\operatorname{d}\xi-\frac{[P_\nu(0)]^{2}}{s-1}\right\}.\label{eq:Pnu_Qnu_int_x_s_0sum}\end{aligned}$$ As we differentiate Eq. \[eq:Pnu\_Qnu\_int\_x\_s\_0sum\] in $s$ and send it to the limit of $ s\to 2$, we obtain the first equality in Eq. \[eq:Pnu\_sqr\_diff\_int\_log\_over\_sqr\].
We may use the Hobson coupling formula for $ P_\nu(\xi)P_{\nu}(-\xi)$ [@Zhou2013Pnu Eq. 11] and the Mellin inversion formula for $ P_\nu(\xi)$ [@Marichev1983 §10.11, Eq. 3(1)] to deduce$$\begin{aligned}
&
P_{\nu}(\xi)P_{\nu}(-\xi)=\int_0^{2\pi}P_\nu(-\xi^2+(1-\xi^2)\cos\phi)\frac{\operatorname{d}\phi}{2\pi}\notag\\={}&\frac{\sin^2(\nu\pi)}{\pi^2}\int_0^{2\pi}\left\{\frac{1}{2\pi i}\int_{c-i\infty}^{c+i\infty}\frac{[\Gamma(s)]^{2}\Gamma(\nu+1-s)\Gamma(-\nu-s)2^s\operatorname{d}s}{(1-\xi^2)^s(1+\cos\phi)^{s}}\right\}\frac{\operatorname{d}\phi}{2\pi}\notag\\={}&\frac{\sin^2(\nu\pi)}{\pi^2}\frac{1}{2\pi i}\int_{c-i\infty}^{c+i\infty}\frac{[\Gamma(s)]^{2}\Gamma(\nu+1-s)\Gamma(-\nu-s)\Gamma(\frac{1}{2}-s)}{\sqrt{\pi}\Gamma(1-s)}\frac{\operatorname{d}s}{(1-\xi^2)^s},\quad 0<c<\min\{\nu+1,-\nu\},-1<\xi<1.\end{aligned}$$ Consequently, we have $$\begin{aligned}
\int _{0}^{1}\xi^{t}P_{\nu}(\xi)P_{\nu}(-\xi)\operatorname{d}\xi={}&\frac{\sin^2(\nu\pi)}{\pi^2}\frac{\Gamma(\frac{1+t}{2})}{2\pi i}\int_{-c-i\infty}^{-c+i\infty}\frac{[\Gamma(-s)]^{2}\Gamma(\nu+1+s)\Gamma(-\nu+s)\Gamma(\frac{1}{2}+s)}{2\sqrt{\pi}\Gamma(\frac{3+t}{2}+s)}\operatorname{d}s\notag\\={}&\frac{1}{1+t}{_3}F_2\left(\left.\begin{array}{c}\nu+1,-\nu,\frac{2+t}{2}\\1 ,\frac{3+t}{2}\end{array}\right|1\right),\quad t>-1,\label{eq:teeth_moment_3F2_original}\end{aligned}$$according to Barnes’s second lemma [@Slater Eq. 4.2.2.1]. One can then show that the identity$$\begin{aligned}
\int _{0}^{1}\xi^{t}\{P_{\nu}(\xi)P_{\nu}(-\xi)-[P_{\nu}(0)]^{2}\}\operatorname{d}\xi=\frac{1}{1+t}\left\{{_3}F_2\left(\left.\begin{array}{c}\nu+1,-\nu,\frac{2+t}{2}\\1 ,\frac{3+t}{2}\end{array}\right|1\right)-[P_{\nu}(0)]^{2}\right\}\label{eq:teeth_moment_3F2}\end{aligned}$$admits an analytic continuation to $\operatorname{Re}t>-3$, and incorporates Eq. \[eq:minus2\_id’\] as a special case. Exploiting the fact that $$\begin{aligned}
\left.\frac{\partial}{\partial t}\right|_{t=-2}\frac{\left(\frac{2+t}{2}\right)_n}{\left(\frac{3+t}{2}\right)_n}=\frac{\sqrt{\pi}\Gamma(n)}{2\Gamma(n+\frac{1}{2})},\quad n\in\mathbb Z_{>0},\end{aligned}$$we may differentiate Eq. \[eq:teeth\_moment\_3F2\] in $t$ at $ t=-2$ and obtain$$\begin{aligned}
&
\int_{0}^1\frac{P_{\nu}(\xi)P_{\nu}(-\xi)-[P_\nu(0)]^{2}}{\xi^{2}}\log \xi\operatorname{d}\xi=\int_{0}^1\frac{P_{\nu}(\xi)P_{\nu}(-\xi)-[P_\nu(0)]^{2}}{\xi^{2}}\operatorname{d}\xi-\frac{1}{2}\sum_{n=1}^\infty\frac{(\nu+1)_{n}(-\nu)_n}{n!(\frac{1}{2})_{n}}\frac{1}{n}\notag\\={}&[P_\nu(0)]^{2}-1-\frac{1}{2}\int_{0}^1\left[ {_2}F_1\left(\left.\begin{array}{c}\nu+1,-\nu\\\frac{1}{2}\end{array}\right|\xi\right)-1 \right]\frac{\operatorname{d}\xi}{\xi}=[P_\nu(0)]^{2}-1+\frac{1}{2}\int_0^{\pi/2}\frac{\cos\theta-\cos((2\nu+1)\theta)}{\sin\theta}\operatorname{d}\theta\notag\\={}&[P_\nu(0)]^{2}-1+\frac{1-\cos ( \nu\pi )}{\nu }+\psi ^{(0)}(\nu )+\gamma_{0} +\log 2\notag\\{}&+\frac{\pi\cos ( \nu\pi ) \tan\frac{ \nu\pi }{2}}{4} +\frac{\cos ( \nu\pi )\left[\psi ^{(0)}\left(\frac{\nu +1}{2}\right)+\psi ^{(0)}\left(\frac{1-\nu}{2}\right)-2 \psi ^{(0)}\left(\frac{\nu }{2}\right)\right]}{4} ,\label{eq:Pnu_refl_over_sqr_log_int}\end{aligned}$$ as one can compute the integral over $ \theta$ following the procedures in Eq. \[eq:angle\_int\_psi\]. Thus, substituting Eq. \[eq:Pnu\_refl\_over\_sqr\_log\_int\] back into the first equality of Eq. \[eq:Pnu\_sqr\_diff\_int\_log\_over\_sqr\], we obtain the evaluation declared in the last step of Eq. \[eq:Pnu\_sqr\_diff\_int\_log\_over\_sqr\].
We can modify Eq. \[eq:Pnu\_Qnu\_int\_x\_s\_0sum\] into a form that analytically continues to a neighborhood of $ s=0$:$$\begin{aligned}
0={}&\frac{4}{\pi}\sin(\nu\pi)\sin(s\pi)\left\{\cos(\nu\pi)\int_1^\infty\left[\frac{P_\nu(\xi)Q_{\nu}(\xi)}{\xi^s}-\frac{1}{(2\nu+1)\xi^{s+1}}\right]\operatorname{d}\xi-\frac{\sin(\nu\pi)}{\pi}\int_1^\infty\frac{[Q_\nu(\xi)]^2}{\xi^s}\operatorname{d}\xi\right\}\notag\\&+\frac{4\sin(\nu\pi)\cos(\nu\pi)}{(2\nu+1)\pi}\frac{\sin(s\pi)}{s}+\sin(s\pi)\int_0^1\frac{[P_\nu(\xi)]^2-[P_\nu(-\xi)]^2}{\xi^{s}}\operatorname{d}\xi\notag\\{}&-2\sin(\nu\pi)[1+\cos(s\pi)]\int_{0}^1\frac{P_{\nu}(\xi)P_{\nu}(-\xi)}{\xi^{s}}\operatorname{d}\xi.\label{eq:Pnu_Qnu_int_x_s_0sum'}\tag{\ref{eq:Pnu_Qnu_int_x_s_0sum}$'$}\end{aligned}$$We can then differentiate Eq. \[eq:Pnu\_Qnu\_int\_x\_s\_0sum’\] in $s$ at $ s=0$ to verify the first equality of Eq. \[eq:PQ-1\_over\_x\_int\]. For $ \nu>-1/2$, the following integral formula [@GradshteynRyzhik item 7.114.3]$$\begin{aligned}
\int_1^\infty[Q_{\nu}(\xi)]^2\operatorname{d}\xi=\frac{\psi^{(1)}(\nu+1)}{2\nu+1}\label{eq:Qnu_sqr_1toinf_int}\end{aligned}$$is a standard result, and can be derived from simple applications of Legendre differential equations. The evaluation $$\begin{aligned}
\int_0^1[P_\nu(\xi)]^2\operatorname{d}\xi-\int_{0}^{1}[P_\nu(-\xi)]^2\operatorname{d}\xi={}&2\int_0^1[P_\nu(\xi)]^2\operatorname{d}\xi-\int_{-1}^1[P_\nu(\xi)]^2\operatorname{d}\xi\notag\\={}&\frac{2\sin(\nu\pi)}{(2\nu+1)\pi}\left[ \psi^{(0)}\left( \frac{\nu+2}{2} \right) - \psi^{(0)}\left( \frac{\nu+1}{2} \right) +\frac{2\sin(\nu\pi)}{\pi}\psi^{(1)}(\nu+1)\right].\label{eq:Pnu_sqr_antisymm_0to1_int}\end{aligned}$$follows immediately from Eqs. \[eq:Pnu\_sqr\_0to1\] and \[eq:P\_nu\_sqr\]. By Kummer’s $ _3F_2$ transformation (Eq. \[eq:Kummer3F2\]) with $ a_1=-\nu,a_2=\nu+1,a_3=(2+t)/2,b_1=1,b_2=(3+t)/2$, we may turn Eq. \[eq:teeth\_moment\_3F2\_original\] into$$\begin{aligned}
\int _{0}^{1}\xi^{t}P_{\nu}(\xi)P_{\nu}(-\xi)\operatorname{d}\xi={}&\frac{1}{1+t}{_3}F_2\left(\left.\begin{array}{c}-\nu,\nu+1,\frac{2+t}{2}\\1 ,\frac{3+t}{2}\end{array}\right|1\right)\notag\\={}&\frac{\sqrt{\pi } \Gamma (\frac{1+t}{2})}{2 \Gamma ( \frac{1}{2}-\nu) \Gamma (\frac{3+t}{2} +\nu)}{_3}F_2\left(\left.\begin{array}{c}-\nu,-\nu,-\frac{t}{2}\\1 ,\frac{1}{2}-\nu\end{array}\right|1\right),\quad t>-1,\tag{\ref{eq:teeth_moment_3F2_original}$'$}\end{aligned}$$whose derivative at $ t=0$ yields the evaluation$$\begin{aligned}
&
\int _{0}^{1}P_{\nu}(\xi)P_{\nu}(-\xi)\log \xi\operatorname{d}\xi=-\frac{\cos ( \nu\pi ) \left[\psi ^{(0)}\left(\frac{3}{2}+\nu \right)+\gamma _{0}+2\log 2\right]}{2(2\nu+1)}-\frac{\cos(\nu\pi)}{2(2\nu+1)}\sum_{n=1}^\infty\frac{(-\nu)_n(-\nu)_n}{n!(\frac{1}{2}-\nu)_{n}}\frac{1}{n}\notag\\={}&-\frac{\cos ( \nu\pi ) }{2(2\nu+1)}\left[\psi ^{(0)}\left(\frac{3}{2}+\nu \right)+\gamma _{0}+2\log 2+\frac{2\nu^{2}}{1-2\nu}{_4}F_3\left(\left.\begin{array}{c}1,1,1-\nu ,1-\nu\ \\[4pt]2,2,\frac{3}{2}-\nu\end{array}\right|1\right)\right]\notag\\={}&\frac{\cos (\nu\pi )}{\nu (2 \nu +1)}-\frac{\cos ( \nu\pi ) \left[\psi ^{(0)}\left(\frac{1}{2}-\nu \right)-2 \psi ^{(0)}(\nu )+\psi ^{(0)}\left(\nu +\frac{3}{2}\right)+2 \log 2\right]}{2 (2 \nu +1)}+\frac{\psi ^{(0)}\left(\frac{\nu+1 }{2}{}\right)-\psi ^{(0)}\left(\frac{\nu+2 }{2}\right)-\pi \sin ( \nu\pi )}{2 (2 \nu +1)}.\label{eq:PnuPnu_logx_int}\end{aligned}$$ Here, in Eq. \[eq:PnuPnu\_logx\_int\], the infinite series in $n$ matches the expansion of a special $_4F_3 $, which in turn, can be computed from a comparison of Eqs. \[eq:PnuPnu\_xx\_int\_closed\_form\] and \[eq:PnuPnu\_xx\_int\_4F3\]. Combining the results in Eqs. \[eq:Qnu\_sqr\_1toinf\_int\]–\[eq:PnuPnu\_logx\_int\], one can confirm the last step of Eq. \[eq:PQ-1\_over\_x\_int\].
Integral Representations of Weight-4 Renormalized Automorphic Green’s Functions\[subsec:int\_repn\_G2\_GZ\_rn\]
---------------------------------------------------------------------------------------------------------------
The automorphic Green’s function $ G_s^{\mathfrak H/\overline {\varGamma}}(z,z')$, $\operatorname{Re}s>1$ tends to infinity, as $\hat\gamma z'\to z $ for a certain $ \hat\gamma\in\varGamma$. It is possible to appropriately subtract the logarithmic divergence of the automorphic Green’s function on the diagonal of $(\varGamma\backslash\mathfrak H)\times(\varGamma\backslash\mathfrak H) $, and define the remaining finite part as an arithmetically meaningful “self-energy” $ G_s^{\mathfrak H/\overline{\varGamma}}(z)$. When $ \varGamma=\varGamma_0(N)$ is the Hecke congruence group, such a subtraction scheme is the Gross–Zagier renormalization [@GrossZagier1985; @GrossZagierI]. In [@GrossZagierI Chap. II, §5], the “automorphic self-energy” (or “renormalized automorphic Green’s function”) $ G_s^{\mathfrak H/\overline{\varGamma}_0(N)}(z)$ was prescribed as follows: $$\begin{aligned}
G_s^{\mathfrak H/\overline{\varGamma}_0(N)}(z)={}&-2\sum_{\hat \gamma\in\overline {\varGamma}_0(N),\hat \gamma z\neq z}Q_{s-1}
\left( 1+\frac{\vert z -\hat \gamma z\vert ^{2}}{2\operatorname{Im}z\operatorname{Im}(\hat\gamma z)} \right)\notag\\{}&-2\left\{ \log\left\vert 4\pi y[\eta(z)]^4 \right\vert+\psi^{(0)}(1)-\psi^{(0)}(s) \right\}m^{\overline {\varGamma}_0(N)}_{z},\label{eq:GZ_rn_defn}\end{aligned}$$where $ \eta(\cdot)$ is the Dedekind eta function, $ \psi^{(0)}(\cdot)$ is the digamma function, and $ m^{\overline {\varGamma}}_{z}:=\#\{\hat \gamma\in\overline{\varGamma}|\hat \gamma z=z\}$ is the number of inequivalent transformations in the projective congruence subgroup $ \overline{\varGamma}$ that fix the point $ z$. The Gross–Zagier algebraicity conjecture for $ G_s^{\mathfrak H/\overline{\varGamma}_0(N)}(z)$ can be stated as [@GrossZagierI p. 317, Conjecture 4.4]$$\begin{aligned}
\exp \left[ (\operatorname{Im}z)^{k-2}G^{\mathfrak H/\overline {\varGamma}_0(N)}_{k/2}(z)\right]\in\overline{\mathbb Q},\quad \text{if }\dim\mathcal S_{k}(\varGamma_0(N))=0\text{ and }[\mathbb Q(z):\mathbb Q]=2.\label{eq:GZ_conj_rn}\end{aligned}$$
As we have $ m_z^{\overline{\varGamma}}=1$ (*i.e.* $z$ is a non-elliptic point) for almost every $ z\in\mathfrak H$, we may reformulate the Gross–Zagier renormalization (Eq. \[eq:GZ\_rn\_defn\]) for weight-4 automorphic Green’s functions as follows:$$\begin{aligned}
G_2^{\mathfrak H/\overline{\varGamma}_0(N)}(z)={}&\lim_{z'\to z}\left[G_2^{\mathfrak H/\overline{\varGamma}_0(N)}(z,z')+2Q_{1}
\left( 1+\frac{\vert z -z'\vert ^{2}}{2yy'} \right)\right]-2\left\{ \log\left\vert 4\pi y[\eta(z)]^4 \right|-1 \right\}\notag\\={}&\lim_{z'\to z}\left[G_2^{\mathfrak H/\overline{\varGamma}_0(N)}(z,z')-2\log|z-z'|\right]-2\log\left\vert 2\pi[\eta(z)]^4 \right\vert,\quad \text{a.e. }z\in\mathfrak H.\label{eq:G2_rn_non_ell_defn}\end{aligned}$$
In the next proposition, we derive integral representations of weight-4 Gross–Zagier renormalized Green’s functions (Eq. \[eq:G2\_rn\_non\_ell\_defn\]) for non-elliptic points.
\[prop:G2\_rn\_non\_ell\_pts\]With the notations $ \rho_{2,-1/6}$, $ \varrho_{2,-1/4}$, $ \varrho_{2,-1/3}$ and $ \varrho_{2,-1/2}$ defined in Proposition \[prop:G2\_HeckeN\_Pnu\], we have [$$\begin{aligned}
&
G_2^{\mathfrak H/PSL(2,\mathbb Z)}(z)\notag\\={}&{}\log\frac{2^{10}3^6}{|j(z)|^{5/3}\left\vert 1-\sqrt{\frac{j(z)-1728}{j(z)}} \right\vert^2}+\operatorname{Re}\int_{\sqrt{(j(z)-1728)/{j(z)}}}^1\left\{ \vphantom{\left[\frac{iP_{-1/6}(-\xi)}{P_{-1/6}(\xi)}-\vphantom{\overline{\frac{\sqrt{j}}{}}}\frac{iP_{-1/6}(-\sqrt{(j(z)-1728)/j(z)})}{P_{-1/6}(\sqrt{(j(z)-1728)/j(z)})}\right]} \right.\frac{2}{\smash[b]{\xi-\sqrt{\frac{j(z)-1728}{j(z)}}}}+\frac{\pi[P_{-1/6}(\xi)]^2\rho_{2,-1/6}(\xi|z)}{\smash[b]\operatorname{Im}\frac{iP_{-1/6}(-\sqrt{(j(z)-1728)/j(z)})}{P_{-1/6}(\sqrt{(j(z)-1728)/j(z)})}}\times\notag\\&\times\left.\left[\frac{iP_{-1/6}(-\xi)}{P_{-1/6}(\xi)}-\vphantom{\overline{\frac{\sqrt{j}}{}}}\frac{iP_{-1/6}(-\sqrt{(j(z)-1728)/j(z)})}{P_{-1/6}(\sqrt{(j(z)-1728)/j(z)})}\right]\left[\frac{iP_{-1/6}(-\xi)}{P_{-1/6}(\xi)}-\overline{\left(\frac{iP_{-1/6}(-\sqrt{(j(z)-1728)/j(z)})}{P_{-1/6}(\sqrt{(j(z)-1728)/j(z)})}\right)}\right]\right\}\operatorname{d}\xi\notag\\&+\frac{\pi}{\operatorname{Im}\frac{iP_{-1/6}(-\sqrt{(j(z)-1728)/j(z)})}{P_{-1/6}(\sqrt{(j(z)-1728)/j(z)})}
}\operatorname{Re}\int_{-1}^1[P_{-1/6}(-\xi)]^2\rho_{2,-1/6}(\xi|z)\operatorname{d}\xi+2,\quad \emph{a.e. }z\in\mathfrak H;\label{eq:G2_PSL2Z_z_rn}\\&G_2^{\mathfrak H/\overline{\varGamma}_0(2)}(z)\notag\\={}&\frac13\log\frac{2^{4}}{|1-2\lambda(2z+1)|^{6}|1-\lambda(2z+1)||\lambda(2z+1)|}+\operatorname{Re}\int_{1-2\alpha_{2}(z)}^1\left\{ \vphantom{\left[\frac{iP_{-1/6}(-\xi)}{P_{-1/6}(\xi)}-\vphantom{\overline{\frac{iP_{-1/3}(2\alpha(z)-1)}{P_{-1/3}(1-2\alpha(z))}}}\right]} \right.\frac{2}{\smash[b]{\xi-1+2\alpha_2(z)}}+\frac{\pi[P_{-1/4}(\xi)]^2\varrho_{2,-1/4}(\xi|z)}{\sqrt{2}\operatorname{Im}\frac{iP_{-1/4}(2\alpha_2(z)-1)}{P_{-1/4}(1-2\alpha_2(z))}}\times\notag\\&\times\left.\left[\frac{iP_{-1/4}(-\xi)}{P_{-1/4}(\xi)}-\vphantom{\overline{\frac{1}{}}}\frac{iP_{-1/4}(2\alpha_2(z)-1)}{P_{-1/4}(1-2\alpha_2(z))}\right]\left[\frac{iP_{-1/4}(-\xi)}{P_{-1/4}(\xi)}-\overline{\left(\frac{iP_{-1/4}(2\alpha_2(z)-1)}{P_{-1/4}(1-2\alpha_2(z))}\right)}\right]\right\}\operatorname{d}\xi\notag\\&+\frac{\pi}{\sqrt{2}\operatorname{Im}\frac{iP_{-1/4}(2\alpha_2(z)-1)}{P_{-1/4}(1-2\alpha_2(z))}
}\operatorname{Re}\int_{-1}^1[P_{-1/4}(-\xi)]^2\varrho_{2,-1/4}(\xi|z)\operatorname{d}\xi+2,\quad \emph{a.e. }z\in\mathfrak H;\label{eq:G2_Hecke2_z_rn}\\&G_2^{\mathfrak H/\overline{\varGamma}_0(3)}(z)\notag\\={}&\log\frac{3|1-\alpha_{3}(z)|}{|\alpha_{3}(z)|^{1/3}}+\operatorname{Re}\int_{1-2\alpha_{3}(z)}^1\left\{ \vphantom{\left[\frac{iP_{-1/6}(-\xi)}{P_{-1/6}(\xi)}-\vphantom{\overline{\frac{iP_{-1/3}(2\alpha(z)-1)}{P_{-1/3}(1-2\alpha(z))}}}\right]} \right.\frac{2}{\smash[b]{\xi-1+2\alpha_{3}(z)}}+\frac{\pi[P_{-1/3}(\xi)]^2\varrho_{2,-1/3}(\xi|z)}{\sqrt{3}\operatorname{Im}\frac{iP_{-1/3}(2\alpha_{3}(z)-1)}{P_{-1/3}(1-2\alpha_{3}(z))}}\times\notag\\&\times\left.\left[\frac{iP_{-1/3}(-\xi)}{P_{-1/3}(\xi)}-\vphantom{\overline{\frac{1}{}}}\frac{iP_{-1/3}(2\alpha_{3}(z)-1)}{P_{-1/3}(1-2\alpha_{3}(z))}\right]\left[\frac{iP_{-1/3}(-\xi)}{P_{-1/3}(\xi)}-\overline{\left(\frac{iP_{-1/3}(2\alpha_{3}(z)-1)}{P_{-1/3}(1-2\alpha_{3}(z))}\right)}\right]\right\}\operatorname{d}\xi\notag\\&+\frac{\pi}{\sqrt{3}\operatorname{Im}\frac{iP_{-1/3}(2\alpha_{3}(z)-1)}{P_{-1/3}(1-2\alpha_{3}(z))}
}\operatorname{Re}\int_{-1}^1[P_{-1/3}(-\xi)]^2\varrho_{2,-1/3}(\xi|z)\operatorname{d}\xi+2,\quad \emph{a.e. }z\in\mathfrak H;\label{eq:G2_Hecke3_z_rn}\\&G_2^{\mathfrak H/\overline{\varGamma}_0(4)}(z)\notag\\={}&\frac13\log\frac{2^{4}|1-\lambda(2z)|^2}{|\lambda(2z)|}+\operatorname{Re}\int_{1-2\lambda(2z)}^1\left\{ \vphantom{\left[\frac{iP_{-1/6}(-\xi)}{P_{-1/6}(\xi)}-\vphantom{\overline{\frac{iP_{-1/3}(2\alpha(z)-1)}{P_{-1/3}(1-2\alpha(z))}}}\right]} \right.\frac{2}{\smash[b]{\xi-1+2\lambda(2z)}}+\frac{\pi[P_{-1/2}(\xi)]^2\varrho_{2,-1/2}(\xi|z)}{2\operatorname{Im}\frac{iP_{-1/2}(2\lambda(2z)-1)}{P_{-1/2}(1-2\lambda(2z))}}\times\notag\\&\times\left.\left[\frac{iP_{-1/2}(-\xi)}{P_{-1/2}(\xi)}-\vphantom{\overline{\frac{1}{}}}\frac{iP_{-1/2}(2\lambda(2z)-1)}{P_{-1/2}(1-2\lambda(2z))}\right]\left[\frac{iP_{-1/2}(-\xi)}{P_{-1/2}(\xi)}-\overline{\left(\frac{iP_{-1/2}(2\lambda(2z)-1)}{P_{-1/2}(1-2\lambda(2z))}\right)}\right]\right\}\operatorname{d}\xi\notag\\&+\frac{\pi}{2\operatorname{Im}\frac{iP_{-1/2}(2\lambda(2z)-1)}{P_{-1/2}(1-2\lambda(2z))}
}\operatorname{Re}\int_{-1}^1[P_{-1/2}(-\xi)]^2\varrho_{2,-1/2}(\xi|z)\operatorname{d}\xi+2,\quad \emph{a.e. }z\in\mathfrak H.\label{eq:G2_Hecke4_z_rn}\end{aligned}$$]{}
We first give a detailed derivation for Eq. \[eq:G2\_PSL2Z\_z\_rn\]. As $ \xi\to\sqrt{(j(z')-1728)/{j(z')}}$ and $ z'\to z$, one has[$$\begin{aligned}
&\frac{\pi[P_{-1/6}(\xi)]^2\rho_{2,-1/6}(\xi|z)}{\operatorname{Im}\frac{iP_{-1/6}(-\sqrt{(j(z')-1728)/j(z')})}{P_{-1/6}(\sqrt{(j(z')-1728)/j(z')})}}\left[\frac{iP_{-1/6}(-\xi)}{P_{-1/6}(\xi)}-
\vphantom{\overline{\frac{\sqrt{j}}{}}}\frac{iP_{-1/6}(-\sqrt{(j(z')-1728)/j(z')})}{P_{-1/6}(\sqrt{(j(z')-1728)/j(z')})}\right]\times\notag\\&\times
\left[\frac{iP_{-1/6}(-\xi)}{P_{-1/6}(\xi)}-\overline{\left(\frac{iP_{-1/6}(-\sqrt{(j(z')-1728)/j(z')})}{P_{-1/6}(\sqrt{(j(z')-1728)/j(z')})}\right)}
\right]\notag\\\sim&-2\pi[P_{-1/6}(\xi)]^2\rho_{2,-1/6}(\xi|z)\left( \xi-\sqrt{\frac{j(z)-1728}{j(z)}}+ \sqrt{\frac{j(z)-1728}{j(z)}}-\sqrt{\frac{j(z')-1728}{j(z')}}\right)\frac{\operatorname{d}}{\operatorname{d}\xi}\frac{P_{-1/6}(-\xi)}{P_{-1/6}(\xi)}\notag\\\sim&-\frac{2}{\xi-\sqrt{\frac{j(z)-1728}{j(z)}}}-\left(\sqrt{\frac{j(z)-1728}{j(z)}}-\sqrt{\frac{j(z')-1728}{j(z')}}\right)\frac{2\rho_{2,-1/6}(\xi|z)}{1-\xi^{2}}+O(1),\label{eq:P_sixth_rn_asym1}\end{aligned}$$]{}and $$\begin{aligned}
&-\lim_{z'\to z}\left(\sqrt{\frac{j(z)-1728}{j(z)}}-\sqrt{\frac{j(z')-1728}{j(z')}}\right)\int_{\sqrt{(j(z')-1728)/{j(z')}}}^1\frac{2\rho_{2,-1/6}(\xi|z)}{1-\xi^{2}}\operatorname{d}\xi\notag\\={}&-\lim_{z'\to z}\left(\sqrt{\frac{j(z)-1728}{j(z)}}-\sqrt{\frac{j(z')-1728}{j(z')}}\right)\int_{\sqrt{(j(z')-1728)/{j(z')}}}^1\frac{2\operatorname{d}\xi}{\left[ \xi-\sqrt{\frac{j(z)-1728}{j(z)}} \right]^2}=2.\label{eq:P_sixth_rn_asym2}\end{aligned}$$With the asymptotic behavior displayed above in Eqs. \[eq:P\_sixth\_rn\_asym1\] and \[eq:P\_sixth\_rn\_asym2\], we may insert the integral representation for $G_2^{\mathfrak H/PSL(2,\mathbb Z)}(z,z') $ (Eq. \[eq:G2\_z\_z’\_arb\]) into the following expression$$\begin{aligned}
&
G_2^{\mathfrak H/PSL(2,\mathbb Z)}(z,z')-2\log\left\vert \frac{\sqrt{\frac{j(z')-1728}{j(z')}}-\sqrt{\frac{j(z)-1728}{j(z)}}}{1-\sqrt{\frac{j(z)-1728}{j(z)}}} \right\vert\notag\\={}&G_2^{\mathfrak H/PSL(2,\mathbb Z)}(z,z')+\operatorname{Re}\int_{\sqrt{(j(z')-1728)/{j(z')}}}^1\frac{2\operatorname{d}\xi}{\xi-\sqrt{\frac{j(z)-1728}{j(z)}}}
,\end{aligned}$$combine the integrals with respect to $\xi\in(\sqrt{(j(z')-1728)/j(z')},1)$, and take the limit $ z'\to z$. So far, we have accounted for the two integrals on the right-hand side of Eq. \[eq:G2\_PSL2Z\_z\_rn\], and the trailing constant 2. As we note that$$\begin{aligned}
&6\lim_{z'\to z}\log\left\vert \frac{\sqrt{\frac{j(z')-1728}{j(z')}}-\sqrt{\frac{j(z)-1728}{j(z)}}}{z'-z} \right\vert\notag\\={}&36\log2+18\log3+6\log\pi+24\log|\eta(z)|-5\log|j(z)|,\quad \text{a.e. }z\in\mathfrak H,\label{eq:log_deriv_P_sixth_inv}\end{aligned}$$the leading logarithmic term on the right-hand side of Eq. \[eq:G2\_PSL2Z\_z\_rn\] is also explained.
The proofs of Eqs. \[eq:G2\_Hecke2\_z\_rn\]–\[eq:G2\_Hecke4\_z\_rn\] are essentially similar to that of Eq. \[eq:G2\_PSL2Z\_z\_rn\]. It would suffice to establish analogs of Eq. \[eq:log\_deriv\_P\_sixth\_inv\] for the modular lambda function $ \lambda(z)$, the Ramanujan cubic invariant $ \alpha_{3}(z)$, and the expression $ \alpha_{2}(\frac{z-1}{2})=[1-2\lambda(z)]^{-2}$, applicable to $\text{a.e. }z\in\mathfrak H $:[$$\begin{aligned}
\lim_{z'\to z}\log\left\vert \frac{\lambda(z')-\lambda(z)}{z'-z} \right\vert={}&\frac{4\log 2}{3}+\log\pi+4\log|\eta(z)|+\frac{2\log|1-\lambda(z)|}{3}+\frac{2\log|\lambda(z)|}{3}\notag\\={}&\frac{2\log 2}{3}+\log\pi+4\log\left|\eta\left(\frac{z}{2}\right)\right|+\frac{\log|1-\lambda(z)|}{3}+\frac{5\log|\lambda(z)|}{6};\\
\lim_{z'\to z}\log\left\vert \frac{\alpha_{3}(z')-\alpha_{3}(z)}{z'-z} \right\vert={}&\log 2+\frac{\log3}{2}+\log\pi+4\log|\eta(z)|+\frac{5\log|\alpha_{3}(z)|}{6}+\frac{\log|1-\alpha_{3}(z)|}{2};\\\lim_{z'\to z}\log\left\vert \frac{\alpha_{2}(\frac{z'-1}{2})-\alpha_{2}(\frac{z-1}{2})}{z'-z} \right\vert={}&\frac{10\log 2}{3}+\log\pi+4\log|\eta(z)|+\frac{2\log|1-\lambda(z)|}{3}\notag\\{}&+\frac{2\log|\lambda(z)|}{3}-3\log|1-2\lambda(z)|\notag\\={}&\frac{8\log 2}{3}+\log\pi+4\log\left|\eta\left(\frac{z-1}{2}\right)\right|+\frac{5\log|1-\lambda(z)|}{6}\notag\\{}&+\frac{5\log|\lambda(z)|}{6}-3\log|1-2\lambda(z)|.\end{aligned}$$]{}These formulae, of course, can be verified using Ramanujan’s differentiation formula (Eq. \[eq:P\_nu\_ratio\_deriv\]) along with the relations between the Weierstra[ß]{} discriminant $ \Delta(z)=[\eta(z)]^{24}$ and fractional degree Legendre functions (Eqs. \[eq:alpha\_N\_ratio\_ids\], \[eq:P\_sixth\_eta\] and \[eq:P\_nu\_sqr\_E2\_diff\]).
\[rmk:spec\_G2\_rn\]In the proposition above, the constraint “a.e. $z\in\mathfrak H $” has two connotations: (i) the point $z$ has to be non-elliptic such that $m_z^{\overline \varGamma}=1 $; (ii) the paths of integration (being straight lines in the complex $ \xi$-plane connecting the ends of designated intervals) do not intersect the branch cut of $ P_\nu(\xi)$ or $ P_\nu(-\xi)$. Therefore, the integral representations in Eqs. \[eq:G2\_Hecke2\_z\_rn\]–\[eq:G2\_Hecke4\_z\_rn\] are applicable to $ G_2^{\mathfrak H/\overline\varGamma_0(N)}(i/\sqrt{N})$ for $ N\in\{2,3,4\}$. One can also exploit Eq. \[eq:Pnu\_sqr\_diff\_int\_log\_over\_sqr\] to evaluate these three integrals exactly: [$$\begin{aligned}
&G_2^{\mathfrak H/\overline{\varGamma}_0(2)}\left( \frac{i}{\sqrt{2}} \right)\notag\\={}&\log2+\int_{0}^1\left\{\frac{2}{\smash[b]{\xi}}+\frac{\pi[P_{-1/4}(\xi)]^2}{\sqrt{2}\xi^{2}}-\frac{\pi[P_{-1/4}(-\xi)]^2}{\sqrt{2}\xi^{2}}\right\}\operatorname{d}\xi+\frac{\pi}{\sqrt{2}
}\operatorname{Re}\int_{-1}^1\frac{[P_{-1/4}(\xi)]^2}{\xi^{2}}\operatorname{d}\xi+2\notag=-3\log2;\\&
G_2^{\mathfrak H/\overline{\varGamma}_0(3)}\left( \frac{i}{\sqrt{3}} \right)\notag\\={}&\log 3-\frac23\log2+\int_{0}^1\left\{\frac{2}{\smash[b]{\xi}}+\frac{\pi[P_{-1/3}(\xi)]^2}{\sqrt{3}\xi^{2}}-\frac{\pi[P_{-1/3}(-\xi)]^2}{\sqrt{3}\xi^{2}}\right\}\operatorname{d}\xi+\frac{\pi}{\sqrt{3}
}\operatorname{Re}\int_{-1}^1\frac{[P_{-1/3}(\xi)]^2}{\xi^{2}}\operatorname{d}\xi+2\notag\\={}&\frac{4}{3}\log2-2\log3;\\&G_2^{\mathfrak H/\overline{\varGamma}_0(4)}\left( \frac{i}{\sqrt{4}} \right)\notag\\={}&\log2+\int_{0}^1\left\{\frac{2}{\smash[b]{\xi}}+\frac{\pi[P_{-1/2}(\xi)]^2}{\sqrt{4}\xi^{2}}-\frac{\pi[P_{-1/2}(-\xi)]^2}{\sqrt{4}\xi^{2}}\right\}\operatorname{d}\xi+\frac{\pi}{\sqrt{4}
}\operatorname{Re}\int_{-1}^1\frac{[P_{-1/2}(\xi)]^2}{\xi^{2}}\operatorname{d}\xi+2=-\log2.\label{eq:G2Hecke4_rn_Fricke_pt}\end{aligned}$$]{}One may wish to compare the last integral identity with the generic formula for $ G_2^{\mathfrak H/\overline \varGamma_0(4)}(z)$ (to be derived in Proposition \[prop:G2\_Hecke4\_z\_KZ\_rn\]).
The fundamental domains of the congruence subgroups $ \varGamma_0(1)=SL(2,\mathbb Z)$, $ \varGamma_0(2)$ and $ \varGamma_0(3)$ contain elliptic points. We need to single out these cases for the evaluations of renormalized Green’s functions $ G_2^{\mathfrak H/\overline{\varGamma}}(z)$. (It is worth noting that the renormalized Green’s function $ G_2^{\mathfrak H/\overline{\varGamma}}(z)$ is *not* continuous at the elliptic points: it diverges logarithmically as $z$ approaches an elliptic point; it nonetheless assumes a well-defined finite value at any elliptic point.) In the following proposition, we shall derive integral representations for $ G_2^{\mathfrak H/\overline{\varGamma}}(z)$ ($ \varGamma=\varGamma_0(1)$, $ \varGamma_0(2)$ and $ \varGamma_0(3)$) at elliptic points $z$, and evaluate these integrals in closed form.
We have the following identities:[$$\begin{aligned}
G_2^{\mathfrak H/PSL(2,\mathbb Z)}(i)={}&2\log3+\int_{0}^1\left\{\frac{4}{\xi}+\frac{2\pi[P_{-1/6}(\xi)]^2}{\xi^{2}}-\frac{2\pi[P_{-1/6}(-\xi)]^2}{\xi^{2}}\right\}\operatorname{d}\xi-\frac{40 G}{\pi}+4\notag\\={}&-4(\log 2+\log3);\label{eq:G2_PSL2Z_i_rn}\\
G_2^{\mathfrak H/PSL(2,\mathbb Z)}\left( \frac{1+i\sqrt{3}}{2} \right)={}&3\log\frac{2^2}{3}+\int_{1}^\infty\left\{\frac{4}{\xi}-\frac{8 P_{-1/6}(\xi)Q_{-1/6}(\xi)}{3}-\frac{8[ Q_{-1/6}(\xi)]^{2}}{3\sqrt{3}\pi}\right\}\operatorname{d}\xi-\frac{30\sqrt{3} L(2,\chi_{-3})}{\pi}+6\notag\\={}&-3(2\log2+\log3);\label{eq:G2_PSL2Z_ell3_rn}\\G_2^{\mathfrak H/\overline{\varGamma}_0(2)}\left( \frac{i-1}{2} \right)={}&-2\log2+\int_{1}^\infty\left\{\frac{2}{\xi}- P_{-1/4}(\xi)Q_{-1/4}(\xi)-\frac{[ Q_{-1/4}(\xi)]^{2}}{\pi}\right\}\operatorname{d}\xi-\frac{16G}{\pi}+4\notag\\={}&-4\log2;\label{eq:G2_Hecke2_ell2_rn}\\G_2^{\mathfrak H/\overline{\varGamma}_0(3)}\left( \frac{3+i\sqrt{3}}{6} \right)={}&\log\frac{1}{2^{2}3^3}+\int_{1}^\infty\left\{\frac{2}{\xi}- \frac{2P_{-1/3}(\xi)Q_{-1/3}(\xi)}{3}-\frac{2[ Q_{-1/3}(\xi)]^{2}}{\sqrt{3}\pi}\right\}\operatorname{d}\xi-\frac{9\sqrt{3} L(2,\chi_{-3})}{\pi}+6\notag\\={}&-3\log3,\label{eq:G2_Hecke3_ell3_rn}\end{aligned}$$]{}where $$G\equiv L(2,\chi_{-4}):=\sum_{n=0}^\infty\frac{(-1)^n}{(2n+1)^2},\quad L(2,\chi_{-3}):=\sum_{n=0}^\infty\left[\frac{1}{(3n+1)^2}-\frac{1}{(3n+2)^2}\right].$$
Recalling that $ m_i^{PSL(2,\mathbb Z)}=2$, we may compute$$\begin{aligned}
G_2^{\mathfrak H/PSL(2,\mathbb Z)}(i)={}&\lim_{z\to i}\left[G_2^{\mathfrak H/PSL(2,\mathbb Z)}(z,i)-4\log|z-i|\right]-4\log\left\vert 2\pi[\eta(i)]^4 \right|\end{aligned}$$ using the Legendre–Ramanujan form (Eq. \[eq:G2\_z\_z’\_arb\]) of the integral representation for $ G_2^{\mathfrak H/PSL(2,\mathbb Z)}(z,i)$ (see Eq. \[eq:G2\_PSL2Z\_z\_i\]), which results in the integral formula in Eq. \[eq:G2\_PSL2Z\_i\_rn\]. We can subsequently evaluate the integral by considering Eq. \[eq:Pnu\_sqr\_diff\_int\_log\_over\_sqr\] in the special case where $ \nu=-1/6$, in parallel to what we did in Remark \[rmk:spec\_G2\_rn\].
We can modify Eq. \[eq:G2\_z\_z’\_arb\] as follows:$$\begin{aligned}
&G_{2}^{\mathfrak H/PSL(2,\mathbb Z)}(z,e^{\pi i/3})\notag\\={}&2\log\left\vert \frac{j(z)}{j(z)-1728} \right\vert-4\operatorname{Re}\int_{\sqrt{\frac{j(z)-1728}{j(z)}}}^1\left\{\frac{1}{\xi}+\frac{\pi [P_{-1/6}(\xi)]^2}{3y}\left[ \frac{iP_{-1/6}(-\xi)}{P_{-1/6}(\xi)} -z\right]\left[ \frac{iP_{-1/6}(-\xi)}{P_{-1/6}(\xi)} -\overline{z}\right]\right\}\operatorname{d}\xi\notag\\&-\frac{4\pi}{3y}\operatorname{Re}\int_{-1}^1[P_{-1/6}(\xi)]^2\operatorname{d}\xi,\quad |z|\geq1,-\frac12<\operatorname{Re}z\leq\frac{1}{2},z\neq \frac{1}{2}+i\frac{\sqrt{3}}{2}.\label{eq:G2PSL2Z_ell3_alt}\end{aligned}$$One may recall an integral formula for $ L(2,\chi_{-3})$ from Eq. \[eq:L2chi3\], subtract $ 6\log|z-e^{\pi i/3}|+6\log|2\pi[\eta(e^{i \pi/3})]^4|$ from Eq. \[eq:G2PSL2Z\_ell3\_alt\], and take the limit $ z\to e^{\pi i/3}$. This leads to the integral representation in Eq. \[eq:G2\_PSL2Z\_ell3\_rn\], where we have already used Eq. \[eq:Pnu\_Qnu\_rln\] to recast the integrand in terms of two real-valued functions $ P_\nu(t),Q_\nu(t),t>1$. One can then evaluate the integral representation in Eq. \[eq:G2\_PSL2Z\_ell3\_rn\] with the help of the integral formulae in Eqs. \[eq:PQ-1\_over\_x\_int\] and \[eq:Qnu\_sqr\_1toinf\_int\], as well as the polygamma identities in Eqs. \[eq:psi1\_id1\] and \[eq:psi1\_id2\].
By Eq. \[eq:G2Hecke234\_Pnu\], we can verify the first equality in Eq. \[eq:G2\_Hecke2\_ell2\_rn\] after some routine computations similar to what was described in the last paragraph. The integral formulae in Eqs. \[eq:PQ-1\_over\_x\_int\] and \[eq:Qnu\_sqr\_1toinf\_int\] will then result in the closed-form expression for $ G_2^{\mathfrak H/\overline{\varGamma}_0(2)}((i-1)/2)$. Alternatively, we can exploit the evaluation in Eq. \[eq:G2Hecke4\_rn\_Fricke\_pt\] as follows:$$\begin{aligned}
G_2^{\mathfrak H/\overline{\varGamma}_0(2)}\left(\frac{z-1}{2},\frac{i-1}{2} \right)={}&2G_2^{\mathfrak H/\overline{\varGamma}(2)}(z,i)=2G_2^{\mathfrak H/\overline{\varGamma}_0(4)}\left(\frac{z}{2},\frac{i}{2} \right)\notag\\={}&2G_2^{\mathfrak H/\overline{\varGamma}_0(4)}\left(\frac{i}{\sqrt{4}} \right)+4\log\left\vert \frac{z-i}{2} \right\vert+4\log\left\vert 2\pi\left[ \eta\left( \frac{i}{2} \right) \right]^4 \right\vert+o(1)\notag\\\Longrightarrow\qquad G_2^{\mathfrak H/\overline{\varGamma}_0(2)}\left(\frac{i-1}{2} \right)={}&2G_2^{\mathfrak H/\overline{\varGamma}_0(4)}\left(\frac{i}{\sqrt{4}} \right)+16\log\left\vert \frac{\eta(\frac{i}{2})}{\eta(\frac{i-1}{2})} \right\vert=-2\log2+16\log\frac{1}{\sqrt[8]{2}}=-4\log2.\end{aligned}$$
The proof of Eq. \[eq:G2\_Hecke3\_ell3\_rn\] is similar to that of Eq. \[eq:G2\_PSL2Z\_ell3\_rn\].
A Few Applications of Ramanujan Rotations and Jacobi Elliptic Functions\[subsec:Ramanujan\_Jacobi\]
---------------------------------------------------------------------------------------------------
In the following lemma, we present a geometric transformation that will facilitate the evaluation of certain multiple elliptic integrals by “separating the variables”.
\[lm:R\_rot\]Let $ \mathcal R(s,t),s\in[0,1],t\in[0,+\infty)$ be a bivariate function with suitable regularity, then we have the following integral identity for $ \lambda\in(0,1)$:$$\begin{aligned}
&
\int_0^{\pi/2}\left[ \int_0^{\pi/2}\frac{\mathcal R(\cos\theta,\tan\phi)\operatorname{d}\phi}{\sqrt{\smash[b]{1-\lambda\cos^2\phi}}} \right]\frac{\operatorname{d}\theta}{\sqrt{\smash[b]{1-\lambda\cos^2\theta}}}\notag\\={}&\int_0^{\pi/2}\left[ \int_0^{\pi/2}\mathcal R\left( \frac{\cos\theta}{\sqrt{\smash[b]{1-\lambda\sin^2\theta\sin^2\phi}}}, \sqrt{\frac{1-\lambda(1-\sin^2\theta\cos^2\phi)}{\smash[b]{1-\lambda\sin^2\theta\sin^2\phi}}}\tan\phi\right)\frac{\operatorname{d}\phi}{\sqrt{\smash[b]{1-\lambda+\lambda^{2}\sin^2\theta\sin^2\phi\cos^2\phi}}} \right]\operatorname{d}\theta\notag\\={}&\int_{S^2_{+++}}\mathcal R\left(\frac{Z}{\sqrt{1-\lambda Y^{2}}},\sqrt{\frac{1-\lambda(1-X^{2})}{1-\lambda Y^{2}}} \frac{Y}{X}\right)\frac{\operatorname{d}\sigma}{\sqrt{(1-\lambda)(1-Z^2)+\lambda^{2}Y^2X^2}}.\label{eq:Ramanujan_rotation}\end{aligned}$$Here in the last step, the integral is carried out on the $ (+,+,+)$-octant of the unit sphere $ S^2_{+++}:=\{(X,Y,Z)\in\mathbb R^3|X^2+Y^2+Z^2=1,X>0,Y>0,Z>0\}$, equipped with the standard spherical coordinates $ (X,Y,Z)=(\sin\theta\cos\phi,\sin\theta\sin\phi,\cos\theta)$ and surface element $ \operatorname{d}\sigma=\sin\theta\operatorname{d}\theta\operatorname{d}\phi$.
The following proof is a modest extension of our previous work [@Zhou2013Pnu Proposition 4.3], which handled a special case (where the function $ \mathcal R(s,t)\equiv 1$ was held a constant) in an effort to recover the geometric motivations behind Entry 7(x) in Chapter 17 of Ramanujan’s second notebook [@RN3 pp. 110–111].
With the spherical coordinates and a rotation of the coordinate axes, it is straightforward to verify that $$\begin{aligned}
&
\int_0^{\pi/2}\left[ \int_0^{\pi/2}\frac{ \mathcal R(\cos\theta,\tan\phi)\operatorname{d}\phi}{\sqrt{\smash[b]{1-\lambda\cos^2\theta}}\sqrt{\smash[b]{1-\lambda\cos^2\phi}}} \right]\operatorname{d}\theta\notag\\={}&\int_{S^{2}_{+++}}\frac{ \mathcal R(Z,Y/X)}{\sqrt{1-\lambda Z^{2}}\sqrt{1-Z^2-\lambda X^2}}\operatorname{d}\sigma=\int_{S^{2}_{+++}}\frac{ \mathcal R(X,Y/Z)}{\sqrt{1-\lambda X^{2}}\sqrt{1-X^2-\lambda Z^2}}\operatorname{d}\sigma\notag\\={}&\int_0^{\pi/2}\left[ \int_0^{\pi/2}\frac{ \mathcal R(\sin\theta\cos\phi,\tan\theta\sin\phi)\operatorname{d}\phi}{\sqrt{\smash[b]{1-\lambda\sin^{2}\theta\cos^{2}\phi}}\sqrt{\smash[b]{1-\sin^{2}\theta\cos^{2}\phi-\lambda\cos^2\theta}}} \right]\sin\theta\operatorname{d}\theta.\label{eq:Ramanujan_rotation1}\end{aligned}$$Now, substituting $ \phi=\arctan(\sqrt{1-\lambda\sin^2\theta}\tan\psi)$, we can verify that$$\begin{aligned}
&
\int_0^{\pi/2}\frac{ \mathcal R(\sin\theta\cos\phi,\tan\theta\sin\phi)\operatorname{d}\phi}{\sqrt{\smash[b]{1-\lambda\sin^{2}\theta\cos^{2}\phi}}\sqrt{\smash[b]{1-\sin^{2}\theta\cos^{2}\phi-\lambda\cos^2\theta}}}\notag\\={}&\int_0^{\pi/2}\frac{ \mathcal R\left(\frac{\sin\theta\cos\psi}{\sqrt{\smash[b]{1-\lambda\sin^{2}\theta\sin^2\psi}}},\sqrt{\frac{1-\lambda\sin^2\theta}{\smash[b]{1-\lambda\sin^{2}\theta\sin^2\psi}}} \frac{\sin\theta\sin\psi}{\cos\theta} \right)\operatorname{d}\psi}{\sqrt{\smash[b]{(1-\lambda)(1-\sin^{2}\theta\cos^{2}\psi)+\lambda^2\cos^2\theta\sin^2\theta\sin^2\psi}}}.\label{eq:Ramanujan_rotation2}\end{aligned}$$ Therefore, we may use Eq. \[eq:Ramanujan\_rotation2\] to convert Eq. \[eq:Ramanujan\_rotation1\] into$$\begin{aligned}
\int_0^{\pi/2}\left[ \int_0^{\pi/2}\frac{ \mathcal R(\cos\theta,\tan\phi)\operatorname{d}\phi}{\sqrt{\smash[b]{1-\lambda\cos^2\theta}}\sqrt{\smash[b]{1-\lambda\cos^2\phi}}} \right]\operatorname{d}\theta={}&
\int_{S^2_{+++}}\frac{\mathcal R\left(\frac{X}{\sqrt{1-\lambda Y^{2}}}, \sqrt{\frac{1-\lambda(1-Z^{2})}{\smash[b]{1-\lambda Y^{2}}}} \frac{Y{}}{Z}\right)}{\sqrt{(1-\lambda)(1-X^2)+\lambda^{2}Y^2Z^2}}\operatorname{d}\sigma\notag\\={}&\int_{S^2_{+++}}\frac{\mathcal R \left(\frac{Z}{\sqrt{1-\lambda Y^{2}}}, \sqrt{\frac{1-\lambda(1-X^{2})}{\smash[b]{1-\lambda Y^{2}}}} \frac{Y}{X}\right)}{\sqrt{(1-\lambda)(1-Z^2)+\lambda^{2}Y^2X^2}}\operatorname{d}\sigma,\end{aligned}$$which confirms Eq. \[eq:Ramanujan\_rotation\].
We recall the definition of the Jacobi $ \Theta$-function:$$\begin{aligned}
&
\Theta(u|\lambda)\notag\\:={}&\prod_{n=1}^\infty\left\{\left[ 1-e^{-2n{\pi\mathbf K(\sqrt{1-\lambda})}/{\mathbf K(\sqrt{\lambda})}} \right]\left[1-2e^{-(2n-1){\pi\mathbf K(\sqrt{1-\lambda})}/{\mathbf K(\sqrt{\lambda})}}\cos\frac{\pi u}{{\mathbf K(\sqrt{\lambda})}}+e^{-2(2n-1){\pi\mathbf K(\sqrt{1-\lambda})}/{\mathbf K(\sqrt{\lambda})}}\right]\right\}\notag\\\equiv{}&\sum_{n\in\mathbb Z}(-1)^n e^{-n^{2}\pi\mathbf K(\sqrt{1-\lambda})/\mathbf K(\sqrt{\lambda})}\cos\frac{n\pi u}{\mathbf K(\sqrt{\lambda})},\quad u\in\mathbb C,\lambda\in(\mathbb C\smallsetminus\mathbb R)\cup(0,1).\end{aligned}$$ Its logarithmic derivative with respect to $u$ is the Jacobi $ \operatorname{Z}$-function: $$\begin{aligned}
\operatorname{Z}(u|\lambda):=\frac{\partial}{\partial u}\log\Theta(u|\lambda).\end{aligned}$$ The Jacobi $ \Theta$-function can be used to introduce the Jacobi elliptic functions $ \operatorname{sn}(u|\lambda)$, $\operatorname{cn}(u|\lambda)$, $\operatorname{dn}(u|\lambda)$ as follows:$$\begin{aligned}
\operatorname{sn}(u|\lambda):={}&-\frac{i}{\sqrt[4]{\lambda}}\frac{\Theta(u+i\mathbf K(\sqrt{1-\lambda})|\lambda)}{\Theta(u|\lambda)}\exp\left\{{\frac{\pi i}{2\mathbf K(\sqrt{\lambda})}}\left[ u+\frac{i\mathbf K(\sqrt{1-\lambda})}{2} \right]\right\},\\\operatorname{cn}(u|\lambda):={}&-i\sqrt[^4\!\!]{\frac{1-\lambda}{\lambda}}\frac{\Theta(u+\mathbf K(\sqrt{\lambda})+i\mathbf K(\sqrt{1-\lambda})|\lambda)}{\Theta(u|\lambda)}\exp\left\{{\frac{\pi i}{2\mathbf K(\sqrt{\lambda})}}\left[ u+\mathbf K(\sqrt{\lambda})+\frac{i\mathbf K(\sqrt{1-\lambda})}{2} \right]\right\},\\\operatorname{dn}(u|\lambda):={}&\sqrt[4]{1-\lambda}\frac{\Theta(u+\mathbf K(\sqrt{\lambda})|\lambda)}{\Theta(u|\lambda)}.\end{aligned}$$All the aforementioned functions named after Jacobi can be also defined for $ \lambda\in(-\infty,0)\cup(1,+\infty)$ by analytic continuation. The Jacobi elliptic functions satisfy the following properties:$$\begin{aligned}
u=\int_0^{\operatorname{sn}(u|\lambda)}\frac{\operatorname{d}t}{\sqrt{1-t^2}\sqrt{1-\lambda t ^2}},\quad \operatorname{sn}^2(u|\lambda)+\operatorname{cn}^2(u|\lambda)=1,\quad \lambda\operatorname{sn}^2(u|\lambda)+\operatorname{dn}^2(u|\lambda)=1\end{aligned}$$for $ 0<\lambda<1,0\leq u\leq\mathbf K(\sqrt{\lambda})$.
We have the following integral identities for $ 0<\lambda<1$:$$\begin{aligned}
\int_0^{\pi/2}\frac{\log\cos\phi\operatorname{d}\phi}{\sqrt{\smash[b]{1-\lambda\cos^2\phi}}}={}&-\frac{\pi}{4}\mathbf K(\sqrt{1-\lambda})-\frac{\mathbf K(\sqrt{\lambda})}{4}\log\lambda\label{eq:log_sn_int};\\\int_0^{\pi/2}\frac{\log(1-\lambda\cos^{2}\phi)\operatorname{d}\phi}{\sqrt{\smash[b]{1-\lambda\cos^2\phi}}}={}&\frac{\mathbf K(\sqrt{\lambda})}{2}\log(1-\lambda);\label{eq:log_dn_int}\\\int_0^{\pi/2}\left[\int_0^{\pi/2}\frac{\log(1-\lambda\cos^2\theta\cos^{2}\phi)\operatorname{d}\phi}{\sqrt{\smash[b]{1-\lambda\cos^2\phi}}}\right]\frac{\operatorname{d}\theta}{\sqrt{\smash[b]{1-\lambda\cos^2\theta}}}={}&-\frac{\pi}{6}\mathbf K(\sqrt{1-\lambda})\mathbf K(\sqrt{\lambda})+\frac{[\mathbf K(\sqrt{\lambda})]^{2}}{3}\log\frac{4(1-\lambda)}{\sqrt{\lambda}}.\label{eq:log_sn_sn_int_int}\end{aligned}$$
With a special case of Jacobi’s Fourier expansions [@WhittakerWatson1927 §22.5, Example 3]:$$\begin{aligned}
\log\frac{\operatorname{sn}(u|\lambda)}{\sin\frac{\pi u}{2\mathbf K(\sqrt{\lambda})}}={}&\frac{1}{4}\log\frac{16e^{-\pi\mathbf K(\sqrt{1-\lambda})/\mathbf K(\sqrt{\lambda})}}{\lambda}+\sum_{n=1}^\infty\frac{2}{n}\frac{e^{-n\pi\mathbf K(\sqrt{1-\lambda})/\mathbf K(\sqrt{\lambda})}}{1+e^{-n\pi\mathbf K(\sqrt{1-\lambda})/\mathbf K(\sqrt{\lambda})}} \cos\frac{n\pi u}{\mathbf K(\sqrt{\lambda})},\end{aligned}$$we may compute $$\begin{aligned}
\int_0^{\pi/2}\frac{\log\cos\phi\operatorname{d}\phi}{\sqrt{\smash[b]{1-\lambda\cos^2\phi}}}={}&\int_0^{\mathbf K(\sqrt{\lambda})}\log\operatorname{sn}(u|\lambda)\operatorname{d}u\notag\\={}&\int_0^{\mathbf K(\sqrt{\lambda})}\log\sin\frac{\pi v}{2\mathbf K(\sqrt{\lambda})}\operatorname{d}v+\frac{\mathbf K(\sqrt{\lambda})}{4}\log\frac{16e^{-\pi\mathbf K(\sqrt{1-\lambda})/\mathbf K(\sqrt{\lambda})}}{\lambda}\notag\\={}&-\mathbf K(\sqrt{\lambda})\log2+\frac{\mathbf K(\sqrt{\lambda})}{4}\log\frac{16e^{-\pi\mathbf K(\sqrt{1-\lambda})/\mathbf K(\sqrt{\lambda})}}{\lambda},\end{aligned}$$where the integration over $v$ is elementary. This verifies Eq. \[eq:log\_sn\_int\], which was also formerly mentioned in [@WhittakerWatson1927 §22.5, Example 4], [@GradshteynRyzhik item 4.386.3] and [@ByrdFriedman item 800.01].
Likewise, one can verify Eq. \[eq:log\_dn\_int\] [@ByrdFriedman item 800.03] with Jacobi’s Fourier series for $ \log\operatorname{dn}(u|\lambda)$.
To demonstrate Eq. \[eq:log\_sn\_sn\_int\_int\], we need an identity$$\begin{aligned}
\int_0^{\pi/2}\frac{\log(1-\lambda\operatorname{sn}^2(v|\lambda)\sin^2\varphi)}{\sqrt{\smash[b]{1-\lambda\sin^2\varphi}}}\operatorname{d}\varphi=-2\mathbf K(\sqrt{\lambda})\int_0^v\operatorname{Z}(u|\lambda)\operatorname{d}u.\label{eq:log_1_minus_sn_sqr_int}\end{aligned}$$The proof of Eq. \[eq:log\_1\_minus\_sn\_sqr\_int\] is straightforward: one simply differentiates both sides with respect to $v$, and compares the result with a standard integral representation for the Jacobi $ \operatorname{Z}$-function $ \operatorname{Z}(v|\lambda)$ [@ByrdFriedman item 140.03]. Thus, we can compute$$\begin{aligned}
&\int_0^{\pi/2}\left[\int_0^{\pi/2}\frac{\log(1-\lambda\cos^2\theta\cos^{2}\phi)\operatorname{d}\phi}{\sqrt{\smash[b]{1-\lambda\cos^2\phi}}}\right]\frac{\operatorname{d}\theta}{\sqrt{\smash[b]{1-\lambda\cos^2\theta}}}=-2\mathbf K(\sqrt{\lambda})\int_0^{\mathbf K(\sqrt{\lambda})}\left[\int_0^v\operatorname{Z}(u|\lambda)\operatorname{d}u\right]\operatorname{d}v\notag\\={}&-2\mathbf K(\sqrt{\lambda})\int_0^{\mathbf K(\sqrt{\lambda})}\log\frac{\Theta(v|\lambda)}{\Theta(0|\lambda)}\operatorname{d}v=-2[\mathbf K(\sqrt{\lambda})]^2\log\prod_{n=1}^\infty \frac{1}{[1-e^{-(2n-1){\pi\mathbf K(\sqrt{1-\lambda})}/{\mathbf K(\sqrt{\lambda})}}]^2}\notag\\={}& -4[\mathbf K(\sqrt{\lambda})]^2\log\prod_{n=1}^\infty \frac{1-e^{-2n{\pi\mathbf K(\sqrt{1-\lambda})}/{\mathbf K(\sqrt{\lambda})}}}{1-e^{-n{\pi\mathbf K(\sqrt{1-\lambda})}/{\mathbf K(\sqrt{\lambda})}}}.\label{eq:log_Theta_int}\end{aligned}$$ Here in Eq. \[eq:log\_Theta\_int\], we have made use of the fact that$$\begin{aligned}
\int_0^{\mathbf K(\sqrt{\lambda})}\log\left[1-2e^{-(2n-1){\pi\mathbf K(\sqrt{1-\lambda})}/{\mathbf K(\sqrt{\lambda})}}\cos\frac{\pi v}{{\mathbf K(\sqrt{\lambda})}}+e^{-2(2n-1){\pi\mathbf K(\sqrt{1-\lambda})}/{\mathbf K(\sqrt{\lambda})}}\right]\operatorname{d}v=0,\quad 0<\lambda<1,n\in\mathbb Z_{>0},\end{aligned}$$which originates from an elementary application of Cauchy’s integral formula:$$\begin{aligned}
\int_0^{ \pi}\log(1-2a\cos \theta+a^2)\operatorname{d}\theta=\operatorname{Re}\counterint_{|w|=1}\log(1-aw)\frac{\operatorname{d}w}{iw}=0,\quad 0<a<1.\end{aligned}$$Thus, as we compare Eq. \[eq:log\_Theta\_int\] with the following consequence of Eqs. \[eq:E4\_Ell\_Ramanujan\]–\[eq:E6\_Ell\_Ramanujan\] and Eqs. \[eq:Landen\_2\]–\[eq:double\_half\_lambda\]:
$$\begin{aligned}
e^{-2{\pi\mathbf K(\sqrt{1-\lambda})}/{\mathbf K(\sqrt{\lambda})}}\prod_{n=1}^\infty\left[ 1-e^{-2n{\pi\mathbf K(\sqrt{1-\lambda})}/{\mathbf K(\sqrt{\lambda})}} \right]^{24}={}&\left[ \frac{2\mathbf K(\sqrt{\lambda})}{\pi} \right]^{12}\frac{\lambda^{2}(1-\lambda)^2}{256},\\e^{-{\pi\mathbf K(\sqrt{1-\lambda})}/{\mathbf K(\sqrt{\lambda})}}\prod_{n=1}^\infty\left[ 1-e^{-n{\pi\mathbf K(\sqrt{1-\lambda})}/{\mathbf K(\sqrt{\lambda})}} \right]^{24}={}&\left[ \frac{2(1+\sqrt{\lambda})\mathbf K(\sqrt{\lambda})}{\pi} \right]^{12}\frac{\frac{16\lambda}{(1+\sqrt{\lambda})^{4}}\left[1-\frac{4\sqrt{\lambda}}{(1+\sqrt{\lambda})^{2}}\right]^2}{256},\end{aligned}$$
where $ 0<\lambda<1$, we see that the statement in Eq. \[eq:log\_sn\_sn\_int\_int\] is true.
For $0<\lambda<1 $, we have $$\begin{aligned}
\int_0^\pi\frac{\mathbf K(\sin\theta)\sin\theta\operatorname{d}\theta}{\sqrt{(2-\lambda)^2-\lambda^2\sin^2\theta}}={}&[\mathbf K(\sqrt{\lambda})]^2,\label{eq:R_rot1}\\\int_0^\pi\frac{\mathbf K(\sin\theta)\sin\theta}{\sqrt{(2-\lambda)^2-\lambda^2\sin^2\theta}}\log\frac{\sqrt{(2-\lambda)^2-\lambda^2\sin^2\theta}}{2\sqrt{\sin\theta}}\operatorname{d}\theta={}&\frac{\pi}{3}\mathbf K(\sqrt{\lambda})\mathbf K(\sqrt{1-\lambda})+\frac{[\mathbf K(\sqrt{\lambda})]^2}{3}\log\frac{\lambda(1-\lambda)}{2}\notag\\{}&-\frac{\pi}{2}\int_0^1\frac{\mathbf K(k)\operatorname{d}k}{\sqrt{4(1-\lambda)+\lambda^2k^2}},\label{eq:R_rot2}\\\int_0^1\frac{\mathbf K(k)\operatorname{d}k}{\sqrt{4(1-\lambda)+\lambda^2k^2}}+\int_0^1\frac{\mathbf K(k)\operatorname{d}k}{\sqrt{\lambda^2+4(1-\lambda)k^2}}={}&\mathbf K(\sqrt{\lambda})\mathbf K(\sqrt{1-\lambda}).\label{eq:R_rot3}\end{aligned}$$
The identity in Eq. \[eq:R\_rot1\] appeared as [@Zhou2013Pnu Eq. 40$'$], and was proved by an application of Eq. \[eq:Ramanujan\_rotation\] in the simplest case $ \mathcal R(s,t)\equiv1$.
Now, we work out Eq. \[eq:R\_rot2\] in detail, starting with the transformations[$$\begin{aligned}
&\int_0^\pi\frac{\mathbf K(\sin\theta)\sin\theta}{\sqrt{(2-\lambda)^2-\lambda^2\sin^2\theta}}\log\frac{\sqrt{(2-\lambda)^2-\lambda^2\sin^2\theta}}{2}\operatorname{d}\theta=\int_0^{\pi/2}\frac{\mathbf K(\sin\theta)\sin\theta}{\sqrt{1-\lambda +\frac{\lambda^2}{4}\cos^2\theta}}\log\sqrt{1-\lambda+\frac{\lambda^2}{4}\cos^2\theta}\operatorname{d}\theta\notag\\={}&\int_0^{\pi/2}\sin\theta\operatorname{d}\theta\int_0^{\pi/2}\operatorname{d}\phi\frac{\log\sqrt{1-\lambda+\frac{\lambda^2}{4}\cos^2\theta}}{\sqrt{1-\sin^2\theta\sin^2\phi}\sqrt{1-\lambda+\frac{\lambda^2}{4}\cos^2\theta}}=\int_{S^2_{+++}}\frac{\log\sqrt{1-\lambda+\frac{\lambda^2}{4}Z^{2}}\operatorname{d}\sigma}{\sqrt{(1-Y^{2})(1-\lambda+\frac{\lambda^{2}}{4}Z^{2})}}\notag\\={}&\int_{S^2_{+++}}\frac{\log\sqrt{1-\lambda+\frac{\lambda^2}{4}Y^{2}}\operatorname{d}\sigma}{\sqrt{(1-Z^{2})(1-\lambda+\frac{\lambda^{2}}{4}Y^{2})}}=\frac{1}{2}\int_0^{\pi/2}\operatorname{d}\theta\int_0^\pi\operatorname{d}\phi\frac{\log\sqrt{1-\lambda+\frac{\lambda^2}{4}\sin^2\theta\sin^2\phi}}{\sqrt{1-\lambda+\frac{\lambda^2}{4}\sin^2\theta\sin^2\phi}}\notag\\={}&\int_0^{\pi/2}\operatorname{d}\theta\int_0^{\pi/2}\operatorname{d}\phi\frac{\log\sqrt{\smash[b]{1-\lambda+\lambda^{2}\sin^2\theta\sin^2\phi\cos^2\phi}}}{\sqrt{\smash[b]{1-\lambda+\lambda^{2}\sin^2\theta\sin^2\phi\cos^2\phi}}}=\int_{S^2_{+++}}\frac{\log\sqrt{\frac{(1-\lambda)(1-Z^{2})+\lambda^{2}Y^2X^2}{1-Z^{2}}}\operatorname{d}\sigma}{\sqrt{(1-\lambda)(1-Z^{2})+\lambda^{2}Y^2X^2}}\notag\\={}&\int_0^{\pi/2}\left[ \int_0^{\pi/2}\frac{1}{\sqrt{\smash[b]{1-\lambda\cos^2\phi}}}\log\sqrt{\frac{\cos^2\phi(1-\lambda\cos^2\phi)[1-\lambda\cos^2\theta+(1-\lambda)\tan^2\phi]}{1-\lambda\cos^2\theta\cos^2\phi}}\operatorname{d}\phi \right]\frac{\operatorname{d}\theta}{\sqrt{\smash[b]{1-\lambda\cos^2\theta}}},\label{eq:K_over_sqrt_log_sqrt_int}\end{aligned}$$where we have employed Eq. \[eq:Ramanujan\_rotation\] in the last step. Meanwhile, we have the following computations:$$\begin{aligned}
&\int_0^\pi\frac{\mathbf K(\sin\theta)\sin\theta}{\sqrt{(2-\lambda)^2-\lambda^2\sin^2\theta}}\log\sin\theta\operatorname{d}\theta=\int_0^{\pi/2}\sin\theta\operatorname{d}\theta\int_0^{\pi/2}\operatorname{d}\phi\frac{\log\frac{\sin\theta\sin\phi}{\sin\phi}}{\sqrt{1-\sin^2\theta\sin^2\phi}\sqrt{1-\lambda+\frac{\lambda^2}{4}\cos^2\theta}}\notag\\={}&\int_{S^2_{+++}}\frac{\log Z\operatorname{d}\sigma}{\sqrt{(1-\lambda)(1-Z^{2})+\lambda^{2}Y^2X^2}}-\int_0^{\pi/2}\sin\theta\operatorname{d}\theta\int_0^{\pi/2}\operatorname{d}\phi\frac{\log\sin\phi}{\sqrt{1-\sin^2\theta\sin^2\phi}\sqrt{1-\lambda+\frac{\lambda^2}{4}\cos^2\theta}}\notag\\={}&\int_0^{\pi/2}\left[ \int_0^{\pi/2}\frac{1}{\sqrt{\smash[b]{1-\lambda\cos^2\phi}}}\log\sqrt{\frac{\cos^2\theta\cos^2\phi[1-\lambda\cos^2\theta+(1-\lambda)\tan^2\phi]}{1-\lambda\cos^2\theta}}\operatorname{d}\phi \right]\frac{\operatorname{d}\theta}{\sqrt{\smash[b]{1-\lambda\cos^2\theta}}}\notag\\{}&+\int_0^{\pi/2}\frac{\frac{\pi}{4}\mathbf K(\cos\theta)+\frac{\mathbf K(\sin\theta)}{2}\log\sin\theta}{\sqrt{1-\lambda+\frac{\lambda^2}{4}\cos^2\theta}}\sin\theta\operatorname{d}\theta,\label{eq:K_over_sqrt_log_sin_int}\end{aligned}$$]{}where the last step involves an application of Ramanujan’s rotations (Eq. \[eq:Ramanujan\_rotation\]) to the integral over $S^2_{+++}$, and a reference to Eq. \[eq:log\_sn\_int\] for the integration over $\phi\in(0,\pi/2)$. It is then clear that Eqs. \[eq:K\_over\_sqrt\_log\_sqrt\_int\] and \[eq:K\_over\_sqrt\_log\_sin\_int\] combine into$$\begin{aligned}
&\int_0^\pi\frac{\mathbf K(\sin\theta)\sin\theta}{\sqrt{(2-\lambda)^2-\lambda^2\sin^2\theta}}\log\frac{\sqrt{(2-\lambda)^2-\lambda^2\sin^2\theta}}{2\sqrt{\sin\theta}}\operatorname{d}\theta\notag\\={}&\int_0^{\pi/2}\left[ \int_0^{\pi/2}\frac{1}{\sqrt{\smash[b]{1-\lambda\cos^2\phi}}}\log\sqrt{\frac{(1-\lambda\cos^2\theta)(1-\lambda\cos^2\phi)}{\cos^2\theta(1-\lambda\cos^2\theta\cos^2\phi)}}\operatorname{d}\phi \right]\frac{\operatorname{d}\theta}{\sqrt{\smash[b]{1-\lambda\cos^2\theta}}}\notag\\{}&-\frac{\pi}{2}\int_0^{\pi/2}\frac{\mathbf K(\cos\theta)\sin\theta\operatorname{d}\theta}{\sqrt{(2-\lambda)^2-\lambda^2\sin^2\theta}}.\label{eq:K_over_sqr_log_quotient_int}\end{aligned}$$As we simplify the last line of Eq. \[eq:K\_over\_sqr\_log\_quotient\_int\] with the integral formulae in Eqs. \[eq:log\_sn\_int\]–\[eq:log\_sn\_sn\_int\_int\], we can confirm Eq. \[eq:R\_rot2\].
We shall prove Eq. \[eq:R\_rot3\] by verifying the following chain of identities: $$\begin{aligned}
\int_0^1\frac{\mathbf K(k)\operatorname{d}k}{\sqrt{4(1-\lambda)+\lambda^2k^2}}={}&\iint_{1<|z|<{1/\sqrt[4]{1-\lambda}},\operatorname{Im}z>0}\frac{\operatorname{d}\operatorname{Re}z\operatorname{d}\operatorname{Im}z}{|(1-z^{2})[1-(1-\lambda)\ z^2]|}\notag\\={}&\iint_{-\mathbf K(\sqrt{1-\lambda})<u<\mathbf K(\sqrt{1-\lambda}),0<v<\frac{1}{2}\mathbf K(\sqrt{\lambda}),\left\vert \operatorname{sn}(u+iv| 1-\lambda)\right|>1}\operatorname{d}u\operatorname{d}v\notag\\={}&\mathbf K(\sqrt{\lambda})\mathbf K(\sqrt{1-\lambda})-\iint_{|z|<1,\operatorname{Im}z>0}\frac{\operatorname{d}\operatorname{Re}z\operatorname{d}\operatorname{Im}z}{|(1-z^{2})[1-(1-\lambda)z^2]|}\notag\\={}&\mathbf K(\sqrt{\lambda})\mathbf K(\sqrt{1-\lambda})-\int_0^1\frac{\mathbf K(k)\operatorname{d}k}{\sqrt{\lambda^2+4(1-\lambda)k^2}}.\label{eq:Jacobi_area_comp1}\end{aligned}$$ For convenience, we write $ \lambda_*=1-\lambda$. To prove the first equality in Eq. \[eq:Jacobi\_area\_comp1\], we employ the polar coordinates $ z=re^{i\theta}$ to compute[$$\begin{aligned}
&
\iint_{1<|z|<{1/\sqrt[4]{\lambda_*}},\operatorname{Im}z>0}\frac{\operatorname{d}\operatorname{Re}z\operatorname{d}\operatorname{Im}z}{|(1-z^{2})(1-\lambda_* z^2)|}=\int_1^{1/\sqrt[4]{\lambda_*}}r\left[ \int_0^\pi\frac{\operatorname{d}\theta}{|(1-r^2e^{2i\theta})(1-\lambda_* r^2e^{2i\theta})|}\right]\operatorname{d}r\notag\\={}&\int_1^{1/\sqrt[4]{\lambda_*}}r\left[ \int_0^\pi\frac{\operatorname{d}\phi}{\sqrt{\smash[b]{(1+r^{2})^{2}(1-\lambda_* r^{2})^{2}-4(1-\lambda_*) r^2(1-\lambda_* r^4)\sin^2\phi}}}\right]\operatorname{d}r\notag\\={}&2\int_1^{1/\sqrt[4]{\lambda_*}}\frac{r}{(1+r^{2})(1-\lambda_* r^{2})}\mathbf K\left( \sqrt{1-\left(\frac{r^2-1}{r^2+1}\right)^2 \left(\frac{1+\lambda_* r^2}{1-\lambda_* r^2}\right)^2} \right)\operatorname{d}r\notag\\={}&2\int_1^{1/\sqrt[4]{\lambda_*}}\frac{1}{(1-\lambda_*) r}\mathbf K\left( \frac{1-\lambda_* r^{4}}{(1-\lambda_*)r^{2}}\right)\operatorname{d}r=\int_0^1\frac{\mathbf K(k)\operatorname{d}k}{\sqrt{4\lambda_*+(1-\lambda_*)^2k^2}}=\int_0^1\frac{\mathbf K(k)\operatorname{d}k}{\sqrt{4(1-\lambda)+\lambda^2k^2}}.\label{eq:Jacobi_repn_a}\end{aligned}$$]{}Here in Eq. \[eq:Jacobi\_repn\_a\], we have used an angular transformation (which is reminiscent of the derivation for Ramanujan rotations in Lemma \[lm:R\_rot\])$$\begin{aligned}
\theta=\arctan\left( \frac{r^2-1}{r^2+1}\tan\phi \right)\end{aligned}$$to complete the integral in $\phi$ with the standard integral representation of $ \mathbf K$, before invoking Landen’s transformation (Eq. \[eq:Landen\_1\]) and a variable substitution $ k=(1-\lambda_* r^4)/[(1-\lambda_*)r^2]$. To deduce the second equality in Eq. \[eq:Jacobi\_area\_comp1\] from the first one, we recall that the conformal mapping $ u+iv\mapsto \operatorname{sn}(u+iv|\lambda_*)$ establishes a bijection between the open rectangle $ -\mathbf K(\sqrt{\lambda_*})<u<\mathbf K(\sqrt{\lambda_*}),0<v<\mathbf K(\sqrt{1-\lambda_*})$ and the upper half-plane [@SteinII p. 234], and that $$\begin{aligned}
\left\vert \operatorname{sn}\left(\left.u+i\frac{\mathbf K(\sqrt{1-\lambda_*})}{2}\right| \lambda_*\right)\right|=\frac{1}{\sqrt[4]{\lambda_*}},\quad \forall u\in\mathbb R,\lambda_*\in(0,1)\label{eq:sn_abs_fourth_root}\end{aligned}$$ follows from the addition formula for $ \operatorname{sn}(u+w|\lambda_*)$ [@ByrdFriedman item 123.01] and special values of $ \operatorname{sn}(w|\lambda_*)$, $ \operatorname{cn}(w|\lambda_*)$, $ \operatorname{dn}(w|\lambda_*)$ at $ w=i\mathbf K(\sqrt{1-\lambda_*})/2$ [@ByrdFriedman item 122.11]. We may rewrite the second line in Eq. \[eq:Jacobi\_area\_comp1\] as$$\begin{aligned}
&\iint_{-\mathbf K( \sqrt{\lambda_*} )<u<\mathbf K( \sqrt{\lambda_*} ),0<v<\frac{1}{2}\mathbf K( \sqrt{1-\lambda_*})}\operatorname{d}u\operatorname{d}v-\iint_{-\mathbf K( \sqrt{\lambda_*} )<u<\mathbf K( \sqrt{\lambda_*} ),0<v<\frac{1}{2}\mathbf K( \sqrt{1-\lambda_*}),\left\vert \operatorname{sn}\left(u+iv| \lambda_*\right)\right|<1}\operatorname{d}u\operatorname{d}v\notag\\={}&\mathbf K(\sqrt{\lambda})\mathbf K(\sqrt{1-\lambda})-\iint_{-\mathbf K( \sqrt{\lambda_*} )<u<\mathbf K( \sqrt{\lambda_*} ),0<v<\frac{1}{2}\mathbf K( \sqrt{1-\lambda_*}),\left\vert \operatorname{sn}\left(u+iv| \lambda_*\right)\right|<1}\operatorname{d}u\operatorname{d}v,\end{aligned}$$ which is conformally equivalent to the third line. To complete the verification of Eq. \[eq:Jacobi\_area\_comp1\], we compute $$\begin{aligned}
&\iint_{|z|<1,\operatorname{Im}z>0}\frac{\operatorname{d}\operatorname{Re}z\operatorname{d}\operatorname{Im}z}{|(1-z^{2})(1-\lambda_* z^2)|}=\int_0^{1}r\left[ \int_0^\pi\frac{\operatorname{d}\theta}{|(1-r^2e^{2i\theta})(1-\lambda_* r^2e^{2i\theta})|}\right]\operatorname{d}r\notag\\={}&\int_0^{1}r\left[ \int_0^\pi\frac{\operatorname{d}\phi}{\sqrt{\smash[b]{(1+r^{2})^{2}(1-\lambda_* r^{2})^{2}-4(1-\lambda_*) r^2(1-\lambda_* r^4)\sin^2\phi}}}\right]\operatorname{d}r\notag\\={}&2\int_0^{1}\frac{r}{(1+r^{2})(1-\lambda_* r^{2})}\mathbf K\left( \sqrt{1-\left(\frac{1-r^2}{1+r^2}\right)^2 \left(\frac{1+\lambda_* r^2}{1-\lambda_* r^2}\right)^2} \right)\operatorname{d}r=2\int_0^1\frac{r}{1-\lambda_* r^4}\mathbf K\left(\frac{(1-\lambda_*) r^{2}}{1-\lambda_* r^4} \right)\operatorname{d}r \notag\\={}&\int_{0}^{1}\frac{\mathbf K(k)\operatorname{d}k}{\sqrt{(1-\lambda_*)^2+4\lambda_* k^{2}}}=\int_0^1\frac{\mathbf K(k)\operatorname{d}k}{\sqrt{\lambda^2+4(1-\lambda)k^2}}.\label{eq:Jacobi_repn_d}\end{aligned}$$ (We also note that Eq. \[eq:R\_rot3\] can be proved without invoking Jacobi elliptic functions [@Zhou2013Spheres Eq. 91].)
\[subsec:G2\_Hecke4\_GZ\_rn\]Elementary Evaluation of $ G_2^{\mathfrak H/\overline{\varGamma}_0(4)}(z)$ with Gross–Zagier Renormalization
-----------------------------------------------------------------------------------------------------------------------------------------
The Kontsevich–Zagier integral representation for $ G_2^{\mathfrak H/\overline{\varGamma}_0(4)}(z)$ has appeared in Eq. \[eq:G2\_Hecke4\_z\_rn\]. To prepare for the main goal of this subsection (evaluation of $ G_2^{\mathfrak H/\overline{\varGamma}_0(4)}(z),\forall z\in\mathfrak H$ in Proposition \[prop:G2\_Hecke4\_z\_KZ\_rn\]), we explore certain integrals whose integrands involve complete elliptic integrals and logarithms, in the next lemma.
\[lm:GZ\_lim\]
1. For any $\lambda\in\mathbb C\smallsetminus\mathbb R$ and $ \mu\in\mathbb (1,+\infty)$, we have the following identity:$$\begin{aligned}
&
\frac{2}{\pi}\int_0^1\frac{\mathbf K(\sqrt{t})\mathbf K(\sqrt{1-t})}{t-\lambda}\log(\mu -t)\operatorname{d}t\notag\\={}&\int_{0}^1\frac{[\mathbf K(\sqrt{t})]^2}{t-\lambda}\operatorname{d}t-\int_{1/\mu}^{1}\frac{[\mathbf K(\sqrt{t})]^2}{\lambda t-1}\operatorname{d}t- \frac{1}{\lambda}\left[ \mathbf K\left( \sqrt{\frac{1}{\lambda}} \right) \right]^2\log(\lambda-\mu)+\frac{\pi}{\lambda} \mathbf K\left( \sqrt{\frac{1}{\lambda}} \right)\mathbf K\left( \sqrt{1-\frac{1}{\lambda}} \right);\label{eq:int_K'K_log_etc}\end{aligned}$$while the following formula holds for $\lambda\in\mathbb C\smallsetminus\mathbb R$ and $ \mu\in\mathbb (0,1)$:$$\begin{aligned}
\frac{2}{\pi}\int_0^1\frac{\mathbf K(\sqrt{t})\mathbf K(\sqrt{1-t})}{1-\lambda t}\log(1-\mu t)\operatorname{d}t=-\int_0^\mu\frac{[\mathbf K(\sqrt{t})]^2\operatorname{d}t}{t-\lambda}+[\mathbf K(\sqrt{\lambda})]^2\log\frac{\mu-\lambda}{\lambda}+\pi i\frac{\operatorname{Im}\lambda}{|\operatorname{Im}\lambda|}[\mathbf K(\sqrt{\lambda})]^2.\label{eq:int_K'K_log_etc_alt_form}\end{aligned}$$Here, all the integrations run along subsets of the open unit interval $ t\in(0,1)$.
2. For $ \operatorname{Im}\lambda\neq0$, we have the following limit formulae:[$$\begin{aligned}
&
\lim_{\varepsilon\to0^{+}}\left\{ \int^{\lambda+\varepsilon}_0\frac{[\mathbf K(\sqrt{t})]^2}{ t-\lambda}\operatorname{d}t-[\mathbf K(\sqrt{\lambda})]^2\log\varepsilon \right\}\notag\\={}&\pi i\frac{\operatorname{Im}\lambda}{|\operatorname{Im}\lambda|}[ \mathbf K( \sqrt{\lambda} ) ]^2+\frac{\pi}{3}\mathbf K(\sqrt{\lambda})\mathbf K(\sqrt{1-\lambda})-\frac{2[\mathbf K(\sqrt{\lambda})]^2}{3}\log[4\lambda(1-\lambda)];\label{eq:GZ_lim1}\\
&\lim_{\varepsilon\to0^{+}}\left[ \int^{\lambda+\varepsilon}_0\frac{2\mathbf K(\sqrt{t})\mathbf K(\sqrt{1-t})}{ t-\lambda}\operatorname{d}t-2\mathbf K(\sqrt{\lambda})\mathbf K(\sqrt{1-\lambda})\log\varepsilon \right]\notag\\
={}&-\frac{\pi}{\lambda}\left[ \mathbf K\left( \sqrt{\frac{1}{\lambda}} \right) \right]^2+\frac{2\pi[\mathbf K(\sqrt{\lambda})]^2}{3}-\frac{\pi[\mathbf K(\sqrt{1-\lambda})]^2}{3}-\frac{4}{3}\mathbf K(\sqrt{\lambda})\mathbf K(\sqrt{1-\lambda})\log[4\lambda(1-\lambda)];\label{eq:GZ_lim2}\\&\lim_{\varepsilon\to0^{+}}\left\{ \int_{\lambda+\varepsilon}^1\frac{[\mathbf K(\sqrt{1-t})]^2}{ t-\lambda}\operatorname{d}t+[\mathbf K(\sqrt{1-\lambda})]^2\log\varepsilon \right\}\notag\\={}&-\frac{\pi}{3}\mathbf K(\sqrt{\lambda})\mathbf K(\sqrt{1-\lambda})+\frac{2[\mathbf K(\sqrt{1-\lambda})]^2}{3}\log[4\lambda(1-\lambda)],\label{eq:GZ_lim3}\end{aligned}$$]{}where all the integrals are computed along straight line segments joining the end points.
<!-- -->
1. Using the imaginary modulus and inverse modulus transformations (Eqs. \[eq:im\_mod\]–\[eq:inv\_mod\]), one may verify that the expression $$\begin{aligned}
f_{\mu}(t):=[\mathbf K(\sqrt{t/\mu})]^2\log\left(\frac{\mu}{t}-\mu\right)-\pi \mathbf K(\sqrt{t/\mu})\mathbf K(\sqrt{1-(t/\mu)}),\quad t\in\mathbb C\smallsetminus((-\infty,0]\cup[1,+\infty))\end{aligned}$$can be extended as an analytic function for $ t\in\mathbb C\smallsetminus[1,+\infty)$, so long as $ \mu\in(1,+\infty)$. Hence, Eq. \[eq:f\_slit\_plane\_Cauchy\_int\] is applicable, and we have$$\begin{aligned}
&
[\mathbf K(\sqrt{s})]^2\log\left(\frac{1}{s}-\mu\right)-\pi \mathbf K(\sqrt{s})\mathbf K(\sqrt{1-s})=f_{\mu}(\mu s)=\frac{1}{2\pi i}\int_{0}^1\frac{f_{\mu}\big(\frac{1}{t}+i0^+\big)-f_{\mu}\big(\frac{1}{t}-i0^+\big)}{1-\mu st}\frac{\operatorname{d}t}{t}\notag\\={}& -\int_{0}^1\frac{[\mathbf K(\sqrt{t})]^2}{1-st}\operatorname{d}t-\int_{1/\mu}^{1}\frac{[\mathbf K(\sqrt{t})]^2}{t-s}\operatorname{d}t+\frac{2}{\pi}\int_{0}^1\frac{\mathbf K(\sqrt{t})\mathbf K(\sqrt{1-t})}{1-st}\log(\mu-t)\operatorname{d}t,\quad s\in\mathbb C\smallsetminus\mathbb R.\label{eq:K_sqr_log_Cauchy}\end{aligned}$$Setting $ s=1/\lambda$ in Eq. \[eq:K\_sqr\_log\_Cauchy\], we can verify the identity claimed in Eq. \[eq:int\_K’K\_log\_etc\].
Before establishing Eq. \[eq:int\_K’K\_log\_etc\_alt\_form\] in its stated form, we may first consider the scenarios where $ 0<\mu<\lambda<1$, and read off the imaginary part of the following vanishing integral:$$\begin{aligned}
0=\int_{-\infty+i0^{+}}^{+\infty+i0^+}\frac{[\mathbf K(\sqrt{1-t})]^{2}+2i\mathbf K(\sqrt{1-t})\mathbf K(\sqrt{t})-[\mathbf K(\sqrt{t})]^2}{1-\lambda t}\log(1-\mu t)\operatorname{d}t.\label{eq:int0_res_sum_1}\end{aligned}$$ Here, the right-hand side of Eq. \[eq:int0\_res\_sum\_1\] can be decomposed into three parts. With the transformation $ t=s/(s-1)$ and Eqs. \[eq:im\_mod\]–\[eq:inv\_mod\], we see that the portion $ \operatorname{Re}t<0$ has vanishing contribution to the imaginary part of Eq. \[eq:int0\_res\_sum\_1\]:
$$\begin{aligned}
\operatorname{Im}\int_{-\infty+i0^{+}}^{0+i0^{+}}(\cdots)\operatorname{d}t={}&\operatorname{Im}\int_{0-i0^+}^{1-i0^+}\frac{[\mathbf K(\sqrt{1-s})]^2}{1-(1-\lambda)s}\log\left( 1-\frac{\mu s}{s-1} \right)\operatorname{d}s=0.\label{eq:int0_res_sum_1a}
\intertext{The second part is simply}
\operatorname{Im}\int_{0+i0^{+}}^{1+i0^{+}}(\cdots)\operatorname{d}t={}&2\int_0^1\frac{\mathbf K(\sqrt{t})\mathbf K(\sqrt{1-t})}{1-\lambda t}\log(1-\mu t)\operatorname{d}t.\label{eq:int0_res_sum_1b}\intertext{With the transformation $ t=1/s$ and Eqs.~\ref{eq:im_mod}--\ref{eq:inv_mod}, we obtain}\operatorname{Im}\int_{1+i0^{+}}^{+\infty+i0^{+}}(\cdots)\operatorname{d}t={}&\operatorname{Im}\int_{0-i0^{+}}^{1-i0^{+}}\frac{[\mathbf K(\sqrt{s})]^2}{\lambda-s}\log\left( 1-\frac{\mu }{s} \right)\operatorname{d}s\notag\\={}&\pi\int_0^\mu\frac{[\mathbf K(\sqrt{t})]^2\operatorname{d}t}{t-\lambda}-\pi[\mathbf K(\sqrt{\lambda})]^2\log\frac{\lambda-\mu}{\lambda},\label{eq:int0_res_sum_1c}\end{aligned}$$
upon application of the Plemelj jump formula. Adding through Eqs. \[eq:int0\_res\_sum\_1a\]–\[eq:int0\_res\_sum\_1c\], we see that$$\begin{aligned}
{}&\frac{2}{\pi}\int_0^1\frac{\mathbf K(\sqrt{t})\mathbf K(\sqrt{1-t})}{1-\lambda t}\log(1-\mu t)\operatorname{d}t=-\int_0^\mu\frac{[\mathbf K(\sqrt{t})]^2\operatorname{d}t}{t-\lambda}+[\mathbf K(\sqrt{\lambda})]^2\log\frac{\lambda-\mu}{\lambda},\quad 0<\mu<\lambda<1
$$ is true. Thus, after a judicious analytic continuation of the equation above in the variable $\lambda$, one can verify Eq. \[eq:int\_K’K\_log\_etc\_alt\_form\].
2. We may also examine the left-hand side of Eq. \[eq:int\_K’K\_log\_etc\] from a different perspective. Without loss of generality, we momentarily assume that $ \lambda>1$ and $ \mu>1$, and use [@Zhou2013Pnu Eq. 43] to deduce$$\begin{aligned}
\frac{2}{\pi}\int_0^1\frac{\mathbf K(\sqrt{t})\mathbf K(\sqrt{1-t})}{t-\lambda}\log(\mu -t)\operatorname{d}t={}&-\frac{1}{2\pi}\int_{0}^{\pi}\mathbf K(\sin\theta)\sin\theta\left[ \int_0^{2\pi}\frac{\log\frac{2\mu-1-\sin\theta\cos\phi}{2}}{2\lambda-1-\sin\theta\cos\phi} \operatorname{d}\phi\right]\operatorname{d}\theta.\label{eq:int_K'K_log_etc_alt}\end{aligned}$$ For $ \lambda,\mu\in(1,+\infty)$ and $ \theta\in(0,\pi)$, the integration over $\phi$ can now be completed with residue calculus: $$\begin{aligned}
&
\int_0^{2\pi}\frac{\log\frac{2\mu-1-\sin\theta\cos\phi}{2}}{2\lambda-1-\sin\theta\cos\phi} \operatorname{d}\phi\notag\\={}&\operatorname{Re}\counterint_{|w|=1}\frac{\log\frac{\sin^2\theta}{4[2\mu-1+\sqrt{(2\mu-1)^2-\sin^2\theta}]}+2\log\left( \frac{2\mu-1+\sqrt{(2\mu-1)^2-\sin^2\theta}}{\sin\theta}-w\right)}{2\lambda-1-\frac{\sin\theta}{2}(w+\frac{1}{w})}\frac{\operatorname{d}w}{iw}\notag\\={}&\frac{2\pi}{\sqrt{(2\lambda-1)^{2}-\sin^2\theta}}\log\frac{\big[\sqrt{\smash[b]{(2\lambda-1)^2-\sin^2\theta}}+\sqrt{\smash[b]{(2\mu-1)^2-\sin^2\theta}}-2(\lambda-\mu)\big]^{2}}{4\big[2\mu-1+\sqrt{\smash[b]{(2\mu-1)^2-\sin^2\theta}}\big]},\label{eq:log_angle_int}\end{aligned}$$ where we have collected the residue at the simple pole $z=(2\lambda-1-\sqrt{(2\lambda-1)^2-\sin^2\theta})/\sin\theta\in(-1,1) $. Combining Eqs. \[eq:int\_K’K\_log\_etc\_alt\] with \[eq:log\_angle\_int\], we obtain an identity$$\begin{aligned}
&
\frac{2}{\pi}\int_0^1\frac{\mathbf K(\sqrt{t})\mathbf K(\sqrt{1-t})}{t-\lambda}\log(\mu -t)\operatorname{d}t\notag\\={}&\int_{0}^{\pi}\frac{\mathbf K(\sin\theta)\sin\theta}{\sqrt{(2\lambda-1)^{2}-\sin^2\theta}}\log\frac{4\big[2\mu-1+\sqrt{\smash[b]{(2\mu-1)^2-\sin^2\theta}}\big]}{\big[\sqrt{\smash[b]{(2\lambda-1)^2-\sin^2\theta}}+\sqrt{\smash[b]{(2\mu-1)^2-\sin^2\theta}}-2(\lambda-\mu)\big]^{2}}\operatorname{d}\theta,\quad \lambda>1,\mu>1.\label{eq:K'K_log_to_K_log_id}\end{aligned}$$
In Eq. \[eq:K’K\_log\_to\_K\_log\_id\], one may trade $ \lambda$ and $ \mu$ for their respective inverses $1/\lambda $ and $ 1/\mu$, to produce a formula $$\begin{aligned}
&
\frac{2}{\pi}\int_0^1\frac{\mathbf K(\sqrt{t})\mathbf K(\sqrt{1-t})}{1-\lambda t}\log(1 -\mu t)\operatorname{d}t\notag\\={}&-\int_0^\pi\frac{\mathbf K(\sin\theta)\sin\theta}{\sqrt{(2-\lambda)^{2}-\lambda^{2}\sin^2\theta}}\log\frac{4\big[2-\mu+\sqrt{\smash[b]{(2-\mu)^2-\mu^{2}\sin^2\theta}}\big]}{\big[\frac{\mu}{\lambda}\sqrt{\smash[b]{(2-\lambda)^2-\lambda^{2}\sin^2\theta}}+\sqrt{\smash[b]{(2-\mu)^2-\mu^{2}\sin^2\theta}}-2(\frac{\mu}{\lambda}-1)\big]^{2}}\operatorname{d}\theta\label{eq:K'K_log_mu_lambda_new}\end{aligned}$$for $ 0<\lambda<1$ and $0<\mu<1 $, upon exploiting the following identity (see Eqs. \[eq:K2\] and \[eq:R\_rot1\]):$$\begin{aligned}
\frac{2}{\pi}\int_0^1\frac{\mathbf K(\sqrt{\smash[b]{\vphantom{1}t}})\mathbf K(\sqrt{\smash[b]{1-t}})}{1-\lambda t}\operatorname{d}t=
\int_0^\pi\frac{\mathbf K(\sin\theta)\sin\theta\operatorname{d}\theta}{\sqrt{(2-\lambda)^2-\lambda^2\sin^2\theta}},\quad 0<\lambda<1.\end{aligned}$$Especially, setting $ \mu=\lambda$ in Eq. \[eq:K’K\_log\_mu\_lambda\_new\], we obtain$$\begin{aligned}
&
\frac{2}{\pi}\int_0^1\frac{\mathbf K(\sqrt{t})\mathbf K(\sqrt{1-t})}{1-\lambda t}\log(1 -\lambda t)\operatorname{d}t=-\int_0^\pi\frac{\mathbf K(\sin\theta)\sin\theta}{\sqrt{(2-\lambda)^{2}-\lambda^{2}\sin^2\theta}}\log\frac{2-\lambda+\sqrt{\smash[b]{(2-\lambda)^2-\lambda^{2}\sin^2\theta}}}{(2-\lambda)^2-\lambda^{2}\sin^2\theta}\operatorname{d}\theta\notag\\={}&-\int_0^\pi\frac{\mathbf K(\sin\theta)\sin\theta}{\sqrt{(2-\lambda)^{2}-\lambda^{2}\sin^2\theta}}\log\frac{2-\lambda+\sqrt{\smash[b]{(2-\lambda)^2-\lambda^{2}\sin^2\theta}}}{4\sin\theta}\operatorname{d}\theta+\frac{2\pi}{3}\mathbf K(\sqrt{\lambda})\mathbf K(\sqrt{1-\lambda})\notag\\{}&+\frac{2[\mathbf K(\sqrt{\lambda})]^2}{3}\log\frac{\lambda(1-\lambda)}{2}-\pi\int_0^1\frac{\mathbf K(k)\operatorname{d}k}{\sqrt{4(1-\lambda)+\lambda^2k^2}},\quad 0<\lambda<1,\label{eq:K'K_log_lambda_lambda_new}\end{aligned}$$ after referring to Eq. \[eq:R\_rot2\]. By an elementary identity$$\begin{aligned}
\log\frac{2-\lambda+\sqrt{\smash[b]{(2-\lambda)^2-\lambda^{2}\sin^2\theta}}}{4\sin\theta}=\log\frac{\lambda}{4}+\tanh^{-1}\frac{\sqrt{\smash[b]{(2-\lambda)^2-\lambda^{2}\sin^2\theta}}}{2-\lambda},\end{aligned}$$we can further simplify the last line of Eq. \[eq:K’K\_log\_lambda\_lambda\_new\]:$$\begin{aligned}
&-\int_0^\pi\frac{\mathbf K(\sin\theta)\sin\theta}{\sqrt{(2-\lambda)^{2}-\lambda^{2}\sin^2\theta}}\log\frac{2-\lambda+\sqrt{\smash[b]{(2-\lambda)^2-\lambda^{2}\sin^2\theta}}}{4\sin\theta}\operatorname{d}\theta\notag\\={}&-[\mathbf K(\sqrt{\lambda})]^2\log\frac{\lambda}{4}-2\int_0^1\frac{\mathbf K(\sqrt{1-\kappa^2})}{\sqrt{4(1-\lambda)+\lambda^{2}\kappa^2}}\tanh^{-1}\sqrt{\frac{4(1-\lambda)+\lambda^{2}\kappa^2}{4(1-\lambda)+\lambda^{2}}}\operatorname{d}\kappa\notag\\={}&-[\mathbf K(\sqrt{\lambda})]^2\log\frac{\lambda}{4}-2\int_{0}^{1}\left[\int_0^1\frac{\mathbf K(\sqrt{1-\kappa^2})}{1-k^2\kappa^2}\operatorname{d}\kappa\right]\frac{\operatorname{d}k}{\sqrt{\lambda^{2}+4(1-\lambda)k^2}}\notag\\={}&-[\mathbf K(\sqrt{\lambda})]^2\log\frac{\lambda}{4}-\pi\int_0^1\frac{\mathbf K(k)\operatorname{d}k}{\sqrt{\lambda^2+4(1-\lambda)k^2}},\label{eq:K'K_artanh_redn}\end{aligned}$$ where we have resorted to Eq. \[eq:K1\] for the reduction of the integral in $ \kappa\in(0,1)$. Recalling the integral identity in Eq. \[eq:R\_rot3\], we can merge Eqs. \[eq:K’K\_log\_lambda\_lambda\_new\] and \[eq:K’K\_artanh\_redn\] into $$\begin{aligned}
&
\frac{2}{\pi}\int_0^1\frac{\mathbf K(\sqrt{t})\mathbf K(\sqrt{1-t})}{1-\lambda t}\log(1 -\lambda t)\operatorname{d}t\notag\\={}&-\frac{\pi}{3}\mathbf K(\sqrt{\lambda})\mathbf K(\sqrt{1-\lambda})+\frac{2[\mathbf K(\sqrt{\lambda})]^2}{3}\log\frac{4(1-\lambda)}{\sqrt{\lambda}},\quad \forall\lambda\in(\mathbb C\smallsetminus\mathbb R)\cup(0,1),\label{eq:K'K_xlogx_int_id}\end{aligned}$$ after analytic continuation.
By an analytic continuation of Eq. \[eq:int\_K’K\_log\_etc\_alt\_form\], the limit expressed on the left-hand side of Eq. \[eq:GZ\_lim1\] amounts to $$\begin{aligned}
&\pi i\frac{\operatorname{Im}\lambda}{|\operatorname{Im}\lambda|}[ \mathbf K( \sqrt{\lambda} ) ]^2-[\mathbf K(\sqrt{\lambda})]^2\log\lambda-\frac{2}{\pi}\int_0^1\frac{\mathbf K(\sqrt{t})\mathbf K(\sqrt{1-t})}{1-\lambda t}\log(1 -\lambda t)\operatorname{d}t.\end{aligned}$$ As we simplify the last integral with Eq. \[eq:K’K\_xlogx\_int\_id\], we see that both sides of Eq. \[eq:GZ\_lim1\] are indeed equal.
For $\operatorname{Re}\mu\in(0,1),\operatorname{Im}\mu>0$, we may employ the inverse modulus transformation (Eq. \[eq:inv\_mod\]) to verify that $$\begin{aligned}
\int_{1/\mu}^{1}\frac{[\mathbf K(\sqrt{t})]^2}{\lambda t-1}\operatorname{d}t=\int_\mu^1\frac{[\mathbf K(\sqrt{t})]^2-2i\mathbf K(\sqrt{t})\mathbf K(\sqrt{1-t})-[\mathbf K(\sqrt{1-t})]^2}{t-\lambda}\operatorname{d}t.\label{eq:K_sqr_1to3}\end{aligned}$$Therefore, for $\operatorname{Re}\lambda\in(0,1),\operatorname{Im}\lambda>0$, we have the following integral identity, by courtesy of Eqs. \[eq:int\_K’K\_log\_etc\], \[eq:GZ\_lim1\], \[eq:K’K\_xlogx\_int\_id\] and \[eq:K\_sqr\_1to3\]:[$$\begin{aligned}
&\frac{\pi}{3\lambda} \mathbf K\left( \sqrt{\frac{1}{\lambda}} \right)\mathbf K\left( \sqrt{1-\frac{1}{\lambda}} \right)-\frac{2}{3\lambda}\left[ \mathbf K\left( \sqrt{\frac{1}{\lambda}} \right) \right]^2\log[4\lambda(\lambda-1)]\notag\\={}&\int_{0}^1\frac{[\mathbf K(\sqrt{t})]^2}{t-\lambda}\operatorname{d}t+\frac{\pi}{\lambda} \mathbf K\left( \sqrt{\frac{1}{\lambda}} \right)\mathbf K\left( \sqrt{1-\frac{1}{\lambda}} \right)\notag\\{}&-\lim_{\mu\to\lambda+0^{+}}\left\{\int_\mu^1\frac{[\mathbf K(\sqrt{t})]^2-2i\mathbf K(\sqrt{t})\mathbf K(\sqrt{1-t})-[\mathbf K(\sqrt{1-t})]^2}{t-\lambda}\operatorname{d}t+ \frac{1}{\lambda}\left[ \mathbf K\left( \sqrt{\frac{1}{\lambda}} \right) \right]^2\log(\lambda-\mu)\right\}\notag\\={}&-\frac{2\pi}{3}\mathbf K(\sqrt{\lambda})\mathbf K(\sqrt{1-\lambda})-\frac{2[\mathbf K(\sqrt{\lambda})]^2}{3}\log[4\lambda(1-\lambda)]\notag\\&+\lim_{\mu\to\lambda+0^{+}}\left\{\int_\mu^1\frac{2i\mathbf K(\sqrt{t})\mathbf K(\sqrt{1-t})}{t-\lambda}\operatorname{d}t+2i\mathbf K(\sqrt{\lambda})\mathbf K(\sqrt{1-\lambda})\log(\mu-\lambda)\right\}\notag\\&+\lim_{\mu\to\lambda+0^{+}}\left\{\int_\mu^1\frac{[\mathbf K(\sqrt{1-t})]^2}{t-\lambda}\operatorname{d}t+[\mathbf K(\sqrt{1-\lambda})]^2\log(\mu-\lambda)\right\},\label{eq:GZ_lim1'}\end{aligned}$$]{}where $ \log(-1)=\pi i$.
Now, we examine Eq. \[eq:GZ\_lim1’\] in the limit scenario where $ \operatorname{Im}\lambda\to0^+$, and $ \mu$ tends to $\lambda+0^{+}$ along the real axis. Reading off the imaginary part, we arrive at the following identity:$$\begin{aligned}
&
\lim_{\mu\to\lambda+0^+}\left[\int_\mu^1\frac{2\mathbf K(\sqrt{t})\mathbf K(\sqrt{1-t})}{t-\lambda}\operatorname{d}t+2\mathbf K(\sqrt{\lambda})\mathbf K(\sqrt{1-\lambda})\log(\mu-\lambda)\right]\notag\\={}&\frac{\pi [\mathbf K(\sqrt{1-\lambda}) ]^2}{3}-\frac{2\pi [\mathbf K(\sqrt{\lambda}) ]^2}{3}+\frac{4\mathbf K(\sqrt{\lambda})\mathbf K(\sqrt{1-\lambda})}{3}\log[4\lambda(1-\lambda)],\quad 0<\lambda<1.\label{eq:GZ_lim2'}\end{aligned}$$Clearly, analytic continuations of Eq. \[eq:GZ\_lim2’\] lead to Eq. \[eq:GZ\_lim2\]. Subtracting Eq. \[eq:GZ\_lim2’\] from Eq. \[eq:GZ\_lim1’\], we arrive at Eq. \[eq:GZ\_lim3\].
\[prop:G2\_Hecke4\_z\_KZ\_rn\]For all $ z\in\mathfrak H$, we have $$\begin{aligned}
G_2^{\mathfrak H/\overline{\varGamma}_0(4)}(z)=-\frac{1}{3}\log\left\vert\frac{\Delta(z)}{\Delta(2z)}\right\vert=-8\log\left\vert \frac{\eta(z)}{\eta(2z)} \right\vert=-\frac{1}{3}\log\frac{2^4|1-\lambda(2z)|^{2}}{|\lambda(2z)|},\label{eq:G2_Hecke4_z_rn_eval}\end{aligned}$$which verifies the corresponding Gross–Zagier algebraicity conjecture (Eq. \[eq:GZ\_conj\_rn\]).
Without loss of generality, we first suppose that the point $ z\in\mathfrak H$ satisfies$$\begin{aligned}
0<|\operatorname{Re}z|<\frac{1}{2},\left\vert z+\frac{1}{4} \right\vert>\frac{1}{4},\left\vert z-\frac{1}{4} \right\vert>\frac{1}{4},\label{eq:z_narrow_range}\end{aligned}$$so that $ \operatorname{Im}\lambda(2z)\operatorname{Re}z>0$, and pick an infinitesimal $ \delta\to0$ such that $ \varepsilon:=\lambda(2(z+\delta))-\lambda(2z)\to0^+$. The infinitesimal parameter $\varepsilon $ is related to the Gross–Zagier renormalization procedure (Eq. \[eq:G2\_rn\_non\_ell\_defn\]) via the following limit:$$\begin{aligned}
\lim_{\delta\to0}\log\left\vert \frac{\lambda(2(z+\delta))-\lambda(2z)}{2\delta} \right\vert={}&\frac{4\log 2}{3}+\log\pi+4\log|\eta(2z)|+\frac{2\log|1-\lambda(2z)|}{3}+\frac{2\log|\lambda(2z)|}{3}\notag\\={}&\frac{2\log 2}{3}+\log\pi+4\log|\eta(z)|+\frac{\log|1-\lambda(2z)|}{3}+\frac{5\log|\lambda(2z)|}{6}.\label{eq:lambda_deriv_abs}\end{aligned}$$
Naturally, the Kontsevich–Zagier integral representation in Eq. \[eq:G2Hecke234\_Pnu\] and the limit formulae in Eqs. \[eq:GZ\_lim1\]–\[eq:GZ\_lim3\] allow us to compute [$$\begin{aligned}
&
G_2^{\mathfrak H/\overline{\varGamma}_0(4)}(z,z+\delta)\notag\\={}&\frac{2}{\pi \operatorname{Im}(z+\delta)}\operatorname{Re}\int_0^{\lambda(2(z+\delta))}[\mathbf K(\sqrt{t})]^2\varrho_{2,-1/2}(1-2t|z)\left[\frac{i\mathbf K(\sqrt{1-t})}{\mathbf K(\sqrt{t})}-\vphantom{\overline{\frac{1}{}}}2(z+\delta)\right]\left[\frac{i\mathbf K(\sqrt{1-t})}{\mathbf K(\sqrt{t})}-2\overline{(z+\delta)}\right]\operatorname{d}t\notag\\&+\frac{2}{\pi \operatorname{Im}(z+\delta)}\operatorname{Re}\int_{0}^1[\mathbf K(\sqrt{1-t})]^2\varrho_{2,-1/2}(1-2t|z)\operatorname{d}t\notag\\={}&-\frac{1}{4\operatorname{Im}(z+\delta)}\operatorname{Re}\int_0^{\lambda(2(z+\delta))}[\mathbf K(\sqrt{t})]^2\left[\frac{i\mathbf K(\sqrt{1-t})}{\mathbf K(\sqrt{t})}-\vphantom{\overline{\frac{1}{}}}2(z+\delta)\right]\left[\frac{i\mathbf K(\sqrt{1-t})}{\mathbf K(\sqrt{t})}-2\overline{(z+\delta)}\right]\times\notag\\&\times y\frac{\partial}{\partial y}\left\{\frac{1}{t-\lambda(2z)}\frac{1}{[\mathbf K(\sqrt{\lambda(2z)})]^2y}\right\}\operatorname{d}t\notag\\&-\frac{1}{4\operatorname{Im}(z+\delta)}y\frac{\partial}{\partial y}\operatorname{Re}\left\{\int_{0}^1\frac{[\mathbf K(\sqrt{1-t})]^2}{t-\lambda(2z)}\frac{\operatorname{d}t}{[\mathbf K(\sqrt{\lambda(2z)})]^2y}\right\}\notag\\={}&-\frac{|z+\delta|^{2}}{ \operatorname{Im}(z+\delta)}y\frac{\partial}{\partial y}\operatorname{Re}\left\{\int^{\lambda(2(z+\delta))}_0\frac{[\mathbf K(\sqrt{t})]^2}{t-\lambda(2z)}\frac{\operatorname{d}t}{[\mathbf K(\sqrt{\lambda(2z)})]^2y}\right\}\notag\\&+\frac{\operatorname{Re}(z+\delta)}{ \operatorname{Im}(z+\delta)}y\frac{\partial}{\partial y}\operatorname{Re}\left\{\int_0^{\lambda(2(z+\delta))}\frac{i\mathbf K(\sqrt{1-t})\mathbf K(\sqrt{t})}{t-\lambda(2z)}\frac{\operatorname{d}t}{[\mathbf K(\sqrt{\lambda(2z)})]^2y}\right\}\notag\\&-\frac{1}{4 \operatorname{Im}(z+\delta)}y\frac{\partial}{\partial y}\operatorname{Re}\left\{\int_{\lambda(2(z+\delta))}^1\frac{[\mathbf K(\sqrt{1-t})]^2}{t-\lambda(2z)}\frac{\operatorname{d}t}{[\mathbf K(\sqrt{\lambda(2z)})]^2y}\right\}\notag\\={}&-\frac{|z+\delta|^{2}}{ \operatorname{Im}(z+\delta)}y\frac{\partial}{\partial y}\operatorname{Re}\left\{\frac{\log\varepsilon+o(1)}{y}+\pi i\frac{x}{|x|}\frac{1}{y}-\frac{2\pi i z}{3y}-\frac{2\log[4\lambda(2z)(1-\lambda(2z))]}{3y}\right\}\notag\\{}&+\frac{\operatorname{Re}(z+\delta)}{ \operatorname{Im}(z+\delta)}y\frac{\partial}{\partial y}\operatorname{Re}\left\{\frac{2z\log\varepsilon+o(1)}{y}-\pi i\frac{(1-2z\frac{x}{|x|})^2}{2y}+\frac{\pi i(1+2z^{2})}{3y}-\frac{4z\log[4\lambda(2z)(1-\lambda(2z))]}{3y}\right\}\notag\\&-\frac{1}{4 \operatorname{Im}(z+\delta)}y\frac{\partial}{\partial y}\operatorname{Re}\left\{\frac{4z^{2}\log\varepsilon+o(1)}{y}+\frac{2\pi iz}{3y}-\frac{8z^{2}\log[4\lambda(2z)(1-\lambda(2z))]}{3y}\right\}\notag\\={}&2\log\varepsilon-\frac{4}{3}\log|4\lambda(2z)(1-\lambda(2z))|+o(1),\quad\text{as }\delta\to0.\label{eq:G2_Hecke4_close_pts}\end{aligned}$$]{}Here, to arrive at the last step of Eq. \[eq:G2\_Hecke4\_close\_pts\], we have replaced all the occurrences of $ z+\delta$ in the penultimate step with $ z$ (which introduces at most $ O(|\delta|\log|\delta|)$ error).
For points $ z\in\mathfrak H$ in the range specified by Eq. \[eq:z\_narrow\_range\], one can directly check that Eq. \[eq:G2\_Hecke4\_z\_rn\_eval\] is a consequence of Eq. \[eq:lambda\_deriv\_abs\] and \[eq:G2\_Hecke4\_close\_pts\]:$$\begin{aligned}
&
G_2^{\mathfrak H/\overline{\varGamma}_0(4)}(z,z+\delta)-2\log|\delta|-2\log\left\vert 2\pi[\eta(z)]^4 \right|\notag\\={}&-\frac{4}{3}\log|4\lambda(2z)(1-\lambda(2z))|+\frac{4\log 2}{3}+\frac{2\log|1-\lambda(2z)|}{3}+\frac{5\log|\lambda(2z)|}{3}+o(1).\end{aligned}$$ The rest of our claim follows from continuity and $ \varGamma_{0}(4)$-invariance with respect to the variable $z$, which are manifested by the integral representation in Eq. \[eq:G2\_Hecke4\_z\_rn\].
As a side note, we mention that many explicit formulae for the function $ \Delta(z)/\Delta(2z)$ at CM points $z$ have been produced in Weber’s treatise on elliptic functions [@WeberVol3; @WeberErrata], as well as Ramanujan’s notebooks [@RN5 Chap. 34]. To put their work into the context of automorphic Green’s functions, we give a couple of examples involving $ G_2^{\mathfrak H/\overline{\varGamma}_0(4)}(z)$: Here in Eq. \[eq:ex\_47\], the quintic polynomial equation satisfied by $\xi$ has a unique positive root and a solvable Galois group $D_5$ (the pentagon symmetry group). An explicit radical form for $ \xi$ is known [@WatsonQuintic]: $$\begin{aligned}
\xi=\sqrt{\frac{10}{\sqrt[5]{A}-\sqrt[5]{B}-\sqrt[5]{C}-\sqrt[5]{D}}}\end{aligned}$$where$$\begin{aligned}
A={}&+39000+18200\sqrt{5}+(1720+920\sqrt{5})\sqrt{\smash[b]{\vphantom{\tfrac12}\smash[t]{235+94\sqrt{5}}}};\notag\\B={}&-39000-18200\sqrt{5}+(1720+920\sqrt{5})\sqrt{\smash[b]{\vphantom{\tfrac12}\smash[t]{235+94\sqrt{5}}}};\notag\\C={}&-39000+18200\sqrt{5}-(1720-920\sqrt{5})\sqrt{\smash[b]{\vphantom{\tfrac12}\smash[t]{235-94\sqrt{5}}}};\notag\\D={}&-39000+18200\sqrt{5}+(1720-920\sqrt{5})\sqrt{\smash[b]{\vphantom{\tfrac12}\smash[t]{235-94\sqrt{5}}}},\end{aligned}$$ and all the radicals represent positive roots. The evaluation in Eq. \[eq:ex\_1848\] paraphrases the last entry in Weber’s list [@WeberVol3 Tabelle VI].
An interesting consequence of the explicit evaluation for $ G_2^{\mathfrak H/\overline{\varGamma}_0(4)}(z)$ (Eq. \[eq:G2\_Hecke4\_z\_rn\_eval\]) is the following logarithmic addition formula$$\begin{aligned}
G_2^{\mathfrak H/\overline{\varGamma}_0(4)}\left( \frac{z+1}{2} ,\frac{z}{2}\right)=G_2^{\mathfrak H/\overline{\varGamma}_0(4)}\left( -\frac{1}{2(2z+1)} ,z+\frac{1}{2}\right)-\frac{1}{3}\log\frac{|1-\lambda(2z)|^{2}}{4|\lambda(2z)|},\quad \alpha_2(z)\in\mathbb C\label{eq:G2_Hecke4_log_add_1}\end{aligned}$$ connecting two (un-renormalized) weight-4 automorphic Green’s functions on $ \overline\varGamma_0(4)$. To prove Eq. \[eq:G2\_Hecke4\_log\_add\_1\], we first reformulate the addition formulae in Eqs. \[eq:Hecke2\_Gamma2\_add\] and \[eq:G\_s\_Theta\_add\_form\] as$$\begin{aligned}
G_2^{\mathfrak H/\overline{\varGamma}_0(2)}(z,z')={}&G_2^{\mathfrak H/\overline{\varGamma}_0(4)}\left( \frac{z}{2} ,\frac{z'}{2}\right)+G_2^{\mathfrak H/\overline{\varGamma}_0(4)}\left( \frac{z+1}{2} ,\frac{z'}{2}\right),\tag{\ref{eq:Hecke2_Gamma2_add}$'$}\\G_2^{\mathfrak H/\overline{\varGamma}_0(2)}(z,z')={}&G_2^{\mathfrak H/\overline{\varGamma}_0(4)}\left( z+\frac{1}{2} ,z'+\frac{1}{2}\right)+G_2^{\mathfrak H/\overline{\varGamma}_0(4)}\left( -\frac{1}{2(2z+1)} ,z'+\frac{1}{2}\right);\tag{\ref{eq:G_s_Theta_add_form}$'$}\label{eq:G_s_Theta_add_form'}\end{aligned}$$ then, the $ z'\to z$ limit brings us to $$\begin{aligned}
G_2^{\mathfrak H/\overline{\varGamma}_0(2)}(z)={}&G_2^{\mathfrak H/\overline{\varGamma}_0(4)}\left( \frac{z}{2} \right)+G_2^{\mathfrak H/\overline{\varGamma}_0(4)}\left( \frac{z+1}{2} ,\frac{z}{2}\right)-2\log2+\frac{1}{3}\log\frac{2^4|1-\lambda(z)|^{2}}{|\lambda(z)|}\notag\\={}&G_2^{\mathfrak H/\overline{\varGamma}_0(4)}\left( \frac{z+1}{2} ,\frac{z}{2}\right)-2\log2,\label{eq:G2_Hecke2_rn_add_1}\\G_2^{\mathfrak H/\overline{\varGamma}_0(2)}(z)={}&G_2^{\mathfrak H/\overline{\varGamma}_0(4)}\left( z+\frac{1}{2} \right)+G_2^{\mathfrak H/\overline{\varGamma}_0(4)}\left( -\frac{1}{2(2z+1)} ,z+\frac{1}{2}\right)-\log|1-\lambda(2z)|\notag\\={}&G_2^{\mathfrak H/\overline{\varGamma}_0(4)}\left( -\frac{1}{2(2z+1)} ,z+\frac{1}{2}\right)-\frac{1}{3}\log\frac{2^4|1-\lambda(2z)|^{2}}{|\lambda(2z)|},\label{eq:G2_Hecke2_rn_add_2}\end{aligned}$$so long as $ \alpha_2(z)\in\mathbb C$. Eliminating $ G_2^{\mathfrak H/\overline{\varGamma}_0(2)}(z)$ from Eqs. \[eq:G2\_Hecke2\_rn\_add\_1\] and \[eq:G2\_Hecke2\_rn\_add\_2\], we arrive at Eq. \[eq:G2\_Hecke4\_log\_add\_1\].
\[app:algebraicity\]Algebraicity of Some Classical Modular Functions
====================================================================
In Eq. \[eq:fns\_alg\_val\_at\_CM\_pts\], we supplied a brief list of special functions that assume solvable algebraic values at CM points. We also mentioned that these algebraic numbers could generate abelian extensions of imaginary quadratic fields. These facts are well known to experts. For the sake of completeness, we outline the arithmetic theory that supports all these classical statements.
When $ z$ is a CM point, the algebraicity of the $ j$-invariant (*i.e. *$ j(z)\in\overline{\mathbb Q}$) follows from isomorphism classes of an elliptic curve $ E_z(\mathbb C)\cong \mathbb C/(2\pi i\mathbb Z+2\pi i z\mathbb Z)$ [@Shimura1994 Theorem 4.14]. Two CM points $ z_1$ and $ z_2$ generate the same field extension $ \mathbb Q(z_1,j(z_1))=\mathbb Q(z_2,j(z_2))$ if the discriminants of their respective minimal polynomials coincide: $ z_1\in\mathfrak Z_{D_1}=\mathfrak Z_{D_2}\ni z_2$. (Here, $ \mathfrak Z_D:=\{z\in\mathfrak H|\exists a,b,c \in\mathbb Z,a>0,\gcd(a,b,c)=1,b^2-4ac=D,az^2+bz+c=0\}$, as in Remark \[rmk:GKZ\_class\_number\_one\].) Via isomorphism to a (necessarily commutative) class group of the ring $ \mathbb Z[(D+i\sqrt{|D|})/2]$, one can show that the (finite) Galois group $ \operatorname{Gal}(\mathbb Q(z,j(z))/\mathbb Q(z)),z\in\mathfrak Z_D$ is abelian [@Shimura1994 Theorem 5.7(i)], and consequently solvable.
For any given integers $ a,b,c\in\mathbb Z$ satisfying $ ac>0$, we have two claims: (i) The expression$$\begin{aligned}
\frac{\Delta\left( \frac{az+b}{c} \right)}{\Delta(z)}\label{eq:Delta_abc_ratio}\end{aligned}$$can be locally identified with an algebraic function $A_{a,b,c} (j(z))$; (ii) For all CM points $z$, both the function $A_{a,b,c} (j(z))$ and its logarithmic derivative $$\begin{aligned}
\frac{A_{a,b,c} '(j(z))}{A\vphantom{'}_{a,b,c} (j(z))},\quad\text{where }A'_{a,b,c}(j):=\frac{\partial A\vphantom{'}_{a,b,c}(j)}{\partial j}\label{eq:Aabc_deriv}\end{aligned}$$assume algebraic values solvable in radical form. Based on a factorization argument [@LangGTM112 p. 167], it would suffice to demonstrate the truthfulness of these two claims for the cases where $ (a,b,c)=(p,0,1)$ and $(a,b,c)=(1,m,p),m\in\mathbb Z\cap[0,p-1] $, with $p$ being a prime number. Concretely speaking, the expression in Eq. \[eq:Delta\_abc\_ratio\] factorizes into a finite product of functions in the forms of $$\begin{aligned}
\frac{\Delta(p z')}{\Delta(z')}\in\mathbb Q(j(z'),j(pz')),\quad \text{and}\quad \frac{\Delta\left( \frac{z'+m}{p} \right)}{\Delta(z')}=\frac{\Delta\left( \frac{z'+m}{p} \right)}{\Delta\left(p \frac{z'+m}{p} \right)}\in\mathbb Q\left( j\left( \frac{z'+m}{p} \right) ,j(z')\right),\;m\in\mathbb Z\cap[0,p-1],\label{eq:appl_Shimura}\end{aligned}$$ with $ z'=(a'z+b')/(c'z+d')$ for some integers $ a',b',c',d'\in\mathbb Z$ (dependent on $a$, $b$, $c$, $p$ and $m$, but not on $z$) such that $ a'd'-b'c'>0$. Here, all the members in the field extension $ \mathbb Q(X,Y)$ are rational functions of variables $X$ and $Y$ with rational coefficients (dependent on $p$ but not on $z'$ or $m$); the two $ j$-invariants $ j(z)$ and $ j(pz)$ are tied to each other via an algebraic equation $ \Phi_p(j(z),j(pz))=0$ — the modular equation of $ p$-th degree (which is a bivariate polynomial equation with integer coefficients that depend on $p$ but not on $z$). The field extension relations in Eq. \[eq:appl\_Shimura\] are well-known results in the theory of automorphic functions (see [@Shimura1994 Proposition 6.9(2)] and [@LangGTM112 Chap. 11, §2, Corollary 2]). Thus far, we have constructed the algebraic function $ A_{a,b,c}(j)$ along with its logarithmic derivative $ A'_{a,b,c}(j)/A\vphantom{'}_{a,b,c}(j)$, and have demonstrated the algebraicity of their CM values. This confirms the two claims at the beginning of the current paragraph.
In the light of the foregoing arguments, we know that for any integer $N$, $$\begin{aligned}
\frac{\Delta(Nz)}{\Delta(z)}\quad \text{and}\quad\left[\frac{1}{\partial j(z)/\partial z}\frac{\partial}{\partial z}\log\frac{\Delta(Nz)}{\Delta(z)}\right]^2=\frac{[NE_{2}(Nz)-E_2(z)]^{2}}{j(z)[j(z)-1728]E_{4}(z)}
=\frac{\Delta(z)[NE_{2}(Nz)-E_2(z)]^{2}}{j(z)E_{4}(z)[E_{6}(z)]^2}\label{eq:Delta_ratios_deriv}\end{aligned}$$ assume solvable algebraic values (possibly infinity, as one has $ j(e^{\pi i/3})=E_4(e^{\pi i/3})=E_6(i)=0$) at CM points. These algebraic numbers (if finite) are expressible as multivariate rational functions (of rational coefficients) for certain CM values of the $j$-invariant: they belong to a field extension $ \mathbb Q(j(z_1),\dots,j(z_n))$ where all the $n$ numbers $z_1,\dots,z_n$ reside in the imaginary quadratic field $ \mathbb Q(z)$. In fact, any such algebraic number $ \alpha_z\in \mathbb Q(j(z_1),\dots,j(z_n))$ generates an abelian extension of the field $K=\mathbb Q(z)$, whose ring of algebraic integers is $ \mathfrak o_{\raisebox{-0em}{$_K$}}=\mathbb Z[z_{\raisebox{-0em}{$_K$}}]$ for a certain $ z_{\raisebox{-0em}{$_K$}}\in K$. To show this, we may suppose that $ z_\ell\in\mathfrak Z_{D_\ell},\ell\in\mathbb{Z}\cap[1,n]$ and pick positive integers $ f_\ell\in\mathbb Z_{>0}$ such that $\mathfrak o_\ell=\mathbb Z[ z_{\raisebox{-0em}{$_K$}}f_\ell]=\mathbb Z[(D_\ell+i\sqrt{\smash[b]{|D_\ell|}})/2] $, with $ f_\ell$ being the conductor of $ \mathfrak o_\ell$, $ \ell\in\mathbb{Z}\cap[1,n]$ [@LangGTM112 p. 91]. Take a ring $\mathfrak o =\mathbb Z[z_{\raisebox{-0em}{$_K$}}f]$ where $ f=\operatorname{lcm}(f_1,\dots,f_n)$, then we have $ \mathfrak o\subseteq\bigcap_{\ell=1}^n \mathfrak o_\ell$, which entails the field inclusion relations $ \mathbb Q(z_{\raisebox{-0em}{$_K$}},j(z_{\raisebox{-0em}{$_K$}}f))\supseteq\mathbb Q(z_\ell,j(z_\ell))$, $ \ell\in\mathbb{Z}\cap[1,n]$ [@LangGTM112 p. 134]. Thus, the tower of fields $K\subseteq K(\alpha_z)\subseteq K(j(z_{\raisebox{-0em}{$_K$}}f)) $ reveals the commutativity of the Galois group $ \operatorname{Gal}(K(\alpha_z)/K)\cong\operatorname{Gal}(K(j(z_{\raisebox{-0em}{$_K$}}f))/K)/\operatorname{Gal}(K(j(z_{\raisebox{-0em}{$_K$}}f))/K(\alpha_z))$.
Then, we show that the same algebraic property holds if one replaces $ NE_{2}(Nz)-E_2(z)$ by $ E_2(z)$ in the last expression of Eq. \[eq:Delta\_ratios\_deriv\]. Suppose that a CM point $ z_*\in\mathfrak H$ solves the quadratic equation $ az_*^2+bz_*+c=0$ with integer coefficients $ (a,b,c)$. Then it is straightforward to compute that $$\begin{aligned}
&
\left.\frac{\partial}{\partial z}\right|_{z=z_*}\log\frac{\Delta\left( \frac{az+b}{c} \right)}{\Delta(z)}+\left.\frac{1}{z_*^2}\frac{\partial}{\partial z}\right|_{z=-1/z_*}\log\frac{\Delta\left( \frac{cz-b}{a} \right)}{\Delta(z)}\notag\\={}&\frac{b^2-4ac}{ac}\left[\frac{\Delta'(z_{*})}{\Delta(z_{*})}-\frac{6i}{\operatorname{Im}z_*}\right]=2\pi i\frac{b^2-4ac}{ac}E_2(z_*),\end{aligned}$$so the assertion follows from the properties of the logarithmic derivative that appeared in Eq. \[eq:Aabc\_deriv\].
Thus, we have confirmed that the following arithmetic functions$$\begin{aligned}
j(z),\quad &\frac{E_2(Nz)}{E_2(z)},\quad \left[ \frac{E_4(Nz)}{E_4(z)} \right]^3=\frac{\Delta(Nz)j(Nz)}{\Delta(z)j(z)},\notag\\ \left[ \frac{E_6(Nz)}{E_6(z)} \right]^2={}&\frac{\Delta(Nz)[j(Nz)-1728]}{\Delta(z)[j(z)-1728]},\quad \frac{[E_2(z)]^2}{E_4(z)},\quad \frac{[E_2(z)]^3}{E_6(z)}\end{aligned}$$all possess the desired algebraic attributes: when $ [\mathbb Q(z):\mathbb Q]=2$, a number from the list above is either infinity or an algebraic number that generates an abelian extension of $ \mathbb Q(z)$.
Solutions to $ \dim\mathcal S_k(\varGamma)=0$ on Congruence Subgroups $ \varGamma$\[app:dimSk\_vanish\]
=======================================================================================================
For completeness, we present in this appendix an expository note on the solutions to the cusp-form-free condition $ \dim\mathcal S_k(\varGamma)=0$, where $ \varGamma$ is a congruence subgroup of $ SL(2,\mathbb Z)$ that has arithmetic interest, and $ k$ is an even number greater than or equal to $4$.
For $ \varGamma=SL(2,\mathbb Z)$, the dimension formulae for even weights $ k\geq4$ are familiar (see [@Shimura1994 Proposition 2.26] or [@DiamondShurman Theorem 3.5.2]):$$\begin{aligned}
\dim\mathcal S_k(SL(2,\mathbb Z))=\begin{cases}\left\lfloor\dfrac{k}{12}\right\rfloor-1, & k\equiv2\pmod {12}; \\[8pt]
\left\lfloor\dfrac{k}{12}\right\rfloor, & k\not\equiv2\pmod {12}. \\
\end{cases}\label{eq:dimSk_SL2Z}\end{aligned}$$ Here, $ \lfloor x\rfloor$ stands for the greatest integer less than or equal to $ x\in\mathbb R$. It is clear from Eq. \[eq:dimSk\_SL2Z\] that for even weights $ k\geq4$, the only solutions to $ \dim\mathcal S_k(SL(2,\mathbb Z))=0$ are $ k=4,6,8,10,14$.
If we shift our attention to $\varGamma(N)$, the principal congruence group of level $N\geq2$, which consists of transformations in $ SL(2,\mathbb Z)$ that are congruent to the identity matrix $ \left(\begin{smallmatrix}1&0\\0&1\end{smallmatrix}\right)$ modulo an integer $ N\geq2$, then we have the following dimension formula for even weights $ k\geq4$ (as one may compute from [@Shimura1994 Eqs. 1.6.2–1.6.3, Proposition 1.40, Theorem 2.24] or [@DiamondShurman Theorems 3.1.1, 3.5.1, Figure 3.4]):$$\begin{aligned}
\dim\mathcal S_k(\varGamma(N))=\begin{cases}\dfrac{k-4}{2}, & N=2 \\[8pt]
\dfrac{[(k-1)N-6]N^{2}}{24}{ \mathop{ \raisebox
{-1.0\depthofsumsign+1\depthofsumsign}
{\scalebox
{1.0}
{$\displaystyle\prod$} }
}
}\limits_{p\mid N}\left( 1-\dfrac{1}{p^{2}} \right), & N>2 \\
\end{cases}\end{aligned}$$ and $ \dim\mathcal S_k(\varGamma(N))=0$ admits just one solution: $ k=4,N=2$.[^8]
The compact Riemann surfaces $ X_0(N)(\mathbb C)=\varGamma_0(N)\backslash\mathfrak H^*$ and $ X_1(N)(\mathbb C)=\varGamma_1(N)\backslash\mathfrak H^*$ are important to the arithmetic studies of elliptic curves. Here, the Hecke congruence groups of level $N$ are defined by $$\begin{aligned}
\varGamma_0(N):=\left\{ \left.\begin{pmatrix}a & b \\
Nc & d \\
\end{pmatrix}\right|a,b,c,d\in\mathbb Z;ad-Nbc=1 \right\},\end{aligned}$$and their projective counterparts are $\overline {\varGamma}_0(N)=\varGamma_0(N)/\{I,-I\} $; for the congruence groups $$\begin{aligned}
\varGamma_1(N):=\left\{ \left.\begin{pmatrix}1+Na & b \\
Nc & 1+Nd \\
\end{pmatrix}\right|a,b,c,d\in\mathbb Z;(1+Na)(1+Nd)-Nbc=1 \right\},\end{aligned}$$the respective projective versions are $ \overline {\varGamma}_1(2)={\varGamma}_1(2)/\{I,-I\}$ and $ \overline {\varGamma}_1(N)={\varGamma}_1(N)$ for $ N>2$. For $ N=2,$ $3$ or $4$, it is well known that the projective congruence group $ \overline{\varGamma}_1(N)$ coincides with the projective Hecke congruence group $\overline{ \varGamma}_0(N)$ of the same level $ N$ [@DiamondShurman Exercise 3.9.3].
In this work, we are concerned with weight-4 automorphic Green’s functions $ G_2^{\mathfrak H/\overline{\varGamma}_0(N)}(z,z')=G_2^{\mathfrak H/\overline{\varGamma}_1(N)}(z,z')$ for $ N\in\{2,3,4\}$, as well as a weight-6 automorphic Green’s function $ G_3^{\mathfrak H/\overline{\varGamma}_0(2)}(z,z')=G_3^{\mathfrak H/\overline{\varGamma}_1(2)}(z,z')$. These cases account for all the “cusp-form-free” scenarios relevant to the Gross–Kohnen–Zagier algebraicity conjecture on congruence subgroups of levels $N\geq2$, as explained in the next lemma.
1. For even weights $ k\geq4$, there are only four isolated cases free from cusp forms on $ \varGamma_0(N)$ (see [@Miyake1989J1 Table A]):$$\begin{aligned}
\dim\mathcal S_4(\varGamma_0(2))=0,\quad \dim\mathcal S_4(\varGamma_0(3))=0,\quad \dim\mathcal S_4(\varGamma_0(4))=0,\quad \dim\mathcal S_6(\varGamma_0(2))=0.\label{eq:Gamma_0_cusp_dim}\end{aligned}$$
2. The solutions to the equation $ \dim{\mathcal S_k(\varGamma_1(N))}=0$ for even weights $k\geq4 $ are exhausted by$$\begin{aligned}
\dim\mathcal S_4(\varGamma_1(2))=0,\quad \dim\mathcal S_4(\varGamma_1(3))=0,\quad \dim\mathcal S_4(\varGamma_1(4))=0, \quad\dim\mathcal S_6(\varGamma_1(2))=0.\label{eq:Gamma_1_cusp_dim}\end{aligned}$$
<!-- -->
1. For even weights $ k\geq4$, we may compute $ \dim\mathcal S_k(\varGamma_0(N))$ explicitly, drawing on some classical results.
Let $ \mathfrak H^*:=\mathfrak H\cup\mathbb Q\cup\{i\infty\}$ be the extended upper half-plane, then the genus of the compact Riemann surface $ X_{0}(N)(\mathbb C):=\varGamma_0(N)\backslash\mathfrak H^*$ is given by (see [@Shimura1994 Proposition 1.40] or [@DiamondShurman Theorem 3.1.1])$$\begin{aligned}
g(\varGamma_0(N)\backslash\mathfrak H^*)=1+\frac{N}{12}\prod_{p\mid N}\left( 1+\frac{1}{p} \right)-\frac{\nu_2(\varGamma_0(N))}{4}-\frac{\nu_3(\varGamma_0(N))}{3}-\frac{\nu_\infty(\varGamma_0(N))}{2}.\end{aligned}$$Here, $ \nu_2(\varGamma_0(N))$ (resp. $ \nu_3(\varGamma_0(N))$) counts the number of inequivalent elliptic fixed points of period $2$ (resp. $3$), explicitly quantified by the formulae (see [@Shimura1994 Proposition 1.43] or [@DiamondShurman Figure 3.3]):$$\begin{aligned}
\nu_{2}(\varGamma_0(N))=\begin{cases}{ \mathop{ \raisebox
{-1.0\depthofsumsign+1\depthofsumsign}
{\scalebox
{1.0}
{$\displaystyle\prod$} }
}
}\limits_{p\mid N}\left( 1+\left( \dfrac{-1}{p} \right) \right), & 4\nmid N, \\
0, & 4\mid N, \\
\end{cases}\qquad \qquad\nu_3(\varGamma_0(N))=\begin{cases}{ \mathop{ \raisebox
{-1.0\depthofsumsign+1\depthofsumsign}
{\scalebox
{1.0}
{$\displaystyle\prod$} }
}
}\limits_{p\mid N}\left( 1+\left( \dfrac{-3}{p} \right) \right), & 9\nmid N, \\
0, & 9\mid N, \\
\end{cases}\label{eq:nu2_nu3}\end{aligned}$$ where the Kronecker–Legendre symbols are evaluated as [@Shimura1994 Proposition 1.43]$$\begin{aligned}
\left( \frac{-1}{p} \right)=\begin{cases}0, & p=2, \\
1, & p\equiv1\bmod4, \\
-1, & p\equiv3\bmod4, \\
\end{cases}\qquad\qquad\left( \frac{-3}{p} \right)=\begin{cases}0, & p=3, \\
1, & p\equiv1\bmod3, \\
-1, & p\equiv2\bmod3. \\
\end{cases}\end{aligned}$$In the meantime, $$\begin{aligned}
\nu_\infty(\varGamma_0(N))=\sum_{d\mid N}\varphi(\gcd(d,N/d))\label{eq:nu_inf}\end{aligned}$$ counts the number of inequivalent cusps in the fundamental domain of $ \varGamma_0(N)$, where the sum “$ \sum_{d\mid N}$” runs over every positive divisor $d$ of $N$, “$ \gcd(n,m)$” denotes the greatest common divisor for a pair of integers $ (n,m)$, and $ \varphi(n):=n\prod_{p|n}(1-\frac1p)$ is Euler’s totient function.
For an even weight $ k\geq4$, we may quote the dimension formula (see [@Shimura1994 Theorem 2.24] or [@DiamondShurman Theorem 3.5.1])$$\begin{aligned}
\dim\mathcal S_k(\varGamma_0(N))=(k-1)(g(\varGamma_0(N)\backslash\mathfrak H^*)-1)+\left\lfloor\frac{k}{4}\right\rfloor\nu_2(\varGamma_0(N))+\left\lfloor\frac{k}{3}\right\rfloor\nu_3(\varGamma_0(N))+\left( \frac{k}{2} -1\right)\nu_\infty(\varGamma_0(N)).\end{aligned}$$For $ N\in\{2,3,4\}$, this specializes to the formulae$$\begin{aligned}
\dim\mathcal S_k(\varGamma_0(2))=\left\lfloor\frac{k}{4}\right\rfloor-1,\quad\dim\mathcal S_k(\varGamma_0(3))=\left\lfloor\frac{k}{3}\right\rfloor-1, \quad \dim\mathcal S_k(\varGamma_0(4))=\frac{k-4}{2},\end{aligned}$$ which confirm the equalities displayed in Eq. \[eq:Gamma\_0\_cusp\_dim\].
To rule out solutions to the equation $\dim\mathcal S_k(\varGamma_0(N))=0 $ with an integer $N>4$ for even weights $ k\geq4$, we need to assess the lower bound of$$\begin{aligned}
\dim\mathcal S_k(\varGamma_0(N))={}&\frac{(k-1)N}{12}\prod_{p\mid N}\left( 1+\frac{1}{p} \right)+\left(\frac{1}{4}+\left\lfloor\frac{k}{4}\right\rfloor-\frac{k}{4}\right)\nu_2(\varGamma_0(N))\notag\\{}&+\left(\frac{1}{3}+\left\lfloor\frac{k}{3}\right\rfloor-\frac{k}{3}\right)\nu_3(\varGamma_0(N))-\frac{\nu_\infty(\varGamma_0(N))}{2},\end{aligned}$$starting from the following estimates of G. Martin [@Martin2005]:$$\begin{aligned}
\max\{\nu_2(\varGamma_0(N)),\nu_3(\varGamma_0(N))\}\leq2^{4-\frac{\log16}{\log11}}N^{\frac{\log2}{\log11}},\quad \nu_\infty(\varGamma_0(N))\leq\sqrt{N}\prod_{p\mid N}\left( 1+\frac{1}{p} \right).\end{aligned}$$
Now that we have the lower bound$$\begin{aligned}
\left(\frac{1}{4}+\left\lfloor\frac{k}{4}\right\rfloor-\frac{k}{4}\right)+\left(\frac{1}{3}+\left\lfloor\frac{k}{3}\right\rfloor-\frac{k}{3}\right)\geq-\frac{7}{12}\end{aligned}$$valid for even integers $k$, we obtain the inequality $$\begin{aligned}
\dim\mathcal S_k(\varGamma_0(N))\geq\frac{(k-1)N-6\sqrt{N}}{12}\prod_{p\mid N}\left( 1+\frac{1}{p} \right)-\frac{7}{12}2^{4-\frac{\log16}{\log11}}N^{\frac{\log2}{\log11}}>\frac{3N-6\sqrt{N}}{12}-\frac{7}{12}2^{4-\frac{\log16}{\log11}}N^{\frac{\log2}{\log11}}>0\end{aligned}$$for $ N\geq77$.
Since $ 2\times3\times 5\times7=210$, any integer in the range $ 5\leq N\leq 76$ contains at most 3 distinct prime factors. Thus, within the range $ 30\leq N\leq 76$, a more stringent upper bound$$\begin{aligned}
\max\{\nu_2(\varGamma_0(N)),\nu_3(\varGamma_0(N))\}\leq2^3=8\end{aligned}$$allows us to refine the former inequality on dimensions as$$\begin{aligned}
\dim\mathcal S_k(\varGamma_0(N))>\frac{3N-6\sqrt{N}}{12}-\frac{7}{12}\times8>0.\end{aligned}$$
Then, for the regime $ 12\leq N\leq 29$, the improvement$$\begin{aligned}
\max\{\nu_2(\varGamma_0(N)),\nu_3(\varGamma_0(N))\}\leq2^{\sum_{p\mid N,p\geq5} 1}=\prod_{p\mid N,p\geq5}2\leq2\end{aligned}$$ leads to$$\begin{aligned}
\dim\mathcal S_k(\varGamma_0(N))>\frac{3N-6\sqrt{N}}{12}-\frac{7}{12}\times2>0.\end{aligned}$$
The remaining scenarios $ N\in\{5,6,7,8,9,10,11\}$ can be studied case by case. If $N=6$, $8$ or $9 $, then we have $\nu_2(\varGamma_0(N))=\nu_3(\varGamma_0(N))=0 $ and $ \nu_\infty(\varGamma_0(N))=4$, thus$$\begin{aligned}
\dim\mathcal S_k(\varGamma_0(N))=\frac{(k-1)N}{12}\prod_{p\mid N}\left( 1+\frac{1}{p} \right)-2\geq\dim\mathcal S_4(\varGamma_0(N))=\frac{N}{4}\prod_{p\mid N}\left( 1+\frac{1}{p} \right)-2=1.\end{aligned}$$ The explicit formulae $$\begin{aligned}
\dim\mathcal S_k(\varGamma_0(5))={}&2\left\lfloor \frac{k}{4} \right\rfloor-1,&\dim\mathcal S_k(\varGamma_0(7))={}&2\left\lfloor \frac{k}{3} \right\rfloor-1, \notag\\\dim\mathcal S_k(\varGamma_0(10))={}&2\left\lfloor\frac{k}{4}\right\rfloor+k-3,&\dim\mathcal S_k(\varGamma_0(11))={}&k-2\end{aligned}$$rule out the solutions to $ \dim\mathcal S _k(\varGamma_0(N))=0$ for even weights $ k\geq4$ and $ N\in\{5,7,10,11\}$.
2. When it comes to the numbers of inequivalent elliptic fixed points $ \nu_2,\nu_3$, the number of inequivalent cusps $ \nu_\infty$, and the genus $g$, the data for $ \varGamma_1(N)$ coincide with those of $ \varGamma_0(N)=\{I,-I\}\varGamma_1(N)$ for $N\in\{2,3,4\}$ (see [@DiamondShurman Figure 3.3]), so the relations in Eq. \[eq:Gamma\_1\_cusp\_dim\] follow from Eq. \[eq:Gamma\_0\_cusp\_dim\].
It is sensible to restrict the rest of our discussions to $ \varGamma_1(N),N>4$, where we have [@DiamondShurman Figure 3.4]$$\begin{aligned}
\nu_{2}(\varGamma_1(N))=0,\quad \nu_3(\varGamma_1(N))=0,\quad \nu_\infty(\varGamma_1(N))=\frac{1}{2}\sum_{d\mid N}\varphi(d)\varphi(N/d)\end{aligned}$$and$$\begin{aligned}
g(\varGamma_1(N)\backslash\mathfrak H^*)=1+\frac{N^{2}}{24}\prod_{p\mid N}\left( 1-\frac{1}{p^{2}} \right)-\frac{1}{4}\sum_{d\mid N}\varphi(d)\varphi(N/d).\end{aligned}$$Consequently, for even weights $ k\geq4$, we obtain [@DiamondShurman Figure 3.4]$$\begin{aligned}
\dim\mathcal S_k(\varGamma_1(N))={}&(k-1)(g(\varGamma_1(N)\backslash\mathfrak H^*)-1)+\left( \frac{k}{2} -1\right)\nu_\infty(\varGamma_1(N))\notag\\={}&\frac{(k-1)N^{2}}{24}\prod_{p\mid N}\left( 1-\frac{1}{p^{2}} \right)-\frac{1}{4}\sum_{d\mid N}\varphi(d)\varphi(N/d).\end{aligned}$$If $ N=p$ turns out to be a prime number greater than or equal to $5$, we immediately have $$\begin{aligned}
\dim\mathcal S_k(\varGamma_1(p))\geq\dim\mathcal S_4(\varGamma_1(p))=\frac{3(p^{2}-1)}{24}-\frac{p-1}{2}=\frac{(p-1)(p-3)}{8}>0.\end{aligned}$$For composite numbers $N$, we may use the inequalities$$\begin{aligned}
\prod_{p\mid N}\left( 1-\frac{1}{p^{2}} \right)>\prod_p\left( 1-\frac{1}{p^{2}} \right)=\frac{6}{\pi^2}\end{aligned}$$and $$\begin{aligned}
\dim\mathcal S_k(\varGamma_1(N))\geq\dim\mathcal S_4(\varGamma_1(N))>\frac{3N^{2}}{4\pi^{2}}-\frac{1}{4}\sum_{d\mid N}\varphi(d)\varphi(N/d).\end{aligned}$$Here, by virtue of the relations $ \varphi(n)\leq n$ and $ n=\sum_{d\mid n}\varphi(d)$, we may estimate$$\begin{aligned}
\sum_{d\mid N}\varphi(d)\varphi(N/d)={}&\sum_{\substack{d\mid N\\0<d\leq\sqrt{N}}}\varphi(d)\varphi(N/d)+\sum_{\substack{d\mid N\\\sqrt{N}< d\leq N}}\varphi(d)\varphi(N/d)\notag\\\leq{}&\left(\max_{n\in\mathbb Z\cap(0,\sqrt{N}]}\varphi(n)\right)\sum_{\substack{d\mid N\\0<d\leq\sqrt{N}}}\varphi(N/d)+\left(\max_{n\in\mathbb Z\cap(0,\sqrt{N}]}\varphi(n)\right)\sum_{\substack{d\mid N\\\sqrt{N}< d\leq N}}\varphi(d)\leq 2N^{3/2}.\end{aligned}$$ For $ N\geq44$, we thus have$$\begin{aligned}
\dim\mathcal S_k(\varGamma_1(N))\geq\dim\mathcal S_4(\varGamma_1(N))>\frac{3N^{2}}{4\pi^{2}}-\frac{1}{4}\sum_{d\mid N}\varphi(d)\varphi(N/d)\geq\frac{3N^{2}}{4\pi^{2}}- \frac{N^{3/2}}{2}>0.\end{aligned}$$For integers in the range $ 27\leq N<44$, we may exploit the relation $ \max_{n\in\mathbb Z\cap(0,\sqrt{44}]}\varphi(n)=\varphi(5)=4$ to improve the estimate:$$\begin{aligned}
\sum_{d\mid N}\varphi(d)\varphi(N/d)\leq 8N,\quad \dim\mathcal S_k(\varGamma_1(N))\geq\dim\mathcal S_4(\varGamma_1(N))>\frac{3N^{2}}{4\pi^{2}}-2N>0.\end{aligned}$$By direct computation, we can verify that the inequality$$\begin{aligned}
\dim\mathcal S_k(\varGamma_1(N))\geq\dim\mathcal S_4(\varGamma_1(N))>\frac{3N^{2}}{4\pi^{2}}-\frac{1}{4}\sum_{d\mid N}\varphi(d)\varphi(N/d)>0\end{aligned}$$remains valid for $ N=25,\sum_{d\mid 25}\varphi(d)\varphi(25/d)=56$ and $ N=26,\sum_{d\mid 26}\varphi(d)\varphi(26/d)=48$. Next, within the regime $ 14\leq N\leq 24$, we have $ \max_{n\in\mathbb Z\cap(0,\sqrt{24}]}\varphi(n)=\varphi(3)=\varphi(4)=2$, and accordingly, $$\begin{aligned}
\sum_{d\mid N}\varphi(d)\varphi(N/d)\leq 4N,\quad \dim\mathcal S_k(\varGamma_1(N))\geq\dim\mathcal S_4(\varGamma_1(N))>\frac{3N^{2}}{4\pi^{2}}-N>0.\end{aligned}$$There are only a few composite numbers within the range $ 5\leq N\leq 13$, for which we may directly compute as follows:$$\begin{aligned}
\dim\mathcal S_4&(\varGamma_1(6))=1,\quad \dim\mathcal S_4(\varGamma_1(8))=3,\quad \dim\mathcal S_4(\varGamma_1(9))=5,\notag\\& \dim\mathcal S_4(\varGamma_1(10))=5,\quad \dim\mathcal S_4(\varGamma_1(12))=7.\end{aligned}$$Therefore, for all integers $ N\geq5$ and even weights $ k\geq4$, we have $ \dim\mathcal S_k(\varGamma_1(N))\geq\dim\mathcal S_4(\varGamma_1(N))>0$.
[^1]: In the current work, we use the notation $A\smallsetminus B $ with a “lowercase” backslash ($ \smallsetminus$) to indicate set minus operation, while writing the “uppercase” backslash ($\backslash $) for orbit spaces such as $ SL(2,\mathbb Z)\backslash\mathfrak H$.
[^2]: Hereafter, the abbreviation “a.e.” that precedes a complex variable refers to “almost every” complex number in question. An equality that holds for “almost every” complex number in a specified domain is allowed to fail in a subset of measure zero.
[^3]: The identities listed in Eq. \[eq:alpha\_N\_ratio\_ids\] are classical. They can be traced back to Ramanujan’s alternative-base representations for the Weierstra[ß]{} discriminant $ \Delta(z)=[\eta(z)]^{24}$ [@RN5 Chap. 33, Corollary 3.4, Theorem 9.10 and Theorem 11.6] and weight-2 Eisenstein series [@RN5 Chap. 33, Theorem 9.11], [@RN5 Chap. 33, Corollary 2.11], [@RN3 Chap. 17, Entries 13(viii)–(ix)]. For an arithmetic perspective on such identities and their generalizations, see [@Maier2009].
[^4]: Exploiting the symmetry of the integral in question, one may assume that the points $ z$ and $z'$ are both in the closure of the fundamental domain $ \mathfrak D_N\cup\partial\mathfrak D_N$. For sufficiently large $ y=\operatorname{Im}z$, one can then designate the integration paths in the $ \zeta$-plane as follows: the path joining $ z'$ to $ z-\frac i3$ is a straight line segment; the path connecting $ z-\frac i3$ to $ z+\frac i3$ is a semi-circle with radius $ 1/3$; the path extending from $ z+\frac i3$ to $ i\infty$ runs parallel to the $ \operatorname{Im}\zeta$-axis. It is ready to check that $ \alpha_N(z)\neq \alpha_N(\zeta)$ along these integration paths.
[^5]: In our notation, the upright $ \Gamma$ is reserved for the Euler integral of the second kind $ \Gamma(\xi):=\int_0^\infty t^{\xi-1}e^{-t}\operatorname{d}t,\xi>0$ and its analytic continuation, in contrast to the congruence subgroup $ \varGamma$ set in slanted typeface.
[^6]: Hereafter, an integration “$ \int_{a+i0^+}^{b+i0^+}$” (resp. “$ \int_{a-i0^+}^{b-i0^+}$”) for $ a,b\in\mathbb R\cup\{-\infty,+\infty\}$ is carried along a path parallel to the real axis, whose imaginary part is a positive (resp. negative) infinitesimal.
[^7]: We have implemented in *Mathematica* all the Kontsevich–Zagier integral representations of automorphic Green’s functions declared in Theorem \[thm:KZ\_int\_repns\]. The source code (NumSuppAGF1.nb) for the numerical implementation can be downloaded from <http://arxiv.org/format/1312.6352>.
[^8]: The product “$ \prod_{p\mid N}$” runs through distinct prime numbers $ p$ that divide $ N$, so the result never vanishes. The equation $ (k-1)N-6=0$ has no integer solutions such that $N>2,k\geq4$.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We derive the spectrum and thermodynamics of the 1D supersymmetric t-J model with long range hopping and spin exchange using a set of maximal spin eigenstates. This spectrum confirms the recent conjecture that the asymptotic Bethe-ansatz spectrum is exact. By empirically determining the spinon degeneracies of each state, we are able to explicitly construct the free energy.'
author:
- 'D. F. Wang'
- 'James T. Liu'
- 'P. Coleman'
---
Spectrum and Thermodynamics of the 1D Supersymmetric t-J\
Model with $1/r^2$ Exchange and Hopping
Joseph Henry Laboratories of Physics, Princeton University,\
Princeton, New Jersey 08544
Institute of Field Physics, University of North Carolina,\
Chapel Hill, North Carolina 27599-3255
Serin Laboratories, Rutgers University,\
P.O. Box 849, Piscataway, New Jersey 08854
ß[[*\_h*]{}]{} The explicit construction of low dimensional models with Jastrow ground-state wavefunctions has attracted considerable recent interest[@laughlin; @girvin; @arovas; @mele; @hald88; @shas88]. In one dimension, Shastry and Haldane[@hald88; @shas88] have demonstrated that the ground-state of the 1D Heisenberg model with a $1/r^2$ exchange interaction is a Gutzwiller state for the half filled infinite-$U$ Hubbard model. Haldane has shown how the spectrum of this model can be written in terms of a generalized type of Jastrow wavefunction with excitations of novel statistics[@hald91].
Kuramoto and Yokoyama[@kura91] have recently extended these results to include holes, demonstrating that the corresponding 1D supersymmetric t-J model is also characterized by a Gutzwiller ground-state. Most recently, Kawakami has obtained an asymptotic Bethe-ansatz (ABA) solution for the model, based on the observation that the ground-state wavefunction is a product of two-body functions[@kawa91]. Assuming factorizability, he derived the spectrum of the system, which was conjectured to be exact. The low-energy critical behavior of the model has been identified as a Luttinger liquid[@yang91; @Kawakami92]; the spin and charge excitations are described independently by $c=1$ conformal field theories.
In the case of the $1/r^2$ Bose gas[@suth72], and the Shastry-Haldane $1/r^2$ Heisenberg chain[@hald91], the ABA has been shown to furnish the correct spectrum, despite the long-range nature of the interactions. A remarkable feature of these models is that excited states are obtained from the ground-state by introducing zeros into the Jastrow wavefunction, in a manner reminiscent of Laughlin’s description of quasiparticles in the fractional quantum Hall effect. This motivates us to examine the $1/r^2$ supersymmetric t-J model in a similar vein. Here, we show how this philosophy can be used to construct the excited state Jastrow wavefunctions of the $1/r^2$ supersymmetric t-J model and indeed, the spectrum confirms Kawakami’s conjecture. In addition to the spectrum, we are able to obtain the spin degeneracies of each state, permitting us to write the the free energy in closed form.
The Hamiltonian for the one-dimensional t-J model is given by $$H=\sum_{{i\ne j}, \sigma}\left[ -t_{ij} c_{i\sigma}^\dagger
c_{j\sigma}\right] +\sum_{i\ne j}\left[ J_{ij}({\bf S}_i\cdot{\bf S}_j
-{\textstyle{1\over4}}n_i n_j) \right],
\label{eq:hamil}$$ where we implicitly project out any double occupancies. We take $t_{ij}=J_{ij}=t/d^2(i-j)$ where $d(n)={N\over\pi}\sin(n\pi/N)$ is the chord distance consistent with periodic boundary conditions on $N$ lattice sites[@Rucken91].
States in the Hilbert space can be represented by spin and hole excitations from the fully-polarized up-spin state $|P\rangle$[@anderson89]. If we let $Q$ denote the number of holes and $M$ denote the number of down-spins, then $S_z$ is given by $S_z=(N-Q)/2-M$. The wavefunctions are given by $$|\psi\rangle=\sum_{x,y}\psi(x,y)
\prod_\alpha S^{-}_{x_\alpha} \prod_i h_{y_i}^\dagger
|P\rangle,
\label{eq:wave}$$ where the amplitude $\psi(x,y)$ is symmetric in $x \equiv (x_1,x_2,\ldots,x_M)$, the positions of the down-spins, and antisymmetric in $y \equiv ( y_1,y_2,\ldots,y_Q)$, the positions of the holes. $S^{-}_{x_{\alpha}}=c_{x_\alpha\down}^\dagger
c_{x_\alpha\up}^{\vphantom{\dagger}}$ is the spin-lowering operator at site $x_\alpha$ and $h_{y_i}^\dagger=c_{y_{i\up}}$ creates a hole at site $y_i$.
We can construct a general class of states corresponding to states of uniform motion and spin polarization. To describe these states, we generalize Kuramoto and Yokoyama’s Jastrow ground-state[@kura91] as follows $$\begin{aligned}
\psi_G(x,y;J_s,J_h)=&&\exp\left[{2\pi i\over N}
\left({J_s\sum_\alpha x_\alpha}+{J_h\sum_i y_i}\right)\right]{\bf
\Psi}_0(x,y)\nonumber\\
\quad {\bf \Psi}_0(x,y)=&&
\prod_{\alpha<\beta}d^2(x_\alpha-x_\beta)\prod_{i<j}d(y_i-y_j)
\prod_{\alpha,i}d(x_\alpha-y_i).
\label{eq:gutz}\end{aligned}$$ Here, $J_s$ and $J_h$ govern the (uniform) momenta of down-spins and holes respectively. $J_s$ and $J_h$ take on either integral or half-integral values as appropriate to insure that $\psi_G$ has the correct periodicities under $x_\alpha \to x_\alpha+N$ and $y_i \to y_i+N$.
The Hamiltonian can be broken up into four parts, $H=T^\up+T^\down+H^0+H^{\rm int}$, where $T^\up$ ($T^\down$) is the up (down) spin transfer operator, $H^0$ is the spin exchange operator and $H^{\rm int}$ is the diagonal interaction term. When $H$ acts on $\psi_G$, $T^\up$ only affects the $y$ variables and $H^0$ only affects the $x$ variables. As a result, these operators are easy to treat and yield only two and three body terms when appropriate conditions on $J_s$ and $J_h$ are met[@hald88; @shas88; @kura91].
However, because $T^\down$ exchanges pairs of $x_\alpha$ and $y_i$, this term must be treated differently. In general, it is [*not*]{} true that $T^\up|\psi_G\rangle
=T^\down|\psi_G\rangle$ because $T^{\up}$ does not commute with the spin raising operator. This difficulty was overlooked in earlier work[@kura91]. To deal with $T^\down$, we use an alternate representation for $|\psi_G\rangle$ in terms of up-spins and holes. Let us introduce the $N-M-Q$ coordinates $u \equiv
(u_1,u_2,\ldots,u_{N-M-Q})$ which give the location of the up-spins. Wavefunctions in this representation are given by the spin rotated version of Eq. (\[eq:wave\]) where the $x$ are replaced by $u$ and $M$ is replaced by $N-M-Q$. Making this transformation, we find $$\psi_G(x,y;J_s,J_h)=A\psi_G(u,y;N-J_s,J_h-J_s+{N\over2}),
\label{eq:dual}$$ where the set of $N$ coordinates $(x,y,u)$ exhausts the entire lattice. $A$ is a constant independent of the spin and hole coordinates. Using this identity, the down-spin transfer operator gives $${T^\down\psi_G(x,y)\over\psi_G(x,y)}=
{T^\down\psi_G(u,y)\over\psi_G(u,y)},$$ and can thus be treated in a similar manner as $T^\up$. The result gives two and three body terms in the variables $u$ and $y$. These terms can then be converted into sums over the $x$ and $y$ variables by making use of the fact that $(x,y,u)$ runs over the entire lattice.
When the separate terms that contribute to the Hamiltonian are combined, we find that the two body terms drop out and the three body terms combine to give constants. As a result, $\psi_G$ with total momentum $P={2\pi\over N}(J_s M + J_h Q)$ is an exact eigenstate of $H$ with energy $$\begin{aligned}
{N^2\over\pi^2t}E=&&{2\over3}M(M^2-1)-2MJ_s(N-J_s)\nonumber\\
&&+Q\biggl[{1\over3}(N^2-1)+{2\over3}(Q^2-1)+{1\over2}(M+Q)(2M-Q)\nonumber\\
&&\quad\quad-2J_h(N-J_h)+2(J_s-J_h)^2\biggr].
\label{eq:energy}\end{aligned}$$ The cancellation of the many body terms, and thus this result, is only valid under the conditions $|J_s-N/2|\le N/2-(M-1+Q/2)$, $|J_h-N/2|\le N/2-(M+Q-1)/2$ and $|J_h-J_s|\le(M+1)/2$. For a given $S_z$, the minimum energy is given when $J_s$ and $J_h$ are as close to $N/2$ as possible. The ground state is given when $S_z$ is either $0$ or $1/2$ and is a singlet whenever possible[@kura91]. When $Q=0$ this reduces to the result for the Heisenberg chain[@hald88; @shas88]. These energy levels have also been found by Ha and Haldane[@ha92], where $J_{\uparrow} = J_h -N+(M+Q+1)/2$ and $J_{\downarrow} =
J_h - J_s - (M-1)/2$. From these energy levels we find the spin and charge velocities to be identical to the previous results[@kura91; @ha92].
To investigate the other excited states of the system, we introduce zeros into the wavefunction by premultiplying it with polynomials of $X_\alpha=\exp (2\pi i x_{\alpha} /N)$ and $Y_i=\exp (2\pi i y_i /N)$. The wavefunctions thus take the following modified Kalmeyer-Laughlin form[@kalm87]: $$\psi(x,y)=\hh (X,Y)\ss (Y) {\bf \Psi_0},$$ where $\hh $ and $\ss$ are completely symmetric under pairwise interchange of their arguments. These states will be termed “fully-polarized spinon states”. Loosely speaking, the polynomials $\hh $ and $\ss $ can be regarded as spin and charge quasiparticle wavefunctions respectively.
When the Hamiltonian acts on this wavefunction, once again all three-body terms combine to give constants. However, in this case, some two-body terms remain and we are left with the eigenvalue equation $${N^2\over\pi^2t}E\hh \ss = E_0\hh \ss + H_1 + H_2 + H_3,
\label{eq:eval}$$ where $$\begin{aligned}
H_1 &=&2\ss \left[\sum_\mu\partial_\mu^2+\sum_{\mu<\nu} {W_\mu+W_\nu
\over W_\mu-W_\nu}(\partial_\mu-\partial_\nu)\right]\hh \nonumber\\
&&+4\hh \left[\sum_i\partial_i^2+{1\over2}\sum_{i<j} {Y_i+Y_j \over
Y_i-Y_j}(\partial_i-\partial_j)\right]\ss \nonumber\\
H_2 &=&4\sum_i\partial_i\hh \partial_i\ss \nonumber\\
H_3 &=&2\ss \sum_{\alpha<\beta}{X_\alpha+X_\beta\over
X_\alpha-X_\beta}(\partial_\alpha-\partial_\beta)\hh .\end{aligned}$$ Here we denote $W\equiv(X,Y)\equiv(X_1 \dots X_M, Y_1 \dots Y_{Q+M})$ and $\partial_\mu\equiv
W_\mu\partial/\partial W_\mu$. In deriving this, we have shifted $\hh $ by the ground state configuration, $\prod_\mu W_\mu^{N/2}$. As a result, $E_0$ is given by Eq. (\[eq:energy\]) with $J_s=J_h=N/2$. We require that $|{\rm degree}\>\hh |\le N/2-(M-1+Q/2)$, $|{\rm degree}\>\hh \ss |\le N/2-(M+Q-1)/2$ and $|{\rm degree}\>\ss |\le(M+1)/2$, which is to hold for each variable $X_\alpha$ or $Y_i$ independently.
The first term, $H_1$, does not mix $\hh $ and $\ss $ and has been solved by Sutherland[@suth72]. However, $H_2$ mixes $\hh $ and $\ss $ and $H_3$ does not act symmetrically on $\hh $. As a result, they are harder to deal with. We follow Sutherland and start by choosing the following symmetric basis functions: $$\begin{aligned}
\hh(W;\nn)&=&\sum_{\{P_\mu\}}
\prod_\mu W_\mu^{n_{P_\mu}}\nonumber\\
\ss(Y;\mm)&=&\sum_{\{{P_i}\}}
\prod_i Y_i^{m_{{P_i}}},
\label{eq:basis}\end{aligned}$$ where the quantum numbers $\{n_1,\ldots,n_{M+Q}\}$ and $\{m_1,\ldots,m_Q\}$ are taken to be in increasing order and $\{P_{\mu}\}$ and $\{ P_i\}$ denote permutations of the indices. These quantum numbers are integral or half-integral as required by periodic boundary conditions.
In this basis, labeled by the two sets of quantum numbers $\{n_\mu\}$ and $\{m_i\}$, the Hamiltonian, considered as a matrix, can be shown to be upper-triangular. Eigenvalues are found by reading the diagonal-elements labeled in terms of the quantum numbers $\{n_\mu\}$ and $\{m_i\}$[@unpub]. The result simplifies when written in terms of a conjugate set of quantum numbers $\{J_1,J_2,\ldots,J_{M+Q}\}$ and $\{I_1,I_2,\ldots,I_Q\}$ defined by $$\begin{aligned}
J_\mu&=&n_\mu+n_\mu^0\qquad\quad\ \, n_\mu^0={1\over2}(2\mu-(M+Q)-1)\nonumber\\
I_i&=&m_i+m_i^0\qquad\quad m_i^0={1\over2}(2i-Q-1),\end{aligned}$$ where $\{n_\mu\}$ and $\{m_i\}$ must satisfy the conditions specified before. This translates into the conditions $|J_\mu|\le
(N-M+1)/2$ and $|I_i|\le (M+Q)/2$. The energy is $${E\over t}={\pi^2\over3}Q(1-{1\over N^2})+{1\over2}\sum_{\mu=1}^{M+Q}
(p_\mu^2-\pi^2),
\label{eq:eigen}$$ where the pseudomomenta, $p_\mu$, are given by the following equations: $$\begin{aligned}
p_\mu N=2\pi J_\mu-\pi&&\sum_{i=1}^Q \sgn(p_\mu-q_i)
+\pi\sum_{\nu=1}^{M+Q}\sgn(p_\mu-p_\nu)\nonumber\\
2\pi I_i=\pi&&\sum_{\mu=1}^{M+Q}\sgn(q_i-p_\mu).
\label{eq:const}\end{aligned}$$ The above equations correspond to the asymptotic Bethe-ansatz equations obtained by Kawakami[@kawa91]. Our result thus confirms that the ABA spectrum is exact.
Here the resulting $\{p_\mu\}$ and $\{q_i\}$ must lie between $-\pi$ and $\pi$. The set of $M+Q$ distinct quantum numbers $J_{\mu}$ is in ascending order and governs the spin excitations. We restrict them to take values in the range $[-(N-M-1)/2,(N-M-1)/2]$ to guarantee that they generate fully spin-polarized states. There are $N-M$ values in this range, of which $M+Q$ are occupied and $2S_z$ are empty. A spin configuration can be represented by a sequence of $N-M$ digits such as $\{S\}=(0111001011)_s$, where 1 represents an occupied quantum number and 0 an unoccupied quantum number. These empty values are identified as spinons[@hald91]; a sequence of $2j_r$ consecutive zeros corresponds to a symmetric bound-state of $2j_r$ spinons, thereby creating an excitation of spin $j_r$ with spin degeneracy $2j_r+1$. On these physical grounds, we anticipate a spin degeneracy in the thermodynamic limit given by $$w_S= \prod_{j} (2j+1)^{n(j)}
\label{eq:degeneracy}$$ where $n(j)$ is the number of sequences of zeros of length $2j$. The set of $Q$ distinct quantum numbers $I_i$, in ascending order and taking values in the range $[-(M+Q)/2, (M+Q)/2]$, governs charge excitations.
To complete the study of the model and confirm our interpretation of the quasiparticle degeneracies, we looked at exact diagonalization of small systems ($N\le10$ with holes). As an example, the low-lying states of the $N=10$, $Q=2$ model is shown in Fig. \[figone\]. We summarize the numerical result as follows:
1. The spectrum described in terms of the real pseudomomenta $\{p_\mu\}$ and $\{q_i\}$ span the $\it full$ set of energy levels of the system.
2. The real pseudomomentum states are all highest weight states when $\{p_\mu\}\ne \pm\pi$.
3. The spin degeneracy rule is obeyed for all internal sequences of zeros.
Certain small corrections to the spin degeneracy rule apply when there are zeros at either end of $\{S\}$ which we shall not enumerate here, and which are not important in the thermodynamic limit[@unpub].
Finally, we may use the spectrum generated by the “fully-polarized spinon states” and the supermultiplicity rule to obtain the free energy of the model in the thermodynamic limit. Besides the “particle-state” solutions of equations (\[eq:eigen\]) and (\[eq:const\]), we have to take into account the “hole-state” solutions[@yang71]. At thermal equilibrium, the distribution functions of these solutions are determined by minimizing the free energy functional[@taka71; @babu83], $F = E - T S - \mu (N -Q)$, with the constraint that each “fully-polarized spinon state” described by quantum numbers $\{ J_{\mu} \}$ and $\{ I_i \}$ is associated with a spin degeneracy $w_S$ as given in (\[eq:degeneracy\]), where $\mu$ is the chemical potential. Following the standard methods of Takahashi[@taka71], minimizing the free energy for a given quantum number distribution gives the following free energy $$F(T)/N=-\mu-{T\over2\pi}
\int_{-\pi}^{\pi} dp\,\ln \left[ 1 + e^{-\beta \eps_s (p) } \right],$$ where $\epsilon_s$ is determined by the coupled equations $$\begin{aligned}
2 \eps_s(p) &=& \eps_0(p) -2a - T \ln \left[ 1 + e^{-\beta\eps_c(p)}
\right]\nonumber\\
\eps_c(q) &=&2a - T \ln \left[ 1 + e^{ -\beta\eps_s (q)}
\right].\end{aligned}$$ Here $\eps_0 (p) ={1\over 2}t(p^2 -{\pi^2\over 3})+\mu$ and $a={1 \over 6}t \pi^2 + {1 \over 2} \mu$. In the limit of half filling, $\mu\rightarrow \infty$, $\eps_s\rightarrow t(p^2 - \pi^2)
/4$, and the free energy reverts to the form obtained by Haldane for the corresponding Heisenberg model[@hald91; @tsvelik]. For general $\mu$, elimination of $\eps_c(p)$ yields the result $$\eps_s(p)=
\eps_o(p)-T
\ln \left[ {1 \over 2} +
\left({1 \over 4} + 2e^{\beta (\eps_0(p)+a)} \cosh (\beta a )\right)^{1/2}
\right],\quad$$ We have verified that high temperature expansion of this free energy in powers of $\beta$ correctly reproduces the first two non-trivial terms in the high temperature perturbation theory.
From the free energy, it is not clear whether we may make a unique identification of the statistics of the spin and charge excitations. We note that the $S=1/2$ spinon excitations always combine into a state with a symmetric spin wavefunction, thus $2S$ spinons form a state with total spin $S$. In the limit of zero doping, the free energy is that of spinless fermions[@hald91; @tsvelik]. We can equally well regard the spinon excitations as $S=1/2$ fermions in a state with a fully antisymmetric spatial wavefunction; or alternatively, as hardcore $S=1/2$ bosons in a fully symmetric spatial state.
In summary, we have derived the spectrum of the 1D t-J model with $1/r^2$ long-range exchange and hopping by the introduction of zeros into Jastrow ground-state wavefunctions. Our solution confirms Kawakami’s conjecture that the ABA provides the exact spectrum, suggesting that despite the long-range nature of the interactions, two-body scattering dominates the long-wavelength physics. By interpreting multiple occupancy of momentum states in the spinon wavefunction as symmetric bound-complexes of spinons, we have been able to determine the degeneracies of the states needed to construct the free energy. Further work is required to determine the integrability conditions of this model. There are also several possible generalizations: most notably, SU(N) generalizations and the appealing possibility of Jastrow-integrable impurity models.
We would like to thank Z. Ha and F. D. M. Haldane for informative discussions. This work was supported in part by the U.S. Department of Energy under Grant No. DE-FG05-85ER-40219, the National Science Foundation under Grant NSF-DMR-13692, and by a Sloan Foundation Grant.
R. B. Laughlin, Phys. Rev. Lett. [**50**]{} ,1395 (1983). D. Arovas and S. M. Girvin, Proceedings of the 7th International Conference on Recent Progress in Many Body Theory, Minneapolis, 1991 (to be published). A. Auerbach, D. Arovas and D. Haldane, Phys. Rev. Lett. [**60**]{}, 531 (1988). C. S. Hellberg and E. J. Mele, Phys. Rev. B [**44**]{}, 1360 (1991), Phys. Rev. Lett. [**67**]{}, 2080 (1991). [*Lecture Notes in Physics*]{}, Vol. 242, eds. B. S. Shastry et al. (Spring-Verlag, 1985) p. 1. F. D. M. Haldane, Phys. Rev. Lett. [**60**]{}, 635 (1988). B. S. Shastry, Phys. Rev. Lett. [**60**]{}, 639 (1988). F. D. M. Haldane, Phys. Rev. Lett. [**66**]{}, 1529 (1991). Y. Kuramoto and H. Yokoyama, Phys. Rev. Lett. [**67**]{}, 1338 (1991). N. Kawakami, Yukawa Institute preprint YITP/K-953 (1991). N. Kawakami and S. K. Yang, Phys. Rev. Lett. [**67**]{}, 2493 (1991). N. Kawakami, Phys. Rev. B [**45**]{}, 7525 (1992), Max-Planck Institute preprint (1992). B. Sutherland, Phys. Rev. A [**5**]{}, 1372 (1972), Phys. Rev. A [**4**]{}, 2019 (1971), J. Math. Phys. [**12**]{}, 251 (1971), J. Math. Phys. [**12**]{}, 246 (1971), Note that a similar model with $1/r$ long-range hopping and a finite local repulsive $U$ has recently been considered by F. Gebhard and A. E. Ruckenstein, Phys. Rev. Lett. P. W. Anderson, B. S. Shastry and D. Hristopulos, Phys. Rev. B [**40**]{}, 8939 (1989). Z. Ha and F. D. M. Haldane, Princeton University preprint (1992). V. Kalmeyer and R. B. Laughlin, Phys. Rev. Lett. [**59**]{}, 2095 (1987). James T. Liu, (unpublished). C. N. Yang and C. P. Yang, J. Math. Phys. [**10**]{}, 1115 (1969). M. Takahashi, Prog. Theor. Phys. [**46**]{}, 401 (1971). H. M. Babudjan, Nucl. Phys. [**B215**]{} \[FS7\], 317 (1983). F. D. M. Haldane and A. Tsvelik, Princeton University preprint (1992). [**68**]{}, 244 (1992).
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'Nir Elkayam, Meir Feder, '
bibliography:
- 'bibfile.bib'
title: Information Spectrum Approach to the Source Channel Separation Theorem
---
Abstract {#abstract .unnumbered}
--------
*A source-channel separation theorem for a general channel has recently been shown by Aggrawal et. al [@agarwal2013universal]. This theorem states that if there exist a coding scheme that achieves a maximum distortion level $d_{max}$ over a general channel [[**W**]{}]{}, then reliable communication can be accomplished over this channel at rates less then $R(d_{max})$, where $R(\cdot)$ is the rate distortion function of the source. The source, however, is essentially constrained to be discrete and memoryless (DMS). In this work we prove a stronger claim where the source is general, satisfying only a “sphere packing optimality” feature, and the channel is completely general. Furthermore, we show that if the channel satisfies the strong converse property as define by Han & [ Verd[ú]{} ]{} [@DBLP:journals/tit/VerduH94], then the same statement can be made with $d_{avg}$, the average distortion level, replacing $d_{max}$. Unlike the proofs in [@agarwal2013universal], we use information spectrum methods to prove the statements and the results can be quite easily extended to other situations.*
Introduction {#sec:intro}
============
The source channel separation theorem, first proved by Shannon [@shannon1959coding] for the transmission of discrete memoryless source (DMS) over discrete memoryless channel (DMC) states that the *separation strategy* is optimal. This means optimal performance can be attained by first compressing the source output to the desired distortion level and then reliably communicating the compressed bits over the channel. The separation theorem was later extended to indecomposable channels [@gallager1968information]. The almost lossless case (transmission codes) was handled at [@vembu1995source] and a general condition is given for the separation theorem to hold.
Joint Source-Channel Coding (JSCC) refers to the case where such a separation is not used. In some cases, e.g. binary sources over BSC or Gaussian source over AWGN channel, separation can attain the optimal performance yet a simpler JSCC strategy can be used, [[*i.e.*]{}]{}, uncoded transmission [@gastpar2003code]. In some cases separation is suboptimal. A simple example would be the symmetric binary source and a compound memoryless BSC where the flipping probability is drawn ahead of the block and stays fixed for the whole block. In this case uncoded transmission is optimal and it is strictly suboptimal to use separation.
Some cation is needed here, because there are two senses of optimality when a distortion measure is given - the **maximal distortion** level and **average distortion** level. The separation relative to the maximal distortion level is easier to accomplish, as this allows us to increase the distortion as long as we do not exceed the desired distortion level. So even if the average distortion in the original scheme is much less then the maximum distortion, the separation strategy yield the desired maximum distortion but the average distortion might be increased.
When average distortion level is used, we should follow the specific distortion, which can be large or small as the channel condition are good or bad. This is much like the variable rate channel capacity [@DBLP:journals/tit/VerduS10] which tries to capture the whole spectrum of channel conditions - when the channel provides good conditions lower distortion level can be achieved or more bits can be reliable transmitted over the channel. A separation strategy in that case will use successive refinement of the source [@equitz1991successive], [@rimoldi1994successive] and then transmission of the bits over a channel with variable rate channel capacity so that the better the channel, the lower the distortion [@tian2008successive] can be made. In many cases this separation strategy is suboptimal.
In transmitting a DMS over a DMC the cases of average distortion level and maximal distortion level coincide [@YuvalKochmanConverse]. However, for other channels this is not always the case even for DMS’s. We will see that for indecomposable channels, this is always true.
Information spectrum methods [@DBLP:journals/tit/VerduH94], [@koga2002information] provide a very simple formalization and intuition into channel capacity in almost every communication situation including unicast, multiple access, broadcast and other situations [@somekh2006general]. The problem is that the expressions given with these methods are usually not useful when we want to compute the channel capacity. Nevertheless, we use information spectrum methods in this work and attain the desired results.
In this paper we deal with sources for which the sphere packing bound is tight. This means that the sphere of radius $d$ around any reconstruction point contains at most $2^{-nR(d)}$ of the typical space. For these source we first prove a generalization of the Han-Verd[ú]{} converse lemma [@DBLP:journals/tit/VerduH94] that connects the *rate distortion spectrum* (that is, the probability that the random variable $D$, the instantaneous distortion level exceed some level $d$) to the information spectrum of the channel. Then we use this to prove a very general separation theorem for the case where maximum distortion level criteria is given. For channels that satisfy the strong converse, which includes DMC and ergodic channels, we prove a separation theorem for the average distortion level as well. Actually, for these kind of channels we prove that the average and maximal distortion level coincide.
Preliminaries
=============
Notation
--------
This paper uses lower case letters (e.g. $x$) to denote a particular value of the corresponding random variable denoted in capital letters (e.g. $X$). Calligraphic fonts (e.g. [[$\mathcal X$]{}]{}) represent a set.
The random variable $D$ represents the instantaneous distortion level. ${\mathbb{E}(D)}=d_{avg}$ is the average distortion level, and $d_{max}$ will be used to denote the maximum distortion level.
We will use the $o(\cdot)$ notation to denote terms that goes to $0$ w.r.t the argument. Mainly, $o(n)$ will denote a sequence $\epsilon_n$ such that $ {\lim_{n \to \infty }}\epsilon_n/n=0 $ and $o(1)$ denote a sequence that converges to 0. Throughout this paper $\log$ will be defined to the base 2 unless otherwise indicated. ${Pr\left\{ A \right\}}$ will denote the probability of the event $A$.
Definition
----------
A general source [[**S**]{}]{} is defined as an infinite sequence of random variable on ${{\mathcal S}}_n$. The reproduction alphabet is defined over the set ${{\hat{\mathcal S}}}_n$. A distortion measure is a function $ d_n: {{\mathcal S}}_n \times {{\hat{\mathcal S}}}_n \rightarrow \mathbb{R_{+}}$. There are several rate distortion functions that can be defined according to the different performance requirements (see [@koga2002information] ch.5.3).
- $R_{fm}(d)$ - fixed length code, maximum distortion criterion.
- $R_{fa}(d)$ - fixed length code, average distortion criterion.
- $R_{vm}(d)$ - variable length code, maximum distortion criterion.
- $R_{va}(d)$ - variable length code, average distortion criterion.
A general source [[**S**]{}]{} is said to be **Sphere packing optimal** if there exist a subsets ${{\mathcal A}}_n \subset {{\mathcal S}}_n$ and $k_n$, such that: $$\begin{aligned}
\label{def:spo_source}
&{\lim_{n \to \infty }}{Pr\left\{ {\textbf{s}_n}\notin {{\mathcal A}}_n \right\}} = 0 \\
&{Pr\left\{ {\textbf{s}_n}\in {{\mathcal A}}_n, d({\textbf{s}_n}, {\hat{\textbf{s}}_n}) \leq d \right\}} \leq 2^{-n{\left( R(d)+k_n \right)}} \\
&{\lim_{n \to \infty }}k_n = 0\end{aligned}$$ for each $d \geq 0$ and $ {\hat{\textbf{s}}_n}\in {{\hat{\mathcal S}}}_n $.
In the appendix we demonstrate that DMS’s are SPO. The set ${{\mathcal A}}_n$ will be the strong typical sequences relative to the source distribution. The Gaussian source with mean square distortion is also SPO as can be shown by a straight forward calculation, given in the appendix as well.
For SPO sources that have a reference word (see [@koga2002information] Theorem 5.3.1), the following different notions of rate functions are equal $$R_{fm}(d) = R_{fa}(d) = R_{vm}(d).$$ The common value will be denoted $R(d)$ without the subscript. This is also shown in the appendix.
A general channel is a sequence of transition matrices $ {{\textbf W}}= {\left\{ W_n : {{\mathcal X}}_n \rightarrow {{\mathcal Y}}_n \right \}} $ where $ W^n(y|x) $ denotes the conditional probability of $y$ given $x$. Throughout this paper we will assume that the channel has finite input and output space. [^1]
The source and channel input and output will be written as ${{\mathcal S}}_n, {{\mathcal X}}_n, {{\mathcal Y}}_n, {{\hat{\mathcal S}}}_n$ to indicate that it is usually not the $n^{th}$ order cartesian product. As an example let ${{\mathcal S}}_n={{\mathcal S}}^{nR}, {{\hat{\mathcal S}}}_n={{\hat{\mathcal S}}}^{nR}$ and ${{\mathcal X}}_n={{\mathcal X}}^n, {{\mathcal Y}}_n={{\mathcal Y}}^n$. This allows us to simplify the notation and to avoid unnecessary “rate” indication. We will use $x$ and $y$ to denote the input and output. Occasionally, we will omit the subscript $n$ and assume that it is understood from the context.
Throughout the paper we assume that the sets ${\left( {{\mathcal S}}_n, {{\mathcal X}}_n, {{\mathcal Y}}_n, {{\hat{\mathcal S}}}_n \right)}$ are discrete. However, the results are valid for any abstract spaces for which information spectrum method can be applied see [@koga2002information].
A general JSCC Scheme of the source ${{\textbf S}}$ over the channel ${{\textbf W}}$, includes:
- Encoding function: $E_n: {{\mathcal S}}_n \rightarrow {{\mathcal X}}_n$
- Decoding function: $D_n: {{\mathcal Y}}_n \rightarrow {{\hat{\mathcal S}}}_n $.
There are several random variables defined:
- ${{\textbf X}}_n = E_n({{\textbf S}}_n)$ - The random variable of the channel input.
- ${{\textbf Y}}_n$ - The output from the channel ${{\textbf W}}$ resulted from the input ${{\textbf X}}_n$.
- $D = d({{\textbf S}}_n, D_n({{\textbf Y}}_n))$ - The instantaneous distortion.
Sometimes we will write $x({\textbf{s}_n})$ for $E_n({\textbf{s}_n})$. The maximum distortion level $d_{max}$ is the infimum of the set of numbers $\alpha$, such that ${Pr\left\{ D \geq \alpha \right\}} {\underset{n \rightarrow \infty}{\rightarrow}}0$. The average distortion level is ${\mathbb{E}(D)}=d_{avg}$.
Information Spectrum Notation
-----------------------------
If ${{\mathcal A}}_n$ is a sequence of random variables, its **liminf in probability** is the supremum of all the reals $\alpha$ for which ${Pr\left\{ {{\mathcal A}}_n \geq \alpha \right\}} {\underset{n \rightarrow \infty}{\rightarrow}}0$.
We will use $i(a;b)$ to denote the information density between two outcomes from correlated random variables $A$ and $B$. Specifically, $i(a;b) = \frac{1}{n}\log{\left( \frac{p(a|b)}{p(a)} \right)} $. We will omit the $n$ and the indication to which random variable produce the outcomes as it will be understood from the context.
Given random variables ${{\textbf X}}_n, {{\textbf Y}}_n$ with joint distribution $p(x,y)$ on ${{\mathcal X}}_n \times {{\mathcal Y}}_n$, the **information density** is the function defined on ${{\mathcal X}}_n \times {{\mathcal Y}}_n$: $$i_{{{\textbf X}}_n, {{\textbf Y}}_n}(x;y) = \frac{1}{n}\log{\left( \frac{p_{{{\textbf Y}}_n|{{\textbf X}}_n}(y|x)}{p_{{{\textbf Y}}_n}(y)} \right)}$$
we will write $i{\left( x,y \right)}$ instead of $i_{{{\textbf X}}_n, {{\textbf Y}}_n}(x;y)$ to avoid cumbersome notation, as it will be understood which random variables is in use.
The liminf in probability of the sequence of random variables $\frac{1}{n}i{\left( {{\textbf X}}_n, {{\textbf Y}}_n \right)}$ will be referred to as the the inf-information rate of the pair $ {{\textbf X}}_n, {{\textbf Y}}_n $ and will be denoted as $ \underline{I}{\left( {{\textbf X}}_n; {{\textbf Y}}_n \right)}$.
In [@DBLP:journals/tit/VerduH94] it is shown that the capacity of a general channel is given by: $$C = \sup_{{{\textbf X}}_n}\underline{I}{\left( {{\textbf X}}_n; {{\textbf Y}}_n \right)}$$ where the supremum is over the prior distribution ${{\textbf X}}_n$.
The epsilon-capacity $C_{\epsilon}$ is the supremum of all rates $R$, for which there exist a coding scheme with rate $R$ and error probability less then $ \epsilon$. A channel is said to satisfy the strong converse property, if $C = \lim_{\epsilon \to 0 } C_{\epsilon}$. This means that is there exist a coding scheme of rate $R$ with error less then 1, then there exist a coding scheme of rate $R$ with error probability approach 0.
Throughout the sequel we will need to use the following lemma which say that if there exist a probability mass to the right of the mean of a non-negative random variable then there must be a probability mass left to it to.
\[lem:pos\_D\_mean\] Let $D$ be a non-negative random variable with $\mu = {\mathbb{E}(D)}< \infty$. If there exist $d_1>0$ and $\epsilon_1>0$ such that: ${Pr\left\{ D > \mu+d_1 \right\}} > \epsilon_1$, then there exist $d_2>0$ and $\epsilon_2>0$ such that: ${Pr\left\{ D < \mu-d_2 \right\}} > \epsilon_2$.
The proof is given in the appendix \[App:AppendixB\].
Joint Source Channel lemma (Unicast) {#sec:lemmas}
====================================
In this section we state and prove a generalization of the Han -[ Verd[ú]{} ]{}converse lemma [@DBLP:journals/tit/VerduH94] which relates the error rate probability to the information spectrum. Here we can connect the instantaneous distortion level $D$ with the information spectrum of the channel. A similar, but different converse result was given in [@DBLP:journals/tit/KostinaV13 Theorem 1] for a general source. We, however, provide the following lemma that holds for an SPO source which is enough for our purposes. Let: $$L_{R(d)}^{\gamma} = {\left\{ {\left( {\textbf{s}_n},y \right)} : i {\left( x({\textbf{s}_n});y \right)} \leq R(d)-\gamma \right \}}$$ This is exactly the set which is used in the definition of channel capacity in terms of information spectrum.
\[lem:JSC\_Converse\] If the source [[**S**]{}]{} is SPO, then: $${Pr\left\{ D > d \right\}} \geq {Pr\left\{ L_{R(d)}^{\gamma} \right\}} -2^{-n {\left( \gamma+k_n \right)}} - {Pr\left\{ {\textbf{s}_n}\notin {{\mathcal A}}_n \right\}}$$
The term ${Pr\left\{ L_{R(d)}^{\gamma} \right\}}$ can be bounded by: $$\begin{aligned}
\label{eq:con_uni_1}
&{Pr\left\{ L_{R(d)}^{\gamma} \right\}} \leq {Pr\left\{ D > d \right\}} + {Pr\left\{ {\textbf{s}_n}\notin {{\mathcal A}}_n \right\}} + {Pr\left\{ L_{R(d)}^{\gamma} \cap {\left( D \leq d \right)}\cap{\left( {\textbf{s}_n}\in {{\mathcal A}}_n \right)} \right\}}\end{aligned}$$ Continues with the last term in : $$\begin{aligned}
\label{eq:con_uni_2}
&{Pr\left\{ L_{R(d)}^{\gamma} \cap {\left( D \leq d \right)}\cap{\left( {\textbf{s}_n}\in {{\mathcal A}}_n \right)} \right\}} \notag \\
& = \sum_{\substack{{\textbf{s}_n}\in {{\mathcal A}}_n\\ {\left( x({\textbf{s}_n}),y \right)} \in L_{R(d)}^{\gamma}\\ D \leq d}}p({\textbf{s}_n})p(y|x({\textbf{s}_n})) \notag \\
&\overset{(a)}{\leq} \sum_{\substack{{\textbf{s}_n}\in {{\mathcal A}}_n, y \in {{\mathcal Y}}_n \\ D \leq d}}p({\textbf{s}_n})p(y) \cdot 2^{nR(d)-n \gamma} \\
&= \sum_{y \in {{\mathcal Y}}_n }p(y) \cdot 2^{nR(d)-n \gamma} \cdot \sum_{\substack{{\textbf{s}_n}\in {{\mathcal A}}_n\\ \hat{s}^n=D_n(y), d({\textbf{s}_n},\hat{s}^n) \leq d}}p({\textbf{s}_n}) \notag \\
&\overset{(b)}{\leq} \sum_{y \in {{\mathcal Y}}_n }p(y) \cdot 2^{nR(d)-n \gamma} \cdot 2^{-n{\left( R(d)+k_n \right)}} \notag \\
&= 2^{-n{\left( \gamma+k_n \right)}} \notag\end{aligned}$$
where $(a)$ follows because ${\left( {\textbf{s}_n},y \right)} \in L_{R(d)}^{\gamma} \Rightarrow p(y|x({\textbf{s}_n})) \leq p(y) \cdot 2^{nR(d)-n\gamma}$ and $(b)$ follows from the SPO assumption of the source [[**S**]{}]{}. Combining and complete the proof of the inequality.
Source Channel Separation Theorems (Unicast) {#sec:uni_thrm}
============================================
Maximum distortion
-------------------
\[sec:uni\_thrm1\] If the maximum distortion in the joint source channel coding of an SPO source [[**S**]{}]{} is $d_{max}$ then any rate less then $R(d_{max})$ is achievable, [[*i.e.*]{}]{} the capacity of the channel [[**W**]{}]{} is at least $R(d_{max})$.
Fix a sequence $\gamma_n {\underset{n \rightarrow \infty}{\rightarrow}}0$ such that $n{\left( \gamma_n+k_n \right)} {\underset{n \rightarrow \infty}{\rightarrow}}\infty$. Let $X_n=E_n(S_n)$ and $Y_n$ the output of $X_n$ through the channel [[**W**]{}]{}. From lemma \[lem:JSC\_Converse\] we have: $${Pr\left\{ D > d_{max} \right\}} \geq {Pr\left\{ L_{R(d_{max})}^{\gamma_n} \right\}}-2^{-n {\left( \gamma_n+k_n \right)}}+o(1)$$ Since $ {\lim_{n \to \infty }}{Pr\left\{ D > d_{max} \right\}} = 0 $ and $2^{-n{\left( \gamma_n+k_n \right)}} {\underset{n \rightarrow \infty}{\rightarrow}}0 $ we get $ {\lim_{n \to \infty }}{Pr\left\{ L_{R(d_{max})}^{\gamma_n} \right\}} = 0 $. Since $\gamma_n {\underset{n \rightarrow \infty}{\rightarrow}}0$ this implies that: $${\lim_{n \to \infty }}{Pr\left\{ i {\left( x({\textbf{s}_n});y \right)}\leq R(d) \right\}} = 0$$ which proves that $ \underline{I}{\left( {{\textbf X}}_n; {{\textbf Y}}_n \right)} \geq R(d)$ so that channel capacity is at least $R(d)$ and separation is possible.
Average distortion
-------------------
\[sec:uni\_thrm2\] For an optimal JSCC scheme, of an SPO source over a channel that satisfy the strong converse property, the notion of average distortion and maximum distortion coincide.
Suppose not. Then there exist $\tau_1$ and $ \epsilon_1 $ such that: $${\limsup_{n \to \infty }}{Pr\left\{ D > {\mathbb{E}(D)}+\tau_1 \right\}} \geq \epsilon_1$$ Fix a sequence $\gamma_n {\underset{n \rightarrow \infty}{\rightarrow}}0$ such that $n{\left( \gamma_n+k_n \right)} {\underset{n \rightarrow \infty}{\rightarrow}}\infty$. By lemma \[lem:pos\_D\_mean\] we have: ${\liminf_{n \to \infty }}{Pr\left\{ D > {\mathbb{E}(D)}-\tau \right\}} > 1-\epsilon$ for some $\tau>0$ and $\epsilon>0$). Now: $$\begin{aligned}
\label{eq:uni_thm_1}
1-\epsilon &\geq {\liminf_{n \to \infty }}{Pr\left\{ D > {\mathbb{E}(D)}-\tau \right\}} \notag \\
&\geq {\liminf_{n \to \infty }}{\left( {Pr\left\{ L_{R({\mathbb{E}(D)}-\tau)}^{\gamma_n} \right\}}-2^{-n{\left( \gamma_n+k_n \right)}}+o(1) \right)}\end{aligned}$$ For $n$ large enough we have that ${\liminf_{n \to \infty }}{Pr\left\{ L_{R({\mathbb{E}(D)}-\tau)}^{\gamma_n} \right\}}$ bound away from 1, so we can transmit at rate $R({\mathbb{E}(D)}-\tau)-\gamma_n$ with error probability less then 1. Since the channel satisfy the strong converse, this means that we can transmit at rate $R({\mathbb{E}(D)}-\tau)$ with error approach 0, so the JSCC is not optimal because with separation we can transmit at maximum distortion level ${\mathbb{E}(D)}-\tau$ which is better the average ${\mathbb{E}(D)}$. (We used here the fact that $R_{fm}(d) = R_{fa}(d)$).
For the average distortion rate we can integrate the bound in lemma \[lem:JSC\_Converse\] to get:
\[sec:uni\_thrm2\] If there is a bound on the maximum distortion $d_{max}$ then: $${\mathbb{E}(D)}\geq \int_{0}^{d_{max}}{Pr\left\{ L_{R(\alpha)}^{\gamma} \right\}}d\alpha -d_{max} \cdot 2^{-n \gamma+o(n)} - d_{max} \cdot {Pr\left\{ {\textbf{s}_n}\notin {{\mathcal A}}_n \right\}}$$
Use ${\mathbb{E}(D)}= \int_{0}^{d_{max}}{Pr\left\{ D>\alpha \right\}}d\alpha $ and lemma \[lem:JSC\_Converse\].
The Multiple Access Case
========================
In this section we demonstrate in a loose manner how the same method can applied to more general communication situations. Specifically, a 2-users multiple access channel. Let ${{\textbf W}}$ now represent a multiple access channel with 2 inputs - $ {{\textbf W}}= {\left\{ W_n : {{\mathcal X}}_{1,n} \times {{\mathcal X}}_{2,n}\ \rightarrow {{\mathcal Y}}_n \right \}} $ and the sources ${{\textbf S}}_1, {{\textbf S}}_2$ are two uncorrelated SPO sources. Let: $$L_{R_1(d)}^{\gamma} = {\left\{ {\left( {\textbf{s}_n^1},{\textbf{s}_n^2},y \right)} : i {\left( x_1({\textbf{s}_n^1});y|x_2({\textbf{s}_n^1}) \right)} \leq R_1(d)-\gamma \right \}}$$ $$L_{R_2(d)}^{\gamma} = {\left\{ {\left( {\textbf{s}_n^1},{\textbf{s}_n^2},y \right)} : i {\left( x_2({\textbf{s}_n^2});y|x_1({\textbf{s}_n^1}) \right)} \leq R_2(d)-\gamma \right \}}$$ $$L_3^{\gamma} = {\left\{ {\left( {\textbf{s}_n^1},{\textbf{s}_n^2},y \right)} : i {\left( x_1({\textbf{s}_n^1}), x_2({\textbf{s}_n^2});y \right)} \leq R_1(d)+R_2(d)-\gamma \right \}}$$ Also, let $T=L_{R_1(d)}^{\gamma}\cup L_{R_2(d)}^{\gamma} \cup L_3^{\gamma}$ This is exactly the event which is used in the definition of channel capacity in term of information spectrum.
\[lem:JSC\_Converse\_MA\] If the source ${{\textbf S}}_1, {{\textbf S}}_2$ are SPO, then: $${Pr\left\{ D_1 > d_1 \cup D_2 > d_2 \right\}} \geq {Pr\left\{ T \right\}} -2^{-n \gamma+o(n)} - {Pr\left\{ {\textbf{s}_n^1}\notin A_{n1} \right\}} - {Pr\left\{ {\textbf{s}_n^2}\notin A_{n2} \right\}}$$
The term ${Pr\left\{ T \right\}}$ can be bounded by: $$\begin{aligned}
\label{eq:conv_multi_1}
&{Pr\left\{ T \right\}} \leq {Pr\left\{ D_1 > d_1 \cup D_2 > d_2 \right\}} \notag \\
&+{Pr\left\{ {\textbf{s}_n^1}\notin A_{n1} \right\}} + {Pr\left\{ {\textbf{s}_n^2}\notin A_{n2} \right\}} \\
&+ {Pr\left\{ T \cap {\left( D_1 \leq d_1 \right)} \cap {\left( D_2 \leq d_2 \right)}\cap{\left( {\textbf{s}_n^1}\in A_{n1} \right)} \cap{\left( {\textbf{s}_n^2}\in A_{n2} \right)} \right\}} \notag\end{aligned}$$ Using the union bound on the $T$ term we get 3 terms which can be bounded like before. We’ll demonstrate for the term which contains $ L_{R_1(d)}^{\gamma}$, the others follow the same lines. $$\begin{aligned}
\label{eq:conv_multi_2}
&{Pr\left\{ L_{R_1(d)}^{\gamma} \cap {\left( D_1 \leq d_1 \right)} \cap {\left( D_2 \leq d_2 \right)}\cap{\left( {\textbf{s}_n^1}\in A_{n1} \right)} \cap{\left( {\textbf{s}_n^2}\in A_{n2} \right)} \right\}} \notag \\
&\leq {Pr\left\{ L_{R_1(d)}^{\gamma} \cap {\left( D_1 \leq d_1 \right)} \cap{\left( {\textbf{s}_n^1}\in A_{n1} \right)} \right\}} \notag \\
& = \sum_{...}p({\textbf{s}_n^1})p({\textbf{s}_n^2})p(y|x({\textbf{s}_n^1}), x({\textbf{s}_n^2})) \notag \\
& \overset{(a)}{\leq} \sum_{...}p({\textbf{s}_n^1})p({\textbf{s}_n^2})p(y|x({\textbf{s}_n^2})) \cdot 2^{nR_1(d)-n \gamma} \notag \\
& = \sum_{...}p({\textbf{s}_n^2})p(y|x({\textbf{s}_n^2})) \cdot 2^{nR_1(d)-n \gamma} \sum_{...}p({\textbf{s}_n^1}) \notag \\
& \overset{(b)}{\leq} \sum_{...}p({\textbf{s}_n^2})p(y|x({\textbf{s}_n^2})) \cdot 2^{nR_1(d)-n \gamma} \cdot 2^{-nR_1(d)+o(n)} \notag \\
& = 2^{-n\gamma+o(n) }\end{aligned}$$
where $(a)$ follows because $p(y|x({\textbf{s}_n^1}), x({\textbf{s}_n^2})) \leq p(y|x({\textbf{s}_n^2})) \cdot 2^{nR_1(d)-n \gamma} $ for ${\left( {\textbf{s}_n^1}, {\textbf{s}_n^2},y \right)} \in L_{R_1(d)}^{\gamma}$ and $(b)$ follows from the SPO assumption of the source ${{\textbf S}}_1$. Combining and complete the proof of the inequality.
We can use this lemma to prove 2 JSCC Seperation analog to \[sec:uni\_thrm1\] and \[sec:uni\_thrm2\].
Further Research
================
There are several ways to continue with this research.
- SPO sources - Find out which sources are SPO’s. Furthermore, examine whether the SPO condition can be relaxed.
- Another interesting question is whether sources for which the average distortion rate equals the maximum distortion rate are SPO’s. If this result holds, it will provide a source channel separation for ergodic stationary sources with sub-additive distortion measure.
- Examine the average distortion case for source transmission over channels that do not satisfy the strong converse.
- Network case: Examine the correlated source [@tian2010optimality].
These problems and probably additional questions are left for further research.
SPO Sources {#App:AppendixA}
===========
Gaussian Source with mean square distortion
-------------------------------------------
Here we provide sketch to prove that gaussian source with mean square distortion is SPO. Let ${{\textbf S}}$ be a gaussian source with variable $\sigma^2$. Let $ \varepsilon_n \geq 0$ be a sequence such that $ {\lim_{n \to \infty }}\varepsilon_n = 0$ and $ {\lim_{n \to \infty }}n \cdot \varepsilon_n^2 = \infty $. ${{\mathcal A}}_n = {\left\{ X^n: \left| \frac{1}{n}\sum_{i=1}^n X_i^2 - \sigma^2 \right| \leq \varepsilon_n \right \}} $. By the Chebyshev’s inequality we have ${Pr\left\{ {{\mathcal A}}_n \right\}} {\underset{n \rightarrow \infty}{\rightarrow}}1$. For any reproduction vector $y^n={\left( y_i \right)}$ define $ d = \frac{1}{n}\sum_{i=1}^{n} (X_i -y_i )^2 $. We need to prove that: $${Pr\left\{ {\left( d \leq D \right)} \cap {{\mathcal A}}_n \right\}} \leq 2^{-nR(D)+o(n)} = 2^{-\frac{n}{2}\log{\left( \frac{\sigma^2}{D} \right)}+o(n)}$$
$$\begin{aligned}
\label{eq:gauss_spo1}
&{Pr\left\{ {\left( d \leq D \right)} \cap {{\mathcal A}}_n \right\}} \notag \\
& = \int_{X^n \in {{\mathcal A}}_n, d \leq D}{\left( 2\pi\sigma^2 \right)}^{-\frac{n}{2}}e^{-\frac{\sum_{i=1}^n X_i^2}{2\sigma^2}} dX \notag \\
& \leq \int_{X^n \in {{\mathcal A}}_n, d \leq D}{\left( 2\pi\sigma^2 \right)}^{-\frac{n}{2}}e^{-n\frac{\sigma^2 - \varepsilon_n}{2\sigma^2}} dX \notag \\
& \leq {\left( 2\pi\sigma^2 \right)}^{-\frac{n}{2}}e^{-n\frac{\sigma^2 - \varepsilon_n}{2\sigma^2}} \int_{d \leq D} dX \notag \\
& \overset{(a)}{=} {\left( 2\pi\sigma^2 \right)}^{-\frac{n}{2}} e^{-\frac{n}{2}+n\frac{\varepsilon_n}{2\sigma^2}} \frac{\pi^{\frac{n}{2}}}{\Gamma {\left( \frac{n}{2}+1 \right)}} {\left( \sqrt{nD} \right)}^{n} \notag \\
& \overset{(b)}{=} {\left( 2\sigma^2 \right)}^{-\frac{n}{2}} e^{-\frac{n}{2}+n\frac{\varepsilon_n}{2\sigma^2}} \frac{1}{{\left( \frac{n}{2} \right)}!} {\left( nD \right)}^{\frac{n}{2}} \notag \\
& \overset{(c)}{\approx} {\left( 2\sigma^2 \right)}^{-\frac{n}{2}} e^{-\frac{n}{2}+n\frac{\varepsilon_n}{2\sigma^2}} \frac{1}{{\left( \frac{n}{2e} \right)}^{\frac{n}{2}}} {\left( nD \right)}^{\frac{n}{2}} \notag \\
& = {\left( \sigma^2 \right)}^{-\frac{n}{2}}e^{n\frac{\varepsilon_n}{2\sigma^2}} {\left( D \right)}^{\frac{n}{2}} \notag \\
& = 2^{-\frac{n}{2} \log {\left( \frac{\sigma^2}{D} \right)}} e^{n\frac{\varepsilon_n}{2\sigma^2}} \notag\end{aligned}$$
$(a)$ follows from the well known formula for the volume of $n$-dimensional sphere with radius $\sqrt{nD}$.\
$(b)$ follows because we canceled $\pi$, and assuming $n$ is even for which value the formula for $\Gamma$ because simpler.\
$(c)$ is the stirling’s approximation where we ignored the $\sqrt{2 \pi n}$ term which is not contributing to the exponent.
DMC with finite distortion
--------------------------
A proof that uses the method of type can be given along the lines of lemma 1 in [@lomnitz2011communication]. We’ll just note that we can control the type’s of $x$ by the intersection with ${{\mathcal A}}_n$.
Notions of distortions
----------------------
For a source that satisfy the SPO property we can show that the average $R_{fa}(d)$ and the maximum rate function $R_{fm}(d)$ coincide.
To see this, let $Y$ be the result of encoding and decoding with average distortion $d$ and assume that $R_{fa}(d) < R_{fm}(d)$. Then we have: $$\begin{aligned}
{Pr\left\{ {\left( d{\left( X,Y \right)} \leq \alpha \right)} \cap {{\mathcal A}}_n \right\}} &\leq \sum_{i=1}^{2^{nR_{fa}(d)}} {Pr\left\{ {\left( d{\left( X,y_i \right)} \leq \alpha \right)} \cap {{\mathcal A}}_n \right\}} \\
&\leq 2^{nR_{fa}(d)} \cdot 2^{-n{\left( R_{fm}(\alpha)+k_n \right)}} \\
&\leq 2^{n{\left( R_{fa}(d)-R_{fm}(d)+k_n \right)}} {\underset{n \rightarrow \infty}{\rightarrow}}0\end{aligned}$$
Continue: $$\begin{aligned}
& 1-{Pr\left\{ {\left( d{\left( X,Y \right)} \leq \alpha \right)} \cap {{\mathcal A}}_n \right\}} \\ &= {Pr\left\{ {\left( d{\left( X,Y \right)} > \alpha \right)} \cup {{\mathcal A}}_n^c \right\}} \\
& \leq {Pr\left\{ d{\left( X,Y \right)} > \alpha \right\}} + {Pr\left\{ {{\mathcal A}}_n^c \right\}}\end{aligned}$$ Rearranging we get: $$\begin{aligned}
{Pr\left\{ d{\left( X,Y \right)} > \alpha \right\}} \geq 1-{Pr\left\{ {\left( d{\left( X,Y \right)} \leq \alpha \right)} \cap {{\mathcal A}}_n \right\}}-{Pr\left\{ {{\mathcal A}}_n^c \right\}}\end{aligned}$$
From: $$\begin{aligned}
{\mathbb{E} \left( d{\left( X,Y \right)} \right)} &= \int_{0}^{\infty}{Pr\left\{ d{\left( X,Y \right)} > \alpha \right\}} d\alpha \\
& \geq \int_{0}^{d}{Pr\left\{ d{\left( X,Y \right)} > \alpha \right\}} d\alpha \\
& \geq \int_{0}^{d}{\left( 1-{Pr\left\{ {\left( d{\left( X,Y \right)} \leq \alpha \right)} \cap {{\mathcal A}}_n \right\}}-{Pr\left\{ {{\mathcal A}}_n^c \right\}} \right)} d\alpha \\
& = d - \int_{0}^{d}{Pr\left\{ {\left( d{\left( X,Y \right)} \leq \alpha \right)} \cap {{\mathcal A}}_n \right\}} d\alpha - d \cdot {Pr\left\{ {{\mathcal A}}_n^c \right\}}\\\end{aligned}$$ The results then follows because ${Pr\left\{ {{\mathcal A}}_n^c \right\}} {\underset{n \rightarrow \infty}{\rightarrow}}0$ and ${Pr\left\{ {\left( d{\left( X,Y \right)} \leq \alpha \right)} \cap {{\mathcal A}}_n \right\}} {\underset{n \rightarrow \infty}{\rightarrow}}0$ for $ \alpha < d $.
For a source that satisfy the SPO property we can show that the average $R_{fm}(d)$ and the maximum rate function $R_{vm}(d)$ coincide.
To see this, let $l_i$ be the length of the $i^{th}$ code word ordered according to their probability. By using kraft’s inequality it can be shown that $R = \frac{1}{n} E {\left( l_i \right)} \geq \frac{1}{n}H(l_i)$, [[*e.g.*]{}]{} [@koga2002information Theorem 5.6.1]. Since ${Pr\left\{ l_i \right\}} \leq 2^{-nR_{fm}(d)}$ it follows that $H(l_i) \geq nR_{fm}(d)$.
Proof of lemma \[lem:pos\_D\_mean\] {#App:AppendixB}
===================================
For a non-negative random variable we have: $ \mu = \int_{0}^{\infty}{Pr\left\{ D>x \right\}}dx $. Now: $$\begin{aligned}
\int_{\mu}^{\infty}{Pr\left\{ D>x \right\}}dx
&\geq \int_{\mu}^{\mu+d_1}{Pr\left\{ D>x \right\}}dx \\
&\geq \int_{\mu}^{\mu+d_1}\epsilon_1dx = d_1 \epsilon_1\end{aligned}$$ So we have: $ \int_{0}^{\mu}{Pr\left\{ D>x \right\}}dx < \mu - d_1 \epsilon_1$. Let $d_2$ and $\epsilon_2$ be such that $(\mu-d_2)(1-\epsilon_2) > \mu-d_1 \epsilon_1$. We must have: ${Pr\left\{ D < \mu-d_2 \right\}} > \epsilon_2$ because otherwise: $$\begin{aligned}
\mu - d_1 \epsilon_1 & > \int_{0}^{\mu}{Pr\left\{ D>x \right\}}dx \notag \\
&> \int_{0}^{\mu-d_2}{Pr\left\{ D>x \right\}}dx \\
&> \int_{0}^{\mu-d_2}{\left( 1-\epsilon_2 \right)}dx \\
& = {\left( \mu-d_2 \right)}{\left( 1-\epsilon_2 \right)} \notag\end{aligned}$$
[^1]: Extension of the results to abstract input and output spaces ${\left( {{\mathcal X}}_n, {{\mathcal Y}}_n \right)}$ require subtle handling, see [@koga2002information], but is possible to whenever situation which information spectrum can be used.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We point out that it is possible to define CFT duals to AdS theories with massive gravitons only. This is achieved by considering the product of two or more CFTs on the boundary, each possessing a large-N expansion and a corresponding bulk dual. It is possible to couple the CFTs through a marginal perturbation that makes all but one of the bulk gravitons massive. In the infinite-N limit for one of the CFTs the massless graviton decouples.'
author:
- Luis Apolo
- Massimo Porrati
title: 'On AdS/CFT without Massless Gravitons'
---
In the AdS/CFT correspondence [@Maldacena:1997re; @Witten:1998qj; @Gubser:1998bc] fields in the bulk act as sources to dual operators on the boundary. In particular, massless gauge fields are dual to conserved currents in the CFT. If the gauge fields are massive their dual operators are not conserved. A remarkable example of this is the correspondence between higher spin theories in AdS$_4$ and 3D $O(N)$ vector models [@Klebanov:2002ja; @Sezgin:2002rt]. The free $O(N)$ model is dual to Vasiliev’s theory [@Vasiliev:1992av; @Vasiliev:1995dn] with massless higher spin fields of even spin. These fields are dual to the infinite number of conserved currents of the free vector model. In the critical $O(N)$ model [@Brezin:1972se; @Wilson:1973jj] these currents are no longer conserved. Accordingly, its bulk dual is Vasiliev’s theory where the higher spin fields are now massive [@Girardello:2002pp; @Giombi:2009wh; @Giombi:2011ya]. On the other hand the graviton in the AdS/CFT correspondence is massless since its dual, the stress-energy tensor of the boundary theory, is always conserved as a consequence of conformal symmetry.
One may think that a CFT dual to an AdS theory with massive gravitons is only possible if it contains an additional massless graviton, but there exist several ways in which one may attempt to modify a pair of AdS/CFT theories to yield duals with only massive gravitons. One possibility is to spontaneously break diffeomorphism invariance and make the graviton massive via quantum corrections from matter fields. This is accomplished by giving the scalars in AdS$_{d+1}$ non-standard, *transparent* boundary conditions characterized by the propagator [@Porrati:2001db; @Aharony:2006hz] $$\begin{aligned}
G(x,x') = a_1\,G_1 (x,x') + a_2\,G_2 (x,x'). \label{eq:propagator}
\end{aligned}$$ Here $G_k(x,x')$, $k=1,2$, is the propagator corresponding to the standard $D(\D_k,0)$ representation of $SO(d,2)$ where $\sum_k \D_k = d$ and $d/2 > \D_1 > d/2-1$. The $a_k$ are constants satisfying $\sum_k a_k = 1$. Scalars with these boundary conditions can form bound states, vectors in the $D(d+1,1)$ representation of AdS$_{d+1}$, that provide the extra three degrees of freedom necessary for the graviton to become massive. The mass of the graviton must thus be proportional to $a_1 a_2$.
Now recall that a $D(\D,0)$ scalar in AdS$_{d+1}$ with $\D > d/2$ behaves near the boundary as $$\begin{aligned}
\phi(z,x) \approx z^{d-\D} \a(x) + z^{\D} \b(x),
\end{aligned}$$ where $z$ is the radial coordinate of AdS in Poincaré coordinates. If we denote by $\W$ a deformation on the CFT and by $\O$ the operator dual to $\phi$, we have [@Witten:2001ua] $$\begin{aligned}
& \a = \frac{\p \W}{\p \b}, & \b = \frac{1}{2\D - d}\Vev{\O}.
\end{aligned}$$ In principle it is possible to change the boundary conditions of the bulk scalars, and give a mass to the graviton, by deforming the dual CFT with an appropriate $\W$. This, however, will likely result in a pathological theory at the boundary. For example, in the correspondence between Vasiliev’s theory and $O(N)$ vector models the deformation is non-local and breaks conservation of the (boundary) stress-energy tensor. Changing the boundary conditions also introduces problems in the interpretation of the bulk theory since the scalars are now able to cross the boundary of AdS – it is in this sense that the boundary conditions are transparent.
Instead, in this Letter we will consider the AdS/CFT duals of Aharony, Clark, Karch, and Kiritsis [@Aharony:2006hz; @Kiritsis:2006hy], which contain massive gravitons and have enough parameters so that a limit exists where the massless graviton decouples. This yields a non-trivial duality between a theory in AdS with only massive gravitons and a CFT on the boundary. In [@Aharony:2006hz; @Kiritsis:2006hy] the product of two or more $d$-dimensional CFTs interacting with one another via a double-trace deformation was studied. Here we consider only the product of two CFTs since their deformation is (almost) unique. Each CFT, denoted by CFT$_i$ with $i=1,2$, lives in the adjoint or vector representations of a group of degree $N_i$ and contains a scalar operator $\O_i$ of dimension $\D_i$ such that $\sum_i\D_i = d$ and $d/2 > \D_1 > d/2-1$. It is possible to have more than one $\O_i$ operator in each CFT and we will consider this case later on.
Each CFT$_i$ has a corresponding dual, a theory in AdS$_{d+1} \times X^{(i)}_{9-d}$ denoted by AdS$^{(i)}$. The AdS theories are not mutually interacting and their geometries are identified at the boundary. This makes transparent boundary conditions feasible. Each theory contains a massless graviton and a scalar $\phi_i$ dual to $\O_i$ in the (standard) $D(\D_i,0)$ representation of $SO(d,2)$. Since $\sum_i \D_i = d$ and $\D_i \ne d/2$ the mass of the scalar fields lies in the range $-d^2/4 < m^2 < -d^2/4 + 1$ which allows for two different solutions corresponding to Dirichlet and Neumann boundary conditions [@Breitenlohner:1982bm]. Hence the two AdS theories may be related by a Legendre transformation and correspond to the UV and IR fixed points of two CFTs [@Witten:2001ua; @Klebanov:1999tb]. The scalar fields are given transparent boundary conditions by deforming the theory at the boundary by $$\begin{aligned}
\W = g\,\O_1 \O_2, \label{eq:deformation}
\end{aligned}$$ where $g$ is a dimensionless coupling. Here the operators $\O_i$ are normalized so that their two-point functions are independent of $N_i$. With this marginal deformation the boundary (bulk) theories become mutually interacting. In particular, the propagators of the bulk scalars become [@Aharony:2005sh] $$\begin{aligned}
\Vev{\phi_i\phi_k} = \frac{1}{1+\tilde{g}^2} &\(\begin{array}{rr}
G_1 + \tilde{g}^{2} G_2 & \tilde{g}\,G_1 - \tilde{g}\,G_2 \\
\tilde{g}\,G_1 - \tilde{g}\,G_2 & G_2 + \tilde{g}^{2} G_1
\end{array}\), \label{eq:propagatormatrix}
\end{aligned}$$ where $\tilde{g} = (2\D_1-d)g$. Each propagator in shares the structure of with $\a_1 \a_2 = \pm \tilde{g}^2/ (1+\tilde{g}^2)^2$ and contributes to the mass of the gravitons. If we denote by $\ell$ the radius of AdS and assume Newton’s constant is the same for each AdS$^{(i)}$ theory, $\k^2_i = \k^2 = 8\pi G_N$, the graviton mass matrix is given by $$\begin{aligned}
M^2 = \ell^{-d-1}\k^2\s \(\begin{array}{rr}
1 & -1 \\
-1 & 1
\end{array}\), \label{eq:massmatrix}
\end{aligned}$$ and is obtained by isolating the contribution of the $D(d+1,1)$ bound state to the self-energy of the transverse-traceless modes of the gravitons (see [@Porrati:2001db; @Duff:2004wh] for details). Here $\s$ is a positive number given in [@Aharony:2006hz] by $$\begin{aligned}
\s = -\frac{\tilde{g}^2}{\(1+\tilde{g}^{2}\)^2}\frac{2(4\pi)^{(3-d)/2}}{(d+2)\G\(\frac{d+3}{2}\)}\prod_i\frac{\D_i\G(\D_i)}{\G\(\D_i-\frac{d}{2}\)}.
\end{aligned}$$ Let us omit the unnecessary Lorentz indices and denote by $h_i$ the graviton in AdS$^{(i)}$. Diagonalizing yields a massless graviton $\tilde{h}_{o} = (1/2)^{1/2}(h_1 + h_2)$, and a massive spin-2 field $\tilde{h}_{m} = (1/2)^{1/2}(h_1 - h_2)$ with mass $m^2 = 2 \ell^{-d-1}\k^2\s$. The graviton is dual to the conserved stress-energy tensor of the boundary CFT, $$\begin{aligned}
T_o = t_1 + t_2 - g\eta \,\O_1\O_2 + \dots, \label{eq:tmunu}
\end{aligned}$$ where we have omitted improvement terms that make $T_o$ traceless. Here $\eta$ is the AdS background metric and $t_i$ is the stress-energy tensor of CFT$_i$. On the other hand the massive spin-2 is dual to a linear combination of $t_i$ and $g\eta\,\O_1\O_2$ that is not conserved and orthogonal to $T_o$.
This summarizes the construction of [@Aharony:2006hz; @Kiritsis:2006hy] in the simplest case. We now show that it is possible to obtain AdS/CFT duals with only massive gravitons by letting $\k_1 \ne \k_2$ and taking a limit where either $\k_1$ or $\k_2$ vanishes. To see that the massless graviton decouples let us consider the quadratic Lagrangian for the gravitons in the bulk[^1] $$\begin{split}
\L_2 = - \frac{1}{2}\(h_1 \;\; h_2 \) & \(\begin{array}{cc} \k_1^{-2} \square & 0 \\ 0 & \k_2^{-2}\square \end{array}\) \(\begin{array}{c} h_1 \\ h_2 \end{array}\) \\
- & \frac{\sigma}{2} \(h_1 \;\; h_2 \)\(\begin{array}{rr} 1 & -1 \\ -1 & 1 \end{array}\) \(\begin{array}{c} h_1 \\ h_2\end{array}\), \label{eq:L}
\end{split}$$ where the mass term is given by without the factor of $\k^{2}$ and we have set $\ell = 1$. Here we have greatly simplified the notation so that $h_{(i)} \square h_{(i)}$ stands for the quadratic Lagrangian of a massless graviton in AdS, namely $$\begin{aligned}
h_{(i)}^{\mu\nu}(-2\n^\a\n_\mu\,\d^\b_\nu + \n^2\,\d^\a_\mu\,\d^\b_\nu + \dots) h_{\a\b(i)},
\end{aligned}$$ with $\n_{\a}$ the background covariant derivative; while $\s\,h_{(i)}\,h_{(k)}$ stands for the Fierz and Pauli [@Fierz:1939ix] mass term $$\begin{aligned}
h_{\mu\nu(i)}h^{\mu\nu}_{(k)}-h^{\mu}_{\mu(i)}h^{\nu}_{\nu(k)}
\end{aligned}$$ In terms of the canonically normalized field $H_i = \k_i^{-1}h_i$ the Lagrangian becomes $$\begin{split}
\L_2 = - & \frac{1}{2}\(H_1 \;\; H_2 \)\(\begin{array}{cc} \square & 0 \\ 0 & \square \end{array}\)\(\begin{array}{c} H_1 \\ H_2\end{array}\) \\
- & \frac{\s}{2}\(H_1 \;\; H_2 \)\(\begin{array}{cc} \k_1^2 & -\k_1\k_2 \\ -\k_1\k_2 & \k_2^2 \end{array}\)\(\begin{array}{c} H_1 \\ H_2\end{array}\). \label{eq:canonicalL}
\end{split}$$ Diagonalizing the mass term then yields $$\begin{aligned}
\L_2 = - \frac{\k_1^2 + \k_2^2}{4\k_1^2\k_2^2}\,\tilde{h}_o \square \tilde{h}_o - \frac{1}{\k_1^{2} + \k_2^{2}} \,\tilde{h}_m \square \tilde{h}_m - \s \tilde{h}_m^2 \label{eq:diagonalL}
\end{aligned}$$ where $\tilde{h}_o$ and $\tilde{h}_m$ are given by $$\begin{aligned}
\tilde{h}_o & \equiv \frac{\sqrt{2}}{\k_1^2 + \k_2^2} \( \k_2^2\,h_1 + \k_1^2\,h_2 \) \\
\tilde{h}_m & \equiv \frac{1}{\sqrt{2}}\( h_1 - h_2 \).
\end{aligned}$$ It is clear from that for $\k_1 = \k_2$ there is no limit in which only $\tilde{h}_o$ decouples. With $\k_1 \ne \k_2$ taking the $\k_i \ra 0$ limit for either value of $i$, say $i = 1$, leads to a theory with only a massive graviton. Its mass is now given by $m^2 = \k_2^2\s$ and is half of that of the theory with $\k_1 = \k_2$. Now recall that in the AdS/CFT correspondence Newton’s constant is related to $N_i$ by $$\begin{aligned}
G_N^{(i)} \propto \ell^{d-1}(1/N_i)^p,
\end{aligned}$$ where $p = 1\,(2)$ for a dual CFT living in the vector (adjoint) representation. Taking the $\k_i \rightarrow 0$ limit in the bulk corresponds to taking the $N_i\rightarrow \infty$ limit in the boundary. Thus the dual theory at the boundary is given by the product of CFTs with deformation where $N_1 \ra \infty$ and $N_2$ is large but finite.
The decoupling of the massless graviton in the bulk is equivalent to the vanishing of all correlators, other than the two-point function, of the boundary stress-energy tensor. To see this recall that in the AdS/CFT correspondence the on-shell partition function in the bulk is the generating functional of operators in the dual CFT. In particular, two-point functions are determined by the quadratic action. Let us denote by $T_o$ and $T_m$ the operators dual to $\tilde{h}_o$ and $\tilde{h}_m$, respectively. Since $\k^2_i \propto \ell^{d-1}(1/N_i)^p$, inspection of yields two-point functions with the following dependence on $N_i$ when $N_i$ is large $$\begin{aligned}
& \Vev{T_o\,T_o} \simeq N_1^p + N_2^p, & \Vev{T_m\,T_m} & \simeq \frac{(N_1 N_2)^p}{N_1^p + N_2^p}.
\end{aligned}$$ This can be verified by inspection of the two-point functions using and the definition of $T_m$ given by eq. (2.9) of [@Aharony:2006hz], with the latter normalized by $(N_1^p + N_2^p)^{1/2}$.
Normalizing $T_o$ and $T_m$ so that their two-point functions are independent of $N_i$ yields, in the limit $N_1 \ra \infty$, $$\begin{aligned}
& T_o \ra (1/N_1)^{p/2}\,T_o, & T_m \ra (1/N_2)^{p/2}\,T_m.
\end{aligned}$$ With all other operators $\J_i \in$ CFT$_i$ normalized in a similar fashion, $$\begin{aligned}
& \J_i \ra (1/N_i)^{p/2}\J_i, & \O_i \ra (1/N_i)^{p/2}\O_i,
\end{aligned}$$ the $n$-point functions of $T_o$ scale with $N_1$ as follows $$\begin{aligned}
\big\langle T_o\,\F[T_o,T_m,\J_i,\O_i]\big\rangle \simeq \frac{1}{N_1^q},
\end{aligned}$$ where $\F$ is any polynomial of degree $n-1$ and $q > 0$ for $n > 2$. Hence the $n$-point functions of $T_o$ vanish in the $N_1 \ra \infty$ limit. On the other hand correlation functions of $T_m$ are finite with all terms of $\O(g^3)$ vanishing.
It is not difficult to find and example where this construction is applicable. Consider again the correspondence between higher spin theories in AdS$_4$ and $O(N)$ vector models on the boundary. As mentioned earlier there are two faces to the correspondence. On the one hand we have Vasiliev’s theory with massless gauge fields of even spin on the bulk and the free vector model on the boundary. This theory contains a scalar in the $D(1,0)$ representation of $SO(3,2)$. Accordingly its boundary dual, $\O_1$, has dimension $\Delta_1 = 1$. On the other hand the scalar in Vasiliev’s theory with massive gauge fields lives in the $D(2,0)$ representation of AdS$_4$; its dual operator $\O_2$ has dimension $\Delta_2 = 2 + \O(1/N_2)$. Since $\Delta_1 + \Delta_2 = 3$, the product of the two vector models with deformation makes one of the bulk gravitons massive. By sending $N_1 \ra \infty$ with $N_2$ large but finite we obtain an explicit example of a correspondence with only massive gravitons in the bulk.
In this Letter we have considered the simplest deformation of the product of two CFTs. In general there may be $\rho$ operators in CFT$_{i}$ of dimension $\Delta_i$ denoted by $\O^{(a)}_i$. In this case eq. may be generalized to $$\begin{aligned}
\W = \sum_{a,b}^\rho g_{ab}\, \O^{(a)}_1\O^{(b)}_2. \label{eq:newdef}
\end{aligned}$$ If for simplicity we take $g_{ab}$ to be symmetric it may be diagonalized so that $\W$ becomes $$\begin{aligned}
\W = \sum_{a}^\rho \l_a \, \widetilde{\O}^{(a)}_1 \widetilde{\O}^{(a)}_2,
\end{aligned}$$ where $\l_a$ are the eigenvalues of $g_{ab}$ and $\widetilde{\O}_i^{(a)}$ is a linear combination of $\O^{(a)}_i$. The pair of operators $\{\widetilde{\O}_1^{(a)}, \widetilde{\O}_2^{(a)}\}$ changes the boundary conditions of the corresponding pair of dual scalars. Thus each pair of scalars contributes to the mass of the graviton and their propagators are given by with $\tilde{g} \ra \tilde{\l}_a = (2\Delta_1 -d)\l_a$. With $N_1 \ra \infty$ and $N_2$ large but finite, the mass of the graviton is given by $m^2 = \k^2_2\, \tilde{\s}$ where $$\begin{aligned}
\tilde{\s} = \sum_a \frac{-\tilde{\l}_a^2}{(1 + \tilde{\l}_a^{2})^2} \frac{2(4\pi)^{(3-d)/2}}{(d+2)\G\(\frac{d+3}{2}\)}\prod_i\frac{\D_i\G(\D_i)}{\G\(\D_i-\frac{d}{2}\)}, \label{eq:newmass}
\end{aligned}$$ and $\tilde{\l}_a^2/(1 + \tilde{\l}_a^{2})^2$ has a maximum at $\tilde{\l}_a = \pm 1$.
Generically, for $g_{ab} \sim \O(1)$ deformation leads to a graviton mass smaller than that induced by . It is possible to choose $g_{ab}$ such that $\tilde{\l}_a \sim \O(1)$. In this case leads to an enhancement of the graviton mass by a factor of $\rho$. On the other hand the presence of $\rho$ species of scalars in the bulk will contribute to the wave function renormalization of the gravitons. This will reduce Newton’s constant $G^{(i)}_N$ by a factor of $\rho^{-1}$ [@Dvali:2007hz]. Thus the more general case given in may reduce the mass of the graviton but leaves the main result unchanged.
We have seen that it is possible to have a correspondence between a theory in AdS with only massive gravitons and a CFT on the boundary. The construction of [@Aharony:2006hz; @Kiritsis:2006hy] was necessary so that (1) the field dual to the massive graviton is some operator other than the conserved stress-energy tensor of the CFT; and (2) there is a limit in which the massless graviton decouples. Using the methods of refs [@Giombi:2009wh; @Giombi:2010vg; @Giombi:2011ya] it should be possible to check explicitly whether Vasiliev’s theory with massive gravitons is dual to the product of vector models with deformation where we take either $N_1$ or $N_2$ to infinity.
We would like to thank Sergei Dubovsky and Matthew Kleban for useful comments. M.P. is supported in part by NSF grant PHY-0758032, and by ERC Advanced Investigator Grant No. 226455 [*Supersymmetry, Quantum Gravity and Gauge Fields (Superfields)*]{}.
[137]{}
J. M. Maldacena, Adv. Theor. Math. Phys. [**2**]{}, 231 (1998) \[Int. J. Theor. Phys. [**38**]{}, 1113 (1999)\] \[hep-th/9711200\]. E. Witten, Adv. Theor. Math. Phys. [**2**]{}, 253 (1998) \[hep-th/9802150\]. S. S. Gubser, I. R. Klebanov and A. M. Polyakov, Phys. Lett. B [**428**]{}, 105 (1998) \[hep-th/9802109\]. I. R. Klebanov and A. M. Polyakov, Phys. Lett. B [**550**]{}, 213 (2002) \[hep-th/0210114\]. E. Sezgin and P. Sundell, Nucl. Phys. B [**644**]{}, 303 (2002) \[Erratum-ibid. B [**660**]{}, 403 (2003)\] \[hep-th/0205131\]. M. A. Vasiliev, Phys. Lett. B [**285**]{}, 225 (1992). M. A. Vasiliev, Int. J. Mod. Phys. D [**5**]{}, 763 (1996) \[hep-th/9611024\]. K. G. Wilson and J. B. Kogut, Phys. Rept. [**12**]{}, 75 (1974). E. Brezin and D. J. Wallace, Phys. Rev. B [**7**]{}, 1967 (1973).
L. Girardello, M. Porrati and A. Zaffaroni, Phys. Lett. B [**561**]{}, 289 (2003) \[hep-th/0212181\]. S. Giombi and X. Yin, JHEP [**1009**]{}, 115 (2010) \[arXiv:0912.3462 \[hep-th\]\]. S. Giombi and X. Yin, arXiv:1105.4011 \[hep-th\]. M. Porrati, JHEP [**0204**]{}, 058 (2002) \[hep-th/0112166\]. O. Aharony, A. B. Clark and A. Karch, Phys. Rev. D [**74**]{}, 086006 (2006) \[hep-th/0608089\]. E. Witten, hep-th/0112258. E. Kiritsis, JHEP [**0611**]{}, 049 (2006) \[hep-th/0608088\]. P. Breitenlohner and D. Z. Freedman, Phys. Lett. B [**115**]{}, 197 (1982). I. R. Klebanov and E. Witten, Nucl. Phys. B [**556**]{}, 89 (1999) \[hep-th/9905104\]. O. Aharony, M. Berkooz and B. Katz, JHEP [**0510**]{}, 097 (2005) \[hep-th/0504177\]. M. J. Duff, J. T. Liu and H. Sati, Phys. Rev. D [**69**]{}, 085012 (2004) \[hep-th/0207003\]. M. Fierz and W. Pauli, Proc. Roy. Soc. Lond. A [**173**]{}, 211 (1939). G. Dvali, Fortsch. Phys. [**58**]{}, 528 (2010) \[arXiv:0706.2050 \[hep-th\]\]. S. Giombi and X. Yin, JHEP [**1104**]{}, 086 (2011) \[arXiv:1004.3736 \[hep-th\]\].
[^1]: Note that there maybe corrections to the kinetic terms that render the corresponding matrix non-diagonal. These corrections are subleading in $1/N$ so they may be ignored.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'M. Hochster defines an invariant namely $\Theta(M,N)$ associated to two finitely generated module over a hyper-surface ring $R=P/f$, where $P=k\{x_0,...,x_n\}$ or $k[X_0,...,x_n]$, for $k$ a field and $f$ is a germ of holomorphic function or a polynomial, having isolated singularity at $0$. This invariant can be lifted to the Grothendieck group $G_0(R)_{\mathbb{Q}}$ and is compatible with the chern character and cycle class map, according to the works of W. Moore, G. Piepmeyer, S. Spiroff, M. Walker. They prove that it is semi-definite when $f$ is a homogeneous polynomial, using Hodge theory on Projective varieties. It is a conjecture that the same holds for general isolated singularity $f$. We give a proof of this conjecture using Hodge theory of isolated hyper-surface singularities when $k=\mathbb{C}$. We apply this result to give a positivity criteria for intersection multiplicty of proper intersections in the variety of $f$.'
address: 'Centro de Investigacion en Matematicas, A.C.\'
author:
- Mohammad Reza Rahmati
title: 'Positivity of Hochster Theta over $\mathbb{C}$'
---
Introduction {#introduction .unnumbered}
============
One of the important ways to generalize known facts about smooth algebraic varieties in algebraic geometry, is to try to extend them over singular varieties. Most of the valid methods in the smooth category are hard to be worked out in the singular category. An example of this is the definition of intersection multiplicity as the $Tor$-formula in algebraic geometry as an Euler characteristic. The Riemann-Roch formula provides a definition of this characteristic class by the chern character map on the $K$-group. The useful observation here is the intersection theory is in fact a theory on $K$-groups. In the singular set up both of the definitions of $K$-groups and the chern character become very complicated in the first glance. although trying to generalize the homology theory over singular varieties provides some type of solution for this case, trying to extend the natural concepts from the smooth category to the singular one remains still complicated and difficult.
In this short note we provide a very basic example of this situation and try to approach to the solution using asymptotic Hodge theory. In this way we employ some standard methods in the extensions of polarized mixed Hodge structure to answer questions about singularities in algebraic geometry. Thus the reader may divide the text into three part, the first is a singular set up in algebraic geometry, the second is on variation of mixed Hodge structures, and the third is an application of the results mentioned in part 2.
Modules over Hypersurface rings
===============================
A hyper-surface ring is a ring of the form $R:=P/(f)$, where $P$ is an arbitrary ring and $f$ a non-zero divisor. Localizing we may assume $P$ is a local ring of dimension $n+1$. As according to the title we assume $P=\mathbb{C}\{x_0,...,x_n\}$ and $f$ a holomorphic germ, or $P=\mathbb{C}[x_0,...,x_n]$ and then $f$ would be a polynomial. Then we are mainly interested to study finitely generated modules over these rings. Consider $f:\mathbb{C}^{n+1} \to \mathbb{C}$ in this form, and choose a representative for the Milnor fibration as $f:X \to T$, where $T$ is the disc around $0$.
*Then, through all the rest of this text we assume $0 \in \mathbb{C}^{n+1}$ is the only singularity of $f$*.
A matrix factorization of $f$ in $P$ is a pair of matrices $A$ and $B$ such that $AB=BA=f.\ id$. It is equivalent to the data of a pair of finitely generated free $P$-modules
$$d_0:X^0 \leftrightarrows X^1:d_1 , \qquad d_0d_1=d_1d_0=f. \ id$$
It is a basic fact, discovered by D. Eisenbud, that the $R$-modules have a minimal resolution that is eventually 2-periodic. Specifically, in a free resolution of such a module $M$, we see that after n-steps we have an exact sequence of the following form.
$$0 \to M' \to F_{n-1} \to F_{n-2} \to ... \to F_0 \to M \to 0$$
where the $F_i$ are free $R$-modules of finite rank and $depth_R(M')=n$. If $M'=0$ then $M$ has a free resolution of finite length., If $M' \ne 0$, then $M'$ is a maximal Cohen-Macaulay module, that is $depth_R(M')=n$. So ”up to free modules” any $R$-module can be replaced by a maximal Cohen-Macaulay module. If $M$ is a maximal Cohen-Macaulay $R$-module that is minimally generated by $p$ elements, its resolution as $P$-module has the form
$\begin{array}[c]{ccccccccc}
0 &\rightarrow &P^p &\stackrel{A}{\rightarrow} &P^p &\rightarrow &M &\rightarrow& 0\\
&&\downarrow&\stackrel{B}{\swarrow}&\downarrow&&\downarrow{0}&&\\
0 &\rightarrow &P^p &\stackrel{A}{\rightarrow} &P^p &\rightarrow &M &\rightarrow& 0
\end{array}$
where $A$ is some $p \times p$ matrix with $det(A)=f^q$. The fact that multiplication by $f$ acts as $0$ on $M$ produces a matrix $B$ such that $A.B=B.A=f.I$, where $I$ is the identity matrix. In other words, we find a matrix factorization $(A,B)$ of $f$ determined uniquely up to base change in the free module $P^p$, by $M$. The matrix factorization not only determines $M$ but also a resolution of $M$ as $R$-module.
$....\to R^p \to R^p \to R^p \to M \to 0$.
So a minimal resolution of $M$ looks in general as follows
$... \to G \to F \to G \to F_{n-1} \to ... \to F_0 \to M \to 0$.
As a consequence all the homological invariants like $Tor_k^R(M,N), Ext_R^k(M,n)$ are 2-priodic, [@BVS], [@EP].
The category of matrix factorizations of $f$ over $R$, namely $MF(R,f)$; is defined to be the differential $\mathbb{Z}/2$-graded category, whose objects are pairs $(X,d)$, where $X=X^0 \oplus X^1$ is a free $\mathbb{Z}/2$-graded $R$-module of finite rank equipped with an $R$-linear map $d$ of odd degree satisfying $d^2=f.\ id_X$. Here the degree is calculated in $\mathbb{Z}/2$. Regarding to the first definition
$$d= \left(
\begin{array}{cc}
0 & d_0\\
d_1 & 0
\end{array} \right), \qquad d^2=f.\ id.$$
The morphisms $MF(X,X')$ are given by $\mathbb{Z}/2$-graded $R$-module maps from $X$ to $X'$ (or equivalent between the components $X^0$ and $X^1$) provided that the differential is given by
$$d(f)=d_{X'} \circ f - (-1)^{\mid f \mid} f \circ d_X.$$
Here $d_{X}$ or $d_X^{\prime}$ may be considered as the matrix given above or to be separately $d_0$ and $d_1$, and also it is evident that $d(f)^2=0$. By choosing bases for $X^0$ and $X^1$ we reach to the former definition, [@EP].
Hochster Theta Function
=======================
M. Hochster in his study of direct summand conjecture defined the following invariant for the hypersurface ring $R=P/(f)$, namely $\Theta$-invariant.
(Hochster Theta pairing) The theta pairing of two $R$-modules $M$ and $N$ over a hyper-surface ring $R/(f)$ is
$\Theta(M,N):= l(Tor_{2k}^R(M,N))-l(Tor_{2k+1}^R(M,N)), \qquad k>>0$
The definition makes sense as soon as the lengths appearing are finite. This certainly happens if $R$ has an isolated singular point.
[@BVS] Take $f=xy-z^2, M=\mathbb{C}[[xyz]]/(x,y)$. A matrix factorization $(A,B)$ associated to $M$ is given by\
$$A=\left( \begin{array}{cc}
y & -z \\
-z & x
\end{array} \right), \qquad
B=\left( \begin{array}{cc}
x & z \\
z & y
\end{array} \right).$$
The $Tor_k^R(M,M)$ is the homology of the complex
$.....\to \mathbb{C}[[y]]^2 \to \mathbb{C}[[y]]^2 \to \mathbb{C}[[y]] \to 0$
where\
$$\alpha=\left( \begin{array}{cc}
y & 0 \\
0 & 0
\end{array} \right), \qquad
\beta=\left( \begin{array}{cc}
0 & 0 \\
0 & y
\end{array} \right).$$\
So we find that $\Theta(M,M)=0$.
Hochster theta pairing is additive on short exact sequences in each argument, and thus determines a $\mathbb{Z}$-valued pairing on $G(R)$, the Grothendieck group of finitely generated $R$-modules. One loses no information by tensoring with $\mathbb{Q}$ and so often theta is interpreted as a symmetric bilinear form on the rational vector space $G(R)_{\mathbb{Q}}$. It is basic that Theta would vanish if either $M$ or $N$ be Artinian or have finite projective dimension [@MPSW], [@BVS]. The $\Theta$-invariant has different interpretations as intersection multiplicity in the singular category.
[@BVS] When $M=\mathcal{O}_Y=R/I,N=\mathcal{O}_Z=R/J$, where $Y,Z \subseteq X_0$ are the sub-varieties defined by the ideals $I,J$ respectively, then
$\Theta(\mathcal{O}_Y,\mathcal{O}_Z)=i(0;Y,Z)$
in case that $Y \cap Z={0}$. Here $i(0; , )$ is the ordinary intersection multiplicity in $\mathbb{C}^{n+1}$.
By additivity over short exact sequences and the fact that any module admits a finite filtration with sub-quotients of the form $R/I$, knowing $\Theta(\mathcal{O}_Y,\mathcal{O}_Z)$ determines $\Theta(M,N)$ for all modules $M,N$.
[@BVS] Assume $f \in \mathbb{C}[[x_1,...,x_{2m+2}]$ is a homogeneous polynomial of degree $d$, and $X_0=f^{-1}(0) \in \mathbb{C}^{2m+2}$ and $T=V(f) \in \mathbb{P}^{2m+1}$ the associated projective cone of degree $d$. Let $Y$ and $Z$ be also co-dimension $m$ cycles in $T$. If $Y,Z$ intersect transversely, then
$\Theta(\mathcal{O}_Y,\mathcal{O}_Z)=-\frac{1}{d}[[Y]].[[Z]]$
where $[[Y]]:=d[Y]-deg(Y).h^m$ is the primitive class of $[Y]$, with $h \in H^1(T)$ the hyperplane class.
The primitive class of a cycle $Y$ is the projection of its fundamental class $[Y] \in H^m(T)$ into the orthogonal complement to $h^m$ with respect to the intersection pairing into $H^{2m}(T)=\mathbb{C}$. As $h^m. h^m=d=deg(T)$ and $[Y].h^m=deg(Y)$ the description of the primitive class follows. Substituting the claim can be reformulated
$\Theta(\mathcal{O}_Y,\mathcal{O}_Z)=-\frac{1}{d}[[Y]].[[Z]]=-d[Y].[Z]+deg(Y)deg(Z)$
Where $[Y].[Z]$ denotes the intersection form on the cohomology of the projective space, [@BVS].
When $f$ in consideration is a homogeneous polynomial of degree $d$, such that $X:=Proj(R)$ is a smooth $k$-variety, the Theta pairing is induced, via chern character map, from the pairing on the primitive part of de Rham cohomology
$\displaystyle{\frac{H^{(n-1)/2}(X,\mathbb{C})}{\mathbb{C}. \gamma^{(n-1)/2}} \times \frac{H^{(n-1)/2}(X,\mathbb{C})}{\mathbb{C}. \gamma^{(n-1)/2}} \to \mathbb{C}}$
given by
$(a,b) \to (\int_{X} a \cup \gamma^{(n-1)/2})(\int_{X} a \cup \gamma^{(n-1)/2})-d(\int_{X} a \cup b)$
where $\gamma$ is the class of a hyperplane section and Theta would vanish for rings of this type having even dimensions. When $n=1$ by $\gamma^0$ we mean $1 \in H^0(X,\mathbb{C})$, [@MPSW].
[@MPSW] For $R$ and $X$ as above, if $n$ is odd there is a commutative diagram
$$\begin{CD}
G(R)_{\mathbb{Q}}^{\otimes 2} @<<< \displaystyle{(\frac{K(X)_{\mathbb{Q}}}{\alpha})^{\otimes 2}}\\
@V{\Theta}VV @VV{(ch^{n-1/2})^{\otimes 2}}V\\
\mathbb{C} @<<{\theta}< ( \displaystyle{\frac{H^{(n-1)/2}(X,\mathbb{C})}{\mathbb{C}. \gamma^{(n-1)/2}}} )^{\otimes 2}
\end{CD}$$
[@MPSW] For $R$ and $X$ as above and $n$ odd the restriction of the pairing $(-1)^{(n+1)/2} \Theta $ to
$im(ch^{\frac{n-1}{2}}): K(X)_{\mathbb{Q}}/{\alpha} \to \displaystyle{\frac{H^{(n-1)/2}(X,\mathbb{C})}{\mathbb{C}. \gamma^{\frac{n-1}{2}}}}$
is positive definite. i.e. $(-1)^{(n+1)/2}\Theta(v,v) \geq 0$ with equality holding if and only if $v=0$. In this way $\theta$ is semi-definite on $G(R)$.
[@MPSW] Define
$W=H^{n-1}(X(\mathbb{C}),\mathbb{Q}) \cap H^{\frac{n-1}{2},\frac{n-1}{2}}(X(\mathbb{C}))$.
It is classical that the image of $ch^{(n-1)/2}$ is contained in $W$. Define $e:W/\mathbb{Q}. \gamma^{(n-1)/2} \hookrightarrow H^{n-1}(X,\mathbb{Q})$ by
$e(a)=a-\displaystyle{\frac{\int_X a \cup \gamma^{(n-1)/2}}{d}. \gamma^{(n-1)/2}} \in W$.
We know that $\theta(a,b)=-d.I^{coh}(e(a),e(b))$ Now the theorem follows from the polarization properties of cup product on cohomology of projective varieties.
Hodge theory and Residue form
=============================
Asuume $f:\mathbb{C}^{n+1} \to \mathbb{C}$ is a germ of isolated singularity. We choose a representative $f:X \to T$ over a small disc $T$ according to the milnor fibration theorem. It is possible to embed the Milnor fibration $f:X \to T$ into a compactified (projective) fibration $f_Y : Y \to T$ such that the fiber $Y_t$ sits in $\mathbb{P}^{n+1}$ for $t \neq 0$. The projective fibration $f_Y$ has a unique singularity at $0 \in Y_0$ over $t = 0$. Then, there exists a short exact sequence
$$0 \to H^n(Y_0,\mathbb{Q}) \to H^n(Y_t,\mathbb{Q}) \stackrel{i^*}{\rightarrow} H^n(X_t,\mathbb{Q}) \to 0 , \qquad t \neq 0.$$
We have $H^n(Y_0,\mathbb{Q})=\ker(M_Y -id)$, by the invariant cycle theorem, where $M_Y$ is the monodromy of $f_Y$. The form $S_Y :=(-1)^{n(n-1)/2} I_Y^{coh} : H^n(Y_t,\mathbb{Q}) \times H^n(Y_t,\mathbb{Q}) \to \mathbb{Q}$ is the polarization form of pure Hodge structure on $H^n(Y_t,\mathbb{C})$, $t \in T^{\prime}$ (the punctured disc). W. Schmid (resp. J. Steenbrink) has defined a canonical MHS on $H^n(Y_t,\mathbb{Q})$ (resp. on $H^n(Y_t,\mathbb{Q})$) namely limit MHS, which make the above sequence an exact sequence of MHS’s.
In the short exact sequence, the map $i^*$ is an isomorphism on $H^n(Y_t,\mathbb{Q})_{\neq 1} \to H^n(X_t,\mathbb{Q})_{\neq 1}$ giving $S=(-1)^{n(n-1)/2} I^{coh} =(-1)^{n(n-1)/2} I_Y^{coh}= S_Y$ on $H^n(X_t,\mathbb{Q})_{\neq 1}$.\
The above short exact sequence restricts to the following one,
$$0 \to \ker \{N_Y:H^n(Y_t,\mathbb{Q})_{1} \to H^n(Y_t,\mathbb{Q})_{1}\} \to H^n(Y_t,\mathbb{Q})_{1} \to H^n(X_t,\mathbb{Q})_{1} \to 0$$
So $a,b \in H^n(X_t,\mathbb{Q})_{1}$ have pre-images $a_Y , b_Y \in H^n(Y_t,\mathbb{Q})_{1}$ and $$S(a,b)=S_Y(a_Y,(-N_Y)b_Y)$$
is independent of the lifts of $a_Y , b_Y$, by the fact that $N_Y$ is an infinitesimal isometry for $S_Y$. The equation (10) defines the desired polarization on $H^n(X_t,\mathbb{Q})_1$. The polarization form $S$ is $M$-invariant, non-degenerate, $(-1)^n$-symmetric on $H^n(X_t,\mathbb{Q})_{\neq 1}$ and $(-1)^{n+1}$-symmetric on $H^n(X_t,\mathbb{Q})_{1}$, [@H1].
Suppose,
$H^n(X_{\infty}, \mathbb{C})= \displaystyle{\bigoplus_{p,q,\lambda}} I^{p,q}_{\lambda}$
is the Deligne-Hodge bigrading, and generalized eigen-spaces of vanishing cohomology, and also $\lambda=\exp(-2\pi i \alpha)$ with $\alpha \in (-1,0]$. Consider the isomorphism obtained by composing the three maps,
$$\Phi_{\lambda}^{p,q}: I^{p,q}_{\lambda} \stackrel{\hat{\Phi}_{\lambda}}{\longrightarrow}Gr_V^{\alpha+n-p}H'' \stackrel{pr}{\longrightarrow} Gr_V^{\bullet} H^{\prime \prime}/\partial_t^{-1}H^{\prime \prime} \stackrel{\cong}{\longrightarrow} \Omega_f$$
where
$\hat{\Phi}_{\lambda}^{p,q}:= \partial_t^{p-n} \circ \psi_{\alpha}| I^{p,q}_{\lambda}$\
$\Phi=\bigoplus_{p,q,\lambda} \Phi_{\lambda}^{p,q}, \qquad \Phi_{\lambda}^{p,q}=pr \circ \hat{\Phi}_{\lambda}^{p,q}$
where $I^{p,q}$ stands for the bigrading, $\partial_t$ is the Gauss-Manin connection and $\psi_{\alpha}$ is the nearby map cf. [@R]. The map $\Phi$ is obviously an isomorphism because both of the $\psi_{\alpha}$ and $\partial_t^{-1}$ are isomorphisms, cf. [@H1], [@V].
(MHS on $\Omega_f$) We define a mixed Hodge structure on $\Omega_f$ using the isomorphism $\Phi$. This means that all the data of the Steenbrink MHS on $H^n(X_{\infty},\mathbb{C})$ such as the $\mathbb{Q}$ or $\mathbb{R}$-structure, the weight filtration $W_{\bullet}\Omega_{f,\mathbb{Q}}$ and the Hodge filtration $F^{\bullet}\Omega_{f,\mathbb{C}}$ is defined via the isomorphism $\Phi$. Specifically; in this we obtain a conjugation map
$$\bar{.}:\Omega_{f,\mathbb{Q}} \otimes \mathbb{C} \to \Omega_{f,\mathbb{Q}} \otimes \mathbb{C}, \qquad \Omega_{f,\mathbb{Q}}:=\Phi^{-1}H^n(X_{\infty},\mathbb{Q})$$
defined from the conjugation on $H^n(X_{\infty},\mathbb{C})$ via this isomorphism.
Recall that the limit (Steenbrink) mixed Hodge structure, is defined by
$$F^p H^n(X_{\infty},\mathbb{C})_{\lambda}=\psi_{\alpha}^{-1}(Gr_V^{\alpha} \partial_t^{n-p} H^{\prime \prime})$$
This justifies the power of $\partial_t^{-1}$ applied in the definition of $\Phi$.
[@R] Assume $f:(\mathbb{C}^{n+1},0) \to (\mathbb{C},0)$, is a holomorphic germ with isolated singularity at $0$. Then, the isomorphism $\Phi$ makes the following diagram commutative up to a complex constant;
$$\begin{CD}
\widehat{Res}_{f,0}:\Omega_f \times \Omega_f @>>> \mathbb{C}\\
@VV(\Phi^{-1},\Phi^{-1})V @VV \times *V \\
S:H^n(X_{\infty}) \times H^n(X_{\infty}) @>>> \mathbb{C}
\end{CD} \qquad \qquad * \ne 0$$
where,
$$\widehat{Res}_{f,0}=\text{res}_{f,0}\ (\bullet,\tilde{C}\ \bullet)$$
and $\tilde{C}$ is defined relative to the Deligne decomposition of $\Omega_f$, via the isomorphism $\Phi$. If $J^{p,q}=\Phi^{-1} I^{p,q}$ is the corresponding subspace of $\Omega_f$, then
$$\Omega_f=\displaystyle{\bigoplus_{p,q}}J^{p,q} \qquad \tilde{C}|_{J^{p,q}}=(-1)^{p}$$
In other words;
$$S(\Phi^{-1}(\omega),\Phi^{-1}(\eta))= * \times \ \text{res}_{f,0}(\omega,\tilde{C}.\eta), \qquad 0 \ne * \in \mathbb{C}$$
The proof of the Theorem 3.2 is a generalization of an argument in [@CIR], in the quasi-homogeneous case.
(Riemann-Hodge bilinear relations for Grothendieck pairing on $\Omega_f$) [@R] Assume $f:\mathbb{C}^{n+1} \to \mathbb{C}$ is a holomorphic germ with isolated singularity. Suppose $\mathfrak{f}$ is the corresponding map to $N$ on $H^n(X_{\infty})$, via the isomorphism $\Phi$. Define
$$P_l=PGr_l^W:=\ker(\mathfrak{f}^{l+1}:Gr_l^W\Omega_f \to Gr_{-l-2}^W\Omega_f)$$
Going to $W$-graded pieces; $$\widehat{Res}_l: PGr_l^W \Omega_f \otimes_{\mathbb{C}} PGr_l^W \Omega_f \to \mathbb{C}$$
is non-degenerate and according to Lefschetz decomposition
$$Gr_l^W\Omega_f=\bigoplus_r \mathfrak{f}^r P_{l-2r}$$
we will obtain a set of non-degenerate bilinear forms,
$$\widehat{Res}_l \circ (id \otimes \mathfrak{f}^l): P Gr_l^W \Omega_f \otimes_{\mathbb{C}} P Gr_l^W \Omega_f \to \mathbb{C},$$
$\widehat{Res}_l=res_{f,0}\ (id \otimes \tilde{C} .\ \mathfrak{f}^l)$
where $\tilde{C}$ is as in 3.2, such that the corresponding hermitian form associated to these bilinear forms is positive definite. In other words,
- $\widehat{Res}_l(x,y)=0, \qquad x \in P_r, \ y \in P_s, \ r \ne s $
- If $x \ne 0$ in $P_l$,
$$Const \times res_{f,0}\ (C_lx,\tilde{C} .\ \mathfrak{f}^l .\bar{x})>0 , \ \ \ \ \ \ \ \ Const \in \mathbb{C}$$
where $C_l$ is the corresponding Weil operator, and the conjugation is as in (7).
Note that the map
$$A_f=\dfrac{\mathcal{O}_X}{\partial f} \to \Omega_f ,\qquad f \mapsto fdx_0...dx_n$$
is an isomorphism. Thus, the above corollary would state similarly for $A_f$.
Chern character (Denis trace map) for isolated hypersurface singularities
=========================================================================
The Hochschild chain complex of $MF(R,f)$ is quasi-isomorphic to the Koszul complex of the regular sequence $\partial_0 f,..., \partial_n f$. In particular the Hochschild homology (and also the Hochschild cohomology) of 2-periodic dg-category $MF(R,f)$ is isomorphic to the module of relative differentials or the Jacobi ring of $f$, [@D].
(T. Dykerhoff) [@D], [@PV] The canonical bilinear form on the Hochschild homology of category of matrix factorizations $\mathcal{C}=MF(P,f)$ of $f$, after the identification $$HH_*MF(P,f) \cong A_f \otimes dx [n]$$
coincides with $$\langle g \otimes dx, h \otimes dx \rangle =(-1)^{n(n-1)/2}res_{f,0}(g,h)$$
The chern character or Denis trace map is a ring homomorphism $$ch:K'_0(X) \to HH_0(X) \cong \Omega_f$$
where $K_0^{\prime}$ is free abelian group on the isomorphism classes of finitely generated modules modulo relations obtained from short exact sequences. The construction of the chern character map or chern classes is functorial w.r.t flat pull back. In the special case of $i:X \hookrightarrow Y$ the compactification, the following diagram commutes,
$$\begin{CD}
K_0^{\prime}(Y_0) @>ch_Y>> HH_0(Y_0) \cong \Omega_f^Y @>\Phi_Y^{-1}>> H^n(Y_{\infty})\\
@Vi^*VV @VVi^*V @VVi^*V\\
K_0^{\prime}(X_0) @>>ch_X> HH_0(X_0) \cong \Omega_f^X @>>\Phi_X^{-1}> H^n(X_{\infty}).
\end{CD}$$
Given a matrix factorization $(A,B)$ for a maximal Cohen-Macaulay $M$, one can find de Rham representatives for the chern classes. Consider $\mathbb{C}[[x_0,...,x_n]]$ as a $\mathbb{C}[[t]]$-module with $t$ acting as multiplication by $f$. Denote by $\Omega^p$ the module of germs of $p$-forms on $\mathbb{C}^{n+1}$, and let $\Omega_f^p=\Omega^p/(df \wedge \Omega^{p-1})$. One puts $\omega(M)=dA \wedge dB$. The components of the chern character
$$ch_M:=tr(\exp(\omega(M)))= \sum_i \frac{1}{i!}{\omega}^i(M)$$
are well-defined classes
$$\omega^i(M)=tr((dA \wedge dB)^i) \in \Omega_f^{2i}/(df \wedge \Omega^{2i-1})$$
There are also odd degree classes
$\eta^i(M):=tr(AdB(dA \wedge dB)^i) \in \Omega_f^{2i+1}/{\Omega_f^{2i}}$.
The group $\Omega_f^{2i+1}/d\Omega_f^{2i}$ can be identified with the cyclic homology $HC_i(P/\mathbb{C}\{t\})$. They fit into the following short exact sequence such that $d\eta^{i-1}=\omega^i(M)$.
$0 \to \Omega_f^{2i-1}/\Omega_f^{2i-2} \to \Omega^{2i}/(df \wedge \Omega^{2i-1}) \to \Omega^{2i}/\Omega^{2i-1} \to 0$.
If the number of variables $n+1$ is even, then a top degree form sits in the Brieskorn module
$\mathcal{H}_f^{(0)}=\Omega^n/(df \wedge d\Omega^{n-1})$
a free $\mathbb{C}[[t]]$-module of rank $\mu$. The higher residue pairing
$$K:\mathcal{H}_f^{(0)} \times \mathcal{H}_f^{(0)} \to \mathbb{C}[t, t^{-1}]$$
of K. Saito can be seen as the de Rham realization of the Seifert form of the singularity, [@BVS].
Positivity of Theta pairing-Main Result
=======================================
The following theorem was conjectured in [@MPSW].
Let $S$ be an isolated hypersurface singularity of dimension $n$. If $n$ is odd, then $(-1)^{(n+1)/2}\Theta$ is positive semi-definite on $G(R)_\mathbb{Q}$, i.e $(-1)^{(n+1)/2}\Theta(M,M) \geq 0$.
By additivity of $\Theta$ on each variable, we may replace $M,N$ by maximal Cohen-Macaulay modules. According to 4.1 and 3.2 the determination of the sign of $\Theta$ amounts to understanding how the image of chern classes look like in the MHS of $\Omega_f$. By Theorem 3.2 it amounts to the same things for the image in $H^n(X_{\infty})$ under the isomorphism $\Phi$. The following diagram is commutative by the functorial properties of chern character.
$$\begin{CD}
K_0^{\prime}(Y_0) @>\Phi_Y^{-1} \circ ch_Y>> H^n(Y_{\infty})\\
@Vi^*VV @VVi^*V\\
K_0^{\prime}(X_0) @>>\Phi_X^{-1} \circ ch_X> H^n(X_{\infty}).
\end{CD}$$
We are assuming that $i^*$ is surjective. By what was said, the chern class we are concerned with, is a Hodge cycle. The commutativity of the above diagram allows us to replace the pre-image of the chern character for $X$, with similar cycle upstairs. Because the polarization form $S_X$ was defined via that of $S_Y$. Thus, if
$$H^n(Y_{\infty})=\oplus_{p+q=n} H^{p,q}$$
is the Hodge decomposition, the only non trivial contribution in the cup product will be for the $H^{n/2,n/2}$, and the polarization form is evidently definite on this subspace (Hodge cycles). Because the map $N_Y$ is of type $(-1,-1)$ for the Hodge structure of $H^n(Y_{\infty})$ and the polarization $S_Y( H^{n/2,n/2}, H^{n/2-1,n/2-1})=0$ for obvious reasons, the corresponding chern class should lie in $H_{\ne 1}^n$. In this way one only needs to prove the positivity statement for Hochster $\Theta$ when the chern character is in $H_{Y,\ne 1}$, and this is the content of Theorem 2.8.
[99]{}
V. Arnold, S. Gusein Zade, A. Varchenko; Singularities of differentiable maps, Vol 2, Monodromy and asymptotics of integrals, 1984
R. O. Buchweitz, D. Van Straten An index theorem for modules on a hypersurface singularity, Moscow Math. Journal, Vol 12, No 2, 2011
O. Celikbas, M. Walker, Hochster theta pairing and algebraic equivalence, Mathematische Annalen June 2012, Volume 353, Issue 2, pp 359-372
A. Chiodo, H. Iritani, Y. Ruan, Landau-Guinzburg/Calabi-Yau correspondence, Global mirror symmetry and Orlov equivalence, Publications mathématiques de l’IHES June 2014, Volume 119, Issue 1, pp 127-216
T. Dyckerhoff Compact generators in category of matrix factorization, Duke Math. J. Volume 159, Number 2 (2011), 223-274.
D. Eisenbud, I. Peeva, Matrix factorizations for complete intersections, and minimal free resolutions, arxiv:math 2013
P. Griffiths , J. Harris. Principles of algebraic geometry. Wiley Classics Library. John Wiley-Sons Inc., New York, 1994. Reprint of the 1978 original.
C. Hertling; Classifying spaces for polarized mixed hodge structures and Brieskorn lattices, Composithio Mathematica 116, 1-37, 1999
W. Moore, G. Piepmeyer, S. Spiroff, M. Walker, Hochster theta pairing and Riemann-Hodge bilinear relations arxiv 0910.1289v2
A. Polishchuk, A. Vaintrob, Chern characters and Hirzebruch-Riemann-Roch formula for matrix factorization, Duke Math. J. Volume 161, Number 10 (2012), 1863-1926.
J. Scherk, J. Steenbrink: On the mixed Hodge structure on the cohomology of the Milnor fibre. Math. Ann. 271 (1985) 641-665
W. Schmid: Variation of Hodge structure: the singularities of the period mapping. Invent, math. 22 (1973) 211-320
\[R\] M. Rahmati, On the mixed Hodge structure associated to isolated hypersurface singularities, arxiv.org, Feb. 2014.
A. Varchenko, On the local residue and the intersection form on the vanishing cohomology, Math. USSR Izv. Volume 26, Number 1, 1986
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
Novel properties arising at interfaces between transition metal oxides, particularly the conductivity at the interface of LaAlO$_{3}$ (LAO) and SrTiO$_{3}$ (STO) band insulators, have generated new paradigms, challenges, and opportunities in condensed matter physics. Conventional transport measurements have established that intrinsic conductivity appears in LAO/STO interfaces when the LAO film matches or exceeds a critical thickness of 4 unit cells (uc). Recently,
a number of experiments raise important questions about the role of the LAO film, the influence of photons, and the effective differences between vacuum/STO and LAO/STO, both above and below the standard critical thickness. Here, using angle-resolved photoemission spectroscopy (ARPES) on *in situ* prepared samples, as well as resonant inelastic x-ray scattering (RIXS), we study how the metallic STO surface state evolves during the growth of a crystalline LAO film. In all the samples, the character of the conduction bands, their carrier densities, the Ti$^{3+}$ crystal fields, and the responses to photon irradiation bear strong similarities. However, LAO/STO interfaces exhibit intrinsic instability toward in-plane
folding of the Fermi surface at and above the 4-uc thickness threshold. This ordering distinguishes these heterostructures from bare STO and sub-critical-thickness LAO/STO and coincides with the onset of unique properties such as magnetism and built-in conductivity.
author:
- 'N. C. Plumb'
- 'M. Kobayashi'
- 'M. Salluzzo'
- 'E. Razzoli'
- 'C. E. Matt'
- 'V. N. Strocov'
- 'K.-J. Zhou'
- 'C. Monney'
- 'T. Schmitt'
- 'M. Shi'
- 'J. Mesot'
- 'L. Patthey'
- 'M. Radovic'
bibliography:
- 'citations.bib'
title: 'From the SrTiO$_{3}$ surface to the LaAlO$_{3}$/SrTiO$_{3}$ interface: How thickness is critical'
---
Unraveling the origins of the electron gases at the STO surface[@Santander-Syro2011; @Meevasana2011; @DiCapua2012] and LAO/STO interfaces[@Ohtomo2004], as well as the mechanisms leading to interesting and useful behaviors such as magnetism[@Brinkman2007; @Kalisky2012; @Salman2012] and persistent photo-enhanced/-induced conductivity[@Ristic2012; @diUccio2012], are crucial steps towards fully exploiting the technological promise[@Mannhart2010] of these systems.
Theories about the existence of the metallic interface in LAO/STO have largely centered around growth-induced defects and doping such as ion intermixing[@Willmott2007] and oxygen vacancies, as well as the polar catastrophe model in which, at the critical thickness of 4 uc[@Thiel2006], carriers transfer across LAO to the STO interface in order to quench the electrostatic potential arising from the growth of polar LAO on nonpolar STO[@Nakagawa2006]. While several arguments can be made in favor of the polar catastrophe model[@Schlom2011], the issue is not yet settled.
Meanwhile the origin of interfacial magnetism in LAO/STO remains an open question as well.
A number of recent observations enrich the story surrounding the origins of conductivity at oxide interfaces and surfaces. For instance, interfacial conducting states have been found in other STO-based heterostructures where no polar discontinuity should occur[@Annadi2012a; @Chen2011b; @Chen2013; @Herranz2012].
These results come amidst the backdrop of the discovery of a metallic state at the surface of bare STO.
Such observations draw attention to the role of the STO substrate and raise the possibility that STOs surface and interface gases are intimately connected; therefore each may be able to teach us about the origins of metallicity and other properties in the other. With this in mind, we set out to investigate the evolution of the low-dimensional metallic state on STO starting from the bare surface and progressing through 4 or more unit cells of LAO over STO.
We begin our study using RIXS to probe the crystal field excitations in bare STO and a sample with 8 uc of LAO grown over STO. As a photon in, photon out technique, RIXS penetrates into the bulk of the sample to probe electron transitions between unoccupied states.
By performing RIXS on resonance with Ti$^{3+}$ absorption features, the technique can be tuned to focus on the surface and interface regions in STO and LAO/STO, respectively, where the low-dimensional electron gases reside[@DiCapua2012; @Plumb2013; @Sing2009; @Cancellieri2013; @Zhou2011]. Figure \[fig:fig1\]a compares RIXS spectra from a bare TiO$_{2}$-terminated STO substrate and one with an LAO film grown on top (8 uc).
In order to selectively probe the conducting states, we performed RIXS using $h\nu=459.2$ eV, corresponding to $2p_{3/2}\text{--}3d\: e_{g}$ absorption by Ti$^{3+}$ sites. Both samples show a peak near the elastic line at an energy loss of -110 meV and a weak shoulder at roughly -200 meV, as well as features centered at approximately -2.5 eV and -1.7 eV. Based on our previous RIXS study of LAO/STO multilayers[@Zhou2011], we attribute the near-elastic features as intra-$t_{2g}$ transitions, the structure at -2.5 eV as $e_{g}\text{--}t_{2g}$ transitions, and the one at -1.7 eV as arising from delocalized electron excitations. Overall, the spectra clearly demonstrate that cubic symmetry is broken[@Zhou2011], and the correspondence between these spectral features in both bare STO and LAO/STO is a clear indication that Ti$^{3+}$ sites in both systems are associated with essentially the same structural distortion most likely the polar Ti-O buckling seen by several studies of LAO/STO interfaces[@Pauli2011; @Cantoni2012] and the bare STO surface[@Bickel1989; @Hikita1993; @Ikeda1999].
Next we studied STO perturbed under 2 uc of LAO film using *in situ* ARPES at conventional UV photon energies. Even though this thickness of LAO/STO
does not have a conducting interface in ambient, dark conditions, scanning tunneling
measurements have successfully been performed on such samples when illuminated, implying the existence of conductivity[@Ristic2012].
Moreover, a Fermi surface appears when the sample is
irradiated in ultrahigh vacuum (UHV, $\sim10^{-11}$ mbar). This behavior is similar to how the electron gas found on bare STO can form under analogous conditions[@Meevasana2011; @Plumb2013]. Fermi surfaces of the bare STO and 2 uc LAO/STO systems obtained using photon energies of 85 eV and 47 eV (corresponding to different cuts through the out-of-plane momentum axis $k_{z}$[@Plumb2013]) are shown in Figs. \[fig:fig1\]b-e. Figs. \[fig:fig1\]f-g show side-by-side comparisons of energy-vs.-$k_{x}$ dispersion data from bare STO and 2 uc LAO/STO obtained at $k_{y}=0$ in the first Brillouin zone using $h\nu=85$ eV. The ellipsoidal components (heavy bands) of the FS extend slightly farther in $k$ than in bare STO (a little over 0.5 $\pi/a$ along the long axes from the Brillouin zone center, instead of $\sim0.4$ $\pi/a$ for bare STO[@Plumb2013]), qualitatively consistent with calculations[@Popovic2008]. The overall nature of the near-$E_{F}$ electronic structure of the LAO/STO system is
similar to the bare surface of STO, whose photon energy dependence in ARPES
is attributed to a combination of confined, non-bulk-like quasi-2D and 3D electronic states associated with Ti $3d_{xy}$ and Ti $3d_{xz}$/$3d_{yz}$ bands, respectively[@Plumb2013]. This naturally suggests that the conducting $3d_{xy}$ and $3d_{xz}$/$3d_{yz}$ electrons comprising the photo-induced metal in the 2 uc LAO/STO interface region have similar spatial distributions to those in STO. Qualitatively, this result is in agreement with several experimental[@Copie2009; @Dubroka2010] and theoretical[@Popovic2008; @Delugas2011; @Stengel2011] studies of thicker ($\geq4$ uc) LAO/STO interfaces, which find that the electron gas actually extends several unit cells into STO. Moreover, the distinct dimensionalities and effective mass characteristics of the $3d_{xy}$ and $3d_{xz}$/$3d_{yz}$ electrons naturally explain the observation of two types of carriers in many experiments[@Zhou2011; @Brinkman2007; @Savoia2009; @BenShalom2010].
Studying the electronic structure of buried interface states presents a significant challenge for photoemission experiments, since the inelastic mean free path
of photoelectrons in a solid is generally considered to be under 1 nm over a broad range of kinetic energies
.
Thus we continued *in situ* ARPES measurements for samples with 4 and 5 uc of LAO by using higher photon energies where resonance enhancement at the Ti $L$ edge occurs,
thereby allowing us to view the momentum-resolved electronic structure associated with Ti.
Figure \[fig:fig2\] presents results from resonant angle-integrated photoemission spectroscopy (resonant PES), absorption measured by the total electron yield (TEY), and core level x-ray photoemission spectroscopy (XPS) performed on 5 uc of LAO grown over STO. The sample underwent post-growth annealing in oxygen ($P_{O_{2}}=200$ mbar, $T=55$ $^{\circ}$C, 1 h). Over the course of the measurement, exposure to synchrotron radiation causes the spectral characteristics of the LAO/STO sample to evolve in a manner indicating that the metallicity at the interface becomes enhanced. In particular, we observe a transfer of spectral weight within the TEY absorption from $ $Ti$^{4+}$ to Ti$^{3+}$ peaks (Fig. \[fig:fig2\]a), concurrent with a Ti$^{4+}$-to-Ti$^{3+}$ redistribution of counts in the Ti $2p$ XPS doublet (Fig. \[fig:fig2\]b).
Complimentary
resonant PES data (Fig. \[fig:fig2\]c–e) were obtained by tuning the photon energy over the full Ti $2p\text{--}3d$ absorption range. The high intensity at the deepest binding energy range of Fig. \[fig:fig2\]e comes from the O $2p$ valence band,
which resonates on both Ti$^{3+}$ and Ti$^{4+}$ energies. This demonstrates a noticeable degree of covalency/hybridization in the Ti-O bonds, which may be relevant to theorectical descriptions of the LAO/STO interface. The features that appear at a binding energy of approximately -1.3 eV come from the “in-gap” states previously observed in the case of both STO single crystals[@Meevasana2011; @Plumb2013] and LAO/STO heterostructures[@Koitzsch2011a; @Ristic2012; @Drera2011], while the spectral weight at the Fermi level indicates that metallicity is present in the sample. Interestingly, the resonance energies of the in-gap states match well with the peak structure of Ti$^{3+}$ (known from LaTiO$_{3}$[@Abbate1991; @Salluzzo2009]), whereas states at the Fermi edge resonate slightly off these energies. We speculate that this slight discrepancy might indicate the existence of different varieties of nominal Ti$^{3+}$ — perhaps located at different depths in the near-interface region and/or with different effective valence Ti$^{3+\delta(z)}$. Example sketches of resonant photoemission processes corresponding with the level structures of the Ti$^{4+}$ and Ti$^{3+}$ states are depicted in Fig. \[fig:fig2\]g. The photon energy ranges where the in-gap and Fermi level states resonate overlap well with the Ti$^{3+}$ features of the TEY absorption spectrum that become enhanced during photon irradiation. This behavior is very similar to the evolution seen on the surface of bare STO during illumination, where the enhancement
of in-gap and metallic states go hand-in-hand[@Meevasana2011; @Plumb2013].
At this point the similarities between the metallic states seen on the surface of irradiated bare STO and at the LAO/STO interface — whether below (2 uc of LAO) or above the standard critical thickness (4–8 uc of LAO) — are striking. All exhibit similar electronic structure, Ti$^{3+}$ crystal fields, correspondence between the existence of in-gap and metallic electronic states, and creation/enhancement of such states via irradiation. However, the Fermi surfaces of LAO/STO where LAO is at least 4 uc thick show a new characteristic not found in thinner samples or on bare STO. Figure \[fig:fig3\] depicts *in situ* resonant ARPES measurements on bare STO and LAO/STO bilayers with LAO thicknesses of 2, 4, and 5 uc. The data were acquired on resonance with Ti$^{3+}$ absorption features. Figure \[fig:fig3\]a shows Fermi surface mapping on bare STO using resonant and conventional ARPES
.
Putting aside
fine details concerning the differences between the Fermi surfaces seen by conventional and resonant ARPES, we notice a substantive change occurring at the standard critical LAO thickness. Although all the studied samples show $1\times1$ ordering of the LAO surfaces (see Supplementary Information), in those where the thickness of LAO is at least 4 uc (Fig. \[fig:fig3\]c–d), the Fermi surface of the interface shows folding along the $k_{x}$ and $k_{y}$ directions — possibly of a twinned $2\times1$ order.
Such folding is not seen for either bare STO or 2 uc LAO/STO (Fig. \[fig:fig3\]a–b). This can also been seen by dispersion cuts along $k_{y}=2\pi/a$ in the second Brillouin zone for each of the Fermi surface maps (Fig. \[fig:fig3\]f–h). We observed the same folding pattern (Fig. \[fig:fig4\]) after annealing the 5 uc LAO/STO sample *in situ* under in high oxygen pressure (200 mbar at 500 C for 1 h). This strongly suggests that the in-plane
ordering is not directly associated with oxygen vacancies or other defects. With this in mind, the simplest interpretation is that a $2\times1$ folding occurs due to twinned rotations of the TiO$_{6}$ octahedra about the [\[]{}100[\]]{}/[\[]{}010[\]]{} axes. The proposed rotation pattern is sketched in Fig. \[fig:fig3\]e, which also qualitatively depicts the ferroelectric Ti-O buckling distortion.
The occurrence of octahedral tilting about [\[]{}100[\]]{}/[\[]{}010[\]]{} or a similar in-plane reconstruction in sufficiently thick LAO/STO interfaces may be connected to unique properties arising in these systems.
In particular, various tilting distortions (sometimes combined with ferroelectric Ti-O buckling) in LAO/STO interfaces might correspond with different charge- and/or magnetically-ordered phases, which may even compete in the ground state[@Zhong2008]. Indeed, the onset of the in-plane
ordering seen here after depositing 4 or more uc of LAO coincides with the appearance of dilute magnetism in samples studied by scanning superconducting quantum interference device microscopy[@Kalisky2012].
Similarly, LAO/STO superlattices were found to exhibit magnetic behavior only when the LAO spacer layers have a minimum thickness of $\sim4\text{--}6$ uc[@Salman2012].
Concerning the origin of built-in conductivity in LAO/STO with at least 4 uc of LAO, it has been shown that the LAO core levels do not exhibit strong thickness dependent shifts or broadening, as would be expected on the basis of the polar catastrophe model[@Segal2009; @Slooten2013].
Furthermore, hole pockets from LAO that should occur at the Brillouin zone corners are absent in our ARPES data, which is in agreement with an
*ex situ* study[@Berner2013]. Since the experiments here are performed on samples not exposed to the atmosphere, arguments in favor of an extrinsic passivation of the surface to explain this discrepancy become less likely. Our results also do not favor a direct connection between oxygen vacancies and the built-in conductivity in LAO/STO. The 5 uc LAO/STO sample shows little qualitative change before and after oxygen annealing (Fig. \[fig:fig4\]), consistent with the idea that oxygen vacancies, while perhaps providing some additional doping, do little to fundamentally alter the interface electron gas[@Basletic2008].
The electronic structure and doping of the metallic state on the surface of bare STO similarly shows no direct link to the concentration of oxygen vacancies[@Santander-Syro2011; @Plumb2013].
While the photoemission experiments do not provide specific confirmation of the origin of the carriers at the interface, they make it clear that distortions in the surface/interface region of STO strongly influence the electronic structure. The polar, ferroelectric-type Ti-O buckling distortion frequently reported on the STO surface and in the interfacial region of STO with LAO is generally thought to play a key role in the spatial confinement of the conducting electrons and compensating the polarity from the LAO
film [@Schwingenschlogl2009; @Stengel2011]. Being common to all these systems, Ti-O buckling is likely to account for the shared signatures of symmetry-breaking in the electronic structure: namely the split $3d_{xy}\text{--}3d_{xz}/3d_{yz}$ conduction bands with indications of both 2D and 3D-like characteristics and the non-cubic crystal field seen in Ti$^{3+}$ RIXS spectra. The polar buckling region should naturally accommodate carriers[@Stengel2011; @Plumb2013], and STO would therefore seem to be a particularly suitable host for low-dimensional electron gases in large part due to its innate propensity for forming this structure near its surface/interface.
It is interesting to consider how $2\times1$ or similar in-plane ordering might be connected to the Ti-O buckling and how, in turn, such ordering might influence the conducting properties of LAO/STO interfaces. Recently, theoretical work has explored the concept of “improper ferroelectricity” in the context of oxide heterostructures, in which ferroelectric displacements occur via coupling to a second order parameter e.g., various forms of octahedral tilting which are not by themselves polar[@Stengel2012]. In principle, even some tilting patterns that in bulk would normally compete with ferroelectric distortions can in certain cases give rise to improper ferroelectricity when the symmetry of the system is broken by an interface[@Stengel2012; @Bousquet2008].
Moreover, it has been proposed that the walls between domains of tilted octahedra (e.g., $2\times1$ twin boundaries) in STO may induce polarization[@Morozovska2012].
Hence the in-plane
ordering might strongly influence the nature and/or strength of polar distortions near the interface, which in turn are closely linked to the extent of the confinement region and thus the transport properties.
Multiple effects (e.g., defects, photons, strain) may trigger or otherwise alter the surface/interface structure, thereby affecting the conductivity or other properties. The role of photons in inducing/enhancing the surface and interface conductivity of STO and LAO/STO, respectively, is particularly
intriguing. The appearance and/or increase of spectral weight at $E_{F}$ and the enhanced signatures of Ti$^{3+}$ in the core level and absorption spectra appear to be long-lived; we can move the synchrotron beam away from a heavily exposed spot for hours, then go back, and find the spectral changes largely intact. This is consistent with the persistent photo-induced changes seen in the electronic structure of the STO surface[@Plumb2013] and in the transport behavior of LAO/STO and related interfaces[@diUccio2012]. The extremely long
lifetimes of the changes imply that photons do more than merely generate carriers by exciting electrons to the conduction band. We therefore propose that light can directly and/or indirectly play some role in altering the structure of the surface/interface region of STO in terms of the polar Ti-O buckling[@Plumb2013], $2\times1$ or other ordering, and/or the domain structure all of which could impact the conductivity and spectroscopic properties
as discussed above.
Whatever the origin of the carriers themselves on bare STO and in LAO/STO, the results here emphasize the important role of the near-surface/-interface structure of STO for the realization of a confined electron gas. Indeed, if there is one unifying characteristic of the STO surface and LAO/STO interfaces, it is that conductivity appears to be strongly linked to the existence of ferroelectric Ti-O buckling in the surface/interface region of STO[@Plumb2013; @Pauli2011; @Stengel2011; @Cantoni2012; @Salluzzo2013]. The similar Ti$^{3+}$ crystal fields and near-$E_{F}$ electronic structure of bare STO and LAO/STO illustrate the intimate connection between these systems owing to the distortions in their metallic regions. These commonalities, however, belie important differences between the various STO-based systems, and the in-plane
ordering seen in the 4- and 5-uc thick LAO/STO interfaces may be an important new insight into how properties such as built-in conductivity and weak/dilute magnetism emerge
above the 4-uc threshold.
Acknowledgments {#acknowledgments .unnumbered}
---------------
The resonant ARPES and RIXS experiments were performed at the Advanced Resonant Spectroscopies (ADRESS) beamline of the Swiss Light Source (SLS) at the Paul Scherrer Institut in Villigen, Switzerland. The SAXES instrument used for the RIXS experiment was jointly built by the Paul Scherrer Institut and Politecnico di Milano, Italy. The high-resolution ARPES measurements were carried out at the Surface/Interface Spectroscopy (SIS) beamline of the SLS. We are grateful for especially valuable conversations with H. Dil and F. Miletto-Granozio. We also shared informative discussions with R. Claessen, M. Sing, and G. Berner, C. Cancellieri, M. L. Reinle-Schmitt, and C. W. Schneider.
Author contributions {#author-contributions .unnumbered}
--------------------
M. R. conceived the project. N. C. P., M. Salluzzo, E. R., M. K., C. E. M., C. M., K.-J. Z. and M. R. collected data. All authors discussed the results. N. C. P. and M. R. wrote the paper with valuable input from all the coauthors.
{width="70.00000%"}
{width="70.00000%"}
{width="100.00000%"}
![\[fig:fig4\]**Comparison of Fermi surfaces of LAO/STO before (left) and after (right) annealing in oxygen atmosphere.** The data were obtained using $h\nu=$465.3 eV. The arrows point to signatures of an in-plane reconstruction. ](fig4_v2){width="1\columnwidth"}
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We study the surface shape of two-dimensional piles using experiments and a continuum theory for surface flows of granular materials (the BCRE equations). We first obtain an analytical solution to the BCRE equations with a simple transformation and show that the surface shapes thereby predicted are in good agreement with the experimental results. By means of such an analytical solution, we find that the formation of the curved tails at the bottom of such piles depends not only on the properties of the granular materials but also on the drift velocity of the grains within the rolling layer.'
author:
- |
Bo Jin$^{\dag}$, $^{1,2}$ Fang Xu$^{\ddag}$ $^{1,2}$ and Hern$\acute{{\mathrm a}}$n A. Makse$^{\S}$ $^{1,3}$\
[, [$^{2}$Department of Chemical Engineering]{}, [$^{3}$Physics Department]{}]{}\
[, [New York, NY 10031]{}]{}\
\
\
\
title: 'Surface Shape of Two-Dimensional Granular Piles'
---
Granular materials have received increasing attention in recent years on account of the fascinating granular flow patterns and their important practical applications. It is well-known that the angle of repose, a characteristic property of granular materials, is determined by measuring the steepest angle of the surface of granular piles. However, the surface of such piles, formed by pouring granular materials on a horizontal table, is not always the straight line implied by the constant angle of repose, in the sense that a curved tail is often present at the bottom of the piles. Alonso and Herrmann [@Herrmann] first investigated the shape of such a curved tail of a two-dimensional static sandpile. By introducing the notion of kinks, they built a model to describe the local slope of the two-dimensional sandpile surface (cf. Eq. (1) in Ref. [@Herrmann]), and then, using the observed translational invariance, derived the following expression for the shape of the tail, $$x=\frac{h_{m}-h(x)}{\theta_{r}}+l_{e}\ln\frac{h_{m}}{h(x)}, \label{eq: aH}$$ where $\theta_{r}\equiv \tan\theta_{c}$ with $\theta_{c}$ being the angle of repose, $h(x)$ is the height of the sandpile surface, $x$ represents the corresponding horizontal coordinate, $h_{m}$ is the maximum height of the sandpile and $l_{e}$$(\equiv l/r)$ is a fitting parameter with $l$ being the horizontal length of the kink (which typically is of the size of a grain) and $r$ the corresponding accumulating rate of the grains on the kinks. The above expression shows that the existence of tails introduces a logarithmic correction to the straight slope given by the angle of repose, and can be used to fit the experimental surface shape quite accurately by adjusting the magnitude of $l_{e}$. However, the derivation of Eq. (\[eq: aH\]) entails three intermediate parameters, i.e. $\rho$, $l$ and $r$ (where $\rho$ is the density of kinks at height $h$), but none of them can be determined experimentally(cf. Ref. [@Herrmann] for details), so that the final fitting parameter $l_{e}$ cannot clearly identify which factors have an effect on the formation of the curved tail at the bottom of the pile.
In order to answer this query, we study the same problem described above on the basis of a continuum theory of surface flows of granular materials proposed by Bouchaud, Cates, Ravi Prakash and Edwards (the BCRE model) [@Edwards]. This model makes use of two coupled variables to describe the dynamics of the two-dimensional sandpile surface: the height of the static sandpile, $h(x, t)$, and the thickness of the rolling grain layer, $R(x, t)$, both of which are a function of the horizontal coordinate $x$ and the time $t$. On assuming that the drift velocity, $v$, of the grains within the rolling layer is independent of $x$, and that diffusion effects are negligible, the original governing equation of motion for the rolling grains in the BCRE model can be simplified into $$\frac{\partial R(x, t)}{\partial t} =-v\frac{\partial R(x, t)}{\partial x}+ \Gamma(R, h), \label{eq: 1}$$ where the interaction term $\Gamma$ accounts for the conversion of immobile grains into rolling grains, and vice versa. Given that the deviation of the local surface slope $\theta(\equiv-\partial h(x, t)/\partial x)$ of the pile from the critical value $\theta_{r}(\equiv \tan\theta_{c}$ with $\theta_{c}$ being the angle of repose) is everywhere small, the simplest form of $\Gamma$ is, to a first approximation, $$\Gamma(h, R) = -\gamma(\theta_{r}-\theta)R(x, t)\equiv -\gamma \left(\theta_{r}+\frac{\partial h(x, t)}{\partial x}\right) R(x, t), \label{eq: 2}$$ where $\gamma$ is a positive constant relative to the property of granular materials, and can be interpreted as equal to the rate of collision between rolling and static grains. Moreover, the principle of mass conservation requires that $$\frac{\partial h(x, t)}{\partial t}=-\Gamma.$$
In what follows, we shall first directly apply the simplified BCRE model, i.e. Eqs. (\[eq: 1\]) and (\[eq: 2\]) (henceforth referred to as the BCRE equations), to investigate the surface shape of two-dimensional piles.
As a granular pile reaches steady growth, the shape of the pile is found to be essentially independent of time ($t$), i.e. the front of the pile propagates along the $x$ direction with a constant velocity ($u$). Therefore, we can combine the two independent variables, $x$ and $t$, into one (say $z$) by means of the following transformation, $$z=x-ut. \label{eq: tran}$$ Consequently, the BCRE Eqs. (\[eq: 1\]) and (\[eq: 2\]) become $$-u\frac{ {\mathrm d} R(z)}{ {\mathrm d} z} + v\frac{{\mathrm d} R(z)}{{\mathrm d} z}= u\frac{{\mathrm d} h(z)}{{\mathrm d} z}, \label{eq: 1t}$$ and $$-u\frac{{\mathrm d} h(z)}{{\mathrm d} z} = \gamma R(z) \left(\theta_{r} + \frac{{\mathrm d} h(z)}{{\mathrm d} z}\right). \label{eq: 2t}$$ We can rearrange Eq. (\[eq: 1t\]) to yield $$\frac{{\mathrm d} h(z)}{{\mathrm d} z} = \left(\frac{v-u}{u}\right) \frac{{\mathrm d} R(z)}{{\mathrm d} z}, \label{eq: RH}$$ which, when substituted into Eq. (\[eq: 2t\]), yields an ordinary differential equation for the thickness of the rolling grain layer $R(z)$, $$(v-u) \left(1 + \frac{\gamma}{u} R(z)\right)\frac{{\mathrm d} R(z)}{{\mathrm d} z} = - \gamma \theta_{r} R(z). \label{eq: eR}$$ We integrate Eq. (\[eq: eR\]) and, after applying the boundary condition which determines the flux of pouring grains (i.e. $R=R_{m}$ at $z=0$), obtain the expression for $R(z)$, $$z \theta_{r} = \frac{v-u}{\gamma}\ln\frac{R_{m}}{R(z)}+\frac{v-u}{u}\left(R_{m}-R(z)\right), \label{eq: R1}$$ where $R_{m}$ is the maximum thickness of the rolling layer.
Similarly, on integrating Eq. (\[eq: RH\]) subject to the the boundary condition ($h=0$ at $R=0$), we find the relationship between $R(z)$ and $h(z)$, i.e. $$R(z) = \frac{u}{v-u} h(z). \label{eq: RHs}$$ Substituting Eq. (\[eq: RHs\]) into Eq. (\[eq: R1\]), we therefore obtain the expression for $h(z)$, $$z = \frac{h_{m}-h(z)}{\theta_{r}} + \frac{v-u}{\theta_{r}\gamma}\ln\frac{h_{m}}{h}, \label{eq: H}$$ where $h_{m}$ is the maximum height of the static pile which must satisfy $$h_{m} = \frac{v-u}{u} R_{m}. \label{eq: r1}$$
Actually, the maximum thickness of the rolling layer ($R_{m}$) usually is much smaller than the maximum height of the static pile ($h_{m}$), which, in view of Eq. (\[eq: r1\]), leads to $$v \gg 2u.$$ Therefore, Eq. (\[eq: H\]) simplifies to $$z = \frac{h_{m}-h(z)}{\theta_{r}}+\frac{v}{\theta_{r}\gamma}\ln\frac{h_{m}}{h(z)}, \label{eq: Hs1}$$ which reduces to, $$x = \frac{h_{m}-h(x)}{\theta_{r}} + \frac{v}{\theta_{r}\gamma}\ln\frac{h_{m}}{h(x)}, \label{eq: Hs2}$$ provided that the growth of the pile is slow enough to satisfy the quasi-static assumption, i.e. $ut \sim 0$.
It is interesting to note that Eq. (\[eq: Hs2\]) has the same form as Eq. (\[eq: aH\]) in that both equations have two terms on the right-hand side: the first one just corresponds to the linear part implied by the angle of repose $\theta_{c}$, while the second one represents the non-linear part which predicts the existence of the logarithmic tail at the bottom of the pile. In fact, the only difference between Eqs. (\[eq: aH\]) and (\[eq: Hs2\]) is the coefficient of the logarithmic term: one is $l_{e}$ and the other is $v/\gamma \theta_{r}$. In contrast to the former (a fitting parameter in Eq. (\[eq: aH\])), the latter consists of an operational parameter ($v$) and two phenomenological constants ($\theta_{r}$ and $\gamma$) characterizing granular materials, all of which, in principle, can be obtained directly by experimental measurements [@Makes]. Therefore, according to Eq. (\[eq: Hs2\]), we clearly see that, for given $h_{m}$, the length of the curved tail depends on the rolling velocity ($v$), the angle of repose ($\theta_{c}$) and the rate of collision between rolling and static grains ($\gamma$). Moreover, since $v$ as well as $\gamma$ are relevant to the dynamics of the rolling grains on the top of the static pile, their presence in Eq. (\[eq: Hs2\]) throws new light on the formation of the curved tail at the bottom of the pile. Thus, for a quasi-static pile, the flow of rolling grains is due to very small perturbations on the pile surface where the length, time, and velocity are respectively scaled with respect to $a$, $1/\gamma$ and $a\gamma$ (where $a$ is the size of the grain) so that $v$ must be proportional to $a\gamma$. However, since $a\gamma$ is essentially independent of the perturbations, so is the value of $v$. In this case, the formation of the tail is predicted to depend mainly on the size and the shape of granular materials. On the other hand, if the external disturbances are not so small, the value of $v$ may be relevant to these disturbances rather than being simply scaled with $a\gamma$. Under this condition, the formation of the tail depends not only on the properties of the granular materials but also on the operational condition, i.e. the drift velocity of the grains within the rolling layer.
In order to check the reliability of Eq. (\[eq: Hs2\]), we also performed experiments with five different granular materials in a rectangular two-dimensional cell made of two vertical Plexiglas plates of size $22$cm $\times$ $45$cm with a fixed gap of $2.5$mm. Granular materials were poured through a funnel placed on the up-left corner of the cell. Several pictures of static piles are shown in Fig. \[fig1r\], and the qualitative comparison between them is listed in Table 1. For example, no obvious tail exists in the static pile of spherical glass beads (cf. Fig. 1(a) where $a=0.3$mm and $\theta_{c}=30^{\circ}$), but a long tail is present at the bottom of the static pile of cubic sugar grains (cf. Fig. 1(c) where $a=0.8$mm and $\theta_{c}=45^{\circ}$). These experimental observations of the static piles show that the formation of the tail is relevant to the properties of granular materials, which is consistent with the previous discussion following Eq. (\[eq: Hs2\]).
As mentioned earlier, all the parameters in Eq. (\[eq: Hs2\]), i.e. $v$, $\theta_{c}$, $\gamma$ and $h_{m}$, can be obtained by experimental measurements [@Makes]. However, compared with the last three parameters, it is more difficult to accurately measure $v$ for the case of non-uniform particle velocity profiles within a varying thickness of rolling layer $R(x,t)$. Under this condition, we can fit the experimental results by taking $v$ to be the only adjustable parameter and using the values of $\theta_{c}$, $\gamma$ and $h_{m}$ as determined from the precise experimental measurements. Here, $\theta_{c}$ and $h_{m}$ are measured in our experiments while $\gamma$ is obtained from early measurements [@Makes] where the experimental setup as well as granular materials were similar to our case. In Fig. \[fig2\], a quantitive comparison of the surface shape is made between experiment and theory: the dark part is the experimental sugar grain pile, and the solid line represents the surface shape given by Eq. (\[eq: Hs2\]). Clearly, the agreement is satisfactory in both linear and non-linear parts with $v=11.5$cm and $\gamma=23/$s [@Makes]. The value of $v$ is consistent with previous estimates [@Makes] under the same flux condition as in the present experiments.
Next, we investigate the dynamic behavior of pile growth using sand in Fig. 1(b). During the experiments with sand, we recorded the whole process of the sandpile growth, and thereby obtained a family of sandpile surface profiles at different growth stages (cf. Fig. \[fig3\](a)). At the early growth stage (e.g. $t=13s$), the corresponding sandpile was small and the tail was clearly seen; but when the sandpile grew larger (e.g. $t=37s$), the corresponding tail became hardly discernible. Since the position of the funnel was fixed while pouring granular materials into the cell, the distance between the outlet of the funnel and the top of the sandpile, called $H$, decreased with the growth of the sandpile. Consequently, the initial velocity of the rolling particles also decreased with time, which may result in the decrease of the tail length. To examine this idea, we can still make use of Eq. (\[eq: Hs2\]) to fit the experimental profiles of the sandpile surface at different growth stages by choosing different values of $v$ but keeping $\gamma$ and $\theta_{r}$ constant. The fitting results are shown in Fig. 3(a) with solid lines. It is surprising to see that the theoretical predictions and the experimental data are in good agreement even in this dynamic case where the shape of the pile is no longer independent of time. Moreover, in light of Fig. \[fig3\](b), the value of $v$ chosen in Fig. \[fig3\](a) is found to be an increasing function of $H$, which is qualitatively consistent with our experimental observation. In fact, the corresponding value of kinetic energy $v^{2}/2$, as expected by conservation of mechanical energy, is decreasing linearly with a decrease in $H$ for the relative larger $H$ cases and may approach a constant as $H\rightarrow 0 $. This result implies that the value of $v$ is dominated by the external disturbances from the falling grains through the distance $H$ if the latter is larger enough; otherwise, it becomes essentially independent of $H$, and thereby further justifies our choosing different values of $v$ for different growth stages.
In summary, we have solved the BCRE equations analytically with a simple transformation. The analytical solution has been used to describe the surface shape of the granular pile successfully in both the static and the dynamic cases. This expression for the surface shape not only predicts a logarithmic tail as in the early work [@Herrmann], but also reveals a clear physical picture for its formation.
[**Acknowledgments**]{}
The authors would like to thank A. Acrivos and H. Herrmann for discussions. H. Makse acknowledges support from NSF-DMR-0239504.
[99]{} J.J. Alonso and H.J. Herrmann, Shape of the Tail of a Two-Dimensional Sandpile, 1996 [*Phys. Rev. Lett.*]{} [**76**]{} 4911. J.-P. Bouchaud, M.E. Cates, J.R. Prakash and S.F. Edwards, Hysteresis and Metastability in a Continuum Sandpile Model, 1995 [*Phys. Rev. Lett.*]{} [**74**]{} 1982. H.A. Makse, R.C. Ball, H.E. Stanley and S. Warr, Dynamics of Granular Stratification, 1998 [*Phys. Rev, E*]{} [**58**]{} 3357 .
----------- --------------- --------------------- ---------------- -------------- --------------
Granular Shape Angle of size Existence of Condition of
materials repose $\theta_{c}$ $a$(mm) curved tail pile surface
Glass Spherical $30^{\circ}$ $\cong 0.3$ No Smooth
Glass Non-Spherical $39^{\circ}$ $\cong1.25$ Yes Rough
PMMA Spherical $37^{\circ}$ $0.45\sim 0.5$ Yes Smooth
Sugar Cubic $45^{\circ}$ $\cong0.8$ Yes Rough
Sand Irregular $39^{\circ}$ $\cong 0.3$ Yes Smooth
----------- --------------- --------------------- ---------------- -------------- --------------
: [Summary of the experimental results relative to the static granular piles.]{} Note that the typical size of the sugar grains is obtained by measuring the volume of the cubic grains and calculating the typical size as the cubic root averaged over $10$ different grains. The values of the angle of repose are obtained by pouring the grains in the cell mentioned earlier and then measuring the resulting angles of piles. The cell gap is smaller $(0.25$cm) than that used in Ref. [@Makes] so that the values of the angle of repose here are greater than those in Ref. [@Makes] due to the extra friction induced from the vertical walls.
[**Fig 1**]{} Pictures of static granular piles. (a) Spherical glass bead pile; (b) Irregular sand grain pile; (c) Cubic sugar grain pile.
[**Fig 2**]{} The surface shape given by Eq. (\[eq: Hs2\]) is in good agreement with the experimental result using sugar grains with $\theta_{c}=45^{\circ}$, $v=11.5$cm/s and $\gamma=23$/s.
[**Fig 3**]{} (a) The surface profiles of the sandpile at different growth stages. The scatters are the experimental data and the solid lines represent the theoretical results given by Eq. (\[eq: Hs2\]) with $\theta_{c}=39^{\circ}$, $\gamma=20$/s and different values of $v$. Note that $t=0$ represents the start point of pouring the sand into the cell. (b) The value of $v^{2}/2$ with $v$ being the velocity of the gains within the rolling layer used to fit the pile profiles at different growth stages is decreasing with a decrease in $H$, the corresponding distance from the outlet of the funnel to the top of the sandpile.
![[]{data-label="fig1r"}](fig1a.eps)
![[]{data-label="fig1r"}](fig1b.eps)
![[]{data-label="fig1r"}](fig1c.eps)
![[]{data-label="fig2"}](fig2.eps)
![[]{data-label="fig3"}](fig3a.eps)
(a)
![[]{data-label="fig3"}](fig3b.eps)
(b)
| {
"pile_set_name": "ArXiv"
} |
---
author:
- |
Luis H. Gallardo - Olivier Rahavandrainy\
Mathematics, University of Brest\
6, Avenue Le Gorgeu, C.S. 93837,\
29238 Brest Cedex 3, France.\
e-mail : [email protected] - [email protected]\
\
AMS Subject Classification: 11T55, 11T06.\
\
Keywords: Sum of divisors, polynomials, finite fields, characteristic $2.$\
\
running head: binary perfect polynomials.\
\
write correspondence to: Luis H. Gallardo.
title: 'Even perfect polynomials over $\F_2$ with four prime factors'
---
\
[**[Abstract]{}**]{}\
A perfect polynomial over the binary field $\F_2$ is a polynomial $A \in \F_2[x]$ that equals the sum of all its divisors. If $\gcd(A,x^2-x) \neq 1$ then we call $A$ even. The list of all even perfect polynomials over $\F_2$ with at most $3$ prime factors in known. The object of this paper is to give the list of all even perfect polynomials over $\F_2$ with four prime factors. These are all the known perfect polynomials with four prime factors over $\F_2$.
Introduction
============
As usual, we denote by $\F_2$ the finite field with two elements $\{0, 1\}$.\
For a polynomial $A \in \F_2[x]$, let $\displaystyle{\sigma(A) =
\sum_{D|A} D}$ be the sum of divisors of $A.$ We denote also, as usual, by $\omega(A)$ the number of distinct prime (irreducible) polynomials that divide $A.$ These two functions are multiplicative, a fact that we shall use without more reference in the rest of the paper. If $\sigma(A) = A,$ then we call $A$ a perfect polynomial.
The notion of perfect polynomial (over $\F_2$) was introduced by Canaday [@Canaday], the first doctoral student of Leonard Carlitz.\
He studied mainly the case in which $\gcd(A,x^2+x) \neq 1.$
We may think $x^2+x \in \F_2[x]$ as being the analogue of $2 \in \Z$ so that the “even” polynomials are the polynomials with linear factors and the “odd" ones are such that $\gcd(A,x^2+x) =1.$ Canaday (among other results in [@Canaday]) classifies the even perfect polynomials with three irreducible factors and gives without proof [@Canaday Theorem 11] the list of all even perfect polynomials $A$ with $\omega(A) = 4$.\
The object of this paper (see Theorem \[evenwA4F2\] ) is to prove Canaday’s results in [@Canaday Theorem 11]: The following polynomials are the only even perfect polynomials $A \in \F_2[x]$ with $\omega(A)=4$ prime factors : $$\begin{array}{l}
C_1(x) = x^2(x+1)(x^2+x+1)^2(x^4+x+1), \ C_2(x) = C_1(x+1),\\
C_3(x) = C_3(x+1) = x^4(x+1)^4(x^4+x^3+x^2+x+1)(x^4+x^3+1),\\
C_4(x) = x^6(x+1)^3(x^3+x^2+1)(x^3+x+1), \ C_5(x) = C_4(x+1).
\end{array}$$ Observe that the two latter polynomials
are also perfect over $\F_4$ (see [@Gall-Rahav2]).
The complete list of all even perfect polynomials over $\F_2$ with $\omega(A) \leq 4$ is then: $$0,1,\,\,\, (x^2+x)^{2^{n}-1},\,\,\, T_1(x) = x^2(x+1)(x^2+x+1),\,\,\, T_1(x+1),$$ $$T_2(x) = x^3(x+1)^4(x^4+x^3+1),\,\,\, T_2(x+1),\,\,\,C_1(x), \ldots, C_5(x),$$ in which $n >0$ is a positive integer.\
In fact this list is the list of all perfect polynomials over $\F_2$ with $\omega(A) \leq 4.$ (see [@Gall-Rahav4]).\
There are only two more known perfect polynomials over $\F_2$, both even, with $\omega(A)=5$ and with degree $20$, namely: $$S_1(x) = x^6(x+1)^4(x^3+x+1)(x^3+x^2+1)(x^4+x^3+1),\,\,\,S_1(x+1).$$
It may have some interest to know whether or not there are perfect polynomials over $\F_2$ with degree moderately bigger that $20$ (so that we may compute them with a computer). These have been investigated [@Gall-Rahav3 Theorem 5.5] (no solutions up to degree $28$) in the special case in which all exponents are equal to $2$ and the polynomial is odd.\
Some useful facts
=================
We denote, as usual by $\N$ the set of nonnegative integers. In this section we recall, and we present, some necessary results for the next sections.\
First of all, we recall some definitions and lemmata.\
\
[**[Definitions]{}**]{}\
- We define (following Canaday’s terminology) as the inverse of a polynomial $P(x)$ of degree $m$, the polynomial $\displaystyle{P^*(x)= x^m P(\frac{1}{x})}$.\
- We say that $P$ inverts into itself if $P= P^*$.\
- A polynomial $P$ is complete if there exists $h \in \N$ such that: $$P = \sigma(x^h) = 1 + x + \cdots + x^h.$$
The following lemma essentially based on a result of Dickson (see proof of [@Canaday Lemma 2]) is key.
\[inversion\] \
i): Let $P \in \F_2[x]$ be such that $P(0)=1.$ We have: $(P^*)^* = P$.\
ii): Any complete polynomial inverts into itself.\
iii): If $1+ x + \cdots + x^m = PQ$, where $P, Q$ are irreducible, then either $(P=P^*, Q = Q^*)$ or $(P=Q^*, Q = P^*)$.\
iv): If $P=P^*$, $P$ irreducible and if $P=x^{a}(x+1)^b +1$, then: $$P \in \{1+x+x^2, 1+x+ \cdots + x^4\}.$$
[**[Proof]{}**]{}:\
i) and ii) are obvious.\
iii) follows by ii).\
iv) is the corollary of Lemma 7 in [@Canaday], (that follows from Lemma 2 of ibid.).
\[translation\] \
If $A(x)$ is a perfect polynomial over $\F_2$, then $A(x+1)$ is also perfect.
[(Lemma 5 in [@Canaday])]{} \[lemma5canaday\] \
Let $P, Q \in \F_2[x]$ and $n, m \in \N$ such that $P$ is irreducible and $\sigma(P^{2n}) = 1 + \cdots + P^{2n} = Q^m$. Then $m \in \{0,1\}$.
[(Lemma 6 in [@Canaday])]{} \[lemma6canaday\] \
Let $P, Q \in \F_2[x]$ and $n, m \in \N$ such that $P$ is irreducible and $\sigma(P^{2n}) = 1 + \cdots + P^{2n} = Q^m A, \ m>
1$. If $m$ is odd (resp. even) then $deg(P) > (m-1) deg(Q)$ (resp. $deg(P) > m \ deg(Q)$).
[(Lemma 4 in [@Canaday])]{} \[lemma4canaday\] \
If $PQ = 1+\cdots+x^{2h}$ and $P =1+ \cdots + (x+1)^{2k}$, then $h =
4$ and $k=1$; that is: $P = 1+x+x^2, \ Q = P(x^3) = 1+x^3+x^6$.
The proof of the following lemma in [@Canaday] uses the properties i) to iii) in Lemma \[inversion\]:
[(Theorem 8 in [@Canaday])]{} \[theorem8canaday\] \
Let $A = 1+\cdots+x^{2h} \in \F_2[x]$ such that any irreducible factor of $A$ is of the form $x^a (x+1)^b + 1$. Then $h \in
\{1,2,3\}$.
The following crucial lemma follows from Lemma 2.5 in [@Gall-Rahav2] that says that the number of minimal primes dividing a perfect polynomial is even:
\
Every even perfect polynomial $A$ over $\F_2$ with $\omega(A)=4,$ is of the form $x^h
(x+1)^k P^{l} Q^{m}$, for some odd prime polynomials $P,Q$ and for some positive integers $h,k,l,m$.
We provide proofs of the following two lemmata claimed but not proved by Canaday:
[(Lemma 10 in [@Canaday])]{} \[lemma10canaday\] \
Let $P \neq Q$ be two odd polynomials in $\F_2[x].$ If $x^h (x+1)^k P^{l} Q^{2n-1}$ is a perfect polynomial over $\F_2$, and if $l \not= 2^r - 1$, then $2n-1 = 2^s - 1$.
[**[Proof]{}**]{}: If $l \not= 2^r - 1$ and $2n-1 \not= 2^s - 1$, then put: $$2n-1 = 2^s u - 1, \mbox{ where $u \geq 3$ is odd}.$$ We can write: $$1+\cdots+Q^{2n-1} = (Q+1)^{2^s-1} (1+\cdots + Q^{u-1})^{2^s}.$$ Since $u-1 \geq 2$ is even, we have by Lemma \[lemma5canaday\]: $$1+\cdots + Q^{u-1} = P.$$ So, $${\rm{deg}}(Q) < {\rm{deg}}(P).$$ If $l$ is even, then by the same argument, ${\rm{deg}}(Q) < {\rm{deg}}(P).$ It is impossible. So $l$ is odd. We can write: $$l = 2^r v - 1, \ 1+\cdots+P^l = (P+1)^{2^r-1}(1+\cdots +
P^{v-1})^{2^r}, \mbox{ where $v\geq 3$ is odd}.$$ Since $v-1 \geq 2$ is even, we have by Lemma \[lemma5canaday\]: $$1+\cdots + P^{v-1} = Q.$$ So, $${\rm{deg}}(P) < {\rm{deg}}(Q).$$ It is impossible.
$\Box$
[(Lemma 11 in [@Canaday])]{} \[lemma11canaday\] \
Let $P \neq Q$ be two odd polynomials in $\F_2[x].$ If $x^h (x+1)^k P^{2l} Q^{2n-1}$ is a perfect polynomial over $\F_2$, then $2l = 2^m$ and $m=n$.
[**[Proof]{}**]{}: We can write: $$\begin{array}{l}
1+\cdots+P^{2l} = Q,\\
1+\cdots+Q^{2^n-1} = (Q+1)^{2^n-1}.
\end{array}$$ So, $P$ divides $Q+1$ and $P^2$ does not. Thus, $$Q+1 = x^a (x+1)^b P, \mbox{ for some $a,b \in \N$}.$$ Since $\sigma(A) = A$, we obtain: $$(1+\cdots+x^h)(1+\cdots+(x+1)^k)(x^a (x+1)^b)^{2^n-1} P^{2^n-1}Q =
x^h(x+1)^kP^{2l}Q^{2^n-1}.$$ - If $h$ and $k$ are even, then by lemma \[lemma5canaday\]: $$(1+\cdots+x^h)(1+\cdots+(x+1)^k)=P^{\alpha}Q^{\beta}, \ 0 \leq \alpha,
\beta \leq 2.$$ Therefore, we must have: $$\alpha = 1.$$ We are done.\
\
- If $h$ and $k$ are both odd, then by considering exponents of $P$, we see that it is impossible.\
\
- If $h$ is even and $k$ odd, then by considering exponents of $Q$, we must have: $$1+\cdots+x^h = P.$$ Put: $$k+1 = 2^r u, \mbox{ where $u$ is odd}.$$ We have: $$1+\cdots+(x+1)^k = x^{2^r-1}(1+\cdots+(x+1)^{u-1})^{2^r} =
x^{2^r-1}(P^{\gamma}Q^{\delta})^{2^r}, \ 0 \leq \gamma, \delta \leq
1.$$ - If $\gamma = 0$, then we are done.\
\
- If $\gamma = 1$ and $\delta = 0$, then $u-1 \geq 2$ and $n = 1$. Thus, by considering exponents of $P$, we get: $$l-1 = 2^{r-1}.$$ Furthermore, we can write: $$Q+1 = P+\cdots+P^{2l} = P(1+P)(1+\cdots+P^{l-1})^2.$$ So, $l$ must be equal to $2$, and then $r = 1$, $a = 3$, $h = 4$.\
\
Thus: $$P = 1+\cdots + x^4, \ Q = 1 + \cdots + P^4 =
(1+x+x^4)(1+x+x^2+x^4+x^6+x^7+x^8+x^9+x^{12}.$$ It is impossible since $Q$ is irreducible.\
\
- If $\gamma = \delta = 1$, then by Lemma \[lemma4canaday\], ${\rm{deg}}(P) = 2$ and ${\rm{deg}}(Q) = 6$. It is impossible since $Q = 1+\cdots + P^{2l}$.\
\
- If $h$ is odd and $k$ even, analogous proof.
$\Box$
In the next section we prove our main result:
\[evenwA4F2\] The complete list of even perfect polynomials over $\F_2$ with $4$ prime factors consists of the five polynomials $C_1(x), \ldots, C_5(x).$
Perfects of the forms: $A = x^h (x+1)^k P^m Q^n$
================================================
We may reduce (see lemmata \[lemma10canaday\] and \[lemma11canaday\]) our study to the following cases:\
(a) $A = x^h (x+1)^k P^{2m} Q^{2n}$\
(b) $A = x^h (x+1)^k P^{2^n} Q^{2^n-1}$\
(c) $A = x^{2h} (x+1)^{2k} P^{2m-1} Q^{2^n-1}$\
(d) $A = x^{2h} (x+1)^{2k-1} P^{2m-1} Q^{2^n-1}$\
(e) $A = x^{2h-1} (x+1)^{2k-1} P^{2m-1} Q^{2^n-1}$.\
Compare with [@Canaday page 733].
Case (a)
--------
Since $x$ and $x+1$ do not divide $\sigma(P^{2m})$, we obtain by Lemma \[lemma5canaday\]: $$\sigma(P^{2m}) = 1 + \cdots + P^{2m} = Q.$$ Analogously, $$\sigma(Q^{2n}) = P.$$ Therefore, considering degrees, we have: $$4mn = 1,$$ which is impossible.
Case (e)
--------
Since $\sigma(A) = A$, we obtain: $$x(x+1)(P+1)(Q+1) B^2 = x^{2h-1} (x+1)^{2k-1} P^{2m-1} Q^{2^n-1},
\ \mbox{for some polynomial } B.$$ It follows that $P$ (respectively $Q$) must divide $Q+1$ (resp. $P+1$). So $P = Q+1$, which is impossible.
Case (b)
--------
We obtain: $$\begin{array}{l}
1 + \cdots + P^{2^n} = Q, \mbox{ by Lemma \ref{lemma5canaday} }
\mbox{ and since } x, \ x+1 \mbox{ do not divide } \sigma(P^{2^n}),\\
1 + \cdots + Q^{2^n-1} =(Q+1)^{2^n-1}. \end{array}$$ Thus, $P$ divides $Q+1$ and $P^2$ does not. So, $Q$ does not divide $P+1$. We may write: $$\begin{array}{l}
Q+1 = P(1+P)^{2^n-1},\\
P+1 = x^{\alpha}(x+1)^{\beta}, \ \alpha, \beta \geq 1 \end{array}$$
### Case $h$, $k$ even
The two monomials $x$ and $x+1$ do not divide $\sigma(x^h), \sigma((x+1)^k)$. So: $$\begin{array}{l}
1+ \cdots + x^{h} = P^{a_0}Q^{b_0}, \ a_0, b_0 \in \{0,1\},\\
1+ \cdots + (x+1)^{k} = P^{a_1}Q^{b_1}, \ a_1, b_1 \in \{0,1\}.
\end{array}$$ Since $\sigma(A) = A$, we obtain: $$\begin{array}{l}
Q(Q+1)^{2^n-1}P^{l}Q^{r} = x^h (x+1)^k P^{2^n} Q^{2^n-1},\\
l = a_0+a_1, \ r = b_0+b_1, \ l, r \in \{0, 1, 2\}.
\end{array}$$ Considering the exponents of $P$ and $Q$, we have: $$2^n-1 + l =
2^n, \ r+1 = 2^n - 1.$$ So, $$l = 1, \ n \in \{1,2\}.$$ (i)- :\
We have: $$r = 0, \ 1+P+P^2 = Q, \ 1+ \cdots + x^{h} = P = 1+ \cdots
+ (x+1)^{k}, \ h = k.$$ Since $P = x^{\alpha}(x+1)^{\beta} +1$, by Lemma \[theorem8canaday\], $P \in \{1+x+x^2, 1+
\cdots+x^4\}$.\
- If $P= 1+x+x^2$, then $h=k=2, \ Q = 1+x+x^4$. Thus $A =
x^2(x+1)^2P^2Q$ which is not perfect.\
- If $P= 1+\cdots +x^4$, then $Q = 1+P+P^2 =
(1+x+x^2)(1+x^2+x^4+x^5+x^6)$ is reducible. It is impossible.\
\
(ii)- :\
We have: $$\begin{array}{l} r = 2, \ 1+\cdots+P^4 = Q,\\
1+ \cdots +
x^{h} = PQ,\\
1+ \cdots + (x+1)^{k} = Q. \end{array}$$ By Lemma \[lemma4canaday\], we have: $$h = 8, \ k = 2, \ Q = 1+x+x^2, \ P = 1 + x^3 +x^6.$$ So, $Q \not= 1+\cdots + P^4$. It is impossible.
### Case $h$, $k$ odd
Since $\sigma(A) = A$, we obtain: $$\label{hkodd}
x(x+1)Q(Q+1)^{2^n-1}B^2 = x^h (x+1)^k P^{2^n} Q^{2^n-1}.$$ Since $P$ divides $Q+1$ and $P^2$ does not, by considering the exponent of $P$, we see that the equality (\[hkodd\]) is impossible.
### Case $h$ odd, $k$ even
Put $h = 2l-1$ and $k = 2r$.\
By Lemma \[lemma6canaday\], we have: $$1+\cdots+(x+1)^k = 1 + \cdots + (x+1)^{2r} = P^aQ^b,
\mbox{ for some } a,b\in \{0,1\}.$$ Since $\sigma(A) = A$, we obtain: $$(x+1)(1+\cdots+x^{l-1})^2Q(Q+1)^{2^n-1}P^aQ^b =
x^{2l-1} (x+1)^{2r} P^{2^n} Q^{2^n-1}.$$ Since $P$ divides $Q+1$ and $P^2$ does not, if $b = 1$ (resp. $a = 0$), then the exponent of $Q$ (resp. of $P$) in the right hand side is even (resp. odd). It is impossible. So, $b = 0$ and $a = 1$. Therefore: $$P=1+\cdots
+(x+1)^{2r} = x^{\alpha}(x+1)^{\beta} +1.$$ By Lemma \[theorem8canaday\], $P \in \{1+x+x^2, 1+x^3+x^4\}$.\
\
(i)- :\
We have $k = 2r = 2$, and by considering the exponent of $x+1$ we get: $$n=1, \ Q = 1+P+P^2 = 1+x+x^4.$$ So, $$l=1, \ h = 1.$$ We obtain the polynomial $C_1(x)$, and by Lemma \[translation\], we get the polynomial $C_1(x+1)$.\
\
(ii)- :\
We have: $$2r = 4, \ Q+1 = (1+P)^{2^n-1}P =
x^{3(2^n-1)}(x+1)^{2^n-1}P.$$ By considering the exponent of $x+1$, we have: $$(2^n-1)^2 +1 \leq 4$$ So, $$n = 1,$$ and $$Q = 1+P+P^2 = 1+x^3+x^4+x^6+x^8
=(1+x+x^2)(1+x+x^4+x^5+x^6).$$ It is impossible.
Case (c)
--------
By Lemma \[lemma6canaday\], we obtain: $$\begin{array}{l}
1 + \cdots + x^{2h} = P^{a_0}Q^{b_0}, \\
1 + \cdots + (x+1)^{2k} = P^{a_1}Q^{b_1},\\
a_0, b_0, a_1, b_1 \in \{0,1\}. \end{array}$$ Since $\sigma(A) = A$, we obtain: $$(P+1)(Q+1)^{2^n-1}P^{a_0+a_1}Q^{b_0+b_1}(1 + \cdots + P^{m-1})^2 =
x^{2h} (x+1)^{2k} P^{2m-1} Q^{2^n-1}.$$ Thus: $$\begin{array}{l}
1+P =x^{\alpha_1} (x+1)^{\beta_1}Q^{\gamma_1},\\
1+Q =x^{\alpha_2} (x+1)^{\beta_2}P^{\gamma_2},\\
\alpha_1, \beta_1, \gamma_1, \alpha_2, \beta_2, \gamma_2 \in \N.
\end{array}$$ We can reduce the work to three cases, since the integers $h$ and $k$ play symmetric roles (by Lemma \[translation\]).
### Case $a_0 = b_0 = b_1 = 1, a_1 = 0$
We have: $$\begin{array}{l}
1 + \cdots + x^{2h} = PQ, \\
1 + \cdots + (x+1)^{2k} = Q. \end{array}$$ So, by Lemma \[lemma4canaday\], we obtain: $$Q = 1+x+x^2, \ P = 1+x^3+x^6, \ h = 4, \ k = 1.$$ Since $\sigma(A) = A$, by considering the exponent of $x+1$, we obtain: $$n = 1,$$ and thus: $$x^4(x+1)^2PQ^3(1+\cdots+P^{m-1})^2 = x^{8} (x+1)^{2} P^{2m-1}
Q.$$ Thus, $x$ must divide $B = 1+\cdots+P^{m-1}$. So, $x+1$ must divide $B$. It is impossible.
### Case $a_0 = b_0 = a_1 = 1, b_1 = 0$
We have: $$\begin{array}{l}
1 + \cdots + x^{2h} = PQ, \\
1 + \cdots + (x+1)^{2k} = P. \end{array}$$ So, by Lemma \[lemma4canaday\], we obtain: $$P = 1+x+x^2, \ Q = 1+x^3+x^6, \ h = 4, \ k = 1.$$ We obtain the same contradiction as in the previous case.
### Case $a_0 = b_1 = 1, a_1 = b_0 = 0$
We have: $$\begin{array}{l}
1 + \cdots + x^{2h} = P, \\
1 + \cdots + (x+1)^{2k} = Q \end{array}$$ \
Therefore, the monomials $x, \ x+1$ divide $P+1$ and $Q+1$. But $x^2$ (resp. $(x+1)^2$) does not divide $P+1$ (resp. $Q+1$).\
Since $\sigma(A) = A$, we have: $$\label{casc1}
PQ(P+1)(Q+1)^{2^n-1}(1+\cdots+P^{m-1})^2 = x^{2h} (x+1)^{2k}
P^{2m-1} Q^{2^n-1}$$ (i)- :\
\
In this case, $P$ does not divide $Q+1$ and $Q$ does not divide $P+1$.\
We obtain: $$P+1 = x (x+1)^{\beta_1}, \ Q+1 = x^{\alpha_2} (x+1).$$ Therefore, by relation (\[casc1\]): $$m = 1 \mbox{ and } n = 1.$$ So, by Lemma \[theorem8canaday\]: $$P \in \{1+x+x^2, 1+\cdots +x^4\}, \ Q \in \{1+x+x^2,
1+x^3+x^4\}.$$ We must have: $$P = 1 + \cdots +x^4, \ Q = 1+x^3+x^4.$$ So, $$h = k =
2.$$ We get the polynomial $C_3(x)$, and thus the polynomial $C_3(x+1) = C_3(x)$.\
\
(ii)- :\
\
The polynomial $P$ divides $Q+1$, and by relation (\[casc1\]), the integer $\gamma_2$ must be even. So: $$Q+1 = x^{\alpha_2} (x+1) P^{2u}, \ u \geq 1.$$ In particular, $P^2$ divides $Q+1$.\
\
Furthermore, $Q$ does not divide $P+1$. So, $$P=x(x+1)^{\beta_1} + 1.$$ So, by Lemma \[theorem8canaday\], $P \in
\{1+x+x^2, 1+\cdots +x^4\}$.\
\
- If $P= 1+x+x^2$, then $2h = 2, \ n=1$ and $\alpha_2 = 1$ (consider the exponents of $x$ in the relation (\[casc1\])). We can write: $$Q+1 = x (x+1)P^{2u}.$$ By considering the exponent of $P$, we have: $$u = m-1.$$ and thus: $$m \geq 2.$$ Moreover, the relation (\[casc1\]) becomes: $$(1 + \cdots +
P^{m-1})^2 = (x+1)^{2k-2}.$$ So, $$k = 1, \ m = 1.$$ It is impossible.\
\
- If $P= 1+\cdots+x^4$, then $2h = 4$.\
We can write: $$Q+1 = x^{\alpha_2}(x+1)P^{2u},
\mbox{ where $\alpha_2$ is odd and $u \geq 1$}.$$ By considering the exponent of $x$ in relation (\[casc1\]), we have: $$\mbox{ either $(n = 2, \ \alpha_2 = 1)$~or~$(n=1,~
\alpha_2~\in~\{1,3\})$.}$$ :\
By considering the exponent of $P$, we have: $$m = 3u+1 \geq 4.$$ Moreover, we must have: $$1+\cdots + P^{m-1} = (x+1)^{k-3}Q.$$ So, $$k = 3, \ 1+\cdots + P^{m-1} = Q.$$ Thus, $P^2$ does not divide $Q+1$. It is impossible.\
\
:\
By considering the exponent of $P$, we have: $$u = m-1 \geq 1.$$ Moreover, we must have: $$(1+\cdots + P^{m-1})^2 = x^{4-\alpha_2-1}(x+1)^{2k-4}.$$ Thus, $m-1$ is odd, $\alpha_2 = 1$.\
By writing: $$1+\cdots + P^{m-1} = (1+P)(1+\cdots + P^{m/2-1})^2.$$ We must have: $m=2$, $u=1$, and $k=5$. So, $$\begin{array}{l}
Q = 1 + \cdots + (x+1)^{10} = 1+x+x^2+x^7+x^8+x^9+x^{10},\\
Q+1 = x(x+1)P^2 = x(x+1)(1+\cdots + x^4)^2 = x + \cdots + x^{10}.
\end{array}$$ It is impossible.\
\
(iii)- :\
\
In this case, $P$ does not divide $Q+1$, and $Q$ divides $P+1$.\
So, $$Q=x^{\alpha_2}(x+1) + 1.$$ So, by Lemma \[theorem8canaday\]: $$Q \in
\{1+x+x^2, 1+ x^3 +x^4\}.$$ Therefore, by relation (\[casc1\]): $$m = 1.$$ So: $$(P+1)(Q+1)^{2^n-1}=
x^{2h} (x+1)^{2k}Q^{2^n-2}.$$ - If $Q= 1+x+x^2$, then $k=1$, $\gamma_1 = 2u = 2^n - 2$ is even, and $\beta_1$ is odd. We can write: $$P+1 = x(x+1)^{\beta_1} Q^{\gamma_1}, \ \gamma_1 = 2u \geq 2.$$ Considering the exponent of $x+1$, we have: $$2 = 2^{n}-1 + \beta_1.$$ So: $$n= \beta_1 = 1.$$ Thus: $$\gamma_1 = 2^n - 2 = 0.$$ It is impossible.\
\
- If $Q= 1+x^3+x^4$, then $$2k=4 = 2^{n}-1 + \beta_1.$$ So: $$\mbox{ either $(n=1, \ \beta_1 = 3)$ or $(n=2, \ \beta_1 = 1)$}.$$ The first case is impossible since $\gamma_1 = 2^n - 2 \geq 2$.\
So, $$n=2, \ \beta_1 = 1, \gamma_1 = 2, \ 2h = 3.(2^n - 1) + 1 =
10.$$ Thus: $$\begin{array}{l}
P = 1 + \cdots + x^{10},\\
P+1 = x(x+1)Q^2 = x(x+1)(1+ x^3 + x^4)^2 = x+x^2+x^7+x^8+x^9+x^{10}.
\end{array}$$ It is impossible.
Case (d)
--------
We obtain: $$\begin{array}{l}
1 + \cdots + x^{2h} = P^{a_0}Q^{b_0}, \ a_0, b_0 \in \{0,1\}
\mbox{
by
Lemma \ref{lemma6canaday}},\\
1 + \cdots + (x+1)^{2k-1} = x(1+\cdots+(x+1)^{k-1})^2.
\end{array}$$ Since $\sigma(A) = A$, we obtain: $$\label{casd1}
x(P+1)(Q+1)^{2^n-1}B^2P^{a_0}Q^{b_0} = x^{2h} (x+1)^{2k-1} P^{2m-1}
Q^{2^n-1}$$ Thus: $$\begin{array}{l}
1+P =x^{\alpha_1} (x+1)^{\beta_1}Q^{\gamma_1},\\
1+Q =x^{\alpha_2} (x+1)^{\beta_2}P^{\gamma_2}. \end{array}$$ By considering degrees, we obtain: $$\gamma_1 \gamma_2 \leq 1.$$ If $\gamma_1 = \gamma_2 = 1$, then $Q = P+1$. It is impossible.\
So, $\gamma_1 \gamma_2 = 0$. We have three cases:
### Case: $\gamma_1 = \gamma_2 = 0$
In this case, $Q$ (resp. $P$) does not divide $P+1$ (resp. $Q+1$). We may write: $$P = x^{\alpha_1} (x+1)^{\beta_1} + 1, \ Q =
x^{\alpha_2} (x+1)^{\beta_2} + 1.$$ - If $1 + \cdots + x^{2h} = P$, then the relation (\[casd1\]) becomes: $$x(P+1)(Q+1)^{2^n-1}B^2P =
x^{2h} (x+1)^{2k-1} P^{2m-1} Q^{2^n-1}.$$ It is impossible (consider the exponent of $Q$).\
- If $1 + \cdots + x^{2h} = Q$, then the relation (\[casd1\]) becomes: $$x(P+1)(Q+1)^{2^n-1}B^2Q =
x^{2h} (x+1)^{2k-1} P^{2m-1} Q^{2^n-1}.$$ It is impossible (consider the exponent of $P$).\
- If $1 + \cdots + x^{2h} = PQ$, then by Lemma \[theorem8canaday\]: $$P, Q \in \{x^3 +x^2 +1, x^3+x+1\}, \ h = 3.$$ We get the polynomial $C_4(x)$ and thus also the polynomial $C_5(x) = C_4(x+1)$.
### Case: $\gamma_1 = 0, \ \gamma_2 \geq 1$
In this case, we may write: $$P = x^{\alpha_1} (x+1)^{\beta_1} + 1,
\ Q = x^{\alpha_2} (x+1)^{\beta_2}P^{\gamma_2} + 1.$$ So ${\rm{deg}}(P) < {\rm{deg}}(Q)$.\
\
- If $1 + \cdots + x^{2h} = P$, then it is impossible as in the above case (consider the exponent of $Q$).\
- If $1 + \cdots + x^{2h} = Q$, then: $$\begin{array}{l}
a_0 = 1, \ b_0 = 0,\\
\mbox{ $x$ divides $Q+1$, $x^2$ does not},\\
Q+1 = x(x+1)(1+ \cdots + x^{h-1})^2.
\end{array}$$ So, $\alpha_2 = 1$ and $\gamma_2$ is even.\
By considering the exponent of $P$, we see that the relation (\[casd1\]) does not hold. It is impossible.\
- If $1 + \cdots + x^{2h} = PQ$, then by Lemma \[inversion\], since ${\rm{deg}}(P) < {\rm{deg}}(Q)$, the polynomial $P$ (resp. $Q$) inverts into itself, and $P \in \{1+x+x^2, 1+
\cdots + x^4\}$.\
Therefore, $\alpha_1 = 1$ and $\beta_1 \in\{1,3\}$ is odd. Thus, by considering the equality: $$x(P+1)(Q+1)^{2^n-1}B^2PQ =
x^{2h} (x+1)^{2k-1} P^{2m-1} Q^{2^n-1},$$ we obtain that the integers $\alpha_2,
\beta_2$ and $\gamma_2$ must be even. So, $Q+1$ is a square. It is impossible by the irreducibility of $Q$.
### Case: $\gamma_1 \geq 1, \ \gamma_2 = 0$
In this case, we may write: $$P = x^{\alpha_1}
(x+1)^{\beta_1}P^{\gamma_1} + 1, \ Q = x^{\alpha_2} (x+1)^{\beta_2}
+ 1.$$ The proof is analogous to that of the previous case, by switching $P$ and $Q$.\
\
- If $1 + \cdots + x^{2h} = Q$, then it is impossible (consider the exponent of $P$).\
- If $1 + \cdots + x^{2h} = P$, then: $$\begin{array}{l}
\mbox{ $x$ divides $P+1$, $x^2$ does not},\\
P+1 = x(x+1)(1+ \cdots + x^{h-1})^2.
\end{array}$$ So, $\alpha_1 = 1$ and $\gamma_1$ is even.\
By considering the exponent of $Q$, we see that the following equality does not hold: $$x(P+1)(Q+1)^{2^n-1}B^2P =
x^{2h} (x+1)^{2k-1} P^{2m-1} Q^{2^n-1}.$$ It is impossible.\
\
- If $1 + \cdots + x^{2h} = PQ$, then by Lemma \[inversion\], since ${\rm{deg}}(Q) < {\rm{deg}}(P)$, the polynomial $P$ (resp. $Q$) inverts into itself, and $Q \in \{1+x+x^2, 1+x+
\cdots + x^4\}$.\
Therefore, $\alpha_2 = 1$ and $\beta_2 \in\{1,3\}$ is odd. Thus, by considering the equality: $$x(P+1)(Q+1)^{2^n-1}B^2PQ =
x^{2h} (x+1)^{2k-1} P^{2m-1} Q^{2^n-1},$$ the integers $\alpha_1,
\beta_1$ and $\gamma_1$ must be even. So, $P+1$ is a square. It is impossible by the irreducibility of $P$.\
This finishes the proof of Theorem \[evenwA4F2\].
\#1
{#section .unnumbered}
‘=1000
=
=
\#1\#2
\#1= \#1
0=
[‘ =]{} [‘ =]{} [‘=]{} [‘ =]{} \#1[0=1 ]{} >0pt >0pt >0pt .
==*=*==**=====****
[99]{}
[Canaday]{} 721 - 737 1941
[Beard]{} 283 - 291 1977
[Gall-Rahav]{} 109- 122 2005
[Gall-Rahav2]{} 21-38 2007
[Gall-Rahav3]{} 167-176 2007
[Gall-Rahav4]{}
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Nonlocal properties of an ensemble of diagonal random unitary matrices of order $N^2$ are investigated. The average Schmidt strength of such a bipartite diagonal quantum gate is shown to scale as $\log N$, in contrast to the $\log N^2$ behavior characteristic to random unitary gates. Entangling power of a diagonal gate $U$ is related to the von Neumann entropy of an auxiliary quantum state $\rho=AA^{\dagger}/N^2$, where the square matrix $A$ is obtained by reshaping the vector of diagonal elements of $U$ of length $N^2$ into a square matrix of order $N$. This fact provides a motivation to study the ensemble of non-hermitian unimodular matrices $A$, with all entries of the same modulus and random phases and the ensemble of quantum states $\rho$, such that all their diagonal entries are equal to $1/N$. Such a state is contradiagonal with respect to the computational basis, in sense that among all unitary equivalent states it maximizes the entropy copied to the environment due to the coarse graining process. The first four moments of the squared singular values of the unimodular ensemble are derived, based on which we conjecture a connection to a recently studied combinatorial object called the “Borel triangle". This allows us to find exactly the mean von Neumann entropy for random phase density matrices and the average entanglement for the corresponding ensemble of bipartite pure states.'
author:
- Arul Lakshminarayan
- Zbigniew Puchała
- Karol Życzkowski
date: 'July 4, 2014'
title: Diagonal unitary entangling gates and contradiagonal quantum states
---
Introduction
============
Entanglement has been at the focus of recent researches in quantum information – for a review see [@HHHHRMP09], as it enables a range of uniquely quantum tasks such as teleportation, quite apart from being a singular nonclassical phenomenon and therefore of fundamental interest. It is well appreciated now that many particle pure states are typically highly entangled [@Lu78; @Pa93; @ZS01; @Hayden06], and share entanglement in a manner that is almost wholly of a multipartite nature. Here typicality refers to ensembles of pure states selected according to the uniform (Haar) measure. If two distinct subsystems of a pure state are such that together they make up the entire system in a typical pure state, then the two subsystems will be largely entangled. In early works, Page and others [@Pa93; @Sen96] had found the average entanglement, which in this case is simply the mean von Neumann entropy of the reduced state, and showed that it is nearly the maximum possible.
More recent studies have explored the distribution of entanglement in such complete bipartite partitions of random states [@Nadal11]. If the two subsystems do not comprise the entire systems, for example two particles in a three particle state, the average entanglement depends on the dimensionality of the subsystems. Roughly speaking if the complementary space of the subsystems (say $A$ and $B$) is smaller, the density matrices $\rho_A$ and $\rho_B$ will be typically negative under partial transpose and therefore $A$ and $B$ are entangled [@Uday03].
Ways to generate entangled states from initially unentangled ones via unitary operators is of natural interest, and in the context of quantum computation implies the construction of appropriate gates. Investigations of entangling power of a unitary quantum gate were initiated by Zanardi and co-workers [@ZZF00; @Za01], while some measures of non–locality were analyzed in [@HVC02; @ZVWS02; @WSB03; @NDDGMOBHH02; @Sc04; @BS09]. A typical quantum gate acting on a composed system consisting of two $N$–level systems can be represented by a random unitary matrix of order $N^2$. Nonlocal properties of such random gates were investigated in [@MKZ13].
In this work we shall analyze a simpler ensemble of diagonal unitary random matrices and will characterize nonlocality of the corresponding quantum gates. It is also naturally related to an ensemble of pure bipartite states in $N^2$ dimensional space whose components in some fixed basis are of the form $e^{i \phi_j}/N^2$, and $\phi_i$ are uniformly distributed random numbers. Such an ensemble has been recently studied as phase-random states [@Nakata2012], and in fact connections to diagonal quantum circuits has been pointed out [@Nakata2014]. Part of the motivation for the study of diagonal quantum gates is that many Hamiltonians have the structure that the basis in which the interaction is diagonal can be chosen to be unentangled. Time evolution is then governed by unitary operators whose entangling parts are diagonal. Recently studies of measurement-based quantum computation has also used diagonal unitary gates and shown that it can still remain superior to classical computation [@Bremner2011; @Hoban2014].
Also explicitly, unitary operators such as $(U_A\otimes U_B) U_{AB}$ occur in the study of coupled systems including kicked quantum tops – see the book of Haake [@Haake]. The coupling could be of the form $J^z_AJ^z_B$, where $J_z^{A,B}$ are spin operators. Thus the nonlocal part of the evolution is diagonal again. It is found numerically that the eigenstates of such operators as well as the time evolution engendered by repeated applications of such operators can create large entanglement well approximated by that of random states [@JNBL02]. However such operators have much fewer number of possible independent elements than Haar distributed unitaries on ${\cal H}^N \otimes {\cal H}^N$. Thus it is of interest to study the origin of such large entanglement.
Furthermore, we are going to study the related ensemble of “unimodular random matrices", comprising complex matrices whose all entries have the same modulus and randomly chosen phases. Such matrices arise from reshaping a pure phase random state as defined in [@Nakata2012]. Note that the usage of the term ’unimodular’ concerns all the entries of a matrix, so such a matrix is [*not*]{} unimodular in the sense of being integer matrices with determinants $\pm1$.
Although the unimodular ensemble differs from the Ginibre ensemble of complex, non-hermitian matrices, with independent, normally distributed elements, it displays the same asymptotic behavior of the level density, which covers uniformly the unit disk. On the other hand the squared singular values of unimodular matrices coincide with eigenvalues of certain specific quantum states of size $N$, the diagonal entries which are equal to $1/N$. As the notion of an “antidiagonal matrix" has entirely different meaning, the density matrices with all diagonal elements equal will be called [*contradiagonal*]{}. Observe that reduced density matrices of random phase pure states are thus contradiagonal. We investigate properties of such an ensemble of quantum states and discuss the contra-diagonalization procedure, which brings any hermitian matrix to such a basis, that all their diagonal elements do coincide.
One may expect that random contradiagonal states correspond to the large entanglement of the pure bipartite states and we show that indeed these states have larger von Neumann entropy than those sampled according to the Ginibre ensemble. The unimodular ensemble, while having no obvious invariance properties, seems to also have remarkable underlying mathematical structure. For example, the average of the moments of the matrices in the ensemble are connected to polynomials with combinatorial interpretations. We evaluate the first four moments and use this to conjecture an exact expression for [*all*]{} of them. We numerically show that this is more than likely to be correct. Analytical continuation of the moments to non-integer powers allows us to evaluate the average von Neumann entropy for this ensemble which appears to be [*exact*]{} for all dimensions.
This paper is organized as follows. In Section II the ensemble of diagonal unitary gates is introduced, while in Section III the related ensemble of unimodular matrices is analyzed. Low order moments are calculated analytically, and a natural conjecture is made for exact expressions for all moments. Numerical evidence is presented for this. Nonlocality and entangling powers of random unitary gates is studied in Sec. IV. An exact expression based on a continuation of moments is presented for the average Schmidt strength of diagonal unitaries, or equivalently of the von Neumann entropy of the random phase states. Procedure of contra-diagonalization of a hermitian matrix, which makes all diagonal elements equal, is introduced in Sec. V. In this section we study in particular the cognate ensemble of random contradiagonal states and show their particular properties concerning the transfer of quantum information. The paper is concluded in Sec VI, and an Appendix reviewing the operator Schmidt decomposition and entangling entropy.
Diagonal bipartite quantum gates
================================
Consider a diagonal unitary matrix $U$ of order $N^2$. Each entry is assumed to be random, so that $U_{\mu \nu}=\delta_{\mu \nu}\exp(i \phi_{\nu})$, where $\phi_{\nu}$ are independent random phases distributed uniformly in $[0,2\pi)$. Such a matrix represents a diagonal unitary gate acting on a bipartite quantum system, a state in ${\cal H}^N \otimes {\cal H}^N$.
Any matrix $U$ acting on the composed Hilbert space ${\cal H}_N \otimes {\cal H}_N$, can be represented in its operator Schmidt form, $$U = \sum_{k=1}^{K} \sqrt{\Lambda_{k}} \; B_k^{\prime}
\otimes B_k^{\prime\prime} ,
\label{VSchmidt}$$ where the Schmidt rank $K \le N^2$. Note that the matrices $B_k^{\prime}$ and $B_k^{\prime\prime}$ of order $N$ in general are non unitary.
It can be shown [@ZB04] that the Schmidt coefficients $\Lambda_k$, $k=1,\dots, K$, can be obtained as squared singular values of the reshuffled matrix $U^R$. This fact is briefly recalled in Appendix A, where the notation is explained. A generic diagonal matrix of size four, after reshuffling forms a non-hermitian matrix of rank two, $$U = \left[\begin{array}{cccc}
U_{11} & 0 & 0 & 0 \\
0 & U_{22} & 0 & 0 \\
0 & 0 & U_{33} & 0 \\
0 & 0 & 0 & U_{44} \\
\end{array}\right], \quad
U^R = \left[\begin{array}{cccc}
U_{11} & 0 & 0 & U_{22} \\
0 & 0 & 0 & 0 \\
0 & 0 & 0 & 0 \\
U_{33} & 0 & 0 & U_{44} \\
\end{array}\right] . \quad
\label{u4} $$ For a diagonal matrix $U$ of order $N^2$ the reshuffled matrix $U^R$ contains $N(N-1)$ columns and rows with all entries equal to zero. Hence the non-zero singular values of $U^R$ are equal to the singular values of a square matrix $A$ of size $N$, obtained by reshaping the diagonal of the unitary gate, $A_{jk}=\exp(i \phi_{\nu})$, where $\nu=(j-1)N+k$. As all entries of $A$ are unimodular and have a random phase, this construction defines an ensemble of random unimodular matrices.
Random unimodular matrices
==========================
Consider a complex square matrix $A$ of size $N$ from the [*unimodular ensemble*]{}, so that a) all entries have the same modulus, $|A_{jk}|=1$, and b) the phases are drawn independently from a uniform distribution, $$A_{jk} \ = \ \exp ( i \phi) , \quad P(\phi)=
\frac{1}{2\pi}, {\rm \quad for \quad}
\phi \in [0,2\pi).
\label{unimod}$$ Such a random matrix $A$ could be called a [*pre–Hadamard*]{} as all entries have the same modulus, so choosing an appropriate set of phases it may become unitary, and thus belong to the class of complex Hadamard matrices [@TZ06]. In our model all phases are random and non-correlated, so a typical matrix from this ensemble exhibits effects of strong non-unitarity. The ensemble of unimodular matrices $A$ with all independent, well–behaved, identically distributed entries is of the Wigner type. Thus this non-hermitian ensemble or random matrices satisfies asymptotically the circular law of Girko [@TV11].
As shown in Fig. \[fig\_100\] already for $N=100$ the spectral density for the unimodular ensemble is close to uniform in the unit disk. Furthermore, the distribution of rescaled squared singular values, $x={\rm eig} (AA^{\dagger})/N$, is asymptotically described by the Marchenko–Pastur (MP) distribution $P_{MP}(x)= \frac{1}{\pi}\sqrt{1/x-1/4}$ for $x \in [0,4]$, characteristic to the Ginibre ensemble. The moments of this distribution are given by the Catalan numbers, while its entropy reads $-\int_0^4 x \log x P_{MP}(x)dx=-1/2$. As $x=N \lambda$, where $\lambda$ denotes the eigenvalue of a normalized density matrix $$\label{AAdag}
\rho=AA^{\dagger}/N^2$$ satisfying ${\mathrm{Tr}}\rho=1$, the average entropy of spectrum of $\rho$ behaves asymptotically as $\log N-1/2$. Note that this behavior is characteristic to random quantum states distributed uniformly with respect to the Hilbert–Schmidt measure [@SZ04] in the entire set of quantum states of a given dimension.
Although for large $N$ statistical properties of the unimodular ensemble coincide with those of complex Ginibre ensemble, deviations are visible for small matrix size. To visualize these effects we studied the moments of the distribution of squared singular values $M_m =\int x^m P(x)dx$. Fig. \[fig\_moment\] shows a comparison of the moments $M_2$, $M_3$ and $M_4$ for random matrices of the unimodular ensemble and the Hilbert–Schmidt ensemble of order $N$. In the later case analytical predictions for the moments are known as the traces of the random states $\rho_N$ of size $N$ distributed according to the Hilbert–Schmidt measure read [@Lu78; @SZ04], $$\br {{\mathrm{Tr}}} \rho_N^2\kt_{HS} = \frac{2N}{N^2+1} , \quad
\br {{\mathrm{Tr}}} \rho_N^3 \kt_{HS}= \frac {5N^2+1}{(N^2+1)(N^2+2)} ,
\label{momHS}$$ and due to rescaling of the variable $x$ one has $M_m=N^{m-1} {{\mathrm{Tr}}} \rho^m$. Numerical data, such as that presented in Fig. (\[fig\_moment\]), show that for a given $N$ the moments for the unimodular ensemble are smaller, so the corresponding distribution are narrower, even though for large $N$ both distributions tend to the limiting Marchenko–Pastur distribution.
Note that the averages moments for the distribution of squared singular values for random Ginibre matrices of a given size $N$, derived recently in [@AKW13], coincide with the predictions (\[momHS\]) for the HS ensemble only in the asymptotic case $N\to \infty$. In the former ensemble the constraint concerns the average trace $\langle {{\mathrm{Tr}}} GG^{\dagger}\rangle$, while in the latter each random matrix has a fixed trace, ${\mathrm{Tr}}\rho=1$, so that this difference asymptotically vanishes.
Lower order moments and a conjecture for all orders
---------------------------------------------------
The lower order moments of the unimodular ensemble can be exactly evaluated and compared to the above case. In fact we will calculate exactly moments till the fourth and conjecture an exact formula for any moment. The density matrix elements are \_[\_1 \_2]{}=\_[l\_1=1]{}\^[N]{}. \[momUni\]
### The second moment $\br {\mathrm{Tr}}\rho_N^2 \kt_{UE} $
The second moment while being the simplest, also serves as a measure of purity of a given quantum mixed state. We have \^2= \_[\_1 \_2]{}\_[l\_1 l\_2]{}. \[trrho2A\] On averaging over the uniform phases the only terms that would survive are those cases for which the phase vanishes. This happens if $\alpha_1=\alpha_2$ for arbitrary $l_1$ and $l_2$. Thus this case contributes \_[\_1]{}\_[l\_1l\_2]{} 1==. Indeed this is the “diagonal" contribution. The phase also vanishes if $\alpha_1 \neq \alpha_2$, but $l_1=l_2$. This “off-diagonal" contribution is \_[\_1\_2]{}\_[l\_1]{} 1= N(N-1)N=. As these exhaust the exclusive possibilities, the average second moment for the level density of related to the unimodular ensemble reads \_N\^2 \_[UE]{} = . \[trrho2B\] Equivalently $M_2=(2N-1)/N$, which indeed tends to $2$ for large $N$. Also note that $\br {\mathrm{Tr}}\rho^2 \kt_{HS}-\br {\mathrm{Tr}}\rho^2 \kt_{UE}=(N-1)^2/[N^2 (N^2+1)]>0$, indicating that the density matrices constructed from the unimodular ensemble are on average more mixed than those from the Ginibre ensemble.
### The third moment $\br {\mathrm{Tr}}\rho_N^3 \kt_{UE} $
The third moment is obtained by considering all those cases when $$\phi_{\alpha_1 l_1}-\phi_{\alpha_1 l_2}+\phi_{\alpha_2l_2}-\phi_{\alpha_2 l_3}+\phi_{\alpha_3 l_3} -\phi_{\alpha_3 l_1}$$ vanishes for arbitrary sets of phases, where $\alpha_i$ and $l_i$ take values in $1, \cdots , N$. The indexes are ordered in such a way that a bijection to a standard counting problem becomes possible. Starting from the first pair, the sign is reversed while the second index is new in the second pair. The third pair is obtained by again reversing the sign of the second but now the first index becomes new, and so on. Finally when there are $6$ pairs, the last index is the same as that of the first pair. It is clear that this generalizes to the moment of order $k$, where there are $2k$ such pairs.
At this level a bijection to several standard problems in counting that involve the Catalan numbers $C_k$ [@Koshy] is possible. For example that of matching 3 pairs of parentheses: $()()(),(())(),()(()),((())),(()())$ is relevant to the third moment, each parenthesis represents the pair of indexes at the corresponding place. The matched pair of parentheses imply that the pair of indexes are equal. The first possibility and its translation in terms of indexes is: $()()()$: $\alpha_1 l_1=\alpha_1 l_2$, $\alpha_2 l_2=\alpha_2 l_3$, $\alpha_3 l_3=\alpha_3 l_1$, or $l_1=l_2=l_3$. The second $(())():$ $\alpha_1 l_1=\alpha_2 l_3$, $\alpha_1 l_2=\alpha_2 l_2$, $\alpha_3 l_3=\alpha_3 l_1$, or $\alpha_1=\alpha_2$, $l_1=l_3$. Similarly $()(())$: $\alpha_2=\alpha_3$, $l_1=l_2$, $((()))$: $\alpha_1=\alpha_3$, $l_2=l_3$, $(()())$: $\alpha_1=\alpha_2=\alpha_3$. The unconstrained indexes can take arbitrary values between $1$ and $N$. Another bijection is between the indexes and points that are joined by noncrossing semicircles. Thus the contributions from $()()()$ and $(()())$ are respectively $$\sum_{l_1}\sum_{\alpha_1,\alpha_2,\alpha_3} 1= N^4,\; \sum_{\alpha_1}\sideset{}{'}\sum_{l_1,l_2,l_3} 1= N^4-N^2.$$ The last sum is restricted in the sense that it does not include the case $l_1=l_2=l_3$ that is already included in the first count. The other three cases contribute equally $$\sum_{\alpha_1\ne \alpha_3} \sum_{l_1\ne l_2}1=N^2(1-N)^2,$$ where the distinct indexes are always unequal; if they are equal it will reduce to a term considered in the first two cases. Thus putting them all together we get \^3 \_[UE]{} = (5N\^2-6N+2), and $M_3=N^2 \br {\mathrm{Tr}}\;\rho_N^3 \kt =(5-6/N+2/N^2)$.
The $k$–[th]{} moment $\br {\mathrm{Tr}}\rho^k\kt$ is of the form $P_k(N)/N^{2(k-1)}$, where $P_k(N)$ is a degree $k-1$ polynomial in $N$ whose leading term’s coefficient is $C_k=\frac{1}{k+1}\binom{2k}{k}$, the $k$–[th]{} Catalan number. Thus it follows that the moments $M_k=N^{k-1} \br {\mathrm{Tr}}\rho^k\kt$ tend to $C_k$ and hence the asymptotic density is described by the universal Marchenko-Pastur distribution. Writing explicit expressions for the moments $M_k$ for the unimodular ensemble we are in position to quantify the deviations from the asymptotic universal MP distribution.
### The fourth moment $\br {\mathrm{Tr}}\rho_N^4 \kt_{UE} $
We will now evaluate explicitly the fourth moment that presents some challenges and then [*conjecture*]{} an [*exact*]{} expression for $P_{k}(N)$. There are 14 different parenthesizations for the $k=4$ case with $8$ pairs of indexes involved. There are 3 “contractions" in each case. For instance the contractions corresponding to $()(())()$ are $l_1=l_2$, $\alpha_2=\alpha_3$, $l_4=l_1$. Thus there are $8-3=5$ sums that are unconstrained from each of the $14$ parenthesizations and hence the leading term in $N^8 \br {\mathrm{Tr}}\rho_N^4\kt$ is $14 N^5$. Generalizing, to the $k$–th moment, the number of contractions is $k-1$. This is seen by writing the alternating indexes as actual products and sums while respecting their distinctness and requiring $\sum_{i=1}^k \alpha_{i}(l_i-l_{i+1})=0$ with $l_{k+1}=l_1$. Setting say $\alpha_{n}l_n =\alpha_{m}l_{m+1}$ for some $n$ and $m$ reduces the number of terms by $1$. Due to periodic boundary conditions, the number of contractions that will set the whole sum to 0 is $k-1$. Thus it follows that in general the leading term in $N^{2k} \br {\mathrm{Tr}}\rho^k\kt$ is $C_k N^{2k-(k-1)}=C_k N^{k+1}$.
Overcounting however lowers the value of $N^8 \br {\mathrm{Tr}}\rho_N^4\kt$ from $14 N^5$, and in general from $C_k N^{k+1}$. Sticking to the $k=4$ case, the combined contributions from $()()()()$ and $(()()())$ which either contract all $l_i$ or all $\alpha_i$ respectively is $2N^5-N^2$, $N^2$ being the double count of all $\alpha_i$ being the same and all $l_i$ being the same. The other $12$ contributions involve sums such as $$\sum_{\alpha_1=\alpha_2}\sum_{\alpha_3=\alpha_4 \ne \alpha_1}\sum_{l_1=l_3} \sideset{}{'}\sum_{l_2,l_4}1= (N^2-1)(N-1)N^2,$$ where the restricted sum eliminates the cases when $l_2=l_4=l_1=l_3$, which has already been considered. While this case corresponds to the parenthesization $(())(())$, the other $11$ have sums over 5 indexes with similar restrictions.
Thus these contribute $12N^2(N-1)(N^2-1)$, however there still remains some overcounting to be accounted for. For example the above case includes instances when $l_2=l_1=l_3$ but $\ne l_4$ which is also possible when the contracted indexes are $\alpha_2=\alpha_4$, and $l_1=l_2=l_3$ and corresponds to the paranthesization $()()(())$. Exhaustive enumeration of these terms that originate from contracting $4$ indexes reveals that there are $16$ such terms each of which gives $$\sum_{\alpha_1=\alpha_2}\sum_{\alpha_3=\alpha_4\ne \alpha_1} \sum_{l_1=l_2=l_3}\sum_{l_4\ne l_1} 1= N^2(N-1)^2.$$ There is no further overcounting as the cases with $>4$ contractions have already been properly included. Finally in total the contribution is $2N^5-N^2+12(N^2-1)(N-1)N^2 - 16 N^2(N-1)^2=14N^5-28 N^4 +20N^3 -5N^2$ and we get \_N\^4\_[UE]{}=(14N\^3-28 N\^2+20N-5).
### A conjecture for all moments $\br {\mathrm{Tr}}\rho_N^n \kt_{UE} $
The complexity of the counting problem is naturally increasing. However having found the polynomials $P_2(N)=2N-1$, $P_3(N)=5N^2-6N+2$ and $P_4(N)=14N^3-28N^2+20N^3-5$ exactly, the following are evident: they have alternating signs, satisfy $P_k(1)=1$, and the constants (coefficients of $N^0$) have the absolute value $1$, $2$ and $5$ which are themselves Catalan numbers. That $P_k(1)=1$ follows simply as for $N=1$, there is a only a single pure phase $e^{i \phi}$ and the “density matrix" is simply $e^{i \phi} e^{-i \phi}=1$.
Based on the triangle of coefficients $\{1,\{2,1\},\{5,6,2\},\{14,28,20,5\}\}$ a search in the On-line Encyclopedia of Integer Sequences [@Sloane] (OEIS) returns two sequences A062991 and A234950. In the first of which is the signed version (that we encounter), it arises as generalizations of Pascal’s triangles via Riordan arrays [@Barry2013]. In the second the unsigned version, which the authors call [*Borel’s triangle*]{} [@Francisco2013] and make connections to new counting problems in commutative algebra and discrete geometry. Going beyond what we have above, the first seven rows of the unsigned triangle reads: $$\begin{array}{lllllll}
1&&&&&&\\
2&1&&&&&\\
5&6&2&&&&\\
14&28&20&5&&&\\
42&120&135&70&14&&\\
132& 495& 770& 616& 252& 42&\\
429&2002& 4004& 4368& 2730& 924& 132\\
\ldots&&&&&& \end{array}$$ The left and rightmost entries are Catalan numbers. The entry $f_{n,k}$ ($n\ge 0,\, k \ge 0)$ of the Borel triangle is f\_[n,k]{}=\_[s=0]{}\^[n]{} C\_[n,s]{} where $C_{n,s}$ is [*Catalan’s triangle*]{}: $$\begin{array}{lllll}
1&&&&\\
1&1&&&\\
1&2&2&&\\
1&3&5&5&\\
1&4&9&14&14\\
\ldots&&&& \end{array}$$ which satisfies the recursion $C_{n,k}=C_{n-1,k}+C_{n,k-1}$, that is the entries are sums of the one to the left and the one above. The first column of all $1$ is the “boundary condition" $C_{n,0}=1$. Explicit formula for $C_{n,k}$ and $f_{n,k}$ are available [@Barry2013]: $$C_{n,k}=\dfrac{(n+k)!(n-k+1)}{k!(n+1)!},$$ $$f_{n,k}=\frac{1}{n+1}\binom{2n+2}{n-k}\binom{n+k}{k}.$$ It is then natural to [*conjecture*]{} that for $n\ge 1$ the average moments of the level density of the random matrices of size $N$ are \[momentUE\] \_N\^n \_[UE]{}= \_[k=0]{}\^[n-1]{} (-1)\^[k]{} f\_[n-1,k]{}N\^[n-k-1]{}. The smallest unproven case is $n=5$, which can be simply read off from Borel’s triangle: \^5 \_[UE]{} = ( 42N\^4-120N\^3+135 N\^2-70N+14). \[Eq:tr5\] Fig. \[fig\_tr5\] displays the numerical calculations of moment $M_5=N^4 \br {\mathrm{Tr}}\rho^5 \kt $, as well as $M_6$ and $M_7$ from a million random realizations of the ensemble and shows how well this conjecture fares. There seems to be no room for doubting the correctness of the conjecture.
A calculation of the first few cumulants from the moments $M_n$ of the scaled variables $N \lambda$ results in (apart from $\kappa_1=1$)
\_2=(N-1),\
\_3=(N-1)(N-2),\
\_4=-(N-1)(4N-5),\
\_5=-(N-1)(N-2)(4N\^2+2N-7).
As $N\rarrow \infty$ these tend to $\{1,1,1,0,-4, \ldots\}$, the initial cumulants corresponding to the moments being the Catalan numbers $\{1,2,5,14,42, \cdots\}$.
The moments of the unimodular ensemble themselves also seem to have combinatorial significance. For example the moments for $N=3$ are such that $3^{(n-1)}M_n(N=3)$ is the integer sequence $\{1,5,29,181, 1181, \cdots\}$. If we include an additional $1$ corresponding to $M_0$, this sequence is found as a column in the entry A183134 of the OEIS. Indeed the other columns of the square array of this entry are similarly the moments for different values of $N$, $N=1,2, \cdots$. This prompts the additional conjecture that the $N^{2(n-1)}{\mathrm{Tr}}(\rho^n_{N}) $ is the same as the number of $N$-alphabet words of length $2n$ beginning with the first character of the alphabet by repeatedly inserting doublets into the initially empty word, as this is the counting problem that is stated in the OEIS entries (see also [@Kassel2013]). For example in the case of $n=2$ and $N=2$, the alphabet set is binary $\{ab,\}$. Doublets are repeated alphabets. Thus inserting two doublets (for $n=2$) one gets $aaaa$, $aabb$, $abba$ as the three possibilities that start with $a$. This coincides with $2N-1$ that we derived above.
Having explored the moments of the unimodular ensemble we now turn to one of our central motivations, finding the entangling power of diagonal unitaries.
Entropy of the unimodular ensemble and nonlocality of random diagonal gates,
=============================================================================
By construction, the average entropy of squared singular values of random unimodular matrices $A$ is equal to the average entropy of entanglement for the corresponding unitary gates $U=A^R$. As squared singular values of $A$ are asymptotically described by the Marchenko–Pastur distribution, making use of the Page formula [@Pa93] we infer that the mean entropy of entanglement (\[shannon\]) of a random diagonal gate behaves as $$\langle S(U) \rangle_{\rm diag} \ = \
\langle S(\rho=AA^{\dag}/N^2) \rangle_{UE} \ \approx \ \log N-1/2.
\label{entropy1}$$ This result forms approximately a half of the entropy of generic Haar random unitary matrices [@MKZ13] $$\langle S(U) \rangle_{\rm Haar}
\ \approx \ 2 \log N-1/2.
\label{entropy2}$$
Based on the discussion of moments in the previous section, and taking them to be exact allows us to find what appears to be an exact expression for the average entropy of entanglement, which can be interpreted as entanglement in a random ensemble of states $\br S(\rho)\kt_{UE}$ where $\rho$ is defined in Eq. (\[AAdag\]) or the average entanglement of diagonal unitary gates via Eq. (\[shannon\]). Using the view of state entanglement, from the last section, the expression in Eq. (\[momentUE\]) can be used to write the $n^{th}$ moment as
\_N\^n \_[UE]{}= \_2F\_1(n,1-n;2+n;1/N).
Using this we can continue the moments to noninteger powers, so that we have $\sum_i \lambda_i^x$ for $x$ real. Observing that the average entropy $\br S(\rho)\kt_{UE}$ is the limit of $(1-\sum_i \lambda_i^x)/(x-1)$ as $x \rarrow 1^+$, we have that S()\_[UE]{} =-df(x)/dx \_[x=1]{} where $f(x)=\br {\mathrm{Tr}}\rho_N^x \kt=$ \_2F\_1(x,1-x;2+x;1/N). \[eq\_momx\] While it is not evident that such a continuation be exact, that it is indeed very likely to be so is illustrated in Fig. (\[fig\_momx\]), where the moments are plotted for $1\le x \le 2$ for small values of $N$. This gives us confidence that the entropy found from such a procedure is also [*exact*]{}.
To evaluate $-f'(1)$ one needs to evaluate the derivatives of gamma functions and the hypergeometric function with respect to their parameters. Using that $\Gamma'(z)=\Gamma(z) \psi_0(z)$, where $\psi_0(z)$ is the digamma function we get that - |\_[x=1]{}=N -, \[eq\_ent\_der1\] where the origin of $1/2$ is due to the fact that $\psi_0(3)-\psi_0(2)=1/2$. The derivative of the hypergeometric function can be evaluated using its definition as an infinite series. We have to compute \_[m=0]{}\^ , for which we need the derivatives of the Pochhammer symbols which are defined as $(a)_n =a (a+1) \cdots (a+n-1)$. Due to the fact that we need to evaluate the derivative at $x=1$ (and $(0)_m=0$ for $m>1$, while $(0)_0=1$), we only need that $d (1-x)_m/dx|_{x=1}=-(m-1)!$ which is easy to see from the definition of the symbol. Putting these together we get that
&- \_2F\_1(x,1-x,x+2;1/N)|\_[x=1]{}=\
&\_[m=1]{}\^ =\
&-N-(N-1)\^2 (1-). \[eq\_ent\_der2\]
The last equality is obtained as the infinite sum can be evaluated by elementary means, integrating thrice the identity $1/(1-x)= 1+x+x^2+ \cdots$. and thus finally the average entropy is the sum of the results in Eqs. (\[eq\_ent\_der1\]) and \[eq\_ent\_der2\]: S()\_[UE]{}=N -(N-1)-(N-1)\^2(). \[entropyUE\]
Numerical results presented in Fig. (\[fig\_ent\]) provide further arguments that the above expressions for the average entropy are exact for any dimension. The differences between the formula and numerical simulations are shown to be smaller than $1/\sqrt{N_S}$, where $N_S$ denotes the size of the numerical sample. For large $N$ it is easy to see that this approaches $\log N-1/2$ with the neglected terms being of order $1/N$, in agreement with what is expected from the Marchenko-Pastur distribution and from the Page formula [@Pa93] for the HS ensemble. For instance, if $N=2$, the exact density of the eigenvalues in $[0,1]$ reads $1/(\pi\sqrt{y(1-y)})$, as discussed in the next section. Hence the average entropy is \_0\^1 =4 -1, which agrees with Eq. (\[entropyUE\]). We may compare this with the exact entropy from the HS ensemble, which is $\br S(\rho)\kt _{HS}=\sum_{k=N+1}^{N^2}1/k-(N-1)/(2N)$ [@Pa93]. For example for $N=2$ this gives $1/3$ which is smaller than that for the UE ensemble that is $\log 4-1 \approx 0.39$. Although the difference decreases with the matrix size, the relation $\br S(\rho)\kt _{HS}<\br S(\rho)\kt _{UE}$ holds true, which indicates again the enhanced average entanglement in the unimodular ensemble.
Another measure of nonlocality of gates is the so-called [*entangling power*]{} based on the ability of operators to create subsystem mixed states from originally unentangled pure bipartite states. If $|\psi_1\kt \otimes |\psi_2\kt $ is an unentangled state in ${\cal H}_N \otimes {\cal H}_N$, and $U$ is an unitary operator on this space, its entangling power as defined by Zanardi, Zalka and Faoro [@ZZF00] is e\_p(U)=\^[\_!,\_2]{}, the average being over all product states. Any entanglement measure can be used for $E$, the one that was used in [@ZZF00] being the simplest useful one, the linear entropy: $E(|\psi\kt)= 1-{\mathrm{Tr}}_1 \mu^2,$ where $\mu\equiv {\mathrm{Tr}}_2 |\psi\kt \br \psi|$ is the reduced density matrix of the subsystem labeled by 1.
The case of interest in the present work is one where the matrix $U$ is diagonal and hence the following is obtained: | l |U|\_1 |\_2= e\^[i \_[l]{}]{} |\_1 l |\_2, where we have used $\br \alpha l |U |\beta m \kt = e^{i \phi_{\alpha l}} \delta_{\alpha \beta} \delta_{lm}$. The reduced density matrix is \_[\_1 \_2]{}=\_[l]{}e\^[i (\_[\_1 l]{}- \_[\_2 l]{})]{} \_1|\_1l|\_2\_1|\_2\_2|l , and therefore
\^2 =\_[\_1,\_2,l\_1,l\_2]{}e\^[i(\_[\_1 l\_1]{}-\_[\_1 l\_2]{}+\_[\_2 l\_2]{}-\_[\_2 l\_1]{})]{}\
|\_1|\_1|\^2 |l\_1|\_2|\^2 |\_2|\_1|\^2 |l\_2 |\_2 |\^2.
Averaging over the states $|\psi_{1,2}\kt$ can be done assuming that they are random vectors distributed uniformly according to the invariant Haar measure. Further assuming the general case of complex random states [@Brody81], the following are the average of the products of two intensity components: =. Using this, it follows that =. The connection to ${\mathrm{Tr}}\rho^2$ of the last section follows from Eq. (\[trrho2A\]). Thus the entangling power of diagonal unitaries $e_p(U)=1-\overline{{\mathrm{Tr}}\mu^2}$ is directly related to the second moment of the squared singular values of the reshaped matrix. The average entangling power, now averaged over all phases in the diagonal unitary gates is e\_p(U)\_[diag]{} = 1-= ()\^2, where the result $\br {\mathrm{Tr}}\rho^2\kt_{UE} = (2N-1)/N^2$ has been used from the previous section. This can be compared with the average entangling power of unitary gates [@ZZF00]: $\br e_p(U)\kt = (N-1)^2/(N^2+1)$, which is only marginally larger. The entropies of full unitaries were almost twice as large as the diagonal ones. At the level of the purity however one still sees the difference in that $\br \overline{{\mathrm{Tr}}\mu^2}\kt=4N/(N-1)^2 \approx 4/N$ is double that of the reduced density matrix of typical random states in ${\cal H}_N \otimes {\cal H}_N$, which reads $2/N$ [@Lu78].
Contradiagonal hermitian matrices
=================================
Any ensemble of random matrices, allows one to generate an ensemble of quantum states [@ZPNC11]. Taking a random matrix $A$, one writes $\sigma=AA^{\dagger}/{{\mathrm{Tr}}} AA^{\dagger}$ to get a random density matrix: a hermitian, positive operator normalized by the trace condition ${{\mathrm{Tr}}} \sigma=1$. In the case of unimodular random matrices (\[unimod\]) one has ${{\mathrm{Tr}}} AA^{\dagger}=N^2$, hence $\sigma=AA^{\dagger}/N^2$.
Observe that by construction of the unimodular matrix $A$ the corresponding positive Wishart matrix $AA^{\dagger}$ has all diagonal elements [*equal*]{}. Thus the diagonal elements of the corresponding random density matrix $\rho$ read $\rho_{ii}=1/N$, where $i=1,\dots,N$. In other words, the state $\rho$ is represented in such a particular basis $\{|1\rangle, |2\rangle, \dots, |N\rangle\}$ that the expectation values among each of the basis states are equal. Thus the entropy of an orthogonal measurement in this basis is maximal and equal to $\log N$. We show below that such a basis is dual to the basis in which a given state is diagonal, in sense that the norm of all the off-diagonal elements is maximal. Thus any density matrix $\sigma=AA^{\dagger}/N^2$ constructed out of a random unimodular matrix $A$ has all diagonal elements equal and therefore will be called [*contradiagonal*]{}.
Procedure of contra–diagonalization of a matrix
-----------------------------------------------
Let $H$ be a Hermitian matrix of order $N$ and let $G=VHV^{\dagger}$ be a unitarily similar matrix, as $VV^{\dagger} = V^{\dagger}V = {\mathbb I}_N$. All matrices from this orbit share the same spectrum and posses the same trace, ${{\mathrm{Tr}}} G = {{\mathrm{Tr}}} H =:t$. For any fixed $H$ we are going to analyze the sum of the squared moduli of off-diagonal elements of $G$ and define a function $f(V)=\sum_{i \ne j}
|G_{ij}|^2$. Let us denote by $D$ any hermitian matrix $VHV^{\dagger}$ for which the function $f$ becomes minimal, $$D=U_{min}HU_{min}^{\dagger} \; : f(U_{min})= \min_{V \in U(N)} \sum_{i \ne j} |(VHV^{\dagger})_{ij}|^2 .
\label{minD}$$ In a similar way let $A$ represent a hermitian matrix $VHV^{\dagger}$ for which the function $f$ becomes maximal, $$A=U_{max}HU_{max}^{\dagger} \; : f(U_{max})= \max_{V \in U(N)} \sum_{i \ne j} |(VHV^{\dagger})_{ij}|^2 .
\label{maxA}$$ It is clear that the minimum $f(U_{min})=0$ is achieved for a matrix $U=U_{min}$ consisting of eigenvectors of $H$, so $D$ is a diagonal matrix of eigenvalues of $H$ or any other matrix similar with respect to permutations $D'=PDP^T$. As the standard procedure to find $D=UHU^{\dagger}$ is called diagonalization, the transformation of $H$ by $U_{max}$, leading to the maximum in (\[maxA\]), will be called [*contra–diagonalization*]{}. For any given hermitian $H$ we shall show below how to find its contra–diagonalizing matrix $U_{\rm max}$.
It is well known [@hj2] that any matrix can be unitarily transformed to a form, in which all the diagonal entries are equal. As the trace of a Hermitian matrix $H$ is unitarily invariant this constant reads $H_{jj}={{\mathrm{Tr}}} H/N$ for $j=1,\dots, N$.
Let $F$ denote a complex Hadamard matrix [@TZ06] of order $N$, so that $F$ is unitary and the moduli of all its entries are equal, $|F_{jk}|=1/\sqrt{N}$. As a typical example let us mention the [*Fourier matrix*]{} of size $N$ with entries $(F_N)_{jk}=\exp(i jk \pi/N)/\sqrt{N}$ for $j,k=0,\dots, N-1$. Two complex Hadamard matrices are called equivalent, written $H_2 \sim H_1$, if they are equivalent up to enphasing and permutations, $H_2=P_1E_2H_1 E_2P_2$. Here $E_1$ and $E_2$ denote diagonal unitary matrices while $P_1$ and $P_2$ represent permutation matrices of order $N$. For $N=2,3$ and $N=5$ all complex Hadamard matrices are equivalent to the Fourier matrices $F_2$, $F_3$ and $F_5$ respectively, see [@TZ06].
In order to compare spectra of hermitian matrices it is convenient to use the notion of majorization. Consider vectors of size $N$ ordered decreasingly, $x_1 \ge x_2 \ge \dots x_N$. A vector $y$ is said to [*majorize*]{} [@MO79] vector $x$, written $x \prec y$, if partial sums satisfy following inequalities $\sum_{i=1}^m {x}_i \leq \sum_{i=1}^m {y}_i$ for $m=1,\dots, N-1$ and additionally $\sum_{i=1}^N {x}_i = \sum_{i=1}^N {y}_i$. A function $f:\mathbb{R}^N \to R$ is said to be Schur convex when $x \prec y$ implies $f(x) \leq f(y)$.
Now we can formulate the following result.
\[lemma:anti\] Let $H$ be a Hermitian matrix of order $N$ and let $V$ be unitary. Then
a\) the maximum in (\[maxA\]) is obtained for $U_{\max}=F U_{min}$ where $U_{min}$ denotes the matrix of eigenvectors of $H$ and $F$ is a complex Hadamard matrix,
b\) the matrix $ A=U_{max}HU_{max}^{\dagger}$ obtained in this way is contradiagonal, $A_{jj}={{\mathrm{Tr}}}H/N$ for $j=1,\dots, N$,
c\) the maximum reads $f_{max}(V)={{\mathrm{Tr}}} H^2 - ({{\mathrm{Tr}}} H)^2/N $.
First to show item b) we consider first diagonal matrix $H=D$ and take an arbitrary complex Hadamard matrix $F$ and find that $A = F D F^{\dagger}$ is contradiagonal, as all its diagonal elements are equal, $$\begin{split}
(F D F^{\dagger})_{ii} = \sum_{k l} F_{ik} D_{kl} (F^{\dagger})_{l i} =
\sum_{k} F_{ik} D_{k k} \overline{F_{i k}} =\\
\sum_{k} |F_{ik}|^2 D_{k k}
=\frac{1}{N} {\mathrm{Tr}}D.
\end{split}$$ Consider now any hermitian matrix $H$ and denote by $U$ the matrix of its eigenvectors. Thus taking a unitary matrix $U_{\max}=F U_{min}$ we see that the transformed matrix $$\label{contr1}
A=U_{max}HU_{max}^{\dagger}=F U_{min} H U_{min}^{\dagger} F^{\dagger}$$ is contradiagonal, as all its diagonal elements are equal, as stated in item b).
To prove item a) we note, that the sum in (\[maxA\]) is maximized, if and only if the sum $$\sum_{i} |(VHV^{\dagger})_{ii}|^2$$ is minimized, since the vector of diagonal elements majorizes the constant vector of the same sum, and the sum of squares is a Schur-convex function we obtain the result.
The last item c) is obtained from the definition of the function $f$ by computing the trace of $H^2$ and subtracting the sum of squared elements at the diagonal.
A generalization of this procedure allowing to find a basis in which Hermitian matrix with spectrum $y$ has diagonal $x \prec y$ is presented in Appendix \[sec:robin-hood\].
\[lemma:max-dist\] Consider a family of unitarily similar hermitian matrices $$G = V D V^{\dagger},$$ where $D$ is a given diagonal matrix, then the maximal Hilbert-Schmidt distance between the orbit of the unitarily similar matrices to diagonal matrix $D$ optimized with respect to all permutation matrices is given by $$\max_{V \in U(N)} \min_{P \in Perm}\! \| D - PVDV^{\dagger}P^T\|^2_{\mathrm{HS}}
=
2 \left( {\mathrm{Tr}}D^2 - \frac{({\mathrm{Tr}}D)^2}{N}\right),$$ and for the optimal matrix $V_{\mathrm{opt}}$ one can take any complex Hadamard matrix $F$.
Consider an arbitrary hermitian matrix $H$, such that $H= U D U^\dagger$, so its diagonalization corresponds to transforming the basis by the matrix $U$ consisting of eigenvectors. The above lemma explains why representing it in the basis $W = F U ^\dagger$ can be called contra–diagonalization, as the matrix $A = W H W^\dagger$ has all diagonal elements equal and is as far from the diagonal matrix as possible.
To prove the lemma we write $$\begin{split}
\max_{V \in U(N)} \min_{P \in Perm} \| D - PVDV^{\dagger}P^T\|^2_{\mathrm{HS}}
=\\
\max_{V \in U(N)} \min_{P \in Perm} 2 {\mathrm{Tr}}D^2 - 2 {\mathrm{Tr}}D P V D V^{\dagger}P^T.
\end{split}$$ Next we note, that $$\max_{P \in Perm} {\mathrm{Tr}}D P V D V^{\dagger}P^T =
\max_{P \in Perm} \langle d | P q^{(V)}\rangle,$$ where $d$ is a diagonal of matrix $D$ and $q^{(V)}$ is a diagonal of matrix $ V
D V^{\dagger}$. It is easy to see that one obtain the maximum value if the vectors are ordered in the same way, i.e. $$\max_{P \in Perm} \langle d | P q^{(V)}\rangle =
\langle d^{\downarrow} | (q^{(V)})^{\downarrow}\rangle.$$ To perform minimization over the set of unitary matrices $V\in U(N)$ we note that the minimum value for the above inner product is achieved, if vector $q$ is minimal in the majorization partial order, $$\min_{V \in U(N)}\langle d^{\downarrow} | (q^{(V)})^{\downarrow}\rangle
\leq \frac{\sum d_i}{N} \sum d_i = \frac{({\mathrm{Tr}}D)^2}{N}.$$ The above minimum can be achieved if the unitary matrix $V$ is complex Hadamard for instance the Fourier matrix $F_N$.
To summarize the proof we write $$\begin{split}
\max_{V \in U(N)} \min_{P \in Perm} \| D - PVDV^{\dagger}P^T\|^2_{\mathrm{HS}}
\\
= 2 \left( {\mathrm{Tr}}D^2 - \frac{1}{N} ({\mathrm{Tr}}D)^2\right),
\end{split}$$ and every complex Hadamard matrix $V$ gives the maximum and in this case one can take any permutation matrix $P$. $\Box$
Contradiagonal density matrices
-------------------------------
The statements on contra-diagonalization introduced above for arbitrary hermitian (or normal) matrices can be now used for a positive definite density matrices $\rho^{\dagger}=\rho\ge 0$ normalized as ${\mathrm{Tr}}\rho=1$. Thus a quantum state $\sigma$ of size $N$ will be called [*contradiagonal*]{} if $\sigma_{ii}=1/N$ for $i=1,\dots, N$.
Spectral density for the ensemble of contradiagonal states obtained form a random unimodular matrix $A$ by Eq. (\[AAdag\]) is shown Fig. \[fig\_234\] for $N=2,3,4$. For $N=2$ a random contradiagonal density matrix takes the form $ \sigma= \frac{1}{2}
\left[ \begin{array}{c c}
1 & z \\
{\bar z} & 1
\end{array} \right] , $ where $z=e^{i \psi_1} +e^{i \psi_2}$ and the phases $\psi_1$ and $\psi_2$ are random. Thus the rescaled eigenvalues of $\sigma$ are distributed according to the arcsin law, $P_{\rm As}(x)=1/ \pi \sqrt{x(2-x)}$ for $x \in (0,2)$, as shown in Fig. \[fig\_234\].
Note oscillations of the level density $P(x)$ present for $N \ge 3$. Observe that the conjectures above yield all the moments for any value of the matrix size $N$. It is then a classic moment problem to find the corresponding density. Curiously, there exists densities that have the exact same moments on different intervals and are found in [@ZPNC11] as a sequence of densities that converges to the MP distribution. We were however unable to solve this moment problem, to find the actual oscillatory one that is found for the unimodular ensemble. While the appearance of multiple densities with the same moments is known in the literature [@Korner], this seems to be a curious case as the densities have support in $(0,1]$. The fact that the densities diverge at the origin makes the current moment problem not belong to the class of Haussdorf moment problems that treat compact intervals and absolutely continuous densities [@Korner].
The Schur–Horn theorem states [@BZ06] that for any density matrix $\rho$ its diagonal is majorized by the spectrum, ${\rm diag}(\rho) \prec {\rm eig}(\rho)$. The uniform vector $x_*=\{1/N, \dots, 1/N\}$ is majorized by any other probability vector. This observation implies the following fact
\[prop3\] Let $\sigma$ denotes a contradiagonal state of order $N$, so that $\sigma_{ii}=1/N$, and let $U$ be a unitary matrix of order $N$. Then the following majorization relation holds $${\rm diag}(\sigma) \ \prec \ {\rm diag}(U\sigma U^{\dagger})
\ \prec \ {\rm eig}(\sigma) .
\label{prec2}$$
The above result provides an additional argument in favor of usage of the notion of a contra–diagonal form of a matrix, as $\sigma$ is distinguished by the majorization order (\[prec2\]) and is opposite to the diagonal form of a density matrix.
Let $\rho$ denote an arbitrary density matrix, $U_{\rm min}$ the matrix of its eigenvectors and $U_{\rm max}=F U_{\rm min}$ the matrix defining the bases in which the state is contradiagonal. Then the entropy of the projective measurement of $\rho$ with respect to the basis $U_{\rm max}$ is maximal and equal to $\ln N$. Note that this basis is hence dual to the eigenbasis of $\rho$ for which the entropy of the projective measurement is minimal and equals to the von Neumann entropy $S(\rho)$. As each projective measurement induces the decoherence to the system and copies the information on the eigenstates of the density matrix to the environment, the information copied in the case of the measurement in the contra-diagonal basis is the largest and reads $\ln N- S(\rho)$. In other words, performing a coarse–graining map, $\rho \to \rho'={\rm diag}(\rho)$ on any pure state $\rho=|\psi\rangle\langle \psi|$, the exchange entropy [@Schum] is the largest if the state is represented in the contra-diagonal basis.
Consider, for instance, a single–qubit pure state written its eigenbasis as $H={\rm diag}(1,0)$. Making use of the real Hadamard matrix $F_2$ and putting it into Eq. (\[contr1\]) one gets the contradiagonal state $\sigma$ where $$F_2 = \frac{1}{\sqrt{2}}
\left[ \begin{array}{c c}
1 & 1 \\
1 & -1
\end{array} \right]
{\ \ \rm and \ \ }
\sigma= \frac{1}{2}
\left[ \begin{array}{c c}
1 & 1 \\
1 & 1
\end{array} \right] .$$ Choosing a complex Hadamard matrix $F_2'$ enphased with an arbitrary complex phase $e^{\i \phi}$ we obtain a more general contradiagonal state ${\sigma}'$ with $$F_2' = \frac{1}{\sqrt{2}}
\left[ \begin{array}{c c}
e^{i \phi} & e^{i \phi} \\
1 & -1
\end{array} \right],
\ \ \
\sigma' = \frac{1}{2}
\left[ \begin{array}{c c}
1 & e^{i \phi} \\
e^{-i \phi} & 1
\end{array} \right] .$$ In higher dimensions, a density matrix $\vert \psi \rangle \langle \psi \vert$ corresponding to a pure state of size $N$ with spectrum ${\rm diag}(1,0, \dots, 0)$ transformed as in (\[contr1\]) by the Fourier matrix $F_N$ leads to a flat contradiagonal state $\sigma$ with all elements equal, $\sigma_{ij}=1/N$. Multiplying $F_N$ from left and right by two arbitrary diagonal unitary matrices one obtains an enphased, complex Hadamard $F_N'$ [@TZ06], which leads to a more general form of a complex Hermitian contradiagonal state ${\sigma}'$ with all elements of the same modulus, $|{\sigma}'_{ij}|=1/N$, and all diagonal elements equal, ${\sigma}'_{jj}=1/N$.
As stated in Proposition \[lemma:anti\] the sum of squared moduli of the off diagonal elements is maximal if and only if the matrix is in its contra–diagonal form. The above is equivalent to the fact, that the sum of squared diagonal elements is minimal. On the other hand, since the geometric mean is a Schur concave function, the product of diagonal elements of a semi positive matrix $H$ is maximal if $H$ is contradiagonal. This fact was used in the analysis of an entanglement measure called *collectibility* [@rudnicki2012collectibility].
Concluding remarks
==================
In this work we showed that a generic random diagonal gate acting on a symmetric bipartite system is strongly non-local and is amazingly efficient in generating quantum entanglement. Its average Schmidt strength [@NDDGMOBHH02] is on average smaller than this characteristic of a Haar random unitary gate by the factor of two, while the mean entangling powers [@ZZF00] for both ensembles are only marginally different.
Investigation of the ensemble of diagonal unitary gates of size $N^2$ leads to the unimodular ensemble of matrices of order $N$ with all entries of the same modulus and independent random phases. We computed first moments of the distribution of the squared singular values of these matrices and showed that it asymptotically converges to the universal Marchenko-Pastur form. The moments have remarkable connections to combinatorial structures that have been recently studied. This allowed us to find the mean entanglement, or von Neumann entropy exactly for the ensemble of unimodular matrices. However, for a finite $N$ we reported the differences with respect to the Hilbert-Schmidt ensemble, which produces, on average, less mixed states.
Squared singular values of a matrix from the unimodular ensemble after a suitable normalization coincide with the spectrum of a density matrix $\sigma$ of order $N$, such that all their diagonal elements are equal. This form of a matrix is called contradiagonal, as it is shown to be opposite to the diagonal form of a matrix concerning the majorization order, the norm of the off-diagonal elements and the entropy of the projective measurement performed on a mixed state in such a basis.
In general, for any Hermitian matrix $H$ of order $N$ we have shown how to find a unitary matrix $U_{\rm max}$ which brings it to the contra–diagonal form, $H'=U_{\rm max}H U_{\rm max}^{\dagger}$ such that $H'_{jj}={{\mathrm{Tr}}}H/N$. This method based on diagonalization and complex Hadamard matrices [@TZ06] can be easily applied for any mixed state $\rho$ to transform it to its contradiagonal form, distinguished form the perspective of quantum information processing.
From a mathematical perspective the problem of constructing a Hermitian matrix with a given spectrum and prescribed diagonal entries was studied in [@Chu95; @DHST05] and more recently in [@FMPS13]. Although several general algorithms for this task were analyzed in these papers, the limiting problem of a constant diagonal was not shown to be reducible to the standard diagonalization procedure followed by a unitary transformation with an arbitrary complex Hadamard matrix.
It is a pleasure to thank Sarika Jalan for the invitation to the CNSD, at IIT Indore, where this work has been initiated. We are thankful to Pawe[ł]{} Horodecki for stimulating discussions and fruitful remarks. We acknowledge support by the grants number DEC-2011/02/A/ST1/00119 and DEC-2012/04/S/ST6/00400 (ZP) financed by Polish National Science Center.
Reshuffling and operator entanglement entropy
==============================================
In this appendix a link between the operator Schmidt decomposition and an the reshuffling of a matrix [@ZB04] is reviewed.
Consider a given unitary matrix $U$ of size $N^2\times N^2$. It belongs to the composite Hilbert–Schmidt space ${\cal H}_{HS} \otimes {\cal
H}_{HS}$. Let us write down its representation in a product basis in the space of matrices, $$U = \sum_{m=1}^{N^2} \sum_{n=1}^{N^2} C_{mn}
B_m \otimes B_n ,
\label{Vmatrix}$$ where $C_{mn}={{\mathrm{Tr}}}( (B_m \otimes B_n)^{\dagger} U)$.
The complex matrix $C$ of order $ N^2\times N^2$ need not be Hermitian nor normal. The usual Schmidt decomposition hold for this vector space and therefore the Schmidt decomposition of $U$ given in Eq.(\[VSchmidt\]) consists of $K$ terms entering with the weights $\sqrt{\Lambda_k}$ equal to the singular values of $C$. It will be convenient to work with the product bases in the HS space of matrices, generated by the identity matrix, of size $N^2\times N^2$. Each of the $N^2$ basis matrices $B_n$ of size $N\times N$ has only a single non vanishing element equal to unity. Let’s denote $B_{k}=B^{m\mu}=|m\rangle\langle \mu |$, where .
For this choice of the basis the matrix of the coefficients $C$ in Eq. (\[Vmatrix\]) takes a particularly simple form, $$C_{\stackrel{\scriptstyle m \mu }{n
\nu}}= {{\mathrm{Tr}}}(B^{m\mu}\otimes B^{n\nu})U= U_{\stackrel{ \scriptstyle mn}{\mu\nu}}.
\label{reschuff}$$ Note that both matrices $U$ and $C$ consist of the same entries, ordered in a different way. This particular reordering of a matrix, called [*reshuffling*]{} [@ZB04], will be denoted as $U^R:=C$. In general the notion of reshuffling is well defined if a matrix $X$ acts on a composite Hilbert space, ${\cal H}_M \otimes {\cal H}_N$. The symbol $U^R$ has a unique meaning if a concrete decomposition of the total dimension, $L=MN$, is specified. Similar reorderings of matrices were considered by Hill et al. [@OH85; @YH00] while investigating CP maps and also in [@Rud02; @CW03; @HHH02] to analyze separability of mixed quantum states and in [@Shudd13] to generate local unitary invariants. This operation in these latter contexts is also referred to as [*realignment*]{}.
To get a better feeling of the transformation of reshuffling observe that reshaping each row of an arbitrary matrix $X$ of length $N^2$ into a submatrix of size $N$ and placing it according to the lexicographical order block after block produces the reshuffled matrix $X^R$ as defined in (\[reschuff\]). Let us illustrate this procedure for the simplest case $N=2$, in which any row of the matrix $X$ is reshaped into a $2 \times 2$ matrix $$C_{kj}=X_{kj}^R :=\left[
\begin{array}{c|c}
{\bf {X_{11}\ \ X_{12}}} & X_{21}{\rm ~ ~ ~ }X_{22} \\
X_{13} {\rm ~ ~ ~ }X_{14} & {\bf X_{23} \ \ X_{24}} \\
\hline
{\bf X_{31}\ \ X_{32}} & X_{41} {\rm ~ ~ ~ }X_{42} \\
X_{33}{\rm ~ ~ ~ }X_{34} & {\bf X_{43}\ \ X_{44} }
\end{array}
\right] .
\label{reshuf1}$$ It is easy to see that $(X^{R})^{R}=X$. In general, $N^{3}$ elements of $X$ do not change their position during the operation of reshuffling, these are typeset in bold in Eq. (\[reshuf1\]). the other $N^{4}-N^{3}$ elements do. It is worth to emphasize that if a matrix $X$ is Hermitian the reshuffled matrix $X^R$ needs not to be Hermitian.
The Schmidt coefficients of $U$ are thus equal to squared singular values $\Lambda_i$ of the reshuffled matrix, $U^R$, equal to the eigenvalues of a positive matrix $H=(U^R)^{\dagger}U^R$. The gate $U$ is local if and only if the rank $K$ of $H$ is equal to one, so that the matrix can be factorized into a product form, $U=U_A \otimes U_B$.
The squared Hilbert-Schmidt norm of any unitary matrix of order $N^2$ is $||U||^2=N^2$, which implies that $ \sum_{k=1}^{N^2} \Lambda_k=N^2$. To characterize nonlocality of a gate $U$ one can then use the normalized vector $\vec \lambda$ of the squared singular values, $\lambda_k:=\Lambda_k/N^2$, which may be interpreted as a probability vector of length $N^2$.
In general, the vector of the Schmidt coefficients of an unitary matrix $U$ acting on a composite $N \times N$ system conveys information concerning the non-local properties of $U$. To characterize quantitatively the distribution of $\vec \lambda$ one uses the Shannon entropy, $$S(U):= S({\vec \lambda}) = - \sum_{k=1}^{N^2} \lambda_k \ln(\lambda_k)
\label{shannon}$$ called in this context [*entropy of entanglement*]{} of $U$ [@Za01], (or [*Schmidt strength*]{} [@NDDGMOBHH02]), and the generalized, R[é]{}nyi entropies $$S_q(U):= S_q({\vec \lambda}) =
-\frac{1}{1-q} \ln \Bigl[ \sum_{k=1}^{N^2} (\lambda_k)^q \Bigr],
\label{Renyi}$$ which tend to $S$ in the limit $q\to 1$. The entropy $S_0$, sometimes called [*Hartley entropy*]{}, is equal to $\ln K$, where $K$ denotes the number of positive coefficients $\lambda_i$, and is called [*Schmidt rank*]{} (or Schmidt number). The second order R[é]{}nyi entropy $S_2$ is closely related to the linear entropy $E(U)=1-\exp(-S_2)$ used by Zanardi in [@Za01].
The generalized entropies $S_q$ are equal to zero if and only if the gate $U$ has a product structure, so it can be obtained by performing local gates. The upper bound, $S_q^{\rm max}=2\log N$ is achieved e.g. for the Fourier unitary matrix of size $N^2$ defined by $$F_{kl}^{(N^2)}\ := \ \frac{1}{N}\exp\bigl(i2\pi kl/N^2\bigr)\ .
\label{Four}$$ To show this fact it is sufficient to notice that the reshuffled matrix $F^R$ remains unitary, so all its singular values are equal to unity, hence the Schmidt vector contains $N^2$ equal components and is maximally mixed.
Hermitian matrices with prescribed spectrum {#sec:robin-hood}
===========================================
We begin with a fact, known as a Horn lemma [@BZ06]
Assume that $x \prec y$ then there exist an orthostochastic matrix $\mathcal{O}$ such, that $x = \mathcal{O} y$.
The matrix $\mathcal{O}$ is said to be orthostochastic, if there exist an orthogonal matrix $W$, such that $\mathcal{O}_{ij} = W_{ij}^2$.
The above lemma in the case of bistochastic matrix instead of orthostochastic is well known [@hj2], and sometimes used in a definition of majorization.
The Horn lemma allows us to formulate a simple lemma, which is a generalization of Proposition \[lemma:anti\](b).
Let $H$ be a Hermitian matrix with spectrum $y$ and let $x \prec y$, then there exist a unitary matrix $V$, such that the diagonal of $V H V^{\dagger}$ is given by $x$.
To prove it we assume, without loss of generality, that $H$ is in its diagonal form with vector $y$ on diagonal, and let $\mathcal{O}$ be an orthostochastic matrix such that $x = \mathcal{O} y$. By $W$ we denote an orthogonal matrix such that $\mathcal{O}_{ij} = W_{ij}^2$. Now we write $$(W H W^{T})_{ii} = \sum_{k} W_{ik} H_{kk} W_{ik} =\sum_{k} \mathcal{O}_{ik} y_{k} = x_i.$$ In other words the unitary matrix $V$ which describes the unitary transformation can be represented as a product of a unitary matrix $W$ appearing in the Horn lemma and the matrix $U^{\dagger}$ containing eigenvectors of $H$. In particular if $x$ is flat, i.e. all $x_i$ are equal, the matrix $W$ present in the Horn lemma can be taken as a complex Hadamard matrix and item (b) in Proposition \[lemma:anti\] is recovered.
This allows us to obtain an alternative solution to the problem of constructing Hermitian matrices with prescribed spectrum, studied in [@Chu95; @DHST05; @FMPS13].
[99]{} R. Horodecki, P. Horodecki, M. Horodecki, and K. Horodecki, Rev. Mod. Phys. [**81**]{}, 865 (2009).
E. Lubkin, J. Math. Phys. [**19**]{}, 1028 (1978).
D. Page, Phys. Rev. Lett. [**71**]{}, 1291 (1993).
K. [Ż]{}yczkowski and H.-J. Sommers, Induced measures in the space of mixed quantum states, J. Phys. A: Math. Theory [**34**]{}, 7111 (2001).
P. Hayden, D. W. Leung, and A. Winter, Commun. Math. Phys. [**265**]{}, 95 (2006).
S. Sen, Phys. Rev. Lett. [**77**]{}, 1 (1996).
C. Nadal, S.N. Majumdar, and M. Vergassola, J. Stat. Phys. [**142**]{}, 403 (2011).
U. T. Bhosale, S. Tomsovic and A. Lakshminarayan, Phys. Rev. A [**85**]{}, 062331 (2012).
P. Zanardi, C. Zalka and L. Faoro, On the entangling power of quantum evolutions, Phys. Rev. A [**62**]{}, 030301(R) (2000).
P. Zanardi, Entanglement of quantum evolution, Phys. Rev. A[**63**]{}, 040304(R), (2001).
K. Hammerer, G. Vidal and J.I. Cirac, Characterization of non-local gates, Phys. Rev. A [**66**]{}, 062321 (2002).
J. Zhang, J. Vala, K.B. Whaley and S. Sastry, A geometric theory of non-local two-qubit operations, Phys. Rev. A [**67**]{}, 042313 (2003).
X. Wang, B. C. Sanders, and D. W. Berry, Entangling power and operator entanglement in qudit systems, Phys. Rev. A [**67**]{}, 042323 (2003).
M. A. Nielsen, C. M. Dawson, J. L. Dodd, A. Gilchrist, D. Mortimer, T. J. Osborne, M. J. Bremner, A. W. Harrow and A. Hines, Quantum dynamics as physical resource, Phys. Rev. A [**67**]{}, [052301]{} (2003).
A. J. Scott, Multipartite entanglement, quantum-error-correcting codes, and entangling power of quantum evolutions, Phys. Rev. A [**69**]{}, 052330 (2004).
S. Balakrishnan, R. Sankaranarayanan, Characterizing the geometrical edges of nonlocal two-qubit gates, Phys. Rev. A [**79**]{}, 052339 (2009).
M. Musz, M. Ku[ś]{}, and K. [Ż]{}yczkowski, Unitary quantum gates, perfect entanglers and unistochastic maps, Phys. Rev. A [**87**]{}, 022111 (2013).
Y. Nakata, P. S. Turner, and M. Murao, Phase-random states: Ensembles of states with fixed amplitudes and uniformly distributed phases in a fixed basis, Phys. Rev. A [**86**]{}, 012301 (2012).
Y. Nakata, M. Koashi, and M. Murao, Generating a state t-design by diagonal quantum circuits, New Journal of Physics [**16**]{}, 053043 (2014).
M. J. Bremner, R. Jozsa, and D. J. Shepherd, Proc. R. Soc. A [**465**]{} 459 (2011).
M. J. Hoban, J. J. Wallman, H. Anwar, N. Usher, R. Raussendorf, and D. E. Browne, Phys. Rev. Lett. [**112**]{}, 140505 (2014).
F. Haake, [*Quantum Signatures of Chaos*]{}, 3rd. ed., Springer, Berlin, 2010.
J. N. Bandyopadhyay and A. Lakshminarayan, Phys. Rev. Lett. [**89**]{}, 060402 (2002).
K. [Ż]{}yczkowski and I. Bengtsson, On duality between quantum states and quantum maps, Open Syst. Inf. Dyn. [**11**]{}, 3-42 (2004).
W. Tadej and K. [Ż]{}yczkowski, A concise guide to complex Hadamard matrices, Open Syst. Inf. Dyn. 13, 133-177 (2006).
T. Tao, V. Vu, Random matrices: Universality of local eigenvalue statistics, Acta Mathematica [**206**]{}, 127, (2011).
H.–J. Sommers and K. [Ż]{}yczkowski, Statistical properties of random density matrices, J. Phys. A: Math. Theory. [**37**]{} 8457-8466 (2004).
G. Akemann, M. Kieburg and L. Wei, Singular value correlation functions for products of Wishart random matrices, J. Phys. A: Math. Theory. [**46**]{}, 275205 (2013).
T. Koshy, [*Catalan Numbers with Applications*]{}, Oxford University Press, New York, 2008.
N. J. A. Sloane, On-line Encyclopedia of Integer Sequences, http://www.research.att.com/$\sim$njas/sequences (2013).
P. Barry, A note on a family of generalized Pascal matrices defined by Riordan arrays, J. Integer Seq. [**16**]{}, Article 13.5.4, (2013).
C. Francisco, J. Mermin, and Jay Schweig, https://www.math.okstate.edu/$\sim$jayjs/ppt.pdf, (2013).
Christian Kassel, and Christophe Reutenauer, Algebraicity of the zeta function associated to a matrix over a free group algebra, arXiv:1303.3481v5 \[math.CO\].
T. A. Brody, J. Flores, J. B. French, P. A. Mello, A. Pandey, and S. S. M. Wong, Rev. Mod. Phys. [**53**]{}, 385 (1981).
K. [Ż]{}yczkowski, K. A. Penson, I. Nechita, B. Collins, Generating random density matrices, J. Math. Phys. [**52**]{}, 062201(20) (2011).
A. W. Marshall and O. Olkin, [*Inequalities: Theory of Majorization and Its Applications*]{} New York: Academic, 1979.
I. Bengtsson and K. [Ż]{}yczkowski, [*Geometry of Quantum States. An introduction to quantum entanglement*]{} (Cambridge University Press, Cambridge, 2006).
T. W. Korner, [*Fourier Analysis*]{} (Cambridge University Press, Cambridge, 1988).
B. Schumacher, Sending entanglement through noisy quantum channels, Phys. Rev. A [**54**]{}, 2614–2628 (1996).
. Rudnicki, Z. Pucha[ł]{}a, P. Horodecki, K. [Ż]{}yczkowski, Collectibility for mixed quantum states, [*Phy. Rev. A*]{}, [86]{}, ([6]{}), [062329]{}, ([2012]{}).
M. T. Chu, Constructing a Hermitian matrix from its diagonal entries and eigenvalues, SIAM J. Matrix Anal. Appl. [**16**]{}, 207217 (1995).
I.S. Dhillon, R. W. Heath Jr. M. A. Sustik and J.A. Tropp, Generalized finite algorithms for constructing hermitian matrices with prescribed spectrum, SIAM J. Matrix Anal. 27, 6171 (2005).
M. Fickus, D.G. Mixon, M.J. Poteet, N. Strawn, Constructing all self-adjoint matrices with prescribed spectrum and diagonal, Adv. Comput. Math. [**39**]{}, 585-609 (2013).
C. J. Oxenrider and R. D. Hill, On the matrix reordering ${\Gamma}$ and ${\Psi}$, Linear Alg. Appl. [**69**]{}, 205 (1985).
D. A. Yopp and R. D. Hill, On completely copositive and decomposable linear transformations, Linear Alg. Appl. [**312**]{}, 1 (2000).
O. Rudolph, On the cross norm criterion for separability, J. Phys. A: Math. Theory. [**36**]{}, [5825]{} (2003).
K. Chen and L.-A. Wu, A matrix realignment method for recognizing entanglement, Quant. Inf. Comp. [**3**]{}, 193 (2003).
M. Horodecki, P. Horodecki, and R. Horodecki, Characterization of separable states: Linear contractions, and permutattion criteria, Open Systems Inform. Dynamics [**13**]{}, 103 (2004).
U. T. Bhosale, K. V. Shuddhodan, and A. Lakshminarayan, Phys. Rev. A [**87**]{}, 052311 (2013).
R. A. Horn, C. R. Johnson, *Topics in Matrix Analysis* Cambridge University Press, New York (1991).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We propose that the Hawking radiation energy and entropy flow rates from a black hole can be viewed as a one-dimensional (1D), *non-equilibrium* Landauer transport process. Support for this viewpoint comes from previous calculations invoking conformal symmetry in the near-horizon region, which give radiation rates that are identical to those of a single 1D quantum channel connected to a thermal reservoir at the Hawking temperature. The Landauer approach shows in a direct way the particle statistics independence of the energy and entropy fluxes of a black hole radiating into vacuum, as well as one near thermal equilibrium with its environment. As an application of the Landauer approach, we show that Hawking radiation gives a net entropy production that is 50% larger than that obtained assuming standard three-dimensional emission into vacuum.'
address:
- 'Advanced Science Institute, RIKEN, Wako-shi, Saitama 351-0198, Japan'
- 'Department of Physics, University of Michigan, Ann Arbor, Michigan 48109-1040, USA'
- 'Department of Physics and Astronomy, Dartmouth College, Hanover, New Hampshire 03755-3528, USA'
- 'Advanced Science Institute, RIKEN, Wako-shi, Saitama 351-0198, Japan'
- 'Department of Physics, University of Michigan, Ann Arbor, Michigan 48109-1040, USA'
author:
- 'P. D. Nation'
- 'M. P. Blencowe'
- Franco Nori
bibliography:
- 'text.bib'
title: 'Non-equilibrium Landauer transport model for Hawking radiation from a black hole'
---
Introduction
============
One of the main achievements of quantum field theory in curved spacetime is the verification of the equilibrium thermodynamic description of black hole mechanics [@wald:1994]. In using this formalism [@hawking:1974; @hawking:1975], Hawking was able to provide a physical interpretation of the black hole temperature through the discovery of particle pair production at the horizon, while also solidifying the connection between black hole entropy and horizon surface area predicted a few years earlier [@bekenstein:1972]. Subsequently, there has been a large body of work devoted to understanding this thermodynamic description of black holes and its deeper implications [@jacobson:1995; @bousso:2002; @susskind:2006]. Yet at the same time, the *non-equilibrium* thermodynamic properties of black holes, namely the steady-state flow, or transport, of energy and entropy via Hawking radiation, has received markedly less attention [@saida:2006; @saida:2007].
When viewed by an observer at spatial infinity, the metric of a non-rotating, uncharged black hole is given by the (1+3)-dimensional Schwarzschild metric. Therefore, in the thermodynamic description of black holes, it is natural to assume that the emission of Hawking radiation corresponds to that of a three-dimensional (3D) thermal body obeying the Stefan-Boltzmann law. However, recently there has been an increasing body of evidence suggesting that black hole emission is instead a 1D radiative process. One indicator is the well-known near-horizon approximation under which the Schwarzschild metric of a black hole can be reduced to a (1+1)-dimensional Rindler space possessing infinite-dimensional conformal symmetry [@fabbri:2005]. The ability to calculate the stress-energy tensor using conformal symmetry is the basis for standard derivations of the Hawking flux [@birrell:1982; @brout:1995]. More recently, it has been suggested that this conformal symmetry is responsible for the Hawking effect [@agullo:2010], as it has been shown that this symmetry alone is sufficient to determine both the Hawking thermal spectrum [@jacobson:1993; @iso:2007] and radiation flux [@robinson:2005]; the Hawking radiation is an inherently (1+1)-dimensional process. This near-horizon conformal symmetry also reproduces the Bekenstein-Hawking form of the black hole entropy [@carlip:2002], thus connecting to the other familiar dimensional reduction in black hole physics, namely the holographic principle [@bousso:2002].
The first to focus on the entropic and information implications of a 1D evaporation process was Bekenstein [@bekenstein:2001], who proposed that the entropy flow rate from a black hole is of the same form as that of a 1D quantum channel [@pendry:1983], thus constraining the information flow from a black hole. This same 1D channel description applies in the context of laboratory analogues of Hawking radiation [@schutzhold:2005; @philbin:2008; @nation:2009], and it was noted that the power output from the analogue Hawking process coincides with the optimal energy current through a single quantum channel [@nation:2009]. The concept of a 1D quantum channel was first considered by Landauer and others [@imry:1999; @imry:2008] in the modeling of electrical transport in mesoscopic circuits. The Landauer approach expresses the conductance of a 1D system $G_{c}$ in terms of its scattering properties [@imry:2008] via the relation $$\label{eq:landauer}
G_{c}=\frac{I}{\mu_{1}-\mu_{2}}=\frac{e^{2}}{\pi\hbar}T,$$ where $I$ is the current through the 1D channel, $\mu_{1}$ and $\mu_{2}$ are the chemical potentials of the channel reservoirs, and $T$ is the transmission coefficient. For perfect transmission, $T=1$, the channel conductance is given by $e^2/(\pi\hbar)$, a value that is independent of the microscopic, material nature of the channel, due to the mutual cancellation of the group velocity and density of states factors entering the current formula in 1D. This Landauer formalism was subsequently extended to describe multiple channels [@buttiker:1985; @sivan:1986], as well as thermal transport [@sivan:1986; @rego:1998; @blencowe:1999; @schwab:2000; @meshke:2006], where the currents are generated by temperature differences rather than by chemical potential differences. Quantum mechanics places upper limits on the energy and entropy currents in 1D channels. These upper limits are attained in the absence of backscattering for bosonic channels [@pendry:1983; @blencowe:2000], and are again independent of the material properties of the channel. Furthermore, for thermal transport, these upper limits can be independent of whether the particles are bosons or fermions, and thus are termed “universal" [@blencowe:2000; @blencowe:2004].
Motivated by these connections, in this paper we argue that a non-equilibrium Landauer-transport model can be applied to black hole entropy flow and energy production rates, describing the Hawking effect in terms of currents flowing in 1D quantum channels connecting thermal reservoirs at each end. We thus relate the emission of Hawking radiation of astrophysical black holes to 1D thermal transport in mesoscopic devices; systems that differ by orders of magnitude in energy. In particular, we emphasize the conditions under which the 1D currents are independent of particle statistics. In contrast to the emitted power, the black hole entropy current cannot be obtained directly from the stress-energy tensor, and is rarely touched on in the literature without a priori assuming the validity of the 3D Stefan-Boltzmann law [@saida:2006; @saida:2007; @zurek:1982]. Therefore, a theory that is capable of providing *both* the black hole energy and entropy currents is required for the correct description of black hole evaporation [@thooft:1993].
Assuming the validity of 1D Landauer transport theory enables the description of certain non-equilibrium, steady state emission processes for black holes, without necessarily requiring knowledge of their microscopic physics. In essence, the Landauer approach allows us to extend the methodology of applying thermodynamic principles to black holes [@bekenstein:1974]. Moreover, the Landauer model gives a physical insight into the transport of energy and entropy from a black hole that is lacking in existing field-theoretic derivations.
This paper is organized as follows. In Sec. \[sec:near\] we review the well-known near-horizon approximation and the resulting conformal symmetry that leads to the standard derivation of the stress-energy tensor and the energy flow rate for Hawking radiation. Next, in Sec. \[sec:1d\] we introduce the Landauer transport description for 1D quantum channels, and highlight the statistics-independent properties of the energy and entropy transport in these channels. Section \[sec:app\] establishes the Landauer transport model to the emission of Hawking radiation, for both bosonic and fermionic particles, to a black hole in vacuum. Charged and rotating black holes are also addressed. As an application of the 1D Landauer approach, In Sec. \[sec:entropy\] we obtain the net entropy production of a black hole and compare with the standard 3D calculation given in Ref. [@zurek:1982]. The special case of a black hole near thermal equilibrium with its environment is also highlighted. Finally, Sec. \[sec:conclusion\] ends with a brief discussion of the results.
Near-horizon conformal symmetry and the Hawking flux {#sec:near}
====================================================
For an observer near the horizon of a spherically symmetric Schwarzschild black hole of mass $M$, the original 4D metric $(G=c=1)$, $$\label{eq:schwar}
ds^{2}=-\left(1-\frac{2M}{r}\right)dt^{2}+\frac{dr^{2}}{\left(1-\frac{2M}{r}\right)}+r^{2}d\Omega^{2},$$ can be reduced to that of a (1+1)-dimensional spacetime through the coordinate transformation $r=2M+x^{2}/8M$, where near $x=0$, $1-2M/r\approx x^{2}/16M^{2}$. Near the horizon, excitations and dimensional quantities transverse to the $t$-$x$ plane are redshifted and can be ignored [@carlip:2007] (i.e. effective potentials for partial wave modes vanish exponentially fast at the horizon [@robinson:2005]). Thus, the near-horizon form of the metric is given by [@fabbri:2005] $$\label{eq:rindler}
ds^{2}=-\left(\kappa x\right)^{2}dt^{2}+dx^{2},$$ where $\kappa=1/4M$ is the surface gravity and the $t$-$x$ portion of the metric defines the flat (1+1)-dimensional Rindler spacetime. Equation (\[eq:rindler\]) can be brought into conformal form by defining the coordinate $x=\kappa^{-1}\exp\left(\kappa \xi\right)$ and forming null coordinates, $u=t-\xi$ and $v=t+\xi$, under which the metric takes the form $$\label{eq:conformal}
ds^{2}=-C(u,v)du\,dv=-e^{\kappa\left(v-u\right)}du\,dv,$$ where $C(u,v)$ is the conformal factor. Here we ignore the effects of the radial potential as it is blue-shifted away by the conformal symmetry [@agullo:2010]. The regularized expectation values for the stress-energy tensor can be immediately evaluated from the conformal structure of Eq. (\[eq:conformal\]): [@padmanabhan:2005] $$\label{eq:conformal-expect}
\left<T^{2\mathrm{D}}_{ii}\right>=-\frac{1}{12\pi}C^{1/2}\partial_{i}^{2}C^{-1/2},$$ for $i=u,v$. For a Schwarzschild black hole, the expectation value with respect to the Unruh vacuum at the horizon, for a single photon polarization, is given as [@brout:1995] $$\label{eq:influx}
\left.\left<T^{2\mathrm{D}}_{vv}\right>_{U}\right|_{r=2M}=-\frac{1}{12\pi}\left(\frac{1}{64M^{2}}\right)=-\frac{\pi}{12\hbar}T_{\mathrm{H}}^{2},$$ where $T_{\mathrm{H}}=\kappa/2\pi$. This represents the influx of negative energy across the horizon, responsible for the evaporation of the black hole, corresponding to the outgoing Hawking flux, as may be checked using the conformal factor for the t-r sector of the Schwarzschild metric, $C(r)=\left(1-M/r\right)$, and Eq. (\[eq:conformal-expect\]) $$\label{eq:outflux}
\left<T^{2\mathrm{D}}_{uu}\right>_{U}=\frac{\pi}{12}T_{\mathrm{H}}^{2}\left[1-\frac{2M}{r}\right]^{2}\left[1+\frac{4M}{r}+\frac{12M^{2}}{r^{2}}\right] .$$ The power emitted through Hawking radiation as seen by an inertial observer at $r=\infty$ is obtained from Eq. (\[eq:outflux\]) as $$\label{eq:hflux}
\left<T^{2\mathrm{D}}_{uu}\right>_{U}=\frac{\pi k_{\mathrm{B}}^{2}}{12\hbar}T_{\mathrm{H}}^{2},$$ where, reintroducing dimensional constants for later convenience, we have $T_{\mathrm{H}}=\hbar c^{3}/8\pi k_{\mathrm{B}}GM$. With $\sim 98\%$ of photons, and likewise $\sim 96\%$ of neutrinos, emitted in the radial direction (s-wave) [@page:1976], Eq. (\[eq:hflux\]) is approximately valid in the full 4D spacetime as well, where the stress-energy tensor in the $r$-$t$ plane is given as [@brout:1995] $$\left<T^{4\mathrm{D}}_{\mu\nu}\right>=\frac{1}{4\pi r^{2}}\left<T^{2\mathrm{D}}_{\mu\nu}\right>.$$ The net flux across a spherical surface of radius $r$ is then given by $4\pi r^{2}\left<T^{4\mathrm{D}}_{\mu\nu}\right>$, which results in a net flux that is again expressed though Eq. (\[eq:hflux\]) [@padmanabhan:2005].
One-dimensional quantum channels {#sec:1d}
================================
As a model for a single 1D quantum channel, we will consider two thermal reservoirs characterized by the temperatures $T_{\mathrm{L}}$ and $T_{\mathrm{R}}$ and with chemical potentials $\mu_{\mathrm{R}}$ and $\mu_{\mathrm{L}}$, respectively. The reservoirs are coupled adiabatically through an effectively 1D connection supporting the bidirectional propagation of particles. The subscripts L and R denote the left and right thermal reservoirs respectively. Here we will assume $T_{\mathrm{L}}>T_{\mathrm{R}}$ and that the transport through the 1D-connection is ballistic.
Although our focus is on fundamental fields/particles, for complete generality we will assume interpolating fractional statistics where the distribution function is [@wu:1994] $$f_{g}\left(E\right)=\left[w\left(\frac{E-\mu}{k_{\mathrm{B}}T}\right)+g\right]^{-1},$$ where $w(x)^{g}\left[1+w(x)\right]^{1-g}=\exp(x)$ with $x\equiv\left(E-\mu\right)/k_{\rm B}T$. Here, $g=0$ and $g=1$ describe bosons and fermions respectively. The individual single-channel energy and entropy currents flowing from the left (L) and right (R) reservoirs may be written as [@blencowe:2000; @blencowe:2004] $$\label{eq:edot}
\dot{E}_{\mathrm{L(R)}}=\frac{\left(k_{\mathrm{B}}T_{\mathrm{L(R)}}\right)^{2}}{2\pi\hbar}\int_{x_{\mathrm{L(R)}}^{0}}^{\infty}\!\!\!\!\!\!\!\!dx\left(x+\frac{\mu_{\mathrm{L(R)}}}{k_{\mathrm{B}}T_{\mathrm{L(R)}}}\right)f_{g}(x)$$ and $$\begin{aligned}
\label{eq:sdot}
\dot{S}_{\mathrm{L(R)}}=&&-\frac{k_{\mathrm{B}}^{2}T_{\mathrm{L(R)}}}{2\pi\hbar}\int_{x_{\mathrm{L(R)}}^{0}}^{\infty}\!\!\!\!\!\!\!\!dx\left\{f_{g}\ln f_{g}+\left(1-gf_{g}\right)\ln(1-gf_{g})\right. \cr
&&\left.-\left[1+(1-g)f_{g}\right]\ln\left[1+(1-g)f_{g}\right]\right\},\end{aligned}$$ where $x_{\mathrm{L(R)}}^{0}=-\mu_{\mathrm{L(R)}}/k_{\mathrm{B}}T_{\mathrm{L(R)}}$. We define the zero of energy with respect to the longitudinal component of the kinetic energy. For the case of bosons with $\mu_{\mathrm{L}}=\mu_{\mathrm{R}}=0$ (e.g. photons), the net power and entropy flow through the quantum channel, $\dot{E}^{\leftrightarrow}_{1D}=\dot{E}_{\mathrm{L}}-\dot{E}_{\mathrm{R}}$ and $\dot{S}^{\leftrightarrow}_{1D}=\dot{S}_{\mathrm{L}}-\dot{S}_{\mathrm{R}}$ respectively, become $$\label{eq:power}
\dot{E}^{\leftrightarrow}_{\mathrm{1D}}=\frac{\pi k_{\mathrm{B}}^{2}}{12\hbar}\left(T_{\mathrm{L}}^{2}-T_{\mathrm{R}}^{2}\right)$$ and $$\label{eq:Sflow}
\dot{S}^{\leftrightarrow}_{\mathrm{1D}}=\frac{\pi k_{\mathrm{B}}^{2}}{6\hbar}\left(T_{\mathrm{L}}-T_{\mathrm{R}}\right).$$ The emitted power Eq. (\[eq:power\]) holds for all bosonic quantum channels since the group velocity and density of states mutually cancel in 1D.
The unidirectional power $$\label{eq:unip}
\dot{E}^{\rightarrow}_{1\mathrm{D}}=\frac{\pi k_{\mathrm{B}}^{2}T^{2}_{\mathrm{L}}}{12\hbar}$$ and the entropy current $$\label{eq:uniS}
\dot{S}^{\rightarrow}_{\mathrm{1D}}=\frac{\pi k_{\mathrm{B}}^{2}}{6\hbar}T_{\mathrm{L}}$$ are the maximum possible rates for single-channel bosonic flow. The unidirectional entropy current (\[eq:uniS\]) is in fact the maximum possible rate for single-channel fermionic flow as well, i.e., it is independent of the particle statistics [@anghel:2002; @blencowe:2004]. To see this, we make a change of integration variables in Eq. (\[eq:sdot\]), $x=\left(E-\mu\right)/k_{\mathrm{B}}T\rightarrow w$, upon which the entropy current can be simplified to [@blencowe:2004] $$\label{eq:sw}
\dot{S}_{\mathrm{L}}=\frac{k_{\mathrm{B}}^{2}T_{\mathrm{L}}}{2\pi\hbar}\int_{w_g\left(\frac{-\mu_{\mathrm{L}}}{k_{\mathrm{B}}T_{\mathrm{L}}}\right)}^{\infty}dw\left[\frac{\ln\!\left(1+w\right)}{w}-\frac{\ln\! w}{1+w}\right].$$ We can see that the statistics of the particles shows up only in the lower integration bound of Eq. (\[eq:sw\]). The maximum current (\[eq:uniS\]) is obtained in the degenerate limit where the statistics-dependence vanishes, since $-\mu_{\mathrm{L}}/k_{\mathrm{B}}T_{\mathrm{L}}\rightarrow 0^{+}$, $w_{g=0}(0)=0$ for bosons, and $-\mu_{\mathrm{L}}/k_{\mathrm{B}}T_{\mathrm{L}}\rightarrow -\infty$, $w_{g=1}(-\infty)=0$ for fermions. However, this same statistics independence in the degenerate limit does not hold for the unidirectional power Eq. (\[eq:edot\]). If one instead considers bidirectional current flow for fermions with $\mu_{\mathrm{R}}=\mu_{\mathrm{L}}$ and $T_{\mathrm{R}}=0$, then in the degenerate limit one recovers the same maximum rate (\[eq:unip\]) as for bosons [@blencowe:2000]. If the maximum energy and entropy current expressions, Eqs. (\[eq:unip\]) and (\[eq:uniS\]) respectively, are combined by eliminating $T_{\mathrm{L}}$, then one obtains equality for the bound $$\label{eq:bound}
\left(\dot{S}^{\rightarrow}_{\mathrm{1D}}\right)^{2}\le\frac{\pi k_{\mathrm{B}}^{2}}{3\hbar}\dot{E}^{\rightarrow}_{\mathrm{1D}},$$ which holds for 1D quantum channels with arbitrary reservoir temperatures, chemical potentials, and particle statistics [@pendry:1983; @blencowe:2000]. We note in passing that this bound is similar in form to the conjectured Bekenstein holographic bound [@bekenstein:1981].
Hawking radiation from a black hole in vacuum {#sec:app}
=============================================
The Landauer description of Hawking radiation is not limited to 1D, but also applies equally well to the 3D black hole spacetime viewed by an observer at infinity. There, the entropy and energy flow rates can be characterized by a large ensemble of quantum channels, each labeled by a transverse spatial (i.e. angular momentum) quantum number, with interactions between channels described via a scattering matrix [@buttiker:1985]. Therefore, scattering due to the potential barrier away from the horizon can be accounted for in the Landauer description through its known multichannel generalization with the inclusion of intra and inter-channel scattering (see, e.g., Ref. [@sivan:1986]). Although this seems to suggest that Hawking radiation flows through a vast number of quantum channels, the near horizon region, where Hawking radiation is emitted and absorbed, is not 3D but rather given by the Rindler metric, Eq. (\[eq:rindler\]). With only a single spatial dimension remaining, the (1+1)-dimensional conformal symmetry of the metric near the horizon allows for a single 1D-quantum channel description of the Hawking process (see Fig. \[fig:fig1\]), where the remaining quantum channel corresponds to the lowest possible angular momentum mode. Comparing Eq. (\[eq:power\]) with Eq. (\[eq:hflux\]), we can see immediately that the Landauer 1D channel formula for the zero chemical potential, bosonic power flow coincides with the Hawking radiation flux where $T_{\mathrm{L}}=T_{\mathrm{H}}$ and $T_{\mathrm{R}}=T_{\mathrm{E}}=0$, with $T_{\mathrm{E}}$ defined to be the temperature of the thermal environment surrounding the black hole. The mutual cancellation of the group velocity and density of states factors in the 1D Landauer formula should make Eq. (\[eq:unip\]) valid not just in flat but in arbitrary curved spacetimes [@bekenstein:2001-2], although the conformal symmetry of the near-horizon region suggests that the production of Hawking radiation is itself essentially a flat-space process.
![(Color online) (a) In the equilibrium thermodynamic description of a Schwarzschild black hole, both the entropy $S$ and temperature $T_{\mathrm{H}}$ of the black hole are given by the properties of the two-dimensional horizon surface, a section of which is highlighted, being proportional to the surface area $A$ and surface gravity $\kappa$ respectively. (b) Near the horizon surface, the conformal symmetry results in an effectively (1+1)-dimensional spacetime, allowing for a 1D Landauer description. Here, the power and entropy flow is through the 1D channel formed by the radial Schwarzschild coordinate $r$. For a black hole in a thermal environment with temperature $T_{\mathrm{E}}>0$, the channel supports the bi-directional propagation of energy and entropy to and from the black hole. The net energy $\dot{E}^{\leftrightarrow}_{1\mathrm{D}}$ and entropy $\dot{S}^{\leftrightarrow}_{1\mathrm{D}}$ flow, Eqs. (\[eq:power\]) and (\[eq:Sflow\]) respectively, is away from the black hole when $T_{\mathrm{H}}>T_{\mathrm{E}}$.[]{data-label="fig:fig1"}](figure1){width="5.5in"}
Although we have appealed to conformal symmetry, these 1D emission properties of Hawking radiation are evident in the full 3D spacetime as well. Following the original argument of Bekenstein [@bekenstein:2001] we note that the flat spacetime entropy emission rate for a blackbody in D-dimensions scales with the output power as $$\label{eq:D}
\dot{S}_{\mathrm{D}}\propto \left(\dot{E}_{\mathrm{D}}\right)^{\mathrm{D}/(\mathrm{D}+1)}.$$ As a result, if a black hole were to radiate as a 3D object, one should expect the emitted entropy to scale as the $3/4$ power of the energy flow rate. However, substitution of the Hawking temperature $T_{\mathrm{H}}$ into the Stefan-Boltzmann law, and making use of the black hole surface area $A=16\pi(GM)^{2}/c^{4}$, one finds that the emitted entropy $$\label{eq:bhineq}
\dot{S}^{2}=\frac{1}{90}\frac{\pi k_{\rm B}^{2}}{3\hbar}\dot{E}$$ goes as the $1/2$ power of the energy rate, just as one would expect for a 1D emitter. In fact, Eq. (\[eq:bhineq\]) is identical to Eq. (\[eq:bound\]) up to a numerical factor arising from the assumption of a 3D, as opposed to 1D, emitter. This result is attributable to the inverse dependence of the Hawking temperature on the black hole mass $M$, a property of black holes not shared by other blackbody emitters. Therefore, the thermodynamic properties of a black hole correspond to that of a 1D blackbody emitter, as one might suspect given the ability to derive both the Hawking temperature and black hole entropy from the 1+1-dimensional conformal symmetry in the near horizon region.
In what follows, we will assume the validity of Eqs. (\[eq:edot\]) and (\[eq:sdot\]) \[equivalently Eq. (\[eq:sw\])\] in describing the net energy and entropy outflow rates, respectively, for particles radiating from a black hole into the vacuum (i.e., $T_{\mathrm{E}}=0)$. With the goal of introducing the Landauer description of Hawking radiation in the near horizon region, we will ignore scattering due to the radial potential barrier. However, the full Landauer approach, relating transport to scattering processes [@imry:1999; @imry:2008], can incorporate inter-channel scattering due to particle interactions and back scatter from the radial potential barrier not considered here.
The electrochemical potential of the black hole reservoir is $\mu_{\mathrm{L}}=\mu_{\mathrm{BH}}=q\Phi$, where $q$ is the electric charge of the field under consideration and $\Phi$ is the electrostatic potential corresponding to the charge of the black hole [@fabbri:2005]. For a Schwarzschild black hole with $\Phi=0$, and hence $\mu_{\mathrm{BH}}=0$, bosons such as photons and gravitons have maximum rates given by Eq. (\[eq:unip\]) and (\[eq:uniS\]) with $T_{\mathrm{L}}=T_{\mathrm{H}}$. For fermions such as neutrinos and electrons (i.e. leptons), setting $\mu_{\mathrm{BH}}=0$ gives a lower integration limit of $w_{g=1}(0)=1$ in Eq. (\[eq:sw\]), resulting in entropy and energy rates that are reduced by a factor of $1/2$ from the maximum values (\[eq:uniS\]) and (\[eq:unip\]). This result for the energy rate was established in earlier calculations for massless fermions [@davies:1977], and shows up in the relative values of the conformal and gravitational anomalies [@robinson:2005]. However, as explicitly pointed out in Ref. [@davies:1977], the physical reason behind this result could not be established. In contrast, the Landauer model presented here shows that these reduced fermionic currents are a direct consequence of the vanishing chemical potential of a Schwarzschild black hole and the 1D nature of the emission process. Subsequently, it was pointed out [@davies:1978] that in a (1+1)-dimensional curved spacetime, the fermionic field describing a massless particle plus its antiparticle is equivalent to a single massless bosonic field. From the Landauer viewpoint, the combined fermionic particle/antiparticle single channel currents can therefore be thought of as a single effective bosonic channel that satisfies the maximum rates, Eqs. (\[eq:unip\]) and (\[eq:uniS\]), when $\mu_{\mathrm{BH}}=0$. Although leptons are massive particles, the conformal symmetry removes the length scale set by the particle mass [@agullo:2010]; the particles are effectively massless. In the case of ballistic transport, multiple channels can be treated independently. Thus, the net Schwarzschild black hole energy and entropy outflow rates are bounded by $N\left(T_{\mathrm{H}}\right)$ times the single channel rates given by Eq. (\[eq:unip\]) and Eq. (\[eq:uniS\]), respectively; a Schwarzschild black hole in vacuum radiates energy and entropy at the maximum rate allowed by quantum mechanics in 1D, saturating the bound in Eq. (\[eq:bound\]). Here, $N\left(T_{\mathrm{H}}\right)$ is the total number of effective bosonic channels spontaneously produced by a black hole at temperature $T_{\mathrm{H}}$; a quantity limited by the number of particle species emitted and their corresponding number of polarizations. The temperature dependence of the effective channel number arises due to the requirement that $k_{\mathrm{B}}T_{\mathrm{H}}\gtrsim 2mc^{2}$ for pair production of particles with mass $m$.
For a black hole with nonzero electrochemical potential, charged particle/antiparticle rates differ so as to cause the black hole net charge to decrease over time. The maximum entropy rate for a single charged fermionic channel coincides with the maximum rate for a single bosonic channel as shown above, giving Eq. (\[eq:uniS\]). The extent to which these maximum rates can be achieved depends on how close to degenerate is the thermal Hawking reservoir of the black hole for charged particles. A special case is provided by extremal charged black holes [@fabbri:2005] satisfying $Q^{2}/M^{2}\approx1$, where $Q$ is the non-dimensional black hole charge. In this limit, $T_{\mathrm{H}}\rightarrow 0$ giving $-\mu_{\mathrm{BH}}/k_{\mathrm{B}}T_{\mathrm{H}}\rightarrow -\infty$, the degenerate limit for fermions. Charged fermions then satisfy Eq. (\[eq:uniS\]). It may be possible to reach the degenerate limit for other choices of black hole parameters. Similar reasoning applies to a black hole with angular momentum where, although spherical symmetry is broken, the emission of Hawking radiation is still governed by (1+1)-dimensional conformal symmetry [@agullo:2010]. Here, the $U(1)$ gauge symmetry corresponding to the angular isometry in the (1+1)-dimensional theory may be written as a chemical potential in the same manner as that of a charged black hole [@iso:2006; @iso:2006b]. Therefore, the Landauer model presented here is quite general, being valid for black holes both with or without charge and angular momentum. Finally, we point out that the cancellation of the density of states and group velocity in 1D quantum channels suggests that Eq. (\[eq:unip\]) should also be valid for black holes in other spacetimes, such as BTZ black holes [@banados:1992] in anti de-Sitter space, where conformal methods may still be applied [@carlip:2005].
Net entropy production in (1+1)-dimensions {#sec:entropy}
==========================================
Originally considered by Zurek [@zurek:1982], the rate of net entropy production by a Schwarzschild black hole due to the emission of Hawking radiation into a thermal environment, neglecting backscattering due to the radial potential barrier [@page:1983], is given by $$\label{eq:R}
R=\frac{dS}{dS_{\mathrm{BH}}}=T_{\mathrm{H}}\frac{dS}{dE}=T_{\mathrm{H}}\frac{\dot{S}}{\dot{E}}$$ where we have used the first law of thermodynamics $dE_{\mathrm{BH}}=T_{\mathrm{H}}dS_{\mathrm{BH}}$ and assumed energy conservation, $dE=dE_{\mathrm{BH}}$. For a 3D black hole radiating into a thermal environment with temperature $T_{\mathrm{E}}$, the power and entropy currents are $$\begin{aligned}
\dot{E}^{\leftrightarrow}_{\mathrm{3D}}&\sim& a\left(T_{\mathrm{H}}^{4}-T_{\mathrm{E}}^{4}\right)\\
\dot{S}^{\leftrightarrow}_{\mathrm{3D}}&\sim& \frac{4a}{3}\left(T_{\mathrm{H}}^{3}-T_{\mathrm{E}}^{3}\right),\end{aligned}$$ where $a$ is a constant. Upon substitution into Eq. (\[eq:R\]), this yields the 3D black hole entropy production ratio $$\label{eq:3d}
R_{\mathrm{3D}}=\frac{4}{3}\frac{1-\left(T_{\mathrm{E}}/T_{\mathrm{H}}\right)^{3}}{1-\left(T_{\mathrm{E}}/T_{\mathrm{H}}\right)^{4}},$$ which gives $R_{\mathrm{3D}}=4/3$ for a black hole in vacuum: $T_{\mathrm{E}}/T_{\mathrm{H}}\rightarrow 0$ [@zurek:1982].
However, as we have shown above, the emission properties of a black hole are better characterized as a 1D Landauer process. Therefore, we compare the entropy produced via our 1D model to the standard calculation presented by Zurek. Our focus in this paper is on the near-horizon region, and hence we do not include scattering. The conformal symmetry in this region removes any inherent length scales, allowing the scattering barrier to be blue shifted away [@agullo:2010]. Moreover, it is important to note that the entropy current Eq. (\[eq:sw\]), like the 1D energy flow Eq. (\[eq:unip\]), should hold for a black hole in any spacetime where conformal symmetry may be invoked, even though the corresponding scattering properties may be markedly different. Likewise, the Landauer approach is valid for analogue black hole models [@schutzhold:2005; @philbin:2008; @nation:2009], as well as a moving mirror in (1+1)-dimensional spacetime [@fulling:1976] that can reproduce the emission properties of Hawking radiation from a Schwarzschild black hole in vacuum, even though no scattering barrier is present. The effects of any scattering potential can be incorporated into a multichannel Landauer model [@buttiker:1985; @sivan:1986], and for the current case of a Schwarzschild black hole, will be presented elsewhere. Since scattering can serve to only increase the net entropy produced from a Schwarzschild black hole [@page:1983], the entropy production rates considered in this section may be viewed as lower bounds. However, we note that for 1D transport, scattering will reduce the individual unidirectional energy $\dot{E}_{\mathrm{L(R)}}$ and entropy currents $\dot{S}_{\mathrm{L(R)}}$, Eq. (\[eq:edot\]) and Eq. (\[eq:sdot\]) respectively, to values below the ballistic bosonic channel limits, Eqs. (\[eq:unip\]) and (\[eq:uniS\]).
For comparison, in our Landauer model we set $\mu_{\mathrm{E}}=\mu_{\mathrm{BH}}=0$, and the net energy and entropy currents are given by Eqs. (\[eq:power\]) and (\[eq:Sflow\]) respectively. The factors of 1/2 in the fermion rates will drop out when evaluating the ratio Eq. (\[eq:R\]). The 1D entropy production ratio is then $$\label{eq:1d}
R_{\mathrm{1D}}=2 \frac{1-\left(T_{\mathrm{E}}/T_{\mathrm{H}}\right)^{\phantom{1}}}{1-\left(T_{\mathrm{E}}/T_{\mathrm{H}}\right)^{2}},$$ which yields a larger value of $R_{1D}=2$ when radiating into vacuum; the net entropy production by Hawking radiation into vacuum is 50% larger than that of a corresponding 3D thermal body at the Hawking temperature. Again, this is due to the 1D properties of the near-horizon region, and the emitted radiation, for which Eq. (\[eq:3d\]) is no longer valid. The difference between the 3D and 1D entropy rates, Eqs. (\[eq:3d\]) and (\[eq:1d\]) respectively, for various ratios of $T_{\mathrm{E}}/T_{\mathrm{H}}$ is presented in Fig. \[fig:fig2\].
![(Color online) Entropy production ratio for a black hole characterized as 1D quantum channel $R_{\mathrm{1D}}$ (dashed-blue) compared to the standard 3D answer $R_{\mathrm{3D}}$ (red). Both results agree near thermal equilibrium $T_{\mathrm{H}}\approx T_{\mathrm{E}}$. []{data-label="fig:fig2"}](figure2){width="3.0in"}
In the case where $T_{\mathrm{H}}\approx T_{\mathrm{E}}$, both Eqs. (\[eq:3d\]) and (\[eq:1d\]) give approximately $R\approx 1+\delta/T_{\mathrm{H}}$ to first order in $\delta=\left(T_{\mathrm{H}}-T_{\mathrm{E}}\right)/2$. As to be expected, in equilibrium ($\delta=0$), there is no net entropy production ($R=1$).
Near thermal equilibrium we can make use of linear response for small temperature differences, $\left(T_{\mathrm{H}}-T_{\mathrm{E}}\right)\ll\bar{T}$ where $\bar{T}=\left(T_{\mathrm{H}}+T_{\mathrm{E}}\right)/2$, to relate the energy and entropy flows by $\dot{S}_{1\mathrm{D}}=\dot{E}_{1\mathrm{D}}/\bar{T}$. In this regime the unidirectional entropy rate Eq. (\[eq:uniS\]) allows us to recover the quantum of thermal conductance for a single effective bosonic channel [@blencowe:2004]: $$\label{eq:conductance}
G_{Q}=\frac{\dot{E}_{1\mathrm{D}}}{T_{\mathrm{H}}-T_{\mathrm{E}}}=\frac{\left(\dot{S}_{\mathrm{H}}-\dot{S}_{\mathrm{E}}\right)\bar{T}}{T_{\mathrm{H}}-T_{\mathrm{E}}}=\frac{\pi k_{\mathrm{B}}^{2}}{6\hbar}\bar{T},$$ that, like Landauer’s original expression Eq. (\[eq:landauer\]), relates conductance to transmission via only fundamental constants. From the statistics independence of Eq. (\[eq:uniS\]), it follows that Eq. (\[eq:conductance\]) provides a general upper bound on the thermal conductance of a black hole that is independent of the particle statistics, as discussed in [@krive:1999; @rego:1999].
Conclusion {#sec:conclusion}
==========
Using the conformal symmetry in the near-horizon region of a black hole, we have presented a 1D Landauer transport model for the non-equilibrium transport of both energy and entropy flow from a black hole, valid for particles with arbitrary statistics, and which clarifies the independence of the underlying microscopic physics. Although our focus is on the near horizon region, the 1D nature of the emission properties are evident in the full (1+3)-dimensional spacetime seen by an observer at infinity, and may be derived from the inverse relationship between black hole mass and Hawking temperature. For a Schwarzschild black hole in vacuum, conformal symmetry results in a Hawking radiation energy flux that is identical to the power flowing in a single 1D quantum channel connected to a thermal bath with the Hawking temperature at one end and zero temperature at the other. Including multiple particle-species and polarizations, a Schwarzschild black hole in vacuum radiates power and entropy at the optimal rate, as a collection of effective bosonic channels. This is a direct result of the statistics independence of unidirectional energy and entropy flow in 1D highlighted by the Landauer formalism, and has not been discussed previously. Furthermore, we have shown that the reduced emission rates for fermions from a Schwarzschild black hole are due to the absence of a black hole chemical potential, giving a physical interpretation that is lacking in previous derivations. Moreover, in contrast to field-theory derivations using the stress-energy tensor, our Landauer model directly yields the entropy current from a black hole without assuming the validity of the 3D Stefan-Boltzmann law. Both the charge and angular momentum of a black hole may be represented as an effective black hole chemical potential, and can be fully incorporated into the Landauer description presented here. The unidirectional entropy current leads to a statistics independent heat flow near thermal equilibrium characterized by the quantum of thermal conductance. Again, this property of black hole transport has not been addressed earlier. In addition, the energy and entropy currents in 1D give a Hawking radiation entropy production ratio that is twice the corresponding value lost by the black hole when radiating into vacuum: a 50% higher value when compared to the currently accepted 3D blackbody rate. These results are a direct consequence of the reduced dimensionality in the near-horizon region and its conformal symmetry. Given the intimate connection between entropy and information, the present findings, in particular Eq. (\[eq:bound\]), place strict limits on the rate of information transfer into and out of a black hole [@lloyd:2004], and therefore will play a role in addressing the information loss problem in black hole evaporation [@hawking:1976; @bekenstein:2004]. However, we note that the reliance on conformal symmetry means that the Landauer model, in its present form, is incapable of describing non-thermal Hawking spectrum transport properties required for unitary black hole evaporation [@page:1993; @nation:2010].
PDN thanks S. Carlip and H. Kang for helpful discussions. PDN was supported by JSPS Postdoctoral Fellowship P11202. MPB was partially supported by the National Science Foundation (NSF) under grant No. DMR-0804477. FN acknowledges partial support from DARPA, AFOSR, Laboratory of Physical Science, National Security Agency, Army Research Office, NSF grant No. 0726909, JSPS-RFBR contract No. 09-02-92114, Grant-in-Aid for Scientific Research (S), MEXT Kakenhi on Quantum Cybernetics, and JSPS through the FIRST program.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We apply stochastic hydrodynamics to the study of charge density fluctuations in QCD matter undergoing Bjorken expansion. We find that the charge density correlations are given by a time integral over the history of the system, with the dominant contribution coming from the QCD crossover region where the change of susceptibility per entropy, $\chi T/s$, is most significant. We study the rapidity and azimuthal angle dependence of the resulting charge balance function using a simple analytic model of heavy-ion collision evolution. Our results are in agreement with experimental measurements, indicating that hydrodynamic fluctuations contribute significantly to the measured charge correlations in high energy heavy-ion collisions. The sensitivity of the balance function to the value of the charge diffusion coefficient $D$ allows us to estimate the typical value of this coefficient in the crossover region to be rather small, of the order of $(2 \pi T)^{-1}$, characteristic of a strongly coupled plasma.'
author:
- Bo Ling
- Todd Springer
- Mikhail Stephanov
bibliography:
- 'Bibliography/balancefunction.bib'
date: 'June 3, 2014'
title: Hydrodynamics of charge fluctuations and balance functions
---
=1
Introduction
============
Event-by-event fluctuations and two particle correlations [@Jeon:2003gk] in high-energy heavy-ion collision experiments provide valuable information about the collective dynamics: thermal and transport properties of the hot and dense QCD matter. Much recent effort has been devoted to the measurement and understanding of the correlations observed at RHIC and LHC. In this paper we focus on charge-dependent correlations. The suppression of the event-by-event fluctuations of the net (electric) charge has been proposed as a signature of the QGP formation [@Jeon:2000wg; @Asakawa:2000wh] and studied experimentally [@Abelev:2008jg; @Abelev:2012pv]. More differential measures of charge fluctuations, such as the azimuthal and rapidity dependence of charge-dependent correlations have also attracted interest. The difference between like-sign and unlike-sign correlations, often expressed as balance functions [@Bass:2000az; @Jeon:2001ue; @Bozek:2012en; @Pratt:2012dz], have also been studied experimentally [@Abelev:2009jv; @Timmins:2011um; @Aggarwal:2010ya; @Abelev:2013csa].
In this paper, we apply relativistic stochastic hydrodynamics [@Kapusta:2011gt] to study the balance functions in heavy-ion collisions. Hydrodynamic equations describe the evolution of conserved quantities such as energy, momentum, and charge averaged over a statistical thermodynamic ensemble. Fluctuations around static equilibrium can be described using the fluctuation-dissipation theorem. In order to describe fluctuations around a [*non-static*]{} solution of the hydrodynamic equations (such as, e.g., Bjorken expansion) one can introduce local noise into the hydrodynamic equations, as has been proposed by Landau and Lifshitz [@LandauStatV9], but has only recently been applied in the context of relativistic heavy-ion collisions [@Kapusta:2011gt]. The hydrodynamic evolution of such local noise naturally leads to observable particle correlations.
As emphasized in Ref. [@Kapusta:2011gt] (see also references therein), the hydrodynamic fluctuations are not the only source of observed correlations. Other sources include initial state fluctuations, fluctuations induced by rare hard processes (jets) and final state (freezeout) fluctuations. These contributions remain important in the case of charge correlations. However, the fact that the inital state in the ultra-relativistic heavy-ion collisions is dominated by saturated glue carrying no electric charge might suppress the contribution of initial state fluctuations to charge correlations we discuss here, compared to entropy fluctuations discussed in Ref. [@Kapusta:2011gt].
Under the conditions of the boost-invariant $1+1$ dimensional Bjorken expansion [@Bjorken:1982qr], the effect of stochastic baryon number current at nonzero mean baryon density was studied in Ref. [@Kapusta:2011gt] without considering diffusion. Here we consider the effect of diffusion and a stochastic charge current at zero mean charge density. The analytical simplicity of the Bjorken solution allows us to understand in detail the mechanisms at work while using a phenomenologically reasonable description of a heavy-ion collision. As a step towards adequately addressing azimuthal dependence of correlations we introduce transverse expansion on top of the Bjorken solution using the standard “blast wave” model.
The paper is organized as follows: In Sec. \[FluctuationSection\], we briefly review hydrodynamics with noise. We linearize the stochastic hydrodynamic equations (around the Bjorken solution at zero charge density) and analytically solve them. We find the charge correlations emerging as a superposition of contributions of past local noise sources which have diffused over the time separating the source and the observation. Successive contributions cancel each other unless the quantity $\chi T/s$ (more precisely $\chi T\tau$) is changing with time. Thus we find the dominant contribution coming from the crossover region of the QCD phase diagram where the effective degrees of freedom change from those of the quark-gluon plasma to those of the hadron gas. In Sec. \[LatticeEOSSec\], we use the lattice QCD data [@Borsanyi:2010cj; @Borsanyi:2011sw] to obtain the dependence of susceptibility per entropy $\chi T/s$ on temperature which determines the magnitude of the charge correlations. We apply these results to a simple semianalytical model of expansion with Cooper-Frye freezeout and make an example comparison with experimental data from RHIC in Section \[FreezeOutSec\]. We conclude with a discussion in Sec. \[ConclusionSec\].
Hydrodynamic Fluctuations {#FluctuationSection}
=========================
Hydrodynamics and noise
-----------------------
Hydrodynamics describes the slow evolution of conserved quantities such as energy, momentum, and conserved charges. In the case of QCD the charge could be the baryon number, electric charge, or strangeness. At top energies at RHIC and at LHC most particles in the final state are pions, which carry only electric charge. Therefore in this work we shall focus on electric charge fluctuations. The extension to other conserved charges such as baryon number or strangeness should be straightforward. The five hydrodynamic equations of motion are the conservation equations for energy-momentum and charge $$\begin{aligned}
&&\nabla_\mu (T^{\mu\nu})=0,
\notag
\\
&&
\partial_\mu (\sqrt{-g}J^\mu)=0.
\label{ConservationEqns}\end{aligned}$$ Here $\nabla_\mu$ denotes the covariant derivative with respect to the background metric $g_{\mu \nu}$ and $g\equiv\det[g_{\mu\nu}]$ – we shall only consider flat space-time, but use curvilinear (Bjorken) coordinates. Fluctuations are described by adding stochastic noise terms $S^{\mu\nu}$ and $I^{\mu}$, as explained in [@LandauStatV5] or, in relativistic context, in [@Kapusta:2011gt]: $$\begin{aligned}
T^{\mu\nu} &=& T^{\mu\nu}_{\rm ideal} + \Delta T^{\mu\nu} +
S^{\mu\nu},
\notag
\\
J^{\mu} &=& n u^{\mu} + \Delta J^{\mu} +
I^{\mu}.
\label{EnergyMomentumTensor}\end{aligned}$$ Here, $n$ and $u^\mu$ are the charge density and fluid velocity, $T^{\mu \nu}_{\rm ideal}$ is the stress-energy tensor for an ideal fluid, and $\Delta T^{\mu\nu}$ and $\Delta J^\mu$ are dissipative (gradient) corrections to stress and current. The dissipative correction to the current to the first order in gradients is given by $$\begin{aligned}
\Delta J^{\mu}=\sigma T\Delta^\mu \left(\frac{\mu}{T}\right)\,,
\label{Jviscous}\end{aligned}$$ where $\sigma$ is the charge conductivity, $\mu$ is the chemical potential, and $\Delta^\mu \equiv \Delta^{\mu\nu}\partial_\nu$ ($\Delta^{\mu\nu} \equiv u^\mu u^\nu-g^{\mu\nu}$) is the spatial derivative in the local rest frame of the fluid (whose 4-velocity is $u^\mu$). The diffusion coefficient $D$ is related to the conductivity by the Einstein relation $$\begin{aligned}
D=\frac{\sigma}{\chi},
\label{EinsteinRelation}\end{aligned}$$ where $\chi$ is the electric charge susceptibility $$\begin{aligned}
\chi \equiv \left(\frac{\partial n}{ \partial \mu} \right)_T.\end{aligned}$$
The hydrodynamic equations (\[ConservationEqns\]) are non-linear. However, in the domain of applicability of hydrodynamics these equations can be linearized [@LandauStatV9] in the perturbations around a given solution of the (non-linear) deterministic equations of motion, i.e., Eqs. (\[ConservationEqns\]) without noise. Such a linearized approach is sufficient to study two-point correlations which are the subject of this paper.
To find two-point correlation functions of hydrodynamic variables we need to know the two-point correlation functions of the noise. One-point functions vanish by definition of the noise. The fluctuation-dissipation theorem determines the magnitude of the two-point correlator: $$\begin{aligned}
\label{eq:II}
\langle I^\mu(x) \rangle&=&0, \\\label{eq:II-2}
\langle I^\mu(x)I^\nu(y)\rangle&=&2\sigma T\Delta^{\mu\nu}\delta(x-y),\end{aligned}$$ where $\sigma$ and $T$ are functions of space and time given by the solution of the deterministic (without noise) hydrodynamic Eqs. (\[ConservationEqns\]). The correlators of $S^{\mu \nu}$ are written down in [@Kapusta:2011gt], but we will not need them in this work.
Generalization to non-linear treatment of fluctuations is an interesting problem, potentially relevant for the study of higher-point correlations or fluctuations near a critical point. Although linearized treatment is sufficient for the purposes of this paper, it is worth keeping in mind the issues involved in the non-linear generalization. The most obvious issue is that the noise would become multiplicative since the magnitude of its correlator in Eq. (\[eq:II-2\]) would be a function of the fluctuating hydrodynamic variables. The formal definition in Eq. (\[eq:II-2\]) would have to be supplemented by a prescription (e.g., Ito or Stratonovich) to resolve the well-known equal-time product ambiguity (see, e.g., Ref. [@GardinerStochastic]). The non-linearities also give rise to short-distance singularities [@Fox:1978] reminiscent of the ultraviolet divergences in quantum field theories. Such issues do not arise in the linearized treatment and we leave them outside of the scope of this paper.
Bjorken expansion and linear perturbations {#BjorkenStochasticSec}
------------------------------------------
We shall use the well-known boost-invariant Bjorken solution [@Bjorken:1982qr] of the deterministic hydrodynamics Eqs. (\[ConservationEqns\]) as the background for the linearized fluctuation analysis. It is most convenient to describe the Bjorken flow in the coordinates $(\tau,
\vec{x}_\perp, \eta)$ defined as $$\begin{aligned}
\tau &\equiv& \sqrt{t^2 - z^2}, \\
\eta &\equiv& \tanh^{-1}\left(\frac{z}{t}\right).\end{aligned}$$ The Bjorken time $\tau$ is invariant under boosts along the $z$ axis while Bjorken rapidity $\eta$ shifts by a constant (the boost rapidity). The liquid undergoing boost-invariant expansion is locally at rest in these coordinates $$\bar u^\mu(x) = \{1,\vec 0,0\}.
\label{u1000}$$ while the energy (or entropy) density is a function of $\tau$, which can be found by solving an ordinary differential equation.
We denote the background quantities with an overbar, and consider small perturbations to entropy density, flow velocity and charge density expressed as $\rho \equiv \delta s / \bar{s}$, $\delta
u^x$, $\delta u^y$, $\delta u^\eta$, and $\delta n$: $$\begin{aligned}
\varepsilon(\tau,\vec{x}_\perp,\eta) &=& \bar{\varepsilon}(\tau) +
\bar T(\tau) \bar{s}(\tau)\, \rho(\tau,\vec{x_\perp},\eta)
\nonumber\\
&&+ \bar\mu(\tau)\, \delta n(\tau,\vec{x_\perp},\eta); \\
u^\mu(\tau,\vec{x}_\perp, \eta) &=& \left\{1,\delta \vec{u}_\perp(\tau,\vec{x}_\perp,\eta), \delta u^\eta(\tau,\vec{x}_\perp,\eta)\right\};\\
n(\tau, \vec{x}_\perp, \eta) &=& \bar{n}(\tau) + \delta n(\tau, \vec{x}_\perp, \eta);\end{aligned}$$ where, as in Eq. (\[u1000\]), we are working in the Bjorken coordinates. The quantity $\delta u^\tau$ vanishes at linear order due to the unit norm constraint $u_\mu u^\mu = 1$.
In general, the fluctuations of the charge and the energy density mix in Eq.(\[ConservationEqns\]). However, in the special case of zero background net charge density ($\bar{n} =0$) or, equivalently, zero chemical potential ($\bar\mu=0$) the fluctuations of the charge density $\delta n$ separate, at linear order considered here, from the fluctuations of entropy density and flow velocity. Since we are going to study only fluctuations of charge density, this simplifies our task considerably. For top-energy RHIC collisions and at LHC the chemical potential is very small compared to relevant microscopic (QCD) scale and the approximation $\bar{\mu}=0$ is useful. Since, as far as charge correlations are concerned, we can ignore entropy and flow velocity fluctuations, we shall no longer distinguish between quantities such as $\bar{s}$ and $s$, or $\bar T$ and $T$.
The stochastic charge diffusion equation in Eq.(\[ConservationEqns\]) becomes $$\begin{aligned}
\partial_\tau J^\tau +\frac{J^\tau}{\tau}+\partial_\eta J^\eta+\vec{\nabla}_\perp \cdot\vec{J}_\perp=0
\label{BjorkenCurrentEqn}\end{aligned}$$ Since for the fluid locally at rest (\[u1000\]) the only derivatives in $\Delta^{\mu\nu}\nabla_\nu (\mu/T)$ are spatial and since $T$ depends on $\tau$ only, we can simplify Eq. (\[Jviscous\]) for $\Delta J^\mu$: $$\begin{aligned}
\Delta J^{\mu}= \sigma \Delta^\mu \mu = D \Delta^\mu n,\end{aligned}$$ which is Fick’s law of diffusion. Substituting this into Eq. (\[BjorkenCurrentEqn\]), we find $$\begin{aligned}
\frac{1}{\tau} \partial_\tau \left(\tau \delta n \right) - D \left[\nabla_\perp^2 + \frac{1}{\tau^2}\partial_\eta^2 \right] \delta n = -\nabla_i I^i - \nabla_\eta I^\eta.
\label{BjorkenDiffusion}\end{aligned}$$ We use Latin indices $i,j,...$ to denote the two transverse directions.
To facilitate the analysis of azimuthal correlations it is useful to decompose the noise current in the transverse plane as [^1] $$\begin{aligned}
I^i = \tau \nabla_j \left[g^{ij}I_S - \epsilon^{ij}I_V \right].\end{aligned}$$ Only $I_S$ will contribute to Eq. (\[BjorkenDiffusion\]). In order to solve Eq. (\[BjorkenDiffusion\]), we express $\vec{x}_\perp$ in polar coordinates $r$ and $\phi$, and use a Fourier-Bessel transformation for $\delta n$, $I^\eta$ and $I_S$, which we define, for any function $f$, as $$\begin{aligned}
f(\tau, r, \phi,\eta)&=& \sum_n \frac{e^{in\phi}}{2\pi}
\int_{-\infty}^\infty \frac{dk_\eta}{2\pi}e^{ik_\eta\cdot \eta}
\int_0^\infty dk_\perp k_\perp
\nonumber\\&\times&
J_n(k_\perp r)\tilde{f}_n(\tau,k_\perp,k_\eta), \label{BesselTransform-f}\\
\tilde{f}_n(\tau,q_\perp,q_\eta) &=& \int_0^{2\pi} e^{-i n \phi} d\phi
\int_{-\infty}^{\infty} e^{-i q_\eta \eta} d \eta \int_0^{\infty} r dr
\nonumber\\&\times&
J_n(q_\perp r) f(\tau,r,\phi,\eta).
\label{BesselTransform}\end{aligned}$$ Inverting the transformation requires the closure relation, $$\begin{aligned}
\int_0^{\infty} r dr J_n(k_\perp r) J_n(q_\perp r) = \frac{\delta(k_\perp - q_\perp)}{k_\perp}.
\label{BesselClosure}\end{aligned}$$
Solution and correlations {#sec:solut-corr}
-------------------------
The charge density fluctuation at a time ${{\tau_{\rm f}}}$ sourced by the hydrodynamic noise $I$ is given by (upon Fourier-Bessel transform) $$\begin{gathered}
\label{eq:delta-n}
\delta \tilde{n}_n ({{\tau_{\rm f}}},k_\perp,k_\eta)
\\ \qquad
= \frac{1}{{{\tau_{\rm f}}}}\int_{\tau_0}^{{{\tau_{\rm f}}}} d\tau \left[(\tau k_\perp)^2\tilde{I}_{S,n} - i \tau k_\eta \tilde{I}_n^\eta \right] e^{-H({{\tau_{\rm f}}},\tau,k_\perp, k_\eta)}\end{gathered}$$ where we defined $$\begin{aligned}
H({{\tau_{\rm f}}},\tau, k_\perp,k_\eta) \equiv \int_{\tau}^{{\tau_{\rm f}}}d\tau^{\prime} D\left(\tau^{\prime}\right) \left(\frac{k_\eta^2}{{\tau^{\prime}}^2 }+ k_\perp^2\right).
\label{Hdef}\end{aligned}$$ The coefficients of $k_\perp^2$ and $k_\eta^2$ in Eq. (\[Hdef\]) are the squared diffusion distances in the $x_\perp$ and $\eta$ directions. It is easy to see that by considering the equation $dl^2=Ddt$ for the diffusion (random walk) distance $dl$ in time $dt$ in a locally comoving frame. The length element in Bjorken coordinates is $dl^2=\tau^2d\eta^2+dx_\perp^2$ and the time element is $dt=d\tau$. Thus the diffusion distance squared in the rapidity direction is given by $(\Delta\eta)^2 = \int d\tau D/\tau^2$ and in the transverse direction by $(\Delta x_\perp)^2= \int d\tau D$.
To determine the charge density correlations, one needs Fourier-Bessel transform of the noise correlators in Eq. (\[eq:II\]): $$\begin{gathered}
\left<\tilde{I}_n^\eta(\tau, k_\perp, k_\eta)
\tilde{I}_m^{*\eta}(\tau',q_\perp, q_\eta) \right> \\= \frac{2
\sigma T}{\tau^3} \delta(\tau-\tau')
\hat{\delta}_{nm}(k,q), \label{eq:IIn}\end{gathered}$$ $$\begin{gathered}
\left<\tilde{I}_{S,n}(\tau, k_\perp, k_\eta)
\tilde{I}_{S,m}^{*}(\tau',q_\perp, q_\eta) \right> \\
= \frac{2 \sigma T}{\tau^3 k_\perp^2} \delta(\tau-\tau' ) \hat{\delta}_{nm}(k,q).\label{eq:IIS}\end{gathered}$$ Here we introduced a shorthand $$\begin{aligned}
\hat{\delta}_{nm}(k,q) \equiv (2\pi)^2 \delta_{n,m}\delta(k_\eta-q_\eta)\frac{\delta(k_\perp -q_\perp)}{k_\perp}.
\label{deltahatdef}\end{aligned}$$ Using Eqs. (\[eq:delta-n\]), (\[eq:IIn\]) and (\[eq:IIS\]) we find for the charge density correlations (at equal time ${{\tau_{\rm f}}}$) $$\begin{gathered}
\left<\delta \tilde{n}_n({{\tau_{\rm f}}},k_\perp,k_\eta) \delta \tilde{n}^*_m({{\tau_{\rm f}}},q_\perp, q_\eta) \right> = \hat{\delta}_{nm}(k,q)\\\times \frac{1}{{{\tau_{\rm f}}}^2} \int_{\tau_0}^{{{\tau_{\rm f}}}} \frac{2 \sigma(\tau)T(\tau)}{\tau}
\left[\tau^2 k_\perp^2 + k_\eta^2 \right] e^{-2 H({{\tau_{\rm f}}},\tau,k_\perp, k_\eta)} d\tau
\label{nnsolution}\end{gathered}$$ With the aid of Eq. (\[EinsteinRelation\]), Eq. (\[nnsolution\]) can be written as $$\begin{gathered}
\left<\delta \tilde{n}_n({{\tau_{\rm f}}},k_\perp,k_\eta) \delta
\tilde{n}^*_m({{\tau_{\rm f}}},q_\perp, q_\eta) \right> =
\hat{\delta}_{nm}(k,q) \\\times \frac{1}{{{\tau_{\rm f}}}^2} \int_{\tau_0}^{{{\tau_{\rm f}}}} \chi(\tau)T(\tau) \tau
\frac{d}{d\tau} e^{-2 H({{\tau_{\rm f}}},\tau,k_\perp, k_\eta)} d\tau.\end{gathered}$$ Finally, performing an integration by parts, we find $$\begin{gathered}
\Big\langle\delta\tilde{ n}_n({{\tau_{\rm f}}},k_\perp,k_\eta)\delta
\tilde{n}^*_m({{\tau_{\rm f}}},q_\perp,q_\eta)\Big\rangle =
{\hat{\delta}_{mn}(k,q)} \\\times \frac1{{{\tau_{\rm f}}}}\left[{\chi_{{\rm f}}T_{{\rm f}}}- {s_{{\rm f}}}\tilde{A}({{\tau_{\rm f}}}, k_\perp,k_\eta)\right]
\label{nnfinal}\end{gathered}$$ where we defined a dimensionless function in Fourier-Bessel space as $$\begin{gathered}
\tilde{A}({{\tau_{\rm f}}}, k_\perp, k_\eta)\equiv\frac{1}{s_{{\rm f}}{{\tau_{\rm f}}}}\Bigg(\chi_0 T_0\tau_0e^{-2H({{\tau_{\rm f}}},\tau_0, k_\perp,k_\eta)}
\\+\int_{\tau_0}^{{{\tau_{\rm f}}}}d\tau e^{-2H({{\tau_{\rm f}}},\tau, k_\perp,k_\eta)}\frac{d(\chi T \tau)}{d\tau}\Bigg).
\label{Adefinition}\end{gathered}$$ Note that $$\begin{aligned}
\tilde{A}({{\tau_{\rm f}}},k_\perp=0,k_\eta=0)=\frac{\chi_{{\rm f}}T_{{\rm f}}}{s_{{\rm f}}}\,.\end{aligned}$$ This, according to Eq. (\[nnfinal\]), implies the vanishing of $\left<\delta n(x) \delta n(y) \right>$ integrated over all space (this is trivially seen also in Eq. (\[nnsolution\])), which is a consequence of charge conservation.
It is useful to rewrite the integral over $\tau$ in Eq. (\[Adefinition\]) in terms of an integral over temperature $T$, which is straightforward. It is also useful to rewrite the integral in the definition of $H$ in Eq. (\[Hdef\]) in the same way. For that we need to know $dT/d\tau$. We shall neglect viscous corrections and use the ideal hydro equation $\tau s= {\rm constant}$, or $$\begin{aligned}
\frac{d T}{d \tau} + \frac{T v_s^2(T)}{\tau} = 0,\end{aligned}$$ where $v_s^2 \equiv \partial P / \partial \varepsilon$ is the speed of sound (at zero charge density). Then, $$\begin{gathered}
H(T_{{\rm f}},T,k_\perp,k_\eta)\\=\int_{T_{{\rm f}}}^T \left(\frac{D(T') s(T')
}{{{\tau_{\rm f}}}s_{{\rm f}}}\, k_\eta^2+ \frac{s_{{\rm f}}{{\tau_{\rm f}}}D(T') }{s(T')}\,k_\perp^2\right)\frac{dT^\prime}{v_s^2(T^\prime) T^\prime}\end{gathered}$$ For convenience, we define dimensionless quantities $\hat{D}(T)\equiv D T$, and $\hat{s}(T)\equiv s(T)/T^3$. Then $H(T_{{\rm f}},T,k_\perp,k_\eta)$ can be written as $$\begin{gathered}
H(T_{{\rm f}},T,k_\perp,k_\eta)= \frac{ k_\eta^2 }{{{\tau_{\rm f}}}s_{{\rm f}}}\int_{T_{{\rm f}}}^T \frac{ \hat{D}(T') \hat{s}(T^\prime) T'}{v_s^2(T')}dT^\prime
\\+ s_{{\rm f}}{{\tau_{\rm f}}}k_\perp^2\int_{T_{{\rm f}}}^T \frac{\hat{D}(T')}{v_s^2(T'){ \hat{s}(T^\prime)(T^\prime)}^5 }dT^\prime,
\label{Hdimensionless}\end{gathered}$$ where the function $\hat{s}$ can be taken directly from lattice data (see Section \[LatticeEOSSec\]). The speed of sound (at $ n=0$) can also be found from $\hat{s}$: $$\begin{aligned}
v_s^2(T) =\left(3+\frac{d\ln\hat{s}}{d\ln T}\right)^{-1}.
\label{SoundSquared}\end{aligned}$$ Combining Eq.(\[Adefinition\]) and Eq.(\[Hdimensionless\]), we find
$$\tilde{A}({{\tau_{\rm f}}}, k_\perp,k_\eta)=
\left[
\frac{\chi_0T_0}{s_0} e^{-2H(T_{{\rm f}},T_0,k_\perp,k_\eta)}
-
\int_{T_{{\rm f}}}^{T_0}\!dT\frac{d}{dT}\left(\frac{\chi T}{s}\right)e^{-2H(T_{{\rm f}},T,k_\perp,k_\eta)}
\right].
\label{Afinal}$$
Discussion and Interpretation {#sec:disc-interpr}
-----------------------------
Eq. (\[nnfinal\]) along with the definitions Eqs. (\[Hdimensionless\]), (\[Afinal\]) describes the charge density correlation due to stochastic noise and is the main result of this section.
The correlator in the square brackets in Eq. (\[nnfinal\]) naturally separates into a local part – the first term, which is independent of $k$ and is thus a $\delta$-function in position space, and the non-local part – the second term, given by Eq. (\[Afinal\]), which vanishes at large $k$.
For a Boltzmann gas (which is a good approximation at freezeout) we can identify the local term with the equilibrium self-correlation, which exists even in a non-interacting gas. On general grounds, the correlation function of a gas of particles in equilibrium is expected to have such a delta-function term [@LandauStatV5; @LandauStatV9] $$\begin{aligned}
\left<\delta n(\vec{x}_1) \delta n(\vec{x}_2) \right> = \bar{n} \, \delta^{3}(\vec{x}_1 - \vec{x}_2) + ...
\label{LandauCorrelation}\end{aligned}$$ where $``..."$ denotes the correlations from interactions. This delta function is not due to a correlation between [*two*]{} particles, as it is present even in a free gas. It is a trivial manifestation of statistical *fluctuations* in the gas, a reflection of the fact that particles are trivially correlated with themselves (see §116 of [@LandauStatV5]). In a free Boltzmann gas $\bar{n} = \chi T$ which is exactly the factor appearing in Eq. (\[nnfinal\]). The factor of ${{\tau_{\rm f}}}^{-1}$ in Eq. (\[nnfinal\]) is the volume Jacobian factor, $1/\sqrt{-g}$, for the delta-function in Bjorken coordinates. Because experimental measures count only [*two*]{}-particle correlations it is necessary to separate the self-correlation term before comparing with the data. The separation of such a self-correlation term has been also discussed in [@Pratt:2012dz], but not in [@Kapusta:2011gt; @Kapusta:2012zb]. We now turn to the non-local contribution to the correlator in Eq. (\[Afinal\]). We note that the dimensionless quantity $\chi T/s$ (charge susceptibility per entropy) and its $T$-dependence plays an important role. The first term in $\tilde{A}$ is a three-dimensional negative Gaussian with width (in position space) given by the diffusion distance over the entire expansion history (since $\tau_0$), and a magnitude controlled by the initial value of $\chi_0T_0/s_0$. If $\chi T /s$ were constant, which would be the case in a conformal theory, and is approximately the case in high temperature QGP, all non-trivial correlations resulting from the diffusion history of hydrodynamical fluctuations would be contained in this negative Gaussian.
The second term in Eq. (\[Adefinition\]) is due to the change of $\chi T /s$. This term is a superposition of many Gaussians with different widths and amplitudes. Because this contribution clearly requires the (charge-carrying) constituents of the plasma to change, its main contribution comes from the QCD crossover region. We also find that this term gives the dominant contribution to the charge correlations in heavy-ion collisions. It is, therefore, essential for our calculation to know $\chi T /
s$ throughout the history of a heavy-ion collision, especially in the crossover region, which is the subject of the next section.
The Temperature Dependence of Susceptibility per Entropy {#LatticeEOSSec}
--------------------------------------------------------
The behavior of entropy density $s$ and charge susceptibility $\chi$ as a function of temperature is easy to understand qualitatively and semiquantitatively. In the crossover region the QCD matter undergoes a smooth transition from the hadron gas to the QGP state. This leads to a significant increase in the number of degrees of freedom (liberation of color), i.e., growth of $s/T^3$. Although the number of charged degrees of freedom also increases, their average charge is smaller in QGP, and as a result the growth of $\chi/T^2$ is only moderate. The growth of $s/T^3$ is much more significant (due to the gluons) and as a result the dimensionless ratio $\chi T/s$ decreases with temperature in the crossover region.
To make the above description more quantitative we can estimate $\chi
T/s$ in the QGP by considering ideal massless gases of gluons and quarks (3 massless flavors). Although this approximation is only valid for asymptotically high $T$, it is sufficient for our illustrative purposes. A straightforward calculation leads to $$\label{eq:chisQGP}
\mbox{QGP:}\quad \frac{s}{T^{3}}= \frac{19\pi^2}{9};
\quad\frac{\chi}{T^{2}} = \frac23;
\quad \frac{\chi T}{s} = \frac{6}{19\pi^2}\,.$$ For a rough estimate of these quantities in the hadron gas phase we can take ideal gas of massless pions, for which we find $$\label{eq:chisHG}
\mbox{Pion gas:}\quad \frac{s}{T^{3}}= \frac{2\pi^2}{15};
\quad\frac{\chi}{T^{2}} = \frac13;
\quad \frac{\chi T}{s} = \frac{5}{2\pi^2}\,.$$ We see that around 16-fold increase of $s/T^3$ in QGP relative to the pion gas overwhelms the only 2-fold increase of $\chi/T^2$ leading to a significant decrease of $\chi T/s$.
These simple estimates are in qualitative and semiquantitative agreement with lattice QCD calculations [@Cheng:2007jq; @Cheng:2008zh; @Bazavov:2009zn; @Bazavov:2012jq; @Borsanyi:2010cj] which show that both entropy density and electric charge susceptibility change significantly in the crossover region. Figure \[ChiOverS\] shows our attempt to extract the ratio $\chi T/s$ from these lattice results. In our exploratory analysis we shall ignore statistical or systematic errors on these data and use equation of state shown in Fig. \[ChiOverS\] in our computations.
![(Color online). The temperature dependence (in units of $T$) of entropy ($s/T^3$), charge susceptibility ($\chi/T^2$) taken from the lattice data [@Borsanyi:2010cj; @Borsanyi:2011sw] and the resulting charge susceptibility per entropy (${\chi T}/{s}$) which we use in this paper.[]{data-label="ChiOverS"}](EOS){width="32.00000%"}
![(Color online). The temperature dependence (in units of $T$) of entropy ($s/T^3$), charge susceptibility ($\chi/T^2$) taken from the lattice data [@Borsanyi:2010cj; @Borsanyi:2011sw] and the resulting charge susceptibility per entropy (${\chi T}/{s}$) which we use in this paper.[]{data-label="ChiOverS"}](Chi){width="32.00000%"}
![(Color online). The temperature dependence (in units of $T$) of entropy ($s/T^3$), charge susceptibility ($\chi/T^2$) taken from the lattice data [@Borsanyi:2010cj; @Borsanyi:2011sw] and the resulting charge susceptibility per entropy (${\chi T}/{s}$) which we use in this paper.[]{data-label="ChiOverS"}](ChiOverS){width="32.00000%"}
Towards comparison with experiment {#FreezeOutSec}
==================================
With the lattice data on electric susceptibility $\chi$, the entropy density $s$ and the charge diffusion coefficient[^2] $D$, one can use the results of Sec. \[BjorkenStochasticSec\] to determine the spatial correlations of the net charge due to hydrodynamic fluctuations. These position space correlations need to be translated into particle momentum space correlations which are measured experimentally in a heavy-ion collision. To achieve this goal we need to address several important issues.
Partial chemical equilibrium {#sec:part-chem-equil}
----------------------------
The lattice equation of state, discussed in Sec. \[LatticeEOSSec\], describes QCD matter in full thermal and chemical equilibrium. Although this is a reasonable approximation during much of the expansion history, it breaks down after chemical freezeout. The rate of reactions responsible for chemical equilibration (inelastic collisions) is too slow in the hadronic phase to maintain chemical equilibrium. However, the thermal (kinetic) equilibrium is maintained until later times. In the intermediate region between the chemical and kinetic freezeout the matter can be described using the so-called partial chemical equilibrium (PCE) equation of state [@Huovinen:2009yb].
Rather than using the PCE equation of state we shall use a simpler approach, based on the observation in Ref. [@Teaney:2002aj] that the PCE equation of state expressed as pressure vs energy density is very similar to the full equilibrium (FE) equation of state. Since hydrodynamic equations involve the equation of state $P(\varepsilon)$, their solution under PCE should be similar to their solution under FE. [^3] The difference is manifested when we ask what the temperature is, given a value of the energy density, i.e., at a given point in time in the expansion history. Thus the actual kinetic freezeout temperature ${T_{\rm kf}}$, which determines the final (observed) momentum spectra of the particles, is different from the temperature, ${T_{\rm hf}}$ which would correspond to the final energy density in full equilibrium equation of state. The results of Ref. [@Teaney:2002aj] suggest that for the kinetic freezeout temperature ${T_{\rm kf}}\approx 100$ MeV the reasonable choice of the corresponding temperature at which the FE equation of state gives the same energy density is around ${T_{\rm hf}}\approx 130$ MeV. In this approach one ends hydrodynamic evolution at a final temperature (hydrodynamic freezeout) ${T_{\rm hf}}$ and implements the freezeout procedure with the momentum spectra of particles determined by ${T_{\rm kf}}$ [@Teaney:2002aj].
Additional conservation laws emerging under PCE are reflected in the appearance of corresponding chemical potentials. For our study we will need to use a chemical potential for the total pion number (the sum of the numbers of $\pi^+$ and $\pi^-$), $\mu_\pi$. The value of this chemical potential has also been estimated in Ref. [@Teaney:2002aj] in the range of $\mu_\pi\sim 80$ MeV. We shall use the above values for ${T_{\rm kf}}$, ${T_{\rm hf}}$ and $\mu_\pi$ in our calculations of the balance functions below.
Transverse expansion {#sec:transverse-expansion}
--------------------
Since significant contribution to the balance function comes from the crossover region, we must also take into account the fact that, due to finite transverse size of the colliding nuclei, the expansion does not remain purely Bjorken. Radial flow becomes significant at times $\tau$ of order the initial radius of the nucleus $R_0$ and the expansion approaches isotropic 3D expansion at later times. As a result the entropy density drops much faster, approximately as $\tau^{-3}$ during this later period, as opposed to $\tau^{-1}$ during the 1D Bjorken period [@Song:2007ux; @Teaney:2009qa].
Since analytical treatment of the full 3D expansion is beyond our reach, we consider a simple idealized approximation by assuming that the 3D stage of the expansion is short enough (due to fast drop in entropy, and thus, temperature) that the diffusion during this period can be neglected. We shall refer to this picture as the sudden transverse expansion approximation. In this idealized picture of the collision the expansion follows the 1D Bjorken solution until a point in time which we denote $\tau_{1D}$, upon which it undergoes sudden transverse expansion and freezes out shortly thereafter at time ${{\tau_{\rm f}}}$ with a pattern of flow given by the blast-wave ansatz.
We determine the charge correlator at time $\tau_{1D}$ (instead of ${{\tau_{\rm f}}}$) using Eq.(\[nnfinal\]) and Eq.(\[Adefinition\]) derived under conditions of the $1D$ Bjorken expansion $$\begin{gathered}
\Big\langle\delta\tilde{ n}_n(\tau_{1D},k_\perp,k_\eta)\,\delta \tilde{n}^*_m(\tau_{1D},q_\perp,q_\eta)\Big\rangle =\hat{\delta}_{mn}(k,q) \\\times \frac{s_{1D}}{\tau_{1D}}\cdot \Bigg[\left(\frac{\chi T}{s}\right)_{1D}-\tilde{A}(\tau_{1D},k_\perp,k_\eta)\Bigg],
\label{nnfinal1D}\end{gathered}$$ where $\tilde{A}(\tau_{1D},k_\perp,k_\eta)$ is given by Eq. (\[Afinal\]) with ${{\tau_{\rm f}}}$ replaced with $\tau_{1D}$. We then should treat Eq.(\[nnfinal1D\]) as the initial condition for the period of the “sudden” 3D expansion. This fast expansion is essentially adiabatic and thus the ratio $n/s$ is conserved. Since the entropy density drops from $s_{1D}$ to $s_{{\rm f}}$ during this period, the charge density must also drop by the same factor. This means the charge correlator at time ${{\tau_{\rm f}}}$ obeys $$\begin{aligned}
\Big\langle\delta\tilde{ n}_n\,\delta \tilde{n}^*_m\Big\rangle_{{\rm f}}=\frac{s^2_{{\rm f}}}{s^2_{1D}}\Big\langle\delta\tilde{ n}_n\,\delta\tilde{n}^*_m\Big\rangle_{1D}\,,\end{aligned}$$ where $\left\langle\delta\tilde{
n}_n\,\delta\tilde{n}^*_m\right\rangle_{1D}$ is given by Eq. (\[nnfinal1D\]). Thus $$\begin{gathered}
\Big\langle\delta\tilde{ n}_n\,\delta \tilde{n}^*_m\Big\rangle_{{\rm f}}=\hat{\delta}_{mn}(k,q) \cdot \frac{s_{{\rm f}}}{{{\tau_{\rm f}}}}\cdot\frac{s_{{\rm f}}{{\tau_{\rm f}}}}{s_{1D}\tau_{1D}}\\\times
\Bigg[\left(\frac{\chi T}{s}\right)_{1D}- \tilde{A}(\tau_{1D},k_\perp,k_\eta)\Bigg]\,.
\label{nnfinalFreeze-out1}\end{gathered}$$ The first (local in position space) term in Eq. (\[nnfinalFreeze-out1\]) contains the contribution of the self-correlation which we need to subtract. As discussed in Section \[sec:disc-interpr\], this self-correlation term is given by $\hat\delta_{mn}(k,q)(\chi
T/\tau)_{{\rm f}}$. Thus we write the charge density correlations at freezeout as $$\begin{aligned}
\Big\langle\delta\tilde{ n}_n\,\delta \tilde{n}^*_m\Big\rangle_{{\rm f}}\equiv\hat{\delta}_{mn}(k,q)\left(\frac{\chi_{{\rm f}}T_{{\rm f}}}{{{\tau_{\rm f}}}}
-\frac{s_{{\rm f}}}{{{\tau_{\rm f}}}}\tilde{A}_{{\rm f}}(k_\perp,k_\eta)\right),
\label{nnfinalFreeze-out2}\end{aligned}$$ which defines two-particle hydrodynamic correlator $\tilde{A}_{{\rm f}}(k_\perp,k_\eta)$ at freezeout. Comparing Eq.(\[nnfinalFreeze-out1\]) and Eq.(\[nnfinalFreeze-out2\]), we find $$\begin{gathered}
\tilde{A}_{{\rm f}}(k_\perp,k_\eta)
=\frac{s_{{\rm f}}{{\tau_{\rm f}}}}{s_{1D}\tau_{1D}}\cdot \tilde{A}(\tau_{1D},k_\perp,k_\eta)
\\
+\left(\frac{\chi T}{s}\right)_{{\rm f}}-\frac{s_{{\rm f}}{{\tau_{\rm f}}}}{s_{1D}\tau_{1D}}\cdot\left(\frac{\chi
T}{s}\right)_{1D}
\label{CPractical}\end{gathered}$$ and use it to calculate the balance function later in this section.
From Eq.(\[CPractical\]) we see that the density-density correlations built during the 1D Bjorken expansion period are diluted due to the transverse expansion by a factor ${(s_{{\rm f}}{{\tau_{\rm f}}})
}/{(s_{1D}\tau_{1D})}$ which would be equal to 1 if the system continued pure 1D expansion until freeze-out. Furthermore, the correlator $\tilde A_{{\rm f}}$ contains a local term, independent of $k$, because the last two terms in Eq. (\[CPractical\]) do not cancel. This is the contribution of the noise from the period of the sudden transverse expansion. It is represented by a delta function in position space because the noise is local and we neglected diffusion during this short time, which would otherwise broaden the delta function.
Cooper-Frye freezeout
---------------------
In order to compare our results with experimental measurements, we need to translate the hydrodynamic correlations in position space into correlations in the kinematic (momentum) space of the observed particles. For this purpose we use the standard Cooper-Frye prescription for pions: $$\begin{aligned}
\frac{dN_Q}{dy d\phi}=\frac{1}{(2\pi)^3}\int p_\perp dp_\perp \int d\sigma_\mu p^\mu\, f_Q(\vec{x},\vec{p})\,,
\label{CooperFrye}\end{aligned}$$ where $f_Q= \exp\left\{ Q\mu/{T_{\rm hf}}+\mu_\pi/{T_{\rm kf}}- p_\mu u^\mu /{T_{\rm kf}}\right\}$ is the equilibrium distribution function for pions carrying charge $Q$ (equal to $\pm1$) in the Boltzmann approximation [^4].
We have also defined kinematic rapidity as $y \equiv \tanh^{-1}\left(p^z/E\right) $, kinematic azimuthal angle as $\phi\equiv
\tan^{-1}\left(p^y/p^x\right)$, and denoted the freezeout hypersurface normal 4-vector as $d\sigma_\mu$. The $p_\perp$ integration range is determined by experimental $p_\perp$ cuts. We choose an isochronous freeze-out condition [^5] at $\tau = {{\tau_{\rm f}}}$, thus $$d\sigma_\mu p^\mu ={{\tau_{\rm f}}}m_\perp d^2x_\perp d\eta \cosh(y-\eta),$$ where $m_\perp \equiv \sqrt{p_\perp^2 + m^2}$, with $m$ being the rest mass of the pion.
Since we are interested in the effect of the hydrodynamic fluctuations, we expand the distribution function to linear order in fluctuations of temperature, chemical potential, and fluid velocity. If the average of the net chemical potential $\bar{\mu}$ is 0, then only the chemical potential fluctuation survives in the difference between particles and antiparticles: $$\begin{gathered}
\delta \frac{d{N_{\rm net}}}{d y d \phi}=\frac{2 {{\tau_{\rm f}}}}{(2\pi)^3 {T_{\rm hf}}}\cdot\int m^2_\perp dm_\perp \int d^2x_\perp \int d\eta \\\times
\delta\mu({{\tau_{\rm f}}},\vec x_\perp,\eta)\, \cosh(y-\eta) f_0(\vec{x},\vec{p}),
\label{3.3}\end{gathered}$$ where $f_0$ is the Boltzmann distribution function at $\mu=0$ and $$\begin{aligned}
{N_{\rm net}}\equiv N_+ - N_-.\end{aligned}$$ Fluctuations of chemical potential are related to those of the charge density by $\delta n=\chi\,\delta \mu $.
Blast Wave {#BlastWaveSec}
----------
As we already discussed in Section \[sec:transverse-expansion\], finiteness of the transverse size of the system leads to transverse expansion. We shall describe the transverse flow velocity profile $v_r(r)$ using transverse rapidity $\kappa(r)$ $$\label{eq:kappa}
v_r(r) = \frac{u^r}{u^\tau} \equiv \tanh \kappa(r).$$ The distribution function $f_0$ can be then written as [@Schnedermann:1993ws] $$\begin{gathered}
f_0(\vec{x},\vec{p}) = \exp \big\{\hat\mu_\pi+\hat p_\perp
\cos(\phi - \psi)\sinh \kappa_{{\rm f}}(r) \\- \hat m_\perp \cosh(y-\eta) \cosh \kappa_{{\rm f}}(r) \big\}
\label{f0radialflow}\end{gathered}$$ where $\kappa_{{\rm f}}(r)$ describes the radial flow profile at kinetic freeze-out, $\psi$ is the position space azimuthal angle characterizing the direction of the radius-vector $\vec x$, and we introduced convenient dimensionless variables: $$\label{eq:hatm}
\hat \mu_\pi= \mu_\pi/{T_{\rm kf}}\quad \hat m_\perp= m_\perp/{T_{\rm kf}},
\quad\hat p_\perp= p_\perp/{T_{\rm kf}}.$$
We apply the standard blast-wave approach, i.e., we specify the radial flow profile $\kappa(r)$ at freezeout by hand (as a linear function of $r$) and limit the transverse size of the system: $r<R$. Such an approach is known to provide a reasonable approximation to single particle observables computed using a full hydrodynamic solution which includes transverse expansion [@Teaney:2002zt]. Finally, we have $$\begin{gathered}
\label{eq:dNnet}
\delta \frac{d{N_{\rm net}}}{d y d \phi}=\frac{ {{\tau_{\rm f}}}{T_{\rm kf}}^3 R^2}{
{\chi_{\rm hf}}{T_{\rm hf}}} \int d^2\vec x_\perp \int_{-\infty}^{\infty} d\eta
\\\times \delta n({{\tau_{\rm f}}},\vec x_\perp,\eta)\, F(\vec{x},\vec{p}),\end{gathered}$$ where we introduced the function $$\begin{gathered}
F(\vec{x},\vec{p}) \equiv \frac{1}{4 \pi^3 R^2}\int \hat
m_\perp^2 d\hat m_\perp \cosh(y-\eta)\\\times f_0(\vec{x},\vec{p}) \Theta(R-r).
\label{Fdef}\end{gathered}$$ which acts as a kernel of the transformation from the position variables $\vec x$ to kinematic variables $\vec p$. We normalized $F$ in such a way that its Fourier-Bessel transform is dimensionless (see below).
To proceed, we introduce Fourier-Bessel expansions for both $\delta n$ and $F$ in Eq. (\[eq:dNnet\]). Due to azimuthal and boost invariance (and integration over $m_\perp$ in Eq. (\[Fdef\])) the function $F(\vec x, \vec p)$ depends only on three arguments: $r$ and the differences $\phi - \psi$, and $y -
\eta$. We define the Fourier-Bessel transform with respect to these three variables as $\tilde F_n(k_\perp,k_y)$ in terms of which we find, substituting Eq. (\[BesselTransform-f\]) and using the closure relation (\[BesselClosure\]) $$\begin{gathered}
\label{eq:dNet-Fn}
\delta \frac{d{N_{\rm net}}}{d y d \phi}=\frac{ {{\tau_{\rm f}}}{T_{\rm kf}}^3 R^2}{ {\chi_{\rm hf}}{T_{\rm hf}}} \int_{-\infty}^{\infty} \frac{dk_y e^{i k_y y}}{2\pi}
\sum_{n} \frac{e^{ i n \phi}}{2\pi} \int_0^{\infty} k_\perp dk_\perp \\\times \delta \tilde{n}_n({{\tau_{\rm f}}},k_\perp,k_y) \tilde{F}_n(k_\perp,k_{y}).\end{gathered}$$ For a given transverse flow profile at freezeout $\kappa_{{\rm f}}(r)$ in Eq. (\[eq:kappa\]) we can obtain an expression for $\tilde{F}_n(k_\perp,k_{y})$ by substituting Eq. (\[f0radialflow\]) into Eq. (\[Fdef\]) and integrating over variables $(\phi-\psi)$ and $(y-\eta)$ in the definition of the Fourier-Bessel transform Eq. (\[BesselTransform\]) $$\begin{gathered}
\tilde{F}_n(k_\perp,k_y) =-\frac{e^{\hat\mu_\pi}}{\pi^2}\int {\hat m_\perp}^2
d\hat m_\perp\int_0^1 \!\hat rd\hat r\, J_n(\hat k_\perp \hat
r)\,\\\times\mathcal{I}_n(\hat p_\perp\sinh \kappa_{{\rm f}})\, \mathcal{K}_{ik_y}'(\hat m_\perp\cosh\kappa_{{\rm f}}),\end{gathered}$$ where ${\cal I}$ is a modified Bessel function, ${\cal K}'$ is the derivative of a modified Bessel function with respect to its argument and we used convenient dimensionless variables defined in Eq. (\[eq:hatm\]) as well as $$\label{eq:hatk}
\quad \hat k_\perp = k_\perp R
\quad \mbox{and} \quad \hat r = r/R.$$
It is also useful to note that the average value of the total number of charged pions $$\label{eq:Nch}
N_{\rm ch} \equiv N_+ + N_-$$ per unit rapidity and azimuthal angle given by $$\begin{gathered}
\Big\langle\frac{dN_{\rm ch}}{dy d\phi}\Big\rangle=\frac{2 {{\tau_{\rm f}}}}{(2\pi)^3}\int m_\perp^2 dm_\perp \int_{-\infty}^{\infty} d\eta \int d^2x_\perp \\\times\cosh(y-\eta)f_0(\vec{x},\vec{p}) \Theta(R-r)\end{gathered}$$ can be also expressed as $$\label{eq:F00}
\Big\langle\frac{dN_{\rm ch}}{dy d\phi}\Big\rangle =
{{{\tau_{\rm f}}}{T_{\rm kf}}^3 R^2}\tilde F_0(0,0)\,.$$
Particle Correlations and Balance Function
------------------------------------------
Finally, to determine the particle correlations, we multiply two fluctuations given by Eq. (\[eq:dNet-Fn\]), average over events and express the correlator $\langle \delta \tilde n_n\delta\tilde
n_m\rangle$ using Eq. (\[nnfinal\]), with the self-correlation subtracted. The delta functions in $\hat{\delta}_{mn}$ ensure that the result is only a function of the rapidity difference $\Delta y \equiv y_2 - y_1$, and angular difference $\Delta \phi = \phi_2 - \phi_1$ and we find $$\begin{gathered}
\label{eq:dNnetdNnet}
\left\langle\delta\frac{d{N_{\rm net}}}{dy_1 d\phi_1}\ \delta\frac{d{N_{\rm net}}}{dy_2 d\phi_2} \right\rangle
= -\left(\frac{{T_{\rm kf}}^3}{ {\chi_{\rm hf}}{T_{\rm hf}}}\right)^2 {s_{{\rm f}}{{\tau_{\rm f}}}} R^2\\
\times\int_{-\infty}^{\infty} \frac{dk_y e^{i k_y \Delta y}}{2\pi}
\sum_{n} \frac{e^{ i n \Delta \phi}}{2\pi}
\int_0^{\infty} \hat k_\perp d\hat k_\perp \\
\times \tilde{A}_{{\rm f}}(k_\perp,k_y) \tilde{F}_n(k_\perp,k_{y})\tilde{F}^*_n(k_\perp,k_{y})\,,\end{gathered}$$ where $k_\perp=\hat k_\perp/ R$ as in Eq. (\[eq:hatk\]).
When $\left<N_+ \right> = \left<N_- \right>$, the correlator in Eq. (\[eq:dNnetdNnet\]) is related to the balance function defined in [@Bass:2000az; @Jeon:2001ue] by $$\begin{aligned}
B( \Delta y,\Delta \phi) \equiv -\left\langle\delta\frac{d{N_{\rm net}}}{dy_1 d\phi_1}\ \delta\frac{d{N_{\rm net}}}{dy_2 d\phi_2} \right\rangle
\left< \frac{d N_{\rm ch}}{dy d\phi} \right>^{-1}.
\label{BalanceDefText}\end{aligned}$$ This relationship is derived in Appendix \[BalanceAppendix\]. Finally, putting Eqs. (\[BalanceDefText\]), (\[eq:dNnetdNnet\]) and (\[eq:F00\]) together, we find the expression for the balance function $$\begin{gathered}
\label{BalanceFinal}
B( \Delta y,\Delta \phi) = \frac{ {T_{\rm kf}}^3 s_{{\rm f}}}{ {\chi_{\rm hf}}^2 {T_{\rm hf}}^2 \tilde{F}_0(0,0)} \\
\times \int_{-\infty}^{\infty} \frac{dk_y e^{i k_y \Delta y}}{2\pi}
\sum_{n} \frac{e^{ i n \Delta \phi}}{2\pi}
\int_0^{\infty} \hat k_\perp d\hat k_\perp \\
\times
\tilde{F}_n(k_\perp,k_{y})\tilde{F}^*_n(k_\perp,k_{y})
\tilde{A}_{{\rm f}}(k_\perp,k_y)\,.
$$ We use Eq. (\[BalanceFinal\]) with $\tilde{A}_{{\rm f}}(k_\perp,k_y)$ given by Eq. (\[CPractical\]) to calculate the balance functions in the next section.
We can calculate the rapidity and the azimuthal projections of the balance function $$\begin{aligned}
B(\Delta y)&=&\int_{-\pi}^\pi d\Delta\phi \, B(\Delta y,\Delta \phi),
\notag
\\
B(\Delta\phi)&=&\int_{-\infty}^\infty d\Delta y\, B(\Delta y,\Delta \phi).
\label{BalanceFunctionProjection}\end{aligned}$$ Integration over $\Delta \phi$ is equivalent to only considering the $n=0$ moment in the summation of Eq.(\[BalanceFinal\]), while integration over $\Delta y$ is equivalent to setting $k_y=0$ instead of performing an integral over $k_y$.
Results {#sec:results}
-------
In order to illustrate the typical shape, width and magnitude of the balance function arising due to the hydrodynamic fluctuations we calculate this function using our semianalytical model of expansion described above. For central collisions at top RHIC energies, we set the time when expansion stops being purely one-dimensional to $\tau_{1D}=7$ fm and the corresponding temperature to $T_{1D}=150$ MeV. We set the initial temperature to $T_0=350$ MeV. The hydro freezeout temperature (see Section \[sec:part-chem-equil\]) is taken to be ${T_{\rm hf}}=130$ MeV [@Teaney:2002aj] We use the lattice data on entropy density $s(T)$ [@Borsanyi:2010cj] and electric charge susceptibility $\chi(T)$ [@Borsanyi:2011sw] as in Fig. \[ChiOverS\]. We set the blast-wave transverse flow profile to be linear $v_r=\frac{3}{2}\langle
\beta\rangle {r}/{R}$ with $\langle \beta\rangle=0.6$ and maximum radius $R=12$ fm at ${{\tau_{\rm f}}}=12$ fm [@Shen:2012vn; @Teaney:2002aj; @Abelev:2008ab].
{width="45.00000%"} {width="45.00000%"}
In Fig.\[FinalBalanceFig\], we show the sensitivity of the balance function to the charge diffusion coefficient, taking the dimensionless combination $DT$ to be constant, with other parameters fixed. In particular, we see that, for chosen values of parameters, the data favors small values of the diffusion coefficient, $2\pi D T\sim 1$, which is characteristic of a strongly coupled medium (short mean free path). Clearly, our semiquantitative analysis is not sufficient to pin down the value of the diffusion coefficient with adequate precision, due to the balance function’s sensitivity to parameters which we fixed by hand (using typical values obtained in numerical hydro simulations). However, our results are indicative of the typical resolution one could achieve if a more realistic numerical hydrodynamic simulation were to be used instead of our simplified analytical model. We leave such quantitative investigations to future work.
Conclusions and discussion {#ConclusionSec}
==========================
We showed that intrinsic hydrodynamic noise induces correlations of charge fluctuations which are observable and typically quantified in terms of the charge balance functions. We have shown how to calculate the noise contribution to the balance function and applied our method to a semianalytical model of hydrodynamic expansion. The balance functions we obtain are in reasonable agreement with experiments and our results suggest that a more realistic calculation may allow one to determine or constrain the charge diffusion coefficient $D$. Our semiquantitative analysis indicates that a small value of $D$, characteristic of the strongly-coupled medium, is favored by the data.
Two main observations characterize the effect of the hydrodynamic noise and diffusion on the charge balance functions. We find that the magnitude of the balance function receives the most significant contribution from the time interval during the expansion where the charge susceptibility per entropy $\chi T/s$ changes most. The rapidity width of the balance function is determined by the diffusion distance that the (originally local) correlation induced by noise propagates during the time from its origin to the freezeout time. [^6]
It is easy to understand that a change of the system’s thermodynamic state is necessary to produce a non-local correlation. Indeed, in a static medium the correlations must be [*local*]{} (on hydrodynamic scale) despite diffusion. This requires that the contributions from successive time intervals cancel each other in a static medium, leaving eventually only the (local) contribution from the most recent time. We found that such cancellations could also occur in a medium undergoing boost-invariant longitudinal expansion as long as $\chi
T\tau$ is constant (which is the same as $\chi T/s$ being constant up to small viscous corrections). In general, however, the expansion leads to nonlocal correlations which carry the memory of the expansion.
One can think of this picture as the hydrodynamic description of the mechanism of the suppression of charge fluctuations proposed and analyzed in Refs. [@Asakawa:2000wh; @Jeon:2000wg; @Shuryak:2000pd]. Indeed the D-measure, $D_m$, introduced in Ref.[@Jeon:2000wg] is related to the balance function as (see Appendix \[BalanceAppendix\])
$$\begin{aligned}
D_m\equiv 4\frac{\langle(\delta {N_{\rm net}})^2\rangle}{\langle N_{\rm ch}\rangle}
= 4 \left[1- \int_{-\infty}^{\infty} B(\Delta y) d\Delta y \right].
\label{DMeasureDef}\end{aligned}$$
Therefore a positive balance function corresponds to suppression of net charge fluctuations ($D_m<4$). The balance function provides differential phase-space information on the distribution of the anti-correlation responsible for the suppression. Moreover, the positivity of the balance function can be seen as a direct consequence of the fact that $\chi T/s$ is smaller in QGP, i.e., $d(\chi
T/s)/dT<0$ (see Eq. (\[Afinal\])), which is the starting point of the argument in [@Asakawa:2000wh; @Jeon:2000wg].
One can also view this hydrodynamic picture as effectively representing the qualitative microscopic mechanism of charge balancing described in Refs.[@Bass:2000az; @Jeon:2001ue; @Pratt:2011bc; @Pratt:2012dz]. The advantage of hydrodynamic description is that it does not need to rely on existence of quasiparticles. This is especially important because both quark and hadron quasiparticle descriptions must break down in the crossover region, and this is the region responsible for the major contribution to the balance function. Our approach allows quantitative description of these phenomena from first principles, i.e., from the (lattice) equation of state and information on kinetic coefficients, within a universal hydrodynamic formalism.
One of the many simplifying assumptions in our semianalytic calculation has been the assumption that dimensionless combination $DT$ is temperature independent. It is, perhaps, the easiest assumption to relax, provided information of the temperature dependence of the diffusion coefficient $D$ was available. Unlike the entropy and charge susceptibility which, being static thermodynamic quantities, can be reliably measured on the lattice, the diffusion coefficient is a property of the real-time low-frequency response, which the Euclidean time lattice calculation has well-known difficulties accessing. With this caveat, it would be still interesting to extract the temperature dependence of the charge diffusion coefficient from the existing lattice data.
As a first exploratory step we attempted to combine recent lattice data on electric conductivity [@Amato:2013naa] with the electric susceptibility $\chi$ data from Ref. [@Borsanyi:2011sw]. Using the relation $D=\sigma/\chi$ we can then plot the temperature dependence of the diffusion coefficient, or its dimensionless combination $2\pi DT$ (see Fig.\[DT\]).
![(Color online.) (a) The temperature dependence of electric charge diffusion coefficient in units of $(2\pi T)^{-1}$ obtained by combining the lattice data from Refs. [@Borsanyi:2011sw; @Amato:2013naa]. (b) The balance function using the same parameters as for Fig. \[FinalBalanceFig\] but with the temperature dependent $2\pi D T$ taken from the lattice data. []{data-label="DT"}](DT "fig:"){width="32.00000%"}\
1em ![(Color online.) (a) The temperature dependence of electric charge diffusion coefficient in units of $(2\pi T)^{-1}$ obtained by combining the lattice data from Refs. [@Borsanyi:2011sw; @Amato:2013naa]. (b) The balance function using the same parameters as for Fig. \[FinalBalanceFig\] but with the temperature dependent $2\pi D T$ taken from the lattice data. []{data-label="DT"}](RHIC_B_Y_TrueDT "fig:"){width="42.00000%"}
Despite large error bars one can see that lattice results suggest that the diffusion coefficient $D$ is indeed of order $1/2\pi
T$ in the crossover region, where we now know most of the contribution to the balance function comes from. This is consistent with the results of our comparison with experimental data in Fig. \[FinalBalanceFig\].
Taking the lattice data as given (and ignoring the error bars) we can also calculate the balance function using our semianalytic model. The result plotted in Fig. \[DT\] shows a reasonable agreement with the data.
An important improvement of our approach can be achieved by implementing a more realistic pattern of radial flow, replacing the blast-wave and sudden transverse expansion approximation. An approach based on the analytic solution proposed by Gubser [*et al.*]{} [@Gubser:2010ze; @Gubser:2010ui] is tempting. However, the limitation of this approach to a conformal equation of state is too restrictive for our purpose, since the major contribution to the balance function comes from the non-conformal (crossover) region. A fully numerical hydrodynamic simulation with stochastic noise will, of course, enable a quantitative comparison with experiment. It would also allow extension of our results to non-central (azimuthally asymmetric) collisions.
A natural application of the stochastic hydrodynamic approach is to fluctuations near the QCD critical point [@Stephanov:2004wx] as has been already initiated by [@Kapusta:2012zb]. The conductivity $\sigma$ and susceptibility $\chi$ diverge at the critical point, leading to the expected increase of the charge fluctuations at the critical point. Such fluctuations are important signatures of the critical point in heavy-ion collisions [@Stephanov:1998dy; @Stephanov:1999zu; @Hatta:2003wn] and understanding the effect of the time evolution [@Berdnikov:1999ph; @Stephanov:2009ra] and expansion on these signatures is important for obtaining quantitative predictions. However, such an application requires extension of the formalism to non-zero baryon density where energy and flow velocity fluctuations now mix with charge fluctuations [@Kapusta:2012zb].
Another possible future direction is the examination of the balance functions of kaons or protons [@Pratt:2012dz], where the fluctuation of strangeness or baryon number may become important. It should be straightforward to generalize our work to multiple conserved charges. We leave these topics for future studies.
Acknowledgments {#acknowledgments .unnumbered}
===============
We thank D. Teaney, H.-U. Yee and Y. Yin for discussions. This work is supported by the US Department of Energy under grant No. DE-FG0201ER41195.
A review of balance functions {#BalanceAppendix}
=============================
For completeness we review here the definitions and properties of the balance functions [@Bass:2000az; @Jeon:2001ue].
To define the balance function we divide the phase space occupied by particles produced in a heavy-ion collision into (infinitesimally) small cells. For the purpose of this paper we consider cells in rapidity $y$ and azimuthal angle $\phi$ (but integrated over transverse momentum) and denote the coordinates of the cell $\Gamma=(y, \phi)$ and the volume of the cell $d\Gamma=dy\, d\phi$. We denote the number of particles of charge $a=+,-$ in a cell as $dN^a(\Gamma)$ and its event average $\langle
dN^a(\Gamma)\rangle$. Since $\langle dN^a(\Gamma)\rangle\sim
d\Gamma$ is infinitesimally small, the probability of finding more than one particle in a cell is negligible (${\cal O}(d\Gamma^2)$ for two or more particles) and the average $\langle dN^a(\Gamma)\rangle\ll1$ is also the probability to find a particle of charge $a$ in the cell.
The conditional probability of finding a particle of charge $b$ in another cell $\Gamma_2$ given a particle of charge $a$ in the cell $\Gamma_1$ can be found as $\langle
dN^b(\Gamma_2)\,dN^a(\Gamma_1)\rangle/\langle dN^a(\Gamma_1)\rangle$ which is easy to understand keeping in mind that $dN^a(\Gamma)$ is either $0$ or (rarely) $1$. The balance function defined on a pair of cells is given by: $$\begin{gathered}
\label{eq:B12}
B(\Gamma_2,\Gamma_1)=\frac1{2}\sum_{a=+,-}
\frac{\langle dN^{-a}_2 dN^a_1\rangle-\langle dN^a_2 dN^a_1\rangle}
{d\Gamma_2\langle dN^a_1\rangle}\\
=
\frac{\langle n^{-a}_2 n^a_1\rangle-\langle n^a_2 n^a_1\rangle}
{\langle n^a_1\rangle}\end{gathered}$$ where we used a shorthand $dN^a_i\equiv dN^a(\Gamma_i)$ and introduced density per phase space volume $n\equiv dN/d\Gamma$. The balance function measures a difference in conditional probabilities of finding a particle of the opposite charge $-a$ vs the same charge $a$ in the cell $\Gamma_2$ given a particle of the charge $a$ in cell $\Gamma_1$. This probability is proportional to the volume $d\Gamma_2$ of the cell and is infinitesimally small, while its ratio to $d\Gamma_2$, as in Eq. (\[eq:B12\]), is finite.
Since we are considering a case when $\mu=0$, we can use $\langle
n^-_1\rangle=\langle
n^+_1\rangle$ to simplify Eq. (\[eq:B12\]): $$\label{eq:B12-simple}
B(\Gamma_2,\Gamma_1)=
-\frac
{\left\langle (n^{+}_2 - n^-_2)(n^{+}_1 - n^-_1)\right\rangle}
{2\langle n^+_1\rangle}
= -\frac
{\left\langle n^{\rm net}_2\,n^{\rm net}_1\right\rangle}
{\langle n^{\rm ch}_1\rangle},
$$ Since $n^{\rm net}=d{N_{\rm net}}/dyd\phi$ and ${N_{\rm net}}=\delta {N_{\rm net}}$ ($\langle{N_{\rm net}}\rangle=0$), this gives us equation (\[BalanceDefText\]) used in the text.
One also defines the balance function as a function of the phase space displacement $\Delta\Gamma\equiv
\Gamma_2-\Gamma_1=(y_2-y_1,\phi_2-\phi_1)$ by summing in Eq. (\[eq:B12\]) over all cells $\Gamma_1$ and $\Gamma_2$ separated by $\Delta \Gamma$. To obtain a finite result for infinitely many infinitesimally small cells ($d\Gamma_i\to0$) we multiply by $d\Gamma_1d\Gamma_2$. We can then write this summation as an integral: $$\label{eq:B-Delta}
B(\Delta\Gamma)=\frac1{\int d\Gamma}\,
\int \!d\Gamma_2\int \!d\Gamma_1\
\delta(\Gamma_2-\Gamma_1-\Delta\Gamma)\,B(\Gamma_2,\Gamma_1)\,
.$$ The normalization factor is chosen in such a way that the result tends to a finite limit with increasing total phase-space volume ($\int d\Gamma$).
The expression in Eq. (\[eq:B-Delta\]) simplifies in the case of azimuthal and boost ($\Gamma\to\Gamma+\Delta \Gamma$) invariance. Since in this case the balance function $B(\Gamma_2,\Gamma_1)$ can only depend on $\Delta\Gamma$ we find from Eq. (\[eq:B-Delta\]), simply, $$\label{eq:B-Delta2}
B(\Delta\Gamma)=B(\Gamma+\Delta \Gamma,\Gamma),$$ for any $\Gamma$.
The derivation above assumes that the rapidity acceptance window is infinite: $y\in(-\infty,\infty)$, or more precisely, is much larger than the rapidity range of the balance function $B(\Delta y,\Delta\phi)$. In practice, the rapidity interval has a finite width $Y$. Still assuming boost invariance, but integrating in Eq. (\[eq:B-Delta\]) over the finite rapidity window of width $Y$ we find the balance function in a finite rapidity acceptance: $$\label{eq:BY}
B(\Delta\Gamma)_Y = B(\Delta\Gamma)_\infty\,\frac{Y-\Delta y}{Y}\,$$ where we used $\int dy= Y$ and $\int dy_2\int dy_1\
\delta(y_2-y_1-\Delta y)=Y-\Delta y$.
To express the D-measure $D_m$ [@Jeon:2000wg] in terms of the balance function we substitute ${N_{\rm net}}=\int\! d\Gamma\, n^{\rm
net}(\Gamma)$ and $N_{\rm ch}=\int\! d\Gamma\, n^{\rm
ch}(\Gamma)$ into the definition $$\label{eq:Dm}
D_m \equiv 4\frac{\langle(\delta {N_{\rm net}})^2\rangle}{\langle N_{\rm
ch}\rangle}
= \frac4{\int d\Gamma}\,
\int \!d\Gamma_2\int \!d\Gamma_1\
\frac{\left\langle n^{\rm net}_2\,n^{\rm net}_1\right\rangle}
{\langle n^{\rm ch}\rangle}$$ The integrand is $-B(\Gamma_2,\Gamma_1)$ as given by Eq. (\[eq:B-Delta\]), except for $\Gamma_1=\Gamma_2$, when Eq. (\[eq:B-Delta\]) does not apply (we have only defined $B(\Gamma_2,\Gamma_1)$ for $\Gamma_1\neq\Gamma_2$). We can calculate the contribution from the cells $\Gamma_1=\Gamma_2$ to Eq. (\[eq:Dm\]) separately. We note that since $dN^a$ takes (most of the time) values 0 or 1, $(dN^a)^2=dN^a$ and thus $\langle(dN^+-dN^-)^2\rangle=\langle dN^++dN^-\rangle$, or $\langle \delta
{N_{\rm net}}^2\rangle = \langle\delta N_{\rm ch}\rangle$, up to terms of order ${\cal O}(d\Gamma)^2$. Therefore, since $n=dN/d\Gamma$, the integrand in Eq. (\[eq:Dm\]) for $\Gamma_1=\Gamma_2$ is $\langle
(n^{\rm net})^2\rangle/\langle n^{\rm ch}\rangle =
1/d\Gamma$. Summation over all cells with $\Gamma_1=\Gamma_2$ gives therefore a contribution to $D_m$ equal to (up to infinitesimally small terms ${\cal O}(d\Gamma)$), $4\,{\textstyle\left(\int d\Gamma\right)^{-1}}\! \int d\Gamma d\Gamma
\cdot 1/d\Gamma=4$. This is the value of $D_m$ for completely uncorrelated particles. Adding the contributions from $\Gamma_1\neq
\Gamma_2$ we find therefore $$\begin{gathered}
\label{eq:D-4-B}
D_m = 4 \left( 1 - \,{\textstyle\left(\int d\Gamma\right)^{-1}}\!\int\! d\Gamma_1 \!\int\! d\Gamma_2\,
B(\Gamma_2,\Gamma_1)\right) \\=4 \left( 1 - \int\! d(\Delta\Gamma) \,
B(\Delta\Gamma)\right),\end{gathered}$$ where we used Eq. (\[eq:B-Delta\]) for the last equality.
[^1]: We are splitting a two-component vector into the gradient of a scalar, $I_S$, and a divergenceless two-vector, which, in two dimensions can also be written in terms of a scalar $I_V$.
[^2]: Lattice data on the charge conductivity or the diffusion coefficient $D$ is subject to more uncertainties due to the analytic continuation from imaginary to real time and will be discussed at the end of this section and in Section \[ConclusionSec\].
[^3]: To maintain simplicity and transparency of our results we make an assumption that the same is true for the charge susceptibility $\chi T$. The charge susceptibility was not discussed in Ref. [@Teaney:2002aj], and it would be interesting to test this natural assumption by applying the same methods to calculate $\chi
T$. Deviations from this assumption can be included in our approach if necessary.
[^4]: The factors $1/T$ accompanying $\mu$ and $\mu_\pi$ in $f_Q$ reflect the definitions of these chemical potentials. While $\mu$ is defined in terms of the FE equation of state used in hydrodynamics, the potential $\mu_\pi$ accounts for the pion excess at kinetic freezeout due to PCE.
[^5]: For net charge correlations at zero chemical potential this is equivalent to isothermal freeze-out because fluctuations of temperature do not mix with charge fluctuations.
[^6]: The azimuthal width of the balance function is also sensitive to diffusion, but is strongly affected by the radial flow.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'This empirical paper examines the adoption of PHP releases in the the contemporary world wide web. Motivated by continuous software engineering practices and software traceability improvements for release engineering, the empirical analysis is based on big data collected by web crawling. According to the empirical results based on discrete time-homogeneous Markov chain (DTMC) analysis, (i) adoption of PHP releases has been relatively uniform across the domains observed, (ii) which tend to also adopt either old or new PHP releases relatively infrequently. Although there are outliers, (iii) downgrading of PHP releases is generally rare. To some extent, (iv) the results vary between the recent history from 2016 to early 2017 and the long-run evolution in the 2010s. In addition to these empirical results, the paper contributes to the software evolution and release engineering research traditions by elaborating the applied use of DTMCs for systematic empirical tracing of online software deployments.'
author:
-
-
title: 'How PHP Releases Are Adopted in the Wild?'
---
release engineering, software evolution, continuous delivery, patching, upgrading, downgrading, web crawling
Introduction
============
Programming languages evolve like any other software [@Favre05]. Like most software, also programming languages require release engineering, and as with conventional software, users of a programming language are likely to abandon the language if it is not properly updated and maintained to meet the continuously changing requirements [@Meyerovich12]. In recent years, different continuous software engineering practices have become increasingly popular for the development and maintenance of conventional software. Interestingly, also programming languages such as PHP have adopted a strategy of continuous releases scheduled to occur in a fast and fixed release cycle.
This paper investigates the continuous release engineering of the PHP programming language from a perspective of deployments using the language to serve some of the most popular web sites in the current Internet. While the release engineering practices used by the PHP project establish the practical motivation, the primary scholarly purpose of the investigation is to examine the previously unexplored use of classical DTMCs for studying release engineering. For putting the elaborated DTMCs into work, large web crawling datasets are used to analyze the current PHP release adoption patterns.
Markov chains belong to the classical methodology toolbox in reliability engineering [@Cheung80; @Wang12]. Discrete time-homogeneous Markov chains have recently been also adopted for studying different empirical software engineering problems, including those related to software evolution [@WongCai11]. Thus far, however, empirical applications have been limited in the release engineering domain. In addition to patching this limitation in the literature, this paper contributes to the release-based approaches (as opposed to approaches based on version control and bug tracking systems) for studying software evolution.
There exists a large amount of empirical work using a release-oriented perspective to study software evolution in general and release engineering in particular. Backward-compatibility [@Raemaekers14], library dependencies [@Kula15], and so-called rapid releases [@Karvonen17; @Mantyla13; @daCosta16] are good examples of recent questions examined (for a review of current research challenges see [@Adams16]). Many of these studies deal with upgrading and downgrading question either explicitly or implicitly. There is thus plenty of prior work for framing the questions examined.
However, much of the existing empirical work has literally been release-based, whereas this paper leans towards a deployment-based approach. In other words, much of the prior work is on the producer-side, while the consumer-side has received less attention [@Baysal12]. While bringing these two sides closer to each other remains a major research challenge, it is worthwhile to further note that the release-versus-deployment distinction applies also to studies examining the evolution of programming languages. In particular, there is some prior work examining PHP deployments seen in the wild [@Ruohonen16WIMS], but the evolution of the programming language itself—with its features and flaws—has received more attention [@Amanatidis16; @Hills15]. The same applies to PHP source code analysis, which has mostly concentrated on evaluating “off-the-shelf” PHP applications (e.g., [@Hills13; @Medeiros16]) without attempting to cover custom applications seen in the wild. Although source code analysis is not pursued in this paper, the deployments examined still cover many custom PHP applications. Due to such applications, better knowledge about adoption and patching on the consumer-side is valuable for those on the producer-side. In other words, it is important to consider continuous tracing of software deployments in order to improve the feedback loops required for sound continuous software engineering practices.
The remainder of the paper is structured into four sections. Section \[section: background\] motivates the research background in more detail and formulates the research questions for the empirical analysis. Section \[section: approach\] outlines the DTMC modeling approach. Results are presented in Section \[section: experimental results\] based on large longitudinal datasets compiled from a few third-party web crawling snapshots. Conclusions and discussion follow in Section \[section: discussion\].
Background {#section: background}
==========
The scholarly background can be motivated by considering the feedback channels that are essential for the contemporary continuous software engineering practices. After connecting these practices to the concept of software traceability, research questions are formulated in relation to the current release engineering strategy of the PHP programming language.
Motivation
----------
Continuous software engineering is an umbrella term covering multiple contemporary software engineering tools and methodologies, including but not limited to continuous planning, continuous budgeting, continuous integration, continuous delivery, continuous deployment, continuous testing, continuous evolution, continuous maintenance, continuous feedback, and, ultimately, continuous innovation . These overlapping continuous-prefixed concepts are also well-recognized in the release engineering research domain.
However, much of the existing research has concentrated on traditional software engineering aspects, such as integration, build systems, testing, and maintenance. This emphasis is reflected in the attempts to define the concept of release engineering. For instance, release engineering has been defined as “a software engineering discipline concerned with the development, implementation, and improvement of processes to deploy high-quality software reliably and predictably” [@Dyck15]. Although the word improvement appears in the definition, a little emphasis is placed on feedback from customers, investors, and other stakeholders, which is a fundamental element in the contemporary continuous software engineering practices and processes . By and large, these feedback mechanisms have constituted an enduring challenge for empirical software evolution research in general [@deOliveira16; @Ruohonen15JSEP]. This gap in the literature is noteworthy because the availability of information about releases has increased substantially in recent years.
Different “telemetry” solutions—including crash reports and other “call-home” features—are increasingly popular in many software industry segments. The availability of feedback data is not limited to features explicitly integrated into software, however. Social media, review and rating sites, and related elements of the contemporary world wide web provide a wealth of information for systematic tracing of releases. A major challenge for modern release engineering relates to integration of such data into meaningful solutions that help developers and stakeholders to make informed decisions about the evolution and patching of software deployments [@Adams16]. For summarizing this key challenge, Fig. \[fig: map\] depicts a relational map of a few interconnected continuous software engineering concepts. In this paper, the focus is at the lower-right corner, which is labeled as continuous tracing of the continuously engineered releases that are continuously deployed in the wild.
![A Terminological Map[]{data-label="fig: map"}](fig_continuous.pdf){width="\linewidth" height="3.5cm"}
The concept of continuous tracing can be linked to software traceability, which “refers to the ability to describe and follow the life of a requirement”, release, or other software artifact “in both a forwards and backwards direction” [@Gotel94]. This definition can be used to clarify and frame the scope of this paper.
By appending the word continuous to the traceability term, it is emphasized that tracing of software artifacts should be systematic and continuous throughout the life cycles of the artifacts. In this paper, the software engineering artifacts are releases of the PHP programming language, but the units of analysis are deployments using the releases for serving web pages. The tracing in forward and backward directions is done by observing upgrading (i.e., roll-forward), downgrading (i.e., roll-back or reverting), and release adoption (i.e., either upgrading or downgrading) patterns of PHP deployments.
PHP Releases and Research Questions {#subsec: research questions}
-----------------------------------
The first version of the PHP programming language was announced in 1995. The second, third, fourth, and fifth major versions followed in 1997, 1998, 2000, 2004, respectively. The sixth major version was branched for development in 2010. Instead of evolving into a production-ready major release branch, the controversies regarding Unicode support resulted in backporting of features from PHP 6 to the fifth major branch [@Sturgeon14]. Currently, most PHP deployments still run with PHP 5, while the head of development occurs in the PHP 7 branch, which is not compatible with the previous major branches due to numerous new language features. For the programming language developers involved in the project, it is relevant to know an answer to the following question (\[rq: php7\]).
1. [*How widespread has the transition been to PHP 7?*]{}\[rq: php7\]
The PHP project follows the semantic versioning strategy conveyed via the “*major.minor.maintenance*” versioning scheme. According to this versioning strategy, in essence, a major release should be reserved for incompatible changes to the application programming interfaces (APIs); a minor release should aggregate functionality enhancements that are backward-compatible; and, finally, maintenance releases should be reserved for small backward-compatible bug fixes and reliability improvements [@Raemaekers14; @Kula15]. These versioning principles have also guided the PHP release process since 2010 when a fixed release cycle was agreed upon. According to the current strategy, minor releases are scheduled to occur annually, whereas maintenance versions are released at least once a month [@PHP10]. Backward-compatibility and API stability are guaranteed within major branches. At the time of writing, the 5.6, 7.0, and 7.1 minor branches are still supported [@PHP17b], which is in accordance with the guarantee of three years of support (bug and security fixes) for each minor release.
{width="\linewidth" height="7.5cm"}
For a programming language project, the monthly cycle for maintenance releases is extremely rapid. On paper, this cycle is actually faster than those used for the development of many web browsers, such as Firefox [@Mantyla13]. As is visible from the illustration in Fig. \[fig: releases\], the strategy of monthly maintenance releases has also resulted a large amount of versions from circa 2012 onward. Given the rapid release cycle and the large amount of releases made in recent years, it is relevant to solicit an answer to the research question \[rq: prevalence\].
1. [*How prevalent has the adoption of PHP releases been?*]{}\[rq: prevalence\]
The question about release adoption includes both upgrading and downgrading of PHP versions. From a release engineering perspective, particularly interesting are cases whereby a deployment downgrades its PHP version. For instance: if a web site used a version `5.5.0` at some point in time but then later adopted a version `5.4.0`, perhaps there were difficulties in adopting the new release. If such downgrading is common, it might be worthwhile to revisit the supportive activities [@Mantyla11] associated with release engineering. Actual bug fixes notwithstanding, these activities include sufficient release notes, good and up-to-date documentation, user support, easy installation procedures, pre-install checks, sane defaults, migration instructions, and related release engineering aspects. Given this reasoning, the question (\[rq: downgrading\]) is worth asking.
1. [*How common is downgrading of PHP deployments?*]{}\[rq: downgrading\]
At a more abstract level of thought, it is relevant to know how consistent or uniform release adoption has generally been in recent history. By uniformity, it is meant that most deployments follow the semantic versioning strategy in their upgrades, moving within a major or minor branch in a relatively logical manner. When planning for new releases or supportive activities thereto, it is less relevant to try to support deployments that adopt releases in a chaotic manner. For instance: if a site upgraded from `5.4.0` to `7.1.1` but then moved to `5.5.0` while using a version `5.3.1` in-between, there are likely problems in the maintenance of the site, which cannot be addressed by the means of release engineering. If such chaotic patterns are widespread, on the other hand, it may be relevant to reconsider the appropriateness of a release strategy. Thus, the following question (\[rq: uniform\]) is justified.
1. [*How uniform has the adoption PHP releases been?*]{}\[rq: uniform\]
Finally, the following atheoretical assertion can be placed for controlling the answers to the research questions outlined.
1. [*Do the answer to \[rq: php7\], \[rq: prevalence\], \[rq: downgrading\], and \[rq: uniform\] vary between the recent short-run history and the long-run evolution?*]{}\[rq: short-run vs. long-run\]
The concepts of prevalence, uniformity, short-run, and long-run are further elaborated in the subsequent sections that introduce the Markov chain framework and the empirical data.
Approach {#section: approach}
========
A few remarks about the fundamental properties of DTMCs are required to outline the research approach. After these remarks, computation and operationalization are discussed.
DTMCs in Brief {#subsec: dtmcs in brief}
--------------
A first-order discrete time Markov chain is a finite sequence of random variables $X_1, X_2, \ldots, X_t, \ldots$, satisfying the fundamental Markov property according to which the probability distribution of a forthcoming $X_{t+1}$ depends on the immediately preceding $X_t$ but not on $X_{t-1}, X_{t-2}, \ldots, X_1$. If $S = \lbrace s_1, \ldots, s_n \rbrace$ denotes a set of all possible values of the random variables, the Markov property implies that the probability of moving to a next state in the state space $S$ is $$\begin{aligned}
\label{eq: Markov property}
\Pr&(X_{t+1} = s_{t+1} \vert~X_1 = s_1, X_2 = s_2, \ldots, X_t = s_t)
\\ \notag
&= \Pr(X_{t+1} = s_{t+1}~\vert~X_t = s_t) .
$$
This first-order Markov property implies a “memoryless” model, meaning that predicting a future state depends only on the current state. In addition to (a) assuming that holds, (b) the chains observed are assumed to be time-homogeneous. The latter condition means that a transition probability $$\label{eq: transition probability}
p_{ij} = \Pr(X_{t+1} = s_j ~\vert~ X_t = s_i)$$ from a state $s_i \in S$ to state $s_j \in S$ is independent from $t$, $$\Pr(X_{t+1} = s_j ~\vert~ X_t = s_i)
= \Pr(X_{t} = s_j ~\vert~ X_{t-1} = s_i) .$$
In other words, the transition probabilities do not change as time passes. This assumption can be further accompanied by emphasizing that (c) only discrete chains are considered without explicit linkage to continuous calendar-time. This further restriction implies that the transition probability in does not depend on the calendar-time lag between $s_j$ and $s_i$, irrespective whether the lag is measured in months or years.
Finally, (d) the state changes associated with two distinct (exogenous) sequences, $X_1, X_2, \ldots, $ and $Y_1, Y_2, \ldots$, are assumed to be independent from each other, such that $$\begin{aligned}
\label{eq: exogenous}
\Pr(&X_{t+1} = s_j ~\vert~ X_t = s_i, Y_1 = s_1, \ldots, Y_t = s_k) ,
\\ \notag
&= \Pr(X_{t+1} = s_j ~\vert~ X_t = s_i) .\end{aligned}$$ In other words, the cross-sectional empirical analysis is conducted without considering any potential dependencies between individual sequences and their state changes.
Computation
-----------
The empirical setup is based on a sample of $m$ domains: $$\begin{aligned}
\label{eq: sequences}
X^{(1)}_1, &~X^{(1)}_2, \ldots, X^{(1)}_{r_1}
\\ \notag
&\vdots
\\ \notag
X^{(m)}_1, &~ X^{(m)}_2, \ldots, X^{(m)}_{r_m} \end{aligned}$$ that are exogenous with respect to each other, such that holds for any pair of sequences and their state changes.
Due to practical reasons stemming from data collection, the length of the sequences and state changes are both allowed to vary across domains. For instance the length of the sequence for the $k$:th domain, denoted by $r_k$, may differ from another sequence length $r_{k+1}$. These varying sequence lengths correspond with the times each domain is observed empirically.
As described later in Section \[subsec: snapshots\], the maximum sequence lengths are 14 and 6 for all domains in the short-run and long-run examinations, respectively. In addition, a constraint $r_k \geq 2$ is imposed for all $m$ domains to ensure that state changes are possible to begin with. Even when the $k$:th domain is observed fourteen times, however, the length of the state space may equal one in case the domain in question never changed the PHP version of its deployment. In contrast, the maximum value $r_k$ for $\vert S_k \vert$ is attained by a domain that has changed its PHP version each time the domain is observed.
The transition probabilities are estimated by $$\label{eq: MLE}
\hat{p}^{(k)}_{ij} =
\begin{cases}
0 & \textmd{if}~f^{(k)}_{i.} = 0, \\
f^{(k)}_{ij}~/~f^{(k)}_{i.} & \textmd{if}~f^{(k)}_{i.} \neq 0 ,
\end{cases}$$ where $f^{(k)}_{ij}$ denotes the frequency of $(X_t = s_i, X_{t+1} = s_j)$ PHP version sequences for the $k$:th domain and $$f^{(k)}_{i.} = \sum^{\vert S_k \vert}_{j=1} f^{(k)}_{ij} .$$
The special case $f^{(k)}_{i.} = 0$ occurs when the last observed state denotes a previously unseen PHP version, meaning that there is not enough data to estimate the transition probability for this state. This additional, context-specific alteration notwithstanding, the equation conveys a conventional maximum likelihood estimator (MLE) for a transition probability from the $i$:th to the $j$:th state [@Singer14; @Hill04]. While a small custom implementation is used for the MLE computations, the results were further verified with an existing R implementation [@markovchain].
![An Example DTMC for PHP Release Adoption[]{data-label="fig: example 1"}](fig_example_1.pdf){width="7cm" height="5cm"}
To illustrate the computation in practice, consider the example in Fig. \[fig: example 1\]. A full sequence of fourteen observations is available for this domain, while the state space contains three unique PHP versions. The first state corresponds with the PHP version `5.4.39`. Because the domain used the same deployment also during the three subsequent observations, the probability of upgrading from this version to `5.6.20` was $1/4 = 0.25$. The probability of subsequently upgrading the deployment from `5.6.20` to `5.6.28` is even lower, given the seven times the domain `www.vraymaterials.co.uk` stayed with its `5.6.20` deployment. Thus, the prevalence of release adoption has been modest for this particular domain in the short-run. For evaluating the prevalence among hundreds of thousands of web sites, a few custom metrics can be derived.
Metrics {#subsec: metrics}
-------
The research question about prevalence (\[rq: prevalence\]) can be answered with a metric based on the estimated transition probabilities. Thus, let ${\textup{\textbf{\textrm{P}}}}_k$ denote a $\vert S_k \vert \times \vert S_k \vert$ matrix of estimated transition probabilities for the $k$:th domain. For instance, the $3 \times 3$ transition probability matrix underneath the illustration in Fig. \[fig: example 1\] is defined by $$\underbrace{\left.\left[\begin{array}{ccc}
0.75 & 0.25 & 0.00 \\
0.00 & 0.86 & 0.14 \\
0.00 & 0.00 & 1.00
\end{array}\right]\right\}}_{\texttt{5.4.39}~~\texttt{5.6.20}~~\texttt{5.6.28}~~}
\begin{array}{c}
\scriptstyle\texttt{5.4.39} , \\
\scriptstyle\texttt{5.6.20} , \\
\scriptstyle\texttt{5.6.28} , \\
\end{array}$$ which can be read as an adjacency matrix for a weighted and directed graph. The trace of this matrix (that is, the sum of the diagonal elements) provides a simple measure for the persistence of a DTMC phenomenon [@Hill04]. For answering to the first research question \[rq: prevalence\], this simple but powerful idea allows to operationalize the concept of prevalence with $$\label{eq: prevalence}
\delta_k
= \frac{1}{\vert S_k \vert}\sum^{\vert S_k \vert}_{i=1}
\left(1 - \hat{p}^{(k)}_{ii}\right),$$ where $\delta_k \in [0, 1]$ for all $k$. In other words, the closer a $\delta_k$ is to unity, the more prevalent has the adoption of releases been. If $\delta_k = 0$, the $k$:th web site never changed its PHP deployment. Collecting the scalars to a vector ${{\boldsymbol{\delta}}}= [\delta_1, \delta_2, \ldots, \delta_m]$ allows evaluating the prevalence among the $m$ domains observed.
Answering to the research questions \[rq: downgrading\] and \[rq: uniform\] is better done with the version sequences in rather than with the transition probabilities within the state spaces. Thus, for evaluating how uniform PHP release adoption has generally been among the $m$ domains observed (\[rq: uniform\]), a simple metric is available by counting the unique version sequences, scaling the resulting amount by $m$. Although this metric approaches zero as $m \to \infty$, it still gives a good overall sense about the uniformity of typical PHP release adoption patterns.
Although calendar-time records can be used for comparing release orderings [@Kula15], a metric for downgrading (\[rq: downgrading\]) can be also computed directly from the PHP version sequences. For all domains with $\vert S_k \vert \geq 2$, downgrading can occur via three different scenarios: (a) when $\textit{major}_{i+1} < \textit{major}_i$, that is, when the major version number of a current deployment is larger than the major version number of a subsequent deployment; (b) when $\textit{major}_{i+1} = \textit{major}_{i}$ but $\textit{minor}_{i+1} < \textit{minor}_{i}$; or (c) when both the major and minor version numbers remain the same but the maintenance version number of the $i$:th state is larger than the number of the subsequent state. Given these three distinct cases, all $m$ version sequences are processed by comparing $(r_k - 1)$ times the $i$:th version to the $(i + 1)$:th version, recording the number of downgrades at each step. If $d_k$ denotes the number of downgrades recorded for the $k$:th domain, a vector ${{\boldsymbol{\phi}}}= [ d_1~/~r_1 - 1, \ldots, d_m ~/~ r_m - 1]$ defines a simple metric for evaluating how common PHP downgrading has generally been. Analogous to , values close to unity indicate frequent downgrading. In theory, also different weights could be used for the three different downgrading scenarios, but this simple counting scheme is sufficient because downgrading should be relatively rare in the context of popular web sites.
The transition matrices ${\textup{\textbf{\textrm{P}}}}_1, \ldots, {\textup{\textbf{\textrm{P}}}}_m$ offer another viewpoint to downgrading: whenever states $s_i$ and $s_j$ communicate (such that there is a transition from $s_i$ to $s_j$ and from $s_j$ to $s_i$), there is also downgrading of PHP versions. Given this reasoning, a further metric can be computed by counting the number of communicating state pairs and scaling the result appropriately: $$\label{eq: communicating pairs}
\gamma_k =
\frac{1}{r_k - 1}
\sum^{\vert S_k \vert}_{i=1}\sum^{\vert S_k \vert}_{j=i}
\operatorname{I}\left(\hat{p}^{(k)}_{ij}\right)
\operatorname{I}\left(\hat{p}^{(k)}_{ji}\right)$$ where $\operatorname{I}(\cdot)$ is an indicator function outputting $$\operatorname{I}(x) =
\begin{cases}
0 & \textmd{if}~x = 0, \\
1 & \textmd{if}~x > 0.
\end{cases}$$
![Another Example DTMC for PHP Release Adoption[]{data-label="fig: example 2"}](fig_example_2.pdf){width="7cm" height="5cm"}
As an example, consider the quite messy real-world release adoption pattern visualized in Fig. \[fig: example 2\]. This particular domain downgraded its PHP deployment as many as seven times during 2016 and early 2017. Therefore, $\phi_k = 7~/~(14 - 1) \simeq 0.54$ and $\gamma_k = 4~/~13 \simeq 0.31$, given the four communicating pairs. Both values are rather high, which indicates that the two downgrading metrics can be used also for probing outlying domains that may have problems with maintenance of their PHP deployments.
Experimental Results {#section: experimental results}
====================
The empirical results are disseminated by first introducing the data used for the DTMC computation. The metrics elaborated in the previous Section \[subsec: metrics\] are subsequently used for summarizing the empirical findings.
Data
----
Two datasets are used for the empirical analysis: one for observing short-run release adoption and the other for proxying the long-run evolution of PHP deployments seen in the wild. While calendar-time is not explicitly observed with the DTMCs computed, the definitions for short-run and long-run are still based on calendar-time: with one exception, the short-run dataset covers a period from January 2016 to March 2017 under a monthly sampling frequency, while the long-run dataset is based on annual records in a period between January 2012 and February 2017. Given the PHP release lineage illustrated in Fig. \[fig: releases\], a truly long-run analysis should start already from the year 2000—or even earlier, but the historical periods used are imposed by the source of empirical data. This data source should be also elaborated in more detail.
### Snapshots {#subsec: snapshots}
Both datasets are compiled from a few large web crawling snapshots that contain data on hypertext transfer protocol (HTTP) headers used for identifying PHP versions. These open data snapshots are provided by the HTTP Archive web crawling project [@HTTPArchive17a], which has also been used in previous research [@Ruohonen16WIMS] alongside analogous archives [@Wambach16; @Ainsworth13]. In total, fourteen crawling snapshots are used for compiling the short-run dataset (see Table \[tab: short-run data\]). Due to data availability issues, the long-run dataset is compiled only from six snapshots (see Table \[tab: long-run data\]), the earliest of which dates to January 2012.
----- ------------------- ------------ --------------- -----------------
Start of crawling User-agent Size (GB)$^b$ PHP domains$^c$
1. March 1, 2017 Chrome $\simeq 47$ 206,739
2. February 1, 2017 Chrome $\simeq 45$ 202,772
3. December 2, 2016 Chrome $\simeq 51$ 216,362
4. November 1, 2016 Chrome $\simeq 49$ 220,871
5. October 1, 2016 Chrome $\simeq 51$ 222,645
7. September 1, 2016 Chrome $\simeq 49$ 222,051
7. August 1, 2016 Chrome $\simeq 48$ 227,757
8. July 1, 2016 Chrome $\simeq 48$ 225,623
9. June 1, 2016 Chrome $\simeq 56$ 226,797
10. May 1, 2016 Chrome $\simeq 52$ 217,592
11. April 1, 2016 Chrome $\simeq 50$ 219,564
12. March 1, 2016 Chrome $\simeq 57$ 230,028
13. February 1, 2016 Chrome $\simeq 52$ 226,859
14. January 1, 2016 Chrome $\simeq 50$ 225,362
----- ------------------- ------------ --------------- -----------------
: Characteristics of the Short-Run Dataset [@HTTPArchive17a]$^a$[]{data-label="tab: short-run data"}
[$^a$ Note that the January 1 snapshot from 2017 was empty and had to be thus excluded. Due to this omission, there is a two month calendar-time delay between the second and the third snapshot. $^b$ The size refers to the unpacked snapshots. $^c$ See Section \[subsec: pre-processing\] for a definition of a “PHP domain”.]{}
---- ------------------- ------------ --------------- -----------------
Start of crawling User-agent Size (GB)$^b$ PHP domains$^c$
1. February 1, 2017 Chrome $\simeq 45$ 202,772
2. January 1, 2016 Chrome $\simeq 50$ 225,362
3. January 1, 2015 IE$^d$ $\simeq 41$ 247,568
4. January 1, 2014 IE$^d$ $\simeq 24$ 158,775
5. January 1, 2013 IE$^d$ $\simeq 22$ 165,741
6. January 1, 2012 IE$^d$ $\simeq 3.7$ 31,445
---- ------------------- ------------ --------------- -----------------
: Characteristics of the Long-Run Dataset [@HTTPArchive17a]$^a$[]{data-label="tab: long-run data"}
[$^{a,b,c}$ See the notes in Table \[tab: short-run data\]. $^d$ The abbreviation stands for Internet Explorer.]{}
As shown in the two tables, the raw snapshots used are quite large. Because the crawls are seeded from Alexa’s list of top-million busiest web sites [@HTTPArchive17b], which is updated daily, the snapshot sizes also vary from a crawl to another. Moreover, it should be emphasized that the dates shown are only tentative regarding individual HTTP requests and responses: due to the large seeding list, crawling can take a relatively long amount of time [@Ainsworth13]. Already because calendar-time is not explicitly observed, the issue is a minor concern for this paper, however.
The long-run dataset is affected by a change in the forged user-agent [@Calzarossa14] used for making the requests. Although user-agents can have a substantial empirical effect for measuring web sites due to specific responses for specific browsers [@Pham16], the consequences should be small in this paper because it seems unlikely that PHP version strings in the HTTP response headers would vary according to a user-agent specified in the HTTP request headers. Therefore, it is more important to further remark that the long-run dataset is affected by changes made to the seeding of the web crawls, which is reflected in the smaller amount of PHP domains between 2012 and 2014. In contrast, each snapshot in the short-run dataset contains roughly the same amount of domains. Finally, it should be emphasized that the total amount of PHP-powered domains observed is substantially larger than reported in Tables \[tab: short-run data\] and \[tab: long-run data\] because the snapshots are “pooled” to include all domains that are present in at least two snapshots.
### Pre-processing {#subsec: pre-processing}
The snapshots were pre-processed from the packaged archives delivered as CSV (comma-seperated value) files. Although the files are provided as open data, a couple of remarks should be made to ensure replicability of the datasets. First and foremost, the presence of PHP is identified via s simple (Python) regular expression of the following form: “`PHP/[0-9]{1}\.[0-9]{1}\.[0-9]{1,}`”, where the quotation marks are not part of the expression. Notice that the expression excludes “invalid” versions such as `PHP/3.100`.
Second, unique domains are identified by extracting the network location from the uniform resource locators (URLs) crawled. Because multiple web pages may be crawled for each domain, duplicates are excluded by omitting the parsing of URLs for domains that have already been identified to run with PHP. It should be remarked that the concept of domain is inexplicit in the sense that no attempts are made to lookup the domains via the domain name system. Therefore, in theory, the domains may refer to actual domain names as well as Internet protocol addresses. For the purposes of this paper, the distinction is irrelevant, however.
Results
-------
The dissemination of the results can be started by noting a few characteristics of the two datasets compiled from the web crawling snapshots. First and foremost, according to the numbers shown in Table \[tab: lengths\], about 451 and 220 hundred thousand domains were identified as running with PHP according to the simple pre-processing routines. By implication, well over half a million transition probabilities were estimated via . Second, on average, about a half of the maximum lengths of the version sequences are realized in the two datasets, although the standard deviations are large. In other words, a typical domain is observed a little over six times in the short-run and about three times in the long-run. The reason for not reaching the maximum lengths is simple: because the snapshots are not crawled from a fixed domain set, not all of the PHP domains observed are present across all snapshots. Third, the average size of the state space, $\frac{1}{m}\sum^m_{k=1} \vert S_k \vert$, is less than two in both datasets. Thus, for many domains, the transition probabilities are represented by the value one supplied via a $1 \times 1$ matrix. While there is still a sufficient amount of variance for analysis, already this observation allows to conclude that the prevalence of PHP release adoption has been at a modest level. Before continuing to the actual prevalence metric, a remark should be made about the most common PHP releases in the datasets.
--------------------------------------- ----------- ----------- ----------
Short-run Long-run
Number of domains ($m$) 451,340 220,293
Number of versions ($r$) Mean 6.5 2.9
Std. dev. 4.0 1.1
Size of state space ($\vert S \vert$) Mean 1.5 1.8
Std. dev. 1.2 0.8
--------------------------------------- ----------- ----------- ----------
: Sample Characteristics[]{data-label="tab: lengths"}
------------------------------- ----------------------- ----------- ----------
Subset Short-run Long-run
Number of unique sequences $\vert S_k \vert = 1$ 2,470 568
$\vert S_k \vert > 1$ 62,376 43,585
Share of unique sequences (%) $\vert S_k \vert = 1$ 0.55 0.26
$\vert S_k \vert > 1$ 13.8 19.8
------------------------------- ----------------------- ----------- ----------
: Uniform PHP Version Sequences[]{data-label="tab: uniform sequences"}
![Most Frequent PHP Releases in the Long-Run[]{data-label="fig: long-run releases"}](fig_releases_longrun.pdf){width="\linewidth" height="4cm"}
![Most Frequent PHP Releases in the Short-Run[]{data-label="fig: short-run releases"}](fig_releases_shortrun.pdf){width="\linewidth" height="4cm"}
![Prevalence of PHP Release Adoption[]{data-label="fig: prevalence"}](fig_prevalence.pdf){width="\linewidth" height="6cm"}
![Downgrading of PHP Releases \#1[]{data-label="fig: downgrading"}](fig_downgrading.pdf){width="\linewidth" height="6cm"}
![Downgrading of PHP Releases \#2[]{data-label="fig: communicating"}](fig_communicating.pdf){width="\linewidth" height="6cm"}
According to the datasets, it is clear that the PHP 5 release branch has been the most popular deployment choice from the early 2000s onward. As shown in Fig. \[fig: long-run releases\], the clear majority of the versions observed in the long-run are part of the PHP 5 branch. While there has also been few domains using the PHP 4 branch, the use of other major branches has been negligible in the long-run. The fifth major version has also retained its popularity in 2016 and 2017, as can be concluded from the subsequent Fig. \[fig: short-run releases\]. Adoption of PHP 7 has been modest (\[rq: php7\]). This conclusion does not change when only the last versions in the sequences are used to proxy popularity; in this case, only about 1.7 % of the domains have used a PHP 7 deployment in the short-run
The observation about relatively infrequent release adoption is reinforced by Fig. \[fig: prevalence\], which shows a histogram of the metric in across the domains observed in the two datasets. There exists a difference between short-run release adoption and long-run evolution, however. The upper plot clearly indicates that monthly release adoption has generally been infrequent in 2016 and early 2017. In contrast, the lower plot displays a bimodal distribution: many domains observed have not adopted releases in the long-run, but almost an equal amount of domains have adopted releases annually. All in all, the prevalence of PHP release adoption has been modest in recent years (\[rq: prevalence\]), although less so in the long-run (\[rq: short-run vs. long-run\]).
Moreover, PHP release adoption patterns have been relatively uniform across the domains observed (\[rq: uniform\]), as can be concluded from the summary shown in Table \[tab: uniform sequences\]. In total, only about $0.55 + 13.8 \simeq 14.4$ and $20.1$ percent of the observed PHP version sequences are unique in the short-run and long-run datasets, respectively. While it should be kept in mind that these relative amounts are affected by the large amount of domains observed, these relative amounts still indicate a modest amount of unique release adoption patterns. Most of the patterns in both datasets describe common transition paths within the PHP 5 major release branch. From a release engineering perspective, this observation is a positive finding: most domains follow other domains in their upgrading patterns. Disorderly release adoption patterns are relatively rare.
The observations about infrequent release adoption and uniformity are reinforced by Fig. \[fig: downgrading\], which shows the frequency of the first downgrading metric in the two datasets. Most domains have not downgraded their PHP deployments even once. Although downgrading is slightly more common in the long-run, the standard deviations are generally small. By implication, the same observation applies also for the results regarding the second downgrading metric in . As can be concluded from Fig. \[fig: communicating\], only a very few of the version sequences involve communicating PHP version pairs. The averages and standard deviations are both negligible. To summarize, downgrading has been rare (\[rq: downgrading\]), and there are no notable differences between short-run and long-run (\[rq: short-run vs. long-run\]).
Discussion {#section: discussion}
==========
The remainder of this paper first summarizes the main empirical findings, then enumerates a few threats to validity, and finally concludes with a couple of new research directions.
Summary of Results
------------------
This empirical paper observed PHP release adoption in two datasets covering over a half a million Internet domains and three million PHP versions deployed within these domains. The main findings can be summarized by briefly answering to the five research questions outlined in Section \[subsec: research questions\].
- [Adoption of PHP 7 has been modest (\[rq: php7\]). As of early 2017, only few popular web sites have adopted the new major release branch. Most sites continue to operate with releases made within the PHP 5 major branch.]{}
- [The prevalence of PHP release adoption has been at a modest level: popular web sites tend to upgrade their deployments relatively infrequently (\[rq: prevalence\]). The observation aligns with previous studies; relatively old PHP 5 versions are commonly used in cloud computing services [@WangNappa14].]{}
- [Downgrading has been uncommon; only a few outlying domains have downgraded their deployments (\[rq: downgrading\]).]{}
- [The adoption patterns have been highly uniform across popular domains; most domains tend to follow similar upgrading paths used also by other domains (\[rq: uniform\]).]{}
- [Only the prevalence of adoption (\[rq: prevalence\]) varies between the short-run history (2016 – early 2017) and long-run evolution (2012 – early 2017). Namely, the longer the period observed, the more common has adoption been.]{}
These findings provide also some material for contemplating about the current release engineering strategy of the PHP project. For the developers of the programming language, a pressing question relates to the means by which the currently slow adoption of PHP 7 could be boosted in the future. One option to consider might be a Firefox-style rapid release strategy, which has been suspected to increase user adoption compared to a traditional release strategy [@daCosta16]. Because adoption has generally been infrequent among popular web sites, it might be possible to also debate whether the current release schedule is actually already too rapid for users and stakeholders. Although downgrading is rare and the adoption patterns are generally uniform, a further interesting question relates to the reasons why some domains downgrade, and whether there is anything that could be done to help outlying domains following chaotic release adoption paths.
Threats to Validity
-------------------
Threats to validity can be enumerated by using the conventional threefold classification of construct validity, external validity, and internal validity. Although there exists no uniformly agreed definitions [@Siegmund15], for the purposes of this paper, these three validity concepts can be equated to questions related to generalizability (how results generalize to a different context or population), operationalization (how well a quantification matches a theoretical concept), and systematic computational errors (how well different biases are eliminated), respectively.
### External Validity
Generalizability questions are always present when the theoretical population is the whole world wide web [@Ainsworth13], including the so-called “deep web” not indexed by standard search engines. Even though generalizability toward such a population is practically impossible even for companies such as Google, it is possible to narrow the target of generalization toward a sub-population of the most popular PHP-powered domains. In this regard, HTTP Archive uses Alexa’s popularity list, which is commonly perceived as a good choice for seeding of large-scale web crawling . While external validity is presumably not threatened in this regard, (a) the results reported are likely specific to popular web sites. When considering further applications, such as those motivated by security questions [@Medeiros16], it is likely that more interesting cases are located in the fringes of the world wide web. Given that prevalence of PHP release adoption was observed to be at a modest level in a sample of popular sites, it is more than likely that even lower levels of adoption could be observed in a sample covering WordPress deployments, for instance. A common limitation [@He13] is also present: (b) the results apply only to domains using PHP for serving pages via plain HTTP, excluding sites using HTTPS.
### Construct Validity
A notable threat to construct validity stems from the identification of PHP deployments via a regular expression from HTTP response headers (see Section \[subsec: pre-processing\]). This coarse identification technique is likely to include both false positives (popular domains incorrectly identified as running with PHP) and false negatives (the missing of popular PHP-powered domains). To evaluate the severity of this limitation, at minimum, parallel identification should be attempted from the actual web page content (cf. [@WangNappa14]). Because the primary identification requirement relates to the version of a PHP backend used for serving a particular web content, robust identification is likely challenging also from web page contents, however. Further research is therefore required to continue the work on identifying and fingerprinting PHP applications [@Kozina09], including the PHP interpreter itself.
### Internal Validity
The potential presence of systematic biases is best evaluated against the classical DTMC assumptions that were imposed for the statistical computation (see Section \[subsec: dtmcs in brief\]). There are three notable concerns about these assumptions. First, the assumption in Eq. implies that regardless whether a state change is due to security updates, reliability improvements, or new features, it is always the currently deployed version that defines the reference point for the change, regardless whether the decision to change versions is made by a human or a package manager. While the assumption seems sensible from a release engineering viewpoint, it is easily questioned from a software evolution perspective [@WongCai11; @Ruohonen15JSEP]. If history matters also for PHP release adoption patterns, it would seem reasonable to recommend that further research should focus on higher-order Markov chains that have a memory . The second concern relates to the assumption of independence between domains. Given that a substantial amount of contemporary web sites require connections to two or more servers , PHP deployments may be uniformly managed and upgraded in a cloud computing service or other large deployment farm. Consequently, a PHP version sequence of a domain might be affected by a sequence of another domain. Conditional Markov chains [@Ching08; @Goutte14] may provide a useful tool for evaluating the potential severity of this cross-domain dependence assumption.
The third notable threat to internal validity relates to the PHP version sequences observed, which mandate making an addition assumption about the transition probabilities in . Consequently, by definition [@Seber08], the transition probability matrices computed are not stochastic matrices, that is, the row sums of these matrices do not necessarily equal one. Although this unavoidable limitation does not affect the results reported as such, it does affect additional computations involving eigenvalues [@Hill04], and particularly the stationary distributions toward which all irreducible, aperiodic, and positive recurrent Markov chains converge (for the mathematical background see [@Privault13]). This point should be kept in mind when considering further DTMC applications in the release engineering and software evolution contexts. Such applications are also a good way to point out a couple of new research directions.
Further Work
------------
The primary purpose of this paper was to examine the usefulness of DTMC modeling for systematic tracing of web deployments in order to establish automated continuous feedback channel for server-side programming language developers and stakeholders. The paper fulfilled this goal: DTMCs are useful also in the release engineering context. For pursuing DTMC analysis further, a worthwhile goal would be to translate some of the concepts used in other disciplines to the language of release engineering and software evolution. For instance, simple DTMC metrics have been used to proxy such concepts as colonization, disturbance, and replacement [@Hill04]. With some theoretical and terminological alterations, such metrics and concepts could be adopted for pursuing DTMC modeling further in the release engineering context. While these concepts and metrics are directly applicable to traditional DTMCs, another prolific path forward involves altering the basic assumptions surrounding discrete time-homogeneous Markov chains. Conditional and higher-order chains are good examples in this regard. For continuous tracing of PHP deployments, continuous Markov chains (as opposed to discrete-time chains) seem prolific to consider in further research. For instance, different time-delay models [@Wang12] could be adopted for studying the time delays between successive state changes. The question about time delays is also fundamental in the release engineering context because the empirical transition probabilities depend on the sampling frequency used.
Acknowledgements {#acknowledgements .unnumbered}
================
The authors gratefully acknowledge Tekes – the Finnish Funding Agency for Innovation, DIMECC Oy, and the Cyber Trust research program for their support.
[10]{} \[1\][\#1]{} url@samestyle \[2\][\#2]{} \[2\][[l@\#1=l@\#1\#2]{}]{}
J.-M. Favre, “[L]{}anguages [E]{}volve [T]{}oo! [C]{}hanging the [S]{}oftware [T]{}ime [S]{}cale,” in *Proceedings of the Eighth International Workshop on Principles of Software Evolution (IWPSE 2005)*.1em plus 0.5em minus 0.4emLisbon: IEEE, 2005, pp. 33–42.
L. A. Meyerovich and A. S. Rabkin, “[S]{}ocio-[PLT]{}: [P]{}rinciples for [P]{}rogramming [L]{}anguage [A]{}doption,” in *Proceedings of the ACM International Symposium on New Ideas, New Paradigms, and Reflections on Programming and Software*.1em plus 0.5em minus 0.4emTucson: ACM, 2012, pp. 39–54.
R. C. Cheung, “[A]{} [U]{}ser-[O]{}riented [S]{}oftware [R]{}eliability [M]{}odel,” *IEEE Transactions on Software Engineering*, vol. SE-6, no. 2, pp. 118–125, 1980.
W. Wang, “[A]{}n [O]{}verview of the [R]{}ecent [A]{}dvances in [D]{}elay-[T]{}ime-[B]{}ased [M]{}aintenance [M]{}odeling,” *Reliability Engineering and System Safety*, vol. 106, pp. 165–178, 2012.
S. Wong and Y. Cai, “[G]{}eneralizing [E]{}volutionary [C]{}oupling with [S]{}tochastic [D]{}ependencies,” in *Proceedings of the 26th IEEE/ACM International Conference on Automated Software Engineering (ASE 2011)*. 1em plus 0.5em minus 0.4emLawrence: IEEE, 2011, pp. 293–302.
S. Raemaekers, A. [van Deursen]{}, and J. Visser, “[S]{}emantic [V]{}ersioning versus [B]{}reaking [C]{}hanges: [A]{} [S]{}tudy of the [M]{}aven [R]{}epository,” in *Proceedings of the IEEE 14th International Working Conference on Source Code Analysis and Manipulation (SCAM 2014)*.1em plus 0.5em minus 0.4emVictoria: IEEE, 2014, pp. 215–224.
R. G. Kula, D. M. German, T. Ishio, and K. Inoue, “[T]{}rusting a [L]{}ibrary: [A]{} [S]{}tudy of the [L]{}atency to [A]{}dopt the [L]{}atest [M]{}aven [R]{}elease,” in *Proceedings of the IEEE 22nd International Conference on Software Analysis, Evolution, and Reengineering (SANER 2015)*, Montreal, 2015, pp. 520–524.
T. Karvonen, W. Behutiye, M. Oivo, and P. Kuvaja, “[S]{}ystematic [L]{}iterature [R]{}eview on the [I]{}mpacts of [A]{}gile [R]{}elease [E]{}ngineering [P]{}ractices,” *Information and Software Technology*, vol. 86, pp. 87–100, 2017.
M. V. Mäntylä, F. Khomh, B. Adams, E. Engström, and K. Petersen, “[O]{}n [R]{}apid [R]{}eleases and [S]{}oftware [T]{}esting,” in *Proceedings of the IEEE International Conference on Software Maintenance (ICSME 2013)*. 1em plus 0.5em minus 0.4emMadrid: IEEE, 2013, pp. 20–29.
D. A. [da Costa]{}, S. McIntosh, U. Kulesza, and A. E. Hassan, “[T]{}he [I]{}mpact of [S]{}witching to a [R]{}apid [R]{}elease [C]{}ycle on the [I]{}ntegration [D]{}elay of [A]{}ddressed [I]{}ssues: [A]{}n [E]{}mpirical [S]{}tudy of the [M]{}ozilla [F]{}irefox [P]{}roject,” in *Proceedings of the 13th International Conference on Mining Software Repositories (MSR 2016)*.1em plus 0.5em minus 0.4emAustin: ACM, 2016, pp. 374–385.
B. Adams and S. McIntosh, “[M]{}odern [R]{}elease [E]{}ngineering in a [N]{}utshell – [W]{}hy [R]{}esearchers [S]{}hould [C]{}are,” in *Proceedings IEEE 23rd International Conference on Software Analysis, Evolution, and Reengineering (SANER 2016)*.1em plus 0.5em minus 0.4emOsaka: IEEE, 2016, pp. 78–90.
O. Baysal, R. Holmes, and M. W. Godfrey, “[M]{}ining [U]{}sage [D]{}ata and [D]{}evelopment [A]{}rtifacts,” in *Proceedings of the 9th IEEE Working Conference on Mining Software Repositories (MSR 2012)*.1em plus 0.5em minus 0.4emZurich: IEEE, 2012, pp. 98–107.
J. Ruohonen, S. Hyrynsalmi, and V. Leppänen, “[E]{}xploring the [U]{}se of [D]{}eprecated [PHP]{} [R]{}eleases in the [W]{}ild [I]{}nternet: [S]{}till a [LAMP]{} [I]{}ssue?” in *Proceedings of the 6th International Conference on Web Intelligence, Mining and Semantics (WIMS 2016)*.1em plus 0.5em minus 0.4emNîmes: ACM, 2016, pp. .
T. Amanatidis and A. Chatzigeorgiou, “[S]{}tudying the [E]{}volution of [PHP]{} [W]{}eb [A]{}pplications,” *Information and Software Technology*, vol. 72, pp. 48–67, 2016.
M. Hills, “[E]{}volution of [D]{}ynamic [F]{}eature [U]{}sage in [PHP]{},” in *Proceedings of the IEEE 22nd International Conference on Software Analysis, Evolution, and Reengineering (SANER 2015)*.1em plus 0.5em minus 0.4emMontreal: IEEE, 2015, pp. 525–529.
M. Hills, P. Klint, and J. Vinju, “[A]{}n [E]{}mpirical [S]{}tudy of [PHP]{} [F]{}eature [U]{}sage: [A]{} [S]{}tatic [A]{}nalysis [P]{}erspective,” in *Proceedings of the International Symposium on Software Testing and Analysis (ISSTA 2013)*. 1em plus 0.5em minus 0.4emLugano: ACM, 2013, pp. 325–335.
I. Medeiros, N. Neves, and M. Correia, “[D]{}etecting and [R]{}emoving [W]{}eb [A]{}pplication [V]{}ulnerabilities with [S]{}tatic [A]{}nalysis and [D]{}ata [M]{}ining,” *IEEE Transactions on Reliability*, vol. 65, no. 1, pp. 54–69, 2016.
B. Fitzgerald and K. Stol, “[C]{}ontinuous [S]{}oftware [E]{}ngineering: [A]{} [R]{}oadmap and [A]{}genda,” *Journal of Systems and Software*, vol. 123, pp. , 2015.
C. Pang and A. Hindle, “[C]{}ontinuous [M]{}aintenance,” in *Proceedings of the IEEE International Conference on Software Maintenance and Evolution (ICSME 2016)*.1em plus 0.5em minus 0.4emRaleigh: IEEE, 2016, pp. 458–462.
A. Dyck, R. Penners, and H. Lichter, “[T]{}owards [D]{}efinitions for [R]{}elease [E]{}ngineering and [D]{}ev[O]{}ps,” in *Proceedings of the Third International Workshop on Release Engineering (RELENG 2015)*.1em plus 0.5em minus 0.4emFlorence: IEEE, 2015, pp. 3–3.
M. Leppänen, S. Mäkinen, M. Pagels, V.-P. Eloranta, J. Itkonen, M. V. Mäntylä, and T. Männistö, “[T]{}he [H]{}ighways and [C]{}ountry [R]{}oads to [C]{}ontinuous [D]{}eployment,” *IEEE Software*, vol. 32, no. 2, pp. 64–72, 2015.
R. P. [de Oliveira]{}, A. R. Santos, E. S. [de Almeida]{}, and G. S. da Silva Gomes, “[E]{}valuating [L]{}ehman’s [L]{}aws of [S]{}oftware [E]{}volution [W]{}ithin [S]{}oftware [P]{}roduct [L]{}ines [I]{}ndustrial [P]{}rojects,” *Journal of Systems and Software*, vol. 131, pp. 347–365, 2016.
J. Ruohonen, S. Hyrynsalmi, and V. Leppänen, “[T]{}ime [S]{}eries [T]{}rends in [S]{}oftware [E]{}volution,” *Journal of Software: Evolution and Process*, vol. 27, no. 2, pp. 990–1015, 2015.
O. C. Z. Gotel and A. C. W. Finkelstein, “[A]{}n [A]{}nalysis of the [R]{}equirements [T]{}raceability [P]{}roblem,” in *Proceedings of IEEE International Conference on Requirements Engineering (ICRE 1994)*.1em plus 0.5em minus 0.4emIEEE, 1994, pp. 94–101.
P. Sturgeon, “[T]{}he [N]{}everending [M]{}uppet [D]{}ebate of [PHP 6]{} v [PHP 7]{},” 2014, [A]{}vailable online in March 2017: <https://philsturgeon.uk/php/2014/07/23/neverending-muppet-debate-of-php-6-v-php-7/>.
, “[R]{}equest for [C]{}omments: [R]{}elease [P]{}rocess,” 2010, [A]{}vailable online in March 2017: <https://wiki.php.net/rfc/releaseprocess>.
——, “[S]{}upported [V]{}ersions,” 2017, [A]{}vailable online in March 2017: <http://php.net/supported-versions.php>.
——, “[U]{}nsupported [H]{}istorical [R]{}eleases,” 2017, [A]{}vailable online in March 2017: <https://secure.php.net/releases/>.
M. V. Mäntylä and J. Vanhanen, “[S]{}oftware [D]{}eployment [A]{}ctivities and [C]{}hallenges – [A]{} [C]{}ase [S]{}tudy of [F]{}our [S]{}oftware [P]{}roduct [C]{}ompanies,” in *Proceedings of the 15th European Conference on Software Maintenance and Reengineering (CSMR 2011)*.1em plus 0.5em minus 0.4emOldenburg: IEEE, 2011, pp. .
P. Singer, D. Helic, B. Taraghi, and M. Strohmaier, “[D]{}etecting [M]{}emory and [S]{}tructure in [H]{}uman [N]{}avigation [P]{}atterns [U]{}sing [M]{}arkov [C]{}hain [M]{}odels of [V]{}arying [O]{}rder,” *PLOS ONE*, vol. 9, no. 7, p. e102070, 2014.
M. F. Hill, J. D. Witman, and H. Caswell, “[M]{}arkov [C]{}hain [A]{}nalysis of [S]{}uccession in a [R]{}ocky [S]{}ubtidal [C]{}ommunity,” *The American Naturalist*, vol. 164, no. 2, pp. E46–E61, 2004.
G. A. Spedicato, “[markovchain]{}: [D]{}iscrete [T]{}ime [M]{}arkov [C]{}hains [M]{}ade [E]{}asy,” 2016, [R]{} package version 0.6, available online in March 2017: <https://cran.r-project.org/web/packages/markovchain/index.html>.
, “[D]{}ownloads,” 2017, [A]{}vailable online in March 2017: <http://httparchive.org/downloads.php>.
T. Wambach and K. Bräunlich, “[T]{}he [E]{}volution of [T]{}hird-[P]{}arty [W]{}eb [T]{}racking,” in *Proceedings of the International Conference on Information Systems Security and Privacy (ICISSP 2016)*, O. Camp, S. Furnell, and P. Mori, Eds.1em plus 0.5em minus 0.4emRome: Springer, 2016.
S. G. Ainsworth and M. L. Nelson, “[E]{}valuating [S]{}liding and [S]{}ticky [T]{}arget [P]{}olicies by [M]{}easuring [T]{}emporal [D]{}rift in [A]{}cyclic [W]{}alks [T]{}hrough a [W]{}eb [A]{}rchive,” in *Proceedings of the 13th ACM/IEEE-CS Joint Conference on Digital Libraries (JCDL 2013)*.1em plus 0.5em minus 0.4emIndianapolis: ACM, 2013, pp. 39–48.
, “[FAQ]{},” 2017, [A]{}vailable online in March 2017: <http://httparchive.org/about.php#faq>.
M. C. Calzarossa and L. Massari, “[A]{}nalysis of [H]{}eader [U]{}sage [P]{}atterns of [HTTP]{} [R]{}equest [M]{}essages,” in *Proceedings of the 2014 IEEE Intl Conf on High Performance Computing and Communications, 2014 IEEE 6th Intl Symp on Cyberspace Safety and Security, 2014 IEEE 11th Intl Conf on Embedded Software and Syst (HPCC, CSS, ICESS)*.1em plus 0.5em minus 0.4em Paris: IEEE, 2014, pp. 847–853.
K. Pham, A. Santos, and J. Freire, “[U]{}nderstanding [W]{}ebsite [B]{}ehavior [B]{}ased on [U]{}ser [A]{}gent,” in *Proceedings of the 39th International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR 2016)*.1em plus 0.5em minus 0.4emPisa: ACM, 2016, pp. 1053–1056.
L. Wang, A. Nappa, J. Caballero, T. Ristenpart, and A. Akella, “[W]{}ho[W]{}as: [A]{} [P]{}latform for [M]{}easuring [W]{}eb [D]{}eployments on [IaaS]{} [C]{}louds,” in *Proceedings of the 2014 Conference on Internet Measurement Conference (IMC 2014)*.1em plus 0.5em minus 0.4emVancouver: ACM, 2014, pp. 101–114.
J. Siegmund, N. Siegmund, and S. Apel, “[V]{}iews on [I]{}nternal and [E]{}xternal [V]{}alidity in [E]{}mpirical [S]{}oftware [E]{}ngineering,” in *Proceedings of the IEEE/ACM 37th IEEE International Conference on Software Engineering (ICSE 2015)*.1em plus 0.5em minus 0.4emFlorence: IEEE, 2015, pp. 9–19.
P. Barford, I. Canadi, D. Krushevskaja, Q. Ma, and S. Muthukrishnan, “[A]{}dscape: [H]{}arvesting and [A]{}nalyzing [O]{}nline [D]{}isplay [A]{}ds,” in *International Conference on World Wide Web (WWW 2014)*.1em plus 0.5em minus 0.4emSeoul: ACM, 2014, pp. 597–608.
Y. J. Park, “[A]{} [B]{}roken [S]{}ystem of [S]{}elf-[R]{}egulation of [P]{}rivacy [O]{}nline? [S]{}urveillance, [C]{}ontrol, and [L]{}imits of [U]{}ser [F]{}eatures in [U.S.]{} [W]{}ebsites,” *Policy & Internet*, vol. 6, no. 4, pp. 360–376, 2014.
A. F. Tappenden and J. Miller, “[C]{}ookies: [A]{} [D]{}eployment [S]{}tudy and the [T]{}esting [I]{}mplications,” *ACM Transactions on the Web*, vol. 3, no. 3, pp. 9:1 – 9:49, 2009.
K. He, A. Fisher, L. Wang, A. Gember, A. Akella, and T. Ristenpart, “[N]{}ext [S]{}top, the [C]{}loud: [U]{}nderstanding [M]{}odern [W]{}eb [S]{}ervice [D]{}eployment in [EC2]{} and [A]{}zure,” in *Proceedings of the 2013 Conference on Internet Measurement Conference (IMC 2013)*.1em plus 0.5em minus 0.4em Barcelona: ACM, 2013, pp. .
M. Kozina, M. Golub, and S. Groš, “[A]{} [M]{}ethod for [I]{}dentifying [W]{}eb [A]{}pplications,” *International Journal of Information Security*, vol. 8, no. 6, pp. , 2009.
W.-K. Ching, M. K. Ng, and E. S. Fung, “[H]{}igher-[O]{}rder [M]{}ultivariate [M]{}arkov [C]{}hains and [T]{}heir [A]{}pplicatons,” *Linear Algebra and Its Applications*, vol. 428, no. 2–3, pp. 492–507, 2008.
B. Newton, K. Jeffay, and J. Aikat, “[T]{}he [C]{}ontinued [E]{}volution of [W]{}eb [T]{}raffic,” in *Proceedings of the IEEE 21st International Symposium on Modelling, Analysis and Simulation of Computer and Telecommunication Systems (MASCOTS 2013)*.1em plus 0.5em minus 0.4emIEEE, 2013, pp. 80–89.
S. Goutte, “[C]{}onditional [M]{}arkov [R]{}egime [S]{}witching [M]{}odel [A]{}pplied to [E]{}conomic [M]{}odelling,” *Economic Modelling*, vol. 38, pp. 258–269, 2014.
G. A. F. Seber, *[A]{} [M]{}atrix [H]{}andbook for [S]{}tatisticians*.1em plus 0.5em minus 0.4emNew Jersey: John Wiley & Sons, 2008.
N. Privault, *[U]{}nderstanding [M]{}arkov [C]{}hains: [E]{}xamples and [A]{}pplications*.1em plus 0.5em minus 0.4emHeidelberg: Springer, 2013.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Some changes in a recent convolution formula are performed here in order to clean it up by using more conventional notations and by making use of more referrenced and documented components (namely Sierpiński’s polynomials, the Thue-Morse sequence, the binomial modulo 2 transform and its inverse). Several variants are published here, by reading afterwards summed coefficients in another order; the last formula is then turned back from a summation to a new divide-and-conquer recursive formula.'
author:
- Thomas Baruchel
date: November 2019
title: |
A non-symmetric divide-and-conquer\
recursive formula for the convolution\
of polynomials and power series
---
Introduction
============
In a recently published paper, a new convolution formula was written down by tracking all terms along a recursion tree built from a variant of Karatsuba’s well-known algorithm [@baruchel]. While several variants of the formula were already given there, some unusual notations were heavily used in order to “pack” the required terms into a single summation. Despite the conciseness of these formulas, one could thus argue that they may be too complicated to stand as an inspiring starting point for ensuing researches.
Small changes in one of them can however lead to another more explicit variant, by noticing that three different arbitrary symbols are actually related to Sierpiński’s polynomials. The coefficients of these polynomials are those from the well-documented Sierpiński triangle, and it may be expected that publishing a new simpler formula relying on such polynomials for something as significant as multiplicating two polynomials (or convoluting two power series) could have some benefits. In Sections 2 to Section 4, some elementary properties of Sierpiński’s polynomials are used in order to prove two new variants as Theorems \[theorem1\] and \[theorem2\], one for polynomials and one for infinite power series.
Rewriting now Theorem \[theorem2\] by summing all embedded coefficients “vertically” rather than “horizontally” in Section \[binomial\] allows to take benefit of a little-studied transform called the “binomial modulo 2 transform” (and its inverse) in order to achieve a still more compact variant. The resulting formula in theorem \[theorem4\] being, like all these variants, a summation, it is then turned into to a new *divide-and-conquer* recursive formula in section \[recursive\].
This final recursive formula is said to be *non-symmetric* because both convolved power series are not handled in the same way, one being handled by performing subtractions and the other by performing additions.
Sierpiński’s polynomials
========================
The Sierpiński triangle is best known as a graphical figure (see below); it is a fractal object built by adding at each iteration two new copies of the same whole object at its own bottom (and scaling down the whole figure in order to keep its original size). Thus iterating over its rows from top to bottom is possible: the $n$th row (counting from the top) is always the same whatever the number of previous iterations is.
Iterating over the rows of the triangle and reading them as finite sequences of binary digits ($0$ for “white” and $1$ for “black”) gives another mathematical object which is also refferred to as the Sierpiński triangle but now in some combinatorial context. Such coefficients are those from Pascal’s triangle *modulo* $2$.
The sequential rows may also be read as polynomials by taking the previously described $0$ and $1$ as coefficients, resulting in the sequence $S$ of Sierpiński’s polynomials defined among the comments of the sequence `A047999` in the *On-Line Encyclopedia of Integer Sequences* [@oeis]. The initial polynomials are: $$1,\, 1+x,\, 1+x^2,\, 1+x+x^2+x^3,\, \dots$$
The power series expanding $(1-x)^{-1}S_k^{-1}$ will be referred to as $\delta_k$ in the current paper (the symbol $\delta$ being chosen because $\delta_k$ happens to be the $k$th main *diagonal* in the Sierpiński triangle).
All formulas below rely on the *termwise* product of sequences (or power series), which will be denotated here as $f\odot g$ (meaning that coefficients of the same rank are multiplicated with no convolution).
\[lemma1\] Let $m=(d_{n-1}\dots d_2 d_1 d_0)_2$ some nonnegative integer between $0$ and $2^n-1$ having the finite sequence $d$ as the digits of its binary encoding; then $$S_m = \left(1+x\right)^{d_0}
\left(1+x^2\right)^{d_1}
\left(1+x^4\right)^{d_2}
\,\dots\,
\left(1+x^{2^{n-1}}\right)^{d_{n-1}}
\,\textrm{.}$$
This is obviously true for $m=0$ and $m=1$. Then, we refer to the building rule described as a comment of the sequence `A047999`: $$\left\{
\begin{array}{l}
S_{2n+1}\left(x\right) = \left(x+1\right)S_n\left(x^2\right)
\\[4pt]
S_{2n}\left(x\right) = S_n\left(x^2\right)
\end{array}
\right.$$ in order to prove by induction that if the lemma is true for any $m$ smaller than some power of $2$, it is also true for any value of $m$ smaller than the following power of $2$. The proof is straightforward since the binary encoding of $2n$ is known to be the same as the one of $n$ shifted to the left by one digit (thus performing $d_{j+1}\gets d_j$) while replacing $x$ by $x^2$ is the same as replacing $$\left(1+x^{2^j}\right)
\quad\textrm{by}\quad
\left(1+x^{2^{j+1}}\right)$$ everywhere in the whole product above.
\[lemma2\] Let $n$ be some power of $2$ and $k$ some nonnegative integer smaller than $n$; then: $$S_k\times S_{n-1-k} = S_{n-1}\,\textrm{.}$$
The binary encoding of $n-1$ is $(111\dots 111)_2$ since $n$ is a power of $2$; thus $n-1-k$ and $k$ have complementary binary encodings. Thus, according to Lemma \[lemma1\], $S_k$ and $S_{n-1-k}$ have complementary factors in regards to the whole product defined in that lemma, which soon leads to the above statement.
The Thue-Morse sequence
=======================
Let $\sigma$ be some specific encoding of the Thue-Morse sequence defined as the sequences `A106400` in the *On-Line Encyclopedia of Integer Sequences* [@oeis], namely: $$1, -1, -1, 1, -1, 1, 1, -1, -1, 1, 1, -1, 1, -1, -1, 1, \dots$$ Like the Sierpiński triangle, the sequence may be built from a duplicating process: the $2^k$ initial coefficients are copied to their right with their sign being flipped in order to build the initial $2^{k+1}$ coefficients.
The notation $\bar{f}$ will be used here for representing the *termwise* product of $\sigma$ and some polynomial $f$ (or power series[^1]). We can thus write: $\bar{f}=\sigma\odot f$ with the symbol $\odot$ indicating the *termwise* product of two objects.
Let also $\bar{S}$ be specifically the sequence of polynomials such that $$\bar{S}_m = \left(1-x\right)^{d_0}
\left(1-x^2\right)^{d_1}
\left(1-x^4\right)^{d_2}
\,\dots\,
\left(1-x^{2^{n-1}}\right)^{d_{n-1}}$$ with $m=(d_{n-1}\dots d_2 d_1 d_0)_2$ some nonnegative integer between $0$ and $2^n-1$. This notation is very slightly different from the $\bar{f}$ one since $\bar{S}$ is a sequence of polynomials while $f$ is a polynomial (or a power series), but it is easy to show that $\bar{S}_m=\overline{(S_m)}=\sigma\odot S_m$ by noticing that a coefficient in $\bar{S}_m$ will be $-1$ if and only if an odd number of digits occurs in the binary encoding described above, which exactly matches the corresponding term in $\sigma$ since $\sigma_n=(-1)^{\mathcal{H}_n}$ (with $\mathcal{H}_n$ being the Hamming weight of $n$).
\[lemma3\] Let $n$ be some power of $2$ and $k$ some nonnegative integer smaller than $n$; let also $f$be some polynomials in the indeterminate $x$; then: $$S_{n-1-k} \,x^k \,\odot\, \bar{S}_k \,f
\,=\, \overline{S_{n-1-k} \,x^k \,\odot\, S_k\,\bar{f}}
\,\textrm{.}$$
This is true for $k=0$ and we want to prove by induction that when the identity is true for some $k$ we can also write: $$S_{n-1-k-2^j} \,x^{k+2j} \,\odot\, \left(1-x^{2^j}\right)\bar{S}_k \,f
\,=\, \overline{S_{n-1-k-2^j} \,x^{k+2^j}
\,\odot\, \left(1+x^{2^j}\right)\overline{S_k\,\bar{f}}}$$ with $2^j$ some power of $2$ not already present in the binary expansion of $k$.
The left-hand side of the previous equation means that we want to take consecutive blocks of $2^{j+1}$ coefficients; subtract the initial $2^j$ ones to the following $2^j$ ones; and finally cancel the second half of such blocks (cancelling half of each block being performed by removing one more factor from the *mask* $S_{n-1-k}$).
The right-hand side of the same equation performs the very same thing in another way: we flip the sign of coefficients in such a way that in consecutive blocks of $2^{j+1}$ coefficients, the $2^j$ initial ones will be flipped in an opposite manner than in the following $2^j$ ones (this comes from the building rule of the Thue-Morse sequence); then we add (rather than subtract) the two parts; then we flip back the signs of the coefficients to their initial state.
New variants of the convolution formulas
========================================
The general idea of a previous paper was to study a variant of Karatsuba’s algorithm and “flatten” the recursion tree into a summation formula (see [@baruchel]). Before going further, the latter is rewritten with more expressive notations in order to help manipulating it.
\[theorem1\] Let $f$ and $g$ two polynomials of degree $n-1$ with $n$ some power of $2$ in the same indeterminate $x$; then $$\begin{array}{lcl}
f\times g
&=&\displaystyle
\sum_{k=0}^{n-1}
\sigma_k \, S_{n-1-k}\left( S_{n-1-k}\, x^k \,\odot\, \bar{S}_k\, f \,\odot\, \bar{S}_k\, g \right)
\\[16pt]
&=&\displaystyle
\sum_{k=0}^{n-1}
\sigma_k \, S_{n-1-k}\left( S_{n-1-k}\, x^k \,\odot\, S_k\, \bar{f} \,\odot\, S_k\, \bar{g} \right) \,\textrm{.}
\end{array}$$
We copy the the formula (8) from [@baruchel] as it is typeset in the original paper despite some differences in used notations; it will be translated to the current notations afterwards: $$A\times B\, = \,
\sum_{k=0}^{n-1}
\bar{f}_k\left(\sigma_k \bar{f}_k X^k \odot \dot{f}_k A\odot\dot{f}_k B\right)
\,\textrm{.}$$ In the previous formula, $\dot{f}_k$ means exactly the same thing as $\bar{S}_k$ in the current paper, while $\bar{f}_k$ can be recognized here as $S_{n-1-k}$ with the help of Lemma \[lemma2\]. Thus, the formula can now be translated as: $$f\times g\, = \,
\sum_{k=0}^{n-1}
S_{n-1-k}\left(\sigma_k S_{n-1-k} x^k \odot \bar{S}_k f\odot \bar{S}_k g\right)
\,\textrm{.}$$ According to Lemma \[lemma3\], $\bar{S}_k f$ and $\bar{S}_k g$ can be replaced above by $S_k\bar{f}$ and $S_k\bar{g}$ since all flipped signs will cancel themselves when evaluating the termwise product $S_k\bar{f}\odot S_k\bar{g}$, leading to the stated formula.
\[theorem2\] Let $f$ and $g$ be two power series in the same indeterminate $x$; then $$\begin{array}{lcl}
f\ast g
&=&\displaystyle
\sum_{k=0}^{\infty}
\displaystyle\sigma_k \delta_k \left( x^k \delta_k \,\odot\, \bar{S}_k\, f \,\odot\, \bar{S}_k\, g \right)
\\[16pt]
&=&\displaystyle
\sum_{k=0}^{\infty}
\displaystyle\sigma_k \delta_k\left( x^k \delta_k \,\odot\, S_k\, \bar{f} \,\odot\, S_k\, \bar{g} \right) \,\textrm{.}
\end{array}$$
Formula (9) from [@baruchel] is now taken into account. We can not refer here to Lemma \[lemma2\] any longer for building some complementary polynomial because we do not work on a finite number $n$ of terms, but the theory of generating function is useful for building the relevant power series, since $$\frac{1}{1-x}\quad\textrm{expands to}\quad
\left(1+x\right)
\left(1+x^2\right)
\left(1+x^4\right)
\left(1+x^8\right)
\,\dots\,$$ where the required factors can easily be cancelled by a simple division: the previously defined $\delta_k$ term matches the required infinite product.
The binomial modulo 2 inverse transform {#binomial}
=======================================
In a comment of `A100735` in the *On-Line Encyclopedia of Integer Sequences* [@oeis] are defined the “modulo 2 binomial transform” and its inverse. In the current paper, the name of this transform is slightly changed to “binomial modulo 2 transform”. Both transforms are defined as: $$\begin{array}{lcl@{\qquad}l}
B_n&=&\displaystyle\sum_{k=0}^n T_{n,k} A_k&\textit{(transform)} \\[14pt]
A_n&=&\displaystyle\sum_{k=0}^n \sigma_{n-k} T_{n,k} B_k&\textit{(inverse transform)}
\end{array}$$ for a given sequence $A$, where $\sigma$ is the signed Thue-Morse sequence and $T_{n,k}$ is the relevant coefficient in the Sierpiński triangle. The notations $B=A'$ and $A=B^*$ will be used from now on; the same notations will be applied to power series as well.
\[theorem3\] The power series $f$ being defined as $f= a_0 + a_1 x + a_2 x^2 + \dots$, we define also $f_k=a_k + a_{k+1}x + a_{k+2}x^2 + \dots$, and $f_k^*$ is the power series defined from the binomial modulo 2 inverse transform of the sequence $a_k, a_{k+1}, a_{k+2},\dots$ (while $f'$ is the binomial modulo 2 transform of $f$). Then for two power series $f$ and $g$ in the same indeterminate $x$, $$f\ast g =
\sum_{k=0}^{\infty}
\left(
\,\overline{ \delta_k} \odot f_k^* \odot g_k^*
\right)'
\, x^k
\,\textrm{.}$$
In Theorem \[theorem2\], the whole parenthesis is turned into $$\sum_{j=0}^{\infty}
\delta_j[k] \, f_j^*[k] \, g_j^*[k] \, x^{k+j}$$ where $f[k]$ is the coefficient of term of degree $k$ in $f$.
Then we gather all contributions of some degree $m$ by noticing that some previously defined term of degree $k+j$ is shifted to degree $m$ (by performing the remaining multiplication in Theorem \[theorem2\]) if and only if $m-(k+j)$ and $k$ have no common digit $1$ in their binary encodings: $$(f\ast g)[m] = \sum_{j=0}^{m}
\sum_{k=0}^{m-j}
T_{m-j,k}\,
\sigma_k\,
\delta_j[k] \, f_j^*[k] \, g_j^*[k]$$ where $T$ is Sierpiński triangle with $T_{n,k}=1$ if and only if $(n-k)\&k=0$, the symbol $\&$ being the bitwise `and` operator.
This identity quickly leads to the expected theorem.
\[slemma\] The power series $f$ being defined as $f= a_0 + a_1 x + a_2 x^2 + \dots$, we define also $f_k=a_k + a_{k+1}x + a_{k+2}x^2 + \dots$, and $f_k^*$ is the power series defined from the binomial modulo 2 inverse transform of the sequence $a_k, a_{k+1}, a_{k+2},\dots$ (with of course $f^* = f_0^*$). Then, $$\left( \delta_k \odot f_k^* \right) x^k
= \delta_k \, x^k \odot f^* S_k
\,\textrm{.}$$
We prove this by induction; the statement is obviously true for $k=0$; then we assume it is true for some nonnegative integer $k$, and we show that it is still true for $k+n$ with $n$ being some power of $2$ greater than $k$. Let $g=f_k$ and $g_n=f_{k+n}$ in order to focus on a power of 2; we want to study $\left(\delta_{k+n}\odot g_n^*\right) x^{k+n}$ in order to match the left-hand side above.
Because of the self-similarity in the Sierpiński triangle, we can notice that $\left(1+x^n\right)g^*$ and $x^n g_n^*$ share the same coefficient of degree $m$ if $m \,\textrm{mod}\,\, 2n \geqslant n$. For the same reasons, $\delta_k$ and $x^n \delta_{k+n}$ share the same coefficient of degree $m$ if $m \,\textrm{mod}\,\, 2n \geqslant n$ (all other coefficients of $\delta_{k+n} x^n$ are null). Thus, $$\begin{array}{lcl}
\left(\delta_{k+n}\odot f_{k+n}^*\right) x^{k+n}
&=& \left(\delta_{k+n}\odot g_n^*\right) x^{k+n} \\[4pt]
&=& \left(x^n \delta_{k+n} \odot x^n g_n^*\right) x^{k} \\[4pt]
&=& \left( x^n\delta_{k+n}\odot\left(1+x^n\right) g^* \right)x^k \\[4pt]
&=& x^{k+n}\delta_{k+n}\odot\left(1+x^n\right) x^k f_k^* \\[4pt]
\end{array}$$ where, again, the (rather restrictive) mask $x^{k+n}\delta_{k+n}$ allows to replace $x^k f_k^*$ with $f^* S_k$ (by using the initial assumption) *because all coefficients to be added and kept in the multiplication by $\left(1+x^n\right)$ where also taken into account by the initial (less restrictive mask)*. Of course $\left(1+x^n\right)f^* S_k = f^* S_{k+n}$.
\[slemma2\] Let $f$ be some power series in the indeterminate $x$. Then, $$\left( \delta_k \odot f \right)' x^k
= \left( \left( \delta_k \odot f\right) x^k \right)' S_k
\,\textrm{.}$$
We prove this by induction; the statement is obviously true for $k=0$; then we assume it is true for some nonnegative integer $k$, and we show that it is still true for $k+n$ with $n$ being some power of $2$ greater than $k$.
Because of the self-similarity in the Sierpiński triangle, we can notice that all coefficients of $\left(\delta_{k+n}\odot f\right)'$ in the left-hand part above can be found among the coefficients of $\left(\delta_{k}\odot f\right)'$ according to the following rule: $$\left\{
\begin{array}{l@{\qquad}l}
\left(\delta_{k+n}\odot f\right)'[m]
= \left(\delta_{k}\odot f\right)'[m] & \text{ if $m \,\textrm{mod}\,\, 2n < n$; }\\[6pt]
\left(\delta_{k+n}\odot f\right)'[m]
= \left(\delta_{k}\odot f\right)'[m-n] & \text{ if $m \,\textrm{mod}\,\, 2n \geqslant n$; }
\end{array}
\right.$$ which can also be written as $\left(\delta_{k+n}\odot f\right)' = \left(\delta_n\odot \left(\delta_{k}\odot f\right)'\right)\left(1+x^n\right)$.
The same idea is separately followed for the right-hand part of the identity to be proved: $\left( \left( \delta_{k+n} \odot f\right) x^{k+n} \right)' S_{k+n}$ is also made of coefficients from $\left( \left( \delta_k \odot f\right) x^k \right)' S_k$ according to the very same same rule because of the self-similarity property of the Sierpiński triangle again. Thus we see that both sides of the statement match exactly.
\[theorem4\] Let $f$ and $g$ be two power series in the same indeterminate $x$. Then, $$f\ast g =
\sum_{k=0}^{\infty}
\left(
\,
\overline{ \delta_k}\, x^k \,\odot\, f^* S_k \,\odot\, g^* S_k
\right)'
\, S_k
\,\textrm{.}$$
Theorem \[theorem3\] is rewritten with the help of Lemma \[slemma\]: $$f\ast g =
\sum_{k=0}^{\infty}
\left(
\,\displaystyle\frac
{
\overline{ \delta_k}\, x^k \,\odot\, f^* S_k \,\odot\, g^* S_k
}
{x^k}
\right)'
\, x^k$$ which quickly leads to the expected statement with the help of Lemma \[slemma2\].
A recursive formula for the convolution {#recursive}
=======================================
Summations involving Sierpiński’s polynomials may be easy to convert into a recursive formula because “jumping” from $S_k$ to $S_{k+n}$ (with $n$ being a power of 2 greater than $k$) is merely achieved by using the relation $S_{k+n}=(1+x^n)S_k$.
Furthermore, the parenthesis in Theorem \[theorem4\] contains terms which still follow the previously studied *interleaved splitting scheme* (see Section 3 in [@baruchel]), meaning that they can be computed from another ones by splitting them into two increasingly-sparse terms. An exact definition of the scheme used below is: $$\left\{
\begin{array}{lcl}
u_{\text{\tiny low}}^{\text{\tiny $(n)$}}
&=& u\odot \delta_n \\[5pt]
u_{\text{\tiny high}}^{\text{\tiny $(n)$}}
&=& u\odot x^n \delta_n
\end{array}\right.
\qquad\text{with $n$ being some power of~$2$.}$$
For traversing the recursion tree, an *ad hoc* operator will be used here in order to “pack” the summation from Theorem \[theorem4\] into the following formula: $$\label{recformula}
f\ast g \,=\, \left(\,\overline{f^*}\odot g^*\right)'
\,+\, \overline{f^*}
\underset{\text{\tiny $(1)$}}{\diamond} g^*$$ this parametrized operator symbol $\underset{\text{\tiny $(n)$}}{\diamond}$ being recursively defined as: $$\label{recoperator}
\begin{array}{lcl}
u \underset{\smash{\text{\tiny $(n)$}}}{\diamond} v
&=& \left(1+x^n\right)
\left( x^n u_{\text{\tiny low}}^{\text{\tiny $(n)$}} - u_{\text{\tiny high}}^{\text{\tiny $(n)$}}
\, \odot \,
x^n v_{\text{\tiny low}}^{\text{\tiny $(n)$}} + v_{\text{\tiny high}}^{\text{\tiny $(n)$}}
\right)' \\[4pt]
&&+\, u \underset{\smash{\text{\tiny $(2n)$}}}{\diamond} v \\[4pt]
&&+\, \left(1+x^n\right)
\left(
x^n u_{\text{\tiny low}}^{\text{\tiny $(n)$}} - u_{\text{\tiny high}}^{\text{\tiny $(n)$}}
\, \underset{\smash{\text{\tiny $(2n)$}}}{\diamond} \,
x^n v_{\text{\tiny low}}^{\text{\tiny $(n)$}} + v_{\text{\tiny high}}^{\text{\tiny $(n)$}}
\right)
\end{array}$$ (where the digits in the binary encoding of the variable $k$ occuring in Theorem \[theorem4\] are read from right to left by either choosing a digit 0 or a digit 1, explaining the two recursive calls).
Obiously the previously-defined operator is non-commutative. Building this formula as a non-symmetric one allows to track and preserve the signs from the $\overline{\delta_k}$ term without using an external mask (which is the case in Theorem \[theorem4\]). Because of the properties of the Thue-Morse sequence, subtractions actually embed “hidden” additions (since exactly one of both terms occuring in each subtractions previously had its sign flipped).
Conclusion
==========
The new theorems \[theorem1\] and \[theorem2\] are equivalent to their previous versions in [@baruchel], but they now make use of more documented notations. They are therefore claimed to be of a more general interest for later researches. The binomial modulo 2 transform is used a couple of times in the *On-Line Encyclopedia of Integer Sequences*, but it does not seem to have been actually studied in published papers; thus theorems and lemmas proved in Section \[binomial\] and \[recursive\] may also provide a basis for further investigating the properties of this transform. The conciseness of formulas in Theorems \[theorem3\] and \[theorem4\] show that “transposing” the initial ones suit more the approach being investigated since the previous paper.
Formula (\[recoperator\]) in Section \[recursive\] “packs” the whole computation into a rather simple statement with no need for symbols related to the Sierpiński triangle any longer, allowing to focus henceforth on the binomial modulo 2 transform only. Directly implementing this formula as code or pseudo-code was not considered at this point, because it would certainly not be very efficient as it is; it should be noticed however that while the definition of the binomial modulo 2 transform in Section \[binomial\] obviously is in $O(n^2)$, much better algorithms for computing the whole transform in $O(n\log n)$ are achievable. Once initially computed, the transform could probably be propagated and partially updated during the recursion process if attempting to efficiently implement the formula.
[**Conflict of Interest:** The authors declare that they have no conflict of interest.]{}
[The current article is accessible on `http://export.arxiv.org/pdf/1912.00452`.]{}
[9]{}
*The On-Line Encyclopedia of Integer Sequences*, published electronically at `https://oeis.org`. Accessed 1 Nov. 2019.
Thomas <span style="font-variant:small-caps;">Baruchel</span>, *Flattening Karatsuba’s recursion tree into a single summation*, in *Computer Science*, ed. Springer Nature, Jan. 2020. Also available at `https://arxiv.org/abs/1902.08982` .
[^1]: The same notation could obviously also be used for a sequence of numbers if needed.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Model-based control is a popular paradigm for robot navigation because it can leverage a known dynamics model to efficiently plan robust robot trajectories. However, it is challenging to use model-based methods in settings where the environment is unknown and can only be observed partially through on-board sensors on the robot. In this work, we address this short-coming by coupling model-based control with learning-based perception. The learning-based perception module produces a series of *waypoints* that guide the robot to the goal via a collision-free path. These waypoints are used by a model-based planner to generate a smooth and dynamically feasible trajectory that is executed on the physical system using feedback control. Our experiments in simulated real-world cluttered environments and on an actual ground vehicle demonstrate that the proposed approach can reach goal locations more reliably and efficiently in novel environments as compared to purely geometric mapping-based or end-to-end learning-based alternatives. Our approach does not rely on detailed explicit 3D maps of the environment, works well with low frame rates, and generalizes well from simulation to the real world. Videos describing our approach and experiments are available on the project website$^3$.'
author:
- 'Somil Bansal${}^{*1}$ Varun Tolani${}^{*1}$ Saurabh Gupta${}^{2}$ Jitendra Malik${}^{1,2}$ Claire Tomlin${}^{1}$'
bibliography:
- 'references.bib'
title: |
Combining Optimal Control and Learning for\
Visual Navigation in Novel Environments
---
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We have detected the CO(6-5), CO(7-6), and \[C [I]{}\] 370 $\mu$m lines from the nuclear region of NGC 891 with our submillimeter grating spectrometer ZEUS on the CSO. These lines provide constraints on photodissociation region (PDR) and shock models that have been invoked to explain the H$_{2}$ S(0), S(1), and S(2) lines observed with Spitzer. We analyze our data together with the H$_{2}$ lines, CO(3-2), and IR continuum from the literature using a combined PDR/shock model. We find that the mid-J CO originates almost entirely from shock-excited warm molecular gas; contributions from PDRs are negligible. Also, almost all the H$_{2}$ S(2) and half of the S(1) line is predicted to emerge from shocks. Shocks with a pre-shock density of $2\times10^{4}$ cm$^{-3}$ and velocities of 10 km/s and 20 km/s for C-shocks and J-shocks, respectively, provide the best fit. In contrast, the \[C [I]{}\] line emission arises exclusively from the PDR component, which is best parameterized by a density of $3.2\times10^{3}$ cm$^{-3}$ and a FUV field of $G_{o} = 100$ for both PDR/shock-type combinations. Our mid-J CO observations show that turbulence is a very important heating source in molecular clouds, even in normal quiescent galaxies. The most likely energy sources for the shocks are supernovae or outflows from YSOs. The energetics of these shock sources favor C-shock excitation of the lines.'
author:
- 'T. Nikola, G. J. Stacey, D. Brisbin, C. Ferkinhoff, S. Hailey-Dunsheath, S. Parshley, C. Tucker'
bibliography:
- 'n891\_zeus.bib'
title: 'Mid-J CO Emission From NGC 891: Microturbulent Molecular Shocks in Normal Star Forming Galaxies'
---
Introduction \[sec:intro\]
==========================
Turbulence and shocks play a profound role in the process of star formation and nuclear gas accretion. Shocks compress the interstellar medium (ISM) and can trigger star formation, while turbulence prevent the collapse of the ISM. These mechanisms can also remove angular momentum from the gas, leading to gas infall toward galactic nuclei, a process that is important in interacting galaxies.
Both mechanisms are very important in converting mechanical energy into thermal energy and can dominate the heating budget of molecular clouds. The main coolants of the shock heated molecular gas are H$_{2}$ and CO lines. With their improved sensitivity, recent space observatories (ISO, Spitzer) were able to trace extragalactic rotational H$_{2}$ emission and revealed the importance of this emission and the prevalence of shock excited molecular gas. For example, H$_{2}$ emission associated with shock or microturbulence excited molecular gas was found in the group Stephan’s Quintet [@Appleton.Xu.Reach.2006; @Cluver.Appleton.Boulanger.2010], the radio galaxy 3C 326 [@Ogle.Antonucci.Appleton.2007], nearby edge-on galaxies NGC 4565, NGC 5907 [@Laine.Appleton.Gottesman.2010] and NGC 891 [@Stacey.Charmandaris.Boulanger.2010 hereafter SCB10], as well as in the extended cold neutral medium in the Milky Way [@Falgarone.Verstaete.PineauDesForets.2005]. Ground-base mid-J CO observations are also good probes of shock or turbulence excited molecular gas. @Bradford.Stacey.Nikola.2005 have invoked dissipation of MHD turbulence to explain their mid-J CO observations of the circumnuclear ring in Galactic center. In the nuclear region of NGC 253, the strong mid-J $^{12}$CO [e.g. @Bradford.Nikola.Stacey.2003] and $^{13}$CO(6-5) [@HaileyDunsheath.Nikola.Stacey.2008] emission is consistent with heating of the warm molecular gas by either decay of supersonic turbulence through shocks or by an enhanced cosmic ray density.
Due to its proximity (9.5 Mpc), and edge-on presentation [$i\geq89^{\circ}$; @Sancisi.Allen.1979], NGC 891 is the primary target for studies of the effects of energetic processes associated with star formation on the interstellar medium in normal spiral galaxies. The edge-on view of NGC 891 results in greatly enhanced gas columns so that relatively weak lines such as the low-lying quadrupole rotational transitions of H$_{2}$ become observable. The mid-IR H$_{2}$ rotational transitions in NGC 891 were studied for the first time with ISO [@Valentijn.vdWerf.1999] and then more recently with Spitzer . The latter study combined the H$_{2}$ lines observed with Spitzer (S(2), S(1), and S(0)) together with \[C [II]{}\], \[O [I]{}\] and IR continuum observations from ISO to constrain gas excitation mechanisms. They found that most of the S(0) line emission arises in photodissociation regions (PDRs), and most of the S(2) line emission arises in low velocity microturbulent shocks, with the S(1) line arising from both types of sources in the ratio 70% PDRs, and 30% shocks.
Here, we expand on the analysis of by including submillimeter observations of the CO(6-5), CO(7-6), and \[C [I]{}\] 370 $\mu$m lines. The mid-J CO lines can arise from gas heated by a variety of mechanisms, including FUV photons (PDR), X-rays (XDR), cosmic rays, and shocks. The \[C [I]{}\] emission arises mainly from PDRs. Since the Spitzer H$_{2}$ study showed that the main heating sources of the warm molecular gas in NGC 891 are turbulence and PDRs we will focus on these two heating mechanisms and use our new submillimeter lines to both test and constrain the models of . As in their analysis, we apply the PDR model of @Kaufman.Wolfire.Hollenbach.1999 and the shock model of @Flower.PineauDesForets.2010.
Observations \[sec:obs\]
========================
We have observed the $^{12}$CO $J=6\to5$ (433.56 $\mu$m) and $J=7\to6$ (371.65 $\mu$m) rotational transitions and the \[C [I]{}\] $^3P_2 \to ^3P_1$ (370.41 $\mu$m) fine structure line with our submillimeter grating spectrometer ZEUS [@HaileyDunsheath.2009] on the Caltech Submillimeter Observatory (CSO) on Mauna Kea, Hawaii, in December 2006.
ZEUS employs a $1\times32$ thermistor sensed detector array that is optimized for observations in the submillimeter wavelength regime. It provides 32 spectral elements at one spatial position. For the Dec. 2006 observing run, we mounted two bandpass filter appropriate for the 350 and 450 $\mu$m telluric window adjacent to each other at the entrance slit of the detector housing, each filter covering 16 spectral pixels. This allowed us to observe in the 350 and 450 $\mu$m telluric windows simultaneously and it still provided a full bandwidth coverage of about 5000 km/s in each band. The spectral resolution of ZEUS is $\lambda/\Delta\lambda \sim 1000$ across the 350 $\mu$m and 450 $\mu$m telluric windows. The CO(7-6) and \[C [I]{}\] lines, which are separated in velocity by only 1001 km/s, were observed simultaneously within a single spectrum. We verified the spectral calibration through observation of CO gas absorption lines from a calibration unit mounted in front of the ZEUS dewar.
The CO lines were observed on two different nights. On the first night we observed only the CO(6-5) emission at the position RA=${\rm 02^h22^m33.10^s}$ and Dec=$+42^{\circ}20'55.1''$ (J2000). On the second night we observed the CO(6-5), (7-6), and \[C [I]{}\] lines, but centered $\sim 2''$ further to the north. This difference is, however, within the pointing accuracy of the telescope. All observations were made in chop-nod mode with the secondary chopping at 2 Hz and a chop throw of $30''$.
We established a pointing model from observations of Saturn and Uranus at various elevation and azimuth positions. The derived pointing accuracy is $\lesssim 5''$. We estimated the coupling of the planets to the beam by assuming the planets to be disk-like sources with uniform emissivity and assuming the ZEUS beam to be Gaussian. During the observations the beam size was $10''$ in the 350 $\mu$m band and $11''$ in the 450 $\mu$m band. At the distance of NGC 891, $10''$ corresponds to 460 pc.
All observations were corrected for atmospheric absorption through use of the CSO radiometers. Each spectrum was corrected for the response function of the grating and detector array, then flux calibrated with respect to Saturn, which was assumed to emit like a blackbody with a temperature of $T = 113$ K at 370 $\mu$m and 434 $\mu$m [@Hildebrand.Loewenstein.Harper.1985]. The correction for telluric transmission is the largest source of error. Overall, we estimate a line flux uncertainty of $\sim 25$%.
The CO(6-5), CO(7-6), and \[C [I]{}\] 370 $\mu$m lines were all strongly detected (Fig. \[fig:spec\]), and their observed line intensities are listed in Table \[tab:zeusobs\].
Origin Of The Mid-J CO Emission in NGC 891 \[sec:origin\]
=========================================================
The warm molecular gas traced by mid-J CO emission can be excited by several mechanisms. For spiral galaxies with quiescent star formation, the most relevant are shocks in micro-turbulent gas or far-UV radiation in photodissociation regions (PDRs). Spitzer observations of the H$_{2}$ mid-IR rotational transitions in NGC 891 have revealed the presence of warm, shock-excited molecular gas. The line intensities can only be explained if part of the H$_{2}$ emission arises from this warm, micro-turbulent gas phase . While the H$_{2}$ S(0) likely originates entirely from PDRs, 30% of the H$_{2}$ S(1) and the majority (80%) of the H$_{2}$ (S2) emission likely emerges from gas excited by C-type shocks with a pre-shock gas density of $\sim 2 \times 10^{4}$ cm$^{-3}$, shock velocities between 20-30 km/s, and a shock filling factor of 2.8 . The properties of this shock excited gas should also be imprinted in the line ratio of our CO observations.
We are taking a slightly different approach in our shock-analysis than . For their analysis they considered H$_{2}$ emission from within large areas of the galaxy since they compared their Spitzer observations with far-IR lines obtained with ISO, which had a beam size of $75''$. Here, we focus on the central $11''$ of NGC 891 and with our new ZEUS mid-J CO observations we can refine the shock modeling. We combine the Spitzer H$_{2}$ observations (Table \[tab:litobs\]) with our ZEUS observations and apply the shock models of @Flower.PineauDesForets.2010. The spatial resolution of the Spitzer H$_{2}$ observations are very close to our ZEUS observations. The slit orientation of the long-high IRS Spitzer H$_{2}$ S(0) observation was such that the slit length ($22.3''$) is along the major axis and the slit width ($11.1''$) is perpendicular to the major axis. For the short-high IRS Spitzer observations of the H$_{2}$ S(1) and S(2) lines, the slit is rotated, with the slit length ($11.3''$) oriented perpendicular to the major axis of NGC 891 and the slit width ($4.7''$) along the plane of the galaxy. Thus perpendicular to the plane of the galaxy the slits cover about the same area. This width is also nearly identical to the FWHM of the ZEUS beam. However, there is a gap of $30''$ between the individual positions of the Spitzer observations along the major axis. Hence, to convert the Spitzer H$_{2}$ observations to the ZEUS beam size we have to assume a distribution of the H$_{2}$ emission.
In our analysis, we also include the CO(3-2) line, which is available in the literature [@Mauersberger.Henkel.Walsh.1999; @Dumke.Nieten.Thuma.2001; @Bayet.Gerin.Phillips.2006]. Unfortunately, the line intensities and flux densities are not consistent. Despite very similar beam sizes, the reported values vary by a factor of about 6.5. For our analysis we use the intensity given by @Mauersberger.Henkel.Walsh.1999 (Table \[tab:litobs\]), which are about a factor of three less than the largest reported intensity.
The CO(1-0) [e.g. @Scoville.Thakkar.Carlstrom.1993; @Sofue.Nakai.1993] and CO(3-2) [@Mauersberger.Henkel.Walsh.1999; @Dumke.Nieten.Thuma.2001] emission shows a plateau of extended emission with strong central peak and local maxima roughly $1'$ to the North-East and South-West. The H$_{2}$ Spitzer observations also show extended emission with a central peak, a trough at the next sampling positions $30''$ away from the center, local maxima at the second sampling position at $60''$ from the center on either side, and a decrease in the following sampling positions. Thus, within the limited sampling of the Spitzer observations the distribution of the H$_{2}$ and CO emission appear similar. @Dumke.Nieten.Thuma.2001 estimated a source size of about $17''\times8''$ (deconvolved) for the central gas concentration from their CO(3-2) map. The profile of the CO(1-0) emission along the major axis is similar, but slightly larger, showing a FHWM of about $25''$ above the plateau [Fig. 2 and 3 in @Scoville.Thakkar.Carlstrom.1993; @Sofue.Nakai.1993, respectively]. It is plausible that the higher excitation CO emission is more concentrated toward the center than the CO(1-0) emission. Here, we use the source size estimated from the CO(3-2) emission to convert the H$_{2}$ Spitzer observations and the CO(7-6) and \[C [I]{}\] intensities to the $11''$ beam size of our ZEUS CO(6-5) observations, resulting in conversion factors of 1.4 for the H$_{2}$ S(0) line and 0.9 for the H$_{2}$ S(1) and S(2) lines. The converted intensities are given in Table \[tab:fit\_res\].
To model the shock parameters we use a different method than . They compared the H$_{2}$ emission with the far-infrared ISO lines on the scale of the large ISO beam, determined the PDR model that best fits the data (biased toward the ISO far-IR lines), and attributed the remaining H$_{2}$ emission to shock-excited, micro-turbulent gas. Since the ZEUS observations cover a much smaller region we can not use the ISO observations to constrain the line emission fraction. Scaling the ISO observations to the much smaller beam size of ZEUS is subject to large uncertainties without knowing the morphology of the gas-phase locked in PDRs. While a reasonable scaling factor could be inferred for some far-IR lines, it would be questionable for others. Several observations with the Kuiper Airborne Observatory (KAO) have found that the distribution of the \[C [II]{}\] and CO(1-0) emission are correlated in galaxies [cf. @Crawford.Genzel.Townes.1985; @Stacey.Geis.Genzel.1991]. It is therefore reasonable to use the observed CO(1-0) morphology in NGC 891 to scale the \[C [II]{}\] ISO observation to the smaller ZEUS beam. However, no such correlation has been shown for the \[O [I]{}\] lines. In our approach we therefore started by only using the higher spatial resolution observations, the CO(7-6), (6-5), and CO(3-2) transition together with the the H$_{2}$ S(0), S(1) and S(2) lines, and applied a constrained least squares algorithm to simultaneously fit the individual line emission fraction, beam filling factor, and the shock parameters for just a C-shock and a J-shock model. Unfortunately, trying to fit all these parameter to just a shock model results in an under-constrained problem and no unique solution is possible. This is apparent in unreasonably small $\chi^{2}$ values for a variety of shock parameters, line fractions, and beam filling factors. In addition, the beam filling factor and the line fractions are degenerate.
To obtain useful model fits it is necessary to constrain the fraction of the line emission from shocks. This can be achieved by introducing and simultaneously fit a second gas component for the remaining line emission. An obvious choice for this component are PDRs, which have been traced in NGC 891 through the far-infrared line emission on large scales. In our fitting routine for the PDR phase we include our \[C [I]{}\] 370 $\mu$m line and the infrared continuum.
To scale the far-IR continuum to our $11''$ beam we use the observed morphology of the submillimeter continuum. @Serabyn.Lis.Dowell.1999 obtained a map of NGC 891 in the 350 $\mu$m continuum using SHARC on the CSO. They estimated that about 4.8% of the continuum flux density from the entire galaxy emerges from within the central $12''$. Here we assume that the total IR luminosity of NGC 891 of $L(8-1000\mu{\rm m}) = 1.89 \times 10^{10} L_{\odot}$ [RBGS; @Sanders.Mazzarella.Kim.2003; converted to a distance of 9.5 Mpc] scales the same. We convert the IR luminosity from the central region into an intensity and multiply that value by 1.09 to scale to our $11''$ FWHM beam. A large fraction of the IR continuum emission in NGC 891 likely arises from a diffuse dust component and is probably not associated directly with PDRs. Considering the entire galaxy as much as 69% of the IR continuum might arise from this component [@Popescu.Tuffs.Kylafis.2004]. Given that our observation is focused on the center of NGC 891 the percentage might be lower. Here we follow and assume a value of 50% as the IR-fraction arising from cirrus in NGC 891, and allow an uncertainty of 50%.
We exclude the \[C [I]{}\] 609 $\mu$m observation (Table \[tab:litobs\]) since the intensity appears unreasonably small compared to the \[C [I]{}\] 370 $\mu$m line intensity. The ratio of the (converted to $11''$) \[C [I]{}\] 370/\[C [I]{}\] 609 intensity is 17.3. In a PDR this ratio increases with density and FUV field, and for a density of $n = 10^{6}$ cm$^{-3}$ and a FUV field of $G_{o}=10^{6}$ (in units of the Habing field: $1.6\times10^{3}$ erg/s/cm$^{2}$) this line ratio is only 7.8 [@Kaufman.Wolfire.Hollenbach.1999]. For a PDR density of $n = 10^{3}$ cm$^{-3}$ and a FUV field of $G_{o}=10^{2}$ the expected line ratio would be 2.4. The CO(3-2) intensity quoted by @Bayet.Gerin.Phillips.2006 is also the smallest value compared with the values reported by @Mauersberger.Henkel.Walsh.1999 and @Dumke.Nieten.Thuma.2001. It could be possible that our ZEUS \[C [I]{}\] 370 $\mu$m intensity is too high. However, we have measured the \[C [I]{}\] 370 $\mu$m line simultaneously with the CO(7-6) line. If the \[C [I]{}\] 370 $\mu$m intensity is too high would also mean the CO(7-6) is too high. Both the \[C [I]{}\] lines and the cirrus-subtracted IR continuum are expected to originate in PDRs. The \[C [I]{}\] 370/IR and \[C [I]{}\] 609/IR intensity ratios is $2.7\times10^{-4}$ and $1.6\times10^{-5}$, respectively. For a PDR density of $n = 10^{3}$ cm$^{-3}$ and a FUV field of $G_{o}=10^{2}$ the expected ratios are $3.2\times10^{-4}$ and $1.3\times10^{-4}$ for the \[C [I]{}\] 370/IR and \[C [I]{}\] 609/IR intensity ratios, respectively. Although we can not rule out errors in estimating the IR continuum arising from PDRs in the central region of NGC 891 or a systematic error in our calibration, it appears that the ZEUS \[C [I]{}\] 370 $\mu$m intensity is consistent with other measurement and PDR predictions. A possibility for the smaller than expected \[C [I]{}\] 609 $\mu$m intensity is self-absorption by cold foreground gas. The \[C [I]{}\] 609 $\mu$m transition connects to the ground, has an upper level energy equivalent to 24 K, and atomic carbon is the fourth most abundant element. Thus, in an edge-on galaxy self-absorption of fine-structure line transitions between the lowest energy levels of abundant elements is a plausible scenario. For example, \[O [I]{}\] 63 $\mu$m line also connects to the ground, and self-absorption by colder foreground gas has also been invoked to explain the relatively weak \[O [I]{}\] 63 $\mu$m emission from galaxies [e.g. @Vasta.Barlow.Viti.2010 and references therein].
We compare the observed to the predicted line intensities from the combination of a PDR and shock model and apply a least squares algorithm to fit the individual line fractions and the beam filling factors. The model grid for shocks [@Flower.PineauDesForets.2010] is rather coarse, with just two values for the pre-shock densities ($2\times10^{4}$ cm$^{-3}$ and $2\times10^{5}$ cm$^{-3}$) and velocity ranges from 10-30 km/s for J-shocks and 10-40 km/s for C-shocks, with a resolution of 10 km/s. The model grid for the PDR parameters covers a large range, with densities, $\log(n/{\rm cm^{-3}})$, and far-UV fields, $\log(G_o)$, between 1-5, with a step size of 0.25. In the fitting algorithm, we consider the model intensities from a single PDR surface. We also take into account that the CO emission is usually optically thick [e.g. @Bradford.Stacey.Nikola.2005] and that 50% of the infrared continuum originates from cirrus in NGC 891. For an externally FUV-irradiated molecular cloud PDRs are created on the outer surface. Thus, along the line of sight a beam intercepts two projected PDR surfaces from a single cloud, one at the front of the cloud and one at the back of the cloud. An optically thin line would be observed from both PDRs, while an optically thick line would only be detected from the front side. To compare the predicted line intensities from a single PDR surface with the observations we therefore multiply the predicted intensities of the optically thin lines by a factor of 2. The beam filling factor derived for PDRs is then with respect to single PDR surfaces and not for entire clouds. We apply this fitting routine to two combinations: a PDR and C-type shock combination and a PDR and a J-type shock combination. Note that the beam filling factors can be larger than unity if the beam intersects many PDR surfaces or shocks along the line of sight, with each contributing to the observed intensities. We do not include the far-infrared ISO lines in the fitting algorithm because of the uncertain scaling factors. Instead, we just list the predicted intensities for the \[O [I]{}\], \[C [II]{}\], and \[C [I]{}\] 609 $\mu$m lines from the shock and PDR models. The results are shown in Tables \[tab:fit\_cpdr\], \[tab:fit\_jpdr\], and \[tab:fit\_res\], where the predicted intensities from shocks and PDRs are already multiplied by the appropriate beam filling factor and for the PDRs, the optically thin lines are multiplied by an additional factor of 2.
The fit of combined PDR and shock models to the data is slightly better for J-type than for C-type shocks. However, most of the resulting model parameters are similar and the difference in $\chi^2$ is not enough to be conclusive. In both cases the best fit is for a pre-shock gas density of $n = 2 \times 10^{4}$ cm$^{-3}$, a PDR density of $n_{\rm PDR} = 3.2\times10^{3}$ cm$^{-3}$, and a FUV field of $G_{o}=100$. For the J-shock/PDR combination, the remaining parameters that best fit the data are a shock velocity of 20 km/s, a shock beam filling factor of 0.87, and a PDR filling factor of 1.34. For the C-shock/PDR combination the best shock velocity is 10 km/s and the beam filling factors are 0.25 for the C-shock and 1.28 for the PDR component. The fraction of the individual line emissions is also very similar for both types of shocks. In both cases 100% of the CO(7-6) and 99% of the CO(6-5) intensity would arise from shock-excited gas. Of the CO(3-2) emission, 47% and 32% would arise from shocks in the C-shock/PDR model and J-shock/PDR model, respectively. For the H$_{2}$ emission, 94% and 93% of the S(2) intensity and 45% and 44% of the S(1) intensity would arise from shocks in the C-shock/PDR and J-shock/PDR model, respectively. For both model combinations only about 3% of the H$_{2}$ S(0) intensity would originate in shocks. The main difference between the results of the model combination is in the beam filling factor of the shocks.
The parameter solution for the shock models is very sensitive to a change in observed intensities, and more observations would be very helpful to provide better constraints. This is mainly due to the coarse grid of the shock model. Marginalization of the probability density distribution shows that the PDR solutions are strongly peaked with a 1 sigma range less than a step-size in the model grid for density and far-UV parameters. While the marginalized probability density distribution for the shock velocities also strongly peak at the best fit values with a 1 sigma range less than a step size, running the fitting routine with the CO(3-2) intensity increased by about 10% changes the densities of the C-shock/PDR model to a pre-shock density of $n = 2 \times 10^{5}$ cm$^{-3}$, a shock filling factor of 0.11, and a PDR density of $n_{\rm PDR} = 5.6\times10^{3}$ cm$^{-3}$. In contrast, J-shock/PDR solutions are hardly affected. The fractions of the line emission originating from shocks are quite robust. Given the uncertainty in the estimate of the IR continuum and the observed CO(3-2) intensity, for which we assumed a relative error of 50% in the fitting routine, the predicted intensities are not better than a factor of 2 in most cases. In fact, a change in shock velocity by 10 km/s can especially change the predicted \[O [I]{}\] intensities originating from shocks by an order of magnitude. Nevertheless, except for high-velocity J-shocks the \[O [I]{}\] intensities from shocks would always be smaller compared to the predicted PDR intensities.
We note that the observed \[C [I]{}\] 609 $\mu$m intensity is too weak by roughly an order of magnitude compared to the J-shock/PDR model result and by about a factor of about three compared to the C-shock/PDR model result.
Since there are no high spatial resolution observations available for the far-IR lines we can only compare the model predictions among each other. Including the beam filling factors, the models predict a fraction of $< 1$% of the \[O [I]{}\] 63 $\mu$m and 146 $\mu$m intensity from C-shocks or J-shocks. For our best fit shock/PDR model combinations, the entire \[O [I]{}\] emission is predicted to originate from PDRs. It is nevertheless interesting to notice that the predicted intensity of the \[O [I]{}\] 63 $\mu$m line is similar to the intensity in the H$_{2}$ S(0) line in shocks. However, both intensities are small compared to the intensity in the higher H$_{2}$ transitions.
We now compare the predicted far-IR line emission that likely emerges from the central $11''$ with the ISO observations. According to the CO(1-0) cut along the major axis of NGC 891 [e.g. @Sofue.Nakai.1993], about 20% of the line emission within the $75''$ (FWHM) central region originates from the central $11''$ (FWHM). If the ISO far-IR fine-structure lines scale similarly, as would be a reasonable assumption for the \[C [II]{}\] line, then the predicted line intensities from our PDR/shock model are only about a factor of 2 larger than the scaled ISO far-IR lines.
We can also just fit the PDR model grid to the ISO observations of the center of NGC 891. If the IR continuum would scale similar to the \[C [II]{}\] emission as observed by ISO then about 23% of the IR continuum from the entire galaxy would arise from the inner $75''$ (FWHM). As before, we also take into account that probably only 50% of the IR continuum might arise from PDRs (with the rest originating from cirrus) and that the \[O [I]{}\] 63 $\mu$m line is optically thick. The best PDR fit to the observed \[C [II]{}\], and \[O [I]{}\] 63 $\mu$m and 146 $\mu$m ISO observations and the IR continuum is obtained for a gas density of $n_{\rm PDR} = 1 \times 10^{3}$ cm$^{-3}$, a FUV field of $\log(G_{o}) = 2.2$, and a PDR beam filling factor of 0.07. The observed intensities and the intensities predicted by the PDR model are shown in Table \[tab:isopdrfit\]. Both, the estimated density and far-UV field are similar to the predictions we obtained for the combined shock/PDR model within the much smaller ZEUS beam. In contrast, the beam filling factor is much smaller than the one we obtained previously for the ZEUS observation, as would be expected for the large ISO beam, whose diameter is much larger than the thickness of the edge-on galactic disc. This small PDR beam filling factor within the ISO beam would correspond to a filling factor of 3.26 within the ZEUS beam. It thus appears that the PDR conditions are similar within a factor of a few ($<10$) within small region covered by ZEUS and the large region covered by ISO. If we apply the beam filling factor corresponding to the ZEUS beam (3.26) instead of the ISO beam filling factor, which translates into multiplying the predicted values in Table \[tab:isopdrfit\] by $3.26/0.07$, then the predicted intensities from the PDR are within factors of 2-3 for the \[C [I]{}\] 370 $\mu$m intensity observed with ZEUS and the CO(3-2) and H$_{2}$ S(0) intensities. The PDR model, however, fails to predict the mid-J CO intensities by more than 2 orders of magnitude and the H$_{2}$ S(2) emission by about 1 order of magnitude. In fact, the large observed CO(7-6)/CO(6-5) intensity ratio can hardly be modeled by a PDR at all. There is no single PDR model that can fit both the CO(7-6)/CO(6-5) and CO(7-6)/\[C [I]{}\] 370 $\mu$m intensity ratio (Fig. \[fig:pdr\]), clearly indicating the presence of an additional heating mechanism.
Energy Sources \[sec:esource\]
==============================
We have shown that the warm interstellar medium in the nuclear region of NGC 891 is likely heated by a combination of FUV radiation and shocks, with the mid-J CO transitions probably entirely due to shocks. Since NGC 891 is an isolated galaxy with only a moderate star formation rate (SFR) and no active galactic nucleus we do not consider shock excitation through powerful mechanism that are encountered in ULIRGs or galaxy interactions, where strong X-ray emission from an active nucleus, nuclear jets, or galaxy-wide gas flows likely drive internal or external shock fronts that are suggested by their strong H$_{2}$ emission. Instead, we have considered moderate internal energy sources as heating mechanisms for the PDRs, shocks, and turbulence.
NGC 891 has a modest SFR of 3.8 $M_{\odot}$/yr [@Popescu.Tuffs.Kylafis.2004], which could be responsible for the PDR excitation of the far-IR and submillimeter fine-structure lines. Sources of shocks and turbulence that could excite the H$_{2}$ and CO rotational line emission include cloud-cloud collisions, outflows from young stellar objects, or supernovae driven outflows. For example, the warm molecular gas that strongly emits mid-J CO lines in the circumnuclear ring in the Milky Way is likely heated by dissipation of MHD turbulence [@Bradford.Stacey.Nikola.2005]. Cloud-cloud collisions in the material that falls into the central region is possibly the source for the turbulence there. However, a 3 pc scale circumnuclear ring, like the one in the Milky Way, would not contribute significantly within our 500 pc scale ZEUS beam on NGC 891.
In the center of NGC 253 the mid-J CO emission is probably either due to the decay of turbulent shocks or due to cosmic ray heating [@HaileyDunsheath.Nikola.Stacey.2008]. Both mechanisms could provide enough energy to heat the warm molecular gas in NGC 253. The enhanced cosmic ray flux in the center of NGC 253 is likely due to the high supernova rate (SNRa) of 0.1 yr$^{-1}$ in the central $\sim100$ pc [@Bradford.Nikola.Stacey.2003]. Supernova outflows could also drive turbulence.
@Roussel.Helou.Hollenbach.2007 have studied H$_{2}$ emission in the SINGS sample, which is comprised mainly of “normal” galaxies and some LINERs/Seyferts. For the galaxies powered by star formation they found a good correlation between the combined power in the H$_{2}$ S(0), S(1), and S(2) lines and the strength of the 7.7 $\mu$m PAH emission. In contrast, the H$_{2}$/PAH ratio in the LINER/Seyfert galaxies shows a much larger scatter and the ratio is usually elevated for stronger H$_{2}$ emission. This suggests that if the combined power in the H$_{2}$ S(0), S(1), and S(2) lines is above a certain threshold, in relationship to the strength of the 7.7 $\mu$m PAH feature, then the H$_{2}$ lines are too bright to be excited exclusively within PDRs. For H$_{2}$ sources above this threshold, they suggest alternative sources of heat ranging from the X-rays from a nuclear engine to interactions with nuclear outflows via shocks. have compared the combined H$_{2}$ rotational lines with the 7.7 $\mu$m PAH feature along the plane of NGC 891 and found that the H$_{2}$/PAH ratios are near the top (threshold) values for star forming galaxies in the SINGS sample for regions within $\sim5$ kpc of the nucleus. These inner regions contain most ($>80$%) of the observed H$_{2}$ line luminosity for the galaxy so that most of the line emission is at the threshold between PDR and shock excited origins. However, for the outer regions of NGC 891 the H$_{2}$/PAH ratio appears to rise by a factor of order 2 which places these regions more into the shock excited regime as defined by @Roussel.Helou.Hollenbach.2007. The rise in the H$_{2}$/PAH ratio is due to a faster decline in the PAH emission compared with the H$_{2}$ emission in the outer galaxy, and might be due to a variety of effects including a lessened importance of star formation in the outer galaxy, or lowered metallicity in the outer regions of NGC 891. This later effect is a plausible cause for the enhanced H$_{2}$/PAH ratio seen for the low metallicity dwarf galaxy NGC 6822 in the @Roussel.Helou.Hollenbach.2007 sample.
For our modeled C-type shocks with a velocity of 10 km/s and a pre-shock gas density of $2\times10^{4}$ cm$^{-3}$ the total expected CO intensity (up to $J=20$) is $3.2\times10^{-4}$ erg/s/cm$^{2}$/sr [Table A1 in @Flower.PineauDesForets.2010]. The total intensity of the H$_{2}$ rotational transitions is a factor of 2 higher, and together with CO they provide basically the entire radiative cooling for the C-shock. Flux energy in the neutral oxygen line is less than 1% and therefore negligible. About 54% of the mechanical flow energy is converted into line radiation. Applying the derived beam filling factor of 0.25 for C-shocks, the total combined luminosity of H$_{2}$ and CO within the $11''$ beam is then $2.17\times10^{6} \ L_{\odot}$, which is roughly half the mechanical heating rate.
For our J-shock solution the total expected CO intensity (up to $J=20$) is $1.4\times10^{-3}$ erg/s/cm$^{2}$/sr [Table A2 in @Flower.PineauDesForets.2010], a factor of 4.3 higher than for the C-shocks. The strongest CO transitions are around $J_{\rm up}=16$, so the total CO intensity given above is a lower limit. The energy in all H$_{2}$ lines is a factor of about 8 higher than from all CO lines [@Flower.PineauDesForets.2010]. This results in a combined luminosity of all the H$_{2}$ and CO lines of $9.7\times10^{7} \ L_{\odot}$ within the $11''$ beam and including the shock filling factor of 0.87. Both lines account for about 82% of the cooling, and the combined luminosity is more than 40 times larger than for C-shocks. H$_{2}$O would account for an additional 15% cooling, and about 97% of the flow energy of this J-shock is converted into line radiation. As for the C-type shocks cooling via the neutral oxygen lines is negligible for J-shocks with the above parameters.
In NGC 891, the total SNRa is about 0.042 yr$^{-1}$ [@Strickland.Heckman.Colbert.2004]. Following @Loenen.Spaans.Baan.2008 and assuming an energy output of $10^{51}$ erg per supernova and a transfer efficiency of 10% the expected mechanical heating rate is $L_{\rm SN} = 8.34\times10^{9} \times 0.1 \times SNRa \ L_{\odot}$ [eq.4 in @Loenen.Spaans.Baan.2008], or $L_{\rm SN} = 3.5\times10^{7} \ L_{\odot}$. This is a factor of 10 larger than the required heating rate for the C-shocks, but about a factor of 3 less than required to explain the J-shocks.
Another possibility for the heating source are outflows from young stellar objects (YSOs). Using equation 2 of @Loenen.Spaans.Baan.2008, the ejected energy rate from YSOs can be estimated as $\frac{{\rm d}E_{\rm flow}}{{\rm d}t} \approx 3\times10^{29} r_{\rm pc}^{2} v_{\rm flow}^{3} n$ erg/s, where $v_{\rm flow}$ is the outflow velocity in km/s, $n$ the density of the surrounding medium in cm$^{-3}$, and $r_{\rm pc}$ the size of the bubble (in parsec) that is affected by the outflow. YSOs only affect the molecular ISM locally, so for a cloud with a radius of 0.1 pc [@Loenen.Spaans.Baan.2008] the resulting luminosity is $\sim16 \ L_{\odot}$, and thus more than $10^{5}$ YSO’s would be required within a $11''$ beam ($\sim$ 500 pc) to account for the line luminosity in the (C-type) shock-excited gas. Given the SFR of 3.8 $M_{\odot}$/yr [@Popescu.Tuffs.Kylafis.2004] and a lifetime of the YSO phase of $10^{5}-10^{6}$ yr [@Loenen.Spaans.Baan.2008] the approximate luminosity ranges from 6.1 to $61\times10^{6} \ L_{\odot}$, or 2.4 to 24 times the line luminosity from C-type shocks. @Lada.Lombardi.Alves.2010 suggest a relationship between the star-forming rate and the number of YSO based on observations of Galactic star-forming clouds: $N(YSO) = 4 \times 10^{6} \times SFR~M_{\odot}{\rm yr^{-1}}$. Applying this relationship to the SFR of the entire galaxy NGC 891 we would expect about $1.5\times 10^{7}$ YSO. NGC 891 is often thought of as an edge-on analog to the Milky Way. Withing the central $\approx 400$ pc of the Milky Way the SFR was estimated to be $\sim0.07 M_{\odot}~{\rm yr^{-1}}$ [@An.Ramirez.Sellgren.2011], which would correspond to about $3\times10^{5}$ YSO’s. If the energy transfer efficiency is similar to SN, YSO outflows could provide enough energy to drive C-type or J-type shock excitation.
Conclusions \[sec:concl\]
=========================
We have observed the CO(6-5), CO(7-6), and \[C [I]{}\] 370 $\mu$m lines from the nuclear region of NGC 891 using our submillimeter grating spectrometer ZEUS on the CSO. To analyze the data we have include observations of the CO(3-2) [@Mauersberger.Henkel.Walsh.1999] and H$_{2}$ S(0), S(1), S(2) , and IR continuum from the literature. We find the emission is best explained by a combination of PDRs and shocks. While J-shocks provide a slightly better fit than C-shocks the type of shocks is not conclusive.
The CO(7-6)/CO(6-5) intensity ratios are large for a PDR paradigm and would require extremely high far-UV and density values. In addition the CO ratio and the CO(7-6)/\[C [I]{}\] 370 $\mu$m ratio can not be fitted by a single PDR model. However, the mid-J CO ratios are consistent with shock-excited warm molecular gas, and our ZEUS observations together with Spitzer H$_{2}$ observations can be explained by a combined shock/PDR model. This result confirms recent studies of the H$_{2}$ S(0), S(1), and S(2) transitions with Spitzer , which required C-shocks to explain the higher H$_{2}$ rotational transitions. The best fitting shock models require a a pre-shock gas density of $2\times10^{4}$ cm$^{-3}$ and shock velocities of 10 km/s and 20 km/s and shock filling factors of 0.25 and 0.87 for C-shocks and J-shocks, respectively. The most likely energy sources for the shocks in the center of NGC 891 are supernovae and outflows from YSOs. While both types of shocks fit our data about equally well, from energy considerations it is significantly easier to provide the requisite mechanical energy for C-shock excitation of the lines than it is for J-shock excitation.
We have estimated the fraction of line emission originating from shocks and PDRs and find that for both types of shocks all of the CO(7-6) and 99% of the CO(6-5) intensity arises from shock-excited gas. For the CO(3-2) emission, about 47% and 32% would originate from C-shocks or J-shocks, respectively. Our model fits also suggest a larger fraction of the H$_{2}$ S(2) (94%) and S(1) (45%) emission from shocks than estimated by . Only about 3% of the H$_{2}$ S(0) emission would emerge from shock-excited gas.
The \[C [I]{}\] 370 $\mu$m emission arises from PDRs. When fitting the combined PDR/shock model to the ZEUS and Spitzer data we find a best solution for a PDR density of $3.2\times10^{3}$ cm$^{-3}$ and a FUV field of $G_{o} = 100$ for both types of shocks. The PDR filling factors are 1.28 and 1.34 for a combination with C-shocks and J-shocks, respectively. Our best PDR solution is consistent with \[O [I]{}\] 63 $\mu$m and 146 $\mu$m and \[C [II]{}\] observations on the scale of the larger ISO beam. Comparing the line predictions from our best fitting shock and PDR models shows that the \[O [I]{}\] emission arises almost entirely from PDRs.
We thank the CSO staff for their excellent support of the ZEUS observations. We also thank the anonymous referee whose comments led to a great improvement in the data analysis. This work was supported by NSF grants AST-0096881, AST-0352855, AST-0705256, and AST-0722220, and by NASA grants NGT5-50470 and NNG05GK70H.
[l c c c c]{} CO(6-5) & $11''$ & $31.2 \pm 4.0$ & 1.056E-5 & 3.403E-17\
CO(7-6) & $10''$ & $35.4 \pm 7.8$ & 1.902E-5 & 5.066E-17\
$[$CI$]$ 370$\mu$m & $10''$ & $65.6 \pm 7.8$ & 3.563E-5 & 9.490E-17\
[l c c c c c]{} H$_{2}$ S(0) & Spitzer & $22.3''\times11.1''$ & & 3.39E-5 & 1.97E-16\
H$_{2}$ S(1) & Spitzer & $4.7''\times11.3''$ & & 9.11E-5 & 1.14E-16\
H$_{2}$ S(2) & Spitzer & $4.7''\times11.3''$ & & 4.71E-5 & 5.88E-17\
$[$CI$]$ 609$\mu$m & CSO & $14.55''$ & $11.5 \pm 1.0$ & 1.4E-6 & 7.9E-18\
CO(3-2) & HHT & $21''$ & $24 \pm 2$ & 1.017E-6 & 1.194E-17\
[c c c c c c c]{} 0.25 & 2.0E+4 & 10.0 & 1.28 & 3.16E+3 & 100 & 7.5\
[c c c c c c c]{} 0.87 & 2.0E+4 & 20.0 & 1.34 & 3.16E+3 & 100 & 5.0\
[l c c c c c c c]{} CO(7-6) & 1.74E-5 & 1.00 & 1.279E-5 & 1.508E-8 & 1.00 & 1.693E-5 & 1.580E-8\
CO(6-5) & 1.06E-5 & 0.99 & 1.091E-5 & 1.016E-7 & 0.99 & 9.773E-6 & 1.064E-7\
CO(3-2) & 2.24E-6 & 0.47 & 1.469E-6 & 1.663E-6 & 0.32 & 7.774E-7 & 1.743E-6\
H$_{2}$ S(2) & 4.72E-5 & 0.94 & 3.221E-5 & 2.583E-6 & 0.93 & 3.411E-5 & 2.707E-6\
H$_{2}$ S(1) & 8.07E-5 & 0.45 & 3.964E-5 & 4.987E-5 & 0.44 & 3.935E-5 & 5.226E-5\
H$_{2}$ S(0) & 4.17E-5 & 0.03 & 1.338E-6 & 2.299E-5 & 0.03 & 1.224E-6 & 2.410E-5\
$[$OI$]$63$\mu$m & & & 2.725E-6 & 2.809E-4 & & 1.399E-6 & 2.943E-4\
$[$OI$]$146$\mu$m & & & 2.477E-7 & 2.788E-5 & & 3.935E-8 & 2.922E-5\
$[$CI$]$370$\mu$m & 3.27E-5 & & & 3.138E-5 & & & 3.298E-5\
$[$CI$]$609$\mu$m & 1.89E-6 & & & 9.647E-6 & & & 1.011E-5\
Total IR & 1.21E-1 & & & 6.650E-2 & & & 6.969E-2\
$[$CII$]$ & & & & 6.106E-4 & & & 6.399E-4\
[l c c]{} $[$CII$]$ & 3.586E-5 & 3.639E-5\
$[$OI$]$ 63$\mu$m & 1.377E-5 & 1.437E-5\
$[$OI$]$ 146$\mu$m & 1.802E-6 & 1.561E-6\
Total IR & 6.266E-3 & 6.525E-3\
$[$CI$]$ 370$\mu$m & & 1.249E-6\
$[$CI$]$ 609$\mu$m & & 5.031E-7\
CO(7-6) & & 3.998E-11\
CO(6-5) & & 2.684E-10\
CO(3-2) & & 2.695E-8\
H$_{2}$ S(2) & & 8.030E-8\
H$_{2}$ S(1) & & 1.947E-6\
H$_{2}$ S(0) & & 5.511E-7\
![ZEUS \[C [I]{}\] and CO(7-6) (top), and CO(6-5) (bottom) spectra.[]{data-label="fig:spec"}](ngc891_co76_ci_2006.png){width="90.00000%"}
![ZEUS \[C [I]{}\] and CO(7-6) (top), and CO(6-5) (bottom) spectra.[]{data-label="fig:spec"}](ngc891_co65_2006.png){width="90.00000%"}
![Plot of parameter range of PDR model (FUV field, $G_{o}$, in units of Habing field $1.6\times10^{3}$ erg/s/cm$^{2}$ versus density) depicting the observed CO(7-6)/CO(6-5) (dashed) and CO(7-6)/\[C [I]{}\] 370 $\mu$m (dashed-dotted) intensity ratios. The uncertainties are indicated by dotted lines.[]{data-label="fig:pdr"}](ngc891_co76_65_ci370_pdr_incl_optdepth.png){width="90.00000%"}
| {
"pile_set_name": "ArXiv"
} |
---
address: |
$^1$ Steklov Mathematics Institute, Fontanka 27, 191023, St. Petersburg, Russia\
$^2$ DESY Theory Group, Notkestrasse 85, D-22603, Hamburg, Germany\
February 2006
author:
- 'Andrei G. Bytsko$^{1}$ and J[ö]{}rg Teschner$^{2}$'
title: |
Quantization of models with non–compact quantum group symmetry.\
Modular XXZ magnet and lattice sinh–Gordon model.
---
Introduction
============
Motivation
----------
One of the main motivations for studying integrable lattice models is their role as a lattice regularization of quantum field theories in continuous space–time. Integrable nonlinear sigma models are of particular interest, and recently there has been a growing interest in nonlinear sigma models with [*non–compact*]{} target spaces. This interest is motivated by possible applications to string theory on curved space–times in general, and to gauge theories via the AdS–CFT correspondence in particular.
However, the quantization and the solution of such non–compact nonlinear sigma models still represents a major challenge for the field of integrable models. Compared to the better understood nonlinear sigma models with [compact]{} target spaces one may expect important qualitative differences, which make it problematic to apply the known techniques from the compact cases to the sigma models with non–compact target spaces. This point is exemplified by the relation between the Wess–Zumino–Novikov–Witten (WZNW) models associated to compact and non–compact symmetric spaces respectively. The solution of the latter is possible [@T1], but it is considerably more difficult than the solution of WZNW models associated to compact groups.
In more general sigma models one can not hope to find the powerful Kac–Moody symmetries of the WZNW models but the integrable structure may still survive. In order to enter the next level of complexity one may therefore try to exploit the integrability of some of these models. Turning to a new class of models it is always advisable to look for the simplest member which still exhibits most of the new qualitative features. In the case of the conformal WZNW models it has turned out that Liouville theory already displays many of the relevant differences which distinguish the non–compact WZNW models from rational conformal field theories [@T2]. Moving outside of the class of sigma models soluble thanks to Kac–Moody or similarly powerful [*chiral*]{} symmetries it seems natural to look for a useful counterpart of Liouville theory within this larger class of models.
A natural candidate for such a model exists: the sinh–Gordon model. Indeed, there is some evidence [@ZZ; @Lu] that the sinh–Gordon model can be seen as a “deformation” of Liouville theory which preserves its integrable structure when the conformal symmetry is lost. While there certainly exists a good basis for the study of the sinh–Gordon model in infinite volume — S-matrix and the form factors are known [@VG; @FMS; @KMu; @BL; @Le] and the basic ingredients of the QISM approach were developed [@S1] — there does not seem to exist a sytematic approach to the quantization and solution of the sinh–Gordon model in [*finite*]{} spatial volume yet. Part of the problem is due to the usual divergencies and ordering problems of quantum (field) theory. But the other part of the problem seems to be closely related to the non–compactness of the target space in the sinh–Gordon model.
Our main motivation behind the present project was therefore to find an integrable lattice regularization for the sinh–Gordon model. This not only tames the usual short distance singularities, it will also allow us to take care of the troubles from non–compactness of the target space in a mathematically well–defined framework. One particular feature that directly follows from the non–compactness of the target space will be the failure of the usual algebraic Bethe ansatz method [@F1] for the model at hand. This failure means that we will have to use the more general separation of variables method [@Sk2; @Sk3; @Sm] instead.
Lattice sinh–Gordon and the modular XXZ magnet
----------------------------------------------
A quantum integrable system is a quantum system $({{\mathcal H}},{{\mathcal A}},{{\mathsf H}})$, with Hilbert space ${{\mathcal H}}$, algebra of observables ${{\mathcal A}}$, Hamiltonian ${{\mathsf H}}$, in which there exists a set ${{\mathcal Q}}=\{{{\mathsf T}}_0,{{\mathsf T}}_1,\dots\}$ of self–adjoint operators such that $$\begin{aligned}
{\rm (A)} \quad & [{{\mathsf T}},{{\mathsf T}}']=0\quad\forall\ {{\mathsf T}},\, {{\mathsf T}}'\in{{\mathcal Q}},\\
{\rm (B)} \quad & [{{\mathsf T}},{{\mathsf H}}]=0\quad\forall\ {{\mathsf T}}\in{{\mathcal Q}},\\
{\rm (C)} \quad & {\rm if}\;\,[{{\mathsf T}},{{\mathsf O}}]=0
\;\,{\rm for~all}\;\,{{\mathsf T}}\in{{\mathcal Q}},\;\,
{\rm then}\;\,{{\mathsf O}}={{\mathsf O}}({{\mathcal Q}}).\end{aligned}$$ Property (C) expresses completeness of the set ${{\mathcal Q}}$ of integrals of motion. It is equivalent to the statement that the spectrum of ${{\mathcal Q}}$ is non–degenerate, i.e., that simultaneous eigenstates of ${{\mathsf T}}_k$, $k\in{{\mathbb Z}}^{\geq 0}$ are uniquely determined by the tuple of their eigenvalues. We will consider the so–called one–dimensional lattice models for which one has $${{\mathcal H}}={{\mathcal K}}^{{\otimes}{{\rm N}}}, \qquad {{\mathcal A}}={{\mathcal B}}^{{\otimes}{{\rm N}}},$$ with one copy of Hilbert space ${{\mathcal K}}$ and algebra of local observables ${{\mathcal B}}$ being associated to each of the ${{\rm N}}$ sites of a one–dimensional lattice.
The quantum inverse scattering method (QISM) [@FST; @F1] goes a long way towards the construction of large classes of quantum integrable models of this type. In this framework one usually characterizes ${{\mathcal K}}$ as a representation of a Hopf algebra ${{\mathfrak U}}$ of “symmetries”, and ${{\mathcal B}}$ is generated from the operators which represent the elements of ${{\mathfrak U}}$ on ${{\mathcal K}}$. It is clear that the representation theoretic properties of ${{\mathcal K}}$ will influence the physical properties of the resulting integrable model decisively. Good control over these properties will be crucial in the construction and solution of such models.
In general it is a highly nontrivial problem to find the “right" representation ${{\mathcal K}}$ which leads to a useful lattice regularization of a particular quantum field theory. We will here propose a particular choice for ${{\mathcal K}}$ which will lead to a lattice model with particularly nice properties, and which will be shown to yield the sinh–Gordon Hamiltonian density in the continuum limit of the corresponding classical lattice model. The representations in question will be representations of the non–compact real form ${{\cal U}_q({{\mathfrak s}{\mathfrak l}}(2,{{\mathbb R}}))}$ of ${{\mathcal U}}_q({{\mathfrak s}{\mathfrak l}}_2)$ which have been studied in [@PT1; @F3; @PT2; @BT].
The non–compactness of the target space will be reflected in the infinite–dimensionality of the representation ${{\mathcal K}}$. It is furthermore worth noting that the same representations were previously found to reflect a key internal structure of Liouville theory [@PT1; @T2]. In view of the existing evidence [@ZZ; @Lu] for the connection between Liouville theory and the sinh–Gordon model, it is quite natural that the same class of representations appears in our lattice version of the sinh–Gordon model as well.
The corresponding representations possess a remarkable [*duality*]{} — they are simultaneously representations of ${\cal U}_q({{\mathfrak s}{\mathfrak l}}_2)$ and ${\cal U}_{\tilde{q}}({{\mathfrak s}{\mathfrak l}}_2)$, where $q=e^{i \pi b^2}$ and $\tilde{q}=e^{i \pi b^{-2}}$. One may therefore view them [@F3] as representations of the [*modular double*]{} ${\cal U}_q({{\mathfrak s}{\mathfrak l}}_2){\otimes}{\cal U}_{\tilde{q}}({{\mathfrak s}{\mathfrak l}}_2)$ (see also [@KLS; @BT]). The parameter $b$ turns out to be proportional to the coupling constant $\beta$ of the sinh–Gordon model. The self–duality of our representations will be directly related to the self–duality of the sinh–Gordon model under $b{\to}b^{-1}$ which was previously observed in its scattering theory. The importance of this self–duality for our analysis can hardly be over–emphasized.
It turns out that there is a close relative of our lattice sinh–Gordon model which is simpler in some respects. This integrable lattice model can be seen as a non–compact counterpart of the XXZ model with spins in infinite–dimensional representations of the modular double. We will refer to this model as the [*modular*]{} XXZ magnet. As some technical issues are simpler in the case of the modular XXZ magnet, we will first construct the latter model before we turn to the lattice sinh–Gordon model. In any case, it seems to us that the study of the modular XXZ magnet is of interest in its own right. We note in particular that despite the different underlying representation theory, our model has many structural similarities with the non–compact XXX type magnet based on infinite–dimensional highest weight representations of ${{\mathfrak s}{\mathfrak l}}_2$ which was studied in [@DKM; @KM]. The latter model plays an important role in high energy QCD [@Li; @FK1].
Plan of the paper
-----------------
To make our paper accessible to a reasonably wide audience, we presented the general description of our approach, the main definitions and results in the main body of the paper and collected more technical developments in Appendices. The article is organized as follows.
In Section 2 we define the modular XXZ magnet in terms of the representations ${{\mathcal P}}_s$ describe its Hilbert space of states, construct the corresponding fundamental R–operator ${{\mathsf R}}(u)$, and discuss the construction of its Hamiltonian and the set of integrals of motion ${{\mathcal Q}}$.
The same is done for the lattice sinh–Gordon model in Section 3. We show that the Hamiltonian density of the sinh–Gordon model is recovered in the continuum limit of the corresponding classical model. We also show that the algebraic Bethe ansatz fails due to the non–compactness of the target space in the sinh–Gordon model.
An important first step towards the solution of these models is taken in Section 4. We refine the spectral problem for the integrals of motion ${{\mathsf T}}_k$ by constructing the Q–operator ${{\mathsf Q}}(u)$ which is related to the ${{\mathsf T}}_k$ via the so–called Baxter equation. Analyzing the properties of ${{\mathsf Q}}(u)$, we derive a set of conditions for its eigenvalues $q_t(u)$ which can be seen as quantization conditions and which replace the usual Bethe ansatz equations in our models. The self–duality of our representations furthermore allows us to derive the so–called quantum Wronskian relation for the sinh–Gordon model with odd ${{\rm N}}$, which encodes valuable additional information about the spectrum.
In order to show that the conditions found in Section 4 are also sufficient to characterize the spectrum we apply the separation of variables approach to our models in Section 5.
Section 6 contains concluding remarks on the conditions which characterize the spectrum of our models, the continuum limit, and the relation with the lattice and continuum versions of Liouville theory. We observe in particular that our results are consistent with the results and conjectures of [@Za; @Lu] on the continuum sinh–Gordon model in a nontrivial way.
Appendices contain necessary technical details. Appendix A collects the relevant information on the special functions that we use. Appendix B discusses the precise mathematical nature of the self–duality $b{\to}b^{-1}$ of the representations that we use. Appendic C contains some important technical results on the structure of the monodromy matrix. Appendix D is devoted to the construction of the fundamental R–operator, the key object for the construction of local integrals of motion for the lattice models. Appendix E contains details on the derivation of the properties of the Q–operator ${{\mathsf Q}}(u)$.
Modular XXZ magnet {#XXZ}
==================
In this section we will begin to develop the QISM for the [*modular*]{} XXZ magnet — an XXZ type non–compact spin chain, which has ${{\cal U}_q({{\mathfrak s}{\mathfrak l}}(2,{{\mathbb R}}))}$ as a quantum symmetry.
Quantum group symmetry ${{\cal U}_q({{\mathfrak s}{\mathfrak l}}(2,{{\mathbb R}}))}$
------------------------------------------------------------------------------------
Let $q=e^{i\gamma}$, $\gamma=\pi b^2$, $b\in (0,1)$. We will also use the notation $Q=b+b^{-1}$.
The quantum group ${{\mathcal U}}_q({{\mathfrak s}{\mathfrak l}}_2)$ is a Hopf algebra with generators $E$, $F$, $K$, $K^{-1}$ satisfying the relations $$\label{def}
KE=qEK,\qquad KF=q^{-1}FK, \qquad
[E,F]={{\textstyle \frac{1}{q-q^{-1}} }}(K^2-K^{-2}) \,$$ and equipped with the following co–product: $$\label{De}
\begin{aligned}
{\Delta}(E)=&\, E{\otimes}K+K^{-1}{\otimes}E \,, \\
{\Delta}(F)=&\, F{\otimes}K+K^{-1}{\otimes}F \,,
\end{aligned}
\qquad {\Delta}(K)=K{\otimes}K \,.$$ The relevant real form of ${{\mathcal U}}_q({{\mathfrak s}{\mathfrak l}}_2)$ is ${{\cal U}_q({{\mathfrak s}{\mathfrak l}}(2,{{\mathbb R}}))}$, which is defined by the following star–structure: $$\label{star}
K^*=K, \qquad E^*=E, \qquad F^*=F \,.$$ The center of ${{\cal U}_q({{\mathfrak s}{\mathfrak l}}(2,{{\mathbb R}}))}$ is generated by the $q$–Casimir element: $$\label{Cas}
C = (2 \sin\gamma)^2 \, FE - qK^2 - q^{-1}K^{-2} + 2 \,,
\qquad C^*=C \,.$$
Representations ${\cal P}_s$ — algebra of observables
-----------------------------------------------------
A one–parameter family of unitary representations ${{\mathcal P}}_s$ of ${{\cal U}_q({{\mathfrak s}{\mathfrak l}}(2,{{\mathbb R}}))}$ can be constructed from a pair of self–adjoint operators ${{\mathsf p}}$ and ${{\mathsf x}}$ on $L^2({{\mathbb R}})$ which satisfy as follows: $$\label{EFK1}
\begin{aligned}
\pi_{s}(E)\;\equiv\;{{\mathsf E}}_{s}\;=\;&e^{+\pi b {{\mathsf x}}} \,
\frac{\cosh\pi b({{\mathsf p}}-s)}{\sin\gamma} \, e^{+\pi b {{\mathsf x}}} \,, \\
\pi_{s}(F)\;\equiv\;{{\mathsf F}}_{s}\;=\;&
e^{-\pi b {{\mathsf x}}} \, \frac{\cosh\pi b({{\mathsf p}}+s)}{\sin\gamma} \,
e^{-\pi b {{\mathsf x}}} \,,
\end{aligned}
\qquad \pi_{s}(K)\;\equiv{{\mathsf K}}_s\;=\;e^{-\pi b{{\mathsf p}}} \,.$$ For this representation $$\label{cps}
{{\mathsf C}}_s\;\equiv\; \pi_{s} ( C ) = 4 \cosh^2 \pi b s \,.$$ It is remarkable and important that the operators ${{\mathsf E}}_s$, ${{\mathsf F}}_s$ and ${{\mathsf K}}_s$ are [*positive*]{} self–adjoint. Indeed, the representations ${{\mathcal P}}_s$ are the [*only*]{} “reasonable" representations of ${{\cal U}_q({{\mathfrak s}{\mathfrak l}}(2,{{\mathbb R}}))}$ which have this property. This property will play a key role in much of the following developments. It will in particular ensure seld–adjointness of operators such as the Hamiltonian and the integrals of motion. It is also the mathematical basis for the self–duality of the representations ${{\mathcal P}}_s$, as shown in Appendix \[posapp\] (see, in particular, eq. [(\[bdual\])]{}).
The lattice model that we are about to define will have one of the representations ${{\mathcal P}}_s$ attached to each site of the one–dimensional lattice. This means that we take $${{\mathcal H}}\,=\,\bigl( L^2({{\mathbb R}}) \bigr)^{{\otimes}{{\rm N}}}$$ as the Hilbert space of our model, and let $$\widehat{{{\mathcal A}}}\,=\,\big(\pi_s({{\mathcal U}})\big)^{{\otimes}{{\rm N}}},\quad {{\mathcal U}}\equiv{{\cal U}_q({{\mathfrak s}{\mathfrak l}}(2,{{\mathbb R}}))}$$ be a set of generators for our algebra of observables. Note that the operators in $\widehat{{{\mathcal A}}}$ are all unbounded, but there exists a basis for $\widehat{{{\mathcal A}}}$ whose elements are positive self–adjoint (see Appendix \[posapp\]). The latter fact allows us to construct large classes of [*non–polynomial*]{} operator functions of the generators in $\widehat{{{\mathcal A}}}$ via standard functional calculus for self–adjoint operators and/or pseudo–differential operator calculus.
Integrals of motion
-------------------
As the next step we shall introduce our main ansatz for the set ${{\mathcal Q}}$ of integrals of motion using the usual scheme of the QISM. To this aim let us assemble the generators of $\widehat{{{\mathcal A}}}$ into the following L–matrix acting on ${\mathbb C}^2 {\otimes}{\cal P}_s$: $$\label{Lxxz}
L^{{{\rm{\scriptscriptstyle}XXZ}}} (u)
= \left( \begin{array}{cc}
e^{\pi b u} {{\mathsf k}}_s - e^{-\pi b u} {{\mathsf k}}_s^{-1} &
i \, e^{\pi b u} \, {{\mathsf f}}_s \\
i \, e^{-\pi b u} \, {{\mathsf e}}_s &
e^{\pi b u} {{\mathsf k}}_s^{-1} - e^{-\pi b u} {{\mathsf k}}_s
\end{array} \right) \,, \qquad u\in {\mathbb C}.$$ In the definition of $L^{{{\rm{\scriptscriptstyle}XXZ}}} (u)$ we have used the rescaled generators ${{\mathsf e}}_s$, ${{\mathsf f}}_s$, ${{\mathsf k}}_s$ which are defined by $$\label{efk}
{{\mathsf e}}_s = (2 \sin\gamma) \, {{\mathsf E}}_s \,, \quad
{{\mathsf f}}_s = (2 \sin\gamma) \, {{\mathsf F}}_s \,, \quad
{{\mathsf k}}_s = {{\mathsf K}}_s \,.$$ Occasionally we will omit the superscript ${{\rm{\scriptscriptstyle}XXZ}}$ for the sake of brevity. The defining relations (\[def\]) and (\[De\]) of ${{\cal U}_q({{\mathfrak s}{\mathfrak l}}(2,{{\mathbb R}}))}$ are equivalent to $$\begin{aligned}
\label{rLL}
R_{12}(u) \, L_{13} (u + v) \, L_{23} (v) &=&
L_{23} (v) \, L_{13} (u + v) \, R_{12}(u) \,,\\[0.5mm]
\label{dLL}
\bigl(\mathrm{id} \otimes \Delta \bigr) L^{{\pm}} &=&
L^{{\pm}}_{13} \, L^{{\pm}}_{12} \,,\end{aligned}$$ where $L^{{\pm}}$ arise in the decomposition $$\label{Lpm}
L (u) = e^{\pi b u} \, L^{+} - e^{-\pi b u} \, L^{-} \,,$$ and the auxiliary R–matrix is given by $$\label{R12}
R(u) =
\left( \begin{array}{cccc}
\sinh \pi b(u + i b) & & & \\ [-1mm]
& \sinh \pi b u &
i\sin\pi b^2 \, e^{ \pi b u} & \\
& i\sin\pi b^2 \, e^{- \pi b u}
& \sinh \pi b u & \\ [-1mm]
& & & \sinh\pi b(u + i b)
\end{array} \right) \,.$$ Out of the L–matrices we may then construct the monodromy ${{\mathsf M}}(u)$, $$\label{Mono}
{{\mathsf M}}(u) \,\equiv\, \left(\begin{matrix} {{\mathsf A}}(u) & {{\mathsf B}}(u) \\
{{\mathsf C}}(u) & {{\mathsf D}}(u) \end{matrix}
\right)\,\equiv\,L_{{\rm N}}(u) \cdot \ldots \cdot L_2(u) \cdot L_1(u) \,.$$ Of particular importance is the one–parameter family of operators: $$\label{Monotr}
{{\mathsf T}}(u) \,= \,\mathrm{tr}\, \bigl({{\mathsf M}}(u)\bigr)=
{{\mathsf A}}(u)+{{\mathsf D}}(u)\,.$$ The trace in (\[Monotr\]) is taken over the auxiliary space, which is ${\mathbb C}^2$ for the models we consider.
The operators ${{\mathsf T}}_m$ which appear in the expansion $$\label{Texp}
{{\mathsf T}}(u)\,=\,e^{\pi b{{\rm N}}u}\,
\sum_{m=0}^{{{\rm N}}}(-e^{-2\pi b u})^m\,{{\mathsf T}}_m,$$ are positive self–adjoint and mutually commuting, $[{{\mathsf T}}_m,{{\mathsf T}}_n]=0$.
Commutativity of ${{\mathsf T}}_m$ follows by the standard argument from the relation (\[rLL\]); the proof of their positivity and self–adjointness is given in Appendix \[monoapp\].
Let us define the set of commuting charges as ${{\mathcal Q}}=\{{{\mathsf T}}_0,{{\mathsf T}}_1,\dots,{{\mathsf T}}_{{\rm N}}\}$.
\[posadj\] Self–adjointness of the operators ${{\mathsf T}}_m$, $m=0,\dots, {{\rm N}}$ ensures the existence of a joint spectral decomposition for the family ${{\mathcal Q}}$.
Let us emphasize that the crucial positivity of the operators ${{\mathsf T}}_m$ is a direct consequence of the fact that the generators $E$, $F$ and $K$ of ${{\cal U}_q({{\mathfrak s}{\mathfrak l}}(2,{{\mathbb R}}))}$ are represented by [*positive*]{} operators in the representations ${{\mathcal P}}_s$. This makes clear why these representations are particularly well–suited for defining non–compact analogues of the XXZ spin chains. We will later make a similar observation in the lattice sinh–Gordon model.
Fundamental R–operator and Hamiltonian
--------------------------------------
Our next aim is to construct a [*local*]{} Hamiltonian which commutes with the elements of ${{\mathcal Q}}$. We will adapt the approach from [@FTT] to the case at hand. The main ingredient of this approach is the so–called fundamental R–operator corresponding to (\[Lxxz\]). This operator, ${{\mathsf R}}^{{\rm{\scriptscriptstyle}XXZ}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(u)$, acts on ${{\mathcal P}}_{s_{{\mathfrak 2}}} \otimes {{\mathcal P}}_{s_{{\mathfrak 1}}}$ and is supposed to satisfy the commutation relations $$\label{RLL}
{{\mathsf R}}_{{{\mathfrak 2}}{{\mathfrak 3}}}(u) \, L_{{{\mathfrak 1}}{{\mathfrak 3}}} (u + v) \, L_{{{\mathfrak 1}}{{\mathfrak 2}}} (v) =
L_{{{\mathfrak 1}}{{\mathfrak 2}}} (v) \, L_{{{\mathfrak 1}}{{\mathfrak 3}}} (u + v) \, {{\mathsf R}}_{{{\mathfrak 2}}{{\mathfrak 3}}}(u) \,.$$ For our purposes it will be sufficient[^1] to deal with ${{{\mathsf R}}}(u) \equiv {{{\mathsf R}}}_{s s}^{{\rm{\scriptscriptstyle}XXZ}}(u)$ acting on ${{\mathcal P}}_{s} \otimes {{\mathcal P}}_{s}$.
Let the operator ${{\mathsf R}}(u)$ be defined by the formula $$\label{Rsimpledef}
{{{\mathsf R}}}(u) = {{\mathsf P}}\, w_b(u + {{\mathsf s}}) \, w_b(u - {{\mathsf s}}) ={{\mathsf P}}\, D_{u}({{\mathsf s}}) \,,$$ where ${{\mathsf P}}$ is the operator which just permutes the two tensor factors in ${{\mathcal P}}_{s} \otimes {{\mathcal P}}_{s}$, and ${{\mathsf s}}$ is the unique positive self–adjoint operator such that $$\label{tensC}
4 \cosh^2 \pi b \, {{\mathsf s}}\,=\,
(\pi_{s} {\otimes}\pi_{s}) \Delta (C) \,$$ The special functions $w_b(x)$ and $D_{\alpha}(x)$ are defined in Appendix \[Qdil\].
\[Rthm\] The operator ${{\mathsf R}}(u)$ satisfies the equation [(\[RLL\])]{} where $L(u)$ is given by (\[Lxxz\]).
The proof of this theorem is given in Appendix \[proofSH\], where the construction of the operator ${{\mathsf R}}^{{\rm{\scriptscriptstyle}XXZ}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(u)$ is presented for the general case, $s_{{\mathfrak 1}}\neq s_{{\mathfrak 2}}$, see equation [(\[Rxxz\])]{}.
The operator ${{\mathsf R}}(u)$ has the following further properties $$\begin{aligned}
\label{RGGa}
\text{regularity} \quad& {{{\mathsf R}}}(0) = {{\mathsf P}}\,,\\
\label{RGGb}
\text{reflection property} \quad&
{{{\mathsf R}}}(-u) ={{\mathsf P}}\, {{{\mathsf R}}}^{-1}(u) \, {{\mathsf P}}\,,\\
\label{RGGc}
\text{unitarity} \quad&
{{{\mathsf R}}}^* (u) = {{{\mathsf R}}}^{-1} (u) \quad \text{for } u\in{{\mathbb R}}\,,\end{aligned}$$ which follow from the properties of $w_b(u)$ and $D_{\alpha}(x)$ listed in Appendix \[Qdil\].
The regularity condition (\[RGGa\]) allows us to apply the standard recipe [@FTT] of the QISM in order to construct a Hamiltonian with local (nearest neighbour) interaction of sites: $$\label{HRxxz}
\begin{aligned}
{{\mathsf H}}^{{\rm{\scriptscriptstyle}XXZ}}&=
\frac{i}{\pi b} \, {{\mathsf U}}^{-1} \, \left[
\frac{\partial}{\partial u} \mathrm{tr}_a \,
\bigl( {{\mathsf R}}_{a {{\rm N}}}(u) \cdot \ldots \cdot {{\mathsf R}}_{a 2}(u)
\cdot {{\mathsf R}}_{a 1}(u) \bigl) \right]_{u=0} \\
&= \sum_{n=1}^{{{\rm N}}} \frac{i}{\pi b} \partial_u
D_u({{\mathsf s}}_{n,n+{{\mathfrak 1}}}) \Bigm|_{u=0} = \sum_{n=1}^{{{\rm N}}}
H^{{\rm{\scriptscriptstyle}XXZ}}_{n,n+1} \,.
\end{aligned}$$ We are using the following notation: We identify ${{\mathsf s}}_{{{\rm N}},{{\rm N}}+1}\equiv {{\mathsf s}}_{{{\rm N}},1}$, the cyclic shift operator ${{\mathsf U}}$ is defined by ${{\mathsf U}}\, f(x_{{\mathfrak 1}},x_{{\mathfrak 2}},\ldots,x_{{\rm N}}) = f(x_{{\mathfrak 2}},\ldots,x_{{\rm N}},x_{{\mathfrak 1}})$, and the subscript $a$ stands for an auxiliary copy of the space ${\cal P}_s$. The trace operation is defined for an operator ${{\mathsf O}}: {\cal P}_s \mapsto {\cal P}_s$ in the usual way: if the integral kernel of ${{\mathsf O}}$ in the momentum representation is given by $O(k|k')$, then $\mathrm{tr}\, {{\mathsf O}}= \int_{-\infty}^\infty dk\, O(k|k)$. According to this definition we have .
Substituting the integral representation (\[Dint\]) for $D_u(x)$ into (\[HRxxz\]), we obtain the following local Hamiltonian density $$\label{Hphi}
H^{{\rm{\scriptscriptstyle}XXZ}}_{n,n+1} =
- \frac{1}{\pi} \, \int\limits_{{{\mathbb R}}+i0} dt\,
\frac{\cos (2 b t\, {{\mathsf s}}_{n,n+1} )}{\sinh t \, \sinh{b^2 t} } \,.$$
It may then be shown in the usual manner [@FTT] that ${{\mathsf H}}$ commutes with the trace of the monodromy matrix, ${{\mathsf T}}(u)$, which means that $$\label{HTcomm}
[\,{{\mathsf H}}\,,\,{{\mathsf T}}_k\,]\,=\,0,\;\;{\rm for}\;\;k=0,\dots,{{\rm N}}.$$
As in any quantum mechanical system, the fundamental problem that we would like to solve is the problem to determine the spectrum of ${{\mathsf H}}$. However, thanks to the commutativity [(\[HTcomm\])]{} it seems promising to first solve the following\
[**Auxiliary Spectral Problem:**]{} [*Find the spectrum of the operator ${{\mathsf T}}(u)$, i.e., the joint spectral decomposition for the family of operators*]{} ${{\mathcal Q}}=\{{{\mathsf T}}_0,\dots,{{\mathsf T}}_{{\rm N}}\}$.\
Simple counting of the degrees of freedom suggests that the spectrum of ${{\mathcal Q}}$ may be simple, i.e., that an eigenstate $\Psi_t$ of ${{\mathsf T}}(u)$, $${{\mathsf T}}(u)\, \Psi_t \,=\, t(u) \, \Psi_t \,,$$ is uniquely characterized by the eigenvalue $t(u)\,=\,e^{\pi b{{\rm N}}u}\,\sum_{m=0}^{{{\rm N}}}(-e^{-2\pi bu})^m\,t_m$. This would imply that ${{\mathsf H}}={{\mathsf H}}({{\mathcal Q}})$, so that the solution to the Auxiliary Spectral Problem also yields the spectral decomposition of ${{\mathsf H}}$.
Classical limit {#Hxxz}
---------------
Let us discuss the classical limit of the quantum Hamiltonian (\[Hphi\]). So far we have been working in the units where the Planck constant $\hbar$ was chosen to be unity. In order to recover it explicitly, we have to make the following rescaling $$\label{hbar}
b^2 \to \hbar \, b^2 \,,\quad
{{\mathsf p}}\to \hbar^{-\frac{1}{2}} {{\mathsf p}}\,, \quad
{{\mathsf x}}\to \hbar^{-\frac{1}{2}} {{\mathsf x}}\,, \quad
{{\mathsf s}}\to \hbar^{-\frac{1}{2}} {{\mathsf s}}\,, \quad
s \to \hbar^{-\frac{1}{2}} s \,,$$ so that we have $q=e^{i \hbar \gamma}$, $\gamma= \pi b^2$. The operators ${{\mathsf e}}_s$, ${{\mathsf f}}_s$, ${{\mathsf k}}_s$ are not affected by the procedure (\[hbar\]). In the limit $\hbar \to 0$ they become classical variables ${{\mathsf e}}$, ${{\mathsf f}}$, ${{\mathsf k}}$ with the following Poisson brackets obtained by the correspondence principle, $[\,,] \to -i \hbar \{\,,\}$ $$\label{Pbr}
\{ {{\mathsf e}}, {{\mathsf k}}\}= \gamma \,{{\mathsf k}}{{\mathsf e}}\,, \quad
\{ {{\mathsf f}}, {{\mathsf k}}\}= - \gamma \,{{\mathsf k}}{{\mathsf f}}\,, \quad
2 \gamma \{ {{\mathsf e}}, {{\mathsf f}}\} = {{\mathsf k}}^2 -{{\mathsf k}}^{-2} \,.$$
Using the asymptotics (computed by means of contour integration) $$\label{lim}
\begin{aligned}
\lim\limits_{\hbar \to 0} \frac{\hbar}{2}
\int_{{{\mathbb R}}+i0} \! dt\,
\frac{e^{-i t z}}{\sinh t \, \sinh{\hbar t} } & =
\frac{1}{2}
\int_{{{\mathbb R}}+i0} \! dt\, \frac{e^{-i t z}}{t \, \sinh t}\\
&= \sum_{n\geq 1}^{\infty} \frac{ (-1)^n \, e^{\pi n z}}{n}
= -\log (1+e^{\pi z}) \,,
\end{aligned}$$ we obtain from (\[Hphi\]) the corresponding classical lattice Hamiltonian density, $$\label{Hcl1}
\begin{aligned}
{} H^{{{\rm{\scriptscriptstyle}XXZ}}, \rm cl}_{n,n+1} & =
\lim_{h\to 0} \hbar H^{{\rm{\scriptscriptstyle}XXZ}}_{n,n+1}
= {{\textstyle \frac{1}{\gamma} }} \log (4 \cosh^2 \pi b \, {{\mathsf s}}^{\rm cl}_{n,n+1}) =
{{\textstyle \frac{1}{\gamma} }} \log ({{\mathsf C}}^{\rm cl}_{n,n+1}) \\
{}& = {{\textstyle \frac{1}{\gamma} }} \log \bigl( ({{\mathsf e}}_n {{\mathsf f}}_{n+1} +
{{\mathsf f}}_n {{\mathsf e}}_{n+1}) \, {{\mathsf k}}_n^{-1} {{\mathsf k}}_{n+1} \\
& \qquad \qquad + 2 {{\mathsf k}}_n^{-2} {{\mathsf k}}_{n+1}^2 +
2 \cosh(2\pi b s) ({{\mathsf k}}_n^{-2} + {{\mathsf k}}_{n+1}^2) +2 \bigr) \,.
\end{aligned}$$ Here ${{\mathsf C}}^{\rm cl}_{n,n+1}$ is the classical limit of the tensor Casimir operator given by (\[tensC\]).
Comparison with similar models {#CSM}
------------------------------
L–matrix (\[Lxxz\]) and R–matrix (\[R12\]) are suitable for the usual XXZ model as well. The only (but essential) difference is that in the latter case matrix coefficients of the L–matrix act on a highest weight module of ${{\mathcal U}}_q({{\mathfrak s}{\mathfrak l}}_2)$. We also remark that (\[Lxxz\]) differs from the most commonly used “standard” L–matrix in that it contains extra factors $e^{\pm \pi b u}$ in the off–diagonal elements. The “standard” L–matrix does not satisfy (\[Lpm\]) but is symmetric (if the matrix transposition ${}^{{{\scriptscriptstyle}T}}$ is combined with the operator transposition ${}^t$ such that ${{\mathsf f}}^t_s={{\mathsf e}}_s$ and ${{\mathsf k}}^t_s={{\mathsf k}}_s$) and corresponds to the symmetric auxiliary R–matrix (\[Rlsg\]).
Let $\psi_b(x)$ denote the logarithmic derivative of the function $S_b(x)$ defined by (\[Sbx\]). Properties (\[Sb1\])–(\[Sb3\]) show that $S_b(x)$ can be regarded as a b–analogue of the gamma function. The Hamiltonian density (\[Hphi\]) rewritten in terms of $\psi_b(x)$ looks as follows $$\label{Hpsi}
H^{{\rm{\scriptscriptstyle}XXZ}}_{n,n+1} = {{\textstyle \frac{2}{\pi b} }} \, \psi_b({{\textstyle \frac{Q}{2} }} + i{{\mathsf s}})=
{{\textstyle \frac{2}{\pi b} }} \, \psi_b({{\textstyle \frac{Q}{2} }} - i{{\mathsf s}}) =
{{\textstyle \frac{1}{\pi b} }} \bigl( \psi_b({{\textstyle \frac{Q}{2} }} + i{{\mathsf s}}) +
\psi_b({{\textstyle \frac{Q}{2} }} - i{{\mathsf s}}) \bigr) \,.$$ Equivalence of these expressions is due to (\[Sb3\]). The last of them resembles the form of the Hamiltonian density of the non–compact XXX magnet [@DKM] expressed in terms of the ordinary $\psi$–function.
The special function $w_b(u)$ is closely related (cf. Eq. (\[gw\])) to the non–compact quantum dilogarithm $g_b(u)$. Counterparts of (\[Rsimpledef\]) and (\[Rxxz\]) for the compact XXZ magnet look similar in terms of the q–gamma function which, in turn, is closely related to the compact analogue of $g_b(u)$ given by $s_q(t)=\prod_{n=0}^{\infty}(1+tq^{2n+1})$.
It is also worth noticing that the R–operator (\[Rsimpledef\]) resembles the fundamental R–operator $r({{\mathsf s}},\lambda)$ found in [@FV] for a simpler L–matrix related to the Volterra model [@V1]. The main difference is that the operator argument ${{\mathsf s}}$ of $r({{\mathsf s}},\lambda)$ has a much simpler structure in terms of the variables ${{\mathsf p}}$ and ${{\mathsf x}}$. It would be interesting to clarify the connection between these two R–operators.
Lattice sinh–Gordon model {#LSG}
=========================
\[sectSG\]
Definition of the model
-----------------------
In this section we will begin to develop the QISM for a lattice version of the sinh–Gordon model, which has ${{\cal U}_q({{\mathfrak s}{\mathfrak l}}(2,{{\mathbb R}}))}$ as a quantum symmetry. We are going to keep much of the set–up from Section \[XXZ\], but we will now be using the following L–matrix acting on ${\mathbb C}^2 {\otimes}{\cal P}_s$ $$\label{lSG}
L^{{{\rm{\scriptscriptstyle}SG}}}(u)
= {{\textstyle \frac{1}{i} }} e^{-\pi b s}\, \left( \begin{array}{cc}
i \, {{\mathsf e}}_s &
e^{\pi b u} {{\mathsf k}}_s^{-1} - e^{-\pi b u} {{\mathsf k}}_s \\
e^{\pi b u} {{\mathsf k}}_s - e^{-\pi b u} {{\mathsf k}}_s^{-1} &
i \, {{\mathsf f}}_s
\end{array} \right) \,.$$
L–matrix (\[lSG\]) satisfies the intertwining relation (\[rLL\]) where the auxiliary R–matrix is now given by $$\label{Rlsg}
R(u) =
\left( \begin{array}{cccc}
\sinh \pi b ( u+ ib) & & & \\ [-1mm]
& \sinh \pi b u & i \sin\gamma & \\ [-1mm]
& i \sin\gamma & \sinh \pi b u & \\ [-1mm]
& & & \sinh \pi b ( u+ ib)
\end{array} \right) \,.$$ This R–matrix possesses the following symmetry $$\label{symR}
[ R(u) , \sigma_a \otimes \sigma_a] = 0
\,,$$ where $\sigma_a$, $a=1,2,3$, are the Pauli matrices. We may then proceed the same way as in Section \[XXZ\] to define the operator ${{\mathsf T}}(u)$ and the family ${{\mathcal Q}}=\{{{\mathsf T}}_0,\dots,{{\mathsf T}}_{{{\rm N}}}\}$ of commuting observables. We again find (see Appendix \[monoapp\]) that the corresponding operators ${{\mathsf T}}_m$, $m=0,\dots,{{\rm N}}$ are positive self–adjoint as a direct consequence of the positivity of ${{\mathsf e}}_s$, ${{\mathsf f}}_s$, ${{\mathsf k}}_s$. The existence of a joint spectral decomposition for the family ${{\mathcal Q}}$ is thereby ensured by the spectral theorem (cf. Remark \[posadj\]).
Fundamental R–operator and Hamiltonian {#FRSG}
--------------------------------------
Now our aim is to find the fundamental R–operator ${{\mathsf R}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}^{{{\rm{\scriptscriptstyle}SG}}} (u)$ corresponding to L–matrix (\[lSG\]). Fortunately, it turns out that it can be constructed from the R–operator of the XXZ chain in a simple way. To demonstrate this, we first introduce an automorphism $\theta$ such that $$\label{theta}
\theta({{\mathsf p}})=-{{\mathsf p}}\,, \qquad\theta({{\mathsf x}})=-{{\mathsf x}}\,.$$ It is useful to notice that $\theta$ can be realzied as an inner automorphism, $\theta({{\mathsf O}})= {{\scriptstyle{\mathsf\Omega}}}\, {{\mathsf O}}\, {{\scriptstyle{\mathsf\Omega}}}^{-1}$, where ${{\scriptstyle{\mathsf\Omega}}}$ is the parity operator whose action in the momentum representation is defined by $({{\scriptstyle{\mathsf\Omega}}}f)(k)=f(-k)$. Notice that ${{\scriptstyle{\mathsf\Omega}}}$ is unitary and satisfies ${{\scriptstyle{\mathsf\Omega}}}^{-1}={{\scriptstyle{\mathsf\Omega}}}$. Observe that for the representation ${{\mathcal P}}_s$ we have (cf. (\[EFK1\])) $$\label{thetaEFK}
\theta ({{\mathsf e}}) = {{\mathsf f}}\,, \qquad
\theta ({{\mathsf f}}) = {{\mathsf e}}\,, \qquad
\theta ({{\mathsf k}}) = {{{\mathsf k}}}^{-1} \,.$$
Let the operator ${{\mathsf R}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}^{{{\rm{\scriptscriptstyle}SG}}} (u)$ be defined by the formula $$\label{Rsg2'}
{{\mathsf R}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}^{{{\rm{\scriptscriptstyle}SG}}} (u) \,=\,
({{\mathsf k}}\otimes {{\mathsf k}})^{- i \frac{u}{2b}} \cdot ({{\scriptstyle{\mathsf\Omega}}}{\otimes}1)\cdot
{{{\mathsf R}}}^{{{\rm{\scriptscriptstyle}XXZ}}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(u)\cdot(1{\otimes}{{\scriptstyle{\mathsf\Omega}}})\cdot
({{\mathsf k}}\otimes {{\mathsf k}})^{- i \frac{u}{2b}} \,,$$ where ${{\mathsf R}}^{{{\rm{\scriptscriptstyle}XXZ}}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(u)$ is given by [(\[Rsimpledef\])]{} if $s_{{\mathfrak 1}}=s_{{\mathfrak 2}}$ and by (\[Rxxz\]) otherwise.
\[Rfsg\] The operator ${{\mathsf R}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}^{{{\rm{\scriptscriptstyle}SG}}} (u)$ satisfies the equation [(\[RLL\])]{} where $L(u)$ is given by (\[lSG\]).
It will be convenient to consider $\check{{\mathsf R}}_{{{\mathfrak 1}}{{\mathfrak 2}}}^{{{\rm{\scriptscriptstyle}SG}}}(u)\equiv{{\mathsf P}}_{\!\!{{\mathfrak 2}}{{\mathfrak 1}}}{{\mathsf R}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}^{{{\rm{\scriptscriptstyle}SG}}}(u):
{{\mathcal P}}_{s_{{\mathfrak 2}}}{\otimes}{{\mathcal P}}_{s_{{\mathfrak 1}}}{\to}{{\mathcal P}}_{s_{{\mathfrak 1}}}{\otimes}{{\mathcal P}}_{s_{{\mathfrak 2}}}$ and analogously defined $\check{{{\mathsf R}}}_{{{\mathfrak 1}}{{\mathfrak 2}}}^{{\rm{\scriptscriptstyle}XXZ}}(u) \equiv
{{\mathsf P}}_{\!\!{{\mathfrak 2}}{{\mathfrak 1}}}{{\mathsf R}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}^{{{\rm{\scriptscriptstyle}XXZ}}}(u)$. Equation [(\[RLL\])]{} for ${{\mathsf R}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}^{{{\rm{\scriptscriptstyle}SG}}}(u)$ is then equivalent to $$\label{cRLL}
\check{{\mathsf R}}_{{{\mathfrak 2}}{{\mathfrak 3}}}^{{{\rm{\scriptscriptstyle}SG}}}(u) \, L_{{{\mathfrak 1}}{{\mathfrak 3}}}^{{{\rm{\scriptscriptstyle}SG}}} (u + v) \,
L_{{{\mathfrak 1}}{{\mathfrak 2}}}^{{{\rm{\scriptscriptstyle}SG}}} (v) = L_{{{\mathfrak 1}}{{\mathfrak 3}}}^{{{\rm{\scriptscriptstyle}SG}}} (v) \,
L_{{{\mathfrak 1}}{{\mathfrak 2}}}^{{{\rm{\scriptscriptstyle}SG}}} (u + v) \, \check{{\mathsf R}}_{{{\mathfrak 2}}{{\mathfrak 3}}}^{{{\rm{\scriptscriptstyle}SG}}}(u) \,.$$ Let us also note that, by using $({{\mathsf k}}{\otimes}{{\mathsf k}})^{-1}\check{{\mathsf R}}^{{\rm{\scriptscriptstyle}XXZ}}_{{{\mathfrak 1}}{{\mathfrak 2}}}(u)({{\mathsf k}}{\otimes}{{\mathsf k}})=
\check{{\mathsf R}}^{{\rm{\scriptscriptstyle}XXZ}}_{{{\mathfrak 1}}{{\mathfrak 2}}}(u)$, we may rewrite the expression for $\check{{\mathsf R}}_{{{\mathfrak 1}}{{\mathfrak 2}}}^{{{\rm{\scriptscriptstyle}SG}}}(u)$ which follows from [(\[Rsg2’\])]{} as $$\label{Rsg2}
\begin{aligned}
\check{{{\mathsf R}}}_{{{\mathfrak 1}}{{\mathfrak 2}}}^{{{\rm{\scriptscriptstyle}SG}}} (u) &\,= \,
\big({\rm id}{\otimes}\theta\big)\big(\check{{{\mathsf R}}}'_{{{\mathfrak 1}}{{\mathfrak 2}}}(u)\big),\\
\check{{{\mathsf R}}}'_{{{\mathfrak 1}}{{\mathfrak 2}}}(u)&\,\equiv\,
({{\mathsf k}}\otimes 1)^{- i \frac{u}{b}} \cdot
\check{{{\mathsf R}}}_{{{\mathfrak 1}}{{\mathfrak 2}}}^{{{\rm{\scriptscriptstyle}XXZ}}}(u)\cdot
(1 \otimes {{\mathsf k}})^{ i\frac{u}{b}} \,.
\end{aligned}$$ The key to the proof of the Proposition will then be the following relation between the L–matrices $L^{{{\rm{\scriptscriptstyle}SG}}}(u)$ and $L^{{\rm{\scriptscriptstyle}XXZ}}(u)$ $$\label{Lsg2}
L^{{{\rm{\scriptscriptstyle}SG}}}(u)\,=\, -i e^{-\pi b s} \,
\sigma_1 \,{{\mathsf k}}^{-i\frac{u}{b}}\,L^{{\rm{\scriptscriptstyle}XXZ}}(u)\,
{{\mathsf k}}^{i\frac{u}{b}}\,\equiv\, \sigma_1\,{L}'(u) \,.$$ Inserting [(\[Lsg2\])]{} into [(\[cRLL\])]{}, we get an expression which contains $ \sigma_1 \, {L}_{{{\mathfrak 1}}{{\mathfrak 3}}}'(u) \, \sigma_1$. Observe that (cf. [(\[thetaEFK\])]{}) $$\label{Lt}
\sigma_1 \, {L}'(u) \, \sigma_1 =
(id \otimes \theta) \, {L}'(u) \,.$$ Therefore, by using $\theta({{\mathsf O}}_{{\mathfrak 1}}{{\mathsf O}}_{{\mathfrak 2}})=\theta({{\mathsf O}}_{{\mathfrak 1}})\theta({{\mathsf O}}_{{\mathfrak 2}})$, one finds that [(\[cRLL\])]{} is equivalent to $$\check{{\mathsf R}}_{{{\mathfrak 2}}{{\mathfrak 3}}}'(u) \, L_{{{\mathfrak 1}}{{\mathfrak 3}}}' (u + v) \, L_{{{\mathfrak 1}}{{\mathfrak 2}}}' (v) =
L_{{{\mathfrak 1}}{{\mathfrak 3}}}' (v) \, L_{{{\mathfrak 1}}{{\mathfrak 2}}}'(u + v) \, \check{{\mathsf R}}_{{{\mathfrak 2}}{{\mathfrak 3}}}'(u) \,,$$ which is now easily reduced to the Theorem \[Rthm\] (its general case for ${{\mathsf R}}^{{{\rm{\scriptscriptstyle}XXZ}}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(u)$).
It is easy to see that properties (\[RGGa\])–(\[RGGc\]) of the R–operator of the modular XXZ magnet hold for ${{\mathsf R}}(u)\equiv {{\mathsf R}}^{{{\rm{\scriptscriptstyle}SG}}}_{ss}(u)$ as well. Therefore, using the regularity of ${{\mathsf R}}(u)$, we can construct a Hamiltonian with nearest neighbour interaction of sites by using the same recipe that we used to derive (\[HRxxz\]). This yields $$\label{Hsg}
\begin{aligned}
H^{{{\rm{\scriptscriptstyle}SG}}}_{n,n+1} &
= {{\textstyle \frac{i}{\pi b} }} \partial_u
\check{{{\mathsf R}}}^{{{\rm{\scriptscriptstyle}SG}}}_{n,n+1}(u) \Bigm|_{u=0}
{\stackrel{(\ref{Rsg2})}{=}} (id \otimes \theta) \, H^{{\rm{\scriptscriptstyle}XXZ}}_{n,n+1}
+ {{\textstyle \frac{1}{\gamma} }} \log({{\mathsf k}}_n {{\mathsf k}}_{n+1}) \\
& = - \frac{1}{\pi} \, \int\limits_{{{\mathbb R}}+i0} dt\,
\frac{\cos (2 b t\, \hat{{\mathsf s}}_{n,n+1} )}{\sinh t \, \sinh{b^2 t} }
+ {{\textstyle \frac{1}{\gamma} }} \log({{\mathsf k}}_n {{\mathsf k}}_{n+1}) \,,
\end{aligned}$$ where $\hat{{\mathsf s}}=(1 {\otimes}{{\scriptstyle{\mathsf\Omega}}}) \,{{\mathsf s}}\, (1 {\otimes}{{\scriptstyle{\mathsf\Omega}}})$ is the unique positive self–adjoint operator on ${{\mathcal P}}_s {\otimes}{{\mathcal P}}_s$ such that $$\label{ttC}
4 \cosh^2 \pi b\, \hat{{\mathsf s}}=
\bigl(\pi_{s} {\otimes}(\theta\circ\pi_{s}) \bigr) \Delta (C) \,.$$
Relation with the continuum theory {#CL}
----------------------------------
To begin with, we may first compute the classical limit, $\hbar\to 0$, of (\[Hsg\]) in the same way as we derived (\[Hcl1\]). Using [(\[thetaEFK\])]{}, we obtain $$\label{Hcl2}
\begin{aligned}
H^{{{\rm{\scriptscriptstyle}SG}}, \rm cl}_{n,n+1}
= &\, {{\textstyle \frac{1}{\gamma} }} \log \bigl( {{\mathsf e}}_n {{\mathsf e}}_{n+1} + {{\mathsf f}}_n {{\mathsf f}}_{n+1}
+ 2 ( {{\mathsf k}}_n^{-1} {{\mathsf k}}_{n+1}^{-1} + {{\mathsf k}}_n {{\mathsf k}}_{n+1} ) \\
& + 2 \cosh(2\pi b s) ({{\mathsf k}}_n^{-1} {{\mathsf k}}_{n+1}
+ {{\mathsf k}}_n {{\mathsf k}}_{n+1}^{-1}) \bigr) \,.
\end{aligned}$$ In order to establish the relation with the sinh–Gordon model it will be convenient to change variables as follows: $$\label{px}
2 \pi b \, {{\mathsf p}}_n = -\beta \, \Phi_n \,, \qquad
4 \pi b \, {{\mathsf x}}_n = \beta \, ({{\textstyle \frac{1}{2} }} \Pi_n -\Phi_n) \,, \qquad
\beta=b \sqrt{8\pi} \,.$$ The variables $\Phi_n$ and $\Pi_n$ will then turn out to correspond to the (discretized) sinh–Gordon field and its conjugate momentum, respectively. The classical field and momentum variables defined in [(\[px\])]{} satisfy the Poisson–bracket relations $\{\Pi_n,\Phi_m\}=\delta_{nm}$. The Hamiltonian [(\[Hcl2\])]{} now looks as follows $$\begin{aligned}
\label{Hsgl}
H^{{{\rm{\scriptscriptstyle}SG}}, \rm cl}_{n,n+1}
&= {{\textstyle \frac{1}{\gamma} }}\log {{\textstyle \frac{4}{\mu} }} \Bigl(
{{\textstyle \frac{1}{2} }}\cosh {{\textstyle \frac{\beta}{4} }} \, (\Pi_n + \Pi_{n+1})
+ {{\textstyle \frac{1+\mu^2}{2} }} \,
\cosh {{\textstyle \frac{\beta}{2} }} \, (\Phi_n - \Phi_{n+1}) \\
\nonumber
&+ {{\textstyle \frac{\mu}{2} }} \, \cosh \beta \bigl(\Phi_n -
{{\textstyle \frac{1}{4} }}(\Pi_n +\Pi_{n+1})\bigr)
+ {{\textstyle \frac{\mu}{2} }} \, \cosh \beta \bigl(\Phi_{n+1} -
{{\textstyle \frac{1}{4} }}(\Pi_n +\Pi_{n+1})\bigr) \\
\nonumber
& + \mu \, \cosh {{\textstyle \frac{\beta}{2} }} (\Phi_n + \Phi_{n+1})
+ {{\textstyle \frac{\mu^2}{4} }} \cosh \beta \bigl( \Phi_n + \Phi_{n+1} -
{{\textstyle \frac{1}{4} }}(\Pi_n +\Pi_{n+1}) \bigr) \Bigr)\,,
$$ where $\mu=e^{-2\pi bs}$. In order to define the relevant limit leading to the [*continuous*]{} sinh–Gordon model, let us combine the limit of vanishing lattice spacing ${{\rm N}}\to\infty$, $\Delta\to 0$ ($R ={{\rm N}}\Delta/2\pi$ is kept fixed) with the limit where the representation parameter $s$ goes to infinity in such a way that the mass parameter $m$ defined via $$\label{mds}
{{\textstyle \frac{1}{4} }} m\Delta\,=\,e^{-\pi b s} \,.$$ stays finite. In addition we shall assume the standard correspondence between lattice and continuous variables: $$\label{cont}
\Pi_n \to \Pi(x) \, \Delta \,, \quad
\Phi_n \to \Phi(x) \,, \quad
x = n\Delta \,.$$ We then find the following limiting expression for the Hamiltonian density: $$\label{limH}
\sum_n {{\textstyle \frac{1}{\Delta} }} H^{{{\rm{\scriptscriptstyle}SG}}, \rm cl}_{n,n+1} \to
{\rm const} + \int\nolimits_0^{2\pi R} dx \,
\bigl( {{\textstyle \frac{1}{2} }} \Pi^2 + {{\textstyle \frac{1}{2} }} (\partial_x\Phi)^2
+ {{\textstyle \frac{m^2}{\beta^2} }} \cosh \beta\Phi \bigr)$$ thus recovering the continuous $\sinh$–Gordon model.
It is also instructive to see what happens to the L–matrix in this limit. In the classical continuous limit, i.e., when $m$ in (\[mds\]) is kept fixed and $\hbar,\Delta\to 0$, Eqs. (\[hbar\]) and (\[cont\]) show that L–matrix (\[lSG\]) becomes $$\label{LU}
L^{{{\rm{\scriptscriptstyle}SG}}}({{\textstyle \frac{u}{\pi b} }}) \to
\bigl(\begin{smallmatrix} 1 & 0\\ 0 & 1
\end{smallmatrix} \bigr)
+ \Delta \, U^{{{\rm{\scriptscriptstyle}SG}}}(u) + O(\Delta^2)\,,$$ where $U^{{{\rm{\scriptscriptstyle}SG}}}(u)$ is the well–known $U$–matrix from the Lax pair for the classical continuous sinh–Gordon model [@KBI; @S1], $$\label{Ucc}
U^{{{\rm{\scriptscriptstyle}SG}}}(u) = \left( \begin{array}{cc}
{{\textstyle \frac{\beta}{4} }} \, \Pi(x) &
{{\textstyle \frac{m}{2i} }} \, \sinh \bigl(u - \frac{\beta}{2} \Phi(x) \bigr) \\
{{\textstyle \frac{m}{2i} }} \, \sinh \bigl(u + \frac{\beta}{2} \Phi(x) \bigr) &
- {{\textstyle \frac{\beta}{4} }} \, \Pi(x)
\end{array} \right)\,.$$
The classical lattice Hamiltonian density (\[Hcl2\]) resembles that found in [@Ta] for the lattice sine–Gordon model. However, relation between the [*quantum*]{} Hamiltonians is less clear because the fundamental R–operator proposed in [@Ta] is represented as a product of of the type $r({{\mathsf s}},\lambda)$ which we mentioned at the end of Subsection \[CSM\]. Possibly, recent results on factorization of R–operators [@DKK] will help to clarify the connection between our construction and that used in [@Ta; @FV; @V1].
Failure of the algebraic Bethe ansatz {#ABA}
-------------------------------------
For the sake of clarity it may be worthwhile explaining in some detail why the algebraic Bethe ansatz is not suitable for the solution of the lattice sinh–Gordon model.
To begin with, let us observe that the L–matrix (\[lSG\]) has no pseudo–vacuum state, i.e., a vector $\Psi$ such that $L^{{{\rm{\scriptscriptstyle}SG}}}_{21}(u) \, \Psi =0$. Indeed, this would require that ${{\mathsf k}}_n\Psi=0={{\mathsf k}}_n^{-1}\Psi$ for all $n=1,\dots,{{\rm N}}$. Such a vector does not exist.
For the sine–Gordon model, one circumvents this difficulty by considering the composite , ${\cal L}(u)$, which is product of two L–matrices [@FST; @IK]. For $\gamma=\pi {{\textstyle \frac{m}{n} }}$, $n,m \in {\mathbb N}$, exponential operators $e^{i \beta \Phi_n}$, $e^{i \beta \Pi_n}$ admit finite–dimensional representations. In this case there exists a vector $\Psi$ that is annihilated by ${\cal L}_{21}(u)$. This makes it possible to apply the algebraic Bethe ansatz technique. Let us therefore consider the analogous construction for the sinh–Gordon model. Let ${\cal L}^{{{\rm{\scriptscriptstyle}SG}}}(u) = L^{{{\rm{\scriptscriptstyle}SG}}}_{2}(u) \,
L^{{{\rm{\scriptscriptstyle}SG}}}_{1}(u+\varpi)$, where we introduced the shift by the constant $\varpi\in{{\mathbb R}}$ in order to increase the generality of our consideration. We then have $$\label{CLL}
{\cal L}^{{{\rm{\scriptscriptstyle}SG}}}_{21}(u) =
i e^{\pi b u} \bigl( {{\mathsf k}}{\otimes}{{\mathsf e}}+
e^{\pi b \varpi} \, {{\mathsf f}}{\otimes}{{\mathsf k}}\bigr) -
i e^{-\pi b u} \bigl( {{\mathsf k}}^{-1} {\otimes}{{\mathsf e}}+
e^{-\pi b \varpi} \, {{\mathsf f}}{\otimes}{{\mathsf k}}^{-1} \bigr) \,.$$ The requirement that a vector $\Psi$ is annihilated by ${\cal L}^{{{\rm{\scriptscriptstyle}SG}}}_{21}(u)$ is equivalent to the two equations $$\bigl( {{\mathsf k}}^{\pm 1} {\otimes}{{\mathsf e}}+
e^{\pm \pi b \varpi} \, {{\mathsf f}}{\otimes}{{\mathsf k}}^{\pm 1} \bigr)\Psi=0 \,.$$ We claim that there does not exist a reasonable (even in the distributional sense) state $\Psi$ with such properties. Indeed, note that $${{\mathsf k}}{\otimes}{{\mathsf e}}+ e^{\pi b \varpi} \, {{\mathsf f}}{\otimes}{{\mathsf k}}=({{\mathsf k}}^{i\frac{\varpi}{b}}
{{\scriptstyle{\mathsf\Omega}}}{\otimes}{\sf 1})\cdot({{\mathsf k}}^{-1}{\otimes}{{\mathsf e}}+{{\mathsf e}}{\otimes}{{\mathsf k}})\cdot
({{\mathsf k}}^{i\frac{\varpi}{b}}{{\scriptstyle{\mathsf\Omega}}}{\otimes}{\sf 1})^{-1} \,,$$ where ${{\scriptstyle{\mathsf\Omega}}}$ is the parity operation introduced in Subsection \[FRSG\]. The operator $({{\mathsf k}}^{i\frac{\varpi}{b}} {{\scriptstyle{\mathsf\Omega}}}{\otimes}{\sf 1})$ is unitary, which allows us to conclude that ${{\mathsf k}}{\otimes}{{\mathsf e}}+ e^{\pi b \varpi} \,
{{\mathsf f}}{\otimes}{{\mathsf k}}$ and ${{\mathsf k}}^{-1}{\otimes}{{\mathsf e}}+{{\mathsf e}}{\otimes}{{\mathsf k}}$ have the same spectrum. However, the latter operator represents $(2\sin\gamma){\Delta}(E)$ on ${{\mathcal P}}_{s}{\otimes}{{\mathcal P}}_s$ (cf. eq. [(\[De\])]{}). Unitarity of the Clebsch–Gordan maps (see Appendix \[CGmaps\]) implies that this operator has the same spectrum as ${{\mathsf e}}_s$. The unitary transformation used in the proof of Lemma \[Pslem\] in Appendix B maps ${{\mathsf e}}_s$ to $e^{2\pi b{{\mathsf x}}}$. It is now clear that all these operators do not have an eigenfunction with eigenvalue zero, as would be necessary to construct a Bethe vacuum.
Keeping in mind that the sinh–Gordon variables are just linear combinations of ${{\mathsf p}}$ and ${{\mathsf x}}$, cf. [(\[px\])]{}, we now see quite clearly that the failure of the Bethe ansatz is connected with the fact that the target space (the space in which the fields take their values) is [*non–compact*]{}. We expect this to be a general lesson.
Q–operator and Baxter equation {#QB}
==============================
As an important first step towards the solution of the Auxiliary Spectral Problem we shall now find [*necessary*]{} conditions for a function $t(u)$ to be eigenvalue of the operator ${{\mathsf T}}(u)$. In order to do this we are going to construct an operator ${{\mathsf Q}}(u)$ which satisfies the following properties: $$\label{defQ1}
\begin{aligned}
{\rm (i)} \quad & {{\mathsf Q}}(u) \;\;{\rm is~a~normal~operator},\;\;
{{\mathsf Q}}(u){{\mathsf Q}}^{*}(v)={{\mathsf Q}}^{*}(v) {{\mathsf Q}}(u),\\
{\rm (ii)} \quad & {{\mathsf Q}}(u) \, {{\mathsf Q}}(v) = {{\mathsf Q}}(v) \, {{\mathsf Q}}(u) \,, \\
{\rm (iii)} \quad & {{\mathsf Q}}(u) \, {{\mathsf T}}(u) = {{\mathsf T}}(u) \, {{\mathsf Q}}(u) \,,\\
{\rm (iv)} \quad & {{\mathsf Q}}(u) \, {{\mathsf T}}(u)\,=\,
\bigl(a(u)\bigr)^{{{\rm N}}} \, {{\mathsf Q}}(u-ib)
+\bigl(d(u)\bigl)^{{{\rm N}}} \, {{\mathsf Q}}(u+ib).
\end{aligned}$$ The first and the second property imply that all operators ${{\mathsf Q}}(u)$, $u\in{{\mathbb C}}$ can be simultaneously diagonalized and their eigenvectors form a complete system of states in the Hilbert space. The third and the fourth property imply that ${{\mathsf T}}(u)$ will be diagonal whenever ${{\mathsf Q}}(u)$ is. One may therefore consider the spectral problem for ${{\mathsf Q}}(u)$ as a refinement of the spectral problem for ${{\mathsf T}}(u)$.
Let us now consider an eigenstate $\Psi_t$ for ${{\mathsf T}}(u)$ with eigenvalue $t(u)$, ${{\mathsf T}}(u)\Psi_t=t(u)\Psi_t$. Thanks to property (iii) above we may assume that it is an eigenstate for ${{\mathsf Q}}(u)$ as well, $$\label{Qpq}
{{\mathsf Q}}(u) \, \Psi_t\,=\,q_t(u) \, \Psi_t\,.$$ It follows from property (iv) that the eigenvalue $q_t(u)$ must satisfy the so–called Baxter equation $$\label{BaxterEV}
\boxed{\quad t(u) \, q_t(u) \,=\, \bigl(a(u)\bigr)^{{{\rm N}}}q_t(u-ib)
+ \bigl(d(u)\bigr)^{{{\rm N}}}q_t(u+ib) \,.\quad}$$ We will construct the operator ${{\mathsf Q}}(u)$ explicitly — see Subsection \[SR\]. This will allow us to determine the analytic and – for the lattice sinh–Gordon model with ${{\rm N}}$ odd (the SGo–model) – the asymptotic properties that the eigenvalues $q_t(u)$ must have, namely $$\label{analQ}
\left[ \;\;\begin{aligned}
{\rm (i)} \;\; & q_t(u)\;\,\text{is meromorphic in}\;\,{{\mathbb C}},
\,\;\text{with poles of maximal order ${{\rm N}}$ }
\text{in}\;\,\Upsilon_{-s}\cup
\bar{\Upsilon}_{s},\\
& \text{where}\;\,
\Upsilon_s = \bigl\{ s + i \bigl( {{\textstyle \frac{Q}{2} }} + nb + mb^{-1} \bigl),
\quad n,m \in {\mathbb Z}^{\geq 0} \bigl\}\,,\quad
\bar{\Upsilon}_s\equiv ({\Upsilon}_s)^*\,,\\
{\rm (ii)} \;\; & q_t^{{{\rm{\scriptscriptstyle}SGo}}}(u)\;{\sim} \,\left\{
\begin{aligned}
& \exp\big(+\pi i {{\rm N}}\big(s+{{\textstyle \frac{i}{2} }}Q\big) u\big)\;\;{\rm for}\;\;
|u|{\to}\infty,\;\, |{\rm arg}(u)|<{{\textstyle \frac{\pi}{2} }}, \\
& \exp\big(-\pi i {{\rm N}}\big(s+{{\textstyle \frac{i}{2} }}Q\big) u\big)\;\;{\rm for}\;\;
|u|{\to}\infty,\,\; |{\rm arg}(u)|>{{\textstyle \frac{\pi}{2} }}.
\end{aligned}\right.
\end{aligned}\;\;\right]$$ The derivation of these properties is discussed in Subsection \[AnaQ\]. This means that there is the following [*necessary*]{} condition for a polynomial $t(u)$ to be eigenvalue of the operator ${{\mathsf T}}(u)$: $t(u)$ can only be an eigenvalue of ${{\mathsf T}}(u)$ if there exists a meromorphic function $q_t(u)$ with singular behavior and asymptotic behavior given in [(\[analQ\])]{} which is related to $t(u)$ by the Baxter equation [(\[BaxterEV\])]{}.
The problem to classify the solutions to this condition is of course still rather nontrivial. However, previous experience from other integrable models suggests that the Baxter equation supplemented by the analytic and asymptotic properties [(\[analQ\])]{} is indeed a useful starting point for the determination of the spectrum of the model, see also our Subsections \[Baxtersub\] and \[contSG\] for some further remarks. We will discuss in the next section how the separation of variables method may allow us to show that the conditions above are also [*sufficient*]{} for $t(u)$ to be an eigenvalue of ${{\mathsf T}}(u)$.
[*Convention:*]{} We will use the superscripts ${{\rm{\scriptscriptstyle}SG}}$ and ${{\rm{\scriptscriptstyle}XXZ}}$ to distinguish analogous operators within the two models we consider. However, we will simply omit these superscripts in any equation which holds in the two cases alike.
Explicit form of ${{\mathsf Q}}(u)$ {#SR}
-----------------------------------
Let us now describe explicitly the Q–operators for the models that we introduced in Sections \[XXZ\] and \[LSG\]. For this purpose we will work in the representation where the operators ${{\mathsf x}}_r$, $r=1,\ldots,{{\rm N}}$ are diagonal. This representation will be called the [*Schr[ö]{}dinger representation*]{} for the Hilbert space of a lattice model. Let ${\bf x}\equiv(x_{{\mathfrak 1}},\ldots,x_{{\scriptscriptstyle}{{\rm N}}})$, ${\bf x'} \equiv(x'_{{\mathfrak 1}},\ldots,x'_{{\scriptscriptstyle}{{\rm N}}})$. We will denote the integral kernel of the operator ${{\mathsf Q}}(u)$ in the Schr[ö]{}dinger representation by $Q_u({\bf x},{\bf x'})$. We will also use the following notations $$\label{ssb}
\sigma \equiv s + {{\textstyle \frac{i}{2} }}Q \,, \qquad
\bar\sigma \equiv s - {{\textstyle \frac{i}{2} }}Q \,,$$ where $s$ stands for the spin of the representation ${{\mathcal P}}_s$.
Let the Q–operators ${{\mathsf Q}}^\flat_\pm(u)$, $\flat= {{\rm{\scriptscriptstyle}XXZ}},\, {{\rm{\scriptscriptstyle}SG}}$, be defined in the Schr[ö]{}dinger representation by the following kernels $$\begin{aligned}
\label{Qxxz+}
Q^{\,\flat}_{+;u} ({\bf x},{\bf x'}) \,= &\\
=\bigl( D_{-s}(u) \bigr)^{{\rm N}}& \prod_{r=1}^{{\rm N}}D_{\frac{1}{2}(\bar\sigma -u)}(x_{r} - x'_{r}) \,
D_{\frac{1}{2}(\bar\sigma +u)}(x_{r-1} - {\varepsilon}_\flat \, x'_{r}) \,
D_{-s}(x_{r} - {\varepsilon}_\flat \, x_{r-1}) \,,\nonumber \\
\label{Qxxz-}
Q^{\,\flat}_{-;u} ({\bf x},{\bf x'}) = &\prod_{r=1}^{{\rm N}}D_{\frac{1}{2}(u-\sigma)}(x_{r} - x'_{r}) \,
D_{-\frac{1}{2}(u+\sigma)}(x_{r} - {\varepsilon}_\flat \, x'_{r-1}) \,
D_{s}(x'_{r} - {\varepsilon}_\flat \, x'_{r-1}) \,,\end{aligned}$$ where ${\varepsilon}_{{\rm{\scriptscriptstyle}XXZ}}= 1$, ${\varepsilon}_{{\rm{\scriptscriptstyle}SG}}= -1$, and, in the sinh–Gordon case, $s$ is related to the parameters $m$, $\Delta$ as in (\[mds\]).
\[QXXZSG\] Let ${{\mathsf T}}^\flat(u)$, $\flat= {{\rm{\scriptscriptstyle}XXZ}},\, {{\rm{\scriptscriptstyle}SG}}$ be the transfer–matrices corresponding to the L–matrices (\[Lxxz\]) and (\[lSG\]).
- The operators ${{\mathsf Q}}^\flat_\pm(u)$ satisfy all relations in (\[defQ1\]).
- The Baxter equation holds for ${{\mathsf Q}}^\flat_\pm(u)$ with the following coefficients $$\label{baxxz}
\begin{aligned}
a^{{\rm{\scriptscriptstyle}XXZ}}(u) &= 2 \sinh \pi b (u-\sigma) \,, \quad
d^{{\rm{\scriptscriptstyle}XXZ}}(u) = 2 \sinh \pi b (u+\sigma) \,, \\
d^{{\rm{\scriptscriptstyle}SG}}(u) &= a^{{\rm{\scriptscriptstyle}SG}}(-u) =e^{ \pi b (u+i\frac{b}{2})} +
({{\textstyle \frac{m \Delta}{4} }})^2 \, e^{ -\pi b (u+i\frac{b}{2})} \,.
\end{aligned}$$
- The operators ${{\mathsf Q}}^\flat_\pm(u)$ satisfy the relation $$\label{QQc}
{{\mathsf Q}}^\flat_+(u) \, {{\mathsf Q}}^\flat_-(v) =
{{\mathsf Q}}^\flat_-(v) \, {{\mathsf Q}}^\flat_+(u) \,.$$
The proof of this Theorem is given in Appendix \[PQT\]. It is worth noting that ${{\mathsf Q}}^{\,\flat}_{-}(u)$ and ${{\mathsf Q}}^{\,\flat}_{+}(u)$ are related by hermitian conjugation as follows. $$\label{xqpm'}
{{\mathsf Q}}^{\,\flat}_{-}(u) = \bigl( D_{-s}(u) \bigr)^{{\rm N}}\,
\bigl( {{\mathsf Q}}^{\,\flat}_{+}(\bar{u}) \bigr)^* \,.$$ This allows us to mostly focus on ${{\mathsf Q}}^{\,\flat}(u)\equiv{{\mathsf Q}}^{\,\flat}_{+}(u)$, but it is nevertheless sometimes useful to consider ${{\mathsf Q}}^{\,\flat}_{-}(u)$ as well. The corresponding eigenvalues $q^{\pm}_t (u)$ are consequently related as $$\label{qqpm}
q^{-}_t (u) = \bigl( D_{-s}(u) \bigr)^{{\rm N}}\
\overline{q^{+}_t (\bar{u})} \,.$$ Relation [(\[qqpm\])]{} will imply that $q^{+}_t$ and $q^{-}_t$ have the same analytic and asymptotic properties [(\[analQ\])]{}.
It is sometimes useful to observe (see Subsection \[QYZ\]) that the operator ${{\mathsf Q}}^\flat(u)$ can be factorized as follows: $$\label{qyz'}
{{\mathsf Q}}^\flat(u) = {{\mathsf Y}}^\flat(u) \cdot {{\mathsf Z}}\,,$$ The operators ${{\mathsf Y}}^\flat(u)$ and ${{\mathsf Z}}$ in [(\[qyz’\])]{} are represented by the kernels $$\begin{aligned}
\label{y1'}
Y_u^\flat({\bf x},{\bf x'}) &= \prod_{r=1}^{{\rm N}}D_{\frac{1}{2}(u-\sigma)}(x_r - {\varepsilon}_\flat\, x'_{r+1}) \,
D_{-\frac{1}{2}(u+\sigma)}(x_r - x'_{r}) \,, \\
Z({\bf x},{\bf x'}) &= \bigl(w_b(i{{\textstyle \frac{Q}{2} }} - 2s)\bigr)^{{\rm N}}\,
\prod_{r=1}^{{\rm N}}\, D_{\bar{\sigma}}(x_r -x'_r)\,,\end{aligned}$$ respectively, where ${\varepsilon}_{{\rm{\scriptscriptstyle}XXZ}}= 1$, ${\varepsilon}_{{\rm{\scriptscriptstyle}SG}}= -1$.
\[gaugerem\] The relations [(\[defQ1\])]{} do not define the Q–operator ${{\mathsf Q}}(u)$ uniquely. For instance, for a given ${{\mathsf Q}}(u)$, relations [(\[defQ1\])]{} are also fulfilled for ${{\mathsf Q}}'(u)= \big(\varphi(u)\big)^{{\rm N}}\, {{\mathsf O}}\, {{\mathsf Q}}(u)$, where $\varphi(u)$ is a scalar function and ${{\mathsf O}}$ is a unitary operator that commutes with ${{\mathsf Q}}(u)$ and ${{\mathsf T}}(u)$. The coefficients in Baxter equation for ${{\mathsf Q}}(u)$ and ${{\mathsf Q}}'(u)$ are related via $$\label{baxxztil}
a'(u) = \frac{\varphi(u)}{\varphi(u-ib)}\, a(u)\,,
\qquad d'(u) = \frac{\varphi(u)}{\varphi(u+ib)} \, d(u) \,.$$ Thus, there is no canonical way to fix these coefficients. However, their combination $a(u)d(u-ib)$ remains invariant; its value is related to the quantum determinant if the latter can be defined for the L–matrix of the model in question (see Appendix \[qDet\]).
Analytic properties of eigenvalues of ${{\mathsf Q}}(u)$ {#AnaQ}
--------------------------------------------------------
We now turn to the derivation of the analytic properties of eigenvalues of ${{\mathsf Q}}(u)$. More precisely we shall prove the following:
$\quad$
- The operators ${{\mathsf Q}}^{{{\rm{\scriptscriptstyle}XXZ}}}_\pm(u)$ and ${{\mathsf Q}}^{{{\rm{\scriptscriptstyle}SG}}}_\pm(u)$ are meromorphic functions of $u$ in ${{\mathbb C}}$ with poles of maximal order ${{\rm N}}$ contained in $\Upsilon_{-s}\cup\bar{\Upsilon}_{s}$.
- Denote ${{\mathsf Q}}_\pm^{{\rm{\scriptscriptstyle}SGo}}(u)\equiv{{\mathsf Q}}^{{\rm{\scriptscriptstyle}SG}}_\pm(u)$ for ${{\rm N}}$ odd. This operator has the following asymptotic behavior $$\label{qtas}
{{\mathsf Q}}_+^{{\rm{\scriptscriptstyle}SGo}}(u)\;{\sim} \,\left\{
\begin{aligned}
& {{\mathsf Q}}_{{\scriptscriptstyle}+\infty} \,
\exp\big(+i \pi {{\rm N}}\sigma \, u \big)\qquad {\rm for}\;\;
|u|{\to}\infty,\;\, |{\rm arg}(u)|<{{\textstyle \frac{\pi}{2} }} \,, \\
& {{\mathsf Q}}_{{\scriptscriptstyle}-\infty} \,
\exp\big(-i\pi {{\rm N}}\sigma \, u\big)\qquad {\rm for}\;\;
|u|{\to}\infty,\,\; |{\rm arg}(u)|>{{\textstyle \frac{\pi}{2} }} \,,
\end{aligned}\right.$$ where ${{\mathsf Q}}_{{\scriptscriptstyle}\pm\infty}$ are commuting unitary operators related to each other as follows $$\label{QQpar}
{{\mathsf Q}}_{{\scriptscriptstyle}-\infty} = {{\mathsf \Omega}}\, {{\mathsf Q}}_{{\scriptscriptstyle}+\infty} \,.$$ Here ${{\mathsf \Omega}}$ is the parity operator (its action in the Schr[ö]{}dinger representation is given by [(\[upar\])]{}).
- ${{\mathsf Q}}^{{\rm{\scriptscriptstyle}SGo}}_-(u)$ has asymptotic behavior of the same form [(\[qtas\])]{} with ${{\mathsf Q}}_{{\scriptscriptstyle}\pm\infty}$ replaced by ${{\mathsf Q}}^*_{{\scriptscriptstyle}\pm\infty}$.
In order to prove part (i) of the Theorem it clearly suffices to consider the corresponding statements for the eigenvalues $q^\pm_t(u)$. Let us first explain why the properties of $q^{+}_t (u)$ and $q^{-}_t (u)$ described in the theorem are the same. Given that the theorem holds for $q^{+}_t (u)$, we infer that poles of $\overline{q^{+}_t (\bar{u})}$ are contained in $\Upsilon_{s}\cup\bar{\Upsilon}_{-s}$. But, since they are of maximal order ${{\rm N}}$, they cancel in [(\[qqpm\])]{} against the ${{\rm N}}$–th order zeroes of $\bigl( D_{-s}(u) \bigr)^{{\rm N}}$ (see properties of $D_{\alpha}(x)$ in Appendix \[Qdil2\]). Thus, the only possible poles of $q^{-}_t (u)$ are those of $\bigl( D_{-s}(u) \bigr)^{{\rm N}}$, i.e., they are of maximal order ${{\rm N}}$ and contained in $\Upsilon_{-s}\cup\bar{\Upsilon}_{s}$. The proof of part (i) will be exactly analogous for the cases of the XXZ magnet and the sinh–Gordon model. Only the latter case will therefore be discussed explicitly. We will study the equation (\[Qpq\]), which is equivalent to $$\label{EVphipsi}
\int_{{{\mathbb R}}^{2{{\rm N}}}} d{{\mathbf x}}\, d{{\mathbf x}}' \; {\Phi}({{\mathbf x}})\,
Q_{+;u}^{{{\rm{\scriptscriptstyle}SG}}}({\bf x},{{\mathbf x}}')\,\Psi_t({{\mathbf x}}')\,=\,
q^+_t(u)\,\langle\,\Phi\,|\,\Psi_t\,\rangle,$$ for some test–function $\Phi({{\mathbf x}})\in{{\mathcal T}}_s^{{\otimes}{{\rm N}}}$, where ${{\mathcal T}}_s$ is the space of test–functions which is canonically associated to the representations ${{\mathcal P}}_s$ as shown in Appendix \[posapp\]. In order to find the analytic properties of $q^+_t(u)$ let us use ${{\mathsf Q}}_+^{{\rm{\scriptscriptstyle}SG}}(u)={{\mathsf Y}}^{{{\rm{\scriptscriptstyle}SG}}}(u){{\mathsf Z}}$ to represent the left hand side of [(\[EVphipsi\])]{} as $\langle\,\Phi'\,|\,{{\mathsf Z}}\Psi_t\,\rangle$, where $$\label{EVexpl}
{\Phi}'({{\mathbf x}}')\,\equiv \, \int_{{{\mathbb R}}^{{\rm N}}} d{{\mathbf x}}\; {\Phi}({{\mathbf x}}) \,
Y_{u}^{{{\rm{\scriptscriptstyle}SG}}}({\bf x},{{\mathbf x}}')\,.$$ With the help of the Paley–Wiener theorems one easily finds that the condition $\Phi({{\mathbf x}})\in{{\mathcal T}}_s^{{\otimes}{{\rm N}}}$ implies that ${\Phi}({{\mathbf x}})$ is entire analytic w.r.t. each variable $x_k$ and decays exponentially as $$\label{Psiasym}
|{\Phi}({{\mathbf x}})|\,\sim\, e^{-\pi Q|x_k|} \qquad
{\rm for}\;\; |x_k|{\to}\infty \,.$$ The kernel $Y_{u}^{{{\rm{\scriptscriptstyle}SG}}}({\bf x},{{\mathbf x}}')$ has the same asymptotics w.r.t. its $x_k$ variables, as seen from equation [(\[y1’\])]{} and relation [(\[Das1\])]{}. Therefore the convergence of the integrals does not represent any problem. Combined with the observation that the left hand side of [(\[EVexpl\])]{} is the convolution of two meromorphic functions we conclude that the only source of singular behavior is the possibility that the contours of integration in [(\[EVexpl\])]{} may become pinched between poles of the integrand approaching the contour from the upper and lower half planes, respectively. With the help of [(\[wan\])]{} one easily compiles a list of the relevant poles of the kernel $Y_{u}^{{{\rm{\scriptscriptstyle}SG}}}({\bf x},{{\mathbf x}}')$ as given in [(\[y1’\])]{}: $$\begin{aligned}
\text{Upper half plane}\;\; {\mathbb H}_- :\quad (1)\;\;
{} & x_r\in x_r'-{{\textstyle \frac{1}{2} }}(u+{\sigma})+\Upsilon_0,\\
(2)\;\; & x_r\in -x_{r+1}' + {{\textstyle \frac{1}{2} }}(u-{\sigma})+\Upsilon_0,\\
\text{Lower half plane}\;\;\ {\mathbb H}_+:\quad (1')\;\;
{} & x_r\in x_r'+{{\textstyle \frac{1}{2} }}(u+{\sigma})-\Upsilon_0,\\
(2')\;\; & x_r\in -x_{r+1}'-{{\textstyle \frac{1}{2} }}(u-{\sigma})-\Upsilon_0.\end{aligned}$$ Pinching of the contour between poles from the upper and lower half planes would produce the following series of poles: $$\label{poles}
\begin{aligned}
{} &(11')\quad u+s\in +\Upsilon_{0},\\
{} &(22')\quad u-s\in -\Upsilon_0,
\end{aligned}
\quad
\begin{aligned}
{} &(12')\quad x_r'+x_{r+1}'-s\in -\Upsilon_0,\\
{} &(21')\quad x_r'+x_{r+1}'+s\in +\Upsilon_0.
\end{aligned}$$ We observe in particular that none of the poles of ${\Phi}'({{\mathbf x}}')$ happens to lie on the real axis, which represents the contour of integration for each of the integrals over the variables $x_k'$ in [(\[EVphipsi\])]{}. Taking into account the exponential decay of $Y_{u}^{{{\rm{\scriptscriptstyle}SG}}}({\bf x},{{\mathbf x}}')$ for $|x_k'|{\to}\infty$, we may conclude that the integration over ${{\mathbf x}}'$ in [(\[EVphipsi\])]{} converges nicely. It follows that the left hand side of [(\[EVphipsi\])]{} defines a meromorphic function of $u$ with poles listed on the l.h.s. of [(\[poles\])]{}.
In order to verify part (ii) of the theorem let us note that $Y_{u}^{{{\rm{\scriptscriptstyle}SG}}}({\bf x},{{\mathbf x}}')$ has the asymptotic behavior $$\label{Yasym}
Y_{u}^{{{\rm{\scriptscriptstyle}SG}}}({\bf x},{{\mathbf x}}')\,\sim e^{\pm\pi i{{\rm N}}{\sigma}u}
\prod_{r=1}^{{{\rm N}}} \,e^{\mp 2\pi i x_r^{}(x_{r+1}'+x_{r}')}
\quad{\rm for}\;\; |u|{\to}\infty,\,\; \left\{
\begin{aligned}
& |{\rm arg}(u)|<{{\textstyle \frac{\pi}{2} }}, \\
& |{\rm arg}(u)|>{{\textstyle \frac{\pi}{2} }}.
\end{aligned}\right.$$ as follows straightforwardly from [(\[Das2\])]{}. In order to check that the integral obtained by exchanging the limit for $u{\to}\pm\infty$ with the integrations in [(\[EVphipsi\])]{} is convergent let us note that performing the integration over the variables $x_r$ yields the Fourier transformation $\tilde{\Phi}({{\mathbf k}})$ of ${\Phi}({{\mathbf x}})$, with argument ${{\mathbf k}}\equiv(x'_{{\mathfrak 1}}+x'_{{\mathfrak 2}},\dots,x'_{{{\rm N}}}+x'_{{\mathfrak 1}})$. Note that the change of variables ${{\mathbf k}}={{\mathbf k}}({{\mathbf x}}')$ is invertible for ${{\rm N}}$ odd. We may therefore represent the integration over ${{\mathbf x}}'$ by an integration over ${{\mathbf k}}$. The nice asymptotic properties of $\tilde{\Phi}({{\mathbf k}})$ which follow from our requirement $\Phi\in({{\mathcal T}}_s)^{{\otimes}{{\rm N}}}$ ensure the convergence of the resulting integrals. Part (ii) of the theorem therefore follows from [(\[Yasym\])]{} and [(\[qyz’\])]{}.
The proof of part (iii) of the theorem is immediate, if [(\[xqpm’\])]{} and [(\[Das1\])]{} are taken into account.
Let us comment on the nature of the problems which prevented us to determine the asymptotics of the Q–operators in the remaining cases. In both remaining cases one must observe that the change of variables ${{\mathbf k}}={{\mathbf k}}({{\mathbf x}}')$ is not invertible, which implies divergence of the integral over ${{\mathbf x}}'$. This is closely related to the fact that the leading asymptotics of ${{\mathsf T}}(u)$ for $|u|{\to}\infty$ introduces a quasi–momentum ${{\mathsf p}}_{{\mathfrak 0}}$ which has purely continuous spectrum (see equations [(\[badinf\])]{}, and [(\[eveninf\])]{} in Appendix \[monoapp\]). It follows that ${{\mathsf T}}(u)$ can not have any [*normalizable*]{} eigenstate. Instead one should work with the spectral representation for ${{\mathsf p}}_{{\mathfrak 0}}$, $
{{\mathcal H}}\,\simeq\,\int_{{{\mathbb R}}}dp_{{\mathfrak 0}}\;{{\mathcal H}}_{p_{{\mathfrak 0}}}
$ where the elements of $ {{\mathcal H}}_{p_{{\mathfrak 0}}}$ are represented by wave–functions of the form $
\Psi_{p_{{\mathfrak 0}}}({{\mathbf x}})=e^{2\pi i p_{{{\mathfrak 0}}}x_{{\mathfrak 0}}}
\Psi(x_{{{\rm N}}-1}-x_{{{\mathfrak 0}}},\dots,x_{{\mathfrak 1}}-x_{{\mathfrak 0}}).
$ This seems to complicate the analysis considerably. We nevertheless expect results similar to to hold for the remaining cases as well.
Self–duality and quantum Wronskian relation
-------------------------------------------
The explicit form [(\[Qxxz+\])]{}–[(\[Qxxz-\])]{} of the Q–operators along with the properties of $D_\alpha(x)$ listed in Appendix \[Qdil2\] show that ${{\mathsf Q}}^{\flat}_\pm(u)$ are self–dual with respect to the replacement . Therefore, the Q–operators also satisfy the dual Baxter equations, $$\label{baxdual}
\widetilde{{{\mathsf T}}}^{\flat}(u) \cdot {{\mathsf Q}}^{\flat}_\pm(u) =
\bigl( \tilde{a}^\flat(u) \bigr )^{{\rm N}}\,
{{\mathsf Q}}^{\flat}_\pm(u - ib^{-1}) +
\bigl( \tilde{d}^\flat(u) \bigr)^{{\rm N}}\,
{{\mathsf Q}}^{\flat}_\pm(u + ib^{-1}) \,,$$ where $\widetilde{{{\mathsf T}}}^{\flat}(u)$, $\flat = {{\rm{\scriptscriptstyle}XXZ}},\, {{\rm{\scriptscriptstyle}SG}}$, denote the transfer–matrices corresponding to the L–matrices (\[Lxxz\]) and (\[lSG\]) with $b$ replaced by $b^{-1}$. These are the transfer–matrices of the modular XXZ magnet and lattice sinh–Gordon model with ${\cal U}_{\tilde{q}}({\mathfrak sl}(2,{{\mathbb R}}))$ symmetry, where . The coefficients $\tilde{a}(u)$, $\tilde{d}(u)$ in [(\[baxdual\])]{} are similarly obtained from those in [(\[baxxz\])]{} by the replacement $b{\to}b^{-1}$. In the sinh–Gordon case the mass $m_{b^{-1}}$ is related to the representation parameter $s$ via .
This self–duality has remarkable consequences, which we shall work out explicitly for the case of the sinh–Gordon model with odd ${{\rm N}}$. We will take advantage of the freedom pointed out in Remark \[gaugerem\] to renormalize the operator ${{\mathsf Q}}^{{\rm{\scriptscriptstyle}SG}}(u)$ as follows: $$\label{checkSQdef}
\check{{\mathsf Q}}(u) \,\equiv\,\,
{{\mathsf Q}}_{+\infty}^*\,e^{-i{{\textstyle \frac{\pi }{2} }} {{\rm N}}\left( u^2 +
\sigma^2 + {\delta}_+^2 \right)} \,
{{\mathsf Q}}_+^{{\scriptscriptstyle}SG}(u)\,,$$ where ${\delta}_+=\frac{1}{2}(b+b^{-1})$ and ${{\mathsf Q}}_{+\infty}$ is the unitary operator which appears in the asymptotics [(\[qtas\])]{}. The Baxter equation for $\check{{\mathsf Q}}(u)$ will then take the following form: $$\label{baa2}
\begin{aligned}
{{\mathsf T}}^{{\scriptscriptstyle}SG}(u) \cdot \check{{{\mathsf Q}}}(u) &=
\bigl( \check a(u) \bigr)^{{\rm N}}\, \check{{{\mathsf Q}}}(u - ib) +
\bigl( \check d(u) \bigr)^{{\rm N}}\,
\check{{{\mathsf Q}}}(u + ib) \,,\\[2pt]
\text{where}\quad & \check d(u) = \check a(-u) =
1 + ({{\textstyle \frac{m \Delta}{4} }})^2 \, e^{ -\pi b (2u+ib)} \,.
\end{aligned}$$ The normalization of the operator $\check{{\mathsf Q}}(u)$ was for later convenience chosen in such a way that $\check{{\mathsf Q}}(u)\,\sim\, e^{\pm\pi i \sigma u -i\frac{\pi }{2}
{{\rm N}}\left( u^2 +\sigma^2 + {\delta}_+^2 \right)} \cdot {\sf 1}$ for ${\rm Re}(u) \to \pm\infty$.
The operator $\check{{\mathsf Q}}(u)$ fulfills the following quantum Wronskian relation: $$\label{q-Wr}
\check{{\mathsf Q}}(u+i{\delta}_+)\check{{\mathsf Q}}(u-i{\delta}_+)-
\check{{\mathsf Q}}(u+i{\delta}_-)\check{{\mathsf Q}}(u-i{\delta}_-)
\,=\,\, W_{{\rm N}}(u) \cdot {\sf 1} \,.$$ where $W_{{\rm N}}(u)=
e^{-i\pi{{\rm N}}\left(u^2 +
\sigma^2\right)}
\big(D_{\sigma}(u)\big)^{-{{\rm N}}}$ and ${\delta}_\pm\equiv\frac{1}{2}(b^{-1} \pm b)$.
Let ${{\mathsf W}}(u)$ be the left hand side of [(\[q-Wr\])]{}. A straightforward calculation, using the Baxter equation and its dual form, shows that ${{\mathsf W}}(u)$ satisfies the following two functional relations: $$\label{q-Wr-diff}
{{\mathsf W}}\big(u+{{\textstyle \frac{i}{2} }}b^{\pm 1})\,=\,
\left( e^{2\pi b^{\pm 1} u} \frac{\cosh\pi b^{\pm 1}(u-\sigma)}{\cosh\pi b^{\pm 1}(u+\sigma)}\right)^{{\rm N}}{{\mathsf W}}\big(u-{{\textstyle \frac{i}{2} }}b^{\pm 1})\,.$$ A solution to both functional relations is given by the expression on the right hand side of [(\[q-Wr\])]{}. If $b$ is irrational it suffices to notice that ${{\mathsf W}}(u)$ is meromorphic in order to conclude that the solution to the system [(\[q-Wr-diff\])]{} must be unique up to multiplication by an operator which does not depend on $u$. This freedom can be fixed by comparing the asymptotics of both sides for ${\rm Re}(u)\to\pm\infty$ using equations [(\[qtas\])]{},[(\[QQpar\])]{} (for considering the asymptotics as ${\rm Re}(u)\to-\infty$, it is helpful to notice that ${{\mathsf \Omega}}^2={\sf 1}$). In order to cover the case of rational $b$ let us notice that both sides of the relation [(\[q-Wr\])]{} can be analytically continued from irrational values of $b$ to the case where $b$ is rational.
From the proof of this theorem it is clear that the main ingredient is the self–duality of our representations ${{\mathcal P}}_s$. We therefore expect that a similar result will hold for the remaining cases as well. However, at present we do not control the asymptotics of the Q–operators sufficiently well in these cases.
Parity and cyclic shift
-----------------------
Let us consider the cyclic shift and parity operators ${{\mathsf U}}$ and ${{\mathsf \Omega}}$ defined respectively by $$\begin{aligned}
\label{upar}
{{\mathsf U}}\, f(x_{{\mathfrak 1}},x_{{\mathfrak 2}},\ldots,x_{{\rm N}}) = f(x_{{\mathfrak 2}},\ldots,x_{{\rm N}},x_{{\mathfrak 1}})\,,\qquad
{{\mathsf \Omega}}\, f(x_{{\mathfrak 1}},\ldots,x_{{\rm N}}) = f(-x_{{\mathfrak 1}},\ldots,-x_{{\rm N}})\,.\end{aligned}$$ These operators commute with each other and also with ${{\mathsf Q}}^{\flat}_\pm(u)$ (as can be easily seen from (\[Qxxz+\])–(\[Qxxz-\])). Hence they must commute with ${{\mathsf T}}^{\flat}_\pm(u)$, as can also be verified directly[^2]. It follows that eigenstates $\Psi_t$ may be assumed to be simultaneously eigenstates of ${{\mathsf U}}$ and ${{\mathsf \Omega}}$, $$\label{UWP}
{{\mathsf \Omega}}\, \Psi_t = \pm \Psi_t \,, \qquad
{{\mathsf U}}\, \Psi_t = e^{2\pi i \frac{m}{{{\rm N}}}} \, \Psi_t \,, \qquad
m=1,\ldots,{{\rm N}}\,.$$ It is therefore useful to observe that ${{\mathsf U}}$ and ${{\mathsf \Omega}}$ can be recovered from ${{\mathsf Q}}^\flat_\pm(u)$ as follows. First one may notice that the integral kernel (\[Qxxz+\]) simplifies for the special values $u= \pm \sigma$ thanks to the relation (\[Ddel\]). Explicitely, we have $$\begin{aligned}
\label{Qsp}
{{\mathsf Q}}_+^{\flat}(\sigma) &=
\bigl( w_b(2s{+}{{\textstyle \frac{i}{2} }}Q) \bigr)^{-{{\rm N}}} \cdot {\sf 1} \,, \qquad
{{\mathsf Q}}_+^{\flat}(-\sigma) = \bigl( w_b(2s{+}{{\textstyle \frac{i}{2} }}Q)\bigr)^{-{{\rm N}}}
\cdot {{\mathsf U}}^{-1} \cdot {{\mathsf \Omega}}^\flat \,,\end{aligned}$$ where ${{\mathsf \Omega}}^\flat$ is defined in (\[Wb\]). Combining this observation with (\[UWP\]), we conclude that $$\label{qspm}
q^+_t(\sigma) = \bigl( w_b(2s{+}{{\textstyle \frac{i}{2} }}Q) \bigr)^{-{{\rm N}}} \,, \qquad
q^+_t(-\sigma) = \pm e^{2\pi i \frac{m}{{{\rm N}}}} \,
\bigl( w_b(2s{+}{{\textstyle \frac{i}{2} }}Q)\bigr)^{-{{\rm N}}} \,,
$$ where the $-$ sign in the second expression can occur only in the sinh–Gordon model.
Secondly let us observe that eqs. (\[qqpm\]) and [(\[qspm\])]{} imply that $u=\pm \bar{\sigma}$ are poles of order ${{\rm N}}$ for $q^-_t(u)$. Indeed, using (\[Qxxz-\]), (\[Ddel\]), and (\[wres\]), we find $$\begin{aligned}
\label{Qsm}
{{\mathsf Q}}_-^{\flat}(\bar{\sigma} + \epsilon) \sim
\bigl( {{\textstyle \frac{i}{2\pi\epsilon} }} \bigr)^{{\rm N}}\cdot {\sf 1} \,, \qquad
{{\mathsf Q}}_-^{\flat}(-\bar{\sigma} + \epsilon) \sim
\bigl( {{\textstyle \frac{1}{2\pi i \epsilon} }} \bigr)^{{\rm N}}\cdot
{{\mathsf \Omega}}^\flat \cdot {{\mathsf U}}\,,\end{aligned}$$ as $\epsilon\to 0$. This implies $$\label{qmspm}
q^-_t(\bar{\sigma} + \epsilon) \sim
\bigl( {{\textstyle \frac{i}{2\pi\epsilon} }} \bigr)^{{\rm N}}\,, \qquad
q^-_t(-\bar{\sigma} + \epsilon) \sim
\pm e^{2\pi i \frac{m}{{{\rm N}}}} \,
\bigl( {{\textstyle \frac{1}{2\pi i \epsilon} }} \bigr)^{{\rm N}}\,,
$$ where the $-$ sign in the second expression can occur again only in the sinh–Gordon model.
Separation of variables {#SOV}
=======================
In the previous section we have identified necessary conditions for a function $t(u)$ to be an eigenvalue of ${{\mathsf T}}(u)$. If we were able to show that these conditions are also [*sufficient*]{}, we would have arrived at a useful reformulation of the Auxiliary Spectral Problem.
A promising approach to this problem is offered by the separation of variables method pioneered by Sklyanin [@Sk2; @Sk3]. The basic idea is to introduce a representation for the Hilbert space ${{\mathcal H}}$ of the model in which the off–diagonal element of the monodromy matrix ${{\mathsf M}}(u)$, the operator ${{\mathsf B}}(u)$, is diagonal.
For simplicity of exposition let us temporarily restrict attention to the case of the sinh–Gordon model with ${{\rm N}}$ odd. The operator ${{\mathsf B}}(u)$ has the following form: $$\label{SBexp}
{{\mathsf B}}(u)\,=\,- i^{{{\rm N}}} \, e^{{{\rm N}}\pi b (u-s)} \sum_{m=0}^{{{\rm N}}} (-)^m
e^{-2m\pi b u} \, {{\mathsf B}}_{m}\,.$$ By Lemma \[poslem\], the operators ${{\mathsf B}}_{m}$, $m=0,\dots,{{\rm N}}$ are positive self–adjoint. Basic for the separation of variables method is the validity of the following conjecture.
\[simplicity\] The joint spectrum of the family of operators $\{{{\mathsf B}}_{m};m=0,\dots,{{\rm N}}\}$ is simple. This means that eigenstates of ${{\mathsf B}}(u)$ are uniquely parameterized by the corresponding eigenvalue $b(u)$.
This conjecture can be supported by counting the degrees of freedom. However, it is not easy to provide a rigorous proof (see also Remark \[Bspecrem\] below).
The function $e^{-\pi b {{\rm N}}u}b(u)$ is a polynomial in the variable ${\lambda}=e^{-2\pi b u}$. It can conveniently be represented in the following form $$\label{bfactor}
b(u) \,= \,-(2i)^{{{\rm N}}} \, e^{-\pi b {{\rm N}}s}\,
\prod_{k=1}^{{{\rm N}}} \sinh\pi b(u-y_k)\,.$$ The variables $y_k$, $k=1,\dots,{{\rm N}}$ are uniquely defined up to permutations once we adopt the convention that ${\rm Im}(y_k)\in
(-{{\textstyle \frac{1}{2b} }},{{\textstyle \frac{1}{2b} }}].$ This means that the representation for the Hilbert space ${{\mathcal H}}$ in which ${{\mathsf B}}(u)$ is diagonal may be described by wave–functions $\Psi({{\mathbf y}})$, ${{\mathbf y}}=(y_1^{},\dots,y_{{{\rm N}}}^{})$. This representation for the vectors in ${{\mathcal H}}$ will subsequently be referred to as the SOV representation.
We will then show that the Auxiliary Spectral Problem, ${{\mathsf T}}(u)\Psi_t=t(u)\Psi_t$, gets transformed into the system of Baxter equations $$\label{baxtereqn}
\begin{aligned}
{}& t(y_k) \, \Psi({{\mathbf y}}) = \Big[\,\big(a(y_k)\big)^{{\rm N}}\,
{{\mathsf T}}_k^{-}+\big(d(y_k)\big)^{{\rm N}}\,{{\mathsf T}}_k^{+}\,\Big]\,\Psi({{\mathbf y}})
\,,\quad k =1,\dots,{{\rm N}}\,,
\end{aligned}$$ where the operators ${{\mathsf T}}_k$ are shift operators defined as $$\label{shiftdef}
{{\mathsf T}}_k^\pm\Psi({{\mathbf y}})\,\equiv\,\Psi(y_1,\dots, y_k\pm ib,\dots,
y_{{{\rm N}}}).$$ The coefficients in front of the shift operators in (\[baxtereqn\]) depend only on a single variable $y_k$, which is the crucial simplification that is gained by working in the representation where ${{\mathsf B}}(u)$ is diagonal.
The key observation to be made at this point is that the [*same*]{} finite difference equation [(\[baxtereqn\])]{} was found in the previous Section \[QB\] in connection with the necessary conditions for a function $t(u)$ to represent a point in the spectrum. It now remains to observe that [*any*]{} function $q_t(u)$ that fulfills the necessary conditions [(\[BaxterEV\])]{},[(\[analQ\])]{} can be used to construct $$\label{factorrep}
\Psi_t({{\mathbf y}})\,=\, \prod_{k=1}^{{{\rm N}}}\,q_t(y_k) \,.$$ The fact that [(\[factorrep\])]{} defines an eigenstate of ${{\mathsf T}}(u)$ is verified by comparing [(\[baxtereqn\])]{} with [(\[BaxterEV\])]{}. The main point that needs to be verified is whether the function $\Psi_t({{\mathbf y}})$ actually represents an element of ${{\mathcal H}}$, i.e. whether it has finite norm. The scalar product of vectors in ${{\mathcal H}}$ can be represented in the form $$\label{SOVscprod}
\langle\,\Psi_{{\mathfrak 2}}\,|\,\Psi_{{\mathfrak 1}}\,\rangle\,=\,
\int_{\mathbb Y} d\mu({{\mathbf y}})\;
\langle\,\Psi_{{\mathfrak 2}}\,|\,{{\mathbf y}}\,\rangle\langle\,{{\mathbf y}}\,|\,\Psi_{{\mathfrak 1}}\,\rangle\,$$ We clearly need to know both the range ${\mathbb Y}$ of values ${{\mathbf y}}$ that we need to integrate over, as well as the measure $d\mu({{\mathbf y}})$ of integration to be used. We will be able to determine the measure $d\mu({{\mathbf y}})$ provided that the following conjecture is true.
\[Bspec\] The functions $b(u)$ of the product form [(\[bfactor\])]{} that describe the spectrum of ${{\mathsf B}}(u)$ have only real roots, i.e., $y_k\in{{\mathbb R}}$ for $k=1,\dots,{{\rm N}}$.
We will discuss the status of this conjecture after having explained its consequences. Conjecture \[Bspec\] directly implies that ${\mathbb Y}={{\mathbb R}}^{{{\rm N}}}$ in [(\[SOVscprod\])]{}. Assuming the validity of Conjecture \[Bspec\], we will show in Proposition \[Sklmeasure\] that the measure $d\mu({{\mathbf y}})$ can be represented in the following explicit form $$\label{mexx'}
d\mu({{\mathbf y}})\,=\, \prod_{k=1}^{{{\rm N}}}\,dy_k
\prod_{l<k}\,4 \sinh\pi b(y_k-y_l)\, \sinh\pi b^{-1}(y_k-y_l)\,.$$
Knowing explicitly how to represent the scalar product of ${{\mathcal H}}$ in the SOV representation finally allows us to check that [*any solution of the necessary conditions [(\[BaxterEV\])]{},[(\[analQ\])]{} defines an eigenvector $|\,\Psi_t\,{\rangle}$ of ${{\mathsf T}}(u)$ via [(\[factorrep\])]{}.*]{} In other words: The conditions [(\[BaxterEV\])]{}, [(\[analQ\])]{} are not only necessary but also [*sufficient*]{} for $t(u)$ to be an eigenvalue of a vector $|\,\Psi_t\,{\rangle}\in{{\mathcal H}}$.
\[Bspecrem\] Our claim that the conditions [(\[BaxterEV\])]{}, [(\[analQ\])]{} are also sufficient for a function $t(u)$ to represent a point in the spectrum of ${{\mathsf T}}(u)$ does not seem to depend very strongly on the validity of the Conjecture \[Bspec\]. In this sense the conjecture mainly serves us to simplify the exposition.
In any case it is a problem of fundamental importance for the separation of variable method to determine the spectrum of ${{\mathsf B}}(u)$ precisely. Even in simpler models which have been studied along similar lines like the Toda chain [@KL] or the XXX spin chains [@DKM] there does not seem to exist a rigorous proof of the analogous statements. The explicit construction of the eigenfunctions of ${{\mathsf B}}(u)$, which may proceed along similar lines as followed for the Toda chain in [@KL] or for the XXX chain in [@DKM], should provide us with the basis for a future proof of Conjecture \[Bspec\] or some modification thereof.
Within the Separation of Variables method the Auxiliary Spectral Problem gets transformed into the separated Baxter equations [(\[baxtereqn\])]{}. However, these finite difference equations will generically have many solutions that do [*not*]{} correspond to eigenstates of ${{\mathsf T}}(u)$.
In order to draw a useful analogy let us compare the situation with the spectral problem for a differential operator like ${{\mathsf h}}=-\partial_y^2+V(y)$. One generically has two linearly independent solutions to the second order differential equation like $(-\partial_y^2+V(y))\psi={\cal E}\psi$ for [*any*]{} choice of ${\cal E}$. The spectrum of ${{\mathsf h}}$ is determined by restricting attention to the subset of square–integrable solutions within the set of all solutions to the eigenvalue equation.
From this point of view we may identify the conditions [(\[analQ\])]{} on analyticity and asymptotics of the function $q_t(u)$ as the quantization conditions which single out the subset which constitutes the spectrum of ${{\mathsf T}}(u)$ among the set of [*all*]{} solutions of [(\[baxtereqn\])]{}.
Operator zeros of ${{\mathsf B}}(u)$
------------------------------------
The adaption of Sklyanin’s observation to the case at hand is based on the following observations. First, by Lemma \[poslem\], the operators ${{\mathsf B}}_{m}$ introduced in [(\[SBexp\])]{} are [positive]{} self–adjoint. Taking into account the mutual commutativity (which follows from [(\[ABD\])]{}), $${[}\,{{\mathsf B}}_{m}\,,\,{{\mathsf B}}_{n}\,{]}\,=\,0,$$ leads us to conclude that the family of operators $\{{{\mathsf B}}_m;m=1,\dots {{\rm N}}\}$ can be simultaneously diagonalized[^3].
Conjecture \[simplicity\] implies that the spectral representation for the family $\{{{\mathsf B}}_m;m=1,\dots {{\rm N}}\}$ can be written in the form $$\label{bspec}
|\Psi\rangle\,=\, \int_{{{\mathbb R}}_+^{{\rm N}}}
d\nu({{\mathbf b}})\,|\,{{\mathbf b}}\,\rangle\,\langle\,{{\mathbf b}}\,|\,\Psi\,\rangle\,$$ where $|\,{{\mathbf b}}\,\rangle$ is a (generalized) eigenvector of ${{\mathsf B}}_m$ with eigenvalue $b_m$, and we have assembled the eigenvalues into the vector ${{\mathbf b}}=(b_1^{},\dots,b_{{{\rm N}}}^{})$.
It now turns out to be particularly useful to parameterize the polynomial of eigenvalues $b(u)$ in terms of its roots. This representation may always be written as follows $$\label{b(u)exp}
b(u) \equiv b(u|{{\mathbf y}})\equiv
- (2i)^{{{\rm N}}} \,
e^{- \pi b {{\rm N}}s } \,
\prod\limits_{k=1}^{{{\rm N}}} \sinh\pi b(u-y_k) \,,$$ where ${{\mathbf y}}=(y_1^{},\dots,y_{{\rm N}}^{})$. The variables $y_k$ are either real or they come in pairs related by complex conjugation. The variables $y_k$ are uniquely defined up to permutations if one requires that ${\rm Im}(y_k)\in (-\frac{1}{2b},\frac{1}{2b}]$. We will assume that $y_k\in{{\mathbb R}}$ according to Conjecture \[Bspec\]. It then follows that the spectral representation [(\[bspec\])]{} can be rewritten as $$\label{bspec2}
|\Psi\rangle\,=\,
\int_{{{\mathbb R}}^{{\rm N}}} d\mu({{\mathbf y}})\,|\,{{\mathbf y}}\,\rangle\,
\langle\,{{\mathbf y}}\,|\,\Psi\,\rangle\,.$$ However, points ${{\mathbf y}}$, ${{\mathbf y}}'$ in ${{\mathbb R}}^{{\rm N}}$ which are obtained from each other by the permutation $y_k\leftrightarrow y_l$ will correspond to the same eigenstate of ${{\mathsf B}}(u)$. This means that the spectral representation for ${{\mathsf B}}(u)$ can be used to define an isomorphism $$\label{CHiso}
{{\mathcal H}}\,\simeq\,L^2({{\mathbb R}}^{{\rm N}};d\mu)^{\rm Symm}\,,$$ where $L^2({{\mathbb R}}^{{\rm N}};d\mu)^{\rm Symm}$ is the subspace within $L^2({{\mathbb R}}^{{\rm N}};d\mu)$ which consists of totally symmetric wave–functions.
Despite the fact that ${{\mathcal H}}$ is isomorphic only to a subspace in $L^2({{\mathbb R}}^{{\rm N}};d\mu)$ it will turn out to be useful to extend the definition of the operators ${{\mathsf A}}(u)$, ${{\mathsf B}}(u)$, ${{\mathsf C}}(u)$, ${{\mathsf D}}(u)$ from $L^2({{\mathbb R}}^{{\rm N}};d\mu)^{\rm Symm}$, where it is canonically defined via [(\[CHiso\])]{} to $L^2({{\mathbb R}}^{{\rm N}};d\mu)$. As a first step let us introduce the operators ${{\mathsf y}}_k$ which act as ${{\mathsf y}}_k\,|\,{{\mathbf y}}\,\rangle=y_k\,|\,{{\mathbf y}}\,\rangle$. Substituting $y_k{\to}{{\mathsf y}}_k$ in [(\[b(u)exp\])]{} leads to a representation of the operator ${{\mathsf B}}(u)$ in terms of its [*operators zeros*]{} ${{\mathsf y}}_k$.
Operators ${{\mathsf A}}(u)$ and ${{\mathsf D}}(u)$
---------------------------------------------------
Monodromy matrices of the modular XXZ magnet and the lattice sinh–Gordon model satisfy the exchange relations (\[rLL\]), where the R–matrix is given by (\[R12\]) and (\[Rlsg\]), respectively. Among these relations we have, in particular, the following $$\begin{aligned}
\label{ABD}
{}& [{{\mathsf B}}(u) , {{\mathsf B}}(v)] = 0 \,, \qquad
[{{\mathsf A}}(u) , {{\mathsf A}}(v) ]= 0 \,,\qquad
[{{\mathsf D}}(u) , {{\mathsf D}}(v) ]= 0 \,, \\
\label{AD}
& \sinh \pi b (u-v+ib) \, {{\mathsf B}}(u) \, {{\mathsf A}}(v) \\
\nonumber
& \qquad\qquad\quad = \sinh \pi b (u-v) \, {{\mathsf A}}(v) \, {{\mathsf B}}(u)
+ R^{\,\flat}_{23}(u-v) \, {{\mathsf B}}(v) \, {{\mathsf A}}(u)\,,\\
\label{BD}
& \sinh \pi b (u-v+ib) \, {{\mathsf B}}(v){{\mathsf D}}(u) \\
\nonumber
& \qquad\qquad\quad =
\sinh \pi b (u-v) \, {{\mathsf D}}(u) \, {{\mathsf B}}(v)
+ R^{\,\flat}_{32}(u-v) \, {{\mathsf B}}(u) \, {{\mathsf D}}(v) \,,\\[1ex]
\label{R23}
& R^{{\rm{\scriptscriptstyle}XXZ}}_{23}(u) = R^{{\rm{\scriptscriptstyle}XXZ}}_{32}(-u) =
i e^{\pi b u} \, \sin\gamma \,, \qquad
R^{{\rm{\scriptscriptstyle}SG}}_{23}(u) = R^{{\rm{\scriptscriptstyle}SG}}_{32}(u) =
i \sin\gamma \,.\end{aligned}$$
We are now going to show that there is an essentially unique representation of the commutation relations [(\[ABD\])]{}–[(\[BD\])]{} on wave–functions $\Psi({{\mathbf y}})$ which is such that ${{\mathsf B}}(u)$ is represented as operator of multiplication by $b(u)\equiv b(u|{{\mathbf y}})$, cf. [(\[b(u)exp\])]{}.
To this aim let us consider for $k\geq 1$ the following distributions: $$\langle{{\mathbf y}}|\,{{\mathsf A}}(y_k)\,\equiv\,
\lim_{u{\to}y_k}\langle{{\mathbf y}}|\,{{\mathsf A}}(u)\,,
\qquad
\langle{{\mathbf y}}|\,{{\mathsf D}}(y_k)\,\equiv\,
\lim_{u{\to}y_k}\langle{{\mathbf y}}|\,{{\mathsf D}}(u)\,.$$ These distributions, which will be defined on suitable dense subspaces of $L^2({{\mathbb R}}^{{{\rm N}}},d\mu)$, can be regarded as the result of action on $\langle{{\mathbf y}}|$ by operators ${{\mathsf A}}({{\mathsf y}}_k)$ and ${{\mathsf D}}({{\mathsf y}}_k)$ with operator arguments substituted into (\[aexp\]), (\[dexp\]), (\[aaexp\]), (\[ddexp\]) from the left. The commutation relations (\[AD\])–(\[BD\]) imply that $\langle{{\mathbf y}}|\,{{\mathsf A}}(y_k)$ and $\langle{{\mathbf y}}|\,{{\mathsf D}}(y_k)$ are eigenstates of ${{\mathsf B}}(u)$ with eigenvalues $b(u|{{\mathbf y}}')$, with $y_k'= y_k \mp ib$, respectively, $y'_l=y_l$ otherwise. This, along with relations (\[ABD\]), leads to the conclusion that the action of the operators ${{\mathsf A}}({{\mathsf y}}_k)$ and ${{\mathsf D}}({{\mathsf y}}_k)$ on wave–functions $\Psi({{\mathbf y}})=\langle{{\mathbf y}}|\Psi\rangle$ can be represented in the form $$\label{SADshift}
{{\mathsf A}}({{\mathsf y}}_k)\Psi({{\mathbf y}})\,=\,a_{{{\rm N}},k}(y_k)\,{{\mathsf T}}_k^{-}\Psi({{\mathbf y}})\,,\qquad
{{\mathsf D}}({{\mathsf y}}_k)\Psi({{\mathbf y}})\,=\,d_{{{\rm N}},k}(y_k)\,{{\mathsf T}}_k^{+}\Psi({{\mathbf y}})\,,$$ where ${{\mathsf T}}_k^\pm$ are the shift operators defined in equation [(\[shiftdef\])]{}. The functions $a_{{{\rm N}},k}(y_k)$ and $d_{{{\rm N}},k}(y_k)$ are further restricted by the following identities: $$\begin{aligned}
\det\nolimits_q {{\mathsf M}}^{{\rm{\scriptscriptstyle}SG}}(u) &\equiv
{{\mathsf A}}^{{\rm{\scriptscriptstyle}SG}}(u) \, {{\mathsf D}}^{{\rm{\scriptscriptstyle}SG}}(u-ib) - {{\mathsf B}}^{{\rm{\scriptscriptstyle}SG}}(u) \, {{\mathsf C}}^{{\rm{\scriptscriptstyle}SG}}(u-ib) \\
{} &= \bigl( 4 e^{-2\pi b s}\, \cosh \pi b( s + u - i {{\textstyle \frac{b}{2} }}) \,
\cosh \pi b( s - u + i {{\textstyle \frac{b}{2} }}) \bigr)^{{\rm N}}\,.\end{aligned}$$ These identities are proven in Appendix \[qDet\]. It follows then from [(\[SADshift\])]{} that $\det\nolimits_q {{\mathsf M}}(y_k) =a_{{{\rm N}},k}(y_k)d_{{{\rm N}},k}(y_k-ib)$. Not having specified the measure $\mu({{\mathbf y}})$ yet leaves us the freedom to multiply all wave–functions $\Psi({{\mathbf y}})$ by functions of the form $\prod_{k} f_k(y_k)$. This allows us to choose $$\begin{aligned}
\label{acoeff}
& a_{{{\rm N}},k}(y_k)= \bigl(a(y_k)\bigr)^{{\rm N}}\,, \qquad\qquad
d_{{{\rm N}},k}(y_k) = \bigl( d(y_k) \bigr)^{{\rm N}}\,,\\[1ex]
& \begin{aligned}
\text{where}\;\;a^{{\rm{\scriptscriptstyle}SG}}(u)=
d^{{\rm{\scriptscriptstyle}SG}}(-u)&=e^{-\pi b s}\,2\cosh\pi b (u-s-i{{\textstyle \frac{b}{2} }})\\
&= e^{-\pi b(u-i\frac{b}{2})}+
\big({{\textstyle \frac{m\Delta}{4} }}\big)^2 e^{\pi b(u-i\frac{b}{2})}\,.
\label{adcoeffs}\end{aligned}\end{aligned}$$ We have used that the sinh–Gordon parameters $m$, $\Delta$ are related to $s$ as in (\[mds\]).
In the special case that $\Psi({{\mathbf y}})$ is an eigenfunction of the transfer–matrix ${{\mathsf T}}(u)$ with eigenvalue $t(u)$ we get the Baxter equations [(\[baxtereqn\])]{} from ${{\mathsf T}}({{\mathsf y}}_k)={{\mathsf A}}({{\mathsf y}}_k)+{{\mathsf D}}({{\mathsf y}}_k)$ and equations [(\[SADshift\])]{} and [(\[acoeff\])]{}, as advertised.
It will be useful for us to have explicit formulae for ${{\mathsf A}}(u)$ and ${{\mathsf D}}(u)$ in terms of the operators ${{\mathsf y}}_k$ and ${{\mathsf T}}_k^{\pm}$. In the case of ${{\rm N}}$ odd we may use the following formulae: $$\begin{aligned}
{{\mathsf A}}^{{\rm{\scriptscriptstyle}SG}}(u)&=
\sum_{k=1}^{{{\rm N}}}\,\prod_{l\neq k}\,
\frac{\sinh\pi b(u-{{\mathsf y}}_l)}{\sinh\pi b ({{\mathsf y}}_k-{{\mathsf y}}_l)}\,
\bigl(a^{{\rm{\scriptscriptstyle}SG}}({{\mathsf y}}_k)\bigr)^{{\rm N}}\,{{\mathsf T}}_k^{-}\, , \\
{{\mathsf D}}^{{\rm{\scriptscriptstyle}SG}}(u)&=
\sum_{k=1}^{{{\rm N}}}\,\prod_{l\neq k}\,
\frac{\sinh\pi b(u-{{\mathsf y}}_l)}{\sinh\pi b ({{\mathsf y}}_k-{{\mathsf y}}_l)}\,
\bigl(d^{{\rm{\scriptscriptstyle}SG}}({{\mathsf y}}_k)\bigr)^{{\rm N}}\,{{\mathsf T}}_k^{+}\,.\end{aligned}$$ These formulae are easily verified by noting that the number of variables ${{\mathsf y}}_k$ coincides with the number of coefficients in the expansion (\[aaexp\]) and (\[ddexp\]). It follows that the polynomials ${{\mathsf A}}^{{\rm{\scriptscriptstyle}SG}}(u)$ and ${{\mathsf D}}^{{\rm{\scriptscriptstyle}SG}}(u)$ are uniquely determined by their values ${{\mathsf A}}^{{\rm{\scriptscriptstyle}SG}}({{\mathsf y}}_k)$ and ${{\mathsf D}}^{{\rm{\scriptscriptstyle}SG}}({{\mathsf y}}_k)$, $k=1,\dots,{{\rm N}}$.
Sklyanin measure
----------------
We furthermore know that the operators ${{\mathsf A}}_m$ and ${{\mathsf D}}_m$ which are defined by the expansion $$\begin{aligned}
{{\mathsf A}}^{{\rm{\scriptscriptstyle}SG}}(u)&= -ie^{-\pi bu} i^{ {{\rm N}}} \,
e^{ \pi b {{\rm N}}(u- s)}
\sum_{m=0}^{{{\rm N}}-1} (-)^m
e^{-2m\pi b u} \, {{\mathsf A}}_{m}\,,\\
{{\mathsf D}}^{{\rm{\scriptscriptstyle}SG}}(u)&= -ie^{-\pi bu} i^{ {{\rm N}}} \,
e^{ \pi b {{\rm N}}(u- s)}
\sum_{m=0}^{{{\rm N}}-1} (-)^m
e^{-2m\pi b u} \, {{\mathsf D}}_{m}\,\end{aligned}$$ are positive (Lemma \[poslem\] in Appendix \[monoapp\]).
\[Sklmeasure\] There exists a unique measure $d\mu({{\mathbf y}})$ such that the operators ${{\mathsf A}}_m$ and ${{\mathsf D}}_m$ on $L^2({{\mathbb R}}^{{\rm N}};d\mu)$ are positive. This measure $d\mu$ can be represented explicitly as $$\label{mexx}
d\mu({{\mathbf y}})\,=\, \prod_{k=1}^{{{\rm N}}}\,dy_k \,
\prod_{l<k}\,4 \sinh\pi b(y_k-y_l) \, \sinh\pi b^{-1}(y_k-y_l)\,.$$
The similarity transformation $\Psi({{\mathbf y}})= \chi_{\scriptscriptstyle{{\mathsf A}}}({{\mathbf y}}) \, \Phi({{\mathbf y}})$, where $$\label{similarA}
\chi_{\scriptscriptstyle{{\mathsf A}}}({{\mathbf y}}) \,=\,\prod_{k=1}^{{{\rm N}}}\,
\big(e^{\pi i y_ks} w_b(y_k-s)\big)^{-{{\rm N}}}
\prod_{l<k} \bigl(w_b(y_k-y_l+ {{\textstyle \frac{i}{2} }}Q)\bigr)^{-1} \,,$$ maps to a representation in which the operator ${{\mathsf A}}^{{\rm{\scriptscriptstyle}SG}}(u)$ is represented as $$\begin{aligned}
{{\mathsf A}}^{{{\rm{\scriptscriptstyle}SG}}}(u)\,=\,\sum_{k=1}^{{{\rm N}}}\,\prod_{l\neq k}\,
{\sinh\pi b(u-{{\mathsf y}}_l)}\,
{{\mathsf T}}_k^{-}\, .\end{aligned}$$ Expanding in powers of $e^{\pi b u}$ yields a representation for the coefficients ${{\mathsf A}}^{{{\rm{\scriptscriptstyle}SG}}}_{m}$ that appear in the expansion [(\[aexp\])]{} which takes the form $$\label{SDexp}
{{\mathsf A}}^{{{\rm{\scriptscriptstyle}SG}}}_{m} \,=\,
\sum_{k=1}^{{{\rm N}}}p_{mk}({{\mathbf y}}){{\mathsf T}}_k^{-}\,,$$ The coefficients $p_{mk}({{\mathbf y}})$ in [(\[SDexp\])]{} are positive for all ${{\mathbf y}}\in{{\mathbb R}}^{{{\rm N}}}$, and $p_{mk}({{\mathbf y}})$ does not depend on $y_k$. We are next going to show that the positivity of ${{\mathsf A}}^{{{\rm{\scriptscriptstyle}SG}}}_{m}$ implies that ${{\mathsf T}}_k^{-}$ must be a positive operator in $L^2({{\mathbb R}}^{{\rm N}};d\mu)$. Let us keep in mind that ${{\mathsf T}}_k^{-}$ satisfies the commutation relations $$\label{Cweyl}
e^{-it{{\mathsf y}}_l}\,{{\mathsf T}}_k^{-}\,e^{it{{\mathsf y}}_l}\,=\,e^{bt{\delta}_{kl}}\,{{\mathsf T}}_k^{-} \,.$$ If there was any negative contribution to the expectation value $\langle \Phi\,|\,\sum_{k=1}^{{{\rm N}}}p_{mk}({{\mathbf y}})
{{\mathsf T}}_k^{-}\,|\,\Phi\rangle$ we could make it arbitrarily large by means of the unitary transformation $|\Phi\rangle{\to}e^{it{{\mathsf y}}_k}|\Phi\rangle$. It follows that $\left\langle \Phi\,|\,p_{mk}({{\mathbf y}})
{{\mathsf T}}_k^{-}\,|\,\Phi\right\rangle>0$ for any $k=1,\dots,{{\rm N}}$.
It remains to notice that, since the $p_{mk}({{\mathbf y}})$ are non–vanishing, vectors of the form $\sqrt{p_{mk}({{\mathbf y}})}\,|\,\Phi\,{\rangle}$ form a dense subset in $L^2({{\mathbb R}}^{{\rm N}};d\mu)$. This finally allows us to conclude that ${{\mathsf T}}_k^-$ must be a positive operator. But this furthermore implies that $({{\mathsf T}}^{-}_k)^{il}$ is a unitary operator which satisfies the commutation relations $$({{\mathsf T}}^{-}_k)^{il}\,e^{it{{\mathsf y}}_m}\,=\,
\exp(iblt{\delta}_{km})\,e^{it{{\mathsf y}}_m}\,({{\mathsf T}}^{-}_k)^{il}\,.$$ It is well–known that the representation of these commutation relations by unitary operators is essentially unique. The measure which defines the corresponding Hilbert space is just $d\nu({{\mathbf y}})=\prod_{k=1}^{{{\rm N}}}dy_k$.
It remains to return in (\[bspec2\]) to the original representation via [(\[similarA\])]{}, that is to compute . Using [(\[wdual\])]{}–[(\[wcc\])]{} to simplify the resulting expression yields the formula for $d\mu({{\mathbf y}})$ stated in Proposition \[Sklmeasure\].
For the operator $D^{{\rm{\scriptscriptstyle}SG}}(u)$, a completely analogous consideration applies with the transformation $\Psi({{\mathbf y}})= \chi_{\scriptscriptstyle{{\mathsf D}}}({{\mathbf y}}) \, \Phi({{\mathbf y}})$, where $$\label{similarD}
\chi_{\scriptscriptstyle{{\mathsf D}}}({{\mathbf y}}) \,=\,\prod_{k=1}^{{{\rm N}}}\,
\big(e^{-\pi i y_ks} w_b(y_k+s)\big)^{{{\rm N}}}
\prod_{l<k} \bigl(w_b(y_k-y_l+ {{\textstyle \frac{i}{2} }}Q)\bigr)^{-1} \,.$$ Thanks to the relation (\[wcc\]) and the Conjecture \[Bspec\], we have $|\chi_{\scriptscriptstyle{{\mathsf A}}}({{\mathbf y}})|^2=
|\chi_{\scriptscriptstyle{{\mathsf D}}}({{\mathbf y}})|^2$. This leads to the same measure $d\mu({{\mathbf y}})$ given by [(\[mexx\])]{}.
Remaining cases
---------------
To end this section let us briefly discuss the necessary modification in the cases of the modular XXZ magnet and the lattice sinh–Gordon model with even ${{\rm N}}$. The main new feature that arises in these cases is the existence of a quasi–momentum $y_{{\mathfrak 0}}$ which first appears in the expansions $$\begin{aligned}
\label{opzero2a}
&b^{{\rm{\scriptscriptstyle}XXZ}}(u) = 2^{{{\rm N}}-1} \, i \, e^{\pi b (u + y_{{\mathfrak 0}})}\,
\prod_{k=1}^{{{\rm N}}-1} \sinh\pi b(u-y_k) \,,\\
\label{opzero2b}
&b^{{\rm{\scriptscriptstyle}SG}}(u) = - (2i)^{{{\rm N}}-1} \,
e^{\pi b ( y_{{\mathfrak 0}}- {{\rm N}}s) } \,
\prod\limits_{k=1}^{{{\rm N}}-1} \sinh\pi b(u-y_k) \,,
\;\;\text{${{\rm N}}$ -- even} \,,\end{aligned}$$ The variable $y_{{\mathfrak 0}}$ requires a slightly different treatment compared to the $y_k$, $k\geq 1$. Considering the states (where $\kappa=0$ for the modular magnet and for the sinh–Gordon model with even ${{\rm N}}$) $$\begin{aligned}
{}& \langle{{\mathbf y}}|\,{{\mathsf A}}_{{{\mathfrak 0}}} \equiv
\lim_{u\to +\infty} e^{\pi {{\rm N}}(\kappa -bu)} \,
\langle{{\mathbf y}}|\,{{\mathsf A}}(u)\,,
\\
{}& \langle{{\mathbf y}}|\,{{\mathsf D}}_{{{\mathfrak 0}}} \equiv
\lim_{u\to +\infty} e^{\pi {{\rm N}}(\kappa - bu)} \,
\langle{{\mathbf y}}|\,{{\mathsf D}}(u) \,,\\
{}& \langle{{\mathbf y}}|\,{{\mathsf A}}_{{{\rm N}}} \equiv
\lim_{u\to -\infty} (-)^{{\rm N}}e^{\pi {{\rm N}}(\kappa +bu)} \,
\langle{{\mathbf y}}|\,{{\mathsf A}}(u)\,,\\
{}& \langle{{\mathbf y}}|\,{{\mathsf D}}_{{{\rm N}}} \equiv
\lim_{u\to -\infty} (-)^{{\rm N}}e^{\pi {{\rm N}}(\kappa + bu)} \,
\langle{{\mathbf y}}|\,{{\mathsf D}}(u) \,, \end{aligned}$$ and taking into account the asymptotic behaviour of ${{\mathsf B}}(u)$ and of the coefficients (\[R23\]) at , we infer from the relations (\[AD\])–(\[BD\]) that $$\begin{aligned}
{{\mathsf A}}_{{{\mathfrak 0}}}\Psi({{\mathbf y}})\,=\,
a_{{{\mathfrak 0}}}(y_{{\mathfrak 0}})\,{{\mathsf T}}_{{\mathfrak 0}}^{+}\Psi({{\mathbf y}})\,,\qquad
{{\mathsf D}}_{{{\mathfrak 0}}}\Psi({{\mathbf y}})\,=\,
d_{{{\mathfrak 0}}}(y_{{\mathfrak 0}})\,{{\mathsf T}}_{{\mathfrak 0}}^{-}\Psi({{\mathbf y}})\,,\\
{{\mathsf A}}_{{{\rm N}}}\Psi({{\mathbf y}})\,=\,
a_{{{\rm N}}}(y_{{\mathfrak 0}})\,{{\mathsf T}}_{{\mathfrak 0}}^{-}\Psi({{\mathbf y}})\,,\qquad
{{\mathsf D}}_{{{\rm N}}}\Psi({{\mathbf y}})\,=\,
d_{{{\rm N}}}(y_{{\mathfrak 0}})\,{{\mathsf T}}_{{\mathfrak 0}}^{+}\Psi({{\mathbf y}})\,,\end{aligned}$$ where the shift operators ${{\mathsf T}}_{{\mathfrak 0}}^\pm$ are defined analogously to [(\[shiftdef\])]{} for the variable $y_{{\mathfrak 0}}$. It follows from (\[badinf\]) and (\[eveninf\]) that we have $a_{{{\mathfrak 0}}}(y) = d_{{{\rm N}}}(y)$, $d_{{{\mathfrak 0}}}(y) = a_{{{\rm N}}}(y)$, $a_{{{\mathfrak 0}}}(y) a_{{{\rm N}}}(y+ib) =
d_{{{\mathfrak 0}}}(y) d_{{{\rm N}}}(y-ib) =1$. Noting that $$\begin{aligned}
\Psi({{\mathbf y}}) &= \lim_{u\to \pm\infty} {\bigl[}
e^{ \pi b {{\rm N}}(2\kappa \pm(i b - 2u))} \,
{\rm det}_q {{\mathsf M}}(u){\bigr]}\,\Psi({{\mathbf y}})= \\
\nonumber
&= \begin{cases}
{{\mathsf A}}_{{{\mathfrak 0}}}{{\mathsf D}}_{{{\mathfrak 0}}} \Psi({{\mathbf y}})=a_{{{\mathfrak 0}}}(y_{{\mathfrak 0}}-ib)
d_{{{\mathfrak 0}}}(y_{{\mathfrak 0}})\Psi({{\mathbf y}}), & u \to +\infty; \\
{{\mathsf A}}_{{{\rm N}}}{{\mathsf D}}_{{{\rm N}}} \Psi({{\mathbf y}}) = a_{{{\rm N}}}(y_{{\mathfrak 0}}+ib)
d_{{{\rm N}}}(y_{{\mathfrak 0}})\Psi({{\mathbf y}}), & u \to -\infty \,,
\end{cases}
$$ allows us to choose $$\begin{aligned}
{}& \qquad\qquad a_{{{\mathfrak 0}}} = d_{{{\mathfrak 0}}} = a_{{{\rm N}}} =
d_{{{\rm N}}} = 1 \,.\end{aligned}$$ The resulting equation for the $y_{{\mathfrak 0}}$–dependence may therefore be written as $$\label{y0baxter}
\lim_{u\to \pm\infty} {\bigl[} (\pm)^{{\rm N}}e^{ \pi {{\rm N}}(\kappa \mp bu)} \, t(u) {\bigr]}
\,\Psi({{\mathbf y}}) = \bigl( {{\mathsf T}}_{{\mathfrak 0}}^{+}+{{\mathsf T}}_{{\mathfrak 0}}^{-} \bigr) \, \Psi({{\mathbf y}}) \,.$$ This relation supplements the Baxter equations [(\[baxtereqn\])]{} in the cases of the modular XXZ magnet and the sinh–Gordon model with even ${{\rm N}}$.
In the case of the modular XXZ magnet we furthermore find a small modification in the form of the coefficient functions $a(u)$ and $d(u)$ which appear in the Baxter equations. These follow from the following formula for the q–determinant (see Appendix \[qDet\]): $$\begin{aligned}
\det\nolimits_q {{\mathsf M}}^{{\rm{\scriptscriptstyle}XXZ}}(u) &\equiv
A(u) \, D(u-ib) - q^{-1} \, B(u) \, C(u-ib) \\
{} &= \bigl( 4 \, \cosh \pi b( s + u - i {{\textstyle \frac{b}{2} }}) \,
\cosh \pi b( s - u + i {{\textstyle \frac{b}{2} }}) \bigr)^{{\rm N}}\,, \end{aligned}$$ The resulting expressions for $a^{{{\rm{\scriptscriptstyle}XXZ}}}(u)$ and $d^{{{\rm{\scriptscriptstyle}XXZ}}}(u)$ will be $$a^{{{\rm{\scriptscriptstyle}XXZ}}}(u) \,=\,d^{{{\rm{\scriptscriptstyle}XXZ}}}(-u)\,=\,
-2i\cosh\pi b (u -s-i{{\textstyle \frac{b}{2} }})\,.$$
The existence of the “zero mode” $y_{{\mathfrak 0}}$ also leads to modifications in the formulae for ${{\mathsf A}}(u)$ and ${{\mathsf D}}(u)$. For the modular magnet and the sinh–Gordon model with even ${{\rm N}}$, the number of coefficients in the expansion (\[aaexp\]) and (\[ddexp\]) exceeds by two the number of the operators ${{\mathsf y}}_k$. However, in these cases we know the asymptotics of ${{\mathsf A}}(u)$ and ${{\mathsf D}}(u)$ and therefore we will need the following interpolation formula.
\[interpol\] Let $e^{{{\rm N}}\pi b u}P(u)$ be a polynomial in $e^{2\pi b u}$ such that $$P(u)\sim
\begin{cases} e^{+\pi b ({{\rm N}}u +p_{{\mathfrak 0}})}, &
\text{for $u\to +\infty$}\,,\\
(-)^{{\rm N}}e^{-\pi b ({{\rm N}}u + p_{{\mathfrak 0}})}, &
\text{for $u\to -\infty$} \,.
\end{cases}$$ For an arbitrary set of variables $y_1,\dots,y_{{{\rm N}}-1}$ such that $y_k\neq y_l$ for all $k\neq l$ we may then write $$P(u)=
\sinh\pi b(u+p_{{\mathfrak 0}}+\rho_{{\scriptscriptstyle}{{\rm N}}})\prod_{k=1}^{{{\rm N}}-1}
\sinh\pi b (u-y_k)+
\sum_{k=1}^{{{\rm N}}-1}\,\prod_{l\neq k}\,
\frac{\sinh\pi b(u-y_l)}{\sinh\pi b (y_k-y_l)}\,
P(y_k)\, ,$$ where $\rho_{{\scriptscriptstyle}{{\rm N}}} \equiv\sum_{k=1}^{{{\rm N}}-1}y_k$.
Thus, for the modular magnet and the sinh–Gordon model with even ${{\rm N}}$, we have the following formulae for the operators ${{\mathsf A}}(u)$ and ${{\mathsf D}}(u)$ (recall that $\kappa=0$ for the modular magnet and for the sinh–Gordon model with even ${{\rm N}}$) $$\begin{aligned}
{{\mathsf A}}(u)&= 2^{{\rm N}}e^{-\pi \kappa {{\rm N}}} \,
\sinh\pi b(u+{{\mathsf p}}_{{\scriptscriptstyle}{{\rm N}}} + \rho_{{\scriptscriptstyle}{{\rm N}}})
\prod_{k=1}^{{{\rm N}}-1} \sinh\pi b (u-{{\mathsf y}}_k)\label{Apsi}\\& \qquad +
\sum_{k=1}^{{{\rm N}}-1}\,\prod_{l\neq k}\,
\frac{\sinh\pi b(u-{{\mathsf y}}_l)}{\sinh\pi b ({{\mathsf y}}_k-{{\mathsf y}}_l)}\,
\bigl(a({{\mathsf y}}_k)\bigr)^{{\rm N}}\,{{\mathsf T}}_k^{-}\, ,
{\nonumber}\\
{{\mathsf D}}(u)&= 2^{{\rm N}}e^{-\pi \kappa {{\rm N}}} \,
\sinh\pi b(u+{{\mathsf p}}_{{\scriptscriptstyle}{{\rm N}}}+\rho_{{\scriptscriptstyle}{{\rm N}}})
\prod_{k=1}^{{{\rm N}}-1} \sinh\pi b (u-{{\mathsf y}}_k)\label{Dpsi}\\&\qquad+
\sum_{k=1}^{{{\rm N}}-1}\,\prod_{l\neq k}\,
\frac{\sinh\pi b(u-{{\mathsf y}}_l)}{\sinh\pi b ({{\mathsf y}}_k-{{\mathsf y}}_l)}\,
\bigl(d({{\mathsf y}}_k)\bigr)^{{\rm N}}\,{{\mathsf T}}_k^{+}\, ,
{\nonumber}$$ where now ${\sf \rho}_{{\scriptscriptstyle}{{\rm N}}}\equiv\sum_{k=1}^{{{\rm N}}-1} {{\mathsf y}}_k$; notice that $[{{\mathsf p}}_{{\scriptscriptstyle}{{\rm N}}},{\sf \rho}_{{\scriptscriptstyle}{{\rm N}}}]=0$ and $[{{\mathsf p}}_{{\scriptscriptstyle}{{\rm N}}},{{\mathsf y}}_{{\mathfrak 0}}]={{\textstyle \frac{i}{\pi b} }}$. Furthermore, comparision with (\[badinf\]) and (\[eveninf\]) shows that ${{\mathsf p}}_{{\scriptscriptstyle}{{\rm N}}}=-\sum_{k=1}^{{{\rm N}}} {{\mathsf p}}_k$ for the modular magnet and ${{\mathsf p}}_{{\rm N}}= \sum_{k=1}^{{{\rm N}}} (-)^k {{\mathsf p}}_k$ for the sinh–Gordon model.
Concluding remarks — outlook
============================
On the Baxter equations {#Baxtersub}
-----------------------
Summarizing the results of Sections \[QB\] and \[SOV\], we arrive at the main result of the present article. We will formulate it only for the case of the sinh–Gordon model with ${{\rm N}}$ odd for which our analysis is most complete, but from our previous discussions and remarks it seems clear that very similar results should hold in the other cases as well.\
[**Main result:**]{} [*A function $t(u)$ is an eigenvalue of the transfer–matrix ${{\mathsf T}}^{{{\rm{\scriptscriptstyle}SGo}}}(u)$ if and only if there exists a function $q_t(u)$ which satisfies the following conditions $$\label{summq}
\left[ \;\;
\begin{aligned}
{\rm (i)} \;\; & q_t(u)\;\,\text{is meromorphic in}\;\,{{\mathbb C}},
\,\;\text{with poles of maximal order ${{\rm N}}$ }
\text{in}\;\,\Upsilon_{-s}\cup
\bar{\Upsilon}_{s},\\
{\rm (ii)} \;\; & q_t(u)\;{\sim} \left\{
\begin{aligned}
& \exp\big(+i \pi {{\rm N}}\, \sigma\, u -
i{{\textstyle \frac{\pi }{2} }} {{\rm N}}u^2 \big)\;\;\;{\rm for}\;\;
|u|{\to}\infty,\;\,\; |{\rm arg}(u)|<{{\textstyle \frac{\pi}{2} }} \,, \\
& \exp\big(-i \pi {{\rm N}}\, \sigma\, u -
i{{\textstyle \frac{\pi }{2} }} {{\rm N}}u^2 \big)\;\;\;{\rm for}\;\;
|u|{\to}\infty,\;\,\; |{\rm arg}(u)|>{{\textstyle \frac{\pi}{2} }} \,.
\end{aligned}\right.\\
{\rm (iii)} \;\; & t(u) \, q_t(u)\,=\,
\bigl(\check{a}(u)\bigr)^{{{\rm N}}}q_t(u-ib)
+\bigl(\check{d}(u)\bigr)^{{{\rm N}}}q_t(u+ib) \,,\\
& \text{where}\;\; \check d(u) = \check a(-u) =
1 + ({{\textstyle \frac{m \Delta}{4} }})^2 \, e^{ -\pi b (2u+ib)} \,, \\
{\rm (iv)} \;\; & q_t(u)\;\,\text{satisfies the following
quantum Wronskian relation}\\
& \qquad q_t(u+i{\delta}_+) \, q_t(u-i{\delta}_+)
-q_t(u+i{\delta}_-) \, q_t(u-i{\delta}_-)
\,=\,W_{{\rm N}}(u)\,,\\
& \text{where}\;\;
W_{{\rm N}}(u)= \,
e^{-i\pi{{\rm N}}\left(u^2 +\sigma^2\right)}
\bigl(D_{\sigma}(u)\bigr)^{-{{\rm N}}}\,.
\end{aligned}
\;\;\right]$$ The corresponding eigenstate $\Psi_t$ in the SOV representation defined in Section \[SOV\] is represented as in [(\[factorrep\])]{}.*]{}\
We have therefore succeeded in reformulating the spectral problem for ${{\mathsf T}}(u)$ as the problem to determine the set $\mathfrak S$ of solutions to the Baxter equation which possess the properties (i)-(iv) above.
It should be observed that conditions (i)-(iii) already constrain the possible functions $q_t(u)$ rather strongly. Let us consider $$Q_t(u)\,=\,
\Big(
\Gamma_b\big({{\textstyle \frac{Q}{2} }}-i(u+s)\big)
\Gamma_b\big({{\textstyle \frac{Q}{2} }}-i(u-s)\big)
\Big)^{-{{\rm N}}}
q_t(u)\,,$$ where $\Gamma_b(x)\equiv \Gamma_2(x|b^{-1},b)$, with $\Gamma_2$ being the Barnes Double Gamma function defined in Appendix \[DGF\]. The function $Q_t(u)$ will then have the properties $$\label{summQ} \left[ \;\;
\begin{aligned}
{\rm (a)} \;\; & Q_t(u)\;\,\text{is entire analytic of order 2 in}\;\,{{\mathbb C}},\\
{\rm (b)} \;\; & t(u) \, Q_t(u)\,=\,
\bigl(A(u)\bigr)^{{{\rm N}}}Q_t(u-ib)
+\bigl(D(u)\bigr)^{{{\rm N}}}Q_t(u+ib) \,,
\end{aligned}
\;\;\right]$$ The explicit form of the coefficients $A(u)$ and $D(u)$ can easily be figured out with the help of the formula [(\[baxxztil\])]{} and the functional relations [(\[Ga2funrel\])]{}.
Property $\rm (a)$ combined with the Hadamard factorization theorem (see e.g. [@Ti]) imply that $Q_t(u)$ can be represented by a product representation of the form $$Q_t(u)\,=\,e^{r(u)}\prod_{k=1}^{\infty} {'}
\left(1-\frac{u}{u_k}\right)\,,$$ where the prime indicates the canonical Weierstrass regularization of the infinite product. The function $r(u)$ in the prefactor is a second order polynomial which can be worked out explicitly. The Baxter equation [(\[BaxterEV\])]{} then implies that the zeros $u_k$ must satisfy an infinite set of equations, $$-1\,=\,\frac{(A(u_k))^{{{\rm N}}}}{(D(u_k))^{{{\rm N}}}} \,
\frac{Q_t(u_k-ib)}{Q_t(u_k+ib)},
\qquad k\in{{\mathbb N}}\,,$$ which may be regarded as a generalization of the Bethe ansatz equations. However, as it stands it is not quite clear if these equations represent an efficient starting point for the investigation of the spectrum of our models.
The quantum Wronskian relation (iv) encodes remarkable additional information which can not easily be extracted from the conditions (i)-(iii) above. We plan to discuss its implications elsewhere.
Continuum limit {#contSG}
---------------
It is certainly interesting to discuss the consistency of our results with existing results and conjectures on the sinh–Gordon model in continuous space–time. Let us therefore now show that our findings are consistent with Lukyanov’s remarkable conjecture [@Lu] on the ground state wave–function for the sinh–Gordon model in the SOV representation.
Recall from the Subsection \[CL\] that we are interested in the limit $N\to\infty$, $\Delta{\to}0$, $s{\to}\infty$ such that $m={{\textstyle \frac{4}{{\Delta}} }} e^{-\pi b s}$ and $R={{\rm N}}{\Delta}/2\pi$ are kept finite in the limit. We are interested in the limiting behavior of the Baxter equation and of its solutions. Let us first note that the poles of ${q}_t(u)$ move out to infinity when $s{\to}\infty$. Also note that according to property rapid decay is found only within the strip ${{\mathcal S}}=\{u\in{{\mathbb C}};|{\rm Im}(u)|< Q/2\}$. By noting that $m{\Delta}={{\mathcal O}}(1/N)$ in the limit under consideration one sees that the coefficients $\check{a}(u)$, $\check{d}(u)$ in the Baxter equation become unity when ${{\rm N}}\to\infty$. Most importantly, let us finally observe that the right hand side of the Wronskian relation approaches a constant for ${{\rm N}}\to\infty$.
Our results therefore strongly suggest the following conjecture on the conditions which characterize the spectrum of the continuum sinh–Gordon model in the SOV representation: $$\label{summqlim}
\left[ \;\;
\begin{aligned}
{\rm (i)} \;\; & q_t(u)\;\,\text{is entire analytic},\\
{\rm (ii)} \;\; & q_t(u)\;\,\text{decays rapidly for}\;\,
|{\rm Re}(u)|{\to}\infty,\;\,u\in{{\mathcal S}},\\
{\rm (iii)} \;\; & q_t(u)\;\,\text{satisfies a difference equation
of the form}\\
& \qquad t(u) \, q_t(u)\,=\,
q_t(u-ib) +q_t(u+ib), \\
& \text{where}\;\,t(u)\;\,\text{is periodic under}\;\,u{\to}u+ib^{-1},\\
{\rm (iv)} \;\; & q_t(u)\;\,\text{satisfies the following
quantum Wronskian relation}\\
& \;\; q_t(u+i{\delta}_+) \, q_t(u-i{\delta}_+)-q_t(u+i{\delta}_-) \, q_t(u-i{\delta}_-)
\,=\,1\,.
\end{aligned}\;\;\right]$$
Our next aim will be to show that, by adding one supplementary condition, one gets a complete characterization of the function $q_0^{}(u)$ which was proposed in [@Lu] to describe the ground state for the continuum sinh–Gordon model in the SOV representation: $$\label{summqlim2} {\rm (v)} \qquad
q_0(u)\;\,\text{is nonvanishing within}\;\, {{\mathcal S}}\,.$$ We claim that the solution to conditions (i)–(v) is essentially unique and given by the formula $$\label{Yq}
\log q_0(u)=-\frac{mR}{2}\frac{\cosh\frac{\pi}{Q}u}{\sin\frac{\pi}{Q}b}+
\int\limits_{{{\mathbb R}}}\frac{dv}{2Q}\;
\frac{\log(1+Y(v))}{\cosh\frac{\pi}{Q}(u-v)},$$ which expresses $q_0(u)$ in terms of the solution $Y(u)$ to the nonlinear integral equation $$\label{TBA}
\log Y(u)=-mR\cosh{{\textstyle \frac{\pi}{Q} }}u+
\int\limits_{{{\mathbb R}}}\frac{dv}{2Q}\, S(u-v) \,{\log(1+Y(v))} ,$$ where the kernel $S(u-v)$ is explicitly given as follows: $$S(u)=\frac{2 \, \sin\frac{\pi}{Q}b \, \cosh\frac{\pi}{Q}u}
{\sinh\frac{\pi}{Q}(u+ib)\sinh\frac{\pi}{Q}(u-ib)}\,.$$ These equations form the basis for the calculation of the ground state energy [@Za] and other local conserved quantities of the continuum sinh–Gordon model [@Lu] within the thermodynamic Bethe ansatz framework.
For the reader’s convenience we will present the outline of an argument[^4] which establishes the equivalence between (i)-(v) and [(\[Yq\])]{}, [(\[TBA\])]{}. Let us define an auxilliary function $Y(u)$ by the formula $$\label{yqq}
1+Y(u)\,=\,q_0^{}(u+i{\delta}_+) \, q_0^{}(u-i{\delta}_+)\,.$$ Assuming that $q_0(u)$ satisfies the properties (i), (ii), and (v), one can take the logarithm of [(\[yqq\])]{} and then solve the resulting difference equation by Fourier transform, which leads to the representation [(\[Yq\])]{}. Re–inserting this representation into the Wronskian relation (iv) shows that $Y(u)$ must satisfy [(\[TBA\])]{}.
A proof of Lukyanov’s conjecture [@Lu] therefore amounts to showing that $q_0(u)$ must satisfy the properties (i)–(v). We find it very encouraging that our study of the lattice sinh–Gordon model gave us strong support for the necessity of properties (i)–(iv).
Connection with lattice Liouville model
---------------------------------------
Relations between the compact XXZ chain, lattice sine–Gordon model, and the (imaginary field) Liouville model was investigated in [@FT] from the view point of the QISM. Let us show that similar connections exist between the modular XXZ magent, lattice sinh–Gordon model, and the (real field) Liouville model. Following [@FT], we introduce the L–matrix $$\label{LLi}
L^{\zeta} (u) =
e^{-\frac{1}{2}\pi b \zeta \sigma_3} \, L^{{{\rm{\scriptscriptstyle}SG}}} (u+\zeta)
\, e^{\frac{1}{2} \pi b \zeta \sigma_3} \,,$$ where $\zeta$ is related to the representation parameter $s$ and the lattice spacing $\Delta$ as $e^{\pi b \zeta} = \Delta \, e^{\pi b s}$, and the operators ${{\mathsf p}}$ and ${{\mathsf x}}$ in (\[lSG\]) are related to the discretized field and its conjugate momentum as follows $$\label{pxz}
2 \pi b \, {{\mathsf p}}= 2\pi b \, \zeta -\beta \, \Phi \,, \qquad
4 \pi b \, {{\mathsf x}}= 2\pi b \, \zeta +
\beta \, ({{\textstyle \frac{1}{2} }} \Pi -\Phi) \,, \qquad
\beta=b \sqrt{8\pi} \,.$$ Comparison with [(\[px\])]{} and [(\[mds\])]{} shows that the new variables defined by [(\[pxz\])]{} are related to these of the sinh–Gordon model via a canonical transformation, $$\label{pppp}
\Pi = \Pi^{{{\rm{\scriptscriptstyle}SG}}} \,, \qquad
\beta \, \Phi = \beta \, \Phi^{{{\rm{\scriptscriptstyle}SG}}} + 2\pi b \zeta \,, \qquad
e^{-\pi b \zeta} ={{\textstyle \frac{m}{4} }} \,.$$ In the $\zeta\to+\infty$ limit L–matrix (\[LLi\]) turns into $$\label{LL}
L^{\rm{\scriptscriptstyle}L}(u) \equiv \lim_{\zeta\to+\infty} L^{\zeta} (u) =
\left( \begin{array}{cc}
e^{\frac{\beta}{8}\, \Pi_n} \,
\bigl( 1 +\Delta^2 \, e^{-\beta \Phi_n} \bigr) \,
e^{\frac{\beta}{8}\, \Pi_n} &
-i \Delta\, e^{\pi b u - \frac{\beta}{2} \Phi_n } \\
-2i \Delta\, \sinh \bigl(\pi b u + \frac{\beta}{2} \Phi_n \bigr) &
e^{- \frac{\beta}{4} \, \Pi_n}
\end{array} \right) \,.$$ It is natural to expect that this L–matrix describes some massless limit of the sinh–Gordon model. The corresponding obtained according to formula (\[LU\]), $$\label{UccL}
U^{\rm{\scriptscriptstyle}L}(u) = \left( \begin{array}{cc}
{{\textstyle \frac{\beta}{4} }} \, \Pi(x) &
-i \, e^{u - \frac{\beta}{2} \Phi(x) } \\
-2i \, \sinh \bigl(u + \frac{\beta}{2} \Phi(x) \bigr) &
- {{\textstyle \frac{\beta}{4} }} \, \Pi(x)
\end{array} \right)\,,$$ reproduces the Liouville equations of motion via the zero curvature equation (see [@FT] for details in the case of sine–Gordon model and imaginary Liouville field). This observation suggests that (\[LL\]) is a suitable L–matrix for describing the quantum lattice Liouville model in the QISM framework.
Although the limiting procedure in (\[LL\]) has not been mathematically rigorously developed yet (in particular, there is a subtle question of interchangibility of $\zeta\to +\infty$ limit with the classical limit), the results of the present article provide further support for the proposed connection between the sinh–Gordon and Liouville models. First, observe that the twist by $e^{\frac{1}{2}\pi b \zeta \sigma_3}$ and the shift of the spectral parameter in [(\[LLi\])]{} do not change the corresponding auxiliary R–matrix (\[Rlsg\]) and the fundamental R–matrix given in Proposition \[Rfsg\]. Therefore, the local lattice density of the classical Hamiltonian corresponding to [(\[LLi\])]{} can be obtained by substituting (\[pppp\]) into (\[Hcl2\]). Taking then the limit $\zeta\to +\infty$, we obtain the following lattice Hamiltonian density (up to an additive constant) $$\begin{aligned}
\label{Cclcont2}
H^{\rm {{\scriptscriptstyle}L}, cl}_{n,n+1} &\equiv
\lim_{\zeta\rightarrow\infty} H^{\rm {{\rm{\scriptscriptstyle}SG}}, cl}_{n,n+1}
= {{\textstyle \frac{1}{\gamma} }}\log \Bigl(
{{\textstyle \frac{1}{2} }}\cosh {{\textstyle \frac{\beta}{4} }} \, (\Pi_n + \Pi_{n+1})
+ {{\textstyle \frac{1}{2} }} \,
\cosh {{\textstyle \frac{\beta}{2} }} \, (\Phi_n - \Phi_{n+1}) \\
\nonumber
&+ {{\textstyle \frac{\Delta^2}{2} }} \, e^{-\frac{\beta}{2}(\Phi_n + \Phi_{n+1})} \,
\bigl( 1 + e^{\frac{\beta}{4}(\Pi_n + \Pi_{n+1})} \,
\cosh {{\textstyle \frac{\beta}{2} }} ( \Phi_n - \Phi_{n+1} ) \bigr) \Bigr)\,,\end{aligned}$$ which in the continuum limit (\[cont\]) yields the Liouville Hamiltonian: $$\sum_n {{\textstyle \frac{1}{\Delta} }}
H^{\rm \scriptscriptstyle L, cl}_{n,n+1}
\rightarrow {\rm const} + \int_0^{2\pi R} dx \,
\bigl( {{\textstyle \frac{1}{2} }} \, \Pi^2 + {{\textstyle \frac{1}{2} }} \,
(\partial_x\Phi)^2
+ {{\textstyle \frac{1}{\gamma} }} \, e^{-\beta \,\Phi} \bigr) \,.$$
The second observation that we can make to support the proposed relationship between the lattice sinh–Gordon and Liouville models is the following. The transfer–matrix corresponding to [(\[LLi\])]{} is given by ${{\mathsf T}}^\zeta(u)={{\mathsf T}}^{{\rm{\scriptscriptstyle}SG}}(u +\zeta)$. Therefore, ${{\mathsf Q}}_\pm^\zeta(u) \equiv \check{{\mathsf Q}}_\pm^{{\rm{\scriptscriptstyle}SG}}(u+\zeta)$ satisfy (cf. [(\[baa2\])]{}) the Baxter equation $$\label{baz}
\begin{aligned}
{}& {{\mathsf T}}^\zeta(u) \cdot {{\mathsf Q}}_\pm^\zeta(u) =
\bigl( a_\zeta(u) \bigr)^{{\rm N}}\, {{\mathsf Q}}_\pm^\zeta(u - ib) +
\bigl( d_\zeta(u) \bigr)^{{\rm N}}\,
{{\mathsf Q}}_\pm^\zeta(u + ib) \,,\\[2pt]
{}& \text{where}\quad a_\zeta(u) =
1 + \Delta^2 \, e^{ \pi b (2u-ib)} \,, \quad d_\zeta(u) =
1 + \Delta^2 \, e^{ -\pi b (2u+4\zeta+ib)} \,.
\end{aligned}$$ Hence for ${{\mathsf T}}^{\rm{\scriptscriptstyle}L}(u) \equiv \lim_{\zeta\to +\infty} {{\mathsf T}}^\zeta(u)$ and ${{\mathsf Q}}_\pm^{\rm{\scriptscriptstyle}L}(u) \equiv \lim_{\zeta\to +\infty} {{\mathsf Q}}_\pm^\zeta(u)$ (the limits are meaningful if ${{\mathsf T}}^\zeta(u)$ and ${{\mathsf Q}}_\pm^\zeta(u)$ are expressed in terms of $\Phi$ and $\Pi$), we obtain the Baxter equation $$\label{baL}
{{\mathsf T}}^{\rm{\scriptscriptstyle}L}(u) \cdot {{\mathsf Q}}_\pm^{\rm{\scriptscriptstyle}L}(u) =
\bigl( 1 + \Delta^2 \, e^{ \pi b (2u-ib)} \bigr)^{{\rm N}}\,
{{\mathsf Q}}_\pm^{\rm{\scriptscriptstyle}L}(u - ib) + {{\mathsf Q}}_\pm^{\rm{\scriptscriptstyle}L}(u + ib) \,,$$ which coincides[^5] with the Baxter equation derived for the lattice Liouville model by a different method in [@K1; @FKV].
We finish by noting that in the continuum limit, $N\to \infty$, $\Delta = {{\mathcal O}}(1/N)$, the coefficient $\bigl( 1 + \Delta^2 \, e^{ \pi b (2u-ib)} \bigr)^{{\rm N}}$ becomes unity. This suggests that the Baxter equation for the eigenvalue $q^{\rm{\scriptscriptstyle}L}_t(u)$ of the Q–operator for the continuum Liouville model is $$\label{baLc}
t^{\rm{\scriptscriptstyle}L}(u) \, q^{\rm{\scriptscriptstyle}L}_t(u)\,=\, q^{\rm{\scriptscriptstyle}L}_t(u-ib) +
q^{\rm{\scriptscriptstyle}L}_t(u+ib) \,,
\qquad t^{\rm{\scriptscriptstyle}L}(u+ib^{-1}) = t^{\rm{\scriptscriptstyle}L}(u) \,,$$ which coincides with that for the continuum sinh–Gordon model (\[summqlim\]–iii).
However, it seems to be crucial to observe that the asymptotic properties of the function $q^{\rm{\scriptscriptstyle}L}_t(u)$ will certainly differ from those found in the case of the sinh–Gordon model. Indeed, for any model, the asymptotic properties of $q_t(u)$ are related to these of $t(u)$. Comparing the structure of the L–matrices [(\[lSG\])]{} and [(\[LL\])]{}, we see that the transfer–matrix ${{\mathsf T}}^{\rm{\scriptscriptstyle}L}(u)$ corresponding to the latter L–matrix has asymmetric asymptotics for ${\rm Re}(u){\to}\pm\infty$. This seems to be related to the fact that the sinh–Gordon potential $\cosh \beta \Phi(x)$ is spatially symmetric while the Liouville potential $e^{-\beta \Phi(x)}$ is asymmetric. As a consequence, we expect that the set of solutions to the Baxter equation [(\[baLc\])]{} describing the spectrum of the continuum Liouville model will be quite different from the set of solutions to the same Baxter equation which characterizes the spectrum of the continuum sinh–Gordon model.
These observations seem to offer a key to the understanding of the relation between massive and massless theories from the point of view of their integrable structure.
Special functions
=================
[\[Qdil\]]{}
Double Gamma function {#DGF}
---------------------
All the special functions that we have to deal with can be obtained from the Barnes Double Gamma function $\Gamma_2(x|\omega_{{\mathfrak 1}},\omega_{{\mathfrak 2}})$ [@Ba], which may be defined by $$\label{dgamma}
\log \Gamma_2 (x|\omega_{{\mathfrak 1}},\omega_{{\mathfrak 2}}) =
\Biggl( \frac{\partial}{\partial t} \sum_{n_{{\mathfrak 1}},n_{{\mathfrak 2}}=0}^\infty
(x + n_{{\mathfrak 1}}\, \omega_{{\mathfrak 1}}+ n_{{\mathfrak 2}}\, \omega_{{\mathfrak 2}})^{-t} \Biggr)_{t=0} \,.$$ The infinite sum in [(\[dgamma\])]{} is defined by analytic continuation from its domain of convergence (${\rm Re}(t)>2$) to the point of interest ($t=0$). One may alternatively use the integral representation $$\log\Gamma_{2}(x|\omega_{{{\mathfrak 1}}},\omega_{{{\mathfrak 2}}})=\frac{C}{2}
B_{2,2}(x|\omega_{{{\mathfrak 1}}},\omega_{{{\mathfrak 2}}})
+\frac1{2\pi i}\int_{\mathcal{C}}\frac{e^{-xt}\log(-t)}{\left(
1-e^{-\omega_{{{\mathfrak 1}}}t}\right) \left( 1-e^{-\omega_{{{\mathfrak 2}}}t}\right) }\frac
{dt}t\label{cint}$$ where the contour $\mathcal{C}$ goes from $+\infty$ to $+\infty$ encircling $0$ counterclockwise, $C$ is the Euler’s constant and $$B_{2,2}(x|\omega_{{{\mathfrak 1}}},\omega_{{{\mathfrak 2}}})=\frac{(2x-\omega_{{{\mathfrak 1}}}-\omega_{{{\mathfrak 2}}})^{2}}{4\omega_{{{\mathfrak 1}}}\omega_{{{\mathfrak 2}}}}-\frac{\omega_{{{\mathfrak 1}}}^{2}+\omega_{{{\mathfrak 2}}}^{2}}
{12\omega_{{{\mathfrak 1}}}\omega_{{{\mathfrak 2}}}}\label{B2}$$ The integral is well defined if ${\rm Re}(\omega_{{{\mathfrak 1}}})>0,$ ${\rm Re}(\omega_{{{\mathfrak 2}}})>0$, and ${\rm Re}(x)>0$. It satisfies the basic functional relations $$\label{Ga2funrel}
\frac{\Gamma_2 (x+\omega_{{\mathfrak 1}}|\omega_{{\mathfrak 1}},\omega_{{\mathfrak 2}})}{\Gamma_2 (x|\omega_{{\mathfrak 1}},\omega_{{\mathfrak 2}})}
=\sqrt{2\pi}
\frac{\omega_{{\mathfrak 2}}^{\frac{1}{2}-\frac{x}{\omega_{{\mathfrak 2}}}}}{\Gamma(x/\omega_{{\mathfrak 2}})},
\qquad
\frac{\Gamma_2 (x+\omega_{{\mathfrak 2}}|\omega_{{\mathfrak 1}},\omega_{{\mathfrak 2}})}{\Gamma_2 (x|\omega_{{\mathfrak 1}},\omega_{{\mathfrak 2}})}
=\sqrt{2\pi}
\frac{\omega_{{\mathfrak 1}}^{\frac{1}{2}-\frac{x}{\omega_{{\mathfrak 1}}}}}{\Gamma(x/\omega_{{\mathfrak 1}})}\,.$$ $\big(\Gamma_2 (x|\omega_{{\mathfrak 1}},\omega_{{\mathfrak 2}}))^{-1}$ is an entire analytic function of order 2 w.r.t. its variable $x$ with simple zeros at $x=-m\omega_{{{\mathfrak 1}}}-n\omega_{{{\mathfrak 2}}}$, where $m$ and $n$ are non–negative integers.
Function $w_b(x)$ {#Qdil1}
-----------------
In what follows we will be dealing with $$\label{Gb}
w_b(x) \equiv
\frac{\Gamma_2({{\textstyle \frac{Q}{2} }} - ix|b^{-1},b)} {\Gamma_2({{\textstyle \frac{Q}{2} }} +i x|b^{-1},b)} \,.$$ In the strip $|{\rm Im}(x)| < {{\textstyle \frac{Q}{2} }}$, function $w_b(x)$ has the following integral representation $$\label{wint}
w_b(x)=
\exp \Biggl\{ \frac{i \pi}{2} x^2 {{+}}\frac{i \pi}{24}(b^2 {{+}}b^{-2}) -
\int\limits_{{{\mathbb R}}+i0} \frac{dt}{4\, t} \,
\frac{ e^{-2i t x}}{\sinh b t \, \sinh{\frac{t}{b}} } \Biggr\} \,,$$ where the integration contour goes around the pole $t=0$ in the upper half–plane. This function is closely related (cf. Eq. (\[gw\])) to the remarkable special function introduced under the name of [*quantum dilogarithm*]{} in [@FK2] and studied in the context of quantum groups and integrable models in [@F2; @Ru; @Wo; @PT2; @K1; @K2; @BT; @T2; @V2].
Analytic continuation of $w_b(x)$ to the entire complex plane is a meromorphic function with the following properties $$\begin{aligned}
\text{self--duality} \label{wdual} \quad&
w_b(x)=w_{b^{-1}}(x) \,, \\[0.5mm]
\text{functional equation} \label{wfunrel} \quad&
\frac{w_b(x + {{\textstyle \frac{i}{2} }}b^{\pm 1})}{w_b(x - {{\textstyle \frac{i}{2} }}b^{\pm 1})} =
2 \, \cosh (\pi b^{\pm 1} x) \,, \\[0.5mm]
\text{reflection property} \label{wrefl} \quad&
w_b(x) \; w_b(-x) = 1 \,, \\[0.5mm]
\label{wcc} \text{complex conjugation} \quad&
\overline{w_b(x)} = w_b(-\bar{x}) \,,\\[0.5mm]
\label{wan}
\text{zeros\,/\,poles} \quad&
(w_b(x))^{{\pm}1} = 0 \ \Leftrightarrow \pm x \in \big\{ i{{\textstyle \frac{Q}{2} }}{+}nb{+}mb^{-1};n,m\in{{\mathbb Z}}^{\geq 0}\big\}
\,, \\[0.5mm]
\label{wres}
\text{residue} \quad&
\operatorname*{Res}_{x=-i\frac{Q}{2}} w_b(x)=\frac{i}{2\pi}
\,, \\[0.5mm]
\text{asymptotics} \quad& w_b(x) \sim
\left\{
\begin{aligned}
& e^{- \frac{i \pi}{2}(x^2 + \frac{1}{12}(b^2+b^{-2}))}\;\;{\rm for}\;\,
|x|{\to}\infty,\;\, |{\rm arg}(x)|<{{\textstyle \frac{\pi}{2} }} \,, \\
& e^{+ \frac{i \pi}{2}(x^2 + \frac{1}{12}(b^2+b^{-2}))}\;\;{\rm for}\;\,
|x|{\to}\infty,\;\, |{\rm arg}(x)|>{{\textstyle \frac{\pi}{2} }} \,.
\end{aligned}\right.\end{aligned}$$ Notice that $|w_b(x)|=1$ if $x\in{{\mathbb R}}$. Therefore, $w_b({{\mathsf O}})$ is unitary if ${{\mathsf O}}$ is a self–adjoint operator.
The function $w_b(x)$ allows us to define a whole class of new special functions. In Appendix \[proofSH\] we will use in particular the following b–analogues of the hypergeometric functions defined by $$\label{Phidef}
\Phi^{}_r(U_1\dots U_r;V_1 \dots V_r;x)\;\equiv\;
\frac{1}{i}
\int\limits_{i{{\mathbb R}}-0}d\tau\;e^{\pi\tau x}\,\prod_{k=1}^{r}
\frac{S_b(U_k+\tau)}{S_b(V_k+\tau)},$$ where the special function $S_b(x)$ is defined by $$\label{Sbx}
S_b(x) = w_b(ix -{{\textstyle \frac{i}{2} }}Q)$$ and has the properties $$\begin{aligned}
\label{Sb1}
\text{self--duality} \quad&
S_b(x) = S_{b^{-1}}(x) \,, \\
\label{Sb2}
\text{functional equation} \quad&
S_b(x {{+}}b^{{\pm}1}) = 2 \, \sin (\pi b^{{\pm}1} x) \, S_b(x) \,,\\
\label{Sb3}
\text{reflection property} \quad&
S_b(x) \, S_b(Q-x) = 1 \,.\end{aligned}$$
Function $D_\alpha(x)$ {#Qdil2}
----------------------
Let us also introduce another useful function $$\label{Ds}
D_\alpha(x) = \frac{ w_b(x+\alpha)}{w_b(x-\alpha)} \,.$$ Combining (\[wint\]) with (\[Dpar\]), we derive the integral representation $$\label{Dint}
D_\alpha(x)=
\exp \Biggl\{ i
\int\limits_{{{\mathbb R}}+i0} \frac{dt}{2t} \,
\frac{ \cos(2 t x) \, \sin(2 \alpha t)} {\sinh b t \, \sinh{\frac{t}{b}} } \Biggr\} \,.$$ $D_{{\alpha}}(x)$ is a meromorphic function with zeros at ${\pm}x \in \Upsilon_{-\alpha}$ and poles at ${\pm}x \in \Upsilon_{\alpha}$, where the set $\Upsilon_\alpha$ is defined in (\[ups\]). The function $D_\alpha(x)$ is self–dual in $b$ (but we will omit this index) and has the following properties $$\begin{aligned}
\text{functional equation} \label{Dfunrel} \quad&
\frac{D_\alpha(x + {{\textstyle \frac{i}{2} }}b^{\pm 1})} {D_\alpha(x - {{\textstyle \frac{i}{2} }}b^{\pm 1})} =
\frac{\cosh \pi b^{\pm 1} (x+\alpha)}{\cosh \pi b^{\pm 1} (x-\alpha)}
\,, \\ \text{$x$--parity} \label{Dpar} \quad&
D_\alpha(x) = D_\alpha(-x) \,, \\[0.5mm]
\text{reflection property} \label{Drefl} \quad&
D_\alpha(x) \; D_{-\alpha}(x) = 1 \,, \\ \label{Dcc} \text{complex conjugation} \quad&
\overline{D_\alpha(x)} = D_{-\bar\alpha}(\bar{x}) \,,\\[0.5mm]
\label{Das1}
\text{$x$--asymptotics} \quad D_\alpha(x) \sim &
\left\{
\begin{aligned} e^{- 2 \pi i \alpha x} \;\;{\rm for}\;\,
|x|{\to}\infty,\;\, |{\rm arg}(x)|<{{\textstyle \frac{\pi}{2} }} \,, \\
e^{+ 2 \pi i \alpha x} \;\;{\rm for}\;\,
|x|{\to}\infty,\;\, |{\rm arg}(x)|>{{\textstyle \frac{\pi}{2} }} \,,
\end{aligned}\right. \\ \label{Das2}
\text{$\alpha$--asymptotics} \quad D_\alpha(x) \sim &
\left\{
\begin{aligned} & e^{- i \pi(x^2 + \alpha^2 + \frac{1}{12}(b^2+b^{-2}))}
\;\,{\rm if}\;
|{\alpha}|{\to}\infty,\;\, |{\rm arg}({\alpha})|<{{\textstyle \frac{\pi}{2} }} \,, \\
& e^{+ i \pi(x^2 + \alpha^2 + \frac{1}{12}(b^2+b^{-2}))}
\;\,{\rm if}\;
|{\alpha}|{\to}\infty,\;\, |{\rm arg}({\alpha})|>{{\textstyle \frac{\pi}{2} }} \,.
\end{aligned}\right.\end{aligned}$$ Also, the following identity is obvious from the definition (\[Ds\]) $$\label{Dsp2}
D_\alpha(x) \, D_\beta(y) =
D_{\frac{\alpha+\beta+x-y}{2}}\bigl({{\textstyle \frac{x+y+\alpha-\beta}{2} }}\bigr) \,
D_{\frac{\alpha+\beta-x+y}{2}}\bigl({{\textstyle \frac{x+y-\alpha+\beta}{2} }}\bigr) \,.$$ Notice that $|D_\alpha(x)|=1$ if $\alpha\in{{\mathbb R}}$ and $x\in {{\mathbb R}}$ or $x\in i{{\mathbb R}}$. Therefore, $D_\alpha({{\mathsf O}})$ is unitary if $\alpha\in{{\mathbb R}}$ and ${{\mathsf O}}$ is a self–adjoint or anti–self–adjoint operator.
Integral identities for $D_\alpha(x)$ {#Qdil3}
-------------------------------------
Here we will give some integral identities involving products of D–functions. These identities can be regarded as summation formulae for the b–hypergeometric functions $\Phi^{}_r$ introduced in (\[Phidef\]).
Let us denote $\alpha^\star\equiv -{{\textstyle \frac{i}{2} }}Q -\alpha$ and introduce the function $$\label{Adef}
A(\alpha_1,\alpha_2,\ldots) = w_b(\alpha_1 {-} \alpha^\star_1) \,
w_b(\alpha_2 {-} \alpha^\star_2) \, \ldots \ .$$ Notice that $(\alpha^\star)^\star =\alpha$ and hence $A(\alpha^\star_1,\alpha^\star_2,,\ldots) \,
A(\alpha_1,\alpha_2,,\ldots) = 1$.
Lemma 15 in [@PT2] and Eqs. (26)–(27) in [@FKV] can be rewritten as the following property of the function $D_\alpha(x)$ under the Fourier transform: $$\label{D1}
\int_{{\mathbb R}}dx\, e^{2\pi i x y} \, D_\alpha(x) =
A(\alpha) \, D_{\alpha^\star}(y) \,.$$ Taking into account that $\lim_{\alpha\to 0} D_{\alpha}(x) =1$, we obtain from (\[D1\]) as a special case $$\label{Ddel}
\lim_{\alpha\to -\frac{i}{2}Q}
\,A(\alpha^\star) \, D_{\alpha}(x) = \delta(x) \,.$$ Here $\delta(x)$ on the r.h.s. is the Dirac delta–function and this relation should be understood in the sense of distributions. Indeed $\lim_{\alpha\to -\frac{i}{2}Q} \,A(\alpha^\star)
= w_b(\frac{i}{2}Q)=0$ and the l.h.s. of (\[Ddel\]) vanishes almost everywhere. On the other hand, the only double pole of $D_{-\frac{i}{2}Q}(x)$ is at $x=0$.
Using (\[D1\]), it is easy to derive the following relation $$\label{D2}\begin{aligned}
\int_{{\mathbb R}}dx\, e^{2\pi i z x} & \, D_\alpha(x-u) \, D_\beta(x-v) = \\
&= A(\alpha,\beta) \, e^{\pi i z (u+v)} \,
\int_{{\mathbb R}}dy\, e^{2\pi i y(u-v)} \,
D_{\alpha^\star}(y+{{\textstyle \frac{z}{2} }}) \, D_{\beta^\star}(y-{{\textstyle \frac{z}{2} }}) \,.
\end{aligned}$$ Choosing $z={\alpha^\star}+{\beta^\star}$, we can use (\[Dsp2\]) in order to rewrite the product of $D$’s in the integrand on the r.h.s. as a single function, $D_{\alpha^\star + \beta^\star}(y+\frac{\alpha^\star - \beta^\star}{2})$, and then apply (\[D1\]). This yields $$\label{D21}\begin{aligned}
\int_{{\mathbb R}}dx\, & e^{2\pi i (\alpha^\star + \beta^\star) x} \,
D_\alpha(x-u) \, D_\beta(x-v) = \\
&= A(\alpha,\beta,\alpha^\star {{+}}\beta^\star) \,
e^{2\pi i (v \alpha^\star + u \beta^\star)} \,
D_{\alpha+\beta+\frac{i}{2}Q}(u-v) \,.
\end{aligned}$$ In the case $\alpha^\star = -\beta^\star$, Eq. (\[Ddel\]) can be used and we conclude that $$\label{D2del}
\int_{{\mathbb R}}dx\, D_\alpha(x-u) \, D_\beta(x-v) =
A(\alpha,\beta) \, \delta(u-v)$$ holds in the sense of distributions provided that $\alpha+\beta=-iQ$.
The identities $$\begin{aligned}
\label{D3}
&\int_{{{\mathbb R}}} dx \, D_\alpha(x {-}u) \, D_\beta(x {-} v) \,
D_\gamma(x {-} w) \\ \nonumber &
\hspace{4cm}= A(\alpha,\beta,\gamma) \,
D_{\alpha^\star}(w {-} v) \, D_{\beta^\star}(u {-} w) \,
D_{\gamma^\star}(v {-} u) \,, \\
\label{D4}
&\int_{{{\mathbb R}}} dx \, D_\alpha(x {-}u) \, D_\beta(x {-} v) \,
D_\gamma(x {-} w) \, D_\omega(x {-}z) \\
\nonumber
&= A(\alpha,\beta,\gamma,\omega) \,
\frac{ D_{\alpha+\beta+\frac{i}{2}Q}(u {-} v) } { D_{\alpha+\beta+\frac{i}{2}Q}(w {-} z) } \,
\int_{{{\mathbb R}}} dx\, D_{\alpha^\star}(x {-}v) \, D_{\beta^\star}(x {-} u) \,
D_{\gamma^\star}(x {-} z) \, D_{\omega^\star}(x {-} w)\end{aligned}$$ are valid provided that $\alpha+\beta+\gamma= -iQ$ in (\[D3\]), and $\alpha+\beta+\gamma +\omega= -iQ$ in (\[D4\]).
Relation (\[D3\]) follows straightforwardly from Eq. (19) in [@K2], where function $\varphi_b(x)$ is our $g_b(e^{2\pi b x})$ (cf. Eq. (\[gw\])). Also, in other notations, relation (\[D3\]) is Eq. (11) in [@V2]. Eq. (\[D3\]) provides two expressions for a function which we denote as $I(u,v,w;\alpha,\beta,\gamma)$. In order to prove (\[D4\]) we multiply two copies of (\[D3\]) and compute the following integral $$\label{D3D3}
\int_{{\mathbb R}}dt\, I(t,u,v;\nu,\alpha,\beta) \,
I(t,w,z;\mu,\gamma,\omega) \,,$$ where $\nu+\alpha+\beta=\mu+\gamma+\omega=-iQ$ and we impose an additional condition $\nu+\mu=-iQ$. Then the l.h.s. of (\[D4\]) is obtained if we substitute for the $I$’s the expressions on the l.h.s. of (\[D3\]) and use relation (\[D2del\]). The r.h.s. of (\[D4\]) is obtained directly from (\[D3D3\]) if we substitute for the $I$’s the expressions on the r.h.s. of (\[D3\]) and use also that $\mu^\star = - \nu^\star$.
Positivity versus self–duality of the representations ${{\mathcal P}}_s$
========================================================================
[\[posapp\]]{}
The representations ${{\mathcal P}}_s$ are distinguished by the property that the operators $\pi_s(u)$, $u\in\{E,F,K\}$ are positive self–adjoint. We are now going to show that this property is closely related to the remarkable self–duality of these representations under $b{\to}b^{-1}$ which has such profound consequences for the physics of our models.
To begin with, let us remark that there exists a linear basis ${{\mathcal B}}_q({{\mathfrak s}{\mathfrak l}}(2,{{\mathbb R}}))$ for ${{\cal U}_q({{\mathfrak s}{\mathfrak l}}(2,{{\mathbb R}}))}$ such that all elements $u$ of ${{\mathcal B}}_q({{\mathfrak s}{\mathfrak l}}(2,{{\mathbb R}}))$ are realized by positive operators $\pi_s(u)$. Such a basis is, e.g., given by the monomials $$\begin{aligned}
{}& q^{+\frac{mn}{2}}\,C^{l}\,E^m \,K^n\;\,
{\rm represented ~by}\;\,{{\mathsf C}}_s^{l}\,
({{\mathsf K}}_s)^\frac{n}{2} \, ({{\mathsf E}}_s)^m \, ({{\mathsf K}}_s)^{\frac{n}{2}},\\
{}& q^{-\frac{mn}{2}}\,C^{l}\,F^m \,K^n \;\,{\rm represented ~by}\;\,
{{\mathsf C}}_s^{l}\,({{\mathsf K}}_s)^{\frac{n}{2}}\,({{\mathsf F}}_s)^m\,({{\mathsf K}}_s)^{\frac{n}{2}},
\end{aligned}
\quad l,m,n\in{{\mathbb Z}},\;\;l,m\geq 0.$$
The elements of ${{\cal U}_q({{\mathfrak s}{\mathfrak l}}(2,{{\mathbb R}}))}$ are clearly realized by unbounded operators on $L^2({{\mathbb R}})$. It is therefore useful to consider suitable subspaces ${{\mathcal T}}_s\subset L^2({{\mathbb R}})$ of test–functions on which all operators $\pi_s(u)$, $u\in{{\cal U}_q({{\mathfrak s}{\mathfrak l}}(2,{{\mathbb R}}))}$ are well–defined. In order to describe a canonical choice for ${{\mathcal T}}_s$ let us represent the elements of ${{\mathcal T}}_s$ by functions $f(k)$ such that ${{\mathsf p}}$ acts as $({{\mathsf p}}f)(k)=k f(k)$.
Let ${{\mathcal T}}_s$ be the space of functions ${f} (k)$ which satisfy $e^{a|k|}f\in L^2({{\mathbb R}})$ for all $a>0$, and which have an analytic continuation to ${{\mathbb C}}\setminus (\Upsilon_s\cup\bar{\Upsilon}_{-s})$, where $$\label{ups}
\begin{aligned}
\Upsilon_s &= \Bigl\{ s + i \bigl( {{\textstyle \frac{Q}{2} }} + nb + mb^{-1} \bigl),
\quad n,m \in {\mathbb Z}^{\geq 0} \Bigl\} \,,\\
\bar{\Upsilon}_s &=
\Bigl\{ s - i \bigl( {{\textstyle \frac{Q}{2} }} + nb + mb^{-1} \bigl),
\quad n,m \in {\mathbb Z}^{\geq 0} \Bigl\} \,.
\end{aligned}$$
On the spaces ${{\mathcal T}}_s$ the action of ${{\cal U}_q({{\mathfrak s}{\mathfrak l}}(2,{{\mathbb R}}))}$ is given by $$\label{efjm}
\begin{aligned}
{{\mathsf E}}_s \, {f}(k) &=
\bigl[{{\textstyle \frac{Q}{2} }} + i s - i k \bigr]_b \,
{f}(k + ib) \,, \\
{{\mathsf F}}_s \, {f}(k) &=
\bigl[{{\textstyle \frac{Q}{2} }} + i s + i k \bigr]_b \,
{f}(k - ib) \,,
\end{aligned}
\quad
{{\mathsf K}}_s \, {f}(k) = e^{-\pi bk} \, {f}(k) \,,$$ where $[x]_b \equiv {{\textstyle \frac{\sin \pi b x}{\sin \pi b^2} }}$.
The distinguished role of the space ${{\mathcal T}}_s$ is explained by the following result, which shows that the space ${{\mathcal T}}_s$ is canonically associated to the representation $\pi_s$:
\[Pslem\] ${{\mathcal T}}_s$ is the largest space on which all $\pi_s(u)$, $u\in{{\cal U}_q({{\mathfrak s}{\mathfrak l}}(2,{{\mathbb R}}))}$ are well–defined, i.e., $$\label{intersect}
{{\mathcal T}}_s\,=\,\bigcap_{u\in{{\mathcal B}}_q({{\mathfrak s}{\mathfrak l}}(2,{{\mathbb R}}))} {{\mathcal D}}_u\,,$$ where ${{\mathcal D}}_u$ is the domain of the unbounded operator $\pi_s(u)$, $u\in{{\mathcal B}}_q({{\mathfrak s}{\mathfrak l}}(2,{{\mathbb R}}))$. The spaces ${{\mathcal T}}_s$, $s\in{{\mathbb R}}$ are Fréchet spaces with topology defined by the family of seminorms $$\label{seminorms}
\lVert \,f\,\rVert_{u}^{}\,\equiv\,\sup_{k\in{{\mathbb R}}}\big| \,
(\pi_s(u)f)(k)\,\big|,\quad u\in{{\mathcal B}}_q({{\mathfrak s}{\mathfrak l}}(2,{{\mathbb R}})).$$
It is easy to check that $\pi_s(u)f\in{{\mathcal T}}_s$ for all $f\in{{\mathcal T}}_s$. In order to show that the conditions in the definition of ${{\mathcal T}}_s$ are all necessary, let us first observe that $e^{a|k|}f\in L^2({{\mathbb R}})$ is clearly necessary for ${{\mathsf K}}_s^{n}f$ to be well–defined for all $n\in{{\mathbb Z}}$. In order to determine the conditions on $f$ for ${{\mathsf E}}_s^nf $ to be well–defined, let us consider the unitary operator ${{\mathsf U}}_s\equiv w_b({{\mathsf p}}-s)$, where the special function $w_b(x)$ and its properties are described in Appendix \[Qdil1\]. We then have $${{\mathsf U}}_s^{}\cdot{{\mathsf E}}_s^n\cdot{{\mathsf U}}_s^{-1} =e^{2\pi nb{{\mathsf x}}} .$$ The intersection of the domains of $e^{2\pi nb{{\mathsf x}}}$ for all $n\in{{\mathbb N}}$ consists of functions $g(k)$ that are analytic in the upper half plane ${\mathbb H}_+$, see, e.g., [@S Lemma 1]. The corresponding functions $f(k)=({{\mathsf U}}_s^{-1}g)(k)= (w_b(k-s))^{-1}g(k)$ may have poles in $\Upsilon_s$. Similar arguments applied to ${{\mathsf F}}_s^n$ allow us to complete the proof of the first statement in Lemma \[Pslem\].
In order to verify the second statement, we mainly have to show that the space ${{\mathcal T}}_s$ is complete w.r.t. the topology defined by the seminorms [(\[seminorms\])]{}. This follows from [(\[intersect\])]{} together with the observation that the self–adjoint operators $\pi_s(u)$, $u\in{{\mathcal B}}_q({{\mathfrak s}{\mathfrak l}}(2,{{\mathbb R}}))$ are [*closed*]{} on ${{\mathcal D}}_u$.
We regard Lemma \[Pslem\] as the key to the mathematical understanding of the duality $b{\to}b^{-1}$ of our representations ${{\mathcal P}}_s$. Indeed, let us introduce the operators $\tilde{{{\mathsf E}}}_s$, $\tilde{{{\mathsf F}}}_s$, $\tilde{{{\mathsf K}}}_s$, obtained by replacing $b\to b^{-1}$ in (\[EFK1\]). These operators generate a representation $\tilde{{{\mathcal P}}}_s$ of ${{\mathcal U}}_{\tilde{q}}({{\mathfrak s}{\mathfrak l}}(2,{{\mathbb R}}))$, $\tilde{q}=e^{i\pi b^{-2}}$ on the [*same*]{} space ${{\mathcal T}}_s$. The space ${{\mathcal T}}_s$ is associated to the representation $\tilde{{{\mathcal P}}}_s$ as canonically as it is associated to ${{\mathcal P}}_s$. Moreover, it is easy to see that the representation $\tilde{{{\mathcal P}}}_s$ commutes[^6] with ${{\mathcal P}}_s$ on ${{\mathcal T}}_s$. It is therefore natural to regard ${{\mathcal T}}_s$ as the natural space on which a representation of the [*modular double*]{} ${{\cal U}_q({{\mathfrak s}{\mathfrak l}}(2,{{\mathbb R}}))}{\otimes}{{\mathcal U}}_{\tilde{q}}({{\mathfrak s}{\mathfrak l}}(2,{{\mathbb R}}))$ [@F3] is realized.
Another way to make the self–duality of the representations ${{\mathcal P}}_s$ transparent uses the rescaled generators introduced in [(\[efk\])]{}. These generators and their counterparts $\tilde{{{\mathsf e}}}_s$, $\tilde{{{\mathsf f}}}_s$, $\tilde{{{\mathsf k}}}_s$, obtained by replacing $b$ with $b^{-1}$ are related as [@BT] $$\label{bdual}
({{\mathsf e}}_s)^{\frac{1}{b}} = (\tilde{{{\mathsf e}}}_s)^b \,, \quad
({{\mathsf f}}_s)^{\frac{1}{b}} = (\tilde{{{\mathsf f}}}_s)^b \,, \quad
({{\mathsf k}}_s)^{\frac{1}{b}} = (\tilde{{{\mathsf k}}}_s)^b \,.$$ These observations express quite clearly that the representations of the two halves of the modular double, ${{\mathcal U}}_{q}({{\mathfrak s}{\mathfrak l}}(2,{{\mathbb R}}))$ and ${{\mathcal U}}_{\tilde{q}}({{\mathfrak s}{\mathfrak l}}(2,{{\mathbb R}}))$, are related to each other like the two sides of the same coin.
Structure of the monodromy matrix
=================================
[\[monoapp\]]{}
This appendix is devoted to the derivation of some simple, but important structural properties of the monodromy matrix ${{\mathsf M}}(u)$, $$\label{Mono'}
{{\mathsf M}}(u) \,\equiv\, \left(\begin{matrix} {{\mathsf A}}_{{\rm N}}(u) & {{\mathsf B}}_{{\rm N}}(u) \\
{{\mathsf C}}_{{\rm N}}(u) & {{\mathsf D}}_{{\rm N}}(u) \end{matrix} \right)\,\equiv\,
L_{{\rm N}}(u) \cdot \ldots \cdot L_2(u) \cdot L_1(u) \,.$$
Expansions in the spectral parameter
------------------------------------
Introduce the following notations for $n \in {\mathbb N}$, $$\label{npm}
\lceil n \rceil \equiv
\begin{cases}
n-1 & \text{if $n$ -- odd;} \\
n & \text{if $n$ -- even;}
\end{cases} \,, \quad
\lfloor n \rfloor \equiv
\begin{cases}
n & \text{if $n$ -- odd;} \\
n-1 & \text{if $n$ -- even;}
\end{cases}
\,.$$
\[poslem\] The elements ${{\mathsf A}}_{{{\rm N}}}(u)$, ${{\mathsf B}}_{{{\rm N}}}(u)$, and ${{\mathsf D}}_{{{\rm N}}}(u)$ of ${{\mathsf M}}(u)$ have the following form: $$\begin{aligned}
\label{aexp}
{{\mathsf A}}^{{\rm{\scriptscriptstyle}XXZ}}_{{{\rm N}}}(u)&= e^{{{\rm N}}\pi b u}\sum_{m=0}^{{{\rm N}}} (-)^m
e^{-2m\pi b u} \, {{\mathsf A}}^{{\rm{\scriptscriptstyle}XXZ}}_{{{\rm N}},m}\,,\\
\label{bexp}
{{\mathsf B}}^{{\rm{\scriptscriptstyle}XXZ}}_{{{\rm N}}}(u)&= i \, e^{{{\rm N}}\pi b u} \sum_{m=0}^{{{\rm N}}-1} (-)^m
e^{-2m\pi b u} \, {{\mathsf B}}^{{\rm{\scriptscriptstyle}XXZ}}_{{{\rm N}},m}\,,\\
\label{dexp}
{{\mathsf D}}^{{\rm{\scriptscriptstyle}XXZ}}_{{{\rm N}}}(u)&= e^{{{\rm N}}\pi b u}\sum_{m=0}^{{{\rm N}}} (-)^m
e^{-2m\pi b u} \, {{\mathsf D}}^{{\rm{\scriptscriptstyle}XXZ}}_{{{\rm N}},m}\,, \end{aligned}$$ $$\begin{aligned}
\label{aaexp}
{{\mathsf A}}^{{\rm{\scriptscriptstyle}SG}}_{{{\rm N}}}(u)&= i^{\lceil {{\rm N}}\rceil} \,
e^{ \pi b (\lceil {{\rm N}}\rceil u- {{\rm N}}s)}
\sum_{m=0}^{\lceil {{\rm N}}\rceil} (-)^m
e^{-2m\pi b u} \, {{\mathsf A}}^{{\rm{\scriptscriptstyle}SG}}_{{{\rm N}},m}\,,\\
\label{bbexp}
{{\mathsf B}}^{{\rm{\scriptscriptstyle}SG}}_{{{\rm N}}}(u)&= - i^{\lfloor {{\rm N}}\rfloor} \,
e^{\pi b (\lfloor {{\rm N}}\rfloor u- {{\rm N}}s) }
\sum_{m=0}^{\lfloor {{\rm N}}\rfloor} (-)^m
e^{-2m\pi b u} \, {{\mathsf B}}^{{\rm{\scriptscriptstyle}SG}}_{{{\rm N}},m}\,,\\
\label{ddexp}
{{\mathsf D}}^{{\rm{\scriptscriptstyle}SG}}_{{{\rm N}}}(u)&= i^{\lceil {{\rm N}}\rceil} \,
e^{ \pi b (\lceil {{\rm N}}\rceil u- {{\rm N}}s)}
\sum_{m=0}^{\lceil {{\rm N}}\rceil} (-)^m
e^{-2m\pi b u} \, {{\mathsf D}}^{{\rm{\scriptscriptstyle}SG}}_{{{\rm N}},m}\,,\end{aligned}$$ where ${{\mathsf A}}_{{{\rm N}},m}$, ${{\mathsf B}}_{{{\rm N}},m}$, and ${{\mathsf D}}_{{{\rm N}},m}$ are positive self–adjoint operators.
Let us consider the case of the XXZ chain, the other case being very similar. The definition of the monodromy matrix ${{\mathsf M}}_{{\rm N}}(u)$ yields the following recursion relations $$\begin{aligned}
\label{Arecrel}
{}&{{\mathsf A}}^{{\rm{\scriptscriptstyle}XXZ}}_{{{\rm N}}}(u)=(e^{\pi b u}{{\mathsf k}}_{{{\rm N}}}- e^{-\pi b u}{{\mathsf k}}_{{{\rm N}}}^{-1})
{{\mathsf A}}^{{\rm{\scriptscriptstyle}XXZ}}_{{{\rm N}}-1}(u)+ie^{\pi bu}{{\mathsf f}}_{{{\rm N}}} {{\mathsf C}}^{{\rm{\scriptscriptstyle}XXZ}}_{{{\rm N}}-1}(u),\\
{}&{{\mathsf B}}^{{\rm{\scriptscriptstyle}XXZ}}_{{{\rm N}}}(u)=(e^{\pi b u}{{\mathsf k}}_{{{\rm N}}}-e^{-\pi b u}{{\mathsf k}}_{{{\rm N}}}^{-1})
{{\mathsf B}}^{{\rm{\scriptscriptstyle}XXZ}}_{{{\rm N}}-1}(u)+ie^{\pi bu}{{\mathsf f}}_{{{\rm N}}} {{\mathsf D}}^{{\rm{\scriptscriptstyle}XXZ}}_{{{\rm N}}-1}(u),\\
{}&{{\mathsf C}}^{{\rm{\scriptscriptstyle}XXZ}}_{{{\rm N}}}(u)=(e^{\pi b u}{{\mathsf k}}_{{{\rm N}}}^{-1}-e^{-\pi b u}{{\mathsf k}}_{{{\rm N}}})
{{\mathsf C}}^{{\rm{\scriptscriptstyle}XXZ}}_{{{\rm N}}-1}(u) +ie^{-\pi bu}{{\mathsf e}}_{{{\rm N}}} {{\mathsf A}}^{{\rm{\scriptscriptstyle}XXZ}}_{{{\rm N}}-1}(u),\\
{}&{{\mathsf D}}^{{\rm{\scriptscriptstyle}XXZ}}_{{{\rm N}}}(u)=(e^{\pi b u}{{\mathsf k}}_{{{\rm N}}}^{-1}-e^{-\pi b u}{{\mathsf k}}_{{{\rm N}}})
{{\mathsf D}}^{{\rm{\scriptscriptstyle}XXZ}}_{{{\rm N}}-1}(u) +ie^{-\pi bu}{{\mathsf e}}_{{{\rm N}}}{{\mathsf B}}^{{\rm{\scriptscriptstyle}XXZ}}_{{{\rm N}}-1}(u),
\label{Drecrel}\end{aligned}$$ where ${{\mathsf k}}_{{{\rm N}}}\equiv {{\mathsf k}}_s{\otimes}1{\otimes}\dots 1$ etc. Using these recursion relations one may inductively show that the operators ${{\mathsf A}}_{{{\rm N}},m}$, ${{\mathsf B}}_{{{\rm N}},m}$ and ${{\mathsf D}}_{{{\rm N}},m}$ are linear combinations of monomials of the form $${{\mathsf u}}_s^{({{\rm N}})}{\otimes}\dots{\otimes}{{\mathsf u}}_s^{(1)},\qquad {{\mathsf u}}_s^{k}\in
\big\{{{\mathsf e}}_s^{},{{\mathsf f}}_s^{},{{\mathsf k}}_s^{},{{\mathsf k}}_s^{-1}\big\}$$ with [*positive*]{} integer coefficients. It remains to note that an operator which is the sum of positive self–adjoint operators will be self–adjoint on the intersection of the domains of the individual summands. These obserations reduce our claim to the self–adjointness and positivity of ${{\mathsf e}}_s$, ${{\mathsf f}}_s$, ${{\mathsf k}}_s$.
\[leadterm\] The leading terms of ${{\mathsf A}}_{{{\rm N}}}(u)$, ${{\mathsf B}}_{{{\rm N}}}(u)$, and ${{\mathsf D}}_{{{\rm N}}}(u)$ at $e^{\pi b u} \to \pm \infty$ are given by $$\begin{aligned}
\label{badinf}
{}& {{\mathsf B}}^{{\rm{\scriptscriptstyle}XXZ}}_{{{\rm N}},{{\mathfrak 0}}} = \Delta^{({{\rm N}}-1)} \, {{\mathsf f}}\,,
{}&& {{\mathsf A}}^{{\rm{\scriptscriptstyle}XXZ}}_{{{\rm N}},{{\mathfrak 0}}} = {{\mathsf D}}^{{\rm{\scriptscriptstyle}XXZ}}_{{{\rm N}},{{\rm N}}}=
\Delta^{({{\rm N}}-1)} \, {{\mathsf k}}\,, \\
{}& {{\mathsf B}}^{{\rm{\scriptscriptstyle}XXZ}}_{{{\rm N}},{{\rm N}}-1} = \bar\Delta^{({{\rm N}}-1)} \, {{\mathsf f}}\,, \qquad
{}&& {{\mathsf A}}^{{\rm{\scriptscriptstyle}XXZ}}_{{{\rm N}},{{\rm N}}} = {{\mathsf D}}^{{\rm{\scriptscriptstyle}XXZ}}_{{{\rm N}},{{\mathfrak 0}}} =
\Delta^{({{\rm N}}-1)} \, {{\mathsf k}}^{-1} \,,
\end{aligned}$$ where $\Delta^{(n)}$ is the $n$–fold co–product defined via $\Delta^{(n+1)}=(\Delta^{(n)} \otimes id) \circ \Delta$, with $\Delta^{(0)} \equiv id$ and $\Delta^{(1)} \equiv \Delta$, and $\bar\Delta^{(n)}$ is defined analogously for the opposite co–product $\bar\Delta^{(1)} \equiv \Delta'$. $$\begin{aligned}
\label{oddinf}
\text{${{\rm N}}$ -- odd:\qquad } &
\begin{cases}
{{\mathsf B}}^{{\rm{\scriptscriptstyle}SG}}_{{{\rm N}},{{\mathfrak 0}}} =
\theta_{\rm odd} \, \bigl( \Delta^{({{\rm N}}-1)} \,
{{\mathsf k}}^{-1} \bigr) \,, &
{{\mathsf B}}^{{\rm{\scriptscriptstyle}SG}}_{{{\rm N}},{{\rm N}}} =
\theta_{\rm odd} \, \bigl( \Delta^{({{\rm N}}-1)} \,
{{\mathsf k}}\bigr) \,, \\
{{\mathsf A}}^{{\rm{\scriptscriptstyle}SG}}_{{{\rm N}},{{\mathfrak 0}}} =
\theta_{\rm odd} \, \bigl( \Delta^{({{\rm N}}-1)} \,
{{\mathsf f}}\bigr) \,, &
{{\mathsf A}}^{{\rm{\scriptscriptstyle}SG}}_{{{\rm N}},{{\rm N}}-1} =
\theta_{\rm odd} \, \bigl( \bar\Delta^{({{\rm N}}-1)} \,
{{\mathsf f}}\bigr) \,, \\
{{\mathsf D}}^{{\rm{\scriptscriptstyle}SG}}_{{{\rm N}},{{\mathfrak 0}}} =
\theta_{\rm odd} \, \bigl( \bar\Delta^{({{\rm N}}-1)} \,
{{\mathsf e}}\bigr) \,, &
{{\mathsf D}}^{{\rm{\scriptscriptstyle}SG}}_{{{\rm N}},{{\rm N}}-1} =
\theta_{\rm odd} \, \bigl( \Delta^{({{\rm N}}-1)} \,
{{\mathsf e}}\bigr) \,,
\end{cases}
\\
\label{eveninf}
\text{${{\rm N}}$ -- even:\qquad } &
\begin{cases}
{{\mathsf B}}^{{\rm{\scriptscriptstyle}SG}}_{{{\rm N}},{{\mathfrak 0}}} =
\theta_{\rm even} \, \bigl( \Delta^{({{\rm N}}-1)} \,
{{\mathsf f}}\bigr) \,, &
{{\mathsf B}}^{{\rm{\scriptscriptstyle}SG}}_{{{\rm N}},{{\rm N}}-1} =
\theta_{\rm even} \, \bigl( \bar\Delta^{({{\rm N}}-1)} \,
{{\mathsf f}}\bigr) \,, \\
{{\mathsf A}}^{{\rm{\scriptscriptstyle}SG}}_{{{\rm N}},{{\mathfrak 0}}} =
\theta_{\rm even} \, \bigl( \Delta^{({{\rm N}}-1)} \,
{{\mathsf k}}\bigr) \,, &
{{\mathsf A}}^{{\rm{\scriptscriptstyle}SG}}_{{{\rm N}},{{\rm N}}} =
\theta_{\rm even} \, \bigl( \Delta^{({{\rm N}}-1)} \,
{{\mathsf k}}^{-1} \bigr) \,, \\
{{\mathsf D}}^{{\rm{\scriptscriptstyle}SG}}_{{{\rm N}},{{\mathfrak 0}}} =
\theta_{\rm even} \, \bigl( \Delta^{({{\rm N}}-1)} \,
{{\mathsf k}}^{-1} \bigr) \,, &
{{\mathsf D}}^{{\rm{\scriptscriptstyle}SG}}_{{{\rm N}},{{\rm N}}} =
\theta_{\rm even} \, \bigl( \Delta^{({{\rm N}}-1)} \,
{{\mathsf k}}\bigr) \,,
\end{cases}\end{aligned}$$ where $\theta_{\rm odd} \equiv
\theta_{{{\rm N}}-1} \circ \ldots \theta_{3} \circ \theta_{1}$ and $\theta_{\rm even} \equiv
\theta_{{{\rm N}}} \circ \ldots \theta_{4} \circ \theta_{2}$ are compositions of the automorphism (\[thetaEFK\]) at odd/even sites.
Eqs. [(\[badinf\])]{} follow easily from the decomposition [(\[Lpm\])]{} of the corresponding . Eqs. [(\[eveninf\])]{} are obtained by analogous consideration if $L^{{{\rm{\scriptscriptstyle}SG}}}(u)$ is replaced with $\sigma_1\,{L}'(u)$ (see [(\[Lsg2\])]{}) and formula [(\[Lt\])]{} is used. In order to apply this approach in the ${{\rm N}}$ odd case, one has to multiply the monodromy matrix with an extra $\sigma_1$ from the right (which leads to the interchange ${{\mathsf A}}\leftrightarrow {{\mathsf B}}$ and ${{\mathsf C}}\leftrightarrow {{\mathsf D}}$).
Quantum determinant {#qDet}
-------------------
Let us discuss connection between the so called [*quantum determinant*]{} and coefficients $a(u)$ and $d(u)$ which arise in the Baxter equation (\[baxxz\]). Since for the modular magnet we use R–matrix (\[R12\]) which is not symmetric, the corresponding quantum determinant will differ from the “standard” formula applicable, e.g., for the sinh–Gordon model. Therefore we commence by deriving the required expression.
\[GQD\] Let $L(u)$ be an L–matrix, satisfying the relation (\[rLL\]) with the auxiliary R–matrix of the form $$\label{Rxi}
R(u;\xi) =
\left( \begin{array}{cccc}
\sinh \pi b(u + i b) & & & \\ [-1mm]
& \sinh \pi b u &
i\sin\pi b^2\, e^{ \pi b \xi u} & \\
& i\sin\pi b^2 \, e^{- \pi b \xi u}
& \sinh \pi b u & \\ [-1mm]
& & & \sinh\pi b(u + i b)
\end{array} \right) \,.$$ and let ${{\mathsf M}}(u)$ be the corresponding monodromy matrix defined by (\[Mono’\]). The following element (quantum determinant) $$\begin{aligned}
\label{qdM0}
\det\nolimits_q {{\mathsf M}}(u) &=
{{\mathsf A}}(u) \, {{\mathsf D}}(u-ib) - q^{-\xi} \, {{\mathsf B}}(u) \, {{\mathsf C}}(u-ib)
$$ is central, i.e., $[{{\mathsf M}}(v),\det\nolimits_q {{\mathsf M}}(u)]=0$, and can be written as follows $$\label{qdML}
\det\nolimits_q {{\mathsf M}}(u) = \bigl( \det\nolimits_q L(u) \bigr)^{{\rm N}}\,,$$ where the quantum determinant of $L(u)$ is defined by the same formula (\[qdM0\]) (with ${{\mathsf M}}(u)$ replaced by $L(u)$).
Existence of the quantum determinant is due to the degeneration of the auxiliary R–matrix $\check{R}(u;\xi)\equiv {{\mathsf P}}\, R(u;\xi)$ at $u=-ib$, $$\label{Pxi}
\check{R}(-ib;\xi) = i\sin\pi b^2 \,
\left( \begin{array}{cccc}
0 & & & \\ [-1mm]
& q^{\xi} & -1 & \\
& -1 & q^{-\xi} & \\ [-1mm]
& & & 0
\end{array} \right) \,.$$ It is interesting to notice that this matrix is proportional (in the standard basis) to the one–dimensional projector $P^-_\xi$ onto the spin $0$ representation in the tensor square of spin ${{\textstyle \frac{1}{2} }}$ representations of ${\cal U}_{q^\xi}(\mathfrak{su}(2))$.
For $R(u;0)$ the statement of the Lemma is well–known (see, e.g., [@KBI]). In the generic case $\xi\neq 0$, one can observe that the gauge transformation $$\label{LMg}
\tilde{L}(u) = g_u^{-1} \, L(u) \, g_u \,, \qquad
\tilde{{{\mathsf M}}}(u) = g_u^{-1} \,{{\mathsf M}}(u) \, g_u \,, \qquad
g_u = e^{\frac{1}{2} \pi b \xi u \sigma_3}$$ yields L–matrix and monodromy matrix which satisfy the relation (\[rLL\]) with the auxiliary R–matrix $R(u;0)$. Therefore $$\label{qdMt}
\det\nolimits_q \tilde{{{\mathsf M}}}(u) =
\tilde{{{\mathsf A}}}(u) \, \tilde{{{\mathsf D}}}(u-ib) -
\tilde{{{\mathsf B}}}(u) \, \tilde{{{\mathsf C}}}(u-ib)$$ commutes with entries of $\tilde{{{\mathsf M}}}(u)$ and hence with entries of ${{\mathsf M}}(u)$. Using (\[LMg\]) in order to rewrite (\[qdMt\]) in terms of entries of ${{\mathsf M}}(u)$, we obtain (\[qdM0\]).
In order to prove (\[qdML\]) it suffices to observe that $$\label{detMP}
\det\nolimits_q {{\mathsf M}}(u) \cdot \check{R}_{12}(-ib;\xi) =
{{\mathsf M}}_1(u) \, {{\mathsf M}}_2 (u-ib) \, \check{R}_{12}(-ib;\xi) \,,$$ which yields also three different expressions equivalent to (\[qdM0\]) if we take into account the relation $\check{R}_{12}(-ib;\xi) \, {{\mathsf M}}_1(u-ib) \, {{\mathsf M}}_2 (u) =
{{\mathsf M}}_1(u) \, {{\mathsf M}}_2 (u-ib) \, \check{R}_{12}(-ib;\xi)$ which is a particular case of (\[rLL\]). Now if ${{\mathsf M}}'(u)$, ${{\mathsf M}}''(u)$ satisfy (\[rLL\]) with the same R–matrix and their entries commute, then we have $$\begin{aligned}
\nonumber
{}& \det\nolimits_q \bigl({{\mathsf M}}'(u) \, {{\mathsf M}}''(u) \bigr)
\cdot \check{R}_{12}(-ib;\xi) =
{{\mathsf M}}'_1(u) \, {{\mathsf M}}''_1 (u) \,
{{\mathsf M}}'_2(u-ib) \, {{\mathsf M}}''_2 (u-ib) \, \check{R}_{12}(-ib;\xi) \\
\label{detMM}
{}& = {{\mathsf M}}'_1(u) \, {{\mathsf M}}'_2 (u-ib) \,
{{\mathsf M}}''_1(u) \, {{\mathsf M}}''_2 (u-ib) \, \check{R}_{12}(-ib;\xi) =
\det\nolimits_q {{\mathsf M}}'(u) \cdot \det\nolimits_q {{\mathsf M}}''(u) \,.\end{aligned}$$ Whence (\[qdML\]) follows immediately.
For the models that we consider, Lemma \[GQD\] yields $$\begin{aligned}
\label{qdxxz1}
\det\nolimits_q {{\mathsf M}}^{{\rm{\scriptscriptstyle}XXZ}}(u) &=
{{\mathsf A}}(u) \, {{\mathsf D}}(u-ib) - q^{-1} \, {{\mathsf B}}(u) \, {{\mathsf C}}(u-ib) \\
\label{qdxxz2}
{} &= \bigl( 4 \, \cosh \pi b( s + u - i {{\textstyle \frac{b}{2} }}) \,
\cosh \pi b( s - u + i {{\textstyle \frac{b}{2} }}) \bigr)^{{\rm N}}\,, \\
\label{qdsg1}
\det\nolimits_q {{\mathsf M}}^{{\rm{\scriptscriptstyle}SG}}(u) &=
{{\mathsf A}}(u) \, {{\mathsf D}}(u-ib) - {{\mathsf B}}(u) \, {{\mathsf C}}(u-ib) \\
\label{qdsg2}
{} &= \bigl( 4 e^{-2\pi b s}\, \cosh \pi b( s + u - i {{\textstyle \frac{b}{2} }}) \,
\cosh \pi b( s - u + i {{\textstyle \frac{b}{2} }}) \bigr)^{{\rm N}}\,.\end{aligned}$$
Proving the Proposition \[Yxxz\] in Appendix E, we will deal with monodromy matrices $\tilde{{{\mathsf M}}}(u)$ obtained from ${{\mathsf M}}(u)$ by a gauge transformation with a matrix $G(u)$. Notice that such $\tilde{{{\mathsf M}}}(u)$ satisfies the exchange relation (\[rLL\]) with an R–matrix $$\label{Ruv}
\tilde{R}_{12}(u,v;\xi) = G_1(u) \, G_2(v) \,R(u-v;\xi) \,
G_1^{-1}(u) \, G_2^{-1}(v)$$ which [*is not*]{} of the form (\[Rxi\]). Therefore, for general $G(u)$, Lemma \[GQD\] does not apply to $\tilde{{{\mathsf M}}}(u)$. Nevertheless, there exists a class of gauge transformations which preserve the quantum determinant in the following sense.
\[QD2\] Let ${{\mathsf M}}(u)$ satisfy (\[rLL\]) with the auxiliary R–matrix $R(u;\xi)$ of the form (\[Rxi\]). Let $\tilde{{{\mathsf M}}}(u)=
\Bigl(\begin{smallmatrix} \tilde{A}(u) & \tilde{B}(u)\\
\tilde{C}(u) & \tilde{D}(u) \end{smallmatrix} \Bigr)$ be its gauge transform defined by $$\label{qdM3}
\tilde{{{\mathsf M}}}(u) =
G(u) \cdot {{\mathsf M}}(u) \cdot G^{-1}(u) \,, \qquad
G(u) = \bigl(\begin{smallmatrix} 1 & 0\\ \rho & 1
\end{smallmatrix} \bigr) \,, \qquad
\rho = e^{-\pi b \xi u} \, \rho_0 \,,$$ where $\rho_0$ is a c–number, which does not depend on $u$. Then $$\label{qdMMt}
\det\nolimits_q \tilde{{{\mathsf M}}}(u) \equiv
\tilde{{{\mathsf A}}}(u) \, \tilde{{{\mathsf D}}}(u-ib) -
q^{-\xi} \, \tilde{{{\mathsf B}}}(u) \, \tilde{{{\mathsf C}}}(u-ib) =
\det\nolimits_q {{\mathsf M}}(u) \,,$$ where the r.h.s. is defined according to Lemma \[GQD\].
Since entries of $\tilde{{{\mathsf M}}}(u)$ are linear combinations of entries of ${{\mathsf M}}(u)$, it follows that $\det\nolimits_q \tilde{{{\mathsf M}}}(u)$ is central, i.e., $[\tilde{{{\mathsf M}}}(v),\det\nolimits_q \tilde{{{\mathsf M}}}(u)]=0$. Thus, the l.h.s. defines the quantum determinant corresponding to the R–matrix (\[Ruv\]).
Although $R(u-v;\xi)$ and $\tilde{R}(u,v;\xi)$ are in general not equal, they coincide at $u-v=-ib$. Indeed, using the explicit form of $G(u)$, it is easy to check that $$\label{cRGG}
G_1(u) \, G_2(u-ib) \, \check{R}(-ib;\xi) =
\check{R}(-ib;\xi) \,.$$ Observing also that (\[detMP\]) holds for $\det\nolimits_q \tilde{{{\mathsf M}}}(u)$ as well, we derive $$\begin{aligned}
{}& \det\nolimits_q \tilde{{{\mathsf M}}}(u) \cdot \check{R}_{12}(-ib;\xi) =
\tilde{{{\mathsf M}}}_1(u) \, \tilde{{{\mathsf M}}}_2 (u-ib) \, \check{R}_{12}(-ib;\xi) \\
{}& = G_1(u) \, {{\mathsf M}}_1(u) \, G_1^{-1}(u) \,
G_2(u-ib) \, {{\mathsf M}}_2(u-ib) \, G_2^{-1}(u-ib) \,
\check{R}_{12}(-ib;\xi) \\
{}& = G_1(u) \, G_2(u-ib) \, {{\mathsf M}}_1(u) \,
{{\mathsf M}}_2(u-ib) \, G_2^{-1}(u-ib) \, G_1^{-1}(u) \,
\check{R}_{12}(-ib;\xi) \\
{}& = G_1(u) \, G_2(u-ib) \, {{\mathsf M}}_1(u) \,
{{\mathsf M}}_2(u-ib) \, \check{R}_{12}(-ib;\xi) \\
{}&= G_1(u) \, G_2(u-ib) \, \check{R}_{12}(-ib;\xi) \cdot
\det\nolimits_q {{\mathsf M}}(u) = \check{R}_{12}(-ib;\xi) \cdot
\det\nolimits_q {{\mathsf M}}(u) \,,\end{aligned}$$ which proves the assertion of the Lemma.
It is important that the gauge transformations used in the proof of the Proposition \[Yxxz\] belong to the class of gauge transformations described in Lemma \[QD2\]; they correspond to $\xi=1$ and $\xi=0$, respectively. This fact allows us to relate the quantum determinants of the models in question and the coefficients of the corresponding Baxter equations. A quick inspection of the proof of Proposition \[Yxxz\] shows that $$\label{adqdet}
\bigl( a(u)\, d(u-ib) \bigr)^{{\rm N}}= \det\nolimits_q {{\mathsf M}}(u) \,,$$ where the r.h.s. is given by (\[qdxxz2\]) and (\[qdsg2\]), respectively.
Construction of the fundamental R–operator ${{\mathsf R}}(u)$
=============================================================
[\[proofSH\]]{}
For the proof of Theorem \[Rthm\] we will need the following material from [@PT2; @BT].
Clebsch–Gordan maps {#CGmaps}
-------------------
Let the space ${{\mathcal M}}$ be defined by the direct integral $${{\mathcal M}}\equiv\int_{{{\mathbb R}}^+}^{\oplus}ds\; {{\mathcal P}}_{s}.$$ Realizing elements of ${{\mathcal T}}_s$ as functions $f(k)$ leads us to represent the elements of ${{\mathcal M}}$ by families of functions $f\equiv(\,f_s\,;\,s\in{{\mathbb R}}^+)$, where $f_s\equiv f_s(k)\in{{\mathcal T}}_s$ for all $s\in{{\mathbb R}}^+$. We shall define the multiplication operator ${{\mathsf s}}$ by $$\label{canact0}
{{\mathsf s}}f\,=\,\big(\,sf_s\,;
\,s\in{{\mathbb R}}^+\,\big).$$ To any family $(\,{{\mathsf O}}_s\,;\,s\in{{\mathbb R}}^+)$ of operators on ${{\mathcal T}}_s$ we may then associate an operator ${{\mathsf O}}_{{{\mathsf s}}}$ on ${{\mathcal M}}$ in the obvious manner. We have the corresponding canonical action of ${{\cal U}_q({{\mathfrak s}{\mathfrak l}}(2,{{\mathbb R}}))}$ on ${{\mathcal M}}$ via $$\label{canact}
\hat{\pi}_{{{{\mathsf s}}}}({{X}}) f\,=\,\big(\,{\pi}_{{s}}({{X}}) \, f_s\,;
\,s\in{{\mathbb R}}^+\,\big),\quad\forall\, {{X}}\in{{\cal U}_q({{\mathfrak s}{\mathfrak l}}(2,{{\mathbb R}}))}.$$
The Clebsch–Gordan maps ${{\mathsf C}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}$ were defined in [@PT1; @PT2] as a family of operators $$\begin{aligned}
{{\mathsf C}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}:
{{\mathcal P}}_{s_{{\mathfrak 2}}}{\otimes}{{\mathcal P}}_{s_{{\mathfrak 1}}}&
\longrightarrow {{\mathcal M}}\,.
$$ The Clebsch–Gordan maps ${{\mathsf C}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}$ intertwine the action of ${{\cal U}_q({{\mathfrak s}{\mathfrak l}}(2,{{\mathbb R}}))}$ on ${{\mathcal P}}_{s_{{\mathfrak 2}}}{\otimes}{{\mathcal P}}_{s_{{\mathfrak 1}}}$ with the canonical action on ${{\mathcal M}}$ in the sense that $${{\mathsf C}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}} \cdot (\pi_{s_{{\mathfrak 2}}} {\otimes}\pi_{s_{{\mathfrak 1}}}) \Delta ({{X}})=
\hat{\pi}_{{{{\mathsf s}}}}({{X}})\cdot{{\mathsf C}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}
\quad\forall\, {{X}}\in{{\cal U}_q({{\mathfrak s}{\mathfrak l}}(2,{{\mathbb R}}))}.$$
R–operator and braiding
-----------------------
Let us introduce $$\label{Rdef}
{{\mathsf R}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}} = q^{{{\mathsf H}}_{s_{{\mathfrak 2}}}{\otimes}{{\mathsf H}}_{s_{{\mathfrak 1}}}} \,
g_b\bigl({{\mathsf e}}_{s_{{\mathfrak 2}}}{\otimes}{{\mathsf f}}_{s_{{\mathfrak 1}}}\bigr)\, q^{{{\mathsf H}}_{s_{{\mathfrak 2}}}{\otimes}{{\mathsf H}}_{s_{{\mathfrak 1}}}}
\,.$$ Here the anti–self–adjoint operator ${{\mathsf H}}_s$ is defined by ${{\mathsf k}}_s=q^{{{\mathsf H}}_s}$, and $g_b(x)$ is the [*non–compact quantum dilogarithm*]{} related to the special function $w_b(x)$, defined in Appendix \[Qdil\], via $$\label{gw}
g_b(e^{2 \pi b x})= e^{\frac{\pi i}{24}(b^2 +b^{-2}) +
\frac{\pi i}{2} x^2} \, w_b(-x) \,.$$ The R–operator ${{\mathsf R}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}$ satisfies the following relations [@BT] $$\begin{aligned}
\label{Rrel1}
& {{\mathsf R}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}} \cdot (\pi_{s_{{\mathfrak 2}}} {\otimes}\pi_{s_{{\mathfrak 1}}}) {\Delta}({{X}}) =
(\pi_{s_{{\mathfrak 2}}} {\otimes}\pi_{s_{{\mathfrak 1}}}){\Delta}^\prime({{X}}) \cdot {{\mathsf R}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}} \,, \\
&\label{Rrel2}
{{\mathsf R}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}} \, ({{\mathsf e}}_{s_{{\mathfrak 2}}}^{} \otimes {{\mathsf k}}^{-1}_{s_{{\mathfrak 1}}}) =
({{\mathsf e}}_{s_{{\mathfrak 2}}}^{} \otimes {{\mathsf k}}_{s_{{\mathfrak 1}}}^{}) \, {{\mathsf R}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}} \,,\\
&
{{\mathsf R}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}} \, ({{\mathsf k}}_{s_{{\mathfrak 2}}}^{} {\otimes}{{\mathsf f}}_{s_{{\mathfrak 1}}}^{} ) =
({{\mathsf k}}^{-1}_{s_{{\mathfrak 2}}} {\otimes}{{\mathsf f}}_{s_{{\mathfrak 1}}}^{}) \, {{\mathsf R}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}} \,,\end{aligned}$$ where ${{X}}\in {{\cal U}_q({{\mathfrak s}{\mathfrak l}}(2,{{\mathbb R}}))}$ and $\Delta^\prime$ stands for the opposite co–product.
The braiding operator ${{\mathsf B}}: {{\mathcal P}}_{s_{{\mathfrak 2}}} \otimes {{\mathcal P}}_{s_{{\mathfrak 1}}} \to {{\mathcal P}}_{s_{{\mathfrak 1}}} \otimes {{\mathcal P}}_{s_{{\mathfrak 2}}}$ is defined by ${{\mathsf B}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}} = {{\mathsf P}}\, {{\mathsf R}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}$, where ${{\mathsf P}}$ is the operator that permutes the two tensor factors. In what follows we will need the following statement.
\[TBCC\] The braiding operator is diagonalized by the Clebsch–Gordan maps in the following sense: $$\label{BCC}
{{\mathsf C}}_{s_{{\mathfrak 1}}s_{{\mathfrak 2}}} \cdot {{\mathsf B}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}} =
\Omega^{\;{{\mathsf s}}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}} \cdot {{\mathsf C}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}} \,,$$ where $\Omega^{\;{{\mathsf s}}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}$ is the operator on ${{\mathcal M}}$ associated via [(\[canact0\])]{} to the scalar function $$\label{Om}
\Omega^{\;s}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}} =
e^{\pi i (s_{{\mathfrak 1}}^2 +s_{{\mathfrak 2}}^2 - s^2 +\frac{Q^2}{4})} \,.$$
In the particular case, , one can regard the permutation ${{\mathsf P}}$ as an endomorphism of ${{\mathcal P}}_{s_{{\mathfrak 1}}} \otimes {{\mathcal P}}_{s_{{\mathfrak 1}}}$ and then Proposition \[TBCC\] allows to relate it to the R–operator: $$\label{PRO}
{{\mathsf P}}= {{\mathsf R}}_{s_{{\mathfrak 1}}s_{{\mathfrak 1}}} \,
e^{-\pi i (\frac{Q^2}{4} + 2 s_{{\mathfrak 1}}^2 - {{\mathsf s}}^2) } \,.$$
Proof of Theorem \[Rthm\]
-------------------------
We will consider the more general R–operator defined on ${{\mathcal P}}_{s_{{\mathfrak 2}}}{\otimes}{{\mathcal P}}_{s_{{\mathfrak 1}}}$ by the formula $$\label{Rxxz}
{{\mathsf R}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(u)
= {{\mathsf R}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}} \, e^{+\pi i ({{\mathsf s}}^2_{{{\mathfrak 2}}{{\mathfrak 1}}}
-s_{{\mathfrak 1}}^2 - s_{{\mathfrak 2}}^2 - \frac{Q^2}{4})} \,
D_u({{\mathsf s}}_{{{\mathfrak 2}}{{\mathfrak 1}}})
$$ where the special function $D_\alpha(x)$ is defined by Eq. (\[Ds\]) in Appendix \[Qdil\], and ${{\mathsf s}}_{{{\mathfrak 2}}{{\mathfrak 1}}}$ is the unique positive self–adjoint operator such that $$\label{tC}
4 \cosh^2 \pi b {{\mathsf s}}_{{{\mathfrak 2}}{{\mathfrak 1}}} =
(\pi_{s_{{\mathfrak 2}}} {\otimes}\pi_{s_{{\mathfrak 1}}}) \Delta (C) \,.$$ It follows easily from relation (\[PRO\]) that ${{\mathsf R}}(u)\equiv{{\mathsf R}}_{ss}(u)$ coincides with the fundamental R–operator defined earlier in [(\[Rsimpledef\])]{}: $$\label{RGG}
{{{\mathsf R}}}(u) = {{\mathsf P}}\, w_b(u + {{\mathsf s}}) \, w_b(u - {{\mathsf s}}) ={{\mathsf P}}\, D_{u}({{\mathsf s}}) \,.$$
To begin the proof of Theorem \[Rthm\], which is somewhat more involved than the proofs of the analogous results in the case of highest weight representations [@Ji; @F1; @BD], let us observe that Eq. (\[RLL\]) is equivalent to the following set of equations $$\begin{aligned}
\label{Rdd}
{}& {{\mathsf R}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(u) \, {\Delta}(L^{{\pm}}) =
{\Delta}^\prime(L^{{\pm}}) \, {{\mathsf R}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(u) \,,\\
\label{Rabcd}
{}&{{\mathsf R}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(u) \, \ell_{ij}(u) =
\tau\bigl(\ell_{ij}(-u)\bigr) \, {{\mathsf R}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(u) \,,
\quad i,j=1,2 \,,\end{aligned}$$ where $L^\pm$ were introduced in (\[dLL\])–(\[Lpm\]), $\tau$ is the flip operation: $\tau(a_{s_{{\mathfrak 2}}} {{\otimes}} b_{s_{{\mathfrak 1}}})= b_{s_{{\mathfrak 2}}}{{\otimes}} a_{s_{{\mathfrak 1}}}$, and the operators $\ell_{ij}(u)$ are given by $$\begin{aligned}
\label{l11}
\ell_{{{\mathfrak 1}}{{\mathfrak 1}}}(u) &= e^{\pi b u} \, {{\mathsf e}}_{s_{{\mathfrak 2}}} {\otimes}{{\mathsf f}}_{s_{{\mathfrak 1}}}
+ e^{\pi b u} \, {{\mathsf k}}^{-1}_{s_{{\mathfrak 2}}} {\otimes}{{\mathsf k}}_{s_{{\mathfrak 1}}} +
e^{-\pi b u} \, {{\mathsf k}}_{s_{{\mathfrak 2}}} {\otimes}{{\mathsf k}}^{-1}_{s_{{\mathfrak 1}}} \,, \\
\label{l12}
\ell_{{{\mathfrak 1}}{{\mathfrak 2}}}(u) &= e^{\pi b u} \, {{\mathsf k}}_{s_{{\mathfrak 2}}} {\otimes}{{\mathsf f}}_{s_{{\mathfrak 1}}}
+ e^{-\pi b u} \, {{\mathsf f}}_{s_{{\mathfrak 2}}} {\otimes}{{\mathsf k}}^{-1}_{s_{{\mathfrak 1}}} \,, \\
\label{l21}
\ell_{{{\mathfrak 2}}{{\mathfrak 1}}}(u) &= e^{\pi b u} \, {{\mathsf e}}_{s_{{\mathfrak 2}}} {\otimes}{{\mathsf k}}^{-1}_{s_{{\mathfrak 1}}}
+ e^{-\pi b u} \, {{\mathsf k}}_{s_{{\mathfrak 2}}} {\otimes}{{\mathsf e}}_{s_{{\mathfrak 1}}} \,, \\
\label{l22}
\ell_{{{\mathfrak 2}}{{\mathfrak 2}}}(u) &= e^{-\pi b u} \, {{\mathsf f}}_{s_{{\mathfrak 2}}} {\otimes}{{\mathsf e}}_{s_{{\mathfrak 1}}}
+ e^{-\pi b u} \, {{\mathsf k}}^{-1}_{s_{{\mathfrak 2}}} {\otimes}{{\mathsf k}}_{s_{{\mathfrak 1}}} +
e^{\pi b u} \, {{\mathsf k}}_{s_{{\mathfrak 2}}} {\otimes}{{\mathsf k}}^{-1}_{s_{{\mathfrak 1}}} \,.\end{aligned}$$
The verification of the relations [(\[Rdd\])]{} is easy. Introducing $$\label{rdef}
{{\mathsf r}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(u) = {{\mathsf R}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}^{-1} \cdot
{{\mathsf R}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}^{}(u) \,,$$ it follows from [(\[Rrel1\])]{} that [(\[Rdd\])]{} is equivalent to the system of equations $$\label{rdd}
{{\mathsf r}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(u) \, {\Delta}({{\mathsf X}}) =
{\Delta}({{\mathsf X}}) \, {{\mathsf r}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(u)\, .$$ Validity of the relation [(\[rdd\])]{} follows from $[{{\mathsf s}},{\Delta}({{\mathsf X}}) ]=0$.
The verification of the relations [(\[Rabcd\])]{} is somewhat harder. To begin with, let us observe that it suffices to verify relation [(\[l12\])]{}, say.
Validity of [(\[Rabcd\])]{} for $(i,j)=(1,2)$ implies that ${{\mathsf R}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(u)$ satisfies the three remaining relations in [(\[Rabcd\])]{} as well.
Equivalence of the $(1,2)$ and $(2,1)$ equations in [(\[Rabcd\])]{} can be established by invoking the automorphism $\theta$, defined in [(\[thetaEFK\])]{}. Introduce $\tau{\!\vphantom |}_\theta \equiv (\theta {\otimes}\theta) \circ \tau$. It is easy to see that $\tau{\!\vphantom |}_\theta \bigl({{\mathsf R}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}\bigr)={{\mathsf R}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}$ and $\tau{\!\vphantom |}_\theta \bigl( \Delta(C) \bigr) = \Delta(C)$. Therefore, $\tau{\!\vphantom |}_\theta \bigl({{\mathsf R}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(u)\bigr)
={{\mathsf R}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(u)$. The claimed equivalence of the $(1,2)$ and $(2,1)$ equations in [(\[Rabcd\])]{} follows now by observing that $\tau{\!\vphantom |}_\theta \bigl( \ell_{{{\mathfrak 1}}{{\mathfrak 2}}}(u) \bigr) =
\ell_{{{\mathfrak 2}}{{\mathfrak 1}}}(u)$.
In order to prove that relation [(\[Rabcd\])]{} for $(i,j)=(1,1)$ follows from the validity of relations [(\[Rabcd\])]{} for $(i,j)=(1,2)$, $(i,j)=(2,1)$ and [(\[Rdd\])]{}, let us consider the following object: $$\begin{aligned}
{\nonumber}{{\mathsf X}}(u) &=
q \bigl( {{\mathsf k}}^2_{s_{{\mathfrak 2}}} {\otimes}{{\mathsf k}}^2_{s_{{\mathfrak 1}}} -
{{\mathsf k}}^{-2}_{s_{{\mathfrak 2}}} {\otimes}{{\mathsf k}}^{-2}_{s_{{\mathfrak 1}}} \bigr) \, \ell_{{{\mathfrak 1}}{{\mathfrak 1}}}(u)
+ {{\mathsf k}}_{s_{{\mathfrak 2}}} {\otimes}{{\mathsf k}}_{s_{{\mathfrak 1}}}
\bigl( e^{-\pi b u} {{\mathsf C}}_{s_{{\mathfrak 2}}} {\otimes}{\sf 1}
+ e^{\pi b u} {\sf 1} {\otimes}{{\mathsf C}}_{s_{{\mathfrak 1}}} \bigr) \\
\label{X1}
& + {{\mathsf k}}^{-1}_{s_{{\mathfrak 2}}} {\otimes}{{\mathsf k}}^{-1}_{s_{{\mathfrak 1}}}
\bigl( e^{\pi b u} {{\mathsf C}}_{s_{{\mathfrak 2}}} {\otimes}{\sf 1} +
e^{-\pi b u} {\sf 1} {\otimes}{{\mathsf C}}_{s_{{\mathfrak 1}}} \bigr) \,.\end{aligned}$$ The operator $( {{\mathsf k}}^2_{s_{{\mathfrak 2}}} {\otimes}{{\mathsf k}}^2_{s_{{\mathfrak 1}}} -
{{\mathsf k}}^{-2}_{s_{{\mathfrak 2}}} {\otimes}{{\mathsf k}}^{-2}_{s_{{\mathfrak 1}}})$ does not have a normalizable zero mode. Validity of relation [(\[Rabcd\])]{} for $(i,j)=(1,1)$ therefore follows from $$\label{RX}
{{\mathsf R}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(u) \, {{\mathsf X}}(u) =
\tau \bigl({{\mathsf X}}(-u)\bigr) \, {{\mathsf R}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(u) \,.$$ Validity of this relation follows from the observation that ${{\mathsf X}}(u)$ can be represented in the form $$\begin{aligned}
\label{X2}
{{\mathsf X}}(u) =
({{\mathsf k}}_{s_{{\mathfrak 2}}} {\otimes}{{\mathsf k}}_{s_{{\mathfrak 1}}}) \, \ell_{{{\mathfrak 1}}{{\mathfrak 2}}}(u) \,
(\pi_{s_{{\mathfrak 2}}} {{\otimes}} \pi_{s_{{\mathfrak 1}}}) \Delta(e) -
({{\mathsf k}}^{-1}_{s_{{\mathfrak 2}}} {\otimes}{{\mathsf k}}^{-1}_{s_{{\mathfrak 1}}}) \ell_{{{\mathfrak 2}}{{\mathfrak 1}}}(u) \,
(\pi_{s_{{\mathfrak 2}}} {{\otimes}} \pi_{s_{{\mathfrak 1}}}) \Delta(f) \,.\end{aligned}$$
Exchanging ${{\mathsf k}}\leftrightarrow {{\mathsf k}}^{-1}$ in [(\[X2\])]{}, we can derive relation [(\[Rabcd\])]{} for $(i,j)=(2,2)$ in a completely analogous way.
It remains to verify relation [(\[Rabcd\])]{} for $(i,j)=(1,2)$. This relation may be rewritten in terms of the operator ${{\mathsf r}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(u)$ defined in [(\[rdef\])]{} by using the relation [(\[Rrel2\])]{}. We conclude that validity of [(\[Rabcd\])]{} follows from the validity of $$\begin{aligned}
\label{rek}
{{\mathsf r}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(u) \, &\bigl( e^{2 \pi b u} {{\mathsf O}}_1 + {{\mathsf O}}_2 \bigr)
= \bigl( e^{2 \pi b u} \, {{\mathsf O}}_1 + {{\mathsf O}}_3
\bigr) \, {{\mathsf r}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(u) \,,\end{aligned}$$ where we have introduced the convenient abbreviations $$\label{EB}
\begin{aligned}
{{\mathsf O}}_{{\mathfrak 1}}\equiv &\ {{\mathsf e}}_{s_{{\mathfrak 2}}} \otimes {{\mathsf k}}_{s_{{\mathfrak 1}}}^{-1}\,,\\
{{\mathsf O}}_{{\mathfrak 2}}\equiv &\ {{\mathsf k}}_{s_{{\mathfrak 2}}} \otimes {{\mathsf e}}_{s_{{\mathfrak 1}}} \,,
\end{aligned}\qquad
{{\mathsf O}}_{{\mathfrak 3}}\equiv {{\mathsf R}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}^{-1} \, ({{\mathsf k}}_{s_{{\mathfrak 2}}}^{-1}
\otimes {{\mathsf e}}_{s_{{\mathfrak 1}}}) \, {{\mathsf R}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}^{} \,.$$
In order to prove that ${{\mathsf r}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(u)$ satisfies [(\[rek\])]{} it will be convenient to use the Clebsch–Gordan maps ${{\mathsf C}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}$ introduced in Subsection \[CGmaps\]. Let us observe that (\[rdd\]) implies that the r–matrix is diagonalized by the Clebsch–Gordan maps: $$\label{rCG}
{{\mathsf C}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}} \cdot {{\mathsf r}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(u) =
{r}^{\,{{\mathsf s}}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(u) \cdot {{\mathsf C}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}} \,,$$ where $r^{\,{{\mathsf s}}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(u)$ is the operator on ${{\mathcal M}}$ associated to the scalar function $r^{\,s}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(u)$ via [(\[canact0\])]{}.
In order to further evaluate equation [(\[rek\])]{} we will need to describe the images $\tilde{{{\mathsf O}}}_{\ell}$, $\ell{=}1,2,3$ of the operators ${{\mathsf O}}_\ell$ under the maps ${{\mathsf C}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}$ which are defined by .
\[SH\] The operators $\tilde{{{\mathsf O}}}_{\ell}$ can be represented as follows: If $g\equiv(g_s;s\in{{\mathbb R}}^+)\in{{\mathcal M}}$ then $\tilde{{{\mathsf O}}}_{\ell}
g\equiv \big(\,(\tilde{{{\mathsf O}}}_{\ell} g)_s^{}\,;\,s\in{{\mathbb R}}^+\,\big)$, where $$\label{Ekjn}
(\tilde{{{\mathsf O}}}_{\ell} g)_s^{}(k) \equiv
\! \sum_{\nu=-1}^{1} \, A_{\ell;s}^{\nu;s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(k)\,{{\mathsf T}}_{s}^{\nu}
g_{s}^{}(k {{+}}ib) \,,$$ where ${{\mathsf T}}_{s}^{\nu}g_s=g_{s {{+}}i \nu b}$. The coefficients $A_{\ell;s}^{\nu;s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(k)$ are symmetric under exchange of $s_{{\mathfrak 2}}$ and $s_{{\mathfrak 1}}$, $$\label{AA}
A_{\ell;s}^{\nu;s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(k) = A_{\ell;s}^{\nu;s_{{\mathfrak 1}}s_{{\mathfrak 2}}}(k)$$ and are otherwise related to each other by $$\label{al23'}
\begin{aligned}
A_{2;s}^{\nu;s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(k)
= &\ e^{+2\pi b\nu s + i \pi b^2 \nu^2}
A_{1;s}^{\nu;s_{{\mathfrak 1}}s_{{\mathfrak 2}}}(k)\,, \\
A_{3;s}^{\nu;s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(k)
= &\ e^{-2\pi b\nu s - i \pi b^2 \nu^2}
A_{1;s}^{\nu;s_{{\mathfrak 1}}s_{{\mathfrak 2}}}(k)\,.
\end{aligned}$$
The proof of Proposition \[SH\], which is somewhat technical, is given in Subsection [(\[proppf\])]{}.
Proposition \[SH\] together with equation [(\[rCG\])]{} allows us to rewrite the defining relation (\[rek\]) as a commutation relation satisfied by the corresponding operators on ${{\mathcal M}}$. Applying (\[rek\]) to a function $g_s$ and matching the coefficients in front of $g_{s {{+}}i\nu}(k {{+}}ib)$ in the resulting equation, we derive functional equations on $r^{\,s}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(u)$, $$\label{reqs}
\begin{aligned}
r^{s{{+}}i b\nu }_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(u) \,&\
\big(e^{2\pi b u } \, A_{1;s}^{\nu;s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(k)
+ A_{2;s}^{\nu;s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(k) \big) \\
&\ = \big(e^{2\pi b u } \, A_{1;s}^{\nu;s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(k)
+ A_{3;s}^{\nu;s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(k) \big)\;
r^{\,s}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(u) \,.
\end{aligned}$$ The case $\nu=0$ holds trivially. Taking into account (\[AA\]) and [(\[al23’\])]{}, Eqs. (\[reqs\]) for $\nu=\pm 1$ are equivalent to a single functional equation, $$\label{req}
r^{s + ib}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(u) \,
\bigl(e^{2\pi b u} + \zeta \bigr) =
\bigl(e^{2\pi b u} + \zeta^{-1} \bigr) \,
r^{\,s}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(u) \,,$$ where $ \zeta = e^{i \pi b^2 + 2\pi b s}$. In terms of $\tilde{r}^{\,s}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(u)=
e^{-i\pi s^2 } r^{\,s}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(u)$ one may rewrite this functional relation as follows $$\label{req2}
\tilde{r}^{s+ i \frac{b}{2}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(u)
\, \cosh \pi b (s + u) =
\tilde{r}^{s- i \frac{b}{2}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(u) \,
\cosh \pi b (s - u) \,.$$ Recalling that the special function $w_b(x)$ satisfies (\[wfunrel\]), we conclude that equation (\[req\]) is solved by the expression for $r^{\,s}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}$ which follows from eqn. [(\[Rxxz\])]{} via [(\[rdef\])]{} and [(\[rCG\])]{}, namely $$\label{rl}
r^{\,s}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}} (u) =
e^{\pi i (s^2 -s_{{\mathfrak 1}}^2 - s_{{\mathfrak 2}}^2 - \frac{Q^2}{4})} \,
\frac{w_b(u + s)}{w_b(s - u)} =
e^{\pi i (s^2 -s_{{\mathfrak 1}}^2 - s_{{\mathfrak 2}}^2 - \frac{Q^2}{4})} \,
D_{u}(s) \,,$$ where $D_{\alpha}(x)$ is defined in (\[Ds\]).
Property (\[Dcc\]) implies that $|r^{\, s}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(u)|=1$ if $u, s \in {{\mathbb R}}$. Since ${{\mathsf s}}_{{{\mathfrak 2}}{{\mathfrak 1}}}$ is self–adjoint, we infer that $r^{\,{{\mathsf s}}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(u)$ is a unitary operator for $u\in{{\mathbb R}}$. On the other hand property (\[wcc\]) implies that $g_b(x)$ given by (\[gw\]) satisfies $|g_b(x)|=1$ for $x\in{{\mathbb R}}^+$. Since ${{\mathsf e}}_{{\mathfrak 2}}\otimes{{\mathsf f}}_{{\mathfrak 1}}$ in (\[Rdef\]) is positive self–adjoint, we infer that ${{\mathsf R}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}$ is unitary. The unitarity of ${{\mathsf R}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(u)$ for $u\in{{\mathbb R}}$ follows.$\square$\
Let us finally remark that relation (\[Das2\]) leads to the following asymptotics of ${{\mathsf R}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(u)$ $$\label{Ras}
\begin{aligned}
{{\mathsf R}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(u) &\sim e^{-i\pi (u^2 + \ldots)} \, {{\mathsf R}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}
\quad \text{for } {\rm Re}(u) \to +\infty \\
{{\mathsf R}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(u) &\sim e^{+i\pi (u^2 + \ldots)} \, \bar{{\mathsf R}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}
\quad \text{for } {\rm Re}(u) \to -\infty \,,
\end{aligned}$$ where $\bar{{\mathsf R}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}^{}\equiv
{{\mathsf P}}\cdot{{\mathsf R}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}^{-1}\cdot
{{\mathsf P}}$.
Clebsch–Gordan and Racah–Wigner coefficients for ${{\mathcal P}}_s$
-------------------------------------------------------------------
In order to prove Proposition \[SH\] we will need to describe the Clebsch–Gordan maps ${{\mathsf C}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}$ more explicitly. For the following it will be convenient to use the variables ${\alpha}_r=\frac{Q}{2}+is_r$ in order parameterize the representations ${{\mathcal P}}_{s_r}$, $r=0,1,2\dots$. The Clebsch–Gordan maps ${{\mathsf C}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}$ can then be represented explicitly as an integral transformation [@PT2; @BT]: $$\tilde{f}_s(k)\,\equiv\,
\int dk_2dk_1 \;\big[\begin{smallmatrix}
{\alpha}_3 \\
k_3\end{smallmatrix}\big|
\begin{smallmatrix}{\alpha}_2 & {\alpha}_1 \\ k_2 & k_1\end{smallmatrix}
\big]\,f(k_2,k_1),$$ The kernel which appears on the right hand side, the so–called b–Clebsch–Gordan kernel, was calculated in [@PT2; @BT]. It is of the general form $$\label{CGexpl}
\big[\begin{smallmatrix}
{\alpha}_3 \\
k_3\end{smallmatrix}\big|
\begin{smallmatrix}{\alpha}_2 & {\alpha}_1 \\ k_2 & k_1\end{smallmatrix}
\big]\,=\,{\delta}(k_3-k_2-k_1)\,
\big[\begin{smallmatrix}
{\alpha}_3 \end{smallmatrix}\big|
\begin{smallmatrix}{\alpha}_2 & {\alpha}_1 \\ k_2 & k_1\end{smallmatrix}
\big]$$ where the function $\big[\begin{smallmatrix}
{\alpha}_3
\end{smallmatrix}\big|
\begin{smallmatrix}{\alpha}_2 & {\alpha}_1 \\ k_2 & k_1\end{smallmatrix}
\big]$ is given as $$\big[\begin{smallmatrix}
{\alpha}_3 \end{smallmatrix}\big|
\begin{smallmatrix}{\alpha}_2 & {\alpha}_1 \\ k_2 & k_1\end{smallmatrix}
\big]=
\frac{
e^{-\frac{\pi i}{2}{\alpha}_{123}^{(1)}{\alpha}_{123}^{(0)}}
e^{\pi (k_1{\alpha}_2-k_2{\alpha}_1)}}
{S_b({\alpha}_1+{\alpha}_2+{\alpha}_3-Q)}
\,\Phi^{}_3(R_1,R_2,R_3;S_1,S_2,S_3;-{\alpha}_{123}^{(0)}).$$ In equation [(\[CGexpl\])]{} we have used the special function $\Phi^{}_3(R_1,R_2,R_3;S_1,S_2,S_3;x)$ defined in [(\[Phidef\])]{} whose arguments have been chosen as follows: $$\begin{aligned}
R_1&={\alpha}_1+ik_1\\
R_2&={\alpha}_2-ik_2\\
R_3&={\alpha}_{123}^{(0)}
\end{aligned}
\qquad
\begin{aligned}
S_1&={\alpha}_{123}^{(1)}+{\alpha}_1+ik_1\\
S_2&={\alpha}_{123}^{(2)}+{\alpha}_2-ik_2\\
S_3&=Q.
\end{aligned}$$ Here and below we are using the following notations: $$\begin{aligned}
{} &{\alpha}_{ijk}^{(1)}={\alpha}_j+{\alpha}_k-{\alpha}_i,\\
{} &{\alpha}_{ijk}^{(2)}={\alpha}_k+{\alpha}_i-{\alpha}_j,\\
{} &{\alpha}_{ijk}^{(3)}={\alpha}_i+{\alpha}_j-{\alpha}_k,
\end{aligned}\qquad
{\alpha}_{ijk}^{(0)}={\alpha}_i+{\alpha}_j+{\alpha}_k-Q.$$ The precise relation to the b–Clebsch–Gordan coefficients from [@BT] is as follows: $$\big[\begin{smallmatrix}
{\alpha}_3 \\
k_3\end{smallmatrix}\big|
\begin{smallmatrix}{\alpha}_2 & {\alpha}_1 \\ k_2 & k_1\end{smallmatrix}
\big]_{\rm here}
\equiv \nu^{{\alpha}_3}_{{\alpha}_2{\alpha}_1}
\big[\begin{smallmatrix}
{\alpha}_3 \\
k_3\end{smallmatrix}\big|
\begin{smallmatrix}{\alpha}_2 & {\alpha}_1 \\ k_2 & k_1\end{smallmatrix}
\big]_{\rm BT}$$ with phases $\nu_{{\alpha}_2{\alpha}_1}^{{\alpha}_3}$ chosen as $$\big(\nu_{{\alpha}_2{\alpha}_1}^{{\alpha}_3}\big)^{-1}\,=\,
S_b({\alpha}_3+{\alpha}_1-{\alpha}_2)S_b({\alpha}_3+{\alpha}_2-{\alpha}_1).$$
The b–Racah–Wigner coefficients are then defined by the relation $$\label{RWdef} \begin{aligned}
\big[\begin{smallmatrix}
{\alpha}_s\\
k_s\end{smallmatrix}\big|
\begin{smallmatrix}
{\alpha}_1 & {\alpha}_0\\
k_1 & k_0
\end{smallmatrix}\big]&
\big[\begin{smallmatrix}
{\alpha}_3 \\
k_3\end{smallmatrix}\big|
\begin{smallmatrix}
{\alpha}_2 & {\alpha}_s\\
k_2 & k_s
\end{smallmatrix}
\big]\\
&=\frac{1}{i}\int\limits_{{{\mathbb S}}} d{\alpha}_t\;
\big\{\begin{smallmatrix}
{\alpha}_0 & {\alpha}_1\\
{\alpha}_2 & {\alpha}_3
\end{smallmatrix}\big|\begin{smallmatrix}{\alpha}_s\\{\alpha}_t
\end{smallmatrix}
\big\}\;
\big[\begin{smallmatrix}
{\alpha}_3 \\
k_3\end{smallmatrix}\big|
\begin{smallmatrix}{\alpha}_t & {\alpha}_0 \\ k_t & k_0\end{smallmatrix}
\big]
\big[\begin{smallmatrix}{\alpha}_t\\k_t\end{smallmatrix}
\big|
\begin{smallmatrix}{\alpha}_2 & {\alpha}_1\\ k_2 & k_1
\end{smallmatrix}\big],
\end{aligned}$$ where ${{\mathbb S}}\equiv \frac{Q}{2}+i{{\mathbb R}}_+$. The coefficients $\{\dots\}$ can be represented explicitly by the following formula $$\begin{aligned}
\big\{\begin{smallmatrix}
{\alpha}_0 & {\alpha}_1\\
{\alpha}_2 & {\alpha}_3
\end{smallmatrix}\big|\begin{smallmatrix}{\alpha}_s\\{\alpha}_t
\end{smallmatrix}
\big\}=&\;
\frac{S_b({\alpha}_t+{\alpha}_0+{\alpha}_3-Q)S_b({\alpha}_3+{\alpha}_0-{\alpha}_t)S_b({\alpha}_3+{\alpha}_t-{\alpha}_0)}
{S_b({\alpha}_s+{\alpha}_2+{\alpha}_3-Q)S_b({\alpha}_3+{\alpha}_2-{\alpha}_s)S_b({\alpha}_3+{\alpha}_s-{\alpha}_2)}\\
&\;\times\frac{S_b({\alpha}_t+{\alpha}_2-{\alpha}_1)}
{S_b({\alpha}_s+{\alpha}_0-{\alpha}_1)}\Phi^{}_4(U_1,U_2,U_3,U_4;V_1,V_2,V_3,V_4;0)
\end{aligned}$$ where $$\begin{aligned}
U_1&={\alpha}_s+{\alpha}_0-{\alpha}_1\\
U_2&={\alpha}_s+Q-{\alpha}_0-{\alpha}_1\\
U_3&={\alpha}_s+{\alpha}_2+{\alpha}_3-Q\\
U_4&={\alpha}_s+{\alpha}_3-{\alpha}_2
\end{aligned}
\qquad
\begin{aligned}
V_1&=Q-{\alpha}_t+{\alpha}_s+{\alpha}_3-{\alpha}_1\\
V_2&={\alpha}_t+{\alpha}_s+{\alpha}_3-{\alpha}_1\\
V_3&=2{\alpha}_s\\
V_4&=Q.
\end{aligned}$$ For completeness let us note that the b–Racah–Wigner coefficients $\{\cdots\}$ are related to the corresponding objects from [@PT2] via $$\begin{aligned}
&\big\{\begin{smallmatrix}
{\alpha}_0 & {\alpha}_1\\
{\alpha}_2 & {\alpha}_3
\end{smallmatrix}\big|\begin{smallmatrix}{\alpha}_s\\{\alpha}_t
\end{smallmatrix}
\big\}_{\rm here}
\,=\,\frac{\nu_{{\alpha}_1{\alpha}_0}^{{\alpha}_s}\nu_{{\alpha}_2{\alpha}_s}^{{\alpha}_3}}
{\nu_{{\alpha}_2{\alpha}_1}^{{\alpha}_t}\nu_{{\alpha}_t{\alpha}_0}^{{\alpha}_3}}
\big\{\begin{smallmatrix}
{\alpha}_0 & {\alpha}_1\\
{\alpha}_2 & {\alpha}_3
\end{smallmatrix}\big|\begin{smallmatrix}{\alpha}_s\\{\alpha}_t
\end{smallmatrix}
\big\}_{\rm PT2}\,.\end{aligned}$$
The b–Clebsch–Gordan and b–Racah–Wigner coefficients are meromorphic functions of all of their arguments. The complete set of poles may be described as follows [@PT2; @BT][^7]\
$$\label{RWpoles}
\begin{aligned}
{} & Q-{\alpha}_{32s}^{(\iota)}=-nb-mb^{-1},\\
{} & Q-{\alpha}_{s10}^{(\iota)}=-nb-mb^{-1},
\end{aligned}\qquad
\begin{aligned}
{} & {\alpha}_{3t0}^{(\iota)}=-nb-mb^{-1},\\
{} & {\alpha}_{t21}^{(\iota)}=-nb-mb^{-1},
\end{aligned}\qquad \iota=0,1,2,3,$$ where $n,m\in{{\mathbb Z}}^{\geq 0}$.\
$$\label{CGpoles}
\begin{aligned}
{} &Q-{\alpha}_{321}^{(\iota)}=-nb-mb^{-1},\qquad \iota=0,1,2,3,\\
{} & \pm ik_i={\alpha}_i+nb+mb^{-1},\qquad i=1,2,\\
{} & \pm ik_3=Q-{\alpha}_3+nb+mb^{-1},
\end{aligned}$$ where again $n,m\in{{\mathbb Z}}^{\geq 0}$.
Proof of Proposition \[SH\] {#proppf}
---------------------------
Our proof of Proposition \[SH\] will be based on the following nontrivial identity satisfied by the –Gordan kernel.
\[idlem\] $$\label{CGid} \begin{aligned}
\big[\!\big[\begin{smallmatrix}
{\alpha}_1 \end{smallmatrix}\big|
\begin{smallmatrix}
{\alpha}_1 & -b\\
k_1 & \;\, ib
\end{smallmatrix}
\big]\!\big]&
\big[\begin{smallmatrix}
{\alpha}_3\end{smallmatrix}\big|
\begin{smallmatrix}
{\alpha}_2 & {\alpha}_1\\
k_2 & k_1+ib
\end{smallmatrix}\big]\\
&=\sum_{\tau=-1}^1
F_{\tau}\big[\begin{smallmatrix}
{\alpha}_2 & {\alpha}_1\\
{\alpha}_3 & -b
\end{smallmatrix}
\big]\;
\big[\!\big[\begin{smallmatrix}
{\alpha}_3 \end{smallmatrix}\big|
\begin{smallmatrix}{\alpha}_3+\tau b & -b\\ k_3 & ib\end{smallmatrix}
\big]\!\big]\,{{\mathsf T}}_{{\alpha}_3}^{\tau}
\big[\begin{smallmatrix}{\alpha}_3\end{smallmatrix}
\big|
\begin{smallmatrix}{\alpha}_2 & {\alpha}_1\\ k_2 & k_1
\end{smallmatrix}\big],
\end{aligned}$$ where ${{\mathsf T}}_{{\alpha}_3}^{\tau}f({\alpha}_3)=f({\alpha}_3+\tau b)$, $k_3=k_2+k_1$ and we have furthermore used the notation $$\begin{aligned}
& \big[\!\big[\begin{smallmatrix}
{\alpha}_3 \end{smallmatrix}\big|
\begin{smallmatrix}
{\alpha}_2 & {\alpha}_1\\
k_2 & -i{\alpha}_1
\end{smallmatrix}
\big]\!\big]=
2\pi i\!
\operatorname*{Res}_{k_1= -i{\alpha}_1} \big[\begin{smallmatrix}
{\alpha}_3\end{smallmatrix}\big|
\begin{smallmatrix}
{\alpha}_2 & {\alpha}_1\\
k_2 & k_1
\end{smallmatrix}\big]\,,\\
& F_{\tau}\big[\begin{smallmatrix}
{\alpha}_2 & \;\,{\alpha}_1\\
{\alpha}_3 & -b
\end{smallmatrix}
\big]=2\pi i\operatorname*{Res}_{{\alpha}_\tau={\alpha}_3+b\tau}
\big\{\begin{smallmatrix}
-b & {\alpha}_1\\
\;{\alpha}_2 & {\alpha}_3
\end{smallmatrix}\big|\begin{smallmatrix}{\alpha}_1\\
{\alpha}_\tau
\end{smallmatrix}
\big\}\,,\end{aligned}$$ We have the explicit formulae $$\begin{aligned}
\label{CGres}
\big[\!\big[\begin{smallmatrix}
{\alpha}_3 \end{smallmatrix}\big|
\begin{smallmatrix}
{\alpha}_2 & {\alpha}_1\\
k_2 & -i{\alpha}_1
\end{smallmatrix}
\big]\!\big]&=\frac{
e^{\frac{\pi i}{2}({\Delta}_{{\alpha}_3}-{\Delta}_{{\alpha}_2}-{\Delta}_{{\alpha}_1})}}
{S_b({\alpha}_3+{\alpha}_2-{\alpha}_1)}
e^{-\pi k_2{\alpha}_1}\frac{S_b({\alpha}_2-ik_2)}{S_b({\alpha}_1+{\alpha}_3-ik_2)}\,,\\
F_{-}^{}\big[\begin{smallmatrix}
{\alpha}_2 & \;\,{\alpha}_1\\
{\alpha}_3 & -b
\end{smallmatrix}
\big]
&=\frac{S_b(2{\alpha}_3-2b-Q)}{S_b(2{\alpha}_1+b)}\sin\pi b({\alpha}_2+{\alpha}_1-{\alpha}_3)
\label{RWres1}\,,\\
F_{+}^{}\big[\begin{smallmatrix}
{\alpha}_2 & \;\,{\alpha}_1\\
{\alpha}_3 & -b
\end{smallmatrix}
\big]
&= \frac{S_b(2{\alpha}_3-Q)}{S_b(2{\alpha}_1+b)}\sin\pi b({\alpha}_3+{\alpha}_2+{\alpha}_1-Q)
\label{RWres2}\\
&\quad{\times}\sin\pi b({\alpha}_3+{\alpha}_1-{\alpha}_2)\sin\pi b({\alpha}_3+{\alpha}_2-{\alpha}_1)\,,{\nonumber}\end{aligned}$$ where ${\Delta}_{{\alpha}}={\alpha}(Q-{\alpha})$.
Given that Lemma \[idlem\] holds, it becomes easy to complete the proof of Proposition \[SH\] as follows: Notice that the left hand side of [(\[CGid\])]{} can be written as $$\begin{aligned}
\big[\!\big[\begin{smallmatrix}{\alpha}_1 \end{smallmatrix}\big|
\begin{smallmatrix}
{\alpha}_1 & -b\\
k_1 & \;\,ib
\end{smallmatrix}\big]\!\big]
\big[\begin{smallmatrix}
{\alpha}_3\end{smallmatrix}\big|
\begin{smallmatrix}
{\alpha}_2 & {\alpha}_1\\
k_2 & k_1+ib
\end{smallmatrix}\big]
&\;=\;e^{\pi bk_1}\,\frac{[{\alpha}_1-ik_1-b]_b}{S_b(2{\alpha}_1+b)}\,
\big[\begin{smallmatrix}
{\alpha}_3\end{smallmatrix}\big|
\begin{smallmatrix}
{\alpha}_2 & {\alpha}_1\\
k_2 & k_1+ib
\end{smallmatrix}\big]\\
&=\,
\frac{e^{\pi b k_3}}{S_b(2{\alpha}_1+b)}
\;\big({{\mathsf K}}_{s_{{\mathfrak 2}}}{\otimes}{{\mathsf E}}_{s_{{\mathfrak 1}}}\big)^t
\cdot\big[\begin{smallmatrix}
{\alpha}_3\end{smallmatrix}\big|
\begin{smallmatrix}
{\alpha}_2 & {\alpha}_1\\
k_2 & k_1 \end{smallmatrix}\big],\end{aligned}$$ where ${{\mathsf O}}^t$ denotes the transpose of an operator on ${{\mathcal P}}_{s_{{\mathfrak 2}}}{\otimes}{{\mathcal P}}_{s_{{\mathfrak 1}}}$ defined by $$\int dk_{{\mathfrak 2}}dk_{{\mathfrak 1}}\;f(k_{{\mathfrak 2}},k_{{\mathfrak 1}})\,({{\mathsf O}}g)(k_{{\mathfrak 2}},k_{{\mathfrak 1}})\,\equiv\,
\int dk_{{\mathfrak 2}}dk_{{\mathfrak 1}}\;({{\mathsf O}}^t f)(k_{{\mathfrak 2}},k_{{\mathfrak 1}})\,g(k_{{\mathfrak 2}},k_{{\mathfrak 1}}).$$ Equation [(\[CGid\])]{} may therefore be written in the form $$\big({{\mathsf K}}_{s_{{\mathfrak 2}}}{\otimes}{{\mathsf E}}_{s_{{\mathfrak 1}}}\big)^t\cdot
\big[\begin{smallmatrix}
{\alpha}_3\end{smallmatrix}\big|
\begin{smallmatrix}
{\alpha}_2 & {\alpha}_1\\
k_2 & k_1 \end{smallmatrix}\big]\,=\,
\sum_{\tau =-1}^1 A_{2;s}^{\tau;s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(k_3)\,{{\mathsf T}}_{{\alpha}_3}^{\tau}
\big[\begin{smallmatrix}{\alpha}_3\end{smallmatrix}
\big|
\begin{smallmatrix}{\alpha}_2 & {\alpha}_1\\ k_2 & k_1
\end{smallmatrix}\big],$$ where $$A_{2;s}^{\tau;s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(k_3)=e^{-\pi bk_3}S_b(2{\alpha}_1+b)
F_{\tau}\big[\begin{smallmatrix}
{\alpha}_2 & {\alpha}_1\\
{\alpha}_3 & -b
\end{smallmatrix}
\big]\;
\big[\!\big[\begin{smallmatrix}
{\alpha}_3 \end{smallmatrix}\big|
\begin{smallmatrix}{\alpha}_3+\tau b & -b\\ k_3 & ib\end{smallmatrix}
\big]\!\big].$$ By using the explicit expressions [(\[RWres1\])]{},[(\[RWres2\])]{} one may easily verify that the coefficients $A_{2;s}^{t;s_{{\mathfrak 2}}s_{{\mathfrak 1}}}$ are symmetric under the exchange of $s_{{\mathfrak 2}}$ and $s_{{\mathfrak 1}}$, as claimed. This completes the proof of all the relevant statements of Proposition \[SH\] for the case of the operator ${{\mathsf O}}_2$.
In order to cover the remaining cases let us observe that $$\label{EBapp}
\begin{aligned}
{{\mathsf O}}_2 \equiv &\ {{\mathsf K}}_{s_{{\mathfrak 2}}} \otimes {{\mathsf E}}_{s_{{\mathfrak 1}}} =
{{\mathsf B}}_{s_{{\mathfrak 1}}s_{{\mathfrak 2}}} \, {{\mathsf O}}_1 \, {{\mathsf B}}_{s_{{\mathfrak 1}}s_{{\mathfrak 2}}}^{-1} \,, \\
{{\mathsf O}}_3 \equiv &\ {{\mathsf R}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}^{-1} \, ({{\mathsf K}}_{s_{{\mathfrak 2}}}^{-1}
\otimes {{\mathsf E}}_{s_{{\mathfrak 1}}}) \, {{\mathsf R}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}} =
{{\mathsf B}}^{-1}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}} \, {{\mathsf O}}_1 \, {{\mathsf B}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}} \,,
\end{aligned}$$ where ${{\mathsf B}}= {{\mathsf P}}{{\mathsf R}}$ is the braiding operator.
Therefore, invoking Proposition \[TBCC\], we conclude that ${{\mathsf O}}_1$ and ${{\mathsf O}}_3$ also satisfy Proposition \[SH\] with coefficients $A_{r;s}^{\nu;s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(k)$, $r=1,2,3$ being related by $$\label{al23}
A_{2;s}^{\nu;s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(k) = \frac{\Omega^{s{{+}}ib\nu}_{s_{{\mathfrak 1}}s_{{\mathfrak 2}}}}{\Omega^{\;s}_{s_{{\mathfrak 1}}s_{{\mathfrak 2}}}}
A_{1;s}^{\nu;s_{{\mathfrak 1}}s_{{\mathfrak 2}}}(k)\,, \qquad
A_{3;s}^{\nu;s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(k)= \frac{\Omega^{\;s}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}}{\Omega^{s{{+}}ib\nu}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}}
A_{1;s}^{\nu;s_{{\mathfrak 1}}s_{{\mathfrak 2}}}(k) \,,
$$ respectively. The proof of Proposition \[SH\] is complete. $\square$\
[*Proof of Lemma \[idlem\]*]{}. Our starting point is the defining relation for the b–Racah–Wigner symbols, equation [(\[RWdef\])]{}. Our claim will follow from [(\[RWdef\])]{} as an identity satisfied by the residues of the meromorphic continuation of [(\[RWdef\])]{}. We need to analyze the relevant limits step by step.\
\
Note that in the limit $U_1\equiv {\alpha}_s+{\alpha}_0-{\alpha}_1{\to}-b$ the contour of integration in the definition of $\Phi^{}_4$, eqn. [(\[Phidef\])]{}, gets pinched between the poles of the integrand $s=Q-V_4\equiv 0$ and $s=-U_1-b$ as well as between $s=Q-V_4+b\equiv b$ and $s=-U_1$. This implies that $\Phi^{}_4$ has a pole when $U_1=-b$. In order extract the part which gets singular in the limit under consideration one may deform the contour of integration in [(\[Phidef\])]{} to the sum of two circles around $s=0$ and $s=b$ plus a contour which passes to the right of the pole at $s=b$ and which approaches the imaginary axis at infinity. The residue is given as $$-\frac{1}{2\pi}\sin\pi b^2
\frac{S_b(U_2)S_b(U_3)S_b(U_4)}{S_b(V_1)S_b(V_2)S_b(V_3)}
\left(1+\frac{\sin\pi bU_2\sin\pi bU_3\sin\pi bU_4}
{\sin\pi bV_1\sin\pi bV_2\sin\pi bV_3}
\right).$$ Considering the behavior of $\{\cdots\}$ at $U_1=-b$, one finds that the pole of $\Phi^{}_4$ at $U_1=-b$ is canceled by the zero from the prefactor $(S_b({\alpha}_s+{\alpha}_0-{\alpha}_1))^{-1}$. Taken together one obtains the following special value for $\{\cdots\}$ at ${\alpha}_s+{\alpha}_0-{\alpha}_1=-b:$ $$\begin{aligned}
\big\{\begin{smallmatrix}
{\alpha}_0 & {\alpha}_1\\
{\alpha}_2 & {\alpha}_3
\end{smallmatrix}\big|\begin{smallmatrix}{\alpha}_2-{\alpha}_1-b \\{\alpha}_t
\end{smallmatrix}
\big\}=&\;\frac{S_b(Q-2{\alpha}_0-b)}{S_b(2({\alpha}_1-{\alpha}_0-b))}
\frac{S_b({\alpha}_t+{\alpha}_2-{\alpha}_1)}{S_b({\alpha}_3+{\alpha}_2-{\alpha}_1+{\alpha}_0+b)}
{\nonumber}\\
&\;\frac{S_b({\alpha}_3+{\alpha}_t-{\alpha}_0)}{S_b({\alpha}_3+{\alpha}_t-{\alpha}_0-b)}
\frac{S_b({\alpha}_3+{\alpha}_0-{\alpha}_t)}{S_b(Q+{\alpha}_3-{\alpha}_t-{\alpha}_0-b)}
\label{spRW}
\\ &\;
S_b({\alpha}_3+{\alpha}_0+{\alpha}_t-Q)
\left(1+\frac{\sin\pi bU_2\sin\pi bU_3\sin\pi bU_4}
{\sin\pi bV_1\sin\pi bV_2\sin\pi bV_3} \right).{\nonumber}\end{aligned}$$ The parameters $U_2,U_3,U_4$ and $V_1,V_2,V_3$ are now given by $$\begin{aligned}
U_2&=Q-2{\alpha}_0-b\\
U_3&={\alpha}_1-{\alpha}_0-b+{\alpha}_3+{\alpha}_2-Q\\
U_4&={\alpha}_1-{\alpha}_0-b+{\alpha}_3-{\alpha}_2
\end{aligned}
\qquad
\begin{aligned}
V_1&=Q-{\alpha}_t+{\alpha}_3-{\alpha}_0-b\\
V_2&={\alpha}_t+{\alpha}_3-{\alpha}_0-b\\
V_3&=2({\alpha}_1-{\alpha}_0-b).\\
\end{aligned}$$\
In the same way as in the previous paragraph one may show that the b–Clebsch–Gordan coefficients $\big[\begin{smallmatrix}
{\alpha}_s\\
k_s\end{smallmatrix}\big|
\begin{smallmatrix}
{\alpha}_1 & {\alpha}_0\\
k_1 & k_0
\end{smallmatrix}\big]
$ and $\big[\begin{smallmatrix}
{\alpha}_3 \\
k_3\end{smallmatrix}\big|
\begin{smallmatrix}{\alpha}_t & {\alpha}_0 \\ k_t & k_0\end{smallmatrix}
\big]$ develop poles, with residues given by [(\[CGres\])]{}.\
\
The left hand side of [(\[RWdef\])]{} is analytic in the range under consideration. In order to describe the analytic continuation of the right hand side let us note that in the continuation from ${\rm Re}({\alpha}_0)=Q/2$ to ${\rm Re}({\alpha}_0)=-b$ exactly three poles ${\alpha}_t={\alpha}_t^{(k)}$, $k=-1,0,1$ cross the contour of integration, namely $$\begin{aligned} k=-1: \qquad & {\alpha}_t^{(k)}={\alpha}_3+{\alpha}_0,\\
k=0: \qquad & {\alpha}_t^{(k)}={\alpha}_3,\\
k=1: \qquad & {\alpha}_t^{(k)}={\alpha}_3-{\alpha}_0.
\end{aligned}$$ The analytic continuation of the right hand side of [(\[RWdef\])]{} may therefore be represented by replacing the integration contour ${{\mathbb S}}$ in [(\[RWdef\])]{} by ${{\mathcal C}}={{\mathbb S}}\cup\bigcup_{k=-1}^1{{\mathcal C}}_{k}$, with ${{\mathcal C}}_t$ being a small circle around the poles at ${\alpha}_t={\alpha}_t^{(k)}$.\
\
We observe that the integral over ${{\mathbb S}}$ vanishes due to the factor $S_b(Q-2{\alpha}_1-b)$. This is not the case for contributions from the poles ${\alpha}_t={\alpha}_t^{(k)}$, $k=-1,0,1$. Our claim now follows by straightforward computations. $\square$
Construction of the Q–operator ${{\mathsf Q}}(u)$
=================================================
[\[PQT\]]{}
Preliminaries {#Qprel}
-------------
Let us now enter into the construction of the ${{\mathsf Q}}$-operators. We begin by collecting some useful preliminaries. We will work in the Schr[ö]{}dinger representation where the operators ${{\mathsf x}}_r$, $r=1,\ldots,{{\rm N}}$ are diagonal. We will need operators ${{\mathsf U}}$, ${{\mathsf \Omega}}$, and ${{\mathsf J}}_s$ defined in the Schr[ö]{}dinger representation by the following integral kernels $$\begin{aligned}
\label{wker}
U({\bf x},{\bf x'}) &=
\prod_{r=1}^{{\rm N}}\, \delta(x_{r+1} -x'_{r}) \,, \qquad
\Omega({\bf x},{\bf x'}) =
\prod_{r=1}^{{\rm N}}\, \delta(x_{r} + x'_{r}) \,, \\
\label{jker}
{}& J_s({\bf x},{\bf x'}) = \bigl(w_b(i{{\textstyle \frac{Q}{2} }} - 2s)\bigr)^{{\rm N}}\,
\prod_{r=1}^{{\rm N}}\, D_{s-\frac{i}{2}Q}(x_r -x'_r) \,.\end{aligned}$$ ${{\mathsf U}}$ is the cyclic shift operator defined in [(\[upar\])]{}. ${{\mathsf \Omega}}$ and ${{\mathsf J}}_s$ are products of local operators, $$\label{Wwj}
{{\mathsf \Omega}}= \prod_{r=1}^{{\rm N}}{{\scriptstyle{\mathsf\Omega}}}_r \,, \qquad
{{\mathsf J}}_s = \prod_{r=1}^{{\rm N}}{{\mathsf j}}_r \,.$$ Here ${{\mathsf j}}_r$ is the operator which intertwines at the site $r$ the representations ${\cal P}_s$ and ${\cal P}_{-s}$ of ${{\cal U}_q({{\mathfrak s}{\mathfrak l}}(2,{{\mathbb R}}))}$ (see [@PT2]), and ${{\scriptstyle{\mathsf\Omega}}}_r$ is the operator which realizes at the site $r$ the parity operation: ${{\scriptstyle{\mathsf\Omega}}}_r \, f(x_{{\mathfrak 1}},x_{{\mathfrak 2}},\ldots,x_r,\ldots,x_{{\rm N}}) =
f(x_{{\mathfrak 1}},x_{{\mathfrak 2}},\ldots,-x_r,\ldots,x_{{\rm N}})$ (whence ${{\scriptstyle{\mathsf\Omega}}}_r \, {{\mathsf p}}_r {{\scriptstyle{\mathsf\Omega}}}_r = \theta({{\mathsf p}}_r)$ and ${{\scriptstyle{\mathsf\Omega}}}_r \, {{\mathsf x}}_r {{\scriptstyle{\mathsf\Omega}}}_r = \theta({{\mathsf x}}_r)$, where the automorphism $\theta$ is defined by (\[theta\])).
We will denote the standard scalar product on $L^2({{\mathbb R}})$ by ${\langle}f|g {\rangle}=
\int_{\mathbb R} d{\bf x} \, f({\bf x}) \,g({\bf x})$. For a given operator ${{\mathsf O}}$, its transposed ${{\mathsf O}}^t$ and hermitian–conjugated ${{\mathsf O}}^*$ are defined, respectively, by (the bar denotes complex conjugation) $$\label{trdef}
{\langle}{{\mathsf O}}^t f|g {\rangle}={\langle}f| {{\mathsf O}}g {\rangle}\,, \qquad
{\langle}\overline{{{\mathsf O}}^* f} |g {\rangle}={\langle}\overline{f} | {{\mathsf O}}g {\rangle}\,.$$ This definition extends to a matrix with operator–valued coefficients as follows $$\label{Ltr}
\bigl( L^t \bigr)_{ij} = \bigl( L_{ij} \bigr)^t \,, \qquad
\bigl( L^* \bigr)_{ij} = \bigl( L_{ij} \bigr)^* \,,$$ i.e., component–wise. If ${{\mathsf O}}$ is represented by the integral kernel $O({\bf x},{\bf x'})$, then the kernels of its transposed and hermitian–conjugated are given by $$\label{Ot}
O^t({\bf x},{\bf x'}) = O({\bf x'},{\bf x}) \,, \qquad
O^*({\bf x},{\bf x'}) = \overline{O({\bf x'},{\bf x})} \,.$$ In particular, we have $$\begin{aligned}
\label{uw*}
{{\mathsf U}}^t = &\, {{\mathsf U}}^* = {{\mathsf U}}^{-1} \,, \qquad
{{\mathsf \Omega}}^t = {{\mathsf \Omega}}^* = {{\mathsf \Omega}}\,, \\
\label{j*}
{} {{\mathsf J}}_s^t &= {{\mathsf J}}_s \,, \qquad\qquad
{{\mathsf J}}_{s}^* = {{\mathsf J}}_{-s} = {{\mathsf J}}_{s}^{-1} \,.\end{aligned}$$
In the Schr[ö]{}dinger representation we have ${{\mathsf x}}^t = {{\mathsf x}}$, ${{\mathsf p}}^t = -{{\mathsf p}}$, ${{\mathsf x}}^* = {{\mathsf x}}$, ${{\mathsf p}}^* = {{\mathsf p}}$ and hence (as seen from (\[EFK1\])): $$\label{bb0}
{{\mathsf e}}^t_s = {{\mathsf e}}_{-s} \,, \qquad
{{\mathsf f}}^t_s = {{\mathsf f}}_{-s} \,, \qquad
{{\mathsf k}}^t_s = {{\mathsf k}}^{-1}_{-s} \,,$$
Properties of the transfer–matrices of models in question with respect to the transposition and hermitian–conjugation are described by the following statement.
\[Tt\] For the operations defined by (\[trdef\]) and (\[Ltr\]) we have $$\begin{aligned}
\label{Ttr}
\bigl( {{\mathsf T}}^{{\rm{\scriptscriptstyle}XXZ}}_s(u) \bigr)^t &= (-1)^{{\rm N}}\, {{\mathsf T}}^{{\rm{\scriptscriptstyle}XXZ}}_{-s}(-u) \,,
{}& \bigl( {{\mathsf T}}^{{\rm{\scriptscriptstyle}SG}}_s(u) \bigr)^t &= {{\mathsf T}}^{{\rm{\scriptscriptstyle}SG}}_{-s}(-u) \, \\
\label{T*}
\bigl( {{\mathsf T}}^{{\rm{\scriptscriptstyle}XXZ}}_s(u) \bigr)^* &= {{\mathsf T}}^{{\rm{\scriptscriptstyle}XXZ}}_s(\bar{u}) \,,
{}& \bigl( {{\mathsf T}}^{{\rm{\scriptscriptstyle}SG}}_s(u) \bigr)^* &= {{\mathsf T}}^{{\rm{\scriptscriptstyle}SG}}_s(\bar{u}) \,,\end{aligned}$$ where ${{\mathsf T}}_{-s}(u) \equiv {{\mathsf J}}_{s} \, {{\mathsf T}}_s(u) \, {{\mathsf J}}_{s}^{-1}$.
Taking into account (\[bb0\]), we observe that the L–matrices (\[Lxxz\]) and (\[lSG\]) satisfy $$\label{bb1}
\begin{aligned}
\bigl( L_s^{{\rm{\scriptscriptstyle}XXZ}}(u)\bigr)^t &= - \sigma_3 \, e^{\pi b u \sigma_3} \,
L_{-s}^{{\rm{\scriptscriptstyle}XXZ}}(-u) \, e^{-\pi b u \sigma_3} \, \sigma_3 \,, \\
\bigl( L_s^{{\rm{\scriptscriptstyle}SG}}(u) \bigr)^t &= \sigma_3 \, L_{-s}^{{\rm{\scriptscriptstyle}SG}}(-u) \, \sigma_3 \,,
\end{aligned}$$ Substitution of these relation into $$\label{bb3}
{{\mathsf T}}^t(u) = {\rm tr}\, \bigl(
(L^t_1)^{{{\scriptscriptstyle}T}}\cdot (L^t_2)^{{{\scriptscriptstyle}T}}\cdot \ldots
\cdot (L^t_{{\rm N}})^{{{\scriptscriptstyle}T}}\bigr) =
{\rm tr}\, \bigl( L^t_{{\rm N}}\cdot \ldots
\cdot L^t_2 \cdot L^t_1 \bigr)$$ yields (\[Ttr\]). Relations (\[T\*\]) are derived analogously by noticing that we have $\bigl(L(u)\bigr)^* = \sigma_3 \, L(\bar{u}) \, \sigma_3$ for both models in question.
A consequence of this Lemma is that it suffices to prove Theorem \[QXXZSG\] only for ${{\mathsf Q}}^{\,\flat}_+(u)$. Indeed, using (\[Ot\]) and (\[Dcc\]), it is easy to conclude that $$\label{xqpm}
{{\mathsf Q}}^{\,\flat}_{-}(u) = \bigl( D_{-s}(u) \bigr)^{{\rm N}}\,
\bigl( {{\mathsf Q}}^{\,\flat}_{+}(\bar{u}) \bigr)^* \,.
$$ Therefore relations – for ${{\mathsf Q}}^{\,\flat}_-(u)$ then follow immediately if we take (\[T\*\]) into account. To check the Baxter equation for ${{\mathsf Q}}^{\,\flat}_-(u)$, we take hermitian–conjugation of for ${{\mathsf Q}}^{\,\flat}_+(u)$, using and the property (\[Ttr\]). After replacement of $\bar{u}$ by $u$ this yields for $\tilde{{{\mathsf Q}}}^{\,\flat}_{-}(u) \equiv
\bigl( {{\mathsf Q}}^{\,\flat}_{+}(\bar{u}) \bigr)^*$ the following equation $$\label{baxtil}
{{\mathsf T}}^{\,\flat}(u) \cdot \tilde{{{\mathsf Q}}}^{\,\flat}_{-}(u) =
\bigl( \overline{a( \bar{u})} \bigr )^{{\rm N}}\,
\tilde{{{\mathsf Q}}}^{\,\flat}_{-}(u + ib)
+ \bigl( \overline{d( \bar{u})} \bigr)^{{\rm N}}\,
\tilde{{{\mathsf Q}}}^{\,\flat}_{-}(u - ib ) \,.$$ Using relations (\[Dfunrel\]) and (\[Drefl\]), we observe that $$\label{ads}
\overline{a( \bar{u})} =
\frac{D_{-s}(u+ib)}{D_{-s}(u)} \, d(u) \,, \qquad
\overline{d( \bar{u})} = \frac{D_{-s}(u-ib)}{D_{-s}(u)} \, a(u) \,.$$ Whence we conclude that ${{\mathsf Q}}^{\,\flat}_-(u)$ defined by (\[xqpm\]) satisfies (with the same coefficients $a(u)$, $d(u)$ as ${{\mathsf Q}}^{\,\flat}_+(u)$ does).
Construction of Q–operators {#QYZ}
---------------------------
In order to construct the Q–operators explicitly let us consider the following general ansatz for the Q–operator: $$\label{qyz}
{{\mathsf Q}}(u) = {{\mathsf Y}}(u) \cdot {{\mathsf Z}}\,.$$
We will prove Theorem \[QXXZSG\] for ${{\mathsf Q}}^{\,\flat}_+(u)$ in three steps: first constructing a suitable solution for ${{\mathsf Y}}^{\,\flat}(u)$ by requiring the Baxter equation to hold, then determining the form of ${{\mathsf Z}}^{\,\flat}$, and finally checking that the obtained Q–operator satisfies (\[defQ1\]). The first step in this proof is based on the idea to find such a gauge transformation of the L–matrix that it becomes effectively upper–triangular. This approach was originally applied by Pasquier and Gaudin [@PG] to the Toda chain. Our computation has many similarities with the modification of this approach developed in [@De; @DKM] for the non–compact XXX magnet.
\[Yxxz\] Let ${{\mathsf T}}^\flat(u)$, $\flat= {{\rm{\scriptscriptstyle}XXZ}},\, {{\rm{\scriptscriptstyle}SG}}$ be the transfer–matrices corresponding to the L–matrices (\[Lxxz\]) and (\[lSG\]). Let ${{\mathsf Y}}^\flat(u)$ be defined in the Schr[ö]{}dinger representation by the kernel $$\label{y1}
Y_u({\bf x},{\bf x'}) = \prod_{r=1}^{{\rm N}}D_{\frac{1}{2}(u-\sigma)}(x_r - {\varepsilon}_\flat\, x'_{r+1}) \,
D_{-\frac{1}{2}(u+\sigma)}(x_r - x'_{r}) \,,$$ where ${\varepsilon}_{{\rm{\scriptscriptstyle}XXZ}}= 1$, ${\varepsilon}_{{\rm{\scriptscriptstyle}SG}}= -1$. Then ${{\mathsf Q}}^{\flat}(u)$ of the form (\[qyz\]) satisfies the Baxter equation with coefficients $a(u)$, $d(u)$ as specified in eq. (\[baxxz\]) of Theorem \[QXXZSG\].
Let us introduce the gauge–transformed Lax operators (the transformation depends on the site number $r$): $$\begin{aligned}
\label{LGxxz}
\widetilde{L}^\flat_r(u) &=
G^\flat_{r+1} \cdot L^\flat_r(u)
\cdot \bigl( G^\flat_{r} \bigr)^{-1} \,, \qquad
G^\flat_{r} = \bigl(\begin{smallmatrix} 1 & 0\\ \rho^\flat_r & 1
\end{smallmatrix} \bigr) \,, \\
\rho^{{\rm{\scriptscriptstyle}XXZ}}_r &= e^{\pi b (2 x'_r - u)} \,, \qquad
\rho^{{\rm{\scriptscriptstyle}SG}}_r = e^{2 \pi b x'_r } \,.\end{aligned}$$ The relevant matrix elements of the new Lax matrices are given by $$\begin{aligned}
\Bigl( \widetilde{L}^\flat_r (u) \Bigr)_{21} =& \,
4 \varkappa_\flat\, (\rho_r \rho_{r+1})^{\frac 12} \label{L21}\\
& \times\Bigl(
\cosh \pi b ({{\mathsf x}}_r - {\varepsilon}_\flat\, x'_{r+1} + {{\textstyle \frac{1}{2} }}(\sigma -u)) \,
\cosh \pi b ({{\mathsf x}}_r - x'_{r} + {{\textstyle \frac{1}{2} }}(\sigma +u)) \,{{\mathsf k}}_r {\nonumber}\\
& - \cosh \pi b ({{\mathsf x}}_r - {\varepsilon}_\flat\, x'_{r+1} - {{\textstyle \frac{1}{2} }}(\sigma -u)) \,
\cosh \pi b ({{\mathsf x}}_r - x'_{r} - {{\textstyle \frac{1}{2} }}(\sigma +u))
\,{{\mathsf k}}^{-1}_r \Bigr) \,, {\nonumber}\\
\nonumber
\Bigl( \widetilde{L}^\flat_r (u) \Bigr)_{11} =&\,
2 \varkappa_\flat\, e^{\pi b (x'_r -{\varepsilon}_\flat\, {{\mathsf x}}_r)} \, \Bigl(
e^{\frac{1}{2} {\varepsilon}_\flat\, \pi b (u-\sigma)} \,
\cosh \pi b ({{\mathsf x}}_r - x'_{r} + {{\textstyle \frac{1}{2} }}(\sigma +u)) \,{{\mathsf k}}_r \\
\label{L11}
& -\, e^{\frac{1}{2} {\varepsilon}_\flat\, \pi b (\sigma -u)} \,
\cosh \pi b ({{\mathsf x}}_r - x'_{r} - {{\textstyle \frac{1}{2} }}(\sigma +u))
\,{{\mathsf k}}^{-1}_r \Bigr) \,,\\
\nonumber
\Bigl( \widetilde{L}^\flat_r (u) \Bigr)_{22} =&\,
2 \varkappa_\flat\, e^{\pi b (x'_{r+1} -{{\mathsf x}}_r)} \, \Bigl(
e^{\frac{1}{2} \pi b (\sigma +u)} \,
\cosh \pi b ({{\mathsf x}}_r - {\varepsilon}_\flat\, x'_{r+1} +
{{\textstyle \frac{1}{2} }}(u-\sigma)) \,{{\mathsf k}}^{-1}_r \\
\label{L22}
& -\, e^{-\frac{1}{2} \pi b (u+\sigma)} \,
\cosh \pi b ({{\mathsf x}}_r - {\varepsilon}_\flat\, x'_{r+1} + {{\textstyle \frac{1}{2} }}(\sigma -u))
\, {{\mathsf k}}_r \Bigr) \,,\end{aligned}$$ where $\varkappa_{{\rm{\scriptscriptstyle}XXZ}}=1$, $\varkappa_{{\rm{\scriptscriptstyle}SG}}= -i e^{-\pi b s}$.
In the Schrödinger representation, operators ${{\mathsf k}}_r$, ${{\mathsf k}}^{-1}_r$ act as shifts of $x_r$ by ${\pm}{{\textstyle \frac{i}{2} }}b$. Using the functional relation (\[Dfunrel\]), it is straightforward to apply (\[L21\]) to $Y^\flat_u({\bf x},{\bf x'})$ and verify that the condition $$\label{l21a}
\bigl( \widetilde{L}^\flat_r (u) \bigr)_{21} \,
Y^\flat_u({\bf x},{\bf x'}) = 0$$ is satisfied for all ${\bf x'}\in {\mathbb R}^{{\rm N}}$. This implies that $\widetilde{L}^\flat_r(u)$ becomes upper triangular when acting on $Y^\flat_u({\bf x},{\bf x'})$ so that we can calculate the action of $\widetilde{{{\mathsf M}}}^{\flat}(u)$ on $Y^\flat_u({\bf x},{\bf x'})$ as $$\label{MY1}
\widetilde{{{\mathsf M}}}^{\flat}(u) \, Y^\flat_u({\bf x},{\bf x'}) =
\left( \begin{matrix} \prod_{r=1}^{{\rm N}}\bigl( \widetilde{L}^\flat_r (u) \bigr)_{11} & * \\
0 & \prod_{r=1}^{{\rm N}}\bigl( \widetilde{L}^\flat_r (u) \bigr)_{22} \end{matrix} \right) \,
Y^\flat_u({\bf x},{\bf x'}) \,.$$ Hence, taking into account the periodicity condition, $G^\flat_{{{\rm N}}+1}=G^\flat_1$, we have $$\label{TY1}
{{\mathsf T}}^{\flat}(u) \, Y^\flat_u({\bf x},{\bf x'}) \,=\,
\Bigg(\, \prod_{r=1}^{{\rm N}}\bigl( \widetilde{L}^\flat_r (u) \bigr)_{11} +
\prod_{r=1}^{{\rm N}}\bigl( \widetilde{L}^\flat_r (u) \bigr)_{22} \,\Bigg) \,
Y^\flat_u({\bf x},{\bf x'}) \,.$$ Applying (\[L11\]), (\[L22\]) to $Y^\flat_u({\bf x},{\bf x'})$ and using (\[l21a\]), we derive $$\label{l11a}
\begin{aligned}
\bigl( \widetilde{L}^\flat_r (u) \bigr)_{11} \,
Y^\flat_u({\bf x},{\bf x'})
=&\ 2 {\varepsilon}_\flat\, \varkappa_\flat\, e^{\pi b (x'_r - x'_{r+1})} \,
\sinh \pi b(u-\sigma) \, Y^\flat_{u-ib}({\bf x},{\bf x'})\,, \\
\bigl( \widetilde{L}^\flat_r (u) \bigr)_{22} \,
Y^\flat_u({\bf x},{\bf x'})
=&\ 2 \varkappa_\flat\, e^{\pi b (x'_{r+1}-x'_r)} \,
\sinh \pi b(u+\sigma) \, Y^\flat_{u+ib}({\bf x},{\bf x'})\,.
\end{aligned}$$ Here we have used that $Y^\flat_u({\bf x},{\bf x'}) = \prod_{r=1}^{{\rm N}}Y^\flat(u,x_r)$, where each factor satisfies the relation $$\label{yy}
\frac{Y^\flat(u,x_r {\pm}i \frac{b}{2})}{Y^\flat(u {\pm}ib,x_r)}
= \Bigl( \frac{\cosh \pi b(x_r -x'_r - \frac{1}{2}(\sigma +u))} {\cosh \pi b(x_r - {\varepsilon}_\flat\, x'_{r+1} + \frac{1}{2}(\sigma -u))}
\Bigr)^{{\pm}1} \,.$$ Combining (\[TY1\]) with (\[l11a\]), we obtain $$\label{ty1}
\begin{aligned}
{{\mathsf T}}^{\flat}(u) \, Y^\flat_u({\bf x},{\bf x'}) =\;
&\bigl(2 \varkappa_\flat\, \sinh \pi b (u+\sigma)\bigr)^{{\rm N}}\,
Y^\flat_{u + ib}({\bf x},{\bf x'}) \\& +
\bigl( 2 {\varepsilon}_\flat\, \varkappa_\flat\,
\sinh \pi b (u-\sigma)\bigr)^{{\rm N}}\,
Y^\flat_{u - ib}({\bf x},{\bf x'}) \,,
\end{aligned}$$ which implies that the Baxter equation holds with the coefficients $a(u)$, $d(u)$ as specified in eq. (\[baxxz\]).
The possible form of ${{\mathsf Z}}$ can be found from the requirement that holds.
\[ZXXZ\] Let ${{\mathsf Y}}^\flat(u)$ be chosen as in Proposition \[Yxxz\]. Then the commutativity condition $$\label{qt1}
{{\mathsf Q}}^{\flat}(u) \, {{\mathsf T}}^{\flat}(u) = {{\mathsf T}}^{\flat}(u) \, {{\mathsf Q}}^{\flat}(u)$$ holds for $ {{\mathsf Q}}^{\flat}(u)$ of the form (\[qyz\]) provided that the corresponding operator ${{\mathsf Z}}^\flat$ satisfies the following relation $$\label{zz}
{{\mathsf Z}}^\flat \, {{\mathsf T}}_s^{\flat} (u) =
{{\mathsf T}}_{-s}^{\flat}(u) \, {{\mathsf Z}}^\flat \,.$$
In order to treat both models in a uniform way, let us introduce the operator $$\label{Wb}
{{\mathsf \Omega}}^\flat = \begin{cases} {\sf 1}\,, & \flat = {{\rm{\scriptscriptstyle}XXZ}}\\
{{\mathsf \Omega}}\,, & \flat = {{\rm{\scriptscriptstyle}SG}}\end{cases}\,,$$ where the parity operator was defined in (\[wker\]). Then, using the explicit expressions (\[y1\]) and taking (\[Dpar\]) into account, it is easy to verify that (the subscript $\bf x$ or $\bf x'$ of an operator specifies the argument on which it acts) $$\label{bb7}
{{\mathsf \Omega}}^\flat_{\bf x} \, Y^\flat_u({\bf x},{\bf x'}) =
{{\mathsf \Omega}}^\flat_{\bf x'} \, Y^\flat_u({\bf x},{\bf x'}) =
{{\mathsf U}}_{\bf x}^{-1} \, Y^\flat_{-u}({\bf x'},{\bf x}) \,.$$ Now we derive $$\begin{aligned}
\nonumber
{} & {{\mathsf T}}_{s;\bf x}^{\,\flat}(u) \, Y^\flat_u({\bf x},{\bf x'})
{\stackrel{(\ref{ty1})}{=}} \bigl( a(u) \bigr)^{{\rm N}}\,
Y^\flat_{u - ib}({\bf x},{\bf x'}) +
\bigl( d(u) \bigr)^{{\rm N}}\,
Y^\flat_{u + ib}({\bf x},{\bf x'}) \\
{} & {\stackrel{(\ref{bb7})}{=}} (-{\varepsilon}_\flat)^{{\rm N}}\, {{\mathsf \Omega}}^\flat_{\bf x} \,
{{\mathsf U}}_{\bf x}^{-1} \, \Bigl( \bigl( d(-u) \bigr)^{{\rm N}}\,
Y^\flat_{-u + ib}({\bf x'},{\bf x})
+ \bigl( a(-u) \bigr)^{{\rm N}}\,
Y^\flat_{-u - ib}({\bf x'},{\bf x}) \Bigr) {\nonumber}\\
{} & {\stackrel{(\ref{ty1})}{=}} (-{\varepsilon}_\flat)^{{\rm N}}\, {{\mathsf \Omega}}^\flat_{\bf x} \,{{\mathsf U}}_{\bf x}^{-1} \,
{{\mathsf T}}_{s;\bf x'}^{\,\flat}(-u) \, Y^\flat_{-u}({\bf x'},{\bf x})
{\stackrel{(\ref{bb7})}{=}} (-{\varepsilon}_\flat)^{{\rm N}}\, {{\mathsf T}}_{s;\bf x'}^{\,\flat}(-u) \,
Y^\flat_u({\bf x},{\bf x'}) \\
{} & = (-{\varepsilon}_\flat)^{{\rm N}}\, Y^\flat_u({\bf x},{\bf x'}) \,
\bigl( {{\mathsf T}}_{s;\bf x'}^{{{\rm{\scriptscriptstyle}SG}}}(-u) \bigr)^t
{\stackrel{(\ref{Ttr})}{=}} Y^\flat_u({\bf x},{\bf x'}) \, {{\mathsf T}}_{-s;\bf x'}^{\,\flat}(u) \,.
\end{aligned}$$ Thus, we verified that ${{\mathsf T}}_{s}^{\flat}(u) \, {{\mathsf Y}}(u) =
{{\mathsf Y}}(u) \, {{\mathsf T}}_{-s}^{\flat}(u) $, which is equivalent to (\[qt1\]) if relation (\[zz\]) is satisfied.
Proposition \[ZXXZ\] implies that we can choose $$\label{zzj}
{{\mathsf Z}}^{\flat} = {{\mathsf J}}_s \,,$$ where ${{\mathsf J}}_s$ was defined in (\[jker\]). Then we compute the integral kernel of ${{\mathsf Q}}^{\flat}_+(u)$: $$\begin{aligned}
\label{qker1}
& Q_{+;u}^{\,\flat} ({\bf x},{\bf x'}) =
\bigl(w_b(i{{\textstyle \frac{Q}{2} }} - 2s) \bigr)^{{\rm N}}\int\limits_{{{\mathbb R}}^{{{\rm N}}}}dz_1\dots dz_{{\rm N}}\, \prod_{r=1}^{{\rm N}}\,
D_{\frac{u-\sigma}{2}}(z_{r} - {\varepsilon}_\flat\, x_{r-1}) \, \\[-1ex]
{} & \hspace{7.3cm}\times
D_{-\frac{u+\sigma}{2}}(z_{r} - x_{r}) \,
D_{\bar\sigma}(z_{r} - x'_{r}) {\nonumber}\\
\label{qker2}
& \quad= \bigl( D_{-s}(u) \bigr)^{{\rm N}}\, \prod_{r=1}^{{\rm N}}D_{\frac{\bar\sigma -u}{2}}(x_{r} - x'_{r}) \,
D_{\frac{\bar\sigma +u}{2}}(x_{r-1} - {\varepsilon}_\flat\, x'_{r}) \,
D_{-s}(x_{r} - {\varepsilon}_\flat\, x_{r-1}) \,.\end{aligned}$$ Equivalence of (\[qker1\]) and (\[qker2\]) is due to the identity (\[D3\]).
The Baxter equation along with the self–commutativity relation, which will be proven below, imply that [(\[qt1\])]{} extends to commutativity of ${{\mathsf Q}}_+^{\flat}(u)$ with ${{\mathsf T}}^{\flat}(v)$ for those values of $v$, where ${{\mathsf Q}}_+^{\flat}(v)$ is invertible.
Proof of commutativity relations
--------------------------------
To complete the proof of Theorem \[QXXZSG\], we have to establish relations and .
\[YYB\] Let ${{\mathsf Y}}^\flat(u)$ be chosen as in Proposition \[Yxxz\]. Then the following identities hold $$\begin{aligned}
\label{yyb1}
\bigl( {{\mathsf Y}}^\flat(\bar{u}) \bigr)^* \cdot {{\mathsf Y}}^\flat(v) &=
\bigl( D_{s}(u) \, D_{-s}(v) \bigr)^{{\rm N}}\,
\bigl( {{\mathsf Y}}^\flat(\bar{v}) \bigr)^* \cdot {{\mathsf Y}}^\flat(u) \,, \\
\label{yyb2}
{{\mathsf Y}}^\flat(u) \cdot \bigl( {{\mathsf Y}}^\flat(\bar{v}) \bigr)^* &=
\bigl( D_{s}(v) \, D_{-s}(u) \bigr)^{{\rm N}}\,
{{\mathsf Y}}^\flat(v) \cdot \bigl( {{\mathsf Y}}^\flat(\bar{u}) \bigr)^* \,.\end{aligned}$$
These identities are just particular cases of the integral identity (\[D4\]). Indeed, let us denote $\alpha_u \equiv {{\textstyle \frac{1}{2} }}(u-\sigma)$, $\beta_u \equiv -{{\textstyle \frac{1}{2} }}(u+\sigma)$. We will also use the notation $\alpha^\star\equiv -{{\textstyle \frac{i}{2} }}Q -\alpha$. Let us consider the operator ${{\mathsf V}}^\flat(u,v) =
\bigl( {{\mathsf Y}}^\flat(\bar{u}) \bigr)^* \cdot {{\mathsf Y}}^\flat(v)$. Its kernel is given by $$\begin{aligned}
\label{mker3}
V{\,^\flat}_{u,v}({\bf x},{\bf x'})
= \int\limits_{{{\mathbb R}}^{{{\rm N}}}}dz_1\dots dz_{{\rm N}}\; \prod_{r=1}^{{\rm N}}\,& D_{\alpha^\star_u}(z_r - {\varepsilon}_\flat \, x_{r+1}) \,
D_{\beta^\star_u}(z_{r} - x_{r}) \\[-1ex]
\times & D_{\alpha_v}(z_r - {\varepsilon}_\flat \, x'_{r+1}) \,
D_{\beta_v}(z_{r} - x'_r) \,, {\nonumber}\end{aligned}$$ Now we can apply identity (\[D4\]) choosing $\alpha=\alpha^\star_u$, $\beta=\alpha_v$, $\gamma=\beta^\star_u$, $\omega=\beta_v$, and $u= {\varepsilon}_\flat \, x_{r+1}$, $v= {\varepsilon}_\flat \, x'_{r+1}$, $w= x_{r}$, $z=x'_r$. This yields $$\begin{aligned}
\nonumber
V^{\,\flat}_{u,v}({\bf x},{\bf x'})
=&\, \bigl(A(\alpha^\star_u,\alpha_v,\beta^\star_u,\beta_v)\bigr)^{{\rm N}}\
\prod_{r=1}^{{\rm N}}{ D_{\frac{v-u}{2}} \left( {\varepsilon}_\flat ( x_{r+1} - x'_{r+1}) \right) }
{ D_{\frac{u-v}{2}} ( x_{r}-x'_r ) } \\
\nonumber
{} &\times \int\limits_{{{\mathbb R}}^{{{\rm N}}} }dz_1\dots dz_{{\rm N}}{} \prod_{r=1}^{{\rm N}}\,
D_{\alpha_u}(z_r- {\varepsilon}_\flat \, x'_{r+1}) \, D_{\beta_u}(z_{r} - x'_r) \,
\\{} & \hspace{28mm} \times
D_{\alpha^\star_v}(z_r- {\varepsilon}_\flat \, x_{r+1}) \,
D_{\beta^\star_v}(z_{r} - x_{r}) {\nonumber}\\
\label{Wker1}
=&\, \bigl( D_{s}(u) \, D_{-s}(v) \bigr)^{{\rm N}}\
V^{\,\flat}_{v,u}({\bf x},{\bf x'}) \,.\end{aligned}$$ Here we used (\[Dpar\]), the definition (\[Adef\]) of the function $A(\alpha_1,\alpha_2,\ldots)$, and took into account the periodic boundary conditions. Identity (\[yyb2\]) can be proven absolutely analogously.
\[QQXXZ\] The operators ${{\mathsf Q}}_+^{\flat}(u)$ and ${{\mathsf Q}}_-^{\flat}(u)$ with the kernels given in Theorem \[QXXZSG\] by eqs. (\[Qxxz+\]) and (\[Qxxz-\]), respectively, satisfy the following commutativity and exchange relations $$\begin{aligned}
\label{qq3a}
{} {{\mathsf Q}}_+^{\flat}(u) \, {{\mathsf Q}}_+^{\flat}(v) =
{{\mathsf Q}}_+^{\flat}(v) \, {{\mathsf Q}}_+^{\flat}(u) \,, &\qquad
{{\mathsf Q}}_-^{\flat}(u) \, {{\mathsf Q}}_-^{\flat}(v) =
{{\mathsf Q}}_-^{\flat}(v) \, {{\mathsf Q}}_-^{\flat}(u) \,, \\
\label{qq3b}
{} {{\mathsf Q}}_+^{\flat}(u) \, {{\mathsf Q}}_-^{\flat}(v) =
{{\mathsf Q}}_-^{\flat}(u) \, {{\mathsf Q}}_+^{\flat}(v) &=
{{\mathsf Q}}_+^{\flat}(v) \, {{\mathsf Q}}_-^{\flat}(u) =
{{\mathsf Q}}_-^{\flat}(v) \, {{\mathsf Q}}_+^{\flat}(u) \,.\end{aligned}$$ for all $u,v \in {{\mathbb C}}$.
Observe that, using (\[Ot\]) and (\[Dcc\]), the equality of (\[qker1\]) and (\[qker2\]) can be written in the following operator form: $$\label{qqa1}
{{\mathsf Q}}_+^{\flat}(u) =
{{\mathsf Y}}^\flat(u) \cdot {{\mathsf J}}_s = \bigl( D_{-s}(u) \bigr)^{{\rm N}}\,
\bigl( {{\mathsf X}}^\flat_{s})^{-1} \cdot {{\mathsf \Omega}}^\flat \cdot
{{\mathsf U}}^{-1} \cdot \bigl( {{\mathsf Y}}^\flat(\bar{u}) \bigr)^* \,,$$ where we used the notation (\[Wb\]) and introduced $$\label{Xs}
{{\mathsf X}}^\flat_{s} = \prod_{r=1}^{{\rm N}}D_{s}({{\mathsf x}}_r - {\varepsilon}_\flat {{\mathsf x}}_{r-1}) \,.$$ Using Lemma \[YYB\], we can write down the product of two such Q–operators as follows $$\label{qq1}
\begin{aligned}
{}& {{\mathsf Q}}_+^{\flat}(u) \, {{\mathsf Q}}_+^{\flat}(v) =
\bigl(D_{-s}(u) \bigr)^{{\rm N}}\, \bigl({{\mathsf X}}^\flat_{s}\bigr)^{-1}
\cdot {{\mathsf \Omega}}^\flat \cdot {{\mathsf U}}^{-1} \cdot
\bigl( {{\mathsf Y}}^\flat(\bar{u}) \bigr)^* \cdot {{\mathsf Y}}^\flat(v) \cdot {{\mathsf J}}_s \\
{} & {\stackrel{(\ref{yyb1})}{=}} \bigl(D_{-s}(v) \bigr)^{{\rm N}}\,
\bigl({{\mathsf X}}^\flat_{s}\bigr)^{-1} \cdot {{\mathsf \Omega}}^\flat\cdot {{\mathsf U}}^{-1}
\cdot \bigl( {{\mathsf Y}}^\flat(\bar{v}) \bigr)^* \cdot {{\mathsf Y}}^\flat(u)
\cdot {{\mathsf J}}_s = {{\mathsf Q}}_+^{\flat}(v) \, {{\mathsf Q}}_+^{\flat}(u) \,.
\end{aligned}$$ This proves the first relation in (\[qq3a\]) and hence for ${{\mathsf Q}}_+^{\flat}(u)$. As was explained in Subsection \[Qprel\], relation for ${{\mathsf Q}}_-^{\flat}(u)$ (i.e., the second relation in (\[qq3a\])) follows then as a consequence of the relation (\[xqpm\]) between ${{\mathsf Q}}_+^{\flat}(u)$ and ${{\mathsf Q}}_-^{\flat}(u)$. By the same token, relation is equivalent to (\[qq3b\]). To prove the latter relation, we substitute (\[qqa1\]) into (\[xqpm\]) and use (\[uw\*\])–(\[j\*\]). This yields the operator ${{\mathsf Q}}^\flat_-(u)$ in the following form: $$\label{qqa3x}
{{\mathsf Q}}_-^{\flat}(u) = \bigl( D_{-s}(u) \bigr)^{{\rm N}}\,
{{\mathsf J}}_{s}^{-1} \cdot \bigl( {{\mathsf Y}}^\flat(\bar{u}) \bigr)^* =
{{\mathsf Y}}^\flat(u) \cdot {{\mathsf U}}\cdot {{\mathsf \Omega}}^\flat \cdot {{\mathsf X}}^\flat_s \,.$$ As seen from [(\[qqa1\])]{} and [(\[qqa3x\])]{}, the two expressions on the l.h.s. of [(\[qq3b\])]{} are just two ways to write down $\bigl( D_{-s}(v) \bigr)^{{\rm N}}\,
{{\mathsf Y}}^\flat(u) \cdot \bigl( {{\mathsf Y}}^\flat(\bar{v}) \bigr)^*$. Analogously, the two expressions on the r.h.s. of [(\[qq3b\])]{} are two ways to write down $\bigl( D_{-s}(u) \bigr)^{{\rm N}}\,
{{\mathsf Y}}^\flat(v) \cdot \bigl( {{\mathsf Y}}^\flat(\bar{u}) \bigr)^*$. The middle equality in [(\[qq3b\])]{} is due to the identity (\[yyb2\]) in Lemma \[YYB\].
[11]{}
E.W. Barnes: [*Theory of the double gamma function*]{}, Phil. Trans. Roy. Soc. [**A196**]{} (1901) 265–388
V. Brazhnikov and S. Lukyanov: [*Angular quantization and form factors in massive integrable models*]{} [[ Nucl. Phys. ]{}[**B512**]{} (1998) 616–636]{}
A.G. Bytsko and A. Doikou: [*Thermodynamics and conformal properties of XXZ chains with alternating spins*]{}, [J. Phys.]{} [**A37**]{} (2004) 4465–4492
A. Bytsko and J. Teschner: [*R–operator, co–product and Haar–measure for the modular double of ${{\cal U}_q({{\mathfrak s}{\mathfrak l}}(2,{{\mathbb R}}))}$*]{}, [[ Commun. Math. Phys. ]{}[**240**]{} (2003) 171–196]{}
S.E. Derkachov: [*Baxter’s Q–operator for the homogeneous XXX spin chain*]{}, J. Phys. [**A32**]{} (1999) 5299–5316
S.E. Derkachov, D. Karakhanyan, and R. Kirschner: [*Baxter Q–operators of the XXZ chain and R–matrix factorization*]{}, Nucl. Phys. [**B738**]{} (2006) 368–390
S.E. Derkachov, G.P. Korchemsky, and A. N. Manashov: [*Noncompact Heisenberg spin magnets from high–energy QCD. I. Baxter Q–operator and separation of variables*]{}, Nucl. Phys. [**B617**]{} (2001) 375–440; [*Separation of variables for the quantum $SL(2,{{\mathbb R}})$ spin chain*]{}, JHEP [**0307**]{} (2003) 047
L.D. Faddeev: [*How algebraic Bethe ansatz works for integrable model*]{}. In: [*Symétries quantiques*]{} (North-Holland, 1998), 149–219 \[hep-th/9605187\]
L.D. Faddeev: [*Discrete Heisenberg–Weyl group and modular group*]{}, [[ Lett. Math. Phys. ]{}[**34**]{} (1995) 249–254]{}
L.D. Faddeev: [*Modular double of a quantum group*]{}, Math. Phys. Stud. [**21**]{} (2000) 149–156 L.D. Faddeev and R.M. Kashaev: [*Quantum dilogarithm*]{}, Mod. Phys. Lett. [**A9**]{} (1994) 427–434
L.D. Faddeev, R.M. Kashaev, and A.Yu. Volkov: [*Strongly coupled quantum discrete Liouville theory. I: Algebraic approach and duality*]{}, [[ Commun. Math. Phys. ]{}[**219**]{} (2001) 199–219]{}
L.D. Faddeev and G.P. Korchemsky: [*High energy QCD as a completely integrable model*]{}, Phys. Lett. (1995) 311–322
L.D. Faddeev, E.K. Sklyanin, and L.A. Takhtajan: [*Quantum inverse problem method. I*]{}, Theor. Math. Phys. [**40**]{} (1979) [688–706]{}
L.D. Faddeev, V.O. Tarasov, and L.A. Takhtajan: [*Local Hamiltonians for integrable quantum models on a lattice*]{}, [[ Theor. Math. Phys. ]{}[**57**]{} (1983) 1059–1073]{}
L.D. Faddeev and O. Tirkkonen: [*Connections of the Liouville model and XXZ spin chain*]{}, [[ Nucl. Phys. ]{}[**B453**]{} (1995) 647–669]{}
L.D. Faddeev and A.Yu. Volkov: [*Yang–Baxterization of the quantum dilogarithm*]{}, Zapiski nauch. semin. POMI [**224**]{} (1995) 146–154 \[Engl. transl.: J. Math. Sci. [**88**]{} (1998) [202–207]{}\]
A. Fring, G. Mussardo, and P. Simonetti: [*Form–factors for integrable Lagrangian field theories, the sinh–Gordon theory*]{}, Nucl. Phys. [**B393**]{} (1993) 413–441
A.G. Izergin and V.E. Korepin: [*Lattice versions of quantum field theory models in two dimensions*]{}, [[ Nucl. Phys. ]{}[**B205**]{} (1982) 401–413]{}
M. Jimbo: [*A $q$–difference analogue of $U(gl(N+1))$ and the Yang-Baxter equations*]{}, [Lett. Math. Phys.]{} [**10**]{} (1985) 63–69
R.M. Kashaev: [*The non–compact quantum dilogarithm and the Baxter equations*]{}, J. Stat. Phys. [**102**]{} (2001) 923–936
R.M. Kashaev: [*The quantum dilogarithm and Dehn twists in quantum Teichm[ü]{}ller theory*]{}, In: Integrable structures of exactly solvable two–dimensional models of quantum field theory (Kiev, 2000), 211–221 (NATO Sci.Ser.II Math.Phys.Chem., [**35**]{}, Kluwer Acad.Publ., Dordrecht, 2001)
S. Kharchev, D. Lebedev: [*Integral representation for the eigenfunctions of quantum periodic Toda chain*]{}, Lett. Math. Phys. [**50**]{} (1999) 53–77
S. Kharchev, D. Lebedev, and M. Semenov–Tian–Shansky: [*Unitary representations of ${{\cal U}_q({{\mathfrak s}{\mathfrak l}}(2,{{\mathbb R}}))}$, the modular double, and the multiparticle $q$–deformed Toda chains*]{}, [[ Commun. Math. Phys. ]{}[**225**]{} (2002) 573–609]{}
M. Kirch and A.N. Manashov: [*Noncompact $SL(2,{{\mathbb R}})$ spin chain*]{}, JHEP [**0406**]{} (2004) 035
V. Korepin, N. Bogoliubov, and A. Izergin: [*Quantum inverse scattering method and correlation functions*]{} (Cambridge U. Press, 1993)
A. Koubek and G. Mussardo: [*On the operator content of the sinh–Gordon model*]{}, [[ Phys. Lett. ]{}[**B311**]{} (1993) 193–201]{}
G. Lechner: [*An existence proof for interacting quantum field theories with a factorising S–matrix*]{}, math-ph/0601022
L.N. Lipatov: [*High energy asymptotics of multi–colour QCD and exactly solvable lattice models*]{}, JETP Lett. [**59**]{} (1994) 596–599
S. Lukyanov: [*Finite temperature expectation values of local fields in the sinh–Gordon model*]{}, Nucl. Phys. [**B612**]{} (2001) 391–412
V. Pasquier and M. Gaudin: [*The periodic Toda chain and a matrix generalization of the Bessel function*]{}, J. Phys. [**A25**]{} (1992) 5243–5252
B. Ponsot and J. Teschner: [*Liouville bootstrap via harmonic analysis on a non–compact quantum group*]{}, hep-th/9911110
B. Ponsot and J. Teschner: [*Clebsch–Gordan and Racah–Wigner coefficients for a continuous series of representations of ${{\cal U}_q({{\mathfrak s}{\mathfrak l}}(2,{{\mathbb R}}))}$*]{}, [[ Commun. Math. Phys. ]{}[**224**]{} (2001) 613–655]{}
S.N.M. Ruijsenaars: [*First order analytic difference equations and integrable quantum systems*]{}, J. Math. Phys. [**38**]{} (1997) 1069–1146
K. Schmüdgen: [*Integrable operator representations of ${{{\mathbb R}}}^2_q$, $X_{q,\gamma}$ and ${\rm SL}_q(2,{{{\mathbb R}}})$*]{}, [[ Commun. Math. Phys. ]{}[**159**]{} (1994) 217–237]{}
T. Shintani: [*On a Kronecker limit formula for real quadratic fields*]{}, J. Fac. Sci. Univ. Tokyo Sect. 1A Math. [**24**]{} (1977) 167–199
E.K. Sklyanin: [*Exact quantization of the sinh–Gordon model*]{}, [[ Nucl. Phys. ]{}[**B326**]{} (1989) 719–736]{}
E.K. Sklyanin: [*The quantum Toda chain*]{}, Lect. Notes Phys. [**226**]{} (1985) 196–233
E.K. Sklyanin: [*Quantum inverse scattering method. Selected topics*]{}. In: [*Quantum groups and quantum integrable systems*]{} (World Scientific, 1992) 63–97 \[hep-th/9211111\]; [*Separation of variables – new trends*]{}, Prog. Theor. Phys. Suppl. [**118**]{} (1995) 35–60
F.A. Smirnov: [*Quasi–classical study of form factors in finite volume*]{}. In: [*L.D. Faddeev’s seminar on mathematical physics*]{} (AMS Transl. Ser. 2, 201, AMS, Providence, RI, 2000) 283–307 \[hep-th/9802132\]
F.A. Smirnov: [ *Baxter equations and deformation of abelian differentials*]{}, Int. J. Mod. Phys. [**A19S2**]{} (2004) 396–417.
V.O. Tarasov: [*Irreducible monodromy matrices for the R matrix of the XXZ model and local lattice quantum Hamiltonians*]{}, [[ Theor. Math. Phys. ]{}[**63**]{} (1985) 440–454]{}
J. Teschner: [*On structure constants and fusion rules in the $SL(2,{{\mathbb C}})/SU(2)$–WZNW model*]{}, [[ Nucl. Phys. ]{}[**B546**]{} (1999) 390–422]{}; [*Operator product expansion and factorization in the $H_3^+$–WZNW model*]{}, [[ Nucl. Phys. ]{}[**B571**]{} (2000) 555–582]{}; [*Crossing symmetry in the $H_3^+$ WZNW model*]{}, Phys. Lett. [**B521**]{} (2001) 127–132
J. Teschner: [*Liouville theory revisited*]{}, [Class. Quant. Grav.]{} [**18**]{} (2001) R153–R222; [*A lecture on the Liouville vertex operators*]{}, Int. J. Mod. Phys. [**A19S2**]{} (2004) 436–458
E.C. Titchmarsh: [*The theory of functions*]{}, 2nd ed. (Oxford Univ. Press, Oxford, 1975)
S.N. Vergeles and V.M. Gryanik: [*Two–dimensional quantum field theories which admit exact solutions*]{}, Sov. J. Nucl. Phys. [**23**]{} (1976) 704–709
A.Yu. Volkov: [*Quantum Volterra model*]{}, Phys. Lett. [**A167**]{} (1992) [345–355]{}
A.Yu. Volkov: [*Noncommutative hypergeometry*]{}, [[ Commun. Math. Phys. ]{}[**258**]{} (2005) 257–273]{}
S.L. Woronowicz: [*Quantum exponential function*]{}, Rev. Math. Phys. [**12**]{} (2000) 873–920
Al. Zamolodchikov: [*On the thermodynamic Bethe Ansatz equation in sinh–Gordon model*]{}, hep-th/0005181
A.B. Zamolodchikov and Al.B. Zamolodchikov: [ *Structure constants and conformal bootstrap in Liouville field theory*]{}, Nucl. Phys. [**B477**]{} (1996) 577–605
[^1]: The general solution ${{\mathsf R}}^{{\rm{\scriptscriptstyle}XXZ}}_{s_{{\mathfrak 2}}s_{{\mathfrak 1}}}(u)$ is needed if we wish to construct an inhomogeneous spin chain, for instance the one with alternating spins (see, e.g., [@BD]).
[^2]: Indeed, for ${{\mathsf U}}$ this is obvious from the definition (\[Monotr\]) and for ${{\mathsf \Omega}}$ it follows from the observation that ${{\mathsf \Omega}}\, L^{{\rm{\scriptscriptstyle}XXZ}}(u) \, {{\mathsf \Omega}}= \sigma_1 \, e^{-\pi b u \sigma_3} \,
L^{{\rm{\scriptscriptstyle}XXZ}}(u) \, e^{\pi b u \sigma_3} \, \sigma_1$ and ${{\mathsf \Omega}}\, L^{{\rm{\scriptscriptstyle}SG}}(u) \, {{\mathsf \Omega}}= \sigma_1 \, L^{{\rm{\scriptscriptstyle}SG}}(u) \, \sigma_1$.
[^3]: By Lemma \[leadterm\], we have ${{\mathsf B}}_0=({{\mathsf B}}_{{\rm N}})^{-1}$.
[^4]: This argument is inspired by the considerations in [@Za] and a suggestion of F. Smirnov (private communication, see also [@Sm2]). However, the key point of our argument, namely the origin of the quantum Wronskian relation (iv) seems to be new.
[^5]: modulo notations, in particular, $\Delta=1$ in [@K1; @FKV]
[^6]: Commutativity of ${{\mathcal P}}_s$ and $\tilde{{{\mathcal P}}}_s$ only holds on the dense domain ${{\mathcal T}}_s$ but not in the usual sense of commutativity of spectral projections!
[^7]: We take the opportunity to correct some typos in these references.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Canonical correlation analysis (CCA) is a popular technique for learning representations that are maximally correlated across multiple views in data. In this paper, we extend the CCA based framework for learning a multiview mixture model. We show that the proposed model and a set of simple heuristics yield improvements over standard CCA, as measured in terms of performance on downstream tasks. Our experimental results show that our correlation-based objective meaningfully generalizes the CCA objective to a mixture of CCA models.'
bibliography:
- 'example\_paper.bib'
---
Introduction {#sec:introduction}
============
Multiview, correlation-based representation learning has been shown to be useful on a variety of tasks [@hardoon2004canonical; @wang2015unsupervised; @arora2013multi; @arora2014multi; @benton2016learning; @vasquez2017multi; @holzenberger2019learning]. Its main workhorse, Canonical Correlation Analysis (CCA) [@hotelling1936relations], enjoys several non-linear extensions with theoretical guarantees (kernel CCA and deep CCA) [@wang2015deep; @lai2000kernel], and can be extended to more than 2 views [@horst1961generalized; @rastogi2015multiview; @benton2017deep].
A related approach that yields an arguably richer representation is to learn a mixture model — instead of learning a single, high-dimensional, complex subspace, we can seek to find a union of subspaces. Mixture models, especially Gaussian Mixture Models, have a wide range of applications, from topic modeling [@blei2003latent] to speech recognition [@gales2008application]. In many cases, data can naturally be described as a union of distributions. For instance, phonemes in speech are a superposition of low-dimensional processes, either explained as templates of time-frequency vectors, or as typical articulatory motions [@sugamura1983isolated]. In the domain of information extraction, documents fall into categories such as newswire, blog posts, agency reports or tweets. In machine translation, domain match or mismatch plays an important part in the performance of a translation system [@koehn2017six]. Identifying the underlying components of the mixture, either explicitly or implicitly, can help unsupervised modeling of the data, useful for a host of downstream tasks.
Learning a mixture of subspaces rather than a single subspace also makes personalization simpler. For an unseen example, being able to assign it quickly to a subpopulation can make classification tasks require less data to achieve good performance. For example, in the context of speaker adaptation, being able to assign a speaker to a specific group of speakers improves speech recognition performance [@kuhn2000rapid].
In this paper, we propose to extend the framework of CCA to learn a union of subspaces. Specifically, we assume that each data point belongs to one of a finite number of sources, and we learn a pair of linear CCA transformations for each source. CCA with a single transformation can detect canonical directions that span the entire dataset. A mixture of CCA is also able to detect the main canonical directions, because they can be found in each subpopulation. In addition, a mixture of CCA can pick up correlations which are present at the subpopulation level but cancel each other out at the population level. This argument would suggest we can provide a much finer grained representation, possibly without increasing the dimensionality of the representation.
While the premise might seem straightforward, combining CCA and mixture models poses a number of challenges: what objective is to be maximized? How to simultaneously learn the cluster assignments and the transformations? Are there any guarantees regarding convergence, and recovery of cluster memberships? Once the parameters of the mixture model are learned, how to assign a new point $(x,y)$ to a cluster? How to assign a single $x$ without corresponding $y$ to a cluster? This paper is meant primarily as a proof of concept, and leaves most of the above questions open. We propose a new objective function, as well as a heuristic way of maximizing it, and test its performance in two distinct settings.
Related work {#sec:related_work}
============
We use Canonical Correlation Analysis (CCA) [@hotelling1936relations] to learn representations for a primary view. We assume that at training time, we are given two views of the same data point. For instance, for a given speech utterance, the audio recording and the articulatory measurements. These two views are represented by random variables $X$ and $Y$ ($d_X$- and $d_Y$-dimensional respectively). Linear CCA seeks two linear transformations ${\textrm{U}}\in {\mathbb{R}^{d_X \times k}}$ and ${\textrm{V}}\in {\mathbb{R}^{d_Y \times k}}$ such that the components of ${\textrm{U}}^TX$ and ${\textrm{V}}^TY$ are maximally correlated. Formally, we want to maximize ${\underset{X,Y}{\mathbb{E}}} [ {\textrm{tr }}({\textrm{U}}^TXY^T{\textrm{V}}) ]$ subject to the constraints that ${{\underset{X}{\mathbb{E}}} [ {\textrm{U}}^TXX^T{\textrm{U}}] = {\underset{Y}{\mathbb{E}}} [ {\textrm{V}}^TYY^T{\textrm{V}}] = {\textrm{I}}_k}$.
Given a dataset $\{x_i,y_i\}_{i=1}^N$, we define ${\textrm{C}}^{XY}$ the empirical cross-covariance matrix between $X$ and $Y$, and ${\textrm{C}}^{XX}$ and ${\textrm{C}}^{YY}$ the empirical auto-covariance matrices of $X$ and $Y$, respectively. ${\textrm{U}}$ and ${\textrm{V}}$ are given by the $k$ left and right singular vectors of $({\textrm{C}}^{XX})^{-1/2}{\textrm{C}}^{XY}({\textrm{C}}^{YY})^{-1/2}$ with the largest singular values, multiplied by $(C^{XX})^{-1/2}$ and $(C^{YY})^{-1/2}$.
At test time, we assume that only the primary view is available. Ideally, we would want to reconstruct the second view with the primary view [@ngiam2011multimodal]. However, in general, that is a difficult task — take for example generating speech from text. Instead, it is easier to predict the dependent variate which has the largest multiple correlation. In other words, when no single regression provides a fully adequate solution, CCA is a better objective than predicting one view with the other.
Deep CCA (DCCA) [@andrew2013deep] is a natural extension of linear CCA, where one seeks to maximally correlate ${\textrm{U}}^Tf(X)$ and ${\textrm{V}}^Tg(Y)$. $f$ and $g$ are non-linear feature extractors, which can be learned via gradient descent on the CCA objective. It is also natural to extend CCA to multiple views [@horst1961generalized].
CCA can also be reformulated as a probabilistic model. @bach2005probabilistic show that the solution to the linear CCA objective function is, up to arbitrary rotation and scaling, contained in the maximum-likelihood solution for the parameters of a Gaussian. @podosinnikova2016beyond extend this probabilistic formulation of CCA to fit any type of statistical distribution, including when one or both views are discrete.
@klami2007local place the probabilistic model of @bach2005probabilistic in a Bayesian setting, with a Dirichlet process, which naturally leads to a mixture of Gaussians. They then maximize log-likelihood under this GMM model using methods from variational inference. As a final step, they extract CCA projections from each Gaussian. This amounts to performing GMM-based clustering first, and then performing CCA on each of the learned clusters. In contrast, we define a correlation-based objective for multiple transformations, and propose to maximize it directly. It remains to be examined whether both objective functions are equivalent, or even whether stationary points of both objective functions are equivalent. In fact, the model of @klami2007local might be suboptimal in terms of our objective. Other works [@wang2007variational; @viinikanoja2010variational; @hosino2010high] have used very similar approaches.
Most related to this work is that of @fern2005correlation, where the authors use of a mixture of CCA, and provide an optimization heuristic similar to ours. However, their work differs on a number of points. First, @fern2005correlation do not provide an objective function to optimize. Second, they do not test the CCA projections in a downstream task, and are thus arguably not concerned with questions pertaining to representation learning. Third, @fern2005correlation do not provide any way of assigning a point with a single view to a cluster (i.e., they specify how to assign $(x,y)$ to a cluster, but not how to assign $x$ to a cluster in the absence of $y$). The main focus of our work is the objective function, as a possible extension of CCA and representation learning method.
Mixture model for Canonical Correlation Analysis
================================================
We propose to learn a union of subspaces that best characterizes a set of points with two views, using canonical correlation analysis (CCA). It involves assigning cluster memberships to data points, and we limit ourselves to linear versions of CCA. Throughout this paper, we will refer to this method as mixture of CCA (MCCA).
Objective formulation {#subsec:objective_formulation}
---------------------
We assume that our data consists of two sets of points. Let $X \in {\mathbb{R}^{d_X \times N}}$, $Y \in {\mathbb{R}^{d_Y \times N}}$ be the matrices containing the paired views, with $x_i \in {\mathbb{R}^{d_X}}$ (resp. $y_i \in {\mathbb{R}^{d_Y}}$) being the $i$-th column of $X$ (resp. $Y$). For example, in our application to speech recognition, $X$ represents a set of MFCC frames and $Y$ the corresponding articulatory measurements. We consider a mixture of $R$ subspaces and consider the following CCA formulation. We define the MCCA objective as:
$$O(U,V,\alpha) = \sum \limits_{r=1}^R {\textrm{tr }}(U_r^T C^{XY}_r V_r)$$
subject to the constraints that
$${\forall r \in \{1..R\}, \thickspace U_r^T C^{XX}_r U_r = V_r^T C^{YY}_r V_r = {\textrm{I}}_k}$$
where ${U=\{U_r \in {\mathbb{R}^{d_X \times k}} \mid r=1,\ldots,R\}}$, and ${V = \{V_r \in {\mathbb{R}^{d_Y \times k}} \mid r=1,\ldots,R\}}$ are the CCA transformation matrices associated with each mixture component.
The assignment of data points to mixture components is given by the following scalar weights:
$${\alpha = \{\alpha_{i,r} \in \mathbb{R}_{\geq 0} \mid i \in [N], r\in [R], \forall i, \sum \limits_{r=1}^R \alpha_{i,r} = 1\}},$$
where $[p]$ denotes the set $\{1,\ldots,p\}$.
The weighted empirical covariance and cross-covariance matrices for each mixture component are given as
$$C^{XY}_r = \frac{1}{\sum \limits_{i=1}^N \alpha_{i,r}} \sum \limits_{i=1}^N \alpha_{i,r} (x_i - \mu^{X}_r) (y_i - \mu^{Y}_r)^T,$$
$$C^{XX}_r = \frac{1}{\sum \limits_{i=1}^N \alpha_{i,r}} \sum \limits_{i=1}^N \alpha_{i,r} (x_i - \mu^{X}_r) (x_i - \mu^{X}_r)^T,$$ and
$$C^{YY}_r = \frac{1}{\sum \limits_{i=1}^N \alpha_{i,r}} \sum \limits_{i=1}^N \alpha_{i,r} (y_i - \mu^{Y}_r) (y_i - \mu^{Y}_r)^T,$$
and the weighted means of the mixture components are given as
$$\mu^{X}_r = \frac{1}{\sum \limits_{i=1}^N \alpha_{i,r}} \sum \limits_{i=1}^N \alpha_{i,r} x_i,~~~ \mu^{Y}_r = \frac{1}{\sum \limits_{i=1}^N \alpha_{i,r}} \sum \limits_{i=1}^N \alpha_{i,r} y_i.$$
At training time, the goal is to find
$$U^*, V^*, \alpha^* = \underset{U,V, \alpha}{\textrm{argmax }} O(U,V, \alpha)~~~\textrm{(subject to constraints)},$$ i.e., to find for each mixture component, both CCA transformation matrices and the assignments for each data point $(x_i,y_i)$.
At test time, given $x$ and/or $y$, one needs to estimate the $\{\alpha_r\}_r$ before projecting $x$ (respectively, $y$) to $\sum \limits_{r=1}^{R} \alpha_r U_r^T x$ (respectively, to $\sum \limits_{r=1}^{R} \alpha_r V_r^T y$).
To avoid spurious correlations, we regularize the covariance matrices by adding scaled identity matrices, i.e., replacing $C^{XX}$ (respectively, $C^{YY}$) with $C^{XX} + w_X I$ (respectively, $C^{YY} + w_Y I$).
Learning and inference {#subsec:learning_inference}
----------------------
The optimization problem in Section \[subsec:objective\_formulation\] is nonconvex, and jointly minimizing over all parameters seems daunting from a computational perspective due to intractability. Therefore, we consider the following optimization approach that first estimates $\alpha$, then estimates $U$ and $V$ given $\alpha$, as summarized in Algorithm \[alg:mcca\_alternate\_maximization\]. When $\alpha$ is fixed, it is well known that we can globally maximize the $R$ CCA objectives over the choice of $U$ and $V$, as described in Section \[sec:related\_work\]. Similarly, fixing the CCA subspaces, maximizing over $\alpha$ is essentially learning a mixture model over the shared representation.
MCCA hyperparameters $\{R,k,w_X,w_Y\}$ and dataset $\{X \in {\mathbb{R}^{d_X \times N}}, Y \in {\mathbb{R}^{d_Y \times N}}\}$ Transformation matrices $U$ and $V$, and cluster assignments $\alpha$ initialize $\alpha$ using a clustering algorithm over the native space find $U,V = \underset{\tilde{U}, \tilde{V}}{\textrm{argmax }} O(\tilde{U}, \tilde{V}, \alpha)$ return $U, V, \alpha$
At training time, we first cluster the data points using a CCA projection and k-means clustering, yielding hard assignments for $\alpha$, i.e. ${\alpha \in \{0,1\}^{NR}}$. We then learn $U$ and $V$.
At test time, we need a way of inferring the $\{\alpha_{i,r}\}_{r}$ for a given $(x_i,y_i)$ where $x_i$ or $y_i$ may be missing. In our experimental setting, only $x$ is available at test time. From the properties of the CCA projections, we know that $\mathrm{var}(U^TX) = I_k$. Heuristically, we expect $U^T (x-\mu_X)$ to be drawn from a unit variance Gaussian distribution. Thus, we assign $x$ to the cluster $$\hat{r} = \underset{r \in \{1,\ldots,R\}}{\textrm{argmin}} ||U_r^T(x-\mu^X_r)||_2^2 - \log \pi_r,$$ where $\pi_r$ is the fraction of points belonging to cluster $r$. In some of our speech recognition experiments, $\alpha$ will be given at training time.
Experiments
===========
In this section, we illustrate the possible uses of MCCA, and test its performance against standard, vanilla linear CCA described in Section \[sec:related\_work\], and referred to as VCCA. All non-linear extensions of CCA – deep CCA and kernel CCA – have significantly more representational power than linear CCA. Most likely, deep CCA would outperform all other methods. Thus, for the comparison to non-linear extensions of CCA to be fair, one would have to extend MCCA to non-linear methods. This would pose many more challenges related to optimization. Instead, our intention is to carefully illustrate the linear version, leaving non-linear extensions to future work.
Phoneme classification {#subsec:phoneme_classification}
----------------------
The University of Wisconsin X-ray microbeam database (XRMB) [@westbury1990x] is a set of sound recordings and articulatory measurements, acquired during production of English read speech. The acoustic recordings are processed into sequences of 13 dimensional MFCC frames. The articulatory measurements are the horizontal and vertical displacements of eight pellets affixed to critical articulators (tongue, lips, jaw) at a given point in time, resulting in a 16 dimensional observation vector. Both views are temporally aligned, and thus we have pairs of MFCC frames and articulatory features. This correspondence is the only supervisory signal when learning representations; while the manual annotations of the spoken text are available, we do not use it for feature learning to avoid any task dependence in the learned representations. The location of the pellets in the original dataset go missing for various reasons; we use the completed version of the XRMB dataset due to @wang2014reconstruction. The MFCC frames are augmented with deltas and double deltas, and are mean-centered and variance-normalized per speaker. The articulatory features are also mean-centered and variance-normalized per speaker.
In this section, we illustrate the usefulness of learning a union of subspaces, by partitioning the 39 phonemes of the XRMB dataset into 4 groups, as detailed in Table \[tab:xrmb\_groups\]. For some experiments, we assume access at training time to an oracle providing us with the correct group for each training instance; in those cases, we are not concerned with inferring $\alpha$. At test time, we also experiment with the presence and absence of said oracle. In the absence of the oracle, we use the heuristic described in Section \[subsec:learning\_inference\] to infer the membership of each point.
0.15in
----------- -------------------------------------------------------------------
vowels AA, AH, IH, AO, ER, EH, IY, UW, AE, OW, UH, AY, EY, AW, OY, HH, Y
alveolars CH, D, DH, JH, L, N, R, S, SH, T, TH, Z, ZH
labials B, F, M, P, V, W
velars G, K, NG
----------- -------------------------------------------------------------------
: 4 groups used to partition the English phonemes.[]{data-label="tab:xrmb_groups"}
-0.1in
Using the notation of Section \[subsec:objective\_formulation\], $X$ consists of 7 stacked MFCC frames with deltas and double deltas, centered around the frame of interest, and is thus 273 dimensional vectors. $Y$ consists of 7 stacked articulatory feature vectors, corresponding to the 7 MFCC frames, and is thus 112 dimensional. The training set has 1.4M data points, and each of the 6 cross-validation folds mentioned below has between 79k and 86k data points.
To evaluate the quality of the learned representations, we follow @wang2015unsupervised and leave out 12 speakers. The other 35 speakers are used to learn representations. The 12 left-out speakers are used to measure how well the learned representations for audio can be used to perform phoneme classification on unseen data. To estimate generalization error, the 12 speakers are partitioned into 6 sets of 2 speakers each, and used in a 6-fold cross-validation, each fold composed of 4 training speakers, 2 dev speakers, and 2 test speakers. We use the 4 train and 2 dev speakers to find the best parameters for a k-nearest neighbor classifier [@malkov2018efficient], then measure the score on the dev set. We use the average dev score over the 6 folds to compare hyperparameters of a given method. This procedure allows us to pick the best hyperparameters for a given CCA method (VCCA or MCCA). Methods are then compared on the 6 test sets, with hyperparameters $\Theta^*$ and $\Phi^*$. We report average and standard deviation of the performance over the 6 dev and test sets. Knn classifier hyperparameters $\Phi$ contain the type of distance used (L2 or cosine), the number of neighbors (8, 16, 32, 64, 128 or 256), and whether to append the original MFCC features to the CCA features.
We sweep $k$ over $\{10, 30, 50, 70, 90, 110\}$, and $w_X$ and $w_Y$ independently over $\{1, 0.1, 0.001\}$. For VCCA, we use the projection of the MFCC features, and experiment with appending the original MFCC features to perform knn classification. For MCCA, we also experiment with appending the original MFCC features. In addition, when projecting point $x$, we have the possibility of mapping $x$ to $\sum \limits_{r=1}^R \alpha_{r}U_r^T x \in {\mathbb{R}^{k}}$ or to the concatenation of $U_1^Tx, U_2^Tx, ... , U_R^T x$, which is an $Rk$ dimensional vector. Note that the former requires inferring $\alpha$ at test time, while the latter doesn’t. We experiment with both settings, reporting the former as “projection” and the latter as “concatenation”.
Table \[tab:xrmb\_results\_noracle\] reports our experimental results. We experiment with the oracle of Table \[tab:xrmb\_groups\] being present or absent at training time (denoted by “oracle” and “no oracle” respectively). When the oracle is absent at training time, we sweep $R$ over $\{2, 4, 8, 16\}$. In this setting, both instances of MCCA yield a significant improvement over VCCA. The best dev score for VCCA is achieved with $k=70, w_X=0.001, w_Y=1$, while the best score for MCCA with oracle is achieved with $k=50$ and the same values of $w_X$ and $w_Y$; the best score without oracle is achieved with $k=30$, $w_X=w_Y=0.001$ and $R=8$. There is no reduction in the dimensionality of the representation because the best dev score is achieved by concatenating all $R$ representations.
0.15in
features dev (%) test (%)
--------------------------------- ---------------- ----------------
MFCC $60.6 \pm 2.8$ $60.6 \pm 2.5$
VCCA $65.3 \pm 2.8$ $65.3 \pm 2.9$
MCCA (oracle, projection) $64.1 \pm 2.7$ $64.2 \pm 2.6$
MCCA (oracle, concatenation) $68.0 \pm 2.7$ $68.0 \pm 2.8$
MCCA (no oracle, projection) $64.7 \pm 2.8$ $64.8 \pm 2.7$
MCCA (no oracle, concatenation) $69.2 \pm 2.8$ $69.3 \pm 2.8$
: Phoneme classification accuracy on XRMB dataset without oracle at test time. Oracle (resp. no oracle) indicates access (resp. no access) to the oracle at test time.[]{data-label="tab:xrmb_results_noracle"}
-0.1in
The results in Table \[tab:xrmb\_results\_noracle\] show that MCCA yields a significant improvement over VCCA, with and without oracle at training time. In practice, it is likely that there is no oracle at training time, either because of a lack of labels or because there is no clear partitioning of the data. The bottom two lines in Table \[tab:xrmb\_results\_noracle\] show that the absence of the oracle does not necessarily imply a loss in performance. In fact, learning each point’s assignment to a mixture component yields better results than following the heuristic from Table \[tab:xrmb\_groups\].
If the oracle had been available at test time in addition to training time, the instance of MCCA projection reported in Table \[tab:xrmb\_results\_noracle\] would reach $77.5 \pm 2.2 \%$ on the dev set and $77.5 \pm 2.1 \%$ accuracy on the test set. This shows that while our test-time heuristic is able to outperform standard CCA, there is still room for improvement. It also shows that, although our MCCA projection results are below our MCCA concatenation results, given correct point assignments, projection is enough to guarantee a useful representation.
In Figure \[fig:perplexity\_matrices\], we show perplexity matrices for some of our MCCA models on the test set. On that plot, phonemes are sorted according to their membership in the 4 groups of Table \[tab:xrmb\_groups\]. The first column has access to the oracle at training time, and achieves the best score when projecting points with a single CCA transformation. In that case, ideally, we would want the assignments to be 4 disjoint bands following the 4 clusters. However, each phoneme is mostly assigned to groups 1 and 2, which are also the groups with the largest mixing weights. This mis-assignment can be somewhat alleviated by removing the $\log \pi_r$ term (see Section \[subsec:learning\_inference\]), but worsens the dev score. Columns 2 shows that, given 4 clusters but no access to the oracle, the clustering heuristic used comes up with sharp clusters for consonants. Roughly, cluster 1 contains the labials, cluster 3 contains the alveolars, and the velars are spread out among clusters 1 and 3. Consonants are mostly spread out between clusters. If we do not constrain the number of clusters to be 4, the best performing model without oracle at training time and using projection has 2 clusters, shown in the third column. In that case, we see mostly sharp assignments, i.e. almost all phonemes are assigned to a single cluster, despite no access to any phonemic information. With some exceptions, consonants belong to cluster 1, and vowels fall into either cluster. The fourth column shows cluster assignments for the best performing MCCA model; note that this model concatenates representations, and thus cluster assignments do not matter at test time. Clusters 2 and 7 seem to be unused, and cluster 4 loosely corresponds to labials. Alveolars and velars are spread out among clusters 1 and 8.
0.2in
![Perplexity matrices for various MCCA models on the test set. Rows are labeled with phonemes and columns with mixture components. Rows are normalized to sum to 1. Each model was chosen as the best in its category, as described by the label. Columns 1, 3 and 4 correspond to the models reported in Table \[tab:xrmb\_results\_noracle\].[]{data-label="fig:perplexity_matrices"}](figures/perplexity_matrices.pdf){width="\columnwidth"}
-0.2in
Twitter data
------------
To provide an illustration of MCCA in a different domain, we use the corpus of Twitter data of @benton2016learning. In that paper, various methods are used to build representations of Twitter users, based on their tweets and friend networks. Following the nomenclature of @benton2016learning, we use the ego view, i.e. a PCA-based representation of the user’s tweets, and the friends view, i.e. a PCA-based representation of the user’s friend network or graph. We use the ego view as the primary (available at training and test time) and the friends view as the secondary view (available at training time only). Using the notation of \[subsec:objective\_formulation\], $X$ (resp. $Y$) is the primary (resp. secondary) view. Both are 1000 dimensional. To estimate how much the use of CCA is improving the performance on a given task, we also report results using the primary view at training time, under “raw features”.
### User engagement prediction {#subsec:user_engagement_prediction}
To evaluate the learned representations, we perform a user engagement prediction task, using hashtag as a proxy (following @benton2016learning). We have 2 sets (dev and test) of 200 unseen hashtags each, and for each hashtag, a set of 16k unseen users who have used them. The task is to predict, for each hashtag, which users are likely to use it, based on the first 10 users who have used it. We follow the setup of @benton2016learning, but report slightly different metrics. Note that because of missing views, we have had to restrict the number of users, including in the test set, and thus the results are not comparable. The train, dev and test set contain respectively 79900, 8220 and 8071 users. For a given method, we use the performance on the dev set to select the best hyperparameters, then compare methods on the test set. We decide whether or not to concatenate the original representation to the CCA projection based on the dev set performance. When concatenating, we scale each representation by its average norm taken over the dataset, in order to balance out the weight of each representation in the cosine similarity.
More specifically: we first project all the users $u_1, u_2, ..., u_N$ of the dev (resp. test) set using the learned CCA transformations. For each hashtag $i$ in the dev (resp. test) set, we have a list of users who have used them. We pick the first 10, and compute the average of their embeddings; we take this as the representation $h_i$ of the hashtag. Hashtag representations $\{h_i\}_i$ and user representation $\{u_j\}_j$ are separately mean-centered. Then, for each hashtag $h_i$ and each user $u_j$, we compute the scaled cosine similarity $\frac{1}{2}(1+\frac{h_i\cdot u_j}{||h_i||||u_j||})$, yielding a confidence score between $0$ and $1$. We use that score as a measure of how likely the user $j$ would use the hashtag $i$. We can compare these scores with the ground truth of which user actually used each hashtag.
We report Recall@1000, mean reciprocal rank, and the area under the ROC curve (ROC-AUC) [@fawcett2004roc] of a multiclass classifier mapping users to hashtags. Recall@1000 and ROC-AUC are bounded between $0$ and $100$, and mean reciprocal rank between $0$ and $1$. For all of these metrics, higher is better. For each hashtag, we rank the users based on their confidence score, in decreasing order. Recall@1000 tells us how much the top 1000 users overlap with the ground truth, and so how many relevant answers are present in our top 1000 answers. Mean reciprocal rank tells us how close the first relevant result is to the top of our result list. If we were to build the simplest classifier for hashtag $j$, it would rule that user $i$ is going to use hashtag $j$ if $\frac{1}{2}(1+\frac{h_i\cdot u_j}{||h_i||||u_j||})>\theta_j$ where $\theta_j$ has to be set for each hashtag. The ROC-AUC tells us how well the confidence scores of correct users are separated from those of incorrect users. A ROC-AUC score of $100$ means we could set the $\{\theta_j\}_j$ randomly and still classify each user and hashtag correctly, and $50$ means we are doing as well as a random classifier. The higher the ROC-AUC score, the better the performance of the classifier will be, regardless of the values of the $\{\theta_j\}_j$.
In contrast to the experiments in Section \[subsec:phoneme\_classification\], there is no obvious partitioning of Twitter users on this dataset, and thus we have no oracle at all in this case. We sweep $k$ over $\{200, 400, 600, 800, 1000\}$, and $w_X$ and $w_Y$ independently over $\{1, 0.1, 0.01\}$. In the case of MCCA, we sweep the number of clusters $R$ over $\{2,4,8,16\}$ and follow Section \[subsec:learning\_inference\] for learning and inference.
We summarize the results of the user engagement prediction task in Table \[tab:twitter\_hashtag\]. On all three metrics considered, MCCA outperforms VCCA. For Recall@1000 and ROC-AUC, the hyperparameters achieving the best scores are identical. For VCCA, ${k=600,w_X=0.1,w_Y=1}$; for MCCA projection, ${R=2,k=1000,w_X=0.1,w_Y=0.001}$; for MCCA concatenation, ${R=8,k=400,w_X=0.001,w_Y=1}$. For all three, the best scores are achieved with the CCA representation alone (without appending the raw features). In the case of mean reciprocal rank, the best scores are achieved by concatenating the CCA representation and the raw features. For VCCA, ${k=400,w_X=0.001,w_Y=1}$; for MCCA projection, ${R=8,k=800,w_X=0.001,w_Y=1}$; for MCCA concatenation, ${R=2,k=1000,w_X=0.001,w_Y=1}$. The large difference in best performing hyperparameters shows how important task-based hyperparameter selection is. In that respect, MCCA has 2 more hyperparameters that can be tuned based on the task: the number of mixture components $R$, and whether to project $x$ to ${\sum \limits_{r=1}^R \alpha_r U_r^T x}$ (“projection”) or to the concatenation of ${U_1^Tx, U_2^Tx, ... , U_R^Tx}$ (“concatenation”).
Except for mean reciprocal rank, MCCA projection performs consistently worse than VCCA. This could possibly be because the clusters picked by our heuristic have no intrinsic relationship with the hashtags. It is interesting to note that MCCA projection drastically improves the mean reciprocal rank to perfect or almost perfect score, meaning that the top result is a correct answer, almost always. Higher ROC-AUC scores for MCCA concatenation show that this method is better able to separate, for each hashtag, relevant from irrelevant users. This is consistent with higher Recall@1000 scores, which indicate more relevant users within the top 1000 results.
0.15in
----------------------------------------- ------ ------ ------ ------ ------ ------
(r)[2-3]{}(r)[4-5]{}(r)[6-7]{} features dev test dev test dev test
raw features 53.0 44.0 78.4 72.0 .858 .876
VCCA 62.0 51.8 83.5 77.1 .911 .969
MCCA (projection) 60.1 50.6 82.7 76.4 1.0 1.0
MCCA (concatenation) 64.3 53.2 85.1 78.4 .920 .958
----------------------------------------- ------ ------ ------ ------ ------ ------
: User engagement prediction results on Twitter dataset. REC: Recall@1000, ROC-AUC: area under the ROC curve, MRR: mean reciprocal rank.[]{data-label="tab:twitter_hashtag"}
-0.1in
### Friend recommendation {#tab:friend_recommendation}
In a similar setting, we perform friend recommendation [@benton2016learning]. 500 user accounts were set aside (250 for each dev and test), and it was recorded which of the dev and test users followed those accounts. The friend recommendation task amounts to predicting which user accounts a given user is likely to follow. The train, dev and test set contain respectively 6522, 82608 and 81985 users. The task setup is identical to user engagement prediction. We sweep over the same values of hyperparameters as for user engagement prediction.
We report our results in Table \[tab:twitter\_friend\]. Based on the results using the raw features, this task is much harder than user engagement prediction, described in \[subsec:user\_engagement\_prediction\]. In this setting, MCCA performs better or on par with VCCA. The smaller performance gap between VCCA and MCCA could be explained by the much smaller size of the train set, and possibly the difficulty of obtaining a coherent clustering. For each of the best performing MCCA instances, the clustering at training time shows that one of the clusters collapses to 1, 2 or 3 points. For each method, the best performing hyperparameters vary from one task to the other, without any pattern emerging. Again, this shows the impact of task-based hyperparameter selection.
While the task is quite different from user engagement prediction described in Section \[subsec:user\_engagement\_prediction\], the conclusions regarding VCCA and MCCA remain mostly identical.
0.15in
----------------------------------------- ------ ------ ------ ------ ------ ------
(r)[2-3]{}(r)[4-5]{}(r)[6-7]{} features dev test dev test dev test
raw features 3.21 2.96 59.6 59.3 .821 .788
VCCA 4.02 4.04 62.3 62.1 .924 .909
MCCA (projection) 3.83 3.48 62.2 61.8 .994 .996
MCCA (concatenation) 4.03 4.05 62.6 62.0 .957 .942
----------------------------------------- ------ ------ ------ ------ ------ ------
: Friend recommendation results on Twitter dataset. REC: Recall@1000, ROC-AUC: area under the ROC curve, MRR: mean reciprocal rank.[]{data-label="tab:twitter_friend"}
-0.1in
Conclusion
==========
In this paper, we proposed a novel objective for representation learning, combining CCA and mixture models, in conjunction with simple heuristics to maximize the objective at training time, and use the representations at test time. Evaluating our representations in different settings, we have shown the usefulness of both our objective and our heuristics. Overall, across tasks, our proposed method performs on par or better than the standard version of CCA.
Our results suggest that the proposed method is a valid objective, potentially a good generalization of the standard CCA objective. With more hyperparameters than the standard CCA objective, and the possibility of informing the mixture components with hierarchical structure present in the data, it has the potential to better adapt to downstream tasks. It would further benefit from a more thorough optimization scheme with provable guarantees, and possibly extensions to non-linear methods.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We demonstrate that in supersymmetry (SUSY) with relatively light top superpartners, $h\to b\bar b$ can be a very promising channel to discover the SM-like Higgs resonance at the Large Hadron Collider (LHC), although in general contexts it is thought to be challenging because of its huge QCD background. In this scenario, the SM-like Higgs boson is mainly produced via cascade decays initiated by pair-produced stop or sbottom squarks. The good sensitivity to $h\to b\bar b$ owes a great deal to the application of multiple ($\ge 4$) $b$-jet tagging in removing the QCD background, and color-flow variables for reconstructing the Higgs resonance. We show in two benchmark points that a SM-like Higgs resonance can be discovered at the 14 TeV LHC (with a signal-to-background ratio as high as $0.35$), with $\sim 40$ fb$^{-1}$ of data. Potentially, this strategy can be also applied to non-SUSY theories with cascade decays of top partners for the SM-like Higgs search, such as little Higgs, composite Higgs, and Randall-Sundrum models.'
author:
- David Berenstein
- Tao Liu
- Erik Perkins
title: ' Multiple $b$-jets Reveal Top Super-partners and the 125 GeV Higgs '
---
Deciphering the Higgs mechanism is one of the top priorities of the ATLAS and CMS experiments at the Large Hadron Collider (LHC). Recently, both the CMS and the ATLAS collaborations announced discovery of a Higgs-like resonance, based on a combined analysis for the Standard Model (SM) Higgs searches via $h\to \gamma\gamma$ and $h\to ZZ^* \to 4l$, with the reconstructed invariant mass $\sim 125-126$ GeV [@:2012gu; @:2012gk]. Furthering the identification of this new particle greatly enhances the necessity and emergency of information from complementary Higgs search channels like $h\to b\bar b$.
Unlike $h\to \gamma\gamma$ and $h\to ZZ^* \to 4l$, $h\to b\bar b$ can help test the mass generation mechanism of the SM fermions directly. However, the $h \to b \bar b$ search at the LHC is challenging because of its huge QCD background, except in the highly boosted regime where jet kinematics allows for the successful application of jet-substructure tools [@Butterworth:2008iy; @Plehn:2009rk]. Physics beyond the SM can modify the collider phenomenology of the Higgs search modes if there are new Higgs production mechanisms available. With new strategies designed, the sensitivity of measuring the challenging decay modes could be sizably improved. In this letter we will demonstrate that in supersymmetric scenarios with relatively light top partners, the Higgs discovery via $h\to b\bar b$ could be greatly assisted by a set of dedicated strategies.
Supersymmetry is the prime candidate theory for physics beyond the SM. It provides a natural solution to the hierarchy problem. Among various SUSY scenarios, the ones with relatively light top partners like natural SUSY are theoretically more predictive and experimentally more accessible. For example, in natural SUSY it is predicted that there existed light superparticles [@Papucci:2011wy]: two stop and one left-handed sbottom squarks, with their masses $\lesssim 700$ GeV; two Higgsino-like neutralinos and one Higgsino-like chargino, with their masses $\lesssim 350$ GeV; and a gluino, with its mass $\lesssim 1500$ GeV. (In this letter, however, we do not adhere to a precise measurement of naturalness, but focus on general scenarios with relatively light top partners.)
With conserved R-parity, stop and sbottom squarks are pair-produced at colliders. An interesting feature is that they often decay into a SM-like Higgs boson, in association with a $b$ quark (for discussions on SUSY-assisted Higgs production in various contexts, see [@Baer:1992ef]-[@Kribs:2009yh]). If the Higgs boson decays into a pair of $b$ quarks, the final state typically contains at least four $b$ quarks. There are two main mechanisms for the Higgs production via the stop and sbottom cascade decays. In the first case, the Higgs boson is produced via neutralino or chargino decays where the Yukawa couplings of the SM fermions are not directly involved. In the second case, the Higgs boson is directly produced via the decay of the heavier stop squark where the top Yukawa coupling gets involved. Because these processes are initiated by light squarks, the produced Higgs bosons tend to be less boosted, compared to the case discussed in [@Kribs:2009yh] (for discussions on less boosted Higgs boson, also see [@Gori:2011hj]). Two topologies of these mechanisms are illustrated in Fig. \[mec\]. Given that the QCD events containing multiple $b$-jets are relatively few, this feature provides a new strategy of improving the sensitivity of the $h \to b\bar b$ resonance search at the LHC.
![The primary Higgs-producing SUSY cascades in the first (left) and second (right) benchmark scenarios.[]{data-label="mec"}](light-stop-cascade.pdf "fig:"){width="23.00000%"} ![The primary Higgs-producing SUSY cascades in the first (left) and second (right) benchmark scenarios.[]{data-label="mec"}](heavy-stop-cascade.pdf "fig:"){width="23.00000%"}
To design the minimal strategies that ensure optimal coverage of the full space of models, we perform a collider analysis in two benchmark scenarios which represent these two mechanisms, respectively. Their mass spectra and branching ratios are calculated using SUSY-HIT [@Djouadi:2006bz], in the context of the MSSM with $m_h=125$ GeV set as an input (see Table \[T:BenchmarkPoints\]). Here new contributions to $m_h$ have been implicitly assumed which could be either from new F-terms or from new D-terms (for recent discussions, e.g., see [@Hall:2011aa]). These benchmarks are not fully realistic, but simplified models in essence [@ArkaniHamed:2007fw]. Generalization of the analysis to more realistic cases is straightforward. Though searching for fully realistic benchmark points is not our main motivation, we would like to point out that these benchmark points are still safe and are allowed by current experimental bounds, after comparing them with the publicly available searches, such as the CMS $b$-enriched razor-variable search [@:CMS-PAS-SUS-11-024], the ATLAS light sbottom search [@:ATLAS-CONF-2012-106], and the CMS $b$-jets $+ \slash{E}_T$ search [@:2012rg].
Benchmarks I (GeV or %) II (GeV or %)
---------------------------------------------------- -------------- ---------------
$m_{\tilde g}$ 1281 1264
$m_{\tilde t_1} $ 568 260
$m_{\tilde t_2} $ 682 586
$m_{\tilde b_1}$ 567 555
$m_{\tilde \chi^0_1}$ 87 84
$m_{\tilde \chi^0_2}$ 325 415
$m_{\tilde \chi^0_3}$ 336 433
$m_{\tilde \chi^\pm_1}$ 321 413
$m_{h}$ 125 125
${\rm Br}(\tilde t_2 \to \tilde t_1 + h)$ 0 47
${\rm Br} (\tilde t_1 \to \tilde \chi_1 + h + t )$ 52 0
${\rm Br}(\tilde h \to b\bar b )$ 61 61
: Two benchmark scenarios. The weak-scale mass spectrum and decay branching ratios are calculated using SUSY-HIT [@Djouadi:2006bz], with $m_h=125$ GeV set as an input.[]{data-label="T:BenchmarkPoints"}
At the LHC, the resonance search of the SM-like Higgs boson via multiple $b$-jets has three main backgrounds. The first one is the SM background, mainly $t\bar{t}b\bar{b}$. The $t\bar{t} + \leq 2j$ is potentially important, but the 4$b$-tagging requirement can remove much of this background; this is discussed further below. Other SM processes involving $\geq 4b$ jets in the final state are less important for our purposes; *e.g.* we have checked that contributions from QCD $b b \bar b \bar b$ can be efficiently removed with a $\slashed{E}_T$ and $H_T$ requirement, and that $ZZ\rightarrow 4b$ makes a negligible contribution to our search. The second type of background is SUSY events containing multiple $b$-jets, but with no $h$ produced. The third background is combinatorial, arising from events where a SM-like Higgs boson decaying to $b$ quarks is present in the cascade, but where the $b$-tagged jets paired to reconstruct the Higgs are chosen incorrectly. We stress, however, that the latter two types are background *for the Higgs reconstruction only* - in fact these ‘backgrounds’ constitute a strong signal for SUSY.
The cross sections for all processes here include a $K$ factor; `Prospino2` [@Beenakker:1997ut] determined this for the SUSY processes. The $K$ factor for pure QCD $t\bar{t}b\bar{b}$ and $t\bar{t}+$jets is 2.3, while that for contributions to $t\bar{t}b\bar{b}$ from $t\bar{t}Z$ and $t\bar{t}h$ is 1.6 [@Bredenstein:2009aj]. The next-to-leading-order (NLO) cross section of stop pair production at the $8$ TeV and $14$ TeV LHC is shown in Fig. \[stop\_cross\_section\].
![The NLO cross section of stop pair production at the $8$ TeV and $14$ TeV LHC generated by Prospino2 [@Beenakker:1997ut]. []{data-label="stop_cross_section"}](stop-cross-section.pdf){width="30.00000%"}
Our analysis framework is as follows. We generate events with `MadGraph5 / MadEvent` [@Alwall:2011uj], and perform showering and hadronization with `Pythia` 8.1 [@Sjostrand:2007gs]. For the SM backgrounds, we use the `Pythia-PGS` package in `MadGraph` to do showering and MLM matching. We do not employ any dedicated detector simulation in our analysis, but we do place cuts on the hadron-level kinematics to mimic the response of a generic detector. We subject the 4-momentum of each visible final state particle to a Gaussian smearing, as implemented in `Delphes 2.0` [@Ovyn:2009tx]. After smearing, all visible final state particles are required to have $p_T > 0.9$ GeV, and $|\eta| < 5.0$ to fall within the calorimeter acceptance. The remaining ‘tracks’ are identified using generator-level information.
Jet reconstruction is performed next, using `FastJet 3.0` [@Cacciari:2011ma]. We use the inclusive anti-$k_T$ algorithm with $R = 0.5$ and $p_T > 20$ GeV, performing the clustering using all tracks. This sometimes results in ‘jets’ which are actually isolated leptons and photons. Electrons and photons are identified as isolated if the scalar sum of charged track $p_T$ in a cone of $R = 0.2$ does not exceed 10% of the electron or photon $p_T$; we further require that the electrons have $p_T > 20$ GeV and $|\eta| < 2.47$, while isolated muons are instead required to have $p_T > 10$ GeV and $|\eta| < 2.4$, with the scalar $p_T$ sum of charged tracks within $R = 0.2$ of the muon less than 1.8 GeV [@Ovyn:2009tx]. Once isolated leptons and photons are identified, they are removed from the collection of clustered objects. The remaining jets are then flavor-tagged with generator-level information.
B tagging is done by first determining the parton with highest $p_T$ within the jet cone, and considering this to be the ‘Monte-Carlo-true’ flavor of the jet. True $b$-jets are tagged with an efficiency parameterized by $\varepsilon = 0.6\tanh(p_T/36.0)(1.02-0.02|\eta|)$ [@ATLAS-CONF-2011-089], and true c-jets and light jets are mistagged as $b$-jets at flat 10% and 1% rates [@Aad:2009wy; @CMSTDR], respectively. Jets are only considered for $b$-tagging if they fall within the tracker acceptance of $|\eta| < 2.5$. Finally, each final-state object is assigned an ancestor heavy particle according to the generator-level decay history; jet ancestry is determined by the ancestry of the hardest parton in the jet cone.
After the final state objects have been reconstructed, events are subjected to selection cuts. Our analysis implements no triggering step, but it is useful to consider this issue briefly. The current searches at ATLAS [@:ATLAS-CONF-2012-135] and CMS [@:CMS-PAS-HIG-12-025] for $t \bar{t} h$ (which has similar final states and backgrounds to our scenarios) use a low-level trigger based on isolated leptons. However, such triggers do not have a high acceptance for our benchmark points. The focus of our strategy is on heavy flavor; it is possible to perform $b$-tagging at the trigger level, so one might consider a trigger on multiple $b$-tags. Actually, at 8 TeV b-tagging is already used at the trigger level, although more as a last resort [@TomDanielson]. We would highly suggest explicit multi-object triggers with b-tagging at 14 TeV. Alternatively, simpler $H_T$ and $\slashed{E}_T$ triggers help capture the multiple-jet and SUSY nature of our events without the need for intensive online computation. $H_T > 500$ GeV has a high efficiency for our benchmark points at 14 TeV; $\slashed{E}_T$ has a lower efficiency, but a modest requirement can help reduce much of the QCD background. As a comparison, at 8 TeV hadronic $p_T$ sum (300 GeV) + MET requirement (100 GeV) is assumed in the CMS experiment [@CMS:trigger]. But, pileup potentially may make these simple triggers less robust. For the first benchmark point, we apply the following cuts at 14 TeV: (1) at least 6 jets with $|\eta| < 2.8$, (2) at least 4 $b$ tagged jets, at least one with $p_{T} > 30$ GeV, (3) $\slashed E _T > 150$ GeV, and (4) $H_T > 500$ GeV. For the second benchmark point, we amend the last two cuts to (3) $\slashed E _T > 120$ GeV, and (4) $H_T > 650$ GeV. In addition to the QCD multiple-jets, the $4b$-tag requirement can also efficiently remove the $t\bar t b\bar b$ background. The two non-top $b$ quarks in $t\bar t b\bar b$ events are mainly generated by gluon-splitting. This induces these events to fail the $4b$-tag cut for two reasons. The first one is at the level of MC truth. The $b$ quarks from gluon splitting tend to be softer than others, and so recoil more dramatically during parton showering. This makes the hard-process $b$-quarks less aligned with the resulting jet, leading to a larger failure rate in parton-jet flavor matching. In addition, $b$-quarks from gluon splitting tend to be more collimated, the collimation increasing with the $p_T$ of the $b\bar{b}$ pair; this can be seen at parton level in Figure \[Gsplit\]. This effect causes the resulting jets to overlap and be reconstructed and tagged as a single $b$ jet. These effects are implicitly indicated in the cut flows of Table \[cutflow\].
![$p_T$ vs. $\Delta R$ for non-top $b\bar{b}$ quark pairs in $t\bar{t}b\bar{b}$ events.[]{data-label="Gsplit"}](gluon-split-pt-dr.pdf){width="30.00000%"}
Next, we reconstruct the Higgs resonance from the $b$-jets in the remaining events. At this stage, we encounter a combinatorial problem. Given that this is a dedicated resonance search, we define the signal sample as the correctly reconstructed Higgs bosons or the correctly selected $b$-jet pairs. If a wrong $b$-jet pair is selected as the Higgs candidate for a Higgs event, even if its invariant mass falls into the yet-to-be-known Higgs mass bins, we will count it as “combinatorial background" instead of a particle signal. This is different from the definition of the signal sample in the studies on the SM Higgs search via the $tth$ process [@Aad:2009wy; @:ATLAS-CONF-2012-135] and the SUSY non-standard Higgs search via the processes associated with additional $b$-jets [@Dai:1994vu], where the Higgs event with a wrongly selected $b$-jet pair as the Higgs candidate, if the $b$-jet pair has a “correct" invariant mass, was counted as signal instead of background.
The local significance of a resonance is simplest to interpret when only one dijet per event lies in the local mass window $\Delta m_h$, but this method effectively may result in the loss of some correct $b$-pairs. Multiple pairs can be chosen per event as long as their invariant masses differ by more than $\Delta m_h$. To achieve this, we rank the jet pairs according to various variables, always keeping the first-ranked pair. The second-ranked pair is included if $|m_1 - m_2| > \Delta m_h$, and the $n^{th}$-ranked pair is included if $|m_i - m_n| > \Delta m_h$ with $i$ running over all $b$-jet pairs included before. $\Delta m_h$ is mainly controlled by the resolution for jet reconstruction at the LHC - in our analysis, $\Delta m_h = 40$ GeV is assumed. Though an event may contribute more than one pair of $b$-jets for the resonance reconstruction, within the mass window all $b$-jet pairs are from different events.
![$p_T$ vs. $\theta_{\rm eff}$ ranking for correct (left) and wrong (right) $b$-jet pairs falling inside the mass window $[100, 140]$ GeV in signal events, Case I at 14 TeV.[]{data-label="rank"}](plane-higgs-light.pdf "fig:"){width="23.00000%"} ![$p_T$ vs. $\theta_{\rm eff}$ ranking for correct (left) and wrong (right) $b$-jet pairs falling inside the mass window $[100, 140]$ GeV in signal events, Case I at 14 TeV.[]{data-label="rank"}](plane-wrong-light.pdf "fig:"){width="23.00000%"}
To rank the $b$-jet pairs in each event, the kinematics of jet pairs and jet superstructure [@Gallicchio:2010sw] are employed. We use the $p_T$ of each $b$-jet pair, as well as the ‘pull angle’ of the pair. The ‘pull’ of a jet is the vector in the $y-\varphi$ plane defined by $\vec{t} = \sum_{i\in\textrm{jet}} \frac{p_T^i |r_i|}{p_T^{\textrm{jet}}}\vec{r}_i$ where $\vec{r}_i = (\Delta y_i, \Delta\varphi_i)$ is the displacement of the $i^{th}$ jet component from the jet axis. $\vec{t}$ is a measure of the hadronic energy gradient within the jet, and carries information about how the jet’s ancestral parton hadronized. In particular, a pair of jets originating from quarks pair-produced by a color singlet tend to hadronize together, so the jet pulls tend to point toward each other. For a pair of selected $b$-jets with transverse momenta $p_T^{b_1, b_2}$, we define an effective “pull angle" $\theta_{\rm eff}$ by using the pull angles $\theta_t^{b_1,b_2}$ which the pulls of the two $b$-jets make with the chord joining the two jets in the $y-\varphi$ plane: $\theta_{\rm eff} = \left( ( \theta_t^{b_1} / \sigma(p_T^{b_1}) )^{2} + ( \theta_t^{b_2} / \sigma(p_T^{b_2}) )^{2} \right)^{1/2}$. Here $\sigma(p_T^{b_i}) = a p_T^{b_{i}} + b$ reflects the jet $p_T$ dependence of the standard deviation of the pull angle for the two $b$-jets produced from Higgs decay, as noticed in [@Gallicchio:2010sw]. For the kinematic regime that we are considering, $a=-1$ TeV$^{-1}$ and $b=1.5$ are assumed.
If the $b$-jet pairs in each event are ranked according to these variables, noticeable differences emerge. Figure \[rank\] shows the $p_T$ vs. $\theta_{\rm eff}$ ranks of correct and wrong pairs in Case I at the 14 TeV LHC. The distribution of pairs in the ranking plane can be used to improve the Higgs search sensitivity.
$\sqrt{s}=14$ TeV $t\bar{t}$+jets $t\bar{t}b\bar{b}$ Case I Case II
------------------- ----------------- -------------------- -------- ---------
Events $5.2\times10^7$ $8.2\times10^5$ 26176 822275
Cut 1 $3.5\times10^7$ 474234 20600 406296
Cut 2 88700 12077 961 790
Cut 3 51 / 79 442 / 796 567 411
Cut 4 29 / 23 351 / 366 547 361
Choice A 20 / 11 5+157 / 4+157 99+215 76+126
Choice B 20 / 12 4+166 / 4+159 91+219 95+116
Choice C 13 / 13 5+104 / 3+104 78+147 71+65
Choice D 19 / 22 2+189 / 4+159 89+322 68+239
: Cut flows for the benchmark point at 14 TeV. The cut flow of some SM events is labeled as ‘Case I / Case II’, where different cuts are used at the same stage. $t\bar{t}b\bar{b}$ includes QCD, $t\bar{t}Z$ and $t\bar{t}h$. Choice A, B, C, D rows correspond to the $\theta_{\rm eff}$, $p_T$ and $p_T-\theta_{\rm eff}$ plane, and min$(|m_{bb}-m_h|)$ pair selection methods, respectively. Bins labeled as (N + M) show (Higgs + wrong) $b$-jet pairs with $100 {\rm \ GeV}<m_{bb}<140 {\rm \ GeV}$.[]{data-label="cutflow"}
![B-jet pair invariant mass, Case I at 14 TeV. \[invmass1\]](invmass-light.pdf){width="32.00000%"}
![Signal significance using various methods of choosing pairs, Case I at 14 TeV, with 40 fb$^{-1}$ of data.[]{data-label="significance1"}](sensitivity-light.pdf){width="30.00000%"}
![B-jet pair invariant mass, Case II at 14 TeV. \[invmass2\]](invmass-heavy.pdf){width="32.00000%"}
![Signal significance using various methods of choosing pairs, Case II at 14 TeV, with 40 fb$^{-1}$ of data.[]{data-label="significance2"}](sensitivity-heavy.pdf){width="30.00000%"}
Complete cut flows of both benchmark points at 14 TeV are presented in table \[cutflow\]. The number of events have been rescaled to the cross section of each process. A first observation is a relatively large sensitivity to our chosen benchmark points. From Table \[cutflow\] the sensitivities in both cases are $\gtrsim 20 \sigma$ for 40fb$^{-1}$ data at the 14 TeV LHC. Again, note that the contribution labeled “SUSY: wrong pair" in these figures is a ‘background’ only for the Higgs search, but indicates a large signal for the discovery of SUSY itself. To combine the $p_T$ and $\theta_{\rm eff}$ ranking strategies, we require that the selected pairs fall within the triangular region where rank($p_T$) + rank ($\theta_{\rm eff}$) $\leq$ 5. This strategy is used for the invariant mass plots of the selected $b$-jet pairs in Figure \[invmass1\] and Figure \[invmass2\], for the SM-like Higgs candidates. The local significances for different strategies of the resonance reconstruction are shown in Figure \[significance1\] and Figure \[significance2\]. For the best option, the correctly paired Higgs $b$-jets give a local significance of $> 5\sigma$ in both cases for 40 fb$^{-1}$ of data, with the local mass window taken to be 100 GeV to 140 GeV. Though rank($p_T$) + rank ($\theta_{\rm eff}$) $\leq$ 5 does not increase the values of $S/\sqrt{B}$ significantly, it improves the $S/B$ to be above 0.3 and 0.4 for the two cases, respectively. This can potentially decrease the impact of systematic uncertainties. Each of these strategies is superior to naively selecting the $b$ pair with invariant mass closest to 125 GeV; for both benchmark points, this naive choice gives a sensitivity of $\sim4\sigma$ in Case I and $\sim3.5\sigma$ in Case II, with $S/B\sim0.17$ in both cases.
These results may be improved. Our $b$-tagging efficiency is rather conservative [@:ATLAS-CONF-2011-102]. If the b-tagging efficiency is assumed to be 0.7, similar to the approach in [@Plehn:2009rk; @:ATL-PHYS-PUB-2009-088], then the significances for various strategies are expected to be universally increased by a factor $\sim 1.5$.
Although the foregoing discussion is confined within the supersymmetric scenarios with relatively light top partners - its impact is fairly profound. It provides an independent way to discover the SM-like Higgs boson and to understand the mass origin of the SM fermions. In turn, searching for the Higgs boson using this strategy provides a direct way to test SUSY. In addition, the potential applicable scope of this strategy is broad. The first class of examples are some non-SUSY theories, such as little Higgs models [@ArkaniHamed:2001nc], composite Higgs models [@Contino:2003ve], and Randall-Sundrum models [@Randall:1999ee], where fermionic top partners $t'$ with a mass below 1 TeV are typically predicted [@Matsedonskyi:2012ym], given a 125 GeV SM-like Higgs boson, and the Higgs boson can be produced via $t' \to h t$, a way similar to the second mechanism discussed above. Actually, multiple $b$-tagging has been noticed to be useful in this context [@AguilarSaavedra:2006gw], while the Higgs reconstruction was thought to be a big problem due to the combinatorial background [@Cacciapaglia:2011fx]. Another class of examples are the SM Higgs search via the $t\bar th$ production and the non-standard Higgs ($H, A$) search via the $b\bar bH$ or $b\bar b A$ production in the MSSM. Though the involved kinematics are different, given $h, H, A \to b\bar b$, these analyses are expected to share the feature of suppressed $4j$ and $t\bar t, b\bar b + 2j$ background, leaving the continuum $t\bar t b\bar b$ or $b\bar b b\bar b$ as the main one. Then the Higgs resonance can be reconstructed with color-flow or kinematic variables. We leave the consideration of these interesting possibilities to future work.
[**Acknowledgements**]{}
We would like to thank T. Danielson, J. Gallicchio, D. Krohn, A. Menon, D. Morrissey, J. Richman, M. Schwartz, J. Shelton, J. Shu, S. Su, N. Toro, and L.T. Wang for useful discussions. Work is supported in part by DOE under grant DE-FG02-91ER40618.
[99]{}
S. Chatrchyan [*et al.*]{} \[CMS Collaboration\], Phys. Lett. B [**716**]{}, 30 (2012) \[arXiv:1207.7235 \[hep-ex\]\]. G. Aad [*et al.*]{} \[ATLAS Collaboration\], Phys. Lett. B [**716**]{}, 1 (2012) \[arXiv:1207.7214 \[hep-ex\]\]. J. M. Butterworth, A. R. Davison, M. Rubin and G. P. Salam, Phys. Rev. Lett. [**100**]{}, 242001 (2008) \[arXiv:0802.2470 \[hep-ph\]\]. T. Plehn, G. P. Salam and M. Spannowsky, Phys. Rev. Lett. [**104**]{}, 111801 (2010) \[arXiv:0910.5472 \[hep-ph\]\]. M. Papucci, J. T. Ruderman and A. Weiler, arXiv:1110.6926 \[hep-ph\]. C. Brust, A. Katz, S. Lawrence and R. Sundrum, JHEP [**1203**]{}, 103 (2012) \[arXiv:1110.6670 \[hep-ph\]\]; M. Asano, H. D. Kim, R. Kitano and Y. Shimizu, JHEP [**1012**]{}, 019 (2010) \[arXiv:1010.0692 \[hep-ph\]\].
H. Baer, M. Bisset, X. Tata and J. Woodside, Phys. Rev. D [**46**]{}, 303 (1992); A. Djouadi, J. L. Kneur and G. Moultaka, Phys. Rev. Lett. [**80**]{}, 1830 (1998) \[hep-ph/9711244\]; A. Djouadi, J. L. Kneur and G. Moultaka, Nucl. Phys. B [**569**]{}, 53 (2000) \[hep-ph/9903218\]; A. Datta, A. Djouadi, M. Guchait and Y. Mambrini, Phys. Rev. D [**65**]{}, 015007 (2002) \[hep-ph/0107271\]; A. Datta, A. Djouadi, M. Guchait and F. Moortgat, Nucl. Phys. B [**681**]{}, 31 (2004) \[hep-ph/0303095\]; P. Bandyopadhyay, A. Datta and B. Mukhopadhyaya, Phys. Lett. B [**670**]{}, 5 (2008) \[arXiv:0806.2367 \[hep-ph\]\]; K. Huitu, R. Kinnunen, J. Laamanen, S. Lehti, S. Roy and T. Salminen, Eur. Phys. J. C [**58**]{}, 591 (2008) \[arXiv:0808.3094 \[hep-ph\]\]; P. Bandyopadhyay, JHEP [**1108**]{}, 016 (2011) \[arXiv:1008.3339 \[hep-ph\]\]. M. Asano, H. D. Kim, R. Kitano and Y. Shimizu, JHEP [**1012**]{}, 019 (2010) \[arXiv:1010.0692 \[hep-ph\]\]. O. Stal and G. Weiglein, JHEP [**1201**]{}, 071 (2012) \[arXiv:1108.0595 \[hep-ph\]\]. D. Ghosh, M. Guchait and D. Sengupta, Eur. Phys. J. C [**72**]{}, 2141 (2012) \[arXiv:1202.4937 \[hep-ph\]\].
S. Gori, P. Schwaller and C. E. M. Wagner, Phys. Rev. D [**83**]{}, 115022 (2011) \[arXiv:1103.4138 \[hep-ph\]\]. G. D. Kribs, A. Martin, T. S. Roy and M. Spannowsky, Phys. Rev. D [**81**]{}, 111501 (2010) \[arXiv:0912.4731 \[hep-ph\]\]; Phys. Rev. D [**82**]{}, 095012 (2010) \[arXiv:1006.1656 \[hep-ph\]\]. A. Djouadi, M. M. Muhlleitner and M. Spira, Acta Phys. Polon. B [**38**]{}, 635 (2007) \[hep-ph/0609292\]. L. J. Hall, D. Pinner and J. T. Ruderman, JHEP [**1204**]{}, 131 (2012) \[arXiv:1112.2703 \[hep-ph\]\]; U. Ellwanger, JHEP [**1203**]{}, 044 (2012) \[arXiv:1112.3548 \[hep-ph\]\]; H. An, T. Liu and L. -T. Wang, Phys. Rev. D [**86**]{}, 075030 (2012) \[arXiv:1207.2473 \[hep-ph\]\].
N. Arkani-Hamed, P. Schuster, N. Toro, J. Thaler, L. -T. Wang, B. Knuteson and S. Mrenna, hep-ph/0703088 \[HEP-PH\]. CMS Note, CMS-PAS-SUS-11-024
ATLAS Note, ATLAS-CONF-2012-106
S. Chatrchyan [*et al.*]{} \[CMS Collaboration\], arXiv:1208.4859 \[hep-ex\].
W. Beenakker, M. Kramer, T. Plehn, M. Spira and P. M. Zerwas, Nucl. Phys. B [**515**]{}, 3 (1998) \[hep-ph/9710451\]. A. Bredenstein, A. Denner, S. Dittmaier and S. Pozzorini, Phys. Rev. Lett. [**103**]{}, 012002 (2009) \[arXiv:0905.0110 \[hep-ph\]\]. S. Dawson, C. Jackson, L. H. Orr, L. Reina and D. Wackeroth, Phys. Rev. D [**68**]{}, 034022 (2003) \[hep-ph/0305087\]. A. Lazopoulos, T. McElmurry, K. Melnikov and F. Petriello, Phys. Lett. B [**666**]{}, 62 (2008) \[arXiv:0804.2220 \[hep-ph\]\]. J. Alwall, M. Herquet, F. Maltoni, O. Mattelaer and T. Stelzer, JHEP [**1106**]{}, 128 (2011) \[arXiv:1106.0522 \[hep-ph\]\]. T. Sjostrand, S. Mrenna and P. Z. Skands, Comput. Phys. Commun. [**178**]{}, 852 (2008) \[arXiv:0710.3820 \[hep-ph\]\]. S. Ovyn, X. Rouby and V. Lemaitre, arXiv:0903.2225 \[hep-ph\]. M. Cacciari, G. P. Salam and G. Soyez, arXiv:1111.6097 \[hep-ph\]. ATLAS Note, ATLAS-CONF-2011-089.
G. Aad [*et al.*]{} \[ATLAS Collaboration\], arXiv:0901.0512 \[hep-ex\]; G. L. Bayatian et al. \[CMS Collaboration\], J. Phys. G 34, 995 (2007).
ATLAS Note, ATLAS-CONF-2012-135.
CMS Note, CMS-PAS-HIG-12-025.
Private communications with Tom Danielson.
<http://j2eeps.cern.ch/cms-project-confdb-hltdev/browser/>
J. Dai, J. F. Gunion and R. Vega, Phys. Lett. B [**345**]{}, 29 (1995) \[hep-ph/9403362\]; C. Balazs, J. L. Diaz-Cruz, H. J. He, T. M. P. Tait and C. P. Yuan, Phys. Rev. D [**59**]{}, 055016 (1999) \[hep-ph/9807349\]; C. Kao, S. Sachithanandam, J. Sayre and Y. Wang, Phys. Lett. B [**682**]{}, 291 (2009) \[arXiv:0908.1156 \[hep-ph\]\]; M. Carena, S. Gori, A. Juste, A. Menon, C. E. M. Wagner and L. -T. Wang, JHEP [**1207**]{}, 091 (2012) \[arXiv:1203.1041 \[hep-ph\]\];
J. Gallicchio and M. D. Schwartz, Phys. Rev. Lett. [**105**]{}, 022001 (2010) \[arXiv:1001.5027 \[hep-ph\]\]. ATLAS Collaboration, ATLAS-CONF-2011-102 (2011), https://cdsweb.cern.ch/record/1369219.
ATLAS Note, ATL-PHYS-PUB-2009-088.
J. Gallicchio, J. Huth, M. Kagan, M. D. Schwartz, K. Black and B. Tweedie, “Multivariate discrimination and the Higgs + W/Z search,” JHEP [**1104**]{}, 069 (2011) \[arXiv:1010.3698 \[hep-ph\]\]. N. Arkani-Hamed, A. G. Cohen and H. Georgi, Phys. Lett. B [**513**]{}, 232 (2001) \[hep-ph/0105239\]; N. Arkani-Hamed, A. G. Cohen, E. Katz, A. E. Nelson, T. Gregoire and J. G. Wacker, JHEP [**0208**]{}, 021 (2002) \[hep-ph/0206020\]; N. Arkani-Hamed, A. G. Cohen, E. Katz and A. E. Nelson, JHEP [**0207**]{}, 034 (2002) \[hep-ph/0206021\]. R. Contino, Y. Nomura and A. Pomarol, Nucl. Phys. B [**671**]{}, 148 (2003) \[hep-ph/0306259\]; K. Agashe, R. Contino and A. Pomarol, Nucl. Phys. B [**719**]{}, 165 (2005) \[hep-ph/0412089\]; R. Contino, L. Da Rold and A. Pomarol, Phys. Rev. D [**75**]{}, 055014 (2007) \[hep-ph/0612048\]. L. Randall and R. Sundrum, Phys. Rev. Lett. [**83**]{}, 3370 (1999) \[hep-ph/9905221\]; K. Agashe, A. Delgado, M. J. May and R. Sundrum, JHEP [**0308**]{}, 050 (2003) \[hep-ph/0308036\]. O. Matsedonskyi, G. Panico and A. Wulzer, arXiv:1204.6333 \[hep-ph\]; J. Berger, J. Hubisz and M. Perelstein, JHEP [**1207**]{}, 016 (2012) \[arXiv:1205.0013 \[hep-ph\]\]; M. Redi and A. Tesi, JHEP [**10**]{}, 166 (2012) \[arXiv:1205.0232 \[hep-ph\]\]; D. Marzocca, M. Serone and J. Shu, JHEP [**1208**]{}, 013 (2012) \[arXiv:1205.0770 \[hep-ph\]\]; G. Panico, M. Redi, A. Tesi and A. Wulzer, arXiv:1210.7114 \[hep-ph\]. J. A. Aguilar-Saavedra, JHEP [**0612**]{}, 033 (2006) \[hep-ph/0603200\]; K. Harigaya, S. Matsumoto, M. M. Nojiri and K. Tobioka, arXiv:1204.2317 \[hep-ph\]. G. Cacciapaglia, A. Deandrea, L. Panizzi, N. Gaur, D. Harada and Y. Okada, JHEP [**1203**]{}, 070 (2012) \[arXiv:1108.6329 \[hep-ph\]\].
| {
"pile_set_name": "ArXiv"
} |
---
author:
- |
, for the NOvA Collaboration\
Tufts University\
E-mail:
title: Impact of cross section uncertainties on NOvA oscillation analyses
---
Introduction
============
Neutrino oscillation experiments rely heavily on predictions from Monte Carlo simulations to infer the parameters of interest from their data. Among the most challenging components of the simulation chain for such experiments is typically a neutrino interaction generator, which predicts the rates of neutrino reactions in detector materials as well as the identities and four-momenta of reactions’ outgoing particles. The NOvA experiment, which is a long-baseline neutrino oscillation experiment based at Fermilab in Batavia, IL, currently employs the GENIE generator [@genie] (version 2.12.2) to simulate neutrino interactions in its near detector (ND) at Fermilab and its far detector (FD) in Ash River, MN.
Historically, generators used by experiments, such as GENIE, have made concessions to the difficult task of predicting interactions with the complex nuclear environment by adopting a “factorization” approach. In this scheme, numerous theoretical models for hard-scattering processes from hadrons or quarks at various momentum scales are composed with a relatively simple model for the nuclear dynamics. Though this picture has been sufficient for past work, increasing statistical precision in modern experiments has begun to reveal cracks in the foundation. In particular, dedicated measurements from neutrino scattering experiments (e.g., [@miniboone-qe-1; @miniboone-qe-2; @minerva-lowq3-1; @minerva-lowq3-2; @minerva-xverse-vars; @t2k-incl; @t2k-xverse-vars]) have cast considerable doubt on whether the non-interacting relativistic Fermi gas (RFG) nuclear model used by default in contemporary versions of GENIE is viable. Nontrivial uncertainty is associated with both the details of the nuclear model and the hard scattering processes themselves, which rely on approximations where explicit non-perturbative calculations using QCD are untenable.
While NOvA is designed with the mitigation of these cross section uncertainties in mind, no present or planned experiment is completely insensitive to them. In the following sections we explore the uncertainties noted above, including adjustments to GENIE’s model we find we are forced to make by external data, improvements to available theory, and our own ND data. We then discuss the impact they have on NOvA’s $\nu_{\mu}$ disappearance and $\nu_{e}$ appearance measurements, given NOvA’s design—the detectors are built to be as similar as possible in materials and technologies—and the calorimetric energy reconstruction principle used in the analyses.
GENIE 2.12.2 model and adjustments
==================================
GENIE 2.12.2’s default model divides the total hard-scattering cross section into numerous processes for which independent models exist. The largest ones (and the only ones we will consider here) in charged-current (CC) interactions are, in order of increasing final-state hadronic mass $W$, quasielastic scattering (QE), resonant baryon production (RES), and nonresonant deep inelastic scattering (DIS). In current analyses, NOvA makes adjustments to the axial mass in the QE dipole form factor (setting $M_A = \unit[1.04]{GeV}$ rather than the default $\unit[0.99]{GeV}$) and nonresonant single pion production with $W < \unit[2.0]{GeV}$ (reducing it to 43% of its nominal value) based on reanalysis of bubble chamber data that these parameters were originally tuned to [@bubble-qe; @bubble-nonres-reana]. These make relatively small differences in the prediction. A much larger impact comes from the addition of a new hard-scattering process, that of two-nucleon ejection via a meson-exchange current (MEC) process.[@nustec-whitepaper] Because this is a reaction well known from electron scattering, but no contemporary model is able to describe the extant neutrino data[@minerva-lowq3-1; @minerva-lowq3-2; @minerva-xverse-vars; @t2k-xverse-vars], NOvA has elected to enable the optional “Empirical MEC” model in GENIE[@katori-empMEC] and tune it to NOvA ND data in energy- and three-momentum transfer $(q_0, |\vec{q}|)$. Comparisons of the default and tuned predictions to NOvA ND data, as well as the uncertainties constructed from alternative tunes, and the outcome of a similar procedure performed by the MINERvA Collaboration to their own data[@minerva-mec-tune], are shown in fig. \[fig:MEC tuning\].
[0.45]{}
[0.45]{}
As noted above, the most precarious component of the current generator prediction is the nuclear model. NOvA alters the default GENIE 2.12.2 model in several ways to address shortcomings here. First, there is widespread agreement that long-range interactions of the nuclear potential between nucleons affect QE reactions, significantly suppressing them at low $Q^2$ and mildly enhancing them at higher $Q^2$ relative to the RFG prediction. We adopt the random phase approximation (RPA)-based calculation of the València group [@valencia-rpa] parameterized as reweights in $(q_0, |\vec{q}|)$ to the GENIE QE model by R. Gran [@gran-rpa] and the associated uncertainties. Measurements of delta resonant production in external data [@miniboone-res; @minos-qe; @minerva-pi-1; @minerva-pi-2], as well as our own ND data, also suggest the presence of nuclear dynamics resulting in a similar suppression at low $Q^2$ relative to the free nucleon prediction as the RPA effect, so we also apply the $Q^2$ parameterization of the RPA effect to RES as a placeholder for whatever the true nuclear effect may be. We take the unmodified RES prediction as an uncertainty variation.
Impact on neutrino oscillation measurements
===========================================
As discussed elsewhere in these Proceedings [@erica-nufact], NOvA uses a calorimetric technique for both $\nu_{\mu}$ and $\nu_e$ energy reconstruction in which neutrino energy is estimated using a function of both lepton and hadronic system energies. Uncertainties in cross section modeling can impact the fidelity of these estimators in a number of different ways: for instance, shifting the balance of energy between the better-resolved leptonic and the more-poorly-resolved hadronic systems in CC events; changing the predicted mean energy that is unseen by the detector (due either to the assumed nuclear binding potential or hadronic energy that escapes as neutrons) and must be added back by the estimator; or adjusting the expected frequency of background processes that have different energy responses than the signal.
To mitigate the impact of these uncertainties on the prediction at the FD, NOvA relies on measurements at the ND, which are propagated to predictions for the FD via an “extrapolation” procedure. The latter supposes that discrepancies observed between ND simulation and data distributions can be accounted for in the FD prediction by modifying the ND true event rate in bins of true energy, which can then be multiplied by the simulated ratio of the geometric and oscillation effects between the two detectors to yield the FD true rate. This is conveniently expressed as a matrix equation over the energy bins: $$\vec{N}_{FD} = \vec{N}_{ND}\ \mathbf{R}\ \mathbf{M}_{ND}\ \mathbf{F}\ \mathbf{P}_{osc}\ \mathbf{M}_{FD}^{-1}
\label{eq:fovern}$$ Here, the $\vec{N}_{\alpha}$ are the predicted event yields in bins of reconstructed energy for detector $\alpha$; the diagonal matrix $\mathbf{R}$ contains the bin-by-bin ratios of the observed and predicted ND yields, $R_{ii} = N^{ND}_{\mathrm{data},i} /N^{ND}_{\mathrm{MC},i}$; the $\mathbf{M}_{\alpha}$ are so-called “migration” matrices between reconstructed and true energies for detector $\alpha$, from simulation; the diagonal $\mathbf{F}$ is denoted the “far over near ratio,” $F/N$, which encodes the predicted effect of the neutrino beam dispersion and the difference in acceptance between the detectors; and the diagonal $\mathbf{P}_{osc}$ applies oscillation probabilities for given oscillation parameters. This approach differs from the strategy sometimes employed by other oscillation experiments in which parameters in the model are fitted to the ND data and propagated to the FD prediction via their fitted covariance matrix. While the NOvA strategy is less general (it is only effective when the ND and FD share the same underlying cross section uncertainties, like in NOvA, for instance), it is guaranteed to reproduce the observed ND distribution, even if unknown effects are present in the data that the model cannot account for.
The extent to which the $F/N$ method enables calculation of the effect of changes in the cross section model on the FD prediction using ND data can be illustrated with test cases. In such a test, the ND data is replaced by a modified prediction using a designated cross-section change during the calculation of $\mathbf{R}$, resulting in a modified $\mathbf{R}'$. The $\vec{N}_{FD}'$ obtained from applying eq. \[eq:fovern\] to $\mathbf{R}'$ can then be compared to a different $\vec{N}_{FD}''$ obtained by directly applying the modified cross section model to the FD prediction in simulation. If $\vec{N}_{FD}'$ and $\vec{N}_{FD}''$ coincide, then the extrapolation procedure can perfectly account for the effect of the given cross-section shift using the ND data. If they differ, on the other hand, the residual between $\vec{N}_{FD}''$ (direct FD prediction under shifted model) and $\vec{N}_{FD}'$ (extrapolation of shifted prediction with nominal model) illustrates the fraction of the given shift that is not “canceled” (i.e., is left uncorrelated between the two predictions) by the extrapolation procedure. Fig. \[fig:extrap MEC unc\] shows the comparison resulting from shifts due to two important uncertainties in the MEC model noted above; the extrapolation procedure reduces the original uncertainties of up to 10% to a few percentage points.
[0.45]{}
[0.45]{}
In fits to the FD data, the extrapolation procedure is used first to correct the nominal FD prediction. Known uncertainties are then accounted for using nuisance parameters constructed from bin-by-bin splines fitted to the difference between shifted predictions $\vec{N}_{FD}''$ and the corrected nominal prediction. The reduction of the cross section impact on uncertainties in the $\nu_{e}$ signal and background predictions due to extrapolation is illustrated in fig. \[fig:extrap xsec nue\]. Even after extrapolation is applied, however, neutrino cross section uncertainties retain significant influence on the results, together accounting for 35%, 44%, and 53% of the total systematic error budgets for NOvA’s $\sin^2(\theta_{23})$, $\Delta m_{32}^2$, and $\delta_{CP}$ measurements, respectively. We anticipate that future continued improvements to cross section modeling, particularly in regard to the nuclear dynamics in QE and RES interactions, the detailed nature of 2p2h, and antineutrino reactions, will be essential as the statistical precision of these measurements improves and systematics begin to limit them.
[0.45]{}
[0.45]{}
Conclusions
===========
NOvA relies on strong internal constraints on cross section uncertainties for its oscillation program derived from the functionally identical detector paradigm and a calorimetric neutrino energy reconstruction technique. In addition, a comprehensive program is underway to ensure that all relevant cross section issues are considered. After the constraint from the ND is applied, cross section uncertainties currently comprise 30-50% of the systematic budget on the most important oscillation parameter measurements. We look forward to continued development of models and associated systematic treatments in the community, new measurements of cross sections to help constrain them, and ultimately their integration into improved oscillation parameter measurements in NOvA.
[99]{} C. Andreopoulos [*et al.*]{}, Nucl. Instrum. Meth. A [**614**]{}, 87 (2010);\
C. Andreopoulos [*et al.*]{}, arXiv:1510.05494 \[hep-ph\].
A. A. Aguilar-Arevalo [*et al.*]{} \[MiniBooNE Collaboration\], Phys. Rev. D [**81**]{}, 092005 (2010).
A. A. Aguilar-Arevalo [*et al.*]{} \[MiniBooNE Collaboration\], Phys. Rev. D [**88**]{}, 032001 (2013).
P. A. Rodrigues [*et al.*]{} \[MINERvA Collaboration\], Phys. Rev. Lett. [**116**]{}, 071802 (2016);\
Addendum: \[Phys. Rev. Lett. [**121**]{}, 209902 (2018)\].
R. Gran [*et al.*]{} \[MINERvA Collaboration\], Phys. Rev. Lett. [**120**]{}, 221805 (2018).
X. G. Lu [*et al.*]{} \[MINERvA Collaboration\], Phys. Rev. Lett. [**121**]{}, 022504 (2018).
K. Abe [*et al.*]{} \[T2K Collaboration\], Phys. Rev. D [**98**]{}, 012004 (2018).
K. Abe [*et al.*]{} \[T2K Collaboration\], Phys. Rev. D [**98**]{}, 032003 (2018).
A. S. Meyer, M. Betancourt, R. Gran and R. J. Hill, Phys. Rev. D [**93**]{}, no. 11, 113015 (2016).
P. Rodrigues, C. Wilkinson and K. McFarland, Eur. Phys. J. C [**76**]{}, no. 8, 474 (2016).
MINERvA Collaboration, private communcation, Oct. 11, 2018. Publication in preparation.
For a comprehensive survey, see sec. 6.3 in L. Alvarez-Ruso [*et al.*]{}, Prog. Part. Nucl. Phys. [**100**]{}, 1 (2018).
T. Katori, AIP Conf. Proc. [**1663**]{}, 030001 (2015).
R. Gran, J. Nieves, F. Sanchez and M. J. Vicente Vacas, Phys. Rev. D [**88**]{}, no. 11, 113007 (2013).
R. Gran, arXiv:1705.02932 \[hep-ex\].
A. A. Aguilar-Arevalo [*et al.*]{} \[MiniBooNE Collaboration\], Phys. Rev. D [**83**]{}, 052007 (2011).
P. Adamson [*et al.*]{} \[MINOS Collaboration\], Phys. Rev. D [**91**]{}, no. 1, 012005 (2015).
C. L. McGivern [*et al.*]{} \[MINERvA Collaboration\], Phys. Rev. D [**94**]{}, no. 5, 052005 (2016).
O. Altinok [*et al.*]{} \[MINERvA Collaboration\], Phys. Rev. D [**96**]{}, no. 7, 072003 (2017).
E. Smith \[NOvA Collaboration\], PoS NUFACT [**2018**]{}, 068 (2018).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Some recent developments in the analysis of long-time behaviors of stochastic solutions of nonlinear conservation laws driven by stochastic forcing are surveyed. The existence and uniqueness of invariant measures are established for anisotropic degenerate parabolic-hyperbolic conservation laws of second-order driven by white noises. Some further developments, problems, and challenges in this direction are also discussed.'
address:
- 'Gui-Qiang G. Chen, Mathematical Institute, University of Oxford, Oxford, OX2 6GG, UK'
- 'Peter Pang, Mathematical Institute, University of Oxford, Oxford, OX2 6GG, UK'
author:
- 'Gui-Qiang G. Chen Peter H.C. Pang'
title: Invariant Measures for Nonlinear Conservation Laws Driven by Stochastic Forcing
---
Introduction
============
The analysis of long-time behaviors of global solutions is the second pillar in the theory of partial differential equations (PDEs), after the analysis of well-posedness. For the analysis of solution behaviors in the asymptotic regime, we seek to understand the global properties of the solution map, such as attracting or repelling sets, stable and unstable fixed points, limiting cycles, or chaotic behaviors that are properly determined by the entire system rather than a given path.
The introduction of noises usually serves to model dynamics phenomenologically – dynamics too complicated to model from first principles, or dynamics only the statistics of which are accurately known, or dynamics almost inherently random such as the decision of many conscious agents – or a combination of such behaviors. Mathematically, noises introduce behaviors that differ from deterministic dynamics, displaying much richer phenomena such as effects of dissipation, ergodicity, among others ([*cf.*]{} [@Ver1981; @FGP2010; @DFV2014; @FFPV2017; @KR2005; @DFPR2013] and the references cited therein). These phenomena are of intrinsic interest.
In this paper, we focus our analysis mainly on white-in-time noises. Indeed, they are the most commonly studied class of noises, though space-time white noises (such as in [@DD2002]) and more general rough fluxes ([*e.g.*]{} [@LPS2013; @LPS2014; @PZ2007]) have also been considered. The reason for the prevalence of white noises as a basic model is not difficult to understand. First, Brownian motion occupies the unusual position of being simultaneously a martingale and a Lévy process. More importantly, with increments that are not only independent but also normally distributed, it commands a level of universality by virtue of the central limit theorem. Some of the ideas, techniques, and approaches presented here can be applied to equations with more general or other forms of stochastic forcing.
The organization of this paper is as follows: In §2, the notion of invariant measures is first introduced, then the Krylov-Bogoliubov approach for the existence of invariant measures is presented, and some methods for the uniqueness of invariant measures including the strong Feller property and the coupling method are discussed. In §3, some recent developments in the analysis of long-time behaviors of solutions of nonlinear stochastic PDEs are discussed. In §4, we establish the existence of invariant measures for nonlinear anisotropic parabolic-hyperbolic equations driven by white noises. In §5, we establish the uniqueness of invariant measures for the stochastic anisotropic parabolic-hyperbolic equations. In §6, we present some further developments, problems, and challenges in this research direction.
Invariant Measures
==================
In this section, we first introduce the notion of invariant measures for random dynamic systems, and then present several approaches to establish the existence and uniqueness of invariant measures.
Notion of invariant measures
----------------------------
The notion of invariant measures on a dynamical system is quite straightforward. Let $({\mathfrak{X}},\Sigma,\mu)$ be a measure space, and let $S: {\mathfrak{X}}\to {\mathfrak{X}}$ be a map. System $({\mathfrak{X}},\Sigma,\mu,S)$ is a measure-preserving system if $\mu(S^{-1}A) = \mu(A)$ for any $A \in \Sigma$. Then $\mu$ is called an invariant measure of map $S$.
On a random dynamic system (RDS), there is an added layer of complexity. We follow the standard definitions in [@Arn1998]; see also [@Cra2002] for further references on RDSs and [@Fla1991; @Fla1995] in a specifically parabolic SPDE context.
Let $(\Omega, \mathcal{F}, \mathbb{P})$ be a probability space, and let $\theta_t: \Omega \to \Omega$ be a collection of probability-preserving maps. A [*measurable RDS*]{} on a measurable space $({\mathfrak{X}},\Sigma)$ over quadruple $(\Omega, \mathcal{F}, \mathbb{P}, \theta_t)$ is a map: $$\varphi\,:\, \mathbb{R} \times \Omega \times {\mathfrak{X}}\to {\mathfrak{X}}$$ satisfying the following:
- Measurability: $\,\varphi$ is $(\mathcal{B}(\mathbb{R})\otimes \mathcal{F}\otimes \Sigma, \Sigma)$–measurable[;]{}
- Cocycle property: $\,\varphi(t,\omega) = \varphi(t,\omega,\cdot): {\mathfrak{X}}\to {\mathfrak{X}}$ is a cocyle over $\theta$: $$\begin{aligned}
&\varphi(0,\omega) = \mathrm{id}_{\mathfrak{X}},\\
&\varphi(t + s,\omega) = \varphi(s,\theta_t \omega) \varphi(t,\omega),\end{aligned}$$ where $\mathcal{B}(\mathbb{R})$ denotes the collection of Borel sets in $\mathbb{R}$.
We think of $\Omega \times {\mathfrak{X}}\to \Omega$ as a fibre bundle with fibre ${\mathfrak{X}}$. On the bundle, we have the [*skew product*]{} defined as $\Theta_t = (\theta_t,\varphi)$. Then the invariant measures can be defined as follows:
An [*invariant measure*]{} on a RDS $\varphi$ over $\theta_t$ is a probability measure $\mu$ on $(\Omega \times {\mathfrak{X}}, \mathcal{F}\otimes \Sigma)$ satisfying $$(\Theta_t)_*\mu = \mu,\quad\; \mu(\cdot,{\mathfrak{X}}) = \mathbb{P},$$ where $(\Theta_t)_*\mu:=\mu\circ (\Theta_t)^{-1}$ is the pushforward measure.
Any probability measure $\mu$ on $\Omega \times {\mathfrak{X}}$ admits a disintegration: $$\mu(\omega, u) = \nu_\omega(u) \mathbb{P}(\omega).$$ A measure $\nu_\omega$ is [*stationary*]{} if $$\varphi(t,\omega)_*\nu_\omega = \nu_{\theta_t\omega}.$$ Let $\{\mathscr{F}_t\}_{t \ge 0}$ be a filtration associated with the RDS $\varphi$, [*i.e.*]{} an increasing sequence of $\sigma$–sub-algebras of $\mathscr{F}$ by which $\varphi(t,\cdot,x)$ is measurable (adapted). A [*Markov*]{} invariant measure is an invariant measure for which map: $\omega \mapsto \nu_\omega(\Gamma)$ is $(\mathscr{F}_0,\mathcal{B}(\mathbb{R}))$–measurable for any $\Gamma \in \mathcal{B}({\mathfrak{X}})$ [@KS2012 §4.2.1].
The disintegration of measures is unique. There is a one-to-one correspondence between a Markov invariant measure and a stationary measure [@Cra1991; @Led1986; @LeJ1987]. Associated with an invariant measure is a random attracting set, which is generalized from the deterministic context ([*cf.*]{} [@CF1994; @CDF1997]). In the context of dissipative PDEs perturbed by noises, it can be shown that the Hausdorff dimension of an attracting set is finite by the methods similar to those used in the deterministic case (see [@Tem1997] and the references cited therein) of linearizing the flow and estimating the sums of global Lyapunov exponents ([*cf*]{}. [@CF1994; @Sch1997; @Deb1997; @Deb1998]).
Approaches for the existence of invariant measures {#sec:krylov_bogoliubov}
--------------------------------------------------
There are several approaches to establish the existence of invariant measures. One of the approaches is the Krylov-Bogoliubov approach, as we are going to discuss here. Another approach is via Khasminskii’s theorem [@DZ2008]. Both of them are based on the compactness property provided by the Prohorov theorem.
We first recall that a sequence of probability measures $\{\nu_n\}$ on a measure space ${\mathfrak{X}}$ is *tight* if, for every ${\epsilon}> 0$, there is a compact set $K_{\epsilon}\subseteq {\mathfrak{X}}$ such that $$\nu_n({\mathfrak{X}}\backslash K_{\epsilon}) \leq {\epsilon}\qquad\,\,\, \mbox{uniformly in $n$}.$$
A tight sequence of probability measures $\nu_n$ is weak\*–compact in the space of *probability* measures[;]{} that is, there exist a subsequence [(]{}still denoted[)]{} $\nu_n$ and a probability measure $\nu$ such that $\nu_n \overset{*}{\rightharpoonup} \nu$.
\[thm:KrylovBogoliubov\] Let $\mathscr{P}_t$ be a semigroup satisfying the [*F*eller property]{} that $\phi \in C({\mathfrak{X}})$ implies $\mathscr{P}_s\phi \in C({\mathfrak{X}})$ for any $s > 0$, and let $\mu$ be a probability measure on ${\mathfrak{X}}$ such that the measure sequence[*:*]{} $$\label{2.1-a}
\nu_T = \frac{1}{T} \int_0^T \mathscr{P}_t^*\mu\;{{\rm d}}t$$ is tight. Then there exists an invariant measure for $\mathscr{P}_t$.
The key of its proof is that the invariant measure generated by the Krylov-Bogoliubov theorem is the weak\*–limit of $\nu_T$ as $T\to \infty$. This can be seen as follows: By the Prohorov theorem, the tight sequence has a weakly converging subsequence (still denoted as) $\{\nu_T\}$ for $T$ ranging over a unbounded subset of $\mathbb{R}$, whose limit is $\nu_*$. Then $$\begin{aligned}
{\langle}\mathscr{P}^*_s \nu_*, {\varphi}{\rangle}_{{\mathfrak{X}}}
= &\, {\langle}\nu_*, \mathscr{P}_s{\varphi}{\rangle}_{\mathfrak{X}}\\
= &\,\lim_{T \to \infty} \frac{1}{T}\int_0^T {\langle}\mathscr{P}^*_t \mu, \mathscr{P}_s{\varphi}{\rangle}_{\mathfrak{X}}\;{{\rm d}}t \\
= &\, \lim_{T \to \infty} \frac{1}{T}\int_0^T {\langle}\mathscr{P}^*_{t+s} \mu, {\varphi}{\rangle}_{\mathfrak{X}}\;{{\rm d}}t \\
= &\, \lim_{T \to \infty} \frac{1}{T}\int_0^T {\langle}\mathscr{P}^*_{t} \mu, {\varphi}{\rangle}_{\mathfrak{X}}\;{{\rm d}}t
+ \lim_{T \to \infty} \frac{1}{T}\int_T^{T + s} {\langle}\mathscr{P}^*_{t} \mu, {\varphi}{\rangle}_{\mathfrak{X}}\;{{\rm d}}t
\\
&\, - \lim_{T \to \infty} \frac{1}{T}\int_0^{s} {\langle}\mathscr{P}^*_{t} \mu, {\varphi}{\rangle}_{\mathfrak{X}}\;{{\rm d}}t \\
=&\,\lim_{T\to \infty}\int_{\mathfrak{X}}{\varphi}(u)d\nu_T.\end{aligned}$$ In the above, we require the Feller property to execute the first equality, as $\mathscr{P}_s {\varphi}$ has to remain continuous. With this, the second and third terms after the fourth equality above tend to zero in the limit $T \to \infty$, as $s$ is fixed.
The following lemma provides two sufficient conditions for the tightness of $\{\nu_T\}$.
A measure sequence $\{\nu_T\}$ is tight if one of the following conditions holds[:]{}
1. $\{\mathscr{P}_t^*\mu\}$ is tight[;]{}
2. $\{\mathscr{P}_t\}$ are compact for $t > 0$ so that $$\mathscr{P}_t({\mathfrak{X}}) \subseteq {\mathfrak{Y}}\qquad \,\,\mbox{for almost all $t>0$}$$ for a Banach space ${\mathfrak{Y}}$ such that there is a compact embedding ${\mathfrak{Y}}\hookrightarrow {\mathfrak{X}}$ and there exists $C>0$ independent of $T$ so that $$\begin{aligned}
\label{eq:bbd_avg}
\frac{1}{T}\int_0^T \|\mathscr{P}_tu_0\|_{{\mathfrak{Y}}}\;{{\rm d}}t \le C
\qquad\mbox{for any $u_0\in {\mathfrak{X}}$};\end{aligned}$$ In addition, $\mu=\delta_{u_0}$ for some $u_0\in {\mathfrak{X}}$.
For (i), we know that, for any ${\epsilon}>0$, there is a compact set $K_{\epsilon}\subseteq {\mathfrak{X}}$ such that $$\mathscr{P}_t^*\mu({\mathfrak{X}}\setminus K_{\epsilon})\le {\epsilon}\qquad \mbox{uniform in $t>0$}.$$ Then $$\nu_T({\mathfrak{X}}\setminus K_{\epsilon})\le \frac{1}{T}\int_0^T
\mathscr{P}_t^*\mu({\mathfrak{X}}\setminus K_{\epsilon})\, dt\le {\epsilon}.$$
For (ii), let $K_R = \{ u\in {\mathfrak{X}}: \|u\|_{\mathfrak{Y}}\leq R\}$. Since ${\mathfrak{Y}}\hookrightarrow {\mathfrak{X}}$ is compact, $K_R$ is compact in ${\mathfrak{X}}$.
If $u \in {\mathfrak{X}}\setminus K_R$, then $\|u\|_{\mathfrak{Y}}> R$. Writing $f(\cdot) = \|\cdot\|_{\mathfrak{Y}}$, then $$\begin{aligned}
\nu_T({\mathfrak{X}}\setminus K_R) \leq &\, \nu_T(\{f(u) > R\}).\end{aligned}$$ Applying the Markov inequality to $f$, we have $$\begin{aligned}
\nu_T({\mathfrak{X}}\setminus K_R) \leq &\, \frac{1}{R} \int_{\mathfrak{X}}f(u)\, {{\rm d}}\nu_T(u)\\
= &\, \frac{1}{RT} \int_0^T \int_{\mathfrak{X}}(\mathscr{P}_tf)(u)\, {{\rm d}}\mu(u) \;{{\rm d}}t .\end{aligned}$$
Since $\mu = \delta_{u_0}$ for some $u_0\in {\mathfrak{X}}$, then $$\frac{1}{RT} \int_0^T \int_{\mathfrak{X}}(\mathscr{P}_tf)(u)\, {{\rm d}}\mu(u)\;{{\rm d}}t
= \frac{1}{RT} \int_0^T (\mathscr{P}_tf)(u_0) \;{{\rm d}}t
= \frac{1}{RT} \int_0^T f(u(t))\;{{\rm d}}t .$$
Therefore, if the temporal average is bounded, then, for any ${\epsilon}>0$, we can choose $R>\frac{1}{{\epsilon}}$ to conclude $$\nu_T({\mathfrak{X}}\setminus K_R) \leq \, \nu_T(\{f(u) > R\})<{\epsilon}.$$ In this way, a compact set $K_R$ has been found such that $\nu_T({\mathfrak{X}}\setminus K_R) \leq {\epsilon}$, which implies that $\{\nu_T\}$ is tight.
This framework can be further refined. An example of such an extension can be found in [@CGT2018], in which the Feller property could not be proved in the context of the one-dimensional stochastic Navier-Stokes equations. Whilst the Feller condition is not available, the continuous dependence (without rates) can be shown. By using the continuous dependence, a class of functions, $\mathcal{G} \supseteq C({\mathfrak{X}})$, is defined so that $\mathcal{G}$ is continuous on the elements of the solution space with finite energy, though not necessarily the entire solution space. With these, it has been shown in [@CGT2018] that $\mathscr{P}_t$ is invariant under $\mathcal{G}$. Then the existence of invariant measures is proved in two steps: First, an energy bound is employed to yield the tightness, so that the existence of a limiting measure is shown to exist; then the limiting measure is shown to be invariant (without invoking the Feller property) by using the continuity condition imposed on $\mathcal{G}$ and following the arguments as in the proof of Theorem \[thm:KrylovBogoliubov\].
Approaches for the proof of the uniqueness of invariant measures {#sec:im_uniqueness}
----------------------------------------------------------------
It is well known that the invariant measures of a map form a convex set in the probability space on $X$. By the Krein-Milman theorem, a convex set is the closure of convex combinations of its extreme points. These extreme points $\mu$ happen to be ergodic measures, which are characterized as the property that, for a measurable subset $A \subseteq {\mathfrak{X}}$, $$\mu((\mathcal{S}^{-1}A) \Delta A) = 0\,\,\, \Longleftrightarrow\, \,\,\mu(A) = 0 \mbox{ or } \mu(A) = 1,$$ where $A\Delta B:=(A\setminus B)\cup (B\setminus A)$.
Ergodic measures heuristically carve up the solution space into essentially disjoint subsets, since any two ergodic measures of a process either coincide or are singular with respect to one another. This is a simple consequence of the property stated above.
It also follows from the extremal property of ergodic measures that, if there are more than one invariant measure, then there are at least two ergodic measures.
There are several approaches to establish the uniqueness of invariant measures.
: This is one of the common conditions used to ensure the uniqueness.
\[defin:strongfeller\] A Markov transition semigroup $\mathscr{P}_t$ is of the strong Feller property at time $t$ if $\mathscr{P}_t \varphi$ is continuous for every bounded measurable $\varphi: {\mathfrak{X}}\to \mathbb{R}$.
The strong Feller property guarantees the uniqueness of invariant measures [@Doo1948; @Kha1960]; see also [@DZ2008 Theorem 5.2.1], [@MS1998], and the references cited therein.
The strong Feller property always holds for transition semigroups of processes associated with nonlinear stochastic evolution equations with Lipschitz nonlinear coefficients and nondegenerate diffusion ([*e.g.*]{} [@PZ1995]).
: This method is a powerful tool in probability theory introduced in Doeblin-Fortet[@DF1937; @Doe1940], which can be used to show the uniqueness of invariant measures.
The general argument proceeds as follows: Let $X_t$ be a Markov process with initial distribution $\mu_0$, and let $Y_t$ be an independent copy of the process with an initial distribution that is an invariant measure $\mu$. Then the first meeting time $\mathcal{T}$ is a stopping time, and the process defined by $$\begin{aligned}
Z_t = \begin{cases}
\, X_t \quad &\mbox{for $t < \mathcal{T}$},\\
\, Y_t \quad &\mbox{for $t \geq \mathcal{T}$}
\end{cases}\end{aligned}$$ is also a copy of $X_t$ by the strong Markov property.
Using the definition of $Z_t$, we can write $$\begin{aligned}
\mathscr{P}_t^* \mu_0 - \mu
=&\, (Z_t)_* \mathbb{P} - (Y_t)_*\mathbb{P}\\
=&\, ({\mathbf{1}}_{\{t <\mathcal{T}\}}Z_t)_*\mathbb{P}+({\mathbf{1}}_{\{t \geq \mathcal{T}\}}Z_t)_*\mathbb{P}
-({\mathbf{1}}_{\{t < \mathcal{T}\}}Y_t)_*\mathbb{P} - ({\mathbf{1}}_{\{t \geq \mathcal{T}\}}Y_t)_*\mathbb{P}\\
=&\, ({\mathbf{1}}_{\{t < \mathcal{T}\}}Z_t)_* \mathbb{P} - ({\mathbf{1}}_{\{t < \mathcal{T}\}}Y_t)_*\mathbb{P}.\end{aligned}$$ Then the total variation norm of $\mathscr{P}_t^* \mu_0 - \mu$ can be estimates as $$\begin{aligned}
\left\|\mathscr{P}_t^* \mu_0 - \mu \right\|_{TV}
\leq &\, \int ({\mathbf{1}}_{\{t < \mathcal{T}\}}Z_t)_* {{\rm d}}\mathbb{P}(u) + \int ({\mathbf{1}}_{\{t < \mathcal{T}\}}Y_t)_* {{\rm d}}\mathbb{P}(u)\\
= &\, \mathbb{P}(\{t < \mathcal{T}\}).\end{aligned}$$
Assume that $\mathcal{T}$ can be shown to be almost surely finite. Then, as $t \to \infty$, we see that $\mathscr{P}_t^* \mu_0 \to \mu$, and there is only one invariant measure.
The coupling method has other applications in various different settings and can be implemented in qualitatively different ways; see also [@Lin1992; @Vil2009] and the references cited therein.
In our applications for the uniqueness of invariant measures for stochastic anisotropic parabolic-hyperbolic equations in §5, $\mathcal{T}$ will be slightly modified to be the time of entry into a small ball. Moreover, instead of the use of independent copies, we take two solutions starting at different initial data, since our Markov processes are solutions of the equations with pathwise uniqueness properties.
First, we show in §\[sec:uniquenessI\] that the two solutions $u$ and $v$ enter a given ball in finite time, almost surely. This is a stopping time. From this, by the strong Markov property, we construct a sequence of increasing, almost surely finite stopping times in (\[eq:stopping\_time\]), which are spaced at least $T$ apart, for some $T>0$ later to be fixed.
Then we show in §\[sec:uniquenessII\] that, for a well-chosen $T >0$, if a solution starts within the same given ball, and the noise is uniformly small in $W^{1,\infty}_x$ over a duration of length $T$, then the temporal average of $\|u(t)\|_{L^1_x}$ over that temporal interval can be taken to be smaller than some ${\epsilon}$. Since the noise is $\sigma(x) W$, the uniform smallness in $W^{1,\infty}_x$ over an interval $[\mathcal{T}, \mathcal{T} + T]$ depends entirely on the size of $W$.
We see that, for $T >0$, the probability that the change in the noise remains small between $\mathcal{T}$ and $\mathcal{T} + T$ is strictly positive. By the strong Markov property, we can replace $\mathcal{T}$ with any other stopping time ([*e.g.*]{} the one in the sequence constructed) spaced at least $T$ apart. Using the $L^1$–contraction, we show finally in §\[sec:uniquenessIII\] that the probability that the difference between the two solutions remains large for all intervals $[\mathcal{T}, \mathcal{T} + T]$, with $\mathcal{T}$ in the sequence of increasing stopping times, is bounded by the probability that the noise is large in $W^{1,\infty}$ over all such sequences. This must be vanishingly small, as the probability is strictly less than one on each individual sequence.
Nonlinear Conservation Laws driven by Stochastic Forcing
========================================================
In this section, we discuss one strand of the recent developments in the analysis of long-time behaviors of global solutions of nonlinear conservation laws driven by stochastic forcing.
The stochastic Burgers equation
-------------------------------
The Burgers equation is the archetypal nonlinear transport equation in many ways. The stochastic Burgers equation has also been used in turbulence and interface dynamics modelling; see [@DDT1994; @GN1999; @LNP2000; @HV2011] and the references cited therein.
The existence of a non-trivial invariant measure of the process associated to the one-dimensional Burgers equation driven by an additive spatially periodic white noise was first derived in Sinai [@Sin1991].
The long-time behavior of global solutions of the Burgers equation in one spatial dimension driven by space-time white noise has also been considered in the form: $${\partial}_t u + {\partial}_x \big(\frac{u^2}{2}\big)={\partial}_{xx}^2 u + {\partial}^2_{xt} \tilde{W},$$ where $\tilde{W}:=\tilde{W}(x,t)$ is a zero-mean Gaussian process with a covariance function given by $${\mathbb{E}}[\tilde{W}(x,t) \tilde{W}(y,s)] = (x\wedge y)(t \wedge s).$$ Apart from the global well-posedness in $L^2(\mathbb{R})$, it is known that an invariant measure for the transition semigroup exists, for example, via an argument of [@CF1994; @Fla1994] by using the ergodic theorem [@DDT1994; @GN1999]. Similar techniques have also been applied to study the two-dimensional Navier-Stokes equations driven by space-time white noises ([*e.g.*]{} [@DD2002]).
Attention in the development of the stochastic Burgers equation with vanishing viscosity has also been turned to the question of additive (spatial) noise in an equation of the form: $$\label{3.1a}
{\partial}_t u + {\partial}_x \big(\frac{u^2}{2}\big)
= \sum_{k = 0}^\infty {\partial}_x F_k(x) {{\rm d}}W^k + {\varepsilon}{\partial}^2_{xx} u.$$ The existence of invariant measures for equation with ${\varepsilon}=0$ is known ([*e.g.*]{} [@EKMS2000]). One of the key points is that there is enough energy dissipation in the inviscid limiting solutions as ${\varepsilon}\to 0$ (satisfying the Lax entropy condition) so that such an invariant measure exists.
The argument for the existence proof of invariant measures in [@EKMS2000] is not directly via the general methods discussed in §2 above. Instead, the structure of the equation is exploited to form a variational problem in [@EKMS1997]. The minimizers of the action functional $$\begin{aligned}
\mathcal{A}[y(t)] = \frac{1}{2}\int_{t_1}^{t_2} \dot{y}^2(s) \;{{\rm d}}s + \int_{t_1}^{t_2}\sum_k F_k(y(s))\,{{\rm d}}W^k(s)\end{aligned}$$ are the curves that satisfy Newton’s equations for the characteristics. These minimizers have an existence and uniqueness property with probability one. Through this, a [*one force-one solution*]{} principle has been shown, in which the random attractor consists of a single trajectory almost surely, which in turn leads to the proof of the existence of an invariant measure for .
Kinetic formulation
-------------------
The theory of kinetic formulation has been developed over the last three decades ([*cf.*]{} Perthame [@Per2002] and the references cited therein). In particular, the compactness of entropy solutions of multidimensional scalar hyperbolic conservation laws with a genuine nonlinearity was first established by Lions-Perthame-Tadmor in [@LPT1994a] via combining the kinetic formulation with corresponding velocity averaging. The velocity averaging is a technique whereby a genuine nonlinearity condition ([*i.e.*]{} a non-degeneracy condition on the nonlinearity) can be shown to imply the compactness (or even improved fractional regularity under a stronger condition) of solutions via the kinetic formulation, as seen in subsequent sections, especially in condition .
We discuss the kinetic formulation in the context of scalar hyperbolic conservation laws here.
One of the inspirations for a kinetic formulation originated from the kinetic theory of gases. One starts with a simple step function as the [*kinetic function*]{}: $$\chi^r(\xi):=\chi(\xi,r)
= \begin{cases}
\,1 &\quad\mbox{for $0 < \xi < r$}, \\
\,-1 &\quad\mbox{for $r < \xi< 0$},\\
\, 0 &\quad \mbox{otherwise}.
\end{cases}$$ Then, for any $\eta\in C^1$, the following representation formula holds: $$\begin{aligned}
\label{eq:representation_kinetic}
\int_{\mathbb{R}} \eta'(\xi) \chi^u(\xi)\;{{\rm d}}\xi =\eta(u)-\eta(0).\end{aligned}$$ A simple combination of kinetic functions yields $$\begin{aligned}
\label{eq:kinetic_combination1}
|u - v| = \int \big(|\chi^u| + |\chi^v| - 2\chi^u\chi^v\big)\;{{\rm d}}\xi.\end{aligned}$$ This provides an approach to the derivation of the $L^1$–contraction between two solutions, by estimating the terms on the right.
There are several variations on the form of the kinetic function. Since $|u - v|=(u - v)_+ + (v-u)_+$, it suffices for a variation, or combinations, of the kinetic function to capture $(u - v)_+$, which is simpler than (\[eq:kinetic\_combination1\]). This can be done by considering the following kinetic function: $$\tilde{\chi}^u:=\tilde{\chi}(\xi,u) = 1 - H(\xi - u) = H(u - \xi),$$ where $H = \mathbf{1}_{[0,\infty)}$ is the Heaviside step function. We then have the representation formula: $$\eta(u) = \int_{\mathbb{R}} \eta'(\xi) \tilde{\chi}^u(\xi) \;{{\rm d}}\xi \,\, \qquad \mbox{for $\eta \in C^1$ with $\eta(-\infty) = 0$}.$$ In particular, $$\label{3.4a}
(u - v)_+ = \int \tilde{\chi}^u(\xi) \big(1 - \tilde{\chi}^v(\xi)\big)\;{{\rm d}}\xi.$$ Such a kinetic function has been popularized by [@LPT1994a] and has been used, inter alia, in [@ChenPang-1; @DV2010; @DHV2014; @DV2015], and even as far back as [@GM1989].
The usefulness of the kinetic function can be seen in the [*kinetic formulation*]{} of scalar conservation laws, in which the kinetic variable takes the place of the solution in the nonlinear coefficients so that a degree of linearity is restored for analysis. In this formulation, many powerful linear methods such as the Fourier transform become not only applicable, but also natural.
Following Chen-Pang [@ChenPang-1], we now derive the [*kinetic formulation*]{} of nonlinear anisotropic parabolic-hyperbolic equations of second order: $$\label{3.5a}
{\partial}_t u +\nabla \cdot F(u) = \nabla \cdot ({\mathbf{A}}(u)\nabla u) + \sigma(u, x) {\partial}_t W,$$ where $F$ is a locally Lipschitz vector flux function of polynomial growth, ${\mathbf{A}}$ is a positive semi-definite matrix function with continuous entries of polynomial growth, and $\nabla=\nabla_x:=(\partial_{x_1}, \cdots, \partial_{x_d})$.
Consider the vanishing viscosity approximation to : $${\partial}_t u^{\varepsilon}+ \nabla \cdot F(u^{\varepsilon})
= \nabla \cdot \big(({\mathbf{A}}(u^{\varepsilon}) + {\varepsilon}\mathrm{I})\nabla u^{\varepsilon}\big) + \sigma(u^{\varepsilon},x) {\partial}_t W,$$ where $I$ is the identity matrix. Let $\eta\in C^1$ be an entropy with $\eta(0) = 0$. Using the Ito formula, we have $$\begin{aligned}
{\partial}_t \eta(u^{\varepsilon})
= & -\eta'(u^{\varepsilon}) \nabla \cdot F(u^{\varepsilon}) + \eta'(u^{\varepsilon}) \sigma(u^{\varepsilon}, x) {\partial}_t W+ \frac{1}{2} \eta''(u^{\varepsilon}) \sigma^2(u^{\varepsilon}, x)\\
&+ \nabla \cdot\big(\eta'(u^{\varepsilon}) {\mathbf{A}}(u^{\varepsilon}) \nabla u^{\varepsilon}\big) - \eta''(u^{\varepsilon}) {\mathbf{A}}(u^{\varepsilon}): \big(\nabla u^{\varepsilon}\otimes \nabla u^{\varepsilon}\big)\\
& + {\varepsilon}\Delta \eta(u^{\varepsilon}) - {\varepsilon}\eta''(u^{\varepsilon}) |\nabla u^{\varepsilon}|^2,\end{aligned}$$ where we have used the notation: ${\mathbf{A}}: {\mathbf{B}}=\sum_{i,j} \mathbf{a}_{ij} \mathbf{b}_{ij}$ for matrices ${\mathbf{A}}= (\mathbf{a}_{ij})$ and ${\mathbf{B}}= (\mathbf{b}_{ij})$ of the same size.
Applying the representation formula (\[eq:representation\_kinetic\]) yields $$\begin{aligned}
{\partial}_t \int \eta'(\xi) \chi^{u^{\varepsilon}} \;{{\rm d}}\xi
= & -\nabla \cdot \Big(\int \eta'(\xi) F'(\xi) \chi^{u^{\varepsilon}} \;{{\rm d}}\xi\Big)
+ \langle \sigma(\cdot,x){\partial}_t W(t)\delta(\cdot - u^{\varepsilon}),\, \eta'(\cdot)\rangle \\
& +\nabla^2 : \Big(\int \eta'(\xi){\mathbf{A}}(\xi) \chi^{u^{\varepsilon}} \;{{\rm d}}\xi\Big)
-\langle {\mathbf{A}}(\cdot) : (\nabla u^{\varepsilon}\otimes \nabla u^{\varepsilon}) \delta(\cdot - u^{\varepsilon}),\, \eta''(\cdot)\rangle\\
&-\langle {\varepsilon}|\nabla u^{\varepsilon}|^2\delta(\cdot - u^{\varepsilon}), \; \eta''(\cdot)\rangle
+ \frac{1}{2} \langle \sigma^2(\cdot, x) \delta(\cdot - u^{\varepsilon}), \; \eta''(\cdot)\rangle\\
&+ {\varepsilon}\Delta \Big(\int \eta'(\xi) \chi^{u^{\varepsilon}} \;{{\rm d}}\xi\Big).\end{aligned}$$
Assume that $u^{\varepsilon}(x,t)\to u(x,t)$ [*a.e.*]{} almost surely as ${\varepsilon}\to 0$. Then, taking $\eta'(\xi)$ as a test function and letting ${\varepsilon}\to 0$, we arrive heuristically at the formulation:
$$\begin{aligned}
\label{eq:kinetic_formulation}
{\partial}_t \chi^u + F'(\xi) \cdot \nabla \chi^u = {\mathbf{A}}(\xi) : \nabla^2 \chi^u + \sigma(\xi, x) {\partial}_t W(t)\delta(\xi - u) + {\partial}_\xi( m^u + n^u - p^u),\end{aligned}$$
which holds in the distributional sense, where $m^u$, $n^u$, and $p^u$ are Radon measures that are the limits of the following measure sequences: $$\begin{aligned}
&{\varepsilon}|\nabla u^{\varepsilon}|^2 \delta(\xi - u^{\varepsilon}) \rightharpoonup m^u,\\
&{\mathbf{A}}(\xi) : \big(\nabla u^{\varepsilon}\otimes \nabla u^{\varepsilon}\big) \delta(\xi -u^{\varepsilon}) \rightharpoonup n^u,\\
&\frac{1}{2} \sigma^2(\xi,x) \delta(\xi - u^{\varepsilon}) \rightharpoonup p^u.\end{aligned}$$ The Radon measure $m^u$ is the [*kinetic dissipation measure*]{} and $n^u$ is the [*parabolic defect measure*]{}, which capture the dissipation from the vanishing viscosity terms and the degenerate parabolic terms, respectively. In addition, the Radon measure $$p^u=\frac{1}{2}\sigma^2(\xi,x)\delta(\xi-u)$$ arises from the Itô correction. As ${\mathbf{A}}$ is positive semi-definite, it is manifest that $m^u$, $n^u$, and $p^u$ are all non-negative.
More precisely, the parabolic defect measure $n^u\ge 0$ is determined by the following: For any $\varphi\in C_0({\mathbb{R}}\times {\mathbb{R}}^d\times {\mathbb{R}}_+)$, $$\label{3.6a}
n^u(\varphi)=\int_{{\mathbb{R}}_+}\int_{{\mathbb{R}}^d}\varphi(u(x,t),x,t)\big|\nabla_x\cdot\big(\int_0^u{{\boldsymbol\alpha}}(\zeta)d\zeta\big)\big|^2\,{{\rm d}}x\, {{\rm d}}t .$$ The kinetic dissipation measure $m^u\ge 0$ satisfies the following:
1. For $B_R^c\subset {\mathbb{R}}$ as the complement of the ball of radius $R$, $$\begin{aligned}
\label{3.6b}
\lim_{R\to\infty}{\mathbb{E}}\big[(m^u+n^u)(B^c_R\times {\mathbb{T}}^d \times [0,T])\big] = 0;\end{aligned}$$
2. For any $\varphi \in C_0(\mathbb{R}\times{\mathbb{R}}^d)$, $$\begin{aligned}
\label{3.6c}
\int_{\mathbb{R}\times {\mathbb{R}}^d \times [0,T]} \varphi(\xi, x)\, {{\rm d}}(m^u+n^u)(\omega; \xi, x, t ) \in L^2(\Omega)\end{aligned}$$ admits a predictable representative (in the $L^2$–equivalence classes of functions).
Then, following Chen-Pang [@ChenPang-1], we introduce the notion of kinetic solutions:
\[def:ksolution\] A function $$u \in L^p(\Omega \times [0,T]; L^p({\mathbb{R}}^d)) \cap L^p(\Omega; L^\infty([0,T]; L^p({\mathbb{R}}^d)))$$ is called a *kinetic solution* of with initial data: $u|_{t=0}=u_0$, provided that $u$ satisfies the following:
- $\nabla \cdot \big(\int_0^u {{\boldsymbol\alpha}}(\xi) \; d \xi\big) \in L^2(\Omega \times {\mathbb{R}}^d\times [0,T])$;
- For any bounded $\varphi \in C(\mathbb{R})$, the Chen-Perthame chain rule relation in [@CP2003] holds: $$\begin{aligned}
\label{eq:art_chainrule}
\nabla \cdot \Big(\int_0^u \varphi(\xi) {{\boldsymbol\alpha}}(\xi) \; {{\rm d}}\xi\Big) = \varphi(u)\;\nabla \cdot \Big(\int_0^u {{\boldsymbol\alpha}}(\xi) \;{{\rm d}}\xi\Big)\end{aligned}$$ in $\mathcal{D}'(\mathbb{T}^d)$ and almost everywhere in $(t,\omega)$.
- There is a kinetic measure $m^u\ge 0$ $\mathbb{P}$-[*a.e.*]{} such that, given the parabolic defect measure $n^u$, the following holds almost surely[:]{} For any $\varphi \in C^\infty_c(\mathbb{R}, {\mathbb{R}}^d \times [0,T))$, $$\begin{aligned}
&-\int_0^T \iint \chi(\xi, u) \; {\partial}_t \varphi\, {{\rm d}}\xi\,{{\rm d}}x\,{{\rm d}}t
- \iint \chi(\xi, u_0)\,\varphi(\xi,x,0) \;{{\rm d}}\xi\,{{\rm d}}x\notag\\
&= \int_0^T \iint \chi(\xi, u)\,F'(\xi)\cdot \nabla \varphi \;{{\rm d}}\xi\,{{\rm d}}x\,{{\rm d}}t
+ \int_0^T \iint \chi(\xi, u) {\mathbf{A}}(\xi):\nabla^2\varphi \;{{\rm d}}\xi\,{{\rm d}}x\,{{\rm d}}t \notag\\
&\quad + \int_0^T \iint \varphi_\xi \, {{\rm d}}(m^u+n^u)(\xi,x,t)
- \frac{1}{2} \int_0^T \int \varphi_u(u,x,t)\sigma^2(u,x)\;{{\rm d}}x\,{{\rm d}}t\notag\\
&\quad - \int_0^T \int \varphi(u,x,t)\sigma(u,x) \;{{\rm d}}x\,{{\rm d}}W \qquad\,\,\mbox{almost surely}.
\label{eq:definsol}\end{aligned}$$
Equation is obtained by testing with $\varphi$ and using the chain rule (\[eq:art\_chainrule\]).
General scalar hyperbolic conservation laws driven by stochastic forcing
------------------------------------------------------------------------
In Feng-Nualart [@FN2008], the well-posedness was studied for the one-dimensional scalar conservation laws driven by white noise: $$\begin{aligned}
{\partial}_t u + {\partial}_x F(u) = \int_{z\in Z}\sigma(u,x;z)\,{{\rm d}}_z W(t, z),\end{aligned}$$ where $Z$ is a metric space, and $W$ is a space-time Gaussian noise martingale random measure with respect to a filtration $\{\mathscr{F}_t\}_{t \ge 0}$ satisfying $${\mathbb{E}}\big[W(t,A) \cap W(t,B)\big] = \mu(A \cap B) t$$ for measurable sets $A,B\subset Z$, with a $\sigma$-finite Borel measure $\mu$ on $Z$. The well-posedness theory was developed around the notion of [*strong stochastic entropy solutions*]{} introduced in Definition 2.6 in [@FN2008] when $t\in [0,T)$ for any fixed $T\in (0, \infty)$. In addition to the usual definition of entropy solutions, the following further conditions on the solution, $u=u(x,t)$, for $t\in [0,T]$ are required:
For any smooth approximation function $\beta(u)$ of function $u_+$ on $\mathbb{R}$ and any $\varphi\in C^\infty(\mathbb{R}^d\times \mathbb{R}^d)$ with $\varphi \geq 0$, and for any $\mathscr{F}_t$–adapted function $v$ satisfying $\sup_{0 \leq t \leq T} {\mathbb{E}}[\|v\|_{L^p_x}^p] < \infty$, there exists a deterministic function $\{A(s,t) : 0 \leq s \leq t\}$ such that the functional $$\begin{aligned}
f(r,z,u,y):= \int_{\mathbb{R}^d} \beta'(v(x,r) - u) \sigma(v(x,r),x;z)\varphi(x,y) \;{{\rm d}}x\end{aligned}$$ satisfies $$\begin{aligned}
&{\mathbb{E}}\big[\int \int_{(s,t]\times Z} f(r,z,u(y,t),y) {{\rm d}}W(r,z) \,{{\rm d}}y\big]\\
&\leq {\mathbb{E}}\big[\int_{(s,t]\times Z} \int {\partial}_v f(r,z,v(y,r),y) \sigma(y,u(y,r);z) \;{{\rm d}}y\,{{\rm d}}r{{\rm d}}\mu(z)\big] + A(s,t),\end{aligned}$$ and that there is a sequence of partitions of $[0,T]$ so that $$\lim_{\max|t_{i + 1} - t_i| \to 0} \sum_{i = 1}^m A(t_i,t_{i + 1}) = 0.$$
This notion of a solution addresses the problem that, in any direct adaptation of the deterministic notion of entropy solutions, one encounters the question of adaptiveness of the Itô integral in the noise-noise interaction. With this notion, in [@FN2008], the $L^1$–contraction and comparison estimates of strong stochastic entropy solutions in any spatial dimension were established, while the existence of solutions is limited to the one-dimensional case based on the compensated compactness argument in Chen-Lu [@ChenLu].
In Chen-Ding-Karlsen [@CDK2012], the existence theory for strong stochastic entropy solutions was established for any spatial dimension with the key observation that the following $BV$ bound is a corollary from the $L^1$–contraction inequality: $$\begin{aligned}
{\mathbb{E}}\big[\|u(t)\|_{BV}\big] \leq {\mathbb{E}}\big[\|u_0\|_{BV}\big],\end{aligned}$$ which provides the strong compactness required for the existence theory in any spatial dimension. More precisely, the following theorem holds:
Consider the Cauchy problem of the equation[:]{} $$\label{3.8a}
{\partial}_t u + \nabla \cdot F(u) = \sigma(u) {\partial}_t W$$ with initial condition[:]{} $$\label{3.8b}
u|_{t=0}=u_0,$$ satisfying $$\begin{aligned}
{\mathbb{E}}\big[\|u_0\|_{L^p}^p + \|u_0\|_{BV}\big] < \infty \qquad \mbox{for $p>1$},\end{aligned}$$ where $F$ is a locally Lipschitz function of polynomial growth and $\sigma$ is a globally Lipschitz function. Then there exists a unique strong stochastic entropy solution of the Cauchy problem – satisfying $${\mathbb{E}}\big[\|u(t)\|_{BV}\big] \leq {\mathbb{E}}\big[\|u_0\|_{BV}\big].$$
This existence theory in $L^p \cap BV$ can also been extended to the second-order equations as established in Chen-Pang [@ChenPang-1], including the case with heterogeneous flux functions $F=F(u,x)$ ([*i.e.*]{} the space-translational variant case).
A well-posedness theory can also be developed for kinetic solutions to the multidimensional scalar balance laws with stochastic force , by employing the Gyöngy-Krylov framework where the existence of a martingale solution with pathwise uniqueness guarantees the strong existence; see [@DV2010]. In particular, the existence of martingale solutions can be proved via the notion of kinetic solutions. These results can be extended ([*e.g.*]{} [@Hof2013; @DHV2014]) to encompass degenerate parabolic equations: $${\partial}_t u + \nabla \cdot F(u) - \nabla \cdot ({\mathbf{A}}(u) \nabla u) = \sigma(u) {\partial}_t W.$$
A well-posedness theory has also been established based on the viscosity solutions (such as in [@BVW2012]). To achieve this, the difficulties caused by the noise-noise interaction that has a non-zero correlation for the multiplicative noise case are avoided by directly comparing two entropy solutions to a viscosity solution.
In Karlsen-Størrensen [@KS2015], these different viewpoints have been partially reconciled via a Malliavin viewpoint, in which the constant in the Kruzhkov entropy is interpreted as a Malliavin differentiable variable.
Long-time asymptotic results concerning the existence and uniqueness of invariant measures have followed the well-posedness theory. Concerning the stochastic balance law: $$\begin{aligned}
{\partial}_t u + \nabla \cdot F(u) = \Phi(x) \;{{\rm d}}B,\end{aligned}$$ with evolution on torus $\mathbb{T}^d$, where $B = \sum_k e_k W_k$ is a cylindrical Wiener process, $\{e_k\}$ is a complete orthonormal basis of a Hilbert space, $\Phi$ is a Hilbert-Schmidt operator given by $\Phi(x) = \sum_k g_k(x) e_k$, and $g_k(x)$ satisfies $$\int_{\mathbb{T}^d} g_k(x) \;{{\rm d}}x = 0,$$ the existence and uniqueness of invariant measures were shown in [@DV2015]. In this case, the noise is [*additive*]{}; that is, it depends only on the spatial variable, but is independent of the solution – a point to which we will return.
These results can be summarized as follows:
Let $F$ satisfy the non-degeneracy condition[:]{} For some $b < 1$ and a constant $C >0$, $$\label{3.14a}
\delta({\varepsilon}) := \int_0^\infty e^{-t}\sup_{\tau\in {\mathbb{R}}, |\hat{k}|=1}
\mathcal{L}^1(\{\xi: |F'(\xi)\cdot \hat{k} + \tau| \le {\varepsilon}t\}) \; {{\rm d}}t
\le C{\varepsilon}^b$$ for the Lebesgue measure $\mathcal{L}^1$ on $\mathbb{R}$, in addition to the condition that $|F''(\xi)| \lesssim |\xi| + 1$. Then there exits an invariant measure to the process. Furthermore, if $|F''(\xi)|\lesssim 1$ is bounded, then the invariant measure is unique.
The bounds for the spaces on which the invariant measures are supported have also been derived. This result has been obtained by employing the velocity averaging. It has been built also on the related ideas of kinetic solutions, which is first applied to the velocity averaging in the deterministic context. They avoided the question of the Fourier transforms of the Wiener process by introducing regularizing operators.
Similar results were also derived for $${\partial}_t u + \nabla \cdot (F(u) \circ {{\rm d}}W) = 0,$$ by further employing the conservative form as considered in Lions-Perthame-Souganidis [@LPS2013; @LPS2014]; see Gess-Souganidis [@GS2014]. A generalization of this with a degenerate parabolic term $\nabla \cdot ({\mathbf{A}}(u)\nabla u)$ has also been considered in [@GS2016; @FG-2019]. In particular, Fehrman-Gess [@FG-2019] investigated the well-posedness and continuous dependence of the stochastic degenerate parabolic equations of porous medium type, including the cases with fast diffusion and heterogeneous fluxes.
By using the methods developed in [@FGRW2015; @KS2012; @HM2008; @HM2011] and developing the probabilistic Gronwall inequality based on delicate reasoning about a stopping time, such MHD equations driven by additive noise of zero spatial average in the vanishing Rossby number and vanishing magnetic Reynold’s number limit were also shown to have a unique invariant measure (that is necessarily ergodic) in [@FFGR2016].
Stochastic Anisotropic Parabolic-Hyperbolic Equations\
I: Existence of Invariant Measures {#sec:dp_invariantm}
======================================================
In this section, we present an approach for establishing the existence of invariant measures for nonlinear anisotropic parabolic-hyperbolic equations driven by stochastic forcing: $$\begin{aligned}
\label{eq:Main2}
{\partial}_t u + \nabla \cdot F(u) = \nabla \cdot({\mathbf{A}}(u) \nabla u) + \sigma(x) {\partial}_t W,\end{aligned}$$ where ${\mathbf{A}}$ is positive semi-definite, and $\sigma$ has zero average over $\mathbb{T}^d$. The main focus of this section is on the presentation of the approach, so we do not seek the optimality of the results, while the results presented below can be further improved by refining the arguments and technical estimates required for the approach which is out of the scope of this section. More precisely, we establish the following theorem:
Let $F$ and ${\mathbf{A}}$ satisfy the nonlinearity-diffusivity condition[:]{} There exist $\beta\in (1,2)$, $\kappa\in (0,1)$, and $C>0$, independent of $\lambda$, such that $$\begin{aligned}
\label{decayrate}
\sup_{\tau \in {\mathbb{R}},\, |\hat{k}|=1}\int \frac{\lambda ({\mathbf{A}}(\xi): \hat{k}\otimes \hat{k} + \lambda)}{({\mathbf{A}}(\xi): \hat{k}\otimes \hat{k} +\lambda)^2
+ \lambda^{\beta}|F'(\xi)\cdot \hat{k} + \tau|^2} \;{{\rm d}}\xi
=:\eta(\lambda)\le C \lambda^\kappa \to 0\end{aligned}$$ as $\lambda\to 0$. In addition, let $F$ and ${\mathbf{A}}$ satisfy the condition[:]{} $$\begin{aligned}
\label{eq:conditions_existence}
|F''(\xi)| \lesssim |\xi| + 1, \qquad
|{\mathbf{A}}'(\xi)| \lesssim |\xi| +1.\end{aligned}$$ Then there exists an invariant measure to the process associated with the solutions to .
The approach is motivated by Debussche-Vovelle [@DV2015] by extending the case from first-order scalar balance laws to the second-order degenerate parabolic-hyperbolic equation . The first-order case is handled in [@DV2015], based on the velocity averaging and built on Lemma 2.4 of Bouchut-Desvillettes [@BD1999]. In our approach, we require a modified version of this lemma, which is incorporated into the calculation that allows us to exploit the cancellations in an oscillatory integral in this more general case than the first-order case. We now proceed to prove the theorem as follows:
- First we incorporate regularizing operators into the equation in order to exploit the bounds that can be provided in the Duhamel representation of the solution.
- We separate the Duhamel representation of the solution into four different summands, the $W^{s,q}$ norms of which we estimate.
- Adding these estimates together by the triangle inequality and using the compact inclusion of $W^{s,q}$ into a suitable $L^q$ norm allow us to invoke the Krylov-Bogoliubov machinery described in §\[sec:krylov\_bogoliubov\].
We expound on the nonlinearity condition in a remark below. As the conditions in (\[eq:conditions\_existence\]) are invoked along the way, we also explain their relevance.
Consider the kinetic formulation of equation (\[eq:Main2\]): $$\begin{aligned}
\label{4.2a}
{\partial}_t \chi^u +\big(F'(\xi) \cdot \nabla - {\mathbf{A}}(\xi) :\nabla \otimes \nabla \big)\chi^u
= {\partial}_\xi(m^u + n^u - p^u) + \sigma(x) \delta(\xi - u) {\partial}_t W.\end{aligned}$$
In order to handle the two measures: $m^u + n^u - p^u$ and $\sigma(x) \delta(\xi - u)$, we need to regularize the operators as in [@DV2015], by adding $\gamma (-\Delta)^\alpha + \theta \, I$ to each side: $$\begin{aligned}
&{\partial}_t \chi^u +\big(F'(\xi) \cdot \nabla - {\mathbf{A}}(\xi) :\nabla \otimes \nabla + \gamma(-\Delta)^\alpha + \theta\,I\big)\chi^u \nonumber\\
&= \big(\gamma(-\Delta)^\alpha + \theta I\big)\chi^u + {\partial}_\xi(m^u + n^u - p^u) + \sigma(x) \delta(\xi - u) {\partial}_t W \label{4.3a}\end{aligned}$$ for $\alpha=\frac{\beta-1}{\beta}\in (0, \frac{1}{2})$ for some $\beta\in (1,2)$ required in the nonlinearity-diffusivity condition .
We adapt the semigroup approach. There are specific reasons to include these regularizing operators: In order to estimate the measure, $\sigma(x) \delta(\xi - u){\partial}_t W$, we require a spatial regularization provided by $(-\Delta)^\alpha$ and temporal decay provided by $\theta I$.
More specifically, let $\mathcal{S}(t)$ be the semigroup of operator ${\partial}_t +\big(F'(\xi) \cdot \nabla - {\mathbf{A}}(\xi) :\nabla \otimes \nabla + \gamma(-\Delta)^\alpha + \theta\,I\big)$: $$\begin{aligned}
{\mathcal{S}}(t) f(x)=&\, e^{-(F'(\xi) \cdot \nabla -{\mathbf{A}}(\xi): \nabla \otimes \nabla + \gamma(-\Delta)^\alpha + \theta I)t} f\nonumber\\
= &\, e^{-\theta t} \big(e^{t{\mathbf{A}}(\xi) : \nabla \otimes \nabla -t\gamma(-\Delta)^{\alpha}} f\big) (x - F'(\xi) t)
\qquad\,\,\mbox{for any $f=f(x)$}. \label{4.4a}\end{aligned}$$ Then we can express the solution, $\chi^u$, to the kinetic formulation in the mild formulation: $$\begin{aligned}
\chi^u =& \, {\mathcal{S}}(t) \chi^u(\xi,x,0) + \int_0^t {\mathcal{S}}(s) (\gamma(-\Delta)^\alpha - \theta I) \chi^u(\xi, x, t - s) \;{{\rm d}}s\\
&+\int_0^t {\mathcal{S}}(t - s) {\partial}_\xi(m^u + n^u - p^u) (\xi,x,s)\;{{\rm d}}s + \int_0^t {\mathcal{S}}(t - s) \sigma(x) \delta(\xi - u(x,s)) \;{{\rm d}}W_s.\end{aligned}$$ This leads to the decomposition: $$\label{4.6a}
u = u^0 + u^\flat + M_1 + M_2,$$ where $$\begin{aligned}
u^0(x,t) = & \int {\mathcal{S}}(t)\chi^u(\xi,x,0) \;{{\rm d}}\xi , \label{4.7a}\\
u^\flat(x,t) = &\int \int_0^t {\mathcal{S}}(s) (\gamma(-\Delta)^\alpha - \theta I)\chi^u(\xi,x,t - s)\;{{\rm d}}s\,{{\rm d}}\xi ,\label{4.7b}\\
{\langle}M_1, \varphi {\rangle}= & \int_0^t\int {\langle}{\partial}_\xi(m^u + n^u - p^u)(\cdot,x,t - s),\, {\mathcal{S}}^*(s)\varphi{\rangle}\, {{\rm d}}x\,{{\rm d}}s,\label{4.7c}\\
{\langle}M_2, \varphi {\rangle}= & \int_0^t\int {\langle}\delta(\cdot - u(x,s)),\, {\mathcal{S}}^*(t - s)\varphi{\rangle}\sigma(x) \;{{\rm d}}x \;{{\rm d}}W_s,\label{4.7d}\end{aligned}$$ where ${\mathcal{S}}^*(t)$ is the dual operator of the semigroup operator ${\mathcal{S}}(t)$, and $\varphi\in C(\mathbb{T}^d)$.
We now estimate each of the four terms separately in each subsection: The first two integrals are essentially “deterministic” parts and estimated by the velocity averaging method, and the final two integrals incorporate stochastic elements and are treated by a kernel estimate on semigroup ${\mathcal{S}}(t)$.
Analysis of $u^0$
-----------------
Notice that the local Fourier transform in $x\in \mathbb{T}^d$ for any periodic function $g(x,\cdot)$ in $x$ with period $P=(P_1,\cdots, P_d)$ is: $$\hat{g}(k, \cdot )=
\frac{1}{|\mathbb{T}^d|}
\int_{\mathbb{T}^d} g(x,\cdot)e^{-i x\cdot k}\,{{\rm d}}x,$$ where frequencies $k=(k_1,\cdots, k_d)$ are discrete: $$k_i=\frac{2\pi}{P_i}n_i, \qquad n_i=0, \pm 1, \pm 2, \cdots, \,\,\, i=1,\cdots, d.$$
Taking the Fourier transform in $x$ and integrating in $\xi$, we have $$\begin{aligned}
\widehat{u^0}(k,t) = &\int \hat{{\mathcal{S}}}(t) \widehat{\chi^u}(\xi,k,0) \;{{\rm d}}\xi \\
= & \int e^{-(iF'(\xi) \cdot k + {\mathbf{A}}(\xi) : (k \otimes k)+ \omega_k |k|)t}\, \widehat{\chi^u} (\xi,k,0)\;{{\rm d}}\xi,\end{aligned}$$ where $\omega_k=\gamma|k|^{2\alpha-1}+\theta |k|^{-1}$.
For simplicity, we denote $\hat{k}=\frac{k}{|k|}$ and $\mathcal{A}= \mathcal{A}(\xi,\hat{k})={\mathbf{A}}(\xi) : \hat{k} \otimes\hat{k}$. Then we square the above and integrate in $t$ from $0$ to $T$ to obtain $$\begin{aligned}
\int_0^T |\widehat{u^0}(k,t)|^2 \;{{\rm d}}t
= & \int_0^T \left| \int e^{-(iF'(\xi) \cdot \hat{k}+ \mathcal{A}(\xi,\hat{k})|k|+\omega_k) |k| t}\, \widehat{\chi^u} (\xi,k,0)\;{{\rm d}}\xi \right|^2 \;{{\rm d}}t\notag\\
\leq &\, \frac{1}{|k|} \int \left| \int \mathds{1}_{\{s > 0\}}
e^{-(iF'(\xi) \cdot\hat{k}+\mathcal{A}(\xi,\hat{k})|k|+\omega_k)s}\,\widehat{\chi^u} (\xi,k,0)\;{{\rm d}}\xi \right|^2 \;{{\rm d}}s.\label{eq:model_estimate1}\end{aligned}$$
Notice that it is impossible to extract the entire non-oscillatory part of the exponential from the integral in $\xi$, as was done with the lemma of Bouchut-Desvillettes [@BD1999]. However, by extending the range of integration over all $\mathbb{R}$ to make the function in $s$ smoother so that its transform has better decay properties, we can partially exploit the cancellations later: $$\begin{aligned}
\int_0^T |\widehat{u^0}(k,t)|^2 \;{{\rm d}}t
\leq &\frac{1}{|k|}\int_{-\infty}^\infty
\left|\int e^{iF'(\xi) \cdot \hat{k} s} e^{-(\omega_k +\mathcal{A}|k|)|s|} \widehat{\chi^u} (\xi,k,0)\;{{\rm d}}\xi \right|^2 \;{{\rm d}}s.
\end{aligned}$$
We can evaluate the temporal Fourier transform of the integrand explicitly: $$\begin{aligned}
\mathscr{F}^{-1}\left\{ e^{iF'(\xi) \cdot \hat{k} s} e^{-(\omega_k + \mathcal{A}|k|)|s|} \right\}(\tau)
= - \frac{2(\mathcal{A}|k|+\omega_k)}{(\mathcal{A}|k|+\omega_k)^2
+ |F'(\xi)\cdot \hat{k} + \tau|^2}.\end{aligned}$$
Next, using the Parseval identity in the temporal variable and the Cauchy-Schwarz inequality, we have $$\begin{aligned}
\int_0^T \big|\widehat{u^0}(k,t)\big|^2 \;{{\rm d}}t
\leq &\, \frac{1}{|k|}\int_{-\infty}^\infty
\left|\int e^{iF'(\xi) \cdot \hat{k} s} e^{-(\omega_k + \mathcal{A}|k|)|s|} \widehat{\chi^u} (\xi,k,0)\;{{\rm d}}\xi \right|^2 \;{{\rm d}}s\\
= &\, \frac{1}{|k|}\int_{-\infty}^\infty
\left|\mathcal{F}^{-1}
\left\{\int e^{iF'(\xi) \cdot \hat{k} s} e^{-(\omega_k + \mathcal{A}|k|)|s|} \widehat{\chi^u} (\xi,k,0)\;{{\rm d}}\xi\right\}(\tau)\right|^2 \;{{\rm d}}\tau\\
= &\, \frac{4}{|k|}\int_{-\infty}^\infty
\left|\int \frac{\mathcal{A}|k|+\omega_k}{(\mathcal{A}|k|+\omega_k)^2
+ |F'(\xi)\cdot \hat{k} + \tau|^2} \widehat{\chi^u} (\xi,k,0)\;{{\rm d}}\xi \right|^2 \;{{\rm d}}\tau\\
\le &\, \frac{4}{|k|}\int_{-\infty}^\infty
\left( \int \big|\widehat{\chi^u}(\xi,k,0)\big|^2\frac{\mathcal{A}|k|+\omega_k}{(\mathcal{A}|k|+\omega_k)^2
+ |F'(\xi)\cdot \hat{k} + \tau|^2}\;{{\rm d}}\xi \right)\,\\
&\qquad\quad\quad\times\left(\int \frac{\mathcal{A}|k|+\omega_k}{(\mathcal{A}|k|+\omega_k)^2
+ |F'(\xi)\cdot \hat{k} + \tau|^2} \;{{\rm d}}\xi\right){{\rm d}}\tau\\
\le &\, \frac{4}{|k|\omega_k}
\int \big|\widehat{\chi^u}(\xi,k,0)\big|^2\left(\int \frac{\mathcal{A}|k|+\omega_k}{(\mathcal{A}|k|+\omega_k)^2
+ |F'(\xi)\cdot \hat{k} + \tau|^2} \;{{\rm d}}\tau\right) {{\rm d}}\xi\\
&\qquad \times \sup_\tau\int \frac{\omega_k(\mathcal{A}|k| + \omega_k)}{(\mathcal{A}|k| + \omega_k)^2
+ |F'(\xi)\cdot \hat{k} + \tau|^2} \;{{\rm d}}\xi.\end{aligned}$$
Notice that the integral $$\int \frac{\mathcal{A}|k|+\omega_k}{(\mathcal{A}|k|+\omega_k)^2
+ |F'(\xi)\cdot \hat{k} + \tau|^2} \;{{\rm d}}\tau$$ is a constant for fixed $\xi$ by the translation invariance of ${{\rm d}}\tau$.
Now invoking and setting $\lambda = \frac{\omega_k}{|k|}$, we have $$\int_0^T |\widehat{u^0}(k,t)|^2\;{{\rm d}}t
\le \frac{C}{|k|\omega_k}\eta(\frac{\omega_k}{|k|})
\int |\widehat{\chi^u}(\xi,k,0)|^2\;{{\rm d}}\xi$$ for some constant $C$ depending on $\gamma$ and $\theta$. That is, $$\int_0^T |k|^{1+\kappa}\omega_k^{1-\kappa} |\widehat{u^0}(k,t)|^2\;{{\rm d}}t
\le
C \int |\widehat{\chi^u}(\xi,k,0)|^2\;{{\rm d}}\xi.$$
Since $u_0$ has null average over $\mathbb{T}^d$, $$\label{4.15a}
\widehat{u^0}(0,t)=\int_{\mathbb{T}^d}u^0(x,t)\, {{\rm d}}x
=\int\widehat{\chi^u}(\xi, 0,0)\, {{\rm d}}\xi
=\iint \chi^u(\xi, x,0)\,{{\rm d}}\xi\,{{\rm d}}x
=\int_{\mathbb{T}^d} u_0(x)\, {{\rm d}}x=0.$$
Then, summing over all the discrete frequencies $k$ with $|k|\ne 0$, using the Plancherel theorem again — in space this time — and noting that $\omega_k\ge \gamma|k|^{2\alpha-1}$, we have the estimate $$\begin{aligned}
\label{eq:u_0_estimate}
\int_0^T\|u\|_{H^{(1 - \alpha) \kappa + \alpha}_x}^2 \;{{\rm d}}t \le C\|u_0\|_{L^{1}_x}.\end{aligned}$$
Analysis of $u^\flat$ {#sec:uflat}
---------------------
The calculation is similar: $$\begin{aligned}
&\int_0^T |\widehat{u^\flat}(k,t)|^2\;{{\rm d}}t \\
&=\int_0^T \left|\int \int_0^t \hat{{\mathcal{S}}}(s) (\gamma|k|^{2\alpha} + \theta I)\,\widehat{\chi^u}(\xi,k,t - s)\;{{\rm d}}s\,{{\rm d}}\xi \right|^2 {{\rm d}}t \\
&= \int_0^T \left| \int_0^T {\mathbf{1}}_{\{t - s \geq 0\}} \int e^{-(iF'(\xi) \cdot k + \omega_k|k| + {\mathbf{A}}(\xi) : k\otimes k) s}
\omega_k |k|\,\widehat{\chi^u}(\xi,k,t - s)\;{{\rm d}}\xi \, {{\rm d}}s\right|^2 \;{{\rm d}}t \\
&\leq \Big(\int_0^\infty \omega_k|k| e^{-\omega_k |k| s}\;{{\rm d}}s\Big)\\
&\quad\,\,\,\times \int_0^T\Big(\int_0^T \Big|\int e^{-(iF'(\xi) \cdot k + \omega_k|k|/2 + {\mathbf{A}}(\xi) : k\otimes k) s}
\sqrt{\omega_k |k|}\, \widehat{\chi^u}(\xi,k, t)\,{{\rm d}}\xi\Big|^2{{\rm d}}s \Big){{\rm d}}t,\end{aligned}$$ where we have used the Cauchy-Schwarz inequality and extended the domain of the inner temporal integration to $[0,\infty)$.
This leaves us in the exact position of Eq. with an additional temporal integral in $t$ (applied only to the kinetic function $\widehat{\chi^u}$) and an additional factor of $|k|\omega_k$. Therefore, we can conclude as in Eq. by using the zero-spatial average property and $\omega_k\le (\gamma+\theta)|k|^{2\alpha-1}$ that $$\begin{aligned}
\label{eq:ubestimate}
\int_0^T\|u^\flat(t)\|_{H^{(1 - \alpha)\kappa}_x}^2 \;{{\rm d}}t\leq &C \int_0^T \|u(t)\|_{L^{1}_x}\;{{\rm d}}t.\end{aligned}$$
\[rem:nondegeneracy\] Condition is reminiscent of the nonlinearity condition given in the deterministic setting by Chen-Perthame [@CP2009]. If we discard the regularising operator $(-\Delta)^{2\alpha}$ in , [*i.e.*]{} by setting $\alpha = 0$, then $\beta=1$ in . On the other hand, we can choose $\beta$ sufficiently close to $2$ so that holds, by selecting $\alpha$ close to $\frac{1}{2}$. For both cases, we are able to conclude that the $u^\flat$-part of the solution operator is compact. However, as we will see below, the regularizing effect of $(-\Delta)^{2\alpha}$ is crucial in estimating – in the way as we do, via , in the next subsections. As the two terms and arise from the martingale and the Itô approximation, respectively, this decay requirement beyond $o(1)$ does not appear in the deterministic setting.
Analysis of $M_1$
-----------------
Next we turn to the analysis of the two measures $M_1$ and $M_2$. For this, we follow [@DV2015] closely, since the only difference is the parabolic defect measure, which has the same sign as the kinetic dissipation measure, and the magnitude of the kinetic dissipation measure is never invoked in [@DV2015]. In this and the following sections, we repeatedly apply bound (\[eq:kernel\_estimate\]) in order to pursue the compactness estimates.
From , we see $$\begin{aligned}
\label{eq:pd_xi_se}
&{\partial}_\xi\big({\mathcal{S}}^*(t -s) h(\xi,x)\big) \notag\\
&= (t - s) F''(\xi) \cdot \nabla ({\mathcal{S}}^*(t - s)h) + {\mathbf{A}}'(\xi) : \nabla^2 ({\mathcal{S}}^*(t - s) h)
+{\mathcal{S}}^*(t - s) {\partial}_\xi h.\end{aligned}$$ Then we have $$\begin{aligned}
{\langle}M_1, \varphi {\rangle}&= -\int_0^t\iint {\partial}_\xi({\mathcal{S}}^*(s)\varphi)\; {{\rm d}}(m^u + n^u - p^u)(\xi,x,t - s)
\notag\\
&= \int_0^t\iint (t - s)F''(\xi) \cdot \nabla ({\mathcal{S}}^*(t - s)\varphi)\;{{\rm d}}(m^u + n^u - p^u)(\xi,x,t - s)
\notag\\
&\quad + \int_0^t\iint {\mathbf{A}}'(\xi) : \nabla^2 ({\mathcal{S}}^*(t - s) \varphi)\; {{\rm d}}(m^u + n^u - p^u)(\xi,x,t - s).
\label{eq:M1_expanded}\end{aligned}$$
Now we show the following total variation estimate.
\[thm:TVestimate1\] Let $u: \mathbb{T}^d \times [0,T] \times \Omega$ be a solution with initial data $u_0$. Let $\psi \in C_c(\mathbb{R})$ be any nonnegative and compactly supported continuous function, and $\Psi = \int_0^s \int_0^r \psi(t)\;{{\rm d}}t {{\rm d}}r$. Then $$\begin{aligned}
{\mathbb{E}}\big[\int_{\mathbb{T}^d \times [0,T] \times \mathbb{R}} \psi(\xi)\;{{\rm d}}|m^u + n^u - p^u|(\xi,x,t)\big]
\leq D_0 {\mathbb{E}}\big[\|\psi(u)\|_{L^1_{x,t}}\big] + {\mathbb{E}}\big[\|\Psi(u_0)\|_{L^1_x}\big],\end{aligned}$$ where $D_0:=\|\sigma^2\|_{L^\infty(\mathbb{T})}$.
The proof is the same as that in [@DV2015] and involves bounding $|m^u + n^u - p^u|\leq m^u + n^u + p^u$, so that $$\begin{aligned}
&{\mathbb{E}}\big[\int_0^T \iint \psi(\xi)\; {{\rm d}}\,|m^u + n^u - p^u|(\xi, x, t)\big]\\
&\leq {\mathbb{E}}\big[\int_0^T \iint \psi(\xi)\; {{\rm d}}(m^u + n^u - p^u)(\xi, x,t)\big]
+2 {\mathbb{E}}\big[\int_0^T \iint \psi(\xi) \;{{\rm d}}p^u(\xi, x, t)\big]\\
&= {\mathbb{E}}\big[-\int\Psi(u)\;{{\rm d}}x \bigg|^T_0\big]
+ {\mathbb{E}}\big[\int_0^T\int \sigma^2(x) \psi(u) \;{{\rm d}}x {{\rm d}}t \big],\end{aligned}$$ by using the kinetic equation in the sense of . Now, using the non-negativity of $\psi$, we have $${\mathbb{E}}\big[\int_0^T \iint \psi(\xi) \; {{\rm d}}|m^u + n^u - p^u|(\xi, x,t)\big]
\leq {\mathbb{E}}\big[ \int\Psi(u_0)\;{{\rm d}}x\big] + D_0 {\mathbb{E}}\big[\int_0^T\int \psi(u) \;{{\rm d}}x\,{{\rm d}}t \big].$$
This estimate is quite crude, as one does not take the cancellation between measures $m^u + n^u$ and $p^u$, both non-negative, into account. Since there is no available way to quantify $m^u + n^u$, this is the best possible at the moment.
In addition to a total variation estimate, we also require the kernel estimate: $$\begin{aligned}
\Big\|(-\Delta)^{\frac{\hat{\beta}}{2}}e^{({\mathbf{A}}:\nabla \otimes \nabla - \gamma(-\Delta)^{\alpha})t}\Big\|_{L^p_{x,\xi} \to L^q_{x,\xi}}
\leq C(\gamma t)^{-\frac{d}{2\alpha} \left(\frac{1}{p}- \frac{1}{q}\right) - \frac{\hat{\beta}}{2\alpha}}.\label{eq:kernel_estimate}\end{aligned}$$ The reason for the no additional improvement over the estimate for operator $e^{t{\mathbf{A}}:\nabla \otimes \nabla}$ is that we have not specified how degenerate ${\mathbf{A}}$ is — it may well be simply the zero matrix. It is the use of this kernel estimate that necessitates the inclusion of the regularizations $\gamma(-\Delta)^{\alpha} + \theta I$.
By the kernel estimate , we have $$\begin{aligned}
&\|(-\Delta)^{\frac{\hat{\beta}}{2}}\nabla({\mathcal{S}}^*(t){\varphi})\|_{L^\infty_{x,\xi}}
\leq C (\gamma t)^{-\mu}e^{-\theta\, t}\|\varphi\|_{L^{p}_x},\\
&\|(-\Delta)^{\frac{\hat{\beta}}{2}}\nabla^2 ({\mathcal{S}}^*(t) {\varphi})\|_{L^\infty_{x,\xi}}
\leq C(\gamma t)^{-\mu-\frac{1}{2\alpha}}
e^{-\theta\,t}\|\varphi\|_{L^{p}_x},\end{aligned}$$ where $\mu:=\frac{\hat{\beta} + 1}{2\alpha} + d(\frac{1}{2\alpha} - \frac{1}{2\alpha p'})$ for $p'>1$, and the universal constant $C$ is independent of $\gamma$ and $\theta$.
Inserting these estimates into (\[eq:M1\_expanded\]), we have the estimate: $$\begin{aligned}
&{\mathbb{E}}\big[\int_0^T {\langle}(-\Delta)^{\frac{\hat{\beta}}{2}}M_1, \varphi {\rangle}\;{{\rm d}}t \big]\\
&= {\mathbb{E}}\big[\int_0^T\int_0^t\iint (-\Delta)^{\frac{\hat{\beta}}{2}}F''(\xi) \cdot \nabla ({\mathcal{S}}^*(t - s)\varphi)\;{{\rm d}}(m^u + n^u - p^u)(\xi,x,t - s) \, {{\rm d}}t \\
& \qquad\,\, + \int_0^T\int_0^t\iint (-\Delta)^{\frac{\hat{\beta}}{2}}{\mathbf{A}}'(\xi) : \nabla^2 ({\mathcal{S}}^*(t - s) \varphi)\;{{\rm d}}(m^u + n^u - p^u)(\xi,x,t - s) \;{{\rm d}}t \big]\\
&\leq {\mathbb{E}}\big[\int_0^T\int\|(-\Delta)^{\frac{\hat{\beta}}{2}}\nabla ({\mathcal{S}}^*(t - s)\varphi)\|_\infty |F''(\xi)|(t - s)\;{{\rm d}}|m^u + n^u - p^u|(\xi, x, s)\;{{\rm d}}t \big]\\
&\quad + {\mathbb{E}}\big[\int_0^T\int\|(-\Delta)^{\frac{\hat{\beta}}{2}}\nabla^2 ({\mathcal{S}}^*(t - s)\varphi)\|_\infty |{\mathbf{A}}'(\xi)| \;{{\rm d}}|m^u + n^u - p^u|(\xi, x,s)\;{{\rm d}}t \big].\end{aligned}$$
By the presence of factor $e^{-\theta(t - s)}$, we can also bound the outer temporal integral by using the definition of the Gamma function: $$\Gamma(z) =\int_0^\infty x^{z - 1}e^{-x} \;{{\rm d}}x$$ so that, taking ${\langle}\cdot, \cdot {\rangle}$ as the $L^{p'}({\mathbb{T}}^d)$–$L^{p}({\mathbb{T}}^d)$ pairing, $$\begin{aligned}
&{\mathbb{E}}\big[\int_0^T {\langle}(-\Delta)^{\frac{\hat{\beta}}{2}}M_1, \varphi {\rangle}\;{{\rm d}}t \big]\\
&\leq \int_0^T (\gamma \tau )^{ - \mu} e^{-\theta \tau} \;{{\rm d}}\tau\;
{\mathbb{E}}\big[\int_{\mathbb{R}\times \mathbb{T}^d\times[0,T]} \|\varphi\|_{L^p_x}|F''(\xi)|\;{{\rm d}}|m^u + n^u - p^u|(\xi, x,s)\big]\\
&\quad +\int_0^T (\gamma \tau )^{ - \mu - \frac{1}{2\alpha}} e^{-\theta \tau} \;{{\rm d}}\tau\;{\mathbb{E}}\big[\int_{\mathbb{R}\times \mathbb{T}^d\times[0,T]}
\|\varphi\|_{L^p_x}|{\mathbf{A}}'(\xi)|\;{{\rm d}}|m^u + n^u - p^u|(\xi, x,s)\big]\\
&\leq C \theta^{\mu + 1} \gamma^{-\mu} |\Gamma(-\mu + 1)|
\;{\mathbb{E}}\big[\int_{\mathbb{R}\times \mathbb{T}^d\times[0,T]} \|\varphi\|_{L^p_x}|F''(\xi)|\;{{\rm d}}|m^u + n^u - p^u|(\xi, x, s)\big]\\
&\quad + C \theta^{\mu - 1 - \frac{1}{2\alpha}} \gamma^{-\mu - \frac{1}{2\alpha}}\big|\Gamma(-\mu + 1 - \frac{1}{2\alpha})\big|\\
&\qquad\,\times \;{\mathbb{E}}\big[\int_{\mathbb{R}\times \mathbb{T}^d\times[0,T]} \|\varphi\|_{L^p_x}|{\mathbf{A}}'(\xi)|\;{{\rm d}}|m^u + n^u - p^u|(\xi, x,s)\big].\end{aligned}$$
By duality, the total variation estimate, and the sublinearity of $F''$ and ${\mathbf{A}}'$, we have $$\begin{aligned}
&{\mathbb{E}}\big[\|M_1\|_{L^1_t W^{\hat{\beta},p'}_x}\big] \notag\\
&\leq C\theta^{\mu + 1} \gamma^{-\mu} |\Gamma(-\mu + 1)|
\;{\mathbb{E}}\big[\int_{\mathbb{R}\times \mathbb{T}^d\times[0,T]}|F''(\xi)|\; {{\rm d}}|m^u + n^u - p^u|(\xi, x, s)\big]\notag\\
&\quad + C\theta^{\mu - 1 - \frac{1}{2\alpha}} \gamma^{-\mu - \frac{1}{2\alpha}}\big|\Gamma(-\mu + 1 - \frac{1}{2\alpha})\big|\;
{\mathbb{E}}\big[\int_{\mathbb{R}\times \mathbb{T}^d\times[0,T]}|{\mathbf{A}}'(\xi)|\;{{\rm d}}|m^u + n^u - p^u|(\xi, x, s)\big]\notag\\
&\leq C\Big(\theta^{\mu + 1} \gamma^{-\mu} |\Gamma(-\mu + 1)|
+ \theta^{\mu - 1 - \frac{1}{2\alpha}} \gamma^{-\mu - \frac{1}{2\alpha}}\big|\Gamma(-\mu + 1 - \frac{1}{2\alpha})\big| \Big) \notag\\
&\qquad \times \Big(1 + \int_0^T{\mathbb{E}}\big[\|u(t)\|_{L^1_x}\big]\;{{\rm d}}t + {\mathbb{E}}\big[\|u_0\|_{L^3_x}^3\big]\Big),
\label{eq:M1estimate}\end{aligned}$$ where we have chosen $\gamma$ and $\theta$ such that $$\begin{aligned}
C\Big(\theta^{\mu + 1} \gamma^{-\mu} |\Gamma(-\mu + 1)|
+ \theta^{\mu - 1 - \frac{1}{2\alpha}} \gamma^{-\mu - \frac{1}{2\alpha}}\big|\Gamma(-\mu + 1 - \frac{1}{2\alpha})\big| \Big)
\leq \frac{\epsilon_0}{2},\end{aligned}$$ for sufficiently small $\epsilon_0$ to be determined later.
Analysis of $M_2$
-----------------
$$\begin{aligned}
{\langle}M_2, \varphi {\rangle}= \int_0^t\int {\langle}\delta(\cdot- u(x, s)), \varphi\,({\mathcal{S}}(t - s)\sigma(x)){\rangle}\,{{\rm d}}x \;{{\rm d}}W_s.\end{aligned}$$
We again invoke the kernel estimate. In fact, it is here that the kernel estimate becomes indispensable. In the stochastic setting, with a forcing term given by $\sigma(x) \delta(\xi - u(x,t)) {\partial}_t W$, which does not easily lend itself to the space-time Fourier transform, one may not simply take the Fourier transform on both sides so that the factor, $i(\tau + F'(\xi) \cdot k) + {\mathbf{A}}(\xi) : (k\otimes k)$, on the left side can simply be divided out, with a certain genuine nonlinearity ([*i.e.*]{} the non-degeneracy condition; [*cf.*]{} [@CF1998; @LPT1994a; @TT2007]). Thus, we have to find a different way to handle the forcing term.
Expanding the effect of the semigroup, we have $$\begin{aligned}
{\langle}M_2, \varphi {\rangle}= \int_0^t \int_{\mathbb{T}^d} e^{-\theta (t - s)} \varphi
e^{-({\mathbf{A}}(\xi) : \nabla \otimes \nabla +\gamma(-\Delta)^{\alpha})(t - s)} \sigma(x - F'(u(x,s)) (t - s)) \;{{\rm d}}x\,{{\rm d}}W_s.\end{aligned}$$ Since $\sigma$ is bounded in $\mathbb{T}^d$, we see that $\sigma(\cdot - F'(u(\cdot,s))(t - s))$ is bounded in $x$.
The kernel estimate then gives $$\begin{aligned}
&\|e^{-\theta (t - s)} \varphi e^{-({\mathbf{A}}(u) : \nabla \otimes \nabla + \gamma(-\Delta)^{\alpha})(t - s)} \sigma(\cdot- F'(u(\cdot,s))(t - s))\|_{H^{\hat{\beta}}_x}\\
& \leq C \big(\gamma(t - s)\big)^{-\frac{\hat{\beta}}{2\alpha}} \|\sigma\|_{L^2_x},\end{aligned}$$ just as in [@DV2015]. In the same way, we have $$\begin{aligned}
{\mathbb{E}}\big[\big\|\int_0^t {\langle}\delta(\cdot - u(x,s)), \,{\mathcal{S}}(t - s) \sigma(x) {\rangle}\,{{\rm d}}W_s\big\|_{H^{\hat{\beta}}_x}^2\big]
\leq C
\gamma^{-\frac{\hat{\beta}}{\alpha}} \theta^{\frac{\hat{\beta}}{\alpha} - 1} \big|\Gamma(1 - \frac{\hat{\beta}}{\alpha})\big|.\end{aligned}$$ Then we have $$\begin{aligned}
{\mathbb{E}}\big[\|M_2\|_{H^{\hat{\beta}}_x}^2\big]
\leq C
\gamma^{-\frac{\hat{\beta}}{\alpha}} \theta^{\frac{\hat{\beta}}{\alpha} - 1} \big|\Gamma(1 - \frac{\hat{\beta}}{\alpha})\big|.\end{aligned}$$
Completion of the existence proof
---------------------------------
From (\[eq:u\_0\_estimate\])–(\[eq:ubestimate\]), we have $$\begin{aligned}
{\mathbb{E}}\big[\|u^0 + u^\flat +M_2\|_{L^2_tW^{s,q}_x}^2\big]
\leq {\mathbb{E}}\big[\|u(0)\|_{L^1_x}\big] + {\mathbb{E}}\big[\|u\|_{L^{1 }([0,T],L^{1 }_x)}\big] + CT,\end{aligned}$$ where $q>1$ and $s>0$.
By the standard Jensen and Young inequalities, we obtain $$\begin{aligned}
&\frac{1}{T}{\mathbb{E}}^2\big[\|u^0 + u^\flat + M_2 \|_{L^1_tW^{s,q}_x}\big]
\leq {\mathbb{E}}\big[\|u^0 + u^\flat +M_2\|_{L^2_tW^{s,q}_x}^2\big],\end{aligned}$$ so that $${\mathbb{E}}^2\big[\|u^0 + u^\flat + M_2 \|_{L^1_tW^{s,q}_x}\big]
\leq C T \Big({\mathbb{E}}\big[\|u(0)\|_{L^{1}_x}\big] +{\mathbb{E}}\big[\|u\|_{L^{1}([0,T],L^{1 }_x)}\big] + T\Big).$$ Then we have $${\mathbb{E}}\big[\|u^0 + u^\flat + M_2 \|_{L^1_tW^{s,q}_x}\big]
\leq C\Big({\mathbb{E}}\big[\|u(0)\|_{L^{1 }_x}\big] + T\Big)
+ \frac{\epsilon_0}{2} {\mathbb{E}}\big[\|u\|_{L^{1 }([0,T],L^{1 }_x)}\big].$$
From (\[eq:M1estimate\]), we further have $$\begin{aligned}
{\mathbb{E}}\big[\|M_1\|_{L^1([0,T],W^{\hat{\beta},p'}_x)}\big]
\leq\frac{\epsilon_0}{2}\Big(1 + {\mathbb{E}}\big[\|u\|_{L^{1}([0,T],L^{1}_x)}\big] + {\mathbb{E}}\big[\|u_0\|_{L^3_x}^3\big]\Big).\end{aligned}$$
By the continuous embedding $W^{s,q}_x \hookrightarrow L^{1}_x$, $$\begin{aligned}
\label{eq:wsp_bound}
{\mathbb{E}}\big[\|u\|_{L^1([0,T],W^{s,q}_x)}\big]
\leq C(\alpha,\hat{\beta},\gamma,\theta)
\Big(1 + {\mathbb{E}}\big[\|u(0)\|_{L^3_x}^3\big] + T\Big).\end{aligned}$$
Since $W^{s,q}$ is compactly embedded in $L^1$ for $q \geq 1$, the Krylov-Bogoliubov mechanism (§2.2) leads to the existence of an invariant measure.
Stochastic Anisotropic Parabolic-Hyperbolic Equations\
II: Uniqueness of Invariant Measures
======================================================
In this section, we prove the uniqueness of invariant measures for the second-order nonlinear stochastic equations .
Let $F$ and ${\mathbf{A}}$ satisfy the non-degeneracy condition and the boundedness condition[:]{} $$\begin{aligned}
|F''(\xi)| \lesssim 1,\qquad
|{\mathbf{A}}'(\xi)| \lesssim 1. \label{eq:conditions_uniqueness}\end{aligned}$$ Then the invariant measure established in Theorem [4.1]{} is unique.
To show the uniqueness, we first show that the solutions enter a certain ball in $L^1_x$ in finite time almost surely. Then we show that the solutions, starting on a fixed ball, enter arbitrarily small balls almost surely, if the noise is sufficiently small in $W^{1,\infty}$. This allows us to conclude that any pair of balls enters an arbitrarily small ball of one another, since the noise is sufficiently small for any given duration with positive probability. This is the property of recurrence discussed in the coupling method in §2, which implies the uniqueness of invariant measures. In showing the recurrence, we follow §4 of [@DV2015] quite closely.
Uniqueness I: Finite time to enter a ball {#sec:uniquenessI}
-----------------------------------------
The following lemma is proved in the same way as in [@DV2015], via a Borel-Cantelli argument.
\[thm:enterball\_kappa\] There are both a radius $\hat{\kappa}$ [(]{}depending on the initial conditions[)]{} and an almost surely finite stopping time $\mathcal{T}$ such that a solution enters $B_{\hat{\kappa}}(0) \subseteq L^1(\mathbb{T}^d)$ in time $\mathcal{T}$.
The proof uses the coupling method, where $v$ is another solution to the same equation with initial condition $v(0) = v_0$. It furnishes us with the recursively defined sequence of stopping times, with $\mathcal{T}_0 = 0$ and $$\begin{aligned}
\label{eq:stopping_time}
\mathcal{T}_l = \inf\{ t \geq \mathcal{T}_{l - 1} + T \, :\, \|u(t) \|_{L^1_x} + \|v(t)\|_{L^1_x} \leq 2 \hat{\kappa}\},\end{aligned}$$ which are also almost surely finite.
Uniqueness II: Bounds with small noise {#sec:uniquenessII}
--------------------------------------
We now prove the following key lemma for the pathwise solutions:
\[thm:small\_ball\_lemma\] For any ${\epsilon}> 0$, there are $T > 0 $ and $\tilde{\kappa} > 0$ such that, for the initial conditions $u_0$ satisfying $$\|u_0\|_{L^1_x} \leq 2\hat{\kappa},$$ and the noise satisfying $$\sup_{t \in [0,T]}\|\sigma W\|_{W^{1,\infty}_x} \leq \tilde{\kappa},$$ then $$\begin{aligned}
\fint_0^T \|u(t)\|_{L^1_x} \;{{\rm d}}t \leq {\epsilon},\end{aligned}$$ where we have used the symbol $\fint$ to denote the averaged integral.
One of the differences in our estimates from [@DV2015] is that a kernel estimate is used on $v^\sharp_F + v^\sharp_A$, instead of velocity averaging techniques, since the extra derivatives are required to be handled here. Of course, this method can also be applied to the first-order case so that the need to estimate the average term $\fint v^\sharp \;{{\rm d}}x$ in [@DV2015] can be eliminated. We divide the proof into nine steps.
1\. Let $u$ be a solution of $${\partial}_t u + \nabla \cdot F(u) + \nabla \cdot ({\mathbf{A}}(u) \nabla u) = \sigma(x) {\partial}_t W$$ with initial condition $u(0) = u_0$, and let $\tilde{u}$ be the solution to the same equation with initial condition $\tilde{u}_0$ satisfying $$\begin{aligned}
\|u_0 - \tilde{u}_0\|_{L^1_x} \leq \frac{{\epsilon}}{8}, \qquad \|\tilde{u}_0\|_{L^2_x} \leq C \hat{\kappa} {\epsilon}^{-\frac{d}{2}},\end{aligned}$$ which can be found by convolving $u_0$ with a mollifying kernel, where $\hat{\kappa}$ is the radius constant of Lemma \[thm:enterball\_kappa\].
2\. Consider the difference between solution $\tilde{u}$ and noise $\sigma(x)W$: $v = \tilde{u} - \sigma(x) W$, which is a kinetic solution to $$\begin{aligned}
{\partial}_t v = - \nabla \cdot F(v + \sigma(x) W) + \nabla \cdot \big({\mathbf{A}}(v + \sigma(x) W) \nabla (v + \sigma(x)W)\big).\end{aligned}$$
The kinetic formulation for this equation can be derived as in (\[eq:kinetic\_formulation\]): $$\begin{aligned}
&{\partial}_t \chi^v + F'(\xi) \cdot \nabla \chi^v - {\mathbf{A}}(\xi) : \nabla^2 \chi^v\notag \\
&= \big(F'(\xi) - F'(\xi + \sigma(x) W)\big) \cdot \nabla \chi^v
- \nabla \cdot \big( ({\mathbf{A}}(\xi) - {\mathbf{A}}(\xi + \sigma(x) W)) \nabla \chi^v\big)\notag\\
&\quad - F'(\xi + \sigma(x) W) \delta(\xi - v) \cdot \nabla (\sigma(x) W)
+ \nabla \cdot \big({\mathbf{A}}(\xi + \sigma(x)W) \delta(\xi - v) \nabla (\sigma(x)\, W)\big)\notag\\
&\quad - {\partial}_\xi \big(\delta( \xi - v) {\mathbf{A}}(\xi + \sigma(x) W) :\big(\nabla (\sigma(x) W)\otimes \nabla (\sigma(x) W)\big)\big)
\label{eq:kinetic_equation2}\\
&\quad + {\partial}_\xi (m^v + N^v).\notag\end{aligned}$$ A notable difference here is that the parabolic defect measure $N^v$ is not the limit of $$\delta(\xi - (v^{\varepsilon}+ \sigma(x) W)){\mathbf{A}}(\xi) :\big(\nabla (v^{\varepsilon}+ \sigma W) \otimes \nabla (v^{\varepsilon}+ \sigma(x) W)\big),$$ but rather the limit of $$\begin{aligned}
N^u_{\varepsilon}= &\,\delta(\xi - v^{\varepsilon}){\mathbf{A}}(\xi + \sigma(x) W) : \big(\nabla v^{\varepsilon}\otimes \nabla v^{\varepsilon}\big)\notag\\
&\, + \delta(\xi - v^{\varepsilon}){\mathbf{A}}(\xi + \sigma(x) W) : \big(\nabla (\sigma(x) W) \otimes \nabla (\sigma(x) W)\big)\notag\\
&\, + \delta(\xi - v^{\varepsilon}) {\mathbf{A}}(\xi +\sigma(x) W): \big(\nabla v^{\varepsilon}\otimes \nabla (\sigma(x) W)\big). \label{eq:N_u}\end{aligned}$$ The asymmetry in the cross term in failing to contain both $\nabla v \otimes \nabla (\sigma(x) W)$ and $\nabla (\sigma(x) W)\otimes \nabla v$ arises from the fact that the convex entropy used is $\Phi(v)$, instead of $\Phi(v + \sigma W)$. One of the key insights in [@CP2003] is that, using the symmetry and nonnegativity of ${\mathbf{A}}$, ${\mathbf{A}}$ can be written as the square of another symmetric, positive semi-definite matrix so that (\[eq:N\_u\]) is non-negative. The limit of $N^u_{\varepsilon}$ is the non-negative parabolic defect measure $N^u$.
3\. As before, we insert the regularizing operators: $\gamma(- \Delta)^\alpha + \theta I$ (with fixed $\gamma$ and $\theta$ in this case) on both sides. Again, we can decompose the solution into the following components: $$\begin{aligned}
{\langle}v(t), \varphi {\rangle}= {\langle}v^0 + v^\flat + v^\sharp_F + v^\sharp_A + M_F + M_A + M_1 + M_2, \varphi {\rangle},\end{aligned}$$ with $$\begin{aligned}
&v^0(x,t) = \int {\mathcal{S}}(t) \chi^v(\xi,x,0)\;{{\rm d}}\xi ,\\
&v^\flat(x,t) = \int \int_0^t {\mathcal{S}}(s) (\gamma(-\Delta)^\alpha
+ \theta \mathrm{I}) \chi^v(\xi,x,t - s) \;{{\rm d}}s\,{{\rm d}}\xi ,\\
&v^\sharp_F(x,t) = \int \int_0^t {\mathcal{S}}(t - s) \big(F'(\xi) - F'(\xi + \sigma(x) W)\big) \cdot \nabla \chi^v (\xi,x,s) \;{{\rm d}}s\,{{\rm d}}\xi ,\\
&v^\sharp_A(x,t) = -\int \int_0^t {\mathcal{S}}(t - s) \nabla \cdot \big( ({\mathbf{A}}(\xi) - {\mathbf{A}}(\xi + \sigma(x) W))\nabla \chi^v (\xi, x,s)\big) \;{{\rm d}}s\,{{\rm d}}\xi ,\\
&{\langle}M_F, \varphi{\rangle}= - \int \int_0^t F'(v + \sigma(x) W) \cdot \nabla (\sigma(x) W)\, ({\mathcal{S}}^*(t - s) \varphi)(x,v(x,s))\;{{\rm d}}s\,{{\rm d}}x,\\
&{\langle}M_A, \varphi {\rangle}= - \int \int_0^t {\mathbf{A}}(v + \sigma(x)W) : \big(\nabla (\sigma(x) W) \otimes \nabla ({\mathcal{S}}^*(t - s) \varphi)(v(x,s),x)\big)\;{{\rm d}}s\,{{\rm d}}x,\\
&{\langle}M_1, \varphi{\rangle}= - \iint \int_0^t {\partial}_\xi ({\mathcal{S}}^*(t - s) \varphi)\;{{\rm d}}(m^v + N^v) (\xi,x,s),\\
&{\langle}M_2, \varphi {\rangle}= \int \int_0^t {\partial}_\xi ({\mathcal{S}}^*(t - s) \varphi)(v(x,s),x)
{\mathbf{A}}(v + \sigma(x) W):\big(\nabla (\sigma(x) W)\otimes \nabla (\sigma(x) W)\big){{\rm d}}s {{\rm d}}x.\end{aligned}$$
Now we estimate each of these integrals, with some variations from [@DV2015] especially for the terms involving ${\mathbf{A}}$. For this, $C>0$ is a universal constant, independent of ${\epsilon}, \tilde{\kappa}$, and $T$.
4\. We first have the familiar estimates: $$\int_0^T\|v^0(t)\|_{H^{\alpha}_x}^2 \;{{\rm d}}t \leq C \gamma^r\|u_0\|_{L^1_x},$$ and $$\int_0^T\|v^\flat(t)\|_{L^2_x}^2 \;{{\rm d}}t \leq C \gamma^{r + 1} \int_0^T \|v(t)\|_{L^1_x}\;{{\rm d}}t$$ from the velocity averaging arguments, where $|r| < 1$ (we see that there is an extra power of $\gamma$ in the second estimate from those arguments, no matter what $r$ might be).
These imply $$\begin{aligned}
&\fint_0^T \|v^0 \|_{L^1_x} \;{{\rm d}}t \leq C T^{-\frac{1}{2}}\gamma^{\frac{r}{2}} \|u_0\|_{L^1_x}^{\frac{1}{2}}, \label{eq:smallball1}\\
&\fint_0^T\|v^\flat(t)\|_{L^1_x} \;{{\rm d}}t \leq C \gamma^{\frac{r + 1}{2}} \Big(\fint_0^T \|v(t)\|_{L^1_x}\;{{\rm d}}t \Big)^{\frac{1}{2}}.
\label{eq:smallball2}\end{aligned}$$
5\. For $v^\sharp_F$ and $v^\sharp_A$, we use the fact that $$\begin{aligned}
&\big(F'(\xi) - F'(\xi + \sigma(x) W)\big)\cdot \nabla\chi^v(\xi,x,s)\\
&\quad =\nabla \cdot \big((F'(\xi) - F'(\xi + \sigma(x) W) ) \chi^v(\xi,x,s)\big)
- \big(F''(\xi + \sigma(x) W) \cdot \nabla \sigma(x)W\big)\chi^v(\xi,x,s), \\
&\big({\mathbf{A}}(\xi) - {\mathbf{A}}(\xi + \sigma(x) W)\big) \nabla \chi^v(\xi,x,s) \\
&\quad = \nabla \cdot \big(({\mathbf{A}}(\xi) - {\mathbf{A}}(\xi + \sigma(x) W) ) \chi^v(\xi,x,s)\big)
- \big({\mathbf{A}}'(\xi + \sigma(x) W) \nabla \sigma(x)W \big)\chi^v(\xi,x,s).\end{aligned}$$
Now we apply the kernel estimates. Let $\varphi \in L^2$ be any test function, and let ${\langle}\cdot, \cdot {\rangle}$ be the pairing in $L^2$. Then $$\begin{aligned}
{\langle}v^\sharp_F(t), \varphi {\rangle}= & \iint \int_0^t \varphi {\mathcal{S}}(t - s) \nabla \cdot\big((F'(\xi) - F'(\xi + \sigma(x) W))\, \chi^v(\xi,x,s)\big) \;{{\rm d}}s\,{{\rm d}}\xi \,{{\rm d}}x\notag \\
& - \iint \int_0^t \varphi\, {\mathcal{S}}(t - s) \big((F''(\xi + \sigma(x) W)\cdot\nabla \sigma(x) W)\,\chi^v(\xi,x,s)\big) \;{{\rm d}}s\,{{\rm d}}\xi \,{{\rm d}}x \notag\\
= & \iint \int_0^t\nabla({\mathcal{S}}^*(t - s)\varphi) \cdot\big(F'(\xi) - F'(\xi + \sigma(x) W)\big)\, \chi^v(\xi,x,s)\;{{\rm d}}s\,{{\rm d}}\xi \,{{\rm d}}x \notag\\
& - \iint \int_0^t {\mathcal{S}}^*(t - s) \varphi\,\big(F''(\xi + \sigma(x) W)\cdot \nabla \sigma(x) W\big) \chi^v(\xi,x,s)\;{{\rm d}}s\,{{\rm d}}\xi \,{{\rm d}}x. \label{5.7a}\end{aligned}$$ Similarly, we have $$\begin{aligned}
{\langle}v^\sharp_A(t), \varphi {\rangle}= & \iint \int_0^t \varphi {\mathcal{S}}(t - s) \nabla^2 : \big(({\mathbf{A}}(\xi) - {\mathbf{A}}(\xi + \sigma(x) W)) \chi^v(\xi,x,s)\big) \;{{\rm d}}s\,{{\rm d}}\xi \,{{\rm d}}x\notag\\
& - \iint \int_0^t \varphi {\mathcal{S}}(t - s) \nabla \cdot \big({\mathbf{A}}'(\xi + \sigma(x) W) \nabla (\sigma(x)W) \chi^v(\xi,x,s)\big) \;{{\rm d}}s\,{{\rm d}}\xi \,{{\rm d}}x\notag\\
=& \iint \int_0^t \nabla^2({\mathcal{S}}^*(t - s)\varphi) : \big({\mathbf{A}}(\xi) - {\mathbf{A}}(\xi + \sigma(x) W)\big) \chi^v(\xi,x,s) \;{{\rm d}}s\,{{\rm d}}\xi \,{{\rm d}}x\notag\\
& - \iint \int_0^t \big(\nabla ({\mathcal{S}}^*(t - s)\varphi) \otimes \nabla \sigma(x) W\big): {\mathbf{A}}'(\xi + \sigma(x) W)\, \chi^v(\xi,x,s)\;{{\rm d}}s\,{{\rm d}}\xi \,{{\rm d}}x. \label{5.7b}\end{aligned}$$
Notice that $$\begin{aligned}
&\int_0^T \int_0^t\|\nabla({\mathcal{S}}^*(t - s)\varphi) \cdot\big(F'(\cdot) - F'(\cdot + \sigma(\cdot) W)\big)\, \chi^v(\cdot,\cdot,s) \|_{L^1_{x,\xi}} \;{{\rm d}}s\,{{\rm d}}t \notag\\
&\leq \int_0^T \int_0^t\|\nabla({\mathcal{S}}^*(t - s)\varphi) \|_{L^\infty_{x,\xi}} \|F'(\cdot) - F'(\cdot + \sigma(\cdot) W)\|_{L^\infty_{x,\xi}} \|\chi^v(\cdot,\cdot,s) \|_{L^1_{x,\xi}}\;{{\rm d}}s\,{{\rm d}}t \notag\\
&\leq \int_0^T \int_0^t \|\nabla{\mathcal{S}}^*(t - s) \|_{L^2 \to L^\infty} \|\varphi\|_{L^2_x} \|F'(\cdot) - F'(\cdot + \sigma(\cdot) W)\|_{L^\infty_{x,\xi}} \|\chi^v(\cdot,\cdot,s) \|_{L^1_{x,\xi}}\;{{\rm d}}s\,{{\rm d}}t \notag\\
&\leq C\tilde{\kappa}\|\varphi\|_{L^2_x}\sup_{s \in [0,T]} \Big(\int_0^T e^{\theta (t - s)} (\gamma t)^{-\frac{d + 2}{4\alpha}} \;{{\rm d}}t\Big)
\int_0^T \|v(s)\|_{L^1_x}\;{{\rm d}}s\notag\\
&\leq C\tilde{\kappa}\|\varphi\|_{L^2_x}\gamma^{-\frac{d + 2}{4\alpha}} \theta^{\frac{d + 2}{4 \alpha} - 1}\int_0^\infty e^{-t} t^{-\frac{d + 2}{4\alpha}} \;{{\rm d}}t \;\int_0^T \|v(s)\|_{L^1_x}\;{{\rm d}}s\notag \\
&= C\tilde{\kappa} \|\varphi\|_{L^2_x} \gamma^{-\frac{d + 2}{4\alpha}} \theta^{\frac{d + 2}{4 \alpha} - 1} \big|\Gamma(1 - \frac{d + 2}{4 \alpha})\big|\;\int_0^T \|v(s)\|_{L^1_x}\;{{\rm d}}s; \label{5.7a-1}\end{aligned}$$
$$\begin{aligned}
&\int_0^T \int_0^t\|\big({\mathcal{S}}^*(t - s)\varphi\big)\, \nabla (\sigma(x) W)\cdot\big(F''(\cdot + \sigma(\cdot) W) \chi^v(\cdot,\cdot,s)\big) \|_{L^1_{x,\xi}}\;{{\rm d}}s\,{{\rm d}}t \notag\\
&\leq \int_0^T \int_0^t\|{\mathcal{S}}^*(t - s)\varphi \|_{L^\infty_{x,\xi}} \|F''(\cdot + \sigma(\cdot) W)\|_{L^\infty_{x,\xi}} \|\sigma W\|_{W^{1,\infty}_x} \|\chi^v(\cdot,\cdot,s)\|_{L^1_{x,\xi}}\;{{\rm d}}s\,{{\rm d}}t \notag\\
&\leq C\tilde{\kappa}\|\varphi\|_{L^2_x}\gamma^{-\frac{d}{4\alpha}} \theta^{\frac{d}{4\alpha} -1} \big|\Gamma(1 -\frac{d}{4\alpha})\big|\;\int_0^T \|v(s)\|_{L^1_x}\;{{\rm d}}s;\label{5.7a-2}\end{aligned}$$
$$\begin{aligned}
&\int_0^T \int_0^t\|\nabla^2({\mathcal{S}}^*(t - s)\varphi):\big({\mathbf{A}}(\cdot) - {\mathbf{A}}(\cdot + \sigma(\cdot) W)\big)\, \chi^v(\cdot,\cdot,s)\|_{L^1_{x,\xi}}\;{{\rm d}}s\,{{\rm d}}t\notag\\
&\leq \int_0^T \int_0^t\|\nabla^2({\mathcal{S}}^*(t - s)\varphi) \|_{L^\infty_{x,\xi}} \|{\mathbf{A}}(\cdot) - {\mathbf{A}}(\cdot + \sigma(\cdot) W)\|_{L^\infty_{x,\xi}} \|\chi^v(\cdot,\cdot,s)\|_{L^1_{x,\xi}}\;{{\rm d}}s\,{{\rm d}}t\notag\\
&\leq C \tilde{\kappa} \|\varphi\|_{L^2_x}\gamma^{-\frac{d + 4}{4\alpha}} \theta^{\frac{d + 4}{4 \alpha} - 1}
\big|\Gamma(1 - \frac{d + 4}{4 \alpha})\big| \;\int_0^T \|v(s)\|_{L^1_x}\;{{\rm d}}s;\label{5.7b-1}\end{aligned}$$
$$\begin{aligned}
&\int_0^T \int_0^t\|\big(\nabla({\mathcal{S}}^*(t - s)\varphi) \otimes \nabla (\sigma(\cdot) W)\big) :{\mathbf{A}}'(\cdot + \sigma(\cdot) W)\, \chi^v(\cdot,\cdot,s) \|_{L^1_{x,\xi}}\;{{\rm d}}s\,{{\rm d}}t\notag\\
&\leq \int_0^T \int_0^t\|\nabla({\mathcal{S}}^*(t - s)\varphi) \|_{L^\infty_{x,\xi}} \|\sigma W\|_{W^{1,\infty}_x}
\|{\mathbf{A}}'(\cdot + \sigma(\cdot) W)\|_{L^\infty_{x,\xi}} \|\chi^v(\cdot,\cdot,s) \|_{L^1_{x,\xi}}\;{{\rm d}}s\,{{\rm d}}t\notag\\
&\leq C \tilde{\kappa} \|\varphi\|_{L^2_x}\gamma^{-\frac{d + 2}{4\alpha}} \theta^{\frac{d + 2}{4\alpha} - 1}
\big|\Gamma(1 - \frac{d + 2}{4\alpha})\big| \;\int_0^T \|v(s)\|_{L^1_x}\;{{\rm d}}s.\label{5.7b-2}\end{aligned}$$
Now, by (\[eq:conditions\_uniqueness\]), we have assumed that $$\begin{aligned}
|F''(\xi)| \lesssim 1,\qquad
|{\mathbf{A}}'(\xi)| \lesssim 1,\end{aligned}$$ and $\|\sigma(x) W\|_{W^{1,\infty}} \leq \tilde{\kappa}$, so that we can use the estimates (the second from the first by the Poincaré-Wirtinger inequality, since $\int_{\mathbb{T}^d} \sigma(x) \;{{\rm d}}x = 0$): $$\begin{aligned}
&|F'(\xi) - F'(\xi + \sigma(x) W)|
+|F''(\xi + \sigma(x) W) \cdot \nabla (\sigma W)|\leq C \tilde{\kappa}, \\
&|{\mathbf{A}}(\xi) - {\mathbf{A}}(\xi + \sigma(x) W) |+
|{\mathbf{A}}'(\xi + \sigma(x) W) \nabla (\sigma W) |\leq C \tilde{\kappa}.\end{aligned}$$
Putting these estimate – back into the bound: $\|v^\sharp_\cdot(t)\|_{L^2_x} = \sup_{\|\varphi\|_{L^2_x} = 1} {\langle}v^\sharp_\cdot(t) , \varphi{\rangle}$, we have $$\begin{aligned}
&\int_0^T \|v^\sharp_A(t) + v^\sharp_F(t)\|_{L^2_x} \;{{\rm d}}t \notag \\
&\leq C\tilde{\kappa} \Big(\gamma^{-\frac{d + 2}{4\alpha}} \theta^{\frac{d + 2}{4 \alpha} - 1}|\Gamma(1 - \frac{d + 2}{4\alpha})|
+ \gamma^{-\frac{d}{4\alpha}} \theta^{\frac{d}{4\alpha} -1} |\Gamma(1 -\frac{d}{4\alpha} )|\notag\\
&\quad \qquad + \gamma^{-\frac{d + 4}{4\alpha}} \theta^{\frac{d + 4}{4 \alpha} - 1}|\Gamma(1 - \frac{d + 4}{4\alpha})|\Big)
\int_0^T \|v(t)\|_{L^1_x}\;{{\rm d}}t . \label{eq:smallball3}\end{aligned}$$
6\. For $M_F$ and $M_A$, we employ the kernel estimate and $\|\sigma W\|_{W^{1,\infty}_x} \leq \tilde{\kappa}$ to obtain $$\begin{aligned}
&|{\langle}M_F, \varphi {\rangle}| \leq \int_0^t\|F'(v + \sigma W)\|_{L^1_x} \| {\mathcal{S}}\varphi\|_{L^\infty_x}\|\nabla \sigma W\|_{L^\infty_x}\;{{\rm d}}s,\\
& |{\langle}M_A, \varphi{\rangle}|\leq \int_0^t \|{\mathbf{A}}(v + \sigma W)\|_{L^1_x} \| \nabla ({\mathcal{S}}\varphi)\|_{L^\infty_x}\|\nabla \sigma W\|_{L^\infty_x}\;{{\rm d}}s.\end{aligned}$$
Now, by (\[eq:conditions\_existence\]), we have $$\begin{aligned}
\|F'(v + \sigma W)\|_{L^1_x}+\|{\mathbf{A}}(v + \sigma W)\|_{L^1_x} \leq C \big(1 + \|v(t)\|_{L^1_x} + \|\sigma \|_{L^1_x}|W|\big).\end{aligned}$$
These give $$\begin{aligned}
&\int_0^T \|M_F(t) + M_A(t)\|_{L^1_x} \;{{\rm d}}t \notag \\
&\leq C \tilde{\kappa}\int_0^T \int_0^t \big(1 + \|v(s)\|_{L^1_x}\big) e^{-\theta(t - s)}\big(1 + (\gamma (t - s))^{ -\frac{1}{2\alpha}}\big) \;{{\rm d}}s \;{{\rm d}}t \notag\\
&\leq C\tilde{\kappa} \Big(\theta^{-1} + \gamma^{ -\frac{1}{2\alpha}} \theta^{\frac{1}{2\alpha} - 1} |\Gamma(1 - \frac{1}{2\alpha})|\Big)
\int_0^T \big(1 + \|v(s)\|_{L^1_x}\big) \;{{\rm d}}s.
\label{eq:smallball4}\end{aligned}$$
7\. For $M_2$, we have $$\begin{aligned}
{\langle}M_2, \varphi {\rangle}=\int \int_0^t {\partial}_\xi ({\mathcal{S}}^*(t - s) \varphi)(v(x,s),x) {\mathbf{A}}(v + \sigma(x) W) :\big(\nabla (\sigma(x) W)\otimes \nabla(\sigma(x) W)\big)\,{{\rm d}}s\,{{\rm d}}x.\end{aligned}$$ We notice that $$\begin{aligned}
{\partial}_\xi ({\mathcal{S}}(t-s)\varphi) (v(x,s),x)= (t - s) F''(v(x,s)) \cdot \nabla ({\mathcal{S}}^*(t - s)\varphi) + {\mathbf{A}}'(v(x,s)):\nabla^2 ({\mathcal{S}}^*(t - s) \varphi),\end{aligned}$$ as explained in (\[eq:pd\_xi\_se\]). By (\[eq:conditions\_uniqueness\]), we have assumed that $$\begin{aligned}
|F''(\xi)| \lesssim 1,\qquad
|{\mathbf{A}}'(\xi)| \lesssim 1.\end{aligned}$$
Again we have $$\begin{aligned}
\|{\mathbf{A}}(v + \sigma(x) W) \|_{L^1_x} \leq C\big(1 + \|v(s)\|_{L^1_x} + \|\sigma\|_{L^1_x}|W|\big).\end{aligned}$$
Finally, using the kernel estimate yields $$\begin{aligned}
&|{\langle}M_2,\varphi{\rangle}|\\
&\leq C\int_0^t (t - s)\|F''\|_{L^\infty}\|\nabla S^*(t - s) \varphi\|_{L^\infty_{x,\xi}} \|\nabla \sigma W\|_{L^\infty_x}^2
\big(1 + \|v(s)\|_{L^1_x} + \|\sigma\|_{L^1_x}|W|\big) \;{{\rm d}}s\\
&\quad + C\int_0^t (t - s)\|{\mathbf{A}}'\|_{L^\infty}\|\nabla^2 S^*(t - s) \varphi\|_{L^\infty_{x,\xi}} \|\nabla \sigma W\|_{L^\infty_x}^2
\big(1 + \|v(s)\|_{L^1_x} + \|\sigma\|_{L^1_x}|W|\big) \;{{\rm d}}s\\
&\leq C\tilde{\kappa}^2\int_0^t (t - s)\big(\|\nabla S^*(t - s) \|+\|\nabla^2 S^*(t - s)\|\big)
\|\varphi\|_{L^\infty_x} \big(1 + \|v(s)\|_{L^1_x} + \|\sigma\|_{L^1_x}|W|\big)\;{{\rm d}}s.\end{aligned}$$ Therefore, we have $$\begin{aligned}
&\int_0^T \|M_2(t)\|_{L^1_x} \;{{\rm d}}t \\
&\leq C\tilde{\kappa}^2 \int_0^T\int_0^t (t - s)
\big(\|\nabla S^*(t - s) \|+\|\nabla^2 S^*(t - s)\|\big)
\big(1 + \|v(s)\|_{L^1_x} + \|\sigma\|_{L^1_x}|W|\big) \;{{\rm d}}s\,{{\rm d}}t \\
&\leq C\tilde{\kappa}^2 \int_0^T\int_0^t (t - s)\big((\gamma (t - s))^{-\frac{1}{2\alpha}}+(\gamma (t - s))^{-\frac{1}{\alpha}}\big)
e^{-\theta (t - s)}\big(1 + \|v(s)\|_{L^1_x} + \|\sigma\|_{L^1_x}|W|\big)\,{{\rm d}}s\, {{\rm d}}t,\end{aligned}$$ and $$\begin{aligned}
&\int_0^T \|M_2(t)\|_{L^1_x} \;{{\rm d}}t \label{eq:smallball5}\\
&\leq C \tilde{\kappa}^2 \Big(\gamma^{-\frac{1}{2\alpha}} \theta^{\frac{1}{2\alpha} - 2} \big|\Gamma(2 - \frac{1}{2\alpha})\big|
+ \gamma^{-\frac{1}{\alpha}} \theta^{\frac{1}{\alpha} - 2} \big|\Gamma(2 - \frac{1}{\alpha})\big|\Big)\int_0^T \big(1 + \|v(s)\|_{L^1_x}\big) \;{{\rm d}}s.\notag\end{aligned}$$
8\. For the kinetic measure $M_1$, we use the total variation estimate again. First, with $\varphi \in L^\infty_x$, $$\begin{aligned}
&|{\langle}M_1, \varphi{\rangle}| \\
&= \left| \iint \int_0^t {\partial}_\xi ({\mathcal{S}}^*(t - s) \varphi)\; {{\rm d}}(m^v + N^v)(x,\xi,s) \right|\\
&= \left|\iint \int_0^t \big((t - s) F''(\xi) \cdot \nabla ({\mathcal{S}}^*(t - s)\varphi) + {\mathbf{A}}'(\xi) : \nabla^2 ({\mathcal{S}}^*(t - s) \varphi)\big)\; {{\rm d}}(m^v + N^v)(x,\xi,s)\right|\\
&\leq C\|\varphi\|_{L^\infty_x} \iint \int_0^t \big(\gamma^{-\frac{1}{2\alpha}}(t - s)^{1 - \frac{1}{2\alpha}}
+\gamma^{-\frac{1}{\alpha}}(t - s)^{1 - \frac{1}{\alpha}}\big)e^{-\theta (t - s)} {{\rm d}}|m^v + N^v|(x,\xi,s),\end{aligned}$$ so that $$\begin{aligned}
&\int_0^T \| M_1(t)\|_{L^1_x}\;{{\rm d}}t \\
&\leq C\Big(\gamma^{-\frac{1}{2\alpha}} \theta^{\frac{1}{2\alpha} - 2} \big|\Gamma(2 - \frac{1}{2\alpha})\big|
+\gamma^{-\frac{1}{\alpha}} \theta^{\frac{1}{\alpha} - 2} \big|\Gamma(2 - \frac{1}{\alpha})\big|\Big)|m^v + N^v|({\mathbb{R}}\times \mathbb{T}^d\times [0,T]).\end{aligned}$$
As in Lemma \[thm:TVestimate1\], we test equation (\[eq:kinetic\_equation2\]) against $\xi$ to find $$\begin{aligned}
&\frac{1}{2}\|v(t)\|_{L^2_x}^2 + |m^v + N^v|(\mathbb{R}\times \mathbb{T}^d\times[0,t])\\
&\leq \frac{1}{2} \|\tilde{u}_0\|_{L^2_x}^2 + \left|\int_0^t \iint \xi \big(F'(\xi) - F'(\xi + \sigma(x) W)\big) \cdot \nabla \chi^v \;{{\rm d}}\xi \,{{\rm d}}x\,{{\rm d}}s\right|\\
& \quad + \left|\int_0^t \int v F'(v + \sigma(x) W) \cdot \nabla (\sigma(x) W) \;{{\rm d}}x\,{{\rm d}}s\right|\\
& \quad + \left|\int_0^t \int {\mathbf{A}}(v + \sigma(x) W) : \big(\nabla(\sigma W) \otimes \nabla (\sigma W)\big) \;{{\rm d}}x\,{{\rm d}}s\right|\\
&\leq \frac{1}{2} \|\tilde{u}_0\|_{L^2_x}^2 + \left|\int_0^t \iint \xi F''(\xi + \sigma(x) W)) \cdot \nabla(\sigma(x) W) \chi^v \;{{\rm d}}\xi \,{{\rm d}}x\,{{\rm d}}s\right|\\
&\quad + C\tilde{\kappa}(1+\tilde{\kappa}) \left|\int_0^t \int v \big(1 + |v| + |\sigma(x) W|\big) \;{{\rm d}}x\,{{\rm d}}s\right|\\
&\leq \frac{1}{2} \|\tilde{u}_0\|_{L^2_x}^2 + C\tilde{\kappa} \int_0^t \big(1 + \|v(s)\|_{L^2_x}^2\big)\;{{\rm d}}s.\end{aligned}$$
Then Gronwall’s inequality implies $$\begin{aligned}
|m^v + N^v|\leq C e^{C\tilde{\kappa} t} \big(\|\tilde{u}_0\|_{L^2_x}^2 + 1\big) \leq C e^{C\tilde{\kappa} t} \big( \hat{\kappa}^2 {\epsilon}^{-d} + 1\big).\end{aligned}$$ Therefore, we have $$\begin{aligned}
&\int_0^T \| M_1(t)\|_{L^1_x}\;{{\rm d}}t \label{eq:smallball6}\\
&\leq C\Big(\gamma^{-\frac{1}{2\alpha}} \theta^{\frac{1}{2\alpha} - 2} \big|\Gamma(2 - \frac{1}{2\alpha})\big|
+ \gamma^{-\frac{1}{\alpha}} \theta^{\frac{1}{\alpha} - 2} \big|\Gamma(2 - \frac{1}{\alpha})\big|\Big) e^{C\tilde{\kappa} T} \big( \hat{\kappa}^2 {\epsilon}^{-d} + 1\big).
\notag\end{aligned}$$
9\. *Completion of the estimates*. First, we set $\alpha \leq \frac{1}{2}$ so that the instances of $|\Gamma|$ are never evaluated at a negative integer, where it is infinite. With the finite bound of all the values of $|\Gamma|$ and finitely many instances of $\Gamma$ in estimates (\[eq:smallball1\])–(\[eq:smallball6\]) above, we can write those estimates as $$\begin{aligned}
&\fint_0^T \|v^0(t)\|_{L^1_x}\;{{\rm d}}t \leq CT^{-\frac{1}{2}}\gamma^{\frac{r}{2}} \|u_0\|_{L^1_x}^{\frac{1}{2}},\\
&\fint_0^T \|v^\flat(t)\|_{L^1_x}\;{{\rm d}}t \leq C\gamma^{\frac{r + 1}{2}}
\Big(\fint_0^T \|v\|_{L^1_x}\;{{\rm d}}t \Big)^{\frac{1}{2}}
\leq C\gamma^{\frac{r + 1}{2}} \fint_0^T \big(1 + \|v\|_{L^1_x}\big)\;{{\rm d}}t ,\\
&\fint_0^T \|v^\sharp_F + v^\sharp_A\|_{L^1_x}\;{{\rm d}}t
\leq C\tilde{\kappa} \Big(\gamma^{-\frac{d + 2}{4\alpha}} \theta^{\frac{d + 2}{4\alpha} - 1} + \gamma^{-\frac{d}{4\alpha}} \theta^{\frac{d}{4\alpha} - 1}
+ \gamma^{-\frac{d + 4}{4\alpha}} \theta^{\frac{d + 4}{4\alpha} - 1}\Big) \fint_0^T \|v(t)\|_{L^1_x} \;{{\rm d}}t ,\\
&\fint_0^T \|M_F(t) + M_A(t)\|_{L^1_x}\;{{\rm d}}t \leq C\tilde{\kappa} \left(\theta^{-1} + \gamma^{-\frac{1}{2\alpha}} \theta^{\frac{1}{2\alpha} - 1}\right)
\fint_0^T \big(1 + \|v(t)\|_{L^1_x} \big)\;{{\rm d}}t ,\\
&\fint_0^T \|M_2(t)\|_{L^1_x} \;{{\rm d}}t \leq C\tilde{\kappa}^2 \left( \gamma^{-\frac{1}{2\alpha}} \theta^{\frac{1}{2\alpha} - 2}
+ \gamma^{-\frac{1}{\alpha}} \theta^{\frac{1}{\alpha} - 2}\right) \fint_0^T\big(1 + \|v(t)\|_{L^1_x}\big)\;{{\rm d}}t ,\\
&\fint_0^T \|M_1\|_{L^1_x} \;{{\rm d}}t \leq \frac{C}{T} \left( \gamma^{-\frac{1}{2\alpha}} \theta^{\frac{1}{2\alpha} - 2}
+ \gamma^{-\frac{1}{\alpha}} \theta^{\frac{1}{\alpha} - 2}\right) e^{C_0\tilde{\kappa} T}\big(\hat{\kappa}^2 {\epsilon}^{-d} + 1\big).\end{aligned}$$
Combining all the estimates together yields $$\begin{aligned}
\fint_0^T \|v(t)\|_{L^1_x} \;{{\rm d}}t
\leq&\, C_1T^{-\frac{1}{2}}\gamma^{\frac{r}{2}} \|u_0\|_{L^1_x}^{\frac{1}{2}}
+ \Big(C_2\gamma^{\frac{r + 1}{2}}
+ C_3(\gamma,\theta) (\tilde{\kappa} + \tilde{\kappa}^2)\Big) \Big(1 + \fint_0^T \|v\|_{L^1_x}\;{{\rm d}}s \Big)\\
&\, + C_4(\gamma,\theta) T^{-1}e^{c\tilde{\kappa} T} (\hat{\kappa}^2 {\epsilon}^{-d} + 1).\end{aligned}$$
We can choose $\gamma$, $\theta$, $T$, and $\tilde{\kappa}$ in that order so that, for some $q$ to be determined, $$\begin{aligned}
C_2 \gamma^{\frac{r +1}{2}} \leq q{\epsilon}.\end{aligned}$$ For $\alpha < \frac{1}{4}$, we see that every $\theta$ has positive power above, except in $C_3(\gamma,\theta)$ for the estimate of $\|M_F + M_A\|_{L^1_{x,t}}$, so that $C(\rho,\theta)$ involves $\theta$ with positive power. Therefore, we choose $\theta$ such that $$C_4(\gamma,\theta) < 1,$$ so that we can choose $T$ sufficiently large such that $$C_1T^{-\frac{1}{2}} \|u_0\|_{L^1_x}^{\frac{1}{2}}
+ C_4(\gamma,\theta)T^{-1}\big(\hat{\kappa}^2{\epsilon}^{-d} + 1\big)
\leq q{\epsilon}.$$ Finally, we choose $\tilde{\kappa}$ such that $$C_3(\gamma, \theta) \tilde{\kappa}(1 + \tilde{\kappa}) \leq q{\epsilon},$$ and $$C_0 \tilde{\kappa} T \leq q {\epsilon}.$$
By taking $q$ sufficiently small, we have $$\begin{aligned}
\fint_0^T \|v(t)\|_{L^1_x} {{\rm d}}t \leq \frac{{\epsilon}}{4}, \qquad \fint_0^T \|\tilde{u}(t) \|_{L^1_x} \;{{\rm d}}t \leq \frac{3{\epsilon}}{8},\end{aligned}$$ which leads to $$\begin{aligned}
\fint_0^T \|u(t)\|_{L^1_x} \;{{\rm d}}t \leq \frac{{\epsilon}}{2},\end{aligned}$$ by the following $L^1$-contraction property of the pathwise solutions:
Let ${\mathbf{A}}$ be symmetric positive-semi-definite, and let both ${\mathbf{A}}(\xi)$ and $F(\xi)$ be Hölder continuous and of polynomial growth. Then, for each initial data function $u_0$, there exists a unique measurable $u: \mathbb{T}^d\times [0,T] \times \Omega \to \mathbb{R}$ solving in the sense of Definition [\[def:ksolution\]]{}. Moreover, for $u_0, \tilde{u}_0 \in L^1(\mathbb{T}^d)$, $$\|u(t)-\tilde{u}(t)\|_{L^1_x}\le \|u_0-\tilde{u}_0\|_{L^1_x} \qquad \mbox{almost surely}.$$
For our case, $$\|u(t)-\tilde{u}(t)\|_{L^1_x}\le \|u_0-\tilde{u}_0\|_{L^1_x}\le \frac{{\epsilon}}{8} \qquad \mbox{almost surely}.$$ This completes the proof.
This almost sure $L^1$–contraction property is not available in the multiplicative case. In fact, it is not available in many other situations, such as in systems or for non-conservative equations where the $L^1$-contraction is not ready to provide a stability condition. It is of interest to study the uniqueness and ergodicity properties of invariant measures for equations without this property.
Uniqueness III: Conclusion {#sec:uniquenessIII}
--------------------------
Let $u^1_0$ and $u^2_0$ be in $L^1_x$. For a given ${\epsilon}> 0$, let $\tilde{u}^1_0$ and $\tilde{u}^2_0$ be in $L^3_x$ such that $$\|u^i_0 - \tilde{u}^i_0\|_{L^1_x} \leq \frac{{\epsilon}}{4}.$$ Denote their corresponding solutions $u^1,u^2,\tilde{u}^1$, and $\tilde{u}^2$, respectively. Let us now put $\tilde{u}^1$ and $\tilde{u}^2$, in place of $u$ and $v$ in §\[sec:uniquenessI\], and the corresponding sequence of stopping times constructed recursively in (\[eq:stopping\_time\]): $$\begin{aligned}
\mathcal{T}_l = \inf\{ t \geq \mathcal{T}_{l - 1} + T : \|\tilde{u}^1(t) \|_{L^1_x} + \|\tilde{u}^2(t)\|_{L^1_x} \leq 2 \hat{\kappa}\}.\end{aligned}$$
As in [@DV2015], choosing $T$ and $\tilde{\kappa}$ as above, we obtain by the $L^1$–contraction (for the additive noise, there is the $L^1$-contraction almost sure): $$\begin{aligned}
&\mathbb{P}\big\{\fint_{\mathcal{T}_l}^{\mathcal{T}_l + T} \|u^1(s) - u^2(s)\|_{L^1_x }\;{{\rm d}}s \leq {\epsilon}\,\, \big| \,\mathscr{F}_{\mathcal{T}_l}\big\}\\
&\geq \mathbb{P} \big\{\fint_{\mathcal{T}_l}^{\mathcal{T}_l + T} \|\tilde{u}^1(s) - \tilde{u}^2(s)\|_{L^1_x} \;{{\rm d}}s \leq \frac{{\epsilon}}{2}\,\, \big|\,\mathscr{F}_{\mathcal{T}_l}\big\}\\
&\geq \mathbb{P}\big\{\sup_{t \in [\mathcal{T}_l, \mathcal{T}_l + T]} \|\sigma W(t) - \sigma W(\mathcal{T}_l)\|_{W^{1,\infty}_x} \leq \tilde{\kappa}\,\, \big|\,\mathscr{F}_{\mathcal{T}_l}\big\}.\end{aligned}$$
Since $\tilde{\kappa} > 0$, and $\sigma$ is Lipschitz, we can denote the positive probability of the event as $\lambda$. By the strong Markov property, we know that it does not change with $l$.
This allows us to write $$\begin{aligned}
&\mathbb{P}\big\{\fint_{\mathcal{T}_l}^{\mathcal{T}_l + T} \|u^1(s) - u^2(s)\|_{L^1_x }\;{{\rm d}}s \geq {\epsilon}\,\,\, \mbox{ for } l = l_0, l_0 + 1, \ldots, l_0 + k \big\}
\leq (1 - \lambda)^k,\end{aligned}$$ so that $$\begin{aligned}
&\mathbb{P}\big\{\lim_{l \to \infty} \fint_{\mathcal{T}_l}^{\mathcal{T}_l + T} \|u^1(s) - u^2(s)\|_{L^1_x}\;{{\rm d}}s \geq {\epsilon}\big\}\\
&= \mathbb{P}\big\{\exists l_0 \,\,\forall l \geq l_0 \,:\, \fint_{\mathcal{T}_l}^{\mathcal{T}_l + T} \|u^1(s) - u^2(s)\|_{L^1_x}\;{{\rm d}}s \geq {\epsilon}\big\}\\
&= 0.\end{aligned}$$ This limit exists as $s \mapsto \|u^1(s) - u^2(s)\|_{L^1_x}$ is non-increasing, by the $L^1$–contraction property. Then, by the same property, $$\mathbb{P}\big\{\lim_{t \to \infty} \|u^1(t) - u^2(t) \|_{L^1_x} \geq {\epsilon}\big\} = 0.$$ Therefore, almost surely, $$\lim_{t \to \infty} \|u^1(t) - u^2(t) \|_{L^1_x} = 0,$$ which implies the uniqueness of the invariant measure.
Further Developments, Problems, and Challenges
==============================================
In this section, we discuss some further developments, problems, and challenges in this direction.
Further problems
----------------
There are several natural problems that follow from the analysis discussed above. We restrict ourselves again to nonlinear conservation laws driven by stochastic forcing.
One of the problems is the long-time behavior problem for solutions of nonlinear conservation laws driven by multiplicative noises. The noises of form $\nabla \cdot (F(u) \circ {{\rm d}}W)$ have been considered in [@GS2014; @GS2016], in which the dynamics remains in the zero-spatial-average subspace of $L^1(\mathbb{T}^d)$.
The well-posedness for nonlinear conservation laws driven by multiplicative noises is quite well understood from several different perspectives – the strong entropy stochastic solutions of Feng-Nualart [@FN2008] and of Chen-Ding-Karlsen [@CDK2012], the viscosity solution methods of Bauzet-Vallet-Wittbold[@BVW2012], and the kinetic approach of Debussche-Hofmanovà-Vovelle [@DHV2014; @DV2010], as we have mentioned above. Nevertheless, the long-time behavior problem for solutions is wide open, since there is no effective way to control $\|u(t)\|_{L^1_x}$.
We remark on two aspects of the noises that can affect qualitative long-time behaviors of solutions:
- The question seems to depend heavily on the roots and growth of the noise coefficient function $\sigma(u)$ – If the noise is degenerate (not cylindrical), say $\sigma(u) = 0$ for certain $u=r \in \mathbb{R}$, then $u \equiv r$ is a fixed point of the evolution. By the $L^1$–contraction, it is possible to prove certain long-time behavior results for the solutions for the unbounded noise coefficient function with one root. Both the growth of $\sigma$ and how many roots it possesses affect the long-time behaviors of solutions, as is evident also in the analysis of other equations such as the KPP equation (discussed below). In the case that $\sigma$ has no roots, there are no fixed points. It is possible that the nonlinear conservation laws driven by bounded noises with no roots have non-trivial invariant measures.
- If the noise is $\sigma(u) {{\rm d}}B = \sum_k g_k(u) {{\rm d}}W^k$, where $B = W^k e_k $ is a cylindrical Wiener process, the behaviors are expected to be very different from the case that the noise is simply $\sigma(u) {{\rm d}}W$.
Another natural direction to consider is the case of nonlinear systems of balance laws. For this case, such as for the isentropic Euler system, the kinetic formulation is not “pure" – it contains the instances of the solution mixed with the kinetic operator ([*cf.*]{} [@LPT1994b]). At present, it seems that the methods discussed above are not directly applicable to the systems.
The Navier-Stokes equations {#sec:NSE}
---------------------------
The two-dimensional incompressible Navier-Stokes equations (INSEs) driven by stochastic forcing has been a subject of intense interest. We focus on the analysis of asymptotic behaviors of solutions to keep ourselves from getting sidetracked.
The existence of invariant measures for the 2-D INSEs on a regular bounded domain with a general noise that is a Gaussian random field and white-in-time has been known at least since [@Fla1994]. The uniqueness and ergodicity for the 2-D INSEs have also been established; see [@FM1995] and the references therein for such results and further existence results of invariant measures under different conditions. These results have subsequently been improved, including for the noises that are localized in time and Gaussian in space, in [@BKL2001; @BKL2002; @MY2002; @Mat2002], and in some references cited in this paper.
We remark particularly that the corresponding existence questions for the 2-D INSEs with multiplicative noises have been established, for example in [@FG1995], via the Skorohod embedding and a Faedo-Galerkin procedure, which have shown the existence of martingale solutions and stationary martingale solutions, from which in turn the existence of an invariant measure can be derived.
The asymptotic behaviors of 2-D INSEs driven by white-in-time noises or Poisson distributed unbounded kick noises have been explored, and the existence and uniqueness of invariant measures for these systems are known. See also [@Mat2003; @KS2006; @KS2012] for the related references.
There are also more recent results on INSEs driven by space-time white noises in 2-D or 3-D; see [@DD2002; @DD2003; @ZZ2015] and the references therein. For example, it is known that the transition semigroup of the Kolmogorov equation associated to the 3-D stochastic INSEs driven by a cylindrical white noise has a unique (and hence ergodic) invariant measure.
The existence of invariant measures for the compressible Navier-Stokes equations, even in the 2-D case, is wide open.
See [@EMS2001; @MT2016], and discussions in [@Madja_2016 Chp. 3] as well as references contained there for further treatments on ergodicity results; also see [@MadjaWang].
The asymptotic strong Feller property
-------------------------------------
Using the 2-D INSEs as a springboard, the notion of the [*asymptotic strong Feller*]{} property has been introduced in Hairer-Mattingley [@HM2006] as a weaker and more natural replacement of the sufficient “strong Feller" property (Definition \[defin:strongfeller\]) in dissipative infinite-dimensional systems, the possession of which guarantees the uniqueness of an invariant measure. In the finite-dimensional case of SDEs, there is a related notion of [eventual]{} strong Feller property, for which sufficient conditions are given in [@Bie2011].
The definition of asymptotic strong Feller property depends on a preliminary definition:
A pseudo-metric is a function $d : {\mathfrak{X}}^2 \to \mathbb{R}^+_0$ for which $d(x,x) = 0$ and the triangle inequality is satisfied, and $d_1 \geq d_2$ if the inequality holds for all arguments $(x,y) \in {\mathfrak{X}}^2$. Let $\{d_k\}_{k = 0}^\infty$ be an increasing sequence of pseudo-metrics on a Polish space ${\mathfrak{X}}$. Then $\{d_k\}_{k = 0}^\infty$ is a [*totally separating system*]{} of pseudo-metrics if $$\lim_{k \to 0} d_k = 1$$ pointwise everywhere off the diagonal on ${\mathfrak{X}}^2$.
Then the asymptotic strong Feller property is defined as follows:
A Markov transition semigroup $\mathscr{P}_t$ on a Polish space $X$ is [*asymptotically strong Feller*]{} at $x$ if there exist both a totally separating system of pseudo-metrics $\{d_k\}_{k = 0}^\infty$ and an increasing sequence of times $t_k$ such that $$\inf_{U \in {\mathrm{nb}}(x)} \limsup_{k \to \infty} \sup_{y \in U} \|P_{t_k}(x,\cdot) - P_{t_k}(y,\cdot)\|_{d_k} = 0,$$ where ${\mathrm{nb}}(x)$ is the collection of open sets containing $x$, $P$ is the transition probabilities associated to $\mathscr{P}$, and $\|P_1 - P_2\|_{d_k}$ is the norm given by $$\|P_1 - P_2\|_{d_k} = \inf \int_{\mathfrak{X}^2} d_k(w,z) \Pi({{\rm d}}w, {{\rm d}}z),$$ the infimum being taken over all positive measures on $\mathfrak{X}^2$ with marginals $P_1$ and $P_2$.
The idea behind the asymptotic strong Feller condition is that ergodicity is preserved even if the stochastic forcing is restricted to a few unstable modes, and dissipated in the others. Using this idea, the ergodicity of the 2-D stochastic INSE with degenerate noise has been established (see [@HM2006]). Some results of ergodicity for the 3-D INSEs driven by mildly degenerate noise relying on the strong asymptotic Feller property have also established (see [@RX2011; @RZ2009] and the references cited therein).
The KPP equation and multiplicative noises
------------------------------------------
The Kolmogorov-Petrovsky-Piskunov equation (KPP) is given by $$\begin{aligned}
&{\partial}_t u = \nabla \cdot ({\mathbf{A}}(x,t) \nabla u) + h(u) + g(u) {\partial}_t W,\\
&u|_{t=0}=\varphi.\end{aligned}$$ Attention is often restricted to the case in which $g$ and $h$ both vanish at the two points $a, b\in \mathbb{R}$, and $g, h > 0 $ on $(a,b)$. In this way, the asymptotic size is controlled in $L^1$.
It has been shown in Chueshov-Villermot [@CV1996a; @CV1996b; @CV1998a; @CV1998b; @CV1998c; @CV2000] that, for the semilinear equation with $h(u) = s g(u)$, evolution on a bounded, open domain with zero Neumann boundary condition is bounded in space. Moreover, the notion of stability in probability has also been introduced in [@CV1998b]:
A function $u_f$ is [*stable in probability*]{} if, for every ${\epsilon}> 0$, the following relation holds: $$\lim_{\|\varphi - u_f\|_{L^\infty} \to 0}
\mathbb{P}\big\{ \omega \in \Omega : \sup_{t \in \mathbb{R}^+\setminus \{0\}}\|u_\varphi(s,\cdot,t,\omega) - u_f\|_{L^\infty} > {\epsilon}\big\} = 0.$$ Otherwise, $u_f$ is [*unstable in probability*]{}.
A function $u_f$ is [*globally asymptotically stable in probability*]{} if it is stable in probability and $$\mathbb{P}\big\{ \omega \in \Omega : \lim_{t \to \infty}\|u_\varphi(s,\cdot,t,\omega) - u_f\|_{L^\infty} = 0\big\} = 1.$$
By considering the moments of the spatial average, it has been shown that the constant functions $u_1 = a$ and $u_2=b$ are fixed points whose stability in probability depends on the values of $s$. The results of [@CV2000] have been refined, say in [@BS2005], and the properties of the global attractor, including the computation of exact Lyapunov exponents in a decay scenario have been derived.
As we have remarked, the main reason that multiplicative noises complicate the analysis of stochastic PDEs is that one fails to have much control over the spatial average, except when additional restrictions on the noise and initial conditions are specified. When the noise has a root, that constant is immediately a fixed point. This cannot be avoided even when working over the non-compact domain $\mathbb{R}$ because the $L^p$ boundedness often relies on the space that is compact, and is a difficulty we have to overcome in order to gain a deeper understanding of the asymptotic behaviors of solutions.
Large deviation principles
--------------------------
Beyond the existence and uniqueness of invariant measures, large deviation principles touch on their specific properties. Whilst it goes some way outside the scope of this survey even to introduce the theory of large deviations, which attempts to characterize the limiting behavior of a family of probability measures (in our case, invariant measures) depending on some parameter by using a [*rate function*]{}, we should be remiss to neglect mentioning it altogether; two vintage references to the subject are [@DS1989; @FW1998]. More modern treatments can be found in [@DE1997; @DZ1998; @Hol2008; @Var2016] and the references cited therein. Of particular interest has been the “zero-noise” limit of stochastic equations in which one looks at the stochastic equations with a small parameter ${\varepsilon}$ multiplied to the noise. Questions of large deviation type also arise in stochastic homogenization theory. Each of these subjects can justify an independent survey. Pertaining specifically to stochastic conservation laws, the literature is, however, more sparse. Going some way outside the classical Freidlin-Wentzell theory, some results have been announced pertaining to large deviation estimates for stochastic conservation laws. Specifically, in [@Mar2010], large deviation principles have been investigated and derived in the limit of jointly vanishing noise and viscosity by using delicate scaling arguments. Notably, in [@BBC2017], the bounds for the rate function have also been derived in the vanishing viscosity limit only, so that the noise is allowed in the limit, and in the multidimensional setting. Finally, we mention the more recent work [@DWZZ2018] and the references cited therein, large deviation principles have been derived for the first-order scalar conservation laws with small [*multiplicative*]{} noise on ${\mathbb{T}}^d$ in the zero-noise limit by using the Freidlin-Wentzell theory. Much still remains to be explored in this direction.
**Acknowledgements**. $\,$ The research of Gui-Qiang G. Chen was supported in part by the UK Engineering and Physical Sciences Research Council Award EP/E035027/1 and EP/L015811/1, and the Royal Society–Wolfson Research Merit Award (UK). The research of Peter Pang was supported in part by the UK Engineering and Physical Sciences Research Council Award EP/E035027/1 and EP/L015811/1, and an Oxford Croucher Scholarship.
[99]{}
L. Arnold. . Springer-Verlag: Berlin Heidelberg, 1998.
J. Barré, C. Bernardin, and R. Chertrite. Density large deviations for multidimensional stochastic hyperbolic conservation laws. **170(3)** (2017), 466 – 491.
C. Bauzet, G. Vallet, and P. Wittbold. The [C]{}auchy problem for conservation laws with a multiplicative stochastic perturbation. **9(4)** (2012), 661–709.
B. Bergé and B. Saussereau. On the long-time behavior of a class of parabolic [SPDE]{}s: Monotonicity methods and exchange of stability. **9** (2005), 254–276.
J. Birkens. Sufficient conditions for the eventual strong [F]{}eller property for degenerate stochastic evolutions. **379(2)** (2011), 469–481.
F. Bouchut and L. Desvillettes. Averaging lemmas without time [F]{}ourier transform and application to discretized kinetic equations. , **129** (1999), 19–36.
J. Bricmont, A. Kupiainen, and R. Lefevere. Ergodicity of the [2D]{} [N]{}avier-[S]{}tokes equations with random forcing. **224(1)** (2001), 65–81.
J. Bricmont, A.Kupiainen, and R. Lefevere. Exponential mixing of the [2D]{} [N]{}avier-[S]{}tokes dynamics. **224(1)** (2002), 87–132.
G.-Q. Chen, Q. Ding, and K. Karlsen. On nonlinear stochastic balance laws. **204(3)** (2012), 707–743.
G.-Q. Chen and H. Frid. Large time behavior of entropy solutions in ${L}^\infty$ for multidimensional conservation laws. In: [*Advances in Nonlinear PDEs and Related Areas*]{}, 28–44, World Scientific, 1998.
G.-Q. Chen and Y.-G. Lu. A study of approaches to applying the theory of compensated compactness. *Chinese Sci. Bull.* **34** (1989), 15–19.
G.-Q. Chen and P. Pang. On nonlinear anisotropic degenerate parabolic-hyperbolic equations with stochastic forcing. Preprint arXiv:1903.02693, March 2019.
G.-Q. Chen and B. Perthame. Well-posedness for non-isotropic degenerate parabolic-hyperbolic equations. , **20(4)** (2003), 645–668.
G.-Q. Chen and B. Perthame. Large-time behavior of periodic entropy solutions to anisotropic degenerate parabolic-hyperbolic equations. , **137(9)** (2009), 3003–3011.
I. Chueshov and P. Vuillermot. On the large-time dynamics of a class of parabolic equations subjected homogeneous white noise: Stratonovitch’s case. , **323(1)** (1996), 29–33.
I. Chueshov and P. Vuillermot. On the large-time dynamics of a class of random parabolic equations. , **322(1)** (1996), 1181–1186.
I. Chueshov and P. Vuillermot. Long-time behavior of solutions to a class of quasilinear parabolic equations with random coefficients. , **15(2)** (1998), 191–232.
I. Chueshov and P. Vuillermot. Long-time behavior of solutions to a class of stochastic parabolic equations with homogeneous white noise: Straonovitch’s case. , **112(2)** (1998), 149–202.
I. Chueshov and P. Vuillermot. On the large-time dynamics of a class of parabolic equations subjected to homogeneous white noise: Itô’s case. , **326(1)** (1998), 1299–1304.
I. Chueshov and P. Vuillermot. Long-time behavior of solutions to a class of stochastic parabolic equations with homogeneous white noise: Itô’s case. **18(4)** (2000), 581–615.
M. Coti-Zelati, N. Glatt-Holtz, and K. Trivisa. Invariant measures for the stochastic one-dimensional compressible [N]{}avier-[S]{}tokes equations. , 2018.
H. Crauel. Markov measures for random dynamical systems. , **37(3)** (1991), 153–173.
H. Crauel. . Taylor and Francis: London, 2002.
H. Crauel, A. Debussche, and F. Flandoli. Random attractors. **9(2)** (1997), 307–341.
H. Crauel and F. Flandoli. Attractors for random dynamical systems. , **100(3)** (1994), 365–393.
G. Da Prato and A. Debussche. Two-dimensional Navier-Stokes equations driven by a space-time white noise. **196(1)** (2002), 180–210.
G. Da Prato and A. Debussche. Ergodicity for the 3[D]{} stochastic Navier-Stokes equations. **82(8)** (2003), 877–947.
G. Da Prato, A. Debussche, and R. Temam. Stochastic [B]{}urger’s equation. **1** (1994), 389–402.
G. Da Prato, F. Flandoli, E. Priola, and M. Röckner. Strong uniqueness for stochastic evolution equations in [H]{}ilbert spaces perturbed by a bounded measurable drift. **41(5)** (2013), 3306–3344.
G. Da Prato and J. Zabczyk. . Cambridge University Press: Cambridge, 2008.
A. Debussche. On the finite dimensionality of random attractors. **15(4)** (1997), 473–492.
A. Debussche. Hausdorff dimension of a random invariant set. **77(10)** (1998), 967–988.
A. Debussche, M. Hofmanová, and J. Vovelle. Degenerate parabolic stochastic partial differential equations: Quasilinear case. **44(3)** (2016), 1916–1955.
A. Debussche and J. Vovelle. Scalar conservation laws with stochastic forcing. **259(4)** (2010), 1014–1042.
A. Debussche and J. Vovelle. Invariant measure of scalar first-order conservation laws with stochastic forcing. , **163(3)** (2015), 575–611.
A. Dembo, and O. Zeitouni. New York, Springer, 2nd Ed., 1997.
F. Delarue, F. Flandoli, and D. Vincenzi. Noise prevents collapse of [V]{}lasov-[P]{}oisson point charges. **67(10)** (2014), 1700–1736.
J.-D. Deuschel, and D. Stroock. Large Deviations. Academic Press: London 1989.
W. Doeblin. léments d’une théorie générale des chaînes simples constantes de [M]{}arkoff. , **57** (1940), 61–111.
W. Doeblin and R. Fortet. Sur le cha[î]{}nes [à]{} liaisons compl[è]{}te. , **65** (1937), 132–148.
Z. Dong, J.-L. Wu, R.-R. Zhang, and T.-S. Zhang. Large deviation principles for first-order scalar conservation laws with stochastic forcing. , 2018.
J. Doob. Asymptotic property of [M]{}arkoff transition probability. , **64(3)** (1948), 393–421.
P. Dupuis, R. S. Ellis. New York,Wiley, 1997.
W. E, K. Khanin, A Mazel, and Ya. Sinai. Probability distribution functions for the random forced [B]{}urgers equation. **78(10)** (1997), 1904–1907.
W. E, K. Khanin, A Mazel, and Ya. Sinai. Invariant measures for [B]{}urgers equation with stochastic forcing. **151(3)** (2000), 877–960.
W. E, J. Mattingly, and Ya. Sinai. Gibbsian dynamics and ergodicity for the stochastically forced Navier-Stokes equation. **224(1)** (2001), 83 – 106.
E. Fedrizzi, F. Flandoli, E. Priola, and J. Vovelle. Regularity of stochastic kinetic equations. , **22(48)** (2017), 1–42.
B. Fehrman and B. Gess, Well-posedness of nonlinear diffusion equations with nonlinear, conservative noise. Preprint arXiv:1712.05775, 2019
J. Feng and D. Nualart. Stochastic scalar conservation laws. **255(2)** (2008), 313–373.
F. Flandoli. Stochastic flows and [L]{}yapunov exponents for abstract stochastic [PDE]{}s of parabolic type. In: [*Lyapunov Exponents Proceedings [(]{}L. Arnold, H. Crauel, J.-P. Eckamann eds.[)]{}*]{}, LNM [**1486**]{} (1991), 196–205.
F. Flandoli. Dissipativity and invariant measures for stochastic [N]{}avier-[S]{}tokes equations. **1(4)** (1994), 403–423.
F. Flandoli and D. Gatarek. Martingale and stationary solutions for stochastic [N]{}avier-[S]{}tokes equations. , **102(3)** (1995), 367–391.
F. Flandoli, M. Gubinelli, and E. Priola. Well-posedness of the transport equation by stochastic perturbation. **180(1)** (2010), 1–53.
F. Flandoli and B. Maslowski. Ergodicity of the 2-[D]{} [N]{}avier-[S]{}tokes equations under random perturbations. **172(1)** (1995), 119–141.
Franco Flandoli. . Gordon and Breach Science Publishers: Singapore, 1995.
J. F[ö]{}ldes, S. Friedlander, N. Glatt-Holtz, and G. Richards. Asymptotic analysis for randomly forced [MHD]{}. , 2016.
J. F[ö]{}ldes, N. Glatt-Holtz, G. Richards, and J.P. Whitehead. Ergodicity in randomly forced [R]{}ayleigh-[B]{}énard convection. , 2015.
M.I. Freidlin, and A.D. Wentzell. . Springer: New York, 2nd Ed., 1998.
B. Gess and P. Souganidis. Long-time behavior, invariant measures and regularizing effects for stochastic scalar conservation laws. , 2014.
B. Gess and P. Souganidis. Stochastic non-isotropic degenerate parabolic-hyperbolic equations. , 2016.
Y. Giga and T. Miyakawa. A kinetic construction of global solutions of first order quasilinear equations. **50(2)** (1989), 505–515.
I. Gy[ö]{}ngy and D. Nualart. On the stochastic [B]{}urgers equation in the real line. **27(2)** (1999), 782–802.
M. Hairer and J. Mattingly. Ergodicity of the 2[D]{} [N]{}avier-[S]{}tokes equations with degenerate stochastic forcing. **164** (2006), 993–1032.
M. Hairer and J. Mattingly. Spectral gaps in [W]{}asserstein distances and the [2D]{} stochastic [N]{}avier-[S]{}tokes equations. **36(6)** (2008), 2050–2091.
M. Hairer and J. Mattingly. A theory of hypoellipticity and unique ergodicity for semilinear stochastic [PDE]{}s. **16(23)** (2011), 658–738.
M. Hairer and J. Voss. Approximations to the stochastic [B]{}urgers equation. **21 (6)** (2011), 897 – 920.
M. Hofmanová. Degenerate parabolic stochastic partial differential equations. **123(12)** (2013), 4294–4336.
F. den Hollander. Fields Institute Monographs, A.M.S.: Providence, RJ, 2000.
Y. Le Jan. quilibre statistique pour les produits de difféomorphismes aléatoires indépendants. **23(1)** (1987), 111–120.
K. H. Karlsen and E. Storrøsten. On stochastic conservation laws and [M]{}alliavin calculus. , 2015.
R. Khas’minskii. Ergodic properties of recurrent diffusion processes and stabilization of the solutions to the [C]{}auchy problem for parabolic equations. **5(2)** (1960), 179–196.
N. Krylov and M. Röckner. Strong solutions of stochastic equations with singular time dependent drift , **131(2)** (2005), 154 – 196.
S. Kuksin and A. Shirikyan. . European Mathematical Society: Zürich, 2006.
S. Kuksin and A. Shirikyan. . Cambridge University Press: Cambridge, 2002.
F. Ledrappier. Positivity of the exponent for stationary sequences of matrices. In: [*Lyapunov Exponents Proceedings [(]{}L. Arnold and V. Wihstutz, eds.[)]{}*]{} LNM [**1186**]{} (1986), 56–73.
J. Leòn, D. Nualart, and R. Pettersson The stochastic Burgers equation: Finite moments and smoothness of the density. , **3** (2000), 363 – 385.
T. Lindvall. . John Wiley [&]{} Sons: New York, 1992.
P.-L. Lions, B. Perthame, and P. Souganidis. Scalar conservation laws with rough (stochastic) fluxes. **1** (2013), 664–686.
P.-L. Lions, B. Perthame, and P. Souganidis. Scalar conservation laws with rough (stochastic) fluxes: the spatially dependent case. **2(4)** (2014), 517–538.
P.-L. Lions, B. Perthame, and E. Tadmor. A kinetic formulation of multidimensional scalar conservation laws and related equations. **7(1)** (1994), 169–191.
P.-L. Lions, B. Perthame, and E. Tadmor. Kinetic formulation of the isentropic gas dynamics and $p$-systems. **163** (1994), 415–431.
A. Majda. . Springer: Cham, 2016. A. Majda, and X.-T. Tong Ergodicity of truncated stochastic Navier Stokes with deterministic forcing and dispersion. **26 (5)** (2016), 1483–1506.
A. Majda and X.-M. Wang. . Cambridge University Press: Cambridge, 2006.
M. Mariani. Large deviations principles for stochastic scalar conservation laws. **147** (2010), 607 – 648.
B. Maslowski and J. Seidler. Invariant measures for nonlinear [SPDE]{}’s: uniqueness and stability. , **34(1)** (1998), 153–172.
N. Masmoudi and L.-S. Young. Ergodic theory of infinite dimensional systems with applications to dissipative parabolic [PDE]{}s. **227(3)** (2002), 461–481.
J. C. Mattingly. Exponential convergence for the stochastically forced [N]{}avier-[S]{}tokes equations and other partially dissipative dynamics. **230(3)** (2002), 421–462.
J. C. Mattingly. Recent progress for the stochastic [N]{}avier-[S]{}tokes equations. (2003), Exp. No. 11, 52 p., doi: 10.5802/jedp.625 $\quad$ http://www.numdam.org/article/JEDP\_2003\_\_\_\_A11\_0.pdf
B. Perthame. . Oxford University Press: Oxford, 2002.
S. Peszat and J. Zabczyk. Strong Feller property and irreducibility for diffusions on Hilbert spaces. **23(1)** (1995), 157–172.
S. Peszat and J. Zabczyk. . Cambridge University Press: Cambridge, 2007.
M. Röckner and X.-C. Zhang. Stochastic tamed 3D Navier-Stokes equations: existence, uniqueness and ergodicity. , **145** (2009), 211 – 267.
M. Romito and L.-H. Xu. Ergodicity of the 3[D]{} stochastic [N]{}avier-Stokes equations driven by mildly degenerate noise. **121** (2011), 673 – 700.
B. Schmalfuss. The random attractor of the stochastic [L]{}orenz system. **48(6)** (1997), 951 – 975.
Ya. Sinai. Two results concerning asymptotic behavior of solutions of the [B]{}urgers equation with force. **64(1)** (1991), 1–12.
E. Tadmor and T. Tao. Velocity averaging, kinetic formulations and regularizing effects in quasi-linear pdes. **60(10)** (2007), 1488–1521.
R. Temam. . Springer: New York, 1997.
S.R.S. Varadhan. Courant Lecture Notes, Amer. Math. Soc., 2016.
A. Veretennikov. On strong solutions and explicit formulas for solutions of stochastic integral equations. , **39(3)** (1981).
C. Villani. , , 2009.
R. Zhu and X. Zhu. Three-dimensional [N]{}avier-[S]{}tokes equations driven by space-time white noise. **259(9)** (2015), 4443–4508.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We introduce a formalism for computing bond percolation properties of a class of correlated and clustered random graphs. This class of graphs is a generalization of the Configuration Model where nodes of different types are connected via different types of *hyperedges*, edges that can link more than 2 nodes. We argue that the multitype approach coupled with the use of clustered hyperedges can reproduce a wide spectrum of complex patterns, and thus enhances our capability to model real complex networks. As an illustration of this claim, we use our formalism to highlight unusual behaviors of the size and composition of the components (small and giant) in a synthetic, albeit realistic, social network.'
address: 'Département de physique, de génie physique, et d’optique, Université Laval, Québec (Qc), Canada G1V 0A6'
author:
- 'A Allard, L Hébert-Dufresne, P-A Noël, V Marceau and'
title: Bond percolation on a class of correlated and clustered random graphs
---
Introduction
============
Bond percolation is the study of the size distribution of components in graphs whose edges exist with a given probability. For its theoretical appeal and its varied applications in many contexts, mathematical modelling of bond percolation on random graphs has recently received substantial attention (see [@Dorogovtsev08_RevModPhys; @newman10_Networks], and references therein). Within the Configuration Model (CM) paradigm [@Molloy95_RandomStructAlg; @Molloy98_CombinatoricsProbabComput], many exact results can be obtained using probability generating functions (PGF) [@Newman01_PhysRevE]. This analytic tractability however comes at the price of simplifying assumptions on the structure of the graphs.
We introduce a generalization of the CM that encompasses many of the previous improvements published to this day [@Newman01_PhysRevE; @Allard09_PhysRevE; @Ghoshal09_PhysRevE; @Gleeson09_PhysRevE; @Karrer10_PhysRevE; @Leicht09_arXiv; @Miller09_PhysRevE; @Meyers06_JTheorBiol; @Newman02_PhysRevE; @Newman02_PhysRevLett; @Newman03a_PhysRevE; @Newman03b_PhysRevE; @Newman09_PhysRevLett; @Serrano06_PhysRevLett; @Shi07_PhysicaA; @Zlatic12_EPL], and brings this class of models closer to the behavior of real complex networks. By combining the multitype approach of [@Allard09_PhysRevE], the analytical method of [@Allard12_EPL] and the one-mode projection of [@Newman03b_PhysRevE], we argue that our model is able to reproduce a wide range of complex patterns found in real networks.
On the one hand, the multitype approach allows to explicitly prescribe how nodes are connected to one another in a very detailed fashion. By assigning types to nodes – in other words by knowing who is who, and therefore who is connected to whom – several mixing patterns (e.g., assortativity, degree correlation, node segregation), as well as heterogeneous bond occupation probabilities (e.g., partial and/or uneven directionnality of edges) can be reproduced. On the other hand, the use of the one-mode projection, coupled with the multitype approach, allows the inclusion of clustering through a myriad of nontrivial motifs, i.e. recurrent, significant patterns of interconnections [@Milo02_Science].
This paper is organized as follows. In , we introduce the generalization of the CM that explicitly includes various correlations and clustering. We then develop the analytical framework to obtain the bond percolation properties of this graph ensemble in . In , we validate our formalism — and also illustrate the versatility of our approach — by comparing its predictions with simulation results on a synthetic, but realistic social network. In , we show that many percolation models published in the litterature are special cases of our model. We also highlight how our approach can be useful for studying interdependent or coupled networks [@Leicht09_arXiv; @Buldyrev10_Nature; @Huang11_PhysRevE; @Marceau11_PhysRevE], and for studying the *weak and strong clustering regimes* [@Serrano06a_PhysRevE; @Serrano06b_PhysRevE]. We conclude in and present in 2 Appendices some relevant aspects of the analysis and simulations. \[sec:general\_method\] details how a recent method, to analytically compute the distribution of the composition of components for any small arbitrary graphs [@Allard12_EPL], can be used in our formalism. \[sec:num\_sim\_details\] gives further details on the numerical simulations.
Correlated and clustered graph ensemble {#sec:themodel}
=======================================
We introduce a general class of correlated and clustered random graphs. To preserve the analytical tractability of the CM, we first consider *unclustered multitype bipartite graphs* that are locally tree-like in the large system size limit. Clustering is then incorporated through a projection, analogous to the one-mode projection of [@Newman03b_PhysRevE].
Unclustered multitype bipartite graph ensemble
----------------------------------------------
We call unclustered multitype bipartite graphs a multitype generalization of the bipartite CM [@Newman01_PhysRevE]. These graphs are composed of $M$ types of “regular nodes” and $\Lambda$ types of “group nodes” (hereafter referred to as nodes and groups, respectively). Edges only exist between regular nodes and group nodes. In these graphs, a fraction $w_i$ of nodes are of type $i$, and any given type-$i$ node is connected to $k_\mu$ type-$\mu$ groups (for each $\mu = 1, \ldots, \Lambda$) with a probability $P_i(k_1,\ldots,k_\Lambda) \equiv P_i(\bm{k})$. Likewise, a randomly chosen type-$\nu$ group is connected to $n_j$ type-$j$ nodes (for each $j = 1, \ldots, M$) with a probability $R_\nu(n_1,\ldots,n_M) \equiv R_\nu(\bm{n})$. In other words, $R_\nu(\bm{n})$ is the distribution of the composition of type-$\nu$ groups. gives an example of such graphs. To lighten the notation, it should now be understood that any free latin (resp. greek) index can take any values in $\{1, \ldots, M\}$ (resp. $\{1, \ldots, \Lambda\}$), except if otherwise mentionned.
In the large system size limit, $w_i$, $P_i(\bm{k})$ and $R_\nu(\bm{n})$ fully define a graph ensemble which is totally random in all other respects (stubs are matched randomly). All finite components therefore have a tree-like structure in this limit (the probability of a closed path goes as the inverse of the size of the graph). These quantities are however not independent. To guarantee the consistency of the graph ensemble, they must, for all applicable combinations of $i$, $j$ and $\nu$, satisfy $$\label{eq:consistency_condition}
\frac{w_i \langle k_\nu \rangle_{P_i}}{\langle n_i \rangle_{R_\nu}} = \frac{w_j \langle k_\nu \rangle_{P_j}}{\langle n_j \rangle_{R_\nu}} \ ,$$ where $\langle a \rangle_B$ denotes the mean value of $a$ with respect to the distribution $B$. Simply stated, asks $w_i$, $P_i(\bm{k})$ and $R_\nu(\bm{n})$ to be chosen such that each node type “forces” the same number of type-$\nu$ groups in the unclustered multitype bipartite graph ensemble.
Clustered multitype graph ensemble
----------------------------------
A clustered graph ensemble is obtained from the unclustered multitype bipartite graph ensemble by means of a projection similar to the one-mode projection of [@Newman03b_PhysRevE]. This projection is achieved by replacing the group nodes in the unclustered multitype bipartite graphs by motifs involving the nodes that were linked to a same group. The nature, either quenched (fixed) or annealed (random), and the structure of these motifs is prescribed by the corresponding group type. The resulting graphs then consist of different motifs embedded in a tree-like backbone.
illustrates a resulting clustered multitype graph where every group is replaced by a multitype quenched motif. For instance, type-*green* groups (B, D and E) are replaced by a *triangle* composed of one node of each of the $M=3$ possible types, and whose edges are undirected except for the one between the type-*blue* node and type-*red* one that is directed. Single edges — directed (C and G) or undirected (K, L, and M) — can also correspond to motifs simply composed of two nodes. The type of each of the two nodes and the direction of the edge is prescribed by the type of the group. An example of the use of annealed motifs, where edges exist with a given probablity rather than being *a priori* fixed, is given in .
Bond percolation can exactly be solved for the CM and its numerous variants because graphs in these ensembles have an underlying tree-like structure. Thus to take advantage of the tree-like backbone of the clustered graph ensemble, the outcome of bond percolation must be solved beforehand for each motif appearing in the graph ensemble. This solution is encoded in $Q_{i\nu}(\bm{l}|\bm{n})$ giving the probability that $\bm{l}$ nodes (i.e., $l_j$ type-$j$ nodes, for all $j$) will eventually be reached from an initial type-$i$ node by following existing edges in a type-$\nu$ motif of size $\bm{n}$. In other words, this distribution prescribes the number of nodes (and their type) from which a given motif can be left while navigating on a graph of the clustered ensemble. It therefore “restores” the tree-like structure of the unclustered multitype bipartite graphs while retaining the effect of the clustered motifs. It is this correspondence that allows the derivation of a PGF-formalism which exactly solves the bond percolation properties of the clustered multitype graph ensemble.
In principle, a wide variety of motifs can be incorporated in our model; this variety is only limited by our ability to solve the bond percolation outcome on these motifs. Motifs can be chosen to reproduce recurring patterns of interactions found in real complex networks [@Milo02_Science], to account for local clustering in realistic synthetic networks (see ), or for theoretical investigations (see ). Following the results of [@Allard12_EPL], we give in \[sec:general\_method\] a general method to calculate $Q_{i\nu}(\bm{l}|\bm{n})$ for most, if not all, imaginable motifs of reasonable size (the limits of the method are discussed in [@Allard12_EPL]). This method can handle quenched (fixed structure) or annealed (random structure) motifs in which edges may be directed or not. Also, nodes may belong to types which permits to model (dis-)assortative mixing and heterogeneous bond percolation [@Allard09_PhysRevE; @Allard12_EPL].
\[sec:math\_form\]Bond Percolation Properties
=============================================
We now introduce a PGF-based mathematical formalism to calculate the percolation properties of the correlated and clustered graph ensemble defined in the last section. Since PGF-based percolation formalisms have become fairly standard, the unfamiliar reader should consult recent reviews on complex network modeling (see for example [@Newman03_SIAMRev] and references therein) for further details.
We first define $\theta_{i\nu}(\bm{x})$ as the function generating the distributions $\{Q_{i\nu}(\bm{l}|\bm{n})\}$ of the outcome of bond percolation, from an initial type-$i$ node, on the motifs corresponding to type-$\nu$ groups. As type-$\nu$ groups may not all have the same composition (e.g., household size distribution in social networks), $\theta_{i\nu}(\bm{x})$ is calculated according to $$\label{eq:theta_general}
\theta_{i\nu}(\bm{x}) = \sum_{\bm{n}} \frac{n_i R_{\nu}(\bm{n})}{\langle n_i \rangle_{R_\nu}} \left[ \sum_{\bm{l}=\bm{\delta_i}}^{\bm{n}} Q_{i\nu}(\bm{l}|\bm{n}) \prod_{j} x_{\nu j}^{l_j-\delta_{ij}} \right] \ ,$$ with $\bm{\delta_i} \equiv (\delta_{i1},\ldots,\delta_{iM})$ where $\delta_{il}$ is Kronecker’s delta. In , we average over $\frac{n_i R_{\nu}(\bm{n})}{\langle n_i \rangle_{R_\nu}}$ instead of over $R_{\nu}(\bm{n})$ to account for the fact that groups containing $n_i$ type-$i$ nodes are $n_i$ times more likely to be reached from any type-$i$ node than groups containing only one type-$i$ node. Although is not explicitly labelled in this respect, more than one motifs may be associated with a given group type. In such case, the distribution $R_\nu(\bm{n})$ gives the probability of occurence of each motif, and the left-hand sum in is taken over each possible motif for which a distinct distribution $Q_{i\nu}(\bm{l}|\bm{n})$ is obtained with the method outlined in \[sec:general\_method\].
The function $\theta_{i\nu}(\bm{x})$ is the mathematical implementation of the correspondence between the unclustered and clustered graph ensembles discussed at the end of the last section. By generating the distribution of $\nu \rightarrow j$ edges (i.e., stemming from a type-$\nu$ group and leading to a type-$j$ node) reached by a type-$i$ node, one can then navigate on a unclustered multitype bipartite graph as if one were on a clustered multitype graph.
We define $g_i(\bm{x})$ as the PGF generating the distribution of the number of $\nu \rightarrow j$ edges emerging from a type-$i$ node (i.e., emerging from the groups a type-$i$ node is connected to) $$g_i(\bm{x}) = \sum_{\bm{k}} P_i(\bm{k}) \prod_\nu \big[\theta_{i\nu}(\bm{x})\big]^{k_\nu} \ .$$ It is also convenient to define a PGF that generates the distribution of the number of $\nu \rightarrow j$ edges emerging from a type-$i$ node which has itself been reached via a $\mu \rightarrow i$ edge $$\label{eq:fmui}
f_{\mu i}(\bm{x}) = \sum_{\bm{k}} \frac{k_\mu P_i(\bm{k})}{\langle k_\mu \rangle_{P_i}} \prod_{\nu} \big[\theta_{i\nu}(\bm{x})\big]^{k_\nu - \delta_{\mu\nu}} \ .$$ The averaging term used in is motivated by the same argument as the one in . With these two PGFs, we may now compute the percolation properties of clustered multitype graph ensemble.
Phase transition
----------------
As a class of random graphs, clustered multitype graphs undergo a phase transition corresponding to the emergence of an extensive connected “giant” component. To locate the phase transition, let us define $\xi_{\nu j}(s)$ as the average number of $\nu \rightarrow j$ edges at a distance $s$ from any node in any graphs of the ensemble. Due to the tree-like structure of the underlying unclustered multitype bipartite graph, each $\xi_{\nu j}(s)$ is a linear combination of all $\xi_{\nu j}(s\!-\!1)$ at distance $s\!-\!1$: $$\label{eq:phase_transition1}
\xi_{\nu j}(s) = \sum_{\mu i} \left[ \frac{\partial f_{\mu i}(\bm{x})}{\partial x_{\nu j}}{} \right]_{\bm{x}=\bm{1}} \xi_{\mu i}(s\!-\!1)$$ where $$\frac{\partial f_{\mu i}(\bm{1})}{\partial x_{\nu j}}{} = \sum_{\bm{k}} \frac{(k_\nu\!-\!\delta_{\mu\nu})k_\mu P_i(\bm{k})}{\langle k_\mu \rangle_{P_i}} \frac{\partial \theta_{i\nu}(\bm{1})}{\partial x_{\nu j}}$$ is the average number of $\nu \rightarrow j$ edges emerging from a type-$i$ node that has been reached via a $\mu \rightarrow i$ edge. In vector notation, becomes $$\label{eq:phase_transition2}
\bm{\xi}(s) = \mathbf{B}\, \bm{\xi}(s\!-\!1) \ .$$ We see from that, in general, every $\xi_{\nu j}(s)$ vanishes with increasing $s$ if all eigenvalues of the $(M\Lambda) \times (M\Lambda)$ matrix $\mathbf{B}$ are below 1. Thus the phase transition happens when the largest eigenvalue of $\mathbf{B}$ reaches unity [^1].
Giant Component
---------------
As there may be directed edges in the graphs (through the motifs), the giant component may have a “bow-tie” structure [@Newman01_PhysRevE; @Allard09_PhysRevE]. This implies that the probability $\mathcal{P}$ of reaching the giant component may not be equal to its relative size $\mathcal{S}$. Both quantities must therefore be computed separately.
Let us define $a_{\mu i}$ as the probability that a $\mu \rightarrow i$ edge does not lead to the giant component. Because of the tree-like structure of finite components in the unclustered multitype bipartite graph, we see that $a_{\mu i}$ must satisfy the self-consistency relation $$\label{eq:gc_selfconsistent}
a_{\mu i} = f_{\mu i}(\bm{a}) \ .$$ That is, every edge reached from an edge that is not leading to the giant component must not lead to the giant component either. The probability that any type-$i$ node does lead to the giant component is therefore given by $\mathcal{P}_i \equiv 1-g_i(\bm{a})$, and, averaging over the node type distribution $\{w_i\}$, the probability $\mathcal{P}$ that a randomly chosen node leads to the giant component is $$\label{eq:P}
\mathcal{P} = \sum_i w_i \mathcal{P}_i = 1 - \sum_{i} w_i g_i(\bm{a}) \ .$$
To obtain the size of the giant component, we must calculate the probability that a given node cannot be reached from any node in the giant component. This is equivalent to computing the probability that this node does not lead to the giant component when edges are followed in the reverse direction [@Newman01_PhysRevE; @Allard09_PhysRevE]. Edges in the underlying unclustered multitype bipartite graph being undirected, only $\theta_{\nu i}(\bm{x})$ needs to be modified. For instance, this can be achieved by using $p_{sr}$ instead of $p_{rs}$ in . We denote this new PGF $\bar{\theta}_{\nu i}(\bm{x})$ and we will add a bar ($\bar{\ }$) over every PGF using $\bar{\theta}_{\nu i}(\bm{x})$ instead of $\theta_{\nu i}(\bm{x})$.
Following a similar approach as for computing $\mathcal{P}$, we define $\bar{a}_{\mu i}$ as the probability that a type-$i$ node cannot be reached from the giant component via a $\mu \rightarrow i$ edge. That is, $\bar{a}_{\mu i}$ is the probability that a neighbour of a type-$i$ node in a type-$\mu$ group is not part of the giant component. Self-consistency then requires for $\bar{a}_{\mu i}$ to satisfy $$\bar{a}_{\mu i} = \bar{f}_{\mu i}(\bm{\bar{a}}) \ .$$ The probability that any given type-$i$ node is not part of the giant component is therefore $\bar{g}_i(\bm{\bar{a}})$. Considering that a fraction $w_i$ of the nodes are of type $i$, the fraction of the graph occupied by type-$i$ nodes in the giant component is $$\label{eq:Si}
\mathcal{S}_i = w_i \big[ 1 - \bar{g}_i(\bm{\bar{a}}) \big] \ ,$$ and the relative size of the giant component is $$\mathcal{S} = \sum_i \mathcal{S}_i = 1 - \sum_{i} w_i \bar{g}_i(\bm{\bar{a}}) \ .$$
Distribution of the composition of small components
---------------------------------------------------
To calculate the distribution of the number of nodes of each type expected in small components, we define the PGF $A_{\mu i}(\bm{x})$ that generates the distribution of the number of edges of each type (i.e., $\nu \rightarrow j$ for all $\nu$ and $j$) that are *ahead* of a $\mu \rightarrow i$ edge in small components. In the large system size limit, the small components have a tree-like structure and no finite-size effects are to be expected \[i.e., the joint degree distribution $P_i(\bm{k})$ is constant\]. We therefore expect $A_{\mu i}(\bm{x})$ to be invariant under translation on a small component; the distribution of the number of each edge type ahead, $A_{\mu i}(\bm{x})$, is independent of the position in a small component. This implies that $A_{\mu i}(\bm{x})$ must satisfy $$\label{eq:sc_translation_independence}
A_{\mu i}(\bm{x}) = x_{\mu i} f_{\mu i}\big(\bm{A}(\bm{x})\big)$$ where the extra $x_{\mu i}$ accounts for the $\mu \rightarrow i$ edge that has just been followed. This extra factor guarantees that a finite extent of the distribution generated by $A_{\mu i}(\bm{x})$ can be obtained in a finite number of iterations of starting with the initial conditions $A_{\mu i}(\bm{x})=1$. Replacing $x_{\nu i} = z_i$ for all $\nu$ in $A_{\mu i}(\bm{x})$ generates the distribution of the number of nodes of each type ahead a type-$i$ nodes reached from a type-$\mu$ group. Thus the composition of a small component reached from a type-$i$ node is generated by $z_i g_i\big(\bm{A}(\bm{z})\big)$; again the extra $z_i$ accounts for the initial type-$i$ node. Because any node is of type $i$ with probability $w_i$, the composition of a small component that is reached from a randomly chosen node is therefore generated by $$\label{eq:small_components_size_dist}
K(\bm{z}) = \sum_i \frac{w_i z_i g_i\big(\bm{A}(\bm{z})\big)}{1-\mathcal{P}} \ ,$$ where $1-\mathcal{P}$ ensures the normalization of $K(\bm{z})$. Note that $A_{\mu i}(\bm{1})$ is equal to the probability that a $\mu \rightarrow i$ edge leads to a finite (small) component, and is therefore equal to $a_{\mu i}$.
Solving – can however become tedious when dealing with large number of types of nodes and groups, or large groups. It is therefore worth noting that the first moments of the distribution generated by $K(\bm{z})$ can be calculated in a more direct manner. For instance, let us compute the average number $\langle s_i \rangle$ of type-$i$ nodes in small components. With $$\langle s_i \rangle = \left. \frac{\partial K(\bm{z})}{\partial z_i}{} \right|_{\bm{z}=\bm{1}}$$ inserted in , replacing $x_{\mu i}$ with $z_i$, we get $$\label{eq:sc_average_1}
\langle s_i \rangle = \frac{w_i (1\!-\!\mathcal{P}_i)}{1-\mathcal{P}}
+ \sum_{j \gamma r} \frac{w_j \langle k_\gamma\rangle_{P_j} a_{\gamma j}}{1-\mathcal{P}}
\frac{\partial \theta_{j\gamma}(\bm{a})}{\partial x_{\gamma r}}{}
\frac{\partial A_{\gamma r}(\bm{1})}{\partial z_i}{} \ ,$$ where we have used , and the fact that $g_i(\bm{a}) = 1-\mathcal{P}_i$. In this last result, $\frac{\partial \theta_{j\gamma}(\bm{a})}{\partial x_{\gamma r}}{}$ is the average number of type-$r$ nodes that are accessible from a type-$j$ node in a type-$\gamma$ group in small components. Also, $\frac{\partial A_{\gamma r}(\bm{1})}{\partial z_i}{}$ is the average number of type-$i$ nodes ahead of a $\gamma \rightarrow r$ edge in small components. From , we see that this last quantity is the solution of $$\label{eq:sc_average_2}
\frac{\partial A_{\gamma r}(\bm{1})}{\partial z_i}{}
= a_{\gamma r}\delta_{ir}
+ \sum_{\lambda s} \frac{\partial f_{\gamma r}(\bm{a})}{\partial \theta_{r\lambda}}{}
\frac{\partial \theta_{r\lambda}(\bm{a})}{\partial x_{\lambda s}}{}
\frac{\partial A_{\lambda s}(\bm{1})}{\partial z_i}{}$$ where $\frac{\partial f_{\gamma r}(\bm{a})}{\partial \theta_{r\lambda}}{}$ is the average number of type-$\lambda$ groups to which a type-$r$ node reached via a type-$\gamma$ group is connected in small components. Thus by solving – and then –, it is possible to obtain quite easily the average number of nodes of each type in the small components. Equations for higher moments can be obtained in a similar manner and are straightforward to derive.
Illustration and validation {#sec:validation}
===========================
To illustrate the versatility and the usefulness of our approach, we generated *urban networks* [@Meyers05_JTheorBiol] and used our formalism to predict the outcome of an outbreak of a hypothetical infectious disease. In these graphs, three ($M\!=\!3$) types of nodes – namely adults (type 1), heath-care workers (HCW, type 2) and children (type 3) – interact whithin groups representing households, workplaces, schools and hospitals. In addition, friendship bonds between children are modeled using a nontrivial motif , and directed edges from adults and children to HCW are added to account for the susceptibility of HCW to get infected by people seeking care in hospitals [@Bansal06_PLoSMed]. The disease spreads from infectious nodes to their neighbours with probability $T$ called the *transmissibility* [@Newman02_PhysRevE]. Further details of these graphs and of the associated numerical simulations are relegated to \[sec:num\_sim\_details\]. It should be appreciated that these graphs contain a wide range of properties found in real complex networks such as clustering of several orders (e.g., arbitrary motifs, heterogeneous Erdős-Rényi cliques), (dis)assortative mixing, degree-degree correlation and directed edges.
shows the typical bifurcation diagram of the giant-component-related quantities $\mathcal{P}$ and $\{\mathcal{S}_i\}$. Apart from the excellent agreement between the results of the numerical simulations and the predictions of our formalism, this figure illustrates how the multitype approach can highlight the behavioral differences between different populations — identified by their own node type — within a same graph ensemble. In this specific case, the HCW population has purposely been put in the situation where each HCW has more incoming edges than outgoing edges with adults and children. Also, the average degree inside the Erdős-Rényi cliques corresponding to hospitals (300 nodes connected to one another with probability 0.05) is greater than 1 for $T$ greater than $T^{\prime}\equiv[0.05\times299]^{-1} \simeq 0.067$. This implies that these cliques are increasingly likely to have percolated (i.e., to have a spanning cluster) for $T>T^{\prime}$. Qualitatively, once an outbreak reaches the HCW population, it is likely to stay mostly confined in it and to infect a large proportion of it. Only when $T$ becomes sufficiently large does the outbreak invade other part of the population (schools, workplaces and friendship circles). These insights are corroborated by . It also shows that although the HCW population accounts for only 5% of the total population, it drives the percolation process by pulling down its threshold to $T_c\simeq0.1$; the other node types only significantly join (i.e., $\mathcal{S}_i/w_i>0.01$) the giant component at $T\simeq0.14$ and $T\simeq0.16$, respectively.
shows the distribution of the total number of nodes in small components for various values of $T$. To support our claim that outbreaks are mostly confined within the HCW populations, also displays the distribution of the number of nodes of type 2 in the small components. The small shift between the two curves is due to adults and children being infected mostly in households. Again, we conclude in an excellent agreement between both the numerical simulations and theoretical predictions of the formalism obtained by solving –.
Interestingly, – give evidence of what one may call the “local percolation” of the hospital cliques as $T$ increases. For $T<T^{\prime}$, the size distribution falls rapidly and monotonously as expected for generic CM graphs [@Newman01_PhysRevE; @Newman07_PhysRevE]. For $T^{\prime}<T<T_c$, however, the shape of the distribution changes as local maxima appear. These are due to the growing spanning cluster in the hospital cliques. For $T>T_c$, most of the HCW population is part of the giant component, and the spanning cluster is more and more likely to cover the entire clique as $T$ increases. The HCW nodes that are not part of the giant component are therefore likely to be part of very large small components composed of one or more “locally percolated” cliques. This is confirmed by the multiple maxima seen on –\[fig:allard\_fig\_3c\].
Special cases, generalization and applications {#sec:special_cases}
==============================================
We now demonstrate our claims that our formalism encompasses many percolation models on random graphs published in the litterature. We also succinctly outline a possible generalization and some straightforward applications of our model.
Multitype random graphs {#sec:sc_multitype}
-----------------------
Our formalism naturally falls back on the model introduced in [@Allard09_PhysRevE] describing the heterogeneous bond percolation on multitype random graphs. In this class of graphs, there are $M$ types of nodes, and a $i \rightarrow j$ edge is occupied with probability $T_{ij}$. Type-$i$ nodes occupy a fraction $w_i$ of the graph, and a type-$i$ node is connected to $\tilde{k}_j$ type-$j$ nodes (for each $j \in [1,M]$) with probability $\tilde{P}_i(\tilde{k}_1,\tilde{k}_2,\ldots,\tilde{k}_M)$.
Our formalism reproduces this model by using one group type for each possible (unordered) type of $i \rightarrow j$ edge. To each of the $\Lambda=M(M+1)/2$ group types are associated the functions $$\begin{aligned}
\theta_{i\nu}(\bm{x}) & = [1 + (x_{\nu j} - 1)T_{ij}] \\
\theta_{j\nu}(\bm{x}) & = [1 + (x_{\nu i} - 1)T_{ji}]\end{aligned}$$ depending whether the edge is considered in the $i \rightarrow j$ or in the $j \rightarrow i$ direction. Along with these functions, $P_i(\bm{k})$ can therefore reproduce the degree distribution $\tilde{P}_i(\tilde{k}_1,\tilde{k}_2,\ldots,\tilde{k}_M)$.
As shown in [@Allard09_PhysRevE], multitype random graphs naturally encompasses multipartite graphs, as well as the undirected random graphs introduced in [@Newman01_PhysRevE; @Newman02_PhysRevE; @Newman03b_PhysRevE]. By assigning nodes with a given degree to a same node type, our formalism can also reproduce degree-degree correlation as in [@Newman02_PhysRevLett].
Clustered random graphs
-----------------------
Being a multitype generalization of the *highly clustered random graphs* introduced in [@Newman03b_PhysRevE], our model simplifies to the latter in a straightforward manner with $M=\Lambda=1$ and all groups being Erdős-Rényi cliques. For $\Lambda=1$, the groups to which any given node belongs to is averaged in so that no correlation whasoever can be taken into account.
When considering only $M=1$ type of nodes but an arbitrary number of uniquely configured groups \[$R_\nu(\bm{n})=1$ for all $\nu$\], we retrieve *random graphs containing arbitrary distributions of subgraphs* as introduced in [@Karrer10_PhysRevE]. The unweighted average plays an analogous function as their *role* distribution, with correlation being taken into account by using node types. It is then straightforward to conclude that our formalism also encompasses the *edge-triangle* model introduced in [@Miller09_PhysRevE; @Newman09_PhysRevLett] and the *strong ties* model proposed by [@Shi07_PhysicaA].
The *$\gamma$-theory* model [@Gleeson09_PhysRevE] can be recovered by considering only $M=1$ type of nodes, and by allowing nodes to belong to only one group of size larger than two (Erdős-Rényi cliques) but to belong to an arbitrary number of group of size two (external edges). Also, the *random hypergraphs* introduced in [@Ghoshal09_PhysRevE] can be reproduced by our formalism by considering $M=3$ types of nodes and $\Lambda=1$ type of groups which are triangles composed of one node of each type.
Finally, a class of formalism [@Serrano06_PhysRevLett; @Zlatic12_EPL; @Serrano06b_PhysRevE] uses the *multiplicity* of edges — the number of triangles to which an edge participates — to derive an effective branching process and solve the percolation on clustered graphs using PGFs. Although this approach tackles percolation from a different perspective, its predictions (*i.e.*, the percolation threshold and the size of the giant component) can be reproduced with our model by using fully connected motifs of size $m+2$ to account for links of multiplicity $m$ and by appropriately using node and group types to account for the correlations that this class of models incorporates.
Directed random graphs
----------------------
Our formalism as presented in this paper can only model directed edges between *different* node types. To describe directed edges among a same node type such as in [@Newman01_PhysRevE; @Meyers06_JTheorBiol], we would need to subdivide the group type corresponding to directed edges into an incoming part and an outgoing part (e.g., $\nu \rightarrow \nu_\mathrm{in}, \nu_\mathrm{out}$), and match the complementary parts to form groups in the unclustered multitype bipartite graph. In other words, each group is linked to an incoming and an outgoing stub. The mathematical formalism introduced in remains valid except that we would need to explicitly consider the fact that nodes are reached by their incoming edges and are left by their outgoing edges when writing down the equations (see [@Newman01_PhysRevE; @Meyers06_JTheorBiol] for detailed examples). This adjustment is nevertheless straightforward and does not affect the generality of our approach.
Interdependent or coupled networks
----------------------------------
The use of node and group types naturally permits our formalism to be used in the study of interdependent or coupled networks. In interdependent – or interacting – networks, node types could for instance be used to distinguish the elements of two or more interacting networks [@Leicht09_arXiv; @Buldyrev10_Nature; @Huang11_PhysRevE]. Different group types would then allow to specify precisely the (nontrivial) interactions within and across the networks. In the case of coupled, or overlayed, networks [@Marceau11_PhysRevE; @Funk10_PhysRevE] elements in a single population interact in different ways which is modelled using different edge types. This again can be easily achieved with our formalism by defining multiple group types, one for each level of interaction. Again, the generality of our approach gives us access to a wide variety of complex patterns of interactions in a very detailed fashion.
\[sec:weak\_and\_strong\]Weak and strong clustering regimes
-----------------------------------------------------------
The existence of two regimes of clustering, *weak* and *strong*, has been put forward in [@Serrano06_PhysRevLett; @Serrano06a_PhysRevE; @Serrano06b_PhysRevE] with the conclusion that these two regimes have opposite effect on the bond percolation threshold. In the weak regime, edges have a multiplicity of either 0 or 1 (single edges or disjoint triangles), and the percolation threshold is higher than for equivalent unclustered graphs. In the strong regime, edges may contribute to more than one triangle, and it is argued that the percolation threshold is then lower than for equivalent unclustered graphs.
Contrariwise, the analysis done in [@Miller09_PhysRevE; @Gleeson09_PhysRevE] strongly suggests that clustering always increases the percolation threshold and that the observed lower percolation threshold in the strong regime is due to assortative mixing instead. Hence, according to theses results, there should be no weak and strong clustering regimes. The use of node and group (or edge) types in our model can generate clustered and unclustered graphs with the same correlations (or mixing patterns). It is therefore possible to investigate — both numerically and analytically — the effect of clustering alone on the percolation threshold, shedding some light on this contradiction while extending the analysis and the conclusions of [@Miller09_PhysRevE; @Gleeson09_PhysRevE]. This will be addressed in a future publication.
Conclusion {#sec:conclusion}
==========
We have presented a generalization of the Configuration Model allowing for the inclusion of several nontrivial mixing patterns and clustering. On the one hand, the use of node and group types permits to explicitly prescribe how nodes are connected to one another, hence reproducing (dis-)assortative mixing, and indirectly degree-degree correlation. On the other hand, the use of a one-mode projection can generate a wide range of nontrivial clustered structures through quenched or annealed motifs. Besides the modeling of mixing patterns, the multitype approach permits to identify nodes. This allows to highlight unusual behaviors or susceptibility of sub-population of nodes, as well as to simulate targetted intervention such as attacks, failures, vaccination or quarantine. We have also demonstrated that our formalism encompasses several models published to this day, and we have outlined potential applications.
Bridging the gap between empirical network datasets and theoretical models is surely one the principal tenets of network theory. Since extracting the effective clustered backbone (*i.e.*, motifs) of real networks is still an open problem, our approach can only offer a partial answer. However, it provides a comprehensive synthesis of the many variants of the CM published to date, and it extends considerably the structural complexity of graphs that can be handled theoretically. In these regards, the versatility and generality of the present framework could prove useful even beyond the strict confines of bond percolation on complex graphs. The authors would like to thank Vigé Lebrun for fruitful and sobering discussions. We are also grateful to an anonymous referee for pointing out a number of relevant publications. This work has been supported by the CIHR, the NSERC and the FRQ-NT.
General method to compute $Q_{i\nu}(\bm{l}|\bm{n})$ for arbitrary multitype motifs {#sec:general_method}
==================================================================================
We present a systematic way to compute the outcome of bond percolation, $Q_{i\nu}(\bm{l}|\bm{n})$, on any arbitrary multitype motifs where edges are simple and can be directed or not.
Let us first consider a multitype generalization of Erdős-Rényi random graphs. These are composed of $\bm{n}$ nodes, and a directed edge exists from a type-$i$ node to a type-$j$ node with probability $p_{ij}$. Edges exist independently of one another. Note that the symmetric case $p_{ij}=p_{ji}$ is statistically equivalent to undirected edges. It has been shown [@Allard12_EPL] that $Q_{i\nu}(\bm{l}|\bm{n})$ can be obtained by iterating $$\begin{aligned}
Q_{i\nu}(\bm{l}|\bm{n}) & = Q_{i\nu}(\bm{l}|\bm{l}) \prod_{rs} {n_r\!-\!\delta_{ir} \choose l_r\!-\!\delta_{ir}} (1\!-\!p_{rs})^{l_r(n_s-l_s)} \label{eq:iterative_erdos_renyi_1} \\
\fl \mbox{and} \nonumber \\
Q_{i\nu}(\bm{l}|\bm{l}) & = 1 - \sum_{\bm{m}<\bm{l}} Q_{i\nu}(\bm{m}|\bm{l}) \label{eq:iterative_erdos_renyi_2}\end{aligned}$$ from the initial condition $Q_{i\nu}(\bm{\delta_i}|\bm{\delta_i}) = 1$ with $\bm{\delta_i} \equiv (\delta_{i1},\ldots,\delta_{iM})$. In essence, knowing the probability of finding a component of size $\bm{l}$ from a node of type $i$ in a graph of size $\bm{l}$, computes the probability of finding a sub-component of size $\bm{l}$ but in a graph of size $\bm{n}$ ($>\bm{l}$). This allows to compute every coefficients of the distribution $Q_{i\nu}(\bm{l}|\bm{n})$ except for the last one, the one corresponding to the case where the whole graph is reachable, which is obtained using .
Let $\mathcal{G}$ be a multitype motif composed of $\bm{n}$ nodes with an arbitrary configuration of edges. The associated distribution $Q_{i\nu}(\bm{l}|\bm{n})$ can then be computed by following these simple steps:
1. Consider an equivalent multitype Erdős-Rényi graph $\mathcal{G}'$ of size $n'=\sum_j n_j$ in which each node belongs to its own unique type (i.e., $n'_{j'}=1$ for all $j' \in \{1,\ldots,n'\}$). Note $p'_{i'j'}$ the probability for a directed edge to exist from the type-$i'$ node to the type-$j'$ one.
2. Compute $Q'_{i'\nu}(\bm{l'}|\bm{n'})$ for $\mathcal{G}'$ with –. Without any loss of generality suppose that the initial node from which the graph is probed is of \[i.e., – need to be solved only once\].
3. From $Q'_{i'\nu}(\bm{l'}|\bm{n'})$, derive the intermediate distribution $Q_{i\nu}^{(j)}(\bm{l}|\bm{n})$ of the number of nodes of each type that are accessible from the $j$-th type-$i$ node. This is achieved by replacing the *artificial* node types in $\mathcal{G}'$ by the *actual* node types in $\mathcal{G}$, and by setting the values of $p'_{i'j'}$ according to the configuration of the edges in $\mathcal{G}$, which can include type-dependent probabilities of existence/occupation of edges.
4. Obtain $Q_{i\nu}(\bm{l}|\bm{n})$ by computing the unweighted average of the $n_i$ distributions $Q_{i\nu}^{(j)}(\bm{l}|\bm{n})$ $$\label{eq:unweighted_average}
Q_{i\nu}(\bm{l}|\bm{n}) = \frac{1}{n_i}\sum_{j} Q_{i\nu}^{(j)}(\bm{l}|\bm{n}) \ .$$
A noteworthy point is that the distribution $Q'_{i\nu}(\bm{l'}|\bm{n'})$ computed for a generic graph of size $n'$ can generate every multitype motif of size smaller than $n'$ by appropriate choices of $p'_{i'j'}$. An explicit example of such a calculation is given in [@Allard12_EPL].
Numerical simulations {#sec:num_sim_details}
=====================
[ c c c]{} Group type & Composition & Probability\
& $\bm{n} = (n_1,n_2,n_3)$ & $R_{\nu}(\bm{n})$\
& $(2,0,0)$ & 0.0810\
& $(1,1,0)$ & 0.0180\
& $(0,2,0)$ & 0.0010\
& $(2,0,1)$ & 0.1215\
& $(1,1,1)$ & 0.0270\
& $(0,2,1)$ & 0.0015\
& $(2,0,2)$ & 0.3240\
& $(1,1,2)$ & 0.0720\
& $(0,2,2)$ & 0.0040\
& $(2,0,3)$ & 0.2835\
& $(1,1,3)$ & 0.0630\
& $(0,2,3)$ & 0.0035\
& $(5,0,50)$ & 0.2500\
& $(10,0,100)$ & 0.5000\
& $(15,0,150)$ & 0.2500\
& $(10,0,0)$ & 0.1000\
& $(20,0,0)$ & 0.2500\
& $(30,0,0)$ & 0.3000\
& $(40,0,0)$ & 0.2500\
& $(50,0,0)$ & 0.1000\
Hospitals & (0,300,0) & 1.0000\
Friendships & (0,0,5) & 1.0000\
Directed edges (1 $\rightarrow$ 2) & (1,1,0) & 1.0000\
Directed edges (3 $\rightarrow$ 2) & (0,1,1) & 1.0000\
Details of the graphs used in and of the numerical simulations performed to validate our formalism are presented.
Urban networks
--------------
The graphs generated in were inspired by the *urban networks* used in [@Meyers05_JTheorBiol; @Bansal06_PLoSMed] in which individuals are connected to one another because of their common membership to a social group (e.g., households, schools, workplaces, hospitals, friendship circles). In this case, the population is divided into three categories – identified by node types – namely adults (type 1), health-care workers (HCW, and children (type 3) with $\{w_i\} = \{0.45,0.05, 0.50\}$. Every node belongs to one household, every HCW belongs to one hospital, every child belongs to one school, 1/9 of adults belong to one school (teachers, janitors, etc.) and the remaining 8/9 belong to one workplace. Also every child belongs to one group of friends (see ), and adults and children are connected at most to two randomly chosen HCW via a directed edge.
explicits the group composition distribution $R_{\nu}(\bm{n})$ used to generate the urban networks. Except for friendship circles, the connections between individuals within groups are modeled with multitype Erdős-Rényi graphs with different probabilities of edge existence. In households, every possible edge exists except for the directed edges from HCW to adults, HCW and children that exist with probability 0.2, 0.2 and 0.1, respectively. In schools and workplaces, edges exist with probability 0.01 and they exist with probability 0.05 in hospitals. The use of relatively large cliques with such low probabilities of existence of edges allows to model redundancy in the neighbourhood of nodes while keeping a relatively low clustering. Finally, directed edges from adults and children to HCW exist with probability 0.5.
These graphs can be generated in a fairly straightforward manner. For a given group type $\nu$, we first generate a sequence of groups whose composition is prescribed by $R_{\nu}(\bm{n})$. We then generate, according to $P_i(\bm{k})$, a list of nodes in which a node belonging to $k_\nu$ type-$\nu$ groups appears $k_\nu$ times. We finally randomly assign these nodes to the groups, and create edges between nodes that are members of a same group according to the probabilities given in the last paragraph.
Percolation simulations
-----------------------
Graphs that were used to obtain the results shown in were composed of at least 1.2$\times 10^5$ nodes. For $T$ around $T_c$, larger graphs (up to 9.6$\times 10^6$ nodes) have been generated to faciliate the distinction between small components, which are intensive, from the giant component, which is extensive. At least $10^3$ ($10^6$) graphs were generated for each value of $T$ used in ().
For each generated graph, 100 percolation simulations were performed. These consist in randomly choosing a starting node and then following every possible edges leaving this node – and the subsequently encountered nodes – with probability $T$ until no new node can be reached. The component size is then simply the number of nodes that have been reached. While it would have been straightforward to use a type-specific probability $T$ (see ), we have used a single value to lighten the presentation of the results.
References {#references .unnumbered}
==========
[10]{}
S. N. Dorogovtsev, A. V. Goltsev, and J. F. F. Mendes. . , 80(4):1275–1335, 2008.
M. E. J. Newman. . Oxford University Press, 2010.
M. Molloy and B. Reed. . , 6(2-3):161–180, 1995.
M. Molloy and B. Reed. . , 7(3):295–305, 1998.
M. E. J. Newman, S. H. Strogatz, and D. J. Watts. . , 64(2):026118, 2001.
A. Allard, P.-A. Noël, L. J. Dubé, and B. Pourbohloul. . , 79(3):036113, 2009.
G. Ghoshal, V. Zlatić, G. Caldarelli, and M. E. J. Newman. . , 79(6):066118, 2009.
J. P. Gleeson. . , 80(3):036107, 2009.
B. Karrer and M. E. J. Newman. . , 82(6):066118, 2010.
E. A. Leicht and R. M. D’Souza. . , 2009.
J. C. Miller. . , 80(2):020901(R), 2009.
L. A. Meyers, M. E. J. Newman, and B. Pourbohloul. , 240(3):400–18, 2006.
M. E. J. Newman. . , 66(1):016128, 2002.
M. E. J. Newman. . , 89(20):208701, 2002.
M. E. J. Newman. . , 67(2):026126, 2003.
M. E. J. Newman. . , 68(2):026121, 2003.
M. E. J. Newman. . , 103(5):058701, 2009.
M. Á. Serrano and M. Boguñá. . , 97(8):088701, 2006.
X. Shi, L. A. Adamic, and M. J. Strauss. . , 378(1):33–47, 2007.
V. Zlatić, D. Garlaschelli, and G. Caldarelli. . , 97(2):28005, 2012.
A. Allard, L. Hébert-Dufresne, P.-A. Noël, V. Marceau, and L. J. Dubé. . , 98(1):16001, 2012.
R. Milo, S. Shen-Orr, S. Itzkovitz, N. Kashtan, D. Chklovskii, and U. Alon. , 298(5594):824–7, 2002.
S. V. Buldyrev, R. Parshani, G. Paul, H. E. Stanley, and S. Havlin. , 464(7291):1025–8, 2010.
X. Huang, J. Gao, S. V. Buldyrev, S. Havlin, and H. E. Stanley. . , 83(6):065101(R), 2011.
V. Marceau, P.-A. Noël, L. Hébert-Dufresne, A. Allard, and L. J. Dubé. . , 84(2):026105, 2011.
M. Á. Serrano and M. Boguñá. . , 74(5):056114, 2006.
M. Á. Serrano and M. Boguñá. . , 74(5):056115, 2006.
M. E. J. Newman. . , 45(2):167, 2003.
C. D. Meyer. . SIAM, 2001.
L. A. Meyers, B. Pourbohloul, M. E. J. Newman, D. M. Skowronski, and R. C. Brunham. , 232(1):71–81, 2005.
S. Bansal, B. Pourbohloul, and L. A. Meyers. , 3(10):e387, 2006.
M. E. J. Newman. . , 76(4):045101(R), 2007.
S. Funk and V. A. A. Jansen. . , 81(3):036118, 2010.
[^1]: We see from that $\mathbf{B}$ is a non-negative and, in general, irreducible matrix. Thus the Perron-Frobenius theorem [@Meyer00_MatrixAnalysis] ensures that the largest eigenvalue of $\mathbf{B}$ is simple, real and positive. Moreover, the associated eigenvector is the only nonnegative eigenvector of $\mathbf{B}$.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We propose a method based on optically detected magnetic resonance (ODMR) to measure the decoherence time $T_{2}$ of a single electron spin in a semiconductor quantum dot. The electron spin resonance (ESR) of a single excess electron on a quantum dot is probed by circularly polarized laser excitation. Due to Pauli blocking, optical excitation is only possible for one of the electron-spin states. The photoluminescence is modulated due to the ESR which enables the measurement of electron-spin decoherence. We study different possible schemes for such an ODMR setup.'
author:
- 'Oliver Gywat, Hans-Andreas Engel, and Daniel Loss'
- 'R.J. Epstein, F.M. Mendoza, and D.D. Awschalom'
title: 'Optical Detection of Single-Electron Spin Decoherence in a Quantum Dot '
---
Introduction
============
Quantum information can be encoded in states of an electron spin $1/2$ in a semiconductor quantum dot.[@spintronics] However, information processing is intrinsically limited by the spin lifetime. For single spins, one distinguishes between two characteristic decay times $T_{1}$ and $T_{2}$. The relaxation of an excited spin state in a magnetic field into the thermal equilibrium is associated with the spin relaxation time $T_{1}$, whereas the spin decoherence time $T_{2}$ is related to the loss of phase coherence of a single spin that is prepared in a superposition of its eigenstates. Experimental $T_{2}$ measurements of single spins in quantum dots are highly desirable because $T_{2}$ is the limiting time scale for coherent spin manipulation.
Recent optical experiments have demonstrated the coherent control and the detection of excitonic states of single quantum dots.[@gammon1] Nevertheless, the measurement of the $T_{2}$ time of a single electron spin in a quantum dot using optical methods has turned out to be an intricate problem. This is mainly due to the interaction of the electron and the hole inside an exciton.[@current] The electron and hole spin are decoupled only if the hole spin couples (via spin-orbit interaction) stronger to the environment than to the electron spin. Recent experiments, measuring Faraday rotation, have suggested that this is not the case for excitons in quantum dots.[@guptaT2] Alternatively, if electron-hole pairs are excited inside the barrier material of a quantum dot heterostructure, the carriers diffuse after their creation to the dots and are captured inside them within typically tens of picoseconds.[@ohnesorge; @raymond] By that time, electron and hole spins have decoupled. In such an experiment, the Hanle effect would allow the measurement of electron-spin decoherence. However, this approach[@epstein] has not yet given conclusive results for $T_{2}$.
What is a promising approach to measure the electron-spin decoherence time $T_{2}$ by optical methods? For this, initially some coherence of the electron spin must be produced, preferably in the absence of holes. This can be done using electron spin resonance (ESR). The coherence decays and, after some time, the remaining coherence is measured optically. This implies using optically detected magnetic resonance (ODMR). ODMR schemes have, e.g., been applied to measure the spin coherence of single nitrogen-vacancy centers in diamond.[@gruber] For quantum dots, ODMR has recently been applied to electrons and holes in CdSe dots[@Lifshitz] and to excitons in InAs/GaAs dots.[@zurauskiene] While these two experiments have not considered single spin coherence, the feasibility of the combination of ESR and optical methods in quantum dot experiments has been demonstrated.
In this work, we make use of Pauli blocking of exciton creation[@pauliblocking] in an ODMR setup. We show that the linewidth of the photoluminescence as a function of the ESR field frequency provides a lower bound on $T_{2}$. Further, if pulsed laser and cw ESR excitation are applied, electron spin Rabi oscillations can be detected via the photoluminescence.
We consider quantum dots which confine electrons as **well as **holes (type I dots). We assume a ground state where the dot is charged with one single electron. This can be achieved, e.g., by $n$ doping[@cortez] or by electrical injection.[@abstreiter2] Such a single-electron state can be optically excited, which leads to the formation of a negatively charged exciton, consisting of two electrons and one hole. Recent experiments on InAs dots[@trion1; @trion2] and GaAs dots[@trion3] have shown that in the charged exciton ground state, the two electrons form a spin singlet in the lowest (conduction-band) electron level and the hole occupies the lowest (valence-band) hole level. Note that single-electron level spacings can be relatively large, e.g., on the order of 50 meV for InAs dots.[@fricke] Typically, the level spacing of confined hole states is smaller than the one of electrons.[@schmidt] We assume that the lowest heavy hole (hh) (with total angular momentum projection $J_{z}\!=\!\pm3/2$) and light hole (lh) ($J_{z}\!=\!\pm1/2$) dot levels are split by an energy $\delta_{hh-lh}$. Additionally, mixing of hh and lh states should be negligible.[@mixing] These conditions are satisfied for several types of quantum dots.[@trion1; @trion2; @trion3; @efrosCdSe; @efros2] Then, circularly polarized optical excitation that is restricted to either hh or lh states excites spin-polarized electrons. In this work, we first assume a hh ground state for holes. We discuss then different hole configurations.
The states of a quantum dot can be taken as follows; see also Fig. \[fig:states\]. A single electron in the lowest orbital state is either in the spin ground state ${{|\! \uparrow\rangle}}$ or in the excited spin state ${{|\! \downarrow\rangle}}$. Adding an electron-hole pair, the negatively charged exciton (in the orbital ground state) is either in the excited spin state ${{|{X_{\downarrow}^{-}}\rangle}}$ or in the spin ground state ${{|{X_{\uparrow}^{-}}\rangle}}$. For these excitonic states, the subscripts $\downarrow,\uparrow$ refer to the hh spin and we apply ****the usual time-inverted ****notation for hole spins. For simplicity, we assume $\mathrm{sign}(g_{e}^{z})=\mathrm{sign}\left(g_{hh}^{z}\right)$ for the electron and the hh $g$ factors in $z$ direction. Note that the very same scheme can also be applied if the sign of $g_{hh}^{z}$ is reversed. Then, one would use a $\sigma^{+}$ laser field and all results apply after interchanging ${{|{X_{\downarrow}^{-}}\rangle}}$ and ${{|{X_{\uparrow}^{-}}\rangle}}$.
Hamiltonian
===========
We describe the coherent dynamics of a quantum dot, charged with a single excess electron, in this ODMR setup with the Hamiltonian $$H={H_{\mathrm{dot}}}+H_{\mathrm{ESR}}+{H_{\mathrm{L}}}+H_{\mathrm{d-L}},\label{eq:H}$$ coupling the three states ${{|\! \uparrow\rangle}}$, ${{|\! \downarrow\rangle}}$, and ${{|{X_{\downarrow}^{-}}\rangle}}$. Here, ${H_{\mathrm{dot}}}$ comprises the quantum dot potential, the Zeeman energies due to a constant magnetic field in $z$ direction, and the Coulomb interaction of electrons and holes. It defines the dot energy $E_{n}$ by ${H_{\mathrm{dot}}}{|n\rangle}=E_{n}{|n\rangle}$. Here, the electron Zeeman splitting is $g_{e}^{z}{{\mu_{\mathrm{B}}}}B_{z}=E_{\downarrow}-E_{\uparrow}$, where ${{\mu_{\mathrm{B}}}}$ is the Bohr magneton.[@overhauser] The ESR term $H_{\mathrm{ESR}}(t)$ couples ${{|\! \uparrow\rangle}}$ and ${{|\! \downarrow\rangle}}$ via $\mathbf{B}_{\perp}(t)$, which rotates with frequency ${\omega_{\mathrm{ESR}}}$ in the $xy$ plane.[@linESR; @engel] The ESR Rabi frequency is ${\Omega_{\mathrm{ESR}}}=g_{e}^{\perp}{{\mu_{\mathrm{B}}}}B_{\perp}$, with $g$ factor $g_{e}^{\perp}$. Even if the ESR field is also resonant with the hole Zeeman splitting, it has a negligible effect on the charged exciton states since they recombine quickly. An oscillating field ${{\mu_{\mathrm{B}}}}\tensor{\mathbf{g}}\mathbf{B}$ can also be produced with voltage-controlled modulation of the electron $g$ tensor $\tensor{\mathbf{g}}$.[@kato] A $\sigma^{-}$-polarized laser beam is applied in $z$ direction (typically parallel to $[001]$), with free laser field Hamiltonian ${H_{\mathrm{L}}}=\omega_{{\mathrm{L}}}a_{{\mathrm{L}}}^{\dagger}a_{{\mathrm{L}}}$, where the laser frequency is $\omega_{{\mathrm{L}}}$, $a_{{\mathrm{L}}}^{(\dagger)}$are photon operators, and we set $\hbar=1$. The coupling of ${{|\! \downarrow\rangle}}$ and ${{|{X_{\downarrow}^{-}}\rangle}}$ to the laser field is described by $H_{\mathrm{d}-\mathrm{L}}$ which introduces the complex optical Rabi frequency ${\Omega_{{\mathrm{L}}}}$.[@optRabi] Since the dot is only coupled to a single circularly polarized laser mode via $H_{\mathrm{d-L}}$, the terms that violate energy conservation vanish due to selection rules. If the laser bandwidth is smaller than $\delta_{hh-lh}$, the absorption of a $\sigma^{-}$ photon in the spin ground state ${{|\! \uparrow\rangle}}$ is excluded due to Pauli blocking.[@trionexc] We neglect all multi-photon processes via other levels since they are only relevant to high-intensity laser fields. For this configuration, the $\sigma^{-}$ photon absorption is switched “on” and “off” by the ESR-induced electron-spin flips. Here, the laser bandwidth and the temperature can safely exceed the electron Zeeman splitting. We transform $H$ into the rotating frame with respect to ${\omega_{\mathrm{ESR}}}$ and $\omega_{\mathrm{L}}$. The laser detuning is ${\delta_{{\mathrm{L}}}}=(E_{{X\downarrow}}-E_{\downarrow})-\omega_{\mathrm{L}}$ and the ESR detuning ${\delta_{\mathrm{ESR}}}=g_{e}^{z}{{\mu_{\mathrm{B}}}}B_{z}-{\omega_{\mathrm{ESR}}}$.
Generalized Master Equation
===========================
We next consider the reduced density matrix for the dot, ${\rho_{{}}}={{{\mathrm{Tr}}_{{\mathrm{R}}}\, }}{\rho_{\mathrm{F}}}$, where ${\rho_{\mathrm{F}}}$ is the full density matrix and ${{{\mathrm{Tr}}_{{\mathrm{R}}}\, }}$ is the trace taken over the environment (or reservoir). In the von Neumann equation $\dot{\rho}_{\mathrm{F}}=-i\left[H,\,{\rho_{\mathrm{F}}}\right]$, we treat the interaction with the ESR and laser fields exactly with the Hamiltonian in the rotating frame. We describe the coupling with the environment (radiation field, nuclear spins, phonons, spin-orbit interaction, etc.) with phenomenological rates. We write $W_{nm}\equiv W_{n\leftarrow m}$ for (incoherent) transitions from state ${|m\rangle}$ to ${|n\rangle}$ and ${V}_{nm}$ for the decay of off-diagonal elements of ${\rho_{{}}}$. Note that usually ${V}_{nm}\geq\frac{1}{2}\sum_{k}\left(W_{kn}+W_{km}\right)$. The electron-spin relaxation time[@kimble] is $T_{1}=\left({W}_{{{\uparrow\downarrow}}}+{W}_{{{\downarrow\uparrow}}}\right)^{-1}$, with spin-flip rates ${W}_{{{\uparrow\downarrow}}}$ and ${W}_{{{\downarrow\uparrow}}}$. In the absence of the ESR and laser excitations, the off-diagonal matrix elements of the electron spin decay with the (intrinsic) single-spin decoherence rate ${V}_{{{\downarrow\uparrow}}}=1/T_{2}$. The linewidth of the optical $\sigma^{-}$ transition is denoted by ${{V}_{\mathrm{X}}}={V}_{{{X\downarrow},\downarrow}}$. We use the notation $\rho_{n}={\langlen|}\rho{|n\rangle}$ and $\rho_{nm}={\langlen|}\rho{|m\rangle}$. The master equation is given in the rotated basis ${{|\! \uparrow\rangle}}$, ${{|\! \downarrow\rangle}}$, ${{|{X_{\uparrow}^{-}}\rangle}}$, ${{|{X_{\downarrow}^{-}}\rangle}}$ as ${\dot{\rho}_{{}}}=\mathcal{M}{\rho_{{}}}$, where $\mathcal{M}$ is a superoperator. Explicitly,$$\begin{aligned}
{{{{\dot{\rho}_{\uparrow}}}}}& = & {\Omega_{\mathrm{ESR}}}{\textrm{Im}}{{\rho_{{{\downarrow\uparrow}}}}}\!+\!{{W}_{\mathrm{em}}}\rho_{{{X\uparrow}}}\!+\!{W}_{{{\uparrow\downarrow}}}{{{\rho_{\downarrow}}}}\!-\!{W}_{{{\downarrow\uparrow}}}{{{\rho_{\uparrow}}}},\label{eq:fullmastereq}\\
{{{{\dot{\rho}_{\downarrow}}}}}& = & -{\Omega_{\mathrm{ESR}}}{\textrm{Im}}{{\rho_{{{\downarrow\uparrow}}}}}+{\textrm{Im}}({\Omega_{{\mathrm{L}}}}^{*}\rho_{{{X\downarrow},\downarrow}})+{{W}_{\mathrm{em}}}\,\rho_{{{X\downarrow}}}\nonumber \\
& & +{W}_{{{\downarrow\uparrow}}}\,{{{\rho_{\uparrow}}}}-{W}_{{{\uparrow\downarrow}}}\,{{{\rho_{\downarrow}}}},\\
\dot{\rho}_{{{X\downarrow}}} & = & -{\textrm{Im}}({\Omega_{{\mathrm{L}}}}^{*}\rho_{{{X\downarrow},\downarrow}})+{W}_{{{X\downarrow},{X\uparrow}}}\,\rho_{{{X\uparrow}}}\nonumber \\
& & -\left({{W}_{\mathrm{em}}}+{W}_{{{X\uparrow},{X\downarrow}}}\right)\rho_{{{X\downarrow}}},\\
\dot{\rho}_{{{X\uparrow}}} & = & {W}_{{{X\uparrow},{X\downarrow}}}\,\rho_{{{X\downarrow}}}-\left({{W}_{\mathrm{em}}}+{W}_{{{X\downarrow},{X\uparrow}}}\right)\rho_{{{X\uparrow}}},\\
{{{{\dot{\rho}_{{{\downarrow\uparrow}}}}}}}& = & \frac{i}{2}{\Omega_{\mathrm{ESR}}}\left({{{\rho_{\downarrow}}}}-{{{\rho_{\uparrow}}}}\right)-\frac{i}{2}{\Omega_{{\mathrm{L}}}}^{*}\rho_{{{X\downarrow},\uparrow}}\nonumber \\
& & -\left(i{\delta_{\mathrm{ESR}}}+T_{2}^{-1}\right)\,{{\rho_{{{\downarrow\uparrow}}}}},\\
\dot{\rho}_{{{X\downarrow},\uparrow}} & = & \frac{i}{2}{\Omega_{\mathrm{ESR}}}\,\rho_{{{X\downarrow},\downarrow}}-\frac{i}{2}{\Omega_{{\mathrm{L}}}}{{\rho_{{{\downarrow\uparrow}}}}}\nonumber \\
& & -[i({\delta_{\mathrm{ESR}}}+{\delta_{{\mathrm{L}}}})+{V}_{{{X\downarrow},\uparrow}}]\,\rho_{{{X\downarrow},\uparrow}},\\
\dot{\rho}_{{{X\downarrow},\downarrow}} & = & \frac{i}{2}{\Omega_{\mathrm{ESR}}}\rho_{{{X\downarrow},\uparrow}}-\frac{i}{2}{\Omega_{{\mathrm{L}}}}({{{\rho_{\downarrow}}}}-\rho_{{{X\downarrow}}})\nonumber \\
& & -(i{\delta_{{\mathrm{L}}}}+{{V}_{\mathrm{X}}})\rho_{{{X\downarrow},\downarrow}}.\end{aligned}$$ The remaining matrix elements of ${\rho_{{}}}$ are decoupled and are not important here.
ESR Linewidth in Photoluminescence
==================================
We first consider the photoluminescence for a cw ESR and laser field. For this, we calculate the stationary density matrix ${\bar{\rho}}_{{}}$ with $\dot{{\bar{\rho}}}_{{}}=0$. We introduce the rate $${{{W}_{{\mathrm{L}}}}}=\frac{|{\Omega_{{\mathrm{L}}}}|^{2}}{2}\:\frac{{{V}_{\mathrm{X}}}}{{{V}_{\mathrm{X}}}^{2}+{\delta_{{\mathrm{L}}}}^{2}}\label{eqWeffEx}$$ for the optical excitation, with maximum value ${{{W}_{{\mathrm{L}}}}}^{\mathrm{max}}$ at ${\delta_{{\mathrm{L}}}}=0$. We first solve $\dot{{\bar{\rho}}}_{{{X\downarrow},\uparrow}}=0$ and find that the coupling to the laser field produces an additional decoherence channel to the electron spin. We obtain the renormalized spin decoherence rate ${{V}_{\mathrm{ESR}}}$ which satisfies$${{V}_{\mathrm{ESR}}}\leq{\frac{1}{T_{2}}}+\frac{|{\Omega_{{\mathrm{L}}}}|^{2}}{4{V}_{{{X\downarrow},\uparrow}}}\approx{\frac{1}{T_{2}}}+\frac{1}{2}{{{W}_{{\mathrm{L}}}}}^{\mathrm{max}}.\label{eqndecohESReff}$$ Further, the ESR detuning is also renormalized, $${\tilde{\delta}_{\mathrm{ESR}}}\geq{\delta_{\mathrm{ESR}}}\left[1-\frac{|{\Omega_{{\mathrm{L}}}}|^{2}}{\left({{W}_{\mathrm{em}}}+{{W}_{{{X\uparrow},{X\downarrow}}}}\right)^{2}}\right].$$ We assume that these renormalizations and ${\delta_{{\mathrm{L}}}}$ are small compared to the linewidth of the optical transition, i.e., ${{{W}_{{\mathrm{L}}}}}^{\mathrm{max}},\,|{\tilde{\delta}_{\mathrm{ESR}}}-{\delta_{\mathrm{ESR}}}|<{{V}_{\mathrm{X}}}$. Then, if both transitions are near resonance, ${\delta_{{\mathrm{L}}}}\lesssim{{V}_{\mathrm{X}}}$ and $|{\tilde{\delta}_{\mathrm{ESR}}}|\lesssim{{V}_{\mathrm{ESR}}}$, no additional terms appear in the renormalized master equation. We solve $\dot{{\bar{\rho}}}_{{{X\downarrow},\downarrow}}=0$ and $\dot{{\bar{\rho}}}_{{{\uparrow\downarrow}}}=0$ and introduce the rate$${{{W}_{\mathrm{ESR}}}}=\frac{{\Omega_{\mathrm{ESR}}}^{2}}{2}\:\frac{{{V}_{\mathrm{ESR}}}}{{{{V}_{\mathrm{ESR}}}}^{2}+{{\tilde{\delta}_{\mathrm{ESR}}}}^{2}},\label{eqWeffESR}$$ which together with ${{{W}_{{\mathrm{L}}}}}$ eliminates ${\Omega_{{\mathrm{L}}}}$, ${{V}_{\mathrm{X}}}$, ${\delta_{{\mathrm{L}}}}$, ${\Omega_{\mathrm{ESR}}}$, ${{V}_{\mathrm{ESR}}}$, and ${\tilde{\delta}_{\mathrm{ESR}}}$ from the remaining equations for the diagonal elements of $\rho$. These now contain the effective spin-flip rates ${\tilde{W}_{{{\uparrow\downarrow}}}}={W_{{{\uparrow\downarrow}}}}+{{{W}_{\mathrm{ESR}}}}$ and ${\tilde{W}_{{{\downarrow\uparrow}}}}={W_{{{\downarrow\uparrow}}}}+{{{W}_{\mathrm{ESR}}}}$.
We find the stationary solution$$\begin{aligned}
{\bar{\rho}}_{\uparrow} & = & \eta\,{{{W}_{{\mathrm{L}}}}}\,{{W}_{\mathrm{em}}}\,{{W}_{{{X\uparrow},{X\downarrow}}}}+\eta\,{\tilde{W}_{{{\uparrow\downarrow}}}}\,{{W}_{\mathrm{em}}}\,{{W}_{{{X\uparrow},{X\downarrow}}}}\nonumber \\
& & \;+\eta\,{\tilde{W}_{{{\uparrow\downarrow}}}}\left({{{W}_{{\mathrm{L}}}}}+{{W}_{\mathrm{em}}}\right)\,\left({{W}_{\mathrm{em}}}+{{W}_{{{X\downarrow},{X\uparrow}}}}\right),\qquad\\
{\bar{\rho}}_{\downarrow} & = & \eta\,{\tilde{W}_{{{\downarrow\uparrow}}}}\,\left({{{W}_{{\mathrm{L}}}}}+{{W}_{\mathrm{em}}}\right)\left({{W}_{\mathrm{em}}}+{{W}_{{{X\downarrow},{X\uparrow}}}}\right)\nonumber \\
& & \;+\eta\,{\tilde{W}_{{{\downarrow\uparrow}}}}\,{{W}_{\mathrm{em}}}\,{{W}_{{{X\uparrow},{X\downarrow}}}},\\
{\bar{\rho}}_{{X\downarrow}} & = & \eta\,{{{W}_{{\mathrm{L}}}}}\,{\tilde{W}_{{{\downarrow\uparrow}}}}\,\left({{W}_{\mathrm{em}}}+{{W}_{{{X\downarrow},{X\uparrow}}}}\right),\\
{\bar{\rho}}_{{X\uparrow}} & = & \eta\,{{{W}_{{\mathrm{L}}}}}\,{\tilde{W}_{{{\downarrow\uparrow}}}}\,{{W}_{{{X\uparrow},{X\downarrow}}}},\end{aligned}$$ where the normalization factor $\eta$ is such that $\sum_{n}\rho_{n}=1$. Note that ${\bar{\rho}}_{\uparrow}\geq{\bar{\rho}}_{\downarrow}$ is satisfied for ${W_{{{\uparrow\downarrow}}}}\geq{W_{{{\downarrow\uparrow}}}}$. Thus, electron-spin polarization is achieved due to the hole-spin relaxation channel, analogous to an optical pumping scheme. Now, photons with $\sigma^{-}$ ($\sigma^{+}$) polarization are emitted from the dot at the rate ${\Gamma^{-}}={{W}_{\mathrm{em}}}{\bar{\rho}}_{{X\downarrow}}$ (${\Gamma^{+}}={{W}_{\mathrm{em}}}{\bar{\rho}}_{{X\uparrow}}$). These rates are proportional to ${{{W}_{\mathrm{ESR}}}}/(\gamma+{{{W}_{\mathrm{ESR}}}})$ for a given $\gamma$, up to a constant background which is negligible for ${W_{{{\downarrow\uparrow}}}}<{{{W}_{\mathrm{ESR}}}}$. In particular, the total rate $\Gamma={\Gamma^{-}}+{\Gamma^{+}}$ as a function of ${\tilde{\delta}_{\mathrm{ESR}}}$ is a Lorentzian with linewidth $$w=2\,{{V}_{\mathrm{ESR}}}\sqrt{1+\frac{{{W_{\mathrm{ESR}}^{\mathrm{max}}}}}{\gamma}};$$ see Fig. \[Fig: cwPL\]. Analyzing the expression for $\gamma$, we find the relevant parameter regime with the inequality $$\begin{aligned}
w & \leq & 2\,{{V}_{\mathrm{ESR}}}\,\Bigg[1+\frac{2\,{{W_{\mathrm{ESR}}^{\mathrm{max}}}}}{{{{W}_{{\mathrm{L}}}}}}\,\left(1+\frac{{{W}_{\mathrm{em}}}}{{{W}_{\mathrm{r}}}}+\frac{{{W}_{{{X\downarrow},{X\uparrow}}}}}{{{W}_{\mathrm{r}}}}\right)\nonumber \\
& & \quad+\frac{3\,{{W_{\mathrm{ESR}}^{\mathrm{max}}}}}{{{W}_{\mathrm{r}}}}+\frac{{{W_{\mathrm{ESR}}^{\mathrm{max}}}}}{{{W}_{\mathrm{em}}}}\,\left(1+\frac{3\,{{W}_{{{X\downarrow},{X\uparrow}}}}}{{{W}_{\mathrm{r}}}}\right)\Bigg]^{1/2}\!\!\!\!\!\!\!,\label{eqnLinewidth}\end{aligned}$$ which saturates for vanishing ${W_{{{\downarrow\uparrow}}}}$ and ${W_{{{\uparrow\downarrow}}}}$. Here, the rate ${{W}_{\mathrm{r}}}={{W}_{{{X\uparrow},{X\downarrow}}}}+{W_{{{\uparrow\downarrow}}}}\left(1+{{W}_{\mathrm{em}}}/{{{W}_{{\mathrm{L}}}}}\right)$ describes different relaxation channels, all leading to the ground state ${{|\! \uparrow\rangle}}$, and thus corresponds to “switching off” the laser excitations. If ${{W}_{\mathrm{r}}}$ is large, e.g., due to efficient hole-spin relaxation,[@Flissikowski] $w\approx2\,{{V}_{\mathrm{ESR}}}$. From the linewidth $w$ one can extract a *lower bound for* $T_{2}$: $T_{2}\geq1/{{V}_{\mathrm{ESR}}}\geq2/w$. Further, this lower bound saturates when the expression in brackets in Eq. (\[eqnLinewidth\]) becomes close to 1 and $T_{2}^{-1}\approx{{V}_{\mathrm{ESR}}}$ [\[]{}see Eq. (\[eqndecohESReff\])[\]]{}, i.e., the $T_{2}$ time is given by the linewidth. Comparing with the exact solution, we find that our analytical approximation gives the value of $\Gamma$ within $0.2\%$ for the parameters of Fig. \[Fig: cwPL\]. Due to possible imperfections in this ODMR scheme, e.g., mixing of hh and lh states or a small contribution of the $\sigma^{+}$ polarization in the laser light, also the state ${{|\! \uparrow\rangle}}$ can be optically excited. We describe this with the effective rate ${W}_{{\mathrm{L}},\uparrow}$ which leads to an additional linewidth broadening [\[]{}similar to Eq. (\[eqnLinewidth\])[\]]{}. This effect is small for ${W}_{{\mathrm{L}},\uparrow}<{{{W}_{\mathrm{ESR}}}}$. Detection of the laser stray light can be avoided by only measuring $\Gamma^{+}$. Otherwise, the laser could be distinguished from $\Gamma^{-}$ by using two-photon absorption. As an alternative, the optical excitation could be tuned to an excited hole state (hh or lh), possibly with a reversal of laser polarization. A *pulsed* laser, finally, would enable the distinction between luminescence and laser light by time gated detection.
Spin Rabi Oscillations via Photoluminescence
============================================
For a pulsed $\sigma^{-}$ laser, one can also measure $\Gamma$ as a function of the pulse repetition time ${\tau_{\mathrm{rep}}}$ instead of ${\tilde{\delta}_{\mathrm{ESR}}}$. We still use cw ESR (or, alternatively, a static transverse magnetic field, i.e., in the Voigt geometry). We stress that the same restrictions on the laser bandwidth as in the cw case apply. Due to hole spin flips, followed by emission of a photon, the dot is preferably in the state ${{|\! \uparrow\rangle}}$ rather than ${{|\! \downarrow\rangle}}$ at the end of a laser pulse. The magnetic field then acts on the electron spin until the next laser pulse arrives. Finally, the spin state ${{|\! \downarrow\rangle}}$ is read out optically and, therefore, the Rabi oscillations (or spin precessions) can be observed in the photoluminescence as function of ${\tau_{\mathrm{rep}}}$; see Fig. \[Fig: pulsedPL\]. For simplicity, we consider square pulses of length $\Delta t$. We write in the master equation ${\mathcal{M}}(t)={{\mathcal{M}}_{\mathrm{L}}}$ during a laser pulse and otherwise ${\mathcal{M}}(t)={{\mathcal{M}}_{0}}$, setting ${\Omega_{{\mathrm{L}}}}=0$. We find the steady-state density matrix ${\rho_{\infty}}$ of the dot just after the pulse with $U_{p}{\rho_{\infty}}={\rho_{\infty}}$, where $U_{p}=\exp({{\mathcal{M}}_{\mathrm{L}}}\Delta t)\exp[{{\mathcal{M}}_{0}}({\tau_{\mathrm{rep}}}-\Delta t)]$ describes the time evolution during ${\tau_{\mathrm{rep}}}$.
The photoluminescence rate is now evaluated by $\Gamma={{W}_{\mathrm{em}}}\overline{(\rho_{{X\downarrow}}+\rho_{{X\uparrow}})}$, where the bar designates time averaging over many periods ${\tau_{\mathrm{rep}}}$. For $\Delta t\geq\pi/{\Omega_{{\mathrm{L}}}},\,{W}_{\mathrm{em}}^{-1}$, the spin oscillations become more pronounced; see Fig. \[Fig: pulsedPL\] (b). This results from an enhanced relaxation to the state ${{|\! \uparrow\rangle}}$ during each pulse and thus from a much larger $\rho_{\uparrow}$ than $\rho_{\downarrow}$ just after the pulse.
Conclusions
===========
We have proposed an ODMR setup with ESR and polarized optical excitation. We have shown that this setup allows the optical measurement of the single-electron spin decoherence time $T_{2}$ in semiconductor quantum dots. The discussed cw and pulsed optical detection schemes can also be combined with pulsed instead of cw ESR, allowing spin echo and similar standard techniques. Such pulses can, e.g., be produced via the ac Stark effect.[@ACstark] Further, as an alternative to photoluminescence detection, photocurrent can be used to read out the charged exciton,[@abstreiter2] and the same ODMR scheme can be applied.
Acknowledgments {#acknowledgments .unnumbered}
===============
We thank J. C. Egues, A. V. Khaetskii, B. Hecht, and H. Schaefers for discussions. We acknowledge support from DARPA, ARO, NCCR Nanoscience, and the Swiss and US NSF.
[10]{} *Semiconductor Spintronics and Quantum Computation*, edited by D.D. Awschalom, D. Loss, and N. Samarth, *Series on Nanoscience and Technology*, Vol. XVI (Springer, New York, 2002). N.H. Bonadeo[, J. Erland, D. Gammon, , D. Park, D.S. Katzer and D.G. Steel]{}, Science **282**, 1473 (1998). Alternatively, one can measure $T_{2}$ via currents through quantum dots in an ESR field (Refs. 24 and 31). However, using an optical detection scheme there is no need for contacting dots with current leads (thus reducing decoherence) and one can benefit from the high sensitivity of photodetectors. J.A. Gupta[, D.D. Awschalom, X. Peng, and A.P. Alivisatos]{}, Phys. Rev. B **59**, R10 421 (1999). B. Ohnesorge[, M. Albrecht, J. Oshinowo, A. Forchel, and Y. Arakawa]{}, Phys. Rev. B **54**, 11 532 (1996). S. Raymond[, S. Fafard, P.J. Poole, A. Wojs, P. Hawrylak, S. Charbonneau, D. Leonard, R. Leon, P.M. Petroff, and J.L. Merz]{}, Phys. Rev. B **54**, 11 548 (1996). R.J. Epstein[, D.T. Fuchs, W.V. Schoenfeld, P.M. Petroff, and D.D. Awschalom]{}, Appl. Phys. Lett. **78**, 733 (2001). A. Gruber[, A. Dräbenstedt, C. Tietz, L. Fleury, J. Wrachtrup, and C. von Borczyskowski]{}, Science **276**, 2012 (1997); F. Jelezko[, T. Gaebel, I. Popa, A. Gruber, and J. Wrachtrup]{}, Phys. Rev. Lett. **92**, 076401 (2004); for a recent work on ensembles, see F.T. Charnock and T.A. Kennedy, Phys. Rev. B **64**, 041201 (2001). E. Lifshitz[, I. Dag, I.D. Litvitn, and G. Hodes]{}, J. Phys. Chem. B **102**, 9245 (1998). N. Zurauskiene[, G. Janssen, E. Goovaerts, A. Bouwen, D. Shoemaker, P.M. Koenraad, and J.H. Wolter]{}, Phys. status solidi B **224**, 551 (2001). For a recent discussion, see, e.g., T. Calarco[, A. Datta, P. Fedichev, E. Pazy, and P. Zoller]{}, Phys. Rev. A **68**, 012310 (2003). S. Cortez[, O. Krebs, S. Laurent, M. Senes, X. Marie, P. Voisin, R. Ferreira, G. Bastard, J.M. Gérard, and T. Amand]{}, Phys. Rev. Lett. **89**, 207401 (2002). M. Baier[, F. Findeis, A. Zrenner, M. Bichler, and G. Abstreiter]{}, Phys. Rev. B **64**, 195326 (2001). M. Bayer[, G. Ortner, O. Stern, A. Kuther, A.A. Gorbunov, A. Forchel, P. Hawrylak, S. Fafard, K. Hinzer, T.L. Reinecke, S.N. Walck, J.P. Reithmaier, F. Klopf, and F. Schäfer]{}, Phys. Rev. B **65**, 195315 (2002). J.J. Finley[, D.J. Mowbray, M.S. Skolnick, A.D. Ashmore, C. Baker, A.F.G. Monte, and M. Hopkinson]{}, Phys. Rev. B **66**, 153316 (2002). J.G. Tischler[, A.S. Bracker, D. Gammon, and D. Park]{}, Phys. Rev. B **66**, 081310 (2002). M. Fricke[, A. Lorke, J.P. Kotthaus, G. Medeiros-Ribeiro, and P.M. Petroff]{}, Europhys. Lett. **36**, 197 (1996). K.H. Schmidt[, G. Medeiros-Ribeiro, M. Oestreich, P.M. Petroff, and G.H. Döhler]{}, Phys. Rev. B **54**, 11 346 (1996). For dot shapes of lower than circular symmetry, mixing of different band states can become significant. See, e.g., L.-W. Wang[, J. Kim, and A. Zunger]{}, Phys. Rev. B **59**, 5678 (1999). Al.L. Efros, Phys. Rev. B **46**, 7448 (1992). Al.L. Efros and A.V. Rodina, Phys. Rev. B **47**, 10 005 (1993). In $B_{z}$, we have absorbed the Overhauser field which could possibly arise from dynamically polarized nuclear spins. A linearly oscillating magnetic field, $\mathbf{B}_{x}(t)=B_{x}^{0}\cos({\omega_{\mathrm{ESR}}}t)\mathbf{\hat{x}}$, can also be applied, cf. J.J. Sakurai, *Modern Quantum Mechanics,* (Addison-Wesley, Reading, MA, 1995), **Revised Edition. This leads, in the rotating wave approximation (RWA), to the same result as the rotating field for $B_{\perp}\rightarrow B_{x}^{0}/2$. H.-A. Engel and D. Loss, Phys. Rev. Lett. **86**, 4648 (2001); Phys. Rev. B **65**, 195321 (2002). Y. Kato[, R.C. Myers, D.C. Driscoll, A.C. Gossard, J. Levy, and D.D. Awschalom]{}, Science **299**, 1201 (2003). We use the standard description $H_{\mathrm{d}-\mathrm{L}}\!=\!-e\mathbf{A}\!\cdot\!\mathbf{p}/m_{0}c$, with vector potential $\mathbf{A}$ and charge $e$, bare mass $m_{0}$, and momentum $\mathbf{p}$ of the electron. Since excitonic electric fields are present, we neglect the typically smaller electric fields of the laser that are contained in second order in the term $\propto\mathbf{A}^{2}$. We apply the electric dipole approximation and assume a coherent photon state for the laser. Then, ${\Omega_{{\mathrm{L}}}}\!=\!\alpha e{{\langle{X_{\downarrow}^{-}}|}}\mathbf{e}_{\mathrm{L}}\!\cdot\!\mathbf{p}{{|\! \downarrow\rangle}}\sqrt{{2\hbar/m_{0}^{2}c^{2}V\epsilon_{0}\epsilon_{r}\omega_{\mathrm{L}}}}$, with volume $V$, unit polarization vector $\mathbf{e}_{\mathrm{L}}$, and mean photon number $|\alpha|^{2}$ of the laser mode. The relative permittivity of the semiconductor is $\epsilon_{r}$, and $\epsilon_{0}$ is the dielectric constant. See also L. Mandel and E. Wolf, *Optical Coherence and Quantum Optics* (Cambridge University Press, Cambridge, 1995). Excitation of a charged exciton state containing an electron triplet would require an additional energy of $\sim40$ meV (Ref. 12) and can be excluded by the low laser bandwidth. The single-spin relaxation **time $T_{1}$ can be measured via a similar double resonance scheme as discussed in this work. For this, we assume that the laser polarization is $\sigma^{+}$ and the temperature is sufficiently low that ${{|{X_{\downarrow}^{-}}\rangle}}$ is decoupled from the three-level system ${{|\! \uparrow\rangle}}$, ${{|\! \downarrow\rangle}}$, and ${{|{X_{\uparrow}^{-}}\rangle}}$, cf. Figs. \[fig:states\] and \[fig:transitions\]. In the regime ${\Omega_{{\mathrm{L}}}},\,{{W}_{\mathrm{em}}}\gg{\Omega_{\mathrm{ESR}}},\,{W_{{{\uparrow\downarrow}}}}$, the mean time of fluorescence interruptions due to ESR excitation provides $T_{1}$, similarly as for a single atom [\[]{}H.J. Kimble, R.J. Cook, and A.L. Wells, Phys. Rev. A **34**, 3190 (1986)[\]]{}. T. Flissikowski, I.A. Akimov, A. Hundt, and F. Henneberger, Phys. Rev. B **68**, R161309 (2003); L.M. Woods, T.L. Reinecke, and R. Kotlyar, Phys. Rev. B **69**, 125330 (2004). J.A. Gupta[, R. Knobel, N. Samarth, and D.D. Awschalom]{}, Science **292**, 2458 (2001); C.E. Pryor and M.E. Flatté (unpublished), quant-ph/0211160. I. Martin, D. Mozyrsky, and H.W. Jiang, Phys. Rev. Lett. **90**, 018301 (2003).
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'Van-Nham Phan'
- Holger Fehske
- 'Klaus W. Becker'
title: 'Excitonic resonances in the 2D extended Falicov-Kimball model'
---
Introduction
============
In solids, the Coulomb interaction binds conduction band electrons and valence band holes to excitons. Normally, excitonic quasiparticles do not form the ground state but electron-hole excitations that tend to decay on a very short time scale. At a semimetal-semiconductor transition, however, the conventional ground state of the crystal may become unstable with respect to a spontaneous formation of excitons, provided the overlap or band gap between the valence and conduction bands is small. Then, for low enough temperatures, these composite bosonic quasiparticles will condense into a macroscopic phase-coherent quantum state, thereby transforming the semimetallic or semiconducting configuration into an insulating one (cf. also fig. \[fig:PD\]). This so-called excitonic insulator (EI) state was theoretically proposed more than four decades ago [@Mo61; @Ko68]; for recent reviews see [@LEKMSS04; @MSGMDCMBTBA10]. The EI phase realized below the critical temperature $T_{EI}$ can be perceived either as BCS condensate in the semimetal region or as Bose Einstein condensate in the semiconductor region [@BF06].
The EI state is extremely rare in nature, so far there is no free of doubt realization in any material. At present, the most promising candidates are the quasi-2D transition-metal dichalgogenide $1T$-${\rm TiSe_2}$ and the pressure-sensitive mixed-valence rare-earth chalgogenide $\rm TmSe_{0.45}Te_{0.55}$. In $1T$-${\rm TiSe_2}$, the excitonic condensate exerts a force on the lattice generating periodic ionic displacements [@MBCAB11]. For $\rm TmSe_{0.45}Te_{0.55}$, Hall effect, thermal diffusivity and heat conductivity data give strong support for a Bose condensed state in the pressure range between 5 and 11 kbar below 20 K [@NW90]. The transport anomalies observed at higher temperatures, in particular the strange increase of the electrical resistivity in a narrow pressure range around 8 kbar might be attributed to a free-bound state scattering in an exciton-rich “halo” of an EI [@BF06]. As a basic prerequisite for the validity of this scenario, the existence of free excitons above the EI phase has to be proven, at least for the semiconducting region.
The aim of this paper is to address this issue. The investigation of Falicov-Kimball-type models seems to be minimal in this respect. The original Falicov-Kimball model describes localized $f$ electrons and itinerant $c$ electrons interacting by an on-site Coulomb interaction [@FK69]. For our problem, we have to allow for a possible coherence between conduction band electrons and valence band holes however. This can be achieved either by including an explicit $c$-$f$ hybridization [@KMM76] or a finite $f$ bandwidth [@Ba02]. Indeed, using constrained path Monte Carlo [@BGBL04] and mean-field [@Fa08] techniques, the 2D Falicov-Kimball model with direct $f$-$f$ particle hopping has been shown to exhibit an excitonic ground state for intermediate Coulomb couplings provided that the center of the $c$ and $f$ bands, $\varepsilon^c$ and $\varepsilon^f$, energetically differ. Note that around the symmetric case $\varepsilon^c=\varepsilon^f$ a charge-density-wave phase is energetically more stable [@Ba02; @BGBL04; @Fa08]. Recent Hartree-Fock [@SC08], RPA [@IPBBF08] and slave-boson [@Br08] studies confirm this finding also for the 3D case. In this paper we use the projector-based renormalization method (PRM) [@BHS02; @PBF10] to calculate directly the excitonic pair susceptibility (up to second order in $U$). Analyzing the non-trivial frequency- and momentum-dependence of $\chi({\bf q},\omega)$ we are able to address the problem of exciton formation and condensation.
Theoretical approach
====================
In order to model the generic situation of semiconductors or semimetals with short-ranged attractive Coulomb interaction between conduction band ($c$) electrons and valence band ($f$) holes we consider an extended version of the Falicov-Kimball Hamiltonian (EFKM), $$\label{Hami}
\mathcal{H}=\sum_{\mathbf{k}}\bar{\varepsilon}^c_{\mathbf{k}}c^\dagger_{\mathbf{k}}c^{}_{\mathbf{k}}
+\sum_{\mathbf{k}}\bar{\varepsilon}^f_{\mathbf{k}}f^\dagger_{\mathbf{k}}f^{}_{\mathbf{k}}
+U\sum_{i}n^c_in^f_i\,,$$ with two dispersive tight-binding bands $\bar{\varepsilon}^{c,f}_{\mathbf{k}}=\varepsilon_{}^{c,f}-t^{c,f}_{}
\gamma^{}_{\mathbf k}-\mu\,$. Here $\varepsilon_{}^{c,f}$ are the on-site energies, $t_{}^{c,f}$ are the nearest-neighbor particle transfer amplitudes, $\gamma^{}_{\mathbf k}=2\sum_d^D \cos k_d$ for a D-dimensional hypercubic lattice, and $\mu$ denotes the chemical potential. Accordingly the fermionic operators $c^{(\dagger)}_{\mathbf{k}}$ and $f^{(\dagger)}_{\mathbf{k}}$ annihilate (create) spinless $c$ and $f$ electrons with momentum $\mathbf{k}$, respectively, and $n^c_i$ and $n^f_i$ are the corresponding particle number operators for Wannier site $i$. $U$ parametrizes the local Hubbard attraction. Note that if the $c$ and $f$ bands are degenerate, $\varepsilon^{c}=\varepsilon^{f}$ and $t^{c}=t^{f}$, the EFKM reduces to the standard Hubbard model [@Hu63], whereas for $t^{f}=0$ the genuine FKM arises [@FK69]. In the latter case the local $f$ electron number is strictly conserved [@SC08]. In what follows, we study the half-filled band case, with total electron density $\langle n\rangle =\langle n^c_i\rangle+\langle n^f_i\rangle=1$. Moreover we consider a direct band gap situation with the maximum (minimum) of the $c$ ($f$) band dispersion located at $(\pi,\pi)$, i.e. $t^f<0$. Without loss of generality the $c$ electrons are considered to be ‘light’ while the $f$ electrons are ‘heavy’, i.e. $|t^f|<1$, where the $c$ electron hopping integral is taken to be the unit of energy, $t^c=1$, and $\varepsilon^c=0$.
The projector-based renormalization approach starts from the decomposition of the many-particle Hamiltonian (\[Hami\]) into an “unperturbed” part $\mathcal{H}_0$ ($c$ and $f$ electron band terms) and into a ‘perturbation’ $\mathcal{H}_1$ (Coulomb interaction term), where the unperturbed part $\mathcal{H}_0$ clearly is solvable. Then, in general, $\mathcal{H}_1$ accounts for all transitions between the eigenstates of $\mathcal{H}_0$ with nonzero transition energies. Using a series of unitary transformations to integrate out the perturbation ${\mathcal H}_1$ (for details see Ref. [@BHS02]) one arrives at a final Hamiltonian which is diagonal or at least quasi-diagonal. To evaluate the expectation value $\langle {\mathcal A}\rangle$ of any operator ${\mathcal A}$ also the operator has to be transformed by the same unitary transformation. One of the main advantages of the method is to find broken symmetry solutions of phase transitions [@SB09]. Note that for practical applications the unitary transformations should best be done in small steps in energy. Therefore, the evaluation of the transformation in each small step can be restricted to low orders in ${\cal H}_1$. This procedure usually limits the validity of the renormalization approach to parameters values of $\mathcal H_1$ which are of the order of those of $\mathcal H_0$. In the present case, good agreement with exact results is expected for $U$ values smaller than $t^{c,f}$ or $|\varepsilon^f_{\mathbf k} -\varepsilon^c_{\mathbf k}|$ (cf. eq. (\[Hartreeshift\]) below).
Excitonic insulator phase
=========================
In a first step, let us address the formation of the long-range ordered EI state in the 2D EFKM. To this end we look for a non-vanishing excitonic expectation value $\langle c^\dagger f^{}\rangle$, indicating a spontaneous symmetry breaking due the pairing of $c$ electrons ($t^c>0$) with $f$ holes ($t^f <0$). Employing the normal-ordered representation of fermionic operators $(: \ldots :)$, the Hamiltonian (\[Hami\]) reads $\mathcal H = \mathcal H_0 +\mathcal H_1$ with $$\begin{aligned}
\label{HFourierNO}
\mathcal{H}_0&=&\sum_{\mathbf{k}}\varepsilon^c_{\mathbf{k}}:c^\dagger_{\mathbf{k}}c^{}_{\mathbf{k}}:
+\sum_{\mathbf{k}}\varepsilon^f_{\mathbf{k}}:f^\dagger_{\mathbf{k}}f^{}_{\mathbf{k}}: \nonumber \\
&&-\sum_{\mathbf{k}}\left(\mit{\Delta}:f^\dagger_{\mathbf{k}}c^{}_{\mathbf{k}}:+\textrm{H.c.}\right) \\
\mathcal H_1 &=&
\frac{U}{N}\sum_{\mathbf{k}_1\mathbf{k}_2\mathbf{k}_3}
:a^{}_{\mathbf{k}_1\mathbf{k}_2\mathbf{k}_3}:\,,\nonumber\end{aligned}$$ where $$\mit{\Delta} = \frac{U}{N}\sum_{\mathbf{k}}{d}^{}_{\mathbf{k}} \,.
\label{Delta}$$ Here ${d}^{}_{\mathbf{k}}=\langle c^\dagger_{\mathbf{k}}f^{}_{\mathbf{k}}\rangle$ plays the role of the EI order parameter, and $a^{}_{\mathbf{k}_1\mathbf{k}_2\mathbf{k}_3}=c^\dagger_{\mathbf{k}_1}
c^{}_{\mathbf{k}_2}
f^\dagger_{\mathbf{k}_3}f^{}_{\mathbf{k}_1+\mathbf{k}_3-\mathbf{k}_2}$. Note that in $\mathcal H_0$, the on-site energies are shifted by a Hartree term, $$\varepsilon^{c,f}_{\mathbf{k}}= \bar{\varepsilon}^{c,f}_{\mathbf{k}}
+U\langle n^{f,c}\rangle\,,
\label{Hartreeshift}$$ where $\langle n^c\rangle=\tfrac{1}{N}
\sum_{\mathbf{k}} \langle c^\dagger_{\mathbf{k}}c^{}_{\mathbf{k}}\rangle$ and $\langle n^f\rangle=\tfrac{1}{N}
\sum_{\mathbf{k}} \langle f^\dagger_{\mathbf{k}}f^{}_{\mathbf{k}}\rangle$ are the mean particle number densities of $c$ and $f$ electrons for a system with $N$ lattice sites. Thus, $\mathcal H_0$ alone corresponds to the Hartree Hamiltonian. Vice versa, only fluctuation operators from the $U$ term contribute to ${\mathcal H}_1$. Evaluating the expectation values $\langle \ldots \rangle$, the temperature enters into the calculation via the Fermi function, see [@PBF10].
Following the procedure of the PRM approach [@PBF10] by integrating out all transitions due to $\mathcal H_1$, the Hamiltonian $\cal H$ can be transformed to a fully renormalized Hamiltonian $$\begin{aligned}
\label{ReH}
\tilde{\mathcal{H}}&=&
\sum_{\mathbf{k}}\tilde{\varepsilon}^c_{\mathbf{k}}:c^\dagger_{\mathbf{k}}c^{}_{\mathbf{k}}:
+\sum_{\mathbf{k}}\tilde{\varepsilon}^f_{\mathbf{k}}:f^\dagger_{\mathbf{k}}f^{}_{\mathbf{k}}:\nonumber\\
&&+\sum_{\mathbf{k}}(\tilde{\Delta}^{}_{\mathbf{k}}:f^\dagger_{\mathbf{k}}c^{}_{\mathbf{k}}:+\textrm{H.c.}),\end{aligned}$$ with modified parameters $\tilde{\varepsilon}^c_{\mathbf{k}}$, $\tilde{\varepsilon}^f_{\mathbf{k}}$, and $\tilde \Delta_{\mathbf k}$. Note that they take important correlation effects into account, which enter from the elimination procedure.
Self-evidently the single particle operators have to be transformed in order to evaluate expectation values, i.e. $$\begin{aligned}
:\tilde{c}^\dagger_{\mathbf{k}}:\!\!&=&\!\!\tilde{x}^{}_{\mathbf{k}}:c^\dagger_{\mathbf{k}}:+\frac{U}{N}\sum_{\mathbf{k}_1\mathbf{k}_2}
\tilde{y}^{}_{\mathbf{k}_1\mathbf{k}\mathbf{k}_2}:c^\dagger_{\mathbf{k}_1}
f^\dagger_{\mathbf{k}_2}f^{}_{\mathbf{k}_1+\mathbf{k}_2-\mathbf{k}}:\\
:\tilde{f}^\dagger_{\mathbf{k}}:\!\!&=&\!\!\tilde{x}'_{\mathbf{k}}:f^\dagger_{\mathbf{k}}:\nonumber+\frac{U}{N}\sum_{\mathbf{k}_1\mathbf{k}_2}
\tilde{y}'_{\mathbf{k}_1\mathbf{k}_2,\mathbf{k}-\mathbf{k}_1+\mathbf{k}_2}\nonumber\\&&\hspace*{3cm}\times
:c^\dagger_{\mathbf{k}_1}c^{}_{\mathbf{k}_2}f^\dagger_{\mathbf{k}-\mathbf{k}_1+\mathbf{k}_2}:\,,
\label{ren_eq}\end{aligned}$$ where ${\tilde x}_{\mathbf k}$, ${\tilde y}_{\mathbf k}$, $\cdots$ are also renormalized parameters. In the PRM the renormalization results from integrating difference equations with initial conditions taken over from the original Hamiltonian and the original single particle operators: $\varepsilon^{c(f)}_{\mathbf{k},\Lambda}=\varepsilon^{c(f)}_{\mathbf{k}}$, $\Delta^{}_{\mathbf{k},\Lambda}= 0^+$, $x^{}_{\mathbf{k},\Lambda}(x'_{\mathbf{k},\Lambda})=1$, and $y_{\mathbf{k}_1\mathbf{k}_2\mathbf{k}_3,\Lambda}(y'_{\mathbf{k}_1\mathbf{k}_2\mathbf{k}_3,\Lambda})=0$.
The final Hamiltonian can be diagonalized by a Bogoliubov transformation $$\begin{aligned}
\label{HFanoRe}
{\tilde{\mathcal H}}
=\sum_{{\mathbf k}}E^c_{{\mathbf k}}:c^\dagger_{{\mathbf k}}c^{}_{{\mathbf k}}:
+\sum_{{\mathbf k}}E^f_{{\mathbf k}}:f^\dagger_{{\mathbf k}}f^{}_{{\mathbf k}}:
+\tilde{E}\,,\end{aligned}$$ where the quasiparticle energies are given by $$\label{qpe_c}
E^{c/f}_{\mathbf k}=\frac{\tilde{\varepsilon}^c_{\mathbf k}+\tilde{\varepsilon}^f_{\mathbf k}}{2}
\mp\frac{\textrm{sgn}(\tilde{\varepsilon}^f_{\mathbf k}-\tilde{\varepsilon}^c_{\mathbf k})}
{2}W_{\mathbf k}$$ with $W_{\mathbf k}=[(\tilde{\varepsilon}^c_{\mathbf k}-\tilde{\varepsilon}^f_{\mathbf k})^2
+4|\tilde{\Delta}^{}_{\mathbf k}|^2]^{1/2}$.
A finite $\tilde{\Delta}^{}_{\mathbf k}$ signals $c$–$f$ electron coherence connected with a band gap that stabilizes the EI phase. Outside the EI phase, where $\tilde \Delta_{\mathbf k}=0$, a band gap may also exist, provided that $$E_g^{(0)}= E_{\mathbf 0}^c - E_{\mathbf 0}^f = \tilde{\varepsilon}^c_{\mathbf 0}-\tilde{\varepsilon}^f_{\mathbf 0}\, .
\label{Eg0}$$ Therefore $E_g^{(0)}<0$ ($E_g^{(0)}>0$) may be taken as indication that the system is in the semimetallic (semiconducting) regime (cf. fig. \[fig:PD\]). Let us emphasize that $E_g^{(0)}$ contains the fully renormalized quasiparticle energies $\tilde{\varepsilon}^{c,f}$, not just the Hartree energies $\varepsilon^{c,f}$ given by eq. .
For the 2D tight-binding band case studied in this paper, we work on a discrete set of $N=24\times 24$ lattice sites and determine all quantities with a relative error of less than $10^{-5}$.
Semimetallic region
-------------------
Figure \[fig:Ek\] (upper panel) shows the renormalized quasiparticle bands $E_{\mathbf k}^c$ and $E_{\mathbf k}^f$ along the high-symmetry axes of the 2D Brillouin zone, for $\varepsilon^f=-1$, $t^f=-0.3$, and $U=2$. In this case, both bands overlap ($E_g^{(0)} <0 $) leading to a large Fermi surface, where both types of quasiparticles participate. At low temperatures a gap opens at the Fermi surface due to the formation of an excitonic insulating state. Such an EI state has been viewed before as a BCS condensate of loosely bound electron-hole pairs [@BF06]. Increasing the temperature above some critical temperature $T_{EI}$ the gap vanishes. At this temperature the EI-semimetal transition takes place. The lower panel of fig. \[fig:Ek\] displays the order parameter function $d_{\mathbf k}$ as a function of $\mathbf k$. For low temperatures and $\bf k$ close to the Fermi surface, where both quasiparticle bands overlap, $d_{\mathbf k}$ is strongly peaked. Otherwise $d_{\mathbf k}$ is a rather smooth function of $\mathbf k$. As a matter of course, increasing $T$ above $T_{EI}$, the order parameter function $d_{\mathbf k}$ vanishes.
In fig. \[fig:DT\] \[panel (a)\] the EI order parameter $\Delta$ \[eq. \] is shown as a function of temperature for various values of $\varepsilon^f$ at $U=2$. Clearly seen is the formation of an EI state with nonzero $\Delta$ at low temperatures. The EI state is weakened by lowering $\varepsilon^f$ since the overlap of $c$ and $f$ electron bands is reduced in this case. In panel (b), the temperature-dependence of $\Delta$ is illustrated for various values of the Coulomb interaction $U$. Similar as before, the formation of an EI state is observed for low $T$. The EI region is decimated by lowering the $c$ electron $f$ hole attraction. One can assure oneself that the EI phase only appears in between some lower critical value $U_{c1}$ and some upper critical value $U_{c2}$ (on the semiconductor side, see below). In both panels of fig. \[fig:DT\], the solid blue lines give the variation of the negative quasiparticle gap $E_g^{(0)}$ with $T$, where $\varepsilon^f =-2.5$ in panel (a) and $U= 2.4$ in (b). In either case, the variation of $E_g^{(0)}$ for small $T$ towards lower values goes along with the formation of the EI state. For $U < U_{c1}$ the bare band splitting $\varepsilon^f-\varepsilon^c$ is somewhat reduced but $E^{(0)}_g$ is still negative, so we end up with a semimetallic situation.
Semiconducting region
---------------------
We now discuss the possible appearance of an EI state on the semiconductor-side of the schematic phase diagram shown in fig. \[fig:PD\]. Figure \[fig:ED\_T\] displays the temperature dependence of the quasiparticle gap $E_g^{(0)}$ and the order parameter $\Delta$ for some larger values of $U$ than before. The order parameter $\Delta$ is finite and negative at low $T$ which again signals the existence of an EI phase. The EI phase in this region was interpreted as a BEC of preformed tightly bound excitons [@BF06].
For $T$ above the critical temperature $T_{EI}$ the order parameter vanishes and no broken-symmetry state exists as in the semimetallic case. Note that $E_g^{(0)}$ is now positive at low temperatures, which indicates that we have a situation with a semiconductor-like band structure at least up to some critical temperature $T_{MI}$, where the band gap closes. Obviously, at $T_{MI}$, a semiconductor-semimetal metal-insulator transition occurs. We further note that due to the inclusion of correlation effects the PRM metal-insulator transition temperature $T_{MI}$ strongly deviates from $T_{MI}^H$ obtained by using the Hartree-shifted bare energies $\varepsilon_{\bf k}^{c,f}$ only, where $E_g^{(0),H}= E_{\mathbf 0}^c - E_{\mathbf 0}^f = \varepsilon^c_{\mathbf 0}-\varepsilon^f_{\mathbf 0}\, $ marks the corresponding Hartree band gap. The essential question whether exitonic bound states might possibly exist in the temperature region $T_{EI} < T < T_{MI}$ will be investigated below. In this connection, in ref. [@BF06] the authors proposed a so-called “halo phase”, where individual valence bond holes, conduction band electrons, and bound (but uncondensed) electron-hole pairs (excitons) should coexist.
Excitonic resonances
====================
To address the possible formation of excitonic bound states above $T_{EI}$, we analyze the frequency and momentum dependence of the dynamical excitonic susceptibility $$\chi({\mathbf q},\omega)=\langle\langle b^{}_{\mathbf{q}};b^{\dagger}_{\mathbf{q}} \rangle\rangle^{}(\omega)\,,
\label{suszi1}$$ where the symbol $\langle\langle \ldots\rangle\rangle$ denotes the retarded Green’s function, and the creation operator of an electron-hole excitation with momentum $\mathbf q$ is defined by $b^{\dagger}_{\mathbf{q}}=\frac{1}{\sqrt{N}}\sum_{\mathbf k}c^\dagger_{\mathbf{k}+\mathbf{q}}f^{}_{\mathbf k}$. Using the unitary invariance of expectation values, Eq. can be rewritten as $$\chi({\mathbf q},\omega)=\frac{1}{N}\sum_{\mathbf{k}\mathbf{k}'}\langle\langle \tilde{f}^{\dagger}_{\mathbf k}\tilde{c}^{}_{\mathbf{k}+\mathbf{q}};\tilde{c}^\dagger_{\mathbf{k}'+\mathbf{q}}\tilde{f}^{}_{\mathbf{k}'}\rangle\rangle^{}_{\tilde{\mathcal H}}(\omega)\,.
\label{suszi2}$$ Here the two-particle Green’s function on the right-hand side is formed with $\tilde{\mathcal H}$ and the quantities with tilde symbols are the fully renormalized operators. Taking into account that the EI order parameter vanishes for $T>T_{EI}$ ($\tilde \Delta_{\mathbf{k}}=0$ $\forall \mathbf{k}$), we obtain up to order ${\cal O}(U^2)$ $$\begin{aligned}
\label{chi1}
\chi(\mathbf{q}&,\omega)=\frac{1}{N}\sum_{\mathbf{k}}\frac{\Gamma^{0}_{\mathbf{kq}}}{\omega-\omega^{}_{\mathbf k}(\mathbf q) + i\eta}\nonumber\\
+&\frac{1}{N^3}\sum_{\mathbf{k}\mathbf{k}_1\mathbf{k}_2}
\Big[\frac{\Gamma^{1}_{\mathbf{kq}\mathbf{k}_1\mathbf{k}_2}}{\omega-E^{(1)}_{\mathbf{kq}\mathbf{k}_1\mathbf{k}_2}+i\eta}-\frac{\Gamma^{2}_{\mathbf{kq}\mathbf{k}_1\mathbf{k}_2}}{\omega-E^{(2)}_{\mathbf{kq}\mathbf{k}_1\mathbf{k}_2}+i\eta}\Big]\\end{aligned}$$ with $$\begin{aligned}
\label{phe}
\omega^{}_{\mathbf k}(\mathbf q)&=&E^c_{\mathbf{k}+\mathbf{q}}-E^f_{{\mathbf k}}\,,
\\\label{er1}
E^{(1)}_{\mathbf{kq}\mathbf{k}_1\mathbf{k}_2}&=&E^c_{\mathbf{k}_1}-E^f_{\mathbf{k}}
-E^f_{{\mathbf k}_1+\mathbf{k}_2-\mathbf{k}-\mathbf{q}}+E^f_{\mathbf{k}_2}\,,\\\label{er2}
E^{(2)}_{\mathbf{kq}\mathbf{k}_1\mathbf{k}_2}&=&E^c_{\mathbf{k}+{\mathbf q}}-E^c_{\mathbf{k}_1}+E^c_{{\mathbf k}_2}-E^f_{\mathbf{k}-\mathbf{k}_1+\mathbf{k}_2}\,,\end{aligned}$$ and $\eta=0^+$. The coefficients $\Gamma^i$ are given by $$\begin{aligned}
\label{GammaX}
\Gamma^0_{\mathbf{kq}}=&\Big\{|\tilde{x}'_{\mathbf{k}}\tilde{x}^{}_{\mathbf{k}+\mathbf{q}}|^2\nonumber\\
-&\frac{2U}{N}\sum_{\mathbf{k}_1}\tilde{x}'_{\mathbf{k}}\tilde{x}^{}_{\mathbf{k}+\mathbf{q}}(\tilde{x}_{\mathbf{k}_1+\mathbf{q}}\tilde{y}'_{\mathbf{k}_1+\mathbf{q},\mathbf{k}+\mathbf{q},\mathbf{k}}\langle \tilde{n}^c_{\mathbf{k}_1+\mathbf{q}}\rangle\nonumber\\
&\quad\quad\quad+\tilde{x}'_{\mathbf{k}_1}\tilde{y}_{\mathbf{k}+\mathbf{q},\mathbf{k}_1+\mathbf{q},\mathbf{k}_1}\langle \tilde{n}^f_{\mathbf{k}_1}\rangle) \nonumber\\
+&\frac{U^2}{N^2}\sum_{\mathbf{k}_1\mathbf{k}_2}\Big[2\tilde{x}'_{\mathbf{k}}\tilde{x}^{}_{\mathbf{k}+\mathbf{q}}\tilde{y}_{\mathbf{k}_2,\mathbf{k}_1+\mathbf{q},\mathbf{k}_1-\mathbf{k}_2+\mathbf{k}+\mathbf{q}}\nonumber\\
&\times\tilde{y}'_{\mathbf{k}_2,\mathbf{k}+\mathbf{q},\mathbf{k}_1-\mathbf{k}_2+\mathbf{k}+\mathbf{q}}\langle \tilde{n}^c_{\mathbf{k}_2}\rangle\langle \tilde{n}^f_{\mathbf{k}_1-\mathbf{k}_2+\mathbf{k}+\mathbf{q}}\rangle\nonumber\\
&+2\tilde{x}'_{\mathbf{k}_1}\tilde{x}^{}_{\mathbf{k}_2+\mathbf{q}}\tilde{y}_{\mathbf{k}_2,\mathbf{k}_1+\mathbf{q},\mathbf{k}_1}\tilde{y}'_{\mathbf{k}_2+\mathbf{q},\mathbf{k}+\mathbf{q},\mathbf{k}}\langle \tilde{n}^f_{\mathbf{k}_1}\rangle\langle \tilde{n}^c_{\mathbf{k}+\mathbf{q}}\rangle\nonumber\\
&+\tilde{x}_{\mathbf{k}_1+\mathbf{q}}\tilde{x}^{}_{\mathbf{k}_2+\mathbf{q}}
\tilde{y}'_{\mathbf{k}_2+\mathbf{q},\mathbf{k}+\mathbf{q},\mathbf{k}}\tilde{y}'_{\mathbf{k}_1+\mathbf{q},\mathbf{k}+\mathbf{q},\mathbf{k}}\nonumber\\
&\times\langle \tilde{n}^c_{\mathbf{k}_1+\mathbf{q}}\rangle\langle \tilde{n}^c_{\mathbf{k}_2+\mathbf{q}}\rangle\nonumber\\
&+\tilde{x}'_{\mathbf{k}_1}\tilde{x}'^{}_{\mathbf{k}_2}
\tilde{y}_{\mathbf{k}+\mathbf{q},\mathbf{k}_2+\mathbf{q},\mathbf{k}_2}\tilde{y}_{\mathbf{k}+\mathbf{q},\mathbf{k}_1+\mathbf{q},\mathbf{k}_1}\langle \tilde{n}^f_{\mathbf{k}_1}\rangle\langle \tilde{n}^f_{\mathbf{k}_2}\rangle\Big]\Big\}\nonumber\\
&\times(\langle \tilde{n}^f_{\mathbf{k}}\rangle-\langle \tilde{n}^c_{\mathbf{k}+\mathbf{q}}\rangle)\,,\end{aligned}$$ $$\begin{aligned}
\Gamma^{1}_{\mathbf{kq}\mathbf{k}_1\mathbf{k}_2}=&U^2(|\tilde{x}'_{\mathbf{k}}\tilde{y}_{\mathbf{k}_1,\mathbf{k}+\mathbf{q},\mathbf{k}_2}|^2\nonumber\\
&-\tilde{x}'_{\mathbf{k}}\tilde{x}'_{\mathbf{k}_1+\mathbf{k}_2-\mathbf{k}-\mathbf{q}}
\tilde{y}_{\mathbf{k}_1,\mathbf{k}+\mathbf{q},\mathbf{k}_2}\tilde{y}_{\mathbf{k}_1,\mathbf{k}_1+\mathbf{k}_2-\mathbf{k},\mathbf{k}_2})\nonumber\\
\times &\Big[\langle\tilde{n}^f_{\mathbf{k}}\rangle\langle
\tilde{n}^f_{\mathbf{k}_1+\mathbf{k}_2-\mathbf{k}-\mathbf{q}}\rangle(1-\langle\tilde{n}^c_{\mathbf{k}_1}\rangle
-\langle\tilde{n}^f_{\mathbf{k}_2}\rangle)\nonumber\\
&-\langle\tilde{n}^f_{\mathbf{k}_2}\rangle
\langle\tilde{n}^c_{\mathbf{k}_1}\rangle(1-\langle\tilde{n}^f_{\mathbf{k}}\rangle-
\langle\tilde{n}^f_{\mathbf{k}_1+\mathbf{k}_2-\mathbf{k}-\mathbf{q}}\rangle)\Big]\,,\end{aligned}$$ $$\begin{aligned}
\Gamma^{2}_{\mathbf{kq}\mathbf{k}_1\mathbf{k}_2}
=&U^2(|\tilde{x}_{\mathbf{k}+\mathbf{q}}\tilde{y}'_{\mathbf{k}_1,\mathbf{k}_2,\mathbf{k}-\mathbf{k}_1+\mathbf{k}_2}|^2\nonumber\\
&-\tilde{x}_{\mathbf{k}+\mathbf{q}}
\tilde{x}_{\mathbf{k}_2-\mathbf{q}}\tilde{y}'_{\mathbf{k}_1\mathbf{k}_2,\mathbf{k}-\mathbf{k}_1+\mathbf{k}_2}\tilde{y}'_{\mathbf{k}_1,\mathbf{k}+\mathbf{q},\mathbf{k}-\mathbf{k}_1+\mathbf{k}_2})\nonumber\\
\times &
\Big[\langle\tilde{n}^c_{\mathbf{k}_2}\rangle\langle\tilde{n}^c_{\mathbf{k}+\mathbf{q}}\rangle
(1-\langle\tilde{n}^f_{\mathbf{k}-\mathbf{k}_1+\mathbf{k}_2}\rangle-\langle\tilde{n}^c_{\mathbf{k}_1}\rangle)\nonumber\\
&-\langle\tilde{n}^c_{\mathbf{k}_1}\rangle
\langle\tilde{n}^f_{\mathbf{k}-\mathbf{k}_1+\mathbf{k}_2}\rangle
(1-\langle\tilde{n}^c_{\mathbf{k}_2}\rangle-\langle\tilde{n}^c_{\mathbf{k}+\mathbf{q}}\rangle)\Big] \, .\end{aligned}$$ Here the expectation values $\langle \tilde{n}_{\mathbf k}^c\rangle = \langle c_{\mathbf k}^\dagger c^{}_{\mathbf k}\rangle_{\tilde{\mathcal{H}}}$ and $\langle \tilde{n}_{\mathbf k}^f\rangle = \langle f_{\mathbf k}^\dagger f^{}_{\mathbf k}\rangle_{\tilde{\mathcal{H}}}$, are formed with the renormalized Hamiltonian $\tilde{\cal H}$ and can easily be evaluated due to the diagonal form of $\tilde{\mathcal H}$. Note that the pole structure of the first (coherent) term of Eq. describes the continuum of particle-hole excitations. Of course, we have $\omega_{\bf 0}({\bf 0})=E_g^{(0)}$, and in view of the form of the $c$ and $f$ band dispersions $\omega_{\bf k}({\bf 0}) > \omega_{\bf 0}({\bf 0})\;\forall {\bf k}\neq {\bf 0}$. Therefore the possibility of ${\bf q}={\bf 0}$ excitations with positive energy indicates that the system is in the semiconducting regime. If one tries to determine the semimetal-semiconductor boundary, i.e., $T_{MI}$, from the pole structure of $\chi({\bf q},\omega)$ this assertion is valid to leading order only; the second and third term of might lead to a shift of the lowest excitation energy in the ${\bf q}={\bf 0}$ sector. As shown below, this effect is negligible however: the values of $T_{MI}$ derived from $E_g^{(0)}$ are in accord with the results obtained from the dynamical susceptibility.
The imaginary part of $\chi({\mathbf q},\omega)$ reads $$\begin{aligned}
\label{Imchi}
-\frac{1}{\pi}\textrm{Im}\chi(\mathbf{q},\omega)=&\frac{1}{N}\sum_{\mathbf{k}}\Gamma^{0}_{\mathbf{kq}}\delta[\omega-\omega^{}_{\mathbf k}(\mathbf q)]\nonumber\\
\hspace*{-2.3cm}&+\frac{1}{N^3}\sum_{\mathbf{k}\mathbf{k}_1\mathbf{k}_2}
[\Gamma^{1}_{\mathbf{kq}\mathbf{k}_1\mathbf{k}_2}\delta(\omega-E^{(1)}_{\mathbf{kq}\mathbf{k}_1\mathbf{k}_2})
\nonumber\\
&\hspace*{1.2cm}
-\Gamma^{2}_{\mathbf{kq}\mathbf{k}_1\mathbf{k}_2}
\delta(\omega-E^{(2)}_{\mathbf{kq}\mathbf{k}_1\mathbf{k}_2})],\end{aligned}$$ where possible non-zero excitations outside the particle-hole continuum point to the existence of excitonic resonances.
Figure \[fig:S00P0\] diplays our numerical results for the imaginary part of the exitonic susceptibility $-\textrm{Im}\chi(\mathbf{q},\omega+i0)/\pi$ as a function of $\omega$ for different momenta $(q_x,0)$ between $q_x=0$ and $\pi$ (black solid lines), indicated by the scale attached to the ordinate axis. In all panels the red dashed line represents the boundaries to the particle-hole continuum (indicated by the strong upturns). The small bumps in the figures correspond to excitonic resonances, where the maximum most closely located to the particle-hole continuum refers to an excitonic bound state.
In panels (a) to (d) the temperature is kept to $T=0.08$. For $U=2.5$ respectively $2.85$ (which are still larger than $U_{c1}$ however), $T>T_{MI}$, and the system is in the semimetallic region. No excitonic resonances can be found for ${\bf q}=0$ in this case, since $E_g^{(0)}$ is already negative (cf. fig. \[fig:ED\_T\]). There is a significant increase of the spectral weight of the finite-${\bf q}$ excitonic resonances by going over from $U=2.5$ to $U=2.85$. In contrast, in panel (c), where $U=2.9$, we have $T_{EI} <T<T_{MI}$, and the system realizes a semiconductor (cf. figs. \[fig:PD\] and \[fig:ED\_T\]). Now a weak excitonic resonance is found at momentum ${\mathbf q}=0$ which can more clearly be seen from the inset. For $U=3.0>U_{c2}$ \[panel (d)\] the EI phase is not realized even for $T=0$ (cf. fig. \[fig:ED\_T\]), and again excitons can be formed for all values of $q_x$. Note that excitons with finite momentum ${\mathbf q}$ can be created in both semiconductor and semimetal cases \[cf. panels (a) to (d)\]. Their resonance positions follow from eqs. and .
In the two lowermost panels (e) and (f) the Coulomb interaction is fixed to $U=2.9$. Increasing the temperature from $T=0.12$ (e) to $T=0.16$ (f) the system passes the semiconductor-semimetal transition. Although, at $T=0.12$, the system is very close to the transition point (cf. fig. \[fig:ED\_T\]), excitons with zero momenta may form (see inset). In contrast, no ${\mathbf q}=0$ excitons can exist for the temperature considered in panel (f). Here we observe only excitonic resonances with finite momenta. For the higher temperature case these resonances are more smeared out and their weight is enhanced. Therefore excitonic states with ${\bf q} \neq 0$ can easier be occupied for this case.
Figure \[fig:S00P0\] clearly shows that it is possible to extract the temperature $T_{MI}$ just as well by monitoring the appearance of excitonic resonances in the imaginary part of $\chi(\mathbf{q},\omega+i0)$ at ${\mathbf q}=0$.
Conclusions
===========
In summary, we have performed a detailed investigation of the two-dimensional extended Falicov-Kimball model by means of the projector-based renormalization method. Thereby we established the long-predicted existence of an intervening excitonic insulator phase at the semimetal-semiconductor transition below some critical $T_{EI}$ (see fig. \[fig:PD\]). We derived the renormalized quasiparticle band structure which shows a correlation-induced single-particle gap and $c$-$f$ electron coherence in the low-temperature EI state and reflects the metal-insulator transition at $T_{MI}$ for higher temperatures. Analyzing the imaginary part of the excitonic pair susceptibility, we demonstrate that on the semiconductor side of this phase transition, preformed excitons with zero momentum exist above $T_{EI}$. On the other hand, excitonic bound states (resonances) with finite momentum may appear on both—semiconducting and semimetallic—sides of the metal-insulator transition, but these excitons will not condense for the studied direct band gap situation. We therefore corroborate the scenario, suggested by Bronold and Fehske [@BF06], that in the semiconducting region the EI phase is surrounded by an excitonic halo consisting of free electrons, holes and tightly bound zero-momentum excitons. Forming the EI state, the latter undergo a Bose-Einstein condensation state as the temperature is lowered. Contrariwise there is a well-defined (large) Fermi surface in the semimetallic regime, and the EI state can be envisaged as composed of BCS-type electron-hole pairs. The authors would like to thank F. X. Bronold, D. Ihle, H. Stolz, and B. Zenker for valuable discussions. HF acknowledges a Gordon Godfrey fellowship by the UNSW, where this work was completed. Research was supported by the DFG through SFB 652.
[10]{} url\#1[`#1`]{}
; in , edited by (Academic Press, New York) 1963 p. Suppl. 5 p. 100; ; .
in , edited by (Gordon & Breach, New York) 1968.
.
.
.
.
; ; . ; .
; .
;
.
.
.
.
; ;
; .
.
.
; .
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In this work, an analytical Volume Of Fluid (VOF) implementation of the Generalized Navier Boundary Condition is presented based on the Brackbill surface tension model. The model is validated by simulations of droplets on a smooth surface in a planar geometry. Looking at the static behavior of the droplets, it is found that there is a good match between the droplet shape resolved in the simulations and the theoretically predicted shape for various values of the Young’s angle. Evaluating the spreading of a droplet on a completely wetting surface, the Voinov-Tanner-Cox law ($\theta \propto {\text{Ca}}^{1/3}$) can be observed. At later times scaling follows $r \propto t^{1/2}$, suggesting spreading is limited by inertia. These observations are made without any fitting parameters except the slip length.'
author:
- 'A.M.P. Boelens'
- 'J.J. de Pablo'
title: 'Generalized Navier Boundary Condition for a Volume Of Fluid approach using a Finite-Volume method.'
---
Introduction
============
For any multi-phase flow over a wall where dissipation at the contact line becomes of the same order as the bulk dissipation, a good understanding of contact line behavior is essential [@squires2005]. However, despite many investigations [@dussan1979; @degennes1985; @deconinck2008; @eral2013], for applications including adhesion of liquids to a solid surface [@furmidge1962], transport of liquid water in fuel cells [@zhu2008], liquid infused surfaces [@wexler2015], and coating [@vandre2013], contact line behavior is still not well understood.
The physics of a static contact angle between a liquid and a gas on a smooth solid surface is well established [@young1805; @gauss1830]. However, real surfaces are not completely smooth. They are not chemically homogeneous, and/or have roughness. This causes static contact angle hysteresis and contact line pinning, both of which are difficult to model. When looking at a moving contact line instead of a static contact angle things get even more complicated. Dynamic contact angle behavior is not well understood, even on a completely smooth solid surface. The origin of this poor understanding of the moving contact line is twofold: there is the contact line singularity problem, and the question of how the contact angle depends on contact line velocity.
The contact line singularity problem was first identified by @huh1971. While normally it is a good approximation to use the no-slip condition as boundary condition on the wall, for corner flow this assumption causes the viscous stress and pressure to scale as $r^{-1}$ and thus to diverge as $r \to
0$ at the contact line. Numerous methods have been proposed to resolve this discontinuity or work around it. @hocking1977 showed that, using domain perturbation method in cylindrical coordinates, any slip-velocity model [@navier1823; @thompson1997] resolves the velocity singularity. Another method to circumvent the contact line singularity has been to use the Cahn-Hilliard-van der Waals model [@cahn1977] as the basis for either diffusive interface models [@seppecher1996; @jacqmin2000], or for precursor film models [@hervet1984; @degennes1985]. Precursor film models use a disjoining pressure [@derjaguin1955; @teletzke1987; @herring2010] to model the van der Waals forces that cause the formation of a precursor film ahead of the interface, removing the singularity. More exotic models have suggested local shear-thinning [@cox1998] and non-constant surface tension [@shikhmurzaev1993; @shikhmurzaev1997] as possible solutions.
For the question of how the contact angle depends on the contact line velocity there are also various models. Typically a distinction is made between the local microscopic contact angle at the contact line and the macroscopic apparent contact angle, which is observed in experiments. Due to experimental limitations to access sufficiently small length scales, the apparent contact angle is measured away from the contact line, and curvature of the interface causes this angle to be different from the microscopic contact angle [@chen1989]. Arguably, the easiest method to define the dynamic microscopic contact angle is to simply assume it is fixed and the same as the static contact angle [@eggers2004a]. Instead of a fixed dynamic contact angle, Molecular Kinetic Theory (MKT) [@blake1969; @blake2006] predicts a microscopic contact angle which changes with the contact line velocity. The Voinov-Tanner-Cox [@voinov1976; @tanner1979; @cox1986] law describes the relation between the apparent contact angle and the microscopic contact angle. This law is based on the assumption that it is possible to choose a length scale arbitrary close to the contact line. This makes it impossible to identify a characteristic length scale of the contact line geometry, and reduces its physics to a balance between capillary and viscous stresses [@kafka1979]. Using the lubrication approximation one can now derive the @voinov1976 equation for some specific asymptotic limits, and matching solutions of the Voinov equation with the mesoscopic hydrodynamic solution further away from the wall then gives the Cox-Tanner-Voinov law [@voinov1976; @tanner1979; @cox1986].
Although there are a lot of different models to describe contact lines, there is no consensus on what is the correct description of the physics. Many models have multiple fitting parameters which can be tuned to give the same results [@sibley2014]. Because impurities and surface heterogeneities have a large effect on measurements, experiments are very difficult to reproduce. On top of this, one needs access to microscopic length scales to get to the details and the outcome of experiments on a macroscopic level only depends very weakly on these small length scales [@bonn2009]. With the advent of Molecular Dynamics (MD) [@koplik1988; @koplik1989; @thompson1989; @thompson1990] simulations, and new experimental techniques, such as Atomic Force Microscopy (AFM) [@checco2006], it is now possible to access length and time scales at the contact line that were not accessible before. While MD simulations still can only probe small systems for short times, a couple of fundamental discoveries have been made. Although it had been argued that continuum models break down all together at the contact line [@dussan1979], it was found that both the Navier-Stokes equations and Young’s equation hold up even down to the nanometer scale [@bocquet2010; @seveno2013]. Furthermore, support was found for contact line slip [@thompson1989], precursor films [@he2003], and non-static dynamic microscopic contact angles [@shen1998]. Apart from simulations, precursor films have also been found experimentally [@hardy1919; @kavehpour2003].
In addition to the experimental difficulties and often contradicting findings, another reason that there is no consensus is that all of the above mentioned models have some fundamental short comings and/or work in different regimes. Further analysis of contact line slip models by @huh1977 showed that, even though stresses are not diverging anymore, the pressure still shows a weak singularity i.e. the pressure diverges, but becomes finite when integrated. While using a slip length model is sufficient from a modeling perspective, a divergent pressure cannot be right from a physics perspective. The precursor film on the other hand does successfully work around the contact line singularity for both viscous stresses and pressure. However, they are typically not seen under partial wetting conditions [@snoeijer2013]. The different models for microscopic and macroscopic contact angles have some limitations too. While a model which describes the contact angle as a function of the capillary number might describe dynamic contact angle hysteresis correctly, there is also static contact angle hysteresis, which causes contact line pinning of non-moving contact lines. Any model that lets the contact angle only depend on the local capillary number will not be able to capture static contact angle hysteresis. Assuming a static angle ,on the other hand, does not capture static contact angle hysteresis, is not able to properly describe the flow of a liquid over a chemically patterned surface with different wetting properties, and does not accurately predict contact line velocity [@yamamoto2013].
In this work, a validation study of a Volume Of Fluid (VOF) implementation of a contact line model called the Generalized Navier Boundary Condition (GNBC) is presented. While the non-sharp interface of the VOF method implicitly resolves the contact line singularity problem, even with a no-slip condition at the wall [@seppecher1996; @chen2000; @sibley2013], the question of what is the right contact angle is still valid for this method. In addition to applying the Navier slip condition at the wall, the GNBC uses the reduced Young’s stress as a restoring force when the contact angle deviates from its equilibrium value, and is informed by Molecular Dynamics [@qian2003; @gentner2004]. While previous implementations used a friction factor to link the reduced Young’s stress to a contact line velocity [@gerbeau2009], this approach does not work for the VOF model, because of the implicit slip of the interface [@ashish2009]. Instead, in this work, the reduced Young’s stress is used directly as a source term in the navier stokes equations [@mahady2015]. A consequence of using the GNBC is that the contact angle no longer is a constraint imposed onto the system, but that the contact angle is a self-selecting variable. Using this approach has a couple of advantages over existing models. Because there is no enforcement of a model that relates the contact angle to contact line speed, this model can reproduce static contact angle hysteresis, without artificially fixing the position of the contact line [@fang2008; @dupont2010]. Additionally, a variable contact angle model can describe flow over a chemically patterned surface [@dupuis2004; @wang2008]. While the Generalized Navier Boundary Condition has been implemented for continuum simulations using a diffuse interface Cahn Hillard method [@qian2006; @ren2007; @wang2008; @cai2015], using a Volume Of Fluid approach has the advantage that many less grid points are needed to resolve the interface [@ding2007]. In contrast to the body force term derived by @mahady2015, the model presented in this work describes wetting of a dry surface. Opposed to @deganello2011, the model presented here is a analytical derivation of the line tension force.
In our work, we uncover the following findings: when looking at the static behavior of droplets, it is found there is a good match between the droplet shape resolved in the simulations and the theoretically predicted shape for various values of the Young’s angle. Investigating the spreading of a droplet on a completely wetting surface, the Voinov-Tanner-Cox law [@voinov1976; @tanner1979; @cox1986] ($\theta \propto {\text{Ca}}^{1/3}$) can be observed. Late time scaling follows $r \propto t^{1/2}$, suggesting spreading is limited by inertia. These observations are made without any fitting parameters except the slip length.
Theory
======
Body force reformulation of uncompensated Young’s stress
--------------------------------------------------------
The traditional form of the Generalized Navier Boundary Condition (GNBC) used in, for example, Arbitrary Lagrangian-Eulerian simulations [@gerbeau2009], looks like: $$\beta {\left( \vec{u} - \vec{u}_{w} \right)} \cdot \tau
+ \mu {\left( \nabla \vec{u} + \nabla \vec{u}^{T} \right)} \hat{n}_{w} \cdot \tau
+ \sigma {\left( \cos{\theta} - \cos{\theta_{0}} \right)} \hat{t}_{w} \cdot \tau \delta_{{\text{cl}}}
=
0
\label{eqn:gnbc}$$ Here the first two terms on the left hand side represent the Navier slip boundary condition [@navier1823], and the third term represents the unbalanced Young’s stress. $\beta$ is the slip coefficient, $\vec{u}$ is velocity, $\vec{u}_{w}$ is the velocity of the wall, $\tau$ is any vector tangent to the wall, $\mu$ is the dynamic viscosity, $\hat{n}_{w}$ is the normal of the wall pointing outward, $\sigma$ is the surface tension coefficient, $\theta$ is the dynamic contact angle, $\theta_{0}$ is the equilibrium Young’s contact angle, $\hat{t}_{w}$ is the vector tangent to the wall and normal to the contact line, and the distribution $\delta_{{\text{cl}}}$ is defined as [@gerbeau2009]: $${\left<{\delta_{{\text{cl}}}, \psi}\right>}
=
\int_{{\text{cl}}} \psi d \lambda_{{\text{cl}}}$$ where $\psi$ is any smooth function, and $\lambda_{{\text{cl}}}$ denotes the Lebesgue measure (i.e. the length measure) on the contact line. The slip length in equation \[eqn:gnbc\] is equal to $l_{s} = \mu / \beta$.
Analogous to the methods of @brackbill1992 a Volume Of Fluid expression is derived for the above equation for the Generalized Navier Boundary Condition. However, instead of relating the uncompensated Young’s stress to a velocity using a slip coefficient, the uncompensated Young’s stress is modeled as an extra body force acting at the contact line in the Navier-Stokes equation. The reason for this approach is that the intrinsic contact line slip in a VOF code is large enough that converting the contact line tension directly to a velocity on the wall does not move the contact line. The Navier-slip boundary condition component of the above equation is left unchanged in our implementation of the Generalized Navier Boundary Condition.
The goal of this derivation is to find an expression that rewrites the contact line force as a body or volume force, so it can be treated as a momentum source term in the Navier-Stokes equations. The first step is to find a relation, that rewrites the line force $\vec{F}_{\tau L}$ as surface force $\vec{F}_{\tau A}$ acting on the wall: $$\lim_{h \to 0} \int_{\Delta A_{w}} \vec{F}_{\tau A} {\left( \vec{x} \right)} d^{2} \vec{x}
=
\int_{\Delta L} \vec{F}_{\tau L} {\left( \vec{x}_{L} \right)} d \vec{x}
\label{lineForce}$$ The points, $\vec{x}_{L}$, form the contact line or triple point, $\Delta L$ is the length of the line segment in a the small volume of integration $\Delta V$, and $\Delta A_{w}$ is the side of $\Delta V$ that is part of the wall, and in which the points $\vec{x}_{L}$ lay. An additional constraint for $\vec{F}_{\tau
A} {\left( x \right)}$ is that it is zero outside of the interface region: $$\vec{F}_{\tau A} {\left( x \right)}
=
0
{\text{ for }}
{\left|{
\hat{n}_{2D} {\left( \vec{x}_{L} \right)}
\cdot
{\left( \vec{x} - \vec{x}_{L} \right)}
}\right|}
\geq
h$$ where $\hat{n}_{2D} {\left( \vec{x}_{L} \right)}$ is the normal to the contact line in the plane of the wall, and $\delta {\left[ \hat{n}_{2D} {\left( \vec{x}_{L} \right)} \cdot
{\left( \vec{x} - \vec{x}_{L} \right)} \right]}$ describes the plane $\Delta A_{w}$.
Consider a system of two fluids, fluid $1$, and fluid $2$, separated by an interface, and define a discontinuous function, $c {\left( \vec{x} \right)}$, to distinguish between the two phases: $$c {\left( \vec{x} \right)}
=
\left\{
\begin{array}{ll}
c_{1} & {\text{in liquid 1}} \\
{\left<{c}\right>} = {\left( c_{1} + c_{2} \right)}/2 & {\text{on interface}} \\
c_{2} & {\text{in liquid 2}}
\end{array}
\right.$$ An example of such a function would be the density of two different incompressible liquids, $\rho_{1}$, and $\rho_{2}$. In this case the position of the contact line can be found with: $$\rho {\left( \vec{x}_{L} \right)}
=
{\left<{\rho}\right>}$$ In order to change this problem from a boundary value problem on the contact line to an approximate continuous model, one can define a continuous function $\tilde{c}$, which varies smoothly over thickness $h$ going from $c_{1}$ to $c_{2}$ , $c_{1} \le \tilde{c} \le c_{2}$. $h$ is a length scale of the order of the grid size $\Delta x$, and defines the width of the transition region from $c_{1}$ to $c_{2}$. The two functions $c$ and $\tilde{c}$ are related through the interpolation function $\mathcal{P} {\left( \vec{x} \right)}$: $$\tilde{c} {\left( \vec{x} \right)}
=
\frac{1}{h^{2}} \int_{A} c {\left( \vec{x}' \right)} \mathcal{P} {\left( \vec{x}'-\vec{x} \right)} d^{2} \vec{x}$$ which is normalized as: $$\int_{A} \mathcal{P} {\left( \vec{x} \right)} d^{2} \vec{x}
=
h^{2}$$ is bounded as: $$\mathcal{P} {\left( \vec{x} \right)}
=
0
{\text{ for }}
{\left|{\vec{x}}\right|}
\ge
h/2$$ is differentiable, and decreases monotonically with ${\left|{\vec{x}}\right|}$. The continuous function is defined such that: $$\lim_{h \to 0} \tilde{c} {\left( \vec{x} \right)}
=
c {\left( \vec{x} \right)}$$ i.e. the function $\tilde{c} {\left( \vec{x} \right)}$ approaches $c {\left( \vec{x} \right)}$ as the interface thickness goes to zero. $\tilde{c}$ is differentiable because $\mathcal{P}$ is, and: $$\nabla_{2D} \tilde{c} {\left( \vec{x} \right)}
=
\frac{1}{h^{2}} \int_{A} c {\left( \vec{x}' \right)} \nabla_{2D} \mathcal{P} {\left( \vec{x}'-\vec{x} \right)} d^{2} \vec{x}$$ where $\nabla_{2D}$ is the two-dimensional gradient in the plane of the wall. Using Gauss’ theorem and the realization that $c {\left( \vec{x} \right)}$ is constant within each fluid, the above integral can be written as: $$\nabla_{2D} \tilde{c} {\left( \vec{x} \right)}
=
\frac{{\left[ c \right]}}{h^{2}} \int_{L} \hat{n}_{2D} {\left( \vec{x}_{L} \right)} \mathcal{P} {\left( \vec{x}-\vec{x}_{L} \right)} d L
\label{gauss}$$ where ${\left[ c \right]} = c_{2} - c_{1}$, thus converting the surface integral to a line integral. To pull the normal out of the integral, its weighted mean is calculated. Since $\mathcal{P}$ is bounded, its maximum contribution to the line integral is $\mathcal{O} {\left( h \right)}$. Integral \[gauss\] can thus be approximated as: $$\frac{1}{h^{2}} \int_{L} \hat{n}_{2D} {\left( \vec{x}_{L} \right)} \mathcal{P} {\left( \vec{x}-\vec{x}_{L} \right)} d L
\approx
\frac{1}{h^{2}} \hat{n}_{2D} {\left( \vec{x}_{L0} \right)} \int_{L} \mathcal{P} {\left( \vec{x}-\vec{x}_{L} \right)} d L
+
\mathcal{O} {\left( \frac{h}{R} \right)}
\label{approximate}$$ where $\vec{x}_{L0}$ is the point on $L$ closest to $x$, and $R$ is the radius of the contact line.
The integral in equation \[approximate\] can be bounded by: $$\frac{1}{h} \int_{L} \mathcal{P} {\left( \vec{x}-\vec{x}_{L} \right)} d L
\le
\mathcal{P} {\left( \vec{x}-\vec{x}_{L0} \right)}$$ where in the limit $h \to 0$, $\mathcal{P} {\left( \vec{x}-\vec{x}_{L0} \right)}$ is zero everywhere except for $\vec{x} = \vec{x}_{L0}$. Taking the corresponding limit of $\nabla_{2D} \tilde{c} {\left( \vec{x} \right)}$ over the interface gives: $$\lim_{h \to 0} \int \hat{n}_{2D} {\left( \vec{x}_{L0} \right)} \cdot \nabla_{2D} \tilde{c} {\left( \vec{x} \right)} dx
=
{\left[ c \right]}$$ As $h$ goes to $0$, $\nabla_{2D} \tilde{c} {\left( \vec{x} \right)}$ is thus equivalent to: $$\lim_{h \to 0} \nabla_{2D} \tilde{c} {\left( \vec{x} \right)}
=
\hat{n}_{2D} {\left[ c \right]} \delta_{2D} {\left[ \hat{n}_{2D} \cdot {\left( \vec{x} - \vec{x}_{L} \right)} \right]}
=
\nabla_{2D} c {\left( \vec{x} \right)}
\label{delta}$$
Because the Brackbill surface tension already takes care of the $\cos(\theta)$ component of the reduced Young’s stress, only the $\cos{\theta_{0}}$ component needs to be modeled. Using the delta function, the contact line force can now be written as a surface force as follows: $$\begin{aligned}
\int_{L} \vec{F}_{\tau L} {\left( \vec{x}_{L} \right)} d L
&=
\int_{A} \vec{F}_{\tau L} {\left( \vec{x} \right)} \delta_{2D} {\left[ \hat{n}_{2D} {\left( \vec{x}_{L} \right)} \cdot {\left( \vec{x} - \vec{x}_{L} \right)} \right]} d^{2} \vec{x} \nonumber \\
&=
\int_{A} \sigma \cos{\theta_{0}} \hat{n}_{2D} {\left( \vec{x} \right)} \delta_{2D} {\left[ \hat{n}_{2D} {\left( \vec{x}_{L} \right)} \cdot {\left( \vec{x} - \vec{x}_{L} \right)} \right]} d^{2} \vec{x}\end{aligned}$$ converting the line integral over the contact line into an integral over the wall surface. Equation \[delta\] can be used as an approximation for the delta function when the interface has a finite thickness. Substitution gives: $$\int_{\Delta L} \vec{F}_{\tau L} {\left( \vec{x}_{L} \right)} d L
=
\lim_{h \to 0} \int_{\Delta A_{w}} \sigma \cos{\theta_{0}} \frac{\nabla_{2D} \tilde{c} {\left( \vec{x} \right)}}{{\left[ c \right]}} d^{2} \vec{x}
\label{surfaceForce}$$ and comparing equation \[lineForce\] with \[surfaceForce\], the surface force $\vec{F}_{\tau A}$ can be identified as: $$\vec{F}_{\tau A}
=
\sigma \cos{\theta_{0}} \frac{\nabla_{2D} \tilde{c} {\left( \vec{x} \right)}}{{\left[ c \right]}}$$ As a last step, the surface integral needs to be converted to a volume integral. Since $\nabla_{2D} \tilde{c}$ is independent of the distance away from the wall, integrating over $\Delta V$ is the same as multiplying $\vec{F}_{\tau A}$ with mesh size, i.e. $\Delta A_{w} / \Delta V$: $$\vec{F}_{\tau V}
=
\vec{F}_{\tau A} \frac{\Delta A_{w}}{\Delta V}
=
\sigma \cos{\theta_{0}} \frac{\nabla_{2D} \tilde{c}
{\left( \vec{x} \right)}}{{\left[ c \right]}} \frac{\Delta A_{w}}{\Delta V}
\label{eqn:volumeForce}$$ where $\Delta A_{w}$ is the surface area of the wall in volume $\Delta V$.
While the above equation has the correct limiting behavior, it was found that, due to the diffuse nature of the interface, at too low resolution the interface gets pulled apart. To counter this phenomenon, a modification is proposed to localize the contact line force more to the interface: $$g {\left( \tilde{c} \right)}
=
H {\left( \pi/2-\theta_{0} \right)} \frac{5}{4} H {\left( \tilde{c}-\frac{1}{5} \right)}
+ H {\left( \theta_{0}-\pi/2 \right)} \frac{5}{4} H {\left( \frac{4}{5}-\tilde{c} \right)}$$ Where $H(x)$ is the Heaviside step function. Depending on the value of the Young’s angle, this function truncates the contact line force. For $\theta_{0} <
\pi/2$ this means that the contact line does not get pulled apart, while for $\theta_{0} > \pi/2$ this prevents the gas phase from being pulled into the droplet at the contact line. The value $5/4$ is to normalize the function. The new function for the contact line restoring force at low resolution thus becomes: $$\vec{F}_{\tau V}
=
\vec{F}_{\tau A} \frac{\Delta A_{w}}{\Delta V} g {\left( \tilde{c} \right)}
\label{eqn:volumeForceCor}$$ where the same definition is used for $\vec{F}_{\tau A}$ as in equation \[eqn:volumeForce\].
Numerical implementation
------------------------
This section focuses on the numerical implementation of the above derived equation in the Volume Of Fluid (VOF) solver that comes with OpenFOAM [@jasak1996; @openfoam]. This involves implementing the uncompensated Young’s stress and the Navier slip boundary condition. In this code the general phase parameter $\tilde{c} {\left( \vec{x} \right)}$ is called $\alpha$, and has the following properties: $$\alpha
=
\left\{
\begin{array}{ll}
0 & {\text{in phase 1}} \\
(0, 1) & {\text{on interface}} \\
1 & {\text{in phase 2}}
\end{array}
\right.$$ Phase parameter $\alpha$ is stored as a separate field, just like velocity and pressure, and its evolution is calculated using the following transport equation: $$\frac{\partial \alpha}{\partial t}
+ \nabla \cdot {\left( \alpha \vec{v} \right)}
+ \nabla \cdot {\left( \alpha {\left( 1 - \alpha \right)} \vec{v}_{lg} \right)}
=
0$$ where $\vec{v} = \alpha \vec{v}_{l} + {\left( 1 - \alpha \right)} \vec{v}_{g}$ is the phase averaged velocity, and $\vec{v}_{lg} = \vec{v}_{l} - \vec{v}_{g}$ is the velocity difference between the liquid and the gas phase. This equation is equivalent to a material derivative, but rewritten to minimize numerical diffusion [@rusche2002].
The volume fraction is used to calculate phase-averaged densities, velocities, and viscosities, which are used in the momentum balance $$\frac{\partial \rho \vec{v}}{\partial t}
+ \nabla \cdot {\left( \rho \vec{v} \otimes \vec{v} \right)}
=
- \nabla p
+ \nabla \cdot {\left( \mu \nabla \vec{v} \right)}
+ \rho \vec{g}
+ \vec{f}_{{\text{st}}}
- \vec{f}_{{\text{cl}}}$$ and the continuity equation: $$\nabla \cdot \vec{v}
=
0$$ In the above equations $\rho$ is the density, $\vec{v}$ is the velocity, $t$ time, $p$ is pressure $\mu$ is the viscosity, $g$ is gravity, $\vec{f}_{{\text{st}}}$ is the surface tension force, $\vec{f}_{{\text{cl}}}$ is the contact line tension force, and $\otimes$ is the dyadic product. The density $\rho$, velocity $\vec{v}$, and viscosity $\mu$, are all phase averaged using $\alpha$.
The surface tension force is calculated using the expression: $$\vec{f}_{{\text{st}}}
=
\sigma_{{\text{st}}} \kappa \nabla{\alpha}$$ where $\sigma_{{\text{st}}}$ is the surface tension coefficient, $$\kappa
=
- {\left( \nabla \cdot \vec{n} \right)}$$ is the curvature of the interface, and $$\vec{n}
=
\frac{\nabla \alpha}{{\left|{\nabla \alpha}\right|}}$$ is the normal of the interface [@brackbill1992].
Using $\alpha$, the line tension force defined in equation \[eqn:volumeForceCor\] is rewritten as: $$\vec{f}_{{\text{cl}}}
=
\sigma_{{\text{st}}} \cos{\theta_{0}} \;
g {\left( \alpha \right)}
\frac{\Delta A_{w}}{\Delta V}$$ and $\nabla_{2D} \alpha$ is implemented as: $$\nabla_{2D} \alpha
=
\nabla \alpha - {\left( \hat{n}_{w} \cdot \nabla \alpha \right)} \hat{n}_{w}$$ where $\hat{n}_{w}$ again is the normal of the wall pointing outward. The equilibrium Young’s angle $\theta_{0}$ can be defined uniquely for any grid cell along the wall, so an arbitrary wettability pattern can be created. $\Delta A_{w}$ and $\Delta V$ are properties of the mesh that can be accessed directly in OpenFOAM. The contact line tension source term is solved explicitly along with the surface tension source term.
The Navier slip condition is implemented as: $$\vec{v}
=
l_{s} {\left( {I} - \hat{n}_{w}^{2} \right)} \nabla \vec{v}$$ where $l_{s}$ is the slip length ${I}$ is the identity matrix, and $\hat{n}_{w}^{2}$ is the dyadic product of the wall normal, $\hat{n}_{w}$, with itself. Using this formulation, the velocity perpendicular to the wall is always set to zero.
Results
=======
The first section of the results focuses on the static behavior of the implementation of the Generalized Navier Boundary Condition presented above. The second section covers the dynamic contact angle behavior. To speed up simulation times all results are for 2D droplets in a planar geometry, and using the symmetry of the system only half of the droplet is simulated. All simulations are performed without gravity and represent water droplets in air at room temperature.
Static
------
![The interface of various droplets at time $0.1 \si{\s}$ \[figure:contourEquil\]](fig1){width="60.00000%"}
Figure \[figure:contourEquil\] shows the interface ($\alpha = 0.5$) of various droplets with different Young’s angles at a resolution of $512 \times 256$ in a box of $1.5 \si{\milli\m} \times 0.75 \si{\milli\m}$. The half droplets in the figure have a surface area of about $0.2 \si{\milli\m^{2}}$ (i.e. a radius of $0.5 \si{\milli\m}$ when the Young’s angle is $\pi/2$). As initial condition these droplet where given their equilibrium shape, and it can be seen that after a simulation time of $0.1 \si{\s}$ the droplets have maintained their shape.
![Radius of the base of the droplet and droplet height as a function of different contact angles.\[figure:angleRadiusEquil\]](fig2){width="60.00000%"}
To further quantify these droplets, figure \[figure:angleRadiusEquil\] shows the height and radius of the base of the droplets as a function of the Young’s angle. The radius of the base of the droplet $r_{B}$ is calculated as: $$r_{b} = R \sin{\theta_{0}},$$ and the height of the droplet is equal to: $$h_{d} = R {\left( 1 - \cos{\theta_{0}} \right)}.$$ In the above equations: $$R = \sqrt{\frac{2 A}{\theta_{0} - \cos{\theta_{0}} \sin{\theta_{0}}}}$$ is the radius of the droplet, $\theta_{0}$ is the Young’s angle, and $A$ is the surface area of the droplet. While there are small deviations between the theoretically predicted droplet shapes and the simulations, both match well.
![Error in the radius of the base of the droplet as a function of resolution for a contact angle of $\theta_{0} = \pi/4$. \[figure:resolution\]](fig3){width="60.00000%"}
The convergence of the error as a function of resolution is shown in figure \[figure:resolution\]. The x axis shows the resolution. The values of the time steps at these resolutions are: $0.1 \si{\mu \s}$, $0.05 \si{\mu \s}$, $0.025 \si{\mu \s}$, and $0.01 \si{\mu
\s}$, to keep the Courant number of the same order between simulations. The y axis shows the absolute error, $\epsilon_{{\text{abs}}}$, between the theoretical value for the radius of the base of the droplet and the simulated value for a Young’s angle of $\theta_{0} = \pi/4$. The graph shows that the code is close to 2nd order accurate.
![\[figure:pressureEquil\] Pressure at the wall at $t = 50 \si{\milli
\s}$ for different resolutions and a contact angle of $\theta_{0} = \pi/4$](fig4){width="60.00000%"}
Because the interface gets thinner as resolution increases the Volume Of Fluid method is mesh dependent. As can be seen in figure \[figure:pressureEquil\], this results in the pressure peak on the surface getting sharper with increasing resolution. The increased pressure inside the drop, left of the pressure peak, is the Laplace pressure.
![Maximum of pressure peak at the contact line as a function of different contact angles.\[figure:anglePresEquil\]](fig5){width="60.00000%"}
The reason there is a pressure peak at the contact line for a stationary droplet in the first place can be seen in figure \[figure:anglePresEquil\]. This figure shows the value of the pressure peak as a function of the Young’s angle. From the figure it is clear that there is a strong dependence of the pressure peak on the Young’s angle. Since the line tension force only acts parallel to the surface, this suggest that the pressure peak is the result of the surface tension force calculated by the @brackbill1992 model. If the vertical component of the reduced Young’s stress also were implemented, an extra term, proportional to $\sigma \sin{\theta_{0}}$, would have been present in the plot as an additional contribution to the pressure peak. Since we are only concerned about solid surfaces, this term was not incorporated in the model.
![\[figure:velocityEquil\] Velocity at the wall at $t = 50 \si{\milli \s}$ for different resolutions](fig6){width="60.00000%"}
Due to the large pressure peak there is also a small residual slip velocity at the contact line, which can be observed in figure \[figure:velocityEquil\]. As was the case with the pressure, this velocity peak gets sharper at higher resolutions. As can be seen in figure \[figure:angleRadiusEquil\] the residual slip velocity does not negatively affect the shape of the drop. However, spurious currents are a well known issue of Volume Of Fluid solvers [@deshpande2012] and, if needed, can be controlled by making the time step sufficiently small.
![Integral of pressure peak at the contact line as a function of resolution.\[figure:resolutionPressure\]](fig7){width="60.00000%"}
![Integral of slip velocity at the contact line as a function of resolution.\[figure:resolutionVelocity\]](fig8){width="60.00000%"}
To investigate to what extent the pressure and velocity peaks affect the solution they are both integrated over the wall for various resolutions. Figure \[figure:resolutionPressure\] shows this integral for the pressure, and figure \[figure:resolutionVelocity\] for the slip velocity. The integration limits in both plots are from $\alpha = 0.01$ to $\alpha = 0.99$. For the integral over the pressure peak this approach makes sure that the integral is not affected by the Laplace pressure inside the droplet, and the velocity integral uses the same limits to be consistent with the pressure. It can be appreciated how the pressure integral converges to a constant value and the velocity integral approaches zero as resolution increases, showing a convergent solution for both the pressure and velocity.
Dynamic
-------
For the validation of the dynamic case, the starting point is again a 2D droplet in planar geometry. However, in this case, the equilibrium Young’s contact angle is set to $\theta_{0} = 0$, and the spreading behavior as a function of time is investigated.
![\[figure:contourTanner\] Interface of a droplet at different times at a resolution of $512 \times 256$.](fig9){width="60.00000%"}
Figure \[figure:contourTanner\] shows the interface ($\alpha = 0.5$) at various times for a box of $1.5 \si{\milli\m} \times 0.75 \si{\milli\m}$ and with a resolution of $512 \times 256$. As expected the droplet keeps spreading until the edge of the simulation box is reached and the simulation is stopped.
![\[figure:diameterTanner\] Radius of a droplet as a function of time.](fig10){width="60.00000%"}
Figure \[figure:diameterTanner\] shows how the droplet spreads by showing the radius of the droplet as a function of time for two different resolutions of $256
\times 128$ and $512 \times 256$. For lower resolutions the contact line was pulled apart and the results are not shown in the graph. For reference, the power law $r \propto t^{1/2}$ is also shown. The $r \propto t^{1/2}$ power law describes the late time spreading behavior for low viscosity axisymmetric droplets [@biance2004; @ding2007b]. At a resolution of $256 \times 128$ initially the contact line hardly moves, but then the spreading radius as a function of time becomes proportional to: $r \propto t^{1/2}$. At the larger resolution of $512 \times
256$ it can be seen that the scaling also converges to $r \propto t^{1/2}$.
![\[figure:pressureTanner\] Pressure at the wall for various times at a resolution of $512 \times 256$](fig11){width="60.00000%"}
.
![\[figure:velocityTanner\] Velocity at the wall for various times at a resolution of $512 \times 256$](fig12){width="60.00000%"}
.
Figures \[figure:pressureTanner\] and \[figure:velocityTanner\] show the pressure profile and velocity at the wall for various times. The pressure shows a sharp peak already, but especially at $t=0.5 \si{\milli\s}$ there is quite some noise in both the curves for pressure and velocity. The peaks are expected to become even sharper with increasing resolution [@afkhami2009]. While it is known that for the Navier-slip boundary condition pressure is divergent [@devauchelle2007], whether the reduced Young’s stress provides a cut-off mechanism for the pressure has not yet been analytically determined.
Because of the difference in the spreading curves in figure \[figure:diameterTanner\] an additional simulation was performed at a much larger resolution. The simulation domain for this simulation is $0.4\si{\milli\m} \times 0.2\si{\milli\m}$ with a resolution of $256 \times 128$, but the mesh is refined at the wall to better capture the curvature at the contact line. The grid cells at the wall are about $10 \si{\nano\m}$ cubed. The droplet has an initial radius of $0.15\si{\milli\m}$ and makes an angle to $\theta = \pi/4$ with the surface. The function $g$ which was used in the above simulations to keep the contact line tension localized was omitted in this simulation because it is not needed at larger resolutions.
![\[figure:contourTannerRes\] Interface of a droplet at different times at high resolution.](fig13){width="60.00000%"}
![\[figure:diameterTannerRes\] Radius of a droplet as a function of time at high resolution.](fig14){width="60.00000%"}
Figure \[figure:contourTannerRes\] shows the interface ($\alpha = 0.5$) of the droplet for various times. Figure \[figure:diameterTannerRes\] shows the corresponding radial position of the contact line as a function of time. As was observed in Figure \[figure:diameterTanner\], this smaller droplet also shows inertia dominated spreading.
![\[figure:angleTannerRes\] Angle of the inflection point (i.e. the apparent contact angle) as a function of the capillary number at high resolution.](fig15){width="60.00000%"}
Because of the larger resolution in these simulations the apparent contact angle can accurately be determined. Figure \[figure:angleTannerRes\] shows the apparent contact angle as a function of the capillary number. It can be appreciated that the simulation recovers the Voinov-Tanner-Cox [@voinov1976; @tanner1979; @cox1986] law: ($\theta \propto {\text{Ca}}^{1/3}$).
![\[figure:pressureTannerRes\] Pressure at the wall for various times at high resolution.](fig16){width="60.00000%"}
![\[figure:velocityTannerRes\] Velocity at the wall for various times at high resolution.](fig17){width="60.00000%"}
Figures \[figure:pressureTannerRes\] and \[figure:velocityTannerRes\] show the pressure peak and slip velocity at the wall at various times. Because of the high resolution at the wall both are very sharp peaks confined to the interface. The fluctuations of the pressure and velocity that was observed in figures \[figure:pressureTanner\] and \[figure:velocityTanner\] is no longer there, suggesting the simulations have fully converged.
Conclusions & Discussion
========================
An implementation of the Generalized Navier boundary condition for the Volume Of Fluid method is presented in this work. In analogy with the Brackbill surface tension model, a body force representation is developed for the contact line tension, while the Navier slip condition is applied on the wall. A validation of the code is presented for both a static case of a droplet maintaining its equilibrium shape, and a dynamic case of a spreading droplet.
It is shown how, on a completely smooth solid surface, in a system without gravity, the shape of the simulated droplets matches with their theoretically predicted shape for various Young’s angles. In addition, it is shown how the pressure peak and corresponding velocity peak at the interface converge with increasing resolution.
For the dynamic case it is found that the spreading of the droplet scales as $r
\propto t^{1/2}$. This suggests that spreading is limited by inertia. Also the Voinov-Tanner-Cox law is observed ($\theta \propto {\text{Ca}}^{1/3}$). This behavior is observed without using any fitting parameters.
An interesting future application of this model is flow over pattered surfaces [@dupuis2004; @wang2008]. Another topic of interest is the possibility to investigate both static and dynamic contact angle pinning. One can apply any pattern of equilibrium contact angles on a surface, and study how different patterns pin the contact line, either in a static or in a dynamic system.
Many thanks to Sidney Nagel and the Nagel group at the Department of Physics of the University of Chicago for stimulating discussions and feedback.
[74]{} natexlab\#1[\#1]{}bibnamefont \#1[\#1]{}bibfnamefont \#1[\#1]{}citenamefont \#1[\#1]{}url \#1[`#1`]{}urlprefix\[2\][\#2]{} \[2\]\[\][[\#2](#2)]{}
, ****, ().
, ****, ().
, ****, ().
, ****, ().
, , , ****, ().
, ****, ().
, , , ****, ().
, , , , , , ().
, , , ****, ().
, pp. ().
, ** (, ).
, ****, ().
, ****, ().
, ****, ().
, ****, ().
, ****, ().
, ****, ().
, ****, ().
, ****, ().
, ****, ().
, , , ****, ().
, ****, ().
, ****, ().
, ****, ().
, ****, ().
, ****, ().
, ****, ().
, ****, ().
, ****, ().
, ****, ().
, ****, ().
, ****, ().
, ****, ().
, , , , pp. ().
, , , , , ****, ().
, , , ****, ().
, , , ****, ().
, ****, ().
, ****, ().
, , , , ****, ().
, ****, ().
, , , ****, ().
, ****, ().
, ****, ().
, ****, ().
, , , ****, ().
, ****, ().
, ****, ().
, , , , ****, ().
, , , ****, ().
, , , , ****, ().
, , , ****, ().
, , , , ****, ().
, ****, ().
, ****, ().
, , , ****, ().
, , , , , ****, ().
, ****, ().
, ****, ().
, , , ****, ().
, , , ****, ().
, ****, ().
, , , , ****, ().
, , , ****, ().
, , , , , , ****, ().
, , , ****, ().
, Ph.D. thesis, ().
, ** (), <http://www.openfoam.org>.
, Ph.D. thesis, ().
, , , ****, ().
, , , ****, ().
, ****, ().
, , , ****, ().
, , , ****, ().
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'A hybrid-duplex aeronautical communication system (HBD-ACS) consisting of a full-duplex (FD) enabled ground station (GS), and two half-duplex (HD) air-stations (ASs) is proposed as a direct solution to the spectrum crunch faced by the aviation industry. Closed-form outage probability and finite signal-to-noise ratio (SNR) diversity gain expressions in aeronautical communications over Rician fading channels are derived for a successive interference cancellation (SIC) detector. Similar expressions are also presented for an interference ignorant (II) detector and HD-equivalent modes at GS and ASs. Through outage and finite SNR diversity gain analysis conducted at the nodes, and system level, residual self-interference (SI) and inter-AS interference are found to be the primary limiting factors in the proposed HBD-ACS. Further investigations revealed that the II and SIC detectors in the proposed HBD-ACS are suitable for weak and strong interference scenarios, respectively. When compared to HD-ACS, the proposed HBD-ACS achieves lower outage probability and higher diversity gains at higher multiplexing gains when operating at low SNRs. Finite SNR analysis also showed the possibility of the proposed HBD-ACS being able to attain interference-free diversity gains through proper management of residual SI. Hence, the proposed HBD-ACS is more reliable and can provide better throughput compared to existing HD-ACS at low-to-moderate SNRs.'
author:
- 'Tan Zheng Hui Ernest, A S Madhukumar, Rajendra Prasad Sirigina, and Anoop Kumar Krishna[^1][^2][^3][^4]'
bibliography:
- 'IEEEabrv.bib'
- 'ref.bib'
title: 'On the Outage Analysis and Finite SNR Diversity-Multiplexing Tradeoff of Hybrid-Duplex Systems for Aeronautical Communications'
---
Aeronautical Communications, Spectral Efficiency, Full-Duplex, Hybrid-Duplex, Half-Duplex, Outage Probability, Rician, Finite signal-to-noise ratio (SNR), Diversity.
Introduction
============
2012 and 2032, air travel within the Pacific South East Asia region is projected to record a compounded annual growth rate of 5.3% [@icao2016long]. This air travel growth trend exposes existing aeronautical communication systems (ACSs) to considerable strain due to demand for data communications from legacy, current and future generation avionics systems. Consequently, this places an additional strain on existing Air-to-Ground (A/G) and Air-to-Air (A/A) aeronautical communication links on the congested aeronautical spectrum. With existing ACSs being unable to deliver the needed data capacity [@neji2013survey], various communication technologies have been proposed to improve the capabilities of existing A/G and A/A links [@neji2013survey; @ernest2016efficiency]. However, these solutions do not directly address the issue of spectrum utilization.
A hybrid-duplex (HBD) ACS consisting of half-duplex (HD) air-stations (ASs) operating existing avionics systems with full-duplex (FD) ground stations (GSs) can be an alternative solution to the shortage of available aeronautical spectrum currently faced by the aviation community. Changes to existing/legacy HD avionics systems currently on board aircrafts can be kept to a minimum in HBD-ACS, thus enabling HBD-ACS to be less disruptive to adopt. Wireless communication systems that have adopted the HBD paradigm include cognitive radio systems [@li2014linear] and cellular systems [@mohammadi2015full; @jang2015spatial; @cirik2018robust].
In HBD systems, both FD and HD nodes communicate on the same spectrum since an FD node can simultaneously transmit and receive signals on the same frequency and thereby improve the spectral efficiency [@kim2015survey], [@sabharwal2014band], [@tapio2013system]. Despite extensive studies done on self-interference (SI) mitigation architectures, SI remains the primary challenge faced by FD nodes due to simultaneous signal transmission and reception. SI mitigation architectures can be categorized into either passive suppression or active cancellation [@sahai2013impact]. The former mitigates SI through induced path loss (e.g. antenna separation) while the latter cancels SI in the analog or digital domain. However, residual SI will still be present due to the limited dynamic range of analog-to-digital converters [@sabharwal2014band], [@korpi2014full], inherent carrier phase noise [@syrjala2014analysis] and imperfect SI channel estimation at the FD transceiver [@sahai2013impact]. Effectively managing residual SI in HBD-ACS opens up the possibility of directly addressing the spectrum crunch faced by the aviation industry. In particular, multiple aircrafts and ground stations can communicate on the same aeronautical spectrum, providing motivation for this paper.
Related Literature
------------------
Apart from SI at FD nodes, HD nodes in HBD systems also experience interference due to transmissions from other HD and FD nodes. In the literature, multiple interference management approaches have been presented. However, this paper focuses on two widely known approaches where interference is either ignored, i.e., interference ignorant (II) detector, or successfully canceled, i.e., successive interference cancellation (SIC) detector.
To quantify the effectiveness of the II and SIC detectors, many related works in literature have attempted to determine the closed-form outage probabilities of these detectors under various fading models. Having a closed-form outage probability expression enables a system’s packet error rate, i.e., link availability, to be analyzed if the transmitted signals span over one fading block [@lin2013outage]. For the II detector, closed-form outage expressions for Nakagami-$m$ fading [@yao1992investigations] and composite fading consisting of exponentially distributed signal-of-interest (SOI) and squared ${\mathcal{K}}$-distributed interfering signals [@bithas2015mobile] have been noted. It should be pointed out that [@yao1992investigations] and [@bithas2015mobile] are only applicable to specific fading environments and may not be applicable for all aeronautical scenarios where Rician fading is experienced. To this end, a recent paper by Rached et al. [@rached2017unified] presented generalized outage probability expressions that apply to a wide variety of fading scenarios, including Rician fading.
Multiple works on outage expressions for SIC detectors have been noted. For instance, SIC outage expressions were investigated by Hasna et al. [@hasna2003performance] and Romero-Jerez and Goldsmith [@romero2008receive], but these studies only considered partial SIC where at least one interfering signal remains after interference cancellation. A closed-form outage expression for SIC was studied by Weber et al. [@weber2007transmission] for nodes distributed via a Poisson point process. The work in [@weber2007transmission] did not consider fading and receiver noise in the signal model, and thus, the closed-form expressions are not directly applicable for aeronautical communications. A recent paper by Zhang et al. [@zhang2017full] presented outage probability expressions for a two-stage SIC detector. However, the outage expressions are specific for Rayleigh fading scenarios and are not applicable to Rician fading scenarios that are common in aeronautical communications. From the mentioned studies, hitherto closed-form outage probability expressions for SIC detectors in Rician fading aeronautical scenarios remain an open problem.
Apart from outage probability, both finite signal-to-noise ratio (SNR) diversity gain and finite SNR diversity-multiplexing trade-off (DMT) are metrics that can be used to measure the effectiveness of II or SIC detectors in fixed and variable transmission rate systems, respectively. In particular, both finite SNR diversity gain and finite SNR DMT quantifies the slope of outage probability curves at a particular SNR [@shin2008diversity], with the latter considering multiplexing gain [@narasimhan2006finite]. Finite SNR analysis can reveal outage deviation behaviors, which are not present at asymptotically high SNRs due to fading statistics [@shin2008diversity]. From a practical perspective, analyzing outage probability decay rates, i.e., finite SNR diversity gain, provides an accurate picture of a system’s outage performance since wireless communication systems are typically designed to operate at low-to-moderate SNR ranges. It has also been pointed out by Narasimhan [@narasimhan2006finite] that finite SNR diversity gain analysis can be used to estimate the SNR needed to achieve a particular rate of error decay, which can be done through turbo codes or low-density parity-check codes. More crucially, outage probability and diversity gain can be used to gauge the upper and lower limits of a system’s bit error rate performance [@zheng2003diversity; @nabar2005diversity].
Finite SNR analysis for Nakagami-$m$ [@wang2012finite] and Rayleigh fading [@lin2013outage; @yang2015efficient] scenarios have also been studied. However, the conclusions drawn in these studies are specific to Nakagami-$m$ and Rayleigh fading and are not fully applicable for ACS since Rician fading scenarios, typically encountered by ACS, are not considered. Studies on finite SNR analysis for Rician fading channels have been seen. The impact of Rician $K$ factors on outage behavior and finite SNR DMT for multiple-input multiple-output (MIMO) systems was investigated by Narasimhan [@narasimhan2006finite] and Shin et al. [@shin2008diversity]. A recent paper by Heidarpour et al. [@heidarpour2017finite] saw finite SNR DMT analysis being applied to analyze the performance of a network coded cooperative communication system. Despite the noted studies, there is still room for further work on finite SNR DMT analysis for HBD-ACS.
Main Contributions
------------------
The main contributions of this paper are summarized as follows:
- This paper presents closed-form expressions for outage probability, finite SNR diversity gain, and finite SNR DMT for a II detector and a two-stage SIC detector in a Rician fading environment.
- It is shown that the proposed HBD-ACS attains superior outage performance over existing HD-ACS at low SNRs. At high SNRs, however, the outage performance of the proposed HBD-ACS is eclipsed by HD-ACS as the former becomes interference-limited. Nonetheless, we show through numerical simulations that the HBD-ACS can meet typical Quality-of-Service (QoS) requirements, e.g., frame error rate $\leq 10^{-3}$, at high SNRs for a range of interference levels through II and SIC detectors.
- Unlike [@etkin2008gaussian] and [@sirigina2016symbol], the desired and interfering signal levels are related through a scaling parameter. In contrast to the results in [@sirigina2016symbol], it is shown that the asymptotic diversity gain of the SIC detector is zero for all interference levels.
- The HD-ACS is shown to achieve better diversity gain than the proposed HBD-ACS at low multiplexing gains. However, at high multiplexing gains, the HD-ACS achieves zero diversity gain while the proposed HBD-ACS achieves non-zero diversity gain.
Relevance to Related Literature
-------------------------------
In this work, full interference cancellation is assumed for the two-stage SIC detector. This is unlike in [@hasna2003performance] and [@romero2008receive] where only partial SIC is assumed. In addition, the impact of interference on the proposed HBD-ACS is analyzed from the outage probability and finite SNR DMT perspective, which was not covered in [@li2014linear] - [@cirik2018robust], [@yao1992investigations] - [@rached2017unified], [@weber2007transmission] and [@zhang2017full]. In contrast to [@shin2008diversity], [@narasimhan2006finite] and [@heidarpour2017finite], this work extends upon the outage and finite SNR DMT analysis framework to jointly identify interference scenarios for the proposed single-input-single-output equivalent HBD-ACS. The remainder of this paper is organized as follows. The system model is introduced in Section II, with closed-form outage probability expressions at GS and AS-2 presented in Section III. In Section IV, finite SNR diversity gain expressions for both HBD-ACS and HD-ACS are derived and analyzed. Numerical results are then presented in Section V before the conclusion of the paper in Section VI.
System Model
============
\[tpb\] ![Air-Station 1 (AS-1) and Air-Station 2 (AS-2) operating in HD mode while communicating with the FD ground station (GS).[]{data-label="fig:1"}](block_diagram.eps "fig:"){width="0.8\columnwidth"}
In this paper, A/G communications involving an FD-enabled GS node with two HD ASs in an A/G link is studied. Specifically, a scenario with Air-Station 1 (AS-1) transmitting signals to the GS while Air-Station 2 (AS-2) is receiving signals from the GS is assumed. For the HD transceivers at AS-1 and AS-2, a single-antenna configuration with separate transmit and receive radio frequency chains is assumed. In contrast, the FD transceiver at the GS is assumed to be configured with one transmit antenna and one receive antenna through separate radio frequency chains. Due to the fact that the GS node is FD-capable, the HD AS-1 and HD AS-2 simultaneously transmits and receives, respectively, signals on the same aeronautical spectrum (e.g. VHF, L-band) as the GS. Therefore, AS-1 interferes with communications at AS-2 when the latter receives signals from GS.
In this work, an SI mitigation architecture with a shared local oscillator is assumed at the FD-enabled GS. Such a setup enables lower levels of phase noise to be experienced [@syrjala2014analysis], [@li2018self]. As such, we only consider residual SI at the FD-enabled GS as a result of imperfect SI channel estimation and phase noise [@sahai2013impact]. Furthermore, the SI link ($h_{si}$) is modeled as a Rician fading channel to account for passive and active SI mitigation [@ahmed2015all] .[^5] Thus, an II detector is considered at the FD-enabled GS since signal detection is performed in the presence of residual SI.
Rician fading aeronautical communications channels in an en route scenario is assumed to provide a realistic evaluation of the HBD-ACS [@haas2002aeronautical; @matolak2017air_suburban; @yuan2018capacity]. Following the work in [@haas2002aeronautical] and [@yuan2018capacity], the link between AS-1 and AS-2 is also modeled as a Rician fading channel. Accordingly, we assume that the ASs are communicating with the GS at cruising altitude, with the signal model of this work based on [@sahai2013impact]. Also, the effect of Doppler shift is assumed to be compensated in this work [@lee2018uav]. [^6]
Ground Station
--------------
Let $x_1[t]$ and $x_{gs}[t]$ be the signals transmitted by AS-1 and GS, respectively, and $h_{1,g}[t]$ be the channel from AS-1 to GS. Additionally, let $x_{si}[t]$ be the SI signal at GS and let $h_{si}$ be the SI channel gain. From the perspective of GS, $x_{si}[t]=x_{gs}[t]$. The received signal at GS can be written as $$\begin{aligned}
\label{y_gs}
y_{gs}[t] & = & \sqrt{\Omega_{X}}h_{1,g}[t]x_{1}[t] + \sqrt{\Omega_X\alpha_{g,g}} \cdot |\widetilde{h}_{si}|x_{si}[t] \nonumber \\
& & + \sqrt{\Omega_X\alpha_{g,g}}|h_{si}|\gamma_{\phi}w_{\phi}[t] + w_{g}[t],\end{aligned}$$ where $\widetilde{h}_{si}$ is the error of the imperfect SI channel gain estimate, defined as $\widetilde{h}_{si}=h_{si}-\widehat{h}_{si}$, and $\widehat{h}_{si}$ is the imperfect estimation of the SI channel gain. In addition, let $\widetilde{h}_{si}$ be modeled as a zero mean, circularly symmetric complex Gaussian random variable (RV) with variance $\epsilon$ to quantify the SI channel estimation error [@zlatanov2017capacity]. Modeling $\widetilde{h}_{si}$ as a zero mean Gaussian RV with variance $\epsilon$ enables the system to model the worst case residual SI [@zlatanov2017capacity]. Also, let $w_{g}[t]$ be the GS additive white Gaussian noise (AWGN) with zero mean and variance $\sigma^2_{g}$, and let the phase noise term $w_{\phi}[t]$ follow a Gaussian distribution with zero mean and unit variance, scaled by the scaling factor $\gamma_{\phi}$ [@sahai2013impact]. [^7]
Let $\Omega_{X}$ be the average received signal power of the signal-of-interest (SOI). The average received signal power is defined based on the free space path loss model [@goldsmith2005wireless Eq. (2.7)] and it is defined as $$\begin{aligned}
\label{Omega_x_soi}
\Omega_{X} \propto \frac{P_t}{{\left(\frac{4\cdot\pi\cdot{10^9}}{3\cdot10^8}\right)^2}{\cdot}f_{c}^2{\cdot}d_{1,g}^{2}{\cdot}\sigma_g^2}_,\end{aligned}$$ where $P_{t}$, $d$, and $f_{c}$ are the transmit power (Watts), distance (Km), and carrier frequency (MHz), respectively. The received signal power levels are normalized with the receiver noise variance ($\sigma_g^2$). The channel between AS-1 and GS is selected as the reference link and the average received signal power in the other links are expressed relative to the reference link via the multiplicative factor $\alpha_{i,j}$, defined as $$\begin{aligned}
\label{alpha_i_j}
\alpha_{i,j} & = & \bigg(\frac{d_{1,g}}{d_{i,j}}\bigg)^n, i\in\left\{g,1\right\}, j\in\left\{g,2\right\}, i \neq j.\end{aligned}$$ For the case of $\alpha_{g,g}$, the variable $\alpha_{g,g}$ is treated as a scaling factor for the average residual SI power at GS. From (\[Omega\_x\_soi\]) and (\[alpha\_i\_j\]), the average received power of the residual SI at GS can be expressed as $\Omega_X\alpha_{g,g}$, where it is assumed that the residual SI scaled by $\alpha_{g,g}$ is below the saturation level of the FD transceiver.
From [@zlatanov2017capacity], the overall level of SI suppression, i.e., combination of passive suppression with analog and digital SI cancellation, can be calculated as $\frac{1}{\alpha_{g,g}\epsilon\sigma_g^2}$.
Air-Station 2
-------------
Let $h_{g,2}[t]$ be the channel between GS and AS-2, $h_{1,2}[t]$ be the channel between AS-1 and AS-2, and $w_{2}[t]$ be the AWGN at AS-2 with zero mean and variance $\sigma^2_2$. From the perspective of AS-2, $x_{gs}[t]$ and $x_1[t]$ are the SOI and interfering signal, respectively. The received signal at AS-2 can be expressed as $$\begin{aligned}
\label{y_as2}
y_{2}[t] \hspace{-0.05cm} = \hspace{-0.05cm} \sqrt{\Omega_{X}\alpha_{g,2}}h_{g,2}[t]x_{gs}[t] \hspace{-0.05cm} + \hspace{-0.05cm} \sqrt{\Omega_{X}\alpha_{1,2}}h_{1,2}[t]x_{1}[t] \hspace{-0.05cm} + \hspace{-0.05cm} w_{2}[t], \hspace{-0.15cm}\end{aligned}$$ where $\Omega_{X}\alpha_{g,2}$ and $\Omega_{X}\alpha_{1,2}$ indicate the average received signal powers of the SOI and interfering signal, respectively.
To handle the interference at AS-2, two approaches are studied. The first approach assumes an II detector at AS-2. The II detector treats $x_1[t]$ as noise. Therefore, interference is effectively ignored. The second approach assumes a SIC detector at AS-2. The two-stage SIC detector first tries to detect and cancel $x_1[t]$ before proceeding to detect $x_{gs}[t]$ [@narasimhan2007individual].
Calculation of Outage Probabilities
===================================
To begin the outage analysis at GS and AS-2, we first define the HBD transmission rates of AS-1 and GS as $R^{HBD}_{1}$ and $R^{HBD}_{gs}$, respectively, and the sum rate of the HBD system as $R^{HBD}_{sum} = R^{HBD}_{1}+R^{HBD}_{gs}$. Similarly, the HD transmission rates of AS-1 and GS are defined as $R^{HD}_{1}$ and $R^{HD}_{gs}$, respectively, and the sum rate of the HD system is defined as $R^{HD}_{sum} = R^{HD}_{1}+R^{HD}_{gs}$. For fair comparison between HBD and HD systems, $R_{i}^{HBD}=\frac{1}{2}R_{i}^{HD}$ for $ i \in \{1, gs\}$ [@kwon2010optimal; @baranwal2013outage; @sofotasios2017full]. The respective HBD and HD outage probabilities at GS and AS-2 are defined in the following subsections.
Hybrid-Duplex Outage Probability
--------------------------------
The FD-enabled GS receives $x_1[t]$ while simultaneously transmitting $x_{gs}[t]$ in the same time slot. The simultaneous transmission and reception of signals result in strong SI at GS. Let $X_{1}=\Omega_X|h_{1,g}|^2$ be the instantaneous received signal power of the SOI at GS, modeled as a non-centered chi-squared distributed RV with Rician $K$ factor $K_{X_{1}}$. Let $Y_{si,1}=\Omega_X\alpha_{g,g}\gamma_{\phi}^2|h_{si}|^2$ and $Y_{si,2}=\Omega_X\alpha_{g,g}|\widetilde{h}_{si}|^2$ be the instantaneous received signal power corresponding to SI components. In particular, $Y_{si,1}$ is modeled as a non-centered chi-squared distributed RV with Rician $K$ factor $K_{Y_{si,1}}$ and $Y_{si,2}$ is modeled as a exponentially distributed RV.
Concurrently, AS-2 also experiences interference from AS-1. Let $X_{gs} = \Omega_{X}\alpha_{g,2}|h_{g,2}|^2$ and $Y_{1}=\Omega_{X}\alpha_{1,2}|h_{1,2}|^2$ be the instantaneous received signal power of the SOI and interference at AS-2, respectively, where $X_{gs}$ and $Y_{1}$ are independent non-centered chi-squared distributed RV with respective Rician $K$ factors $K_{X_{gs}}$ and $K_{Y_1}$, respectively. Additionally, let $\alpha\big(q,\Omega,K,\gamma\big)$ be defined as $$\begin{aligned}
\label{alpha_func}
\alpha\big(q,\Omega,K,\gamma\big) & \equiv & (-1)^q \exp(-K) \frac{{L_q}^{(0)}(K)}{(1+q)!} \Bigg(\frac{(1+K)}{\Omega}\gamma\Bigg)^{q+1}_,\end{aligned}$$ where $q$, $\Omega$, $K$ and $\gamma$ represent an arbitrary non-negative integer, average received power of the signal, Rician $K$ factor, and threshold, respectively. The function ${L_q}^{(0)}(\bullet)$ represents the $q$-th degree, zero-order Laguerre polynomials [@andras2011generalized] while $\alpha\big(q,\Omega,K,\gamma\big)$ in (\[alpha\_func\]) represents the Rician power cumulative distribution function (CDF) expansion due to Rician faded signal parameters.
### Ground Station
At GS, let the HBD threshold be $\gamma_{th,gs}^{HBD} = 2^{R_{1}^{HBD}}-1$, with HBD outage event $\mathcal{O}_{gs}^{HBD} = \Big\{ h_{1,g}, h_{si} : R_{1}^{HBD} \geq \log_{2}\Big(1 + \frac{X_{1}}{Y_{si,1} + Y_{si,2} + 1}\Big)\Big\}$. By substituting $X_{1}, Y_{si,1}$ and $Y_{si,2}$ into [@rached2017unified Eq. (12)], the closed-form outage probability at GS can be expressed as $$\begin{aligned}
\label{P_out_gs_II}
Pr\big(\mathcal{O}_{gs}^{HBD}\big) & = & \sum_{q\geq0}\sum_{l_1+l_2+l_3=q+1} \alpha\big(q,\Omega_X,K_{X_1},\gamma_{th,gs}^{HBD}\big) \nonumber\\
& & \hspace{1cm} \times \frac{(q+1)!}{l_1! \cdot l_2! \cdot l_3!} E\{Y_{si,1}^{l_1}\} E\{Y_{si,2}^{l_2}\}_,\end{aligned}$$ where $\alpha\big(q,\Omega_X,K_{X_1},\gamma_{th,gs}^{HBD}\big)$ is the Rician SOI power CDF expansion at GS, as defined in (\[alpha\_func\]). In addition, $E\{Y_{si,1}^{l_1}\}$ and $E\{Y_{si,2}^{l_2}\}$ are the $l_1^{th}$ and $l_2^{th}$ moments of $Y_{si,1}$ and $Y_{si,2}$, respectively. From [@rached2017unified Table II], $E\{Y_{si,1}^{l_1}\} = \Gamma(1+l_1) \Big(\frac{\alpha_{g,g}\gamma_{\phi}^2}{1+K_{Y_{si,1}}}\Big)^{l_1}{}_1{F_1}(-l_1,1;-K_{Y_{si,1}})(\Omega_X)^{l_1}$ and $E\{Y_{si,2}^{l_2}\} = \Gamma(1+l_2) (\alpha_{g,g}\epsilon\big)^{l_2} (\Omega_X)^{l_2}$. The function ${}_1{F_1}(\bullet)$ represents the confluent Hypergeometric function [@gradshteyn2014table Eq. (9.210.1)] and summation on the right hand side (RHS) of (\[P\_out\_gs\_II\]) is convergent if $\gamma_{th,gs}^{HBD} \leq \frac{\Omega_X}{3(1+K_{X_{1}})(\Omega_X\alpha_{g,g}\epsilon)}$ [@rached2017unified Eq. (14)]. In (\[P\_out\_gs\_II\]), $E\{Y_{si,1}^{l_1}\}$ and $E\{Y_{si,2}^{l_2}\}$ quantifies the strength of residual SI due to phase noise and SI channel estimation errors, respectively. We do not expect $\alpha_{g,g}$ to approach infinity as the distance on the SI link ($d_{g,g}$) cannot be zero. However, it is possible for the average received SI power to be strong if $d_{g,g}$ is short. From $E\{Y_{si,1}^{l_1}\}$ and $E\{Y_{si,2}^{l_2}\}$, the impact of residual SI is diminished as $\alpha_{g,g} \to 0$ and hence, proper SI mitigation strategies is crucial at the FD-enabled GS.
### Air-Station 2 (Interference Ignorant Detector) {#AS2_II_subsect}
At AS-2, let the HBD threshold be $\gamma_{th,2}^{HBD} = 2^{R_{gs}^{HBD}}-1$ and the HBD outage event be $\mathcal{O}_{2}^{HBD(II)} = \Big\{ h_{g,2}, h_{1,2} : R_{gs}^{HBD} \geq \log_{2}\Big(1+\frac{X_{gs}}{Y_{1} + 1}\Big)\Big\}$. By substituting $X_{gs}$ and $Y_{1}$ into [@rached2017unified Eq. (12)], the closed-form outage probability at AS-2 can be expressed as $$\begin{aligned}
\label{P_out_as2_II}
& & \hspace{-1.5cm} Pr\big(\mathcal{O}_{2}^{HBD(II)}\big) \nonumber \\
& = & \sum_{q\geq0} \sum_{l=0}^{q+1} \alpha\big(q,\Omega_X \alpha_{g,2}, K_{X_{gs}}, \gamma_{th,2}^{HBD}\big) \binom{q+1}{l} E\{Y_1^l\}_,\end{aligned}$$ where $\alpha\big(q,\Omega_X \alpha_{g,2}, K_{X_{gs}}, \gamma_{th,2}^{HBD}\big)$ is the Rician SOI power CDF expansion at AS-2, as defined in (\[alpha\_func\]), and $E\{Y_1^l\}$ is the $l^{th}$ moment of the interfering signal from AS-1. From [@rached2017unified Table II], $E\{Y_1^l\} = \Gamma(1+l) \left[\frac{\alpha_{1,2}}{1+K_{Y_1}}\right]^{l} {}_1{F_1}(-l,1;-K_{Y_1}) (\Omega_X)^{l}$ and the RHS of (\[P\_out\_as2\_II\]) is convergent if $\gamma_{th,2}^{HBD} \leq \frac{\Omega_{X}\alpha_{g,2}(1+K_{Y_{1}})}{2(1+K_{X_{gs}})\Omega_X\alpha_{1,2}}$ [@rached2017unified Eq. (14)]. In (\[P\_out\_as2\_II\]), $E\{Y_1^l\}$ quantifies the strength of the interference from AS-1 through moment parameters of $Y_1$.
To investigate the impact of inter-AS interference, we evaluate $\lim_{\alpha_{1,2} \to L} E\{Y_1^l\}$ for $L \in \{0,\infty\}$. Although $\alpha_{1,2}$ does not reach infinity in practice, large values of $\alpha_{1,2}$ are possible when $d_{1,2}$ is small and vice-versa. Evaluating $\lim_{\alpha_{1,2} \to L} E\{Y_1^l\}$ for $L \in \{0,\infty\}$ shows that the impact of inter-AS interference reduces as $\alpha_{1,2} \to 0$, and increases as $\alpha_{1,2} \to \infty$. Thus, the II detector operates effectively in low interference scenarios such as over remote airspace where inter-AS distance is long.
### Air-Station 2 (Successive Interference Cancellation Detector)
In the case of SIC, if the first stage is unable to detect the interfering signal or if the SOI cannot be detected at the second stage, then outage occurs. Therefore, the HBD outage event at AS-2 is defined as $$\begin{aligned}
\mathcal{O}_{2}^{HBD(SIC)} & = & \bigg\{h_{g,2}, h_{1,2} : R_{1}^{HBD} > \log_{2}\bigg(1+\frac{Y_{1}}{1+X_{gs}}\bigg) \bigg\} \nonumber\\
& \cup & \bigg\{h_{g,2}, h_{1,2} : R_{1}^{HBD} \leq \log_{2}\bigg(1+\frac{Y_{1}}{1+X_{gs}}\bigg) \nonumber \\
& & \hspace{1.6cm} , R_{gs}^{HBD} > \log_{2}\big(1+X_{gs}\big) \bigg\}_.\end{aligned}$$
The closed-form expression for outage probability with SIC detector at AS-2 is $$\begin{aligned}
\label{P_out_as2_SIC}
& & \hspace{-1.25cm} Pr\big(\mathcal{O}_{2}^{HBD(SIC)}\big) \nonumber\\
& = & \sum_{q\geq0}\sum_{l=0}^{q+1} \alpha\big(q,\Omega_X\alpha_{1,2},K_{Y_{1}},\gamma_{th,gs}^{HBD}\big) \binom{q+1}{l} E\{X_{gs}^l\} \nonumber \\
& & + 1 - Q_1\Bigg( \sqrt{2K_{X_{gs}}}, \sqrt{\frac{2(K_{X_{gs}}+1)\gamma_{th,2}^{HBD}}{\Omega_{X}\alpha_{g,2}}} \Bigg) \nonumber\\
& & - \sum_{n\geq0}\sum_{i=0}^{n}\sum_{j=0}^{i+1} \alpha\big(i,\Omega_X\alpha_{1,2},K_{Y_{1}},\gamma_{th,gs}^{HBD}\big) \nonumber\\
& & \times \alpha\big(n-i,\Omega_X\alpha_{g,2},K_{X_{gs}},1\big) \binom{i+1}{j}\frac{(\gamma_{th,2}^{HBD})^{j+n-i+1}}{j+n-i+1}_,\end{aligned}$$ where $Q_1\left(\cdot,\cdot\right)$ is the Marcum Q function [@andras2011generalized], [@simon2005digital Eq. (4.33)] and $E\{X_{gs}^l\}=\Gamma(1+l)\left(\frac{\Omega_{X}\alpha_{g,2}}{1+K_{X_{gs}}}\right)^{l} {}_1{F_1}(-l,1;-K_{X_{gs}})$ [@rached2017unified Table II].
The proof can be found in Appendix \[SIC\_proof\].
The first term in (\[P\_out\_as2\_SIC\]) is the outage probability due to detecting interference from AS-1. The second term in (\[P\_out\_as2\_SIC\]) is the outage probability due to SOI detection after interference cancellation. From $\alpha\big(q,\Omega_X\alpha_{1,2},K_{Y_{1}},\gamma_{th,2}^{HBD}\big)$ in (\[P\_out\_as2\_SIC\]), it is evident that the SIC detector works effectively in high interference scenarios, such as in congested airspace where inter-AS distance is short, since the effect of interference at the SOI detection stage is diminished when $\alpha_{1,2}$ is large. The closed-form expressions in (\[P\_out\_gs\_II\]), (\[P\_out\_as2\_II\]) and (\[P\_out\_as2\_SIC\]) can shed insights into the impact of residual SI at GS and interference from AS-1 at AS-2. Further discussions on outage performance with respect to the level of interference are presented in Section V.
Half-Duplex Outage Probability
------------------------------
When the GS is operating in HD mode, AS-2 does not experience interference from AS-1. Let the HD threshold at GS and AS-2 be defined as $\gamma_{th,gs}^{HD} = 2^{2R_{1}^{HBD}}-1$ and $\gamma_{th,2}^{HD} = 2^{2R_{gs}^{HBD}}-1$, respectively. Then, the HD outage probabilities at GS and AS-2 are given in (\[P\_out\_hd\_gs\]) and (\[P\_out\_hd\_as2\]), respectively [@rached2017unified Table I]. $$\begin{aligned}
Pr\big(\mathcal{O}_{gs}^{HD}\big) & = & \sum_{m\geq0} \alpha\big(m,\Omega_X,K_{X_1},\gamma_{th,gs}^{HD}\big)_, \label{P_out_hd_gs} \\
Pr\big(\mathcal{O}_{2}^{HD}\big) & = & \sum_{m\geq0} \alpha\big(m,\Omega_X\alpha_{g,2},K_{X_{gs}},\gamma_{th,2}^{HD}\big)_. \label{P_out_hd_as2}\end{aligned}$$ The outage probability expressions in (\[P\_out\_hd\_gs\]) and (\[P\_out\_hd\_as2\]) can be used as a benchmark comparison against HBD mode at GS and AS-2, respectively, which is presented in Section V.
System Level Outage Probability
-------------------------------
For the proposed multi-user system, the overall system level outage probability is used as a performance metric to compare HBD and HD protocols. For $\beta \in \{HBD(II), HBD(SIC)\}$, the system level outage probability is defined as $P_{out,system}^{\beta} = \max\left(Pr\big(\mathcal{O}_{gs}^{HBD}\big),Pr\big(\mathcal{O}_{2}^{\beta}\big)\right)$ and $P_{out,system}^{HD} = \max\left(Pr\big(\mathcal{O}_{gs}^{HD}\big),Pr\big(\mathcal{O}_{2}^{HD}\big)\right)$. The system level outage probability provides the worst case system level outage behavior for the II and SIC detectors and allows the identification of performance bottlenecks in HBD-ACS. Having knowledge of the performance bottleneck in the HBD-UCS enables interference management to be more effective. For instance, if the link between the FD-enables GS and AS-2 has the highest outage probability, then the design requirements of the SI mitigation architecture at the FD-enabled GS can be less stringent which can lead to hardware with lower cost or power requirements.
Finite SNR Analysis
===================
In the following subsections, the mathematical preliminaries and derivations related to finite SNR diversity gain are presented for both fixed and variable transmission rates, with detailed derivation omitted for brevity. As it will be shown, finite SNR analysis is an effective tool in evaluating the performance of the II and SIC detectors in an interference-limited environment.
Mathematical Preliminaries
--------------------------
### Finite SNR Diversity Gain
For a system with outage event $\mathcal{O}$, outage probability $Pr\big(\mathcal{O}\big)$, transmission rate $R$, threshold $\gamma$, and average received power $\Omega$ with unit noise variance, the diversity gain $d$ at high SNR is given by Zheng and Tse [@zheng2003diversity] as $$\begin{aligned}
\label{asymp_diversity_gain}
d = \lim_{\Omega\to\infty} \frac{\log_2(Pr\big(\mathcal{O}\big))}{\log_2(\Omega)}_.\end{aligned}$$ The diversity gain definition in (\[asymp\_diversity\_gain\]) is for systems that operate at high SNR ranges. The finite SNR diversity gain $d_f$, which quantifies the decay rate of the outage probability at low-to-moderate SNRs, is given as [@narasimhan2006finite Eq. (5)] $$\begin{aligned}
\label{df}
d_f = \frac{-\Omega}{Pr\big(\mathcal{O}\big)}\frac{\partial}{\partial\Omega}Pr\big(\mathcal{O}\big)_.\end{aligned}$$ It has since been shown by Shin et al. [@shin2008diversity] and Heidarpour et al. [@heidarpour2017finite] that $\lim_{\Omega\to\infty}d_f = d$. Therefore, (\[df\]) is consistent with the asymptotic diversity definitions in [@zheng2003diversity] at high SNR. Practical wireless systems typically operate at the low-to-moderate SNR range [@narasimhan2006finite]. The outage behavior of these systems may also be different at high and moderate SNRs. Therefore, there is motivation to quantify diversity gains at finite SNRs since (\[asymp\_diversity\_gain\]) does not accurately reflect outage behaviors at low-to-moderate SNRs [@shin2008diversity].
### Finite SNR DMT Parameters
For a system which varies its transmission rate with respect to $\Omega$, i.e., variable transmission rate, the high SNR multiplexing gain $r$ is given by Zheng and Tse [@zheng2003diversity] as $$\begin{aligned}
\label{asymp_mult_gain}
r = \lim_{\Omega\to\infty} \frac{R(\Omega)}{\log_2(\Omega)}\end{aligned}$$ and the finite SNR multiplexing gain $r_f$ for such systems is [@narasimhan2006finite Eq. (4)] $$\begin{aligned}
\label{rf}
r_f = \frac{R(\Omega)}{\log_2(1+\Omega)}_.\end{aligned}$$ It has similarly been shown by Shin et al. [@shin2008diversity] and Heidarpour et al. [@heidarpour2017finite] that $\lim_{\Omega\to\infty}r_f = r$, with $Pr\big(\mathcal{O}\big)$ computed with respect to the threshold $\gamma=(1+\Omega)^{r_f}-1$. The finite SNR diversity gain for such a variable transmission rate system (denoted as $d_{f}^{*}$) can be obtained from (\[df\]) as [@shin2008diversity Eq. (36)] $$\begin{aligned}
d_f^* & = & \frac{-\Omega}{Pr\big(\mathcal{O}\big)} \lim_{\Delta(\Omega)\to0} \bigg[\frac{Pr\big(\mathcal{\widehat{O}}\big) - Pr\big(\mathcal{O}\big)}{\Delta(\Omega)}\bigg]_, \label{df_var_rate1} \end{aligned}$$ where $\Delta(\Omega)=\widehat{\Omega}-\Omega$, $\widehat{\Omega}>\Omega$ and $\mathcal{\widehat{O}}$ is the outage event with respect to $R\big(\Omega+\Delta(\Omega)\big)$. Furthermore, $Pr\big(\mathcal{\widehat{O}}\big)$ is the outage probability with average received power $\widehat{\Omega}=\Omega+\Delta(\Omega)$, threshold $\widehat{\gamma}=[1+\Omega+\Delta(\Omega)]^{r_f}-1$ and $Pr\big(\mathcal{\widehat{O}}\big)=Pr\big(\mathcal{O}\big)$ when $\Delta(\Omega)=0$. Applying L’Hospital’s rule in (\[df\_var\_rate1\]) by differentiating with respect to $\Delta(\Omega)$ and setting $\Delta(\Omega)=0$ yields $$\begin{aligned}
d_f^* & = & \frac{-\Omega}{Pr\big(\mathcal{O}\big)} \frac{\partial}{\partial\Delta(\Omega)} Pr\big(\mathcal{\widehat{O}}\big) {\Bigg|_{\Delta(\Omega)=0}}_. \label{df_var_rate2}\end{aligned}$$ Let $Z$ be a RV with normalized $n^{th}$ moment defined as $M\{Z^{n}\}=\frac{E\{Z^{n}\}}{(\Omega)^{n}}$ and let function $g(i,j,\Omega,r_f)$ be $$\begin{aligned}
g(i,j,\Omega,r_f) & = & \big(\big[1 + \Omega\big]^{r_f}-1\big)^{i} (\Omega)^{j} \nonumber \\
& & \hspace{-2cm} \times \bigg[ \left(\big[1 + \Omega\big]^{r_f} - 1\right) \frac{j}{\Omega} + (i+1)(r_f)(1 + \Omega)^{r_f-1} \bigg]_,\end{aligned}$$ where $i$ and $j$ are integers. The function $M\{Z^{n}\}$ represents the normalized $n^{th}$ moment of an interfering signal while the function $g(i,j,\Omega,r_f)$ reflects the outage probability decay rate of a variable transmission rate scheme due to average received power ($\Omega$) and finite SNR multiplexing gain ($r_f$).
Although [@shin2008diversity Eq. (36)] and [@narasimhan2006finite Eq. (5)] evaluate finite SNR diversity gains using different approaches, the principles underlying them are the same since the latter is an extension of the former. To this end, (\[df\_var\_rate2\]) can be used to evaluate $d_f^*$ for adaptive systems, with $r_f$ indicating the sensitivity of the rate adaptation scheme [@narasimhan2006finite]. It is also of interest to analyze $d_f^*$ as it can lead to better code designs that improve transmission rates at the expense of reliability for adaptive systems and vice-versa.
Finite SNR Diversity Gain for HBD Systems
-----------------------------------------
Let the finite SNR HBD diversity gain at GS and AS-2 be defined as $d_{f,gs}^{HBD}$ and $d_{f,2}^{HBD,i}, i \in \{II,SIC\}$, respectively. Additionally, let $R_1$ and $R_{gs}$ be fixed constants with average received power $\Omega = \Omega_X$. Then, the finite SNR diversity gain at GS and AS-2 are presented in the following propositions.
The finite SNR diversity gain at the FD-enabled GS is $$\begin{aligned}
\label{df_fixed_GS}
d_{f,gs}^{HBD} & = & \frac{-\Omega_X}{Pr\big(\mathcal{O}_{gs}^{HBD}\big)} \sum_{q\geq0}\sum_{l_1+l_2+l_3=q+1} \alpha\big(q,1,K_{X_1},\gamma_{th,gs}^{HBD}\big) \nonumber \\
& & \hspace{-1.9cm} \times \frac{(q+1)!(l_1+l_2-q-1)}{l_1! \cdot l_2! \cdot l_3!} M\{Y_{si,1}^{l_1}\} M\{Y_{si,2}^{l_2}\} \left(\Omega_X\right)^{l_1+l_2-q-2}_.\end{aligned}$$
The finite SNR diversity gain at GS can be obtained by substituting (\[P\_out\_gs\_II\]) into (\[df\]).
At low-to-moderate $\Omega_X$, the outage behavior at GS can be analyzed from (\[df\_fixed\_GS\]). In particular, (\[df\_fixed\_GS\]) allows observation of subtle changes in outage behavior due to the scaling factor associated with the SI strength ($\alpha_{g,g}$) and SI channel estimation error ($\epsilon$) that is not present at high $\Omega_X$. In addition, the asymptotic behavior of $d_{f,gs}^{HBD}$ can be obtained from (\[df\_fixed\_GS\]) as shown in the following corollary.
\[coro\_asymp\_df\_fixed\_GS\] The asymptotic behavior of $d_{f,gs}^{HBD}$ is given by $$\begin{aligned}
\label{asymp_df_fixed_GS}
\lim_{\Omega_X\to\infty} \frac{-\Omega_X}{Pr\big(\mathcal{O}_{gs}^{HBD}\big)}\frac{\partial}{\partial\Omega_X}Pr\big(\mathcal{O}_{gs}^{HBD}\big) & = & 0.\end{aligned}$$
The proof is given in Appendix \[coro\_asymp\_df\_fixed\_GS\_proof\].
From (\[asymp\_df\_fixed\_GS\]), $d_{f,gs}^{HBD} \to 0$ as $\Omega_X \to \infty$ because increasing $\Omega_X$ also causes residual SI to be stronger, hence there is no improvement in the overall SINR. Also, (\[asymp\_df\_fixed\_GS\]) suggests that the tolerance for residual SI in HBD-ACS is progressively diminished as $\Omega_X$ is increased since $d_{f,gs}^{HBD} \to 0$ corresponds to negligible improvements in outage probability at GS.
The finite SNR diversity gain at AS-2 with the II ($d_{f,2}^{HBD(II)}$) and SIC detectors ($d_{f,2}^{HBD(SIC)}$) are $$\begin{aligned}
d_{f,2}^{HBD(II)} & \hspace{-0.2cm} = & \hspace{-0.2cm} \frac{-\Omega_X}{Pr\big(\mathcal{O}_{2}^{HBD(II)}\big)} \sum_{q\geq0} \sum_{l=0}^{q+1} \alpha\big(q,\alpha_{g,2}, K_{X_{gs}}, \gamma_{th,2}^{HBD}\big) \nonumber \\
& & \times \binom{q+1}{l} M\{Y_1^l\} (l-q-1) (\Omega_X)^{l-q-2}_, \label{df_fixed_AS2_II} \\
d_{f,2}^{HBD(SIC)} & \hspace{-0.2cm} = & \hspace{-0.2cm} \frac{-\Omega_X}{Pr\big(\mathcal{O}_{2}^{HBD(SIC)}\big)} \Bigg[ \sum_{q\geq0}\sum_{l=0}^{q+1} \alpha\big(q,\alpha_{1,2},K_{Y_{1}},\gamma_{th,gs}^{HBD}\big) \nonumber \\
& & \times \binom{q+1}{l} M\{X_{gs}^l\} (l-q-1) (\Omega_{X})^{l-q-2} \nonumber\\
& & \hspace{-2cm} + \sum_{m\geq0} \alpha\big(m,\alpha_{g,2},K_{X_{gs}},\gamma_{th,2}^{HBD}\big) (-m-1) (\Omega_{X})^{-m-2} \nonumber \\
& & \hspace{-2cm} - \sum_{n\geq0}\sum_{i=0}^{n}\sum_{j=0}^{i+1} \alpha\big(i,\alpha_{1,2},K_{Y_{1}},\gamma_{th,gs}^{HBD}\big) \alpha\big(n-i,\alpha_{g,2},K_{X_{gs}},1\big) \nonumber\\
& & \hspace{-0.3cm} \times \binom{i+1}{j} \frac{\big(\gamma_{th,2}^{HBD}\big)^{j+n-i+1}}{j+n-i+1} (-n-2) (\Omega_X)^{-n-3} \Bigg]_. \label{df_fixed_AS2_SIC}\end{aligned}$$
At AS-2, $d_{f,2}^{HBD(i)}, i \in \{II,SIC\}$ can be obtained for the II and SIC by respectively substituting (\[P\_out\_as2\_II\]) and (\[P\_out\_as2\_SIC\]) into (\[df\]).
The outage behavior at AS-2 can be analyzed from (\[df\_fixed\_AS2\_II\]) and (\[df\_fixed\_AS2\_SIC\]) at low-to-moderate $\Omega_X$. In particular, (\[df\_fixed\_AS2\_II\]) and (\[df\_fixed\_AS2\_SIC\]) enables the observation of subtle changes in outage behavior for both II and SIC detectors, which are not present at high $\Omega_X$, as inter-aircraft interference varies. Extending upon (\[df\_fixed\_AS2\_II\]) and (\[df\_fixed\_AS2\_SIC\]), the asymptotic behavior of $d_{f,2}^{HBD(i)}, i \in \{II,SIC\}$ can be obtained as follows.
\[coro\_asymp\_df\_fixed\_AS2\] The asymptotic behavior of $d_{f,2}^{HBD(i)}, i \in \{II,SIC\}$ is given by $$\begin{aligned}
\label{asymp_df_fixed_AS2}
\lim_{\Omega_X\to\infty} \frac{-\Omega_X}{Pr\big(\mathcal{O}_{2}^{HBD,i}\big)}\frac{\partial}{\partial\Omega_X}Pr\big(\mathcal{O}_{2}^{HBD,i}\big) & = & 0.\end{aligned}$$
The proof is given in Appendix \[coro\_asymp\_df\_fixed\_AS2\_proof\].
*In [@etkin2008gaussian] and [@sirigina2016symbol], the average received signal powers of the desired ($\Omega_x \alpha_{g,2}$) and interfering ($\Omega_x \alpha_{1,2}$) links are related through an exponent, where a large exponent corresponds to very strong interference. At high SNRs, the SIC-based receiver is shown to achieve full diversity under very strong interference levels. In contrast, this work demonstrates that the SIC detector achieves zero diversity gain at high SNRs when the desired and interfering signal levels are related through a scaling parameter.*
In the presence of interference at AS-2, (\[asymp\_df\_fixed\_AS2\]) shows that improvements to outage probability at AS-2 progressively diminishes since $d_{f,2}^{HBD(i)} \to 0$ as $\Omega_X \to \infty$ for $i \in \{II,SIC\}$. For the II detector, increasing $\Omega_X$ results in strong interference. As a consequence, there is no improvement to the overall SINR. Hence, the II detector is unsuitable in strong interference environments. Similarly, for the SIC detector, increasing $\Omega_X$ causes $x_{gs}[t]$ to be stronger, making the detection and subtraction of $x_1[t]$ increasingly challenging at stage 1 of the SIC detector. Hence, $\alpha_{1,2}$ must either increase (for the II detector) or decrease (for the SIC detector) at high $\Omega_X$ for HBD-ACS to see meaningful improvements in outage probability.
Finite SNR Diversity Gain for HD Systems
----------------------------------------
Let the finite SNR diversity gain at GS and AS-2 be defined as $d_{f,i}^{HD}, i \in \{gs,2\}$, respectively, with $R_1$ and $R_{gs}$ assumed to be constants with average received power $\Omega = \Omega_X$. Then, the finite SNR diversity gain at GS and AS-2 are presented in the following proposition.
The finite SNR diversity gain at GS and AS-2 operating in HD mode are given in (\[df\_fixed\_hd\_GS\]) and (\[df\_fixed\_hd\_as2\]), respectively. $$\begin{aligned}
d_{f,gs}^{HD} & = & \frac{-\Omega_X}{Pr\big(\mathcal{O}_{gs}^{HD}\big)} \sum_{m\geq0} \alpha\big(m,1,K_{X_1},\gamma_{th,gs}^{HD}\big) \nonumber \\
& & \hspace{2.7cm} \times (-m-1) (\Omega_{X})^{-m-2}_, \label{df_fixed_hd_GS}\\
d_{f,2}^{HD} & = & \frac{-\Omega_X}{Pr\big(\mathcal{O}_{2}^{HD}\big)} \sum_{m\geq0} \alpha\big(m,\alpha_{g,2},K_{X_{gs}},\gamma_{th,2}^{HD}\big) \nonumber \\
& & \hspace{2.7cm} \times (-m-1) (\Omega_{X})^{-m-2}_. \label{df_fixed_hd_as2}\end{aligned}$$
The expressions in (\[df\_fixed\_hd\_GS\]) and (\[df\_fixed\_hd\_as2\]) can be obtained by respectively substituting (\[P\_out\_hd\_gs\]) and (\[P\_out\_hd\_as2\]) into (\[df\]).
The HD outage behavior at GS and AS-2 can be analyzed from (\[df\_fixed\_hd\_GS\]) and (\[df\_fixed\_hd\_as2\]), respectively, and it enables the observation of changes in outage probability decay rate that is not visible at high $\Omega_X$. As $\Omega_X \to \infty$, (\[df\_fixed\_hd\_GS\]) and (\[df\_fixed\_hd\_as2\]) can be evaluated to determine the asymptotic diversity gain as follows.
\[coro\_lim\_df\_fixed\_hd\] The asymptotic behavior of $d_{f,i}^{HD}, i \in \{gs,2\}$ is $$\begin{aligned}
\label{lim_df_fixed_hd}
\lim_{\Omega_X \to \infty} d_{f,i}^{HD} & = & 1. \end{aligned}$$
The proof is provided in Appendix \[coro\_lim\_df\_fixed\_hd\_proof\].
From (\[lim\_df\_fixed\_hd\_GS\]), $d_{f,i}^{HD} \to 1$ as $\Omega_X \to \infty$ for $i \in \{gs,2\}$ and it indicates that the HD system achieves full diversity in the absence of interference at high $\Omega_X$, which is consistent with [@shin2008diversity Fig. 3].
Finite SNR DMT Analysis for HBD Systems
---------------------------------------
Let the finite SNR diversity gain at GS for a HBD system be defined as $d_{f,gs}^{HBD*}$, with variable transmission rate $R_1^{HBD}(\Omega_X)=r_f\log_2(1+\Omega_X)$ and threshold $\gamma_{th,gs}^{HBD} = [1+\Omega_X]^{r_f}-1$. Similarly, let the finite SNR diversity gain at AS-2 for a HBD system be denoted as $d_{f,2}^{HBD(i)*}, i \in \{II,SIC\}$, with variable transmission rate $R_{gs}^{HBD}(\Omega_X)=r_f\log_2(1+\Omega_X)$ and threshold $\gamma_{th,2}^{HBD} = [1+\Omega_X]^{r_f}-1$. The finite SNR diversity gains at GS and AS-2 are presented in the following propositions.
At GS, the finite SNR diversity gain is given as $$\begin{aligned}
\label{df_var_gs}
d_{f,gs}^{HBD*} & = & \frac{-\Omega_X}{Pr\big(\mathcal{O}_{gs}^{HBD}\big)} \sum_{q\geq0}\sum_{l_1+l_2+l_3=q+1} \alpha\big(q,1,K_{X_1},1\big) \nonumber \\
& & \hspace{-2.4cm} \times \frac{(q+1)!}{l_1! \cdot l_2! \cdot l_3!} M\{Y_{si,1}^{l_1}\} M\{Y_{si,2}^{l_2}\} g(q,l_1+l_2-q-1,\Omega_X,r_f).\end{aligned}$$
Let $\Omega=\Omega_X$, $\widehat{\Omega}=\widehat{\Omega}_X$ and $\gamma = \gamma_{th,gs}^{HBD}=[1+\Omega_X]^{r_f}-1$ and $\mathcal{O} = \mathcal{O}_{gs}^{HBD}$. Then, $d_{f,gs}^{HBD*}$ can be obtained through algebraic manipulations by substituting (\[P\_out\_gs\_II\]) into (\[df\_var\_rate2\]).
At AS-2, the finite SNR diversity gain with II and SIC detectors are $$\begin{aligned}
d_{f,2}^{HBD(II)*} & = & \frac{-\Omega_X}{Pr\big(\mathcal{O}_{2}^{HBD(II)}\big)} \sum_{q\geq0} \sum_{l=0}^{q+1} \alpha\big(q,\alpha_{g,2}, K_{X_{gs}}, 1\big) \nonumber \\
& & \times \binom{q+1}{l} M\{Y_1^l\} g(q,l-q-1,\Omega_X,r_f)_, \label{df_var_AS2_II} \\
d_{f,2}^{HBD(SIC)*} & = & \frac{-\Omega_X}{Pr\big(\mathcal{O}_{2}^{HBD(SIC)}\big)} \Bigg[ \sum_{q\geq0}\sum_{l=0}^{q+1} \alpha\big(q,\alpha_{1,2},K_{Y_{1}},1\big) \nonumber \\
& & \times \binom{q+1}{l} M\{X_{gs}^l\} g(q,l-q-1,\Omega_X,r_f) \nonumber\\
& & \hspace{-2cm} + \sum_{m\geq0} \alpha\big(m,\alpha_{g,2},K_{X_{gs}},1\big) g(m,-m-1,\Omega_X,r_f) \nonumber \\
& & \hspace{-2cm} - \sum_{n\geq0}\sum_{i=0}^{n}\sum_{j=0}^{i+1} \alpha\big(i,\alpha_{1,2},K_{Y_{1}},1\big) \alpha\big(n-i,\alpha_{g,2},K_{X_{gs}},1\big) \nonumber\\
& & \hspace{-0.6cm} \times \frac{\binom{i+1}{j}}{j+n-i+1} g(j+n+1,-n-2,\Omega_X,r_f) \Bigg]_. \label{df_var_AS2_SIC}\end{aligned}$$
Let $\Omega=\Omega_X$, $\widehat{\Omega}=\widehat{\Omega}_X$ and $\gamma=\gamma_{th,2}^{HBD}=[1+\Omega_X]^{r_f}-1$ and $\mathcal{O} = \mathcal{O}_{2}^{HBD,i}, i \in \{II,SIC\}$. Then, similar to (\[df\_var\_gs\]), $d_{f,2}^{HBD(II)*}$ and $d_{f,2}^{HBD(SIC)*}$ can be obtained through algebraic manipulations by respectively substituting (\[P\_out\_as2\_II\]) and (\[P\_out\_as2\_SIC\]) into (\[df\_var\_rate2\]).
In the presence of interference at GS and AS-2, DMT at low-to-moderate $\Omega_X$ can be analyzed from (\[df\_var\_gs\]), (\[df\_var\_AS2\_II\]) and (\[df\_var\_AS2\_SIC\]). It reveals the interference scenarios in which the II or SIC detectors achieves better diversity gain than HD systems.
Finite SNR DMT Analysis for HD Systems
--------------------------------------
Let the finite SNR HD diversity gain at GS be defined as $d_{f,gs}^{HD*}$. To ensure fair comparison, we let the variable HD date rate be twice the variable HBD data rate. Let $R_1^{HD}(\Omega_X)=2r_f\log_2(1+\Omega_X)$ be the variable transmission rate at AS-1 with threshold $\gamma_{th,gs}^{HD} = [1+\Omega_X]^{2r_f}-1$. Similarly at AS-2, let the finite SNR HD diversity gain at AS-2 be defined as $d_{f,2}^{HD*}$ with variable transmission rate $R_{gs}^{HD}(\Omega_X)=2r_f\log_2(1+\Omega_X)$ and threshold $\gamma_{th,2}^{HD} = [1+\Omega_X]^{2r_f}-1$. The closed-form expressions for the finite SNR diversity gains at GS and AS-2 are presented in the following proposition.
For a variable transmission rate scheme, the finite SNR diversity gain at GS and AS-2 are given in (\[df\_var\_gs\_HD\]) and (\[df\_var\_AS2\_HD\]), respectively. $$\begin{aligned}
d_{f,gs}^{HD*} & = & \frac{-\Omega_X}{Pr\big(\mathcal{O}_{gs}^{HD}\big)} \sum_{m\geq0} \alpha\big(m,1,K_{X_1},1\big) \nonumber \\
& & \hspace{2cm} \times g(m,-m-1,\Omega_X,2r_f)_, \label{df_var_gs_HD} \\
d_{f,2}^{HD*} & = & \frac{-\Omega_X}{Pr\big(\mathcal{O}_{2}^{HD}\big)} \sum_{m\geq0} \alpha\big(m,\alpha_{g,2},K_{X_{gs}},1\big) \nonumber \\
& & \hspace{2cm} \times g(m,-m-1,\Omega_X,2r_f)_. \label{df_var_AS2_HD}\end{aligned}$$
The expressions in (\[df\_var\_gs\_HD\]) and (\[df\_var\_AS2\_HD\]) can be obtained through algebraic manipulations by respectively substituting (\[P\_out\_hd\_gs\]) and (\[P\_out\_hd\_as2\]) into (\[df\_var\_rate2\]).
In the absence of interference, (\[df\_var\_gs\_HD\]) and (\[df\_var\_AS2\_HD\]) can be used to evaluate the DMT at GS and AS-2, providing a benchmark that can be used in evaluating the performance of the II and SIC detectors in HBD systems.
System Level Finite SNR Diversity Gain and DMT
----------------------------------------------
The system level finite SNR diversity gain and DMT for the multi-user system in Fig. \[fig:1\] will be used as a metric to compare HBD and HD systems. For fixed transmission rate schemes, the HBD and HD system level finite SNR diversity gain are defined as $d_{f,system}^{\beta} = \min\left( d_{f,gs}^{HBD}, d_{f,2}^{\beta} \right)$ and $d_{f,system}^{HD} = \min\left( d_{f,gs}^{HD}, d_{f,2}^{HD} \right)$, respectively. Similarly, for variable transmission rate schemes, the HBD and HD system level finite SNR DMT are defined as $d_{f,system}^{\beta*} = \min\left( d_{f,gs}^{HBD*}, d_{f,2}^{\beta*} \right)$ and $d_{f,system}^{HD} = \min\left( d_{f,gs}^{HD*}, d_{f,2}^{HD*} \right)$, respectively. Quantifying the finite SNR diversity gain and DMT provides insights into the degree of improvements in outage performance at the system level, which will be further discussed in Section V.
Numerical Results
=================
In this section, numerical results pertaining to the outage probabilities and finite SNR diversity gains at GS, AS-2 and system level are discussed. Monte Carlo simulations are conducted with $10^{9}$ samples to verify the accuracy of the outage probability computations. In addition, all Rician $K$ factors are fixed at 15, i.e., $K_{X_1}=K_{Y_{si,1}}=K_{X_{gs}}=K_{Y_1}=15$ [@matolak2017air_suburban], with $\sigma_g^2 = \sigma_2^2=-115$ dBm [^8] [@itu2011m2233], $R_{sum}^{HD}=R_{sum}^{HBD}=1$ for fair comparison between the HBD and HD systems. Furthermore, a phase noise strength of $\gamma_{\phi}^2=-130$ dBm is chosen and is subsequently normalized by $\sigma_g^2 = -115$ dBm in the simulations. [^9] At the FD-enabled GS, SI suppression levels of $163$ dB to $175$ dB are considered by choosing $\alpha_{g,g}=\{1, 1.5\}$ and $\epsilon = \{0.01, 0.001\}$. Finally, the subsequent analysis in this section is conducted for $0\text{ dB} \leq \Omega_X \leq 30\text{ dB}$. [^10]
Finite SNR Diversity Gain and Outage Analysis
---------------------------------------------
###
*The FD-enabled GS has near-ideal outage probability and diversity gain at very low SNR, and is interference-limited at high SNR.*
The HBD outage probability at GS is shown in Fig. \[fig:fixed\_pout\_gs\]. The ideal HBD and the HD outage probability are also plotted in Fig. \[fig:fixed\_pout\_gs\] as a benchmark comparison. From Fig. \[fig:fixed\_pout\_gs\], it can be seen that $Pr\big(\mathcal{O}_{gs}^{HBD}\big)$ is close to the ideal HBD case, i.e., no interference, at low-to-moderate average received power ($\Omega_X$) and vice-versa. As expected, $Pr\big(\mathcal{O}_{gs}^{HBD}\big)$ is higher as SI channel estimation error ($\epsilon$) is increased. In addition, increasing the strength of the residual SI ($\alpha_{g,g}$) degrades the outage performance more than the increase in $\epsilon$ since a larger $\alpha_{g,g}$ corresponds to a higher average residual SI power, with phase noise ($\gamma^2_{\phi}$) scaled accordingly. In fact, $Pr\big(\mathcal{O}_{gs}^{HBD}\big)$ approaches the ideal HBD case when $\alpha_{g,g}=1, \epsilon=0.001$ at low $\Omega_X$ in Fig. \[fig:fixed\_pout\_gs\]. Hence, sufficient SI mitigation is needed in order for the FD-enabled GS to outperform the HD-enabled GS.
The finite SNR diversity gain at GS is shown in Fig. \[fig:fixed\_df\_gs\], where it can be seen that $d_{f,gs}^{HBD}$ peaks at $\Omega_X=2$ dB while $d_{f,gs}^{HD}$ peaks at $\Omega_X=6$ dB. [^11] Additionally, (\[asymp\_df\_fixed\_GS\]) and (\[lim\_df\_fixed\_hd\]) are also confirmed in Fig. \[fig:fixed\_df\_gs\] as $\Omega_X \to \infty$ and is also corroborated in Fig. \[fig:fixed\_pout\_gs\], where the slope of the outage probability curves become constant as $\Omega_X \to \infty$. In other words, the FD-enabled GS becomes interference-limited at high $\Omega_X$. Interestingly, in the absence of interference at the FD-enabled GS, $d_{f,gs}^{HBD} \to 1$ as $\Omega_X \to \infty$ since only SNR needs to be considered at GS. From Fig. \[fig:gs\_outage\_diversity\_gain\], residual SI is the performance limiting factor for the FD-enabled GS. Therefore, it is important to sufficiently mitigate SI at each of the cascaded stages in Fig. \[fig:1\] in order to keep the strength of the residual SI low for effective operation of the FD-enabled GS.
###
*The II and SIC detectors achieve lower outage probability and higher diversity gain than HD-mode at low SNR regimes and are interference-limited at high SNR regimes. For the SIC detector, strong interference at low SNR regime enables easy removal of the interfering signal.*
The HBD outage probabilities at AS-2 for both II and SIC detectors are shown in Fig. \[fig:fixed\_pout\_as2\]. It can be seen that the II detector at AS-2 outperforms the HD mode at low-to-moderate $\Omega_X$ when inter-AS interference ($\alpha_{1,2}$) is weak. The trend in Fig. \[fig:fixed\_pout\_as2\] also suggests that the further reduction in $\alpha_{1,2}$ will enable the II detector at AS-2 to attain the ideal HBD outage performance for moderate $\Omega_X$, which is expected since $\alpha_{1,2}\to 0$ corresponds to diminishing levels of interference at AS-2.
The SIC detector performs better than the HD mode at the low-to-moderate $\Omega_X$ when interference is strong, e.g., $\alpha_{1,2}=10$, since stage 1 of the SIC detector is more likely to detect and subtract $x_1[t]$. The resultant signal at stage 2 of the SIC detector is thus almost interference-free. As $\alpha_{1,2}$ increases, the SIC detector performance approaches that of the ideal HBD case due to the near perfect cancellation of interference in the first stage. When $\Omega_X>10$ dB for $\alpha_{1,2}\in\{5,10\}$, an error floor is present which verifies Corollary \[coro\_asymp\_df\_fixed\_AS2\]. Similar error floor observations are also made for the II detector and it indicates that the II and SIC detectors become interference-limited at high $\Omega_X$. From a practical perspective, the trend in Fig. \[fig:fixed\_pout\_as2\] shows that the II detector is well suited for en route scenarios with less congested flight routes such as those over sparsely populated or oceanic regions since the II detector experiences weak interference due to path loss as a result of large inter-aircraft or aircraft to GS distance. On the other hand, the SIC detector is suitable for use in congested airspace scenarios such as the landing or even continental en route scenarios as interference from nearby aircrafts can be effectively removed. Although HD-ACS has superior outage performance compared to the II and SIC detectors at high $\Omega_X$, the interference-limited HBD detectors can meet typical QoS requirements, e.g., frame error rate $\leq 10^{-3}$.
The finite SNR diversity gains, $d_{f,2}^{HBD(II)}$ and $d_{f,2}^{HBD(SIC)}$, at AS-2 are shown in Fig. \[fig:fixed\_df\_as2\]. A trend similar to what was seen in Fig. \[fig:fixed\_df\_gs\] can be found in Fig. \[fig:fixed\_df\_as2\], with $d_{f,2}^{HBD(II)}$ and $d_{f,2}^{HBD(SIC)}$ peaking at $\Omega_X=2$ dB, and $d_{f,2}^{HD}$ peaking at $\Omega_X=6$ dB. As expected, reducing $\alpha_{1,2}$ causes $d_{f,2}^{HBD(II)}$ to perform close to the ideal HBD case at low $\Omega_X$. Fig. \[fig:fixed\_df\_as2\] also confirms (\[asymp\_df\_fixed\_AS2\]) for both the II and SIC detectors. It is clear that the SIC detector can attain an outage probability decay rate that is similar to the ideal HBD case when $\Omega_X\leq5$ dB. Further increasing $\alpha_{1,2}$ will enable $d_{f,2}^{HBD(SIC)}$ to be almost identical to the ideal HBD case at $\Omega_X\leq5$ dB since the system becomes noise-limited rather than interference-limited. The trends in Fig. \[fig:fixed\_df\_as2\] are also reflected in Fig. \[fig:fixed\_pout\_as2\] since the slope of the outage probability curves behave as indicated in (\[asymp\_df\_fixed\_AS2\]) and (\[lim\_df\_fixed\_hd\]) as $\Omega_X \to \infty$.
###
*The system level performance of the HBD-ACS is constrained by inter-AS interference. When the II detector is considered, weak inter-AS interference enables near-ideal system level performance. Likewise for the SIC detector when strong inter-AS interference is present. If the SI suppression level is lower, then it is possible for the GS to be the bottleneck.*
Fig. \[fig:fixed\_pout\_sys\] and Fig. \[fig:fixed\_df\_sys\] respectively shows the outage probability and finite SNR diversity gain at the system level. Through numerical analysis, we observed that $P_{out,system}^{HBD(II)}$ is dominated by the II detector at AS-2 for $\alpha_{1,2} \in \{0.1, 0.5\}$ and $0 \text{ dB} \leq \Omega_X \leq 30 \text{ dB}$, i.e., $Pr\big(\mathcal{O}_{gs}^{HBD}\big)<Pr\big(\mathcal{O}_{2}^{HBD(II)}\big)$ because inter-AS interference at AS-2 is stronger than the residual SI experienced at GS. Thus, although not shown in the figure, increasing $\alpha_{g,g}$ or $\epsilon$ does not affect $P_{out,system}^{HBD(II)}$ unless inter-AS interference is decreased. It can also be observed from Fig. \[fig:fixed\_pout\_sys\] that $P_{out,system}^{HBD(II)} \leq P_{out,system}^{HD}$ when $\Omega_X\leq4$ dB, $\alpha_{1,2}=0.5$. When $\alpha_{1,2}=0.1$, $P_{out,system}^{HBD(II)} \leq P_{out,system}^{HD}$ for $\Omega_X\leq11$ dB. In fact, $P_{out,system}^{HBD(II)}$ approaches that of the ideal HBD case when $\alpha_{1,2}$ is decreased due to the near absence of inter-AS interference at the II detector and it also explains the trend seen in Fig. \[fig:fixed\_df\_sys\] where it can be seen that $d_{f,system}^{HBD(II)}$ approaches that of the ideal HBD case when $\alpha_{1,2}$ is decreased. In other words, the decay of $P_{out,system}^{HBD(II)}$ approaches that of the ideal HBD case when inter-AS interference weakens, as reflected in Fig. \[fig:fixed\_df\_sys\], for $\Omega_X \leq 5$ dB. Therefore, when an II detector is used at AS-2, the inter-AS interference is the limiting factor for both $P_{out,system}^{HBD(II)}$ and $d_{f,system}^{HBD(II)}$.
When AS-2 adopts an SIC detector, $P_{out,system}^{HBD(SIC)}$ is dominated by GS when $\Omega_X\leq4$ dB and $\alpha_{1,2}=5$. Similar trends for the SIC detector are also seen in Fig. \[fig:fixed\_df\_sys\] for $\Omega_X \leq 5$ dB. When $\Omega_X>4$ dB, $P_{out,system}^{HBD(SIC)}$ is dominated by AS-2 and it can be explained from the perspective of the two-stage SIC detector at AS-2. When $\alpha_{1,2}=5$, $x_1[t]$ is five times stronger than the SOI from GS ($x_{gs}[t]$). In addition, at stage 1 of the SIC detector, noise power ($\sigma^2_{2}$) is stronger than $x_{gs}[t]$ when $\Omega_X\leq4$ dB. Thus, the SIC detector is more likely to detect and cancel $x_1[t]$ which results in $Pr\big(\mathcal{O}_{gs}^{HBD}\big)>Pr\big(\mathcal{O}_{2}^{HBD(SIC)}\big)$ due to residual SI at GS. When $\Omega_X>4$ dB, $\sigma_2^2$ will be weaker than $x_{gs}[t]$ at stage 1 of the SIC detector. Consequently, the SIC detector is less likely to detect and cancel $x_1[t]$, leading to $Pr\big(\mathcal{O}_{gs}^{HBD}\big)<Pr\big(\mathcal{O}_{2}^{HBD(SIC)}\big)$. When $\alpha_{1,2}=10$, $P_{out,system}^{HBD(SIC)}$ is dominated by GS for $\Omega_X \leq 10$ dB due to stronger interference at AS-2, with $P_{out,system}^{HBD(SIC)}$ close to that of the ideal HBD case. Further increasing $\alpha_{1,2}$ enables $P_{out,system}^{HBD(SIC)}$ to reach near-ideal HBD performance for a wider $\Omega_X$ range due to the increased likelihood of successfully detecting and canceling $x_1[t]$, thus explaining the trend in Fig. \[fig:fixed\_df\_sys\]. Hence, the strength of the interference from AS-1 ($\alpha_{1,2}$) is the main limiting factor for both $P_{out,system}^{HBD(SIC)}$ and $d_{f,system}^{HBD(SIC)}$ when a SIC detector is used at AS-2.
From Fig. \[fig:fixed\_pout\_sys\] and Fig. \[fig:fixed\_df\_sys\], the outage and finite SNR diversity gain analysis has highlighted the feasibility of HBD-ACS over legacy HD-ACS in weak and strong interference scenarios through the II and SIC detectors, respectively. For instance, weak and strong interference scenarios could involve en route flights over sparely and densely populated airspace, respectively. From the aeronautical perspective, the proposed HBD-ACS has better reliability over HD-ACS while providing more throughput than legacy HD systems.
Finite SNR DMT Analysis
-----------------------
###
*The FD-enabled GS achieves non-zero diversity gain for a larger range of multiplexing gains compared to the HD GS.*
Fig. \[fig:var\_df\_gs\] shows the finite SNR diversity gain at GS, where it is evident that the stronger residual SI due to SI channel estimation error ($\epsilon$) or phase noise ($\gamma_{\phi}^2$) reduces $d_{f,gs}^{HBD*}$. Increasing the strength of the residual SI ($\alpha_{g,g}$) affects $d_{f,gs}^{HBD*}$ more than increasing the SI channel estimation error ($\epsilon$) since the effect of phase noise ($\gamma_{\phi}^2$) on the residual SI is amplified. From the outage probability perspective, increasing residual SI results in a slower decay rate of the outage probability, which lowers $d_{f,gs}^{HBD*}$. However, it does not imply that outage probability is better when a higher maximum value for $d_{f,gs}^{HBD*}$ is attained. Nonetheless, the range of $r_f$ for which $d_{f,gs}^{HBD*}\geq d_{f,gs}^{HD*}$ increases when the strength of the residual SI ($\alpha_{g,g}$) decreases and vice versa. Therefore, FD-enabled GS can experience improved DMT as residual SI decreases, which is evident in Fig. \[fig:var\_df\_gs\] for $\alpha_{g,g}=1$. Although $d_{f,gs}^{HBD*}$ is limited by residual SI, the importance of proper SI mitigation is again emphasized since it is still feasible for GS to be FD-enabled if operating at a higher $r_f$ is the objective of an ACS.
###
*At low multiplexing gains, the II and SIC detectors have lower finite SNR diversity gain. In contrast, at high multiplexing gains, the II and SIC detectors achieve near-ideal finite SNR diversity gain under weak and strong inter-AS interference, respectively.*
Fig. \[fig:var\_df\_as2\] shows the finite SNR diversity gain at AS-2. The trends seen in Fig. \[fig:var\_df\_as2\] are similar to what was seen in [@narasimhan2006finite Fig. 4], with lower $d_{f,2}^{HBD(i)*}, i \in \{II,SIC\}$ and $d_{f,2}^{HD*}$ observed as $r_f \to 0$. It has been pointed out by Narasimhan [@narasimhan2006finite] and Shin et al. [@shin2008diversity] that Rician fading outage probability curves are influenced by Rician $K$ factors. In particular, increasing the Rician $K$ factor causes the slope of outage probability curves to become steeper [@shin2008diversity Fig. 2]. From a finite SNR DMT perspective, $r_f \to 0$ causes $K_{X_{gs}}$ to have less impact on the outage performance at AS-2.
On the other hand, Fig. \[fig:var\_df\_as2\] also suggests that the II and SIC detectors are able to provide better reliability at higher multiplexing gains compare to HD systems. At high multiplexing gains, if the inter-AS interference reduces, then $d_{f,2}^{HBD(II)*} \geq d_{f,2}^{HD*}$. On the other hand, at low multiplexing gains, $d_{f,2}^{HBD(II)*} < d_{f,2}^{HD*}$ even at low inter-AS interference. In fact, $d_{f,2}^{HBD(II)*}$ approaches that of the ideal HBD case as $\alpha_{1,2} \to 0$ since the signal at the II detector is almost interference-free. As a consequence, the resultant outage probability decay rate becomes similar to that of the ideal HBD case. When a SIC detector is adopted at AS-2, $d_{f,2}^{HBD(SIC)*} \geq d_{f,2}^{HD*}$ as inter-AS interference increases (for example, refer to $d_{f,2}^{HBD(SIC)*}$ at $\alpha_{1,2} = 14.3$ in Fig. \[fig:var\_df\_as2\]). As $\alpha_{1,2}\to\infty$, it becomes easier to detect and remove $x_1[t]$ at the two-stage SIC detector. When coupled with the lower threshold requirement of the SIC detector, as compared to HD systems, the SIC detector can potentially achieve superior diversity gains over HD systems in strong interference scenarios. Moreover, at large values of $\alpha_{1,2}$, if the multiplexing gain is high, the achievable $d_{f,2}^{HBD(SIC)*}$ matches the ideal HBD case. As shown in Fig. \[fig:var\_df\_as2\], at low multiplexing gain, the achievable $d_{f,2}^{HBD(SIC)*}$ is close to that of the ideal HBD case. Therefore, the II and SIC detectors provides better reliability at higher multiplexing gains compared to HD-ACS in the presence of weak and strong interference, respectively. However, at low multiplexing gains, HD-ACS exhibited better reliability than the II and SIC detectors.
\[\] ![System level finite SNR DMT (II and SIC detectors) for $\alpha_{g,2}=1$, $\alpha_{g,g}=1$, $\gamma_{\phi}^2=-130 \text{ dBm}$, $\Omega_X=10 \text{ dB}$.[]{data-label="fig:var_df_sys"}](var_df_sys.eps "fig:"){width="0.9\columnwidth"}
###
*At high multiplexing gains, the HBD-ACS achieves better finite SNR diversity gain at the system level than the HD-ACS and is also constrained by inter-AS interference and residual SI.*
Fig. \[fig:var\_df\_sys\] shows the system level finite SNR diversity gain for HBD-ACS ($d_{f,system}^{\beta*}$) and HD-ACS ($d_{f,system}^{HD*}$) for $\beta \in \{HBD(II), HBD(SIC)\}$. From Fig. \[fig:var\_df\_sys\], it is evident that $d_{f,system}^{HBD(II)*} > d_{f,system}^{HD*}$ and $d_{f,system}^{HBD(SIC)*} > d_{f,system}^{HD*}$ as $r_f$ increases, and it enables an HBD-ACS to provide better reliability at higher multiplexing gain than HD-ACS since HBD-ACS requires a lower operating threshold than existing HD-ACS at both GS and AS-2. However, the degree of improvement that HBD-ACS has over HD-ACS is constrained by the strength of interference experienced at GS and AS-2 in the HBD-ACS.
When the II detector is adopted at AS-2 for weak interference scenarios, $d_{f,gs}^{HBD*} > d_{f,2}^{HBD(II)*}$ for $\alpha_{1,2}=0.1$. Reducing the strength of the inter-AS interference ($\alpha_{1,2}=0.01$) causes $d_{f,2}^{HBD(II)*} > d_{f,gs}^{HBD*}$, with lower SI channel estimation error ($\epsilon $) corresponding to higher $d_{f,system}^{HBD(II)*}$. In the presence of strong interference at AS-2 ($\alpha_{1,2}=100$), adopting the SIC detector at AS-2 results in $d_{f,2}^{HBD(SIC)*} > d_{f,gs}^{HBD*}$. However, when interference from AS-1 is not as strong, e.g., $\alpha_{1,2} \in \{14.3, 15\}$, then $d_{f,gs}^{HBD*} > d_{f,2}^{HBD(SIC)*}$. From Fig. \[fig:var\_df\_sys\], the reliability of the HBD-ACS depends on the inter-AS interference at AS-2 for both II and SIC detectors and residual SI at GS. Furthermore, it is possible for the proposed HBD-ACS to attain finite SNR DMT curves that are identical to the ideal HBD case at sufficiently low residual SI.
From Fig. \[fig:var\_df\_sys\], the trends show that the proposed HBD-ACS is a viable alternative to legacy HD-ACS in weak and strong interference scenarios. In particular, the proposed HBD-ACS can operate at a higher multiplexing gain than legacy HD-ACS, thus offering better throughput and reliability compared to the latter.
Conclusion
==========
An HBD-ACS consisting of an FD-enabled GS and two HD ASs simultaneously communicating on the same spectrum is proposed to improve spectrum utilization. To investigate the impact of interference on the proposed HBD-ACS, closed-form outage probability and finite SNR diversity gain expressions are presented in this paper for a SIC detector over Rician fading aeronautical channels. Through outage and finite SNR diversity gain analysis, it is established that residual SI is the main limiting factor at the FD-enabled GS. Therefore, the need for sufficient SI mitigation must be properly addressed in a HBD-ACS. At AS-2, inter-AS interference is the main limiting factor for both II and SIC detectors. At the system level, the proposed HBD-ACS is found to be very suitable for weak and strong interference scenarios for the II and SIC detectors, respectively. The proposed HBD-ACS is also able to achieve superior outage performance and better diversity gains at low-to-moderate SNRs compared to existing HD-ACS for both weak and strong interference scenarios. Finite SNR DMT analysis has also revealed that HBD-ACS can achieve interference-free diversity gain if residual SI is sufficiently suppressed, enabling HBD-ACS to be more reliable than HD-ACS at higher multiplexing gains while operating at low SNR ranges.
Acknowledgment {#acknowledgment .unnumbered}
==============
This research is jointly funded by Airbus Singapore Pte Ltd and the Singapore Economic Development Board (EDB). The authors would also like to thank the editor and the anonymous reviewers whose feedback helped us to improve the quality of this work.
Proof of (\[P\_out\_as2\_SIC\]) {#SIC_proof}
===============================
Let $X_{gs}$ be the average received power of the SOI with non-centered chi-squared probability density function (PDF) $f_{X_{gs}}(x) = \frac{K_{X_{gs}}+1}{\Omega_{X}\alpha_{g,2}} \exp\left(-K_{X_{gs}}-\frac{K_{X_{gs}} + 1}{\Omega_{X}\alpha_{g,2}}x\right) I_{0}\left(2\sqrt{\frac{K_{X_{gs}}(K_{X_{gs}}+1)}{\Omega_{X}\alpha_{g,2}}x}\right)$, where $I_{0}\left(\cdot\right)$ is the modified Bessel function of the first kind with zero order [@gradshteyn2014table Eq. (8.445)]. Similarly, let $Y_1$ be the average received power of the interfering signal with non-centered chi-squared PDF $f_{Y_1}(y) = \frac{K_{Y_1}+1}{\Omega_{X}\alpha_{1,2}}\exp\left(-K_{Y_1}-\frac{K_{Y_1} + 1}{\Omega_{X}\alpha_{1,2}}y\right) I_{0}\left(2\sqrt{\frac{K_{Y_1}(K_{Y_1}+1)}{\Omega_{X}\alpha_{1,2}}y}\right)$.
The closed-form SIC outage probability at AS-2 is equivalent to computing the sum of the areas of outage regions $P_1$ and $P_2$, i.e.,$Pr(\mathcal{O}_{2}^{HBD(SIC)}) = P_1 + P_2$. Let the outage regions be defined as $P_1 = Pr\left\{Y_1 < \gamma_{th,gs}^{HBD}\left(1+X_{gs}\right) \right\}$ and $P_2 = Pr\left\{Y_1 \geq \gamma_{th,2}^{HBD}(1+X_{gs}), X_{gs} < \gamma_{th,2}^{HBD} \right\}$. The expression for $P_1$ can be rewritten as [@rached2017unified]: $$\begin{aligned}
\label{eq_P1_2}
P_1 & \hspace{-0.25cm} = & \hspace{-0.25cm} \int_{0}^{\infty} \int_{0}^{\gamma_{th,gs}^{HBD}(1+X_{gs})} f_{Y_1}(y) f_{X_{gs}}(x) dydx \nonumber \\
& \hspace{-0.25cm} = & \hspace{-0.25cm} \sum_{q\geq0}\sum_{l=0}^{q+1}\alpha(q,\Omega_X\alpha_{1,2},K_{Y_{1}},\gamma_{th,gs}^{HBD})\binom{q+1}{l}E\{X_{gs}^l\}_,\end{aligned}$$ where $E\{\cdot\}$ represents the expectation function. The expression for $P_2$ can be expressed as: $$\begin{aligned}
\label{eq_P2_2}
P_2 & = & \int_{0}^{\gamma_{th,2}^{HBD}} \int_{\gamma_{th,gs}^{HBD}(1+X_{gs})}^{\infty} f_{Y_1}(y) f_{X_{gs}}(x) dydx \nonumber\\
& = & \int_{0}^{\gamma_{th,2}^{HBD}} \hspace{-0.2cm} Q_1\left( \sqrt{2K_{Y_1}}, \sqrt{\frac{2(K_{Y_1}+1)\gamma_{th,gs}^{HBD}(1+x)}{\Omega_{X}\alpha_{1,2}}} \right) \nonumber \\
& & \hspace{4.7cm} \times f_{X_{gs}}(x) dx_. \end{aligned}$$ From [@andras2011generalized], $f_{X_{gs}}(x) = \sum_{j\geq0}\alpha(j,\Omega_X\alpha_{g,2},K_{X_{gs}},1)x^j$. Thus, (\[eq\_P2\_2\]) can be rewritten as: $$\begin{aligned}
\label{eq_P2_3}
P_2 & = & 1 - Q_1\left( \sqrt{2K_{X_{gs}}}, \sqrt{\frac{2(K_{X_{gs}}+1)\gamma_{th,2}^{HBD}}{\Omega_{X}\alpha_{g,2}}} \right) \nonumber \\
& & \hspace{-1.3cm} - \int_{0}^{\gamma_{th,2}^{HBD}} \left(\sum_{j\geq0}\alpha(j,\Omega_X\alpha_{g,2},K_{X_{gs}},1)x^j\right) \nonumber\\
& & \hspace{-0.7cm} \times \Bigg(\sum_{n\geq0}\alpha(n,\Omega_X\alpha_{1,2},K_{Y_1},\gamma_{th,gs}^{HBD})\sum_{i=0}^{n+1}\binom{n+1}{i}x^{i}\Bigg) dx_.\end{aligned}$$ Let $c(n) = \alpha(n,\Omega_X\alpha_{1,2},K_{Y_1},\gamma_{th,gs}^{HBD})\sum_{i=0}^{n+1}\binom{n+1}{i}x^{i}$ and $ d(j) = \alpha(j,\Omega_X\alpha_{g,2},K_{X_{gs}},1)x^j$, then the integral in (\[eq\_P2\_3\]) can be written as [@andras2011generalized; @bartoszewicz2012algebrability]: $$\begin{aligned}
\label{eq_P2_4}
& & \hspace{-0.5cm} \int_{0}^{\gamma_{th,2}^{HBD}}\left(\sum_{n\geq0}c(n)\right)\left(\sum_{j\geq0}d(j)\right) dx \nonumber \\
& & = \int_{0}^{\gamma_{th,2}^{HBD}} \sum_{n\geq0}\sum_{i=0}^{n}c(i)d(n-i) dx \nonumber\\
& & = \sum_{n\geq0}\sum_{i=0}^{n}\alpha(i,\Omega_X\alpha_{1,2},K_{Y_1},\gamma_{th,gs}^{HBD}) \nonumber \\
& & \hspace{0.25cm} \times \alpha(n-i,\Omega_X\alpha_{g,2},K_{X_{gs}},1) \sum_{j=0}^{i+1}\binom{i+1}{j}\int_{0}^{\gamma_{th,2}^{HBD}}x^{j+n-i}dx \nonumber\\
& & = \sum_{n\geq0}\sum_{i=0}^{n}\sum_{j=0}^{i+1} \alpha(i,\Omega_X\alpha_{1,2},K_{Y_1},\gamma_{th,gs}^{HBD}) \nonumber \\
& & \hspace{0.4cm} \times \alpha(n-i,\Omega_X\alpha_{g,2},K_{X_{gs}},1)\binom{i+1}{j}\frac{(\gamma_{th,2}^{HBD})^{j+n-i+1}}{j+n-i+1}_.\end{aligned}$$ Combining (\[eq\_P1\_2\]) and (\[eq\_P2\_4\]), the expression in (\[P\_out\_as2\_SIC\]) can be obtained.
In (\[eq\_P2\_2\]), $Q_1\left( \sqrt{2K_{Y_1}}, \sqrt{\frac{2(K_{Y_1}+1)\gamma_{th,gs}^{HBD}(1+x)}{\Omega_{X}\alpha_{1,2}}} \right) = 1 - \sum_{n\geq0}{\alpha(n,\Omega_X\alpha_{1,2},K_{Y_1},\gamma_{th,gs}^{HBD})(1+x)^{n+1}}$ if $\frac{K_{Y_1}+1}{\Omega_{X}\alpha_{1,2}}(\gamma_{th,gs}^{HBD})(1+x)\geq0$ [@andras2011generalized]. In addition, the PDF $f_{X_{gs}}(x)$ can be expressed as a convergent power series if $\frac{K_{X_{gs}}+1}{\Omega_{X}\alpha_{g,2}}x\geq0$ [@andras2011generalized]. Assuming the power series in (\[eq\_P2\_3\]) is convergent, the resultant product of the power series in (\[eq\_P2\_4\]) will also be convergent [@bartoszewicz2012algebrability]. Similarly in (\[eq\_P1\_2\]), the power series is convergent if $\gamma_{th,gs}^{HBD}\leq \frac{(\Omega_{X}\alpha_{1,2})/(1+K_{Y_1})}{2(\Omega_{X}\alpha_{g,2})/(1+K_{X_{gs}})}$ [@rached2017unified]. Therefore, the closed-form expression in (\[P\_out\_as2\_SIC\]) holds if the power series in (\[eq\_P1\_2\]) and (\[eq\_P2\_4\]) are convergent. This completes the proof.
Proof of Corollary \[coro\_asymp\_df\_fixed\_GS\] {#coro_asymp_df_fixed_GS_proof}
=================================================
From (\[P\_out\_gs\_II\]) and (\[df\_fixed\_GS\]), $\left(\Omega_X\right)^{l_1+l_2-q-1} < 1$ when $l_1 + l_2 + l_3 \leq q$. Thus, $\lim_{\Omega_X\to\infty} \left(\Omega_X\right)^{l_1+l_2-q-1} \\ = 0, l_1 + l_2 + l_3 \leq q$. Therefore, only $l_1 + l_2 + l_3 = q + 1$ needs to be considered, which consequently leads to the numerator in (\[df\_fixed\_GS\]) to be zero, i.e., $l_1 + l_2 - q - 1 = 0$. This completes the proof.
Proof of Corollary \[coro\_asymp\_df\_fixed\_AS2\] {#coro_asymp_df_fixed_AS2_proof}
==================================================
To evaluate $\lim_{\Omega_X\to\infty} d_{f,2}^{HBD(II)}$, the approach seen in (\[asymp\_df\_fixed\_GS\]) can be used. Starting with the denominator of $d_{f,2}^{HBD(II)}$, $\left(\Omega_X\right)^{l-q-1} < 1$ when $l \leq q$. Thus, $\lim_{\Omega_X\to\infty} \left(\Omega_X\right)^{l-q-1} = 0$ when $l \leq q$. In the numerator, $(l-q-1) \left(\Omega_X\right)^{l-q-2} = 0$ when $l=q+1$. Similarly, to evaluate $\lim_{\Omega_X\to\infty} d_{f,2}^{HBD(SIC)}$, we first begin with the denominator of $d_{f,2}^{HBD(SIC)}$. Specifically, $\lim_{\Omega_X\to\infty}(\Omega_X)^{l-q-1}=0 $ when $l\leq{q}$ and $(\Omega_X)^{l-q-1}=1$ when $l=q+1$. For $(\Omega_X)^{-m-1}$, $\lim_{\Omega_X\to\infty}(\Omega_X)^{-m-1}=0 $ for $m\geq0$ and for $(\Omega_X)^{-n-2}$, $\lim_{\Omega_X\to\infty}(\Omega_X)^{-n-2}=0 $ for $n\geq0$. In the numerator, $(l-q-1) \left(\Omega_X\right)^{l-q-2} = 0$ when $l=q+1$. Additionally, $\lim_{\Omega_X\to\infty} (\Omega_X)^{-m-2}=0$ when $m\geq0$ and $\lim_{\Omega_X\to\infty} (\Omega_X)^{-n-3}=0$ when $n\geq0$. This completes the proof.
Proof of Corollary \[coro\_lim\_df\_fixed\_hd\] {#coro_lim_df_fixed_hd_proof}
===============================================
At GS, the asymptotic behavior of $d_{f,gs}^{HD}$ can be easily evaluated after some simplifications as shown below: $$\begin{aligned}
\label{lim_df_fixed_hd_GS}
& & \hspace{-1.4cm} \lim_{\Omega_X \to \infty} d_{f,gs}^{HD} \nonumber \\
& \hspace{-1.1cm} = & \hspace{-0.7cm} \lim_{\Omega_X \to \infty} \frac{ -\sum_{m\geq0} \alpha\big(m,1,K_{X_1},\gamma_{th,gs}^{HD}\big) (-m-1) (\Omega_{X})^{-m} }{\sum_{m\geq0} \alpha\big(m,1,K_{X_1},\gamma_{th,gs}^{HD}\big) (\Omega_{X})^{-m} }_. \end{aligned}$$ From (\[lim\_df\_fixed\_hd\_GS\]), It can be seen that $\lim_{\Omega_X\to\infty} (\Omega_{X})^{-m}=1 $ when $m=0$, and $\lim_{\Omega_X\to\infty} (\Omega_{X})^{-m} = 0 $ when $m>0$. Thus, when evaluating (\[lim\_df\_fixed\_hd\_GS\]), only $m=0$ needs to be considered. The asymptotic behavior of $d_{f,2}^{HD}$ can also be proven using the same approach. This completes the proof.
[^1]: Tan Zheng Hui Ernest is with the School of Computer Science and Engineering, Nanyang Technological University, Singapore e-mail: ([email protected]).
[^2]: A S Madhukumar is with the School of Computer Science and Engineering, Nanyang Technological University, Singapore e-mail: ([email protected]).
[^3]: Rajendra Prasad Sirigina is with the School of Computer Science and Engineering, Nanyang Technological University, Singapore e-mail: ([email protected]).
[^4]: Anoop Kumar Krishna is with Airbus Singapore Pte Ltd, Singapore e-mail: ([email protected]).
[^5]: Depending on the degree of passive and active SI mitigation, the resultant SI channel ($h_{si}$) can be a Rician or Rayleigh fading channel [@ahmed2015all]. Thus, modeling $h_{si}$ as a Rician fading channel enables the degree of passive and active SI mitigation to be defined through the Rician $K$ factor.
[^6]: It is useful to note that Doppler shift is not a performance limitation in the upcoming L-band digital aeronautical communication systems (LDACS) standard [@jamal2017fbmc].
[^7]: The scaling factor $\gamma_{\phi}$ models the jitter present in oscillators due to hardware imperfections [@sahai2013impact]
[^8]: Assuming a noise figure of $6$ dB, $\sigma_g^2 = \sigma_2^2=-115$ dBm results in an effective bandwidth of 200kHz. Such a bandwidth falls within the range of existing VHF datalink (VDL) standards (25kHz) [@stacey2008aeronautical Table 3.16], and the upcoming orthogonal frequency-division multiplexing (OFDM)-based L-band digital aeronautical communication systems-1 (LDACS-1) standard [@jamal2017fbmc; @gligorevic2011ldacs1] and Gaussian minimum shift keying (GMSK)-based LDACS-2 standard (500kHz) [@jamal2017fbmc].
[^9]: It is worth noting that the normalized phase noise strength falls within the range of phase noise seen in Appendix C of [@sahai2013impact].
[^10]: In this work, we consider $0\text{ dBm} \leq P_t \leq 36.4\text{ dBm}$. Taking the GMSK-based LDACS-2 as an example [@jamal2017fbmc], with 200kHz of bandwidth, noise figure of 6 dB, and carrier frequency $f_c = 968$MHz, a transmit power of $P_t = 26.4$ dBm is obtained for $d_{1,g}=9.2$km [@matolak2017air_suburban], $\alpha_{1,2}=0.5$, $d_{1,2}=4.6$km [@caas2014manual], and $\Omega_X = 30$ dB. When $d_{1,g}=29$km [@matolak2017air_suburban], $\alpha_{1,2}=5$, $d_{1,2}=145$km [@caas2014manual], and $\Omega_X = 30$ dB, $P_t = 36.4$ dBm is obtained. The obtained values of $P_t$ is comparable to [@jamal2017fbmc], where a transmit power of $P_t=41$ dBm was used in the performance analysis of LDACS-2.
[^11]: Higher diversity gain does not mean lower outage probability and vice-versa. To get a parametric representation for outage probability from diversity gain and SNR, the array gain, coding gain, or SNR offset, needs to be factored as shown in [@ordonez2012array] and the references therein. Similar analysis is needed from the interference-limited receiver’s perspective to quantify the SNR offsets for different protocols at a given interference level, and it is left as a future extension of the current paper.
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'W. Wang , S. Boudreault , B. Goldman , Th. Henning, J. A. Caballero , and C. A. L. Bailer-Jones ,'
date: 'Received 17 August 2010 / Accepted 03 May 2011'
title: 'The substellar mass function in the central region of the open cluster Praesepe from deep LBT observations [^1]'
---
[Studies of the mass function (MF) of open clusters of different ages allow us to probe the efficiency with which brown dwarfs evaporate from clusters to populate the field. Surveys of older clusters (age$\gtrsim$100Myr) are not affected so severely by several problems encountered in young clusters, such as intra-cluster extinction and large uncertainties in brown dwarf models.]{} [We present the results of a deep photometric survey to study the MF of the central region of the old open cluster Praesepe (age 590$^{+150}_{-120}$Myr, distance 190$^{+6.0}_{-5.8}$pc), down to the substellar regime.]{} [We performed an optical ($riz$ and $Y$-band) photometric survey of Praesepe using the Large Binocular Telescope Camera covering an area of 0.59deg$^2$ in the cluster centre from $i\sim19.0$ mag ($\sim$100$M_{\rm
Jup}$) down to a 5$\sigma$ detection limit at $i\sim$25.6 mag ($\sim$40$M_{\rm Jup}$). The survey is approximately 95% complete at $i=23.8$ mag and $z=22.0$ mag ($\sim$55$M_{\rm Jup}$). ]{} [We identify 59 cluster member candidates, of which 37 are substellar, by comparing with the predictions of a dusty atmosphere model. The MF of those candidates rises from the substellar boundary until $\sim$67$M_{\rm
Jup}$ and then declines. This is quite different from the form inferred for other open clusters older than 50Myr, but seems to be similar to those found in very young open clusters, the MFs of which peak at $\sim$10$M_{\rm Jup}$. Either Praesepe really does have a different MF from other clusters or they had similar initial MFs but a different dynamical evolution. Since most of the candidates are faint, we lack astrometric or spectroscopic follow-ups to test their memberships. However, the contaminations by field dwarfs, galaxies, or giants are found to have little effect on the shape of MF and therefore the MF of ‘real’ cluster members should have similar characteristics. ]{}
Introduction
============
The mass functions (MFs) of stellar and substellar populations have been determined from optical and near-infrared surveys for several open clusters at different ages, such as the Orion Nebula Cluster, $\sigma$ Orionis, $\rho$ Ophiuchi, Taurus, IC 348, IC 2391, M35, the Pleiades, and the Hyades. These MFs show clear heterogeneity (see Fig. \[fig:all-mf\]), which may be partially caused by cluster evolution.
Studies of relatively old open clusters (age$>$100Myr) are important for two particular reasons: first, they allow us to study the intrinsic evolution of basic properties of brown dwarfs (BDs), e.g., luminosity and effective temperature, and to compare the evolution with structural and atmospheric models; second, we may investigate how the BD and low-mass star populations as a whole evolve, e.g., the efficiency with which BDs and low-mass stars evaporate from clusters. These investigations have been carried out for the Hyades (@bouvier2008 and references therein) and for Praesepe (@boudreault2010 and references therein).
The Praesepe open cluster has been surveyed extensively in the past (cf.Table \[tab01\]), but only a few surveys have reached masses below the substellar limit (and then only just). Several BD candidates were detected in those surveys, some of which will be re-examined in the present work. The substellar MF of Praesepe remains uncertain.
[@boudreault2010] observed a significant difference between the MFs of Praesepe and Hyades. While they found that the Hyades MF has a maximum at $\sim$0.6M$_\odot$ (@bouvier2008), the MF of Praesepe continues to rise from 0.8M$_\odot$ down to 0.1M$_\odot$. This is surprising, as both clusters share similar physical properties (ages, mass, metallicity, and tidal radii). Disagreement between the Praesepe and Hyades MFs could arise from variations in the clusters’ initial MFs, or from differences in their dynamical evolution (@bastian2010). Although different binary fractions could cause the observed (system) MFs to differ, there is no clear evidence of any variations in the binary fractions from measurements published in the literature (@boudreault2010).
In this paper, we present a survey of the very low-mass star and substellar populations of Praesepe using the blue and red Large Binocular Cameras, extending down to hitherto unexplored mass regimes ($\sim$40$M_{\rm Jup}$). The main aims of our study are to search for new BDs and determine the MF of the Praesepe for a large coverage of the substellar regime.
The candidate selection procedure, and mass and temperature determination methods employed in this study are similar to those adopted in @boudreault2010. However, we probe a lower mass regime and use an evolutionary model based on a dusty atmosphere instead of a combination of dust-free and dusty models.
[lllll]{} Authors & Telescope / instrument & Area & BD & Completeness limits\
& &(deg$^2$) & candidates & (mag)\
Hambly et al. (1995) & COSMOS / POS & UKSTU & 19 & 0 & $R_F \sim$20, $I_N \sim$19\
Pinfield et al. (1997) & INT / WFC & 1.0 & $\sim$10$^{a}$ & $R$=21.5, $I$=20.0, $Z$=21.5\
[Magazzù]{} et al. (1998) & INT / WFC & 0.22 & 1 & $R$=22.2, $I$=21.2\
@chappelle2005 & INT / WFC & 2.6 & 4 & $I_c$=21.3, $Z$=20.5\
@gonzalez-garcia2006 & 3.5m CAHA / LAICA & 5m Hale / LFI& 0.33 & 1 & $i'$=23.8, $z'$=23.3\
Boudreault et al. (2010) & 3.5m CAHA / $\Omega$2k & 2.2m La Silla / WFI & 3.1 & 6 & $I_c$=23.2, $J$=19.9, $Ks$=18.6\
This work$^{b}$ & LBT / LBC & 0.59 & 37 & $r$=24.1, $i$=25.6, $z$=24.7, $Y$=20.3\
- From Fig. 3 in Pinfield et al. (1997), about 10 of their 26 Praesepe member candidates have masses below 72 $M_{\rm Jup}$.
- Our work was complemented by the the $JK_{\it s}$ 3.5m CAHA / $\Omega$2k data from Boudreault et al. (2010).
\[obs-data-calib\] Observations and analysis
============================================
\[obs\] Observations
--------------------
The observations presented in this paper were carried out with the Large Binocular Telescope (LBT) located on Mount Graham, Arizona (@hill2006), using the Large Binocular Cameras (LBCs, see @Speziali2008). The LBCs are two wide-field, high-throughput imaging cameras, namely Blue (LBCB) and Red (LBCR), located at the prime focus stations of the LBT. Each LBC has a wide field of view (23’$\times$23’), with four CCD detectors of 2048$\times$4608 pixels each, providing images with a sampling of 0.23/pixel.
The optical design and detectors of the two cameras are optimized for different wavelength ranges: one for ultraviolet–blue wavelengths (320–500nm, including the Bessel $U$, $B$, $V$ and Sloan $g$ and $r$ bands), and one for the red–infrared bands (500–1000nm, including the Sloan $i$, $z$ and Fan $Y$ bands). In the full binocular configuration, both cameras are available simultaneously, and both point in the same direction of the sky, thus doubling the net efficiency of the LBT.
To accomplish the entire survey of the inner region of Praesepe, we carried out three observing runs, in March 2008, December 2008, and February 2009. Table \[tab01a\] summarizes the observations and Fig. 1 shows the areas surveyed. The total area covered is 0.59 deg$^2$, about 1percent of the cluster region. The transmission curves of the filters used in this survey is presented in Fig. 2, along with a synthetic spectrum of a brown dwarf with $T_{\rm eff}$ = 2300K, log *g* = 4.5 \[CGS\], and solar metallicity (NextGen model).
![\[f1\] Spatial distribution of the LBT pointings](./fig1.eps){width="8cm"}
![\[fig:passband\] Transmission curves of the filters used in our survey compared to a synthetic spectrum of a brown dwarf with $T_{\rm eff}$ = 2300K, log *g* = 4.5, and solar metallicity (NextGen model). The transmission curves include the quantum efficiency of the detectors. ](./fig2.eps){width="8cm"}
[l cc ccccc]{} Field &RA &DEC &Date & Filter&$t_{\it exp}$ &Seeing &$m_{5\sigma}$\
& & & & &(min) &(arcsec) &(mag)\
PraeA &08:40:53.76 &+19:52:41.9 &2009-02-28 &$r$ &114 &1.0 &25.8\
& & &2009-02-28 &$i$ & 42 &1.0 &25.8\
& & &2009-02-28 &$z$ & 54 &1.0 &24.8\
& & &2009-02-28 &$Y$ & 18 &1.0 &21.3\
PraeB &08:39:14.23 &+19:52:41.9 &2009-02-28 &$r^a$ & 54 &1.0 &25.1\
& & &2008-03-06 &$i$ & 42 &2.2 &25.7\
& & &2009-02-28 &$z$ & 54 &1.0 &25.2\
& & &2008-03-04 &$Y$ & 48 &1.4 &22.2\
PraeC &08:39:14.36 &+19:27:18.0 &2008-03-07 &$r$ & 90 &1.9 &25.6\
& & &2008-03-07 &$i$ & 30 &2.0 &25.8\
& & &2008-03-07 &$z$ & 30 &1.0 &25.1\
& & &2009-02-28 &$Y$ & 18 &1.0 &21.2\
PraeD &08:40:53.63 &+19:27:18.0 &2008-12-29 &$r$ & 99 &1.0 &25.8\
& & &2008-12-29 &$i$ & 42 &1.0 &25.8\
& & &2008-12-29 &$z$ & 39 &1.0 &24.8\
& & &2008-12-29 &$Y$ & 18 &1.0 &21.9\
- We also performed a shallower, 84-min pointing in $r$ on 2008-12-30.
\[data\] Reduction and astrometry
---------------------------------
The standard data reduction steps for the LBT data were performed using the IDL astronomy package and IRAF. The bias subtraction was executed on a nightly basis and for each CCD chip. To correct for pixel-to-pixel variations and global illumination, master flat frames were created for the nights using twilight exposures. For nights when no appropriate sky flat exposures were available, we used a master sky flat in the adjacent night. The individual images of a given field were registered and median combined, resulting in a combined science frame for each CCD, field and filter. To detect faint sources, we subtracted the strong background introduced by very bright stars$\footnote{The area fraction affected by bright stars
is less than 3\% for most of CCD images, and is about 6\% in
the worst case.}$. We then used the IRAF task [daofind]{} to detect sources in the “clean" frames. The sources were extracted from the original science frames and instrumental magnitudes calculated using both aperture and point-spread function photometry with the IRAF tasks [phot]{} and [allstar]{} respectively.
At this stage, weak fringes were still visible for several $Y$-band images. The method described by @Bailer-Jones2001 for removing fringes does not apply well in the present case, since the images are seriously affected by bright stars and no clean fringe images could be created. However, as the $Y$-band images are $\sim$1.5 dex shallower than expected, we decided not to use them in this study (although we still quote some statistics of the photometry below).
An astrometric solution was achieved using the Sloan Digital Sky Survey (SDSS York et al. 2000) catalogue as a reference. The root mean square (rms) accuracy of our astrometric solution is 0.10-0.15arcsec. As with other reduction procedures, astrometry was also performed separately for each CCD, to ensure that the solutions were as robust as possible.
\[calib\] Photometric calibration
---------------------------------
To correct for Earth atmospheric absorption of the photometry, we calibrated the inferred data using the $r$, $i$, and $z$ band values of Sloan Digital Sky Survey (SDSS) objects that were observed in the science fields. Zero point offsets were determined from the difference between the SDSS magnitudes and our instrumental magnitudes. Since these were obtained with objects in the same field of view for each science frame, we did not perform a colour or airmass correction when reducing our $riz$ photometry. The error introduced by this approximation is less than 0.05 mag.
{width="14cm"}
To calibrate our $Y$ band photometry, we used our LBT $i$ and $z$ photometry and the $Y$ band photometry from the United Kingdom Infrared Telescope Infrared Deep Sky Survey (UKIDSS, Lawrence et al. 2007). We found that the differences between $Y$ band LBT magnitudes and $Y$ band UKIDSS magnitudes have a linear dependence on the $i-z$ colours, which can be described by the equation $$Y_{UKIDSS}-Y_{LBT,raw}=a_0+a_1\times(i-z),
\label{calib-eq1}$$
On the basis of about 800 common objects between UKIDSS and our measurements, the a$_{0,1}$ coefficients were determined and the instrumental $Y$ magnitudes were then transferred into the UKIDSS $Y$ photometry system. For the same reasons as for our $riz$ data, we did not (need to) perform a colour or airmass correction for our $Y$-band photometry.
The 5$\sigma$ detection limits of our survey are $\sim$ 25.6mag and 24.7mag for the $i$ and $z$ bands, respectively. However, we do not expect all targets brighter than these limits to be able to be detected. We estimate the survey completeness by comparing the number of objects detected to the number predicted assuming a uniform three-dimensional spatial distribution of stars. As shown in Fig. \[f6\], the number of detected sources in each band deviates from a log-normal relationship at bright and faint limits. From this, we estimate the completeness to our 5$\sigma$ detection limit as 67.0% at $i=25.6$ and 70.3% at $z=24.7$ respectively, which corresponds to $\sim$40$M_{\rm Jup}$ assuming a dusty atmosphere. A similar estimation yields a completeness of 82.0% at $r=24.1$mag and 82.6% at $Y=20.3$mag, respectively. Our survey is approximately 95% complete at $i=23.8$ mag and $z=22.0$ mag.
The relatively low completeness of our survey is possibly caused by the saturation of bright stars. For stars lying near to saturated stars, the photometric uncertainties are relatively large. A significant fraction of detected stars is then excluded because of their large photometric errors, which lowers the completeness. The total area seriously affected by bright stars in $i$ band is 3–6 percent.
\[selection\] Candidate selection procedure
-------------------------------------------
The candidate selection introduced by [@boudreault2010] was adopted in the present work. Candidates were first selected based on the colour–magnitude diagram (CMD) using $iz$ bands from our LBC observations[^2] A second selection was performed using a colour–colour diagram. While our $rY$ bands observations are not deep enough for our present investigations, data from the near-IR photometric survey by [@boudreault2010] – which fully covers our survey area, with a 5$\sigma$ detection limit at $\sim$55$M_{\rm Jup}$ in $J$ and $K_{\it s}$ bands – was used instead for the second selection. In the third and final selection, we used the known distance to Praesepe to reject objects based on the discrepancy between their observed magnitude in $J$ and the magnitude predicted from the isochrones and our estimation of $T_{\rm eff}$. To be considered as a cluster member, an object had to satisfy all three of these criteria.
We use the evolutionary tracks of [@chabrier2000] and the atmosphere models from [@allard01] – assuming a dusty atmosphere (the AMES-Dusty model) – to compute an isochrone for Praesepe using an age of 590$^{+150}_{-120}$Myr (@fossati2008), a distance of 190$^{+6.0}_{-5.8}$pc (@Leeuwen2009), and a solar metallicity (\[Fe/H\]=0.038$\pm$0.039, @friel1992). We neglect the reddening \[$E(B-V)$=0.027$\pm$0.004mag, @taylor2006\]. The transmission curves we used for the filters for these calculations are plotted in Fig. \[fig:passband\]. The effective temperature varies from 500 K to 3900 K in steps of 100 K, while the gravity ranges from 4.0 dex to 6.0 dex in steps of 0.5 dex.
### \[selection-1st\] Colour-magnitude diagram
Candidates were first selected from our CMD by keeping all objects that are no more than 0.28mag redder or bluer than the isochrones in all CMDs. This number accommodates errors in the magnitudes and uncertainties in the model isochrones. We also include the errors from the age estimate and distance to Praesepe. We additionally include objects brighter than the isochrones by 0.753mag in order to include unresolved binaries. In Figure \[fig:cmd-iz\], we show the CMD where candidates were selected based on their values of $z$ versus $i$–$z$.
From a total of 44209 objects above the 5$\sigma$ detection limit in $i$ and $z$ filters and below our saturation limit, 709 objects are retained as candidate cluster members (98.4% are rejected). Of these, 160 were detected in the NIR observations of [@boudreault2010]. These objects are used in the selection process described below.
![\[fig:cmd-iz\] CMD with $i$ and $z$ bands used in the first selection procedure. As solid lines we show the isochrone computed from an evolutionary model with a dusty atmosphere (AMES-Dusty). The numbers indicate the masses (in $M_{\rm Jup}$) on the model sequence for various $z$ magnitudes. The dashed lines delimit our selection band.](./fig4n.eps){width="8cm"}
### \[selection-2nd\] Colour–colour diagram
From this step, The candidate selection is based on both the optical $iz$ data and the NIR $JK_{\it s}$ data. A candidate must be detected in every band.
The second stage of the candidate selection involves retaining only those objects that lie within 0.28mag of the isochrone in the colour–colour diagram. This value accommodates the photometric errors, uncertainties in the model isochrones, and the uncertainty in the age estimation of Praesepe. The colour–colour diagram with the selection limits is shown in Figure \[fig:ccd-izjk\], where we also plot the theoretical colours of red giants [using the atmosphere models of @hauschildt1999b] and the theoretical colours of six galaxies with redshifts from 0 to 2 (Meisenheimer et al. 2011). Neither the red giants nor the galaxies are expected to be a significant source of contamination; most of the low redshift galaxies that were not automatically discarded during PSF photometry with full-width-half-maxima (FWHMs) broader than the average stellar FWHM by $\sim$30%, were rejected by means of the visual inspection of individual cluster member candidates after selection procedures. Of the 160 objects selected in the first step, 88 are kept here.
![\[fig:ccd-izjk\] Colour-colour diagram used in the second selection step. The solid line is the isochrone computed from an evolutionary model with a dusty atmosphere (the AMES-Dusty model; the masses in M$_{\odot}$ for each $z-K_{\it s}$ colour are shifted for clarity). The dashed lines show our selection area. We also show the theoretical colours of six galaxies as thin lines and the theoretical colours of red giants as thick lines. The six galaxies are two starbursts, one Sab, one Sbc, and two ellipticals of 5.5 and 15Gyr, with redshifts from $z$=0 to $z$=2 in steps of 0.25 (evolution not considered). We assume that all red giants have a mass of 5M$_{\odot}$, 0.5 $<$ log *g* $<$ 2.5 and 2000K $<$ $T_{\rm eff}$ $<$ 6000K.](./fig5.eps){width="8cm"}
### \[selection-3rd\] Observed magnitude vs. predicted magnitude
As indicated in Section \[get-mass\], our determinations of $T_{\rm
eff}$ and mass are based on the spectral energy distribution of each object, so are independent of the assumed distance. The membership status of an object can therefore be assessed by comparing its observed magnitude in a band with its magnitude predicted from its $T_{\rm eff}$ and Praesepe’s isochrone (which assumes a distance). The predicted magnitude of a background contaminant would be lower (brighter) than its observed magnitude and higher (fainter) for a foreground contaminant. To avoid removing unresolved binaries that are real members of the cluster, we keep all objects with a computed magnitude of up to 0.753mag brighter than the observed magnitude. We also take into account photometric errors and uncertainties in the age and distance of Praesepe. This selection procedure is illustrated in Figure \[fig:mj\_vs\_mj\].
Of the 88 objects selected through CMDs and colour-colour diagrams in the first two steps, 74 are retained here. After this step, we perform a direct visual inspection of the images to reject resolved galaxies and spurious detections. This inspection removes 15 objects from the photometric selection, of which seven are possibly galaxies and eight are false detections.
The remaining objects constitute our final cluster member candidates, shown as large dots in Fig. \[fig:cmd-iz\_izJK\]. Those selected using only the $iz$ photometry amount to 709, and are presented in Fig. \[fig:cmd-iz\_izJK\] as small dots. We note that employing NIR $JK_{\it s}$ data helps us to remove a significant fraction of contaminations. At each mass bin, we calculate the number of stars removed as a result of including $JK_{\it s}$ data in the selection, and use this to estimate the number of stars in the final mass bin where $JK_{\it s}$ data are unavailable.
@Schmidt2010 and @West2011 investigated the colors of L and M dwarfs, respectively, for every spectral types using the SDSS and Two Micron All Sky Survey (2MASS) catalogues. The observed colour ranges are consistent with our computed colour ranges for M and early L dwarfs. However, the observed $i-J$ color range for individual L dwarfs as shown in Fig.2 in @Schmidt2010 is 3.5–6.0mag, broader and slightly redder than our model ranges (the dashes lines in our Fig.5). We note that the Hammer spectral–typing procedure employed by @Schmidt2010, was developed by @Covey2007, who sought to optimize only for K and M dwarfs. For L dwarfs, the uncertainties should be $\sim$2 subclasses or larger. In addition, the Hammer is designed for solar-metallicity dwarfs, while Praesepe has a metallicity of 0.27$\pm$0.10dex[@Pace2008], hence will be prone to significant, systematic errors[@Covey2007]. We therefore keep the use of our model colours alone for candidate selection for a homogeneous study. We found that if we use the observed color range from @Schmidt2010 instead, the observed rise in the mass function around 60$M_{\rm Jup}$ (cf. Section 4) remains – our main conclusion remains unchanged.
![\[fig:mj\_vs\_mj\] Difference between the observed $J$ magnitude and the model $J$ magnitude computed from the derived mass and $T_{\rm eff}$, as a function of $T_{\rm eff}$. The vertical line marks the location of L0 dwarfs. The dotted line (at $-0.753$mag) represents the offset due to the possible presence of unresolved binaries, the dashed-dotted lines represent the error in the magnitude determination, and the long-dashed lines represent the uncertainties in the age and distance of Praesepe. The horizontal solid line just traces zero.](./fig6.eps){width="8cm"}
![\[fig:cmd-iz\_izJK\]Same as Fig. \[fig:cmd-iz\], but with large dots standing for the 59 final candidates, and small dots for the 709 candidates that pass the first selection procedures. ](./fig7.eps){width="8cm"}
\[results-survey\] Results
==========================
Selected photometric candidates
-------------------------------
We find that 59 photometric candidates survive the selection procedures (based on isochrones assuming dusty atmospheres), a density of about 100 objects per square degree. Details of all photometric candidates are listed in Table 3. The identification number (ID) of a candidate is defined according to the field in which it was found and a sequential number for that field. The last column, $J_{\rm model}$, is the predicted $J$ magnitude based on photometric determination of $T_{\rm eff}$ and mass.
Our survey concentrates on the substellar regime. Saturation occurs at $\sim$18 mag in $z$ band, corresponding to $\sim$100$M_{\rm Jup}$. Therefore, most of the low mass candidates discovered in previous surveys (e.g. @pinfield97, @hambly1995) are saturated in our LBT images. Only a few faint brown dwarfs classified by [@pinfield97], [@gonzalez-garcia2006], and [@boudreault2010] are rediscovered in the current survey (cf. Table 4). These objects are M5–9 dwarfs, according to the photometric relations given by @West2008.
Some of our targets were previously identified as cluster members but rejected by our selection procedures or visual inspection. For example, eleven of the 150 Boudreault et al. (2010) candidates are detected in our LBT survey (the rest are mostly saturated). However, nine of them are rejected in this work (cf. Table 5). Among them, seven are rejected on the basis of the $z$ vs. $i$–$z$ CMD selection, because they are bluer than the isochrone area. Another one is obviously not a point-like source in the LBT image, and another is rejected because its observed $J$ magnitude is inconsistent with its model-predicted magnitude. The remaining two targets are confirmed to be cluster dwarf stars. As our current work employed more photometric bands than [@boudreault2010], it is unsurprising that our selection is more conservative.
Most of our candidates are in the substellar regime, and other than the five targets listed in Table 4, no other accurate photometric observations are available from past epochs. This precludes using proper motions as a means of cluster membership assessment at this time.
\[get-mass\] Photometrically-derived masses and effective temperatures
----------------------------------------------------------------------
From the evolutionary tracks and atmosphere models described previously, we obtained the magnitudes $m_{A}$ from the average flux of a star in a specific band $A$ using the equation
$${m_{A}=-2.5\;\mathrm{log}\;f_{A}\;+\;c_{A}},
\label{eqw:phot-cal}$$
where $m_{A}$ is the magnitude observed in a given passband, $f_{A}$ the flux received on Earth in this passband, and $c_{A}$ is a constant that remains to be determined. The flux $f_{A}$ was obtained using the total transmission function of the passband for a given filter, convolved with the quantum efficiency of the CCDs (we assumed that the telescope and instrumental throughput is flat over each passband). To transform our optical $i$ and $z$ band magnitudes from the models to the Johnson photometric system, we assumed all magnitudes $m_{A}$ to be equal to 0.03 when $f_{A}$ is the average Vega flux received on Earth. The constant $c_{A}$ for each passband is then determined using the Vega flux from [@colina1996]. The fitted values for these constants are $i=-22.180$ and $z=-22.706$mag. The values computed by [@boudreault2010] for $J$ and $K_{\it s}$ were $-23.687$ and $-25.908$mag, respectively.
The masses and effective temperatures were estimated in the way described by [@boudreault2009][^3]. For the faintest objects where only $i$ and $z$ are available, the colour $i$-$z$ is used to compute masses and $T_{\rm eff}$. There are different sources of errors for the estimation of the mass and $T_{\rm eff}$, including the photon noise, the photometric calibration, the least squares fitting (imperfect model), and the uncertainties in the age of and distance to Praesepe. The latter two are the most significant errors and give the uncertainties of 0.008M$_\odot$ and 263K for a 0.05M$_\odot$ substellar object ( $T_{\rm eff}=1\,690$K ), 0.010M$_\odot$ and 260K for a 0.06M$_\odot$ substellar object ( $T_{\rm eff}=1\,990$K ), and 0.008M$_\odot$ and 201K for a 0.072M$_\odot$ object at the hydrogen burning limit ( $T_{\rm eff}=2\,293$K ).
[lllccccccccccc]{} ID &RA(J2000) &DEC(J2000) &$i $ & $z $ & $I_c$ & $J$ &$K_{\rm s}$ & $M$ & $T_{\rm eff}$ & $J_{\rm model}$\
& & &(mag) & (mag) & (mag) & (mag) & (mag) & ($M_\odot)$ & (K) & (mag)\
005 & 8:41:08.50 & +19:54:02.0 & 20.19 & 18.70 & 19.02 & 16.58 & 15.39 & 0.088 & 2636 & 17.06\
007 & 8:39:39.56 & +19:47:54.3 & 18.93 & 17.80 & 17.95 & 16.10 & 15.07 & 0.104 & 2860 & 16.58\
009 & 8:39:55.84 & +19:53:14.3 & 18.05 & 17.78 & 20.29 & 17.50 & 16.54 & 0.081 & 2520 & 17.32\
018 & 8:39:42.79 & +19:35:48.2 & – & – & 18.27 & 16.20 & 15.21 & 0.097 & 2782 & 16.78\
022 & 8:41:04.20 & +19:31:27.8 & 20.12 & 19.15 & 18.89 & 16.67 & 15.75 & 0.092 & 2702 & 16.97\
901 & 8:39:59.45 & +19:43:37.4 & 18.05 & 18.30 & 19.09 & 17.16 & 16.41 & 0.084 & 2574 & 17.20\
902 & 8:39:23.72 & +19:52:01.8 & 20.41 & 19.46 & 20.15 & 17.77 & 16.88 & 0.073 & 2348 & 17.72\
903 & 8:40:00.20 & +19:30:27.0 & 19.59 & 19.01 & 19.74 & 17.50 & 16.62 & 0.076 & 2409 & 17.57\
914 & 8:38:52.02 & +19:35:05.3 & 19.48 & 18.69 & 19.12 & 17.25 & 16.35 & 0.085 & 2591 & 17.16\
Contamination by non-members
----------------------------
As mentioned above, the three main sources of contamination are background red giants, unresolved galaxies, and Galactic M and L dwarfs. Red giants contaminate the high mass end of this study, as seen in the $i-J$ vs. $z-K_{\it s}$ diagram, hence can be ignored. Although some types of galaxies have similar colours to Praesepe cluster members more massive than 60$M_{\rm Jup}$, these low-redshift galaxies are in general extended sources and therefore easily rejected by our visual inspection. Among the 74 candidates that passed our selection procedures, we identify four as galaxies on the basis of their LBT images. Other possible sources are field L dwarfs and high redshift quasars (for instance at redshift $z\sim$6; @caballero2008). However, because these quasars have spectral energy distributions similar to mid-T dwarfs, whereas our faintest candidates have colours of early L dwarfs, and given that they are rare (0.25 quasars at $5.5<z<6.5$ in a 0.59deg$^{2}$ survey, Stern et al. 2007), the MF should not be affected by quasar contamination.
The contamination by field dwarfs is not negligible. @caballero2008 identified possible field dwarf contaminants covering spectral types from M3 to T8 from the literature and presented the spatial density in the solar neighbourhood in their Table 3. From this, the spatial density of field dwarfs in the vicinity of Praesepe can be easily inferred, given the Galactic latitude of Praesepe of $b=+32.5$deg and its distance of 190pc, assuming an exponential decrease for stellar density perpendicular to the Galactic disk with a scale height of 500pc. The absolute $J$ band magnitude range constrained by our selection procedures is $\sim\pm 2$ mag, as shown in Fig. \[fig:mj\_vs\_mj\], corresponding to a certain distance interval and a survey ‘volume’, which is defined by the product of survey area and depth. We calculated the number of contaminants by multiplying the survey volume by spatial density at each mass bin. The result is shown in Table \[dwarf\_contamintes\]. The first column gives the central value of log$M$ in each interval, while the second and third columns present the corresponding $T_{\rm eff}$ and $J_{\rm model}$ values at that mass. The fourth column is the number density of cluster member candidates (also shown Fig. \[fig:mf-prae-us\] as filled triangles) after applying all corrections (except for in the lowest two bins, where contaminations are too high to be corrected). The final two columns give the number density of field dwarf contaminants and corresponding fraction.
We found that the field dwarf contaminants do not affect the MF shape. The contamination is significant for $J_{\rm model}\ga 20$ mag. At this magnitude, the @boudreault2010 $J$ band has a completeness of 88%. Therefore, below this magnitude, the mass functions we derived are probably upper limits.
[cccccc]{} log$M$ & $T_{\rm eff}$& $J_{\rm model}$ & N(cand.) & N(cont.)& Fraction\
($M_{\rm Jup}$)& (K) &(mag) & (deg$^{-2}$) &(deg$^{-2}$)\
1.625 & 1412 & 23.29 & 13 & 195 & 1500%\
1.675 & 1692 & 20.28 & 27 & 41 & 150%\
1.725 & 1839 & 19.27 & 18 & 3.3 & 15%\
1.775 & 1981 & 18.87 & 33 & 4.8 & 13%\
1.825 & 2244 & 18.01 & 38 & 2.3 & 5.7%\
1.875 & 2361 & 17.71 & 17 & 3.2 & 16%\
1.925 & 2479 & 17.43 & 8.4 & 1.7 & 17%\
1.975 & 2668 & 17.01 & 8.4 & 1.6 & 16%\
We conclude that the various contaminants are either negligible, or do not affect the MF shape in the range that we can investigate quantitatively, i.e., from about 53 to 94$M_{\rm Jup}$.
\[mf\] Mass function of very low mass and substellar population of Praesepe
===========================================================================
The mass function, $\xi$(log$_{10}$M), we present here is the total number of objects per square degree in each logarithmic mass interval log$_{10}$M to log$_{10}$M + 0.1. Since we do not make any corrections for binaries, we compute here a *system* MF.
As our candidates have been selected only from their photometric properties, cluster membership confirmation via spectroscopy is desirable. However, these observations will not be feasible in the near future because of the faintness of our candidates. The following discussion is therefore based on the assumption that the MF of candidates is similar to that of ‘real’ cluster members. The assumption is possibly valid because our derived MF is consistent with that given by Boudreault et al. (2010) in the common mass bin, and the contamination by field dwarfs, giants, and galaxies should not affect the shape of the MF significantly.
To account for the survey detection efficiency, we use a simple simulation. For example, to calculate the detection efficiency of candidate A01 in $i$ band, we select a bright but unsaturated star in the $i$ band CCD image in which A01 resides, scale it down to the magnitude of A01 (i.e., $i=20.29$), and randomly cast this ‘fake’ star in the CCD image 100 times. We then search and re-measure the ‘fake’ star with our procedures. The detection efficiency is the fraction of ‘fake’ stars that have been re-discovered. We run this test for each filter ($izJK_{\it s}$) and multiply the detection efficiencies together (as we need a detection in every filter) to evaluate the overall detection efficiency for each candidate. This detection efficiency (or recovery rate) is about 90% for the brightest candidates and drops very quickly to 10% for the faintest candidates.
We mentioned in Section \[obs-data-calib\] that our optical photometry reaches lower masses than the NIR photometry that we used. To compute the MF of Praesepe to the lowest mass bin reached without optical data, we first computed a MF[^4] using only the optical $iz$ photometry. This ‘MF’ is presented in Fig. \[fig:mf-prae-us\] as filled dots. We computed a second MF from the list of candidates that pass the three selection criteria and are also detected in the survey of [@boudreault2010] in the NIR $J$ and $K_{\it s}$ bands (presented on Fig. \[fig:mf-prae-us\] as filled triangles). For each mass bin, we computed the number of objects removed as a result of adding the $J$ and $K_{\it s}$ filters to our selection process and mass determination procedure (plotted as a function of mass in Fig. \[fig:mf-prae-us\], top panel). We fitted a linear function to estimate the number of objects that would be removed *if* we had additional $J$ and $K_{\it s}$ photometry to 40–45$M_{\rm Jup}$, which is our lowest mass data point in Fig. \[fig:mf-prae-us\]. However, as shown in Table 6, the contamination in the two lowest mass bins is so overwhelming that the MFs in these two bins can only be regarded as upper limits and are no longer discussed in the paper.
![\[fig:mf-prae-us\] *Lower panel.* Mass function based on (a) the LBT $i$ and $z$ photometry (filled dots) after detection efficiency corrections, and (b) the LBT photometry plus the $J$ and $K_s$ photometry from [@boudreault2010] (dotted open triangles: the original MF; filled triangles: the MF after detection efficiency corrections and removing contaminants). The selection and mass calibration are based on dusty atmospheres. Error bars are Poissonian arising from the number of objects in each bin, except for the lowest-mass bin, for which the error bar is from the linear fit in the top panel. The vertical thin dotted lines are the mass limits at which detector saturation occurs in the $i$ and $z$-bands. The vertical thin long dashed line is the mass at the 5$\sigma$ detection limit of our optical LBT data while the thick short dashed line is the mass at the 5$\sigma$ detection limit of the NIR data of [@boudreault2010]. *Top panel.* Difference of the log of the number of objects (in each mass bin) between the MF computed using the optical $iz$ data and the MF computed using both $iz$ and NIR $JK_{\it s}$ data from [@boudreault2010]. The dotted line is a linear fit.](./fig8.eps){width="8cm"}
Our derived MF (presented in Fig. 8, 9, & 10) shows a rise from 105$M_{\rm Jup}$ to 67$M_{\rm Jup}$ and then a turn-over at $\sim$67$M_{\rm J}$. This turn-over occurs well above the 5$\sigma$ of either $iz$ bands or $JK_{\it s}$ bands (e.g. at $\sim$67$M_{\rm Jup}$, $i\sim$22) and we note that we have corrected the incompleteness of our survey and field dwarf contaminations. We therefore believe that this feature is genuine. This is the first time a clear rise in the substellar MF in an old open cluster has been observed.
The MF of Praesepe near the hydrogen-burning limit was previously obtained in several studies. However, only [@boudreault2010] provide a common mass range for comparison, as shown in Fig. \[fig:mf-praesepe\]. In the first substellar mass bin ($\sim$80$M_{\rm
Jup}$) we see that, both surveys give consistent results within their error bars. However, for the second bin at $\sim$70$M_{\rm Jup}$, our MF is much higher than that of [@boudreault2010]; the discrepancy is smaller when considering the MF from that work using the dusty atmosphere (open dots), which is still a reasonable model for such low mass stars ($\sim$M9/L0). This may indicate that some faint candidates are missing in the [@boudreault2010] survey, as these authors did not make any corrections for the detection efficiencies.
We emphasize that our LBT survey covers the very central 0.59deg$^2$ of Praesepe, while the $\Omega$2k survey by [@boudreault2010] covers a much wider area ($\sim$3.1deg$^2$). If no significant candidates are missing in the $\Omega$2k survey, this discrepancy may suggest that the very low mass cluster members are mostly concentrated in the cluster centre, in contrast to what is expected from a ‘dynamical evaporation’ of brown dwarf in open clusters. The basic idea of dynamical evaporation is that lower mass stars in a cluster have higher speeds according to equipartition of energy, so are able to move higher in the gravitational potential well of the cluster. Hence the fraction of low mass stars should increase with increasing distance from the cluster centre. By comparing the Praesepe and Hyades MFs, [@boudreault2010] concluded that Praesepe might have been less affected by dynamical evolution.
Owing to its large distance and old age, no other published MF determination of Praesepe has reached masses below 70$M_{\rm Jup}$. We therefore compare our results with those from other clusters in Fig. \[fig:all-mf\]. This includes IC2391 from [@boudreault2009], ONC from [@hillenbrand2000], $\sigma$ Orionis from [@caballero2007; @bihain2009], and the Hyades from [@bouvier2008]. The MF of Praesepe is quite different from both IC2391 (age of $\sim$50Myr) and the Hyades ($\sim$625Myr). Either the ‘dynamical evaporation’ does not have (or has not yet had) the same effect on these three clusters, or they had different initial mass functions. Another possibility is that Praesepe has a different binary fraction. Employing different cluster member selection criteria may also account for the observed MF discrepancies among clusters. Further studies are necessary to clarify these points.
The continuing rise of the MF into the substellar regime that we observe has also been observed in young clusters (as shown in Fig. \[fig:all-mf\]), especially in $\sigma$Orionis (@bihain2009), Trapezium [turn-over at $\sim$10–20$M_{\rm Jup}$, @muench2002], $\rho$Oph [MF rising to $\sim$10$M_{\rm Jup}$, @Marsh2010], and in the very low luminosity young cluster in S106, where the MF increases or at least remains flat down to $\sim$10$M_{\rm Jup}$ [@oasa2006]. If we assumed a universal IMF, then it seems that the substellar MF of Praesepe has not evolved significantly since the cluster formed.
![\[fig:mf-praesepe\] MF of Praesepe based on our survey LBT $iz$ and $\Omega$2k $JK_s$ photometry (triangles), compared with that from [@boudreault2010] (*open dots* assuming a dusty atmosphere and *filled dots* assuming a dust-free atmosphere). Error bars are Poissonian arising from the finite number of objects observed in each bin, except for the last bin, for which the error bar is derived from the linear fit. The vertical thin dotted line is the mass limit above which detector saturation occurs in the $i$ and $z$–bands. The vertical thin long-dashed line is the mass at the 5$\sigma$ detection limit of our optical LBT data.](./fig9.eps){width="8cm"}
![\[fig:all-mf\] Mass functions of various open clusters. From top to bottom: $\rho$Oph ($\sim$1Myr, @Marsh2010); S106 ($\sim1$Myr, @oasa2006, their Fig. 10, middle panel for example); Trapezium ($\sim$0.8Myr, @muench2002); IC2391 ($\sim$50Myr, @boudreault2009); ONC ($\sim$5Myr, @hillenbrand2000); $\sigma$ Ori ($\sim$3Myr, @bihain2009); Hyades ($\sim$625Myr, @bouvier2008); Praesepe ($\sim$590Myr, @boudreault2010 and from this work). We also show the lognormal fit to the Galactic field star MF from [@chabrier2003] as a dashed line. All the other MFs are normalized to this at the substellar limit ($\sim$72$M_{\rm Jup}$). ](./fig10.eps){width="8cm"}
\[conclusions\] Conclusions
===========================
We have carried out the deepest survey to date of the old open cluster Praesepe, covering the central 0.59deg$^2$ in the $rizY$ bands. The survey probed a mass range from $\sim$100 to 40$M_{\rm Jup}$ at 5$\sigma$ detection limit, with which we have derived the very low mass and substellar mass function of this cluster.
We compared our optical $iz$-bands data, combined with the $\Omega$2k NIR ($J$ and $K_{\rm s}$) band observations from @boudreault2010, with theoretical loci of cluster members based on a dusty atmosphere (the AMES-Dusty model), to select cluster member candidates. Our final sample comprises 59 photometric candidates. We estimate that the contamination by field dwarfs is about 15%, and that this does not affect the shape of MF. The contamination by galaxies and red giants is believed to be negligible. About two thirds of our cluster members have theoretical masses below the hydrogen-burning limit at 0.072$M_{\odot}$, and are therefore brown dwarf candidates. We emphasize that to claim cluster memberships for the candidates, follow-up astrometric or spectroscopic observations are required. However, given that the candidates are generally faint and these observations are very time-consuming, none of them has yet been confirmed in this way. The discussion in this contribution therefore refers to the mass function of photometric cluster member [*candidates*]{}.
The mass function we have inferred for the central 0.59deg$^2$ of Praesepe is consistent with that inferred for a wider area by [@boudreault2010] at a mass just below the substellar boundary, but deviates by $\sim$0.6dex in the next lowest mass bin, which may indicate that there is either a significant number of objects missing in the Boudreault et al. 2010 survey, or a higher concentration of substellar objects in the centre of Praesepe (as the Boudreault et al. survey is at a larger cluster radius). The latter possibility suggests that the dynamical evolution of very low mass stars is inefficient in this cluster, as proposed by [@boudreault2010] for explaining the discrepancy between the Praesepe MF and Hyades MF.
The steady rise in the Praesepe MF down to $\sim$70$M_{\rm Jup}$ and a turn-over there were unexpected for this old cluster. Such a significant peak has never been observed in any other cluster older than 50Myr, but has been observed in several very young open clusters such as $\sigma$Orionis or clusters in star–forming regions (e.g., Trapezium). This suggests that the dynamical interactions in Praesepe have very little effect on MFs, if we assume there is a universal initial MF.
This project was supported by DFG-Sonderforschungsbereich 881 “The Milky Way System”. Some of the observations on which this work is based were obtained during LBT programme “LBT-F08-02”. Some data analysis in this article has made use of the freely available R statistical package, http://www.r-project.org. This research has made use of the SIMBAD database, operated at CDS, Strasbourg, France. This publication makes use of data products from the Two Micron All Sky Survey, Sloan Digital Sky Survey, and United Kingdom Infrared Telescope Infrared Deep Sky Survey.
Allard, F., Hauschildt, P. H., Alexander, D. R., et al., 2001, , 556, 357
Bailer-Jones, C. A. L., & Mundt, R. 2001, , 367, 218
Bastian, N., Covey, K. R. & Meyer, M. R., 2010, ARA&A, 48, 339
Bihain, G., Rebolo, R., Zapatero Osorio, M. R., et al., 2009, A&A, 506, 1169
Bouvier, J., Kendall, T. T., Meeus, G., et al., 2008, A&A, 481, 661
Boudreault, S. & Bailer-Jones, C. A. L., 2009, ApJ, 706, 1484
Boudreault, S., Bailer-Jones, C. A. L., Goldman, B., et al., 2010, A&A, 510, 27
Caballero, J. A., Béjar, V. J. S., Rebolo, R., et al. 2007, A&A, 470, 903
Caballero, J. A., Burgasser, A. J., & Klement, R. 2008, , 488, 181
Chabrier, G., 2003, , 115, 763
Chabrier, G., Baraffe, I., Allard, F. & Hauschildt, P., 2000, , 542, 464
Chappelle, R. J., Pinfield, D. J., Steele, I. A., Dobbie, P. D., & Magazz[ù]{}, A. 2005, , 361, 1323
Colina, L., Bohlin, R. & Castelli, F., 1996, Instrument Science Report CAL/SCS, 8, 1
Covey, K. R., et al. 2007, , 134, 2398
Dahn, C. C., Harris, H. C., Verba, F. J., et al., 2002, AJ, 124, 1170
Fossati, L., Bagnulo, S., Landstreet, J., et al., 2008, A&A, 483, 891
Friel, E. D. & Boesgaard, A. M., 1992, , 387, 170
González-García, B. M., Zapatero Osorio, M. R., Béjar, V. J. S., et al., 2006, A&A, 460, 799
Hambly, N. C., Steele, I. A., Hawkins, M. R. S. & Jameson, R. F., 1995, Ap&SS, 109, 29
Hauschildt, P. H., Allard, F., Ferguson, J., Baron, E. & Alexander, D. R., 1999, , 525, 871
Hill, J. M., Green, R. F., Slagle, J. H., 2006, SPIE, 6267, 31
Hillenbrand, L. A.& Carpenter, J. M., 2000, , 540, 236
Hodgkin, S. T., Pinfield, D. J., Jameson, R. F., et al., 1999, , 310, 87
Lawrence, A., et al. 2007, , 379, 1599
van Leeuwen, F. 2009, A&A, 497, 209
, A., Rebolo, R., Zapatero Osorio, M. R., Martín, E. L., Hodgkin, S. T., 1998, & Hodgkin, S. T.1998, , 497, L47
Marsh, K. A., Plavchan, P., Kirkpatrick, J. D., Lowrance, P. J., Cutri, R. M., & Velusamy, T. 2010, arXiv:1006.2506
Meisenheimer, K., Wolf, C. & Nicol, M.-H., 2011, *in prep.*
Muench, A. A., Lada E. A., Lada, C. J., & Alves, J., 2002, ApJ, 573, 366
Oasa, Y., Tamura, M., Nakajima, Y., Itoh, Y., Maihara, T., Iwamuro, F., Motohara, K., Hayashi, S. S., Hayashi, M., Kaifu, N., 2006, AJ, 131, 1608
Pace, G., Pasquini, L., & Fran[ç]{}ois, P. 2008, , 489, 403
Pinfield, D. J., Hodgkin, S. T., Jameson, R. F., et al., 1997, , 287, 180
Schmidt, S. J., West, A. A., Hawley, S. L., & Pineda, J. S. 2010, , 139, 1808
Speziali, R., et al. 2008, , 7014, 158
Stern, D., et al. 2007, , 663, 677
Taylor, B. J. 2006, , 132, 2453
West, A. A., Hawley, S. L., Bochanski, J. J., Covey, K. R., Reid, I. N., Dhital, S., Hilton, E. J., & Masuda, M. 2008, , 135, 785
West, A. A., et al. 2011, , 141, 97
York, D. G., et al. 2000, , 120, 1579
[^1]: The LBT is an international collaboration among institutions in Germany, Italy and the United States. LBT Corporation partners are: LBT Beteiligungsgesellschaft, Germany, representing the Max-Planck Society, the Astrophysical Institute Potsdam, and Heidelberg University; Istituto Nazionale di Astrofisica, Italy; The University of Arizona on behalf of the Arizona university system; The Ohio State University, and The Research Corporation, on behalf of The University of Notre Dame, University of Minnesota and University of Virginia.
[^2]: Our $rY$ bands observations do not reach a similar stellar mass, hence are not used here.
[^3]: We first normalized the energy distribution of each object to the energy distribution of the model using the $J$ filter. The energy distribution was then fitted via a least squares fit of the model magnitudes to the measured ones.
[^4]: Note that this is an inaccurate ‘MF’, because of serious contaminations.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Perturbing a CFT by a relevant operator on a half space and letting the perturbation flow to the far infrared we obtain an RG interface between the UV and IR CFTs. If the IR CFT is trivial we obtain an RG boundary condition. The space of massive perturbations thus breaks up into regions labelled by conformal boundary conditions of the UV fixed point. For the 2D critical Ising model perturbed by a generic relevant operator we find the assignment of RG boundary conditions to all flows. We use some analytic results but mostly rely on TCSA and TFFSA numerical techniques. We investigate real as well as imaginary values of the magnetic field and, in particular, the RG trajectory that ends at the Yang-Lee CFT. We argue that the RG interface in the latter case does not approach a single conformal interface but rather exhibits oscillatory non-convergent behaviour.'
date: '27 October, 2016'
---
.5in [**RG boundaries and interfaces in Ising field theory**]{}\
.250in
.5in [Anatoly Konechny]{} 0.5cm [*Department of Mathematics, Heriot-Watt University\
Edinburgh EH14 4AS, United Kingdom\
and\
Maxwell Institute for Mathematical Sciences\
Edinburgh, United Kingdom\
*]{} E-mail: [email protected]
.5in
Introduction
============
We are interested in RG flows in two-dimensional Euclidean quantum field theories. We will look at flows that originate in a UV fixed point described by a 2D ${\rm CFT}_{\rm UV}$ and arrive at an IR fixed point described by a ${\rm CFT}_{\rm IR}$ which may be trivial if a mass gap develops along the flow. The flows are triggered by perturbations of ${\rm CFT}_{\rm UV}$ by relevant operators $\phi_{i}$. In general it is a hard non-perturbative problem to determine the infrared properties of the perturbed theory, in particular to identify ${\rm CFT}_{\rm IR}$ when it is non-trivial. A technique that allows one to investigate the perturbed theories numerically in the infrared including the flows to non-trivial fixed points is the truncated conformal space approach (TCSA) invented in [@YZ1], [@YZ2]. The basic setup of TCSA is as follows.
In order not to worry about perturbative infrared divergences we put the perturbed theory on a cylinder with spacial periodic coordinate $x\sim x+ R$ and Euclidean time $y$ directed along the axis of the cylinder. In the Hamiltonian formalism the perturbed Hamiltonian on a circle at $y=0$ is \[Ham\_pert\] H=H\_[0]{} + \^[i]{} dx \_[i]{}(x,0) . Here H\_[0]{}=( L\_[0]{} + |L\_[0]{} - ) is the Hamiltonian of ${\rm CFT}_{\rm UV}$ that is expressed via the Virasoro modes $L_{0}$, $\bar L_{0}$ and the central charge $c_{\rm UV}$. The eigenvalues of $H_{0}$ are scaling dimensions of ${\rm CFT}_{\rm UV}$ shifted by $\frac{c_{\rm UV}}{12}$ . Since the state space ${\cal H}_{0}$ of the unperturbed theory on a cylinder is discrete $H$ can be represented by an infinite matrix acting in this space. To regulate the UV divergences we can truncate $H$ to a finite matrix by restricting it to a finite dimensional subspace in ${\cal H}_{0}$. This gives rise to a variety of truncated Hamiltonian techniques in which one numerically calculates the eigenvalues and eigenvectors of the truncated Hamiltonian matrix for various values of the scale set by the circle length $R$. In TCSA the truncated subspace is specified by imposing an upper bound on the scaling dimensions of the allowed states. This upper bound is called the truncation level. The UV divergences show up in the dependence of numerics on the truncation level. This dependence and various improvement techniques have been discussed in [@GW], [@LT_RG], [@HRvR], [@RV].
TCSA has been applied to situations in which a perturbed theory arrives to a non-trivial fixed point (see e.g. [@Cardy_etal], [@GW]). In this case the dimensionless energy eigenvalues $ER/2\pi$ at large $R$ asymptote to constant values that give scaling dimensions in ${\rm CFT}_{\rm IR}$. Moreover the asymptotic eigenvectors that correspond to scaling states $|v_{i}\rangle$ in ${\rm CFT}_{\rm IR}$ are obtained as vectors in the truncated subspace of ${\cal H}_{0}$. The corresponding components of $|v_{i}\rangle\in {\cal H}_{0}$ give us pairings \[TCSA\_pairing\] i | v\_[j]{} between scaling states in both CFT’s.
Pairings of this type are also associated with conformal interfaces. A conformal interface ${\cal I}$ between ${\rm CFT}_{\rm UV}$ and ${\rm CFT}_{\rm IR}$ can be described (via the folding trick [@BBDO]) as a conformal boundary condition in the tensor product ${\rm CFT}_{\rm UV} \otimes {\rm CFT}_{\rm IR}$. If $\phi_{i}^{\rm UV}$ and $\phi_{j}^{\rm IR}$ are scaling fields in the two theories then we can define a pairing \[conf\_pairing\] \_[UV]{}i |j\_[IR]{} = \_[i]{}\^[UV]{} \_[j]{}\^[IR]{} \_[I]{} as a disc one-point function with boundary condition ${\cal I}$. Such pairings are canonically normalised by means of Cardy constraint [@Cardy].
In TCSA there is no canonical way to fix normalisation of the eigenstates $|v_{j}\rangle$ so the natural observables that are free from this ambiguity are component ratios \[gen\_rat\] \^[j]{}\_[i,k]{} = . In (untruncated) quantum field theory perturbed states more often than not do not lie in the unperturbed state space. This has many manifestations such as Haag’s theorem, orthogonality catastrophe, inequivalent representations of canonical commutation relations, etc (see [@H_theorem] for a nice exposition and discussion). Typically[^1] however the problem is with the norm of the perturbed states in interaction representation which is formally infinite, while the component ratios such as (\[gen\_rat\]) are well defined.
To show that the pairing (\[TCSA\_pairing\]) and ratios (\[gen\_rat\]) associated with an RG flow arise from a particular local conformal interface we discuss states in Lagrangian formalism using wave functionals. Suppose ${\rm CFT}_{\rm UV}$ is described via a fundamental field $\varphi(x)$ and an action functional $S_{0}[\varphi]$. States in this theory can be represented by a wave functional $\Psi[\varphi_{0}]$ depending on functions $\varphi_{0}(x)$ defined on the circle $y=0$. The vacuum state is then given by a renormalised functional integral over the left half cylinder $y\le 0$ \_[vac]{}\[\_[0]{}\] = \_[(x,0)=\_[0]{}(x)]{} e\^[-S\_[0]{}\[\]]{} . The functionals describing excited states can be obtained by inserting local operators at positions with $y<0$ inside the above functional integral.
Suppose now that the perturbed theory is described by an action functional S\[\] = S\_[0]{}\[\] + dxdy V((x,y)) where $V(\varphi)$ is some potential. The vacuum of the perturbed theory can be represented by the functional integration over the fields defined \[perturbed\_psi\] \_[vac]{}\^[pert]{}\[\_[0]{}\] = \_[(x,0)=\_[0]{}(x)]{} e\^[-S\[\]]{} . Expanding inside the functional integral the exponent $$\exp\Bigl[{\iint\limits_{y\le 0} \! dxdy\, V(\varphi) }\Bigr]$$ in power series we obtain a formal expansion of the perturbed vacuum functional $\Psi_{\rm vac}^{\rm pert}[\varphi_{0}] $ in terms of unperturbed states (interaction representation). We can extend this pairing to excited states of the deformed theory by inserting into the functional integral (\[perturbed\_psi\]) additional local operators. This gives a pairing between the states in the two theories. In particular if we follow the perturbed theory to the IR fixed point we obtain a pairing of the type (\[TCSA\_pairing\]). We assume in this discussion that all divergences are renormalised including the ones that need additional boundary counterterms at $y=0$. Renormalisation of wave functionals has been discussed in [@Symanzik], [@Luscher], [@Minic_Nair] and more recently in [@FLP]. (More on renormalisation shortly.)
What becomes clear in this picture is that the pairing arises by perturbing the UV fixed point on a half space (a half cylinder) and letting the perturbed theory flow to the IR fixed point. This gives us a local conformal interface between ${\rm CFT}_{\rm UV}$ and ${\rm CFT}_{\rm IR}$ which we call an RG interface. The idea to associate such interfaces with RG flows was put forward in [@BR] (see also [@FQ]). A concrete proposal for such an interface for $\psi_{1,3}$-flows between neighbouring minimal models was put forward in [@Gaiotto] and pairing (\[conf\_pairing\]) associated with it was considered[^2].
Note that we do not need to have fundamental fields and functional integral representation to define the pairing in terms of the RG interface. To elucidate renormalisation and to relate the above functional integral picture to conformal perturbation and TCSA we first recast (\[perturbed\_psi\]) in the language of conformal perturbation. Wave functionals correspond to a particular basis in state space in which the field operator $\varphi$ is diagonal. We can choose instead the conformal basis of scaling states $|i\rangle$ in ${\rm CFT}_{\rm UV}$. Consider the following amplitude on a cylinder of length $L$ \[Z0i\] Z\_[0,i]{}(L) = 0| | i illustrated on the picture below.
(-2,0) – (6,0) arc ( 270:90: 0.6 and 1.1)– (-2,2.2) arc (90:270: 0.6 and 1.1)–cycle; (-2.25,0)–(6.25,0); (-2.25,2.2)–(6.25,2.2); (-2.25,2.2) arc (90: 270: 0.6 and 1.1); (-2.3,0) arc (-90:90:0.6 and 1.1 ); (-2,2.2) arc (90: 270: 0.6 and 1.1); (-2,0) arc (-90:90:0.6 and 1.1 ); (6,0) arc (270:90:0.6 and 1.1); (6.25,1.1) circle \[x radius=0.6, y radius = 1.1\]; (6.0,0) arc (-90:90:0.6 and 1.1);
(7.3,1.1) node [$|i \rangle$]{}; (-3.3,1.1) node [$\langle 0|$]{}; (2.3,1.1 ) node [$e^{\lambda^{k}\! \int\! \phi_{k}}$]{}; (2.2, 0.6) node [perturbed theory]{}; (-2,2.25) – (-2, 2.6); (6,2.25)–(6,2.6); (-2, 2.45) – (6,2.45); (2,2.68) node [$L$]{};
We have added little collars (depicted white) of unperturbed theory at both ends where the states $\langle 0|$, $|i\rangle$ are attached. The actual width of the collars is inessential. The perturbative expansion of (\[Z0i\]) is Z\_[0,i]{}(L) = \_[n]{} \_[i\_1,…, i\_[n]{}]{} \_[-Ly\_10]{} d\^2 z\_[1]{} …\_[-Ly\_n0]{} d\^2 z\_[n]{}0| \_[i\_1]{}(z\_1) …\_[i\_n]{}(z\_n) |i The correlators here are correlators in ${\rm CFT}_{\rm UV}$ and can be calculated by mapping the cylinder onto an annulus on the plane and inserting $\phi_{i}$ at infinity. Divergences arise when operators $\phi_{i_{k}}$ collide. Collisions can happen at a point in the bulk of the cylinder (annulus) or at a point on the boundary: at $y=0$ or $y=-L$. The former are handled by the usual renormalisation of perturbed theory while the latter may give rise to additional boundary counter terms of the type discussed in [@Symanzik][^3]. It is clear from the collisions picture that all divergences are local. Assuming they can be renormalised we obtain a cylinder with two local interfaces between ${\rm CFT}_{\rm UV}$ and the perturbed theory with two external states of ${\rm CFT}_{\rm UV}$ attached. The perturbed theory can be driven towards ${\rm CFT}_{\rm IR}$ that results in having a system with two conformal interfaces. Taking the limit $L\to \infty$ results in the following asymptotic Z\_[0,i]{}(L) \~e\^[-LE\_[0]{}\^[IR]{}]{} 0|v\_[0]{}v\_[0]{}|i where $E_{0}^{\rm IR}=-\pi c_{\rm IR}/6R$ is the vacuum energy of ${\rm CFT}_{\rm IR}$. Removing the divergent exponential factor we obtain up to the overall factor $ \langle 0|v_{0}\rangle$ the overlaps $\langle v_{0}|i\rangle $ of the perturbed vacuum $|v_{0}\rangle$ with the unperturbed scaling states. Overlaps with excited states $|v_{i}\rangle$ can be obtained from the subleading terms in the $L\to \infty$ asymptotic.
To summarise the above discussion, in the Lagrangian formalism the TCSA overlaps (\[TCSA\_pairing\]) are described in terms of a local interface between the perturbed and unperturbed theories. This has implications that are not evident in the Hamiltonian picture (and thus in TCSA). Thus for flows to a trivial fixed point the RG interface is a conformal boundary condition (the RG boundary). In the far infrared the perturbed theory vacuum state is given by the conformal boundary state describing this boundary condition (this was previously observed in [@TWunpub] and [@CEF] for particular models). The boundary state at hand must satisfy Cardy constraint which arises from locality. There are similar constraints for flows to a non-trivial fixed point with local conformal interfaces described by conformal boundary conditions in the tensor product theories as was already mentioned above.
All massive flows from a given ${\rm CFT}_{\rm UV}$ are then labeled by conformal boundary conditions. They appear to be natural labels of infrared phases of massive theories. The unstable manifold is then broken into regions of possibly different dimensions labeled by conformal boundary conditions with particular conditions on boundaries separating these regions. It seems to be an interesting enterprise to investigate this structure for particular two-dimensional CFTs. For each Virasoro minimal model there is a finite number of fundamental conformal boundary conditions, so that we expect their unstable manifolds to be broken into finitely many regions (massive phases). As we will show below superpositions of fundamental boundary conditions may also arise for RG flows in the presence of spontaneous symmetry breaking.
In this paper we investigate the RG boundaries and interfaces for the simplest minimal model - the critical Ising model. It has three primary fields: ${\bf 1}$, $\epsilon$, $\sigma$ and three fundamental conformal boundary conditions with boundary states [@Cardy] \[conf\_bcs\] |= , |F= |0- | . Here $|0\rangle\!\rangle $, $|\epsilon \rangle\!\rangle $, $|\sigma\rangle\!\rangle $ stand for the Ishibashi states built on the respective primaries: ${\bf 1}$, $\epsilon$, and $\sigma$. The boundary states $|\pm\rangle\!\rangle$ describe the fixed spin while $ |F\rangle\!\rangle $ describes the free spin boundary condition. The theory has two relevant perturbations - temperature and magnetic field that are described by mass $m$ and magnetic field coupling $h$ in the free fermion theory: \[IFT\] S\_[IFT]{}=(|+ ||+ im|) d\^2 x + h d\^2 x . For real values of the couplings $m$, $h$ all flows are massive and our results regarding the corresponding RG boundaries are summarised on the diagram presented below. We parameterise an outgoing RG trajectory for Ising field theory by points on a circle on the $m,h$-plane. For all points on the upper semicircle except for the points on the $h=0$ axis the RG boundary condition is $|+\rangle\!\rangle$ and similarly for all points on the lower semicircle we have $|-\rangle\!\rangle$. The exceptional points are those on the $h=0$ axis. For $h=0, m>0$ we obtain a superposition $|+\rangle\!\rangle \oplus |-\rangle\!\rangle$ while for $h=0,m<0$ we obtain $|F\rangle\!\rangle$.
(-3,0) –(3.3,0); (0,-3)–(0,3); (0,0) circle \[radius=2\]; (3.6,0 ) node [$m$]{}; (0,3.3) node [$h$]{}; (1.8,1.8) node [$|+\rangle\!\rangle $]{}; (-1.7,1.8) node [$|+\rangle\!\rangle $]{}; (-1.7,-1.8) node [$|-\rangle\!\rangle$]{}; (1.8,-1.8) node [$|-\rangle\!\rangle$]{}; (3.8,1) – (2.1,0.1); (-2,0) node [$\bullet$]{} ; (2,0) node [$\bullet$]{} ; (3.8,1.2) node [$|+\rangle\!\rangle \oplus |-\rangle\!\rangle$]{}; (-2.4,0.3) node [$|F\rangle\!\rangle$]{};
The fact that we obtain a superposition of boundary states for $h=0, m>0$ reflects the spontaneous symmetry breaking of the low temperature phase.
We have also investigated the imaginary magnetic field. It is convenient to label the RG trajectories by the dimensionless ratio \[y\_def\] y= . As shown in [@FZ] for imaginary magnetic field when $y>y_{\rm cr}=-2.42929$ (the value is taken from [@Zam2]) the vacuum energy is complex with a two-dimensional vacuum subspace spanned by the eigenstates with complex conjugated energy eigenvalues. We find that in all such cases the RG boundary is $|+\rangle\!\rangle \oplus |-\rangle\!\rangle$. For $y<y_{\rm cr}$ the vacuum energy is real and the flows are massive. We argue that in this case the RG boundary does not approach a single conformal boundary condition as we go to the far infrared. For large negative values of $y$ the RG flow of the vacuum vector can be well approximated by the boundary magnetic field model that is exactly solvable and provides further insight into the non-convergent behaviour. At $y=y_{\rm cr}$ the RG trajectory approaches the Yang-Lee edge singularity [@YL1], [@YL2] that is described by a non-unitary conformal minimal model ${\cal M}(2,5)$ [@Cardy2]. In this case our findings point as well to the picture in which the RG interface does not approach a single conformal interface but demonstrates an oscillatory non-convergent behaviour. The case $y=-\infty$ is the exceptional case of no magnetic field and the RG boundary is given by the free boundary condition as before. We summarise these answers on the diagram below.
(-3,0) –(3.3,0); (0,-3)–(0,3); (-1.41,-1.41) arc (225:135:2); (-1.41,-1.41) arc (-135:135:2); (3.6,0 ) node [$m$]{}; (0,3.3) node [$ih$]{}; (1.8,2.1) node [$|+\rangle\!\rangle \oplus |-\rangle\!\rangle$]{}; (1.8,-2.1) node [$|+\rangle\!\rangle\oplus |-\rangle\!\rangle$]{}; (-3.6,-1) – (-2.1,-0.1); (-2,0) node [$\bullet$]{} ; (-1.3,2.2) node ; (-3.95,-1) node [$|F\rangle\!\rangle$]{}; (0,0) parabola (-1.7,2.04); (0,0) parabola (-1.7,-2.04);
The main body of the paper is organised as follows. In section 2 we work out in detail the analytically solvable case of vanishing magnetic field. In section 3 we explain the basic TCSA setup for the perturbed Ising model using free fermions. In section 4 we present numerical results obtained using TCSA for real valued couplings $m$ and $h$. In particular we give plots for some component ratios of the type given by (\[gen\_rat\]). The numerical results together with the analytic ones from section 2 lead to the diagram on Fig. \[diagram\_real\] which is further discussed in section 4.2. While TCSA works really well for the real couplings, for imaginary magnetic field when the vacuum energy is real another numerical method - truncated free fermion space approach (TFFSA) of [@FZ] works much better. This method which we explain in section 5 allows one to treat the mass coupling non-perturbatively. In section 6 we present our numerical results obtained using TFFSA for the case of imaginary magnetic field and the region $y>y_{\rm cr}$ where the vacuum energy is complex. In section 7 we discuss the region $y\le y_{\rm cr}$. We first consider the massive flows with $y<y_{\rm cr}$ in section 7.1. Before discussing in section 7.3 the numerics for component ratios related to the flow to the Yang-Lee fixed point at $y=y_{\rm cr}$ we describe all conformal interfaces between the Ising and Yang-Lee models in section 7.2. We finish off with a brief discussion of some open problems in section 8.
RG flows at zero magnetic field
===============================
Preliminaries. Massive and massless free fermions. {#preliminaries}
--------------------------------------------------
The Ising model at zero magnetic field is a theory of free fermions with Euclidean action (|+ ||+ im|) d\^2 x Here the coupling $m$ measures the deviation from the critical temperature of the classical model: $m=(T_{c}-T)/T_{c}$.
We are going to put this theory on an infinite cylinder of radius $R$. Let $x\sim x + R$ be the periodic coordinate and $y$ - the coordinate along the cylinder. Two sectors arise - with periodic (Neveu-Schwarz or NS-sector for brevity) and anti periodic (Ramond or R-sector). Quantizing on the spacial $x$-circle we get the single-particle frequencies \_n = where $n\in {\mathbb Z}$ in the R-sector and $n\in 1/2 + {\mathbb Z}$ in the NS-sector. Below we will use the index $n$ (or $n_1, n_2, \dots$) for integers and $k$ (or $k_1, k_2, \dots$) for half-integers.
Using the complex coordinate $z=x+iy$ the mode expansion in the NS sector is introduced as (x,y)=\_[k1/2 + [Z]{}]{} (-)\^[1/2]{} |(x,y)=\_[k1/2 + [Z]{}]{} ()\^[1/2]{} where (k) = {
[r@l]{} 1 , &k<0\
-i[sign]{}(m) , &k>0
. We also have similar expressions in the R-sector in terms of creation and annihilation operators $b^{\dagger}_{n}$, $b_{n}$ which we omit for brevity. The physical state space on a circle ${\cal H}_{m}$ is spanned by the states \[mbasis1\] |k\_1, …, k\_[N]{}\_[NS]{} = b\^\_[k\_1]{}…b\^\_[k\_N]{}|0\_[NS]{} , k\_[i]{}1/2 + [Z]{} with $N$ even and by the states \[mbasis2\] |n\_1, …, n\_[M]{}\_[R]{} = b\^\_[n\_1]{}…b\^\_[n\_M]{}|0\_[R]{} , n\_[i]{} where when $m>0$ the number of oscillators $M$ is even and when $m<0$ $M$ is odd.
The Hamiltonian blocks are \[HFF1\] H\^[NS]{} = \_[n=-]{}\^ \_[n+1/2]{}b\^\_[n+1/2]{}b\_[n+1/2]{} + E\_[0]{}\^[([NS]{})]{} , \[HFF2\] H\^[R]{} = \_[n=-]{}\^ \_[n]{}b\^\_[n]{}b\_[n]{} + E\_[0]{}\^[([R]{})]{} where the vacuum energies are E\_[0]{}\^[([NS]{})]{}=( ) -|m|\_[-]{}\^ () ( 1 + e\^[-|m|R()]{} ) , E\_[0]{}\^[([R]{})]{}=( ) -|m|\_[-]{}\^ () ( 1 - e\^[-|m|R()]{} ) . Here $\mu$ is a subtraction scale for the logarithmic divergence that arises from the OPE of two energy operators and is responsible for the Onsager singularity of free energy.
The theory with an arbitrary value of $m$ can be described in the $m=0$ state space using a Bogolyubov transformation. The massless fermionic fields which we denote $\chi(z)$, $\bar \chi(\bar z)$ have mode expansions && (z) = ( -i )\^[1/2]{} \_[k1/2+[Z]{}]{} a\_[k]{}e\^[i(x+iy)k]{} ,\
&& |(|z) = ( i )\^[1/2]{} \_[k1/2+[Z]{}]{} |a\_[k]{}e\^[-i(x-iy)k]{} . The operators $a_{k}, \bar a_{k}, k>0$ are the annihilation operators and $a^{\dagger}_{k}=a_{-k}$, $\bar a^{\dagger}_{k}=\bar a_{-k}$, $k>0$ are the corresponding creation operators. Similarly in the Ramond sector the expansions are && (z) = ( -i )\^[1/2]{} \_[n]{} a\_[n]{}e\^[i(x+iy)n]{} ,\
&& |(z) = ( i )\^[1/2]{} \_[n]{} |a\_[n]{}e\^[-i(x-iy)n]{} where the zero modes satisfy a\_[0]{}\^2 = , |a\_[0]{}\^2 = . We choose conventions in which the zero modes act on the spin and disorder primary states according to a\_[0]{}|= | , |a\_[0]{}|= -| . The physical state space ${\cal H}_{0}$ is spanned by the states |a\_[k\_1]{}\^…|a\_[k\_p]{}\^a\_[k\_[p+1]{}]{}\^…a\_[k\_N]{}\^|0 , k\_[i]{}1/2 + [Z]{} , N - in the NS sector and by the states |a\_[n\_1]{}\^…|a\^\_[n\_q]{}a\^\_[n\_[q+1]{}]{}…a\_[n\_M]{}\^| , n\_[i]{}, , M - in the R sector. Above $|0\rangle$ stands for the conformal vacuum state.
The Bogolyubov transformation relating the massive and massless theories is \[BogNS1\] b\_[k]{}\^ = f(k){
[l@l]{} (|a\^\_[k]{} - i a\_[k]{}) , k>0\
(a\_[-k]{}\^ + i |a\_[-k]{} ) , k<0
. in the NS sector and \[BogR1\] b\_[n]{}\^ = f(n){
[l@l]{} (|a\^\_[n]{} - i a\_[n]{}) , n>0\
(a\_[-n]{}\^ + i |a\_[-n]{} ) , n<0
. \[BogR2\] b\_[0]{}\^ = {
[l@l]{} ( a\_[0]{} + i|a\_[0]{}) , m>0\
(a\_[0]{} - i|a\_[0]{} ) , m<0
. , b\_[0]{} = {
[l@l]{} ( a\_[0]{} - i|a\_[0]{}) , m>0\
(a\_[0]{} + i|a\_[0]{} ) , m<0
. in the R-sector. In the above \_k = \_[k]{} - , f(k) = where $k$ is a half-integer and similar formulas hold with $k$ replaced by an integer $n$ for the R-sector.
The interface between the massless and massive theories {#mass_interface}
-------------------------------------------------------
Bogolyubov transformation (\[BogR1\]), (\[BogR2\]) is a proper Bogolyubov transformation in the terminology of [@Berezin] which means that it is implemented by a unitary operator $U: {\cal H}_{0}\to {\cal H}_{0}$ so that U |a\^\_[k]{} U\^[\*]{} & = & f(k)(|a\^\_[k]{} - i a\_[k]{}) ,\
U a\^\_[k]{}U\^[\*]{}&=&f(k)(a\_[k]{}\^ + i |a\_[k]{} ) and similarly in the R-sector with the zero modes transforming as U a\_[0]{} U\^[\*]{} = a\_[0]{} , U |a\_[0]{} U\^[\*]{} = [sign]{}(m)|a\_[0]{} .
The Bogolyubov transformation at hand allows one to embed the states from ${\cal H}_{m}$ into ${\cal H}_{0}$ and vice versa. More precisely the first embedding is realised by means of an interface operator D\_[m]{} : [H]{}\_[0]{} \_[m]{} , D\_[m]{} = [O]{} U where |a\^\_[k\_1]{} …a\^\_[k\_n]{}…|0= b\^\_[k\_1]{} …b\^\_[-k\_n]{}…|0\_[NS]{} and similar relations are satisfied in the R-sector with the zero modes transforming as |a\^\_[0]{} …a\^\_[0]{}…|= … …|0\_[R]{} .
The inverse operator D\_[m]{}\^[-1]{} : [H]{}\_[m]{} \_[0]{} , D\_[m]{}\^[-1]{} = [O]{}\^[-1]{} V where $V:{\cal H}_{m} \to {\cal H}_{m}$ is a unitary operator that realises the inverse Bogolyubov transformation.
A general formula for operators $U$, $V$ for a given Bogolyubov transformation is known (see e.g. formula (5.15) from Berezin’s book [@Berezin]). Using it we find the following explicit formulas for the two blocks of $\hat D_{m}^{-1}$: D\_[m]{}\^[-1]{} = (
[cc]{} D\^[NS]{} & 0\
0& D\^[R]{}
) , \[IONS\] D\^[NS]{} = [N]{}()|00|\_[NS]{} \_[k>0]{}( i(a\^\_[k]{}|a\^\_[k]{} - b\_[-k]{}b\_[k]{}) + f\^[-1]{}(k)(a\^\_[k]{}b\_[-k]{} + |a\^\_[k]{}b\_[k]{}) ) , \[IOR\] D\^[R]{} = ()\_[n=1]{}\^( i(a\^\_[n]{}|a\^\_[n]{} - b\_[-n]{}b\_[n]{}) + f\^[-1]{}(n)(a\^\_[n]{}b\_[-n]{} + |a\^\_[n]{}b\_[n]{}) ) where = {
[l@l]{} |0|\_[R]{} + |0|\_[R]{}b\_[0]{} , m>0\
|0|\_[R]{} b\_[0]{} + |0|\_[R]{} , m<0
. and all creation operators $a_{n}^{\dagger}, \bar a_{m}^{\dagger}$ are understood to act on $|0\rangle$ and $|\sigma\rangle$ from the left. The normalisation factors ${\cal N}(\nu) $, $ {\tilde {\cal N}}(\nu)$ are given in terms of infinite products () = \_[n=0]{}\^ f(n+1/2) , () = \_[n=1]{}\^ f(n) . where we denoted $\nu = |m|R$ - the dimensionless mass.
Taking the logarithms of ${\cal N}(\nu) $, $ {\tilde {\cal N}}(\nu)$ and applying Abel-Plana summation formula we obtain the following integral expressions \[N1\] [N]{}() = , \[N2\] () = 2\^[1/4]{} ,
The ratio of the normalisation factors $\tilde {\cal N}/{\cal N}$ interpolates between $1$ at $\nu=0$ and $2^{1/4}$ at $\nu=\infty$. Below is a plot of \[Fnu\] f() =
This function monotonically increases from the value $2^{-1/4}\approx 0.84$ to 1. It passes the value $0.9$ around $\nu=3.5$ and the value $0.99$ around $\nu = 40$.
The interface operator can be alternatively described in terms of a local gluing condition at $y=0$: (x,0)=(x,0) , |(x,0)=|(x,0) . This description is manifestly local in the $x$-direction.
RG flows and the corresponding conformal boundary conditions
------------------------------------------------------------
In the decompactification limit $\nu \to \infty $ all excited states in the spectrum acquire infinite energy above the vacuum. As conformal symmetry is restored we expect the vacuum states of the infinitely massive theory to be described as conformal boundary states in the massless theory ${\cal H}_{0}$.
For $m>0$ the system is in the low temperature ordered phase. As $m$ goes to plus infinity the vacuum becomes doubly degenerate. This happens exponentially fast in $\nu$. More precisely the difference of vacuum energies in the NS- and R- sectors is e (E\_[0([R]{})]{} - E\_[0([NS]{})]{} ) = \_[-]{}\^ () that for $\nu\to \infty$ behaves as e \~\^[-3/2]{}e\^[-]{} . At $m = +\infty$ only a two-dimensional vacuum space is left in the spectrum. Using (\[IONS\]), (\[IOR\]) we can find the expressions for the asymptotic vacuum states. As they have infinite norm we should think of them, as we usually do for boundary states, as vectors in the dual space ${\cal H}_{0}^{*}$.
The details are as follows. The normalisation factors ${\cal N}(\nu)$, $\tilde {\cal N}(\nu)$ each goes to zero due to the Casimir energy factors \[Casimir\] in (\[N1\]), (\[N2\]). This reflects the fact that the conformal boundary states to which the massive vacuum approaches have infinite norm. Stripping off this vanishing factor we obtain the following asymptotic representations of the vacuum sectors of ${\cal H}_{\infty}$ in ${\cal H}_{0}$ |0\_[NS]{}\^[+]{}\_[m+]{} D\_[NS]{}|0\_[NS]{} = \_[n=0]{}\^(ia\^\_[n+1/2]{}|a\^\_[n+1/2]{})|0 , |0\_[R]{}\^[+]{}\_[m+]{} D\_[R]{}|0\_[R]{} = 2\^[1/4]{} \_[n=1]{}\^(ia\^\_[n]{}|a\^\_[n]{})| . These states can be decomposed into conformal Ishibashi states[^4] [@Cardy] using the identification of the energy primary state \[eps\_state\] |= ia\^\_[1/2]{}|a\_[1/2]{}\^|0 . We obtain |0\_[NS]{}\^[+]{} = |0+ | , |0\_[R]{}\^[+]{} = 2\^[1/4]{} | . We observe that these states already have correct relative normalisations to be combined into conformal boundary states. By multiplying them by $1/\sqrt{2}$ and choosing the appropriate relative phases between the NS- and R- sector components we combine them into two Cardy states which represent local boundary conditions corresponding to fixed boundary spin: $|+\rangle\!\rangle$, $|-\rangle\!\rangle$ given in (\[conf\_bcs\]). Hence the RG boundary state that corresponds to the $m\to +\infty$ flow is a superposition $$|{\rm RG}\rangle\!\rangle = |+\rangle\!\rangle \oplus |-\rangle\!\rangle \, .$$
For $m<0$ the system is in the high temperature disordered phase. In the R-sector the lowest energy state is the Fock space one-particle excitation $b_{0}^{\dagger}|0\rangle_{\rm R}$ so that all R-sector states disappear from the spectrum as $m\to -\infty$. The remaining vacuum state is |0\_[NS]{}\^[-]{}\_[m-]{} D\_[NS]{}|0\_[NS]{} = \_[n=0]{}\^(-ia\^\_[n+1/2]{}|a\^\_[n+1/2]{})|0 , which as above can be decomposed into the Ishibashi states as |0\_[NS]{}\^[-]{} = |0- | that already corresponds to a normalised Cardy boundary state $|F\rangle\!\rangle = |0\rangle\!\rangle - |\epsilon\rangle\!\rangle $ giving the free boundary condition. Thus in this case |[RG]{}= |F .
In the presence of magnetic field (the case we take up in the next section) the main quantities we will calculate numerically are component ratios of the type (\[gen\_rat\]) discussed in the introduction. Specifically for the massive flows of the Ising field theory it is convenient to introduce two ratios: \[Ising\_ratios\] \_\^[0]{} = , \_\^[0]{} = . Here $|v_{0}\rangle$ stands for the vacuum of perturbed theory as it appears inside the critical Ising state space.
It is instructive to look at these component ratios for the $h=0$ flows where we can get the exact formulas for them. The (unnormalised) vacuum of the perturbed theory (at finite value of $m$) is $$|v_{0}\rangle=\hat D^{\rm NS}|0\rangle_{\rm NS}$$ that always lies in the NS-sector. From (\[IONS\]) we find \_\^[0]{}=0 , \_\^[0]{} = = = [sign]{}(m) g() where \[gnu\] g()= - . The function $g(\nu)$ asymptoticaly goes to 1 very slowly, as a power function. Here is a plot of $g(\nu)$
\[g\_fig\] 
This function monotonically increases and becomes larger than $0.9$ past $\nu=30$ and larger than $0.99$ past $\nu=313$. These values of the dimensionless mass are very large. This shows that it may be much harder to read off the component ratios from TCSA numerics than say the energy spectrum for which the convergence is typically exponential. The further we have to go into the infrared the greater the TCSA errors can be for a fixed truncation level. We will see that while the perturbative corrections grow the numerics demonstrates that the total truncation error remains bounded.
For $m\to +\infty$ the first excited vector remains in the spectrum. It is $|v_{1}\rangle=\hat D^{\rm R}|0\rangle_{\rm R} $ for which we can define a ratio of components | | = g() that again converges as slowly as $\Gamma_{\epsilon}^{0}$.
TCSA for Ising field theory
===========================
We would like now to consider a generic perturbation of the critical Ising model with $m\ne 0 $ and $h\ne 0$. For real-valued $h\ne 0$ the system flows to a trivial theory with a single vacuum. We can study numerically the ratios $\Gamma_{\epsilon,\sigma}^{0}$ using TCSA.
We work in the massless fermion physical space ${\cal H}_{0}$. The perturbed Euclidean action of the Ising field theory is given in (\[IFT\]). Let us choose an orthonormal basis in ${\cal H}_{0}$ described in section \[preliminaries\]. In the NS sector we choose \[basis1\] ||k\_1, …|k\_[p]{}; k\_[1]{}, …k\_[q]{}\_[NS]{} = |a\_[|k\_1]{}\^…|a\_[|k\_p]{}\^a\_[k\_[1]{}]{}\^…a\_[k\_q]{}\^|0 where k\_[i]{}, |k\_[j]{} 1/2 + [Z]{} , k\_i>0 , |k\_j > 0 , p+q - and |k\_1 > |k\_2 >…> |k\_p , k\_[1]{}> k\_[2]{} >…> k\_[q]{} . In the R-sector we choose \[basis2\] ||n\_1, …|n\_[p]{}; n\_[1]{}, …n\_[q]{}\_[R]{} = |a\_[|n\_1]{}\^…|a\^\_[|n\_p]{}a\^\_[n\_[1]{}]{}…a\_[n\_q]{}\^| , p + q - and \[basis3\] ||n\_1, …|n\_[p]{}; n\_[1]{}, …n\_[q]{}\_[R]{} = |a\_[|n\_1]{}\^…|a\^\_[|n\_p]{}a\^\_[n\_[1]{}]{}…a\_[n\_q]{}\^| , p + q - . Here we assume n\_[i]{}, |n\_j , n\_[i]{}>0, |n\_j>0 and |n\_1 > |n\_2 >…> |n\_p , n\_[1]{}> n\_[2]{}> …> n\_[q]{} . We are going to work in a finite-dimensional truncated space ${\cal H}_{0}^{\rm tr}$ is spanned by (\[basis1\]), (\[basis2\]), (\[basis3\]) satisfying the following additional constraints \_[i=1]{}\^[p]{}|k\_i = \_[i=1]{}\^[q]{}k\_i n\_c , \_[i=1]{}\^[p]{}|n\_i = \_[i=1]{}\^[q]{}n\_i n\_c where $n_c$ is the integer that sets the level truncation[^5]. We choose the space ${\cal H}_{0}^{\rm tr}$ to contain only zero spin vectors that is relevant for identifying RG boundary states.
$n_c$ dimension of truncated NS space dimension of truncated R space Total dimension of truncated space Number of diagonal states
------- --------------------------------- -------------------------------- ------------------------------------ ---------------------------
8 91 97 188 58
9 141 161 302 76
10 226 261 487 99
11 354 405 759 127
12 556 630 1186 162
13 844 954 1798 204
$n_c$ 8 9 10 11 12 13 14 15 16 17 18
------------------------------------ ----- ----- ----- ----- ------ ------ ------ ------ ------ ------ -------
dimension of truncated NS space 91 141 226 354 556 844 1296 1909 2838 4139
dimension of truncated R space 97 161 261 405 630 954 1438 2167 3191 4635 6751
Total dimension of truncated space 188 302 487 759 1186 1798 2734 4076 6029 8774 12820
Dimension of diagonal subspace 58 76 99 127 162 204 256 318 393 482 590
In the table below we give dimensions of truncated spaces for a range of values of $n_{c}$. In the last row we give dimension of the truncated subspace of diagonal states - this one is spanned by states of the form (\[basis1\]) with $p=q$ and $k_i=\bar k_i$, $i=1, \dots p$ in the NS sector and similarly by states (\[basis2\]), (\[basis3\]) with $n_{i}=\bar n_i$ in the R-sector. These states are of particular interest because the conformal Ishibashi states are linear combinations of such states.
$n_c$ 8 10 11 12 13 14 16 17 18
------------------------------------ ----- ----- ----- ------ ------ ------ ------ ------ -------
dimension of truncated NS space 91 226 354 556 844 1296 2838 4139 6069
dimension of truncated R space 97 261 405 630 954 1438 3191 4635 6751
Total dimension of truncated space 188 487 759 1186 1798 2734 6029 8774 12820
Dimension of diagonal subspace 58 99 127 162 204 256 393 482 590
In operator quantisation the matrix elements of $\sigma$ factorise into holomorphic and antiholomorphic factors \[sigma\_matrix\] && \_[NS]{}|k\_1, …|k\_p, k\_1 …, k\_[q]{}|(0,0)||n\_1,…|n\_r, n\_1,…, n\_[s]{} \_[R]{}\
&& = ()\^[1/8]{}(-1)\^[q(q-1)/2 + s(s-1)/2 + r(s+1)]{} G(|k\_1, …|n\_1, …,|n\_[r]{} )\
&& G(k\_1, …, k\_q, n\_1, …, n\_s ) where \[G\] && G( k\_1, …k\_p,n\_1, …,n\_[r]{} ) = \_[j=1]{}\^[p]{} g\_[NS]{}( k\_j) \_[i=1]{}\^[r]{}g\_[R]{}( n\_i) (\_[1i<jp]{} )\
&& (\_[1i<j q]{} )(\_[1i p; 1jr]{} ) where g\_[NS]{}(k) = , g\_[R]{}(n) = are the massless leg factors. The antiholomorphic factor $G(\bar k_1, \dots \bar n_1, \dots ,\bar n_{r} )$ is given by the same formula as (\[G\]) with $n_i, k_j$ mode numbers replaced by $\bar n_i, \bar k_j$.
The RG trajectories of the Ising field theory (\[IFT\]) are labeled by a dimensionless parameter y= for which we use the same notation as [@FZ]. We also use a dimensionless distance scale r=R|h|\^[8/15]{} . In terms of these two scaling variables the dimensionless mass can be expressed as = R|m| = |y| r . The TCSA Hamiltonian is \[H\_TCSA\] H\_[TCSA]{} = . Here $H_{0}$, $B$ and $M$ are matrices not containing any dimensional parameters or scaling variables $y$, $r$. They are defined as follows. We have H\_[0]{} = \_[n=0]{}\^ (n + 1/2)\[a\^\_[n+1/2]{}a\_[n+1/2]{} + |a\^\_[n+1/2]{}|a\_[n+1/2]{}\] - in the NS-sector and H\_[0]{} = \_[n=0]{}\^ n\[ a\^\_[n]{}a\_[n]{} + |a\^\_[n]{}|a\_[n]{}\] + in the R-sector. The matrix $B$ has the same matrix elements as (\[sigma\_matrix\]) with the factor $\left(\frac{2\pi}{R}\right)^{1/8}$ removed. The matrix $M$ is the matrix for the operator $i\psi\bar \psi$ in the basis given by (\[basis1\]), (\[basis2\]), (\[basis3\]).
If $E(r)$ is an eigenvalue of $H_{\rm TCSA}$ we also define dimensionless eigenvalues e(r) = E(r) and E(r) = . For a flow to Yang-Lee fixed point the differences of the eigenvalues $e(r)$ interpolate between the differences of scaling dimensions at the fixed points (see a detailed discussion in section \[ycr\_sec\]).
To identify RG boundaries for the massive flows we are going to calculate numerically the component ratios $\Gamma^{0}_{\epsilon}$, $\Gamma^{0}_{\sigma}$ defined in (\[Ising\_ratios\]). Also later we will consider the RG flow to Yang-Lee fixed point which occurs at imaginary magnetic field. In that case we will also look at the ratios \[Ising\_ratios2\] \_\^[i]{} = , \_\^[i]{} = corresponding to the $i$-th excited energy eigenstates $|v_{i}\rangle$, $i=1,2,\dots$. TCSA results at finite $n_{c}$ of course differ from the exact renormalised QFT results at $n_{c}=\infty$. Some of the corrections correspond to redefining the coupling constants and can be taken into account using a certain RG scheme [@GW], while there are also non-local corrections [@HRvR], [@RV]. The redefined coupling constants corresponding to the RG scheme of [@GW] can be approximated by running couplings that depend on continuous $t=1/n_{c}$. Let $\tau=yr={\rm sign}(y)\nu$. The RG equations for the running couplings $\tau(t)$, $r(t)$ then read =-Ar , =-sBt\^[7/4]{}r\^[15/4]{} where A = 0.01 , B=0.004 and $s=1$ for real magnetic field and $s=-1$ for the imaginary one. These equations are to be solved with the bare couplings from (\[H\_TCSA\]) taken as initial condition at $t=0$. Both couplings change quite slowly. For example for $y=y(0)=-3$ and $s=-1$ (close to the Yang-Lee trajectory) we find numerically for $n_c =12$ that for $r=r(0)=12$ the effective values are $r(1/12)\approx 12.49$, $\tau(1/12)\approx -35.98$. The mass coupling $\tau$ changes particularly slowly.
The method for dealing with non-local perturbative corrections to energy eigenvalues has been worked out in [@HRvR], [@RV]. One strategy to implement such corrections in practice is by varying $n_{c}$ and by fitting the variations in numerical results to combinations of negative powers of $n_{c}$. For the Ising field theory this has been done in [@LT_Potts], [@Takacs_quenches] (in particular see Appendix B in [@Takacs_quenches] where RG corrections and fittings in both TCSA and TFFSA are discussed in detail). Perturbative truncation corrections to eigenvectors (which would be of interest for calculating the component ratios $\Gamma^{i}_{\epsilon, \sigma}$) have not been so far systematically investigated. Moreover as Figure \[g\_fig\] shows one can expect a very slow approach of $\Gamma^{i}_{\epsilon, \sigma}$ to their IR asymptotics that does not make perturbative corrections useful for finding their IR values. Nevertheless our TCSA numerical results, that we present in the next section, show that the component ratios for real magnetic field asymptotically approach constant values with (non-perturbative) truncation errors staying bounded. Since the running couplings corrections are really tiny and do not change any qualitative conclusions and also to make direct comparison with the results of [@FZ], [@Zam1], [@Zam2] possible, everywhere in the paper we present uncorrected (raw) TCSA and TFFSA data.
Massive flows. Real magnetic field.
===================================
TCSA data {#TCSA_section}
----------
The component ratios we are going to find numerically are $\Gamma^{0}_{\epsilon}$ and $\Gamma^{0}_{\sigma}$ defined in (\[Ising\_ratios\]). The sign of $\Gamma^{0}_{\sigma}$ is correlated with the sign of $h$. For simplicity we will always choose $h$ so that $\Gamma^{0}_{\sigma}$ is positive.
When a magnetic field perturbation is present it is natural to expect the $|\sigma\rangle\!\rangle$ component to be present in the RG boundary state. As we don’t expect any vacuum degeneration we expect the RG boundary state to be $|+\rangle\!\rangle$ or $|-\rangle\!\rangle$ depending on the sign of the coupling $h$. This means that $\Gamma_{\epsilon}^{0}(r)$ should approach 1 and $\Gamma_{\sigma}^{0}(r)$ should approach $2^{1/4}\approx 1.18921$. Indeed this is what we observe in TCSA numerics presented below for a range of values of $y$. The approach to asymptotic values is always quite slow - of the type the exact solution for $\Gamma_{\epsilon}^{0}$ demonstrates in the $h=0$ case. On the graphs for $\Gamma_{\sigma}^{0}$ below the red dashed line corresponds to the value $2^{1/4}$. The results presented below are not very sensitive to the the cutoff level $n_c$ (see a more detailed discussion below). For illustration we chose $n_c=12$.
We observe that the ratio $\Gamma_{\sigma}^{0}$ converges to its expected limiting value with much better accuracy than $\Gamma_{\epsilon}^{0}$. While for the values of $r\sim 50$ we have $\Gamma_{\sigma}^{0}\approx 1.17$ for $\Gamma_{\epsilon}^{0}$ typical asymptotic values are near $0.9$. In the disordered phase, where $y<0$, the mass term dominates at small distances so that $\Gamma_{\epsilon}^{0}$ starts out with negative values. The magnetic field takes over at sufficiently large distances with $\Gamma_{\epsilon}^{0}$ passing through zero and monotonically increasing to a positive asymptotic value that is less than 1. Thus for $y<0$ the convergence of $\Gamma_{\epsilon}^{0}$ to its asymptotic value is slower than in the ordered phase.
It should be noted that at the large values of $r$ we considered: $30<r<100$, the values of $\nu/n_{c}$ and $hr^{15/8}/n_{c}$ are large. This means that any perturbative corrections are large in that region. However the numerics shows that the perturbative corrections must sum up to a form $f(r)/n_{c}$ where $f(r)$ is a bounded function of $r$. The following plot shows how the numerical values of $\Gamma_{\epsilon}^{0}$ taken for $y=1$ change when we change the truncation level $n_{c}$.
It is not clear to us why the total truncation correction remains bounded for such large values of the scale but this is certainly crucial for getting reasonable numerical results for the ratios $\Gamma_{\epsilon, \sigma}^{0}$ due to their extremely slow asymptotics.
We focused on two particular ratios $\Gamma^{0}_{\epsilon, \sigma}$ for convenience and for practical reasons (these ratios are related to the lowest weight components and we expect them to be best approximated by TCSA). But in general we expect all components of the asymptotic vacuum vector to arrange themselves according to the components of Cardy boundary states (or their superposition). To get a better measure of that we can plot all vacuum components. More precisely let $|i\rangle_{\rm NS}$, $|j\rangle_{\rm R}$ stand for the basis elements in the critical Ising model that correspond to scaling fields of dimension $\Delta_{i}^{\rm NS}$, $\Delta_{j}^{\rm R}$ in the NS and R-sectors respectively. Up to phases the basis elements are given by formulae (\[basis1\]), (\[basis2\]), (\[basis3\]). We fix the phases by state-operator correspondence (see (\[eps\_state\]) and label these basis vectors in such a way that the conformal weights are monotonically non-decreasing: $\Delta_{i}^{\rm NS}\le \Delta_{j}^{\rm NS}$ if $i\le j$ and similarly in the R-sector. Let $C^{\rm NS, R}(i)$ be the vacuum vector components relative to this basis: \[conf\_basis\] |0= \_[i]{} C\^[NS]{}(i)|i\_[NS]{} + \_[j]{} C\^[R]{}(j)|j\_[R]{} . For $n_{c}=12$ we have 556 basis vectors in the NS sector out of which 92 are diagonal. On the graph below all components are plotted, out of which only the diagonal ones are clearly visible while the rest of the components are much smaller and on the graph they are concentrated on the $i$-axis. We see that the diagonal components have the same phase and decrease in amplitude with increasing conformal weight practically forming a continuous curve[^6]. The decrease is a truncation effect. In the untruncated theory we would expect all components to be the same according to the composition of the Cardy states (\[conf\_bcs\]).
The Ramond components form a similar pattern with the diagonal components decoupled from the much smaller non-diagonal ones.
Another way to see the domination of the diagonal states in the vacuum is by calculating the share of the diagonal states in the square of the norm: S= . We found that numerically $S$ is above $0.9999$ for a range of $y$, $n_{c}$ and $r$ between $0$ and $50$. Below is a sample graph
As can be seen from the table given on Fig. \[table\] the number of diagonal states grows significantly slower with $n_c$ than the total number of states. It is remarkable that these states dominate the vacuum with such a high proportion. Even more unexpectedly we found that the first and second excited states have about the same high proportion of the diagonal states in their norm. It remains to be seen whether this observation can be put to use to improve the numerics.
Summary {#massive_summary}
-------
Our analytical results for the vanishing magnetic field together with the numerical results for measuring $\Gamma_{\epsilon}^{0}, \Gamma_{\sigma}^{0}$ are summarised on the diagram depicted on Fig. \[diagram\_real\]. We observe that the two exceptional RG boundaries: $|+\rangle\!\rangle \oplus |-\rangle\!\rangle$, $|F\rangle\!\rangle$, are unstable. There are boundary RG flows from each of them that end up either with $|+\rangle\!\rangle$ or with $|-\rangle\!\rangle$. These end points are precisely the RG boundary conditions corresponding to the one-dimensional regions joined at the exceptional points. This fact has the following physical explanation. Suppose we are at an exceptional point that belongs to a submanifold separating higher-dimensional regions labeled by different RG boundaries. Far in the infrared the RG interface is almost non-transparent and can be well approximated by a conformal boundary condition. If we add now a small bulk perturbation on the massive side of the interface that moves us away from the separating submanifold this will result in an effective perturbation of the RG boundary and a subsequent effective RG flow to a new RG boundary condition. (More precisely this is still a bulk plus boundary flow, however in this representation low energy degrees of freedom survive only on the boundary so it is effectively described by a pure boundary flow.)
In the Ising field theory case we can be more quantitative in describing these effective boundary RG flows that start from the exceptional RG boundaries. These flows are triggered by the $\sigma$-perturbation taken at a large dimensionless mass $\nu$. The matrix elements of $\sigma$ in a massive Fock space are given explicitly in formula (\[matrix\_mass\]). In the limit $\nu \to \infty$ all of these matrix elements go to zero or to a finite constant except for the vacuum-vacuum and the vacuum-one-particle ones that diverge. This means that for $m>0$ effectively the $\sigma$ perturbation at large mass will act as a boundary identity field that mixes the two vacua, while for $m<0$ the one-particle matrix elements give rise to a boundary magnetic field perturbation.
Truncated Free Fermion Space Approach and component ratios
==========================================================
In this section we outline how the truncated free fermion space approach (TFFSA) invented in [@FZ] can be used to find numeric approximations to the ratios $\Gamma_{i}$. This method has the advantage over the TCSA in treating the mass coupling non-perturbatively and also in having greater control over the large $r$ asymptotics. However it has its own subtleties related to UV divergences which we are going to discuss as well.
In TFFSA one uses the massive fermion physical space ${\cal H}_{m}$ described in section \[preliminaries\]. We write the Ising field theory Hamiltonian as \[Ham2\] H = H\_[FF]{} + h\_[0]{}\^[R]{} (x,0) dx where $H_{\rm FF}$ is the free massive fermion Hamiltonian given in (\[HFF1\]), (\[HFF2\]) and the matrix elements of the magnetic field perturbation are given by [@FZ] \[matrix\_mass\] && \_[NS]{}k\_1, k\_2,…, k\_[N]{}|(0,0)|n\_1,n\_2, …, n\_[M]{}\_[R]{}= i\^ |S(R) \_[i=1]{}\^[N]{}g(\_[k\_i]{}) \_[j=1]{}\^[M]{} g(\_[n\_j]{})\
&& \_[1i<j<N]{}() \_[1p<q<M]{}()\
&& \_[1rN; 1sM]{}() . Here $\theta_{n}, \theta_{k}$ are finite size rapidities in the R- and NS-sectors: (\_[n]{}) = , n (\_[k]{}) = , k + [Z]{} , the functions $g(\theta), \tilde g(\theta)$ are the leg factors defined as g() = , g() = where () = \_[-]{}\^[+]{} . The overall factor $ \bar \sigma S(R)$ in (\[matrix\_mass\]) is the vacuum-vacuum matrix element |S(R)= {
[l@l]{} \_[NS]{}0|(0,0)|0\_[R]{} , &m>0 ,\
\_[NS]{}0|(0,0)|0\_[R]{} , &m<0 .
. We have the following explicit expressions |= |m|\^[1/8]{}2\^[1/12]{}e\^[-1/8]{}A\^[3/2]{} where $A$ is Glaisher-Kinkelin constant, and S(R) = ()\^2 d\_1d\_2 .
The Hamiltonian (\[Ham2\]) is restricted to a truncated space ${\cal H}_{m}^{\rm tr}\subset {\cal H}_{m}$ that is spanned by vectors (\[mbasis1\]), (\[mbasis2\]) satisfying \_[i=1]{}\^[N]{} k\_i = \_[j=1]{}\^[M]{}n\_j = 0 , and \_[i=1]{}\^[N]{}|k\_i| 2n\_c , \_[j=1]{}\^[M]{} |n\_j| 2n\_c . Here $n_c$ is an integer that controls the truncation. In contrast with the TCSA, it no longer is related to the energy of the unperturbed Hamiltonian $H_{\rm FF}$. The total dimensions of truncated spaces are the same as the ones given in Figure \[table\].
While the TFFSA eigenvectors lie in ${\cal H}_{m}$ we can use the interface operator $\hat D_{m}^{-1}: {\cal H}_{m}\to {\cal H}_{0}$ to obtain their image in ${\cal H}_{0}$. This can be formally thought of as combining (or fusing) two perturbation interfaces: the mass interface $\hat D^{-1}_{m}$ and the magnetic field interface $\hat D^{-1}_{m,h}$. The last one corresponds to perturbing the free massive theory by the magnetic field.
(3.4,0) – (5.8,0) arc (270:90:0.6 and 1.1) – (3.4, 2.2) arc (90:270: 0.6 and 1.1)–cycle; (5.8,0) – (9,0) arc (-90:90:0.6 and 1.1 ) – (5.8, 2.2) arc (90:270: 0.6 and 1.1)–cycle; (0,0) arc (270:90:0.6 and 1.1); (0,0) arc (-90:90:0.6 and 1.1 ); (0,0)–(9,0); (0,2.2 cm )–(9, 2.2 cm); (9,0) arc (270:90:0.6 and 1.1); (9,0) arc (-90:90:0.6 and 1.1); (3.4,0) arc (270:90:0.6 and 1.1); (3.4,0) arc (-90:90:0.6 and 1.1); (5.8,0) arc (270:90:0.6 and 1.1); (5.8,0) arc (-90:90:0.6 and 1.1 );
(3.6,-0.35) node [${ D}_{m}^{-1}$]{}; (6.0,-0.4) node [${ D}_{m,h}^{-1}$]{}; (1.5,1.2 ) node [$m=0 $]{}; (1.5, 0.85) node [$h=0$]{}; (4.6,1.2 ) node [$m\ne 0 $]{};(4.65,0.85 ) node [$ h=0$]{};(7.5,1.2 cm) node [$m\ne 0$]{}; (7.5,0.85 cm) node [$ h\ne 0$]{}; (3.4, 2.25) – (3.4,2.6); (5.8, 2.25) – (5.8,2.6); (3.4, 2.45) – (5.8,2.45); (4.55,2.64) node [$\epsilon$]{};
The mass interface operator was constructed analytically in section \[mass\_interface\] while we can use the TFFSA numerics to obtain information on the second interface. Fusion of conformal interfaces usually contains multiplicative divergences (see e.g. the discussion in [@BBDO], [@BB]). This divergence is regulated by the truncation of ${\cal H}_{m}$ present in TFFSA. However one may still worry whether the fusion procedure gives the same interface as the TCSA one in the $n_{c}\to \infty$ limit. While we cannot exclude this situation with definiteness we have not observed anything in the numerical results to be presented below that would suggest this scenario[^7].
Let $P_{n_c}: {\cal H}_{m} \to {\cal H}_{m}^{\rm tr}$ be the projector implementing TFFSA truncation at level $n_c$. Then we can write for the component ratios (\[Ising\_ratios\]), (\[Ising\_ratios2\]) \_\^[i]{} = , \_\^[i]{} = . Here and below $i=0,1,2,\dots$. Using (\[IONS\]), (\[IOR\]) we calculate \_\^[i]{}= -[sign]{}(m) , \_\^[i]{} = 2\^[1/4]{}f() where $f(\nu)$ and $g(\nu)$ are given by (\[Fnu\]), (\[gnu\]), and T\_1(v,)= \_[NS]{}0|( -i g() b\_[-1/2]{}b\_[1/2]{} ) \_[n=1]{}\^[n\_c-1]{} ( 1 - i b\_[-n-1/2]{}b\_[n+1/2]{}) |v , T\_2(v,)= \_[NS]{}0| \_[n=1]{}\^[n\_c-1]{} ( 1 - i b\_[-n-1/2]{}b\_[n+1/2]{}) |v , U(v,) = \_[R]{}0| (b\_[0]{})\^[p]{} \_[n=1]{}\^[n\_c]{}( 1 - i b\_[-n]{}b\_[n]{}) |vwhere $p=1$ if $y<0$ and $p=0$ when $y>0$.
We see from these expressions that part of the scale dependence of $\Gamma_{\epsilon, \sigma}^{i}$ comes from the functions $f(\nu)$ and $g(\nu)$ and from $\Delta \omega_{k}$ which we know analytically. In practice we observed a faster rate of convergence of $\Gamma_{\sigma}^{i}$ to its asymptotic value than that of $\Gamma_{\epsilon}^{i}$. This can be at least partially attributed to the faster convergence of $f(\nu)$ than that of $g(\nu)$ to its asymptotic value 1. Having noted this we can formally define quantities in which $\hat D_{\rm NS}$, $\hat D_{\rm R}$ are taken at $\nu = \infty$: \_\^[i]{}=-[sign]{}(m) , \_\^[i]{} = 2\^[1/4]{} . In practice we found that using $\tilde \Gamma_{\epsilon, \sigma}^{i}$ gives a small improvement in the convergence rate towards the asymptotic values. As these quantities do not have a clear physical meaning (the eigenvector $|v_{i}\rangle$ is still taken at finite $\nu$) we are not going to present the numerical results for them in the paper.
Applied to the flows triggered by real $h$ the TFFSA method here described gives data very similar to the TCSA one presented in section \[TCSA\_section\]. The method however has a significant advantage when applied to the imaginary magnetic field flows discussed in the forthcoming sections.
Imaginary magnetic field. Complex vacuum energy. {#Complex_vac_sec}
================================================
The Ising field theory (\[IFT\]) taken at imaginary values of $h$ is not unitary. However the corresponding Hamiltonian $H$ enjoys the following symmetry \[Ssym\] SHS=H\^ where $S$ is the operator that multiplies any Ramond sector vector by $-1$ and leaves any NS sector vector intact. To see the implications of this symmetry consider $H$ as an operator in ${\cal H}_{0}$ and choose a basis in which $H$ is (complex) symmetric.
For example we can take the conformal basis described before equation (\[conf\_basis\]). In this basis the matrices $M$ and $B$ in (\[H\_TCSA\]) are symmetric and hence $H$ is symmetric as well. As before denote these basis vectors as $|i\rangle_{\rm R}$, $|j\rangle_{\rm NS}$. If |v\_= \_[i]{} C\^[NS]{}(i)|i\_[NS]{} + \_[j]{} C\^[R]{}(j)|j\_[R]{} is an eigenvector of $H$ with eigenvalue $\lambda$ then ||v\_\_[i]{} |C\^[NS]{}(i)|i\_[NS]{} + \_[j]{} |C\^[R]{}(j)|j\_[R]{} is an eigenvector of $\bar H$ with eigenvalue $\bar \lambda$. (The bar stands for complex conjugation everywhere.) Then (\[Ssym\]) in this particular basis implies that HS||v\_= |S||v\_that is |v\_[|]{}\_[i]{} |C\^[NS]{}(i)|i\_[NS]{} - \_[j]{} |C\^[R]{}(j)|j\_[R]{} is an eigenvector of $H$ with eigenvalue $\bar \lambda$. Thus the energy eigenvalues are either real or form a pair of complex conjugated values. Moreover the above implies that if the vacuum of $H$ has real energy and is non-degenerate then the vacuum vector must be of the form |v\_[0]{}= |0+ C\^[NS]{}|+ iC\^[R]{}|+ … where $C^{\rm NS}$, $C^{\rm R}$ are real and we show only the three lowest components. Hence in this case $\Gamma_{\epsilon}^{0}$ is real and $\Gamma_{\sigma}^{0}=iC^{\rm R}$ is imaginary.
Alternatively if the vacuum energy is complex and the vacuum space is a two-dimensional subspace corresponding to two conjugate eigenvalues then the corresponding eigenvectors can be written as \[pair\_eig\] && |v\_[0]{}= |0+ C\^[NS]{}e\^[i]{}|+ C\^[R]{} e\^[i]{} |+ …\
&& ||v\_[0]{}= |0+ C\^[NS]{}e\^[-i]{}|- C\^[R]{} e\^[-i]{} |+ …where $C^{\rm NS}$, $C^{\rm R}$ are real and hence $\Gamma_{\epsilon}^{0}=C^{\rm NS}e^{\pm i\chi}$ $\Gamma_{\sigma}^{0}=\pm C^{\rm R}e^{\pm i\phi}$ are both complex in general. Since none of the conformal boundary states (\[conf\_bcs\]) has a complex value of $\Gamma_{\sigma}$ we see[^8] that in the first case [*as long as the vacuum eigenvalue remains real the vacuum vector has no chance of approaching a conformal boundary state as we move along an RG trajectory[^9].*]{} In the second case the only way $\Gamma_{\epsilon}$ and $\Gamma_{\sigma}$ can become real at the end of the RG flow is if the phases $e^{i\chi}$, $e^{i\phi}$ tend to 1 or -1 as $r\to \infty$. If this is the case the eigenvectors given in (\[pair\_eig\]) should either tend to the pair of fixed boundary states $|\pm\rangle\!\rangle$ or to $|F\rangle\!\rangle\oplus |F\rangle\!\rangle$. We will present evidence that supports the first possibility.
We now focus on the case of complex vacuum energy. The spectrum of the Ising field theory for imaginary magnetic field and arbitrary mass was investigated in [@FZ], [@Zam1], [@Zam2] using TFFSA as well as analytic results. It was shown in those papers that for $y>y_{\rm cr}$ where $y_{\rm cr}\approx -2.429$ the vacuum energy becomes complex for sufficiently large values of $r$. The vacuum vectors form a pair with conjugated energy values. As discussed in [@FZ] for positive $y$ and large enough $r$ the vacuum and excited states form complex pairs with ${\rm Re} \tilde E(r)$ asymptotically approaching constant values and ${\rm Im} \tilde E(r)$ asymptoting to straight lines with equal slopes (free energy density) thus furnishing particle-like excitations with complex masses.
Here we present plots of the numerical results (using TFFSA) for the ratios $\Gamma_{\epsilon}$, $\Gamma_{\sigma}$ calculated for the two vacuum vectors at $y=2$.
Comparing these plots to the ratios in conformal boundary conditions (\[conf\_bcs\]) we find that the RG boundary is a superposition $|+\rangle\!\rangle \oplus |-\rangle\!\rangle$.
The case of large positive $y$ can be understood in terms of boundary RG flows (see the discussion in section \[massive\_summary\]). This regime corresponds to starting with a large $m>0$ and adding to it a small imaginary magnetic field perturbation. Thus with a good approximation we start with a vacuum described by the conformal boundary condition $|+\rangle\!\rangle \oplus |-\rangle\!\rangle$ and the perturbation is just the identity field between the two boundary conditions multiplied by an imaginary coupling. The only effect of this is a relative phase factor which keeps rotating as we change the scale but does not change any physical quantities.
We have also checked numerically the region $y_{\rm cr}<y<0$. As long as the vacuum complex pair does not experience collisions with upper level eigenvalues the picture is qualitatively the same as presented on the above plots. Such collisions are sensitive to the truncation level so strictly speaking we do not have a proof that the above picture remains in the extreme $r\to \infty$ asymptotics at infinite truncation level. But it looks highly plausible to us.
To summarise the numerical results support the RG boundary being $|+\rangle\!\rangle \oplus |-\rangle\!\rangle$ for all cases when the vacuum energy is complex and we have a pair of vacuum vectors.
Imaginary magnetic field. Real vacuum energy.
=============================================
Massive flows {#Im_massive}
-------------
For $y\le y_{\rm cr}$ numerical studies [@FZ] show that the vacuum energy remains real until a large positive value if $r$. The latter value at which the vacuum eigenvalue collides with the first excited eigenvalue and forms a conjugate pair is numerically the larger the larger $n_c$ is and presumably such collision is a numerical artefact. We found that TFFSA method works much better than TCSA in this regime. In particular the lowest eigenvalue stays real for much larger value of $r$ when using the TFFSA method.
We start discussing the numerical results for this region by taking a large negative $y$ where the vacuum energy stays real for large $r$’s and where we can use the boundary RG picture discussed in section \[massive\_summary\]. Below we present numerical data for a sample point $y=-3.5$. One finds that the behaviour of the component ratios is very different here from the situations discussed before. Namely the overlap of the perturbed vacuum with the unperturbed one goes to zero and both $\Gamma^{0}_{\epsilon}$ and $\Gamma^{0}_{\sigma}$ go to infinity at a finite value of RG scale $r$. The ratio $\Gamma^{0}_{\epsilon}/\Gamma^{0}_{\sigma}$ remains finite. Define T= , =|| . Here is a plot of these ratios for $y=-3.5$
![Imaginary magnetic field. The component ratios $\Gamma_{\epsilon}^{0}$ and $\Gamma_{\sigma}^{0}$ at $y=-3.5$, $n_{c}=12$.[]{data-label="data1"}](G1blowup_v2_corr.pdf)
![Imaginary magnetic field. The component ratios $\Gamma_{\epsilon}^{0}$ and $\Gamma_{\sigma}^{0}$ at $y=-3.5$, $n_{c}=12$.[]{data-label="data1"}](G2blowup_v2_corr.pdf)
![Imaginary magnetic field. The component ratios $T$ and $\Gamma$ at $y=-3.5$, $n_{c}=12$.[]{data-label="data2"}](Ty-3point5_v2.pdf)
![Imaginary magnetic field. The component ratios $T$ and $\Gamma$ at $y=-3.5$, $n_{c}=12$.[]{data-label="data2"}](Gy-3point5_v2.pdf)
We find that $T$ vanishes and $\Gamma^{0}_{\epsilon,\sigma}$ blow up at $r=r_{1}\approx 25.5$. The ratio $\Gamma$ remains finite at this point with the value $\Gamma(25.5)=\Gamma^{*}_{\rm num} \approx 2.0445$.
For large negative values of $y$ we can approximate the flow of the vacuum vector by an RG boundary flow. For large negative values of $m$ the vacuum is well approximated by the boundary state $|F\rangle\!\rangle$ corresponding to free boundary spin. Switching on a small imaginary magnetic field in the bulk perturbed this boundary condition by an imaginary boundary magnetic field. The critical Ising model with a boundary magnetic field is a Gaussian theory and can be solved exactly [@CZ], [@Chat]. It is not hard to extend this solution to imaginary magnetic field.
The action functional on a cylinder for this model reads (|+ ||) d\^2 x + (|+ aa +h\_[b]{}(+ ||) a )dy where the boundary is located at $x=0$, $\omega = e^{i\pi/4}$ and the boundary magnetic field coupling is taken here to be $ih_{\rm b}$ with $h_{\rm b}$ - real. The boundary fermion $a=a(y)$ accounts for the double degeneracy of the vacuum.
Adopting the boundary state found in [@Chat] for the case of imaginary magnetic field we obtain \[b\_state\_Chat\] && |h\_[b]{}= e\^[-(R)]{}where $\alpha = 2h^2_{\rm b}R$ and the sign in front of the Ramond component is fixed by the sign of $h_{\rm b}$. This exact solution gives us the following component ratios \[GexactC\] \_\^[0]{} = - , \_\^[0]{}= i 2\^[1/4]{} , \[Gh\] = . The vacuum component of $|h_{\rm b}\rangle\!\rangle$ vanishes linearly at $\alpha = 1/2$ that is $T\sim 1/2-\alpha$. This results in $\Gamma^{0}_{\epsilon}$ and $\Gamma^{0}_{\sigma}$ having a simple pole at this point while their ratio is finite and given (in absolute value) by function (\[Gh\]) a portion of which which we depict below
Comparing (\[GexactC\]), (\[Gh\]) with Fig. \[data1\] and Fig. \[data2\] we find qualitatively exactly the same picture. We find that the phases of all ratios are the same as in (\[b\_state\_Chat\]) and that at $r=r_{1}$ in the NS sector all components are small except for the components including the $a_{1/2}^{\dagger}\bar a_{1/2}^{\dagger}$ oscillators as in (\[b\_state\_Chat\]). Since we do not know how the effective boundary $h_{\rm b}$ depends on $y$ and $r$ we cannot match the two curves for $\Gamma$ (they are parameterised differently). However we can still match quantitatively the two special values: $\Gamma_{*}\approx 1.70$ - the minimal value and $\Gamma^{*}=2^{1/4}\sqrt{\pi}\approx 2.107$ - the value at the point at which $\Gamma^{0}_{\epsilon,\sigma}$ blow up. The numerical TFFSA values: $\Gamma_{*}^{\rm num}$, $\Gamma^{*}_{\rm num}$ are approximately 1.573 and 2.044 respectively. These values are within few percent from those predicted by the boundary magnetic field model. The match gets better for larger values of $|y|$ and larger $n_c$.
In the numerics for $y=-3.5$ the data continues past the blow up point $r=r_1$ with $\Gamma$ continuing to grow. In the boundary magnetic field model going past the first special point $\alpha=1/2$ the ratio $\Gamma$ blows up at $\alpha=1$. In the numerics we do not quite get to that second point as the vacuum and the first excited energy levels collide and form a complex pair. The collision point is sensitive to the truncation level $n_c$ and is moved towards larger values of $r$ as $n_c$ is increased. So presumably in the untruncated theory the vacuum remains real at all scales. As explained at the beginning of section \[Complex\_vac\_sec\] as long as the vacuum energy remains real and $\Gamma_{\sigma}^{0}$ does not vanish (and thus remains imaginary) we cannot approach a local conformal boundary condition. So, what happens? Given a good match at the onset of the flow with the imaginary boundary magnetic field model we are going to rely on it in describing what happens as we continue increasing the scale. As we can clearly see from (\[b\_state\_Chat\]) the model keeps going through a sequence of special points at which $\alpha_{n}$ is a (positive) integer or half integer. At half integer points $\alpha_{n}^{*}=n+1/2$ all boundary state components with level less than $2n+1$ in the NS sector vanish and in the higher weight components only those including $a_{n+1/2}^{\dagger}\bar a_{n+1/2}^{\dagger}$ oscillators survive. For $n$ large the low energy components in the R-sector approach those of the $|\sigma\rangle\!\rangle$ Ishibashi state. At integer points $ \alpha_{*}^{n} = n$ similarly the low energy components in the R-sector are wiped out while those in the NS-sector approach those of the $|0\rangle\!\rangle + |\epsilon\rangle\!\rangle$ Ishibashi states combination. If we focus on the low lying components with level smaller than $\alpha$ then asymptoticaly (up to an overall factor) we have |h\_[b]{}\~() \[|0+ |\] i() 2\^[1/4]{}| + … where the ellipsis stands for terms that contain components of level larger than $\alpha$. Thus as $\alpha$ goes to infinity we will see a never ending rotation of the two combinations of Ishibashi states. This cyclic behaviour is of course in violation of $g$-theorem which however is possible here because we are in a non-unitary situation.
Conformal interfaces between Ising and Yang-Lee models
------------------------------------------------------
At any real value of $h$ the Ising field theory flows to a trivial fixed point. For pure imaginary $h$ with a certain value of the scaling parameter $y=y_{\rm cr}$ the theory flows to the Yang-Lee fixed point which is a non-unitary minimal model ${\cal M}(2,5)$ with central charge $c=-22/5$. It has two primary fields: the identity and a field $\phi$ with scaling dimension $\Delta_{\phi}=-2/5$ and correspondingly two conformal boundary conditions with Cardy states $|1\rangle\!\rangle_{\rm YL}\, , |\phi\rangle\!\rangle_{\rm YL} $. The value of $y_{cr}$ was most recently estimated numerically [@Zam2] to be $y_{cr}=-2.42929(2) $.
All conformal interfaces between the Ising and Yang-Lee models are known [@QRW] due to a remarkable fact that the tensor product $({\rm Ising})\otimes (\mbox{Yang-Lee})$ is itself a minimal model ${\cal M}(5,12)$ with $E_{6}$ modular invariant. The interfaces are described as boundary conditions in the tensor product theory ${\cal M}(5,12)$ which were found in [@BPPZ]. The ${\cal M}(5,12)$ theory has 12 primary states and thus 12 elementary conformal boundary conditions. We will use the same conventions as in [@QRW], [@BPPZ] in which chiral primaries $\phi_{r,s}$ are labelled by $r\in \{1,3\}$ and $s\in \{1,4,5,7,8,11\}$ - the set of $E_6$ exponents. The tensor products of primaries are identified as = \_[1,1]{} , = \_[1,5]{} , = \_[1,4]{} , = \_[3,7]{} , = \_[3, 5]{} , = \_[3,8]{} . The conformal boundary states $|\widetilde{(r,a)}\rangle\!\rangle$ are labelled by a pair $r \in \{ 1,3\}$, $a\in \{ 1,2,3,4,5,6 \}$ (the labels of nods of the $E_6$ Dynkin diagram). The decomposition into the Ishibashi states can be written as[^10] [@BPPZ] \[Cardy\_E6\] |= \_[r’, s’]{} |(r’,s’) where \[Psi\] \_[r,a]{}\^[(r’,s’)]{} = S\_[rr’]{}\^[s’]{}\_[a]{} and $\psi^{s'}_{a}$ is the matrix made of the eigenvectors of the $E_6$ adjacency matrix. Explicitly we have \[matrix\] (\^[s]{}\_[i]{}) = (
[rrrrrr]{} a& & b&b & & a\
b& & a& -a & -& -b\
c&0&-d&-d& 0 & c\
b& -& a&-a& &-b\
a&-& b & b & - & a\
d& 0 & -c&c & 0 &-d
) where a = , b= , c = , d= . In (\[matrix\]) the row index $i\in \{1,2,3,4,5,6\}$ labels the boundary states and the column index $s\in \{1,4,5,7,8,11\}$ labels the primaries. The $S$-matrices present in (\[Cardy\_E6\]), (\[Psi\]) are S\_[rs,r’ s’]{} = (-1)\^[(r+s)(r’+s’)]{} ( r r’ ) ( s s’ ) , S\_[rr’]{} = (rr’ ) .
Using the above we calculate the component ratios $\Gamma_{\epsilon}^{i}$, $\Gamma_{\sigma}^{i}$ as ratios of one-point functions of factored primaries in the $E_6$ theory. As they turn out to be independent of the first index - $r$ labelling the boundary states, we present the answers in the form of 6-vectors with components labeled by the second index - $a$. \[InterfaceG1\] \_\^[0]{} = = (
[r]{} 1\
-1\
1\
-1\
1\
-1
) , \_\^[0]{} = = (
[r]{}\
\
0\
-\
-\
0
) ,
\[InterfaceG2\] \_\^[1]{} = = (
[r]{}\
1\
-1\
1\
\
-
) , \_\^[1]{} = = (
[r]{}\
\
0\
-\
-\
0
) where =2\^[1/4]{}1.189 , = 2\^[1/4]{}()\^[1/2]{}0.615 , = 2\^[1/4]{}()\^[1/2]{} 2.297 , = 3.732 . We observe that knowing the above four component ratios determines the index $a$ uniquely. Furthermore the factorised interfaces correspond to $a=2,3,4$. Since a magnetic perturbation is present on the Ising to Yang-Lee trajectory we expect a non-vanishing $|\sigma\rangle$-component in the vacuum. This means that there are essentially two types of candidates for the Ising - Yang-Lee RG interface - the factorised defects $|\pm\rangle\!\rangle\langle\!\langle \phi, 1|_{\rm YL}$ and the non-factorisable defects corresponding to $a=1,5$. Note that the pair $a=1,5$ forms a doublet under $Z_2$ spin reversal symmetry so that there is essentially one non-factorizable candidate interface. To determine the index $r$ one would need to know other ratios of one-point functions e.g. the ratio = i ( )\^[3/4]{} , = -i ( )\^[1/4]{} which is independent of $a$. However this requires knowing how the Ising eigenvector $|\epsilon\rangle$ is realised in the Yang-Lee state space that is inaccessible by TCSA method which only gives the infrared theory eigenvectors in the UV theory space. In the next section we will argue that the RG interface for the Ising to Yang-Lee flow does not asymptote to any single conformal interface.
As a final remark about the conformal interfaces considered in this section we note that using the Cardy constraint we can also find the spectrum of interface fields. We found that all interfaces contain relevant fields (and are thus unstable) except for the factorizable ones: $|\pm\rangle\!\rangle\langle\!\langle 1|_{\rm YL}$.
RG interface for the flow to Yang-Lee fixed point {#ycr_sec}
-------------------------------------------------
The approach to a non-trivial fixed point at $y=y_{\rm cr}$ is marked by the appearance of a “nose” on the plot of dimensionless energies $\tilde E(r)$ depicted below: the vacuum and the first excited level energies get asymptoticaly close without collision. The relative energies $e_{i}-e_{0}$ level to constant values equal to the conformal dimension difference $\Delta_{i}-\Delta_{0}$. We find numerically (see the second graph below) that at truncation level $n_{c}=12$ the difference $e_{1}-e_{0}$ levels at the value $0.391$ while $e_{2}-e_{0}$ approaches $1.90$. These values approximate the dimension gaps for the Yang-Lee operators ${\bf 1}$ and $L_{-1}\bar L_{-1}\phi$ respectively (the lowest dimension state is $\phi$ with dimension -0.4).
![The lowest 3 energy eigenvalues at $y=y_{\rm cr}$, $n_{c}=12$.[]{data-label="nose"}](nose_n12v3.pdf)
![The lowest 3 energy eigenvalues at $y=y_{\rm cr}$, $n_{c}=12$.[]{data-label="nose"}](ycr_spec_v2.pdf)
For the truncation level $n_c=12$ we find numerically that the lowest two levels merge into a complex pair at $r\approx 8.5$. Looking at the behaviour at different values of $n_{c}$ we find that the merging point moves to higher $r$ as we increase $n_c$. So as in the case $y<y_{\rm cr}$ such a merger seems to be merely a finite precision artefact which however limits the domain of $r$ we can investigate numerically.
In the previous section we found explicitly the component ratios $\Gamma^{0}_{\epsilon,\sigma}$, $\Gamma^{1}_{\epsilon,\sigma}$ for all conformal interfaces between the critical Ising and Yang-Lee models (see formulai (\[InterfaceG1\]), (\[InterfaceG2\])). Crucially all possible values are real while, as discussed in section \[Complex\_vac\_sec\], for as long as an eigenvalue (vacuum or excited) stays real the corresponding ratio $\Gamma_{\epsilon}$ is real and $\Gamma_{\sigma}$ is imaginary. This leaves us with two possibilities: [*either the limiting conformal interface does not exist as in the case of massive flows discussed before, or the RG interface approaches a conformal interface with*]{} $\Gamma_{\sigma}^{0}=\Gamma_{\sigma}^{1}=0$ (i.e. an interface symmetric under spin reversal).
The numerical data for the component ratios of the vacuum and the first two excited states is shown on the graphs below.
We see that the ratios for the vacuum and the first excited level each behave qualitatively very similar to the vacuum ratios for the massive flows discussed in section \[Im\_massive\]. The ratios $\Gamma^{0}_{\epsilon,\sigma}$ blow up near $r=6.4$ while $\Gamma^{1}_{\epsilon,\sigma}$ blow up near $r=5.3$. In each case the ratios $\Gamma^{0}_{\epsilon}/\Gamma^{0}_{\sigma}$, $\Gamma^{1}_{\epsilon}/\Gamma^{1}_{\sigma}$ remain finite and increasing functions past the blow up points. The ratios $\Gamma^{2}_{\epsilon,\sigma}$ while remaining finite do not show any tendency to level at a constant value. While we cannot say with certainty that the oscillations between the blow up points will continue everything points towards the non-convergent oscillatory scenario.
In addition to this numerical indications we would like to remark that the alternative scenario in which we approach a conformal interface symmetric under spin reversal would be very hard to envisage from a general point of view. Clearly at large distances the magnetic field perturbation dominates and it breaks this symmetry.
Open problems
=============
In this paper we showed how pairings between states arising in RG interfaces of [@BR] can be calculated numerically using truncated Hamiltonian techniques. As illustrated by the exactly solvable case of the Ising model with zero magnetic case in order to read off the numerical values of the pairings (or rather their ratios) one needs to go to very large values of dimensionless couplings. At these values one cannot a priory trust any results. Perturbative corrections are organised in ratios of couplings to the truncation energy and are very large in this region. Some aberrations in the low-lying spectrum that one can easily spot are the non-linearity of the vacuum energy dependence on scale and non-constance of the mass gap[^11]: $\tilde E_{1}-\tilde E_{0}$. These can be easily spotted in the TCSA sample of data presented on the plots below.
The TFFSA vacuum energy behaves much better even at very large scales $r\sim 50$. The vacuum energy remains linear with a very good accuracy (regression variance is about 0.01). The mass gap calculated using TFFSA deviates from constant at big enough $r$ but the deviation is significantly smaller than in the TCSA data.
This means that in general we cannot trust the energy eigenvalues and eigenvectors for such large values of scale, and yet our numerical results (both TCSA and TFFSA) indicate that ratios of low lying components of the vacuum eigenvector are well behaved at large scales and the non-perturbative errors for them remain well bounded. This suggests that although we need to go to very large scales to read off the asymptotic values, these quantities are more robust against truncation errors than the energy eigenvalues and eigenvectors as a whole.
At the moment we have no conceptual understanding of this fact which is crucial for the TCSA scheme to work for calculating the IR asymptotics of component ratios. Perturbative corrections also need to be understood as well with a suitable quantitative method for their incorporation needs to be worked out.
Another important issue that needs further understanding is renormalisation of wave functionals. As discussed in the introduction in some cases additional boundary counter terms may be needed to renormalise wave functionals in the interaction picture. This may occur when vector fields are present in the OPE of the perturbing operators. In the bulk perturbation such terms do not need any counter terms due to rotational symmetry but from the point of view of the boundary they are scalars and may lead to additional divergences for collisions at the boundary (where rotational symmetry is broken). Such counter terms are local along the boundary but what is their manifestation in the Hamiltonian formalism (and TCSA) is not clear.
There is a good chance, in our opinion, that for superrenormalisable perturbations the above conceptual issues do not impede using the raw TCSA data for identifying RG interfaces and RG boundaries. It would be interesting to explore systematically other examples such as the tricritical Ising model or Potts model. For the tricritical Ising model there are 4 relevant operators and 6 elementary conformal boundary conditions. It would be interesting to find out how the three-dimensional space of massive flows breaks up into regions according to their RG boundaries. Also a conformal interface for the RG flow from the tricritical Ising to the critical Ising was put forward in [@Gaiotto]. It will be interesting to test that proposal numerically using TCSA. We hope to report some answers to these questions in a near future [@inprogress].
[**Acknowledgments** ]{}
The author is grateful to Adam Nahum, Slava Rychkov, Cornelius Schmidt-Colinet, Stefan Sint, G' abor Tak' acs, Balt van Rees, and Gerard Watts for stimulating discussions and comments on the manuscript. He also thanks the organisers of the workshop “Conformal Field Theories and Renormalisation Group Flows in Dimensions $d>2$” at Galileo Galilei Institute (GGI) in Florence and GGI staff for hospitality. This work was supported in part by STFC grant “Particle Theory at the Higgs Centre”, ST/L000334/1. All numerical results presented in the paper were obtained using Wolfram [*Mathematica* ]{} package v. 10.2 .
[99]{} I. Brunner and D. Roggenkamp, [*Defects and bulk perturbations of boundary Landau-Ginzburg orbifolds*]{}, JHEP [**0804**]{} (2008) 00; arXiv:0712.0188.
S. Fredenhagen and T. Quella, [ *Generalised permutation branes*]{}, JHEP [**0511**]{} (2005) 004; arXiv:hep-th/0509153.
D. Gaiotto, [*Domain walls for two-dimensional renormalization group flows*]{}, JHEP 1212 (2012) 103; arXiv:1201.0767.
P. Fonseca and A. Zamolodchikov, [ *Ising field theory in a magnetic field: Analytic properties of the free energy*]{}, Journal of Statistical Physics, Vol. 110, Issue 3 (2003) pp. 527-590; arXiv:hep-th/0112167.
T. Quella, I. Runkel and G. Watts, [*Reflection and Transmission for Conformal Defects*]{}, JHEP [**0704**]{} (2007) 095; arXiv:hep-th/0611296.
V. Yurov and Al. Zamolodchikov, [*Truncated Conformal Space Approach To Scaling Lee-yang Model*]{}, Int. J. Mod. Phys. [**A5**]{} (1990) 3221-3246.
V. Yurov and Al. Zamolodchikov, [ *Truncated fermionic space approach to the critical 2-D Ising model with magnetic field*]{}, Int. J. Mod. Phys. [**A6**]{} (1991) 4557-4578.
P. Giokas and G. Watts, [*The renormalisation group for the truncated conformal space approach on the cylinder*]{}, arXiv:1106.2448.
M. Hogervorst, S. Rychkov, and B. C. van Rees,[*A Cheap Alternative to the Lattice?*]{}, Phys. Rev. [**D91**]{} (2015) 025005; arXiv:1409.1581.
S. Rychkov, L. G. Vitale, [*Hamiltonian Truncation Study of the $\Phi^4$ Theory in Two Dimensions*]{}, Phys. Rev. [**D 91**]{} (2015) 085011; arXiv:1412.3460.
M. Lencs' es, G. Tak' acs, [*Excited state TBA and renormalized TCSA in the scaling Potts model*]{}, JHEP [**09**]{} (2014) 052; arXiv:1405.3157.
M. Lencs' es, G. Tak' acs, [*Confinement in the q-state Potts model: an RG-TCSA study*]{}, JHEP [**09**]{} (2015) 146; arXiv:1506.06477.
T. Rakovszky, M. Mesty' an, M. Collura, M. Kormos, and G. Tak' acs, [*Hamiltonian truncation approach to quenches in the Ising field theory*]{}, arXiv:1607.01068.
M. L" assig, G. Mussardo, and John L. Cardy, [*The scaling region of the tricritical Ising model in two-dimensions*]{}, Nucl. Phys. [**B348**]{} (1991) 591-618.
K. Symanzik, [*Schr" odinger representation and Casimir effect in renormalizable quantum field theory*]{}, Nucl. Phys. [**B190**]{} (1981) 1-44.
M. L" usher, [*Schr" odinger representation in quantum field theory*]{}, Nucl Phys. [**254**]{} (1985) 52.
D. Minic and V.P. Nair, [*Wave Functionals, Hamiltonians and Renormalization Group*]{}, Int. J. Mod. Phys. [**A11**]{} (1996) 2749; arXiv:hep-th/9406074.
J. R. Fliss, R. G. Leigh, and O. Parrikar, [*Unitary Networks from the Exact Renormalization of Wave Functionals*]{}, arXiv:1609.03493.
L. Klaczynski, [*Haag’s theorem in renormalised quantum field theories*]{}, arXiv:1602.00662.
G. Tak' acs (2012), unpublished.
P. Calabrese, F. H. L. Essler, and M. Fagotti, [*Quantum Quench in the Transverse Field Ising chain I: Time evolution of order parameter correlators*]{}, J. Stat. Mech. (2012) P07016; arXiv:1204.3911.
C.N. Yang and T. D. Lee, [*Statistical theory of equations of state and phase transitions. I. Theory of condensation.*]{}, Phys. Rev. [**87**]{} (1952) 404.
C.N. Yang and T. D. Lee, [*Statistical theory of equations of state and phase transitions. II. Lattice gas and Ising model.* ]{}, Phys. Rev. [**87**]{} (1952) 410.
J. L. Cardy, [ *Conformal Invariance and the Yang-lee Edge Singularity in Two-dimensions*]{}, Phys. Rev. Lett. [**54**]{} (1985) 1354.
F. A. Berezin, [*The method of second quantization*]{}, Academic Press, 1966.
J. Cardy, [ *Boundary Conditions, Fusion Rules and the Verlinde Formula* ]{}, Nucl. Phys. [**B324**]{} (1989) 581-596.
R. E. Behrend, P.A. Pearce, V. B. Petkova, and J.-B. Zuber, [*Boundary Conditions in Rational Conformal Field Theories*]{}, Nucl. Phys. [**B579**]{} (2000) 707-773; arXiv:hep-th/9908036.
J. Fuchs, I. Runkel, and C. Schweigert, [*TFT construction of RCFT correlators I: Partition functions*]{}, Nucl. Phys. [**B646**]{} (2002) 353-497; arXiv:hep-th/0204148.
C. Bachas, J. de Boer, R. Dijkgraaf, and H. Ooguri, [*Permeable conformal walls and holography*]{}, JHEP 06 (2002) 027; arXiv:0111210.
C. Bachas and I. Brunner, [*Fusion of conformal interfaces*]{}, JHEP 0802:085,2008; arXiv:0712.0076.
A. Zamolodchikov, [*Ising Spectroscopy I: Mesons at $T\! <\! T_c$*]{}, arXiv:1310.4821.
A. Zamolodchikov, [*Ising Spectroscopy II: Particles and poles at $T\!>\!T_{c}$*]{}, arXiv:1310.4821.
R. Chatterjee and A. Zamolodchikov, [*Local Magnetization in Critical Ising Model with Boundary Magnetic Field*]{}, Mod. Phys. Lett. [**A9**]{} (1994) 2227; arXiv:hep-th/9311165.
Chatterjee, [Exact Partition Function and Boundary State of Critical Ising Model with Boundary Magnetic Field]{}, Mod. Phys. Lett. [**A10**]{} (1995) 973; arXiv:hep-th/9412169.
A. Konechny, [*RG boundaries and interfaces for tricritical Ising field theory*]{}, work in progress.
[^1]: At least for superrenormalizable theories.
[^2]: Unlike [@Gaiotto] we restrict our pairing to states only and do not consider issues of transport of local operators from UV into IR that we believe to be rather subtle.
[^3]: They result in additional functions present in the Callan-Symanzik equation for $Z_{0,i}$ (and $\Psi_{\rm vac}^{\rm pert}[\varphi_{0}]$)
[^4]: The states with an even number of chiral oscillators correspond to the Virasoro tower of the identity while the ones with an odd number to that of the energy primary $\epsilon$.
[^5]: This means that $n_c$ truncates the total Virasoro weight. Note that in some TCSA schemes the truncation is done in the level of descendants only rather than in the total weight.
[^6]: The basis elements come in batches. Vectors in each batch have the same conformal weight. On the graph the red numbers mark the first component with the indicated conformal weight.
[^7]: For example the fusion could trigger an RG flow on the interface. If that was the case we would expect some additional sensitivity to truncation level in the TFFSA numerics which we have not observed.
[^8]: The reality of $\Gamma_{\sigma}$ for conformal boundary states can be seen as a consequence of locality of the boundary condition.
[^9]: Strictly speaking this leaves out the possibility of approaching the free boundary condition. This would be however highly unlikely in the view of the perturbation explicitly breaking the spin reversal symmetry.
[^10]: Formula (\[Cardy\_E6\]) was derived in [@BPPZ] with the assumption that the matrix $\Psi$ is unitary. A more general derivation in which this assumption was not made was later presented in [@TFT].
[^11]: The author thanks V. Rychkov for suggesting to discuss these quantities.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'This paper builds on work of Hochster and Yao that provides nice embeddings for finitely generated modules of finite G-dimension, finite projective dimension, or locally finite injective dimension. We extend these results by providing similar embeddings in the relative setting, that is, for certain modules of finite ${\text{G}_C}$-dimension, finite ${{\mathcal{P}}_C}$-projective dimension, locally finite ${{\mathcal{GI}}_C}$-injective dimension, or locally finite ${{\mathcal{I}}_C}$-injective dimension where $C$ is a semidualizing module. Along the way, we extend some results for modules of finite homological dimension to modules of locally finite homological dimension in the relative setting.'
address: 'Mathematics Department, NDSU Dept \# 2750, PO Box 6050, Fargo, ND 58108-6050 USA'
author:
- 'Sean Sather-Wagstaff'
title: Embedding modules of finite homological dimension
---
[^1]
Introduction {#sec0}
============
Throughout this note, $R$ is a commutative noetherian ring with identity. The purpose of this note is to extend some results of Hochster and Yao, beginning with the following, wherein $\underline z_{[i]}=z_1,\ldots,z_i$ and ${\mathrm{G}\text{-}\!\dim}_R(M)$ is the G-dimension of Auslander and Bridger [@auslander:adgeteac; @auslander:smt].[^2]
\[thm110527a\] Let $M$ be a non-zero finitely generated $R$-module. Then
1. \[thm110527a1\] If ${\mathrm{G}\text{-}\!\dim}_R(M) = r < \infty$, then there are an $R$-regular sequence $\underline z = z_1, \ldots, z_r$, integers $n_0, n_1, \ldots , n_r {\geqslant}0$ with $n_r{\geqslant}1$, and an exact sequence $$0 \to M \to Z \to N \to 0$$ with $Z = \oplus_{i=0}^r(R/(\underline z_{[i]}))^{n_i}$ and ${\mathrm{G}\text{-}\!\dim}_R(N) {\leqslant}r$.
2. \[thm110527a2\] If ${\operatorname{pd}}_R(M) = r < \infty$, then there exist an $R$-regular sequence $\underline z = z_1, \ldots, z_r$, integers $n_0, n_1, \ldots , n_r {\geqslant}0$ with $n_r{\geqslant}1$, and an exact sequence $$0 \to M \to Z \to N \to 0$$ with $Z = \oplus_{i=0}^r(R/(\underline z_{[i]}))^{n_i}$ and ${\operatorname{pd}}_R(N) {\leqslant}r$.
The point of Theorem \[thm110527a\], as Hochster and Yao indicate, is that the given module of finite G-dimension or projective dimension embeds in a module that *obviously* has finite projective dimension, with some extra control on the G-dimension or projective dimension of the cokernel.
The first of our results, stated next, is for modules of finite ${\text{G}_C}$-dimension and modules of finite ${{\mathcal{P}}_C}$-projective dimension, where $C$ is a semidualizing $R$-module, after Golod [@golod:gdagpi], Holm and Jørgensen [@holm:smarghd] and White [@white:gpdrsm].[^3] (See Section \[sec110530a\] for definitions and background material.) On the one hand, this result is a corollary to Theorem \[thm110527a\]. On the other hand, Theorem \[thm110527a\] is the special case $C=R$, so our result generalizes this one. This result is proved in Proofs \[proof110530a\] and \[proof110530b\], which are similar to the proof of [@hochster:etmf Theorem 4.2].
\[thm110527b\] Let $M$ be a non-zero finitely generated $R$-module, and let $C$ be a semidualizing $R$-module. Then
1. \[thm110527b1\] If ${{\mathrm{G}_{C}\text{-}\!\dim}}_R(M) = r < \infty$ and $M$ is in the Bass class ${{\mathcal{B}}_C}(R)$, then there exist an $R$-regular sequence $\underline z = z_1, \ldots, z_r$, integers $n_0, n_1, \ldots , n_r {\geqslant}0$ with $n_r{\geqslant}1$, and an exact sequence $$0 \to M \to Z \to N \to 0$$ with $Z = \oplus_{i=0}^r(C/(\underline z_{[i]})C)^{n_i}$ and ${{\mathrm{G}_{C}\text{-}\!\dim}}_R(N) {\leqslant}r$ and $N\in{{\mathcal{B}}_C}(R)$.
2. \[thm110527b2\] If ${{{\mathcal{P}}_C}\text{-}{\operatorname{pd}}}_R(M) = r < \infty$, then there exist an $R$-regular sequence $\underline z = z_1, \ldots, z_r$, integers $n_0, n_1, \ldots , n_r {\geqslant}0$ with $n_r{\geqslant}1$, and an exact sequence $$0 \to M \to Z \to N \to 0$$ with $Z = \oplus_{i=0}^r(C/(\underline z_{[i]})C)^{n_i}$ and ${{{\mathcal{P}}_C}\text{-}{\operatorname{pd}}}_R(N) {\leqslant}r$.
The next result we extend is the following:
\[thm110530a\] Let $R$ be a Cohen-Macaulay ring with a pointwise dualizing module $\omega$. Then, for any non-zero finitely generated $R$-module $M$ with locally finite injective dimension, there exist an integer $r = {\operatorname{pd}}_R({\operatorname{Hom}_{R}(\omega,M)})$, an $R$-regular sequence $\underline z = z_1, \ldots , z_r$, non-negative integers $n_0, n_1, \ldots , n_r$ with $n_r{\geqslant}1$, and an exact sequence $$0 \to M \to Z \to N \to 0$$ with $Z = \oplus_{i=0}^r(\omega/(\underline z_{[i]})\omega)^{n_i}$, such that $N$ has locally finite injective dimension and ${\operatorname{pd}}_R({\operatorname{Hom}_{R}(\omega,N)}){\leqslant}r$.
We extend this in two directions in Theorems \[thm110608a\] and \[thm110530b\]. First, we have the version for G-injective dimension. Second, we give versions relative to a semidualizing module. These are proved in Proofs \[proof110609a\], \[proof110609b\], and \[proof110609c\].
\[thm110608a\] Let $R$ be a Cohen-Macaulay ring with a pointwise dualizing module $\omega$. For any non-zero finitely generated $R$-module $M$ with locally finite G-injective dimension, there exist an integer $r = {\mathrm{G}\text{-}\!\dim}_R({\operatorname{Hom}_{R}(\omega,M)})$, an $R$-regular sequence $\underline z = z_1, \ldots , z_r$, non-negative integers $n_0, n_1, \ldots , n_r$ with $n_r{\geqslant}1$, and an exact sequence $$0 \to M \to Z \to N \to 0$$ with $Z = \oplus_{i=0}^r(\omega/(\underline z_{[i]})\omega)^{n_i}$ such that $N$ has locally finite G-injective dimension and ${\mathrm{G}\text{-}\!\dim}_R({\operatorname{Hom}_{R}(\omega,N)}){\leqslant}r$.
\[thm110530b\] Let $R$ be a Cohen-Macaulay ring with a pointwise dualizing module $\omega$. Let $M$ a non-zero finitely generated $R$-module, and let $C$ be a semidualizing $R$-module. Set ${C^{\dagger}}={\operatorname{Hom}_{R}(C,\omega)}$. Then
1. \[thm110530b1\] If $M$ has locally finite ${{\mathcal{GI}}_C}\text{-}{\operatorname{id}}$ and $M$ is in the Auslander class ${{\mathcal{A}}_C}(R)$, then there exist an integer $r = {\mathrm{G}\text{-}\!\dim}({\operatorname{Hom}_{R}(\omega,{C\otimes_{R}M})})$, an $R$-regular sequence $\underline z = z_1, \ldots , z_r$, non-negative integers $n_0, n_1, \ldots , n_r$ with $n_r{\geqslant}1$, and an exact sequence $$0 \to M \to Z \to N \to 0$$ such that $N$ has locally finite ${{\mathcal{GI}}_C}\text{-}{\operatorname{id}}$, and $Z = \oplus_{i=0}^r({C^{\dagger}}/(\underline z_{[i]}){C^{\dagger}})^{n_i}$, $N \in {{\mathcal{A}}_C}(R)$ and ${\mathrm{G}\text{-}\!\dim}({\operatorname{Hom}_{R}(\omega,{C\otimes_{R}N})}){\leqslant}r$.
2. \[thm110530b2\] If $M$ has locally finite ${{\mathcal{I}}_C}\text{-}{\operatorname{id}}$, then there exist an integer $$r = {\operatorname{pd}}_R({\operatorname{Hom}_{R}(\omega,{C\otimes_{R}M})})$$ an $R$-regular sequence $\underline z = z_1, \ldots , z_r$, non-negative integers $n_0, n_1, \ldots , n_r$ with $n_r{\geqslant}1$, and an exact sequence $$0 \to M \to Z \to N \to 0$$ such that $N$ has locally finite ${{\mathcal{I}}_C}\text{-}{\operatorname{id}}$, and $Z = \oplus_{i=0}^r({C^{\dagger}}/(\underline z_{[i]}){C^{\dagger}})^{n_i}$ and ${\operatorname{pd}}_R({\operatorname{Hom}_{R}(\omega,{C\otimes_{R}N})}){\leqslant}r$.
We summarize the organization of this paper. Section \[sec110530a\] contains the proof of Theorem \[thm110527b\], along with the necessary background material. Section \[sec110608a\] similarly treats Theorems \[thm110608a\] and \[thm110530b\]; this section also contains several lemmas about modules of locally finite G-injective dimension, locally finite ${\text{G}_C}$-injective dimension, and locally finite ${{\mathcal{I}}_C}$-injective dimension where $C$ is a semidualizing module, as these notions have not been developed in the literature as best we know.[^4] Note that we do not develop the background material in the most optimal manner, focusing instead on accessibility. To be specific, we present background material as it is needed in Sections \[sec110530a\] and \[sec110608a\] instead of putting it all in its own section. Also, we feel that the definition of locally finite ${{\mathcal{GI}}_C}\text{-}{\operatorname{id}}$ is a bit much to swallow on the first bite, so we first discuss the special case of ${{{\mathcal{G}}\text{-}{\operatorname{id}}}}$ and work our way up.
Proof of Theorem \[thm110527b\] {#sec110530a}
===============================
\[notn1xx\] We let ${\operatorname{pd}}_R(-)$ and ${\operatorname{id}}_R(-)$ denote the classical projective dimension and the classical injective dimension, respectively.
The study of semidualizing modules was initiated independently (with different names) by Foxby [@foxby:gmarm], Golod [@golod:gdagpi], and Vasconcelos [@vasconcelos:dtmc]. For example, a finitely generated projective $R$-module of rank 1 is semidualizing.
\[notation08a\] A finitely generated $R$-module $C$ is *semidualizing* if the natural map $R\to{\operatorname{Hom}_{R}(C,C)}$ is an isomorphism, and ${\operatorname{Ext}}_R^{{\geqslant}1}(C,C)=0$. An $R$-module $C$ is *pointwise dualizing* if it is semidualizing and ${\operatorname{id}}_{R_{{{\mathfrak{m}}}}}(C_{{{\mathfrak{m}}}})<\infty$ for all maximal ideals ${{\mathfrak{m}}}\subset R$.[^5] An $R$-module $C$ is *dualizing* if it is semidualizing and ${\operatorname{id}}_R(C)<\infty$.
\[fact110527a\] (a) It is straightforward to show that the semidualizing property is local. That is, if $C$ is a semidualizing $R$-module, then for each multiplicatively closed subset $U\subseteq R$ the localization $U^{-1}C$ is a semidualizing $U^{-1}R$-module; and conversely, if $M$ is a finitely generated $R$-module such that $M_{{{\mathfrak{m}}}}$ is a semidualizing $R_{{{\mathfrak{m}}}}$-module for each maximal ideal ${{\mathfrak{m}}}\subset R$, then $M$ is a semidualizing $R$-module. Similarly, if $C$ is a (pointwise) dualizing module for $R$, then for each multiplicatively closed subset $U\subseteq R$ the localization $U^{-1}C$ is a (pointwise) dualizing module for $U^{-1}R$; and conversely, if $M$ is a finitely generated $R$-module such that $M_{{{\mathfrak{m}}}}$ is a dualizing module for $R_{{{\mathfrak{m}}}}$ for each maximal ideal ${{\mathfrak{m}}}\subset R$, then $M$ is a pointwise dualizing module for $R$. Moreover, an $R$-module $M$ is a dualizing module for $R$ if and only if it is a pointwise dualizing module for $R$ and the Krull dimension of $R$ is finite.
\(b) Let $C$ be a semidualizing $R$-module. The isomorphism $R\cong{\operatorname{Hom}_{R}(C,C)}$ implies that ${\operatorname{Supp}}_R(C)={\operatorname{Spec}}(R)$ and ${\operatorname{Ass}}_R(C)={\operatorname{Ass}}(R)$. Thus, an element $x\in R$ is $C$-regular if and only if it is $R$-regular. When $x$ is $R$-regular, the quotient $C/xC$ is a semidualizing $R/xR$-module. Thus, by induction, a sequence $\underline z=z_1,\ldots,z_r\in R$ is $R$-regular if and only if it is $C$-regular; see [@frankild:rrhffd Theorem 4.5]. Also, from [@holm:fear Proposition 3.1] we know that, for all $R$-modules $M\neq 0$, we have ${C\otimes_{R}M}\neq 0\neq{\operatorname{Hom}_{R}(C,M)}$.
The next categories were introduced by Foxby [@foxby:gdcmr] when $C$ is dualizing, and by Vasconcelos [@vasconcelos:dtmc §4.4] for arbitrary $C$, with different notation.
\[notation08d\] Let $C$ be a semidualizing $R$-module. The *Auslander class* of $C$ is the class ${{\mathcal{A}}_C}(R)$ of $R$-modules $M$ such that ${\operatorname{Tor}}^R_{{\geqslant}1}(C,M)=0={\operatorname{Ext}}_R^{{\geqslant}1}(C,C\otimes_R M)$, and the natural map $M\to{\operatorname{Hom}_{R}(C,C\otimes_R M)}$ is an isomorphism. The *Bass class* of $C$ is the class ${{\mathcal{B}}_C}(R)$ of $R$-modules $N$ such that ${\operatorname{Ext}}_R^{{\geqslant}1}(C,M)=0={\operatorname{Tor}}^R_{{\geqslant}1}(C,{\operatorname{Hom}_{R}(C,M)})$, and the natural evaluation map $C\otimes_R{\operatorname{Hom}_{R}(C,N)}\to N$ is an isomorphism.
\[disc110530a\] It is straightforward to show that the Auslander and Bass classes for $C=R$ are trivial: ${{\mathcal{A}}}_{R}(R)$ and ${{\mathcal{B}}}_R(R)$ both contain all $R$-modules.
The following notion was introduced and studied by Holm and Jørgensen [@holm:smarghd] and White [@white:gpdrsm]. Note that in the special case $C=R$, we recover the class of projective $R$-modules and the classical projective dimension.
\[notation08bxy\] Let $C$ be a semidualizing $R$-module. Let ${{\mathcal{P}}_C}(R)$ denote the class of modules $M\cong P\otimes_R C$ with $P$ projective. Modules in ${{\mathcal{P}}_C}(R)$ are called *$C$-projective*. We let ${{\mathcal{P}}_C}\text{-}{\operatorname{pd}}_R(-)$ denote the homological dimension obtained from resolutions in ${{\mathcal{P}}_C}(R)$, with the convention that ${{{\mathcal{P}}_C}\text{-}{\operatorname{pd}}}_R(0)=-\infty$.[^6]
\[disc01xyz\] Let $C$ be a semidualizing $R$-module. The Bass class ${{\mathcal{B}}_C}(R)$ contains all $R$-modules of finite ${{{\mathcal{P}}_C}\text{-}{\operatorname{pd}}}$; see [@holm:fear Lemmas 4.1 and 5.1, and Corollary 6.3]. Given an $R$-module $M$, one has ${{{\mathcal{P}}_C}\text{-}{\operatorname{pd}}}_R(M)={\operatorname{pd}}_R({\operatorname{Hom}_{R}(C,M)})$ and ${\operatorname{pd}}_R(M)={{{\mathcal{P}}_C}\text{-}{\operatorname{pd}}}_R({C\otimes_{R}M})$ by [@takahashi:hasm Theorem 2.11]. In particular, one has ${{{\mathcal{P}}_C}\text{-}{\operatorname{pd}}}_R(M)<\infty$ if and only if ${\operatorname{pd}}_R({\operatorname{Hom}_{R}(C,M)})<\infty$, and one has ${\operatorname{pd}}_R(M)<\infty$ if and only if ${{{\mathcal{P}}_C}\text{-}{\operatorname{pd}}}_R({C\otimes_{R}M})<\infty$.
\[proof110530a\] Assume that $M$ is a non-zero finitely generated $R$-module such that ${{{\mathcal{P}}_C}\text{-}{\operatorname{pd}}}_R(M) = r < \infty$. Then Fact \[disc01xyz\] implies that ${\operatorname{pd}}_R({\operatorname{Hom}_{R}(C,M)})=r<\infty$. Since $M\neq 0$, we have ${\operatorname{Hom}_{R}(C,M)}\neq 0$, by Fact \[fact110527a\](b), so Theorem \[thm110527a\] provides an $R$-regular sequence $\underline z = z_1, \ldots, z_r$, integers $n_0, n_1, \ldots , n_r {\geqslant}0$ with $n_r{\geqslant}1$, and an exact sequence $$\label{eq110530a}
0 \to {\operatorname{Hom}_{R}(C,M)} \to Z' \to N' \to 0$$ with $Z' = \oplus_{i=0}^r(R/(\underline z_{[i]}))^{n_i}$ and ${\operatorname{pd}}_R(N') {\leqslant}r$. In particular, we have $N'\in{{\mathcal{A}}_C}(R)$ which implies that ${\operatorname{Tor}^{R}_{1}(C,N')}=0$. Thus, an application of ${C\otimes_{R}-}$ to the sequence yields the next exact sequence: $$\label{eq110530b}
0 \to {C\otimes_{R}{\operatorname{Hom}_{R}(C,M)}} \to {C\otimes_{R}Z'} \to {C\otimes_{R}N'} \to 0.$$ Fact \[disc01xyz\] implies that $M\in{{\mathcal{B}}_C}(R)$, so we have ${C\otimes_{R}{\operatorname{Hom}_{R}(C,M)}}\cong M$. The equality $Z' = \oplus_{i=0}^r(R/(\underline z_{[i]}))^{n_i}$ implies that $Z:={C\otimes_{R}Z'} = \oplus_{i=0}^r(C/(\underline z_{[i]})C)^{n_i}$. Because of Fact \[disc01xyz\], the condition ${\operatorname{pd}}_R(N') {\leqslant}r$ implies that ${{{\mathcal{P}}_C}\text{-}{\operatorname{pd}}}_R({C\otimes_{R}N'}){\leqslant}r$. Thus, with $N:={C\otimes_{R}N'}$, the sequence satisfies the conclusion of Theorem \[thm110527b\].
The remainder of the proof of Theorem \[thm110527b\] is similar to the above proof, but it requires a bit more technology.
\[disc01xy\] Let $C$ be a semidualizing $R$-module.
\(a) The Auslander class ${{\mathcal{A}}_C}(R)$ contains all $R$-modules of finite projective dimension, and the Bass class ${{\mathcal{B}}_C}(R)$ contains all $R$-modules of finite injective dimension; see [@holm:fear Lemmas 4.1 and 5.1, and Corollary 6.3]. Moreover, it is straightforward to show that the Bass class satisfies the following local-global principal: For an $R$-module $M$, the following conditions are equivalent:
(i) $M\in{{\mathcal{B}}_C}(R)$;
(ii) $U^{-1}M\in{{\mathcal{B}}}_{U^{-1}C}(U^{-1}R)$ for each multiplicatively closed subset $U\subseteq R$;
(iii) $M_{{{\mathfrak{p}}}}\in{{\mathcal{B}}}_{C_{{{\mathfrak{p}}}}}(R_{{{\mathfrak{p}}}})$ for each ${{\mathfrak{p}}}\in{\operatorname{Spec}}(R)$; and
(iv) $M_{{{\mathfrak{m}}}}\in{{\mathcal{B}}}_{C_{{{\mathfrak{m}}}}}(R_{{{\mathfrak{m}}}})$ for each maximal ideal ${{\mathfrak{m}}}\in{\operatorname{Supp}}_R(M)$.
It follows that ${{\mathcal{B}}_C}(R)$ contains every $R$-module of locally finite injective dimension. The Auslander class satisfies an analogous local-global principal.
\(b) The Auslander and Bass classes also satisfy the two-of-three property by [@holm:fear Corollary 6.3]. That is, given a short exact sequence $0\to M'\to M\to M''\to 0$ of $R$-module homomorphisms, if two of the modules in the sequence are in ${{\mathcal{A}}_C}(R)$, then so is the third module, and similarly for ${{\mathcal{B}}_C}(R)$.
\(c) From [@takahashi:hasm (2.8)] we know that an $R$-module $M$ is in ${{\mathcal{B}}_C}(R)$ if and only if ${\operatorname{Hom}_{R}(C,M)}\in{{\mathcal{A}}_C}(R)$, and that $M\in{{\mathcal{A}}_C}(R)$ if and only if $C\otimes_RM\in{{\mathcal{B}}_C}(R)$. This is known as *Foxby equivalence*.
\[notation08bx\] Let $C$ be a semidualizing $R$-module. A finitely generated $R$-module $G$ is *totally $C$-reflexive* if the natural map $G\to{\operatorname{Hom}_{R}({\operatorname{Hom}_{R}(G,C)},C)}$ is an isomorphism, and ${\operatorname{Ext}_{R}^{{\geqslant}1}(G,C)}=0={\operatorname{Ext}_{R}^{{\geqslant}1}({\operatorname{Hom}_{R}(G,C)},C)}$. Let ${{\mathcal{G}}_C}(R)$ denote the class of totally $C$-reflexive $R$-modules, and set ${{\mathcal{G}}}(C)={{\mathcal{G}}_C}(R)\cap{{\mathcal{B}}_C}(R)$. In the case $C=R$ we use the more common terminology “totally reflexive” and the notation ${{\mathcal{G}}}(R)={{\mathcal{G}}}_R(R)={{\mathcal{G}}}(R)\cap{{\mathcal{B}}}_R(R)$. We abbreviate as follows: $$\begin{aligned}
{{\mathrm{G}_{C}\text{-}\!\dim}}_R(-)
&=\text{the homological dimension obtained from resolutions in ${{\mathcal{G}}_C}(R)$}
\\
{{\mathcal{G}}}(C)\text{-}{\operatorname{pd}}_R(-)
&=\text{the homological dimension obtained from resolutions in ${{\mathcal{G}}}(C)$}
\\
{\mathrm{G}\text{-}\!\dim}_R(-)
&=\text{the homological dimension obtained from resolutions in ${{\mathcal{G}}}(R)$.}\end{aligned}$$
The following facts are included for perspective.
Let $C$ be a semidualizing $R$-module, and let $M$ be a finitely generated $R$-module. Because of the containments ${{\mathcal{P}}_C}(R)\subseteq{{\mathcal{G}}}(C)\subseteq{{\mathcal{G}}_C}(R)$, one has ${{\mathrm{G}_{C}\text{-}\!\dim}}_R(M){\leqslant}{{\mathcal{G}}}(C)\text{-}{\operatorname{pd}}_R(M){\leqslant}{{{\mathcal{P}}_C}\text{-}{\operatorname{pd}}}_R(M)$ with equality to the left of any finite quantity. In particular, the case $C=R$ says that ${\mathrm{G}\text{-}\!\dim}_R(M)={{\mathcal{G}}}(R)\text{-}{\operatorname{pd}}_R(M){\leqslant}{\operatorname{pd}}_R(M)$ since ${{\mathcal{G}}}(R)={{\mathcal{G}}}_R(R)$, with equality holding when ${\operatorname{pd}}_R(M)<\infty$.
The next lemma explains how these homological dimensions are connected.
\[lem0701x\] Let $C$ be a semidualizing $R$-module. For a finitely generated $R$-module $M$, the following conditions are equivalent:
1. \[lem0701xi\] ${{\mathcal{G}}}(C)\text{-}{\operatorname{pd}}_R(M)<\infty$;
2. \[lem0701xii\] ${{\mathrm{G}_{C}\text{-}\!\dim}}_R(M)<\infty$ and $M\in{{\mathcal{B}}_C}(R)$; and
3. \[lem0701xiii\] ${\mathrm{G}\text{-}\!\dim}_R({{\operatorname{Hom}_{R}(},C)}M)<\infty$ and $M\in{{\mathcal{B}}_C}(R)$.
When these conditions are satisfied, we have $$\begin{aligned}
{{\mathcal{G}}}(C)\text{-}{\operatorname{pd}}_R(M)
&={{\mathrm{G}_{C}\text{-}\!\dim}}_R(M)
={\mathrm{G}\text{-}\!\dim}_R({{\operatorname{Hom}_{R}(},C)}M).\end{aligned}$$
Now we are in position to complete the proof of Theorem \[thm110527b\].
\[proof110530b\] Assume that $M$ is a non-zero finitely generated $R$-module such that ${{\mathrm{G}_{C}\text{-}\!\dim}}_R(M) = r < \infty$ and $M\in{{\mathcal{B}}_C}(R)$. Then Lemma \[lem0701x\] implies that ${\mathrm{G}\text{-}\!\dim}_R({\operatorname{Hom}_{R}(C,M)})=r<\infty$. Since $M\neq 0$, we have ${\operatorname{Hom}_{R}(C,M)}\neq 0$, by Fact \[fact110527a\](b), so Theorem \[thm110527a\] provides an $R$-regular sequence $\underline z = z_1, \ldots, z_r$, integers $n_0, n_1, \ldots , n_r {\geqslant}0$ such that $n{\geqslant}1$, and an exact sequence $$\label{eq110530c}
0 \to {\operatorname{Hom}_{R}(C,M)} \to Z' \to N' \to 0$$ with $Z' = \oplus_{i=0}^r(R/(\underline z_{[i]}))^{n_i}$ and ${\mathrm{G}\text{-}\!\dim}_R(N') {\leqslant}r$.
The condition $M\in{{\mathcal{B}}_C}(R)$ implies that ${\operatorname{Hom}_{R}(C,M)}\in{{\mathcal{A}}_C}(R)$ by Fact \[disc01xy\](c), and Fact \[disc01xy\](a) implies that $Z'\in{{\mathcal{A}}_C}(R)$. Also from Fact \[disc01xy\](a)–(b), we conclude that $N'\in{{\mathcal{A}}_C}(R)$, which implies that ${\operatorname{Tor}^{R}_{1}(C,N')}=0$. Thus, an application of ${C\otimes_{R}-}$ to the sequence yields the next exact sequence: $$\label{eq110530d}
0 \to {C\otimes_{R}{\operatorname{Hom}_{R}(C,M)}} \to {C\otimes_{R}Z'} \to {C\otimes_{R}N'} \to 0.$$ The assumption $M\in{{\mathcal{B}}_C}(R)$ gives an isomorphism ${C\otimes_{R}{\operatorname{Hom}_{R}(C,M)}}\cong M$. The equality $Z' = \oplus_{i=0}^r(R/(\underline z_{[i]}))^{n_i}$ implies that $Z:={C\otimes_{R}Z'} = \oplus_{i=0}^r(C/(\underline z_{[i]})C)^{n_i}$.
The condition $N'\in{{\mathcal{A}}_C}(R)$ implies that $N:={C\otimes_{R}N'}\in{{\mathcal{B}}_C}(R)$ by Fact \[disc01xy\](c), and $N'\cong{\operatorname{Hom}_{R}(C,N)}$ by the definition of what it means for $N'$ to be in ${{\mathcal{A}}_C}(R)$. Thus, because of Lemma \[lem0701x\], we have $${{\mathrm{G}_{C}\text{-}\!\dim}}_R(N)={\mathrm{G}\text{-}\!\dim}({\operatorname{Hom}_{R}(C,N)})={\mathrm{G}\text{-}\!\dim}_R(N') {\leqslant}r.$$ Thus, the sequence satisfies the conclusion of Theorem \[thm110527b\].
Proofs of Theorems \[thm110608a\] and \[thm110530b\] {#sec110608a}
====================================================
We continue with a definition due to Enochs and Jenda [@enochs:gipm].
\[notation110609a\] A *complete injective resolution* is an exact complex $Y$ of injective $R$-modules such that ${\operatorname{Hom}_{R}(J,Y)}$ is exact for each injective $R$-module $J$. An $R$-module $N$ is *G-injective* if there exists a complete injective resolution $Y$ such that $N\cong{\operatorname{Ker}}(\partial^Y_0)$, and $Y$ is a *complete injective resolution of $N$*. Let ${{\mathcal{GI}}}(R)$ denote the class of G-injective $R$-modules, and let ${\operatorname{Gid}}_R(-)$ denote the homological dimension obtained from coresolutions in ${{\mathcal{GI}}}(R)$. We say that an $R$-module $M$ has *locally finite G-injective dimension* provided that ${\operatorname{Gid}}_{R_{{{\mathfrak{m}}}}}(M_{{{\mathfrak{m}}}})<\infty$ for each maximal ideal ${{\mathfrak{m}}}\subset R$.
\[disc110726a\] Using existing technology, the modules of finite G-injective dimension behave best when $R$ is Cohen-Macaulay with a dualizing module.[^7] For instance, in general it is not known whether the G-injective dimension localizes; but it is known to localize when $R$ has a dualizing module. This is due to the following connection with Bass classes, the local case of which is from [@enochs:fdgipm].[^8]
\[lem110610a\] Assume that $R$ is Cohen-Macaulay with a pointwise dualizing module. Let $M$ be an $R$-module. Then an $R$-module $M$ has locally finite G-injective dimension if and only if $M\in{{\mathcal{B}}}_{\omega}(R)$.
If $R$ is local, then the result follows from [@enochs:fdgipm Proposition 1.4 and Theorems 1.6 and 2.5]. By definition, the condition “$M$ has locally finite G-injective dimension” is a local condition. Fact \[disc01xy\](b) shows that the condition “$M\in{{\mathcal{B}}}_{\omega}(R)$” is also a local condition. Note that Fact \[fact110527a\](a) implies that for each maximal ideal ${{\mathfrak{m}}}\subset R$, the localization $\omega_{{{\mathfrak{m}}}}$ is a dualizing module for the Cohen-Macaulay local ring $R_{{{\mathfrak{m}}}}$. Thus, the general result follows from the local case.
The next result shows that the G-dimension of a finitely generated module behaves like its projective dimension.
\[lem110610b\] Let $M$ be a finitely generated $R$-module. The following conditions are equivalent:
1. \[lem110610b1\] ${\mathrm{G}\text{-}\!\dim}_{R}(M)<\infty$;
2. \[lem110610b2\] ${\mathrm{G}\text{-}\!\dim}_{R_{{{\mathfrak{m}}}}}(M_{{{\mathfrak{m}}}})<\infty$ for each maximal ideal ${{\mathfrak{m}}}\subset R$.
When $R$ is Cohen-Macaulay with a pointwise dualizing module $\omega$, these conditions are equivalent to the following:
1. \[lem110610b3\] $M\in{{\mathcal{A}}}_{\omega}(R)$.
The implication $\eqref{lem110610b1}\implies\eqref{lem110610b2}$ follows from the inequality ${\mathrm{G}\text{-}\!\dim}_{R_{{{\mathfrak{m}}}}}(M_{{{\mathfrak{m}}}}){\leqslant}{\mathrm{G}\text{-}\!\dim}_R(M)$, which is straightforward to verify. The implication $\eqref{lem110610b2}\implies\eqref{lem110610b1}$ is from [@avramov:rrc1 Theorem 3.3]. When $R$ is Cohen-Macaulay with a pointwise dualizing module $\omega$, the equivalence $\eqref{lem110610b2}\iff$(iii) follows from the local results [@enochs:fdgipm Proposition 1.3 and Theorems 1.6 and 2.1] as in the proof of Lemma \[lem110610a\].[^9]
\[proof110609a\] Assume that $R$ is Cohen-Macaulay with a pointwise dualizing module $\omega$ and that $M$ is a non-zero finitely generated $R$-module with locally finite G-injective dimension. Lemma \[lem110610a\] implies that $M \in \mathcal B_{\omega}(R)$. Using Foxby equivalence (from Fact \[disc01xy\](c)) we conclude that ${\operatorname{Hom}_{R}(\omega,M)}\in{{\mathcal{A}}}_{\omega}(R)$, so Lemma \[lem110610b\] implies that $r:={\mathrm{G}\text{-}\!\dim}_R({\operatorname{Hom}_{R}(\omega,M)})<\infty$. The fact that $R$ is noetherian implies that ${\operatorname{Hom}_{R}(\omega,M)}$ is finitely generated. (The proof concludes as in Proof \[proof110530b\]. We include the details for the sake of thoroughness.)
Since $M\neq 0$, we have ${\operatorname{Hom}_{R}(\omega,M)}\neq 0$, by Fact \[fact110527a\](b), so Theorem \[thm110527a\] implies that there are an $R$-regular sequence $\underline z = z_1, \ldots, z_r$, integers $n_0, n_1, \ldots , n_r {\geqslant}0$ with $n_r{\geqslant}1$, and an exact sequence $$\label{proof110609a1}
0 \to {\operatorname{Hom}_{R}(\omega,M)} \to Z' \to N' \to 0$$ with $Z' = \oplus_{i=0}^r(R/(\underline z_{[i]}))^{n_i}$ and ${\mathrm{G}\text{-}\!\dim}_R(N') {\leqslant}r$. Fact \[disc01xy\](a) implies that $Z'\in{{\mathcal{A}}}_\omega(R)$. From Fact \[disc01xy\](b), we conclude that $N'\in{{\mathcal{A}}}_\omega(R)$, which implies that ${\operatorname{Tor}^{R}_{1}(\omega,N')}=0$. Thus, an application of ${\omega\otimes_{R}-}$ to the sequence yields the next exact sequence: $$\label{proof110609a2}
0 \to {\omega\otimes_{R}{\operatorname{Hom}_{R}(\omega,M)}} \to {\omega\otimes_{R}Z'} \to {\omega\otimes_{R}N'} \to 0.$$ The assumption $M\in{{\mathcal{B}}}_\omega(R)$ gives an isomorphism ${\omega\otimes_{R}{\operatorname{Hom}_{R}(\omega,M)}}\cong M$. The equality $Z' = \oplus_{i=0}^r(R/(\underline z_{[i]}))^{n_i}$ implies that $Z:={\omega\otimes_{R}Z'} = \oplus_{i=0}^r(\omega/(\underline z_{[i]})\omega)^{n_i}$.
The condition $N'\in{{\mathcal{A}}}_\omega(R)$ implies that $N:={\omega\otimes_{R}N'}\in{{\mathcal{B}}}_\omega(R)$ by Fact \[disc01xy\](c), and $N'\cong{\operatorname{Hom}_{R}(\omega,N)}$ by the definition of what it means for $N'$ to be in ${{\mathcal{A}}}_\omega(R)$. Thus, we have ${\mathrm{G}\text{-}\!\dim}({\operatorname{Hom}_{R}(\omega,N)})={\mathrm{G}\text{-}\!\dim}_R(N') {\leqslant}r$. So, the sequence satisfies the conclusion of Theorem \[thm110608a\].
For our next results, we need a better understanding of the relationship between semidualizing modules and a pointwise dualizing module.
\[lem110611a\] Assume that $R$ is Cohen-Macaulay with a pointwise dualizing module $\omega$. If $C$ is a semidualizing $R$-module, then so is ${\operatorname{Hom}_{R}(C,\omega)}$.
In the local case, this result is standard; see, e.g., [@sather:bnsc Facts 1.18–1.19]. Since the semidualizing and (pointwise) dualizing conditions are local by Fact \[fact110527a\](a), the general case of the result follows.
The next two lemmas elaborate on the local-global behavior for our invariants.
\[lem110611d\] Let $M$ be a finitely generated $R$-module, and let $C$ be a semidualizing $R$-module. Then ${{{\mathcal{P}}_C}\text{-}{\operatorname{pd}}}_{R}(M)<\infty$ if and only if $\text{P}_{C_{{{\mathfrak{m}}}}}\text{-}{\operatorname{pd}}_{R_{{{\mathfrak{m}}}}}(M_{{{\mathfrak{m}}}})<\infty$ for each maximal ideal ${{\mathfrak{m}}}\subset R$.
The forward implication follows from the inequality $\text{P}_{C_{{{\mathfrak{m}}}}}\text{-}\dim_{R_{{{\mathfrak{m}}}}}(M_{{{\mathfrak{m}}}}){\leqslant}{{{\mathcal{P}}_C}\text{-}{\operatorname{pd}}}_R(M)$, which is straightforward to verify. For the converse, assume that $\text{P}_{C_{{{\mathfrak{m}}}}}\text{-}{\operatorname{pd}}_{R_{{{\mathfrak{m}}}}}(M_{{{\mathfrak{m}}}})<\infty$ for each maximal ideal ${{\mathfrak{m}}}\subset R$. It follows that the module ${\operatorname{Hom}_{R}(C,M)}_{{{\mathfrak{m}}}}\cong{\operatorname{Hom}_{R_{{{\mathfrak{m}}}}}(C_{{{\mathfrak{m}}}},M_{{{\mathfrak{m}}}})}$ has finite projective dimension over $R_{{{\mathfrak{m}}}}$ for all ${{\mathfrak{m}}}$. Hence, the finitely generated $R$-module ${\operatorname{Hom}_{R}(C,M)}$ has finite projective dimension, so Fact \[disc01xyz\] implies that ${{{\mathcal{P}}_C}\text{-}{\operatorname{pd}}}_{R}(M)$ is finite.
The next lemma is a souped-up version of a special case of a result of Takahashi and White that is documented in [@sather:tate1].
\[lem110612a\] Assume that $R$ is Cohen-Macaulay with a pointwise dualizing module $\omega$. For each finitely generated $R$-module $M$, one has ${{\mathcal{P}}}_{\omega}\text{-}{\operatorname{pd}}_R(M)<\infty$ if and only if $M$ has locally finite injective dimension.
When $\omega$ is a dualizing module for $R$, i.e., when $R$ has finite Krull dimension, this result is from [@sather:tate1 Lemma 2.7]. In particular, this takes care of the local case. The general case follows from the local case, by Lemma \[lem110611d\].
Here are some more notions from [@holm:smarghd].
\[notation08b\] Let $C$ be a semidualizing $R$-module, and let ${{\mathcal{I}}_C}(R)$ denote the class of modules $N\cong {\operatorname{Hom}_{R}(C,I)}$ with $I$ injective. Modules in ${{\mathcal{I}}_C}(R)$ are called *$C$-injective*. We let ${{{\mathcal{I}}_C}\text{-}{\operatorname{id}}}_R(-)$ denote the homological dimension obtained from coresolutions in ${{\mathcal{I}}_C}(R)$. We say that an $R$-module $M$ has *locally finite ${{{\mathcal{I}}_C}\text{-}{\operatorname{id}}}$* provided that $\text{I}_{C_{{{\mathfrak{m}}}}}\text{-}{\operatorname{id}}_{R_{{{\mathfrak{m}}}}}(M_{{{\mathfrak{m}}}})<\infty$ for each maximal ideal ${{\mathfrak{m}}}\subset R$.
\[fact110612a\] Let $C$ be a semidualizing $R$-module. The Auslander class ${{\mathcal{A}}_C}(R)$ contains every $R$-module of finite ${{\mathcal{I}}}_C$-injective dimension; see [@holm:fear Lemmas 4.1 and 5.1]. Given an $R$-module $M$, one has ${{{\mathcal{I}}_C}\text{-}{\operatorname{id}}}_R(M)={\operatorname{id}}_R({C\otimes_{R}M})$ and ${\operatorname{id}}_R(M)={{{\mathcal{I}}_C}\text{-}{\operatorname{id}}}_R({\operatorname{Hom}_{R}(C,M)})$ by [@takahashi:hasm Theorem 2.11]. In particular, one has ${{{\mathcal{I}}_C}\text{-}{\operatorname{id}}}_R(M)<\infty$ if and only if ${\operatorname{id}}_R({C\otimes_{R}M})<\infty$, and one has ${\operatorname{id}}_R(M)<\infty$ if and only if ${{{\mathcal{I}}_C}\text{-}{\operatorname{id}}}_R({\operatorname{Hom}_{R}(C,M)})<\infty$. It follows that $M$ has locally finite ${{{\mathcal{I}}_C}\text{-}{\operatorname{id}}}$ if and only if ${C\otimes_{R}M}$ has locally finite injective dimension, and $M$ has locally finite injective dimension if and only if ${\operatorname{Hom}_{R}(C,M)}$ has locally finite ${{{\mathcal{I}}_C}\text{-}{\operatorname{id}}}$.
\[lem110612b\] Assume that $R$ is Cohen-Macaulay with a pointwise dualizing module $\omega$. Let $C$ be a semidualizing $R$-module, and let $x=x_1,\ldots,x_i\in R$ be an $R$-regular sequence. Then ${\operatorname{Hom}_{R}(C,\omega/(\underline x)\omega)}\cong{C^{\dagger}}/(\underline x){C^{\dagger}}$ where ${C^{\dagger}}={\operatorname{Hom}_{R}(C,\omega)}$.
Let $K$ be the Koszul complex $K^R(\underline x)$, and let $K^+$ denote the augmented Koszul complex $$K^+=(0\to R\to\cdots\to R\to \underbrace{R/(\underline x)}_{\text{deg. $-1$}}\to 0).$$ Since $x$ is $R$-regular, $K^+$ is an exact sequence of $R$-modules of finite projective dimension. Thus, each module $K^+_j$ is in ${{\mathcal{A}}}_{\omega}(R)$, so the induced complex $${\omega\otimes_{R}K^+}=(0\to \omega\to\cdots\to \omega\to \underbrace{\omega/(\underline x)\omega}_{\text{deg. $-1$}}\to 0)$$ is an exact sequence of $R$-modules locally of finite injective dimension. In particular, the modules in this exact sequence are in ${{\mathcal{B}}_C}(R)$, so the next sequence is also exact: $$\begin{aligned}
{\operatorname{Hom}_{R}(C,{\omega\otimes_{R}K^+})}\hspace{-2cm}\\
&=(0\to {\operatorname{Hom}_{R}(C,\omega)}\to\cdots\to {\operatorname{Hom}_{R}(C,\omega)}\to \underbrace{{\operatorname{Hom}_{R}(C,\omega/(\underline x)\omega)}}_{\text{deg. $-1$}}\to 0).\end{aligned}$$ It is straightforward to show that the differential on this sequence in positive degrees is the same as the differential on ${{C^{\dagger}}\otimes_{R}K}$. It follows that $${\operatorname{Hom}_{R}(C,\omega/(\underline x)\omega)}
\cong{\operatorname{H}}_0({{C^{\dagger}}\otimes_{R}K})
\cong{C^{\dagger}}/(\underline x){C^{\dagger}}$$ as desired.
\[proof110609b\] Assume that $R$ is Cohen-Macaulay with a pointwise dualizing module $\omega$ and that $M$ is a non-zero finitely generated $R$-module with locally finite ${{{\mathcal{I}}_C}\text{-}{\operatorname{id}}}$. Fact \[fact110612a\] implies that ${C\otimes_{R}M}$ has locally finite injective dimension. Since $M\neq 0$, we have ${C\otimes_{R}M}\neq 0$, by Fact \[fact110527a\](b), so Theorem \[thm110530a\] provides an integer $r = {\operatorname{pd}}_R({\operatorname{Hom}_{R}(\omega,{C\otimes_{R}M})})$, a proper $R$-regular sequence $\underline z = z_1, \ldots , z_r$, non-negative integers $n_0, n_1, \ldots , n_r$ with $n_r{\geqslant}1$, and an exact sequence $$\label{proof110609b1}
0 \to {C\otimes_{R}M} \to Z' \to N' \to 0$$ with $Z' = \oplus_{i=0}^r(\omega/(\underline z_{[i]})\omega)^{n_i}$, where $N'$ has locally finite injective dimension and ${\operatorname{pd}}_R({\operatorname{Hom}_{R}(\omega,N')}){\leqslant}r$.
As ${C\otimes_{R}M}$ has locally finite injective dimension, we have ${C\otimes_{R}M}\in{{\mathcal{B}}_C}(R)$ by Fact \[disc01xy\](b), so ${\operatorname{Ext}_{R}^{1}(C,{C\otimes_{R}M})}=0$ and ${\operatorname{Hom}_{R}(C,{C\otimes_{R}M})}\cong M$. Thus, an application of ${\operatorname{Hom}_{R}(C,-)}$ to the sequence yields the next exact sequence: $$\label{proof110609b2}
0 \to M \to {\operatorname{Hom}_{R}(C,Z')} \to {\operatorname{Hom}_{R}(C,N')} \to 0.$$ Since $N'$ has locally finite injective dimension, Fact \[fact110612a\] implies that the $R$-module $N:={\operatorname{Hom}_{R}(C,N')}$ has locally finite ${{{\mathcal{I}}_C}\text{-}{\operatorname{id}}}$. With $Z:={\operatorname{Hom}_{R}(C,Z')}$, it remains to show that $Z \cong \oplus_{i=0}^r({C^{\dagger}}/(\underline z_{[i]}){C^{\dagger}})^{n_i}$ and ${\operatorname{pd}}_R({\operatorname{Hom}_{R}(\omega,{C\otimes_{R}N})}){\leqslant}r$. The first of these follows from the next sequence of isomorphisms $$\begin{aligned}
Z
&\cong{\operatorname{Hom}_{R}(C,\oplus_{i=0}^r(\omega/(\underline z_{[i]})\omega)^{n_i})}\\
&\cong\oplus_{i=0}^r{\operatorname{Hom}_{R}(C,{\omega\otimes_{R}(R/(\underline z_{[i]})})}^{n_i}\\
&\cong\oplus_{i=0}^r({C^{\dagger}}/(\underline z_{[i]}){C^{\dagger}})^{n_i}\end{aligned}$$ where the last isomorphism is from Lemma \[lem110612b\]. To complete the proof, observe that since $N'$ has locally finite injective dimension, it is in ${{\mathcal{B}}_C}(R)$, so we have $N'\cong {C\otimes_{R}N}$. This implies that ${\operatorname{pd}}_R({\operatorname{Hom}_{R}(\omega,{C\otimes_{R}N})})={\operatorname{pd}}_R({\operatorname{Hom}_{R}(\omega,N')})
{\leqslant}r$, as desired.
In the next definition, the special case $C=R$ recovers the complete injective resolutions and Gorenstein injective modules.
\[notation08b’\] Let $C$ be a semidualizing $R$-module. A *complete ${{\mathcal{I}}_C}{{\mathcal{I}}}$-coresolu-tion* is a complex $Y$ of $R$-modules such that $Y$ is exact and ${\operatorname{Hom}_{R}(U,Y)}$ is exact for each $U\in{{\mathcal{I}}_C}(R)$, $Y_i$ is injective for $i{\leqslant}0$, and $Y_i$ is $C$-injective for $i>0$. An $R$-module $N$ is *$\text{G}_C$-injective* if there exists a complete ${{\mathcal{I}}_C}{{\mathcal{I}}}$-coresolution $Y$ such that $N\cong{\operatorname{Ker}}(\partial^Y_0)$, and $Y$ is a *complete ${{\mathcal{I}}_C}{{\mathcal{I}}}$-coresolution of $N$*. Let ${{\mathcal{GI}}_C}(R)$ denote the class of $\text{G}_C$-injective $R$-modules, and set ${{\mathcal{G}}}({{\mathcal{I}}_C})={{\mathcal{GI}}_C}(R)\cap{{\mathcal{A}}_C}(R)$. We let ${{\mathcal{GI}}_C}\text{-}{\operatorname{id}}_R(-)$ and ${{\mathcal{G}}}({{\mathcal{I}}_C})\text{-}{\operatorname{id}}_R(-)$ denote the homological dimensions obtained from coresolutions in ${{\mathcal{GI}}_C}(R)$ and ${{\mathcal{G}}}({{\mathcal{I}}_C})$, respectively.
An $R$-module $M$ has *locally finite ${{\mathcal{G}}}({{\mathcal{I}}_C})\text{-}{\operatorname{id}}$* provided that ${{\mathcal{G}}}({{\mathcal{I}}}_{C_{{{\mathfrak{m}}}}})\text{-}{\operatorname{id}}_{R_{{{\mathfrak{m}}}}}(M_{{{\mathfrak{m}}}})<\infty$ for each maximal ideal ${{\mathfrak{m}}}\subset R$. An $R$-module $M$ has *locally finite ${{\mathcal{G}}}{{\mathcal{I}}_C}\text{-}{\operatorname{id}}$* provided that ${{\mathcal{G}}}{{\mathcal{I}}}_{C_{{{\mathfrak{m}}}}}\text{-}{\operatorname{id}}_{R_{{{\mathfrak{m}}}}}(M_{{{\mathfrak{m}}}})<\infty$ for each maximal ideal ${{\mathfrak{m}}}\subset R$.
The next lemma explains the relation between the different Gorenstein injective dimensions.
\[lem0701’\] Let $C$ be a semidualizing $R$-module. For an $R$-module $M$, the following conditions are equivalent:
1. \[lem0701’i\] ${{\mathcal{G}}}({{\mathcal{I}}_C})\text{-}{\operatorname{id}}_R(M)<\infty$;
2. \[lem0701’ii\] ${{\mathcal{G}}}{{\mathcal{I}}_C}\text{-}{\operatorname{id}}_R(M)<\infty$ and $M\in{{\mathcal{A}}_C}(R)$; and
3. \[lem0701’iii\] ${\operatorname{Gid}}_R(C\otimes_RM)<\infty$ and $M\in{{\mathcal{A}}_C}(R)$.
When these conditions are satisfied, we have $$\begin{aligned}
{{\mathcal{G}}}({{\mathcal{I}}_C})\text{-}{\operatorname{id}}_R(M)
&
={{\mathcal{G}}}{{\mathcal{I}}_C}\text{-}{\operatorname{id}}_R(M)
={\operatorname{Gid}}_R(C\otimes_RM). \end{aligned}$$
We actually need the following corollary of the previous result.
\[lem110613a\] Let $C$ be a semidualizing $R$-module. For an $R$-module $M$, the following conditions are equivalent:
1. \[lem110613ai\] $M$ has locally finite ${{\mathcal{G}}}({{\mathcal{I}}_C})\text{-}{\operatorname{id}}$;
2. \[lem110613aii\] $M$ has locally finite ${{\mathcal{G}}}{{\mathcal{I}}_C}\text{-}{\operatorname{id}}$ and $M\in{{\mathcal{A}}_C}(R)$; and
3. \[lem110613aiii\] ${C\otimes_{R}M}$ has locally finite ${\operatorname{Gid}}$ and $M\in{{\mathcal{A}}_C}(R)$.
This follows from the definitions of the locally finite Gorenstein injective dimensions and the local-global principal for Auslander classes from Fact \[disc01xy\](b).
\[proof110609c\] Assume that $R$ is Cohen-Macaulay with a pointwise dualizing module $\omega$ and that $M$ is a non-zero finitely generated $R$-module in ${{\mathcal{A}}_C}(R)$ with locally finite ${{\mathcal{G}}}{{{\mathcal{I}}_C}\text{-}{\operatorname{id}}}$. Lemma \[lem110613a\] implies that ${C\otimes_{R}M}$ has locally finite G-injective dimension. Since $M\neq 0$, we have ${C\otimes_{R}M}\neq 0$, by Fact \[fact110527a\](b), so Theorem \[thm110608a\] provides an integer $r = {\mathrm{G}\text{-}\!\dim}_R({\operatorname{Hom}_{R}(\omega,{C\otimes_{R}M})})$, a proper $R$-regular sequence $\underline z = z_1, \ldots , z_r$, non-negative integers $n_0, n_1, \ldots , n_r$ with $n_r{\geqslant}1$, and an exact sequence $$\label{proof110609c1}
0 \to {C\otimes_{R}M} \to Z' \to N' \to 0$$ such that $Z' = \oplus_{i=0}^r(\omega/(\underline z_{[i]})\omega)^{n_i}$, $N'\in{{\mathcal{B}}}_{\omega}(R)$, and ${\mathrm{G}\text{-}\!\dim}_R({\operatorname{Hom}_{R}(\omega,N')}){\leqslant}r$.
Since $M$ is in ${{\mathcal{A}}_C}(R)$, we have ${\operatorname{Ext}_{R}^{1}(C,{C\otimes_{R}M})}=0$ and ${\operatorname{Hom}_{R}(C,{C\otimes_{R}M})}\cong M$. Thus, an application of ${\operatorname{Hom}_{R}(C,-)}$ to the sequence yields the next exact sequence: $$\label{proof110609c2}
0 \to M \to {\operatorname{Hom}_{R}(C,Z')} \to {\operatorname{Hom}_{R}(C,N')} \to 0.$$ Note that $Z'$ has locally finite injective dimension, so we have $Z'\in{{\mathcal{B}}_C}(R)$, which implies that ${\operatorname{Hom}_{R}(C,Z')}$ and ${\operatorname{Hom}_{R}(C,N')}$ are in ${{\mathcal{A}}_C}(R)$ by Fact \[disc01xy\](b)–(c). With $Z:={\operatorname{Hom}_{R}(C,Z')}$, we have ${\mathrm{G}\text{-}\!\dim}_R({\operatorname{Hom}_{R}(\omega,{C\otimes_{R}N})}){\leqslant}r$ and $Z \cong \oplus_{i=0}^r({C^{\dagger}}/(\underline z_{[i]}){C^{\dagger}})^{n_i}$, as in Proof \[proof110609b\].
Acknowledgments {#acknowledgments .unnumbered}
===============
We are grateful to the referee for her/his thoughtful comments.
[10]{}
M. Auslander, *Anneaux de [G]{}orenstein, et torsion en algèbre commutative*, Séminaire d’Algèbre Commutative dirigé par Pierre Samuel, vol. 1966/67, Secrétariat mathématique, Paris, 1967. [MR ]{}[37 \#1435]{}
M. Auslander and M. Bridger, *Stable module theory*, Memoirs of the American Mathematical Society, No. 94, American Mathematical Society, Providence, R.I., 1969. [MR ]{}[42 \#4580]{}
L. L. Avramov and R.-O. Buchweitz, *Support varieties and cohomology over complete intersections*, Invent. Math. **142** (2000), no. 2, 285–318. [MR ]{}[1794064 (2001j:13017)]{}
L. L. Avramov, S. B. Iyengar, and J. Lipman, *Reflexivity and rigidity for complexes. [I]{}. [C]{}ommutative rings*, Algebra Number Theory **4** (2010), no. 1, 47–86. [MR ]{}[2592013]{}
L. L. Avramov and A. Martsinkovsky, *Absolute, relative, and [T]{}ate cohomology of modules of finite [G]{}orenstein dimension*, Proc. London Math. Soc. (3) **85** (2002), 393–440. [MR ]{}[2003g:16009]{}
L. W. Christensen, A. Frankild, and H. Holm, *On [G]{}orenstein projective, injective and flat dimensions—a functorial description with applications*, J. Algebra **302** (2006), no. 1, 231–279. [MR ]{}[2236602]{}
E. E. Enochs and O. M. G. Jenda, *Gorenstein injective and projective modules*, Math. Z. **220** (1995), no. 4, 611–633. [MR ]{}[1363858 (97c:16011)]{}
E. E. Enochs, O. M. G. Jenda, and J. Z. Xu, *Foxby duality and [G]{}orenstein injective and projective modules*, Trans. Amer. Math. Soc. **348** (1996), no. 8, 3223–3234. [MR ]{}[1355071 (96k:13010)]{}
H.-B. Foxby, *Gorenstein modules and related modules*, Math. Scand. **31** (1972), 267–284 (1973). [MR ]{}[48 \#6094]{}
[to3em]{}, *Gorenstein dimensions over [C]{}ohen-[M]{}acaulay rings*, Proceedings of the international conference on commutative algebra (W. Bruns, ed.), Universität Osnabrück, 1994, pp. 59–63.
A. Frankild and S. Sather-Wagstaff, *Reflexivity and ring homomorphisms of finite flat dimension*, Comm. Algebra **35** (2007), no. 2, 461–500. [MR ]{}[2294611]{}
E. S. Golod, *[$G$]{}-dimension and generalized perfect ideals*, Trudy Mat. Inst. Steklov. **165** (1984), 62–66, Algebraic geometry and its applications. [MR ]{}[85m:13011]{}
R. Hartshorne, *Residues and duality*, Lecture Notes in Mathematics, No. 20, Springer-Verlag, Berlin, 1966. [MR ]{}[36 \#5145]{}
M. Hochster and Y. Yao, *An embedding theorem for modules of finite ([G]{}-)projective dimension*, preprint (2009).
H. Holm and P. J[ø]{}rgensen, *Semi-dualizing modules and related [G]{}orenstein homological dimensions*, J. Pure Appl. Algebra **205** (2006), no. 2, 423–445. [MR ]{}[2203625]{}
H. Holm and D. White, *Foxby equivalence over associative rings*, J. Math. Kyoto Univ. **47** (2007), no. 4, 781–808. [MR ]{}[2413065]{}
S. Sather-Wagstaff, *Bass numbers and semidualizing complexes*, Commutative algebra and its applications, Walter de Gruyter, Berlin, 2009, pp. 349–381. [MR ]{}[2640315]{}
S. Sather-Wagstaff, T. Sharif, and D. White, *Tate cohomology with respect to semidualizing modules*, J. Algebra **324** (2010), no. 9, 2336–2368. [MR ]{}[2684143]{}
R. Takahashi and D. White, *Homological aspects of semidualizing modules*, Math. Scand. **106** (2010), no. 1, 5–22. [MR ]{}[2603458]{}
W. V. Vasconcelos, *Divisor theory in module categories*, North-Holland Publishing Co., Amsterdam, 1974, North-Holland Mathematics Studies, No. 14, Notas de Matemática No. 53. \[Notes on Mathematics, No. 53\]. [MR ]{}[0498530 (58 \#16637)]{}
D. White, *Gorenstein projective dimension with respect to a semidualizing module*, J. Commut. Algebra **2** (2010), no. 1, 111–137. [MR ]{}[2607104]{}
[^1]: This material is based on work supported by North Dakota EPSCoR and National Science Foundation Grant EPS-0814442. The author was supported in part by a grant from the NSA
[^2]: This is also now called “Gorenstein projective dimension” after the work of Enochs, Jenda, and others. However, this is a generalization of Auslander and Bridger’s original notion for non-finitely-generated modules. Since we are focused on finitely generated modules here, we continue with Auslander and Bridger’s terminology.
[^3]: It is worth noting that these notions appeared (more or less) implicitly in several places. However, to the best of our knowledge these are the first places where ${\text{G}_C}$-dimension and ${{\mathcal{P}}_C}$-projective dimension were studied explicitly.
[^4]: Ths can be partially explained by the fact that, at the time of the writing of this article, the localization question for G-injective dimension is still not completely answered. Remark \[disc110726a\].
[^5]: In [@hochster:etmf] the term “global canonical module” is used in place of “pointwise dualizing module”. Our choice follows the terminology of Grothendieck and Hartshorne [@hartshorne:rad].
[^6]: We observe the same convention for any homological dimension of the zero module.
[^7]: Most good behavior is also known when $R$ has a dualizing complex, but we restrict our attention to the Cohen-Macaulay case. For instance, the conclusion of Lemma \[lem110610a\] holds when $R$ is only assumed to have a pointwise dualizing complex; the proof is the same, using results from [@christensen:ogpifd] in place of the results from [@enochs:fdgipm].
[^8]: It is worth noting that the results in [@enochs:fdgipm] assume that $R$ is local, hence with finite Krull dimension; also, the results of [@christensen:ogpifd] assume implicitly that $R$ has finite Krull dimension since the definition of a dualizing complex used there includes an assumption of finite injective dimension. Contrast this with our definition of pointwise dualizing module, and with Grothendieck’s definition of a pointwise dualizing complex from [@hartshorne:rad].
[^9]: Note that this uses the characterization of totally reflexive modules in terms of “complete resolutions” found in [@avramov:svcci (4.4.4)]; see also [@avramov:aratc Theorem 3.1]. Specifically, an $R$-module $M$ is totally reflexive if and only if there is an exact complex $F=\cdots{\xrightarrow}{\partial^F_1}F_0{\xrightarrow}{\partial^F_0}F_{-1}{\xrightarrow}{\partial^F_{-1}}\cdots$ such that ${\operatorname{Hom}_{R}(F,R)}$ is exact and $M\cong{\operatorname{Im}}(\partial^F_0)$. Note that the local result was also announced in [@foxby:gdcmr Corollary 3.3].
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In this paper, we use known camera motion associated to a video sequence of a static scene in order to estimate and incrementally refine the surrounding depth field. We exploit the SO(3)-invariance of brightness and depth fields dynamics to customize standard image processing techniques. Inspired by the Horn-Schunck method, we propose a SO(3)-invariant cost to estimate the depth field. At each time step, this provides a diffusion equation on the unit Riemannian sphere that is numerically solved to obtain a real time depth field estimation of the entire field of view. Two asymptotic observers are derived from the governing equations of dynamics, respectively based on optical flow and depth estimations: implemented on noisy sequences of synthetic images as well as on real data, they perform a more robust and accurate depth estimation. This approach is complementary to most methods employing state observers for range estimation, which uniquely concern single or isolated feature points.'
author:
- 'Nadège Zarrouati [^1]'
- 'Emanuel Aldea [^2]'
- 'Pierre Rouchon [^3]'
title: '**SO(3)-invariant asymptotic observers for dense depth field estimation based on visual data and known camera motion** '
---
Introduction
============
Many vision applications are aimed at assisting in interacting with the environment. In military as well as in civilian applications, moving in an environment requires topographical knowledge: either in order to avoid obstacles or to engage targets. Since this information is often inaccessible in advance, the real-time computation of a 3D map is a goal that has kept the research community busy for many years. For example, environment reconstruction is tightly related to the SLAM problem [@Smith90], which is addressed by nonlinear filtering of observed key feature locations (e.g. [@EKFMono; @Montemerlo03a]), or by bundle adjustment [@StrasdatMD10; @Strasdat-RSS-10]. However, estimating a sparse point cloud is often insufficient, yet the transition between a discrete local distribution of 3D locations to a continuous depth estimation of the surroundings is an ongoing research topic. Dynamical systems provide interesting means for incrementally estimating depth information based on the output of vision sensors, since only the current estimates are required, and image batch processing is avoided.
For our work, we are interested in recovering in real-time the depth field around the carrier under the assumptions of known camera motion and known projection model for the onboard monocular camera. The problem of designing an observer to estimate the depth of *single or isolated keypoints* has raised a lot of interest, specifically in the case where the relative motion of the carrier is described by constant known [@AbdursulIG04; @DahlWLH10], constant unknown [@Heyden09] or time-varying known [@chen02; @Dixon03; @Karagiannis05; @deluca08; @Astolfi10] affine dynamics. From a different perspective, the seminal paper of [@MatthiesKS89] performs incremental depth field refining for the whole field of view via *iconic* (pixel-wise) Kalman filtering. Video systems, typically found on autonomous vehicles, have successfully used this approach for refining the disparity values obtained by stereo cameras in order to estimate the free space ahead [@Hoilund10]. Average optical flow estimations over planar surfaces have also been used for terrain following, in order to stabilize the carrier at a certain pseudo-distance [@Herisse10]. Yet, none of these methods provide an accurate dense depth estimation in a general setting concerning the environment and the camera dynamics.
We propose a novel frame of methods relying on a system of partial differential equations describing the $SO(3)$-invariant dynamics of the brightness perceived by the camera and of the depth field of the environment. Based on this invariant kinematic model and the knowledge of the camera motion, these methods provide dense estimations of the depth field at each time-step and exploit such $SO(3)$-invariance.
The present paper is structured as follows. The invariant equations governing the dynamics of the brightness and depth fields are recalled in section \[sec:probstat\] and their formulation in pinhole coordinates is given. In section \[sec:VarMeth\], we adapt the Horn-Schunck algorithm to a variational method providing depth estimation. In section \[sec:Observer\], we propose two asymptotic observers for depth field estimations: the first one is based on standard optical flow measures and the second one enables the refinement of rough or inaccurate depth estimations; we prove their convergence under geometric assumptions concerning the camera dynamics and the environment. In section \[sec:implement\], we test these methods on synthetic data and compare their accuracy, their robustness to noise and their convergence rate; tested on real data, this approach gives promising results.
The $SO(3)$-invariant model {#sec:probstat}
============================
The partial differential system on ${{\mathbb S}}^2$ {#subsec:PDS}
------------------------------------------------------
The model is based on geometric assumptions introduced in [@Bonnabel-Rouchon2009]. We consider a spherical camera, whose motion is known. Linear and angular velocities $v(t)$ and $\omega (t)$ are expressed in the camera frame. Position of the optical center in the reference frame $\cal R$ is denoted by $C(t)$. Orientation versus $\cal R$ is given by the quaternion $q(t)$: any vector $\varsigma$ in the camera frame corresponds to the vector $q\varsigma q^*$ in the reference frame $\cal R$ using the identification of vectors as imaginary quaternions. We have thus: ${{\tfrac{d}{dt}}}q = \frac{1}{2} q\omega$. A pixel is labeled by the unit vector $\eta$ in the camera frame: $\eta$ belongs to the sphere $\mathbb S^2$ and receives the brightness $y(t,\eta)$. Thus at each time $t$, the image produced by the camera is described by the scalar field $\mathbb S^2\ni \eta \mapsto y(t,\eta)\in{{\mathbb R}}$.
The scene is modeled as a closed, $C^1$ and convex surface $\Sigma$ of ${{\mathbb R}}^3$, diffeomorphic to $\mathbb S^2$. The camera is inside the domain $\Omega\subset {{\mathbb R}}^3$ delimited by $\Sigma=\partial\Omega$. To a point $M\in\Sigma$ corresponds one and only one camera pixel: if the points of $\Sigma$ are labeled by $s\in{{\mathbb S}}^2$, for each time $t$, a continuous and invertible transformation $\mathbb S^2\ni s\mapsto \phi(t,s)\in \mathbb S^2$ enables to express $\eta$ as a function of $s$: $\eta=\phi(t,s)$.
The density of light emitted by a point $M(s)\in\Sigma$ does not depend on the direction of emission ($\Sigma$ is a Lambertian surface) and is independent of $t$ (the scene is static). This means that $y(t,\eta)$ depends only on $s$: thus $y$ can be seen either as a function of $(t,\eta)$ or, via the transformation $\phi$, as a function of $s$. The distance $C(t)M(s)$ between the optical center and the object seen in the direction $\eta=\phi(t,s)$ is denoted by $D(t,\eta)$, and its inverse by $\Gamma = 1/D$. Fig.\[fig:notations\] illustrates the model and the notations. We assume that $s\mapsto y(s)$ is a $C^1$ function. For each $t$, $s\mapsto D(t,s)$ is $C^1$ since $\Sigma$ is a $C^1$ surface of ${{\mathbb R}}^3$.
![Model and notations of a spherical camera in a static environment.[]{data-label="fig:notations"}](notations){width="60.00000%"}
Under these assumptions, we first have: $$\left.{{\frac{\partial y}{\partial t}}}\right|_s=0, \quad
\left.{{\frac{\partial \Gamma}{\partial t}}}\right|_s=\Gamma^2v\cdot\eta
\label{eq:gamma_direction}$$ then $$\begin{gathered}
\left.{{\frac{\partial h}{\partial t}}}\right|_s=\left.{{\frac{\partial h}{\partial t}}}\right|_{\eta}+\left.{{\frac{\partial h}{\partial \eta}}}\right|_{t}\left.{{\frac{\partial \eta}{\partial t}}}\right|_s
=\left.{{\frac{\partial h}{\partial t}}}\right|_{\eta}+\nabla h\cdot\left.{{\frac{\partial \eta}{\partial t}}}\right|_s\end{gathered}$$ where $h$ is any scalar field defined on $\mathbb S^2$ and $\nabla h$ its gradient with respect to the Riemannian metric on $\mathbb S^2$. The value of $\nabla h$ at $\eta\in{{\mathbb S}}^2$ is identified with a vector of ${{\mathbb R}}^3$ tangent to the sphere at the point $\eta$ also identified to a unitary vector of ${{\mathbb R}}^3$ in the camera moving frame. The Euclidean scalar product of two vectors $a$ and $b$ in ${{\mathbb R}}^3$ is denoted by $a\cdot b$ and their wedge product by $a\times b$. By differentiation, the identity $q\eta q^*=\frac{\overrightarrow{C(t)M(s)}}{\left||C(t)M(s)\right||}$, where $^*$ denotes conjugation and $\eta$ is identified to an imaginary quaternion, yields $$\left.{{\frac{\partial \eta}{\partial t}}}\right|_s=\eta \times (\omega +\Gamma\eta\times v)$$ since the vector $\eta\times \omega$ corresponds to the imaginary quaternion $(\omega\eta - \eta\omega)/2$. Therefore, the intensity $y(t,\eta)$ and the inverse depth $\Gamma(t,\eta)$ satisfy the following equations: $${{\frac{\partial y}{\partial t}}} =- \nabla y \cdot(\eta \times (\omega +\Gamma\eta\times v))
\label{eq:flot}$$ $${{\frac{\partial \Gamma}{\partial t}}} =- \nabla \Gamma \cdot(\eta \times (\omega +\Gamma\eta\times v)) + \Gamma^{2}v\cdot\eta
\label{eq:prof}$$ Equations and are $SO(3)$-invariant: they remain unchanged by any rotation described by the quaternion $\sigma$ and changing $(\eta,\omega,v)$ to $(\sigma \eta \sigma^*,\sigma \omega \sigma^*,\sigma v \sigma^* )$. Equation is the well-known optical flow equation that can be found under different forms in numerous papers (see [@Murray09] or [@Astolfi10] for example), while is less standard (see e.g., [@Bonnabel-Rouchon2009]).
The system in pinhole coordinates {#subsec:Pinhole}
---------------------------------
To use this model with camera data, one needs to write the invariant equations and with local coordinates on ${{\mathbb S}}^2$ corresponding to a rectangular grid of pixels. One popular solution is to use the pinhole camera model, where the pixel of coordinates $(z_1,z_2)$ corresponds to the unit vector $\eta\in{{\mathbb S}}^2$ of coordinates in ${{\mathbb R}}^3$: $\left(1+z_{1}^2+z_{2}^2\right)^{-1/2}(z_1,z_2,1)^T$. The optical camera axis (pixel $(z_1,z_2)=(0,0)$) corresponds here to the direction $z_3$. Directions $1$ and $2$ correspond respectively to the horizontal axis from left to right and to the vertical axis from top to bottom on the image frame.
The gradients $\nabla y$ and $\nabla \Gamma$ must be expressed with respect to $z_1$ and $z_2$. Let us detail this derivation for $y$. Firstly, $\nabla y$ is tangent to $\mathbb S^2$, thus $\nabla y\cdot\eta=0$. Secondly, the differential $dy$ corresponds to $\nabla y\cdot d\eta$ and to ${{\frac{\partial y}{\partial z_{1}}}} dz_{1} +{{\frac{\partial y}{\partial z_{2}}}} dz_{2}$. By identification, we get the Cartesian coordinates of $\nabla y$ in ${{\mathbb R}}^3$. Similarly we get the three coordinates of $\nabla\Gamma$. Injecting these expressions in and , we get the following partial differential equations (PDE) corresponding to and in local pinhole coordinates: $$\begin{aligned}
{{\frac{\partial y}{\partial t}}}= &- {{\frac{\partial y}{\partial z_1}}} \left[\begin{array}{c}z_1z_2\omega_1-(1+{z_1}^2)\omega_2+z_2\omega_3\nonumber\\
+\Gamma\sqrt{1+z_{1}^2+z_{2}^2}(-v_1+z_1v_3)\end{array}\right] \nonumber\\
&- {{\frac{\partial y}{\partial z_2}}} \left[\begin{array}{c}(1+{z_2}^2)\omega_1-z_1z_2\omega_2-z_1\omega_3\\
+\Gamma\sqrt{1+z_{1}^2+z_{2}^2}(-v_2+z_2v_3)\end{array}\right]\nonumber
\label{eq:flotcart}
\end{aligned}$$
$$\begin{aligned}
{{\frac{\partial \Gamma}{\partial t}}}= &- {{\frac{\partial \Gamma}{\partial z_1}}} \left[\begin{array}{c}z_1z_2\omega_1-(1+{z_1}^2)\omega_2+z_2\omega_3\nonumber\\
+\Gamma\sqrt{1+z_{1}^2+z_{2}^2}(-v_1+z_1v_3)\end{array}\right]\nonumber\\
&- {{\frac{\partial \Gamma}{\partial z_2}}} \left[\begin{array}{c}(1+{z_2}^2)\omega_1-z_1z_2\omega_2-z_1\omega_3\\
+\Gamma\sqrt{1+z_{1}^2+z_{2}^2}(-v_2+z_2v_3)\end{array}\right]\nonumber\\
&+\Gamma^2(z_1v_1+z_2v_2+v_3)\nonumber
\label{eq:profcart}
\end{aligned}$$
where $v_1$, $v_2$, $v_3$, $\omega_1$, $\omega_2$, $\omega_3$ are the components of linear and angular velocities in the camera frame.
Depth estimation inspired by Horn-Schunck method {#sec:VarMeth}
================================================
The Horn-Schunck variational method {#ssec:HS}
-----------------------------------
In [@HS81], Horn and Schunck described a method to compute the optical flow, defined as “the distribution of apparent velocities of movement of brightness patterns in an image”. The entire method is based on the optical flow constraint written in the compact form $${{\frac{\partial y}{\partial t}}}+V_1{{\frac{\partial y}{\partial z_1}}}+V_2{{\frac{\partial y}{\partial z_2}}}=0
\label{eq:OFC}$$ Identification with yields $$\begin{split}
&V_1(t,z)=f_{1}(z,\omega(t))+\Gamma(t,z) g_{1}(z,v(t))\\
&V_2(t,z)=f_{2}(z,\omega(t))+\Gamma(t,z) g_{2}(z,v(t))
\end{split}
\label{eq:defV}$$ with $$\begin{split}
&f_{1}(z,\omega)=z_1z_2\omega_1-(1+{z_1}^2)\omega_2+z_2\omega_3\\
&g_{1}(z,v)=\sqrt{1+z_{1}^2+z_{2}^2}(-v_1+z_1v_3)\\
&f_{2}(z,\omega)=(1+{z_2}^2)\omega_1-z_1z_2\omega_2-z_1\omega_3\\
&g_{2}(z,v)=\sqrt{1+z_{1}^2+z_{2}^2}(-v_2+z_2v_3)
.
\end{split}
\label{eq:fg}$$ For each time $t$, the apparent velocity field $V=V_1{{\frac{\partial }{\partial z_1}}}+V_2{{\frac{\partial }{\partial z_2}}}$ is then estimated by minimizing versus $W=W_1{{\frac{\partial }{\partial z_1}}}+W_2{{\frac{\partial }{\partial z_2}}}$ the following cost (the image $\mathcal I$ is a rectangle of ${{\mathbb R}}^2$ here) $$\begin{gathered}
I(W)=\iint_{\mathcal I}\bigg(\big({{\frac{\partial y}{\partial t}}}+W_1{{\frac{\partial y}{\partial z_1}}}+W_2{{\frac{\partial y}{\partial z_2}}}\big)^2 \\
+\alpha^{2} ({\nabla W_1}^2+{\nabla W_2}^2)\bigg) dz_1dz_2
\label{eq:integraleOFC1}\end{gathered}$$ where $\nabla$ is the gradient operator in the Euclidian plane $(z_1,z_2)$, $\alpha >0$ is a regularization parameter and the partial derivatives ${{\frac{\partial y}{\partial t}}}$, ${{\frac{\partial y}{\partial z_1}}}$ and ${{\frac{\partial y}{\partial z_2}}}$ are assumed to be known.
Such Horn-Schunk estimation of $V$ at time $t$ is denoted by $${{V_{\text{\tiny HS}}}}(t,z)={{V_{\text{\tiny HS}}}}_1(t,z) {{\frac{\partial }{\partial z_1}}}+{{V_{\text{\tiny HS}}}}_2(t,z) {{\frac{\partial }{\partial z_2}}}.$$ For each time $t$, usual calculus of variation yields the following PDE’s for ${{V_{\text{\tiny HS}}}}$: $$\begin{aligned}
&\left({{\frac{\partial y}{\partial z_1}}}\right)^{2} {{V_{\text{\tiny HS}}}}_1+{{\frac{\partial y}{\partial z_1}}}{{\frac{\partial y}{\partial z_2}}} {{V_{\text{\tiny HS}}}}_2=\alpha^{2} \Delta {{V_{\text{\tiny HS}}}}_1-{{\frac{\partial y}{\partial z_1}}}{{\frac{\partial y}{\partial t}}}
\\
&{{\frac{\partial y}{\partial z_1}}}{{\frac{\partial y}{\partial z_2}}}{{V_{\text{\tiny HS}}}}_1 + \left({{\frac{\partial y}{\partial z_2}}}\right) ^{2} {{V_{\text{\tiny HS}}}}_2=\alpha^{2} \Delta {{V_{\text{\tiny HS}}}}_2-{{\frac{\partial y}{\partial z_2}}}{{\frac{\partial y}{\partial t}}}\end{aligned}$$ with boundary conditions ${{\frac{\partial {{V_{\text{\tiny HS}}}}_1}{\partial n}}}={{\frac{\partial {{V_{\text{\tiny HS}}}}_2}{\partial n}}}=0$ ($n$ the normal to $\partial \mathcal I$). Here $\Delta$ is the Laplacian operator in the Euclidian space $(z_1,z_2)$. The numerical resolution is usually based on
- computations of ${{\frac{\partial y}{\partial z_1}}}$, ${{\frac{\partial y}{\partial z_2}}}$ and ${{\frac{\partial y}{\partial t}}}$ via differentiation filters (Sobel filtering) directly from the image data at different times around $t$.
- approximation of $\Delta {{V_{\text{\tiny HS}}}}_1$ and $\Delta {{V_{\text{\tiny HS}}}}_2$ by the difference between the weighted mean $\bar {{V_{\text{\tiny HS}}}}_1$ and $\bar {{V_{\text{\tiny HS}}}}_2$ of ${{V_{\text{\tiny HS}}}}_1$ and ${{V_{\text{\tiny HS}}}}_2$ on the neighboring pixels and their values at the current pixel;
- iterative resolution (Jacobi scheme) of the resulting linear system in ${{V_{\text{\tiny HS}}}}_1$ and ${{V_{\text{\tiny HS}}}}_2$.
The convergence of this numerical method of resolution was proven in [@convHS]. Three parameters have a direct impact on the speed of convergence and on the precision: the regularization parameter $\alpha$, the number of iterations for the Jacobi scheme and the initial values of ${{V_{\text{\tiny HS}}}}_1$ and ${{V_{\text{\tiny HS}}}}_2$ at the beginning of this iteration step. To be specific, $\alpha$ should neither be too small in order to filter noise appearing in differentiation filters applied on $y$, nor too large in order to have ${{V_{\text{\tiny HS}}}}$ close to $V$ when $\nabla V\neq 0$.
Adaptation to depth estimation {#ssec:HSDepth}
-------------------------------
Instead of minimizing the cost $I$ given by with respect to any $W_1$ and $W_2$, let us define a new invariant cost $J$, $$\begin{gathered}
J(\Upsilon)=
\iint_{\mathcal J} \bigg(\left({{\frac{\partial y}{\partial t}}} +\nabla y \cdot\left(\eta \times \left(\omega +\Upsilon\eta\times v\right)\right)\right)^2\\
+\alpha^{2} {\nabla \Upsilon}^2 \bigg) d\sigma_{\eta}
\label{eq:integraleOFC2}\end{gathered}$$ and minimize it with respect to any depth profile $\mathcal J \ni \eta \mapsto \Upsilon(t,\eta)\in{{\mathbb R}}$. The time $t$ is fixed here and $d\sigma_\eta$ is the Riemannian infinitesimal surface element on ${{\mathbb S}}^2$. $\mathcal J\subset {{\mathbb S}}^2$ is the domain where $y$ is measured and $\alpha >0$ is the regularization parameter.
The first order stationary condition of $J$ with respect to any variation of $\Upsilon$ yields the following invariant PDE characterizing the resulting estimation ${{\Gamma_{\text{\tiny HS}}}}$ of $\Gamma$: $$\begin{gathered}
\alpha^{2}\Delta {{\Gamma_{\text{\tiny HS}}}}=
\left({{\frac{\partial y}{\partial t}}} +\nabla y \cdot\left(\eta \times \left(\omega +{{\Gamma_{\text{\tiny HS}}}}\eta\times v\right)\right)\right) \ldots \\ \ldots \left(\nabla y \cdot\left(\eta \times \left(\eta\times v\right)\right)\right) \phantom{eeeeee} \text{on }\mathcal J
\label{eq:CV}\end{gathered}$$ $$\phantom{eeeeeeeeeeeeeeeeeee}{{\frac{\partial {{\Gamma_{\text{\tiny HS}}}}}{\partial n}}}=0 \phantom{eeeeeeeeeeeee} \text{on }\partial\mathcal J
\label{eq:bord}$$ where $\Delta {{\Gamma_{\text{\tiny HS}}}}$ is the Laplacian of ${{\Gamma_{\text{\tiny HS}}}}$ on the Riemannian sphere $\mathbb S^2$ and $\partial\mathcal J$ is the boundary of $\mathcal J$, assumed to be piece-wise smooth and with unit normal vector $n$.
In pinhole coordinates $(z_1,z_2)$, we have $$\begin{aligned}
&d\sigma_{\eta}=\left(1+z_{1}^2+z_{2}^2\right)^{-3/2}dz_{1}dz_{2}
\\
&\nabla\Upsilon^2= \left(1+z_{1}^2+z_{2}^2\right)\left({{\frac{\partial \Upsilon}{\partial z_{1}}}}^2+{{\frac{\partial \Upsilon}{\partial z_{2}}}}^2+(z_1{{\frac{\partial \Upsilon}{\partial z_{1}}}}+z_2{{\frac{\partial \Upsilon}{\partial z_{2}}}})^2\right)\\
&\left({{\frac{\partial y}{\partial t}}} +\nabla y \cdot\left(\eta \times \left(\omega +\Upsilon\eta\times v\right)\right)\right)^2
=(F+\Upsilon G)^2\end{aligned}$$
where $$\begin{split}
&F={{\frac{\partial y}{\partial t}}}+f_{1}(z,\omega){{\frac{\partial y}{\partial z_1}}}+f_{2}(z,\omega){{\frac{\partial y}{\partial z_2}}}\\
&G=g_{1}(z,v){{\frac{\partial y}{\partial z_1}}}+g_{2}(z,v){{\frac{\partial y}{\partial z_2}}}.
\label{eq:F1F2}
\end{split}$$ Consequently, the first order stationary condition reads in $(z_1,z_2)$ coordinates: $$\begin{split}
{{\Gamma_{\text{\tiny HS}}}}G^2+ FG=&\alpha^2\left[\frac{\partial}{z_1}\left(\frac{1+z_1^2}{\sqrt{1+z_{1}^2+z_{2}^2}}{{\frac{\partial {{\Gamma_{\text{\tiny HS}}}}}{\partial z_1}}}\right)
\right.\\
&+\frac{\partial}{z_2}\left(\frac{1+z_2^2}{\sqrt{1+z_{1}^2+z_{2}^2}}{{\frac{\partial {{\Gamma_{\text{\tiny HS}}}}}{\partial z_2}}}\right)\\
&+\frac{\partial}{z_2}\left(\frac{z_1z_2}{\sqrt{1+z_{1}^2+z_{2}^2}}{{\frac{\partial {{\Gamma_{\text{\tiny HS}}}}}{\partial z_1}}}\right)\\
&\left. +\frac{\partial}{z_1}\left(\frac{z_1z_2}{\sqrt{1+z_{1}^2+z_{2}^2}}{{\frac{\partial {{\Gamma_{\text{\tiny HS}}}}}{\partial z_2}}}\right)\right]
\end{split}
\label{eq:intrinsicEDP}$$ on the rectangular domain $\mathcal I=[-\bar z_1,\bar z_1]\times [-\bar z_2,\bar z_2]$ ($\bar z_1,\bar z_2>0$ with $\bar z_1^2+\bar z_2^2 < 1$).
The right term of corresponds to the Laplacian operator on the Riemannian sphere ${{\mathbb S}}^2$ in pinhole coordinates. The numerical resolution of this scalar diffusion providing the estimation ${{\Gamma_{\text{\tiny HS}}}}$ of $\Gamma$ is similar to the one used for the Horn-Schunck estimation ${{V_{\text{\tiny HS}}}}$ of $V$. The functional $I(W)$ defined in is minimized with respect to two varying parameters $W_1$ and $W_2$ while there is really only one unknown function in this problem: the depth field $\Gamma$. On the contrary, the functional $J(\Upsilon)$ takes full advantage of the knowledge of the camera dynamics since the only varying parameter here is $\Upsilon$.
Depth estimation via asymptotic observers {#sec:Observer}
=========================================
Asymptotic observer based on optical flow measures (${{V_{\text{\tiny HS}}}}$) {#ssec:ObsVHS}
------------------------------------------------------------------------------
From any optical flow estimation, such as ${{V_{\text{\tiny HS}}}}$, it is reasonable to assume that we have access for each time $t$, to the components in pinhole coordinates of the vector field $$\begin{aligned}
\label{eq:varpi}
\varpi_t:~ {{\mathbb S}}^2\ni \eta \mapsto \varpi_t(\eta)= \eta \times (\omega +\Gamma\eta\times v)\in T_\eta {{\mathbb S}}^2\end{aligned}$$ appearing in . This vector field can be considered as a measured output for , expressed as $\varpi_t(\eta)=f_t(\eta)+\Gamma(t,\eta) g_t(\eta)$, where $f_t$ and $g_t$ are the vector fields $$\begin{aligned}
&f_t:~ {{\mathbb S}}^2\ni \eta \mapsto f_t(\eta)= \eta \times \omega \in T_\eta {{\mathbb S}}^2\label{eq:f}
\\
&g_t:~ {{\mathbb S}}^2\ni \eta \mapsto g_t(\eta)= \eta \times (\eta\times v) \in T_\eta {{\mathbb S}}^2\label{eq:g}.\end{aligned}$$ This enables us to propose the following asymptotic observer for $D=1/\Gamma$ since it obeys to ${{\frac{\partial D}{\partial t}}} =- \nabla D \cdot\varpi_t - v\cdot\eta$: $$\label{eq:obserIntrinsic}
{{\frac{\partial {{\widehat{D}}}}{\partial t}}} =- \nabla {{\widehat{D}}}\cdot\varpi_t - v\cdot\eta + k g_t\cdot( {{\widehat{D}}}f_t+ g_t -{{\widehat{D}}}\varpi_t)$$ where $\varpi_t$, $f_t$ and $g_t$ are known time-varying vector fields on ${{\mathbb S}}^2$ and $k>0$ is a tuning parameter. This observer is trivially $SO(3)$ invariant and reads in pinhole coordinates: $$\begin{split}
&{{\frac{\partial {{\widehat{D}}}}{\partial t}}}=-{{\frac{\partial {{\widehat{D}}}}{\partial z_1}}}V_1-{{\frac{\partial {{\widehat{D}}}}{\partial z_2}}}V_2-(z_1v_1+z_2v_2+v_3)\\
&\quad+k \big(g_1({{\widehat{D}}}f_1+g_1-{{\widehat{D}}}V_1)+g_2({{\widehat{D}}}f_2+ g_2-{{\widehat{D}}}V_2) \big)
\end{split}
\label{eq:observerPH}$$ where $V$ is given by any optical flow estimation and $(f_1,f_2,g_1,g_2)$ are defined by .
As assumed in the first paragraphs of subsection \[subsec:PDS\], for each time $t$, there is a one to one smooth mapping between $\eta\in{{\mathbb S}}^2$ attached to the camera pixel and the scene point $M(s)$ corresponding to this pixel. This means that, for any $t\geq 0$, the flow $\phi(t,s)$ defined by $$\label{eq:phi}
\left.{{\frac{\partial \phi}{\partial t}}}\right|_{(t,s)} = \varpi_t(\phi(t,s)),\quad \phi(0,s)=s\in{{\mathbb S}}^2$$ defines a time varying diffeomorphism on ${{\mathbb S}}^2$. Let us denote by $\phi^{-1}$ the inverse diffeomorphism: $\phi(t,\phi^{-1}(t,\eta))\equiv \eta$. Assume that $\Gamma(t,\eta) >0 $, $v(t)$ and $\omega(t)$ are uniformly bounded for $t\geq 0$ and $\eta\in{{\mathbb S}}^2$. This means that the trajectory of the camera center $C(t)$ remains strictly inside the convex surface $\Sigma$ with minimal distance to $\Sigma$. These considerations motivate the assumptions used in the following theorem.
\[thm:error\]
Consider $\Gamma(t,\eta)$ associated to the motion of the camera inside the domain $\Omega$ delimited by the scene $\Sigma$, a $C^1$, convex and closed surface as explained in sub-section \[subsec:PDS\]. Assume that exist $\bar v>0$, $\bar\omega>0$, $\bar\gamma>0$ and $\bar\Gamma>0$ such that $$\forall t\geq 0,~\forall \eta\in{{\mathbb S}}^2,~|v(t)|\leq\bar v,~|\omega(t)|\leq\bar \omega,~\bar\gamma \leq \Gamma(t,\eta) \leq \bar\Gamma.$$ Then, for $t\geq 0$, $\Gamma(t,\eta)$ is a $C^1$ solution of . Consider the observer with a $C^1$ initial condition versus $\eta$, ${{\widehat{D}}}(0,\eta)$. Then we have the following implications:
- $\forall t\geq 0$, the solution ${{\widehat{D}}}(t,\eta)$ of exists, is unique and remains $C^1$ versus $\eta$. Moreover $$t\mapsto \| {{\widehat{D}}}(t,\rule{0.2em}{0.2em})- D(t,\rule{0.2em}{0.2em}) \|_{L^\infty} = \max_{\eta\in{{\mathbb S}}^2} \left|{{\widehat{D}}}(t,\eta) -D(t,\eta) \right|$$ is decreasing ($L^\infty$ stability).
- if additionally for all $s\in{{\mathbb S}}^2$, $\int_0^{+\infty } \|g_\tau(\phi(\tau,s))\|^2 d\tau =+\infty$, then we have for all $p >0$, $$\lim_{t\mapsto +\infty} \int_{{{\mathbb S}}^2} \big|{{\widehat{D}}}(t,\eta)-D(t,\eta)\big|^p d\sigma_\eta = 0$$ (convergence in any $L^p$ topology)
- if additionally there is $\lambda>0$ and $T>0$ such that, for all $t\geq T$ and $s\in{{\mathbb S}}^2$, $\int_0^{t} \|g_\tau(\phi(\tau,s))\|^2 d\tau \geq \lambda t $, then we have, for all $t\geq T$, $$\| {{\widehat{D}}}(t,\rule{0.2em}{0.2em})- D(t,\rule{0.2em}{0.2em}) \|_{L^\infty} \leq
e^{-k\bar\gamma\lambda t}
\| {{\widehat{D}}}(0,\rule{0.2em}{0.2em})- D(0,\rule{0.2em}{0.2em}) \|_{L^\infty}$$ (exponential convergence in $L^\infty$ topology).
Assumptions on $\int \|g_t(\phi(t,s))\|^2 dt $ can be seen as a condition of persistent excitations. It should be satisfied for generic motions of the camera.
The facts that $v$, $\omega$ and $\Gamma$ are bounded and that the scene surface $\Sigma$ is $C^1$, closed and convex, ensure that the mapping $\eta=\phi(t,s)$ and its inverse $s=\phi^{-1}(t,\eta)$ are $C^1$ diffeomorphism on ${{\mathbb S}}^2$ with bounded derivatives versus $s$ and $\eta$ for all time $t>0$. Therefore, $\Gamma$ is also a function of $(t,s)$. Set $\overline{\Gamma}(t,s)=\Gamma(t,\phi(t,s))$: in the $(t,s)$ independent variables the partial differential equation becomes a set of ordinary differential equations indexed by $s$: $\left.{{\frac{\partial \overline{\Gamma}}{\partial t}}}\right|_s=\overline{\Gamma}^2v(t)\cdot\phi(t,s)$ that reads also $ \left.{{\frac{\partial (\overline{D})}{\partial t}}}\right|_s=-v(t)\cdot\phi(t,s)$ with $\overline{D}=1/\overline{\Gamma}$. Thus $\overline{D}(t,s) - \overline{D}(0,s) = -\int_{0}^{t} v(\tau )\cdot\phi(\tau,s)~d\tau$. Consequently, $\overline{\Gamma}$ is $C^1$ versus $s$ and thus $\Gamma$ is $C^1$ versus $\eta$.
Set $\overline{{{\widehat{D}}}}(t,s)={{\widehat{D}}}(t,\phi(t,s))$. Then $$\begin{gathered}
\left.{{\frac{\partial \overline{{{\widehat{D}}}}}{\partial t}}}\right|_s=- v(t)\cdot\phi(t,s)\\+k \|g_t(\phi(t,s))\|^2\overline{\Gamma}(t,s)(\overline{D}(t,s)-\overline{{{\widehat{D}}}}(t,s)).\end{gathered}$$ Set $E={{\widehat{D}}}-D $ and $\overline{E}= \overline{{{\widehat{D}}}}-\overline{D}$. Then $$\label{eq:E}
\left.{{\frac{\partial \overline{E}}{\partial t}}}\right|_s = - k \overline{\Gamma}(t,s) \|g_t(\phi(t,s))\|^2 \overline{E}(t,s)$$ Consequently, $\bar E$ is well defined for any $t>0$ and $C^1$ versus $s$. Thus $E$ and consequently ${{\widehat{D}}}=E+D$ are also well defined for all $t>0$ and are $C^1$ versus $\eta$.
Since for any $s$ and $t_2> t_1 \geq 0$ we have $|\overline{E}(t_2,s)| \leq |\overline{E}(t_1,s)|$, we have also $$|\overline{E}(t_2,s)|\leq \max_{\sigma} |\overline{E}(t_1,\sigma)|= \| {{\widehat{D}}}(t_1,\rule{0.2em}{0.2em})- D(t_1,\rule{0.2em}{0.2em}) \|_{L^\infty}.$$ Thus, taking the max versus $s$, we get $$\| {{\widehat{D}}}(t_2,\rule{0.2em}{0.2em})- D(t_2,\rule{0.2em}{0.2em}) \|_{L^\infty}\leq \| {{\widehat{D}}}(t_1,\rule{0.2em}{0.2em})- D(t_1,\rule{0.2em}{0.2em}) \|_{L^\infty}
.$$
Since $$\overline{E}(t,s) = \overline{E}(0,s) e^{-k \int_0^t \big (\overline{\Gamma}(\tau ,s) \|g_\tau (\phi(\tau,s))\|^2\big) d\tau }$$ we have ($\phi(0,s)\equiv s$). $$|\overline{E}(t,s)|\leq |{E}(0,s)| e^{-k \bar\gamma \int_0^t \|g_\tau (\phi(\tau,s))\|^2~d\tau }
.$$ Take $p>0$. Then $$\int_{{{\mathbb S}}^2} \big|E(t,\eta)\big|^p d\sigma_\eta
=
\int_{{{\mathbb S}}^2} \big|\overline{E}(t,s)\big|^p \det\left({{\frac{\partial \phi}{\partial s}}} (t,s)\right) d\sigma_s
.$$ By assumption ${{\frac{\partial \phi}{\partial s}}}$ is bounded. Thus exists $C>0$ such that $$\|{E}(t,\rule{.2em}{.2em})\|_{L^p}=\int_{{{\mathbb S}}^2} \big|E(t,\eta) \big|^p d\sigma_\eta
\leq
C \int_{{{\mathbb S}}^2} \big|\overline{E}(t,s)\big|^p d\sigma_s
.$$
When $\int_0^{+\infty } \|g_\tau(\phi(\tau,s))\|^2 d\tau =+\infty$, for each $s$ we have $\lim_{t\mapsto +\infty} \overline{E}(t,s)=0$. Moreover $|\overline{E}(t,s)|$ is uniformly bounded by the $L^\infty $ function $E(0,s)$. By Lebesgue dominate convergence theorem $\lim_{t\mapsto +\infty} \|\overline{E}(t,\rule{.2em}{.2em})\|_{L^p}=0$. Previous inequality leads to $\lim_{t\mapsto +\infty}\|{E}(t,\rule{.2em}{.2em})\|_{L^p}=0$.
When, for $t>T$, $\int_0^{T } \|g_\tau(\phi(\tau,s))\|^2 d\tau \geq \lambda t $, we have, for all $s\in{{\mathbb S}}^2$, $|\overline{E}(t,s)|\leq |E(0,s)|e^{-k\bar\gamma\lambda t}$. Thus, for all $s\in{{\mathbb S}}^2$ we get $|\overline{E}(t,s)| \leq \|{E}(0,\rule{.2em}{.2em})\|_{L^\infty}$. Since $\eta\mapsto \phi^{-1}(t,\eta)$ is a diffeomorphism of ${{\mathbb S}}^2$, we get finally, for all $\eta\in{{\mathbb S}}^2$, $|{E}(t,\eta)| \leq \|{E}(0,\rule{.2em}{.2em})\|_{L^\infty}e^{-k\bar\gamma\lambda t}$. This proves $\|{E}(t,\rule{.2em}{.2em})\|_{L^\infty} \leq \|{E}(0,\rule{.2em}{.2em})\|_{L^\infty}e^{-k\bar\gamma\lambda t}$.
Asymptotic observer based on rough depth estimation (${{\Gamma_{\text{\tiny HS}}}}$) {#ssec:ObsGHS}
------------------------------------------------------------------------------------
Instead of relying the observer on estimation ${{V_{\text{\tiny HS}}}}$, we can base it on ${{\Gamma_{\text{\tiny HS}}}}$. Then becomes ($k>0$) $$\label{eq:obserIntrinsic_Bis}
{{\frac{\partial {{\widehat{D}}}}{\partial t}}} =- \nabla {{\widehat{D}}}\cdot(f_t+{{\Gamma_{\text{\tiny HS}}}}g_t)- v\cdot\eta + k ( 1 -{{\widehat{D}}}{{\Gamma_{\text{\tiny HS}}}})$$ that reads in pinhole coordinates $$\begin{gathered}
{{\frac{\partial {{\widehat{D}}}}{\partial t}}}=-{{\frac{\partial {{\widehat{D}}}}{\partial z_1}}}(f_1+{{\Gamma_{\text{\tiny HS}}}}g_1)-{{\frac{\partial {{\widehat{D}}}}{\partial z_2}}}(f_2+{{\Gamma_{\text{\tiny HS}}}}g_2)\\
-(z_1v_1+z_2v_2+v_3)+ k (1-{{\widehat{D}}}{{\Gamma_{\text{\tiny HS}}}})
.
\label{eq:observerPH_Bis}\end{gathered}$$ For this observer we have the following convergence result.
\[thm:error\_Bis\] Take assumptions of theorem \[thm:error\] concerning the scene surface $\Sigma$, $\Gamma=1/D$, $v$ and $\omega$. Consider the observer where ${{\Gamma_{\text{\tiny HS}}}}$ coincides with $\Gamma$ and where the initial condition is $C^1$ versus $\eta$. Then $\forall t\geq 0$, the solution ${{\widehat{D}}}(t,\eta)$ of exists, is unique, remains $C^1$ versus $\eta$ and $$\| {{\widehat{D}}}(t,\rule{0.2em}{0.2em})- D(t,\rule{0.2em}{0.2em}) \|_{L^\infty} \leq
e^{-k\bar\gamma t}
\| {{\widehat{D}}}(0,\rule{0.2em}{0.2em})- D(0,\rule{0.2em}{0.2em}) \|_{L^\infty}$$ (exponential convergence in $L^\infty$ topology)
The proof, similar to the one of theorem \[thm:error\], is left to the reader.
Simulations and numerical implementations {#sec:implement}
=========================================
Sequence of synthetic images and method of comparison {#ssec:sequence}
-----------------------------------------------------
The non-linear asymptotic observers described in section \[sec:Observer\] are tested on a sequence of synthetic images characterized by the following:
- *virtual camera*: the size of each image is 640 by 480 pixels, the frame rate of the sequence is 60 Hz and the field of view is 50 deg by 40 deg;
- *motion of the virtual camera*: it consists of two combined translations in a vertical plane ($v_3=\omega_1=\omega_2=\omega_3=0$), and the velocity profiles are sinusoids with magnitude 1 $m.s^{-1}$, and different pulsations ($\pi$ for $v_1$ and $3\pi$ for $v_2$);
- *virtual scene*: it consists of a 4 $m^2$-plane placed at 3 m and tipped of an angle of 0.3 rad with respect to the plane of camera motion; the observed plane is virtually painted with a gray pattern, whose intensity varies in horizontal and vertical directions as a sinusoid function;
- *generation of the images*: each pixel of an image has an integer value varying from $1$ to $256$, directly depending on the intensity of the observed surface in the direction indexed by the pixel, to which a normally distributed noise varying with mean 0 and standard deviation $\sigma$ is added.
The virtual setup used to generate the sequence of images is represented in Fig.\[fig:synthetic\_film\_setup\].
![Virtual setup used to generate the sequence of images processed in \[sec:implement\][]{data-label="fig:synthetic_film_setup"}](synthetic_film_setup){width="70.00000%" height="70.00000%"}
To compare the performances of both methods, we use the global error rate in the estimation of $D$, defined as $$E=\int_{ \mathcal I}\tfrac{{\lvert{{\widehat{D}}}(t,\eta)-D(t,\eta)\rvert}}{D(t,\eta)} d\sigma_\eta~/ \int_{ \mathcal I} d\sigma_\eta
\label{eq:Error}$$ where $D$ is the true value of the depth field, ${{\widehat{D}}}$ is the estimation computed by any of the proposed methods and $\mathcal I$ is the image frame.
Implementation of the depth estimation based on optical flow measures (${{V_{\text{\tiny HS}}}}$) {#ssec:Obs1}
-------------------------------------------------------------------------------------------------
We test on the sequence described in \[ssec:sequence\] the depth estimation characterized by the partial differential equation . The optical flow input ${{V_{\text{\tiny HS}}}}$ ($V_1$ and $V_2$ components) is computed by a classical Horn-Schunck method. Note that convergence theorem \[thm:error\] assumes that the domain of definition of the image was the entire unit sphere ${{\mathbb S}}^2$. Here the field of view of our virtual camera limits this domain to a portion of the sphere $\mathcal I$. However, the motion of our virtual camera ensures that most of the points of the scene appearing in the first image stay in the field of view of the camera during the whole sequence. The convergence of the method is only ensured for these points, and Neumann boundary conditions are chosen at the borders where optical flow points toward the inside of the image: $${{\frac{\partial {{\widehat{D}}}}{\partial n}}}=0 \text{ if } n \cdot {{V_{\text{\tiny HS}}}}<0$$ The observer gain $k>0$ is chosen in accordance with scaling considerations. Setting $k=500$ $s.m^{-2}$ provides a rapid convergence rate: we see on Fig. \[fig:errObs1sigma1total\] that after a few frames, the initial relative error (blue curve) is reduced by $1/3$. Setting $k=50$ $s.m^{-2}$ is more reasonable when dealing with noisy data: on Fig. \[fig:errObs1sigma20total\] initial relative error is reduced by $1/3$ after around $20$ frames.
More precisely, the standard deviation $\sigma$ of the noise added to the synthetic sequence of images is 1. The gains $k=500$ and $k=100$ are successively tested, and the associated error rates for ${{\widehat{D}}}$ are plotted in Fig. \[fig:errObs1sigma1total\]. As expected, the convergence is more rapid for a larger gain but at convergence, i.e., after 40 frames, the relative errors are similar and below 1.5%.
![Relative estimation errors of the depth field $D$ estimated by the asymptotic observer filtering the optical flow input ${{V_{\text{\tiny HS}}}}$ obtained by Horn-Schunck method, for different correction gains k. The noise corrupting the image data is normally distributed, with mean $\mu=0$ and standard deviation $\sigma=1$.[]{data-label="fig:errObs1sigma1total"}](errObs1sigma1total){width="\textwidth"}
To test robustness when dealing with noisy data, the standard deviation $\sigma$ is magnified by 20. The correction gain is tuned to $k=50$ $s.m^{-2}$. The converged errors after $40$ frames significantly increases and yet stays between 12 and 14 %. Note that such permanent errors can not decrease since such noise level first affects the optical flow estimation ${{V_{\text{\tiny HS}}}}$ that feeds the asymptotic observer. Compared to its true value $V$, the error level for ${{V_{\text{\tiny HS}}}}$ is about 15 %. These results underline the fact that this approach is sensitive to input optical flow measures, but not directly to noise corrupting the image data.
![Relative estimation errors of the depth field $D$ estimated by the asymptotic observer filtering the optical flow input ${{V_{\text{\tiny HS}}}}$ obtained by Horn-Schunck method.The noise corrupting the image data is normally distributed, with mean $\mu=0$ and standard deviation $\sigma=20$.[]{data-label="fig:errObs1sigma20total"}](Obs1sigma20total){width="\textwidth"}
Implementation of the asymptotic observer based on rough depth estimation (${{\Gamma_{\text{\tiny HS}}}}$) {#ssec:Obs2}
----------------------------------------------------------------------------------------------------------
Subsequently, the observer described by was applied to the same sequence. The input depth field ${{\Gamma_{\text{\tiny HS}}}}$ is obtained as the output of . To adapt the numerical method to this model, we make the small-angle approximation by neglecting the second order terms $z$ (we neglect the curvature of ${{\mathbb S}}^2$ and consider that the camera image corresponds to a small part of ${{\mathbb S}}^2$ that can be approximated by a small Euclidean rectangle; the error of this approximation is smaller than 3% for such rectangular image): becomes $$G^2\Gamma+FG=\alpha^{2}\left( {{\frac{\partial^2 \Gamma}{\partial {z_1}^2}}} + {{\frac{\partial^2 \Gamma}{\partial {z_2}^2}}} \right)
\label{eq:CVb}$$ $F$ and $G$ are computed using angular and linear velocities, $\omega$ and $v$, and differentiation (Sobel) filters directly applied on the image data $y(t,z_1,z_2)$ . $\Delta \Gamma$ is approximated by the difference between the weighted mean $\bar \Gamma$ of $\Gamma$ on the neighboring pixels and its value at the current pixel. The resulting linear system in $\Gamma$ is solved by the Jacobi iterative scheme, with an initialization provided by the previous estimation. The regularization parameter $\alpha$ is chosen accordingly to scaling considerations and taking into account the magnitude of noise: $\alpha=300$ $m.s^{-1}$ provides a convergence in about 5 or 6 frames for relatively clean image data. As for the observer , the correction gain $k=50$ $s.m^{-1}$ enables a convergence in around 20 frames.
As in section \[ssec:Obs1\], we test the observer for different levels of noise corrupting the image data. For $\sigma=1$, the error rates associated to the input depth ${{\Gamma_{\text{\tiny HS}}}}$ and to the estimated depth ${{\widehat{D}}}$ are plotted in Fig. \[fig:errObs2sigma1total\]. After only 6 images of the sequence, the error rate for ${{\Gamma_{\text{\tiny HS}}}}$ is smaller than 4% and stays below this upper bound for the rest of the sequence. On the downside, the error rate stays larger than 2.5%. On the contrary, the error rate associated to ${{\widehat{D}}}$ keeps decreasing, and reaches the minimal value of 0.5 %.
![Relative estimation error of ${{\Gamma_{\text{\tiny HS}}}}$ (blue), using the depth estimation inspired by Horn-Schunck, described in \[ssec:HSDepth\] and of $D$ (red) estimated by the asymptotic observer filtering ${{\Gamma_{\text{\tiny HS}}}}$.The noise corrupting the image data is normally distributed, with mean $\mu=0$ and standard deviation $\sigma=1$.[]{data-label="fig:errObs2sigma1total"}](errObs2sigma1total){width="\textwidth"}
For $\sigma=20$, the error rates associated to the input depth ${{\Gamma_{\text{\tiny HS}}}}$ and to the estimated depth ${{\widehat{D}}}$ are plotted in Fig. \[fig:errObs2sigma20total\]. For the computation of ${{\Gamma_{\text{\tiny HS}}}}$, the diffusion parameter $\alpha$ is increased to $1000$ $m.s^{-1}$ to take into account such stronger noise. The observer filters the error associated to ${{\Gamma_{\text{\tiny HS}}}}$ (between 4 and 8 %) to provide a 3 % accuracy. The results show a good robustness to noise for this observer.
![Relative estimation error of ${{\Gamma_{\text{\tiny HS}}}}$ (blue), using the depth estimation inspired by Horn-Schunck, described in \[ssec:HSDepth\] and of $D$ (red) estimated by the asymptotic observer filtering ${{\Gamma_{\text{\tiny HS}}}}$.The noise corrupting the image data is normally distributed, with mean $\mu=0$ and standard deviation $\sigma=20$.[]{data-label="fig:errObs2sigma20total"}](errObs2sigma20total){width="\textwidth"}
Experiment on real data
-----------------------
To realize the experiments, a camera was fixed on a motorized trolley traveling back and forth on a 2 meter-linear track in about 6 seconds. The resolution of the encoder of the motor enables to know the position and the speed of the trolley with a micrometric precision. The camera is a Flea2 - Point Grey Research VGA video cameras (640 by 480 pixels) acquiring data at 20.83 fps, with a Cinegon 1.8/4.8 C-mount lens, with an angular field of view of approximately 50 by 40 deg, and oriented orthogonally to the track. The scene is a static work environment, with desks, tables, chairs, lamps, lit up by electric light plugged on the mains, with frequency 50 Hz. The acquisition frame rate of the cameras produces an aliasing phenomenon on the video data at 4.17 Hz. In other words, the light intensity in the room is variable, at a frequency that can not be easily ignored, which does not comply with the initial hypothesis . However, the impact of this temporal dependence in the equations can be reduced by a normalization of the intensity of the images such as $
y(x,y)=\frac{y(x,y)}{\bar{y}}
$ where $x$ and $y$ are the horizontal and vertical indexes of the pixels in the image, $y(x,y)$ is the intensity of this pixel and $\bar{y}$ is the mean intensity on the entire image.
The depth field was estimated via the asymptotic observer based on optical flow measures. The components of the optical flow were computed using a high quality algorithm based on TV-$L^1$ method (see [@TVL1] for more details). The correction gain was tuned to $k=100$ $s.m^{-2}$. An example of image data is shown in Fig.\[fig:film\], and the depth estimate associated to that image at the same time $t*$ is shown in Fig.\[fig:depth\_film\]. At that specific time $t^\star\approx 8$ s, the trolley already traveled once along the track and is on its way back toward its starting point. Some specific estimates are extracted from the whole depth field (two tables, two chairs, a screen, two walls) and highlighted in black; they are compared to real measures taken in the experimental room (in red): the estimate depth profile ${{\widehat{D}}}(t^\star,\cdot)$ exhibits a strong correlation with these seven punctual reference values of $D(t^\star,\cdot)$; the global appearance of the depth field looks very realistic.
![The static scene as seen by the camera[]{data-label="fig:film"}](film){width="\textwidth"}
![Estimation of the depth field associated with the image shown in Fig.\[fig:film\]. Depth is associated to a gray level, whose scale in meters is on the right. Some estimates are extracted from the entire field (in black) and compared to real measures (in red).[]{data-label="fig:depth_film"}](depth_film_annote){width="\textwidth"}
Conclusions and future works
============================
In section \[sec:probstat\], we recalled a system of partial differential equations, describing the invariant dynamics of brightness and depth smooth fields under the assumptions of a static and Lambertian environment. We proposed in section \[sec:VarMeth\] an adaptation of optical flow algorithms that take the best advantage of the $SO(3)$-invariance of these equations and the knowledge of camera dynamics: it yielded an $SO(3)$-invariant variational method to directly estimate the depth field. In section \[sec:Observer\], we proposed two asymptotic observers, respectively based on optical flow and on depth estimates. We proved their convergence under geometric and persistent excitation assumptions. On synthetic images, we showed in section \[sec:implement\] that the variational method converges rapidly, but its performance is highly dependent of the noise level whereas both asymptotic observers filter this noise. These asymptotic obersevers based on image processing of the entire field of view of the camera seems to be an interesting tool to dense range estimation and a complement to methods based on feature tracking.
[10]{} \[1\][\#1]{} url@samestyle \[2\][\#2]{} \[2\][[l@\#1=l@\#1\#2]{}]{}
R. Smith, M. Self, and P. Cheeseman, *Estimating uncertain spatial relationships in robotics*.1em plus 0.5em minus 0.4emNew York, NY, USA: Springer-Verlag New York, Inc., 1990, pp. 167–193.
J. Civera, A. Davison, and J. Montiel, “Inverse depth parametrization for monocular slam,” *Robotics, IEEE Transactions on*, 2008.
M. Montemerlo, S. Thrun, D. Koller, and B. Wegbreit, “[FastSLAM]{} 2.0: An improved particle filtering algorithm for simultaneous localization and mapping that provably converges,” in *International Joint Conference on Artificial Intelligence IJCAI*, 2003.
H. Strasdat, J. M. M. Montiel, and A. J. Davison, “Real-time monocular slam: Why filter?” in *ICRA*, 2010, pp. 2657–2664.
H. Strasdat, J. M. M. Montiel, and A. Davison, “Scale drift-aware large scale monocular slam,” in *Proceedings of Robotics: Science and Systems*, Zaragoza, Spain, June 2010.
R. Abdursul, H. Inaba, and B. K. Ghosh, “Nonlinear observers for perspective time-invariant linear systems,” *Automatica*.
O. Dahl, Y. Wang, A. F. Lynch, and A. Heyden, “Observer forms for perspective systems,” *Automatica*.
A. Heyden and O. Dahl, “Provably convergent on-line structure and motion estimation for perspective systems,” in *Computer Vision Workshops, 2009 IEEE 12th International Conference on*, 2009.
X. Chen and H. Kano, “A new state observer for perspective systems,” *Automatic Control, IEEE Transactions on*, 2002.
W. Dixon, Y. Fang, D. Dawson, and T. Flynn, “Range identification for perspective vision systems,” *Automatic Control, IEEE Transactions on*, vol. 48, no. 12, pp. 2232 – 2238, dec. 2003.
D. Karagiannis and A. Astolfi, “A new solution to the problem of range identification in perspective vision systems,” *Automatic Control, IEEE Transactions on*, vol. 50, no. 12, pp. 2074 – 2077, dec. 2005.
A. D. Luca, G. Oriolo, and P. R. Giordano, “Feature depth observation for image-based visual servoing: Theory and experiments.” *I. J. Robotic Res.*, pp. 1093–1116, 2008.
M. Sassano, D. Carnevale, and A. Astolfi, “Observer design for range and orientation identification,” *Automatica*.
L. Matthies, T. Kanade, and R. Szeliski, “Kalman filter-based algorithms for estimating depth from image sequences,” *International Journal of Computer Vision*, vol. 3, no. 3, pp. 209–238, 1989.
C. Hoilund, T. Moeslund, C. Madsen, and M. Trivedi, “Improving stereo camera depth measurements and benefiting from intermediate results,” in *IEEE Intelligent Vehicles Symposium*, 2010, pp. 935 –940.
B. Herisse, S. Oustrieres, T. Hamel, R. E. Mahony, and F.-X. Russotto, “A general optical flow based terrain-following strategy for a vtol uav using multiple views,” in *ICRA*, 2010, pp. 3341–3348.
S. Bonnabel and P. Rouchon, “Fusion of inertial and visual : a geometrical observer-based approach,” *2nd Mediterranean Conference on Intelligent Systems and Automation (CISA’09)*, 2009.
A. Censi, S. Han, S. B. Fuller, and R. M. Murray, “A bio-plausible design for visual attitude stabilization,” in *CDC*, 2009, pp. 3513–3520.
B. Horn and B. Schunck, “Determining optical flow,” *Artificial Intelligence*, vol. 17, pp. 185–203, 1981.
A. Mitiche and A. reza Mansouri, “On convergence of the horn and schunck optical-flow estimation method,” *IEEE Transactions on Image Processing*, vol. 13, pp. 848–852, 2004.
A. Chambolle and T. Pock, “A first-order primal-dual algorithm for convex problems with applications to imaging,” *Journal of Mathematical Imaging and Vision*, pp. 1–26, 2010.
[^1]: Nadège Zarrouati is with DGA, 7-9 rue des Mathurins, 92220 Bagneux, France, PHD candidate at Mines-ParisTech, Centre Automatique et Systèmes, Unité Mathématiques et Systèmes 60, boulevard Saint-Michel 75272 Paris Cedex, France [[email protected]]{}
[^2]: Emanuel Aldea is with SYSNAV, 57 rue de Montigny, 27200 Vernon, France [[email protected] ]{}
[^3]: Pierre Rouchon is with Mines-ParisTech, Centre Automatique et Systèmes, Unité Mathématiques et Systèmes 60, boulevard Saint-Michel 75272 Paris Cedex, France [[email protected]]{}
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Various models of charged particles interacting with a quantized, ultraviolet cutoff radiation field (but not with each other) are investigated. Upper and lower bounds are found for the self- or ground state-energies without mass renormalization. For $N$ fermions the bounds are proportional to $N$, but for bosons they are sublinear, which implies ‘binding’, and hence that ‘free’ bosons are never free. Both ‘relativistic’ and non-relativistic kinematics are considered. Our bounds are non-perturbative and differ significantly from the predictions of perturbation theory.'
author:
- |
Elliott H. Lieb[^1]\
Departments of Mathematics and Physics\
Princeton University, Princeton, New Jersey 08544-0708\
[*[email protected]*]{}
- |
Michael Loss[^2]\
School of Mathematics\
Georgia Institute of Technology, Atlanta, Georgia 30332-0160\
[*[email protected]*]{}
date: 'September 15, 1999'
title: 'SELF-ENERGY OF ELECTRONS IN NON-PERTURBATIVE QED[^3][^4]'
---
Introduction {#intro}
============
Quantum electrodynamics (QED), the theory of electrons interacting with photons (at least for small energies) is one of the great successes of physics. Among its major achievements is the explanation of the Lamb shift and the anomalous magnetic moment of the electron. Nevertheless, its computations, which are entirely based on perturbation theory, created some uneasiness among the practitioners. The occurrence of infinities was and is especially vexing. Moreover, a truly nontrivial, 3+1-dimensional example of a relativistically invariant field theory has not yet been achieved.
There are, however, unresolved issues at a much earlier stage of QED that hark back to black-body radiation, the simplest and historically first problem involving the interaction of matter with radiation. The conceptual problems stemming from black-body radiation were partly resolved by quantum mechanics, i.e., by the the non-relativistic Schrödinger equation, which is, undoubtedly, one of the most successful of theories, for it describes matter at low energies almost completely. It is mathematically consistent and there are techniques available to compute relevant quantities. Moreover, it allows us to explain certain facts about bulk matter such as its stability, it extensivity, and the existence of thermodynamic functions. What has not been as successful, so far, is the incorporation of radiation phenomena, the very problem quantum mechanics set out to explain.
It ought to be possible to find a mathematically consistent theory, free of infinities, that describes the interaction of non-relativistic matter with radiation at moderate energies, such as atomic binding energies. It should not be necessary, as some physicists believe, to embed QED as a low energy part of a consistent high energy theory.
From such a theory one could learn a number of things that have not been explained rigorously. i) The decay of excited states in atoms. This problem has been investigated in some ultraviolet cutoff models in [@BFS] and in a massive photon model in [@OY]. See also the review of Hogreve [@H]. ii) Non-relativistic QED could be a playground for truly non-perturbative calculations and it could shed light on renormalization procedures. In fact, this was the route historically taken by Kramers that led to the renormalization program of Dyson, Feynman, Schwinger and Tomonaga. iii) Last but not least, one could formulate and answer the problems of stability of bulk matter interacting with the radiation field.
It has been proved in [@F],[@LLS] that stability of non-relativistic matter (with the Pauli Hamiltonian) interacting with classical magnetic fields holds provided that the fine-structure constant, $\alpha = e^2/\hbar c$, is small enough. It is certain, that the intricacies and difficulties of this classical field model will persist and presumably magnify in QED.
The same may be expected from a relativistic QED since replacing the Pauli Hamiltonian by a Dirac operator leads to a similar requirement on $\alpha$ [@LSS]. Indeed, stability of matter in this model (the Brown-Ravenhall model) requires that the electron (positron) be defined in terms of the positive (negative) spectral subspace of the Dirac operator [*with*]{} the magnetic vector potential $A(x)$, instead of the free Dirac operator without $A(x)$. This observation, that perturbation theory, if there is one, must start from the dressed electrons rather than the electrons unclothed by its magnetic field, might ultimately be important in a non-perturbative QED.
The first, humble step is to understand electrons that interact with the radiation field but which are free otherwise. In order for this model to make sense an ultraviolet cutoff has to be imposed that limits the energy of photon modes. The simplest question, which is the one we address in this paper, is the behavior of the self-energy of the electron as the cutoff tends to infinity (with the bare mass of the electron fixed). The self-energy of the electron diverges as the cutoff tends to infinity and it has to be subtracted for each electron in any interacting theory. The total energy will still depend strongly on the cutoff because of the interactions. This dependence will, hopefully, enter through an effective mass which will be set equal to the physical mass (mass renormalization). The resulting theory should be essentially Schrödinger’s mechanics, but slightly modified by so-called radiative corrections.
Lest the reader think that the self-energy problem is just a mathematical exercise, consideration of the many-body problem will provide a counterexample. Imagine $N$ charged bosons interacting with the radiation field, but neglect any interaction among them such as the Coulomb repulsion. We say that these particles bind if the energy of the combined particles is less than the energy of infinitely separated particles. As we shall show, charged bosons indeed bind and they do it in such a massive way that it will be very likely that this cannot be overcome by the Coulomb repulsion. In particular, the energy of a charged many-boson system is [*not*]{} extensive, and from this perspective it is fortunate that stable, charged bosons do not exist in nature.
The situation is very different for fermions. We are not able to show that they do not bind but we can show — and this is one of the main results of our paper — that the self-energy is extensive, i.e., bounded above and below by a constant times $N$.
We thus have strong evidence that there is [*no*]{} consistent description of a system of stable charged relativistic or non-relativistic bosons interacting with the radiation field, while the Pauli exclusion principle, on the other hand, is able to prevent the above mentioned pathology.
In the remainder of the section we explain our notation and state the results. In the subsequent sections we sketch the proof of some of them but for details we refer the reader to \[LL\].
We measure the energy in units of $mc^2$ where $m$ is the bare mass of the electron, the length in units of the Compton wave length $ \ell_C = \hbar/mc$ of the bare electron. We further choose $\ell_C^{-1} \sqrt{\hbar c}$ as the unit for the vector potential $A$ and $\ell_C^{-2} \sqrt{\hbar c}$ as the unit for the magnetic field $B$. The argument is the dimensionless quantity $ \ell_C^{-1} x$. As usual, $\alpha = e^2 / \hbar c \approx 1/137.04$ is the fine structure constant.
In the expression below, $A(x)$ denotes an ultraviolet cutoff radiation field localised in a box $L\times L\times L$ with volume $V=L^3$,
$$A(x)=\frac {1 }{ \sqrt{2V}} \sum_{|k| < \Lambda} \sum_ {\lambda =1,2} \frac{1}{
\sqrt{|k|}}
\varepsilon_{\lambda}(k)\bigl[ a_{\lambda}(k) e^{ i x\cdot k} +
a_{\lambda}^*(k) e^{- i x\cdot k} \bigr] \ .$$
\[beans\] The index $k=2 \pi n/L$ where $n \in \mathbb{Z}^3$, and the word cutoff refers to the restriction to all values of $k$ with $|k| < \Lambda$.
The vectors $\varepsilon_{\lambda}(k)$ are the polarization vectors and are normalized in such a way that $$\varepsilon_{i}(k)\cdot \varepsilon_{j}(k) = \delta_{i,j}\ ,\
\varepsilon_{i}(k)\cdot k = 0 \ .$$
The operators $a_{\lambda}(k)$ and $a_{\lambda}^{*}(k)$ satisfy the commutation relation $$\bigl[a_{\lambda}(k), a_{\lambda^{\prime}}^{*}(k^{\prime})\bigr]=
\delta_{\lambda,\lambda^{\prime}} \delta(k,k^{\prime}) \ ,$$ while all others commute with each other.
The energy of the radiation field can now be conveniently written as $$H_f = \sum_ {|k| < \Lambda} \sum_{ \lambda =1,2}|k| a_{\lambda}^{*}(k)
a_{\lambda}(k) \ .$$
These operators act on the Hilbert space generated by the polynomials in $a_{\lambda}^{*}(k)$ acting on the vacuum $|0 \rangle$.
The self energy of (one or more) particles is the [**ground state energy**]{} of the Hamiltonian $$H = {\rm kinetic \ energy} \ + \ H_f \ .$$ where, as usual, the ground state energy of $H$ is defined to be $$E_0 = \inf_{\Psi} \frac{ \langle \Psi, H\ \Psi \rangle }{\langle \Psi, \Psi \rangle}\
.$$
Typically, in the inquiry about the self–energy problem, i.e., the problem of computing the self–energy for fixed, albeit small, $\alpha$ and for large $\Lambda$, one proceeds via perturbation theory. First order perturbation theory will predict an energy of the order of $\alpha \Lambda ^2$, and a higher order power counting argument confirms the asymptotically large $\Lambda$ dependence of that calculation. Our theorems below show that the predictions of perturbation theory for the self–energy problem are wrong, if one is interested in the large $\Lambda$ asymptotics of the energy. If perturbation theory works at all, then it works only for a range of $\alpha$ that vanishes as $\Lambda$ increases. In fact we deduce from the upper bound in Theorem \[theo:nonrel\] that the size of this range shrinks at least as $\Lambda^{-2/5}$.
All the theorems below are asymptotic statements for large $\Lambda$ and for fixed $\alpha$. For actual bounds we refer the reader to \[LL\]. The first result concerns the self energy of a nonrelativistic electron interacting with the radiation field. The Hamiltonian is given by
$$H= \frac{1}{2} (p+ \sqrt{\alpha} A(x))^2 +H_{f} \ ,
\label{bacon}$$
where $p =-i\nabla$ and acts on $L^2(\mathbb{R}^3)\otimes {\mathcal F}$, where ${\mathcal F}$ denotes the photon Fock space.
The ground state energy, $E_0$, of the operator (\[bacon\]) satisfies the bounds $$C_1 \alpha^{1/2} \Lambda^{3/2} \ < \ E_0 \ < \ C_2
\alpha^{2/7} \Lambda^{12/7}
\label{eq:nonrel}$$ \[theo:nonrel\]
We do not know how to get upper and lower bounds that are of the same order in $\Lambda$, but we suspect that $\Lambda^{12/7}$ is the right exponent. This is supported by the following theorem in which the $p\cdot A$ term is omitted.
The ground state energy $E_0$ of the operator $$\frac{1}{2} \left[ p^2 + \alpha A(x)^2 \right] + H_f
\label{eq:2.28}$$ satisfies the bounds $$C_1 \alpha^{2/7} \Lambda ^{12/7}
\leq E_0 \leq C_2 \alpha^{2/7} \Lambda ^{12/7}
\label{eq:2.29}$$ \[theo:2.3\]
While these results are not of direct physical relevance (since $E_0$ is not observable), the many-body problem is of importance since it reveals a dramtic difference between bosons and fermions.
The ground state energy of $N$ bosons, $E^{boson}_0(N)$, with Hamiltonian $$H(N)=\sum_{j=1}^N \frac{1}{ 2}(p_j +\sqrt \alpha A(x_j))^2
+ H_f\
\label{manyham}$$ satisfies the bounds $$C_1 \sqrt{N}
\sqrt{\alpha} \Lambda^{3/2} \leq
E^{boson}_0(N) \leq
C_2 N^{5/7}\alpha ^{2/7} \Lambda^{12/7}
\label{eq:nonrelboson}$$ \[theo:nonrelboson\]
Thus, the energy $E^{boson}_0(N)$ is not extensive, i.e., it costs a huge energy to separate bosons. This has to be contrasted with the next theorem about fermions. The Hamiltonian is the same as before but it acts on the Hilbert space $${\mathcal F}\otimes
\wedge_{j=1}^N L^2(\mathbb{R}^3; \mathbb{C}^2) ~ ,$$ where the wedge product indicates the antisymmetric tensor product is taken.
The ground state energy, $E^{fermion}_0(N)$, of $N$ charged fermions interacting with the radiation field satisfies $$C_1 \alpha^{1/2} \Lambda^{3/2} N \ \leq \ E^{fermion}_0(N) \ \leq \ C_2
\alpha^{2/7} \Lambda^{12/7} N
\label{eq:nonrelfer}$$ \[theo:nonrelfer\]
The [**“relativistic” kinetic energy**]{} for an electron is $$T^{\rm rel} = | p + \sqrt{\alpha} A (x) | = \sqrt{[p + \sqrt{\alpha} A(x)]^2}
\label{eq:4.1}$$ with $p=-i\nabla$. (Really, we should take $\sqrt{[p+\sqrt{\alpha}A
(x)]^2 + 1}$, but since $x < \sqrt{x^2 + 1} < x + 1$, the difference is bounded by $N$.)
Consider, first, the $N=1$ body problem with the Hamiltonian $$H =T^{\rm rel} + H_f ~.
\label{eq:4.2}$$
By simple length scaling (with a simultaneous scaling of the volume $V$) we easily see that $E_0 = \inf {\rm{spec}}~ (H) = C \Lambda$. Our goal here is to show that the constant, $C$, is strictly positive and to give an effective lower bound for it. But we would like to do more, namely investigate the dependence of this constant on $\alpha$. We also want to show, later on, that for $N$ fermions the energy is bounded below by a positive constant times $N \Lambda$. Our proof will contain some novel — even bizarre — features.
For the Hamiltonian in \[eq:4.2\] there are positive constants, $C, C',
C''$ such that $$\begin{aligned}
E_0 &\leq& C \sqrt{\alpha} \Lambda\\
E_0 &\geq& C' \sqrt{\alpha} \Lambda~{\rm{for~ small}}~ \alpha\\
E_0 &\geq& C'' \Lambda~{\rm{for ~ large}}~ \alpha ~ .\end{aligned}$$ \[theo:4.1\]
The generalization of this to $N$ fermions is similar to the nonrelativistic generalization, except that the power of $\Lambda$ is the same on both sides of the inequalities.
For $N$ fermions with Hamiltonian $$H_N = \sum_{i=1}^N T^{\rm rel}(x_i) + H_f$$ there are positive constants $C, C', C'',$ independent of $\alpha$ and $N$, such that $$\begin{aligned}
E_0 &\leq& C N \sqrt{\alpha} \Lambda\nonumber \\ E_0 &\geq& C' N \sqrt{\alpha} \Lambda \quad {\rm{for~small}} ~ \alpha\nonumber
\\E_0 &\geq& C'' N \Lambda \quad {\rm{for~large}} ~ \alpha
\label{eq:4.23}\end{aligned}$$ \[theo:4.3\]
We close this introduction by mentioning one last result about the Pauli–operator. The kinetic energy expression is given by $$T^{\rm Pauli}=[\sigma \cdot (p+ \sqrt{\alpha}A(x))]^2 = (p+
\sqrt{\alpha}A(x))^2 + \sqrt{\alpha}\ \sigma \cdot B(x) \ .
\label{eq:4.4}$$ where $\sigma$ denotes the vector consisting of the Pauli matrices. Observe that this term automatically accounts for the spin–field interaction. Our result for the self energy of a Pauli electron is the following.
The ground state energy $E_0$ of the Hamiltonian with Pauli energy, $$\frac{1}{2}[\sigma \cdot (p+ \sqrt{\alpha}A(x))]^2 + H_f ,$$ satisfies the bounds $$\begin{aligned}
E_0 &\leq & C_3 \sqrt{\alpha}\Lambda^{3/2} \\
E_0 &\geq & C_1 \alpha \Lambda \quad
{\rm{for~small}} ~ \alpha\nonumber \\
E_0 &\geq & C_2 \alpha^{1/3} \Lambda \quad
{\rm for~large} ~ \alpha \nonumber \end{aligned}$$ For $N$ fermions, the bounds above are multiplied by $N$ (and the constants are changed). \[Pauli\]
For the detail of the proof, we refer the reader to \[LL\]. We believe that the upper bound is closer to the truth since the main contributions to the self energy should come from the fluctuations of the $A^2$ term.
Theorem \[Pauli\] has the following consequence for stability of matter interacting with quantized fields. It was shown in [@LLS] that a system of electrons and nuclei interacting with Coulomb forces, with the Pauli kinetic energy for the electrons and with a [*classical*]{} magnetic field energy is stable (i.e., the ground state energy is bounded below by $N$) if and only if $\alpha$ is small enough. In [@BFG],[@FFG] this result was extended to [*quantized*]{}, ultraviolet cutoff magnetic fields (as here). Among other things, it was shown in [@FFG] that the ground state energy, $E_0$, of the electrons and nuclei problem is bounded below by $-\alpha^2
\Lambda N$ for small $\alpha$. Theorem \[Pauli\] implies, as a corollary, that for small $\alpha$ the [*total energy*]{} (including Coulomb energies) is bounded below by $+\alpha \Lambda N$. In other words, among the three components of energy (kinetic, field and Coulomb), the first two overwhelm the third — for small $\alpha$, at least.
All of these statements are true without mass renormalization and the situation could conceivably be more dramatic when the mass is renormalized. In any case, the true physical questions concern energy differences, and this question remains to be addressed.
NON-RELATIVISTIC ENERGY BOUNDS
==============================
[*Theorem \[theo:nonrel\]:*]{} We sketch a proof of Theorem \[theo:nonrel\]. It is clear by taking the state $V^{-1/2}/ \otimes |0\rangle$ that the ground state energy is bounded above by ${\rm{(const)}} \alpha \Lambda^2$, which is the same result one gets from perturbation theory. Since the field energy in this state vanishes, such a computation ignores the tradeoff between the kinetic energy of the electron and the field energy. Thus, it is important to quantify this tradeoff. The main idea is to estimate the field energy in terms of selected modes.
Consider the operators (field modes), parametrized by $y \in {\mathbb
R}^3$, $$L(y) = \frac{1}{\sqrt{2V}} \sum_{|k| < \Lambda, \lambda} \sqrt{|k|}
a_{\lambda}(k) \overline{v}_{\lambda} (k)e^{i k \cdot y} \ ,
\label{eq:2.4}$$ with some arbitrary complex coefficients $v_{\lambda}(k)$. The following lemma is elementary
$$H_f \geq \int w(y) L^{*}(y) L(y) {\rm d} y\
\label{eq:2.5}$$
provided that the functions $v_{\lambda}(k)$ and $w$ are chosen such that, as matrices, $$|k|\delta_{\lambda^{\prime}, \lambda} \delta(k^{\prime}, k)
\geq \frac{1}{2V} \overline{v}_{\lambda} (k)
\widehat{w}(k-k^{\prime}) v_{\lambda^{'}} (k^{'}) \ ,
\label{eq:2.6}$$ or equivalently, that $$\sum_{|k| < \Lambda, \lambda} \frac{|f_{\lambda}(k)|^2}{|v_ {\lambda} (k)|^2}
\geq \sum_{|k|,|k^{'}| < \Lambda, \lambda, \lambda ^{'}} \frac{1}{2V}
\overline{f_{\lambda}(k)}f_{\lambda^{\prime}} (k^{\prime})
\widehat{w}(k-k^{\prime})
\label{eq:2.7}$$ for all $f_{\lambda}(k)$, where $\widehat{w}(k) = \int e^{ik \cdot x} w(x) {\rm d} x$. \[lemma:2.2\]
For the proof, one simply notes that both sides of (\[eq:2.5\]) are quadratic forms in the creation and annihilation operators, and hence (\[eq:2.6\]) and (\[eq:2.7\]) are necessary and sufficient conditions for (\[eq:2.5\]) to be true.
$$H_f \geq -\frac{1}{2V} \sum_{|k| < \Lambda, \lambda} |k|
|v_{\lambda} (k)|^2 \ \int w(y) {\rm d} y + \frac{1}{4}
\begin{cases}
\int w(y) (L(y) + L^{*}(y))^2 {\rm d} y \\
-\int w(y) (L(y) - \frac{1}{4} L^{*}(y))^2 {\rm d} y
\end{cases}
\label{eq:2.8}$$
\[corollary:1\]
To prove this, note that $$L^{*} L = L L^{*}- \frac{1}{2V} \sum_{|k| < \Lambda , \lambda}
|k| |v_{\lambda} (k)|^2 \ ,
\label{eq:2.9}$$ and, quite generally for operators, $$LL + L^{\ast} L^{\ast} \leq L^{\ast} L + L L^{\ast} \ . ~~~~~~~~~~~~~~~\blacksquare
\label{eq:2.10}$$
Returning to the proof of Theorem \[theo:nonrel\] we start with the lower bound. Denote by $$\Pi(x) = \frac{-i}{\sqrt{2V}}\sum_{|k| < \Lambda , \lambda}
{\sqrt{|k|} \varepsilon_{\lambda}(k) } \left( a_{\lambda}(k)
e^{ik \cdot x} - a^{*}_{\lambda}(k) e^{-i k \cdot x} \right) \ .
\label{eq:2.11}$$ This operator is canonically conjugate to $A(x)$ in the sense that we have the commutation relations $$i[\Pi_i(x) , A_j(x)] = \delta_{i,j} \frac{1}{(3\pi)^2 } \Lambda ^3 \ .
\label{eq:2.12}$$ For our calculation, it is important to note the $${\rm div} ~ \Pi(x) =0 \ .
\label{eq:2.13}$$ Hence from (\[eq:2.12\]) and (\[eq:2.13\]) we get that (for all $j$) $$[p_j + \sqrt{\alpha} A_j(x) , \Pi_j(x)] = \sqrt{\alpha} \frac{i}{(3\pi)^2 } \Lambda ^3 \ .
\label{eq:2.14}$$ The inequality $$\frac{1}{2 } (p+ \sqrt{\alpha}A(x))^2 + 2a^2 \Pi(x)^2 \geq
-ai \sum^{3}_{j=1}[p_j + \sqrt{\alpha} A_j(x) , \Pi_j(x)] \ ,
\label{eq:2.15}$$ valid for all positive numbers $a$, yields $$\frac{1}{2} (p+ \sqrt{\alpha} A(x))^2 + H_f
\geq a \sqrt{\alpha} \frac{1}{(3 \pi)^2} \Lambda ^3 + H_f - 2a^2 \Pi(x)^2 \ .
\label{eq:2.16}$$ Now, with $$v_ {\lambda} (k) = (3 \pi)\frac{\varepsilon_{\lambda}(k)}{\Lambda ^{3/2}}
\label{eq:2.17}$$ and $$w(y) = \delta (x-y) \ ,
\label{eq:2.18}$$ it is elementary to see that (\[eq:2.7\]) is satisfied. Hence Corollary \[corollary:1\] yields $$H_f \geq \frac{9 \pi^2}{4\Lambda ^3} \Pi(x)^2 - \frac{9}{8} \Lambda
\label{eq:2.19}$$ Choosing $a= (3 \pi)/(\sqrt{2} \Lambda^{3/2})$ yields the [*lower bound*]{} $$H \geq \frac{1}{3 \pi}\sqrt{\frac{\alpha}{2}} \Lambda ^{3/2} -
\frac{9}{8}\Lambda
\label{eq:2.20}$$
The idea of using a commutator, as in (\[eq:2.15\]), (\[eq:2.16\]) to estimate the ground state energy, goes back to the study of the polaron [@LY].
For the upper bound we take a simple trial function of the form $$\phi(x) \otimes \Psi
\label{eq:2.21}$$ where $\Psi \in {\mathcal F}$ is normalized and $\phi(x)$ is a [*real*]{} function normalized in $L^2({\mathbb R}^3)$. An upper bound to the energy is thus given by $$\frac{1}{2} \int|\nabla \phi(x)|^2 {\rm d} x +\frac{ \alpha}{2}
\int \phi(x)^2 \left(\Psi, A(x)^2 \Psi\right) {\rm d} x + (\Psi,H_f
\Psi) ~ .
\label{eq:2.22}$$ It is not very difficult to see that the last two terms can be concatenated into the following expression. $$\frac{1}{2}\int \left(\Psi, \left[ \Pi(x)^2 +\alpha A(x) (-\Delta
+\phi(x)^2) A(x)\right] \Psi\right) {\rm d}x ~ -\frac{1}{2} {\rm{Tr}}
\sqrt{P-\Delta P} \ .
\label{eq:2.23}$$ Here, $P$ is the projection onto the divergence free vector fields with ultraviolet cutoff $\Lambda$. This can be deduced by writing the field energy in terms of $\Pi(x)$ and $A(x)$. The first term in (\[eq:2.23\]) is a sum of harmonic oscillators whose zero point energy is given by $$\frac{1}{2} {\rm{Tr}} \sqrt{ P\left(-\Delta + \alpha
\phi(x)^2 \right)P}
\label{eq:2.24}$$ and hence $$\frac{1}{2} {\rm{Tr}} \sqrt{ P\left(-\Delta + \alpha
\phi(x)^2 \right)P} -\frac{1}{2} {\rm{Tr}} \sqrt{P (-\Delta) P}\ ,
\label{eq:2.25}$$ is an exact expression for the ground state energy. Using the operator monotonicity of the square root we get as an upper bound on (\[eq:2.23\]) $$\frac{1}{2} \int|\nabla \phi(x)|^2 {\rm d}x + \frac{1}{2} \sqrt{\alpha}
\ {\rm{Tr}}\sqrt{ P \phi(x)^2 P} \ .
\label{eq:2.26}$$ As a trial function we use the positive function $$\phi(x) = {\rm{const.}} K^{-3/2} \int \left(1 - \frac{|k|}{K} \right)^3_+ e^{ik \cdot
x} {\rm d} x \ .
\label{eq:2.27}$$ Optimizing the resulting expression over $K$ yields the stated result. For details we refer the reader to [@LL].
It is natural to ask, how good this upper bound is. If we neglect the cross terms in $(p+A)^2$, i.e., we replace the kinetic energy by $p^2 +
\alpha A(x)^2$, then we have Theorem \[theo:2.3\], which we prove next.
[*Theorem \[theo:2.3\]:*]{} The upper bound was already given in Theorem \[theo:nonrel\] because $< p\cdot A > = 0$ in the state (\[eq:2.21\]). Loosely speaking equation (\[eq:2.12\]) expresses the Heisenberg uncertainty principle for the field operators. An uncertainty principle that is quite a bit more useful is the following.
The following inequality holds in the sense of quadratic forms $$\Pi(x)^2 \geq \frac{1}{4}\frac{1}{(3\pi)^4}\Lambda^6 \frac{1}{A(x)^2}\ .
\label{eq:2.30}$$ \[lemma:2.4\]
For the proof note that $[A_j(x), A_k(y)] = 0$ and compute $$i[\Pi(x)_j, \frac{A_j(x)}{A(x)^2}]
=\frac{1}{(3\pi)^2}\Lambda^3 \left[\frac{1}{A(x)^2} -
2 \left( \frac{A_j(x)}{A(x)^2}\right)^2 \right] \ ,
\label{eq:2.31}$$ and summing over $j$ we obtain that $$i \sum_{j=1}^3[\Pi(x)_j, \frac{A_j(x)}{A(x)^2}]
=\frac{1 }{(3\pi)^2}\Lambda^3 \frac{1}{A(x)^2}
\label{eq:2.32}$$
Our statement follows from the Schwarz inequality.
To prove Theorem \[theo:2.3\] we return to Lemma \[lemma:2.2\] and choose $v_{\lambda}(k)=
\varepsilon_{\lambda}(k)$ and $w(x)$ any function $\leq 1$. Corollary \[corollary:1\] applied to each of the 3 components of $\Pi(x)$ then yields $$H_f \geq \frac{ 1}{4} \int w(x-y) \Pi (y)^2 {\rm d} y - \Lambda^4
\frac{3}{8 \pi^2} \int w(y) {\rm d} y \ ,
\label{eq:2.33}$$ for every $x \in {\mathbb R}^3$. By Lemma \[lemma:2.2\] the right side is bounded below by $$\Lambda^6 \int w(x-y) \frac{1}{A(y)^2} {\rm d} y -
\Lambda^4\int w(y) {\rm d} y \ ,
\label{eq:2.34}$$ and hence $$\begin{aligned}
\langle \Psi, H \Psi \rangle &\geq& \frac{1}{2} \int \langle \nabla
\Psi(x), \nabla \Psi(x)
\rangle {\rm d} x + \frac{\alpha}{2} \int \langle \Psi(x), A(x)^2
\Psi(x) \rangle {\rm d} x \nonumber \\
&+&\Lambda^6 \int w(x-y) \langle \Psi(y),
\frac{1}{A(x)^2}\Psi(y) \rangle {\rm d}
y {\rm d} x \nonumber \\
&-& \Lambda^4\int w(y) {\rm d} y \int \langle \Psi(x), \Psi(x)
\rangle {\rm d} x \ .
\label{eq:2.35}\end{aligned}$$ By Schwarz’s inequality the second and third term together are bounded below by $$\sqrt{\frac{\alpha}{2}} \Lambda^3\int \langle \Psi(x), \Psi(y)
\rangle \frac{w(x-y)}{\sqrt{\int w(z)
{\rm d}z}} {\rm d} x {\rm d} y \ .
\label{eq:2.36}$$ If we restate our bound in terms of Fourier space variables we get $$\int \left[\frac{|p|}{2}^2 + \sqrt{\frac{\alpha}{2}} ~ \Lambda^3
\frac{\widehat{w}(p)}{\sqrt{ \widehat{w}(0)}}
\right] \langle \widehat{\Psi}(p), \widehat{\Psi}(p)\rangle {\rm d} p
-\Lambda^4 \widehat{w}(0) \int \langle \widehat{\Psi}(p),
\widehat{\Psi}(p) \rangle {\rm d}p \ .
\label{eq:2.37}$$ Choosing the function $\widehat{w}(p)$ to be $(2 \pi)^3
\Lambda^{-18/7}$ times the characteristic function of the ball of radius $\Lambda^{6/7}$, we have that $w(x)
\leq 1$ and it remains to optimize (\[eq:2.37\]) over all normalized states $\widehat{\Psi}(p)$. This is easily achieved by noting that the function $$\frac{1}{2} |p|^2 + \sqrt{\frac{\alpha}{ 2}} \Lambda^3 \frac{\widehat{w}(p)}{\sqrt{ \widehat{w}(0)}}
\label{eq:2.38}$$ is everywhere larger than $\Lambda ^{12/7}$. Strictly speaking, the function $w(x)$ should be positive in order for the argument that led to (\[eq:2.36\]) to be valid. This can be achieved with a different choice of $w(x)$ that is more complicated but does not change the argument in an essential way.
NON-RELATIVISTIC MANY-BODY ENERGIES {#non-relativistic}
===================================
A problem that has to be addressed is the energy of $N$ particles (bosons or fermions) interacting with the radiation field. If $E_0 =
E_0(1)$ is the energy of one particle (which we estimated in the preceding section) then, ideally the energy, $E_0(N)$, of $N$ particles (which trivially satisfies $E_N
\leq NE$, since the $N$ particles can be placed infinitely far apart) ought to be, [*exactly*]{}, $$E_0(N) = NE_0$$ in a correct QED. In other words, in the absence of nuclei and Coulomb potentials, there should be [*no binding*]{} caused by the field energy $H_f$. This is what we seem to observe experimentally, but this important topic does not seem to have been discussed in the QED literature.
Normally, one should expect binding, for the following mathematical reason: The first particle generates a field, $A(x)$, and energy $E_0$. The second particle can either try to generate a field $A(x+y)$, located very far away at $y$ or the second particle can try to take advantage of the field $A(x)$, already generated by the first particle, and achieve an insertion energy lower than $E_0$.
Indeed, this second phenomenon happens for bosons — as expected. For fermions, however, the Paul principle plays a crucial role (even in in the absence of Coulomb attractions). We show that $E_0(N) \geq C N
E_0$ for fermions, but we are unable to show that the universal constant $C = 1$. Even if $C < 1$, the situation could still be saved by mass renormalization, which drives the bare mass to zero as $\Lambda$ increases, thereby pushing particles apart.
BOSONS
------
[*Theorem \[theo:nonrelboson\];*]{} This theorem concerns the ground state energy of $N$ charged bosons. the Hamiltonian is given by \[manyham\] acting on the Hilbert space of symmetric functions tensored with the photon Fock space $\mathcal F$. It states, basically, that $C_1 \sqrt{N} \sqrt{\alpha} \Lambda^{3/2} \leq
E_0^{boson}(N) \leq C_2 N^{5/7} \alpha^{2/7} \Lambda^{12/7}$.
The proof follows essentially that of the one particle case. The interesting fact is that it implies [*binding*]{} of [*charged bosons*]{} (in the absence of the Coulomb repulsion). The binding energy is defined by $$\Delta E (N) = E_0 (N) - NE_0(1)$$ and satisfies the bounds $$\begin{aligned}
\Delta E(N) &\geq& C_1 \sqrt{N} \sqrt{\alpha}~ \Lambda^{3/2} -
C_2 N \alpha^{2/7} \Lambda^{12/7} \nonumber \\
\Delta E(N) &\leq& C_2 N^{2/7}
\alpha^{2/7} \Lambda^{12/7} - C_1 N \sqrt{\alpha}~ \Lambda^{3/2}
\label{eq:3.6}\end{aligned}$$ which can be made negative for appropriately chosen $N$ and $\Lambda$. There will be binding for all large enough $N$, irrespective of the cutoff $\Lambda$. It also has to be remarked that the Coulomb repulsion will, in all likelihood, not alter this result since it has an effect on energy scales of the order of $\Lambda$ and not $\Lambda^{12/7}$ or $\Lambda^{3/2}$.
FERMIONS
--------
The real issue for physics is what happens with fermions. We cannot show that there is no binding but we can show that the energy is extensive as in Theorem \[theo:nonrelfer\]. The Hamiltonian is the same as (\[manyham\]) but it acts on antisymmetric functions tensored with ${\mathcal F}$. (Spin can be ignored for present purposes.)
[*Rough sketch of the proof of Theorem \[manyham\]*]{}.
The difficulty in proving this theorem stems from the fact that the field energy is not extensive in any obvious way.
Define $\underbar{X} = (x_1, \cdots , x_N)$ and define the function $$n_j (\underbar{X}, R) = \# \{ x_i \neq x_j ~:~ |x_i - x_j| < R \} ~ .$$ This function counts the number of electrons that are within a distance $R$ of the $j^{\rm{th}}$ electron. Note that this function is not smooth, so that all the following computations have to be modified. (See [@LL].) We save half of the kinetic energy and write $$H = \frac{1}{4} \sum^{N}_{j=1} (p_j + \sqrt{\alpha} A (x_j))^2 +
H' ~ .$$ We apply the commutator estimate (\[eq:2.14\]) to the pair $$i [p_j + \sqrt{\alpha} A (x_j), \frac{1}{\sqrt{N_j (\underbar{X}, R) +
1}} \Pi (x_j)]$$ and obtain the bound (with the caveat mentioned above), for all $\alpha > 0$, $$H' \geq a \sqrt{\alpha} \Lambda^3 \sum^N_{j=1}
\frac{1}{\sqrt{N_j(\underbar{X}, R) + 1}} - a^2 \sum^N_{j=1} \frac{1}{N_j
(\underbar{X}, R) + 1} {\rm{F}} (x_j)^2 + H_f
\label{eq:3.8} $$ The next two steps are somewhat nontrivial and we refer the reader to [@LL]. First one notes that the modes $F(x_i)$ and $F(x_j)$ are essentially orthogonal (i.e., they commute) if $|x_i- x_j| > \Lambda^{-1}~ .$ Ignoring the technical details of how this is implemented, the key observation is that the last two terms in (\[eq:3.8\]) can be estimated from below by $ - N\Lambda$ provided $a = \Lambda^{-3/2}$.
The next ingredient is a new Lieb-Thirring type estimate involving the function $N_j(\underbar{X},R)$. It is here and only here that the Pauli exclusion principle is invoked.
On the space $\wedge^{N}_{j=1} L^2 {\mathbb R}^3; {\mathbb C}^q)$ of antisymmetric functions $$\sum_{j=1}^{N} (p_j + \sqrt{\alpha} A (x_j))^2 \geq \frac{C}{q^{2/3}}
\frac{1}{R^2} \sum^N_{j=1} N_j (\underbar{X}, R)^{2/3}$$ with $C \geq 0.00127$. An analogous inequality holds for the relativistic case as well: $$\sum_{j=1}^{N} |p_j + \sqrt{\alpha} A (x_j)| \geq \frac{C}{q^{1/3}}
\frac{1}{R} \sum^N_{j=1} N_j (\underbar{X}, R)^{1/3}$$ \[theo:3.3\]
By using the kinetic energy previously saved together with (\[eq:3.8\]) and the previous discussion, we get $$H \geq \sum^N_{j=1} \left\{ N_j (\underbar{X},R)^{2/3}
+ \sqrt{\alpha} \Lambda^{3/2} \frac{1}{\sqrt{N_j
(\underbar{X},R)+1}} \right\} - N \Lambda ~ .$$ By minimizing over $N_j$ the desired estimate is obtained. The upper bound is fairly elementary and is omitted. $\blacksquare$
RELATIVISTIC ENERGY BOUNDS {#relativistic}
==========================
[*Theorem \[theo:4.1\]:*]{} Sketch of Proof.
An upper bound for $E_0$ is easy to obtain, but it is indirect. Note that $$|p + \sqrt{\alpha} A(x) | \leq \varepsilon [ p + \sqrt{\alpha} A (x)
]^2 + (4 \varepsilon)^{-1}
\label{eq:4.3}$$ for any $\varepsilon > 0$. Take $\Psi = f (x) \otimes | 0\rangle$ with $|0\rangle$ being the Fock space vacuum. Using (\[eq:4.3\])
$$\begin{aligned}
(\Psi, H \Psi) &\leq& \varepsilon ~\int_{\mathbb{R}^{\rm{3}}} \{ \alpha
\langle 0 | A
(x)^2 | 0 \rangle | f (x)|^2 + | \nabla f (x)|^2 \} dx +
\varepsilon^{-1} \nonumber \\
&=& \frac{\varepsilon \alpha \Lambda^2}{4 \pi} ~ + \int |\nabla f|^2 +
\frac{1}{4 \varepsilon}~,
\label{eq:4.5a}\end{aligned}$$
since $\langle 0 | A (x)^2 | 0 \rangle = (2 \pi)^{-3} \int_{|k|<
\Lambda} |k|^{-1}
dk = \Lambda^2/4\pi^2$. We can now let $f(x) \rightarrow
V^{-\frac{1}{2}}$ and take $\varepsilon = (\pi/ \alpha)^{1/2}
\Lambda^{-1}$, whence $$E_0 \leq (\alpha/4 \pi)^{1/2} \Lambda ~.
\label{eq:4.5b}$$
Now we turn to the lower bound for $H$.
[**Step 1**]{}: Since $x \rightarrow \sqrt{x}$ is operator monotone, $$T > T_1 = | p_1 + \sqrt{\alpha} A_1 (x) | ~,
\label{eq:4.6}$$ where the subscript 1 denotes the 1 component (i.e., the $x$-component) of a vector. By replacing $T$ by $T_1$, we are now in a position to remove $A_1$ by a gauge transformation - but it has to be an [*operator-valued gauge transformation*]{}. The use of such a gauge transformation is is a novelty, as far as we are aware, in QED.
To effect the gauge transformation, set $$\varphi (x) = \frac{1}{\sqrt{2V}}~ \sum_{h,\lambda} ~
\frac{\varepsilon^{\lambda}(k)_1}{\sqrt{(k)}} ~ \left[ a_{\lambda} (k)
+ a_{\lambda}^{\ast} (-k) \right] \frac{e^{ik_1 x_1} -1}{ik_1} ~
e^{ik_{\perp}x_{\perp}}
\label{eq:4.7}$$ with $x_{\perp} = (x_2, x_3$). Then $[A_j (x), \varphi (x) ] = 0, ~~j= 1,2,3$ and $p_1 \exp~ [i \varphi (x)] = -A_1(x)$. The unitary $U = \exp~ [i
\varphi (x)]$ is a gauge transformation, but it is [*operator-valued*]{}. We have $$\begin{aligned}
U^{-1} | p_1 + A_1 (x) | U &=& |p_1| \nonumber \\
U^{-1} a_{\lambda} (k) U &=& a_{\lambda}(k) + f_{\lambda} (k, x)
\nonumber \\
U^{-1} a_{\lambda}^{\ast} (k) U &=& a_{\lambda}^{\ast}(k) +
\bar{f}_{\lambda} (k, x) \nonumber \\
\stackrel{\sim}H_f ~=~ U^{-1} H_f U &=& \sum_{k,\lambda} |k|
[ a_{\lambda}^{\ast} (k) + \bar{f}_{\lambda} (k,x) ]
[ a_{\lambda} (k) + f_{\lambda} (k,x) ]
\label{eq:4.8}\end{aligned}$$ with $$f_{\lambda} (k,x) ~=~ \sqrt{\frac{\alpha}{2V}} ~ \sum_{k,\lambda} ~
\frac{\varepsilon_{\lambda}(k)_1}{|k|} ~\frac{e^{-i k_1 x_1}-1}{k_1} ~
e^{-ik_{\perp}x_{\perp}} ~ .
\label{eq:4.9}$$
Since $p_{\perp}$ does not appear in our new Hamiltonian, $$\stackrel{\sim}H ~=~ U^{-1} HU ~=~ |p_1|+~\stackrel{\sim}H_f ~,
\label{eq:4.10}$$ the variable $x_{\perp}$ appears only as a parameter, and thus we can set $x_{\perp}$ = constant = (0,0), by translation invariance, and replace $\mathbb{R}^3$ by $\mathbb{R}^1 = \mathbb{R}$.
From now on $x_1 = x$ and, $p_1 = p = -i ~ d/dx$.
[**Step 2**]{}: The dependence on $x$ now appears in $\stackrel{\sim}H_f$ instead of in the kinetic energy, $|p|$. For each $x$ we can try to put $\stackrel{\sim}H_f$ into its ground state, which is that of a displaced harmonic oscillator. But, since this state depends on $x$, to do so will require a great deal of kinetic energy, $|p_1|$.
Let $\Psi$ be a normalized wave-function, i.e., a function on $L^2(\mathbb R) \otimes {\mathcal{F}}$. We write it as $\psi_x$ where $\psi_x \in {\mathcal{F}}$. Thus, with $\langle \cdot~ ,~ \cdot \rangle$ denoting the inner product on ${\mathcal{F}}, \int_{\mathbb R} \langle \psi_x, \psi_x
\rangle dx = 1$.
Decompose $\mathbb R$ as the disjoint union of intervals of length $\ell/\Lambda$, where $\ell$ is a parameter to be determined later. Denote these intervals by $I_j,~ j = 1, 2, \ldots~.$ A simple Poincaré type inequality gives, for $g ~:~ L^2 (\mathbb R)
\rightarrow
{\mathbb{C}}$, $$(g,|p| g) \geq C_1 \frac{\Lambda}{\ell} \sum_j
\int_{I_j} \{ | g(x) |^2 - |\bar{g}_j |^2 \} dx ~ ,$$ where $\bar{g}_j = \frac{\Lambda}{\ell} \int_{I_j} g (x) dx$ is the average of $g$ in $I_j$. Then $$(\Psi, |p| \Psi) ~ \geq C_1 \frac{\Lambda}{\ell} ~\sum_j ~
\int_{I_j} ~ \{ \langle \psi_x, \psi_x \rangle ~-~ \langle
\bar{\psi}_j, \bar{\psi}_j \rangle~\} dx ~ .
\label{eq:4.11}$$
[**Step 3**]{}: Next, we analyze $\stackrel{\sim}H_f$. We think of this as an operator on ${\mathcal{F}}$, parameterized by $x \in \mathbb R$. We would like $\stackrel{\sim}H_f$ to have a gap so we define $$H_x ~=~ \frac{\Lambda}{2} \sum_{\varepsilon \Lambda
\leq |k| \leq \Lambda} ~ \sum_{\lambda} ~ [a^+_{\lambda} (k) +
\bar{f}_{\lambda} (k,x)]~\cdot~ [~h.c.~]
\label{eq:4.12}$$
Clearly, $\stackrel{\sim}H_f \geq H_x$ and $$(\Psi, \stackrel{\sim}H \Psi) \geq \frac{\Lambda}{\ell} \sum_{j}
\int_{I_j} \langle \psi_x, \psi_x \rangle - \langle \bar{\psi}_j,
\bar{\psi}_j \rangle + \langle \psi_x, H_x \psi_x
\rangle dx ~ .
\label{eq:4.13}$$ For each interval $I_j$ we can minimize (\[eq:4.13\]) subject to $\int_{I_j}\langle \psi_x, \psi_x \rangle dx$ fixed. This leads to $$(h_j ~ \psi)_x = e ~ \psi_x
\label{eigenvalue1}$$ with $$(h_j ~ \psi)_x = \frac{\Lambda}{\ell} ~ \psi_x -
\frac{\Lambda}{\ell} ~ \bar{\psi}_j + H_x ~ \psi_x
\label{eigenvalue2}$$ Obviously, this eigenvalue problem (\[eigenvalue1\], \[eigenvalue2\]) is the same for all intervals $I_j$, so we shall drop the subscript $j$ and try to find the minimum $e$.
A lower bound to $h_j$ (and hence to $e$) can be found by replacing $H_x$ by $$\widehat{H}_x = \frac{\Lambda}{2} ( 1 - \Pi_x) ~,$$ where $\Pi_x = |g_x\rangle \langle g_x|$ is the projector onto the ground state, $|g_x\rangle,~ {\rm{of}}~ H_x$.
If we substitute $\widehat{H}_x$ into (\[eigenvalue2\]) the corresponding eigenvalue equation (\[eigenvalue1\]) becomes soluble. Multiply (\[eigenvalue1\]) on the left by $\langle g_x|$, whence
$$\left( \frac{\Lambda}{\ell} - e \right) \langle g_x, \psi_x \rangle =
\frac{\Lambda}{\ell} \langle g_x, \bar{\psi} \rangle
\label{eq:4.16}$$
Then, substitute (\[eq:4.16\]) into (\[eigenvalue2\]) and integrate $\int_I dx$ to find
$$\frac{1}{2} \Lambda^3 \ell^{-2} \left( \int \Pi_x dx \right)
\bar{\psi} = \left ( \frac{\Lambda}{\ell} - e \right)
\left(\frac{\Lambda}{2} - e \right) \bar{\psi} ~ .
\label{eq:4.17}$$
We know that $e < \Lambda/2$ because we could take $\psi_x$ = constant as a trial function, and then use $\stackrel{\sim}H_x \leq
\Lambda/2$. Also, $e < \Lambda/\ell$, because we could take $\Psi = \delta_{x_{0}} | g_{x_{0}} \rangle ~ .$
[**Step 4**]{}: Eq. (\[eq:4.17\]) will give us a lower bound to $e$ if we can find an upper bound to $Y = (\Lambda/\ell) \int_I \Pi_x dx ~.$ To do this note that $$\begin{aligned}
Y^2 &\leq& {\rm{Trace}} ~ Y^2 = \left( \frac{\Lambda}{\ell} \right)^2
\int_I \int_I |\langle g_x, g_y \rangle |^2 dxdy\nonumber \\
&=& \left( \frac{\Lambda}{\ell} \right)^2 \int_I\int_I \exp \{ -
\frac{\alpha}{2V} \sum_{\varepsilon \Lambda \leq |k| \leq \Lambda}
\sum_{\lambda}|f_{\lambda} (k,x) - f_{\lambda} (k,y)|^2 dxdy \}
\label{eq:4.18}\end{aligned}$$ Noting that $\sum_{\lambda = 1}^{2} e_{\lambda}(k)_1^2 = 1 -
k_1^2/k^2$, the quantity { } in (\[eq:4.18\]) becomes (as $V
\rightarrow \infty$)
$$\{ \quad \} = -\frac{2 \alpha}{(2 \pi)^3} \int_{\Lambda/2 < |k|
< \Lambda} \frac{1}{|k|^3 k_1^2} (k_{\perp}^2) [ \sin
\frac{k_1}{2} (x-y) ]
\label{eq:4.19}$$
After some algebra we find that
$$\left( \frac{1}{\ell} - \frac{e}{\Lambda} \right) \left( \frac{1}{2} -
\frac{e}{\Lambda} \right) \leq \frac{1}{2 \ell} ({\rm{Trace}}~
Y^2)^{1/2} \leq \frac{1}{2 \ell} \sqrt{K_{\ell} (\alpha)}
\label{eq:4.20}$$
where $$\begin{aligned}
K_{\ell} (\alpha) &=& \int_0^1 \int_0^1 \exp \left[ - \alpha
\frac{\ell}{\pi^2} |x-y| \int_0^{|x-y|\ell/4} \left( \frac{\sin t}{t}
\right)^2 ~ dt \right] ~ dxdy ~ .\nonumber \\
&\leq& \int_{-1/2}^{1/2} \exp [ - \alpha x^2 \ell^2 / 8 \pi ] ~ dx ~.
\label{eq:4.21}\end{aligned}$$ We find that $$\begin{aligned}
K_{\ell} (\alpha) &\sim& 1 - \alpha \ell^2 / 96 \pi, \quad \ell^2 \alpha
~ {\rm{small}}\\
&\sim& \sqrt{2} \pi ( \alpha \ell^2)^{1/2}, \quad \ell^2 \alpha ~
{\rm{large}} \nonumber
\label{eq:4.22}\end{aligned}$$ If $\alpha$ is small we take $\ell \sim \alpha^{-1/2}$. If $\alpha$ is large we take $\ell = 2$. This establishes our theorem for $N=1$.
[*Theorem \[theo:4.3\]:*]{} Sketch of Proof.
For $N > 1$ we can decompose $\mathbb{R}^3$ into cubic boxes $B_j,
j=1,2,3, \ldots$ of size $\ell \Lambda$ and “borrow” $\frac{1}{2}
|p+A(x)|^2$ kinetic energy from each particle. That is, $H_N =
H_N^{1/2} + \frac{1}{2} T_N$ with $T_N = \sum_{i=1}^{N} T(x_i)$. The Pauli principle will then yield an energy for $\frac{1}{2}T_N$ that is bounded below by (const.) $(n_{j-1})^{4/3}$, where $n_j$ is the particle number in box $B_j$.
[A]{}
L. Bugliaro, J. Fröhlich and G.M. Graf, *Stability of quantum electrodynamics with nonrelativistic matter*, Phys.Rev. Lett. [**77**]{} (1996), 3494-3497.
V. Bach, J. Fröhlich and I.M. Sigal, *Renormalization group analysis of spectral problems in quantum field theory*, Adv. Math. [**137**]{} (1998),205-298; *Quantum electrodynamics of confined nonrelativistic particles*, [*ibid*]{} 299-395.
C. Fefferman, *Stability of Coulomb systems in a magnetic field*, Proc. Natl. Acad. Sci. USA, [**92**]{} (1995), 5006-5007.
C. Fefferman, J. Fröhlich and G.M. Graf, *Stabilty of nonrelativistic quantum mechanical matter coupled to the (ultraviolet cutoff) radiation field*, Proc. Natl. Acad. Sci. USA [**93**]{} (1996), 15009-15011; *Stability of ultraviolet cutoff quantum electrodynamics with non-relativistic matter*, Commun. Math. Phys. [**190**]{} (1997), 309–330.
H. Hogreve, Math. Reviews [**99e:81051a, b**]{} Amer. Math. Soc. (1999).
E.H. Lieb and M. Loss, *Remarks about the ultraviolet problem in QED*, (in preparation).
E.H. Lieb, M. Loss and J. P. Solovej, *Stability of Matter in Magnetic Fields*, Phys. Rev. Lett. [*75*]{}, 985-989 (1995).
E.H. Lieb, H. Siedentop and J.P. Solovej, *Stability and Instability of Relativistic Electrons in Magnetic Fields*, J. Stat. Phys. [**89**]{} (1997), 37-59.
E.H.Lieb and K. Yamazaki, *Ground State Energy and Effective Mass of the Polaron*, Phys. Rev. [**111**]{} (1958), 728-733.
T. Okamoto and K. Yajima, *Complex scaling technique in nonrelativistic massive QED*, Ann. Inst. H. Poincaré Phys. Théor. [**42**]{} (1985), 311-327.
[^1]: Supported in part by NSF Grant PHY-98 20650.
[^2]: Supported in part by NSF Grant DMS-95 00840.
[^3]: ©1999 by the authors. Reproduction of this work, in its entirety, by any means, is permitted for non-commercial purposes by any means is permitted for non-commercial purposes.
[^4]: This is a preliminary report on our work presented at the University of Alabama,Birmingham - Georgia Tech International Conference on Differential Equations and Mathematical Physics, Birmingham, March 15-19, 1999
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'From the viewpoint of classical electrodynamics, we identify the role of spin-to-orbital angular momentum conversion in spin Hall effect (SHE) of light. We introduce a distinct separation between spin and orbital angular momenta to clarify the spin-orbital interaction in conventional beam refraction. We demonstrate that the refractive index gradient can enhance or suppress the spin-to-orbital angular momentum conversion, and thus can control the SHE of light. We suggest that the metamaterial whose refractive index can be tailored arbitrarily may become a good candidate for amplifying or eliminating the SHE of light, and by properly facilitating the spin-to-orbital angular momentum conversion the SHE may be enhanced dramatically. The transverse spatial shifts governed by the spin-to-orbital angular momentum conversion, provide us a clear physical picture to clarify the role of refractive index gradient in the SHE of light. These findings provide a pathway for modulating the SHE of light and can be extrapolated to other physical systems.'
author:
- Hailu Luo
- Shuangchun Wen
- Weixing Shu
- Dianyuan Fan
title: 'Spin-to-orbital angular momentum conversion in spin Hall effect of light'
---
Introduction {#SecI}
============
The development of spin photonics has taken an important step forwards due to the recently experimental verifications of the spin Hall effect (SHE) of light [@Hosten2008; @Bliokh2008]. The SHE is a transport phenomenon, in which an applied field on the spin particles leads to a spin-dependent shift perpendicular to the electric field [@Murakami2003; @Sinova2004; @Wunderlich2005]. The SHE of light can be regarded as a direct optical analogy in which the spin electrons and electric potential are replaced by spin photons and refractive index gradient. The SHE of light sometimes referred to as the Fedorov–Imbert effect, was predicted theoretically by Fedorov [@Fedorov1965], and was experimentally confirmed by Imbert [@Imbert1972]. The theory was extended in a formulation which shows that the transverse spatial separation of the left and right circularly polarized components on oblique incidence directly from the total angular momentum conservation [@Onoda2004; @Bliokh2006a]. More recently, the interesting effect has also been observed in scattering from dielectric spheres [@Haefner2009], on the direction making an angle with the propagation axis [@Aiello2009a], and in silicon via free-carrier absorption [@Menard2010].
The polarization-dependent transverse shift in the SHE of light is generally believed as a result of an effective spin-orbital interaction, which describes the mutual influence of the spin (polarization) and trajectory of the light beam [@Bliokh2008]. There are two characteristics of the spin-orbit interaction of photons: The first one is the influence of the trajectory upon polarization [@Chiao1986; @Tomita1986]; The second one is the reciprocal influence of the polarization upon the trajectory [@Dooghin1992; @Liberman1992]. These mechanisms have well been understood in gradient refractive index media [@Chiao1986; @Tomita1986; @Dooghin1992; @Liberman1992]. In addition, the spin-orbital interaction in both inhomogeneous anisotropic media [@Marrucci2006] and in tightly focused beams [@Zhao2007] can be explained by spin-to-orbital angular momentum conversion. However, the physical picture of spin-orbit interaction in conventional beam refraction has not yet been fully examined. For example, the relation between refractive index gradient and spin-to-orbital angular momentum conversion is unclear. Whether spin-to-orbital angular momentum conversion can be enhanced (or suppressed) by increasing (or decreasing) the refractive index gradient? Thus, the aim of this paper is to reveal the spin-to-orbital angular momentum conversion in the SHE of light from the viewpoint of classical electrodynamics.
The paper is organized as follows. First, we establish a three dimension beam propagation model for describing the SHE of light in conventional beam refraction. Our result shows that the refractive index gradient can enhance or suppress the transverse spatial shifts in SHE of light. We suggest that the metamaterial whose refractive index can be tailored arbitrarily is a good candidate to modulate the SHE of light. Next, we attempt to obtain a clear physical picture of spin-to-orbital angular momentum conversion in the SHE of light. Within the paraxial approximation, a distinct separation between spin and orbital angular momenta is introduced. We find that the SHE of light may be dramatically enhanced by facilitating the spin-to-orbital angular momentum conversion. Finally, we want to explore what role refractive-index gradient plays in the spin-to-orbital angular momentum conversion. We demonstrate that the refractive index gradient can enhance or suppress the spin-to-orbital angular momentum conversion, and thus can control the SHE of light.
Three dimension beam propagation model {#SecII}
======================================
To reveal the spin-to-orbital angular momentum conversion, we need to establish a three dimension beam propagation model for describing the SHE of light. Figure \[Fig1\] illustrates the beam reflection and refraction in Cartesian coordinate system. The $z$ axis of the laboratory Cartesian frame ($x,y,z$) is normal to the air-glass interface locating at $z=0$. We use the coordinate frames ($x_a,y_a,z_a$) for central wave vector, where $a=i,r,t$ denotes incident, reflection, and transmission, respectively. In addition, we must introduce local Cartesian frames ($X_a,Y_a,Z_a$) to describe an arbitrary wave vector. The electric field of the $a$th beam can be solved by employing the Fourier transformations [@Goodman1996]. The complex amplitude for the $a$th beam can be conveniently expressed as $$\begin{aligned}
\mathbf{E}_a(x_a,y_a,z_a )&=&\int d k_{ax}dk_{ay}
\tilde{\mathbf{E}_a}(k_{ax},k_{ay})\nonumber\\
&&\times\exp [i(k_{ax}x_a+k_{ay}y_a+ k_{az} z_a)],\label{noapr}\end{aligned}$$ where $k_{az}=\sqrt{k_a^2-(k_{ax}^2+k_{ay}^2)}$ and $\tilde{\mathbf{E}_a}(k_{ax},k_{ay})$ is the angular spectrum. The approximate paraxial expression for the field in Eq. (\[noapr\]) can be obtained by the expansion of the square root of $k_{az}$ to the first order [@Lax1975], which yields $$\begin{aligned}
\mathbf{E}_a&=&\exp(i k_a z_a) \int dk_{ax}dk_{ay}
\tilde{\mathbf{E}_a}(k_{ax},k_{ay})\nonumber\\
&&\times\exp \left[i\left(k_{ax}x_a+k_{ay}y_a-\frac{k_{ax}^2+k_{a
y}^2}{2 k_a}z_a\right)\right]\label{apr}.\end{aligned}$$ In general, an arbitrary linear polarization can be decomposed into horizontal and vertical components. In the spin basis set, the angular spectrum can been written as: $$\tilde{E}_i^H=\frac{1}{\sqrt{2}}(\tilde{\mathbf{E}}_{i+}+\tilde{\mathbf{E}}_{i-})\label{SBH},$$ $$\tilde{E}_i^V=\frac{1}{\sqrt{2}}i(\tilde{\mathbf{E}}_{i-}-\tilde{\mathbf{E}}_{i+})\label{SBV}.$$ Here, $H$ and $V$ represent horizontal and vertical polarizations, respectively. The positive and negative signs denote the left and right components, respectively [@Beth1936]. The monochromatic Gaussian beam can be formulated as a localized wave packet whose spectrum arbitrarily narrows, whose angular spectrum can be written as $$\tilde{\mathbf{E}}_{i\pm}=(\mathbf{e}_{ix} \pm
i\mathbf{e}_{iy})\frac{w_0}{\sqrt{2\pi}}\exp\left[-\frac{w_0^2(k_{ix}^2+k_{iy}^2)}{4}\right]\label{asi},$$ where $w_0$ is the beam waist. After the angular spectrum is known, we can obtain the field characteristics for the $a$th beam.
![\[Fig1\] (color online) Schematic illustrating the reflection and refraction of central and local wave vectors at an air-glass interface in Cartesian coordinate system. $x_ay_az_a$ are are reference frames for central wave vector and $X_aY_aZ_a$ are local reference frames for an arbitrary wave vector. $oxyz$ and $OXYZ$ are the interface reference frames for central and local wave vectors, respectively.](Fig1.eps){width="8cm"}
To accurately describe the SHE of light, it is need to determine the reflection and transmission of arbitrary wave-vector components. From the central frame $x_iy_iz_i$ to the local frame $X_iY_iZ_i$, the following three steps should be carried out. First, we transform the electric field from the reference frame $x_iy_iz_i$ around the $y$ axis by the incident angle $\theta_i$ to the frame $xyz$: $\tilde{E}_{xyz}=m_{{x_iy_iz_i}\rightarrow{xyz}}\tilde{E}_{x_iy_iz_i}$, where $$\begin{aligned}
m_{{x_iy_iz_i}\rightarrow{xyz}}=\left[
\begin{array}{ccc}
\cos\theta_i & 0 & -\sin\theta_i\\
0 & 1 & 0\\
\sin\theta_i & 0 & \cos\theta_i
\end{array}
\right].\label{matrixr}\end{aligned}$$ Then, we transform the electric field from the reference frame $xyz$ around the $y$ axis by an angle $k_{iy}/k_0\sin\theta_i$ to the frame $XYZ$, and the correspondingly matrix is given by $$\begin{aligned}
m_{{xyz}\rightarrow{XYZ}}=\left[
\begin{array}{ccc}
1 & \frac{k_{iy} }{k_0\sin\theta_i} & 0\\
-\frac{k_{iy}}{k_0\sin\theta_i} & 1 & 0\\
0 & 0 & 1
\end{array}
\right].\label{matrixr}\end{aligned}$$ Finally, we transform the electric field from the reference frame $XYZ$ around the $y$ axis by an angle $-\theta_i$ to the frame $X_iY_iZ_i$, and the rotation matrix can be written as $$\begin{aligned}
m_{{XYZ}\rightarrow{X_iY_iZ_i}}=\left[
\begin{array}{ccc}
\cos\theta_i & 0 & \sin\theta_i\\
0 & 1 & 0\\
-\sin\theta_i & 0 & \cos\theta_i
\end{array}
\right].\label{matrixr}\end{aligned}$$ The rotation matrix of the reference frame $x_iy_iz_i$ around the $y$ axis by an angle $M_{{x_iy_iz_i}\rightarrow{X_iY_iZ_i}}=m_{{XYZ}\rightarrow{X_iY_iZ_i}}
m_{{xyz}\rightarrow{XYZ}}m_{{x_iy_iz_i}\rightarrow{xyz}}$, and we have $$\begin{aligned}
M_{{x_iy_iz_i}\rightarrow{X_iY_iZ_i}}=\left[
\begin{array}{cc}
1 &\frac{k_{ry} \cot\theta_i}{k_0} \\
-\frac{k_{ry} \cot\theta_i}{k_0} & 1
\end{array}
\right].\label{matrixr}\end{aligned}$$ For an arbitrary wave vector, the reflected field is determined by $\tilde{E}_{X_rY_rZ_r}=r_{p,s}\tilde{E}_{X_iY_iZ_i}$, where $r_p$ and $r_s$ are the Fresnel reflection coefficients. The reflected field should be transformed from $X_rY_rZ_r$ to $x_ry_rz_r$. Following the similar procedure, the reflected field can be obtained by carrying out three steps of transformation: $\tilde{E}_{x_ry_rz_r}=m_{{X_rY_rZ_r}\rightarrow{x_ry_rz_r}}\tilde{E}_{X_rY_rZ_r}$ where $$\begin{aligned}
M_{{X_rY_rZ_r}\rightarrow{x_ry_rz_r}}=\left[
\begin{array}{cc}
1 &\frac{k_{ry} \cot\theta_i}{k_0} \\
-\frac{k_{ry} \cot\theta_i}{k_0} & 1
\end{array}
\right].\label{matrixr}\end{aligned}$$ Here, only the two dimension rotation matrices is taken into account, since the longitudinal component of electric field can be obtained from the divergence equation $\tilde{E}_{az}k_{az}=-(\tilde{E}_{ax}k_{ax}+\tilde{E}_{ay}k_{ay})$. The reflection matrix can be written as $$\begin{aligned}
M_R=M_{{X_rY_rZ_r}\rightarrow{x_ry_rz_r}}\left[
\begin{array}{cc}
r_p &0 \\
0 & r_s
\end{array}
\right]M_{{x_iy_iz_i}\rightarrow{X_iY_iZ_i}}.\label{matrixrI}\end{aligned}$$ The reflected angular spectrum is related to the boundary distribution of the electric field by means of the relation $\tilde{E}_r(k_{rx},k_{ry})=M_R\tilde{E}_i(k_{ix},k_{iy})$, and we have $$\begin{aligned}
\left[\begin{array}{cc}
\tilde{E}_r^H\\
\tilde{E}_r^V
\end{array}\right]
=\left[
\begin{array}{cc}
r_p&\frac{k_{ry} (r_p+r_s) \cot\theta_i}{k_0} \\
-\frac{k_{ry} (r_p+r_s)\cot\theta_i}{k_0} & r_s
\end{array}
\right]\left[\begin{array}{cc}
\tilde{E}_i^H\\
\tilde{E}_i^V
\end{array}\right],\label{matrixrII}\end{aligned}$$ It is well known that $r_p$ and $r_s$ can be expanded as a power series that can be truncated to an appropriate order $N$. By making use of Taylor series expansion based on the arbitrary angular spectrum component, $r_p$ and $r_s$ can be expanded as a polynomial of $k_{ix}$: $$\begin{aligned}
r_{p,s}(k_{ix})&=&r_{p,s}(k_{ix}=0)+k_{ix}\left[\frac{\partial
r_{p,s}(k_{ix})}{\partial
k_{ix}}\right]_{k_{ix}=0}\nonumber\\
&&+\sum_{j=2}^{N}\frac{k_{ix}^N}{j!}\left[\frac{\partial^j
r_{p,s}(k_{ix})}{\partial k_{ix}^j}\right]_{k_{ix}=0}\label{LMD}.\end{aligned}$$ Our analysis is confined to the zero order to obtain a sufficiently good approximation. From the boundary condition, we obtain $k_{rx}=-k_{ix} $ and $k_{ry}= k_{iy}$. In fact, after the incident angular spectrum is known, Eq. (\[apr\]) together with Eqs. (\[asi\]) and (\[matrixrII\]) provides the paraxial expression of the reflected field: $$\begin{aligned}
\mathbf{E}_{r\pm}^H&=&\frac{r_p}{\sqrt{\pi
}w_0}\frac{z_R}{z_R+iz_r}\exp(ik_rz_r)\nonumber\\
&&\times\exp\left[-\frac{k_0}{2}\frac{x_{r}^2+(y_{r}\pm\delta_r^H)^2}{z_R+iz_r}\right]\label{HPR}.\end{aligned}$$ $$\begin{aligned}
\mathbf{E}_{r\pm}^V&=&\frac{\pm i r_s}{\sqrt{\pi
}w_0}\frac{z_R}{z_R+iz_r}\exp(ik_rz_r)\nonumber\\
&&\times\exp\left[-\frac{k_0}{2}\frac{x_{r}^2+(y_{r}\pm\delta_r^V)^2}{z_R+iz_r}\right]\label{VPR}.\end{aligned}$$ where $\delta_r^H=(1+r_s/r_p)\cot\theta_i/k$ and $\delta_r^V=(1+r_s/r_p)\cot\theta_i/k$. Note that the above expression of reflected field coincides with the early results [@Bliokh2007; @Aiello2008; @Menzel2008; @Luo2009] with different methods.
We proceed to consider the transmitted field. Following the similar procedure, we obtain the transform matrix from $X_tY_tZ_t$ to $x_ty_tz_t$ as $$\begin{aligned}
M_{{X_tY_tZ_t}\rightarrow{x_ty_tz_t}}=\left[
\begin{array}{cc}
1 &\frac{k_{ty} \cos\theta_t}{k_0 \sin\theta_i} \\
-\frac{k_{ty} \cos\theta_t}{k_0 \sin\theta_i} & 1
\end{array}
\right],\label{matrixr}\end{aligned}$$ where $\theta_t$ is the transmitted angle. For an arbitrary wave vector, the transmitted field is determined by $\tilde{E}_{X_tY_tZ_t}=t_{p,s}\tilde{E}_{X_iY_iZ_i}$, where $t_p$ and $t_s$ are the Fresnel transmission coefficients. Hence, the transmitted field should be transformed from $X_rY_rZ_r$ to $x_ty_tz_t$, and the transmission matrix can be written as $$\begin{aligned}
M_T=M_{{X_tY_tZ_t}\rightarrow{x_ty_tz_t}}\left[
\begin{array}{cc}
t_p &0 \\
0 & t_s
\end{array}
\right]M_{{x_iy_iz_i}\rightarrow{X_iY_iZ_i}}.\label{matrixtI}\end{aligned}$$ The transmitted angular spectrum is related to the boundary distribution of the electric field by means of the relation $\tilde{E}_t(k_{tx},k_{ty})=M_T\tilde{E}_i(k_{ix},k_{iy})$, and can be written as $$\begin{aligned}
\left[\begin{array}{cc}
\tilde{E}_t^H\\
\tilde{E}_t^V
\end{array}\right]
=\left[
\begin{array}{cc}
t_p&\frac{k_{ty} (t_p-\eta t_s) \cot\theta_i}{k_0} \\
\frac{k_{ty} (\eta t_p-t_s)\cot\theta_i}{k_0} & t_s
\end{array}
\right]\left[\begin{array}{cc}
\tilde{E}_i^H\\
\tilde{E}_i^V
\end{array}\right],\label{matrixtII}\end{aligned}$$ where $\eta=\cos\theta_t/\cos\theta_i$. Based on the Taylor series expansion, the Fresnel transmission coefficients $t_p$ and $t_s$ can be written as $$\begin{aligned}
t_{p,s}(k_{ix})&=&t_{p,s}(k_{ix}=0)+k_{ix}\left[\frac{\partial
t_{p,s}(k_{ix})}{\partial
k_x}\right]_{k_{ix}=0}\nonumber\\
&&+\sum_{j=2}^{N}\frac{k_{ix}^N}{j!}\left[\frac{\partial^j
t_{p,s}(k_{ix})}{\partial k_{ix}^j}\right]_{k_{ix}=0}\label{TPS}.\end{aligned}$$ Note that the Fresnel coefficients are real in the regime of partial reflection and transmission, $\sin\theta_i<n$. From the Snell’s law under the paraxial approximation, we obtain $k_{tx}=k_{ix}/\eta $ and $k_{ty}= k_{iy}$. Substituting Eqs. (\[asi\]) and (\[matrixtII\]) into Eq. (\[apr\]), we obtain the transmitted field: $$\begin{aligned}
\mathbf{E}_{t\pm}^H&=&\frac{t_p}{\sqrt{\pi} w_0}\frac{ z_{Ry} \exp
(ik_tz_t)}{\sqrt{(z_{Rx}+iz_t)(z_{Ry}+iz_t)}}
\nonumber\\&&\times\exp\left[-\frac{n k_0}{2}\left(\frac{x_{t}^2}{
z_{Rx}+iz_t}+\frac{(y_{t}\pm\delta_t^H)^2}{z_{Ry}+iz_t}\right)\right]\label{HPT}.\end{aligned}$$ $$\begin{aligned}
\mathbf{E}_{t\pm}^V&=&\frac{\pm i t_s}{\sqrt{\pi} w_0}\frac{z_{Ry}
\exp (ik_tz_t)}{\sqrt{(z_{Rx}+iz_t)(z_{Ry}+iz_t)}}
\nonumber\\&&\times\exp\left[-\frac{n k_0}{2}\left(\frac{x_{t}^2}{
z_{Rx}+iz_t}+\frac{(y_{t}\pm\delta_t^V)^2}{z_{Ry}+iz_t}\right)\right]\label{VPT}.\end{aligned}$$ Here, $\delta_t^H=(1-\eta t_s/t_p)\cot\theta_i/k$ and $\delta_t^V=(1-\eta t_p/t_s)\cot\theta_i/k$. The interesting point we want to stress is that there are two different Rayleigh lengths, $z_{Rx}=n\eta^2k_0w_0^2/2$ and $z_{Ry}=n k_0w_0^2/2$, characterizing the spreading of the beam in the direction of $x$ and $y$ axes, respectively. Up to now, we have established a general propagation model to describe the reflected and transmitted fields.
Role of the refractive index gradient {#SecII}
=====================================
It is well known that the SHE of light manifests itself as polarization-dependent transverse shifts in the process of reflection and refraction. To reveal the SHE of light, we now determine the transverse spatial shifts of field centroid. The time-averaged linear momentum density associated with the electromagnetic field can be shown to be [@Jackson1999] $$\mathbf{p}_{a}(\mathbf{r})=
\frac{1}{2c^2}\mathrm{Re}[\mathbf{E}_{a}(\mathbf{r})
\times\mathbf{H}_{a}^\ast(\mathbf{r})]\label{LMD},$$ where the magnetic field can be obtained by $\mathbf{H}_{a}=-ik_{a}^{-1} \nabla\times\mathbf{E}_{a}$. The intensity distribution of electromagnetic fields is closely linked to the longitudinal momentum currents $I(x_a,y_a,z_a)\propto\mathbf{p}_a\cdot \mathbf{e}_{az}$.
![\[Fig2\] (color online) The SHE of light manifests itself as polarization-dependent transverse shifts of field centroid. \[(a), (b)\] Intensity distribution of the reflected field for left and right circularly polarized components, respectively. \[(c), (d)\] Intensity distribution of the transmitted field for left and right circularly polarized components, respectively. The refractive index of the glass is $n=1.515$ and the incident angle is chosen as $\theta_i=\pi/6$. The beam waist is chosen as $w_0=10\lambda$. The intensity distributions in the plane $z_a=0$ are plotted in normalized units.](Fig2.eps){width="8cm"}
Figure \[Fig2\] shows the transverse spatial shifts of beam centroid in an air-glass interface. In the case of reflection, the left circularly polarized component exhibits a negative shift \[Fig. \[Fig2\](a)\]. For the right circular polarization, however, presents a positive shift \[Fig. \[Fig2\](b)\]. In the case of transmission, the left circularly polarized component exhibits a positive transverse shift \[Fig. \[Fig2\](c)\]. For the right circularly polarized component, however, presents a negative transverse shift \[Fig. \[Fig2\](d)\]. This interesting phenomenon gives a clear evidence of polarization-dependent splitting of field intensity in the SHE of light. The transverse shifts are polarization-dependent, and thus can be regarded as the influence of the polarization upon trajectory. In the air-glass interface, the transverse shifts are just a few tens of nanometers, which can be observed via weak measurements [@Hosten2008; @Krowne2009]. However, how to amplify this tiny effect is still an open problem.
As we know that the refractive index gradient acts as the electric potential gradient in the electronic systems. Now a question naturally arises: Whether can the refractive index gradient enhance the transverse shift in the SHE of light? To answer this question we need to obtain a relation between the transverse shift and the refractive index gradient. At any given plane $z_a=\text{const.}$, the transverse spatial shift of beam centroid compared to the geometrical-optics prediction is given by $$\Delta y_{a}= \frac{\int \int y_a I(x_a,y_a,z_a) \text{d}x_a
\text{d}y_a}{\int \int I(x_a,y_a,z_a) \text{d}x_a
\text{d}y_a}.\label{centroid}$$ Note that the transverse spatial shift is $z_a$-independent, while the transverse angular shift can be regarded as a small shift inclining from the $z_a$ axis. In addition, the field also experience a longitudinal spatial shift [@Goos1974; @Merano2007; @Lofler2010] and a longitudinal angular shift [@Merano2009; @Aiello2009b]. Note that the angle shift means that the Snell’s law cannot accurately describe the beam refraction phenomenon [@Duval2006].
We first consider the transverse spatial shift of the reflected field. After substituting the reflected field Eqs. (\[HPR\]) and (\[VPR\]) into Eq. (\[centroid\]), we obtain the transverse spatial shifts as $$\Delta y_{r\pm}^H =\mp\frac{\lambda}{2\pi} (1+r_s/r_p)\cot
\theta_i\label{TSRH},$$ $$\Delta y_{r\pm}^V =\mp\frac{\lambda}{2\pi} (1+r_p/r_s)\cot
\theta_i\label{TSRV}.$$ For an arbitrary linearly polarized beam, the transverse spatial shift of the reflected field is given by $$\Delta y_{r\pm}=\cos\gamma_r^2\Delta y_{r\pm}^H+\sin\gamma_r^2\Delta
y_{r\pm}^V\label{TSRHV},$$ where $\gamma_r$ is the reflected polarization angle. In the frame of classical electrodynamics, the reflection polarization angle is determined by: $$\cos\gamma_r=\frac{\cos\gamma_i
r_p}{\sqrt{\cos^2\gamma_i^2r_p^2+\sin^2\gamma_i^2r_s^2}},$$ $$\sin\gamma_r=\frac{\sin\gamma_i
r_s}{\sqrt{\cos^2\gamma_i^2r_p^2+\sin^2\gamma_i^2r_s^2}}.$$ Here, $\gamma_i$ is the incident polarization angle. Under the limit of ultra-high refractive index gradient, the transverse spatial shifts can be written as $\Delta y_{r\pm}=0$. This simple result means that the recent advent of a new class of metamaterial with ultra-high refractive index [@Shin2009] is a possible candidate to eliminate the transverse spatial shift in the reflected field.
![\[Fig3\] (a) Schematically shows the SHE of light manifests itself as spin-dependent splitting. The transverse spatial shifts of transmitted field $\Delta{y_{t\pm}}/\lambda$ versus the thickness of air gap $d_2$. The incident polarization angles are chooser as (b) $\theta_i=0$ , (c) $\theta_i=\pi/4$, and $\theta_i=\pi/2$. Other parameters are chosen to be the same as in Fig. \[Fig2\].](Fig3.eps){width="8cm"}
We next consider the transverse spatial shifts of the transmitted field. After substituting the transmitted field Eqs. (\[HPT\]) and (\[VPT\]) into Eq. (\[centroid\]), we have $$\Delta y_{t\pm}^H =\mp\frac{\lambda}{2\pi} (1-\eta t_s/t_p)\cot
\theta_i\label{TSRH},$$ $$\Delta y_{t\pm}^V =\mp\frac{\lambda}{2\pi} (1-\eta t_p/t_s)\cot
\theta_i\label{TSRV}.$$ For an arbitrary linearly polarized wave-packet, the transverse spatial shifts of the transmitted field are given by $$\Delta y_{t\pm}=\cos\gamma_t^2\Delta y_{t\pm}^H+\sin\gamma_t^2\Delta y_{t\pm}^V\label{TSRHV},$$ where the transmission polarization angle $\gamma_t$ is determined by $$\cos\gamma_t=\frac{\cos\gamma_i
t_p}{\sqrt{\cos^2\gamma_i^2t_p^2+\sin^2\gamma_i^2t_s^2}}\label{COST},$$ $$\sin\gamma_t=\frac{\sin\gamma_i
t_s}{\sqrt{\cos^2\gamma_i^2t_p^2+\sin^2\gamma_i^2t_s^2}}\label{SINT}.$$ Note that the above expression coincides well with the early results [@Hosten2008] with the quantum method. Our scheme shows that the SHE of light can be explained from the classic electrodynamics. Under the limit of ultra-high refractive index gradient, the transmitted field tends to reach a saturation value: $$\Delta y_{t\pm}=\pm\frac{\lambda\sin\theta_i\cos^2\gamma_i^2}
{2\pi(\cos^2\gamma_i^2+\cos^2\theta_i^2\sin^2\gamma_i^2)}\label{TSTL}.$$ Up to now, we have described the polarization-dependent splitting in the SHE of light from the viewpoint of pure classical electrodynamics.
We proceed to examine the role of refractive index gradient (i.e., $\Delta{n}=|n|-1$) in the SHE of light. The normalized transverse spatial shifts of transmitted field for various refractive index gradients as shown in Fig. \[Fig3\]. We first consider the wave packet incident from air to a low-refractive-index medium ($\Delta{n}<0$). For the left circularly polarized component, the field centroid exhibits a negative transverse shift. For the right circularly polarized component, the beam centroid also presents a transverse spatial shift, but in an opposite direction. We next consider the wave packet incident from air to a high-refractive-index medium ($\Delta{n}>0$). For a certain polarized component, we find that the SHE of light is reversed when the refractive index gradient is inverted. It is clearly shown that the transverse spatial shifts increases with the increase of the refractive index gradient $\Delta n$. Within the regime of low refractive index, the transverse spatial shifts are enhanced sharply with the increase of $\Delta n$. When the refractive index $|n|<\sin\theta_i$, the beam is totally reflected. While in the regime of high refractive index, the transverse spatial shifts tend to reach a saturation value. In addition, the transverse spatial shifts are also sensitive to the incident polarization angle as clearly shown in the figure. Hence, we can enhance or suppress the transverse spatial shifts in SHE of light by modulating the refractive index gradient and incident polarization angle.
![\[Fig4\] The normalized transverse spatial shifts of transmitted field $\Delta y_{t\pm}/\lambda$ versus incident angles $\theta_i$ in the metamaterial with $\varepsilon=-1.00+0.028i$ and $\mu=-1.01+0.028i$ (dashed-dotted and dashed-doted-dotted lines). As a comparison, the conventional medium is chosen as the glass with $n=+1.515$ (solid and dashed lines). The incident polarization angle choose as $\gamma_i=0$. Other parameters are chosen to be the same as in Fig. \[Fig2\]. Inset: Zoomed-in view of the transverse spatial shifts at the air-metamaterial interface. ](Fig4.eps){width="8cm"}
How to attenuate the SHE of light is not of pure academic interest, owing to the requirement of eliminating the transverse shifts, such as reflection, refraction, and focusing in optical experiments. We suggest that the metamaterial whose refractive index can be tailored arbitrarily [@Smith2004; @Pendry2006] may become a good candidate. In order to accurately describe how the metamaterial attenuate the SHE of light, it is necessary to include material dispersion and losses. Thus, a certain dispersion relation, such as the Lorentz medium model, should be introduced. The relative constitutive parameters are $$\begin{aligned}
\varepsilon(\omega)&=&1-\frac{\omega_{ep}^2}{\omega^2
-\omega_{eo}^2+i\Gamma_e\omega },\\
\mu(\omega)&=&1-\frac{F\omega_{mp}^2}{\omega^2-\omega_{mo}^2+i\Gamma_m\omega
}.\end{aligned}$$ To avoid the trouble involving in a certain value of frequency, we assume the material parameters are $\omega_{eo}=\omega_{mo}=\omega_o$, $\omega_{ep}^2=\omega_{mp}^2=2
\omega_o^2$, $F=1.005$, and $\Gamma_e=\Gamma_m=0.01\omega_0$. Figure \[Fig4\] shows the normalized transverse spatial shifts of transmitted field versus incident angle at the air-metamaterial interface. For the comparison, we also plot the transverse spatial shifts at the air-glass interface. In a large range of incident angles, the metamaterial is a good candidate for suppressing the SHE of light in the process of refraction. In the ideal case ($\varepsilon=-1$ and $\mu=-1$), the transverse shift can be eliminated completely. This is another reason why a simple planar slab provides us with a perfect lens without aberration [@Pendry2000]. As we will see in the following, the spin-to-orbital angular momentum conversion can be used to explain the inherent physics underlying these intriguing phenomena.
Spin-to-orbital angular momentum conversion {#SecII}
===========================================
It should be noted is that the spin-orbital interaction in both inhomogeneous anisotropic media [@Marrucci2006] and in tightly focused beams [@Zhao2007] can be explained by spin-to-orbital angular momentum conversion. To obtain a clear physical picture of the SHE of light, we introduce a distinct separation between spin and orbital angular momenta to clarify the spin-orbital interaction. The momentum current can be regarded as the combined contributions of spin and orbital parts: $$\mathbf{p}_a=\mathbf{p}_a^O+\mathbf{p}_a^S.$$ Here, the orbital term is determined by the macroscopic energy current with respect to an arbitrary reference point and does not depend on the polarization. The spin term, on the other hand, relates to the phase between orthogonal field components and is completely determined by the state of polarization [@Bekshaev2007]. In a monochromatic optical beam, the spin and orbital currents can be respectively written in the form: $$\mathbf{p}_a^{O}=\mathrm{Im}[\mathbf{E}_a^{\ast}\cdot(\nabla)\mathbf{E}_a],$$ $$\mathbf{p}_a^{S}=\mathrm{Im}[(\mathbf{E}_a\cdot\nabla)\mathbf{E}_a^{\ast}],$$ where $\mathbf{E}_a^{\ast}\cdot(\nabla)\mathbf{E}_a=E_{ax}^{\ast}\nabla
E_{ax}+E_{ay}^{\ast}\nabla E_{ay}+E_{az}^{\ast}\nabla E_{az}$ is the invariant Berry notation [@Berry2009]. It has been shown that both spin and orbital currents originate from the beam transverse inhomogeneity and their components are directly related to the azimuthal and radial derivatives of the beam profile parameters. The orbital currents are mainly produced by the phase gradient, while the spin currents are orthogonal to the intensity gradient.
![\[Fig5\] (color online) The transverse angular momentum density of left circularly polarized component. \[(a) and (b)\] Incident fields: $j_{ix+}^{HS}$ and $j_{ix+}^{HO}$. \[(c) and (d)\] Transmitted fields: $j_{tx+}^{HS}$ and $j_{tx+}^{HO}$. The cross section is chosen as $z_t=0$ and the intensity is plotted in normalized units. Other parameters are the same as those in Fig. \[Fig2\].](Fig5.eps){width="8cm"}
We proceed to analysis the angular momentum density for each of individual wave packet, which can be written as $$\mathbf{j}_a(\mathbf{r})=\mathbf{r}_a\times\mathbf{p}_a(\mathbf{r}_a)\label{AMD}.$$ Within the paraxial approximation, the angular momentum can be divided into the orbital and spin angular parts: $\mathbf{j}_a=\mathbf{j}_a^O+\mathbf{j}_a^S$ [@Allen1992; @Padgett1995], it follows that $$\mathbf{j}_a^O=\mathbf{r}_a\times\mathbf{p}_a^O\label{AMDO},$$ $$\mathbf{j}_a^S=\mathbf{r}_a\times\mathbf{p}_a^S\label{AMDS}.$$ It should be mentioned that this separation still hold beyond the paraxial approximation [@Barnett2002]. The transverse angular momentum density $j_x$ can be regarded as the combined contributions of orbital and spin parts: $$j_{ax}^O=y_a p_{az}^O-z_a p_{ay}^O\label{OAMDX},$$ $$j_{ax}^S=y_a p_{az}^S-z_a p_{ay}^S\label{SAMDX}.$$ The time-averaged linear and angular momenta can be obtained by integrating over the whole $x-y$ plane [@Jackson1999] $$J_{a}=\int\int j_{a}\text{d}x \text{d}y\label{OAM},$$ $$P_{a}=\int\int p_{a}\text{d}x \text{d}y\label{SAM}.$$ The transverse orbital angular momentum would appear if the spin-to-orbital angular momentum conversion occurs in the process of reflection and refraction. It should be mentioned that the transverse orbital angular momentum can lead to a transverse shift of field centroid even in free space [@Luo2010a]. In the plane $z_a=0$, after substituting Eqs (\[OAMDX\]) and (\[SAMDX\]) into Eqs (\[OAM\]) and (\[SAM\]) we obtain $$\Delta
y_{a}=\frac{J_{ax}^S}{P_{az}^S}=\frac{J_{ax}^O}{P_{az}^O}\label{SAMM}.$$ Figure \[Fig5\] plots the transverse spin and orbital momentum currents for incident and transmitted fields. It is clearly shows that the transverse orbital angular momentum plays a dominant role in the SHE of light. Compare Fig \[Fig5\](b) with (d), however, it is not very clear why the transmitted field exhibits the SHE of light. Thus, it is necessary to determine the transverse orbital angular momentum. From Eq. (\[OAM\]), we can obtain $j_{ix}^{O}=0$ and $J_{tx}^{O}\neq0$. The presence of transverse orbital angular momentum $J^O_{tx}$ provide a direct evidence for the spin-to-orbital angular momentum conversion. In the SHE of light associated with refraction [@Hosten2008], reflection [@Qin2009], and tunneling [@Luo2010b], the transverse spatial shifts are sensitive to the incident polarization angles.
We proceed to explore the role of refractive index gradient in spin-to-orbital angular momentum conversion. The $z$ component of total angular momentum $J_{az}$ for the $a$th beam can be represented as a sum of the extrinsic orbital angular momentum $J_{az}^O$ and the intrinsic spin angular momentum $J_{az}^S$, i.e., $J_{az}=J_{az}^O+J_{az}^S$ [@Bliokh2009]. The $z$ component of the orbital angular momenta are given by $J_{iz}=0$, $J_{rz}=-\Delta
y_r k_r \sin\theta_r$, and $J_{tz}=-\Delta y_t k_t \sin\theta_t$. for incident, reflected, and transmitted wave packets, respectively. In the laboratory Cartesian frame ($x,y,z$), the $z$ component of the orbital angular momenta are given by $$J_{iz\pm}^O=0\label{JIZO},$$ $$J_{rz\pm}^O=\pm\frac{(r_p^2+r_s^2-2r_p r_s)\cos\theta_i}{ r_p^2 +
r_s^2}\label{JRZO},$$ $$J_{zt\pm}^O=\pm\frac{(t_p^2+t_s^2-2\eta t_pt_s)\cos\theta_i}{ t_p^2
+t_s^2}\label{JTZO}.$$ The $z_a$ components of spin angular momentum for the $a$th beam is respectively described by $$J_{iz\pm}^S=\pm\cos\theta_i\label{JIZS},$$ $$J_{rz\pm}^S=\pm\frac{2 r_p r_s}{r_p^2 +
r_s^2}\cos\theta_r\label{JRZS},$$ $$J_{tz\pm}^S=\pm\frac{2 t_p t_s}{t_p^2 +
t_s^2}\cos\theta_t\label{JTZS}.$$ From Eqs. (\[JIZO\])-(\[JTZO\]) and Eqs. (\[JIZS\])-(\[JTZS\]), we find that the angular momenta fulfill the conservation law: $$Q_r J_{rz\pm}+ Q_t J_{tz\pm}=J_{iz\pm}.\label{totalam}$$ Here, $Q_r=(r_p^2+r_s^2)/2$ and $Q_t=n\eta(t_p^2+t_s^2)/2$ are the energy reflection and energy transmission coefficients, respectively. Note that the transverse angular shifts, which governed by the linear momentum conservation law, are not discussed in this paper.
![\[Fig6\] The refractive index gradient $\Delta{n}$ induces the spin-to-orbital angular momentum conversion. The incident wave packets are chosen as left circular polarization and right circular polarization. Other parameters are chosen to be the same as in Fig. \[Fig2\].](Fig6.eps){width="8cm"}
To obtain a clear physical picture of the SHE of light, the spin-to-orbital angular momentum conversion induced by refractive index gradient is depicted in Fig. \[Fig6\]. For the left circularly polarized component, the spin angular momentum $J^S_{tz+}$ monotonously increase while the orbital angular momentum $J^O_{tz+}$ monotonously decreases with the increase of the refractive index gradient $\Delta n$. For the right circular polarization, both the spin angular momentum $J^S_{tz-}$ and the orbital one $J^O_{tz-}$ present oppositive features. When the refractive index gradient continues increasing, the spin-to-orbital angular momentum conversion appears to reach a saturation value: $$J^O_{tz\pm}=\mp\frac{\cos\theta_i\sin^2\theta_i}
{1+\cos^2\theta_i}\label{JOTZM}.$$ This gives a good explanation of the intriguing phenomena why the transverse spatial shifts tends to saturation values in the Fig. \[Fig3\]. The metamaterial with $n=-1$ is a good candidate to eliminate the SHE of light in the refraction, since the spin-to-orbital angular momentum conversion is impressed completely. The spin-to-orbital angular momentum conversion can be enhanced in the region of low refractive index gradient as shown in Fig. \[Fig6\]. As a result, the transverse spatial shifts are also be amplified in this region. However, when the refractive index $|n|<\sin\theta_i$ the wave pack is totally reflected. By properly facilitating the spin-to-orbital angular momentum conversion the SHE may be enhanced dramatically.
We now give a very simple way to understand how the refractive index gradient enhance the spin-to-orbital angular momentum conversion in the SHE of light. We attempt to perform analyses on the $z$ component of the total angular momentum for each of individual photons, i.e., $J_{iz}=J_{tz}$ [@Onoda2006; @Nasalski2006]. The total angular momentum conservation law for single photon is given by $$J^O_{tz\pm}\pm\cos\theta_{t}=\pm\cos\theta_i,$$ where $J^O_{tz\pm}=-\Delta y_{t\pm} k_t \sin\theta_t$. When the photons penetrate from air into a low-refractive-index medium ($\Delta n<0$), the incident angle is less than the transmitted angle $\theta_i<\theta_t$. For the left circularly polarized photons, the $z_t$ component of spin angular momentum $+\cos\theta_{t}$ decreases after entering the medium. Because of the conservation law, the total angular momentum must remain unchanged. To conserve the total angular momentum, the photons must move to the direction $-y$ ($\Delta y_{t+}<0$) and thus generate a positive orbital angular momentum ($J_{tz+}^O>0$). For the right circularly polarized photons, the $z$ component of spin angular momentum $-\cos\theta_t$ increases. In this case, the photons must move to the direction $+y$ ($\Delta y_{t-}>0$) and induce a negative orbital angular momentum ($J_{tz-}^O<0$). When the photons enter into a high-refractive-index medium ($\Delta n>0$), the incident angle is larger than the transmitted angle $\theta_i>\theta_t$. As a result, the orbital angular momentum reverses its direction.
Conclusions
===========
In conclusion, we have identified the role of spin-to-orbital angular momentum conversion in SHE of light. We have demonstrated that the refractive index gradient can enhance or suppress the spin-to-orbital angular momentum conversion, and thus can control the SHE of light. The recent advent of metamaterial whose refractive index can be tailored arbitrarily seems to be an available candidate to amplify or eliminate the SHE of light. However, the spin-to-orbital angular momentum conversion in a ultra-large refractive index gradient is limited by a saturation value. Fortunately, the SHE of light can be dramatically amplified by plasmonic nanostructure [@Gorodetski2008; @Gorodetski2009; @Vuong2010; @Herrera2010]. In addition, the SHE of light can also be noticeably enhanced when the beam carries orbital angular momentum [@Bliokh2006b; @Okuda2006; @Fadeyeva2009; @Merano2010]. Hence, the exploration of spin-to-orbital angular momentum conversion in these systems would be very interesting. The transverse spatial shifts governed by the spin-to-orbital angular momentum conversion, give us a clear physical picture to clarify the role of refractive index gradient in the SHE of light. These findings provide a pathway for modulating the SHE of light, and thereby open the possibility for developing new nano-photonic devices. Because of the close similarity in optical physics [@Hosten2008; @Bliokh2008], condensed matter [@Murakami2003; @Sinova2004; @Wunderlich2005], and high-energy physics [@Berard2006; @Gosselin2007], by properly facilitating the spin-to-orbital angular momentum conversion, the SHE may be enhanced dramatically in these physical systems.
One of authors (H.L.) thanks Dr. W. Löffler for helpful discussions. This research was partially supported by the National Natural Science Foundation of China (Grants Nos. 61025024, 11074068, and 10904036).
O. Hosten and P. Kwiat, Science **319**, 787 (2008).
K. Y. Bliokh, A. Niv, V. Kleiner, and E. Hasman, Nature Photon. **2**, 748 (2008).
S. Murakami, N. Nagaosa, and S. C. Zhang, Science **301**, 1348 (2003).
J. Sinova, D. Culcer, Q. Niu, N. A. Sinitsyn, T. Jungwirth, and A. H. MacDonald, Phys. Rev. Lett. **92**, 126603 (2004).
J. Wunderlich, B. Kaestner, J. Sinova, and T. Jungwirth, Phys. Rev. Lett. **94**, 047204 (2005).
F. I. Fedorov, Dokl. Akad. Nauk SSSR **105**, 465 (1955).
C. Imbert, Phys. Rev. D **5**, 787 (1972).
M. Onoda, S. Murakami, and N. Nagaosa, Phys. Rev. Lett. **93**, 083901 (2004).
K. Y. Bliokh and Y. P. Bliokh, Phys. Rev. Lett. **96**, 073903 (2006).
D. Haefner, S. Sukhov, and A. Dogariu, Phys. Rev. Lett. **102**, 123903 (2009).
A. Aiello, N. Lindlein, C. Marquardt, and G. Leuchs, Phys. Rev. Lett. **103**, 100401 (2009).
J.-M. Ménard, A. E. Mattacchione, H. M. van Driel, C. Hautmann, and M. Betz, Phys. Rev. B **82**, 045303 (2010).
R. Y. Chiao and Y. S. Wu, Phys. Rev. Lett. **57**, 933 (1986).
A. Tomita and R. Y. Chiao, Phys. Rev. Lett. **57**, 937 (1986).
A. V. Dooghin, N. D. Kundikova, V. S. Liberman, and B. Y. Zeldovich, Phys. Rev. A **45**, 8204 (1992).
V. S. Liberman and B. Y. Zeldovich, Phys. Rev. A **46**, 5199 (1992).
L. Marrucci, C. Manzo, and D. Paparo, Phys. Rev. Lett. **96**, 163905 (2006).
Y. Zhao, J. S. Edgar, G. D. M. Jeffries, D. McGloin, and D. T. Chiu, Phys. Rev. Lett. **99**, 073901 (2007).
J. W. Goodman, *Introduction to Fourier Optics* (McGraw-Hill, New York, 1996).
M. Lax, W. H. Louisell, and W. McKnight, Phys. Rev. A **11**, 1365 (1975).
R. A. Beth, Phys. Rev. **50**, 115 (1936).
K. Y. Bliokh and Y. P. Bliokh, Phys. Rev. E **75**, 066609 (2007).
A. Aiello and J. P. Woerdman, Opt. Lett. **33**, 1437 (2008).
C. Menzel, C. Rockstuhl, T. Paul, S. Fahr, and F. Lederer, Phys. Rev. A **77**, 013810 (2008).
H. Luo, S. Wen, W. Shu, Z. Tang, Y. Zou, and D. Fan, Phys. Rev. A **80**, 043810 (2009).
J. D. Jackson, *Classical Electrodynamics* (Wiley, New York, 1999).
C. M. Krowne, Phys. Lett. A **373**, 466 (2009).
F. Goos and H. Hächen, Ann. Phys. **1**, 333 (1947).
M. Merano, A. Aiello, G. W. ’t Hooft, M. P. van Exter, E. R. Eliel, and J. P. Woerdman, Opt. Express **15**, 15928 (2007).
W. Löfler, M. P. van Exter, G. W. ’t Hooft, E. R. Eliel, K. Hermans , D. J. Broer, J. P. Woerdman, Opt. Commun. **283**, 3367 (2010).
M. Merano, A. Aiello, M. P. van Exter, and J. P. Woerdman, Nature Photon. **3**, 337 (2009).
A. Aiello, M. Merano, and J. P. Woerdman, Phys. Rev. A **80**, 061801(R) (2009).
C. Duval, Z. Horváth, and P. A. Horváthy, Phys. Rev. D **74**, 021701(R) (2006).
J. Shin, J. T. Shen, and S. Fan, Phys. Rev. Lett. **102**, 093903 (2009).
D. R. Smith, J. B. Pendry, and M. C. K. Wiltshire, Science **305**, 788 (2004).
J. Pendry, D. Schurig, and D. Smith, Science **312**, 1780 (2006).
J. B. Pendry, Phys. Rev. Lett. **85**, 3966 (2000).
A. Y. Bekshaev and M. S. Soskin, Opt. Commun. **271**, 332 (2007).
M. V. Berry, J. Opt. A: Pure Appl. Opt. **11**, 094001 (2009).
L. Allen, M. W. Beijersbergen, R. J. C. Spreeuw, and J. P. Woerdman, Phys. Rev. A **45**, 8185 (1992).
M. J. Padgett and L. Allen, Opt. Commun. 121, **36** (1995).
S. M. Barnett, J. Opt. B **4**, S7 (2002).
H. Luo, S. Wen, W. Shu, and D. Fan, Phys. Rev. A **81**, 053826 (2010).
Y. Qin, Y. Li, H. Y. He, and Q. H. Gong, Opt. Lett. **34**, 2551 (2009).
H. Luo, S. Wen, W. Shu, and D. Fan, Phys. Rev. A **82**, 043825 (2010).
K. Y. Bliokh, I. V. Shadrivov, and Y. S. Kivshar, Opt. Lett. **34**, 389 (2009).
M. Onoda, S. Murakami, and N. Nagaosa, Phys. Rev. E **74**, 066610 (2006).
W. Nasalski, Phys. Rev. E **74**, 056613 (2006).
Y. Gorodetski, A. Niv, V. Kleiner, and E. Hasman, Phys. Rev. Lett. **101**, 043903 (2008).
Y. Gorodetski, N. Shitrit, I. Bretner, V. Kleiner, and E. Hasman, Nano Lett. **9**, 3016 (2009).
L. T. Vuong, A. J. L. Adam, J. M. Brok, P. C. M. Planken, and H. P. Urbach, Phys. Rev. Lett. **104**, 083903 (2010).
O. G. Rodríguez-Herrera, D. Lara, K. Y. Bliokh, E. A. Ostrovskaya, and C. Dainty, Phys. Rev. Lett. **104**, 253601 (2010).
K. Y. Bliokh, Phys. Rev. Lett. **97**, 043901 (2006).
H. Okuda and H. Sasada, Opt. Express **14**, 8393 (2006).
T. A. Fadeyeva, A. F. Rubass, and A. V. Volyar, Phys. Rev. A **79**, 053815 (2009).
M. Merano, N. Hermosa, J. P. Woerdman, and A. Aiello, Phys. Rev. A **82**, 023817 (2010).
A. Bérard and H. Mohrbach, Phys. Lett. A **352**, 190 (2006).
P. Gosselin, A. Bérard, and H. Mohrbach, Phys. Rev. D **75**, 084035 (2007).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
We describe the equation of motion of two charged spherical shells with tangential pressure in the field of a central Reissner-Nordstrom (RN) source. We solve the problem of determining the motion of the two shells *after* the intersection by solving the related Einstein-Maxwell equations and by requiring a physical continuity condition on the shells velocities.
We consider also four applications: post-Newtonian and ultra-relativistic approximations, a test-shell case, and the ejection mechanism of one shell.
This work is a direct generalization of Barkov-Belinski-Bisnovati-Kogan paper.
address:
- |
Physics Department, Rome University “La Sapienza”,\
Piazzale A. Moro, 00185 Rome, Italy, and\
ICRANet, Pescara, Italy 65122.\
[email protected]
- |
Physics Department, Rome University “La Sapienza”,\
Piazzale A. Moro, 00185 Rome, Italy.
author:
- 'M. PIZZI'
- 'A. PAOLINO'
title: 'Intersections of self-gravitating charged shells in a Reissner-Nordstrom field'
---
Introduction
============
The mathematical model that we analyze in this paper describes the dynamic evolution of two spherical shells of charged matter which freely move outside the field of a central Reissner-Nordstrom (RN) source. Microscopically these shells are assumed to be composed by charged particles which move on elliptical orbits with a collective variable radius. The angular motion, distributed uniformly and isotropically on the shell surfaces, is mathematically described by a tangential-pressure term in the energy momentum tensor of the Einstein equations. The definition of the shell implies that all the particles have the same following three ratios: energy/mass, angular momentum/mass, and charge/mass. Indeed, since the equations of motion for any singled-out particle “a” are $$\begin{aligned}
& \frac{dt_a}{ds}= \frac{1}{-m_a c^2 g_{tt}(r_a)}(E_a+e_a A_0(r_a)) \label{mot1} \\
& \left(\frac{dr_a}{ds}\right)^{2}=\frac{1}{m_a^2 c^4}(E_a+e_a A_0(r_a))^2\left(\frac{1}{-g_{tt}(r_a)g_{rr}(r_a)}\right) - \left(\frac{l_a^2}{m_a^2c^2}\frac{1}{r^2}+1\right) \frac{1}{g_{rr}(r_a)} \label{mot2}\\
& \left(\frac{d\theta_a}{ds}\right)^2=\frac{l_a^2}{m_a^2c^2}\frac{1}{r^4}-\frac{k_a^2}{m_a^2c^2}\frac{1}{r^4 \sin^2\theta_a} \label{mot3}\\
& \frac{d\varphi_a}{ds}=\frac{k_a}{m_a c}\frac{1}{r^2 \sin^2\theta_a}\end{aligned}$$ ($g_{tt}$ and $g_{rr}$ are the components of a spherical symmetric metric and $A_0$ is the electric potential; $k_a$ and $l_a$ are arbitrary constants), it is easy to see that the radial motion for all particles is the same if $$\begin{aligned}
\frac{E_a}{m_a}=const, \ \ \ \ \ \frac{e_a}{m_a}=const, \ \ \ \ \ \frac{|l_a|}{m_a}=const, \ \ \ \ \ \forall a, \end{aligned}$$ where each $const.$ does not depend on the index $a$. Therefore, if at the beginning the particles are on the same radius $r_a=R_0$, then the shell will evolve “coherently”, i.e. all particles will evolve with the same radius.
Now the problem we are interested in is to find the exchange of energy between the two shells after the intersection. Indeed the motion of the shells before and after the crossing can be easily deduced from the equation of motion for just one shell, which equation has been found many years ago by Chase[@Cha] with a geometrical method first used by Israel[@Isr]. Instead, the intersection problem was considered first by Langer-Eid[@LE]: they applied it to the particular case of electrically neutral and pressureless shells (neutral dust), then Barkov et al.[@BBB] considered the more general case of shells intersections when the shells have also tangential pressure (so Langer-Eid’s results follows from Barkov et al. results as particular case). Now we generalize this problem for the shells with tangential pressure and also electric charge.
What we achieve in the present paper is the determination of the constant parameters after the intersection knowing just the parameters before the intersection. Actually the unknown parameter is only one, ${m_{21}}$, which is the Schwarzschild mass parameter measured by an observer between the shells after the intersection. This parameter is strictly related to the energy transfer which takes place in the crossing, and it is found imposing a proper continuity condition on the shells velocities.
In the model we assume that there are no other interactions between the two shells apart the gravitational and electrostatic ones. In particular the shells, during the intersection, are assumed to be “transparent” each other (i.e. no scattering processes).
The paper is divided as follows: in Sec.2 we preliminarily discuss the one-shell case; in Sec.3, which is the central part of this article, we find the unknown parameter $m_{21}$; then, Secs.4-7 are devoted to some applications: post-Newtonian approximation, zero effective masses case (i.e. ultra-relativistic case), test-shell case, and finally the ejection mechanism.
In this paper we deal only with the mathematical aspects of the problem; some astrophysical applications of charged shells in the field of a RN black hole have been considered in Ref.\[\].
A gravitating charged shell with tangential pressure
====================================================
The motion of a thin charged dust-shell with a central RN singularity was firstly studied by De La Cruz and Israel[@DI], while the case with tangential pressure was achieved by Chase[@Cha] in 1970. All these authors used the extrinsic curvature tensor and the Gauss-Codazzi equations. However we followed a different way, indeed the same solution can be found also by using $\delta$ and $\theta$ distributions and then by direct integration of the Einstein-Maxwell equations (see Ref.\[\] and the appendix in Ref.\[\]). This method has the advantage of a clearer physical interpretation, and it is also straightforward in the calculations; however in the following we will give only the main passages.
Let there be a central body of mass $m_{in}$ and charge $e_{in}$ and let a spherical massive charged shell with charge $e$ move outside this body. It is clear in advance that the field internal to the shell will be RN, while externally we will have again a RN metric but with different mass and charge parameters $m_{out}$ and $e_{out}={e_{in}}+e$. Using the coordinates $x^0=ct$ and $r$, which are continuous when passing through the shell, we can write the intervals inside, outside, and on the shell as
$$\begin{aligned}
&-(ds)_{in}^{2}=-e^{T(t)} f_{in}(r)c^2dt^2+f_{in}^{-1}(r)dr^2 +r^2d\Omega^2 \label{ds1} \\
&-(ds)_{out}^{2}=-f_{out}(r)c^2dt^2+f^{-1}_{out}(r) dr^2 +r^2d\Omega^2 \label{ds2}\\
&-(ds)_{on}^{2}=-c^2d\tau^2+r_0(\tau)^2d\Omega^2 \label{ds3}\end{aligned}$$
where we denoted $$d\Omega^2=d\theta^2+\sin^2\theta d\phi^2 \nonumber$$ and $$\label{4}
{f_{in}}=1-2\frac{G\,m_{in}}{c^2r}+\frac{G e_{in}^2}{c^4r^2}\,,\ \ \
{f_{out}}=1-2\frac{G\,m_{out}}{c^2r}+\frac{G({e_{in}}+e)^2}{c^4r^2}.$$ In the interval (\[ds3\]), $\tau$ is the proper time of the shell. The “dilaton” factor $e^{T(t)}$ in (\[ds1\]) is required to ensure the continuity of the time coordinate $t$ through the shell. If the equation of motion for the shell is $$\label{rR}
r=R_0(t),$$ then joining the angular part of the three intervals (\[ds1\])-(\[ds3\]), one has $$\label{rR2}
r_0(\tau)=R_0[t(\tau)],$$ where the function $t(\tau)$ describes the relationship between the global time and the proper time of the shell. Joining the radial-time parts of the intervals (\[ds1\])-(\[ds2\]) on the shell requires that the following relations hold: $${f_{in}}(r_0)\left({\frac{d\,t}{d\tau}}\right)^2e^{T(t)}-{f_{in}}^{-1}(r_0)\left({\frac{d\,r_0}{cd\tau}}\right)^2=1\,, \label{6}$$ $${f_{out}}(r_0)\left({\frac{d\,t}{d\tau}}\right)^2-{f_{out}}^{-1}(r_0)\left({\frac{d\,r_0}{cd\tau}}\right)^2=1\,. \label{7}$$ If the equation of motion for the shell —i.e. the function $r_0(\tau)$— is known, then the function $t(\tau)$ follows from (\[7\]) and consequently $T(t)$ can be deduced by (\[6\]). Thus the problem consist only in determining $r_0(\tau)$, which can be done by direct integration of the Einstein-Maxwell equations $$\left\{\begin{array}{l}\label{A.1}
R_i^k-\frac{1}{2}R g_i^k=\frac{8\pi G}{c^4}T_i^k\\\\
(\sqrt{-g}F^{ik})_{,k}=\sqrt{-g}\frac{4\pi}{c} \rho u^{i}
\end{array}
\right.$$ with the energy-momentum tensor given by: $$\begin{aligned}
\label{A.2}
&T_i^k=\epsilon \,u_i u^k+(\delta_i^2\delta_2 ^k+\delta_i^3\delta_3^k)p+T^{(el)}\,_i^k
\\
&T^{(el)}\,_i^k =\frac{1}{4\pi}(F_{il}F^{kl}-\frac{1}{4} \delta_i^kF_{lm}F^{lm})\ .\end{aligned}$$
Here on we employ the following notations: $$\nonumber
\begin{array}{ll}
-ds^2=g_{ik} dx^idx^k , & g_{ik} \ $has signature$\ (-,+,+,+)\\\\
x^k=(ct,r,\theta,\varphi) & \ \ i,j,k...=0,1,2,3\\\\
p\equiv p(R_0)= p_{\theta}=p_{\varphi} = $tangential pressure$ & \ \ \ (p_r=0)\\\\
F_{ik}=A_{k,i}-A_{i,k}
\end{array}$$ The above equations are to be solved for the metric $$\label{8}
-ds^{2}=g_{00}(t,r)c^2dt^2+g_{11}(t,r)dr^2 +r^2d\Omega^2,$$
and for the potential $$\label{10}
A_0=A_0(t,r),\ \ A_1=A_2=A_3=0,$$
As follows from the Landau-Lifshitz approach \[\] (see Ref.\[\]) the energy distribution of the shell is $$\label{11}
{\epsilon}=\frac{M(t)c^2\delta[r-R_0(t)]}{4\pi r^2 u^0 \sqrt{-g_{00}g_{11}}},$$ while its charge density is $$\label{12}
\rho=\frac{c\,e \delta[r-R_0(t)]}{4\pi r^2u^0\sqrt{-g_{00}g_{11}}},$$ where $\delta$ is the standard $\delta$-function. In the absence of tangential pressure $p$, the quantity $M$ in Eqn.(\[11\]) would be a constant, but in presence of pressure, $Mc^2$ includes the rest energy along with the energy (in the radially comoving frame) of the tangential motions of the particles that produce this pressure.
It can be checked that the Einstein part of (\[A.1\]) actually lead to the solution (\[ds1\])-(\[ds3\]) with, in addition, the “joint condition” $$\label{joint}
\sqrt{{f_{in}}(r_0)+\left({\frac{d\,r_0}{cd\tau}}\right)^2}+\sqrt{{f_{out}}(r_0)+\left({\frac{d\,r_0}{cd\tau}}\right)^2}=2\frac{({m_{out}}-{m_{in}})}{\mu(\tau)}
-\frac{e^2+2ee_{in}}{\mu(\tau)c^2 r_0},$$ where we denoted $$\label{mu}
\mu(\tau)=M[t(\tau)],$$ while $$\label{Emo}
{m_{out}}-{m_{in}}=E/c^2$$ is a constant which can be interpreted as the total amount of energy of the shell. Then, from the Maxwell side of (\[A.1\]) the only non-vanishing component of the electric field is $$\label{13}
F_{01}=-\frac{\sqrt{-g_{00}g_{11}}}{r^2}\{e_{in}+e\theta[r-R_0(t)]\}$$ ($\theta(x)$ is the standard step function). Finally, the equations $T^k_{i;k}=0$ can be reduced to the only one relation: $$\label{14}
p=-\frac{dM}{dt}\frac{c^2\delta[r-R_0(t)]}{8\pi r u^1\sqrt{-g_{00}g_{11}}}$$
We will not treat here the steady case (i.e. $r_0=const$) which should be treated separately; thus in the following we will assume always $r_0\neq const.$ .
The joint condition (\[joint\]) can be written in several different forms: two of them, which will be useful in the following, are $$\label{18}
\sqrt{{f_{in}}(r_0)+\left({\frac{d\,r_0}{cd\tau}}\right)^2}=\frac{({m_{out}}-{m_{in}})}{\mu(\tau)}
+\frac{G\mu^2(\tau)-e^2-2ee_{in}}{2\mu(\tau)c^2 r_0}$$ and $$\label{19}
\sqrt{{f_{out}}(r_0)+\left({\frac{d\,r_0}{cd\tau}}\right)^2}=\frac{({m_{out}}-{m_{in}})}{\mu(\tau)}
-\frac{G\mu^2(\tau)+e^2+2ee_{in}}{2\mu(\tau)c^2 r_0}.$$ As in Ref. , all the radicals encountered here are taken positive, since for astrophysical considerations only these cases are meaningful. To proceed further, we must specify the equation of state, i.e. the function $\mu(\tau)$. Here we consider a particle-made shell, therefore the sum of kinetic and rest energy of all the particles is $$\label{21}
M c^2=\sum_a \left(m_ac^2\sqrt{1+\frac{p_a^2}{m_a^2c^2}}\right),$$ where $p_a$ is the tangential momentum of each particle (the electric interaction between the particles is already taken into account by the self-energy term of, e.g., (\[18\]), thus one has not to include it in $M$ too). From the definition of the shell (see Introduction) it follows: $$\label{22}
\frac{p_a^2}{m_a^2}=\frac{l_a^2}{m_a^2R_0^2}=\frac{const}{R_0^2},$$ the square root in (\[21\]) does not depend on the index $a$; then defining $$\nonumber
\sum_a{m_ac^2}=m c^2,\ \ \ \sum_a{|l_a|}=L,$$ formula (\[21\]) can be re-written (remembering definition (\[mu\]) too) as $$\label{24}
\mu(\tau)=\sqrt{m^2+\frac{L^2}{c^2 r_0^2(\tau)}}.$$ Thus, now, one can determine the function $r_0(\tau)$ from equation (\[joint\]) \[or from one of the equivalent forms (\[18\])-(\[19\])\] if the initial radius of the shell and the six arbitrary constants ${m_{in}},\,{m_{out}},\,m,\,e_{in},\,e$ and $L$ are specified. Accordingly with (\[11\]), (\[14\]), (\[mu\]) and (\[24\]), the equation of state that relates the shell energy density ${\epsilon}$ to the tangential pressure $p$ is $$\label{p}
p=\frac{{\epsilon}}{2}\frac{L^2}{m^2c^2R_0^2}\left(1+\frac{L^2}{m^2c^2R^2_0}\right)^{-1}$$ as in the uncharged case, i.e. the presence of the charges do not modify the relation between energy density and pressure (indeed the presence of the charge is hidden in the equation of motion). Note that when the shell expands to infinity ($R_0\rightarrow\infty$) the angular momentum becomes irrelevant and the equation of state tends to the dust case $p<<{\epsilon}$.
The shells intersection {#inters}
=======================
Let us now consider the case of two shells which move in the field of a central charged mass. The generalization from the previous (single-shell) case is straightforward if the shells do not intersect: indeed the outer shell do not affect the motion of the inner one, while the inner one appears from outside just as a RN metric. Therefore the principal aim of this section is to consider the intersection eventuality and to predict the motion of the two shells *after* the crossing, having specified the initial conditions before the crossing. After the intersection one has a new unknown constant that has to be found by imposing opportune joining conditions as now we are going to explain (the analysis follows step by step the Ref.’s one).
![The four region in which it is divided the spacetime; the two lines represent the trajectories of shell-1 and shell-2.[]{data-label="f.1"}](ABCDO){width="9cm" height="7cm"}
Let us previously analyze the space-time in the $(t,r)$ coordinates (which are continuous through the shells). We define the point $O\equiv (t_*,r_*)$ as the intersection point; then the space-time is divided in four regions (see Fig.1): $$\begin{array}{lr}
COB & (r>R_1,r>R_2),\\
COA & (R_1<r<R_2), \\
AOD & (r<R_1,r<R_2), \\
BOD & (R_2<r<R_1).
\end{array}$$ Correspondingly to these regions we have the metric in form (\[7\]) but with different coefficients $g_{00}$ and $g_{11}$: $$\begin{aligned}
&g_{00}^{(COB)}=-{f_{out}}(r) \,, &g_{11}^{(COB)}={f_{out}}^{-1}(r) \label{25}\\
&g_{00}^{(COA)}=-e^{T_1(t)}f_{12}(r) \,, &g_{11}^{(COA)}=f_{12}^{-1}(r) \label{26}\\
&g_{00}^{(AOD)}=-e^{T_0(t)}f_{in}(r)\,, &g_{11}^{(AOD)}=f_{in}^{-1}(r) \label{27}\\
&g_{00}^{(BOD)}=-e^{T_2(t)}f_{21}(r)\,, &g_{11}^{(BOD)}=f_{21}^{-1}(r) \label{28}\end{aligned}$$ The dilaton factor $T(t)$ allows to cover all the space-time with only one $t$-coordinate; here, ${f_{in}}$ and ${f_{out}}$ are the same as those in (\[4\]) while ${f_{12}}$ and ${f_{21}}$ are given by similar expressions: $$\begin{aligned}
{f_{12}}=1-\frac{2G{m_{12}}}{c^2r}+\frac{G({e_{in}}+e_1)^2}{c^4r^2} \label{29}\\
{f_{21}}=1-\frac{2G{m_{21}}}{c^2r}+\frac{G({e_{in}}+e_2)^2}{c^4r^2} \label{29b}\end{aligned}$$ As we said, the parameters ${m_{in}},\,{m_{12}},\,{m_{out}},\,e_{in},\,e_1,\,e_2$ are assumed to be specified at the beginning, while ${m_{21}}$ is the actual unknown constant which has yet to be determined from the joining conditions on $(t_*,r_*)$.
Before the intersection {#sec:BeforeTheIntersection .unnumbered}
-----------------------
Let us write the equation of motion for the two shells before the intersection (shell-1 inner and shell-2 outer). This can be made easily adapting the (\[19\]) and (\[18\]) to the present case: $$\label{30}
\sqrt{{f_{12}}(r_1)+\left({\frac{d\,r_1}{cd\tau_1}}\right)^2}=\frac{({m_{12}}-{m_{in}})}{M_1}
-\frac{GM_1^2+e_1^2+2e_{in}e_1}{2M_1c^2 r_1}$$ for shell 1, while for shell 2 $$\label{31}
\sqrt{{f_{12}}(r_2)+\left({\frac{d\,r_2}{cd\tau_2}}\right)^2}=\frac{({m_{12}}-{m_{in}})}{M_2}
+\frac{GM_2^2-e_2^2-2(e_{in}+e_1)e_2}{2M_2c^2 r_2}$$ with $$\label{32}
M_1=\sqrt{m_1^2+\frac{L_1^2}{c^2 r_1^2}}, \ \ \ M_2=\sqrt{m_2^2+\frac{L_2^2}{c^2 r_2^2}}.$$ Here, $\tau_1$ and $\tau_2$ are the proper times of the first and second shells respectively, while $r_1(\tau_1)=R_1[t(\tau_1)]$ and $r_2(\tau_2)=R_2[t(\tau_2)]$. Now we have to impose the joining conditions for the intervals on both the shells. For the first shell (on curve AO) one has: $$\label{33}
e^{T_1(t)}{f_{12}}(r_1)\left({\frac{d\,t}{d\tau_1}}\right)^2-{f_{12}}^{-1}(r_1)\left({\frac{d\,r_1}{cd\tau_1}}\right)^2=1$$ $$\label{34}
e^{T_0(t)}{f_{in}}(r_1)\left({\frac{d\,t}{d\tau_1}}\right)^2-{f_{in}}^{-1}(r_1)\left({\frac{d\,r_1}{cd\tau_1}}\right)^2=1;$$ while for the second shell: $$\label{35}
{f_{out}}(r_2)\left({\frac{d\,t}{d\tau_2}}\right)^2-{f_{out}}^{-1}(r_2)\left({\frac{d\,r_2}{cd\tau_2}}\right)^2=1$$ $$\label{36}
e^{T_1(t)}{f_{12}}(r_2)\left({\frac{d\,t}{d\tau_2}}\right)^2-{f_{12}}^{-1}(r_2)\left({\frac{d\,r_2}{cd\tau_2}}\right)^2=1.$$
If all free parameters and initial data to Eqs.(\[30\])-(\[32\]) were specified and if the functions $r_1(\tau_1)$ and $r_2(\tau_2)$ were derived, then their substitution in (\[33\])-(\[36\]) gives the functions $\tau_1(t)$, $\tau_2(t)$ and $T_1(t)$, $T_0(t)$, which is enough for determining the motion of the shells before the intersection. Therefore the intersection point $(t_*,r_*)$ can be found by solving the system $$\label{37}
\left\{\begin{array}{l}
r_*=r_1(\tau_1(t_*))\\
r_*=r_2(\tau_2(t_*))\ ,
\end{array}
\right.$$ which we assume that has a solution.
After the intersection {#sec:AfterTheIntersection .unnumbered}
----------------------
The equation of motion for the shells after the intersection time $t_*$ can be constructed in the same way again by turning to Eqns.(\[18\]) and (\[19\]), and introducing the new parameter ${m_{21}}$ which characterize the “Schwarschild mass” seen by an observer in the region $BOD$. We use Eq.(\[18\]) for (now outer) shell 1 and Eq.(\[19\]) for (now inner) shell 2: $$\label{38}
\sqrt{{f_{21}}(r_1)+\left({\frac{d\,r_1}{cd\tau_1}}\right)^2}=\frac{({m_{out}}-{m_{21}})}{M_1}
+\frac{GM_1^2-e_1^2-2e_1(e_{in}+e_2)}{2M_1c^2 r_1} \ ,$$
$$\label{39}
\sqrt{{f_{21}}(r_2)+\left({\frac{d\,r_2}{cd\tau_2}}\right)^2}=\frac{({m_{21}}-{m_{in}})}{M_2}
-\frac{GM_2^2+e_2^2+2e_2e_{in}}{2M_2c^2 r_2} \ .$$
Naturally, $M_1(r_1)$ and $M_2(r_2)$ are given by the same expression of (\[32\]) but now they have to be calculated on $r_1(\tau_1)$ and $r_2(\tau_2)$ after the intersection.
Joining the intervals on the first shell (on curve $OB$) yields $$\label{40}
{f_{out}}(r_1)\left({\frac{d\,t}{d\tau_1}}\right)^2-{f_{out}}^{-1}(r_1)\left({\frac{d\,r_1}{cd\tau_1}}\right)^2=1$$ $$\label{41}
e^{T_2(t)}{f_{21}}(r_1)\left({\frac{d\,t}{d\tau_1}}\right)^2-{f_{21}}^{-1}(r_1)\left({\frac{d\,r_1}{cd\tau_1}}\right)^2=1.$$ Then, joining the second shell (on curve $OB$) we obtain: $$\label{42}
e^{T_2(t)}{f_{21}}(r_2)\left({\frac{d\,t}{d\tau_2}}\right)^2-{f_{21}}^{-1}(r_2)\left({\frac{d\,r_2}{cd\tau_2}}\right)^2=1$$ $$\label{43}
e^{T_0(t)}{f_{in}}(r_2)\left({\frac{d\,t}{d\tau_2}}\right)^2-{f_{in}}^{-1}(r_2)\left({\frac{d\,r_2}{cd\tau_2}}\right)^2=1.$$ Since the initial data to Eqs.(\[38\]) and (\[39\]) have already been specified (from the previous evolution), then the evolution of the shells after the intersection would be determined from Eqs.(\[38\])-(\[43\]) if parameter ${m_{21}}$ were known. Thus we need an additional physical condition from which we could determine ${m_{21}}$.
This condition follows from the fact that the Christoffel symbols (i.e. the accelerations) of the shells have only finite discontinuities (finite jumps), therefore the relative velocity of the shells must remain continuous through the crossing point.
In the presence of two shells, we can construct one more invariant than in the single shell case (where only $u_i u^i=-1$ was possible): the scalar product between the two 4-velocities of the shells. We can also avoid to apply the parallel transport if we evaluate the 4-velocities on the intersection point $(t_*,r_*)$. The continuity condition can be found imposing that the scalar product has to have the same value when evaluated in both the two limits $t\rightarrow t_*^-$ and $t\rightarrow t_*^+$.
Determination of Q. {#determination-of-q. .unnumbered}
-------------------
Let us start determining the quantity $$\label{52}
Q\equiv \{g_{00}^{(COA)}u^0_{AO}u^0_{CO}+g_{11}^{(COA)}u^1_{AO}u^1_{CO}\}_{t=t^*,r=r_1=r_2=r_*},$$ which is the scalar product of the two 4-velocities evaluated in the intersection point from the region $AOC$ (along the curves $AO$ and $CO$). Written explicitly, the unit tangent vector to trajectory $AO$ is
$$\begin{aligned}
u^i_{AO}&=(u^0_{AO},u^1_{AO},u^2_{AO},u^3_{AO}) \nonumber \\
&=\left({\frac{d\,t}{d\tau_1}},{\frac{d\,r_1}{cd\tau_1}},0,0\right)_{t\leq t_*}, \label{44}\end{aligned}$$
while for the trajectory $CO$ we have
$$\begin{aligned}
u^i_{CO}&=(u^0_{CO},u^1_{CO},u^2_{CO},u^3_{CO})\nonumber\\
&=\left({\frac{d\,t}{d\tau_2}},{\frac{d\,r_2}{cd\tau_2}},0,0\right)_{t\leq t_*}.\label{45}\end{aligned}$$
The fact that these are actually unit vectors follows from the joining equations (\[33\]) and (\[36\]).
The components of the vector (\[44\]) can be easily expressed from Eqs.(\[30\]) and (\[33\]) as $$\label{46}
\left({\frac{d\,t}{d\tau_1}}\right)_{t\leq t_*}=\frac{e^{-T_1(t)/2}}{M_1(r_1){f_{12}}(r_1)}\left({m_{12}}-{m_{in}}-\frac{GM_1^2(r_1)+e_1^2+2e_1e_{in}}{2c^2r_1}\right)$$
$$\begin{aligned}
\left({\frac{d\,r_1}{cd\tau_1}}\right)_{t\leq t_*}&= \nonumber\\
=&\frac{\delta_1}{M_1(r_1){f_{12}}(r_1)}\sqrt{\left({m_{12}}-{m_{in}}-\frac{GM_1^2(r_1)+e_1^2+2e_1e_{in}}{2c^2r_1}\right)^2-M_1^2(r_1){f_{12}}(r_1)} \label{47}
\end{aligned}$$
where $$\label{48}
\delta_1=\texttt{sgn}\left({\frac{d\,r_1}{cd\tau_1}}\right)_{t\leq t_*}.$$ Analogously, for the components of vector (\[45\]), we obtain the following expressions from Eqs.(\[31\]) and (\[36\]): $$\label{49}
\left({\frac{d\,t}{d\tau_2}}\right)_{t\leq t_*}=\frac{e^{-T_1(t)/2}}{M_2(r_2){f_{12}}(r_2)}\left({m_{out}}-{m_{12}}+\frac{GM_2^2(r_2)-e_2^2-2e_2(e_{in}+e_1)}{2c^2r_2}\right)$$
$$\begin{aligned}
\left({\frac{d\,r_2}{cd\tau_2}}\right)_{t\leq t_*}&= \frac{\delta_2}{M_2(r_2){f_{12}}(r_2)}\, \cdot \nonumber \\
\cdot \, & \sqrt{\left({m_{out}}-{m_{12}}+\frac{GM_2^2(r_2)-e_2^2-2e_2(e_{in}+e_1)}{2c^2r_2}\right)^2-M_2^2(r_2){f_{12}}(r_2)} \label{50}
\end{aligned}$$
$$\label{51}
\delta_2=\texttt{sgn}\left({\frac{d\,r_2}{cd\tau_2}}\right)_{t\leq t_*}.$$
Thus, from the preceding results, we obtain: $$\label{53}
\begin{array}{l l}
Q=&\frac{-1}{M_1M_2{f_{12}}}\,\cdot\\\\
&\,\cdot\Bigl\{
\left({m_{12}}-{m_{in}}-\frac{GM_1^2+e_1^2+2e_1e_{in}}{2c^2r_*}\right)
\left({m_{out}}-{m_{12}}+\frac{GM_2^2-e_2^2-2e_2(e_{in}+e_1)}{2c^2r_*}\right)+\\\\
& -\delta_1\delta_2\sqrt{\left({m_{12}}-{m_{in}}-\frac{GM_1^2+e_1^2+2e_1e_{in}}{2c^2r_*}\right)^2-M_1^2{f_{12}}}
\\\\
& \sqrt{\left({m_{out}}-{m_{12}}+\frac{GM_2^2-e_2^2-2e_2(e_{in}+e_1)}{2c^2r_*}\right)^2-M_2^2{f_{12}}} \Bigl\};
\end{array}$$ here and in the following we omit the coordinate dependence of $f_a$, $M_a$ etc., implicitly assuming that they have to be evaluated on $(t_*,r_*)$ where not differently indicated.
Determination of Q$'$. {#sec:DeterminationOfQ .unnumbered}
----------------------
It is possible to apply the same procedure to the region $BOD$ (i.e. after the intersection time), finding the quantity $$\label{56}
Q'\equiv \{g_{00}^{(BOD)}u^0_{OB}u^0_{OD}+g_{11}^{(BOD)}u^1_{OB}u^1_{OD}\}_{t=t^*,r=r_1=r_2=r_*}\ .$$
Now the unit tangent vectors to trajectories $OB$ and $OD$ are[^1]:
$$\begin{aligned}
u^i_{OB}&=(u^0_{OB},u^1_{OB},u^2_{OB},u^3_{OB}) \nonumber\\
&=\left({\frac{d\,t}{d\tau_1}},{\frac{d\,r_1}{cd\tau_1}},0,0\right)_{t\geq t_*}, \label{54}\end{aligned}$$
and
$$\begin{aligned}
u^i_{OD}&=(u^0_{OD},u^1_{OD},u^2_{OD},u^3_{OD})\nonumber\\
&=\left({\frac{d\,t}{d\tau_2}},{\frac{d\,r_2}{cd\tau_2}},0,0\right)_{t\geq t_*}; \label{55}\end{aligned}$$
from the joining conditions (\[41\]) and (\[42\]) it is possible to see that these are actually unit vectors. The components of $u^i_{OB}$ can be deduced from Eqs.(\[38\]) and (\[41\]), while the components of $u^i_{OD}$ from Eqs.(\[39\]) and (\[42\]). Then, using the metric in the region $BOD$, it is possible to calculate the scalar product $$\label{56b}
\begin{array}{l l}
Q'=&\frac{-1}{M_1M_2{f_{21}}}\,\cdot\\\\
&\,\cdot\Bigl\{
\left({m_{out}}-{m_{21}}+\frac{GM_1^2-e_1^2-2e_1(e_{in}+e_2)}{2c^2r_*}\right)
\left({m_{21}}-{m_{in}}-\frac{GM_2^2+e_2^2+2e_2e_{in}}{2c^2r_*}\right)+\\\\
& -\delta_1'\delta_2'\sqrt{\left({m_{out}}-{m_{21}}+\frac{GM_1^2-e_1^2-2e_1(e_{in}+e_2)}{2c^2r_*}\right)^2-M_1^2{f_{21}}}
\\\\
& \sqrt{\left({m_{21}}-{m_{in}}-\frac{GM_2^2+e_2^2+2e_2e_{in}}{2c^2r_*}\right)^2-M_2^2{f_{21}}}\, \Bigl\} \ ,
\end{array}$$ where $\delta_1'$ and $\delta_2'$ have been defined as in (\[48\]) and (\[51\]), but for $t\geq t_*$. We introduced these symbols only for generality, but actually we are interested only in the case with[^2]
$$\delta_1'= \delta_1,\ \ \ \ \delta_2'= \delta_2 \ .$$
The necessary continuity requirement is thus $$\label{58}
Q=Q',$$ then, since $r_*$ is assumed to be known, this equation allows to find ${m_{21}}$.
Physical meaning of Q and Q$'$. {#sec:PhysicalInterpretationOfQAndQ .unnumbered}
-------------------------------
Using standard definition for the shell velocities before the intersection one has $$\label{59}
\left(\frac{v_1}{c}\right)^2=\frac{g_{11}^{(COA)}(r_1)}{-g_{00}^{(COA)}(r_1)}\left({\frac{d\,r_1}{cdt}}\right)^2$$
$$\label{59.b}
\left(\frac{v_2}{c}\right)^2=\frac{g_{11}^{(COA)}(r_2)}{-g_{00}^{(COA)}(r_2)}\left({\frac{d\,r_2}{cdt}}\right)^2\ ,$$
and similarly for the velocities after the intersection, $$\label{63}
\left(\frac{v_1'}{c}\right)^2=\frac{g_{11}^{(BOD)}(r_1)}{-g_{00}^{(BOD)}(r_1)}\left({\frac{d\,r_1}{cdt}}\right)^2$$
$$\label{63.b}
\left(\frac{v_2'}{c}\right)^2=\frac{g_{11}^{(BOD)}(r_2)}{-g_{00}^{(BOD)}(r_2)}\left({\frac{d\,r_2}{cdt}}\right)^2\ .$$
Then it is easy to obtain from the definitions (\[52\]) and (\[56\]), that[^3] $$\label{62}
Q=\left\{\frac{v_1 v_2/c^2-1}{\sqrt{1-v_1^2/c^2}\sqrt{1-v_2^2/c^2}}\right\}_{t=t_*,r_1=r_2=r_*}$$ and $$\label{66}
Q'=\left\{\frac{v_1' v_2'/c^2-1}{\sqrt{1-(v_1')^2/c^2}\sqrt{1-(v_2')^2/c^2}}\right\}_{t=t_*,r_1=r_2=r_*}.$$
Determination of P and P$'$. {#sec:DeterminationOfPAndP .unnumbered}
----------------------------
First of all it is convenient to introduce new symbols to simplify the expressions of Q and Q’. With
$$\begin{aligned}
q_1&\equiv -\frac{GM_1^2+e_1^2+2e_1e_{in}}{2c^2r_*} \nonumber\\
q_2&\equiv \frac{GM_2^2-e_2^2-2e_2(e_{in}+e_1)}{2c^2r_*}\nonumber \,,\end{aligned}$$
and
$$\begin{aligned}
q_1'&\equiv \frac{GM_1^2-e_1^2-2e_1(e_{in}+e_2)}{2c^2r_*} \nonumber \\
q_2'&\equiv -\frac{GM_2^2+e_2^2+2e_2e_{in}}{2c^2r_*}\nonumber\,,\end{aligned}$$
then $Q$ and $Q'$ can be re-written as $$\label{70}
\begin{array}{l l}
Q=&\frac{-1}{M_1M_2{f_{12}}}\,\cdot\\\\
&\,\cdot\Bigl\{
\left({m_{12}}-{m_{in}}+q_1\right)
\left({m_{out}}-{m_{12}}+q_2\right)+\\\\
& -\delta_1\delta_2\sqrt{\left({m_{12}}-{m_{in}}+q_1\right)^2-M_1^2{f_{12}}}
\\\\
& \sqrt{\left({m_{out}}-{m_{12}}+q_2\right)^2-M_2^2{f_{12}}}\, \Bigl\} \,
\end{array}$$ and $$\label{71}
\begin{array}{l l}
Q'=&\frac{-1}{M_1M_2{f_{21}}}\,\cdot\\\\
&\,\cdot\Bigl\{
\left({m_{out}}-{m_{21}}+q_1'\right)
\left({m_{21}}-{m_{in}}+q_2'\right)+\\\\
& -\delta_1'\delta_2'\sqrt{\left({m_{out}}-{m_{21}}+q_1'\right)^2-M_1^2{f_{12}}}
\\\\
& \sqrt{\left({m_{21}}-{m_{in}}+q_2'\right)^2-M_2^2{f_{12}}}\, \Bigl\} \ ,
\end{array}$$ Now, in principle is possible to find ${m_{21}}$ by squaring and solving $Q=Q'$ (which is a quartic equation). However the procedure is cumbersome and moreover it is not possible with Eq.(\[58\]) alone to determine the sign of the roots. Fortunately, as in the non-charged case, it is possible to follow another easier way. Indeed, it is possible to introduce two other invariants, say $P$ and $P'$, similar to $Q$ and $Q'$, which are constructed using the scalar products of the 4-velocities of the two shell, but now taking the limit to $(t_*,r_*)$ from the $AOD$ and $COB$ regions respectively. More explicitly, we define $$\label{P}
P\equiv \{g_{00}^{(AOD)}u^0_{AO}u^0_{OD}+g_{11}^{(AOD)}u^1_{AO}u^1_{OD}\}_{t=t^*,r=r_1=r_2=r_*} \ ,$$ and $$\label{Pp}
P'\equiv \{g_{00}^{(COB)}u^0_{CO}u^0_{OB}+g_{11}^{(COB)}u^1_{CO}u^1_{OB}\}_{t=t^*,r=r_1=r_2=r_*} \ .$$ Then, the same continuity requirement of Eq.(\[58\]) implies that it must hold also that $$\label{74}
Q=P\ ,\ \ \ \ \ P=P'\ .$$ Following the same method used to find $Q$ and $Q'$, after some calculations, one arrives to $$\label{75}
\begin{array}{l l}
P=&\frac{-1}{M_1M_2{f_{in}}}\,\cdot\\\\
&\,\cdot\Bigl\{
\left({m_{12}}-{m_{in}}+p_1\right)
\left({m_{21}}-{m_{in}}+p_2\right)+\\\\
& -\delta_1\delta_2'\sqrt{\left({m_{12}}-{m_{in}}+p_1\right)^2-M_1^2{f_{in}}}
\\\\
& \sqrt{\left({m_{21}}-{m_{in}}+p_2\right)^2-M_2^2{f_{in}}}\, \Bigl\}
\end{array}$$ and $$\label{76}
\begin{array}{l l}
P'=&\frac{-1}{M_1M_2{f_{in}}}\,\cdot\\\\
&\,\cdot\Bigl\{
\left({m_{out}}-{m_{21}}+p_1'\right)
\left({m_{out}}-{m_{12}}+p_2'\right)+\\\\
& -\delta_1'\delta_2\sqrt{\left({m_{out}}-{m_{21}}+p_1'\right)^2-M_1^2{f_{out}}}
\\\\
& \sqrt{\left({m_{out}}-{m_{12}}+p_2'\right)^2-M_2^2{f_{out}}}\, \Bigl\} \ ,
\end{array}$$ where we have denoted
$$\begin{aligned}
p_1&\equiv \frac{GM_1^2-e_1^2-2e_1e_{in}}{2c^2r_*} \nonumber\\
p_2&\equiv \frac{GM_2^2-e_2^2-2e_2e_{in}}{2c^2r_*}\ ,\nonumber\end{aligned}$$
and
$$\begin{aligned}
p_1'&\equiv -\frac{GM_1^2+e_1^2+2e_1(e_{in}+e_2)}{2c^2r_*} \nonumber\\
p_2'&\equiv -\frac{GM_2^2+e_2^2+2e_2(e_{in}+e_1)}{2c^2r_*}\ .\nonumber\end{aligned}$$
Determination of ${m_{21}}$; the energy transfer {#sec:DeterminationOfMto .unnumbered}
------------------------------------------------
Thus the complete set of continuity conditions at the point of intersection can be written as $$\label{77}
Q=Q',\ \ \ Q=P,\ \ \ Q=P'.$$ It turns out that this three quartic equations for the unknown parameter ${m_{21}}$ have only one common root[^4] . It is possible to find the solution using hyperbolic functions. The final result is remarkably simple: $$\label{78}
{m_{21}}={m_{in}}+{m_{out}}-{m_{12}}-\frac{e_1e_2}{c^2r_*}-\frac{GM_1M_2}{c^2r_*}Q \ ,$$ or equivalently, in terms of ${f_{21}}$: $$\label{79}
{f_{21}}={f_{in}}+{f_{out}}-{f_{12}}+2\frac{G^2M_1M_2}{c^4r_*^2}Q \ .$$ It can be easily seen from Eqn.(\[78\]) that the charge $e_{in}$ of the central singularity does not affect the result (but it affects the equation of the motion of the shells and thus $Q$). Formula (\[78\]) solves the problem of determining the mass parameter ${m_{21}}$ from the quantities specified at the evolutionary stage before intersection. It is then possible to determine the energy transfer between the shells. Indeed the energy of shell 1 and 2 before the intersection are, respectively $$\label{79E}
E_1=({m_{12}}-{m_{in}})c^2\ ,\ \ \ E_2=({m_{out}}-{m_{12}})c^2\ ,$$ while, after the intersection $$\label{80}
E_1'=({m_{out}}-{m_{21}})c^2\ ,\ \ \ E_2'=({m_{21}}-{m_{in}})c^2 \ .$$ The conservation of total energy is automatically ensured by the above formulas, indeed $$\label{81}
E_1+E_2=E_1'+E_2'\ .$$ Then it is natural to define the exchange energy as $$\label{82}
\Delta E=E_2'-E_2=-(E_1'-E_1) \ .$$ Then, from Eqn.(\[78\]) and the above definitions, it follows that $$\label{83}
\Delta E=-\frac{e_1e_2}{r_*}-\frac{GM_1M_2}{r_*}Q \ .$$ It is also useful (especially for the Newtonian approximation) to use Eqn.(\[62\]) and re-express $\Delta E$ as: $$\label{84}
\Delta E =-\frac{e_1e_2}{r_*}
-\frac{GM_1M_2}{r_*}\left\{\frac{v_1 v_2/c^2-1}{\sqrt{1-v_1^2/c^2}\sqrt{1-v_2^2/c^2}}\right\}_{r=r_*} \ .$$
Post-Newtonian approximation {#sec:NewtonianApproximation}
============================
For slow velocities of the shells it is interesting to consider the Post-Newtonian limit of Eqn.(\[84\]):
$$\begin{aligned}
\Delta E=&\frac{Gm_1m_2-e_1e_2}{r_*}+ \nonumber\\
&+\frac{1}{2c^2}\left\{\frac{Gm_1m_2}{r_*}[v_1(r_*)-v_2(r_*)]^2+\frac{Gm_2L_1^2}{m_1r_*^3}+ \frac{Gm_1L_2^2}{m_2r_*^3} \right\}+o\left(\frac{1}{c^4}\right) \ .\label{99}\end{aligned}$$
It is worth noting that only the zeroth order in $1/c^2$ changes with respect to the uncharged case (because of the Coulomb term $-e_1e_2/r_*$), while all the other orders remain unchanged, being of kinetic origin; $m_1$ and $m_2$ are the rest masses of the shells, indeed we have used for the masses $M_1$ and $M_2$ the definitions (\[32\]).
It can be also useful to re-express all the quantities in a Newtonian language and consider only the zeroth order in $1/c^2$, e.g. we can expand the energy as $$\label{E}
E=mc^2+\mathcal E+o\left(\frac{1}{c^2}\right)\ ,$$ where $m$ and $\mathcal E$ do not depend on $c$. Therefore, similarly, we can define at the first order in $1/c^2$ $$\begin{aligned}
\label{87}
{m_{12}}-{m_{in}}=m_1+\frac{\mathcal E_1}{c^2}, \ \ \ \ {m_{out}}-{m_{12}}=m_2+\frac{\mathcal E_2}{c^2},\\
{m_{out}}-{m_{21}}=m_1+\frac{\mathcal E_1'}{c^2}, \ \ \ \ {m_{21}}-{m_{in}}=m_1+\frac{\mathcal E_2'}{c^2}.\end{aligned}$$ Then it follows also that the energy conservation law takes the form $$\label{89}
\mathcal E_1+\mathcal E_2=\mathcal E_1'+\mathcal E_2' \ ,$$ and Eqn.(\[82\]) becomes $$\label{90}
\mathcal E_1'=\mathcal E_1-\Delta \mathcal E,\ \ \ \ \mathcal E_2'=\mathcal E_2+\Delta \mathcal E\,,$$ where $\Delta\mathcal E=(\Delta E)_{c\rightarrow\infty}$. Thus from the above formulas and definitions it is clear that $$\label{91}
\Delta \mathcal E=\frac{{Gm_1m_2-e_1e_2}}{r_*} \ .$$
Pressureless shells with zero effective masses ($L_1=L_2=0$ and $M_1=M_2=0$) {#sec:ShellWithZeroEffectiveMasses}
============================================================================
It is interesting also to consider the case in which the motion of the particles of the shells is only radial (i.e. $L_1=L_2=0$) and the rest masses are negligible with respect to the kinetic energies and to the charges —indeed this is the case for two shells composed by (ultra)rela-tivistic electrons and positrons. In this case the effective masses can be replaced by $$M_1=M_2=\lambda\,,$$ where $\lambda$ is a parameter arbitrary small. From Eqn.(\[83\]), with $Q$ expressed by formula (\[53\]), it is easy to find that the energy transfer in this case is $$\begin{aligned}
\label{120}
\Delta E=-\frac{e_1e_2}{r_*}+\frac{c^4 r_*}{2G{f_{12}}}({f_{in}}-{f_{12}})({f_{12}}-{f_{out}})+o(\lambda^2)\,,\end{aligned}$$ having assumed that the shells have opposite-directed velocities, i.e. $$\begin{aligned}
\label{dd}
\delta_1\delta_2=-1\ .\end{aligned}$$ Otherwise, if the shells goes in the same direction, i.e. $$\begin{aligned}
\label{dd1}
\delta_1\delta_2=1\ ,\end{aligned}$$ then Eqn.(\[83\]) becomes simply $$\begin{aligned}
\label{120l}
\Delta E=-\frac{e_1e_2}{r_*}+o(\lambda^2)\ ;\end{aligned}$$ obviously the previous formulas make sense only if $r_*$ exists. We want to underline the presence of the term $o(\lambda^2)$, because, strictly speaking, a charge cannot have zero rest mass, therefore we are in the case of just *small* effective masses. As expected, in the case of vanishing charges ($e_1=e_2=0$), Eqn.(\[120l\]) gives zero at $\lambda=0$ because this is the case of two photon-shells which go in the same direction and therefore cannot never intersect.
The intersection of a test shell with a gravitating one {#sec:TheIntersectionOfATestShellWithAGravitatingOne}
=======================================================
One-shell case {#sec:OneShellCase .unnumbered}
--------------
Let us consider firstly the case of a test shell on the RN field. This limit has the only aim to show that the shell’s equation of motion (\[18\]) actually reduce to the simple test-particle case; the limit can be obtained by putting $$\begin{aligned}
\label{l}
m \rightarrow\lambda m\ ,\ \ e \rightarrow\lambda m \ , \ \ L \rightarrow\lambda L \ ,\ \ ({m_{out}}-{m_{in}})c^2 \rightarrow\lambda E\end{aligned}$$ with $\lambda\rightarrow 0$. Then, considering also (\[24\]), we find that Eqn.(\[18\]) becomes $$\begin{aligned}
\label{128}
E=\mu c^2\sqrt{{f_{in}}(r_0)+\left({\frac{d\,r_0}{cd\tau}}\right)^2}+\frac{e{e_{in}}}{r_0}-\lambda \frac{G\mu^2-e^2}{2r_0} \ ,\end{aligned}$$ now, putting $\lambda =0$ the self-energy term is killed; then re-writing Eqn.(\[128\]) using the more familiar Schwarzschild time $t$ and Eqn.(\[rR2\]), $$\begin{aligned}
\label{129}
E=c^2\sqrt{m^2+\frac{L^2}{c^2 R_0^2(t)}} \sqrt{\frac{{f_{in}}^3(R_0)}{{f_{in}}^2(R_0)-\left(\frac{dR_0}{cdt}\right)^2}}+\frac{e{e_{in}}}{R_0}+o(\lambda)\ ,\end{aligned}$$ it is easy to recognize that Eqn.(\[129\]) coincides with the first integral of motion of a test-charge particle on the Reinssner-Nordstrom background, where $E$ is the conserved energy of the particle, $m$ the rest mass, $e$ the charge and $L$ the angular momentum.
Two-shell case, with one test-shell {#sec:TwoShellCaseWithOneTestShell .unnumbered}
-----------------------------------
Now we can deal with the more interesting two-shell case, in which shell-2 is considered “test”. To gain this limit we have to put $$\begin{aligned}
\label{l2}
m_2 \rightarrow\lambda m_2\ ,\ \ e_2 \rightarrow\lambda m_2 \ , \ \ L_2 \rightarrow\lambda L_2 \ ,\\ \nonumber ({m_{out}}-{m_{12}})c^2 \rightarrow\lambda E_2 \ , \ \ ({m_{21}}-{m_{in}})c^2 \rightarrow\lambda E_2' \ .\end{aligned}$$ Then, using Eqn.(\[78\]) with $Q$ given by formula (\[53\]), one obtains $$\label{137}
\begin{array}{l l}
\Delta E=&-\frac{e_1e_2}{r_*}+\frac{1}{r_*{f_{12}}}\,\cdot\\\\
&\,\cdot\Bigl\{
\left(E_1-\frac{GM_1^2+e_1^2+2e_1e_{in}}{2c^2r_*}\right)
\left(E_2-\frac{e_2(e_{in}+e_1)}{c^2r_*}+\lambda\frac{GM_2^2-e_2^2}{2c^2r_*}\right)+\\\\
& -\delta_1\delta_2\sqrt{\left(E_1-\frac{GM_1^2+e_1^2+2e_1e_{in}}{2c^2r_*}\right)^2-M_1^2{f_{12}}}
\\\\
& \sqrt{\left(E_2-\frac{e_2(e_{in}+e_1)}{c^2r_*}+\lambda\frac{GM_2^2-e_2^2}{2c^2r_*}\right)^2-M_2^2{f_{12}}}\ \Bigl\}\ .
\end{array}$$ Thus, only the self-energy terms of shell-2 are killed by $\lambda=0$.
Now, it is worth noting the following fact: shell-1 does not have any discontinuity when it intersect the shell-2 (this is natural because shell-2 is “test” and does not affect the metric), on the other hand shell-2 undergoes a discontinuity in the metric when it cross shell-1 and consequently it has an actual discontinuity in the velocity. It is easy to calculate this gap; indeed using the definition (\[59.b\]) of velocity $v_2$ \[with the time ${\frac{d\,t}{d\tau_2}}$ given by the joint condition (\[36\])\], with metric coefficient (\[26\]), and with the help the first integral of motion (\[31\]), one finds $$\label{138}
v_2^2(r_2)=
1-{f_{out}}(r_2)\left(\frac{E_2}{M_2(r_2)}-\frac{e_2(e_1+{e_{in}})^2}{M_2(r_2)r_2}\right)^{-2}+o(\lambda) \ ,\ \ \ t\leq t_* \ ,$$ where we have used ${f_{12}}={f_{out}}+o(\lambda)$; in the same way, using (\[63.b\]), (\[42\]), (\[28\]), and (\[39\]), the velocity $v_2'$ (after the intersection) is $$\label{139}
[v_2'(r_2)]^2=
1-{f_{in}}(r_2)\left(\frac{E_2'}{M_2(r_2)}-\frac{e_2(e_1+{e_{in}})^2}{M_2(r_2)r_2}\right)^{-2}+o(\lambda) \ ,\ \ \ t\geq t_*\ ,$$ where $E_2'$ can be expressed in function of $E_2$ with the help of (\[137\]). From the previous formulas it is clear that in general $$\label{139b}
v_2'(r_*)-v_2(r_*)\neq 0 \ .$$
Shell ejection {#sec:ShellEjection}
==============
The exchange in energy of the shells during the intersection makes possible that one initially bounded shell can acquire enough energy to escape to infinity. The shell ejection mechanism can take place also in the Newtonian regime. In this case, from Eqs.(\[90\])-(\[91\]) it results that $$\begin{aligned}
\label{140}
\mathcal E_1'=\mathcal E_1-\frac{{Gm_1m_2-e_1e_2}}{r_*'}\ ,\ \ \ \ \mathcal E_2'=\mathcal E_2+\frac{{Gm_1m_2-e_1e_2}}{r_*'}\ ,\end{aligned}$$ and then, after the first intersection $$\label{141}
\begin{cases}
\mathcal E_1''=\mathcal E_1'+\frac{{Gm_1m_2-e_1e_2}}{r_*''}=\mathcal E_1-({Gm_1m_2-e_1e_2})\left(\frac{1}{r_*'}-\frac{1}{r_*''}\right)\\\\
\mathcal E_2''=\mathcal E_2'-\frac{{Gm_1m_2-e_1e_2}}{r_*''}=\mathcal E_2+({Gm_1m_2-e_1e_2})\left(\frac{1}{r_*'}-\frac{1}{r_*''}\right),
\end{cases}$$ where we have denoted the radius of the first and second intersection with $r_*'$ and $r_*''$ respectively. In the following we will consider only the case $$\begin{aligned}
\label{ga}
{Gm_1m_2-e_1e_2}>0 \ ,\end{aligned}$$ this is e.g. the case in which the two shells have opposite charges. Thus, also in the case $\mathcal E_1, \mathcal E_2<0$, if $$\begin{aligned}
\label{r'r''}
r_*''>r_*' \ ,\end{aligned}$$ and if the initial condition were in such a way that $r_*'$ is enough small and $r_*''$ not too much close to $r_*'$, then it is possible to have $\mathcal E_2''>0$, i.e. the ejection of the second shell.
Let us now assume that $r_*''>r_*'$, and consider a “semi-relativistic” case in which at the first intersection we use the full relativistic formulas[^5], $$\label{142}
\begin{cases}
E_1'=E_1-\frac{M_1(r_*')M_2(r_*')}{r_*'}(-Q)+\frac{e_1e_2}{r_*'}\\\\ E_2'=E_2+\frac{M_1(r_*')M_2(r_*')}{r_*'}(-Q)-\frac{e_1e_2}{r_*'}\ ,
\end{cases}$$ while at the second intersection we use the Newtonian approximation,
$$\label{143}
\begin{cases}
E_1''&= E_1'+\frac{{Gm_1m_2-e_1e_2}}{r_*''} \\
&=E_1-\left[\frac{M_1(r_*')M_2(r_*')(-Q)-e_1e_2}{r_*'}-\frac{{Gm_1m_2-e_1e_2}}{r_*''}\right]
\\\\ E_2''&= E_2'-\frac{{Gm_1m_2-e_1e_2}}{r_*''} \\
&= E_2+\left[\frac{M_1(r_*')M_2(r_*')(-Q)-e_1e_2}{r_*'}-\frac{{Gm_1m_2-e_1e_2}}{r_*''}\right]\,.
\end{cases}$$
This approximation is always justified if the radius of the second intersection $r_*''$ is enough large. Now, it is remarkable that whatever the value of $r_*'$ is, the first term in the square brackets in Eqn.(\[143\]) satisfies the inequality $$\begin{aligned}
\label{144}
\frac{M_1(r_*')M_2(r_*')(-Q)-e_1e_2}{r_*'}>\frac{{Gm_1m_2-e_1e_2}}{r_*'}\ .\end{aligned}$$ Comparing the expressions (\[143\]), (\[144\]) and (\[141\]) it is possible to see that in the relativistic regime the shell ejection possibility is even greater than in the Newtonian case. Furthermore, it is worth noting that the presence of the charge do not change qualitatively the pure gravitational analysis, but just magnifies the ejection effect.
${Gm_1m_2-e_1e_2}<0$ case {#sec:Ga0Case}
-------------------------
Let us consider also briefly the case in which the shells are equal-signed charged and the repulsion overcome the gravity attraction, i.e. ${Gm_1m_2-e_1e_2}<0$. In this case the ejection can happen only after an odd number of intersections.
E.g. after three intersections, from the previous formulas we have, in the Newtonian approximation: $$\begin{aligned}
\label{145}
\mathcal E_1'''=\mathcal E_1-({Gm_1m_2-e_1e_2})\left(\frac{1}{r_*'}-\frac{1}{r_*''}+\frac{1}{r_*'''}\right) \ .\end{aligned}$$ Obviously this formula has a meaning only if $$\begin{aligned}
\label{146}
\frac{1}{r_*'}<\frac{1}{r_*'}-\frac{1}{r_*''}+\frac{1}{r_*'''} \ ,\end{aligned}$$ otherwise the ejection happens at the first intersection (and then there would not be other crossings, and no $r_*'',\ r_*'''$), or never more; if Eqn.(\[146\]) is true, then it means that the barycenter of the two shells is falling into the center singularity.
Conclusions {#sec:Conclusions}
===========
We have found the energy exchange between two charged crossing shells (formula (\[84\])). Then we have studied special cases of physical interest in which the formulas simplify: the non relativistic case, the massless shells, the test shell, and finally the ejection mechanism in a semi-Newtonian regime: we found that the ejection mechanism is more efficient in the charged case than in the neutral one if the charges have opposite sign (because the energy transfer is larger due to the Coulomb interaction).
[1]{}
J.E. Chase. , (1970).
W. Israel. , (1966).
C. Cherubini, R. Ruffini, and L. Vitagliano. , (2002).
V. De La Cruz and W. Israel. , (1967).
J. Langer and A. Eid. , (2000).
M.V. Barkov, V.A. Belinski, and G.S. Bisnovatyi-Kogan. , (2002). arXiv:astro-ph/0210296v.2.
V. Belinski, M. Pizzi, and A. Paolino. .
L.D. Landau and E.M. Lifshitz. . Pergamon Press, (1962).
[^1]: Obviously, when we say $t\geq t_*$, we tacitly assume before a (possible) second intersection.
[^2]: This is the only possible case if one excludes $v_1(t^*)=v_2(t^*)=0$, because there are non discontinuities in the velocities.
[^3]: It is also worth noting that $\sqrt{Q^2-1}/Q=-|v_1/c-v_2/c|/(1-v_1v_2/c^2)$, which is the relative velocity definition of two “particles” in relativistic mechanics.
[^4]: In the pure gravitational case it is also possible to use just Eqn. $Q=P'$, and then choose the correct root solution by requiring that $m_{12}$ has to be positive, see Ref. .
[^5]: Remember that $-Q=1+o(1/c^2)$.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'This work builds on the following result of a previous article (quant-ph/0509044): the matter field can be naturally eliminated from the equations of the scalar electrodynamics (the Klein-Gordon-Maxwell electrodynamics) in the unitary gauge. The resulting equations describe independent dynamics of the electromagnetic field (they form a closed system of partial differential equations). An improved derivation of this surprising result is offered in the current work. It is also shown that for this system of equations, a generalized Carleman linearization (Carleman embedding) procedure generates a system of linear equations in the Hilbert space, which looks like a second-quantized theory and is equivalent to the original nonlinear system on the set of solutions of the latter. Thus, the relevant local realistic model can be embedded into a quantum field theory. This model is equivalent to a well-established model - the scalar electrodynamics, so it correctly describes a large body of experimental data. Although it does not describe the electronic spin and possibly some other experimental facts, it may be of great interest as a “no drama quantum theory”, as simple (in principle) as classical electrodynamics. Possible issues with the Bell theorem are discussed.'
author:
- 'A. Akhmeteli'
bibliography:
- 'akhijqi.bib'
title: 'No Drama Quantum Theory?'
---
Introduction
============
Is it possible to offer a “no drama” quantum theory? Something as simple (in principle) as classical electrodynamics - a local realistic theory described by a system of partial differential equations in 3+1 dimensions, but reproducing unitary evolution of quantum theory in the configuration space?
Of course, the Bell inequalities cannot be violated in such a theory. This author has little, if anything, new to say about the Bell theorem, and this article is not about the Bell theorem. However, this issue cannot be “swept under the carpet” and will be discussed in Section IV using other people’s arguments.
In a previous article (Ref. [@Akhm10]), the equations of the scalar electrodynamics (the Klein-Gordon-Maxwell electrodynamics) with Lagrangian $$\begin{aligned}
\label{eq:pr6}
\nonumber
-\frac{1}{4}F^{\mu\nu}F_{\mu\nu}+\frac{1}{2}(\psi^*_{,\mu}-ieA_\mu\psi^*)(\psi^{,\mu}+ieA^\mu\psi)-\\
-\frac{1}{2}m^2\psi^*\psi\end{aligned}$$ were considered: $$\label{eq:pr7}
(\partial^\mu+ieA^\mu)(\partial_\mu+ieA_\mu)\psi+m^2\psi=0,$$ $$\label{eq:pr8}
\Box A_\mu-A^\nu_{,\nu\mu}=j_\mu,$$ $$\label{eq:pr9}
j_\mu=ie(\psi^*\psi_{,\mu}-\psi^*_{,\mu}\psi)-2e^2 A_\mu\psi^*\psi.$$ Schrödinger (Ref. [@Schroed]) noted that the complex charged matter field can be made real by a gauge transform (at least locally), and the equations of motion in the relevant gauge (unitary gauge) for the transformed 4-potential of electromagnetic field $B^{\mu}$ and real matter field $\varphi$ are as follows: $$\label{eq:pr10}
\Box\varphi-(e^2 B^\mu B_\mu-m^2)\varphi=0,$$ $$\label{eq:pr11}
\Box B_\mu-B^\nu_{,\nu\mu}=j_\mu,$$ $$\label{eq:pr12}
j_\mu=-2e^2 B_\mu\varphi^2.$$ It should be noted that these equations can be obtained from the following Lagrangian (Ref. [@Takabayasi]): $$\label{eq:pr12a}
-\frac{1}{4}F^{\mu\nu}F_{\mu\nu}+\frac{1}{2}e^2 B_\mu B^\mu \phi^2+\frac{1}{2}(\varphi_{,\mu}\varphi^{,\mu}-m^2\varphi^2).$$ Actually, it coincides with the Lagrangian of Eq. (\[eq:pr6\]) up to the replacement of the complex scalar field by a real one.
The following surprising result was proven in Ref. [@Akhm10]: the equations obtained from Eqs. (\[eq:pr10\],\[eq:pr11\],\[eq:pr12\]) after natural elimination of the matter field form a closed system of partial differential equations and thus describe independent dynamics of electromagnetic field. The detailed wording is as follows: if components of the 4-potential of the electromagnetic field and their first derivatives with respect to time are known in the entire space at some time point, the values of their second derivatives with respect to time can be calculated for the same time point, so the Cauchy problem can be posed, and integration yields the 4-potential in the entire space-time. Thus, the broad range of quantum phenomena described by the scalar electrodynamics can be described in terms of electromagnetic field only. This result not only permits mathematical simplification, as the number of fields is reduced, but can also be useful for interpretation of quantum theory. For example, in the Bohm (de Broglie-Bohm) interpretation (Refs. [@BohmHiley; @Holland; @Goldstein]), the electromagnetic field can replace the wave function as the guiding field. This may make the interpretation more attractive, removing, for example, the reason for the following criticism of the Bohm interpretation: “If one believes that the particles are real one must also believe the wavefunction is real because it determines the actual trajectories of the particles. This allows us to have a realist interpretation which solves the measurement problem, but the cost is to believe in a double ontology. [@Smolin]” Independent of the interpretation, quantum phenomena can be described in terms of electromagnetic field only.
The proof in Ref. [@Akhm10] was somewhat awkward. For example, the expression for the matter field contains an ugly looking square root, suggesting sign ambiguities. Therefore, an improved and proof is offered in this work.
It is also shown (using other people’s results) how the “one-particle” theories can be turned into “many-particle” theories, which look very much like quantum field theory.
Elimination of Matter Field from Scalar Electrodynamics
=======================================================
Elimination of the matter field $\varphi$ from Eqs. (\[eq:pr10\],\[eq:pr11\],\[eq:pr12\]) can be simplified compared to Ref. [@Akhm10], using a substitution $\Phi=\varphi^2$ first. For example, as $$\label{eq:pr1q}
\Phi_{,\mu}=2\varphi\varphi_{,\mu},$$ we obtain $$\label{eq:pr2q}
\Phi_{,\mu}^{,\mu}=2\varphi^{,\mu}\varphi_{,\mu}+2\varphi\varphi^{,\mu}_{,\mu}=
\frac{1}{2}\frac{\Phi^{,\mu}\Phi_{,\mu}}{\Phi}+2\varphi\varphi^{,\mu}_{,\mu}.$$ Multiplying Eq. (\[eq:pr10\]) by $2\varphi$, we obtain the following equations in terms of $\Phi$ instead of Eqs. (\[eq:pr10\],\[eq:pr11\],\[eq:pr12\]): $$\label{eq:pr3q}
\Box\Phi-\frac{1}{2}\frac{\Phi^{,\mu}\Phi_{,\mu}}{\Phi}-2(e^2 B^\mu B_\mu-m^2)\Phi=0,$$ $$\label{eq:pr4q}
\Box B_\mu-B^\nu_{,\nu\mu}=-2e^2 B_\mu\Phi,$$ To prove that these equations describe independent evolution of the electromagnetic field $B^\mu$, it is sufficient to prove that if components $B^\mu$ of the potential and their first derivatives with respect to $x^0$ ($\dot{B}^\mu$) are known in the entire space at some time point $x^0=\rm{const}$ (that means that all spatial derivatives of these values are also known in the entire space at that time point), Eqs. (\[eq:pr3q\],\[eq:pr4q\]) yield the values of their second derivatives, $\ddot{B}^\mu$, for the same value of $x^0$. Indeed, $\Phi$ can be eliminated using Eq. (\[eq:pr4q\]) for $\mu=0$, as this equation does not contain $\ddot{B}^\mu$ for this value of $\mu$: $$\label{eq:pr5q}
\Phi=(-2e^2 B_0)^{-1}(\Box B_0-B^\nu_{,\nu 0})=(-2e^2 B_0)^{-1}(B^{,i}_{0,i}-B^i_{,i 0})$$ (Greek indices in the Einstein sum convention run from $0$ to $3$, and Latin indices run from $1$ to $3$). Then $\ddot{B}^i$ ($i=1,2,3$) can be determined by substitution of Eq. (\[eq:pr5q\]) into Eq. (\[eq:pr4q\]) for $\mu=1,2,3$: $$\label{eq:pr6q}
\ddot{B}_i=-B^{,j}_{i,j}+B^\nu_{,\nu i}+(B_0)^{-1} B_i(B^{,j}_{0,j}-B^j_{,j 0}).$$ Thus, to complete the proof, we only need to find $\ddot{B}^0$. Conservation of current implies $$\label{eq:pr7q}
0=(B^\mu \Phi)_{,\mu}=B^\mu_{,\mu}\Phi+B^\mu\Phi_{,\mu}.$$ This equation determines $\dot{\Phi}$, as spatial derivatives of $\Phi$ can be found from Eq. (\[eq:pr5q\]). Differentiation of this equation with respect to $x^0$ yields $$\begin{aligned}
\label{eq:pr8q}
\nonumber
0=(\ddot{B}^0+\dot{B}^i_{,i})\Phi+(\dot{B}^0+B^i_{,i})\dot{\Phi}+\\
+\dot{B}^0\dot{\Phi}+\dot{B}^i\Phi_{,i}+B^0\ddot{\Phi}+B^i\dot{\Phi}_{,i}.\end{aligned}$$ After substitution of $\Phi$ from Eq. (\[eq:pr5q\]), $\dot{\Phi}$ from Eq. (\[eq:pr7q\]), and $\ddot{\Phi}$ from Eq. (\[eq:pr3q\]) into Eq. (\[eq:pr8q\]), the latter equation determines $\ddot{B^0}$ as a function of $B^\mu$, $\dot{B}^\mu$ and their spatial derivatives (again, spatial derivatives of $\Phi$ and $\dot{\Phi}$ can be found from the expressions for $\Phi$ and $\dot{\Phi}$ as functions of $B^\mu$ and $\dot{B}^\mu$). Thus, if $B^\mu$ and $\dot{B}^\mu$ are known in the entire space at a certain value of $x^0$, then $\ddot{B}^\mu$ can be calculated for the same $x^0$, so integration yields $B^\mu$ in the entire space-time. Therefore, we do have independent dynamics of electromagnetic field.
Apparently, it is possible to introduce a Lorentz-invariant Lagrangian with higher derivatives that does not include the matter field, but is largely equivalent to the Lagrangian of Eq. (\[eq:pr12a\]) (the significance of some special cases, e.g., $\varphi=0$ and $B^\mu B_\mu=0$ (see below) is unclear (different particles?)). To this end, the latter Lagrangian can be expressed in terms of $\Phi=\varphi^2$, rather than $\varphi$, using, e.g., the following: $$\label{eq:pr16a}
\varphi_{,\mu}\varphi^{,\mu}=\frac{1}{4}\frac{\Phi_{,\mu}\Phi^{,\mu}}{\Phi},$$ and then $\Phi$ can be replaced by the following expression obtained from the equations of motion Eq. (\[eq:pr4q\]): $$\label{eq:pr16b}
\Phi=-\frac{1}{2e^2} \frac{B^\mu(\Box B_\mu-B^\nu_{,\nu\mu})}{B^\mu B_\mu}.$$
Transition to Many-Particle Theory
==================================
The theory we considered in the previous section is not second-quantized, so, on the face of it, it cannot describe many particles. On the other hand, nightlight (Ref. [@nightlight2]) indicated that, rather amazingly, second-quantized theories (or at least theories that look like second-quantized ones) can be obtained from nonlinear partial differential equations by a generalization of the Carleman linearization (Carleman embedding) procedure (Ref. [@Kowalski]). This generalized procedure generates for a system of nonlinear partial differential equations a system of linear equations in the Hilbert space, which looks like a second-quantized theory and is equivalent to the original nonlinear system on the set of solutions of the latter.
Following Ref. [@Kowalski2], let us consider a nonlinear differential equation in an (s+1)-dimensional space-time (the equations describing independent dynamics of electromagnetic field for the scalar electrodynamics are a special case of this equation) ${\partial_t}\boldsymbol{\xi}(x,t) = \boldsymbol{F}(\boldsymbol{\xi},{D^\alpha}\boldsymbol{\xi};x,t)$ , $\boldsymbol{\xi}(x,0)=\boldsymbol{\xi}_0(x)$, where $\boldsymbol{\xi}:\mathbf{R}^s\times\mathbf{R}\rightarrow\mathbf{C}^k$, $D^\alpha\boldsymbol{\xi}=\left(D^{\alpha_1}\xi_1,\ldots ,D^{\alpha_k}\xi_k\right)$, $\alpha_i$ are multiindices, ${D^\beta}={\partial^{|\beta|}}/\partial x_1^{\beta_1}\ldots\partial x_s^{\beta_s}$, with $ |\beta|=\sum\limits_{i=1}^{s}\beta_i$, is a generalized derivative, $\boldsymbol{F}$ is analytic in $\boldsymbol{\xi}$, $D^\alpha\boldsymbol{\xi}$. It is also assumed that $\boldsymbol{\xi_0}$ and $\boldsymbol{\xi}$ are square integrable. Then Bose operators $\boldsymbol{a^\dag(x)}=\left(a^\dag_1(x),\ldots,a^\dag_k(x)\right)$ and $\boldsymbol{a(x)}=\left(a_1(x),\ldots,a_k(x)\right)$ are introduced with the canonical commutation relations: $$\begin{aligned}
\label{eq:ad1}
\nonumber
\left[a_i(x),a^\dag_j(x')\right]=\delta_{ij}\delta(x-x')I,\\
\left[a_i(x),a_j(x')\right]=\left[a^\dag_i(x),a^\dag_j(x')\right]=0,\end{aligned}$$ where $x,x'\in\mathbf{R}^s$, $i,j=1,\ldots,k$. Normalized functional coherent states in the Fock space are defined as $|\boldsymbol{\xi}\rangle =\exp\left(-\frac{1}{2}\int d^sx|\boldsymbol{\xi}|^2\right)\exp\left(\int d^sx\boldsymbol{\xi}(x)\cdot\boldsymbol{a}^\dagger(x)\right)|\boldsymbol{0}\rangle$. They have the following property: $$\label{eq:ad1a}
\boldsymbol{a}(x)|\boldsymbol{\xi}\rangle =\boldsymbol{\xi}(x)|\boldsymbol{\xi}\rangle,$$. Then the following vectors in the Fock space can be introduced: $$\begin{aligned}
\label{eq:ad2}
\nonumber
|\xi,t\rangle = \exp\left[\frac{1}{2}\left(\int {d^s}x|\boldsymbol{\xi}|^2-\int {d^s}x|\boldsymbol{\xi}_0|^2\right)\right]|\boldsymbol{\xi}\rangle\\
\nonumber
=\exp\left(-\frac{1}{2}\int d^sx|\boldsymbol{\xi}_0|^2\right)\\
\times\exp\left(\int d^sx\boldsymbol{\xi}(x)\cdot\boldsymbol{a}^\dagger(x)\right)|\boldsymbol{0}\rangle.\end{aligned}$$ Differentiation of Eq. (\[eq:ad2\]) with respect to time $t$ yields, together with Eq. (\[eq:ad1a\]), a linear Schrödinger-like evolution equation in the Fock space: $$\begin{aligned}
\label{eq:ad3}
\nonumber
\frac{d}{dt}|\xi,t\rangle = M(t)|\xi,t\rangle,\\
|\xi,0\rangle=|\boldsymbol{\xi}_0\rangle,\end{aligned}$$ where the boson “Hamiltonian” $M(t) = \int {d^s}x{\boldsymbol{a}^\dagger}(x)\cdot F(\boldsymbol{a}(x),{D^\alpha}\boldsymbol{a}(x))$.
Can similar results be obtained for the Dirac-Maxwell electrodynamics, based on the results of Ref. [@Akhm10]? Probably yes, but this has not been done yet, and it may require the use of the fermionic coherent states (Ref. [@Cahill]). It should also be emphasized that the results of Ref. [@Akhm10] are less general for the Dirac-Maxwell electrodynamics than for the scalar electrodynamics.
Bell Theorem
============
In Section III, it was shown that a theory similar to quantum field theory (QFT) can be built that is basically equivalent to non-second-quantized scalar electrodynamics on the set of solutions of the latter. However, the local realistic theory violates the Bell inequalities, so this issue is discussed below using other people’s arguments. Most of them were outlined by nightlight in various forums (see, e.g., Ref. [@nightlight]) and by E. Santos (see, e.g., Ref. [@Santos]), and can be summarized as follows.
While the Bell inequalities cannot be violated in local realistic theories, there are some reasons to believe these inequalities cannot be violated either in experiments or in quantum theory. Indeed, there seems to be a consensus among experts that “a conclusive experiment falsifying in an absolutely uncontroversial way local realism is still missing” [@Gen]. For example, A. Shimony offers the following opinion:
“The incompatibility of Local Realistic Theories with Quantum Mechanics permits adjudication by experiments, some of which are described here. Most of the dozens of experiments performed so far have favored Quantum Mechanics, but not decisively because of the “detection loophole” or the “communication loophole.” The latter has been nearly decisively blocked by a recent experiment and there is a good prospect for blocking the former. [@Shimony]”
M. Aspelmeyer and A. Zeilinger agree:
“But the ultimate test of Bell’s theorem is still missing: a single experiment that closes all the loopholes at once. It is very unlikely that such an experiment will disagree with the prediction of quantum mechanics, since this would imply that nature makes use of both the detection loophole in the Innsbruck experiment and of the locality loophole in the NIST experiment. Nevertheless, nature could be vicious, and such an experiment is desirable if we are to finally close the book on local realism.” [@Aspel]
The popular argument of the latter quote that the loopholes were closed in separate experiments does not look conclusive either. Otherwise one could argue, for example, that the sum of the angles of a triangle in planar Euclidian geometry can differ from 180 degrees because experiments demonstrate that the sum of angles can differ from 180 degrees for planar quadrangles and for triangles on a sphere. The Bell inequalities for local realistic theories can only be guaranteed if all conditions of the Bell theorem are fulfilled simultaneously. Therefore, if one of the assumptions of the Bell theorem is not fulfilled in an experiment, the violation of the Bell inequalities in that experiment cannot rule out local realistic theories.
On the other hand, to prove theoretically that the inequalities can be violated in quantum theory, one needs to use the projection postulate (loosely speaking, the postulate states that if some value of an observable is measured, the resulting state is an eigenstate of the relevant operator with the relevant eigenvalue). However, such postulate, strictly speaking, is in contradiction with the standard unitary evolution of the larger quantum system that includes the measured system and the measurement device (and the observer, if needed), as such postulate introduces irreversibility, whereas there is no irreversibility for the larger system (see, e.g., Ref. [@alla] or the references to journal articles there), and, according to the quantum recurrence theorem, the larger system will return to a state that can be arbitrarily close to its initial, pre-measurement state, at least in a very large, but finite volume (Ref. [@Bocc]). Furthermore, unitary evolution cannot generate a mixture of states (the well-known measurement problem in quantum theory). The standard argument that collapse takes place during measurements and unitary evolution takes place between measurements does not seem convincing, as there is no obvious reason why unitary evolution should not be applicable to an instrument or an observer. For example, based on an analysis of experimental data, Schlosshauer (Ref. [@Schloss]) believes that “(i) the universal validity of unitary dynamics and the superposition principle has been confirmed far into the mesoscopic and macroscopic realm in all experiments conducted thus far; (ii) all observed ‘‘restrictions’’ can be correctly and completely accounted for by taking into account environmental decoherence effects; (iii) no positive experimental evidence exists for physical state-vector collapse; (iv) the perception of single ‘‘outcomes’’ is likely to be explainable through decoherence effects in the neuronal apparatus.”
Therefore, it seems that mutually contradictory assumptions (e.g., unitary evolution and the projection postulate) are required to prove the Bell theorem, so it is on shaky grounds both theoretically and experimentally. On the other hand, the local realistic theory of this work reproduces unitary evolution of a theory that looks like a quantum field theory, so it may need a modification of the theory of measurement (cf. Ref. [@Santos2]).
Conclusion
==========
Natural elimination of the matter field from the non-second-quantized scalar electrodynamics yields a theory that is, on the one hand, as simple (in principle) as classic electrodynamics, on the other hand, can be embedded into a quantum field theory.
Acknowledgments {#acknowledgments .unnumbered}
===============
The author is grateful to A.E. Allahverdyan, A.V. Gavrilin, A.Yu. Kamenshchik, A.Yu. Khrennikov, nightlight, H. Nikoli, W. Struyve, and H. Westman for their interest in this work and valuable remarks.
| {
"pile_set_name": "ArXiv"
} |
KEK-TH-1645
[ Kaoru Hagiwara$^{a, b}$, Toshifumi Yamada$^a$ ]{}
**
$^a$ KEK Theory Center,\
1-1 Oho, Tsukuba, Ibaraki 305-0801, Japan\
$^b$ SOKENDAI,\
1-1 Oho, Tsukuba, Ibaraki 305-0801, Japan\
We consider a light supersymmetric top quark (*stop*) scenario, where a *stop* and a neutralino are the lightest supersymmetric particles and the *stop* mass is close to the sum of the neutralino mass and the top quark mass. In this scenario, the top quark and the neutralino coming from the decay of a *stop* have almost equal velocity vectors, and hence the missing transverse momentum in *stop* pair production events at colliders is suppressed, which makes the *stop* search challenging, as in the degenerate scenario where the *stop* mass and the neutralino mass are nearby. In this paper, we propose a novel analysis technique aiming at discovering the *stop* pair production signal in such an ‘equal-velocity’ scenario. The key is to look for events where the missing transverse momentum vector is proportional to the transverse momentum vector of the top quark pair with specific values for the proportionality coefficient, as this coefficient equals to the ratio of the neutralino mass over the top quark mass in the equal-velocity limit. We examine the possibility that the *stop* signal can be identified at the $\sqrt{s}=14$ TeV LHC by our analysis method.
Although the gauge hierarchy problem suggests that the standard model (SM) should be extended at $O(100)$ GeV scale, the 8 TeV LHC has reported no signal of new physics so far. One possible explanation to this gap is that new physics signatures are hidden behind SM backgrounds due to the lack or smallness of missing transverse momentum, which is often utilized for tagging new physics events with a dark matter candidate. In this letter, we consider a scenario based on the R-parity conserving minimal supersymmetric standard model (MSSM), where the lightest R-parity odd particle is a dark matter particle, and the next-to-lightest R-parity odd particle has the mass which approximately equals to the sum of the masses of its SM partner and the dark matter particle (the lightest R-parity odd particle). In MSSM, we can consider the following four cases: $$\begin{aligned}
m_{\tilde{t}_1} \ &\simeq \ m_t \ + \ m_{\tilde{\chi}_1^0} , & {\rm (1a)}
\\
m_{\tilde{\chi}_2^0} \ &\simeq \ m_h \ + \ m_{\tilde{\chi}_1^0} , & {\rm (1b)}
\\
m_{\tilde{\chi}_2^0} \ &\simeq \ m_Z \ + \ m_{\tilde{\chi}_1^0} , & {\rm (1c)}
\\
m_{\tilde{\chi}_2^{\pm}} \ &\simeq \ m_W \ + \ m_{\tilde{\chi}_1^0} , & {\rm (1d)}
$$ where $\tilde{\chi}_1^0$ denotes the lightest neutralino which is a dark matter particle. A common feature of these cases is that the heavy SM particle $t, \, h, \, Z$ or $W$ and the dark matter particle $\tilde{\chi}_1^0$ have similar velocity vectors once they are produced from the decay of the next-to-lightest R-odd particle. At colliders, therefore, the missing transverse momentum coming from the dark matter particles is suppressed in events where next-to-lightest R-odd particles are pair-produced and then decay. Nevertheless, such events can be identified through the information that the missing transverse momentum vector due to the dark matter particles and the transverse momentum vector of the heavy SM particles are parallel, and that the ratio of their absolute values equals to the ratio of the dark matter mass and the SM particle mass. This is the key observation of this letter. Although the mass spectra like eqs. (1a-1d) are realized only for very particular combinations of the model parameters, we have an example in the SM where $D^*$ particles can be identified by requiring that a $D$ candidate and a $\pi$ have the same velocity vector because the mass of $D^*$ is close to the sum of the $D$ mass and the $\pi$ mass. We hereafter call the scenario the ‘equal-velocity scenario’.
In this letter, particular attention is paid to the mass spectrum of eq. (1a). This spectrum is motivated by the light SUSY top (*stop*) scenario [@light; @stop], which is inspired by the fact that the top quark and its SUSY partners have the largest couplings with the Higgs boson and hence the SUSY top partners should be light in order to solve the gauge hierarchy problem. Also, since the *stop*s are colored particles, it is relatively easy to find the *stop* signal at the LHC, compared to the cases with other mass spectra eqs. (1b-1d). Search strategies and techniques for the light *stop* scenario have been discussed in refs. [@stop; @search]. In this study, we focus on a mass spectrum where the *stop* mass is close to the sum of the top quark mass and the mass of the lightest R-odd particle, and propose a novel search method optimized for such a mass spectrum.
Before describing our new search method, we review the status of light *stop* searches in the LHC 7 TeV and 8 TeV runs, concentrating on the case where a *stop* and a neutralino are the only light SUSY particles and their mass difference is nearly equal to or larger than the top quark mass. Searches for such a mass spectrum have been done by observing events with hadronic [@lhchad], semi-leptonic [@lhcsemi] or di-leptonic decays of a top quark pair [@lhcdi] associated with large missing transverse momentum. With about 20 fb$^{-1}$ of data at the 8 TeV LHC, the parameter region with 200 GeV $< m_{\tilde{t}_1} < 300$ GeV and $m_{\tilde{\chi}_1^0} \lesssim (m_{\tilde{t}_1} - m_t - 15$ GeV) and the region with 300 GeV $<m_{\tilde{t}_1} \lesssim 600$ GeV and $m_{\tilde{\chi}_1^0}$ much smaller than $(m_{\tilde{t}_1} - m_t - 15$ GeV) are excluded. Additionally, a search using the spin correlation of a top quark pair has been conducted [@lhcspin], and the region with $m_t<m_{\tilde{t}_1} \lesssim 195$ GeV and $m_{\tilde{\chi}_1^0}< (m_{\tilde{t}_1} - m_t)$ is excluded.\
In the equal-velocity limit where $m_{\tilde{t}_1}=m_t+m_{\tilde{\chi}^0_1}$ holds, the top quark and the neutralino originating from the decay of a *stop* have the same velocity vector. Hence, in a *stop* pair production event at hadron colliders $$\begin{aligned}
p p &\rightarrow& \tilde{t}_1 \bar{\tilde{t}}_1 \ \rightarrow \ t \tilde{\chi}^0_1 \bar{t} \tilde{\chi}^0_1 \ ,
$$ the top quarks and the neutralinos have the following three-momenta: $$\begin{aligned}
\vec{p}_t \ &= \ m_t \gamma_1 \vec{\beta}_1 \ , \ \ \ \vec{p}_{\bar{t}} \ = \ m_t \gamma_2 \vec{\beta}_2 \ ,
\ \ \ \vec{p}_{(\tilde{\chi}^0_1)^a} \ = \ m_{\tilde{\chi}^0_1} \gamma_1 \vec{\beta_1} \ , \ \ \
\vec{p}_{(\tilde{\chi}^0_1)^b} \ = \ m_{\tilde{\chi}^0_1} \gamma_2 \vec{\beta}_2 \ ,
$$ where the indices $a$ and $b$ label the two neutralinos, $\vec{\beta}_1$ and $\vec{\beta}_2$ respectively denote the velocity vectors of $\tilde{t}_1$ and $\bar{\tilde{t}}_1$ in the laboratory frame, and $\gamma_1$ and $\gamma_2$ denote their corresponding boost factors. It follows that the missing transverse momentum vector coming from the neutralinos, $\vec{{p\hspace{-0.45em}/ }}_T$, and the transverse momentum vector of the top quark pair, $\vec{p}_{t\bar{t} \, T}$, satisfy the following relation in the equal-velocity limit: $$\begin{aligned}
\vec{{p\hspace{-0.45em}/ }}_T \ &= \ (\vec{p}_{(\tilde{\chi}^0_1)^a}+\vec{p}_{(\tilde{\chi}^0_1)^b})_T \ = \
\frac{m_{\tilde{\chi}^0_1}}{m_t} (\vec{p}_t+\vec{p}_{\bar{t}})_T \ = \
\frac{m_{\tilde{\chi}^0_1}}{m_t} \vec{p}_{t\bar{t} \, T} \ .
\label{momenta relation}
$$ To utilize the relation eq. (\[momentum relation\]) to distinguish *stop* events from SM backgrounds in the equal-velocity scenario, we plot each event on the 2-dimensional plane spanned by the azimuthal angle between the missing transverse momentum vector and the reconstructed top-pair transverse momentum vector, $\Delta \phi( \vec{{p\hspace{-0.45em}/ }}_T, \, \vec{p}_{t\bar{t} \, T} )$, and the ratio of their absolute values, ${p\hspace{-0.45em}/ }_T/p_{t\bar{t} \, T}$. As the mass spectrum approaches to the equal-velocity limit, *stop* pair production events gather around the point of $\Delta \phi( \vec{{p\hspace{-0.45em}/ }}_T, \, \vec{p}_{t\bar{t} \, T} )=0$ and ${p\hspace{-0.45em}/ }_T/p_{t\bar{t} \, T}=m_{\tilde{\chi}^0_1}/m_t$, whereas SM background events that mainly come from top quark pair production events do not show such a tendency.
When the mass spectrum deviates from the equal-velocity limit, with the value of $\Delta m \equiv m_{\tilde{t}_1} - m_t - m_{\tilde{\chi}^0_1}$ being as large as $O$(1) GeV, the top quark and the neutralino from one *stop* decay have slightly different velocity vectors. Even in the equal-velocity limit, the sizable top quark decay width of about 2 GeV causes separation of the velocity vectors of the reconstructed top quark and the neutralino because the *stop* decay via off-shell top quark does not satisfy the kinematic condition that leads to eq. (\[momenta relation\]). These departures from the equal-velocity limit can be parametrized by the 3-momentum of the reconstructed top quark in the rest frame of the mother *stop*, which we denote by $\vec{P}$. The absolute value of $\vec{P}$, $P$, is the solution to eq. (\[P-sol\]), $$\begin{aligned}
m_{\tilde{t}_1} \ = \ m_t + m_{\tilde{\chi}^0_1} + \Delta m \ &= \
\sqrt{P^2 + m_{\tilde{\chi}^0_1}^2} + \sqrt{P^2 + m_t^2 - x \Gamma_t m_t}
\label{P-sol}\end{aligned}$$ where $x$ measures the off-shellness of the top quark propagator in the *stop* decay and follows the probability density function $f(x)$ and the range of eq. (\[prob\]), $$\begin{aligned}
f(x) \, {\rm d}x \ &\propto \ P(x)^2 \, \frac{{\rm d}P(x)}{{\rm d}x} \, \frac{1}{x^2+1} \, {\rm d}x,
\nonumber \\
- \frac{2 m_t \Delta m + \Delta m^2}{m_t \Gamma_t} \, &< \, x \, < \, \frac{2 m_{\tilde{t}_1} m_t + 2 m_{\tilde{t}_1} \Delta m - 2 m_t \Delta m - \Delta m^2}{m_t \Gamma_t}
\label{prob}\end{aligned}$$ where $P(x)$ in turn depends on $x$ through eq. (\[P-sol\]). For $\Delta m \ll m_t, m_{\tilde{\chi}^0_1}$ and $\vert x \vert \lesssim 10$, the solution to eq. (\[P-sol\]) is approximately given by $$\begin{aligned}
P \ &\simeq \ \sqrt{ \frac{ m_t }{ m_{\tilde{t}_1} } ( m_{\tilde{t}_1} - m_t ) } \sqrt{ 2\Delta m - x \Gamma_t } \ ,\end{aligned}$$ and the probability density function for $P$ is expressed as $$\begin{aligned}
\bar{f}(P) \, {\rm d}P \ &\propto \ P^2 \, \frac{1}{\left(P^2 - 2\Delta m \, \frac{m_t(m_{\tilde{t}_1}-m_t)}{m_{\tilde{t}_1}}\right)^2 + \left(\Gamma_t \, \frac{m_t(m_{\tilde{t}_1}-m_t)}{m_{\tilde{t}_1}}\right)^2 } \, {\rm d}P \ .\end{aligned}$$ For example, for $\Delta m=2$ GeV, $m_{\tilde{t}_1}=223$ GeV, $m_t=173$ GeV, $\Gamma_t=1.56$ GeV, $\bar{f}(P)$ is maximized at $P\simeq 13$ GeV; for $\Delta m=0$ GeV, $m_{\tilde{t}_1}=223$ GeV, $m_t=173$ GeV, $\Gamma_t=1.56$ GeV, $\bar{f}(P)$ is maximized at $P\simeq 8$ GeV.
In the frame where the mother *stop* is boosted transversely by the 2-velocity $(\beta, 0)$, the transverse momenta of the reconstructed top quark and the neutralino are given by $$\begin{aligned}
\vec{p}_{t \, T} \ &= \ ( \, \gamma \beta \sqrt{P^2+m_t^2}+\gamma P \cos \theta, \, P \sin \theta \cos \phi \, ), \nonumber \\
\vec{p}_{\tilde{\chi}^0_1 \, T} \ &= \ ( \, \gamma \beta \sqrt{P^2+m_{\tilde{\chi}^0_1}^2}-\gamma P \cos \theta, \, -P \sin \theta \cos \phi \, ),\end{aligned}$$ where $\gamma=1/\sqrt{1-\beta^2}$ and ($P \cos \theta, \, P \sin \theta \cos \phi$) is the transverse component of the top quark momentum vector in the *stop* rest frame. We find that as the mother *stop* is highly boosted with $\beta \rightarrow 1$, the momentum vectors $\vec{p}_{t \, T}$ and $\vec{p}_{\tilde{\chi}^0_1 \, T}$ become collinear while the ratio of their absolute values, $p_{\tilde{\chi}^0_1 \, T} / p_{t \, T}$, only approaches to $$\begin{aligned}
\frac{\sqrt{P^2+m_{\tilde{\chi}^0_1}^2}+ P \cos \theta}{\sqrt{P^2+m_t^2}+ P \cos \theta}\end{aligned}$$ and never converges to $m_{\tilde{\chi}^0_1}/m_t$. As for *stop* pair production events, the departure of $p_{\tilde{\chi}^0_1 \, T} / p_{t \, T}$ from $m_{\tilde{\chi}^0_1}/m_t$ gives rise to the deviation of $\Delta \phi( \vec{{p\hspace{-0.45em}/ }}_T, \, \vec{p}_{t\bar{t} \, T} )$ from zero. However, we can alleviate such a deviation by selecting events with small azimuthal angle between the two top quarks. On the other hand, the deviation of ${p\hspace{-0.45em}/ }_T/p_{t\bar{t} \, T}$ from $m_{\tilde{\chi}^0_1}/m_t$ cannot be lessened by any event selection criteria. To summarize, we have *stop* events gathered near the line of $\Delta \phi( \vec{{p\hspace{-0.45em}/ }}_T, \, \vec{p}_{t\bar{t} \, T} )=0$ even with the sizable top quark decay width and even when the mass spectrum is slightly different from the one in the equal-velocity limit, if we select those events where the two reconstructed top quarks have large transverse momenta and small azimuthal angle separation. Still, the *stop* events are scattered along the axis of ${p\hspace{-0.45em}/ }_T/p_{t\bar{t} \, T}$ and only possess a broad peak around the point of ${p\hspace{-0.45em}/ }_T/p_{t\bar{t} \, T}=m_{\tilde{\chi}^0_1}/m_t$.\
First we simulate *stop* pair production events assuming an ideal detector with which top quark momenta are reconstructed perfectly and missing transverse momentum solely comes from neutralinos when the top quarks decay hadronically, and study the distribution of the simulated events on the plane of $$\begin{aligned}
(\Delta \phi( \vec{{p\hspace{-0.45em}/ }}_T, \, \vec{p}_{t\bar{t} \, T} ), \, {p\hspace{-0.45em}/ }_T/p_{t\bar{t} \, T}) \ .\end{aligned}$$ We thereby examine to what extent *stop* pair production events are concentrated around the point of $$\begin{aligned}
(\Delta \phi( \vec{{p\hspace{-0.45em}/ }}_T, \, \vec{p}_{t\bar{t} \, T} ), \, {p\hspace{-0.45em}/ }_T/p_{t\bar{t} \, T}) \, &= \, (0, \, m_{\tilde{\chi}^0_1}/m_t) \ .\end{aligned}$$ By using MadGraph5 and MadEvent [@mg] with the parton distribution function set NN23LO1 [@nn23lo1], we calculate the matrix elements for the process eq. (\[parton\]) with $\sqrt{s}=14$ TeV, $$\begin{aligned}
p p \ &\rightarrow \ \tilde{t}_1 \bar{\tilde{t}}_1 \ + \ 1 \ {\rm jet}, \ \ \
\tilde{t}_1 \ \rightarrow \ b \, j \, j \, \tilde{\chi}^0_1, \ \ \
\bar{\tilde{t}}_1 \ \rightarrow \ \bar{b} \, j \, j \, \tilde{\chi}^0_1
\label{parton}
$$ where $b$ denotes a bottom quark jet and $j$ denotes a light flavor jet, and thus generate parton-level events. We here focus on events with the hadronic decay of the top quark pair, to realize full reconstruction of the top quark momenta and avoid contamination of missing transverse momentum from neutrinos due to top quark decays. We require one additional jet along with the *stop* pair in the event generation because in the equal-velocity limit, missing transverse momentum from the neutralinos arises only when the *stop* pair is boosted. We test the following four benchmark mass spectra: $$\begin{aligned}
(m_{\tilde{t}_1}, \ m_{\tilde{\chi}^0_1}, \ \Delta m = m_{\tilde{t}_1}-m_t-m_{\tilde{\chi}^0_1})
& \ = \ (223 {\rm GeV}, \ 50 {\rm GeV}, \ 0 {\rm GeV}) & {\rm (11a)} \\
& \ = \ (223 {\rm GeV}, \ 48 {\rm GeV}, \ 2 {\rm GeV}) & {\rm (11b)} \\
& \ = \ (323 {\rm GeV}, \ 150 {\rm GeV}, \ 0 {\rm GeV}) & {\rm (11c)} \\
& \ = \ (323 {\rm GeV}, \ 148 {\rm GeV}, \ 2 {\rm GeV}) & {\rm (11d)}
$$ with the top quark mass and width set at $m_t=173$ GeV and $\Gamma_t=1.56$ GeV. For simplicity, it is assumed that the lighter *stop* $\tilde{t}_1$ is purely composed of SU(2)$_L$ singlet *stop* and the lightest neutralino $\tilde{\chi}^0_1$ is composed of Bino, and that the other SUSY particles are decoupled from the mass spectrum. Hence the *stop* decays into the neutralino and the top quark with 100% branching ratio. The *stop* decay width is calculated and found to be $\Gamma_{\tilde{t}_1}=$0.00753 GeV, 0.0183 GeV, 0.0145 GeV and 0.0371 GeV for the mass spectra eqs. (11a), (11b), (11c) and (11d), respectively. For events generated, we require that
- the transverse momentum of each of the reconstructed top quarks should satisfy $p_T > 200$ GeV. [(12a)]{}
- the azimuthal angle between the reconstructed top quarks should satisfy $\Delta \phi < \pi/2$. [(12b)]{}
They reflect our observations mentioned in the previous paragraph.
In Figure 1, we plot events of the process eq. (\[parton\]) on the 2-dimensional plane spanned by the azimuthal angle between the missing transverse momentum vector $\vec{{p\hspace{-0.45em}/ }}_T$ and the top-quark-pair transverse momentum vector $\vec{p}_{t\bar{t} \, T}$, and the ratio of their absolute values. Here the top-quark-pair transverse momentum vector is reconstructed from the six jets coming from the hadronic decay of the top quark pair. The four subfigures, (1a) to (1d), show results for the four mass spectra, eqs. (11a) to (11d), respectively. The leading order perturbative QCD calculation predicts the cross section for the process eq. (\[parton\]) after the selection cuts eqs. (12a) and (12b) to be 30.9 fb for $(m_{\tilde{t}_1}, \ m_{\tilde{\chi}^0_1})=$(223 GeV, 50 GeV), 33.6 fb for $(m_{\tilde{t}_1}, \ m_{\tilde{\chi}^0_1})=$(223 GeV, 48 GeV), 2.71 fb for $(m_{\tilde{t}_1}, \ m_{\tilde{\chi}^0_1})=$(323 GeV, 150 GeV), 3.19 fb for $(m_{\tilde{t}_1}, \ m_{\tilde{\chi}^0_1})=$(323 GeV, 148 GeV), at $\sqrt{s}=14$ TeV. Note that the cross sections do depend on $m_{\tilde{\chi}^0_1}$ because the selection cut eq. (12b) is sensitive to the proportion of the on-shell top quark contribution to the *stop* decay. Each dot corresponds to 1 event for 10 fb$^{-1}$ of data in subfigures (1a) and (1b), and to 1 event for 100 fb$^{-1}$ of data in subfigures (1c) and (1d).
![The scatter plots of events of the process eq. (\[parton\]) after the selection cuts eqs. (12a) and (12b) in $\sqrt{s}=14$ TeV $pp$ collisions, on the plane spanned by the azimuthal angle between the missing transverse momentum vector $\vec{{p\hspace{-0.45em}/ }}_T$ and the top-pair transverse momentum vector $\vec{p}_{t\bar{t} \, T}$, $\Delta \phi (\vec{{p\hspace{-0.45em}/ }}_T, \vec{p}_{t\bar{t} \, T})$, and the ratio of their absolute values, ${p\hspace{-0.45em}/ }_T / p_{t\bar{t} \, T}$. It is assumed that the top-quark-pair momentum is perfectly reconstructed and that missing transverse momentum solely comes from neutralinos. The four subfigures are for $(m_{\tilde{t}_1}, \, m_{\tilde{\chi}^0_1}) =$(223 GeV, 50 GeV) (1a), (223 GeV, 48 GeV) (1b), (323 GeV, 150 GeV) (1c) and (323 GeV, 148 GeV) (1d), respectively. The thick horizontal line inside each figure corresponds to the line of ${p\hspace{-0.45em}/ }_T / p_{t\bar{t} \, T} = m_{\tilde{\chi}^0_1}/m_t$. ](22350par.eps){width="80mm"}
![The scatter plots of events of the process eq. (\[parton\]) after the selection cuts eqs. (12a) and (12b) in $\sqrt{s}=14$ TeV $pp$ collisions, on the plane spanned by the azimuthal angle between the missing transverse momentum vector $\vec{{p\hspace{-0.45em}/ }}_T$ and the top-pair transverse momentum vector $\vec{p}_{t\bar{t} \, T}$, $\Delta \phi (\vec{{p\hspace{-0.45em}/ }}_T, \vec{p}_{t\bar{t} \, T})$, and the ratio of their absolute values, ${p\hspace{-0.45em}/ }_T / p_{t\bar{t} \, T}$. It is assumed that the top-quark-pair momentum is perfectly reconstructed and that missing transverse momentum solely comes from neutralinos. The four subfigures are for $(m_{\tilde{t}_1}, \, m_{\tilde{\chi}^0_1}) =$(223 GeV, 50 GeV) (1a), (223 GeV, 48 GeV) (1b), (323 GeV, 150 GeV) (1c) and (323 GeV, 148 GeV) (1d), respectively. The thick horizontal line inside each figure corresponds to the line of ${p\hspace{-0.45em}/ }_T / p_{t\bar{t} \, T} = m_{\tilde{\chi}^0_1}/m_t$. ](22348par.eps){width="80mm"}
![The scatter plots of events of the process eq. (\[parton\]) after the selection cuts eqs. (12a) and (12b) in $\sqrt{s}=14$ TeV $pp$ collisions, on the plane spanned by the azimuthal angle between the missing transverse momentum vector $\vec{{p\hspace{-0.45em}/ }}_T$ and the top-pair transverse momentum vector $\vec{p}_{t\bar{t} \, T}$, $\Delta \phi (\vec{{p\hspace{-0.45em}/ }}_T, \vec{p}_{t\bar{t} \, T})$, and the ratio of their absolute values, ${p\hspace{-0.45em}/ }_T / p_{t\bar{t} \, T}$. It is assumed that the top-quark-pair momentum is perfectly reconstructed and that missing transverse momentum solely comes from neutralinos. The four subfigures are for $(m_{\tilde{t}_1}, \, m_{\tilde{\chi}^0_1}) =$(223 GeV, 50 GeV) (1a), (223 GeV, 48 GeV) (1b), (323 GeV, 150 GeV) (1c) and (323 GeV, 148 GeV) (1d), respectively. The thick horizontal line inside each figure corresponds to the line of ${p\hspace{-0.45em}/ }_T / p_{t\bar{t} \, T} = m_{\tilde{\chi}^0_1}/m_t$. ](323150par.eps){width="80mm"}
![The scatter plots of events of the process eq. (\[parton\]) after the selection cuts eqs. (12a) and (12b) in $\sqrt{s}=14$ TeV $pp$ collisions, on the plane spanned by the azimuthal angle between the missing transverse momentum vector $\vec{{p\hspace{-0.45em}/ }}_T$ and the top-pair transverse momentum vector $\vec{p}_{t\bar{t} \, T}$, $\Delta \phi (\vec{{p\hspace{-0.45em}/ }}_T, \vec{p}_{t\bar{t} \, T})$, and the ratio of their absolute values, ${p\hspace{-0.45em}/ }_T / p_{t\bar{t} \, T}$. It is assumed that the top-quark-pair momentum is perfectly reconstructed and that missing transverse momentum solely comes from neutralinos. The four subfigures are for $(m_{\tilde{t}_1}, \, m_{\tilde{\chi}^0_1}) =$(223 GeV, 50 GeV) (1a), (223 GeV, 48 GeV) (1b), (323 GeV, 150 GeV) (1c) and (323 GeV, 148 GeV) (1d), respectively. The thick horizontal line inside each figure corresponds to the line of ${p\hspace{-0.45em}/ }_T / p_{t\bar{t} \, T} = m_{\tilde{\chi}^0_1}/m_t$. ](323148par.eps){width="80mm"}
From Figure 1, we have confirmed that the *stop* pair production events gather around the point of ($\Delta \phi( \vec{{p\hspace{-0.45em}/ }}_T, \, \vec{p}_{t\bar{t} \, T} ), \, {p\hspace{-0.45em}/ }_T/p_{t\bar{t} \, T}) \, = \, (0, \, m_{\tilde{\chi}^0_1}/m_t)$. Most of the *stop* events are distributed in the region with $\Delta \phi( \vec{{p\hspace{-0.45em}/ }}_T, \, \vec{p}_{t\bar{t} \, T} )<\pi/10$, which suggests that $\Delta \phi( \vec{{p\hspace{-0.45em}/ }}_T, \, \vec{p}_{t\bar{t} \, T} )$ can be an efficient discriminating variable for selecting *stop* events in the presence of SM backgrounds. On the other hand, the event distribution along the axis of ${p\hspace{-0.45em}/ }_T/p_{t\bar{t} \, T}$ can be useful for estimating the mass of the *stop* and neutralino.
In subfigures (1a) and (1c), the events are scattered due to the finite width of the top quark, whereas in subfigures (1b) and (1d), they are scattered due to the combination of the finite top quark width and the deviation of the mass spectrum from the equal-velocity limit by $\Delta m=2$ GeV. As a reference, we have also checked the event distributions for the mass spectra eqs. (1b) and (1c) with the top quark width set to zero, namely, with the assumption that only on-shell top quark contributes to the *stop* decay. Such distributions are unphysical, but useful for studying how a finite value of $\Delta m \equiv m_{\tilde{t}_1} - m_t - m_{\tilde{\chi}^0_1}$ alone causes the scattering of events. It turns out that the distributions of *stop* events on the plane of ($\Delta \phi( \vec{{p\hspace{-0.45em}/ }}_T, \, \vec{p}_{t\bar{t} \, T} ), \, {p\hspace{-0.45em}/ }_T/p_{t\bar{t} \, T}$) are similar for the cases with $(\Delta m, \, \Gamma_t)=(0, \, 1.56)$ GeV, $(\Delta m, \, \Gamma_t)=(2, \, 1.56)$ GeV and $(\Delta m, \, \Gamma_t)=(2, \, 0)$ GeV. Based on this observation, in the detector-level analysis later on, we neglect off-shell top quark contribution and consider mass spectra with $(\Delta m, \, \Gamma_t)=(2, \, 0)$ GeV.\
Next we perform a detector-level analysis, taking into account parton showering, hadronization and detector responses. We again focus on *stop* events with the hadronic decay of the top quark pair, where the momenta of the top quarks can be reconstructed from six jets. In addition, we consider the dominant background that originates from SM top quark pair production events. We examine how effective the plot on the plane of ($\Delta \phi( \vec{{p\hspace{-0.45em}/ }}_T, \, \vec{p}_{t\bar{t} \, T} ), \, {p\hspace{-0.45em}/ }_T/p_{t\bar{t} \, T}$) is for distinguishing *stop* events from SM backgrounds in the equal-velocity scenario. The benchmark mass spectra are eqs. (11b) and (11d), and we take into account only on-shell top quark contribution to the *stop* decay, which has been justified in the previous paragraph. It is again assumed that the lighter *stop* $\tilde{t}_1$ is purely composed of SU(2)$_L$ singlet *stop*, the lightest neutralino $\tilde{\chi}^0_1$ is composed of Bino and the other SUSY particles are decoupled.
Before event simulation, we evaluate the cross sections for the *stop* pair production and the SM top quark pair production to assess the viability of the analysis. In Figure 2, the cross sections for the *stop* pair production, $\sigma(\tilde{t}_1 \bar{\tilde{t}}_1)$, the *stop* pair production associated with one parton whose transverse momentum $p_{jT}$ is larger than 300 GeV, $\sigma(\tilde{t}_1 \bar{\tilde{t}}_1 j; p_{jT}>300 {\rm GeV})$, and the *stop* pair production associated with one parton with $p_{jT}>$500 GeV, $\sigma(\tilde{t}_1 \bar{\tilde{t}}_1 j; p_{jT}>500 {\rm GeV})$, in $\sqrt{s}=14$ TeV $pp$ collisions are plotted. These cross sections are calculated at the leading order in perturbative QCD. Also shown in Figure 2 are the cross sections for the SM top quark pair prodcution, $\sigma(t \bar{t})_{LO}$, $\sigma(t \bar{t})_{NLO}$, and the SM top quark pair production associated with one parton with $p_{jT}>300$ GeV, $\sigma(t \bar{t}; p_{jT}>300 {\rm GeV})_{LO}$, $\sigma(t \bar{t}; p_{jT}>300 {\rm GeV})_{NLO}$, which are calculated at the leading order for $\sigma(t \bar{t})_{LO}$, $\sigma(t \bar{t}; p_{jT}>300 {\rm GeV})_{LO}$ and at the next-to-leading order for $\sigma(t \bar{t})_{NLO}$, $\sigma(t \bar{t}; p_{jT}>300 {\rm GeV})_{NLO}$ in perturbative QCD. Here the next-to-leading order calculation is done with MadGraph5$\underline{ \ }$aMC@NLO [@mg]. The K-factor for the SM top quark pair production cross section is evaluated to be 1.34, and that for the production cross section of a SM top quark pair + one parton with $p_{jT}>300$ GeV is evaluated to be 1.09. The process of *stop* pair production associated with one parton with $p_{jT}>300$ GeV resembles the situation where a pair of *stop*s with the mass of 223 GeV are produced with a transverse boost balanced by a hard jet, and each of the top quarks coming from their decays has a transverse momentum above 200 GeV and their azimuthal angle is below $\pi/2$, which is the situation where the selection cuts eqs. (16g) and (16h) below are imposed. Likewise, the process of *stop* pair production associated with one parton with $p_{jT}>500$ GeV resembles the situation where a pair of *stop*s with the mass of 323 GeV are produced with the same selection cuts. The process of top quark pair production associated with one parton with $p_{jT}>300$ GeV resembles the situation where a top quark pair is produced with a transverse boost balanced by a hard jet, with the transverse momentum of each top quark above 200 GeV and with their azimuthal angle below $\pi/2$.
{width="100mm"}
Let us proceed to event simulation with full detector-level analysis. By MadGraph5 and MadEvent with the parton distribution function set NN23LO1, we generate signal events from the following processes $$\begin{aligned}
p p \ &\rightarrow \ \tilde{t}_1 \bar{\tilde{t}}_1 \ + \ 1 \ {\rm jet}, \ \
\tilde{t}_1 \ \rightarrow \ t \, \tilde{\chi}^0_1, \ \
\bar{\tilde{t}}_1 \ \rightarrow \ \bar{t} \, \tilde{\chi}^0_1; \nonumber
\\
p p \ &\rightarrow \ \tilde{t}_1 \bar{\tilde{t}}_1 \ + \ 2 \ {\rm jets}, \ \
\tilde{t}_1 \ \rightarrow \ t \, \tilde{\chi}^0_1, \ \
\bar{\tilde{t}}_1 \ \rightarrow \ \bar{t} \, \tilde{\chi}^0_1,
\label{sig}
$$ with the precuts that
- each top quark should have a transverse momentum above 150 GeV. (14a)
- the azimuthal angle between the two top quarks shoule be smaller than $2\pi/3$. (14b)
- the $k_T$ distance of the jets produced in association with the *stop* pair should be above 100 GeV. (14c)
Parton showering and hadronization are simulated by PYTHIA8 [@pythia], and the 1-jet and 2-jet events are matched by MLM matching scheme [@mlm] with the matching scale of 100 GeV. The top quark decay is simulated by PYTHIA8, with the off-shell top quark contribution to the *stop* decay neglected. Detector simulation and jet clustering are performed by PGS4 [@pgs], where we use anti-$k_T$ algorithm [@antikt] with the distance parameter of $\Delta R=0.4$. We also generate background events from the following processes by MadGraph5 and MadEvent with the same precuts, $$\begin{aligned}
p p \ &\rightarrow \ t \bar{t} \ + \ 1 \ {\rm jet}; \ \ \nonumber
\\
p p \ &\rightarrow \ t \bar{t} \ + \ 2 \ {\rm jets}. \ \
\label{bkg}
$$ Parton showering, hadronization, the top qurak decay, event matching, detector responses and jet clustering are simulated in the same way as the signal events. The next-to-leading order QCD corrections are taken into account only for the background process, by multiplying the number of background events by the K-factor evaluated in Figure 2 for the production of a SM top pair + one parton with $p_T>300$ GeV, which has been found to be 1.09.
We impose the following selection cuts on the signal and background events generated:
- Require six jets whose pseudo-rapidities satisfy $\vert \eta_j \vert < 2.8$ and whose transverse momenta satisfy $p_{jT} > 30$ GeV. (16a)
- Exactly two jets should be b-tagged. (16b)
- Require missing transverse momentum ${p\hspace{-0.45em}/ }_T > 30$ GeV. (16c)
- Azimuthal angle between the missing transverse momentum vector and the transverse momentum vector of each of the three leading jets should be larger than $0.1\pi$,
$\Delta \phi(\vec{{p\hspace{-0.45em}/ }}_T, \, \vec{p}_{j_{1,2,3}T}) > 0.1\pi$. (16d)
- The event is vetoed if it contains an electron with $\vert \eta \vert < 2.47$ and $p_T > 10$ GeV, or a muon with $\vert \eta \vert < 2.4$ and $p_T > 10$ GeV. (16e)
- Hadronic decays of two top quarks should be reconstructed successfully by the method below. (16f)
- The reconstructed top quarks, $t_1, t_2$, should satisfy
$p_{t_1 \, T} > 200$ GeV, $p_{t_2 \, T} > 200$ GeV. (16g)
- The azimuthal angle between the reconstructed top quarks should satisfy
$\Delta \phi(\vec{p}_{t1 \, T}, \, \vec{p}_{t2 \, T}) < 0.5\pi$. (16h)
In eq. (16f), we use the following method to reconstruct the two top quarks that decay hadronically. First we select a pair of non-b-tagged jets, $j_1, \, j_2,$ whose invariant mass is most close to the $W$ boson mass. If the invariant mass is separated from the $W$ boson mass by more than 20 GeV, the method fails and the event is vetoed: $$\begin{aligned}
{\rm Look \ for \ } j_1, \, j_2 \ {\rm that \ give \ the \ minimum \ of} \ \vert m(j_1, \, j_2) - M_W \vert.
\\ {\rm The \ event \ is \ vetoed \ if} \ {\rm min}\{ \vert m(j_1, \, j_2) - M_W \vert \} \, > \, 20 \, {\rm GeV}.\end{aligned}$$ Next we select one b-tagged jet, $b_1$, for which the three-jet invariant mass for this jet and the two non-b-tagged jets selected above, $\bar{j}_1, \, \bar{j}_2,$ is most close to the top quark mass. If the invariant mass is separated from the top quark mass by more than 50 GeV, the method fails and the event is vetoed: $$\begin{aligned}
{\rm Look \ for \ } b_1 \ {\rm that \ gives \ the \ minimum \ of} \ \vert m(\bar{j}_1, \, \bar{j}_2, \, b_1) - m_t \vert.
\\ {\rm The \ event \ is \ vetoed \ if} \ {\rm min}\{ \vert m(\bar{j}_1, \, \bar{j}_2, \, b_1) - m_t \vert \} \, > \, 50 \, {\rm GeV}.\end{aligned}$$ The two non-b-tagged jets $j_1, \, j_2$ and one b-tagged jet $b_1$ are considered to constitute one hadronically decaying top quark. We repeat the same procedures on the remaining jets to reconstruct the other top quark. This method can be regarded as a simplified version of the one used in ref. [@atlas; @sm; @hadronic; @top], where a likelihood function is utilized to select two sets of two non-b-tagged jets and one b-tagged jet that constitute a hadronically decaying top quark. Note that in our method, we do not rely on information about the angular distance between a pair of jets or combined jets This is because the decay products of the two top quarks are likely to overlap in the $\phi-\eta$ space, as eq. (16h) demands that the two top quarks have small azimuthal angular distance.
Our simulation of b-tagging in eq. (16b) is based on “loose-b-tag” implemented in PGS4 [@pgs], which gives a conservative estimate on the power of b-tagging compared to the performance achieved by the ATLAS and CMS collaborations. Here the efficiency and mis-tagging rate are simulated to depend on the energy and pseudo-rapidity of the jet. The efficiency of b-tagging is simulated as above 40% and below 60% for b-jets with $p_T > 20$ GeV and $\vert \eta_j \vert < 1.0$. The rate for mis-tagging a light flavor jet coming from up, down or strange quark or gluon is simulated to be suppressed below 4%, and that for mis-tagging a charm jet is set to be 26% of the b-tagging efficiency.
We have further comments on the selection cuts. The mild requirement on ${p\hspace{-0.45em}/ }_T$ in eq. (16c) is because missing transverse momentum is suppressed in the equal-velocity scenario compared to the case with general mass spectra. The cut eq. (16d) is to reduce fake missing transverse momentum due to jet energy mismeasurement. The cut eq. (16e) is to reduce SM top quark pair background events where missing transverse momentum originates from their semi-leptonic decays. The cuts eqs. (16g) and (16h) are special cuts that ameliorate the concentration of *stop* signal events on the plane of ($\Delta \phi( \vec{{p\hspace{-0.45em}/ }}_T, \, \vec{p}_{t\bar{t} \, T} ), \, {p\hspace{-0.45em}/ }_T/p_{t\bar{t} \, T}$).
We find that the cross section of the *stop* pair production process after the selection cuts eqs. (16a)-(16h) is 0.60 fb and 0.077 fb for the mass spectra $(m_{\tilde{t}_1}, \, m_{\tilde{\chi}_1^0})=$(223, 48) GeV and (323, 148) GeV, respectively, and the cross section of the SM top quark pair production process after the same cuts is 5.95 fb. Here, the K-factor of 1.09 derived in Figure 2 has already been multiplied to the top quark pair production cross section. Table 1 shows the cross sections of *stop* pair production events with $(m_{\tilde{t}_1}, \, m_{\tilde{\chi}_1^0})=$(223, 48) GeV and (323, 148) GeV and top quark pair production events, at each stage of event selection.
$(m_{\tilde{t}_1}, \, m_{\tilde{\chi}_1^0})=$(223, 48) GeV $(m_{\tilde{t}_1}, \, m_{\tilde{\chi}_1^0})=$(323, 148) GeV SM top quark
------------------------ ------------------------------------------------------------ ------------------------------------------------------------- ----------------------
Before the cuts 3.8 $\times 10^4$ fb 6.6 $\times 10^3$ fb 7.5 $\times 10^5$ fb
After eqs. (16a,b,e,f) 4.2 $\times 10^2$ fb 1.0 $\times 10^2$ fb 9.1 $\times 10^3$ fb
After eqs. (16c,d) 1.9 $\times 10^2$ fb 63 fb 1.3 $\times 10^3$ fb
After eqs. (16g,h) 0.60 fb 0.077 fb 5.95 fb
: The cross sections of *stop* pair production events with $(m_{\tilde{t}_1}, \, m_{\tilde{\chi}_1^0})=$(223, 48) GeV and (323, 148) GeV and top quark pair production events at four stages of event selection that are, before the cuts, after the reconstruction of a hadronically-decaying top quark pair eqs. (16a, 16b, 16e, 16f), after the cuts on missing transverse momentum eqs. (16c, 16d), and after the cuts on the momenta of reconstructed top quarks eqs. (16g, 16h).
Note that in Table 1, the cross sections before the selection cuts eqs. (16g, 16h) are evaluated by generating *stop* pair or top quark pair production events with 0 jet and 1 jet with the precut of eq. (14c) only, passing them to parton showering and matching, and doing the rest of the simulations. Also, the K-factor of 1.34, instead of 1.09, derived in Figure 2 has been multiplied to the top quark pair production cross sections before the cuts eqs. (16g, 16h).
We plot in Figure 3 *stop* signal events and SM top quark background events on the plane spanned by the azimuthal angle between missing transverse momentum vector $\vec{{p\hspace{-0.45em}/ }}_T$ and the reconstructed top quark pair transverse momentum vector $\vec{p}_{t\bar{t} \, T}$, $\Delta \phi(\vec{p}_{t\bar{t} \, T}, \vec{{p\hspace{-0.45em}/ }}_T)$, and the ratio of their absolute values, ${p\hspace{-0.45em}/ }_T / p_{t\bar{t} \, T}$.
{width="90mm"}
{width="90mm"}
{width="90mm"}
From Figure 3, we find that *stop* pair production events in the equal-velocity scenario are concentrated around the point of $(\Delta \phi( \vec{{p\hspace{-0.45em}/ }}_T, \, \vec{p}_{t\bar{t} \, T} ), \, {p\hspace{-0.45em}/ }_T/p_{t\bar{t} \, T}) \, = \,
(0, \, m_{\tilde{\chi}^0_1}/m_t)$ even at the detector level. We thus confirm that the kinematic relations of the top quark pair momentum and the missing transverse momentum which we observe at the parton-level in Figure 1 are preserved through parton showering, hadronization, detector simulations, jet clustering and the reconstruction of hadronically-decaying top quarks. On the other hand, SM top pair production background events are distributed broadly along the axis of $\Delta \phi( \vec{{p\hspace{-0.45em}/ }}_T, \, \vec{p}_{t\bar{t} \, T} )$ and concentrated in the region with small ${p\hspace{-0.45em}/ }_T/p_{t\bar{t} \, T}$. Such a distribution is explained in the following manner. There are two principal sources of missing transverse momentum in the background events, which are the semi-leptonic decay of top quark pair where the charged lepton is a hadronically-decaying tau or evades detection, and the leptonic decay of B meson formed by a b-quark from the top quark decay. The direction of such missing transverse momentum has no strong correlation with that of the momentum of wrongly-reconstructed hadronically-decaying top quark pair, for which an initial or final state radiation is misidentified as a top quark decay product. Also, the absolute value of such missing transverse momentum remains small no matter how boosted the wrongly-reconstructed top quark pair is. Comparison of Subfigures (4a) and (4b) with Subfigure (4c) implies that the selection cut on $\Delta \phi( \vec{{p\hspace{-0.45em}/ }}_T, \, \vec{p}_{t\bar{t} \, T} )$ after the selection cuts eqs. (16a-16h) is useful for enhancing the signal-to-background ratio in the search for *stop*s in the equal-velocity scenario. Then the distribution of ${p\hspace{-0.45em}/ }_T/p_{t\bar{t} \, T}$ gives a further criterion for distinguishing *stop* events from SM top quark events if the neutralino mass is larger than about 50 GeV. As a bonus, the same distribution makes it possible to estimate the neutralino mass and the *stop* mass with the assumption of the equal-velocity scenario.
We finally demonstrate how the analysis using the scatter plots of Figure 3 has an advantage over the analysis without the plots, by deriving the signal significance, $S/\sqrt{S+B}$, by both analyses. In the latter analysis, we simply count the numbers of simulated signal and background events that pass the cuts eqs. (16a-16h). With 300 fb$^{-1}$ of data, the significance in the cases with $(m_{\tilde{t}_1}, \, m_{\tilde{\chi}_1^0})=$(223, 48) GeV and (323, 148) GeV is found to be $$\begin{aligned}
S/\sqrt{S+B} \ &= \ 4.1 \ \ \ {\rm for} \ (m_{\tilde{t}_1}, \, m_{\tilde{\chi}_1^0})=(223,~48)~{\rm GeV},
\nonumber \\
S/\sqrt{S+B} \ &= \ 0.54 \ \ \ {\rm for} \ (m_{\tilde{t}_1}, \, m_{\tilde{\chi}_1^0})=(323,~148)~{\rm GeV},
\ {\rm with \ } 300~{\rm fb}^{-1} \ {\rm of \ data.} \end{aligned}$$ On the other hand, in the analysis that makes use of the scatter plots, we impose the selection cuts eq. (18a, 18b) in addition to eqs. (16a-16h): $$\begin{aligned}
\Delta \phi( \vec{{p\hspace{-0.45em}/ }}_T, \, \vec{p}_{t\bar{t} \, T} ) \ &< \ 0.5. \ \ \ \ \ \ \ \ \ \ (18a)
\\
{p\hspace{-0.45em}/ }_T/p_{t\bar{t} \, T} \ &> \ 0.2. \ \ \ \ \ \ \ \ \ \ (18b)\end{aligned}$$ These cuts are based on the scatter plots of Figure 3, which indicate that the region defined by eqs. (18a, 18b) is rich with signal events for cases with $m_{\tilde{\chi}_1^0}\gtrsim 50$ GeV, and contains considerably less background events than the region outside. After the cuts eqs. (18a, 18b), we find $$\begin{aligned}
S/\sqrt{S+B} \ &= \ 8.4 \ \ \ {\rm for} \ (m_{\tilde{t}_1}, \, m_{\tilde{\chi}_1^0})=(223,~48)~{\rm GeV},
\nonumber \\
S/\sqrt{S+B} \ &= \ 1.6 \ \ \ {\rm for} \ (m_{\tilde{t}_1}, \, m_{\tilde{\chi}_1^0})=(323,~148)~{\rm GeV},
\ {\rm with \ } 300~{\rm fb}^{-1} \ {\rm of \ data.} \end{aligned}$$ We can further improve the significance by taking advantage of the peak distribution of signal events on ($\Delta \phi( \vec{{p\hspace{-0.45em}/ }}_T, \, \vec{p}_{t\bar{t} \, T} ), \, {p\hspace{-0.45em}/ }_T/p_{t\bar{t} \, T}$) plane, but this is beyond the scope of our theoretical study.\
We have studied the equal-velocity scenario, in which a *stop* and a neutralino are the lightest SUSY particles and the *stop* mass is close to the sum of the SM top quark mass and the neutralino mass, which forces the top quark and the neutralino coming from the decay of a *stop* to have similar velocity vectors. In this case, it is difficult to distinguish *stop* pair production events from the SM top quark pair production background because missing transverse momentum due to neutralinos is suppressed. We have proposed a novel analysis technique aiming at discovering the *stop* signal in the equal-velocity scenario, which takes advantage of the information that, in the equal-velocity limit, missing transverse momentum vector due to the neutralinos and the transverse momentum vector of the SM top quark pair have the same direction and the ratio of their absolute values equals to the ratio of the neutralino mass and the top quark mass. We have investigated the distributions of *stop* pair production events and SM top quark pair production events that pass special selection cuts on the reconstructed top quarks, on the plane of $( \Delta \phi(\vec{p}_{t\bar{t} \, T}, \vec{{p\hspace{-0.45em}/ }}_T), \ {p\hspace{-0.45em}/ }_T / p_{t\bar{t} \, T} )$. We have found that even after a full detector simulation, *stop* events gather around the point of $\Delta \phi(\vec{p}_{t\bar{t} \, T}, \vec{{p\hspace{-0.45em}/ }}_T)=0$ and ${p\hspace{-0.45em}/ }_T / p_{t\bar{t} \, T} = m_{\tilde{\chi}_1^0}/m_t$, while the SM top quark background does not show such a tendency. Hence we expect that the event scatter plot on the plane of $( \Delta \phi(\vec{p}_{t\bar{t} \, T}, \vec{{p\hspace{-0.45em}/ }}_T), \ {p\hspace{-0.45em}/ }_T / p_{t\bar{t} \, T} )$ leads to the discovery of a *stop* signal if the mass spectrum is in accord with the equal-velocity scenario.\
[99]{} R. Kitano and Y. Nomura, Phys. Rev. D 73 (2006) 095004 \[hep-ph/0602096\]; M. Asano, H. D. Kim, R. Kitano and Y. Shimizu, JHEP [**1012**]{} (2010) 019 \[arXiv:1010.0692 \[hep-ph\]\]; M. Papucci, J. T. Ruderman and A. Weiler, JHEP [**1209**]{} (2012) 035 \[arXiv:1110.6926 \[hep-ph\]\].
N. Desai and B. Mukhopadhyaya, JHEP [**1205**]{} (2012) 057 \[arXiv:1111.2830 \[hep-ph\]\]; Z. Han, A. Katz, D. Krohn and M. Reece, JHEP [**1208**]{} (2012) 083 \[arXiv:1205.5808 \[hep-ph\]\]; G. Belanger, R. M. Godbole, L. Hartgring and I. Niessen, JHEP [**1305**]{} (2013) 167 \[arXiv:1212.3526\]; D. S. M. Alves, M. R. Buckley, P. J. Fox, J. D. Lykken and C. -T. Yu, Phys. Rev. D [**87**]{} (2013) 3, 035016 \[arXiv:1205.5805 \[hep-ph\]\]; X. -Q. Li, Z. -G. Si, K. Wang, L. Wang, L. Zhang and G. Zhu, Phys. Rev. D [**89**]{} (2014) 077703 \[arXiv:1311.6874 \[hep-ph\]\]; M. R. Buckley, T. Plehn and M. J. Ramsey-Musolf, Phys. Rev. D [**90**]{} (2014) 1, 014046 \[arXiv:1403.2726 \[hep-ph\]\].
ATLAS Collaboration, JHEP [**1409**]{} (2014) 015 \[arXiv:1406.1122 \[hep-ex\]\]; CMS Collaboration, CMS-PAS-SUS-13-015, http://cds.cern.ch/record/1635353.
ATLAS Collaboration, JHEP [**1411**]{} (2014) 118 \[arXiv:1407.0583 \[hep-ex\]\]; CMS Collaboration, EPJ C [**73**]{} (2013) 2677 \[arXiv:1308.1586 \[hep-ex\]\].
ATLAS Collaboration, JHEP [**1406**]{} (2014) 124 \[arXiv:1403.4853 \[hep-ex\]\].
ATLAS Collaboration, Phys. Rev. Lett. [**114**]{} (2015) 142001 \[arXiv:1412.4742 \[hep-ex\]\].
J. Alwall et al., JHEP [**0709**]{} (2007) 028 \[arXiv:0706.2334\[hep-ph\]\]; J. Alwall, M. Herquet, F. Maltoni, O. Mattelaer and T. Stelzer, JHEP [**1106**]{} (2011) 128 \[arXiv:1106.0522\[hep-ph\]\]; J. Alwall [*et al.*]{}, arXiv:1405.0301 \[hep-ph\].
R. D. Ball, V. Bertone, S. Carrazza, C. S. Deans, L. Del Debbio, S. Forte, A. Guffanti and N. P. Hartland [*et al.*]{}, Nucl. Phys. B [**867**]{} (2013) 244 \[arXiv:1207.1303 \[hep-ph\]\].
T. Sjostrand, S. Mrenna and P. Z. Skands, JHEP [**0605**]{} (2006) 026 \[arXiv:hep-ph/0603175\].
J. Conway et al., PGS (Pretty Good Simulation) (2009), http://physics.ucdavis.edu/ conway/research/software/pgs/pgs4-general.htm.
M. L. Mangano, M. Moretti and R. Pittau, Nucl. Phys. B 632 (2002) 343 \[arXiv:hep-ph/0108069\]; M. L. Mangano, M. Moretti, F. Piccinini and M. Treccani, JHEP [**0701**]{} (2007) 013 \[arXiv:hep-ph/0611129\].
M. Cacciari, G. P. Salam and G. Soyez, JHEP [**0804**]{}, 063 (2008) \[arXiv:0802.1189 \[hep-ph\]\].
ATLAS Collaboration, ATLAS-CONF-2012-031, http://cds.cern.ch/record/1432196.
| {
"pile_set_name": "ArXiv"
} |
addtoreset[equation]{}[section]{}
[ ]{}[ ]{}[\
]{}
As a further elaboration of the proposal of Ref.[@aaj] we address the construction of Standard-like models from configurations of stacks of orientifold planes and D-branes on an internal space with the structure ${( Gepner \,\, model)^{c=6} \times \IT^{2}}/\IZ_N$.
As a first step, the construction of $D=6$ Type II B orientifolds on Gepner points, in the diagonal invariant case and for both, odd and even, affine levels is discussed. We build up the explicit expressions for B-type boundary states and crosscaps and obtain the amplitudes among them. From such amplitudes we read the corresponding spectra and the tadpole cancellation equations.
Further compactification on a $\IT^{2}$ torus, by simultaneously orbifolding the Gepner and the torus internal sectors, is performed. The embedding of the orbifold action in the brane sector breaks the original gauge groups and leads to ${\cal N}=1$ supersymmetric chiral spectra. Whenever even orbifold action on the torus is considered, new branes, with worldvolume transverse to torus coordinates, must be included. The detailed rules for obtaining the $D=4$ model spectra and tadpole equations are shown. As an illustration we present a 3 generations Left-Right symmetric model that can be further broken to a MSSM model.
Introduction
============
The quest of the Standard Model like vacua, from open string interacting conformal field theories, received considerable attention in last years. In particular, much progress has been achieved in the context of orientifolds of Type II string compactified on Gepner models [@bw9806; @gm0306; @aaln; @blumen0310; @bw0401; @bhhw0401; @bw0403; @dhs0403; @dhs0411]. Gepner models [@gepnerlect] are special points of Calabi Yau manifolds, at string scale, that allow for a description in terms of an exactly solvable rational CFT. First string model building on rational conformal field theories was performed in heterotic string theories in the middle 80’ s [@fkss:1989; @rcftheterotic]. First preliminary studies of Type II orientifolds on Gepner points were presented in [@angelantonj] for six dimensions, and in [@bw9806] for $D=4$ dimensions.
Recent studies of open strings models on Gepner points have been based on two alternative (but equivalent) descriptions, namely, the partition function approach or the boundary state approach (see for instance [@asopen] for a review).
In the partition function approach consistent Type II orientifold partition functions are built up. Once Klein-bottle closed string partition function is identified, Möbius strip and cylinder amplitudes are included for consistency. The string spectrum can, therefore, be read out from them. Consistency implies factorization, tadpole cancellation and integer particle states multiplicities (see, for instance [@aaln] for details). On the other hand, one loop open string amplitudes can be expressed in terms of closed strings propagating among boundary and crosscap states. Once such states are identified, tadpole cancellation conditions and spectrum can be found in terms of the quantum numbers labeling those states (see for instance, [@ReS; @bhhw0401; @bw0403]). Either approach has lead to considerable progress. The rules for computing spectra and the tadpole cancellation equations have been derived for generic situations. Moreover, connections with a geometric large volume descriptions were established [@ABbranes; @bhhw0401]. Nevertheless, even if concise and rather simple generic expressions can be obtained, the computation of spectra for specific models can become rather cumbersome due to the, generically, huge number of open string states involved. Only solving the tadpole consistency equations can represent a difficult task even for a fast computer. Therefore a systematics is needed in order to be able to extract any useful information. In this sense a remarkable computing search for models with Sandard like model spectra was performed in [@dhs0403; @dhs0411; @Anastasopoulos:2006da] by restricting the scan to four stacks of SM branes, by following the ideas advanced in [@imr] in the context of intersecting brane models [@intersec] on toroidal like manifolds. In fact, thousands of SM like models were found. It is worth mentioning that even the simplest of these models requires to introduce a huge number of projections and to solve several tadpole equations.
In Ref. [@aaj] a hybrid Type IIB orientifold construction was proposed where the internal sector is built up from a Gepner sector times a torus. By choosing a torus invariant under some of the known $\IZ_N$ phase symmetries of Gepner models, an orbifold by such symmetries was then performed. Thus, schematically, in $D=4+2n$ the internal sector is given by ${( Gepner \,\, model)^{c_{int}=9-3n} \times \IT^{2n}}/\IZ_N$ (where $c_{int}$ is the internal central charge). The orbifold action is simultaneously embedded as a twist on Chan Paton factors on the open string sector resulting in a breaking of the starting $D$ dimensional Gepner orientifold gauge groups. In particular, such constructions lead to ${\cal N}=1$ $D=4$ chiral models. Illustrating examples were presented for odd affine Kac-moody levels. Hybrid Gepner-torus models have some interesting features. An important, practical, observation [@aaj] is that the number of Gepner models (see [@fkss:1989]) involved, 3 in $D=8$ or 16 in $D=6$, is remarkably lower than the 168 models in $D=4$ (without including moddings) and so it is the number of internal states. Also, many features can be studied analytically without the need of computers. From the phenomenological point of view, the possibility of having large extra dimensions, in the torus directions, could be an appealing feature allowing for some control over the string scale.
In this note we elaborate on this proposal of hybrid models. We concentrate on $D=6$ Gepner models, with diagonal invariant couplings, and extend the results of [@aaj] to include both, odd and even, affine Kac-Moody levels. $D=6$ models present particular features that make them interesting *per se* (see for instance [@fkllsw]). Moreover, due to the presence of potential gravitational and gauge anomalies these models are particularly useful to test the consistency of the construction.
We build up the explicit expressions for B-type boundary states and crosscaps and obtain the corresponding amplitudes for strings propagating among them. From such expressions we read the tadpole cancellation equations and the rules for reading the spectra. An explicit example (the $6620$ model) is developed in detail. Results for the 16 six dimensional models are summarized in [@aajd6gepners]. As far as we are aware of, besides the first examples of Gepner orientifolds in Ref.[@angelantonj], only some other $D=6$ spare examples (see for instance [@aaln; @aaj; @Braun:2005eg]) appear in the accessible literature.
Following [@aaj] we further compactify on a $\IT^{2}$ torus by simultaneously orbifolding the Gepner and the torus internal sectors and by embedding the orbifold action on the brane sector. Interestingly enough, whenever even orbifold action on the torus is considered, new branes, with worldvolume transverse to torus coordinates, must be generically included for consistency requirements[^1]. Detailed rules for obtaining the $D=4$ model spectra and tadpole equations are shown.
As an illustration we show how to obtain a 3 generations Left-Right symmetric model (which can be further broken into a MSSM model) from a $Z_4$ orbifold of the, $D=6$, $6620$ diagonal Gepner model times a torus..
The article is organized as follows. Section 2 contains a generic introduction to Type IIB orientifolds, crosscaps and boundary states. In Section 3 orientifold of $D=6$ Gepner models are discussed and crosscap and boundary states are constructed. The rules for computing the spectra and tadpole cancellation equations are derived. The explicit example $6620$ is discussed in detail. Section 4 provides a generic discussion of hybrid compactifications ${( Gepner \,\, model)^{c_{int}=6} \times \IT^{2}}/\IZ_N$. In Section 5 we construct a MSSM like example as a $\IZ_4 $ modding of $6620\times \IT^{2}$ and discuss some, generic, phenomenological features. Computation details are collected in the Appendices.
Type II orientifolds, crosscaps and boundary states
===================================================
In this section we briefly review the basic steps in the construction of orientifold models. Essentially, an orientifold model is obtained by dividing out the orientation reversal symmetry of Type II string theory (see for instance [@asopen; @aaln]). Schematically, Type IIB torus partition function is defined as $$\label{2bpf}
{\cal Z}_{T} (\tau, {\bar \tau})= \sum_{a,b} \chi_a(\tau )
{\cal N}^{ab} {\bar \chi}_b({\bar \tau })$$ where the characters $\chi_a(\tau )= \Tr_{{\cal H}a}
q^{L_0-\frac{c}{24} }$, with $q= e^{2i\pi \tau}$, span a representation of the modular group of the torus generated by [S]{}: $\tau \to -\frac1{\tau}$ and [T]{}: $\tau
\to\tau+1$ transformations. ${\cal H}_a$ is the Hilbert space of a conformal field theory with central charge $c=15$, generated from a conformal primary state $\phi _a$ (similarly for the right moving algebra). In particular $\chi_a( -\frac1{\tau} )= S_{aa'}\chi_{a'}(\tau )$ and modular invariance require $ S {\cal N} S^{-1}= {\cal N}$ (for left -right symmetric theories ${\cal N}^{ab}={\cal N}^{ba}$). Generically, the characters can be split into a spacetime piece, contributing with $c_{st}= {\bar c }_{st}= {\frac32} D$ and an internal sector with $c_{int} = {\bar c}_{int}= {\frac32}(10-D)$.
Let $\Omega $ be the reversing order (orientifolding) operator permuting right and left movers. Modding by order reversal symmetry is then implemented by introducing the projection operator $\frac12 (1+\Omega)$ into the torus partition function. The resulting vacuum amplitude reads $$\label{clomega}
{\cal Z}_{\Omega} (\tau, {\bar \tau})= {\cal Z}_{T} (\tau, {\bar \tau})
+ {\cal Z}_{K} (\tau- {\bar \tau}) .$$ The first term is just the symmetrization (or anti-symmetrization in case states anticommute) of left and right sector contributions indicating that two states differing in a left-right ordering must be counted only once. The second term is the Klein bottle contribution and takes into account states that are exactly the same in both sectors. In such case, the operator $e^{2i \pi \tau {L_0}}
e^{-2i \pi {\bar \tau}{{\bar L}_0}}$, when acting on the same states, becomes $e^{2i \pi 2it_K {L_0}}$ with $\tau -{\bar {\tau}}= 2it_K$ and thus $$\label{kbd}
{\cal Z}_{K} (2it_K) = \frac 12 \sum_{a} {\cal K}^{a} \chi_a(2it_K) ,$$ where $|{\cal K}^{a}|={\cal N}^{aa}$. The Klein bottle amplitude in the [*transverse channel*]{} is obtained by performing an [S]{} modular transformation such that $$\label{kbo}
\tilde {\cal Z}_{K} (i l)= \frac 12 \sum_{a} O^2_a \chi_a(il )$$ with $l=\frac1{2t_K}$ and (O\^a)\^2= 2\^D [K]{}\^[b]{}S\_[ba]{} This notation for the closed channel coefficients highlights the fact that the Klein bottle transverse channel represents a closed string propagating between two crosscaps (orientifold planes) states. Namely, a quantum state $|C\rangle $, describing the crosscap can be found such that the KB amplitude can be expressed as $$\tilde {\cal Z}_{K} (i l)=\frac12 \langle C |q^{\frac12 H_{cl}}|C\rangle .$$ with $H_{cl}=L_0-\tilde L_0-\frac c{12}$.
Indeed, crosscap states can be formally expanded in terms of Ishibashi states [@ishi; @ooy] such that $$|C\rangle= O_a |a\rangle\rangle _C$$ with $$_C\langle\langle b |q^{\frac12 H_{cl}}|a\rangle\rangle_C=\delta_{a,b}\chi^{a}({\tilde q})~,$$ and ${\tilde q}=e^{2i\pi l}$.
When integrated over the tube length, such amplitude leads, for massless states, to tadpole like divergences. In particular, for RR massless states, such tadpoles must be cancelled for the theory to be consistent. Notice that, for such fields, $O_a$ represents the charge of the orientifold plane (crosscap) under them and, therefore, inclusion of an open string sector with D-branes carrying $-O_a$ RR charge provides a way for having a consistent theory [@pol; @gp; @pchj] with net vanishing charge.
Therefore, we introduce stacks of boundary states $|\alpha\rangle$ (referred to as “brane-$\alpha$") $$|\alpha\rangle= \sum_{a} D_{\alpha}^a |a\rangle\rangle_B$$
such that the amplitude, describing propagation of strings between “intersecting” stacks ${ \alpha}$ and ${ \beta}$ can be written as $$\tilde {\cal Z}_{{ \beta}, { \alpha}}(i l)=
\langle \beta |q^{\frac12 H_{cl}}|\alpha\rangle=
\sum_{a} D_{\alpha}^a D_{\beta}^a\chi^{a}(l)=\sum_{b} C_{{ \beta}, { \alpha}}^b\chi^{b}(t/2)
\label{alfabeta}$$ where in the last step we have perform an $S$ modular transformation to direct channel.
Here $$C_{{ \beta}, { \alpha}}^b=\sum_{a} D_{\beta}^a D_{\alpha}^a S_{ab}$$ is the multiplicity of open string states contained in $ \chi_a$. Namely, it counts open string sector states of the form $$|\Phi_a ; {\beta},{\alpha}\rangle $$ where $\Phi_a$ is a world sheet conformal field and ${\alpha},{\beta}$ label the type of “branes" where the string endpoints must be attached to. $C_{ \beta, { \alpha}}^a$ are positive integers (actually $C_{ \beta, { \alpha}}^a= 0,1,2$ [@aaln]) generated when the trace over open states $|\Phi_a;{\beta},{\alpha}\rangle$ is computed.
The full cylinder partition function is obtained when summing over all possible stacks of $n_{ \alpha}$ branes, namely $${\tilde {\cal Z}}_C (i l)=\sum_{a} {D_a} ^2\chi^{a}(l)$$ with $D^a=\sum_{\alpha} n_{ \alpha} D_{\alpha}^a$.
In a similar manner, strings propagating between branes and orbifold planes give rise to strip amplitude [^2]
$${\tilde {\cal Z}}_M (il)=2 D_a O_a{\hat\chi}^{a}(l+\frac12)$$
By modular transforming to direct channel we obtain multiplicities of open string states between a brane and its orientifold image O\_a ( n\_[ ]{} D\_\^b)P\_[ba]{}= [M]{}\_[a]{} = M\_\^bn\_[ ]{} where we have used the fact that characters in the direct and transverse channels of the Möbius strip are related by the transformation [@bs1] [P]{}: $it_M+\frac{1}{2}
\to \frac{i}{4t_M}+\frac{1}{2}$ generated from the modular transformations [S]{} and [T]{} as ${\rm \sf P}={\rm {\sf TST}}^{2}{\sf S}$.
For indices $a$ representing massless RR fields $ D_a$ is the D-brane RR charge. Therefore zero net RR charge requires the $$O_a +n_{ \alpha} D_{\alpha}^a =0 .
\label{tadpolecg}$$ tadpole cancellation equations.
Orientifolds of $D=6$ Gepner models
===================================
In this section we briefly summarize the main ingredients involved in the construction Gepner model orientifolds in six spacetime dimensions. We refer the reader to the appendices and references for a survey of the details. In Gepner models [@gepnerlect], in $D$ space time dimensions, the internal sector is given by a tensor product of $r$ copies of $N=2$ superconformal minimal models with levels $k_j$, $j=1,...,r$ and central charge $$c_{j} = \frac{3{k_j}}{{k_j}+2} \quad , \quad {k_j}=1,2,... \label{mm}$$ such that internal central charge $c_{int}=\sum _{j=1}^r c_{j}^{int}=12-3(D-2)/2$.
Unitary representations of $N=2$ minimal models are encoded in primary fields labelled by three integers $(l,m,s)$ such that $l=0,1,...,k$; $l+m+s=0$ mod 2. These fields belong to the NS or R sector when $l+m$ is even or odd respectively[^3]. Spacetime supersymmetry and modular invariance are implemented by keeping in the spectrum only states for which the total $U(1)$ charge is an odd integer.
The primary field information of the complete theory can be conveniently encoded in the vectors $\lambda $ and $\mu $ defined in appendix B. Thus, the index $a$ in the previous section amounts here for $a=(\lambda,\mu )$ in Gepner’s case.
In six dimensions $c_{int}=6$ and 16 different possible Gepner models exist, which are associated to $K3$ surfaces [@fkss:1989; @Braun:2005eg]. Namely,
------------------------------- ------------------------------- ---------------------------- -----------------------------
$\overline{k}=(1,1,1,1,1,1)$, $\overline{k}=(0,1,1,1,1,4)$, $\overline{k}=(2,2,2,2)$, $\overline{k}=(1,2,2,4)$,
$\overline{k}=(1,1,4,4)$, $\overline{k}=(1,1,2,10)$, $\overline{k}=(0,4,4,4)$, $\overline{k}=(0,3,3,8)$,
$\overline{k}=(0,2,6,6)$, $\overline{k}=(0,2,4,10)$, $\overline{k}=(0,2,3,18)$, $\overline{k}=(0,1,10,10)$,
$\overline{k}=(0,1,8,13)$, $\overline{k}=(0,1,7,16)$, $\overline{k}=(0,1,6,22)$, $\overline{k}=(0,1,5,40)$
------------------------------- ------------------------------- ---------------------------- -----------------------------
: Gepner models associated to $K3$
. \[tab:K3\]
Notice that, in some cases, $k=0$ blocks have been added. Even if such terms are irrelevant in a closed string theory (for instance the central charge remains invariant), they have been shown to have a non trivial (K-theory) effect when open string sector is included. In fact, an even (odd) number of internal minimal blocks is required (see for instance [@bhhw0401; @fkllsw]) in $D=6$ ($D=4$) for consistency[^4].
Actually, for the sake of simplicity we will consider the case where the internal sector is a tensor product of $r = 6 $ conformal blocks. This will allow us to simultaneously consider cases with 3, 4, 5, and 6 conformal blocks such as $(6)^2(2)$, $(2)^4$, $1^4$ or $(1)^6$ by adding, if necessary, conformal blocks with level $k=0$. On the other hand, as it is shown in [@bhhw0401], the formulae for the total crosscap states contain the sign factor $(-1)^{\mu}$ where the parameter $\mu$ is given by
= \_[i=1]{}\^r ( 1- )= When $r=6$ there are no extra signs due to $(-1)^{\mu}$ and hence the expressions for the crosscap, that we will derive below, become somewhat simpler.
The Klein bottle amplitude is determined from that of the torus up to signs representing different ways of “dressing” the world-sheet parity $\Omega$. We will denote the dressed parity (we closely follow the notations in references [@aaj; @blumen0310]) as $\Omega_{\Delta, \omega_j}^{B}$, where $B$ means we are dealing with $B$-type orientifolds and $\Delta$, $\om_j$ label the quantum and phase symmetries respectively.
Recall that in four-dimensional Gepner models the B-parity is related to the $A$-parity $\Om_{\om, \Delta_i}^A$ via the Green/Plesser [@greenplesser] mirror construction
\_[, \_i]{}\^[B]{} && \_[, \_i]{}\^[A]{}\
= H \_[i=1]{}\^r && =\_[i=1]{}\^r \_i
where $H = lcm \{ k_i+2\}$.
Following [@blumen0310] we define an orientifold projection $\Om_{\Delta, \omega_j}$ by including the sign factors (-1)\^[\_j \_0]{} (-1)\^[b/H]{}\_j (-1)\^[\_j m\_j]{} \[signos\] where $\Delta, \om_j =0, 1$. These signs or parity dressings are chosen so that they preserve supersymmetry. By introducing these signs and by computing the trace in (\[kb\]) we are lead to Z\^B\_[K \_j,]{} &=& \_0\^ \_[, ]{}\^ \_[\_1,,\_r=0]{}\^1 \_[b=0]{}\^[-1]{}(-1)\^[\_0]{}(-1)\^[\_j \_0]{}\
&&(-1)\^[b/H]{}\_j (-1)\^[\_j m\_j]{}( \_[k<l]{}(-1)\^[\_k \_l]{} )\
&&\_j \_[b , \_j (k\_j+2)]{}\^[(2k\_j+4)]{} \_j \_[\_j l\_j, \_j ]{} \_\^
where $K={\rm lcm}(4,2k_j+4)$ (see appendix B for notation).
The factor $\prod_{k<l}(-1)^{\eta_k \eta_l}$ is introduced for convenience and arises naturally from the definition of the crosscap state below.
From the Klein bottle amplitude in the transverse channel we can read the expression for the crosscap state up to signs that can be fixed from the Möbius strip amplitude. The result is that the crosscap state is given by[^5]
$$\begin{aligned}
\label{crosscapbf}
| C\rangle^{NS}_B &=& \frac1{\kappa_c}
{\sum_{\lambda',\mu'}}^{ev} \sum_{\nu_0=0}^{\frac{K}2-1 }
\sum_{{\nu_j} , \tilde \nu_{1}=0}^1 \sum_{\epsilon_j=0}^1
\,\, \left(\prod_{k<l} (-1)^{\nu_k\nu_l} \right) (-1)^{\nu_0} \,(-1)^{\sum_j \nu_j}
\\ \nonumber
& & \,
(-1)^{\sum \omega_j \Lambda'_0 /2} e^{2\pi i \nu_0 \sum
\frac{\Delta_j}{k_j+2}}\, \delta_{\Lambda'_0,2+2\nu_0 +2\tilde \nu_{1}
+2\sum \nu_j+2\sum \om_j }^{(4)} \delta_{\Lambda'_{1},2\nu_0 +2\tilde \nu_{1}
}^{(4)} \\\nonumber
&& \prod_{j=1}^r \Biggl( \sigma_{(l_j',m_j',s_j')}\,
\frac{P_{l'_j,\epsilon_j k_j}}{\sqrt{S_{l'_j,0} }} \,
\delta_{m_j',2\nu_0+(1-\epsilon_j+\omega_j)(k_j+2)}^{(2k_j+4)}\\\nonumber
&& \delta_{s_j',2\nu_0 +2 \nu_j+2(1-\epsilon_j)}^{(4)}
(-1)^{\epsilon_j {\frac{(m'_j+s'_j)} 2} } \Biggr)
|\lambda',\mu'\rangle\rangle_c\end{aligned}$$
The normalization is chosen so that the overlap of the crosscap with itself yields the transverse Klein amplitude
\[overlap\] \_K\^B= \_0\^ C|e\^[-2l H\_[cl]{}]{} | C\_B .
In order to cancel tadpole-like divergences, boundary states must be introduced. We consider the B-type RS-boundary states [@ReS]
\[borde\] |\_B&=&|S\_0, S\_[1]{};(L\_j,M\_j,S\_j)\_[j=1]{}\^r \_B= [1[ \^B\_]{}]{} [\_[’,’]{}]{}\^[,b]{} (-1)\^[2]{} e\^[-i]{} e\^[-i]{}\
&&\_[j=1]{}\^r e\^[i]{} e\^[-i]{} |’,’
where the “b” in the summatory implies that
m\_j = b .
In fact, due to supersymmetry and field identifications these B-type boundary states only depend on ${\bf L}=(L_1,\dots,L_r)$ with $L_i \leq
k_i/2$, $M=H \sum \frac{M_i}{k_i+2}$ and $S=\sum S_i$. However, whenever a label $L_i$ reaches $k_i/2$, extra copies of the gauge field may appear propagating on the brane world-volume. In this case, itº is necessary to resolve the branes into elementary branes such that only a single gauge field is propagating on the world-volume. The details depend on the values of $|\mathbf{S}|$ counting the number of $i$ such that $L_i=k_i/2$. It can be shown [@bhhw0401] that when $|\mathbf{S}|$ is an odd integer the elementary branes are given by
|\_[ele]{} \_B = |\_B
Instead, if $|\mathbf{S}|$ is even there is an extra $\mathbf{Z}2$-valued label $\psi$ taking values $\pm$ so that the elementary boundary states are now labelled by $\bf{L},M,S, \psi$. The original boundary states can be written in terms of the elementary ones as
|\_B = { |+ \_B+|-\_B } where $\a$ stands for all labels different from $\psi$. The two boundary states $|\a\pm
\rangle_B$ contain new states from the twisted $(c,c)$ RR sector
$$\begin{aligned}
|\alpha\pm\rangle _B &=&|S_0,S_{-1};(L_j,M_j,S_j)_{j=1}^r,\pm \rangle_B \\ \nonumber
& = & {\frac{1}{ \kappa^B_\alpha}}
{\sum_{\lambda',\mu'}}^{\beta,b} \,e^{-i\pi{\mu' \mu_B}}
\{\otimes_{j=1}^r \frac{S_{l'_j,L_j}}{\sqrt{S_{l'_j,0}}} \, \delta_{m'_j,b}^{(k_j+2)}\,
\\ &&\pm\,\otimes_{j\in S}\delta_{l',\frac{k_j}2} \,e^{-i\frac{\pi}2 M_j} \,
\delta_{m'_j,b+\frac12(k_j+2)}^{(k_j+2)}\otimes_{j\notin S}{\frac{S_{l'_j,L_j}}{\sqrt{S_{l'_j,0}}} \, \delta_{m'_j,b}^{(k_j+2)}\, \}
|\lambda',\mu'\rangle\rangle }
\label{btypebs}\end{aligned}$$
where $\mathbf{S}=\{i : l'_i=\frac{k_i}2\}$ and $\mu_B= (S_0, S_{-1};M_1,...,M_r;S_1,...,S_r)$.
Actually, the $|\a\pm
\rangle_B$ branes are necessary in order to have D-branes charged under all RR fields in the theory. Geometrically the situation is as follows [@brunschom]. The twisted RR fields are related to singular curves of the associated Calabi-Yau spaces. Then the elementary D-branes $|\a\pm\rangle_B$ can wrap on the new homological cycles arising from the resolution of the singularities.
Let us now look at the supersymmetric spectrum in the open sector. The boundary states (\[borde\]) preserve the same supersymmetry than the crosscap (\[crosscapbf\]) if the following condition is satisfied
\[susy\_brana\] M=+
The massless fields in the 6D spacetime theory are the vector field $(2,0)(0,0,0)^6$ and the hypermultiplets $(0,0)\prod_j(l_j, l_j, 0)$ with $\sum_j \frac{l_j}{k_j +2} = 1$. They are contained in the cylinder and Möbius amplitude which we present next. The bosonic and massless part of the cylinder amplitude between two D-branes $|\mathbf{L},M \rangle$ and $|\mathbf{L'},M'\rangle$ is generally given by
$$\label{cilindro}
\frac12\frac{1}{N_{\mathbf{S}}}\frac{1}{N_{\mathbf{S}'}}\sum_{\lambda,\mu}^{sr}\sum_{\epsilon_1,...,\epsilon_r=0}^1\left(\prod_{i=1}^{r}N_{L_j,{
{L'_j}}}^{|\epsilon_jk_j-l_j|} \right) \delta_{\sum_i \epsilon_i =
1+\frac{s_0}{2}}^{(2)}\chi_{\mu}^{\lambda}$$
where $N_{L_1,L_2}^{l}$ are the $SU(2)$ fusion coefficients (\[su2fusion\]) and $$N_{\mathbf{S}} =\left \{ \begin{array}{ll}
2^{|\mathbf{S}|/2} & \textrm{if } |\mathbf{S}| \textrm{ even} \\
2^{[|\mathbf{S}|-1]/2} & \textrm{if } |\mathbf{S}| \textrm{ odd} \\
\end{array}\right.
\label{ns}$$ eliminates any extra counting when some of the D-branes are short. We have already taken into account the condition (\[susy\_brana\]) and therefore the labels $M, M'$ do not appear explicitely in this expression. Besides, we have defined an extra label $s_0= \Lambda_0 + \Lambda_{1} \mod{4}$ (see Appendix A) taking the values $0, 2$ whenever the fields are in the scalar and vector representations, respectively (Note that in six dimensional spacetime $s_0 = 0, 2$ also for the spinor representations, so this definition strictly makes sense when we restrict ourselves to bosonic representations). When the amplitude between two short-orbit branes $|\mathbf{L}, \psi\rangle$ and $|\mathbf{L'}, \psi'\rangle$ such that $\mathbf{S}=\mathbf{S}'$ and $|\mathbf{S}|\in 2\IZ^{+}$ is considered, an additional projection must be taken into account, due to the $\psi$ labels, leading to
$$\label{cilindroshort}
\frac{1}{2^{|\mathbf{S}|}}\sum_{\lambda,\mu}^{sr}\sum_{\epsilon_1,...,\epsilon_r=0}^1\left(\prod_{i=1}^{r}N_{L_j,{\tilde
{L_j}}}^{|\epsilon_jk_j-l_j|} \right) \delta^{(2)}_{\sum_{i \in
\mathbf{S}}^{} \epsilon_i = \frac12(1-\psi \psi')} \delta_{\sum_i
\epsilon_i = 1+\frac{s_0}{2}}^{(2)}\chi_{\mu}^{\lambda}$$
On the other hand, the massless states in the bosonic Möbius strip amplitude are given by \[moebius\]-\_[k<l]{}(-1)\^[\_k \_l]{} \_[\_j + 1 + ,\_j]{}\^[(2)]{} (-1)\^[\_j ]{} e\^[i \_j \_j(m\_j-2L\_j+\_j(k\_j+2))]{}(-1)\^[\_j]{} N\_[L\_j L\_j]{}\^[|\_j k\_j-l\_j|]{} \_\^
where $\rho_j = \frac{s_0 }{2} +1 + \epsilon_j + \sum \om_j$.
In particular, we see the vector ($s_0=2$) has the sign
-(-1)\^[\_j]{}(-1)\^[\_j L\_j]{} \_[\_1, …, \_r=0]{}\^[1]{} \_[k<l]{} (-1)\^[\_k\_l]{} (-1)\^[\_j \_j ]{} \_[\_jL\_j,\_j]{} \_[\_j , \_j]{} \[signo\_vector\]
A plus (minus) sign indicates a symplectic (orthogonal) gauge group while a zero leads to a unitary gauge group. In a similar manner, the gauge group representations in which matter states transform, can be identified (an example is given in next section).
The action of $\Om_{\D,\om_j}$ on these elementary boundary states can be obtained by comparing (\[moebiloopdr\]) to the cylinder amplitude between a D-brane $|\mathbf{L},M \rangle$ and its $\Om-$image $|\mathbf{L'},M'\rangle$. They coincide if the action is given by (see [@blumen0310] for instance)
\_[,\_j]{}: | ,M,S |,2-M,-S
Furthermore, consistency of (\[signo\_vector\]) with the cylinder amplitude (\[cilindroshort\]) between a given brane with a label $\psi$ and its image under $\Om$ with a label $\psi'=\Om(\psi)$ requires
\[psi\]\_[,\_j]{}: (-1)\^ (-1)\^[||/2]{} \_[i]{} (-1)\^[\_j ]{} .
To see it we use that
\_[\_1, …, \_r=0]{}\^[1]{} \_[k<l]{} (-1)\^[\_k\_l]{} \_[\_jL\_j,\_j]{} \_[\_j , 0]{} = (1+ (-1)\^ (-1)\^[||/2]{} )
Even though we are dealing with the case $r=6$ we have introduced the factor $(-1)^{\mu}$ to make contact with the case $r=4$. Its origin is simple. When we go from $r=6$ to $r=4$ subtracting two $k=0$ factors leads to $(-1)^{|\mathbf{S}|/2} \to (-1)^{\mu}(-1)^{|\mathbf{S}|/2}$.
Thus, for instance, in the case $\D = \om_j = 0$ we find that, according to $|{\bf S}|$ values and specifying for $r=6$ , the groups shown in Table \[groups\] arise.
$|\mathbf{S}|$ $|B\rangle$ Group r=6
---------------- -------------------------------------- -----------------------
0 $*$ $SO(N)$
1 $|\hat B\rangle$ SO(N)
2 $|\hat B+\rangle+|\hat B-\rangle$ U(N)
3 $2|\hat B\rangle$ SP(2N)
4 $2(|\hat B+\rangle+|\hat B-\rangle)$ $SP(2N)\times SP(2N)$
5 $4|\hat B\rangle$ SP(4N)
6 $4(|\hat B+\rangle+|\hat B-\rangle)$ U(4N)
: Groups that arise from introducing a given number of reducible D-branes are shown. Gauge symmetry is enhanced in some cases since these branes can be decomposed into a set of elementary D-branes. However, nothing prevents us from considering simply one copy of an elementary D-brane (and its image), thus yielding a gauge group with unity range, ie. $U(1), Sp(2), SO(2)$.
\[groups\]
The tadpole cancellation conditions can be easily read from the expressions for the crosscap (\[crosscapbf\]) and boundary states (\[borde\]). They take the general form Tad$_{D}(\lambda,\mu)-8$Tad$_{O}(\lambda,\mu)=0$. For the massless fields $(2,0)(0,0,0)^6$ and $(0,0)\prod_j(l_j, l_j, 0)$ the NS-NS tadpoles of the orientifold plane read
$$\begin{aligned}
\label{tadorien}
{\rm Tad}_{O}(\lambda,\mu)_B&=&
\sum_{\nu_0=0}^{{\frac K2}-1}
\sum_{\epsilon_1,\ldots,\epsilon_r=0}^1
\, \, \left(\prod_{k<l} (-1)^{\epsilon_k\epsilon_l} \right)\,(-1)^{\nu_0 \sum \epsilon_j}\,\\ \nonumber
& & \,
\delta_{s_0/2,1+\sum \epsilon_j + \sum \om_j}^{(2)}(-1)^{\sum
\om_j (s_0/2)}(-1)^{\D_j (1-\epsilon_j)}
\\\nonumber
&& \prod_{j=1}^r \Biggl( \, {\rm sin}\left[\frac{1}{2}(l_j, \epsilon_j k_j)\right]\,
\delta^{(2)}_{l_j+(1-\epsilon_j) k_j,0} \,
\delta_{m_j',2\nu_0+(1-\epsilon_j +\om_j)(k_j+2)}^{(2k_j+4)}\\\nonumber
&& (-1)^{\epsilon_j {\frac{m_j} 2} } \Biggr).
\label{tadod=6}\end{aligned}$$
Also, collecting all terms from the boundary states and their $\Omega_{{\Delta}_j,\omega,\omega_\alpha}$ images, we obtain, for their massless tadpoles
\_[D]{}(,)& =& \_[a=1]{}\^N [cos]{} \_j [sin]{}(l\_j,L\_j\^a). \[taddd=6\] These expressions are valid up to common phases. We have also renormalized the tadpole equations by introducing the factor $N_{\mathbf{S}}$ so that the Chan-Paton factors $N_a$ truly represent the multiplicity of elementary D-branes.
$(6)^2(2)(0^3)$ model
---------------------
We exemplify the construction presented in the preceeding section for the specific Gepner model $(6)^2(2)(0^3)$. We will later consider this example to discuss model building in four dimensions. Results for the other six dimensional models are presented in [@aajd6gepners]. The allowed branes and corresponding gauge groups and matter representations living on them (see \[\[groups\]\]) are given in Table \[662table\].
\# $(L_1,L_2,L_3,L_4)$ Group \# ( [-0.4pt ]{} [-6.9pt ]{} )
----------------- --------------------- -------------------------------- ------ ------------- ------------- ---
${\bf L_1}$ 0 0 0 $0^3$ $Sp(N_1)$ 0 0
${\bf L_2}$ 1 1 0 $0^3$ $Sp(N_2)$ 0 3
${\bf L_3}$ 3 1 0 $0^3$ $Sp(N_3)\times Sp(N_3)$ 0 2
${\bf L_4}$ 3 3 0 $0^3$ $Sp(N_4)$ 0 3
${\bf L_5}$ 2 0 0 $0^3$ $Sp(N_5)$ 0 2
${\bf L_6}$ 2 2 0 $0^3$ $Sp(N_6) $ 1 7
${\bf L_7}$ 0 0 1 $0^3$ $Sp(N_7)\times Sp(N_7)$ 0 0
${\bf L_8}$ 1 1 1 $0^3$ $Sp(N_8)\times Sp(N_8)$ 0 2
${\bf L_9}$ 3 1 1 $0^3$ $Sp(N_9)$ 0 3
${\bf L_{10}}$ 3 3 1 $0^3$ $U(N_{10})$ 0 6
${\bf L_{11}}$ 2 0 1 $0^3$ $Sp(N_{11})\times Sp(N_{11})$ 0 1
${\bf L_{12} }$ 2 2 1 $0^3$ $Sp(N_{12}) \times Sp(N_{12})$ 0 4
: The gauge groups and matter content living on their world volume of each possible boundary state $\bf L_{I}$ is indicated.
\[662table\]
This spectrum is obtained from Eqs. (\[cilindroshort\]) and (\[moebius\]). For instance, we see that brane ${\bf L_{10}}= \,3\, 3 \,1 \,0^3$ is short, with $|\mathbf{S}|=6$. Thus, for the vector ($s_0=2$), a non vanishing contribution in (\[cilindroshort\]) implies $ \delta^{(2)}_{\sum_i \epsilon_i \frac12(1-\psi \psi')} \delta_{\sum_i
\epsilon_i = 0}^{(2)} \ne 0$, namely $\psi =\psi'$. Moreover, for such choice of $\epsilon_i' s$ we see that (\[moebius\]) vanishes thus leading to the unitary group shown in the Table \[662table\]. In a similar way, for the scalars ($s_0=0$) the states propagating between a boundary state and its orientifold image are selected, $\psi =-\psi'$. Möbius amplitude (\[moebius\]) is non vanishing in this case and produces a minus sign thus leading to antisymmetric representations.
The tadpole equations (\[tadod=6\],\[taddd=6\]) for this set of branes reads $$\begin{aligned}
N_2+2N_3 + N_4 + N_5 +2N_6 +N_7 +2N_8 +N_9 +2N_{10}+N_{11}+3 N_{12} = 16&&\\
N_1+2N_2 + 2N_3 + 2N_4 + N_5 +3N_6 +2N_8 +2N_9 +2N_{10}+2N_{11}+4 N_{12} = 24&&
\label{662d6tadpoles}\end{aligned}$$
States propagating between branes can be easily computed from (\[cilindro\]) and (\[cilindroshort\]). Two tensor multiplets are found in the internal sector (see for instance [@aaln]). It can be checked that all gauge and gravitational anomalies cancel.
At this point it may be instructive and useful for our subsequent calculations to illustrate this in a detailed example containing only $({\bf L_1},{\bf L_6},{\bf L_{10}})$ states.
The tadpole equations for the reduced set of D-branes lead to N\_[10]{} &=&8-N\_6=N1 \[tadL1L6L10\]
The gauge group is $Sp( N_1)\times Sp( N_6)\times U( N_{10})$
with matter hypermultiplets in
&& 3\[(1,1,[-6.9pt ]{})+(1,1,[|[-6.9pt ]{}]{})\]+7(1,[-6.9pt ]{},1)+(1,[-0.4pt ]{},1)\
&+& ( N\_1, N\_6,1)+(N\_1,1,N\_[10]{}) +(N\_1,1,[|N\_[10]{}]{})+\
&& 3\[(1, N\_6, N\_[10]{})+(1, N\_6, [|N\_[10]{}]{})\]
$${\renewcommand{\arraystretch}{1.3}
\begin{tabular}{|c|c|c|c|}
\hline
S-T & Internal & mult. & irrep.\\
\hline v & $(0,0)^6$ & 1 & $Sp(N_1) \otimes
Sp(N_6) \otimes U( N_{10})$\\
s & $ \underline{(6,6)(2,2)}(0,0)(0,0)^3$ & 2 &
$2(1,{\raisebox{-3.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt}}}\hskip-6.9pt \raisebox{3pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt}}}},1) + (1,1,{\raisebox{-3.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt}}}\hskip-6.9pt \raisebox{3pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt}}}})+(1,1,{\bar {\raisebox{-3.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt}}}\hskip-6.9pt \raisebox{3pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt}}}}}) $\\
s & $ (4,4)^2(0,0)(0,0)^3 $ & 1 &
$( N_1,N_6,1)+ 3(1,{\raisebox{-3.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt}}}\hskip-6.9pt \raisebox{3pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt}}}},1) +(1,{\raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt}}}\hskip-0.4pt \raisebox{-.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt}}}},1)+ (1,1,{\raisebox{-3.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt}}}\hskip-6.9pt \raisebox{3pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt}}}})+ (1,1,{\bar{\raisebox{-3.5pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt}}}\hskip-6.9pt \raisebox{3pt}{{\hbox{\rule{0.4pt}{6.5pt}\hskip-0.4pt\rule{6.5pt}{0.4pt}\hskip-6.5pt\rule[6.5pt]{6.5pt}{0.4pt}}\rule[6.5pt]{0.4pt}{0.4pt}\hskip-0.4pt\rule{0.4pt}{6.5pt}}}}})$\\
s & $(3,3)^2(1,1)(0,0)^3$ & 1 &
$(1,N_6,N_{10})+(1,N_6,{\bar N}_{10}) + (N_1,1,N_{10}) + (N_1,1,{\bar N_{10}})$\\
s & $ {(5,5)(1,1)}(1,1)(0,0)^3 $& 1 &
$2(1,N_6,N_{10})+2(1,N_6,{\bar N}_{10})$\\[.5ex]
\hline
\end{tabular}
}$$ \[spectL1L6L10\]
.
It is easy to check that this spectrum (plus a closed sector containing two tensor multiplets and nineteen hypermultiplets) leads to vanishing of gauge and gravitational anomalies if tadpole equations (\[tadL1L6L10\]) are satisfied.
Orbifolding ${( Gepner \,\, model)}^{c=6} \times \IT^{2}$
=========================================================
Orbifolds of Gepner models are also easily implemented in the language of boundary and crosscap states. The internal sector described by ${( Gepner \,\, model)}^{c=6} \times \IT^{2}$ has a discrete symmetry acting on fields in the following way
g: Z e\^[2i v]{} Z g: \^[l\_i]{}\_[m\_i s\_i]{} e\^[2i ]{}\^[l\_i]{}\_[m\_i s\_i]{} i=1,…,r \[phasesymm\]
where $Z=X^4 + i X^5$ denotes the complex coordinate on $\IT^{2}$ and $(v; \g_i)$ are labels for the generator $\hat g \in G$. For a torus with symmetry $\IZ_N$ we have $N v \in \IZ$. The labels $(v; \g_i)$ are conveniently encoded in terms of a simple current vector $j$
j=(0, v; 2\_1,…,2\_r;0,…,0).
which satisfies $2\b_0 \bullet j \in \mathbf{Z}$ or in components
- + = 0 . \[susycurrent\]
As it is well known, twisted sectors must be included in order to ensure the modular invariance of the torus partition function. As a consequence, new tadpoles are expected to appear, in the transverse channel, due to RR fields propagating in the twisted sectors.
The boundary states required to cancel the tadpoles include the RR fields in the twisted sector of the theory.
When the internal symmetry group is $\IZ_N$, which is the case we are mainly interested in, we can write an expression for the boundary state in the simple case $v=0$ that would correspond to a four-dimensional compactification with $N=2$ supersymmetries. The case $v \ne 0$ will be considered later on in this section.
For a symmetry group $\IZ_N$ the twisted boundary states read \[bound-orb\] |\_B&=&|S\_0, S\_[-1]{};(L\_j,M\_j,S\_j)\_[j=1]{}\^r \_B= [1[C\_B]{}]{} \_[x=0]{}\^[N]{}[\_[’,’]{}]{}\^[,b]{} (-1)\^[2]{} e\^[-i]{} e\^[-i]{}\
&& \_[j=1]{}\^r e\^[i]{} e\^[-i]{} |’,’\[twbs\]
where now
\[carganula2\] m\_j = b + 2 \_j x
with $x=0,1,\dots,N-1$ and $N$ is the order of the symmetry group generated by the simple current $j$. The branes are labelled as $\mathbf{L}, \mathbf{M}, S$ with $M=(M_1,...,M_r)$ modulo group identifications and space-time labels $S_0, S_{-1}$ are defined in Appendix A . Short-orbit D-branes also include a $\psi$-label.
Replacing (\[carganula2\]) into (\[bound-orb\]) we see that the boundary state depends on $M_1,\dots,M_r$ only through the phase
\[fase\]e\^ e\^[2i ]{}
and therefore an independent set of labels for $|\a \rangle$ is given by M= H \_[i=1]{}\^r V = \_[i=1]{}\^r .
In this way, $M$ represents nothing but the action of the symmetry group on the Chan-Paton factor.
In other words, if we begin with a configuration of N coincident D-branes defined by the set $\mathcal{M}=\{(M^{\a}_1,\dots,M^{\a}_r) ~~\a = 1,\dots,N \,|\, M = H \sum \frac{M^{\a}_i}{k_i+2}\} $, then the modding by $\G=(\g_1,...,\g_r)$ divides $\mathcal{M}$ into classes \_I ={ := V\_I }, each with $N_I$ elements such that $\sum N_I = N$. From (\[fase\]) the action on the Chan-Paton class $\mathcal{C}_I$ is given by the matrix $(\g)_{II} = e ^{2\pi i V_I}$ and the character of this representation
Tr \^x = N\_I e\^[2i x V\_I]{} .
Successive modding by simple currents $j_2, j_3,...$ will introduce extra labels $W, X, \dots$, which at the end, if conveniently chosen, will be in one-to-one correspondence with the labels $(M_1,\dots, M_r)$ of the A-type boundary states. This is expected because of the Green/Plesser construction of the mirror theory relating A- and B-type models.
The spectrum of massless particles in the orbifolded theory is read from the annulus amplitude. Given two boundary states with labels $\a =(\mathbf{L}, M, S)$ in class $I$ and $\tilde \a =( \mathbf{\tilde L}, \tilde M,\tilde S) $ in class $J$, the amplitude between them in tree channel reads
\[cylin\] Z\_[I, J]{}\^B(q)=1C \^[NS]{}\_[’’]{} \^[(H)]{}\_[+]{} e\^[2 x i[([V\_I-V\_J]{}]{}+)]{} \_[j=1]{}\^rN\^[l\_j’]{}\_[L\_j,L\_j]{} \^[’]{}\_[’]{}(q)
In this case, with the symmetry acting only on the Gepner sector, it is possible to sum over $x$ leading to the condition
\[proj\]V\_I-V\_J+ =0.
which implies that in general some states will be projected out of the original spectrum.
Under the orbifold projection the original full cylinder amplitude changes as follows N\_ N\_Z\_ \_[x=0]{}\^[N-1]{}\_ Tr \_\^x Tr \_\^x e\^[2 x i ]{}Z\_
Interestingly enough, it is possible to rewrite the projection by simple currents in such a way that its relation to the usual orbifolds of toroidal manifolds is much more evident. To see this we recall that open string states formally read $$|\Phi_k ; i,j\rangle \lambda^k_{ji}$$ where $\lambda^k $ encodes the gauge group representation into which the state $\Phi_k$ transforms. For instance, if the state $\Phi_0$ corresponds to gauge bosons, $ \lambda^0$ represents gauge group $G$ generators [^6].
Let us assume that such Chan-Paton factors have been determined already and that we further act on string states with a generator $\theta$ of a $Z_N$ symmetry group. Such action which manifests as a phase $e^{2\pi i \frac{\g_i m_i}{k_i+2}}$ on world sheet field $\Phi_k$ should, in principle, be accompanied by corresponding representation of group action such that $$\begin{aligned}
\hat \theta|\Phi_k ; i,j\rangle \lambda_{ji}& = &\g_{ii'}|\hat \theta\Phi_k ; i',j'\rangle \g_{j'j}\lambda_{ji} \\
& = &e^{2\pi i \frac{\g_i
m_i}{k_i+2}}(\g^{-1}\lambda\g)_{j'i'}|\Phi_k ; i',j'\rangle\end{aligned}$$ Therefore, invariance under such action requires $$e^{2\pi i \frac{\g_i m_i}{k_i+2}} \g^{-1}\lambda ^k \g = \lambda
^k \label{thetaprojection}$$
By following the same steps as in Ref. [@afiv], we can represent $Z_N$ Chan-Paton twist in terms of Cartan generators as $\g=e^{2\pi i VH}$ where $V$ is a “shift" eigenvalues vector of the generic form $$V=\frac{1}{N}(0^{N_0},1^{N_1},\dots,(N-1)^{N_{N-1}}) \label{twist}$$ (ensuring $\g^N=1$) and Cartan generators are represented by $2
\times 2$ $\sigma_3$ submatrices.
On this basis, projection equation (\[thetaprojection\]) reduces to the simple condition $$\rho _k V = \frac{\g_i m_i}{k_i+2} \label{proyeccion}$$ where $\rho _k$ is the weight vector associated to the corresponding $\lambda ^k$ representation. This should be compared to (\[proj\]).
In this latter framework the extension to the case $v \ne 0$ is easily written down. (\[proj\]) is replaced with
\[proj2\]V\_I-V\_J- + =0.
where now $V_I $ represents the action on the Chan-Paton factor due to the symmetry that acts simultaneously on the torus $\IT^2$ and the Gepner piece.
A last comment about the action on Chan-Paton factors $\g_{I,J}$ is due. In the orientifold theory we must introduce a boundary state and its image under $\Omega$. For long-orbit D-branes $|B_{\mathbf{L},M}\rangle$ this yields an effective action
\_[II]{} + \_[II]{}\^\* = e\^[2i ]{}+e\^[-2i ]{}= 2
which is real. This is simply the orientifold condition $\g_{\Omega} \g \g_{\Omega}^{-1} = \g^*$ which identifies Chan-Paton factors.
For short-orbit D-Branes such that $\Omega \psi \to -\psi$, however, we have
\_[II]{} |, + +\_[II]{}\^\* |, - = e\^[2i ]{} |, + +e\^[-2i ]{} |, -
Tadpole conditions can be generalized for orbifolded hybrid models $T^{2} \times Gepner$ in the following way [@aaj] $$D_{\mu}^{\lambda}\left( {\Tr
\g_{N,2x}} +{\sqrt{\mathnormal{f}}} \Tr{\g_{D,2x,I}}\right) + O_{\mu}^{\lambda} \cos \pi x v =0 \label{tadhyb1}$$
$$D_{\mu}^{\lambda}\left( {\Tr
\g_{N,2x+1}} +{\sqrt{\mathnormal{f}}} \Tr{\g_{D,2x+1,I}}\right)= 0
\label{tadhyb2}$$
for all states $(\lambda,\mu)$ such that $\chi^{\lambda}_{\mu+2 x \G}$ is massless.
Here $O_{\mu}^{\lambda} $ is the orientifold charge we have in six dimensions for the state $(\lambda,\mu)$ while the factor ${f}=4 \sin^2{\pi x v}$ is a non-trivial contribution from the fixed points in the complex torus $\mathrm{T}^2$ in the NN sector. The labels N and D are used to distinguish between D-branes with Neumann and Dirichlet boundary conditions in the torus $T^2$.
Before closing this section let us recap the general steps to be followed in order to build up four dimensional models. Our construction proceeds through two consecutive stages. A first step is to build a six dimensional model out of the possible Gepner models showed in Table \[tab:K3\]. It is also necessary to choose an orientifold projection as indicated in \[signos\]. This gives rise to tadpoles which must be cancelled according to equations (\[tadorien\]) and (\[taddd=6\]). For each configuration of tadpole-canceling D-branes, the spectrum, the matter content and the gauge group, can be read from (\[cilindro\]), (\[cilindroshort\]), ( \[moebius\]) and (\[psi\]). This completes the building of six-dimensional gauge theories. Further compactification to four dimensions is achieved by choosing an orbifold action on ${( Gepner \,\, model)}^{c=6} \times \IT^{2}$ as shown in (\[phasesymm\]). Interestingly enough, spectra in the orbifold can be easily read using the simple expression (\[proyeccion\]). Tadpole cancellation conditions (\[tadhyb1\]) and (\[tadhyb2\]) will in general require the presence of additional D-branes with Dirichlet boundary conditions on $T^2$ sitting at fixed points. The great advantage of this method is that six dimensional Gepner models are clearly easier to solve. If we are able to identify some of the distinctive features of the Standard Model in this first stage, say the number of generations or the gauge group, then the steps down to four dimensions are quite direct and easy to implement.
A MSSM example
==============
As an illustration of the general techniques discussed above we concentrate here on a $\IZ_4$ modding of the $[(6)^2(2)(0)]^{c=6}\times T^2$ model[^7]. Let us notice that inspection of allowed internal states indicates that only 3 massless chiral ($l_i=m_i, s_i=0$) states, namely those such $(m_1,m_2,m_3,m_4)=(3,3,1,0),\, (5,1,1,0),\, (1,5,1,0)$, do propagate between brane ${\bf L_{10}}$ (with a $U(N_{10})$ gauge group living on its worldvolume) and ${\bf L_6}$ (with an $Sp(N_6)$ gauge group). Therefore, an internal modding of the form $\Gamma=(0,0,1,0)$ acting on the Gepner model will allow such states to remain in the spectrum and, by appropriately embedding it as a twist $\gamma^{10}, \gamma^{6},\dots$ on the D-brane sector, the original $U(N_{10})\times Sp(N_6)$ gauge group could be broken into a Standard-like model with 3 generations. Moreover, in order to have ${\cal N}=1$ supersymmetry in four dimensions, we must accompany this modding with a $\IZ_2$ modding on $ T^2$, namely $v_3=1/2$, so as to satisfy Eq.(\[susycurrent\]). Thus, our starting point is $$\Gamma=(0,0,1,0)(\frac12)\otimes \g^{a}
\label{gmodding}$$
Note that the actual internal modding (see (\[phasesymm\])) is $\g_i/(k_i+2)$ so it represents a $Z_4$ action.
As we stressed in the previous section, performing a $\IZ_2$ modding on the torus will require the introduction of a new set of branes having Dirichlet boundary conditions on the open string ends living on $T^2$. We quote them with an index $D$ while introducing an index $N$ to label the original branes with Neumann conditions on the third complex coordinate $Z$. We will refer to them as $D_Z$ and $N_Z$ branes respectively. The generic tadpole equations (see Eq.(\[tadhyb1\],\[tadhyb2\]) for this model thus read
\[t1\]\_a D\^a(ł,) Tr \_[0,N,a]{}+ O(ł,)&=0&\
\_a D\^a(ł,) Tr \_[0,D,a]{}+ O(ł,)&=0&\
\_a D\^a(ł,) Tr \_[2,N,a]{}&=0&\
\_a D\^a(ł,) Tr \_[2,D,a]{}&=0&\
\_a D\^a(ł,) ( Tr \_[1,N,a]{}+2Tr \_[1,D,a]{} )&=0& \[gtadpoles\]
The indices indicate the order of the twist, the $D$ or $N$ sector on the torus, and the label $a$ for a brane ${\bf L_a}$ (see Table \[662table\]). It is easy to see that any extra massless state is introduced in the closed sector by twisted internal states. Therefore, $D^a(\l,\mu)$ coefficients are just the coefficients appearing in the untwisted tadpole equations (see (\[662d6tadpoles\])) corresponding to the vector $(2;0000;0)$ and the scalar state $(0;2,2,2,0;0)$ $$\begin{aligned}
N_2 + 2N_3 + 2N_4 + N_5 +2N_6 +N_7 +2N_8 +2N_9 +2N_{10}+N_{11}+3 N_{12} & = & 16\\
N_1+2N_2 + 2N_3 + 4N_4 + N_5 +3N_6 +2N_8 +4N_9 +2N_{10}+2N_{11}+4 N_{12} & = & 24\end{aligned}$$ and similarly for the $D_3$-branes sector.
As mentioned, ${\bf L_{10}}$ and ${\bf L_6}$ constitute the basic branes which, after splitting under modding action, will give rise to our model. It is interesting to remark that, both boundary states can be placed on the same NN sector, or either ${\bf L_{10}}$ in NN sector and ${\bf L_6}$ in the DD sector (or viceversa) or both in the DD sector. The basic features, discussed below, will be independent of the sector choice. However phenomenological details will be different, mainly due to the extra branes that must be added to satisfy RR tadpole cancellation.
We choose two (minimal) $N_{10}=N_{6}=4$ stacks of ${\bf L_6}$ and ${\bf L_{10}}$ branes to start with a $U(4)\times Sp(4)$ gauge group. The modding $\Gamma$ in ( \[gmodding\]) is embedded as twists $\gamma_6$ and $\gamma_{10}$, on each respective stack, as
$$\begin{aligned}
\gamma_6& \rightarrow& V_6 =\uq(0,2) \\
\gamma_{10}&\rightarrow &V_{10}=\uq(1,1,1,3)\end{aligned}$$
For the vector $\Gamma.\mu =0$ and therefore from (\[proyeccion\]) we find $$\begin{aligned}
U(4) & \rightarrow& SU(3) \times U(1)^2 \\\nonumber
Sp(4)&\rightarrow &SU(2)\times SU(2)\nonumber
\label{groupbreaking}\end{aligned}$$ where $Sp(2)\equiv SU(2)$. Thus, a LR symmetric-like model group is obtained.
Moreover, the correct LR spectrum with 3 massless generations is found. Namely, massless chiral states propagating in between ${\bf L_{10}}- {\bf L_{6}}$ $$\begin{aligned}
&&(0; 3 3 1 0;0 )\\\nonumber
&&(0; 1 5 1 0 0)\\\nonumber
&&( 0;5 1 1 0;0 )\nonumber
\label{3ciralstates}\end{aligned}$$ satisfy $\Gamma \mu=\frac14$ and therefore we find the spectrum representations under $SU(3)\times SU(2)_L\times SU(2)_R \times Q_{B-L}$ to be $$3[(3,2,1)_{\frac13}+(\bar3,1,2)_{-\frac13}+(1,1,2)_{1}+(1,2,1)_{-1}]$$ where the subindex indicates the charge eigenvalue of $$Q_{B-L}=\frac13 Q_3+Q$$ $Q_3$ being the generator of the $U(1)$ in $U(3)$ and $Q$ the other $U(1)$ generator in (\[groupbreaking\]).
Actually, it is possible to establish a correspondence with an intersecting brane model picture in toroidal manifolds (see for instance [@cim:2003] or [@cim:2002]). Namely, under the action of $\gamma_{10}$ and $\gamma_{6}$, boundary states ${\bf L_{10}}$ and ${\bf L_{6}}$ intersecting at a six dimensional manifold, split into four stacks of boundary states as $$\begin{aligned}
{\bf L_{10}} [\,U(4)]&\rightarrow &{\bf L^a_{10}}[\,U(3)]+{\bf L^d_{10}}[\,U(1)]\\
{\bf L_{6}}[\,Sp(4)]& \rightarrow& {\bf L^b_{6}}[\,Sp(2)]+{\bf L^c_{6}}[\,Sp(2)]\end{aligned}$$
where we have indicated in brackets the gauge group living on the corresponding brane. Thus, boundary states ${\bf L^a_{10}},{\bf L^d_{10}},{\bf L^b_{6}},{\bf L^c_{6}}$ do match with the basic branes $a,b,c,d$ arising in intersecting brane models on toroidal constructions ([@imr; @cim:2002; @Gmeiner:2005vz]).
Thus, drawing boundary states as lines and interpreting multiplicities as intersection numbers we are lead to a graphic representation as the one given in Figure \[LR\].
1.0cm
Besides states propagating between different branes we must consider states along the same type of branes. They lead to vector like matter.
Interestingly enough, massless states $( 4 0 2 0 ), ( 0 4 2 0 )$ and $( 2 2 2 0 )$ do propagate in ${\bf L_{6}}-{\bf L_{6}}$ sector. They satisfy $\Gamma.\mu=\frac12$ and thus, together with $(1)( 0 0 0 0 )$, descending from the six dimensional vector, lead to $$9(1,2,2)_0
\label{L6L6}$$ candidates to LR Higgses[^8]. There is also a pair of states $(1,2,1)_0+(1,1,2)_0$ descending from the symmetric representation.
Notice that this sector is non chiral and that states fill up an ${\cal N}=2$ hypermultiplet
Branes $c$ and its image $c*$ are placed here on top of each other and on top of an orientifold point (leading to $Sp(2)$). Since such branes are parallel in the torus, following similar steps as discussed [@cim:2003], we can think into separating them away from the orientifold point in the torus. Thus, $SU(2)_R \rightarrow U(1)_c$ where $U(1)_c$ charges are given by $T^3_R$ eigenvalues. Therfore, by introducing the weak hypercharge $$Y=-T^3_R+\frac12 Q_{B-L}$$ we find that the original LR symmetric model breaks down to $SU(3)\times SU(2)_L\times U(1)_Y$ MSSM with three chiral generations $$3[(3,2,1)_{\frac16}+(\bar3,1)_{-\frac23}+(\bar3,1)_{\frac13}+
(1,1)_{\frac12}+(1,2)_{-\frac12}+(1,1)_{0}]$$ including three right handed neutrini. Moreover, LR chiral states $(1,2,2)_0$ decompose into $(1,2)_{-1/2}+(1,2)_{1/2}$ with the correct MSSM Higgs charges.
A pictorial representation is presented in Figure \[LRSM\].
0.5cm
Besides these basic boundary states leading to the MSSM structure, additional stacks of branes must be added in order to satisfy tadpole cancellation equations. Different choices are possible and each of them will give rise to particular phenomenological features. Here we just want to show a simple choice that allows to complete the above construction to a fully consistent supersymmetric model.
With this aim we introduce a stack of $N_1$ ${\bf L_1}$ “$N_Z$ branes" and three stacks of $N'_5$ ${\bf L'_5}$, $N'_1$ ${\bf L'_1}$, $N'_{10}$ ${\bf L'_{10}}$ “$D_Z$-branes". Therefore the starting gauge group structure is NN&:&SP(N\_6)U(N\_[10]{})SP(N\_1)\
DD&:&SP(N’\_5)U(N’\_[10]{})SP(N’\_1)
When performing the modding given in Eq.(\[gmodding\]) tadpole equations (\[gtadpoles\]) become
$$\begin{aligned}
2N_6 +2N_{10}& = & 16\\
N_1+ 3N_6 +2N_{10} & = & 24\\
2Tr\gamma^{2}_6 +2Tr\gamma^{2}_{10}& = & 0\\
Tr\gamma^{2}_1+ 3Tr\gamma^{2}_6 +2Tr\gamma^{2}_{10} & = & 0\\
N'_5 +2N'_{10}& = & 16\\
N'_1+ N'_5 +2N'_{10} & = & 24\\
Tr{\gamma'}^{2}_5 +2Tr{\gamma'}^{2}_{10}& = & 0\\
Tr{\gamma'^{2}}_1+ Tr{\gamma'}^{2}_5 +2Tr{\gamma'^{2}}_{10} & = & 0\\
2Tr{\gamma}_6+2 Tr{\gamma'}_5 +2 (Tr{\gamma}_{10}+2Tr{\gamma'}_{10})& = & 0\\
Tr{\gamma}_1+ 2 Tr{\gamma'}_1+ 3 Tr{\gamma }_6 +2(Tr{\gamma}_{10}+2Tr{\gamma'}_{10}) & = & 0\end{aligned}$$
A solution is obtained by choosing $N_1=N_6=N_{10}=4$ in the DD sector and $N'_{10}=1$, $N'_{1}=8$, $N'_{5}=14$ with the corresponding twists embedding (and induced gauge symmetry breaking) given in Table \[tabtwists\]. Massless states propagating at intersection of different pairs of branes are shown in Table \[tabpnnspectrum\]
Sector Brane twist Group
-------- ----------------- --------------------------------- --------------------------------------------------------
NN ${\bf L_{10}}$ $V_{10}=\frac14(1,1,1,3)$ $U(4)\rightarrow U(3)\times U(1)$
${\bf L_{6}}$ $V_6=\frac14(0,2)$ $Sp(4)\rightarrow SU(2)_L\times SU(2)_R$
$ {\bf L_{1}}$ $V_1 = \frac14(1,1) $ $Sp(4)\rightarrow U(2)$
[DD]{} ${\bf L'_{10}}$ $V'_{10}=\frac14(3)$ $U(1) $
$ {\bf L'_{1}}$ $V'_{1}=\frac14(0,2,1,1)$ $SP(8)\rightarrow SP(2)\times \times SP(2)\times U(2)$
$ {\bf L'_{5}}$ $V'_{5}=\frac14(0,0;2,2;1,1,1)$ $SP(14)\rightarrow SP(4)\times SP(4)\times U(3)$
: Original ${\bf L_{I}}$ branes do split due to twist $V_{I}$. The original gauge group living on ${\bf L_{I}}$ world volume breaks accordingly.[]{data-label="tabtwists"}
-----------------------------------------------------------------------------------------------------------------------------
Sector Branes States IRREP
-------- ------------------------------- -------------------------------- ---------------------------------------------------
NN ${\bf L_{10}}- {\bf L_{6}}$ $(3310)$, $(5110)$,$(1510)$ SM
N $ {\bf L_{10}}- {\bf L_{1}}$ $(3310)$ none
ND ${\bf L_{10}}- {\bf L'_{1}}$ $(3310)$ $(3,1,1;2,1,1)_{1/3}+(1,1,1;2,1,1)_{-1/3}+$
$(3,1,1;1,2,1)_{-1/3}+
(1,1,1;1,2,1)_{1/3}$
ND ${\bf L_{10}}- {\bf L'_{5}}$ $(3310)$ $(3,1,1;4,1,1)_{1/3}+(1,1,1;4,1,1)_{-1/3}+$
$(3,1,1;1,4,1)_{-1/3}+
(1,1,1;1,4,1)_{1/3}$
NN ${\bf L_{6}}- {\bf L_{1}}$ $(4400)+(2200)$ none
ND ${\bf L_{6}}- {\bf L'_{1}}$ $(4400)+(2200)$ $(1,{\underline {2,1}};{\underline {2,1}},1)_{0}$
ND ${\bf L_{6}}- {\bf L'_{10}}$ $(3310), (5110),(1510)$ none
ND ${\bf L_{6}}- {\bf L'_{5}}$ $(6200),(4400),(2220),(0420)$, $(1,{\underline {2,1}};{\underline {4,1}},1)_{0}$
ND ${\bf L_{1}}- {\bf L'_{10}}$ $(3310)$ 3(1,2,1;1)+3(1,1,2;1)
ND ${\bf L_{1}}- {\bf L'_{5}}$ $(2600),(4020)$ $(2,1,3,1)+({\bar 2},1,{\bar 3},1)$
$(2,1,{\bar 3},1)+({\bar 2},1,{ 3},1)$
DD ${\bf L'_{10}}- {\bf L'_{1}}$ $(3310)$ $(1;2,1,1)+(1;1,1,2)$
DD ${\bf L'_{10}}- {\bf L'_{5}}$ $(3310)$ $(1,1,1;4,1,1)+(1,1,1;1,4,1)$
-----------------------------------------------------------------------------------------------------------------------------
: Massless chiral primary states, denoted by $(m_1,\dots,m_6)$ propagating between a pair of branes ${\bf L_I}-{\bf L_J}$ are indicated. After performing the orbifold twist, the original gauge group in each brane breaks into several subgroups as in Table \[tabtwists\] above. The last column shows the representations, with respect to such subgroups, in which chiral superfields accommodate.[]{data-label="tabpnnspectrum"}
A study of interactions among LR states and extra matter is beyond the scope of the present work. Nevertheless, some general remarks about Yukawa couplings can be advanced.
As a general observation notice that a Yukawa coupling will have the form $$Y_{ijk}\Phi^i _{ba} \Phi^j_{ac} \Phi^k_{cb}$$ where $\Phi _{ab}$ is the chiral superfield insertion connecting boundaries $a$ and $b$ and $i,j,k$ refers to internal CFT labels. Such a term should be a singlet of the gauge group and invariant under $\Gamma$ modding. Moreover, it must be allowed by the fusion rules (\[su2fusion\]) of the internal conformal field theory [@dg; @brunschom]. Namely,
$$Y_{ijk}\propto \langle i j k\rangle\propto {\cal N}_{ij}^k$$
For instance, couplings like $$[( {\underline{ 5 1}} 1 0 )](3,2,1)_{1/3}^{ab} \times [( 3 3 1 0) )]({\bar 3},1,2)_{-1/3}^{ac}[( 2 2 2 0 )](1,2,2)_0^{bc}$$
(where we have indicated the internal charges in brackets) are non vanishing and lead to degenerate masses for two quark generations. Fusion rules forbid masses for the first quark generation (see (\[3ciralstates\])). A similar result is obtained for lepton masses since the same internal states are involved for leptonic Yuakawa couplings.
The general pattern is very similar (the number of Higgses is different) to the one found in Ref.[@aiq] in the context of branes at singularities.
It is interesting to notice that couplings of quarks or leptons to states $[( {\underline{4 0}} 2 0 )](1,2,2)_0$ and $[(1)(0000)](1,2,2)_0$, discussed in (\[L6L6\]), are not allowed by fusion rules. Thus, the model contains four effective LR Higgses.
In particular, as addressed in in [@aiq], the full picture of mass structures becomes more complicated due, for instance, to the presence of Yukawa couplings of quarks with colored triplets present at other intersections. For instance, D quarks will couple to triplets in the ${\bf L_{10}}-{\bf L'_1}$ sector
$$[( 3 3 1 0) )]({\bar 3},1,2)_{-1/3}^{ac}\times
[( 3 3 1 0) )]({ 3},1,1;2,1,1)_{1/3}^{a1'}\times [( 2 2 0 0 )](1,1,2;2,1,1)_0^{1'c}$$
and therefore D quarks and triplets mix once $SU(2)_R $ doublet acquires a vev. Through similar terms all the three quarks would become massive.
Notice also that, from the 9 candidates to be interpreted as Higgs particles coming from ${\bf L_6}-{\bf L_6}$ sector, only those with CFT quantum numbers $( 2 2 2 0 )$ are allowed in Yukawa couplings. For all of them, on the other hand, mass term like couplings are allowed. Thus, we can imagine a scenario where some of the $(1,2,2)$ multiplets become very massive.
### An alternative with the LR week sector on DD branes
In the example discussed above the basic branes ${\bf L_{10}}$, containing strong group, and ${\bf L_6}$, where $SU(2)_L \times SU(2)_R$ lives, were placed in the same NN sector. However, it might be useful for future phenomenological applications, to place the part of the spectrum containing the $SU(2)_L \times SU(2)_R$ gauge theory on the branes in the DD sector.
An interesting possibility of this kind is shown in Table \[tabL6DD\]. In this case, even if ${\bf L_6}$ is placed in DD sector, tadpole cancellation requires to place a similar stack in NN sector thus leading to two alternatives realizations of (3 generations) LR models. Extra boundary states, required by consistency, are of the same kind we introduced in previous example, thus, states propagating between different pairs of branes can read directly from third column in Table \[\[tabpnnspectrum\]\].
Sector Brane twist Group
-------- ----------------- --------------------------- -------------------------------------------
NN ${\bf L_{10}}$ $V_{10}=\frac14(1,1,1,3)$ $U(4)\rightarrow U(3)\times U(1)$
${\bf L_{6}}$ $V_6=\frac14(0,2)$ $Sp(4)\rightarrow SU(2)_L\times SU(2)_R$
$ {\bf L_{1}}$ $V_1 = \frac14(1,1) $ $Sp(4)\rightarrow U(2)$
[DD]{} ${\bf L'_{10}}$ $V'_{10}=\frac14(3)$ $U(1) $
${\bf L'_{6}}$ $V'_6=\frac14(0,2)$ $Sp(4)\rightarrow SU(2)_L\times SU(2)_R$
$ {\bf L'_{1}}$ $V'_{1}=\frac14(1,1)$ $Sp(4)\rightarrow U(2)$
$ {\bf L'_{5}}$ $V'_{5}=\frac14(1,1,1)$ $SP(6)\rightarrow U(3)$
: An alternative construction leading to a duplicated LR structure. The group $ SU(2)_L\times SU(2)_R$ can be chosen to be in the DD sector or in the NN sector.[]{data-label="tabL6DD"}
It can be easily verified that this solution satisfies tadpole equations.
### Massless U(1) and K-theory constraints
Anomalous $U(1)$ generators acquire mass through the Green-Schwarz mechanism. However, a non-anomalous U(1) may also become massive if there is an effective coupling $ B\wedge F$. We must therefore ensure that $Q_{B-L}$ is not one of them and remains massless.
For a $U(1)_a$ gauge group on a brane $a$, we will have the coupling $$\int _{[M_4]} (C_2^a- {C_2^a}')\wedge F^a
\label{BF}$$ where, in a geometrical setting, $C_2^a$ (${C_2^a}'$ is its $\Om$-image) come from the reduction of a $C_p$ form on a supersymmetric cycle $a$ and $F_a$ is the $U(1)_a$ gauge field.
Therefore, by expanding $ C_2^a$ forms, or analogously their corresponding cycles, into Ishibashi states, with expansion coefficients $D_a^i$ (and their $\Om$-images ${D'}_a^{i}$) ([@dhs0403]), and requiring $Q_{B-L}=\sum x_a Q^a$ coupling to vanish leads to
$$N_a(D_a^i-{D_a^i}')x_a=0$$
for each Ishibashi state $|i \rangle\rangle=|\lambda \mu ;x \rangle \rangle$ in the orbifold theory.
For the Ishibashi state $I=|(33)(33)(11)(00)^3; x=1 \rangle\rangle$ we obtain $$3 x_{a}i(D_{10,+}^I-D_{10,-}^I)- x_{d}i(D_{10,+}^I-D_{10,-}^I)=0$$ where $D_{10,\pm}$ are the expansion coefficients for the parts $a$ or $d$ of the brane ${\bf L_{10}}$ (\[twbs\]) in terms of Ishibashi states.
Then a solution with $x_{a}=1/3$ and $x_{d}=1$ corresponds to having $$Q_{B-L}=\frac13 Q_3+Q$$ massless . It can be shown that this is the only nontrivial condition.
Finally, there are additional constraints on the compactified theory coming from the fact that D-brane charges are classified by K-theory [@Witten:1998cd]. One particular constraint is the vanishing of the Witten global anomaly which means that the number of massless fermions in the fundamental representation of a symplectic group is even. We have verified that the Witten anomaly vanishes in the example we presented in the last section.
Generically, however, K-theory might impose additional constraints. It would be interesting to further check consistency using maybe the method of probe-branes where additional constraints might appear [@U2000xp].
Summary and outlook
===================
In the first part of the present work we have addressed the construction of six dimensional Type IIB orientifold models based on a *Gepner models* internal space
Six dimensional models were constructed by considering stacks of B-type boundary states, required by a diagonal invariant partition function. Such boundary states would correspond to D branes wrapping even cycles of $K3$ [@eoty; @ooy]. We have found the explicit expressions for these boundary states and the rules to compute their massless states spectra (associated to open strings propagating among them). Tadpole cancellation equations were also derived. Explicit computations for the sixteen diagonal Gepner models present in $D=6$ will be collected in [@aajd6gepners].
We have also shown how moddings by internal discrete symmetries and the so called parity and quantum dressings can be included in this context. In particular, A-type boundary models, corresponding to a charge conjugation invariant, should be obtainable [@greenplesser] by performing possible moddings on B-type construction.
As shown in Ref.[@eoty; @ooy], more general boundary states, corresponding $SO(3)$ rotations, and including A and B-type cases, can be constructed in $K3$. It would be interesting to explore how such states could be obtained in the present context
Following the ideas presented in Ref.[@aaj], four dimensional chiral models were built by further compactifying on a $T^2$ torus, sharing some of the symmetries of the $D=6$ models, and by modding out by such symmetries. The projection is realized as the combined action of a phase symmetry modding of the Gepner sector and a rotation of the torus lattice accompanied by a twist on Chan-Paton factors. The twist on Chan Paton factors can be viewed as a breaking of the original boundary states into component states with specific monodromy under the twist. Generically, when even order moddings are considered, new sets of branes are required for tadpole cancellation.
Interestingly enough, inspection of six dimensional spectra allows to identify phenomenologically appealing models without the use of a computer scanning.
As an example of the construction, we described a $Z_4$ modding of the model $6620 \times \IT^2$. In such a model a basic structure of two stacks of four boundary states , which we call ${\bf L}_{10}$ and ${\bf L}_{6}$, exist, with gauge groups $U(4)$ and $Sp(4)$ respectively, living on their world volumes with three hypermultiplets propagating between them. These two stacks of boundary states constitute the basic, six dimensional, “bulding blocks" of the MSSM. In fact, we showed that further compactification on a torus, accompanied by $Z_4$ modding, leads to the breaking $$\begin{aligned}
{\bf L_{10}} [\,U(4)]&\rightarrow &{\bf L^a_{10}}[\,U(3)]+{\bf L^d_{10}}[\,U(1)]\\
{\bf L_{6}}[\,Sp(4)]& \rightarrow& {\bf L^b_{6}}[\,Sp(2)]+{\bf L^c_{6}}[\,Sp(2)]\end{aligned}$$ namely, into a four stacks of branes, giving rise to a Left-Right symmetric model with three massless generations living at the boundary states intersections. Further breaking to a MSSM (with the expected three right handed neutrinos) can be achieved.
The four stacks $a,b,c,d$ of boundary states, possess the “basic building block" properties used in intersecting brane realizations [@imr; @Marchesanothesis] of the Standard Model. They can be further embedded into a fully consistent supersymmetric orientifold model.
We have indicated in the example some appealing features of the basic Yuakawa couplings structure. For instance, the fact that vertex operators must connect different boundary states, the requirement of gauge and discrete twist invariance and the CFT fusion rules allow to discard several terms. A detailed investigation of the structure of Yuakawa couplings remains to be done. In particular, it would be interesting to see if a more systematic study, like in [@cim:2003] for intersecting branes, can be pursued in this context of RCFT.
An interesting feature of the hybrid construction is that lowering of the string scale [@Arkani-Hamed:1998rs; @Antoniadis:1998ig; @imr; @CIMlower; @Kokorelis:2002qi; @Marchesanothesis] could be achieved by considering large extra dimensions in the $T^2$ torus, transverse to the whole configuration of intersecting boundary states.
Indeed, in the present examples of the type $Gepner \times T^2 /\IZ_N$, boundary states would correspond to branes wrapping cycles on $K_3$ and stuck at a ${\bf C/\IZ_N}$ singularity. Thus, if we denote by $V_4$ the volume of the Gepner piece, which should be of the order of the string scale $V_4\propto 1/M_s^4$, and by $V_2$ that of two dimensional manifold. Then we expect the Planck scale, after dimensional reduction to four dimensions, to be $$M_{Planck} \ =\ \frac2{\lambda } M_s^4 \sqrt{V_4V_2}$$ where $\lambda$ is the string coupling. Therefore the string scale $M_s$ can be lowered by choosing the volume $V_2$ ($V_2 \ =\ \frac{ M_{Planck}^2 \lambda ^2 }{4 M_s^4}$) sufficiently large. Recall that the models constructed here are fully supersymmetric and though lowering the scale could be phenomenologically attractive in some cases it is not as compelling as in non supersymmetric models.
Presumably, having these large extra dimension could allow for the introduction of dilute fluxes in a supergravity limit of some of these hybrid construction [@Grana:2005jc].
**Acknowledgments**
We are grateful to A. Font, L.E. Ibañez and A. Uranga for stimulating observations and suggestions. G.A. work is partially supported by ANPCyT grant PICT 11064 and CONICET PIP. J. J. is grateful to the Particle and Field Group of the Bariloche Atomic Center (CAB) for partial support and hospitality during the first part of this work.
SO(2d) Space- time partition functions
======================================
We show the basic ingredients needed for the computation of modular transformation matrices of the space- time part of the partition functions of closed and open sectors. Even if we are mainly interested in $D=6$ dimensions we present here the general result in $D$ dimensions Consider $SO(2d)$ $d=\frac{D-2}2$ for dimensions $D=4,6,8$. There exist four representations $\Lambda>=|0>,|v>,|s>,|c>$ whose fundamental weights are encoded as $$\begin{aligned}
|0>&=& (0,0,\dots,0)\\\nonumber
|v>&=& (1,0,\dots,0)\\\nonumber
|s> &=& (\frac12,\frac12,,\dots,\frac12)\\\nonumber
|c> &=& (\frac12,\frac12,,\dots,-\frac12)\end{aligned}$$
Scalar product between two representations $ \Lambda$ and $ \Lambda '$ is given by $$\Lambda.\Lambda'= \sum_{l=0}^{d-1} \Lambda_l\Lambda_l$$
Recall that we need to redefine $\Lambda_{Gepner}=2\Lambda$ in order to have the normalization usually used in Gepner models.
The character that is associated to highest weight $\Lambda$, at level one, is given by $\chi_{\Lambda,1}(\tau)=\theta_{\Lambda,1}(\tau)$ [@goddardolive]. It leads to modular transformation matrices
$$\begin{aligned}
S_{\Lambda,\Lambda'}&=& e^{-2i\pi \Lambda.\Lambda' }\\
T_{\Lambda,\Lambda'}&=& e^{2i\pi (\Lambda^2-\frac{d}{24 })}\delta({\Lambda,\Lambda'})\end{aligned}$$
Therefore, the space time matrix $P$ [@bs1] can be obtained as $$\hat{P}_{\Lambda,\Lambda'}=T_{\Lambda}^{\frac12}S_{\Lambda,\Lambda''}T_{\Lambda'',\Lambda''}^2
S_{\Lambda'',\Lambda}T_{\Lambda''}^{\frac12}$$
where $T_{\Lambda}^{\frac12}=e^{i\pi( h_{\Lambda}-\frac{d}{24 })}$ is the phase factor that is introduced in order to construct a real character from $\theta_{\Lambda,1}(\tau+\frac12)$ and $h_{\Lambda}$ is the conformal weight. It coincides with $\Delta= \frac{\Lambda^2}{2}$ only in the case in which quantum numbers $\Lambda$ are given in the standard range above. Thus, $P$ reads $${\hat P}_{\Lambda,\Lambda'}=e^{-i\pi\frac{d}{4 }}
e^{i\pi( h_{\Lambda}+h_{\Lambda'})}
\sum _{\Lambda''} e^{-2i\pi \Lambda.\Lambda'' }
e^{2i\pi (\Lambda'')^2 }e^{-2i\pi \Lambda''.\Lambda' }$$
It is easy to see that, when all states are in the range above, the matrix $P$ is given by $${P}=\left(
\begin{array}{cccc}
c & s & 0 & 0 \\
s & -c & 0 & 0 \\
0 & 0 & \zeta c & i\zeta s \\
0 & 0 & i\zeta s & \zeta c
\end{array}
\right)$$ with $c=cos\pi d/4$ and $s=sin\pi d/4$.
Since in the actual computation of Möbius amplitude weights are shifted from the standard range by $\beta_i,\beta_0$ vectors, it appears useful to rewrite $P$ as (see for instance ([@blumen0310] for $d=1$))
$${\hat P}_{\Lambda,\Lambda'}=\sigma(\Lambda)\sigma(\Lambda')
e^{-i\pi\frac{d}{4 }}
e^{-i\pi {\Lambda \Lambda'}
}
\sum _{\Lambda''}
e^{2i\pi (\Lambda''-\frac{\Lambda +\Lambda'}2)^2 }
\label{pfinal}$$
where
$$\sigma(\Lambda)=(-1)^{(\frac{\Lambda ^2}2 -h_{\Lambda})}$$
and $${\tilde P}_{\Lambda,\Lambda'}=
\sum _{\Lambda''}
e^{2i\pi (\Lambda''-\frac{\Lambda +\Lambda'}2)^2 }
\label{ptilde}$$ which, for NS weights (scalar or vector) reads
$${\tilde P_{(\Lambda_0+\Lambda'_0 +\nu_i\beta_i )}^{NS}}=\left \{ \begin{array}{lll}
& 1 \quad &d=1\\
&-e^{i\pi(\Lambda_0+\Lambda'_0 +\sum \nu_i)}\quad & d=3\\
& \frac12(1-e^{i\pi(\Lambda_0+\Lambda'_0 +\sum \nu_i )}&d=2
\end{array}\right.$$
The crosscap state in D=6
==========================
The $D=6$ spacetime bosons and fermions realize a $(2,2)$ superconformal algebra. The four world-sheet fermions $\psi^{2,3,4,5}$ have an $SO(4)$ symmetry which requires them to be organized into unitary representations of the affine transverse Lorentz algebra at level $k=1$. These are the scalar, vector, spinor and conjugate spinor representations labelled respectively by $s_0=-1,
0, 1, 2$. It proves convenient to split the representations of $SO(4)$ at level $1$ into those of $SO(2)\times SO(2)$. The latter are labelled by two numbers $\Lambda_0, \Lambda_1 = -1, 0, 1, 2$ subject to $\Lambda_0+\Lambda_{1}=0 \mod{2\mathbf{Z}}$.
In order to implement a GSO projection we define the vectors $\mu$’s and the inner product between them as $$\begin{aligned}
&& \lambda=(l_1, \dots, l_r)\\
&& \mu = (\Lambda_{0},\Lambda_1;s_1,\dots,s_r;m_1,\dots,m_r),\\
&& \mu\bullet\mu'\equiv-\frac{\Lambda_{0}\Lambda_{0}'}{4}-\frac{\Lambda_{1}\Lambda_{1}'}{4}
-\sum_{j}\frac{s_{j}s_{j}'}{4}+\sum_{j}\frac{m_jm_j'}{2(k_j+2)} .\end{aligned}$$ It is convenient to introduce special vectors $\b_{0}$, $\b _{j}$ and $\tilde \b_{1}$ $$\begin{aligned}
&& \b_{0}=(1,1;1,\dots,1;1,\dots,1),\\
&& \b_{j}=(0,2;0,\dots,0,\underset{j}{2}
,0,\dots,0;0,\dots,0),\ (j=1,\dots,r),\\
&& \tilde \b_{1}=(2,2;0,\dots,0;0,\dots,0).\\\end{aligned}$$ By using these vectors, we can construct the building blocks $\chi^{\lambda}_{\mu}(\tau)$ as $$\begin{aligned}
&&\chi^{\lambda}_{\mu}(\tau) =
\chi_{\Lambda_{0}}(\tau)\chi_{\Lambda_{1}}(\tau)\chi_{m_1 s_1}^{l_1}(\tau)
\dots \chi_{m_r s_r}^{l_r}(\tau)\end{aligned}$$ where $\chi_{\Lambda_{0}}(\tau)$ and $\chi_{\Lambda_{1}}(\tau)$ are $\widehat{SO(2)}_1$ characters. Then the GSO conditions and the condition of fermionic sectors are $$\begin{aligned}
&& 2\b_{0}\bullet \mu\in 2 \mathbf{Z}+1,\qquad
\b_{j}\bullet \mu \in \mathbf{Z},\qquad
\tilde \b_{1}\bullet \mu \in \mathrm{Z}, \label{BetaCondition4}\end{aligned}$$
The type-B, GSO-projected partition function is then given by Z\_D\^[B]{}(, )=1[2\^r]{} \_[b\_0=0]{}\^[K-1]{} \_[b\_[1]{}, b\_1,…,b\_r=0]{}\^1 [\_[,]{}]{}\^ (-1)\^[s\_0]{} \^\_(q) \^\_[+b\_0\_0+b\_[1]{} \_[1]{} +b\_1 \_1 +…b\_r \_r]{} (q) .Here $K={\rm lcm}(4,2k_j+4)$ and ${\sum}^\beta$ means that the sum is restricted to those $\lambda$ and $\mu$ satisfying (\[BetaCondition4\]).
The Klein bottle partition function is obtained from that of the torus by keeping states with equal left and right oscillators. In the direct channel it is given by
\[kb\]Z\_K\^B &= &\_0\^\_[cl]{}( e\^[-4t (L\_0-)]{} ) where ${\rm Tr'}_{cl}$ denotes the trace over the oscillator modes in the closed string sector. The integration over the bosonic zero modes yields the factor $(4\pi^2 \alpha')^3$.
The Klein bottle amplitude in the transverse channel is obtained by performing an $S$ modular transformation
\_K\^B &=&2\^7 \_[j]{} \_0\^ 1[(2il)\^3 ]{} [\_[’,’]{}]{}\^[ev]{} \_[\_0=0]{}\^[K-1]{} \_[\_1,…,\_r=0]{}\^1 \_[\_1,…,\_r=0]{}\^1\_[\_1,…,\_r=0]{}\^1\
&& (-1)\^[\_0]{} \_[\_0’,2+\_0 +2\_j]{}\^[(4)]{}\_[\_1’,\_0]{}\^[(4)]{}\
&&\_[j=1]{}\^r ( \_[m\_j’,\_0+(1-\_j)(k\_j+2)]{}\^[(2k\_j+4)]{} \_[s\_j’,\_0 +2 \_j+2(1-\_j)]{}\^[(4)]{} ) \^[’]{}\_[’]{} (2il)
where $l=1/t$.
From the Klein bottle amplitude in the transverse channel we can read the expression for the crosscap state up to signs which are contained in the Möbius strip amplitude. The result is that the crosscap state is given by
$$\begin{aligned}
\label{crosscapbf2}
| C\rangle^{NS}_B&=&
\frac1{\kappa_c}
{\sum_{\lambda',\mu'}}^{ev} \sum_{\nu_0=0}^{\frac{K}2-1}
\sum_{\nu_1,\ldots,\nu_r=0}^1 \sum_{\epsilon_1,\ldots,\epsilon_r=0}^1
\,\, \\ \nonumber
& & \eta(\nu_0,\nu_i,\epsilon_j)
\delta_{\Lambda'_0,2+2\nu_0
+2\sum \nu_j}^{(4)}\delta_{\Lambda'_{1},2\nu_0}^{(4)}
\\\nonumber
&& \prod_{j=1}^r \Biggl(
\frac{P_{l'_j,\epsilon_j k_j}}{ \sqrt{S_{l'_j,0} }} \,
\delta_{m_j',2\nu_0+(1-\epsilon_j+\omega_j)(k_j+2)}^{(2k_j+4)}
\delta_{s_j',2\nu_0 +2
\nu_j+2(1-\epsilon_j)}^{(4)}
\Biggr) |\lambda',\mu'\rangle\rangle_c\end{aligned}$$
We still have to fix the signs of the crosscap state. As in [@blumen0310], the condition that GSO orbits of hatted characters transforms,under the $P$-transformation, into themselves will be used as an ansatz to fix the signs in the crosscap state.
We want to compute the modular transformation of $${ M^\lambda_\mu=\sum_{\nu_0=0}^{\frac{K}2-1}
\sum_{\nu_1,\ldots,\nu_r=0}^1 (-1)^{\left[ h^\lambda_{\mu}(\nu_0,\nu_j)-
h^\lambda_{\mu}\right]}
\,\, \widehat\chi^\lambda_{\mu+2\nu_0\beta_0+ \sum \nu_j \beta_j }
(it+{\textstyle{\frac12}}) ,}
\label{msorbit}$$
Thus, when we perform the $P$ transformation in (\[msorbit\]) we get
$$\begin{aligned}
\nonumber
M^\lambda_\mu &=&\sum _{\mu',\lambda'}^{\beta}
\sum_{\nu_1,\ldots,\nu_r=0}^1
\prod_{k<l}(-1)^{\nu_l\nu_k}\sigma'_{(l',m',s')} e^{i\pi \sum
\nu_i(\Lambda_0+s_i-\Lambda'_0-s'_i+1)} e^{-i\pi{\mu.\mu'}}
\delta_{s_i+s_i',0}^{(1)} \\\nonumber
& & \tilde P_{(\Lambda_0+\Lambda'_0 +\nu_i\beta_i )}
\sum_{\epsilon=0}^1 P_{l,|\epsilon k-l'|}\, \delta_{m+m'+(1-\epsilon)(k+2)}^{(2)}\\&&
(-1)^{\epsilon_i (\frac{l_i'+m_i'}2+s_i')}(-1)^{\epsilon_i
(\frac{m_i}2+s_i+\nu_i)} \,\, \widehat\chi^\lambda_{\mu'}
(it+\frac12) \label{mstr3}\end{aligned}$$
where the spacetime $\tilde P$-matrix is given in (\[ptilde\]) for $d=2$.
By summing over $\nu_i$ we are lead to
$$\begin{aligned}
\nonumber
M^\lambda_\mu &=&\sum _{\mu',\lambda'}^{\beta}
\sum_{\nu_1,\ldots,\nu_r=0}^1 \sigma'_{(l',m',s')}\\\nonumber
&& \prod_{k<l}(-1)^{\eta_l\eta_k} \delta^{(2)}_{\Lambda_0+\Lambda_1+\Lambda'_0+\Lambda'_1+\sum
\nu_i,0}
e^{-i\pi{\mu.\mu'}} \delta_{s_i+s_i',0}^{(1)} \\\nonumber
& &
\sum_{\epsilon=0}^1 P_{l_i,|\epsilon k_i-l_i'|}\, \delta_{m_i+m_i'+(1-\epsilon_i)(k_i+2)}^{(2)}\\&&
(-1)^{\epsilon_i (\frac{l_i'+m_i'}2+s_i')}(-1)^{\epsilon_i (\frac{m_i}2+s_i)}
\,\, \widehat\chi^\lambda_{\mu'}
(it+\frac12)
\label{mstreta}\end{aligned}$$
$$\eta_i=\Lambda_0 +s_i-\Lambda'_0-s'_i+1+\epsilon_i$$
Using (\[mstreta\]) we determine the unknown signs $\eta(\nu_0,\nu_i)$ in (\[crosscapbf2\]) to obtain the expression (\[crosscapbf\]), for the crosscap state.
MS amplitude in the direct channel
===================================
We present here the expression for the tree-channel Möbius amplitude required to extract the gauge and matter field content. It can be computed from the amplitude of closed strings propagating between a boundary state (\[borde\]) and the crosscap state (\[crosscapbf\]) and then performing a modular transformation to open string channel[^9].
$$\begin{aligned}
\label{moebiloopdr}
{{\cal Z}_M}^{B,NS}_{\alpha}&=&\langle C|q^{H}|{\alpha}\rangle _B =\\ \nonumber
&=& -\frac1{2^{r+1}}
\int_0^\infty \frac{dt}{t^5}\,
\frac{1}{\eta(it+\frac12)^3 }
{\sum_{\lambda,\mu}}^{ev}
\sum_{\epsilon_1,\ldots,\epsilon_r=0}^1
\left( \prod_{k<l} (-1)^{\rho_k\rho_l}\right)
\\ \nonumber
& & e^{i(\pi/2) \sum \omega_j(m_j-2L_j +\epsilon_j(k_j+2))}
(-1)^{\sum \omega_j \Lambda_0 /2} \\ \nonumber
& & \delta_{\sum_j \rho_j, 1+ \frac{\Lambda_0+\Lambda_{1}}{2}+\sum \om_j}^{(2)}
\, \delta_{\Lambda_0,0}^{(2)}\delta_{\Lambda_{1},0}^{(2)}
\delta_{\sum \frac{K'}{ 2k_j+4}\left[ 2(M_j- \Delta_j)-m_j-
\epsilon_j(k_j+2)\right],0}^{(K')}\\ \nonumber
& & \, \prod_{j=1}^r \Biggl( \sigma_{(l_j,m_j,s_j)}\, \,
Y_{L_j,\epsilon_j\, k_j}^{l_j}\, \,
\delta_{m_j+\epsilon_j(k_j+2),0}^{(2)}\\\nonumber
&& \delta_{s_j,0}^{(2)}\,
(-1)^{\frac{\epsilon_j}{2}\left[2S_j-s_j-2\epsilon_j\right]}\,
(-1)^{\frac{(1-\epsilon_j)}{ 2}
\left[2M_j-m_j-\epsilon_j(k_j+2)\right]} \Biggr)\\ \nonumber
&& \, \widehat\chi^{\lambda}_{\mu} (it+{\textstyle\frac{1}{2}})\end{aligned}$$
where $\rho_j= \frac{\L_0+s_j}{2} +\epsilon_j-1+\sum\om_j$.
Here $$Y_{l_1,l_2}^{l_3}=\sum_{l=0}^k \frac{S_{l_1,l}\, P_{l_2,l} \, P_{l_3,l}}{
S_{0,l} }=(-1)^{\frac{1-\epsilon_j}2(2L_j+lj)}N_{L_j,L_j}^{|\epsilon_jk_j-l_j|}$$ in terms $SU(2)_k$ fusion coefficients [@difran]
$$N_{L_1,L_2}^{l}=\left \{ \begin{array}{lll}
&&1 \quad {\rm if} \quad |L_1-L_2|\le l\le {\rm min} \{{L_1+L_2,2k-(L_1+L_2)}\}\\
&& L_1+L_2+l= \,{\rm even }\\
&& N_{L_1,L_2}^{l}=0\, \quad {\rm otherwise}
\end{array}\right.
\label{su2fusion}$$
[99]{} G. Aldazabal, E. C. Andrés and J. E. Juknevich, [*Particle models from orientifolds at Gepner-orbifold points,*]{} JHEP [**0405**]{}, 054 (2004), hep-th/0403262.
R. Blumenhagen and A. Wisskirchen, *Spectra of 4D, N = 1 type I string vacua on non-toroidal CY threefolds*, Phys. Lett. B438 (1998) 52, hep-th/9806131. S. Govindarajan and J. Majumder, [*Crosscaps in Gepner models and type IIA orientifolds,*]{} JHEP [**0402**]{} (2004) 026, hep-th/0306257. G. Aldazabal, E. C. Andrés, M. Lestón and C. Nuñez,[*Type IIB orientifolds on Gepner points*]{} JHEP [**0309**]{}, 067 (2003), hep-th/0307183. R. Blumenhagen, [*Supersymmetric orientifolds of Gepner models*]{} JHEP [**0311**]{}, 055 (2003), hep-th/0310244. R. Blumenhagen and T. Weigand, [*Chiral supersymmetric Gepner model orientifolds*]{}, JHEP [**0402**]{} (2004) 041, hep-th/0401148. I. Brunner, K. Hori, K. Hosomichi and J. Walcher,[*Orientifolds of Gepner models*]{}, hep-th/0401137. R. Blumenhagen and T. Weigand, *A Note on Partition Function of Gepner Model Orientifolds* Phys.Lett. B591 (2004) 161-169, hep-th/0403299. T. P. T. Dijkstra, L. R. Huiszoon and A. N. Schellekens, [*Chiral supersymmetric standard model spectra from orientifolds of Gepner models,*]{} Phys. Lett. B [**609**]{}, 408 (2005), hep-th/0403196. T.P.T. Dijkstra, L.R. Huiszoon, A.N. Schellekens, [*Supersymmetric Standard Model Spectra from RCFT Orientifolds* ]{}, Nucl.Phys.B710:3-57,2005, hep-th/0411129. C. Angelantonj, M. Bianchi, G. Pradisi, A. Sagnotti and Y. Stanev, [*Comments on Gepner models and Type I vacua in string theory*]{}, Phys. Lett. [**B387**]{} (1996) 743, hep-th/960722.
D. Gepner, *Lectures On N=2 String Theory*, [*Lectures at Spring School on Superstrings,*]{} Trieste, Italy, Apr 3-14, 1989.
J. Fuchs, A. Klemm, C. Scheich and M. G. Schmidt, *Spectra And Symmetries Of Gepner Models Compared To Calabi-Yau Compactifications*, Annals Phys. [**204**]{}, 1 (1990). J. Fuchs, A. Klemm, C. Scheich and M. G. Schmidt, *Gepner Models With Arbitrary Affine Invariants And The Associated Calabi-Yau Spaces*, Phys. Lett. B [**232**]{}, 317 (1989).
A. Font, L. E. Ibanez, M. Mondragon, F. Quevedo and G. G. Ross, *(0,2) Heterotic String Compactifications From N=2 Superconformal Theories,* Phys. Lett. B [**227**]{}, 34 (1989).
D. Gepner, *String Theory On Calabi-Yau Manifolds: The Three Generations Case*, hep-th/9301089. C. Angelantonj, A. Sagnotti, *Open Strings* Phys.Rept. 371 (2002) 1-150; Erratum-ibid. 376 (2003) 339-405, hep-th/0204089. A. Recknagel and V. Schomerus, *D-branes in Gepner models*B Nucl. Phys. B531 (1998) 185, hep-th/9712186.
J. M. Maldacena, G. W. Moore, N. Seiberg, [*Geometrical interpretation of D-branes in gauged WZW models*]{}, JHEP [**0107**]{}:046,2001, hep-th/0105038.
L. E. Ibanez, F. Marchesano and R. Rabadan, [*Getting just the standard model at intersecting branes,*]{} JHEP [**0111**]{} (2001) 002, hep-th/0105155.
R. Blumenhagen, L. Goerlich, B. Kors, D. Lust, [*Noncommutative compactifications of type I strings on tori with magnetic background flux*]{}. JHEP [**0010**]{}:006, 2000,hep-th/0007024.\
R. Blumenhagen, B. Kors, D. Lust; [*Type I strings with F flux and B flux*]{}. JHEP [**0102**]{}:030, 2001, hep-th/0012156.\
M. Berkooz, M. R. Douglas, R. G. Leigh, [*Branes at angles*]{}, Nucl.Phys. [**B480**]{}:265-278,1996, hep-th/9606139.\
G. Aldazabal, S. Franco, Luis E. Ibanez, R. Rabadan, A.M. Uranga, [*Intersecting brane worlds*]{}. JHEP [**0102**]{}:047,2001, hep-ph/0011132.\
M. Cvetic, G. Shiu and A. M. Uranga, *Three-family supersymmetric standard like models from intersecting brane worlds,* Phys. Rev. Lett. [**87**]{}, 201801 (2001),hep-th/0107143.\
M. Cvetic, G. Shiu and A. M. Uranga, *Chiral four-dimensional N = 1 supersymmetric type IIA orientifolds from intersecting D6-branes*, Nucl. Phys. B [**615**]{}, 3 (2001),hep-th/0107166.\
A. M. Uranga, [*Chiral four dimensional string compactifications with intersecting D-branes*]{}. Class.Quant.Grav. [**20**]{}:S373-S394,2003, hep-th/0301032.\
R. Blumenhagen, M. Cvetic, P. Langacker and G. Shiu, *Toward realistic intersecting D-brane models*,hep-th/0502005. G. Honecker and T. Ott, [*Getting just the supersymmetric standard model at intersecting branes on the Z(6)-orientifold,*]{} Phys. Rev. D [**70**]{} (2004) 126010 \[Erratum-ibid. D [**71**]{} (2005) 069902\], hep-th/0404055\]. J. Fuchs, P. Kaste, W. Lerche, C. A. Lutken, C. Schweigert and J. Walcher, *Boundary fixed points, enhanced gauge symmetry and singular bundles on K3,* Nucl. Phys. B [**598**]{} (2001) 57, hep-th/0007145.
G. Aldazabal, E.C. Andr[é]{}s and J.E. Juknevich, Tables containing the allowed branes with the gauge groups and the massless matter living on them, as well as the tadpole equations for six dimensional Gepner models are available by requested to the authors.
V. Braun and S. Schafer-Nameki, *D-brane charges in Gepner models*, hep-th/0511100
N. Ishibashi, *The boundary and crosscap states in conformal field theories*, Mod. Phys. Lett. A4 (1989) 251; N. Ishibashi, T. Onogi, *Conformal field theories on surfaces with boundaries and cross- caps*, Mod. Phys. Lett. A4 (1989) 161. H. Ooguri, Y. Oz and Z. Yin, *D-branes on Calabi-Yau spaces and their mirrors*, Nucl. Phys. B477 (1996) 407, hep-th/9606112. J. Polchinski, [*Dirichlet-Branes and Ramond-Ramond Charges,*]{} Phys. Rev. Lett. [**75**]{}, 4724 (1995), hep-th/9510017.
E. G. Gimon, J. Polchinski, [*Consistency conditions for orientifolds and D manifolds*]{}, Phys.Rev.[**D54**]{}:1667-1676,1996; hep-th/9601038.
J. Polchinski, S. Chaudhuri and C. V. Johnson, [*Notes on D-branes*]{}, hep-th/9602052.
M. Bianchi and A. Sagnotti,*On The Systematics Of Open String Theories*, Phys. Lett. [**B247**]{} (1990) 517.
B.R. Greene and M.R. Plesser, *Duality In Calabi-Yau Moduli Space*, Nucl. Phys. B338 15 (1990).
I. Brunner and V. Schomerus, *On superpotentials for D-branes in Gepner models*, JHEP [**0010**]{}, 016 (2000), hep-th/0008194. D. Cremades, L. E. Ibanez and F. Marchesano, [*Yukawa couplings in intersecting D-brane models,*]{} JHEP [**0307**]{}, 038 (2003), hep-th/0302105\].
D. Cremades, L. E. Ibanez and F. Marchesano, *Intersecting brane models of particle physics and the Higgs mechanism*, JHEP [**0207**]{}, 022 (2002) \[arXiv:hep-th/0203160\]. F. Gmeiner, R. Blumenhagen, G. Honecker, D. Lust and T. Weigand, [*One in a billion: MSSM-like D-brane statistics,*]{} JHEP [**0601**]{}, 004 (2006), hep-th/0510170\]. P. Di Francesco, P. Mathieu and D. Senechal, “Conformal field theory,” 1997. New York, USA: Springer (1997) 890 p. J. Distler and B. R. Greene, *Some Exact Results On The Superpotential From Calabi-Yau Compactifications*, Nucl. Phys. B [**309**]{}, 295 (1988). G. Aldazabal, L. E. Ibanez and F. Quevedo, *A D-brane alternative to the MSSM,* JHEP [**0002**]{}, 015 (2000), hep-ph/0001083. G. Aldazabal, A. Font, L. E. Ibanez and G. Violero, *D = 4, N = 1, type IIB orientifolds*, Nucl. Phys. B [**536**]{} (1998) 29, hep-th/9804026.
T. Eguchi, H. Ooguri, A. Taormina and S. K. Yang, [*Superconformal Algebras And String Compactification On Manifolds With SU(N) Holonomy,*]{} Nucl. Phys. B [**315**]{}, 193 (1989).
F. G. Marchesano Buznego, [*Intersecting D-brane models*]{}, hep-th/0307252.
N. Arkani-Hamed, S. Dimopoulos and G. R. Dvali, [*“The hierarchy problem and new dimensions at a millimeter,”*]{} Phys. Lett. B [**429**]{}, 263 (1998), hep-ph/9803315. I. Antoniadis, N. Arkani-Hamed, S. Dimopoulos and G. R. Dvali, [*“New dimensions at a millimeter to a Fermi and superstrings at a TeV,”*]{} Phys. Lett. B [**436**]{}, 257 (1998), hep-ph/9804398.
D. Cremades, L. E. Ibanez and F. Marchesano, *Standard model at intersecting D5-Branes: lowering the string scale*, Nucl. Phys. B643 (2002) 93, hep-th/0205074. C. Kokorelis, *Exact standard model structures from intersecting D5-branes*,” Nucl. Phys. B [**677**]{} (2004) 115,hep-th/0207234.
I. Brunner and V. Schomerus, “D-branes at singular curves of Calabi-Yau compactifications,” JHEP [**0004**]{}, 020 (2000), hep-th/0001132.
M. Billo, B. Craps and F. Roose, *Orbifold boundary states from Cardy’s condition*, JHEP [**0101**]{} (2001) 038, hep-th/0011060.
M. Bianchi, *A note on toroidal compactifications of the Type I superstring and other superstring vacuum configurations with sixteen supercharges*,\
Nucl.Phys.[**B528**]{}:73-94,1998, hep-th/9711201.
P. Goddard and D. I. Olive, [*Kac-Moody And Virasoro Algebras In Relation To Quantum Physics*]{}, Int. J. Mod. Phys. A [**1**]{} (1986) 303. E. Witten, [*D-branes and K-theory,*]{} JHEP [**9812**]{}, 019 (1998),hep-th/9810188. A. M. Uranga, [*D-brane probes, RR tadpole cancellation and K-theory charge,*]{} Nucl. Phys. B [**598**]{}, 225 (2001), hep-th/0011048. M. Graña, [*Flux compactifications in string theory: A comprehensive review,*]{} Phys. Rept. [**423**]{}, 91 (2006), hep-th/0509003\]. P. Anastasopoulos, T. P. T. Dijkstra, E. Kiritsis and A. N. Schellekens, [*Orientifolds, hypercharge embeddings and the Standard Model,*]{} arXiv:hep-th/0605226.
[^1]: This is in fact expected. It parallels the inclusion of a 55 sector, besides a 99 brane sector, when even twists are present in $\IZ_{2N}$ orientifold compactifications.
[^2]: In order to obtain the above expressions we have used that $$\begin{aligned}
_B\langle\langle a|q^{\frac12 H_{cl}}|b\rangle\rangle_B&=&\delta_{a,b}\chi^{a}(l)~,\\
_C\langle\langle b |q^{\frac12 H_{cl}}|a\rangle\rangle_B=\delta_{a,b}\chi^{a}({\tilde q})&=&\delta_{a,b}{\hat\chi}^{a}(l+\frac12)~,\end{aligned}$$ where ${\hat
\chi}_a(il+\oh )=e^{i\pi
(h_a-c/24)} \chi_a (it_{M}+\frac{1}{2})$ is real.
[^3]: Recall that two representations labelled by $(l^{\prime},m^{\prime},s^{\prime})$ and $(l,m,s)$ are equivalent if $ l^{\prime}=l$ and $m^{\prime}= m ~{\rm mod} ~ 2(k+2)$ and $s^{\prime}=s ~ {\rm mod}~4$ or $(l^{\prime},m^{\prime},s^{\prime})=(k-l,m+k+2,s+2)$.
[^4]: Although in [@Braun:2005eg] the extra case $(0,1,1,1,1,1,1)$ is suggested to be associated to a different K3 surface we have not explored this possibility. Also the $(0,1,1,1,2,2)$ models correspond to tori surfaces.
[^5]: For explicit expressions for modular matrices $P_{l',l}$ and $S_{l',l}$ see [@aaj; @blumen0310].
[^6]: Which generically will be a product of unitary, orthogonal and symplectic groups.
[^7]: We will write the internal sector in terms of four theories in what follows.
[^8]: They come from the seven antisymmetric and one symmetric representations of $Sp(N_6)$ in Table(\[spectL1L6L10\]) and the vector.
[^9]: The corresponding amplitude when the boundary state is short is essentially the same with a change in the normalization.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We study the longitudinal instabilities of two interpenetrating fluids interacting only through gravity. When one of the constituents is of relatively low density, it is possible to have a band of unstable wave numbers well separated from those involved in the usual Jeans instability. If the initial streaming is large enough, and there is no linear instability, the indefinite sign of the free energy has the possible consequence of explosive interactions between positive and negative energy modes in the nonlinear regime. The effect of dissipation on the negative energy modes is also examined.'
---
epsf [](\hspace{.2cm})
[**NEGATIVE ENERGY MODES AND GRAVITATIONAL**]{}
[**INSTABILITY OF INTERPENETRATING FLUIDS**]{}
A. R. R. CASTI$,^{a}$ P. J. MORRISON$,^{b}$ AND E. A. SPIEGEL$^{c}$
${ }^{a}$[*Department of Applied Physics/*]{}
[*Division of Applied Mathematics*]{}
${ }^{b}$[*Department of Physics and Institute for Fusion Studies*]{}
${ }^{c}$[*Department of Astronomy*]{}
—— P. A. M. Dirac, 1930
BACKGROUND {#sec-gravback}
==========
It used to be thought that the stellar component of two spiral galaxies would pass right through each other in the event of a collision and that only the gaseous components would merge. However, simulations over the past twenty years or so$^{1,2}$ have shown that the macroscopic energy of such a collision is quickly converted into internal energy and that merger of the stellar systems is a common natural outcome of a collision. How is this conversion effected?
An answer to this may lie in the physics of streaming instabilities. In the context of plasma physics, interpenetrating electron beams produce the two-stream instability$^{3}$ whose gravitational analog has long been recognized, beginning with the investigations of Sweet$^{4}$ and Lynden-Bell.$^{5}$ Of course, in the case of galaxy collisions, which occur quickly, the conventional two-stream instability may operate too slowly to be effective. However, it is known that even when two streams of interacting particles are not linearly unstable, they may collectively produce negative energy modes that lead to an explosive nonlinear growth of perturbations for arbitrarily small disturbances. There are well-developed criteria for the occurrence of this explosive growth in plasma physics$^{6}$ and, as Lovelace [*et al.*]{}$^{7}$ have suggested in their analysis of counter-rotating galaxies, we may expect something analogous in the gravitational setting. Our aim here is to briefly develop this topic of negative energy modes for the case of gravitational interaction in the expectation that the phenomena involved will be found significant in a variety of astrophysical processes.
Even in the event of linear instability, the case of counterstreaming populations is significantly different from the standard gravitational instability, which occurs only for perturbation scales greater than the Jeans length.$^{8}$ When there are two interpenetrating fluids such as stars and gas, modes of arbitrary wavelength can be rendered unstable. Numerous investigators have reported on these issues. Most of them (such as Ikeuchi et al.,$^{9}$ Fridman & Polyachenko,$^{10}$ and Araki$^{11}$) focused primarily on the symmetric situation of identical fluids in counterstreaming motion. In that case one finds that the spectrum of any instabilities arising from the relative motion is wholly contained within the Jeans instability band, and this blurs the distinction between the two processes. This need not be true when this symmetry is broken, and indeed not all authors restricted themselves entirely to the symmetric case. The present venture into the asymmetric problem is intended to focus on the possibility of well-separated instability bands, which has not been elucidated in the gravitational context, as far as we are aware.
EQUATIONS OF MOTION {#sec-graveqns}
===================
To see the problem in its simplest version, it is useful to have a uniform medium as the unperturbed state. Rather than formulate the problem inconsistently to achieve this end, as Jeans did, we prefer the Einstein device of introducing a cosmological repulsion term. In the Newtonian setting we readily see how to redefine the gravitational potential so that, instead of introducing such a repulsion term, we fill space with a fluid of [*negative*]{} gravitational mass of density $\rho_{_\Lambda}$. As in the one-fluid plasma model, we treat this density as a constant since its purpose is to allow a gravitationally neutral background state. One may also contemplate the analog of the two-fluid plasma model in which this background antigravitational fluid has its own dynamics, but we do not do that here. The two dynamically active fluids we consider are gravitationally ordinary and polytropic. Thus, the Poisson equation is written here as, \^2 V=4G(\_1+\_2-\_[\_]{}), \[eq:3Dgrav\] where $V$ is the gravitational potential, $\rho_1$ and $\rho_2$ are the source densities of the conventional fluids, and $-\rho_{_\Lambda}$ is the cosmological background density.
The equations of motion for the two fluids ($j=1,2$) are, \_j(\_t+[**u**]{}\_j)[**u**]{}\_j&=& -p\_j-\_jV \[eq:3Dmom\]\
\_t \_j+(\_j[**u**]{}\_j)&=&0, \[eq:3Dmass\] where we do [*not*]{} sum over repeated indices. Each fluid has a sound speed, $c_{j}^{2}=$, and a Jeans wavenumber, $k_{{Jj}}^{2}=$[${\frac{8\pi G{\widehat\rho_j}}{c_{j}^{2}}}$]{}, where the hat signifies the equilibrium value and an uncustomary factor of $2$ appears in the definition of the Jeans wavenumbers.
We shall use natural units with $k_{_{{J1}}}^{-1}$ as the length scale and $(k_{_{{J1}}}c_1)^{-1}$ as the time scale. We further simplify the description by considering only longitudinal motions in one-dimension, so each single-component velocity field may be expressed as the gradient of a velocity potential: $u_j=\partial_x\phi_j$. The fundamental equations (\[eq:3Dgrav\])-(\[eq:3Dmass\]) take the dimensionless form, M\_[1]{}\_t\_1+M\_[1]{}\^[2]{}(\_x\_1)\^2+ +V&=&B\_[\_1]{} \[eq:bern1\]\
cM\_[2]{}\_t\_2+c\^2 M\_[2]{}\^[2]{}(\_x\_2)\^2+ +V&=&B\_[\_2]{} \[eq:bern2\]\
\_t\_1+M\_[1]{}\_x(\_1\_x\_1)&=&0 \[eq:mass1\]\
\_t\_2+cM\_[2]{}\_x(\_2\_x\_2)&=&0 \[eq:mass2\]\
\_[x]{}\^[2]{}V-\_1- \_2+ &=&0,\[eq:grav\] where the $M_{j}=$[${\frac{{\cal U}_j}{c_j}}$]{} are Mach numbers, ${\cal U}_j$ measures the initial streaming velocities, $\beta=$[${\frac{\widehat\rho_2}{{\widehat\rho_1}}}$]{}, $c=$[${\frac{c_2}{c_1}}$]{}, and the Bernoulli constants $B_{_j}$ are chosen to balance the basic state.
LINEAR THEORY {#sec-gravlin}
=============
We perturb from the state of uniform densities and constant velocities by setting $\phi_j=(-1)^{j+1}x+\delta\phi_j$, $\rho_j = 1+ \delta \rho_j$, and $V={\widehat V}+\delta V$. The density terms of the Poisson equation (\[eq:grav\]) combine to vanish and ${\widehat V}$ is a constant. Since the linearized equations are separable we may decompose the perturbations into normal modes proportional to $\exp(i\omega t-ikx)$ to find the dispersion relation, (,k) = 1 ++ 1+\_1+\_2=0. \[eq:diagravic\] The quantity $\Gamma$, which we call the diagravic function by analogy with the dielectric function of electrodynamics, measures the collective response of the fluid to disturbances in the gravitational field and will serve to indicate the energy signature of any normal mode (section \[sec-gravnonlin\]).
For real $k$ the solutions of (\[eq:diagravic\]) with complex $\omega$ correspond to instability; if $\omega$ is real, then solutions of (\[eq:diagravic\]) with complex $k$ can give rise to wave amplification instability. Here we analyze only the case of real $k$. However, we have to deal with both the traditional Jeans instability as well as the two-stream instability, the latter of which involves a sympathetic bunching of particles and is effective for creating instability when the phase speed of the disturbance conspires to create a resonance between different modes.
**Symmetric Case** {#sec-symm}
------------------
If both fluids have the same basic properties ($c=1$ and $\beta=1$), a frame exists in which $M_{_1}=M_{_2}=M$. The dispersion relation then simplifies into a manageable biquadratic with solutions, \_\^[2]{}= -+k\^2(M\^2+1). \[eq:symroots\]
For $M=0$, we recover a simple version of the previously studied two-fluid Jeans problem.$^{10,12,13}$ We find $\omega_{+}^{2}=k^2$, corresponding to sound waves at all $k$, and $\omega_{-}^{2}=k^2-1$, which is the conventional Jeans dispersion relation. The new acoustic modes arise because the aggregate fluid now allows motions unaffected by the gravitational field; for these modes the perturbed gravitational potential is zero.
With relative velocity in the subsonic regime ($0<M<1$), there is only a single unstable mode that branches continuously from the Jeans mode at $M=0$. This mode is unstable for all wavenumbers below a critical value that approaches infinity as $M$ tends to unity from below (see Table 1). To study this limit, we let $M^2=1-\alpha/k^2$ with $0<\alpha<1$. As $k\rightarrow\infty$ we find the approximate solution $\omega_{-}^{2}\sim -$[${\frac{\alpha(1-\alpha)}{4k^2}}$]{}, which reveals a weak instability at large $k$. Thus, weak relative streaming allows gravitational instability at arbitrarily small wavelengths. These large-$k$ instabilities do not arise for Maxwellian velocity distributions within the context of the Vlasov equation.$^{9}$
For supersonic motion ($M>1$) the large-$k$ gravitational instability is no longer present, but a new instability emerges that we call a two-stream instability since it owes its presence to the energy contained in the initial streaming motion. As $M$ ranges from $1$ to $\infty$, the critical wavenumber for instability increases from $k_{{crit}}=1/2$ to $k_{{crit}}=\sqrt{2}/2$.
The upper half of figure \[f:quarticM2\] shows that near $k=0$ the two-stream modes are wholly contained within the Jeans band. This fact coupled with the larger growth rates of the Jeans modes has led some to believe that the two-stream instability is swamped by the Jeans instability and is essentially unimportant.$^{11}$ As $k$ increases the two-stream and Jeans modes collapse upon each other and together bifurcate into growing and damped oscillations. At still larger wavenumbers all motions are stable, propagating waves. The critical wavenumbers below which growth is possible at any Mach number are shown in Table 1.
[**TABLE 1**]{}[^1]
---------------------------------------------------------------------------------------------------------------------------------------------------------
Mach Range Mode Type $k_{{crit}}^{2}$ $\lim_{M\rightarrow 1} $\lim_{M\rightarrow\infty} k_{{crit}}^{2}$
k_{{crit}}^{2}$
-------------- ------------ ------------------------------------------------------- ------------------------ --------------------------------------------
$0\leq M< 1$ Jeans [${\frac{1}{1-M^2}}$]{} $\infty$ Not Applicable
$1\leq M$ Two-Stream [ ${\frac{\sqrt{M^2-1}}{4M(M^2-1+M\sqrt{M^2-1})}}$]{} ${\frac{1}{4}}$ $0$
$1\leq M$ Jeans [ ${\frac{\sqrt{M^2-1}}{4M(1-M^2+M\sqrt{M^2-1})}}$]{} ${\frac{1}{4}}$ ${\frac{1}{2}}$
---------------------------------------------------------------------------------------------------------------------------------------------------------
**Asymmetric Case** {#sec-asymm}
-------------------
When we relax the constraint of identical conditions in the two fluids, one of the more interesting consequences is the possibility of large wavenumber two-stream instability bands well-separated from the Jeans instabilities clustered at small $k$. For illustration we consider the effect of changing the initial relative streaming $M_1+cM_2$ for fixed $\beta$ and $c$. In fact, $\beta$ turns out to be the crucial parameter in achieving the spectral separation; variations in the sound speed ratio $c$ widens both bands together. The distancing of a bubble of two-stream modes from the Jeans band is illustrated in figure \[f:wispyM1.45\].
The large-$k$ two-stream modes can be explained qualitatively by examining the separate pieces of the dispersion relation (\[eq:diagravic\]). Since the streaming instability is related to resonant motions, it is revealing to examine the solutions to $1+\Gamma_1(\omega,k)=
1+\Gamma_2(\omega,k)=0$ in isolation and see where the curves intersect. The frequencies of these non-interacting modes are given by, \_1&=&kM\_1 \[eq:sepmode1\]\
\_2&=&-ckM\_2 \[eq:sepmode2\] When $\omega_1=\omega_2$, the assumed independent frequencies match one another for wavenumbers satisfying, k(M\_1+cM\_2)={
[l]{} (+)\
(-)
. \[eq:rescon\]
In the symmetric case where $M_1=M_2\equiv M$, $c=\beta=1$, we see that two of the resonances are lost except in the irrelevant cases $M=0$ and $k=0$. The other pair of possibilities, $kM=\pm\sqrt{k^2-{\frac{1}{2}}}$, just restate the critical wavenumber condition for what we know to be the modified Jeans instability when $M<1$. In the general case, we can expect another pair of intersections that account for the two-stream bubbles of figure \[f:wispyM1.45\].
NONLINEAR THEORY {#sec-gravnonlin}
================
**Hamiltonian Formulation and Energy Signature** {#sec-ham}
------------------------------------------------
The dynamical equations (\[eq:bern1\])-(\[eq:grav\]) derive from a variational principle and a conserved Hamiltonian functional. The variational formulation has the advantage of shedding light on the relation between the energy content of the disturbances and nonlinear stability. Here we present results for the symmetric case, though the formalism follows through for the asymmetric case as well.
The Hamiltonian and associated equations are, H&=&\_[j=1]{}\^[2]{}\_[0]{}\^[L]{} dx(\_j \_[jx]{}\^[2]{} +\_j[U]{}\_j -V\_[x]{}\^[2]{} ), \[eq:ham\]\
\_t\_j&=&{\_j,H}=-\[eq:bernpoisson\]\
\_t\_j&=&{\_j,H}=,\[eq:masspoisson\] where ${U}_j (\rho_j)=$[${\frac{\rho_{j}^{\gamma-1}}
{\gamma(\gamma-1)}}$]{} is the internal energy for the $j^{\rm th}$ fluid and the Poisson bracket is defined by, {F,G}=\_[j=1]{}\^[2]{}\_[0]{}\^[L]{}dx\^( - ). \[eq:poissonbracket\] For definiteness we have chosen a box-geometry of length $L$.
The energy content of a particular mode when the perturbation amplitude is small is given by the second variation of $H$ evaluated at equilibrium (the free energy). After some calculation this is seen to be, \^2 H=\_[0]{}\^[L]{}dx(M\^2+2M\^2+\_[1]{}\^[2]{} +\_2\^2-2V\_[x]{}\^[2]{}). \[eq:freeenergy\] It may be verified that this functional is conserved by the equations of motion. Suppose we now insert into (\[eq:freeenergy\]) an eigenfunction corresponding to a single stable mode with Im($\omega$)$=0$. Employing overbars to denote eigenvector components and $*$ for complex conjugation, we write, V=e\^[i(t-kx)]{}+[\_j\^\*]{} e\^[-i(\^[\*]{} t-kx)]{}, \[eq:rhoefunk\] and similarly for the other perturbation variables. Upon effecting the integrations, we can make use of the dispersion relation, (,k)=1++ =0, \[eq:nondimdia\] to express the modal free energy in the compact form, \^2 H=-2Lk\^2\^2 . \[eq:freediagravic\]
Wherever [$\omega{\frac{\partial\Gamma}{\partial\omega}}$]{}$<0$ a positive energy mode (PEM) is implied by (\[eq:freediagravic\]), while the condition [$\omega{
\frac{\partial\Gamma}{\partial\omega}}$]{}$>0$ defines a negative energy mode (NEM). This possibility of modes of either signature has been elucidated in the plasma physics literature$^{14}$ and a gravitational analog was suggested by Lovelace et al.$^{7}$ in the context of thin, counter-rotating stellar disks.
Figure \[f:diagravic\_SymmetricCase\] shows the diagravic function for both a subsonic and a supersonic case of stable modes with $k=3$. In the subsonic regime, we find the Hamiltonian to be positive definite near equilibrium since [$\omega{\frac{\partial\Gamma}{\partial\omega}}$]{}$<0$ at every crossing of $\Gamma$ on the $\omega$-axis. Right at the border of supersonic streaming ($M=1$), concomitantly with the appearance of the two-stream instability, the $\Gamma$ curves undergo a topological transition that allows the coexistence of positive and negative energy modes. The NEMs are [*slow modes*]{} in that they have smaller frequencies than their PEM counterparts. This is the typical situation; $\omega$ must pass through zero if the energy signature changes.$^{15}$ We expect from the precedents of plasma physics$^{6}$ that the simultaneous presence of positive and negative energy modes has dramatic consequences on the nonlinear stability of the system.
**Reduction to Action Angle Variables** {#sec-canon}
---------------------------------------
In the rest of this section we will concentrate on the nonlinear interactions between linearly stable modes in the symmetric problem (see figure \[f:diagravic\_SymmetricCase\]). From a physical standpoint, attention is focused on situations where the disturbances are of sufficiently small scale so that the Jeans instability can be ignored, though there are no compelling reasons why this ought to be the case. We will further assume supersonic motion in order to examine the interaction of positive and negative energy modes, a situation we expect to be the most interesting. Under these assumptions the equations of motion achieve their simplest form in action-angle coordinates that we now develop.
First we Fourier transform the field variables: \_j=\_[m=-]{}\^\_m\^[(j)]{}(t)e\^[ik\_m x]{} ,\_[[-m]{}]{}\^[(j)]{}=[\_[[m]{}]{}\^[(j)]{}]{}\^\*,k\_m=.\[eq:fourier\] We can then write the free energy in terms of real variables as, \^2 H=\_[m=1]{}\^([**q**]{}\^[T]{} [**A**]{}[**q**]{}+[**p**]{}\^[T]{}[**B**]{}[**p**]{}), \[eq:freeenergypq\] where ${\bf q}\equiv(q_1,q_2,q_3,q_4)^{T}$, ${\bf p}\equiv(p_1,p_2,p_3,p_4)^{T}$, are linear combinations of the complex modal amplitudes, ${\bf A}$ and ${\bf B}$ are symmetric matrices (given in Casti$^{16}$), and the “$T$” indicates transpose.
Defining the configuration variables ${\bf z}=(q_1,\ldots,p_4)^{T}$, we recast the linearized equations in the form, =[**J**]{}\_[[**z**]{}]{}[\^2 H]{},=(
[clcr]{} 0 & [**B**]{}\
-[**A**]{} & 0
) ,=(
[clcr]{} 0 &[**I**]{}\
-[**I**]{} & 0
), \[eq:Hameqns\] where ${\bf J}$ is the canonical $8\times 8$ cosymplectic form. The next order of business is to construct a symplectic transformation that puts $\delta^2 H$ in its normal form.$^{17,18}$ This can be achieved by writing ${\bf z}={\bf S}{\bf Z}$, where the matrix ${\bf S}$ consists of suitably ordered eigenvectors of ${\bf L}$ satisfying the symplectic condition, ${\bf S}^{T}{\bf J}{\bf S}={\bf J}$. After a final transformation to action-angle coordinates, the free energy expression (\[eq:freeenergypq\]) becomes a superposition of harmonic oscillators, \^2 H=\_[m=1]{}\^(\_+ J\_1+\_+ J\_2 -\_- J\_3-\_- J\_4). \[eq:freeaa\] The free energy is thus manifestly composed of two pairs each of positive and negative energy modes.
**Three-Wave Resonance and Explosive Growth** {#sec-3wave}
---------------------------------------------
Energy conservation forbids nonlinear runaway growth if $H$ is definite (Dirichlet’s theorem), as is the case here for subsonic motion. When the relative streaming is supersonic, however, interacting PEMs and NEMs can circumvent this restriction since they contribute energy of opposite sign.
We demonstrate the possibility of explosive growth with a three-wave resonant interaction between two NEMs and one PEM. Since the energy signature of a mode is not Galilean invariant, the existence of a reference frame in which all three modes have the same signature implies nonlinear stability. It may be shown that there is no reference frame in which all three modes have the same energy signature if the highest frequency wave has opposite signature to that of the other two.$^{19}$ This provides a criterion for three-wave interactions leading to instability.
The third-order resonance conditions for a triplet of modes are, m\_[\_1]{}k\_[\_1]{}+m\_[\_2]{}k\_[\_2]{}+m\_[\_3]{}k\_[\_3]{}&=&0\
m\_[\_1]{}\_[\_1]{}m\_[\_2]{}\_[\_2]{}m\_[\_3]{}\_[\_3]{}&=&0\[eq:rescon3\]\
m\_[\_1]{}+m\_[\_2]{}+m\_[\_3]{}&=&3(m\_[\_1]{},m\_[\_2]{},m\_[\_3]{}).which here may be satisfied by $(m_{_1},m_{_2},m_{_3})=(1,1,-1)$, $(k_{_1},k_{_2},k_{_3})=(k_m,k_m,2k_m)$, and $(\omega_{_1},\omega_{_2},\omega_{_3})=(\omega_+,\omega_-,\omega_-)$, where the $\omega_{_j}$ are taken to be positive. One can see from figure \[f:quarticM2\] that $\omega_{_1}>\omega_{_2},\omega_{_3}$, so the relative signatures of this triplet are immune to a Galilean shift. Note from figure \[f:quarticM2\] that a resonant triplet involving two PEMs and one NEM would not have robust relative signatures under a frame shift since the PEMs have larger frequencies.
The lowest order nonlinear terms come from the third variation of $H$ expanded about the dynamical equilibrium, \^3 H=\_[j=1]{}\^[2]{}\_[0]{}\^[L]{} dx(\_[\_j]{}\_[\_[jx]{}]{}\^[2]{}+ \_[\_j]{}\^[3]{}). \[eq:hamnon\] In terms of the Fourier amplitudes this expression is, \^3 H=\_[j=1]{}\^[2]{}\_[[m,n=1]{}]{}\^ \[-Mk\_[\_m]{}k\_[\_n]{}(\_[\_[m+n]{}]{}\^[(j)]{}\_[\_[-m]{}]{}\^[(j)]{} \_[\_[-n]{}]{}\^[(j)]{}-\_[\_[m-n]{}]{}\^[(j)]{}\_[\_[-m]{}]{}\^[(j)]{} \_[\_n]{}\^[(j)]{}+[c.c.]{})\[eq:fourierham3\]\
+(\_[\_m]{}\^[(j)]{}\_[\_n]{}\^[(j)]{} \_[\_[-m-n]{}]{}\^[(j)]{}+\_[\_m]{}\^[(j)]{}\_[\_[-n]{}]{}\^[(j)]{} \_[\_[n-m]{}]{}\^[(j)]{}+[c.c.]{})\].We then effect the same transformations on (\[eq:fourierham3\]) that led to the diagonalized free energy (\[eq:freeaa\]). This spawns a myriad of nonlinear terms, only some of which survive an averaging process that leads to the Birkhoff normal form.$^{20}$
For a three-wave resonance, one finds after near-identity transformations that the only higher order terms contributing to the normal form are of the type,$^{14}$ (3)[Terms]{}\~J\_[\_1]{}\^[|l|/2]{} J\_[\_2]{}\^[|m|/2]{} J\_[\_3]{}\^[|n|/2]{} ( l\_[\_1]{}+m\_[\_2]{}+n\_[\_3]{}), \[eq:cubicterm\] with $|l|+|m|+|n|=3$. The Hamiltonian up to $3^{\rm{rd}}$ order terms for the resonant NEM/PEM triplets can be written as, H=\_1 J\_[\_1]{}-\_2 J\_[\_2]{}-\_3 J\_[\_3]{} +(\_[\_1]{}+\_[\_2]{}+\_[\_3]{}),\[eq:3waveham\] where $\alpha=\alpha(k_m,M,L)$ is a nonlinear coupling constant that is neither especially large or small in the parameter regime of the three-wave resonance considered here.
Since the angles $(\theta_{_1},\theta_{_2},\theta_{_3})$ appear in only one combination in $H$, further simplification of (\[eq:3waveham\]) is possible via the generating function, F\_[\_2]{}([**I**]{},[****]{})=I\_[\_1]{}( \_[\_1]{}+\_[\_2]{})+I\_[\_2]{}\_[\_2]{}+I\_[\_3]{}\_[\_3]{}, \[eq:genfunk\] with ${\psi_{_j}}=$ and ${J_{_j}}=$. This canonical transformation yields, H=[\_[\_1]{}]{}I\_[\_1]{}-\_[\_3]{}I\_[\_3]{}+ (2\_[\_1]{}+\_[\_3]{}), \[eq:3waveham2\] with $2{\tilde\omega_{_1}}\equiv\omega_{_1}-\omega_{_2}$. If one chooses initial conditions satisfying $I_{_2}\equiv J_{_2}-J_{_1}=0$, then $H$ is identical to the normal form of a two-wave interaction originally presented by Cherry.$^{21,22}$ In terms of the $({\bf q},{\bf p})$ variables, Cherry’s Hamiltonian is H=[\_[\_1]{}]{}(p\_[\_1]{}\^2+q\_[\_1]{}\^2)- [\_[\_3]{}]{}(p\_[\_3]{}\^2+q\_[\_3]{}\^2)+ (2q\_[\_1]{} p\_[\_1]{} p\_[\_3]{}-q\_[\_3]{}), \[eq:cherry\] where $\epsilon=$[${\frac{\sqrt{2}\alpha}{4}}$]{}.
The dynamical system generated by the Cherry Hamiltonian is integrable. In the special case of a third-order resonance with $\omega_{_3}=2{\tilde\omega_{_1}}$, there exists a family of two-parameter solutions, q\_[\_1]{}=([\_[\_1]{}]{}t+) &,&p\_[\_1]{}=-([\_[\_1]{}]{}t+) \[eq:rescherrysolns\]\
q\_[\_3]{}=-(2[\_[\_1]{}]{}t+2) &,&p\_[\_3]{}=-(2[\_[\_1]{}]{}t+2),where $\xi$ and $\eta$ are constants depending on the initial conditions.
The solutions (\[eq:rescherrysolns\]) show the possibility of [*finite-time density singularities*]{} when two negative energy modes interact resonantly with a positive energy mode. A system exhibiting this behavior is said to undergo [*explosive growth*]{}, and it could be an important mechanism for structure formation in galactic and cosmological settings when relative motion between different fluid species is involved. If the resonance is detuned, separatrices bounding stable orbits emerge in phase space, but the dynamics are still prone to finite-amplitude instability.
DISSIPATIVE INSTABILITY {#sec-dissipation}
=======================
We close our investigation of the consequences of negative energy modes by examining the effects of dissipation on the linear stability of the system. With negative energy modes propagating through a dissipative medium, we may expect new instabilities since the damping can pump more negative energy into the wave. This somewhat counterintuitive effect of frictional forces in other contexts was first pointed out by Kelvin and Tait$^{23}$ (see also Zajac$^{24}$).
Suppose that collisions are important at some stage in the development of a gravitationally bound structure. A simple model of this effect incorporates a dynamical friction term $(-1)^j\nu\left({\bf u}_1-{\bf u}_2\right)$ on the right hand side of the momentum equations (\[eq:3Dmom\]), where $\nu$ is a positive damping coefficient.
In the dimensionless symmetric case the dispersion relation becomes, \^4-2i\^3 +\^2 &+&2i\[eq:dampeddis\]\
& &+k\^2(M\^2-1)=0. If we assume the damping is weak, $\nu\ll 1$, we may develop (\[eq:dampeddis\]) in a regular perturbation series ($\omega=\omega_{_0}+\nu\omega_{_1}+\ldots$) to find the lowest order corrections to the frequencies (\[eq:symroots\]), \_[\_1]{}\^=(1), \[eq:freqcorrection\] where we assume $k^2\gg{\frac{M\pm\sqrt{M^2-1}}{4M}}$ to avoid the singularity accompanying the vanishing denominator in (\[eq:freqcorrection\]) (see Casti$^{16}$ for the details). If we assume $M>1$ so that negative energy modes are present, then a close examination of the corrections reveals that the dissipation promotes instability in the wavenumber band ${\frac{M\pm\sqrt{M^2-1}}{4M}}\ll k<{\frac{\sqrt{2}}{2}}$ for any $M\ge1$ [*no matter how weak the damping*]{}. Since the instability as $k^2\rightarrow{\frac{1}{2}}$ is realized only in the $M\rightarrow\infty$ limit of the undamped problem, we see that the dissipation indeed has the effect of destabilizing modes that were stable in the conservative case. A numerical investigation revealed that this result holds for any $\nu>0$.
The modal bands destabilized by the damping become more significant in the asymmetric case. As remarked in section \[sec-asymm\], bubbles of unstable two-stream modes can pinch off from the Jeans-unstable bubble and result in well-separated instability bands. The inclusion of dissipation can destabilize the entire band of modes separating the bubbles, as well as some higher-$k$ modes beyond the undamped two-stream bubble. This is illustrated in figure \[f:damwispyM1.45\].
One should not assume that any form of dissipation will destabilize negative energy modes. For instance, if each fluid feels only a drag proportional to its own velocity, there are no new instabilities even with relative motion. In other words, the dissipation must in some sense project onto the eigenspace spanned by the negative energy modes in a way that decreases their energies. This depends not only on the nature of the dissipation, but also upon the initial equilibrium about which one perturbs.
The effect of damping can be understood by examining the time evolution of the free energy. If the dissipation acts to increase the energy of a positive energy mode or decrease the energy of a negative energy mode, then one can show that pure imaginary eigenvalues take on a positive real part.$^{25}$ To see that this is possible here, consider the temporal change in the free energy, which in the symmetric case can be written, =-M\_[0]{}\^[L]{} dx( \_[[1x]{}]{}-\_[[2x]{}]{})\^2 -M\_[0]{}\^[L]{} dx(\_1-\_2) (\_[[1x]{}]{}+\_[[2x]{}]{}). \[eq:hdot2\] Since the first term of ${\delta^2{\dot H}}$ is negative definite, the conditions for which the free energy decays or grows in time is determined by the relative phasings of the velocity and density perturbations comprising the second term. One may deduce the effect of the dissipation on any particular mode of the conservative problem by inserting the undamped modes into (\[eq:hdot2\]), which yields a formula for ${\delta^2{\dot H}}$ valid up to ${\cal O}(\nu^2)$. For subsonic relative motion, $M<1$, the expression (\[eq:hdot2\]) is negative definite and the damping lives up to its name and causes the PEMs to decay in time. When $M>1$, ${\delta^2{\dot H}}$ can be either positive or negative for an NEM depending on the value of $k$, which explains why some NEMs are destabilized and others are damped in the usual sense.
DISCUSSION {#sec-discussion}
==========
The formation of structures through the action of gravity is much analyzed in cosmology, galactic structure and cosmogony. Most of this analysis is centered on the operation of gravitational instability, though streaming fluids can resonantly interact via the gravitational field to cause linear instability in spectral ranges inaccessible to the traditional Jeans instability. As we have brought out here, the distinction between the two types of unstable modes, the Jeans and the two-stream, becomes sharper when one constituent is far denser than the other. Much of the previous work on the subject failed to take advantage of this crucial feature by focusing attention on situations where each component exists in equal abundance. Even when the two-stream instability does not occur, if the total energy of the gravitational two-stream interaction is indefinite, the positive and negative energy modes that are [*stable*]{} in the linear theory can interact to produce explosive development of disturbances of arbitrarily small amplitude. This can be a significant aspect of the theory of structure formation.
There are many clear instances where the dynamics of interpenetrating fluids may play a role in developing structures, but we close here by suggesting that even when the streaming is not apparent, two-stream dynamics may be relevant. An interesting example is provided by the coexistence of dark matter and luminous (baryonic) matter that is generally believed to occur throughout the cosmos. The locations of the two kinds of matter seem to be well correlated, which would not be the case if they were now streaming through each other. On the other hand, it might be reasonable to ask why there is this apparent correlation (or anticorrelation in the case of negative gravitational density) of the two kinds of material. Even if they had once been in relative motion this situation would not long persist, as we have seen. But the outcome, as far as large-scale structure is concerned, could be quite different if the kinematic history of the interaction of the two matters had been richer than has been supposed hitherto. Given the indefiniteness of the free energy if the initial streaming is large enough, waves of short length scale could have interacted in an explosive manner to quickly produce highly nonlinear density fluctuations. This is a feature of gravitational structure development that could be profitably studied, particularly in situations where the background Hubble expansion cannot be ignored. The dynamics of a two-fluid system with initially [*time-dependent*]{} relative motion is currently under investigation.
[**REFERENCES**]{}
1\. Theys, J. C. & E. A. Spiegel. 1977. Ap. J. [**212:**]{} 616.
2\. Barnes, J. E. 1992. Ap. J. [**393:**]{} 484.
3\. Buneman, O. 1959. Phys. Rev. [**115:**]{} 503.
4\. Sweet, P. A. 1963. Monthly Notices Roy. Astron. Soc. [**125:**]{} 285.
5\. Lynden-Bell, D. 1967, in Relativity Theory and Astrophysics 2. Galactic Structure.
J. Ehlers, ed. Providence: American Math. Soc.
6\. Davidson, R. C. 1972. Methods in Nonlinear Plasma Theory. Academic Press,
New York.
7\. Lovelace, R. V. E., K. P. Jore & M. P. Haynes. 1997. Ap. J. [**475:**]{} 83.
8\. Jeans, J. 1902. Phil. Trans. R. Soc. [**199A:**]{} 49.
9\. Ikeuchi, S., T. Nakamura & F. Takahara. 1974. Prog. Theor. Phys. [**52:**]{} 1807.
10\. Fridman, A. M. & V. L. Polyachenko. 1984. Physics of Gravitating Systems II.
Springer-Verlag, New York.
11\. Araki, S. 1987. Astron. J. [**94:**]{} 99.
12\. Spiegel, E. A. 1972, in Symposium on the Origin of the Solar System: 165. H. Reeves,
ed. Editions du Centre National de la Recherche Scientifique, Paris.
13\. de Carvalho, J. P. M. & P. G. Macedo. 1995. Astron. Astrophys. [**299:**]{} 326.
14\. Kueny, C. S. & P. J. Morrison. 1995. Phys. Plasmas. [**2:**]{} 1926.
15\. Morrison, P. J. 1998. Rev. Mod. Phys. [**70:**]{} 467.
16\. Casti, A. R. R. 1998. Ph. D. Thesis. Columbia University.
17\. Moser, J. K. 1958. Comm. Pure Appl. Math. [**11:**]{} 81.
18\. Meyer, K. R. & G. R. Hall. 1991. Introduction to Hamiltonian Dynamical Systems
and the N-Body Problem. Springer-Verlag New York Inc.
19\. Weiland, J. & H. Wilhelmsson. 1977. Coherent Nonlinear Interaction of Waves in
Plasmas. Pergamon, Oxford.
20\. Ozorio de Almeida, A. M. 1988. Hamiltonian Systems: Chaos and Quantization.
Cambridge University Press, Cambridge.
21\. Cherry, T. M. 1925. Trans. Cambridge Philos. Soc. [**23:**]{} 199.
22\. Whittaker, E. T. 1937. Analytical Dynamics: 142. Cambridge, London.
23\. Thompson, W. (Lord Kelvin) & P. G. Tait. 1921. Treatise on Natural Philosophy,
Part I:388. Cambridge University Press, Cambridge.
24\. Zajac, E. E. 1964. Journal of the Astronautical Sciences [ **11:**]{} 46.
25\. Mackay, R. S. 1991. Phys. Lett. A. [**155:**]{} 266.
=4.5in
=4.5in
=4.5in
=4.5in
[^1]: The value of $k_{{crit}}^{2}$ for the two-stream modes with $M\geq 1$ more accurately refers to the $k$ value at which the unstable two-stream and Jeans branches merge.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We lift the splicing formula of Némethi and Veys, which deals with polynomials in two variables, to the motivic level. After defining the motivic zeta function and the monodromic motivic zeta function with respect to a differential form, we prove a splicing formula for them, which specializes to this formula of Némethi and Veys. We also show that we cannot introduce a monodromic motivic zeta functions in terms of a (splice) diagram since it does not contain all the necessary information. In the last part we discuss the generalized monodromy conjecture of Némethi and Veys. The statement also holds for motivic zeta functions but it turns out that the analogous statement for monodromic motivic zeta functions is not correct. We show some examples illustrating this.'
address: 'KU Leuven, Dept. Wiskunde, Celestijnenlaan 200B, 3001 Leuven, Belgium'
author:
- Thomas Cauwbergs
bibliography:
- 'ref.bib'
title: Splicing for motivic zeta functions
---
Introduction {#introduction .unnumbered}
============
We will consider here a polynomial $f\in {\ensuremath{\mathbb{C}}}[x,y]$, which has at most a singularity at the origin.
Splicing {#splicing .unnumbered}
--------
The topology of a plane curve singularity $\{f = 0\}$ at the origin is closely related to its link $\{(z_1,z_2) \in {\ensuremath{\mathbb{C}}}^2 \mid f(z_1,z_2) = 0, \abs{z_1}^2 + \abs{z_2}^2 = \varepsilon\}$, where $\varepsilon$ is sufficiently small. This link can be studied by looking at its splice diagram.
Splicing itself is originally a technique from link theory which constructs a new link out of two given links. We can also use splicing to decompose complicated links into easier links. This decomposition procedure has a nice description for our case of plane curve singularities. Fix an embedded resolution of singularities $\pi: X\to {\ensuremath{\mathbb{A}^{2}_{{\ensuremath{\mathbb{C}}}}}}$ of $f^{-1}(0)$. The splice diagram $\Gamma$ associated to $f$ and $\pi$ is then constructed by taking the dual graph of the exceptional curves of $\pi$, removing nodes if necessary, where these nodes correspond to exceptional curves, and adding decorations to the edges.
The splicing of a splice diagram $\Gamma$ along an edge $e$ produces two new splice diagrams $\Gamma_R$ and $\Gamma_L$. As shown in Figure \[fig:splicingintro\], it divides $\Gamma$ into two pieces and then makes them again into a splice diagram by adding appropriate multiplicities $M$ and $M'$. Splicing the links of $\Gamma_L$ and $\Gamma_R$ together, we obtain the link of $\Gamma$ (see [@EisenbudNeumann] for more details).
[cc]{}\
[0.3]{}
(sup) at (0,0); (dOne) at (-[2]{},[.75]{}); (dN) at (-[2]{},-[.75]{}) ; (daOne) at ([2]{},0) ; (vL) at (-[0]{},0) ; (vR) at ([0]{},0) ; (dOne)–(vL) node \[midway, above \] [$d_1$]{}; (dN)–(vL) node \[midway, above \] [$d_n$]{}; ($(vL)+(sup)$)–($(daOne) + (sup)$) node \[ near start, above \] [$d$]{} node \[very near end, above\] [$(M)$]{}; node at ($ 1/2*(dOne) + 1/2*(dN) +(1/3,0) $) ; (vL) circle \[radius=3pt\];
&
[0.3]{}
(sup) at (0,0); (dOne) at (-[2]{},0); (daOne) at ([2]{},[.75]{}) ; (daN) at ([2]{},-[.75]{}) ; (vL) at (-[0]{},0) ; (vR) at ([0]{},0) ; ($(dOne)+(sup)$)–($(vR) + (sup)$)node \[near end, above \] [$d'$]{} node \[very near start, above\] [$(M')$]{}; (vR)–(daOne) node \[midway, above \] [$d'_1$]{}; (vR)–(daN) node \[midway, above \] [$d'_{n'}$]{}; node at ($ 1/2*(daOne) + 1/2*(daN) +(-1/3,0) $) ; (vR) circle \[radius=3pt\];
Némethi and Veys in [@NemethiVeys2012] applied this splicing technique to the topological zeta function.
Topological zeta functions {#topological-zeta-functions .unnumbered}
--------------------------
Denef and Loeser introduced the topological zeta function in [@DenefLoeser92] as follows: let $E_j, j\in J$, be the irreducible components of $\pi^{-1}( f^{-1}(0) )$, $N_j$ the multiplicity of $\pi^*f$ along $E_j$ in and $\nu_j-1 $ the multiplicity of $\pi^*(dx\wedge dy)$ along $E_j$. The (local) topological zeta function is then $${Z_{f}^{\text{top}}(s)} = \sum_{\emptyset\neq I \subseteq J}{\ensuremath{\chi_{\text{top}}}}({\ensuremath{E_I^\circ}}\cap \pi^{-1}(0)) \prod_{i\in I}
\frac{1}{N_is +\nu_i} \in {\ensuremath{\mathbb{Q}}}(s), \label{eq:definitiontopzetafintro}$$ where ${\ensuremath{E_I^\circ}}= \cap_{i\in I} E_i \setminus\left( \cup_{i \in J\setminus I} E_i \right)$. Using the existence of a minimal resolution, it is obvious that that this is independent of the chosen resolution. In general dimension the independence was originally proven by a limit argument using the $p$-adic zeta functions $\int_{p{{\ensuremath{\mathbb{Z}}}}_p^2} \abs{f}^s_p \abs{dx}$, where we assumed that $f\in {\ensuremath{\mathbb{Q}}}[x,y]$. This can also be shown by using the weak factorization theorem [@MR2013783][@MR1896232] or by considering the topological zeta function as a specialization of the motivic zeta function, which we will discuss further on.
To obtain a splicing formula, Némethi and Veys incorporated a differential form $\omega$ into the splice diagram and defined a topological zeta function ${Z_{f,\omega}^{\text{top}}(s)} = {Z_{\Gamma}^{\text{top}}(s)}$ for such splice diagrams (and thus with respect to $\omega$) by using , where we redefine $\nu_i$ in terms of $\omega$. This corresponds to considering the $p$-adic zeta functions $\int_{p{{\ensuremath{\mathbb{Z}}}}_p^2} \abs{f}^s_p \abs{\omega}$, given $f\in {\ensuremath{\mathbb{Q}}}[x,y]$. These zeta functions with respect to a differential form were already introduced in [@ACLM1], [@ACLM2] and [@Veyszetafunctionsonsingularvarieties] with a restriction on the support of $\omega$. In [@VeysMEAndzetafunctionswithdifferential] and [@NemethiVeys2012] there is no restriction on the support but only the topological zeta function is considered.
Splicing formula and generalized monodromy conjecture {#splicing-formula-and-generalized-monodromy-conjecture .unnumbered}
-----------------------------------------------------
Némethi and Veys showed then that there is a nice splicing formula connecting the involved diagrams in Figure \[fig:splicingintro\] and their topological zeta functions: $${Z_{\Gamma}^{\text{top}}(s)} = {Z_{\Gamma_L}^{\text{top}}(s)} + {Z_{\Gamma_R}^{\text{top}}(s)} -
\frac{1}{ (Ms + i)( M's +i' )} \label{eq:splicingformula:intro}.$$ These $i$ and $i'$ are also introduced when we splice $\Gamma$ and the involved zeta functions are with respect to some differential form, whose information is contained in the diagram. They used this to prove the generalized monodromy conjecture, which predicts the existence of a class of ‘allowed’ differential forms such that the following holds:
- for every allowed form $\omega$ and every pole $s_0$ of ${Z_{f,\omega}^{\text{top}}(s)}$, $\exp(2\pi i s_0)$ is a local monodromy eigenvalue of $f$;
- $dx\wedge dy$ is allowed;
- every local monodromy eigenvalue of $f$ is obtained as a pole of ${Z_{f,\omega}^{\text{top}}(s)}$ for some allowed $\omega$.
Motivic zeta functions {#motivic-zeta-functions .unnumbered}
----------------------
We consider here the Grothendieck ring of varieties ${\ensuremath{K_0(\operatorname{Var}_{{{\ensuremath{\mathbb{C}}}}})}}$, its localization ${\ensuremath{\mathcal{M}_{{\ensuremath{\mathbb{C}}}}}}$ with respect to ${\ensuremath{\mathbb{L}}}= {\ensuremath{ [{\ensuremath{\mathbb{A}^{1}_{{\ensuremath{\mathbb{C}}}}}}]}}$ and the completion ${\ensuremath{\hat{\mathcal{M}}_{{\ensuremath{\mathbb{C}}}}}}$. We denote by ${\ensuremath{ [X]}}$ the class of the variety $X$. Also ${\ensuremath{\mathcal{L}_{n} (X)}}$ will be the scheme of $n$-jets of a variety $X$ and ${\ensuremath{\mathcal{L} (X)}}$ will be the arc space of $X$. All of this will be introduced in more detail in Section \[section:motivicrings\]. As mentioned before, the topological zeta function can also be considered as an avatar of the (local) motivic zeta function ${Z_{f}(T)} \in {\ensuremath{\mathcal{M}_{{\ensuremath{\mathbb{C}}}}}}[[T]]$. This motivic zeta function is defined by $
{Z_{f}(T)} := \sum_{n> 0} {\ensuremath{ [{\ensuremath{\mathfrak{X}_{n}}}]}} {\ensuremath{\mathbb{L}}}^{-dn } T^n \in {\ensuremath{\mathcal{M}_{{\ensuremath{\mathbb{C}}}}}}[[T]]
$ where $ {\ensuremath{\mathfrak{X}_{n}}} := \{\varphi \in {\ensuremath{\mathcal{L}_{n} ({\ensuremath{\mathbb{A}^{2}_{{\ensuremath{\mathbb{C}}}}}})}} \mid \operatorname{ord}_t f_n(\varphi) = n, \ \pi^n_0(\varphi) = 0\}$. There is also an explicit formula for ${Z_{f}(T)}$ in terms of a log resolution similar to the one in .
We introduce here a motivic zeta function with respect to a differential form. We do this by observing that $ {\ensuremath{ [{\ensuremath{\mathfrak{X}_{n}}}]}} {\ensuremath{\mathbb{L}}}^{-dn }$ is the (naive) measure of a cylinder ${\ensuremath{\mathcal{Z}_{n}}}$ in the arc space ${\ensuremath{\mathcal{L} ({\ensuremath{\mathbb{A}^{2}_{{\ensuremath{\mathbb{C}}}}}})}}$. After defining a new motivic measure $\mu_\omega$ with values in ${\ensuremath{\hat{\mathcal{M}}_{{\ensuremath{\mathbb{C}}}}}}$, we obtain a motivic zeta function ${Z_{f, \omega}(T)} :=
\sum_{n > 0} \mu_\omega( {\ensuremath{\mathcal{Z}_{n}}}) T^n \in {\ensuremath{\hat{\mathcal{M}}_{{\ensuremath{\mathbb{C}}}}}}[[T]]$. For this motivic zeta function ${Z_{f, \omega}(T)}$ there exists again a formula: $${Z_{f, \omega}(T)} = \sum_{\emptyset\neq I \subset J} ({\ensuremath{\mathbb{L}}}- 1)^{\lvert I\rvert }
{\ensuremath{ [{\ensuremath{E_I^\circ}}\cap \pi^{-1}(0)]}} \prod_{i\in I} \frac{T^{N_i}}{{\ensuremath{\mathbb{L}}}^{v_i} - T^{N_i}} \in
{\ensuremath{\hat{\mathcal{M}}_{{\ensuremath{\mathbb{C}}}}}}[[T]] \label{eq:defmotivic:intro} .$$
We want to extend the definition of the motivic zeta function to the case of splice diagrams, but it turns out to be a bit hard. To ease our notation and to simplify our proofs, we introduce the notion of a [*diagram*]{}: this is actually the same as a splice diagram except we can choose how many nodes (of valency two) we remove (instead of all of them). A diagram where no nodes were removed is called realizable and refining a diagram is adding nodes again to the diagram to obtain a new diagram. The definition in for realizable diagrams then extends to all diagrams by choosing any refinement.
This leads us to our theorem.
Consider a diagram $\Gamma$ and the splicing of $\Gamma$ into $\Gamma_L$ and $\Gamma_R$. Then we have $${Z_{\Gamma}(T)} = {Z_{\Gamma_L}(T)} + {Z_{\Gamma_R}(T)} - \frac{({\ensuremath{\mathbb{L}}}- 1)^2 T^{M + M'}}{ ({\ensuremath{\mathbb{L}}}^i - T^M)({\ensuremath{\mathbb{L}}}^{i'} - T^{M'})}.$$
This formula specializes to .
Splice diagrams and monodromic motivic zeta functions {#splice-diagrams-and-monodromic-motivic-zeta-functions .unnumbered}
-----------------------------------------------------
Consider ${\ensuremath{\hat{\mu }}}= {\ensuremath{\varprojlim}}_n\mu_n$, where $\mu_n$ is the group of $n$-th roots of unity and the localized monodromic Grothendieck ring ${{\ensuremath{\mathcal{M}^{{\ensuremath{\hat{\mu }}}}_{{\ensuremath{\mathbb{C}}}}}}}$. This ${{\ensuremath{\mathcal{M}^{{\ensuremath{\hat{\mu }}}}_{{\ensuremath{\mathbb{C}}}}}}}$ will be defined in more detail in Section \[section:motivicrings\]. The next step is to add a ${\ensuremath{\hat{\mu }}}$-action to the motivic zeta function, which is done by considering the monodromic motivic zeta function ${{\ensuremath{Z^{{\ensuremath{\hat{\mu }}}}_{f}(T)}}} :=
\sum_{n > 0} {\ensuremath{ [{\ensuremath{\mathfrak{X}_{n,1}}}]}} {\ensuremath{\mathbb{L}}}^{-dn} T^n \in
{{\ensuremath{\mathcal{M}^{{\ensuremath{\hat{\mu }}}}_{{\ensuremath{\mathbb{C}}}}}}}[[T]]$, where $ {\ensuremath{\mathfrak{X}_{n,1}}} := \{\varphi \in {\ensuremath{\mathcal{L}_{n} ({\ensuremath{\mathbb{A}^{2}_{{\ensuremath{\mathbb{C}}}}}})}} \mid
f_n(\varphi) \equiv t^n \mod(t^{n+1}), \pi^n_0(\varphi) = 0\}$. The action of $\mu_n$ on ${\ensuremath{\mathfrak{X}_{n,1}}}$ is defined by $a \cdot \varphi(t) = \varphi(at)$, where $a\in \mu_n $ and $\varphi \in {\ensuremath{\mathcal{L}_{m} ({\ensuremath{\mathbb{A}^{n}_{{\ensuremath{\mathbb{C}}}}}})}}$, which induces an action of ${\ensuremath{\hat{\mu }}}$.
In section \[section:motivicrings\] we define ${{\ensuremath{Z^{{\ensuremath{\hat{\mu }}}}_{f,\omega}(T)}}}$, which is the monodromic motivic zeta function with respect to a differential form $\omega$. It turns out that we cannot define ${{\ensuremath{Z^{{\ensuremath{\hat{\mu }}}}_{f,\omega}(T)}}}$ in terms of a (splice) diagram: we will show that there exist $\lambda, \lambda' \in {\ensuremath{\mathbb{C}}}\setminus\{0,1\}$ such that $${{\ensuremath{Z^{{\ensuremath{\hat{\mu }}}}_{f_\lambda}(T)}}} \neq {{\ensuremath{Z^{{\ensuremath{\hat{\mu }}}}_{f_{\lambda'}}(T)}}},$$ where $
f_\lambda = xy^2(x-y)(x-\lambda y) \in {\ensuremath{\mathbb{C}}}[x,y]$. For this $\lambda$ and $\lambda'$, we find that the associated splice diagrams are the same but the monodromic motivic zeta functions are not. Hence we cannot define ${{\ensuremath{Z^{{\ensuremath{\hat{\mu }}}}_{f,\omega}(T)}}}$ in terms of a diagram.
To construct such $\lambda$ and $\lambda'$ we use the Picard morphism constructed by Ekedahl in [@Ekedahl2009]. In the appendix we prove and explain the details of this construction since [@Ekedahl2009] was never published.
Generalized monodromy conjecture for motivic zeta functions {#generalized-monodromy-conjecture-for-motivic-zeta-functions .unnumbered}
-----------------------------------------------------------
The straightforward generalization to the motivic zeta functions turns out to be true. We will specify what we mean by a pole in Section \[section:mc\].
Consider the set of allowed forms for a diagram $\Gamma$ of $f \in {\ensuremath{\mathbb{C}}}[x,y]$. It satisfies the following conditions:
- for every allowed form $\omega$, every pole of $ {Z_{f, \omega}(T)}$ induces a monodromy eigenvalue. More specifically Theorem \[thm:generalizedmotivicmonodromyconjecture\] holds.
- $dx\wedge dy$ is allowed;
- every monodromy eigenvalue is obtained as a pole of the motivic zeta function of $f$ with respect to $\omega$.
However in the case of the monodromic motivic zeta function we show that this does not hold. First we give an easy example where an allowed form gives a non-desired pole of the twisted topological zeta function. The twisted topological zeta function will also be introduced in Section \[section:motivicrings\]. This means that it must occur as a pole of the monodromic motivic zeta function and thus the generalized monodromy conjecture cannot be valid. Secondly we produce an example which shows that a subset of allowed forms is not sufficient: there exists no allowed differential form such that all the poles of the monodromic motivic zeta function induce monodromy eigenvalues and such that a particular monodromy eigenvalue is obtained.
I am very grateful to Wim Veys and Johannes Nicaise for their valuable suggestions.
Motivic zeta functions and Grothendieck rings\[section:motivicrings\]
=====================================================================
In this section we introduce the necessary Grothendieck rings and zeta functions. A variety will be a complex algebraic variety, i.e. a reduced separated scheme of finite type over ${\ensuremath{\mathbb{C}}}$. The associated category will be denoted by $\operatorname{Var}_{\ensuremath{\mathbb{C}}}$. Also fix a polynomial $f\in {\ensuremath{\mathbb{C}}}[x_1,\ldots, x_d]$, which we will also consider as a morphism ${\ensuremath{\mathbb{A}^{n}_{{\ensuremath{\mathbb{C}}}}}} \to {\ensuremath{\mathbb{A}^{1}_{{\ensuremath{\mathbb{C}}}}}}$. See [@MR1923998] and [@denefloeser2001] for more background information.
Grothendieck rings
------------------
The Grothendieck ring ${\ensuremath{K_0(\operatorname{Var}_{{{\ensuremath{\mathbb{C}}}}})}}$ of complex varieties is the abelian group generated by isomorphism classes of varieties, where we denote the class of a variety $X$ by ${\ensuremath{ [X]}}$, subject to the relations
${\ensuremath{ [X]}} = {\ensuremath{ [X \setminus Z]}} + {\ensuremath{ [Z]}}$, where $X$ is a variety and $Z$ is a closed subvariety of $X$.
The ring structure is induced by defining ${\ensuremath{ [X]}} \cdot {\ensuremath{ [Y]}} = {\ensuremath{ [X \times Y]}} $ for varieties $X$ and $Y$. We denote by ${\ensuremath{\mathbb{L}}}$ the class of ${\ensuremath{\mathbb{A}^{1}_{{\ensuremath{\mathbb{C}}}}}}$ and the localization ${\ensuremath{K_0(\operatorname{Var}_{{{\ensuremath{\mathbb{C}}}}})}}[{\ensuremath{\mathbb{L}}}^{-1}]$ is denoted by ${\ensuremath{\mathcal{M}_{{\ensuremath{\mathbb{C}}}}}}$. Consider for every $i \in {\ensuremath{\mathbb{N}}}$ the subgroup $F^i$ of ${\ensuremath{\mathcal{M}_{{\ensuremath{\mathbb{C}}}}}}$ generated by the elements ${\frac{{\ensuremath{ [X]}}}{{\ensuremath{\mathbb{L}}}^n}}$, where $X$ is a variety and $i \in {\ensuremath{\mathbb{N}}}$ such that ${\ensuremath{\operatorname{dim}X}} - n \leq -i$. These subgroups form a descending filtration on ${\ensuremath{\mathcal{M}_{{\ensuremath{\mathbb{C}}}}}}$ and its completion $\varprojlim_n ({\ensuremath{\mathcal{M}_{{\ensuremath{\mathbb{C}}}}}}\slash F^n)$ is denoted by ${\ensuremath{\hat{\mathcal{M}}_{{\ensuremath{\mathbb{C}}}}}}$.
The group of $n$-th roots of unity is denoted by $\mu_n$. A [*good*]{} $\mu_n$-action on a variety $X$ is an algebraic group action $\mu_n \times X \to X$ such that each orbit is contained in an affine open subvariety. Consider the group ${\ensuremath{\hat{\mu }}}:=\varprojlim_n \mu_n$. A [*good*]{} ${\ensuremath{\hat{\mu }}}$-action on $X$ is an action of ${\ensuremath{\hat{\mu }}}$ on $X$ which factors through a good $\mu_n$-action for some $n\in {\ensuremath{\mathbb{N}}}$.
Call two ${\ensuremath{\hat{\mu }}}$-actions on varieties $X$ and $Y$ isomorphic if there is an isomorphism between $X$ and $Y$ which also preserves the ${\ensuremath{\hat{\mu }}}$-action. The monodromic Grothendieck ring of complex varieties with good ${\ensuremath{\hat{\mu }}}$-action ${\ensuremath{K_0^{{\ensuremath{\hat{\mu }}}}(\operatorname{Var}_{{{\ensuremath{\mathbb{C}}}}})}}$ is defined as the abelian group generated by isomorphism classes of varieties with a good ${\ensuremath{\hat{\mu }}}$-action, where the class of a variety $X$ with action $\alpha:{\ensuremath{\hat{\mu }}}\times X \to X$ is denoted by ${\ensuremath{ [X,\alpha]}}$ or ${\ensuremath{ [X]}}$, subject to the relations
- ${\ensuremath{ [X]}} = {\ensuremath{ [X\setminus Z]}} + {\ensuremath{ [Z]}}$, where $X$ is a variety with ${\ensuremath{\hat{\mu }}}$-action, $Z$ is a closed subvariety invariant under the action, and the actions on $X\setminus Z$ and $Z$ are induced by the one on $X$;
- ${\ensuremath{ [ V ]}}= {\ensuremath{ [X \times {\ensuremath{\mathbb{A}^{n}_{{\ensuremath{\mathbb{C}}}}}}]}}$, where $V \to X$ is vector bundle of rank $n$ over a variety $X$ with a ${\ensuremath{\hat{\mu }}}$-action which is linear over the action on $X$. (We are following [@MR2106970] here)
The ring structure can be defined in the same way as before. Again we denote by ${\ensuremath{\mathbb{L}}}$ the class of ${\ensuremath{\mathbb{A}^{1}_{{\ensuremath{\mathbb{C}}}}}}$, where we equip it with the trivial ${\ensuremath{\hat{\mu }}}$-action and we define ${{\ensuremath{\mathcal{M}^{{\ensuremath{\hat{\mu }}}}_{{\ensuremath{\mathbb{C}}}}}}}$ to be the localization ${\ensuremath{K_0^{{\ensuremath{\hat{\mu }}}}(\operatorname{Var}_{{{\ensuremath{\mathbb{C}}}}})}}[{\ensuremath{\mathbb{L}}}^{-1}]$. Analogously for every $i\in {\ensuremath{\mathbb{N}}}$ we define $\hat{F}^i$ as the subgroup generated by the elements ${\frac{{\ensuremath{ [X]}}}{{\ensuremath{\mathbb{L}}}^n}}$ where $X$ is a variety such that ${\ensuremath{\operatorname{dim}X}} - n \leq -i$. Its completion $\varprojlim_n ({{\ensuremath{\mathcal{M}^{{\ensuremath{\hat{\mu }}}}_{{\ensuremath{\mathbb{C}}}}}}}\slash \hat{F}^n)$ is the monodromic completed Grothendieck ring $ {\ensuremath{\hat{\mathcal{M}}^{{\ensuremath{\hat{\mu }}}}_{{\ensuremath{\mathbb{C}}}}}}$.
$n$-jets and arcs
------------------
Fix a variety $X$ of dimension $d$ in this subsection and let $n\in {\ensuremath{\mathbb{N}}}$. Recall that the functor $$\operatorname{Sch}_{\ensuremath{\mathbb{C}}}\to \operatorname{Set}: Y \mapsto X\left(Y\times_{\ensuremath{\mathbb{C}}}\operatorname{Spec}\left( \frac{{\ensuremath{\mathbb{C}}}[t]}{(t^{n+1})}\right) \right) = \operatorname{Hom}_{\ensuremath{\mathbb{C}}}\left(Y\times_{\ensuremath{\mathbb{C}}}\operatorname{Spec}\left( \frac{{\ensuremath{\mathbb{C}}}[t]}{(t^{n+1})}\right),X \right)$$ is representable by a scheme ${\ensuremath{\mathcal{L}_{n} (X)}}$. In particular we have ${\ensuremath{\mathcal{L}_{n} (X)}} (F) = X\left(\frac{F[t]}{(t^{n+1})}\right)$ for any field extension $F$ of ${\ensuremath{\mathbb{C}}}$. This scheme is called the scheme of $n$-jets of $X$. Remark also that ${\ensuremath{\mathcal{L}_{0} (X)}} = X$.
The truncation morphisms $\pi_m^n : {\ensuremath{\mathcal{L}_{n} (X)}} \to {\ensuremath{\mathcal{L}_{m} (X)}}$ are affine, where $n \geq m$, and thus we can consider the scheme ${\ensuremath{\mathcal{L} (X)}} = \varprojlim_n {\ensuremath{\mathcal{L}_{n} (X)}}$, which is called the arc space of $X$. It is equipped with projection morphisms $\pi_n : {\ensuremath{\mathcal{L} (X)}} \to {\ensuremath{\mathcal{L}_{n} (X)}}$. A morphism $f: X\to Y$ of algebraic varieties induces morphisms $f_n : {\ensuremath{\mathcal{L}_{n} (X)}} \to {\ensuremath{\mathcal{L}_{n} (Y)}}$ and $f: {\ensuremath{\mathcal{L} (X)}} \to {\ensuremath{\mathcal{L} (Y)}}$, which are compatible with the projection morphisms.
Motivic zeta function
---------------------
We consider $${\ensuremath{\mathfrak{X}_{n}}} := \{\varphi \in {\ensuremath{\mathcal{L}_{n} ({\ensuremath{\mathbb{A}^{d}_{{\ensuremath{\mathbb{C}}}}}})}} \mid \operatorname{ord}_t f_n(\varphi) = n, \ \pi^n_0(\varphi) = 0\}$$ for $n \in {\ensuremath{\mathbb{N}}}$. This is a locally closed subset of ${\ensuremath{\mathcal{L}_{n} ({\ensuremath{\mathbb{A}^{d}_{{\ensuremath{\mathbb{C}}}}}})}} $ and thus defines a class ${\ensuremath{ [{\ensuremath{\mathfrak{X}_{n}}} ]}}$ in ${\ensuremath{K_0(\operatorname{Var}_{{{\ensuremath{\mathbb{C}}}}})}}$. The (local) motivic zeta function of $f$ (at the origin 0) is then defined as $${Z_{f}(T)} := \sum_{n> 0} {\ensuremath{ [{\ensuremath{\mathfrak{X}_{n}}}]}} {\ensuremath{\mathbb{L}}}^{-dn } T^n \in {\ensuremath{\mathcal{M}_{{\ensuremath{\mathbb{C}}}}}}[[T]].$$
Our goal now is to incorporate a (regular) differential form $\omega$ of maximal degree in this definition. This is done in [@ACLM1], [@ACLM2] and [@Veyszetafunctionsonsingularvarieties] with a restriction on the support of $\omega$. In [@VeysMEAndzetafunctionswithdifferential] and [@NemethiVeys2012] there is no restriction on the support but only the topological zeta function is used. We will show how we can see the coefficients of the motivic zeta function as the motivic measure of a subset of ${\ensuremath{\mathcal{L} ({\ensuremath{\mathbb{A}^{d}_{{\ensuremath{\mathbb{C}}}}}})}}$. Assume that $X$ is a smooth variety of dimension $d$. Let $C$ be the boolean algebra of cylindrical subsets of ${\ensuremath{\mathcal{L} (X)}}$, i.e. subsets of the form $\pi_n^{-1}(A')$, where $A'$ is a constructible subset of ${\ensuremath{\mathcal{L}_{n} (X)}}$ for some $n\in {\ensuremath{\mathbb{N}}}$. Recall that for cylindrical subsets $A $ we have that $\pi_n(A) {\ensuremath{\mathbb{L}}}^{-dn} \in {\ensuremath{\mathcal{M}_{{\ensuremath{\mathbb{C}}}}}}$ stabilizes if $n$ tends to infinity; we denote this element by $\mu(A)$. This is called the (naive) motivic measure of $A$. Define $ {\ensuremath{\mathcal{Z}_{n}}}$ to be $ \{ \varphi \in {\ensuremath{\mathcal{L} ({\ensuremath{\mathbb{A}^{d}_{{\ensuremath{\mathbb{C}}}}}})}} \mid \operatorname{ord}_t f(\varphi) = n, \pi_0(\varphi) = 0\}$ and note that $\mu( {\ensuremath{\mathcal{Z}_{n}}}) = {\ensuremath{ [{\ensuremath{\mathfrak{X}_{n}}}]}} {\ensuremath{\mathbb{L}}}^{-dn}$. Hence the coefficients of the motivic zeta function are the measures of ${\ensuremath{\mathcal{Z}_{n}}}$.
Consider now a regular differential form $\omega$ of maximal degree on $X$. Remark that $\operatorname{div}(\omega)$ is a divisor and consider ${\ensuremath{\Delta_{e}}} = \{ \varphi \in {\ensuremath{\mathcal{L} (X)}} \mid \operatorname{ord}_t \operatorname{div}(\omega)(\varphi) = e\}$, which is a cylindrical subset. We define $$\mu_\omega : C \to {\ensuremath{\hat{\mathcal{M}}_{{\ensuremath{\mathbb{C}}}}}} : A \mapsto \sum_{e\in \mathbb{N} } \mu(A \cap \Delta_e)
{\ensuremath{\mathbb{L}}}^{-e}$$ and we call this the motivic measure with respect to $\omega$. This sum converges in ${\ensuremath{\hat{\mathcal{M}}_{{\ensuremath{\mathbb{C}}}}}}$ since there exists an $i\in {\ensuremath{\mathbb{N}}}$ such that $ \mu(A \cap \Delta_e){\ensuremath{\mathbb{L}}}^{-i} \in F^0$ for all $e \in {\ensuremath{\mathbb{N}}}$.
The (local) motivic zeta function of $f$ with respect to $\omega$ (at 0) is $${Z_{f, \omega}(T)} :=
\sum_{i > 0} \mu_\omega( {\ensuremath{\mathcal{Z}_{n}}}) T^i \in {\ensuremath{\hat{\mathcal{M}}_{{\ensuremath{\mathbb{C}}}}}}[[T]].$$
We find that ${Z_{f, \omega}(T)}$ coincides with ${Z_{f}(T)}$ if $\omega $ is the standard form $dx_1 \wedge \cdots \wedge dx_n$ and if we consider the coefficients in ${\ensuremath{\hat{\mathcal{M}}_{{\ensuremath{\mathbb{C}}}}}}$.
We also have a formula in terms of an embedded resolution. Let $\pi: X \to {\ensuremath{\mathbb{A}^{n}_{{\ensuremath{\mathbb{C}}}}}}$ be an embedded resolution of singularities of $f$ and $\omega$, i.e. a proper morphism such that $\pi^{-1}(f^{-1}(0) \cup \operatorname{Supp}{\omega})$ is a strict normal crossing divisor and such that is an isomorphism outside $ f^{-1}(0) \cup \operatorname{Supp}{\omega}$. Let $E_j, j\in J$, be the irreducible components of $\pi^{-1}(f^{-1}(0)\cup \operatorname{Supp}{\omega})$, $N_j$ the multiplicity of $E_j$ in $\pi^*f$ and $\nu_j-1 $ the multiplicity of $\pi^*\omega$ along $E_j$. Remark that $(N_j,\nu_j) \neq (0,0)$ for all $j\in J$, but $N_j=0$ is possible for some $j\in J$ since it can happen that $\operatorname{Supp}(\omega)\not \subset f^{-1}(0)$.
\[theorem:denefformulamotiviczetafunction\]
Define ${\ensuremath{E_I}}= \cap_{i\in I} E_i$ and ${\ensuremath{E_I^\circ}}= \cap_{i\in I} E_i \setminus\left( \cup_{i \in J\setminus I} E_i \right)$ for $I \subset J$. Then we have $${Z_{f, \omega}(T)} = \sum_{\emptyset\neq I \subset J} ({\ensuremath{\mathbb{L}}}- 1)^{\lvert I\rvert }
{\ensuremath{ [{\ensuremath{E_I^\circ}}\cap \pi^{-1}(0)]}} \prod_{i\in I} \frac{T^{N_i}}{{\ensuremath{\mathbb{L}}}^{\nu_i} - T^{N_i}} \in
{\ensuremath{\hat{\mathcal{M}}_{{\ensuremath{\mathbb{C}}}}}}[[T]]$$
Using the techniques of [@MR1923998 Theorem 2.4] one can show this easily.
Monodromic motivic zeta function
--------------------------------
Consider $${\ensuremath{\mathfrak{X}_{n,1}}} := \{\varphi \in {\ensuremath{\mathcal{L}_{n} ({\ensuremath{\mathbb{A}^{d}_{{\ensuremath{\mathbb{C}}}}}})}} \mid
f_n(\varphi) \equiv t^n \mod\ (t^{n+1}), \pi^n_0(\varphi) = 0\}.$$ for $n \in {\ensuremath{\mathbb{N}}}$. This is a closed subset of $ {\ensuremath{\mathcal{L}_{n} ({\ensuremath{\mathbb{A}^{d}_{{\ensuremath{\mathbb{C}}}}}})}}$. We have a (natural) $\mu_n$-action defined by $a \cdot \varphi(t) = \varphi(at)$, where $a\in \mu_n $ and $\varphi \in {\ensuremath{\mathcal{L}_{n} ({\ensuremath{\mathbb{A}^{d}_{{\ensuremath{\mathbb{C}}}}}})}}$. Hence we can consider ${\ensuremath{ [{\ensuremath{\mathfrak{X}_{m,1}}}]}}$ in ${{\ensuremath{\mathcal{M}^{{\ensuremath{\hat{\mu }}}}_{{\ensuremath{\mathbb{C}}}}}}}$. The monodromic motivic zeta function of $f$ is then $$\label{eq:monodromicmotiviczetafunctionwithoutdiff}
{{\ensuremath{Z^{{\ensuremath{\hat{\mu }}}}_{f}(T)}}} :=
\sum_{n > 0} {\ensuremath{ [{\ensuremath{\mathfrak{X}_{n,1}}}]}} {\ensuremath{\mathbb{L}}}^{-dn} T^n \in
{{\ensuremath{\mathcal{M}^{{\ensuremath{\hat{\mu }}}}_{{\ensuremath{\mathbb{C}}}}}}}[[T]].$$
Look at ${\ensuremath{\mathcal{Z}_{n,1}}} = \{\varphi \in {\ensuremath{\mathcal{L} ({\ensuremath{\mathbb{A}^{d}_{{\ensuremath{\mathbb{C}}}}}})}} \mid f(\varphi) \equiv t^n \mod(t^{n+1}), \pi_0(\varphi) = 0 \}$. This has an action of $\mu_n$ like before. Then ${\ensuremath{\mathcal{Z}_{n,1}}}\cap \Delta_e$ is a cylindrical subset for every $e\in \mathbb{N}$ and the action induces an action on $\pi_m( {\ensuremath{\mathcal{Z}_{n,1}}}\cap \Delta_e)$ for every $m\in {\ensuremath{\mathbb{N}}}$. The sequence $\pi_m({\ensuremath{\mathcal{Z}_{n}}}\cap \Delta_e){\ensuremath{\mathbb{L}}}^{-md}$ stabilizes in ${\ensuremath{\hat{\mathcal{M}}^{{\ensuremath{\hat{\mu }}}}_{{\ensuremath{\mathbb{C}}}}}}$ if $m$ tends to $\infty$ and we denote this element by $\mu( {\ensuremath{\mathcal{Z}_{n,1}}}\cap \Delta_e)$. This leads us to $$\mu_\omega ({\ensuremath{\mathcal{Z}_{n,1}}}) = \sum_{e\in \mathbb{N} } \mu({\ensuremath{\mathcal{Z}_{n,1}}} \cap \Delta_e)
{\ensuremath{\mathbb{L}}}^{-e} \in {\ensuremath{\hat{\mathcal{M}}^{{\ensuremath{\hat{\mu }}}}_{{\ensuremath{\mathbb{C}}}}}}.$$ This sum converges in ${\ensuremath{\hat{\mathcal{M}}^{{\ensuremath{\hat{\mu }}}}_{{\ensuremath{\mathbb{C}}}}}}$ by the same argument as before.
The (local) monodromic motivic zeta function of $f$ with respect to $\omega$ (at the origin) is $${{\ensuremath{Z^{{\ensuremath{\hat{\mu }}}}_{f, \omega}(T)}}} := \sum_{i> 0}\mu_\omega ({\ensuremath{\mathcal{Z}_{i,1}}}) T^i \in {\ensuremath{\hat{\mathcal{M}}^{{\ensuremath{\hat{\mu }}}}_{{\ensuremath{\mathbb{C}}}}}} [[T]].$$
Remark that this definition coincides again with the one in (\[eq:monodromicmotiviczetafunctionwithoutdiff\]) if $\omega $ is the standard form and we consider the coefficients in $ {\ensuremath{\hat{\mathcal{M}}^{{\ensuremath{\hat{\mu }}}}_{{\ensuremath{\mathbb{C}}}}}}$.
We also have a formula for this zeta function. Consider again the situation of Theorem \[theorem:denefformulamotiviczetafunction\]. Suppose $\emptyset \neq I\subseteq J$. Define $m_I = \gcd_{i \in I}(N_i)$. We will introduce ${\ensuremath{\tilde{E}_I^\circ}}$ as a unramified Galois cover of ${\ensuremath{E_I^\circ}}$ with Galois group $\mu_{m_I}$. Let $U$ be an affine Zariski open such that $f\circ h = u v^{m_I}$, where $u$ is a unit and $v$ a regular function on $U$. Then the restriction of ${\ensuremath{\tilde{E}_I^\circ}}$ above ${\ensuremath{E_I^\circ}}\cap U$ is defined as $$\{(z,y) \in {\ensuremath{\mathbb{A}^{1}_{{\ensuremath{\mathbb{C}}}}}} \times U \mid z^{m_I} = u^{-1}\}.$$ Since another choice of $u$ and $v$ induces an isomorphism, these covers glue to a finite Galois cover ${\ensuremath{\tilde{E}_I^\circ}}$ of ${\ensuremath{E_I^\circ}}$. The (natural) $\mu_{m_I}$-action is obtained by multiplying the $z$-coordinates with elements of $\mu_{m_I}$, which gives us an element ${\ensuremath{ [{\ensuremath{\tilde{E}_I^\circ}}]}}$ in ${\ensuremath{K_0^{{\ensuremath{\hat{\mu }}}}(\operatorname{Var}_{{{\ensuremath{\mathbb{C}}}}})}}$.
\[theorem:denefformulamonodromicmotiviczetafunction\] We have the following equality: $${Z_{f, \omega}(T)} = \sum_{\emptyset\neq I \subset J} ({\ensuremath{\mathbb{L}}}- 1)^{\lvert I\rvert -1}
{\ensuremath{ [{\ensuremath{\tilde{E}_I^\circ}}\cap \pi^{-1}(0)]}} \prod_{i\in I} \frac{T^{N_i}}{{\ensuremath{\mathbb{L}}}^{\nu_i} - T^{N_i}}
\in {\ensuremath{\hat{\mathcal{M}}^{{\ensuremath{\hat{\mu }}}}_{{\ensuremath{\mathbb{C}}}}}}[[T]].$$
Using the techniques of [@MR1923998 Theorem 2.4] one can show this easily.
Topological and other zeta functions
------------------------------------
Given a ring morphism $\chi: {\ensuremath{\hat{\mathcal{M}}_{{\ensuremath{\mathbb{C}}}}}} \to R$, we can consider the specialization of these motivic zeta functions, i.e. $${Z_{f, \omega}(T)} = \sum_{i > 0} \chi(\mu_\omega( {\ensuremath{\mathcal{Z}_{n}}})) T^i \in R[[T]].$$ In this sense we can obtain several other and already known zeta functions such as $ p$-adic zeta functions and Hodge zeta functions.
### Topological zeta function
The topological zeta function will be used and thus we discuss this incarnation in more detail. The topological Euler characteristic ${\ensuremath{\chi_{\text{top}}}}(X) \in {{\ensuremath{\mathbb{Z}}}}$ of a variety $X$ has the following properties:
- ${\ensuremath{\chi_{\text{top}}}}(X) = {\ensuremath{\chi_{\text{top}}}}(X\setminus Z) + {\ensuremath{\chi_{\text{top}}}}(Z) $ for all varieties $X$ and closed subvarieties $Z$ of $X$,
- ${\ensuremath{\chi_{\text{top}}}}(X\times Y) = {\ensuremath{\chi_{\text{top}}}}(X) \cdot {\ensuremath{\chi_{\text{top}}}}(Y)$ for all varieties $X$ and $Y$,
- ${\ensuremath{\chi_{\text{top}}}}({\ensuremath{\mathbb{A}^{1}_{{\ensuremath{\mathbb{C}}}}}}) = 1$
This implies that we can consider it as a ring morphism ${\ensuremath{\chi_{\text{top}}}}: {\ensuremath{\mathcal{M}_{{\ensuremath{\mathbb{C}}}}}} \to {{\ensuremath{\mathbb{Z}}}}$. Since ${\ensuremath{\chi_{\text{top}}}}({\ensuremath{\mathbb{L}}}) = 1$, we cannot extend it to a morphism from ${\ensuremath{\hat{\mathcal{M}}_{{\ensuremath{\mathbb{C}}}}}}$. As discussed in [@denefloeser2001], we can still apply ${\ensuremath{\chi_{\text{top}}}}$ to elements of ${\ensuremath{\hat{\mathcal{M}}_{{\ensuremath{\mathbb{C}}}}}}$ which are the image of an element of ${\ensuremath{\mathcal{M}_{{\ensuremath{\mathbb{C}}}}}}$.
The local topological zeta function of $f$ is defined as the rational function $${Z_{f,\omega}^{\text{top}}(s)} = \sum_{\emptyset\neq I \subseteq J}{\ensuremath{\chi_{\text{top}}}}({\ensuremath{E_I^\circ}}\cap \pi^{-1}(0)) \prod_{i\in I}
\frac{1}{N_is +\nu_i} \in {\ensuremath{\mathbb{Q}}}(s),$$ and the twisted local topological zeta function as $${Z_{f,\omega}^{\text{top}, (e)}(s)} =
\sum_{\stackrel{\emptyset\neq I \subseteq J,}{ e\vert m_I}}{\ensuremath{\chi_{\text{top}}}}({\ensuremath{E_I^\circ}}\cap \pi^{-1}(0))
\prod_{i\in I} \frac{1}{N_i s+ \nu_i} \in {\ensuremath{\mathbb{Q}}}(s),$$ where $e\in {\ensuremath{\mathbb{N}}}$.
Given a character $\alpha$ of ${\ensuremath{\hat{\mu }}}$, there is a natural ring homomorphism $${\ensuremath{\chi_{\text{top}}}}(\cdot, \alpha) : {{\ensuremath{\mathcal{M}^{{\ensuremath{\hat{\mu }}}}_{{\ensuremath{\mathbb{C}}}}}}} \to {{\ensuremath{\mathbb{Z}}}}: X \mapsto \sum_{q\geq 0} \dim H^q(X,{\ensuremath{\mathbb{C}}})_\alpha$$ where $H^*(X,{\ensuremath{\mathbb{C}}})_\alpha$ is the part of $H^*(X,{\ensuremath{\mathbb{C}}})$ on which ${\ensuremath{\hat{\mu }}}$ acts by multiplication by $\alpha$.
Remark that there always exists a character $\alpha$ of given order $e$ and that ${\ensuremath{\chi_{\text{top}}}}(X,\alpha)$ only depends on $\alpha$. We denote this by ${\ensuremath{\chi_{\text{top}}}}^{(e)}(X) = {\ensuremath{\chi_{\text{top}}}}(X,\alpha) $. We can apply ${\ensuremath{\chi_{\text{top}}}}$ to $ {Z_{f, \omega}({\ensuremath{\mathbb{L}}}^{-n})}$ where $n\in {\ensuremath{\mathbb{N}}}$ since it is equal to $$\sum_{\emptyset\neq I \subset J}
{\ensuremath{ [{\ensuremath{E_I^\circ}}\cap \pi^{-1}(0)]}} \prod_{i\in I}
\frac{ ({\ensuremath{\mathbb{L}}}- 1){\ensuremath{\mathbb{L}}}^{-nN_i}}{{\ensuremath{\mathbb{L}}}^{\nu_i} - {\ensuremath{\mathbb{L}}}^{-nN_i}}
= \sum_{\emptyset\neq I \subset J} ({\ensuremath{\mathbb{L}}}- 1)^{\lvert I\rvert }
{\ensuremath{ [{\ensuremath{E_I^\circ}}\cap \pi^{-1}(0)]}} \prod_{i\in I} \frac{1}{{\ensuremath{ [{\ensuremath{\mathbb{P}^{ N_in + \nu_i -1}_{{\ensuremath{\mathbb{C}}}}}}]}}}.$$ Another definition of ${Z_{f,\omega}^{\text{top}}(s)}$ is then the unique rational function such that $${Z_{f,\omega}^{\text{top}}(n)} = {\ensuremath{\chi_{\text{top}}}}( {Z_{f, \omega}({\ensuremath{\mathbb{L}}}^{-n})})$$ for all $n\in {\ensuremath{\mathbb{N}}}$. Analogous we have that ${Z_{f,\omega}^{\text{top}, (e)}(s)}$ is the unique rational function such that $${Z_{f,\omega}^{\text{top}, (e)}(n)} = {\ensuremath{\chi_{\text{top}}}}^{(e)}(({\ensuremath{\mathbb{L}}}-1){{\ensuremath{Z^{{\ensuremath{\hat{\mu }}}}_{f,\omega}({\ensuremath{\mathbb{L}}}^{-n })}}}).$$ for all $n\in {\ensuremath{\mathbb{N}}}$. In this sense the (twisted) topological zeta functions are avatars of the motivic zeta functions.
Splice diagrams
===============
This section is dedicated to the notion of splice diagrams as described in [@NemethiVeys2012]. We will not discuss it in full generality but rather stick to the case of plane curve singularities. Consider a polynomial $f \in {\ensuremath{\mathbb{C}}}[x,y]$, a (regular) differential $2$-form $\omega$ on ${\ensuremath{\mathbb{A}^{2}_{{\ensuremath{\mathbb{C}}}}}}$ and an embedded resolution of singularities $\pi : X \to {\ensuremath{\mathbb{A}^{2}_{{\ensuremath{\mathbb{C}}}}}}$ for $(f,\omega)$.
The topology of the singularity can be described by means of the dual graph $G = G_\pi $ associated to $\pi$, $f$ and $\omega$. Let $E_i, i\in J$, be the irreducible components of $\pi^{-1}( f^{-1}(0) \cup \operatorname{Supp}\omega )$. We have then three types of components: exceptional curves, strict transforms of components of $\{f=0\}$ and strict transforms of components of $\operatorname{Supp}(\omega)$. This type is not unique, i.e. a component can be a strict transform of a component of $\{f=0\}$ and a strict transform of a component of $\operatorname{Supp}(\omega)$, but this is the only case where a component can have multiple types.
Each exceptional curve $E_i$ determines a vertex of $G$ and edges correspond to the intersection points of the exceptional curves. Note that this $G$ is a tree, each exceptional curve is rational and $\det (-I(G)) = 1$, where $I(H)$ is the negative definite intersection matrix $(E_i\cdot E_j)_{i,j\in H}$ if $H$ is a subset of the nodes of $G$. These are exactly the conditions of an integral homology sphere and thus we can use all the machinery developed in [@NemethiVeys2012].
We can now talk about the splice diagram $\Gamma = \Gamma_\pi(f,\omega)$. The underlying graph is the one obtained by removing all nodes of $G$ of valency $2$. We add to this graph some decorations. On each pair $(v,e)$ where $v $ is a node of $\Gamma$ and $e$ is an edge starting at $v$, we have the decoration $d_{ve} = \det(-I(G_{ve}))$, where $G_{ve}$ is the component of $G\setminus\{v\}$ in the direction of $e$.
Each irreducible component of the strict transform of $\{f=0\}$ intersecting in the exceptional component $E$ corresponding to the node $v$ is represented by an arrow $a$ attached at $v$ and has the multiplicity $N_a$ of $f$ along $E$ as a decoration. Similarly a component of a strict transform of $\operatorname{Supp}\omega $ is being displayed by a dotted arrow $a$ and again the multiplicity $\nu_a-1$ is the associated decoration.
It is important to stress that we are only dealing here with plane curve singularities. All these can easily be extended to the case of integral homology spheres as in [@NemethiVeys2012].
Consider the cusp $f=x^3 - y^2$ , its minimal embedded resolution $\pi$ and differential form $\omega= x^4y^5dx\wedge dy$. Then its splice diagram is the following:
(een) at (-[3]{}, 0); (twee) at (0, 0); (drie) at ([3]{}, 0); (vier) at (0, [2]{}); (vijf) at ( -[3]{}-[2]{},0); (zes) at ( +[3]{}+[2]{},0); (een)–(twee) node\[very near start, above\][$1$]{} node\[very near end, above\][$3$]{}; (twee)–(drie) node\[very near start, above\][$2$]{} node\[very near end, above\][$2$]{}; (een) circle \[radius=3pt\]; (twee) circle \[radius=3pt\]; (drie) circle \[radius=3pt\]; (twee)–(vier); node\[right\] at (vier) [(1)]{}; node\[above\] at (vijf) [(4)]{}; node\[above\] at (zes) [(5)]{}; (een)–(vijf); (drie)–(zes);
We have the following properties of the decorations $\{d_{ve}\}_{e,v}$ of $\Gamma$: $d_{ve} \geq 1$, $\{d_{ve}\}_e$ are pairwise coprime for a fixed node $v$ and any edge determinant $q_e$ is positive. Recall that this edge determinant is $\det(-I(G_e))$, where $G_e$ are the exceptional curves of $G$ that lie on $e$. There is an easier formula. Consider the situation in Figure \[fig:edgesituation\]. The edge determinant $q_e$ is then equal to $
dd' - DD'
$ where $D = \prod_{i=1}^nd_i$ and $D'=\prod_{i=1}^{n'}d'_i$.
(dOne) at (-[4]{},[1]{}); (dN) at (-[4]{},-[1]{}) ; (daOne) at ([4]{},[1]{}) ; (daN) at ([4]{},-[1]{}) ; (vL) at (-[2]{},0) ; (vR) at ([2]{},0) ; (dOne)–(vL) node \[midway, above \] [$d_1$]{}; (dN)–(vL) node \[midway, above \] [$d_n$]{}; (vL)–(vR) node\[very near start, above\][$d$]{} node\[very near end, above\][$d'$]{}; (vR)–(daOne) node \[midway, above \] [$d'_1$]{}; (vR)–(daN) node \[midway, above \] [$d'_{n'}$]{}; node at ($ 1/2*(dOne) + 1/2*(dN) +(1/3,0) $) ; node at ($ 1/2*(daOne) + 1/2*(daN) +(-1/3,0) $) ; node at ($1/2*(vR) + 1/2*(vL) + (0,-.7) $)[$e$]{}; (vR) circle \[radius=3pt\]; (vL) circle \[radius=3pt\];
It turns out that this splice diagram is very useful for computing the multiplicities of all the exceptional components. We only need the multiplicities at the strict transforms and the decorations on the splice diagram. For a node $v$ of $\Gamma$, which corresponds to an exceptional curve, we have $$\label{eq:formulafornodes}
N_v = \sum_{a \text{ arrow}} N_al_{va }$$ where $l_{va}$ is the product of the edge decorations adjacent to the path from $v$ to $a$ but not on it. Analogously, we have $$\label{eq:formulafordifferentialform}
\nu_v = \sum_{w \text{ node}} (2-\delta_w)l_{vw} + \sum_{a \text{ dotted arrow }} l_{va}(\nu_a-1).$$ This $\delta_v$ is the valency of the node $v$ considered without arrows.
Diagrams
--------
We introduce now what we will call a diagram. While creating splice diagrams, we are deleting nodes of valency two in the graph. But the formulas still work without doing so. Consider now dual graphs of resolutions as before, were we may choose how many valency two nodes we have removed. The decorations are still the same. From now on we will call this a [*diagram*]{}.
We call a diagram $\Gamma$ [*realizable*]{} if it is a dual graph of a resolution of singularities. A diagram with no nodes of valency 2 is called [*minimally reduced*]{}.
So a splice diagram is a minimally reduced diagram. Of course every diagram can be reduced to a minimally reduced diagram by deleting all the nodes of valency two.
1. If a diagram satisfies $q_e = 1$ for all edges $e$ and $d_{va}=1$ for all (dotted) arrows $a$ attached at a node $v$, then it must be realizable.
2. Every reduced diagram can be ‘extended’ or ‘refined’ to a realizable diagram. This can be done by (re)adding nodes (of valency 2) on those edges with $q_e > 1$ and arrows with $d_{va} >1$.
3. The possible ways to add nodes to the edge $e$ is not unique and corresponds to smooth refinements of the fan associated to the cone $(D,d) \mathbb{R}_{\geq 0} + (d',D')\mathbb{R}_{\geq 0} \subseteq \mathbb{R}^2$ where $D = \prod_{i=1 }^n d_i$ and $D'= \prod_{i=1}^{n'} d'_i$, where we consider the situation as in Figure \[fig:edgesituation\].
This follows from interpreting toric concepts described in [@MR2810322 Chapter 10].
Zeta functions of diagrams
--------------------------
Recall that we have a formula for our zeta functions in terms of an embedded resolution. In [@NemethiVeys2012] Némethi and Veys define a topological zeta function in terms of the splice diagram. We will give here another definition which is easily seen to be equivalent and is actually just the formula as in Theorem \[theorem:denefformulamotiviczetafunction\].
Let $\Gamma$ be a realizable diagram. We define the [*topological zeta function*]{} of $\Gamma$ to be $$\begin{aligned}
{Z_{\Gamma}^{\text{top}}(s)} &:=&
\sum_{v \text{ is a node}} \frac{2-\delta_v}{ ( N_vs + \nu_v )} \\
& & + \sum_{e= (v,w) \text{ is an edge}}
\frac{1}{( N_vs + \nu_v)({N_w}s + {\nu_w})} \\
& & + \sum_{a \text{ (dotted) arrow at $v$}}
\frac{1}{( {N_v}s + {\nu_v})( {N_a}s + {\nu_a} )}
\in {\ensuremath{\mathbb{Q}}}(s).\end{aligned}$$ If $\Gamma$ is not realizable, let $\Gamma' $ be a realizable refinement of $\Gamma$ and define $ {Z_{\Gamma}^{\text{top}}(s)} := {Z_{\Gamma'}^{\text{top}}(s)}$. To prove that this is well-defined, we remark two things:
- If you add a node on a realizable diagram $\Gamma$ such that the resulting diagram is still realizable, the topological zeta function does not change. This coincides with blowing up in an intersection point of the two exceptional curves corresponding to the adjoining nodes.
- For any diagram, you can go from a realizable refinement to any other realizable refinement by blowing up points and doing the opposite operation.
In what follows we will only give the definition of the considered zeta functions in the case of a realizable diagram since these remarks will also imply that it is well-defined in those cases.
[cc]{}\
[0.3]{}
(sup) at (0,.04); (dOne) at (-[2]{},[.5]{}); (dN) at (-[2]{},-[.5]{}) ; (daOne) at ([2]{},0) ; (vL) at (-[1]{},0) ; (vR) at ([1]{},0) ; (dOne)–(vL) node \[midway, above \] [$d_1$]{}; (dN)–(vL) node \[midway, above \] [$d_n$]{}; (vL)–(vR) node\[very near start, above\][$d$]{} node\[very near end, above\][$d'$]{}; ($(vR)+(sup)$)–($(daOne) + (sup)$) node \[ near start, above \] [$D'$]{} node \[very near end, above\] [$(M)$]{}; ($(vR)-(sup)$)–($(daOne) - (sup)$) node \[very near end,below\] [$(i-1)$]{}; node at ($ 1/2*(dOne) + 1/2*(dN) +(1/3,0) $) ; (vR) circle \[radius=3pt\]; (vL) circle \[radius=3pt\];
&
[0.3]{}
(sup) at (0,.04); (dOne) at (-[2]{},0); (daOne) at ([2]{},[.5]{}) ; (daN) at ([2]{},-[.5]{}) ; (vL) at (-[1]{},0) ; (vR) at ([1]{},0) ; ($(dOne)+(sup)$)–($(vL) + (sup)$)node \[near end, above \] [$D$]{} node \[very near start, above\] [$(M')$]{}; ($(dOne)-(sup)$)–($(vL) - (sup)$) node \[very near start, below \] [$(i'-1)$]{}; (vL)–(vR) node\[very near start, above\][$d$]{} node\[very near end, above\][$d'$]{}; (vR)–(daOne) node \[midway, above \] [$d'_1$]{}; (vR)–(daN) node \[midway, above \] [$d'_{n'}$]{}; node at ($ 1/2*(daOne) + 1/2*(daN) +(-1/3,0) $) ; (vR) circle \[radius=3pt\]; (vL) circle \[radius=3pt\];
Let $\Gamma$ be a realizable diagram. We define the [*motivic zeta function*]{} of $\Gamma$ as $$\begin{aligned}
{Z_{\Gamma}(T)} &:=& \sum_{v \text{ is a node}} \frac{({\ensuremath{\mathbb{L}}}- 1) ({\ensuremath{\mathbb{L}}}+ 1 - \delta_v)
T^{N_v}}{{\ensuremath{\mathbb{L}}}^{\nu_v} - T^{N_v}} \\
& & + \sum_{e= (v,w) \text{ is an edge}} \frac{({\ensuremath{\mathbb{L}}}- 1)^2 T^{N_v + N_w}}{({\ensuremath{\mathbb{L}}}^{\nu_v} - T^{N_v})({\ensuremath{\mathbb{L}}}^{\nu_w} - T^{N_w})} \\
& & + \sum_{a \text{ (dotted) arrow at $v$}} \frac{({\ensuremath{\mathbb{L}}}- 1)^2 T^{N_v + N_a}}{({\ensuremath{\mathbb{L}}}^{\nu_v} - T^{N_v})({\ensuremath{\mathbb{L}}}^{\nu_a} - T^{N_a})}
\in{\ensuremath{\hat{\mathcal{M}}_{{\ensuremath{\mathbb{C}}}}}}[[T]].\end{aligned}$$
It is possible to define motivic zeta functions for splice diagrams using similar formulas as described in [@MR1425327]. These formulas for the topological zeta function are used in [@NemethiVeys2012] to prove . However it is unlikely that a proof of Theorem \[theorem:splicedecompositionformulaformotiviczetafucntions\] can be constructed using these formulas without our notion of diagrams because of the complexity of the formulas.
A monodromic motivic zeta function is not available for our notion of diagrams. However we can define a twisted topological zeta function. Consider an $e\in {\ensuremath{\mathbb{N}}}$ and define $$\begin{aligned}
{Z_{\Gamma}^{\text{top}, (e)}(s)} &:=&
\sum_{\overset{ v \text{ is a node},}{ e\mid N_v}} \frac{2-\delta_v}{ ( sN_v + \nu_v )} \\
& & + \sum_{\overset{e= (v,w) \text{ is an edge},}{ e\mid N_v, e\mid N_w}}
\frac{1}{( sN_v + \nu_v )(s{N_w} + {\nu_w} )} \\
& & + \sum_{\overset{a \text{ (dotted) arrow at $v$},}{ e\mid N_v, e\mid N_a}}
\frac{1}{(s {N_v} + {\nu_v} )(s{N_a} + {\nu_a})}
\in {\ensuremath{\mathbb{Q}}}(s)\end{aligned}$$ if $\Gamma$ is a realizable diagram.
[0.6]{}
(dOne) at (-[2]{},[.5]{}); (dN) at (-[2]{},-[.5]{}) ; (daOne) at ([2]{},[.5]{}) ; (daN) at ([2]{},-[.5]{}) ; (down) at (0,-[.5]{}) ; (vL) at (-[1]{},0) ; (vR) at ([1]{},0) ; (vLL) at (-[2]{},0) ; (vRR) at ([2]{},0) ; (labelup) at (0,.3); (labeldown) at (0,-.3); (down) at (0,-[.5]{}) ;
(smalldown) at (0,-.1); (smallup) at (0,.05);
(vLu) at ($(vL) + (smallup)$); (vLLu) at ($(vLL) + (smallup)$); (vRu) at ($(vR) + (smallup)$); (vRRu) at ($(vRR) + (smallup)$);
(vLd) at ($(vL) + (smalldown)$); (vLLd) at ($(vLL) + (smalldown)$); (vRd) at ($(vR) + (smalldown)$); (vRRd) at ($(vRR) + (smalldown)$);
(vL) – (vR);
(vL)–(vR) node\[very near start, above\][$d$]{} node\[very near end, above\][$d'$]{}; (vR) circle \[radius=3pt\]; (vL) circle \[radius=3pt\]; ($1/4*(vL) + 3/4*(vR)$) circle \[radius=3pt\]; ($2/3*(vL) + 1/3*(vR)$) circle \[radius=3pt\]; node\[color=[green]{}, rectangle, rounded corners, draw=[green]{}\] at ($1/2*(vL) + 1/2*(vR)+(down)$)[$B$]{} ;
($(dOne) + (-.5,.5)$) – ($(dOne)+ (.5,.5) $) – ($(vL)$) – ($(dN) + (0,-.25)$) – ($(dN) + (-.5,-.5)$) – cycle;
(dOne)–(vL) node \[midway, above \] [$d_1$]{}; (dN)–(vL) node \[midway, above \] [$d_n$]{}; node at ($ 1/2*(dOne) + 1/2*(dN) +(1/3,0) $) ; ($1/4*(vL) + 3/4*(dOne)$) circle \[radius=3pt\]; ($2/3*(vL) + 1/3*(dOne)$) circle \[radius=3pt\]; ($1/4*(dN) + 3/4*(vL)$) circle \[radius=3pt\]; ($2/3*(dN) + 1/3*(vL)$) circle \[radius=3pt\]; node\[color=[red]{}, rectangle, rounded corners, draw=[red]{}\] at ($1/2*(vL) + 1/2*(dN)+(down)$)[$A$]{} ;
($(vR) + (0,-0.5)$) –($(vR) + (0,.5)$) – (daOne) – (daN) – cycle;
(vR)–(daOne) node \[midway, above \] [$d'_1$]{}; (vR)–(daN) node \[midway, above \] [$d'_{n'}$]{}; node at ($ 1/2*(daOne) + 1/2*(daN) +(-1/3,0) $) ; ($1/4*(vR) + 3/4*(daOne)$) circle \[radius=3pt\]; ($2/3*(vR) + 1/3*(daOne)$) circle \[radius=3pt\]; ($1/4*(daN) + 3/4*(vR)$) circle \[radius=3pt\]; ($2/3*(daN) + 1/3*(vR)$) circle \[radius=3pt\]; node\[color=[blue]{}, rectangle, rounded corners, draw=[blue]{}\] at ($1/2*(vR) + 1/2*(daN)+(down)$)[$C$]{} ;
[0.6]{}
(dOne) at (-[2]{},[.5]{}); (dN) at (-[2]{},-[.5]{}) ; (daOne) at ([2]{},[.5]{}) ; (daN) at ([2]{},-[.5]{}) ; (down) at (0,-[.5]{}) ; (vL) at (-[1]{},0) ; (vR) at ([1]{},0) ; (vLL) at (-[2]{},0) ; (vRR) at ([2]{},0) ; (labelup) at (0,.3); (labeldown) at (0,-.3); (down) at (0,-[.5]{}) ;
(smalldown) at (0,-.1); (smallup) at (0,.05);
(vLu) at ($(vL) + (smallup)$); (vLLu) at ($(vLL) + (smallup)$); (vRu) at ($(vR) + (smallup)$); (vRRu) at ($(vRR) + (smallup)$);
(vLd) at ($(vL) + (smalldown)$); (vLLd) at ($(vLL) + (smalldown)$); (vRd) at ($(vR) + (smalldown)$); (vRRd) at ($(vRR) + (smalldown)$);
(vL) – (vR);
(vL)–(vR) node\[very near start, above\][$d$]{} node\[very near end, above\][$d'$]{}; (vR) circle \[radius=3pt\]; (vL) circle \[radius=3pt\]; ($1/4*(vL) + 3/4*(vR)$) circle \[radius=3pt\]; ($2/3*(vL) + 1/3*(vR)$) circle \[radius=3pt\]; node\[color=[green]{}, rectangle, rounded corners, draw=[green]{}\] at ($1/2*(vL) + 1/2*(vR)+(down)$)[$B$]{} ;
($(vLL) + (-1,-.2)$) – ($(vLL) + (-1,.2)$) – ($(vL)+(0,1)$)– ($(vL)+(0,-1)$) –cycle;
node\[color=[black]{}, rectangle, rounded corners, draw=[black]{}\] at ($1/2*(vL) + 1/2*(dN)+(down)$)[$D$]{}; (vLLu) – (vLu) node\[ near end, above\][$D$]{}; (vLLd) – (vLd); at ($(vLLu) + (labelup)$) [$(M')$]{}; at ($(vLLd) + (labeldown)$) [$i'-1$]{}; ($2/3*(vL) + 1/3*(vLL)$) circle \[radius=3pt\];
($(vRR) + (1,-.2)$) – ($(vRR) + (1,.2)$) – ($.5*(vRR) + .5*(vR) + (0,1.5)$) – ($(vR)+(0,.75)$)– ($(vR)+(0,-.75)$) –cycle;
(vRu) – (vRRu) node\[ near start, above\][$D'$]{}; (vRd) – (vRRd);
at ($(vRRd) + (labeldown)$) [$i-1$]{}; at ($(vRRu) + (labelup)$) [$(M)$]{}; ($1/4*(vR) + 3/4*(vRR)$) circle \[radius=3pt\]; node\[color=[darkbrown]{}, rectangle, rounded corners, draw=[darkbrown]{}\] at ($1/2*(vR) + 1/2*(daN)+(down)$)[$E$]{} ;
[0.6]{}
(dOne) at (-[2]{},[.5]{}); (dN) at (-[2]{},-[.5]{}) ; (daOne) at ([2]{},[.5]{}) ; (daN) at ([2]{},-[.5]{}) ; (down) at (0,-[.5]{}) ; (vL) at (-[1]{},0) ; (vR) at ([1]{},0) ; (vLL) at (-[2]{},0) ; (vRR) at ([2]{},0) ; (labelup) at (0,.3); (labeldown) at (0,-.3); (down) at (0,-[.5]{}) ;
(smalldown) at (0,-.1); (smallup) at (0,.05);
(vLu) at ($(vL) + (smallup)$); (vLLu) at ($(vLL) + (smallup)$); (vRu) at ($(vR) + (smallup)$); (vRRu) at ($(vRR) + (smallup)$);
(vLd) at ($(vL) + (smalldown)$); (vLLd) at ($(vLL) + (smalldown)$); (vRd) at ($(vR) + (smalldown)$); (vRRd) at ($(vRR) + (smalldown)$);
(vL) – (vR);
(vL)–(vR) node\[very near start, above\][$d$]{} node\[very near end, above\][$d'$]{}; (vR) circle \[radius=3pt\]; (vL) circle \[radius=3pt\]; ($1/4*(vL) + 3/4*(vR)$) circle \[radius=3pt\]; ($2/3*(vL) + 1/3*(vR)$) circle \[radius=3pt\]; node\[color=[green]{}, rectangle, rounded corners, draw=[green]{}\] at ($1/2*(vL) + 1/2*(vR)+(down)$)[$B$]{} ;
($(dOne) + (-.5,.5)$) – ($(dOne)+ (.5,.5) $) – ($(vL)$) – ($(dN) + (0,-.25)$) – ($(dN) + (-.5,-.5)$) – cycle;
(dOne)–(vL) node \[midway, above \] [$d_1$]{}; (dN)–(vL) node \[midway, above \] [$d_n$]{}; node at ($ 1/2*(dOne) + 1/2*(dN) +(1/3,0) $) ; ($1/4*(vL) + 3/4*(dOne)$) circle \[radius=3pt\]; ($2/3*(vL) + 1/3*(dOne)$) circle \[radius=3pt\]; ($1/4*(dN) + 3/4*(vL)$) circle \[radius=3pt\]; ($2/3*(dN) + 1/3*(vL)$) circle \[radius=3pt\]; node\[color=[red]{}, rectangle, rounded corners, draw=[red]{}\] at ($1/2*(vL) + 1/2*(dN)+(down)$)[$A$]{} ;
($(vRR) + (1,-.2)$) – ($(vRR) + (1,.2)$) – ($.5*(vRR) + .5*(vR) + (0,1.5)$) – ($(vR)+(0,.75)$)– ($(vR)+(0,-.75)$) –cycle;
(vRu) – (vRRu) node\[ near start, above\][$D'$]{}; (vRd) – (vRRd);
at ($(vRRd) + (labeldown)$) [$i-1$]{}; at ($(vRRu) + (labelup)$) [$(M)$]{}; ($1/4*(vR) + 3/4*(vRR)$) circle \[radius=3pt\]; node\[color=[darkbrown]{}, rectangle, rounded corners, draw=[darkbrown]{}\] at ($1/2*(vR) + 1/2*(daN)+(down)$)[$E$]{} ;
[0.6]{}
(dOne) at (-[2]{},[.5]{}); (dN) at (-[2]{},-[.5]{}) ; (daOne) at ([2]{},[.5]{}) ; (daN) at ([2]{},-[.5]{}) ; (down) at (0,-[.5]{}) ; (vL) at (-[1]{},0) ; (vR) at ([1]{},0) ; (vLL) at (-[2]{},0) ; (vRR) at ([2]{},0) ; (labelup) at (0,.3); (labeldown) at (0,-.3); (down) at (0,-[.5]{}) ;
(smalldown) at (0,-.1); (smallup) at (0,.05);
(vLu) at ($(vL) + (smallup)$); (vLLu) at ($(vLL) + (smallup)$); (vRu) at ($(vR) + (smallup)$); (vRRu) at ($(vRR) + (smallup)$);
(vLd) at ($(vL) + (smalldown)$); (vLLd) at ($(vLL) + (smalldown)$); (vRd) at ($(vR) + (smalldown)$); (vRRd) at ($(vRR) + (smalldown)$);
(vL) – (vR);
(vL)–(vR) node\[very near start, above\][$d$]{} node\[very near end, above\][$d'$]{}; (vR) circle \[radius=3pt\]; (vL) circle \[radius=3pt\]; ($1/4*(vL) + 3/4*(vR)$) circle \[radius=3pt\]; ($2/3*(vL) + 1/3*(vR)$) circle \[radius=3pt\]; node\[color=[green]{}, rectangle, rounded corners, draw=[green]{}\] at ($1/2*(vL) + 1/2*(vR)+(down)$)[$B$]{} ;
($(vLL) + (-1,-.2)$) – ($(vLL) + (-1,.2)$) – ($(vL)+(0,1)$)– ($(vL)+(0,-1)$) –cycle;
node\[color=[black]{}, rectangle, rounded corners, draw=[black]{}\] at ($1/2*(vL) + 1/2*(dN)+(down)$)[$D$]{}; (vLLu) – (vLu) node\[ near end, above\][$D$]{}; (vLLd) – (vLd); at ($(vLLu) + (labelup)$) [$(M')$]{}; at ($(vLLd) + (labeldown)$) [$i'-1$]{}; ($2/3*(vL) + 1/3*(vLL)$) circle \[radius=3pt\];
($(vR) + (0,-0.5)$) –($(vR) + (0,.5)$) – (daOne) – (daN) – cycle;
(vR)–(daOne) node \[midway, above \] [$d'_1$]{}; (vR)–(daN) node \[midway, above \] [$d'_{n'}$]{}; node at ($ 1/2*(daOne) + 1/2*(daN) +(-1/3,0) $) ; ($1/4*(vR) + 3/4*(daOne)$) circle \[radius=3pt\]; ($2/3*(vR) + 1/3*(daOne)$) circle \[radius=3pt\]; ($1/4*(daN) + 3/4*(vR)$) circle \[radius=3pt\]; ($2/3*(daN) + 1/3*(vR)$) circle \[radius=3pt\]; node\[color=[blue]{}, rectangle, rounded corners, draw=[blue]{}\] at ($1/2*(vR) + 1/2*(daN)+(down)$)[$C$]{} ;
Splicing formula for motivic zeta function
==========================================
In this section we will prove a splicing formula for motivic zeta functions for (splice) diagrams. This immediately generalizes [@NemethiVeys2012 Theorem 3.2.4 (1)].
Consider a diagram $\Gamma$ with an edge $e$ between $v_L$ to $v_R$ as in Figure \[fig:splicingexpl:Gamma\]. Recall that $D = \prod_{i=1}^{n} d_i$ and $D' = \prod_{i=1}^{n'} d'_i$. We define the multiplicities $$\label{eq:splicing:multiplicityoff}
M = \sum_{a \text{ arrow at right side}} N_a l_{ea } \qquad \text{ and } \qquad M' = \sum_{a \text{ arrow at left side}} N_a l_{ea },$$ and the multiplicities $$\label{eq:splicing:multiplicityofomega}
i = \sum_{w \text{ node at right side}} (2-\delta_w)l_{ew} + \sum_{a \text{ dotted arrow at right side}} l_{ea}(i_a-1)$$ and $$\label{eq:splicing:multiplicityofomegap}
i' = \sum_{w \text{ node at left side}} (2-\delta_w)l_{ew} + \sum_{a \text{ dotted arrow at left side}} l_{ea}(i_a-1),$$ where $l_{ea}$ is the product of the edge decorations adjacent to the path from $e$ to $a$ but not on it. This product does not use the decorations on $e$ itself. We obtain the diagram $\Gamma_L$ by removing all the nodes and edges at the side of $v_R$ and add two arrows and a node as in Figure \[fig:splicingexpl:GammaL\]. Analogously we have $\Gamma_R$ as in Figure \[fig:splicingexpl:GammaR\]. This procedure is called splicing (of $\Gamma$ along $e$).
We now state and prove our result.
\[theorem:splicedecompositionformulaformotiviczetafucntions\] Consider a diagram $\Gamma$ and the splicing of $\Gamma$ into $\Gamma_L$ and $\Gamma_R$. Then we have $${Z_{\Gamma}(T)} = {Z_{\Gamma_L}(T)} + {Z_{\Gamma_R}(T)} - \frac{({\ensuremath{\mathbb{L}}}- 1)^2 T^{M + M'}}{ ({\ensuremath{\mathbb{L}}}^i - T^M)({\ensuremath{\mathbb{L}}}^{i'} - T^{M'})}.$$
Define the diagram $\tilde{\Gamma}$ as follows:
(vL) at (-[2]{},0) ; (vR) at ([2]{},0) ; (vLL) at (-[4]{},0) ; (vRR) at ([4]{},0) ; (labelup) at (0,.3); (labeldown) at (0,-.3);
(down) at (0,-.1); (up) at (0,.05);
(vLu) at ($(vL) + (up)$); (vLLu) at ($(vLL) + (up)$); (vRu) at ($(vR) + (up)$); (vRRu) at ($(vRR) + (up)$);
(vLd) at ($(vL) + (down)$); (vLLd) at ($(vLL) + (down)$); (vRd) at ($(vR) + (down)$); (vRRd) at ($(vRR) + (down)$);
(vLLu) – (vLu) node\[ near end, above\][$D$]{}; (vLLd) – (vLd); (vL) – (vR) node\[very near start, above\][$d$]{} node\[very near end, above\][$d'$]{} node\[midway, below\][$\tilde{e}$]{}; (vRu) – (vRRu) node\[ near start, above\][$D'$]{}; (vRd) – (vRRd); (vR) circle \[radius=3pt\]; (vL) circle \[radius=3pt\];
at ($(vLLu) + (labelup)$) [$(M')$]{}; at ($(vRRu) + (labelup)$) [$(M)$]{};
at ($(vLLd) + (labeldown)$) [$i'-1$]{}; at ($(vRRd) + (labeldown)$) [$i-1$]{};
where $\tilde{e}$ is the edge between the nodes. Remark that this is the dual graph of a (toric) resolution of the polynomial $f = x^My^{M'}$ and the differential form $\omega = x^{i-1}y^{i'-1} {dx} \wedge {dy}$. Hence ${Z_{\tilde{\Gamma}}(T)} = \frac{({\ensuremath{\mathbb{L}}}- 1)^2 T^{M + M'}}{ ({\ensuremath{\mathbb{L}}}^i - T^M)({\ensuremath{\mathbb{L}}}^{i'} - T^{M'})}$ and thus we need to show that $$\label{eq:relbetweenmotivc}
{Z_{\Gamma}(T)} + {Z_{\tilde{\Gamma}}(T)} = {Z_{\Gamma_L}(T)} + {Z_{\Gamma_R}(T)}.$$ We prove this by considering suitable realizable refinements of these diagrams. First, take a realizable refinement $\Gamma'$ of $\Gamma$. This is drawn in Figure \[fig:Gamma\], where you have the division into parts
node\[color=[red]{}, rectangle, rounded corners,inner sep=2,outer sep=0, draw=[red]{}\] at (0,0)[A]{} ;
,
node\[color=[green]{}, rectangle, rounded corners,inner sep=2,outer sep=0, draw=[green]{}\] at (0,0)[B]{} ;
and
node\[color=[blue]{}, rectangle, rounded corners,inner sep=2,outer sep=0, draw=[blue]{}\] at (0,0)[C]{} ;
. Remark that the edges crossing the border belong to
node\[color=[green]{}, rectangle, rounded corners,inner sep=2,outer sep=0, draw=[green]{}\] at (0,0)[B]{} ;
.
Second, take a realizable refinement $\tilde{\Gamma}'$ of $\tilde{\Gamma}$ such that the edge $e$ of $\Gamma$ has the same refinement as the edge $\tilde{e}$ of $\tilde{\Gamma}$. Hence we can consider Figure \[fig:GammaT\] where we have a division into
node\[color=[black]{}, rectangle, rounded corners,inner sep=2,outer sep=0, draw=[black]{}\] at (0,0)[D]{} ;
,
node\[color=[green]{}, rectangle, rounded corners,inner sep=2,outer sep=0, draw=[green]{}\] at (0,0)[B]{} ;
and
node\[color=[darkbrown]{}, rectangle, rounded corners,inner sep=2,outer sep=0, draw=[darkbrown]{}\] at (0,0)[E]{} ;
where the subdiagram
node\[color=[green]{}, rectangle, rounded corners,inner sep=2,outer sep=0, draw=[green]{}\] at (0,0)[B]{} ;
is the same as in Figure \[fig:Gamma\].
We glue these refinements together to obtain refinements $\Gamma_L'$ and $\Gamma_R'$ of $\Gamma_L$ and $\Gamma_R$. This is shown in Figures \[fig:GammaL\] and \[fig:GammaR\].
We compare the diagrams $\Gamma'$ and $\tilde{\Gamma}'$ to the diagrams $\Gamma_R'$ and $\Gamma_L'$. Both groups contain the subdiagrams
node\[color=[red]{}, rectangle, rounded corners,inner sep=2,outer sep=0, draw=[red]{}\] at (0,0)[A]{} ;
,
node\[color=[green]{}, rectangle, rounded corners,inner sep=2,outer sep=0, draw=[green]{}\] at (0,0)[B]{} ;
,
node\[color=[green]{}, rectangle, rounded corners,inner sep=2,outer sep=0, draw=[green]{}\] at (0,0)[B]{} ;
,
node\[color=[blue]{}, rectangle, rounded corners,inner sep=2,outer sep=0, draw=[blue]{}\] at (0,0)[C]{} ;
,
node\[color=[black]{}, rectangle, rounded corners,inner sep=2,outer sep=0, draw=[black]{}\] at (0,0)[D]{} ;
and
node\[color=[darkbrown]{}, rectangle, rounded corners,inner sep=2,outer sep=0, draw=[darkbrown]{}\] at (0,0)[E]{} ;
. Hence if we calculate the motivic zeta function, where we take the sum over all nodes and edges in these subdiagrams, we have proven (\[eq:relbetweenmotivc\]) if we show that the corresponding nodes have the same multiplicities. But this is easily seen by using (\[eq:formulafornodes\]), (\[eq:formulafordifferentialform\]), (\[eq:splicing:multiplicityoff\]), (\[eq:splicing:multiplicityofomega\]) and (\[eq:splicing:multiplicityofomegap\]).
We can connect our proof to the splicing of links. Recall that splicing consists of taking two links, where in each a knot is selected, removing a tubular neighborhood around these knots and glueing the remainders together. However if you glue these removed tubular neighbourhoods together, you find the link of $\Gamma'$.
Algebraic dependence of the monodromic motivic zeta function
============================================================
Following Theorem \[theorem:splicedecompositionformulaformotiviczetafucntions\], we want to define a monodromic zeta function for a diagram. But it turns out that this is not possible. Consider for this the family of polynomials $$f_\lambda = xy^2(x-y)(x-\lambda y) \in {\ensuremath{\mathbb{C}}}[x,y],$$ where $\lambda \in {\ensuremath{\mathbb{C}}}\setminus \{0,1\}$. The splice diagram associated to this family of polynomials is independent of $\lambda$. But we do have the following result.
\[proposition:algebraicdependenceofmonodromiczetafunctions\] There exist $\lambda, \lambda' \in {\ensuremath{\mathbb{C}}}\setminus\{0,1\}$ such that $${{\ensuremath{Z^{{\ensuremath{\hat{\mu }}}}_{f_\lambda}(T)}}} \neq {{\ensuremath{Z^{{\ensuremath{\hat{\mu }}}}_{f_{\lambda'}}(T)}}}.$$
This results shows that a monodromic motivic zeta function cannot be defined for a diagram since the diagrams for this family of polynomials are the same.
- Define the Grothendieck group of abelian varieties ${\ensuremath{{A}_{{\ensuremath{\mathbb{C}}}}}}$ as the abelian group generated by isomorphism classes of abelian varieties with the relations ${\ensuremath{ [A {\oplus}B ]}} = {\ensuremath{ [ A]}} + {\ensuremath{ [B]}} $ and
- define ${\ensuremath{{A^0_{{\ensuremath{\mathbb{C}}}}}}}$ as the abelian group generated by isogeny classes of abelian varieties and relations ${\ensuremath{ [A{\oplus}B]}} = {\ensuremath{ [A]}} + {\ensuremath{ [B]}} $.
The structure of ${\ensuremath{{A^0_{{\ensuremath{\mathbb{C}}}}}}}$ is easier since Poincaré’s complete irreducibility theorem [@mumford1974abelian p. 173] implies that ${\ensuremath{{A^0_{{\ensuremath{\mathbb{C}}}}}}}$ is isomorphic to the free abelian group on simple abelian varieties. Hence equality of the classes of two abelian varieties in ${\ensuremath{{A^0_{{\ensuremath{\mathbb{C}}}}}}}$ implies that they are isogenous.
In the appendix we describe a group morphism $\widetilde{\operatorname{Pic}} : {\ensuremath{\hat{\mathcal{M}}_{{\ensuremath{\mathbb{C}}}}}} \to {\ensuremath{{A}_{{\ensuremath{\mathbb{C}}}}}} $. This morphism sends the class of a smooth complete variety to the class of its Jacobian. We will use this morphism in the following proof, where we compose it with the forgetful morphism ${\ensuremath{{A}_{{\ensuremath{\mathbb{C}}}}}} \to {\ensuremath{{A^0_{{\ensuremath{\mathbb{C}}}}}}}$.
Let $\pi : X \to {\ensuremath{\mathbb{A}^{2}_{{\ensuremath{\mathbb{C}}}}}}$ be the blowup at the origin. This is an embedded resolution of singularities for $f_\lambda$. The formula of Denef-Loeser then tells us that we need to prove that the class of $$\tilde{ E}^\circ_\lambda = \{(x,y) \in {\ensuremath{\mathbb{C}}}^2 \mid y^5 = x(x-1)(x-\lambda), y \neq 0\}$$ does depend on $\lambda$.
We will now prove that if two varieties are equal in ${\ensuremath{\hat{\mathcal{M}}_{{\ensuremath{\mathbb{C}}}}}}$, then their jacobians are isogenous. Combining this with [@dejongNoot Proposition 2.7] we find that the set $\{{\ensuremath{ [\tilde{ E}^\circ_\lambda]}} \in {\ensuremath{\hat{\mathcal{M}}_{{\ensuremath{\mathbb{C}}}}}} \mid \lambda \in {\ensuremath{\mathbb{C}}}\setminus \{0,1\} \}$ is infinite.
We have a map ${\ensuremath{\hat{\mathcal{M}}_{{\ensuremath{\mathbb{C}}}}}} \stackrel{\tilde\operatorname{Pic}}{\to} {\ensuremath{{A}_{{\ensuremath{\mathbb{C}}}}}} \to {\ensuremath{{A^0_{{\ensuremath{\mathbb{C}}}}}}}$ which sends a complete smooth variety to the class of its jacobian. Using the remark before the start of the proof, we find that equality in ${\ensuremath{\mathcal{M}_{{\ensuremath{\mathbb{C}}}}}}$ implies that their jacobians are isogenous.
This has implications on how to formulate a splicing formula for the monodromic motivic zeta functions. The same proof of Theorem \[theorem:splicedecompositionformulaformotiviczetafucntions\] will work if you work with a more general notion of a diagram. In this generalization you need to encode the information of ${\ensuremath{\tilde{E}_I^\circ}}$ more carefully, for example by remembering the locations of the branch points of the cover ${\ensuremath{\tilde{E}_I^\circ}}\to {\ensuremath{\mathbb{P}^{1}_{{\ensuremath{\mathbb{C}}}}}}$.
Another way is to formulate the splicing formula in terms of $f$ and $\omega$.
Let $f\in {\ensuremath{\mathbb{C}}}[x,y]$ and let $\omega$ be a differential form. Also fix an edge $e$ in an associated diagram $\Gamma$ for some embedded resolution and denote by $\Gamma_L$ and $\Gamma_R$ the resulting diagrams after splicing along $e$. Then there exists $f_L, f_R \in {\ensuremath{\mathbb{C}}}[x,y]$ and differential forms $\omega_L, \omega_R$, whose splice diagrams are $\Gamma_L$ and $\Gamma_R$ for some embedded resolutions, such that $${{\ensuremath{Z^{{\ensuremath{\hat{\mu }}}}_{f, \omega}(T)}}} = {{\ensuremath{Z^{{\ensuremath{\hat{\mu }}}}_{f_1, \omega_1}(T)}}} + {{\ensuremath{Z^{{\ensuremath{\hat{\mu }}}}_{f_2, \omega_2}(T)}}} -
\frac{({\ensuremath{\mathbb{L}}}- 1)^2 T^{M + M'}}{ ({\ensuremath{\mathbb{L}}}^i - T^M)({\ensuremath{\mathbb{L}}}^{i'} - T^{M'})}.$$
Monodromy conjecture\[section:mc\]
==================================
In this section we discuss how we can lift the results in [@NemethiVeys2012] about the generalized monodromy conjecture to the motivic level. Recall that we consider our polynomial $f$ as a germ $({\ensuremath{\mathbb{A}^{2}_{{\ensuremath{\mathbb{C}}}}}},0) \to ({\ensuremath{\mathbb{A}^{1}_{{\ensuremath{\mathbb{C}}}}}},0)$. Let $F_0$ be the Milnor fiber of this germ, $h_i : H_i(F_0, {\ensuremath{\mathbb{C}}}) \to H_i(F_0, {\ensuremath{\mathbb{C}}})$ the algebraic monodromy ($i=0,1$), $\Delta_i = \det(tI - h_i)$ the characteristic polynomial of $h_i$ and $\zeta(t) = \Delta_1\slash \Delta_0$.
Given a diagram $\Gamma$, we define the formula $$\zeta_\Gamma(t) = \prod_{ {v \text{ is a node}}} \left( t^{N_v} - 1 \right)^{\delta'_v}$$ where $\delta'_v$ is the valency of $v$ considered in the graph including the arrows corresponding to strict transforms of $f$, but without the dotted arrows corresponding to the strict transforms of $\operatorname{Supp}\omega$. This recovers the monodromy zeta function of $f$. Remark that $\Delta_0 = t^d-1$ where $d = \gcd_{a \text{ is an arrow}} N_a $, and thus we can discuss monodromy eigenvalues of a diagram.
In what follows it will be implicit that monodromy eigenvalues are the monodromy eigenvalues of our fixed $f$.
Motivic zeta function
---------------------
Némethi and Veys defined a differential form $\omega$ on a diagram $\Gamma $ to be allowed if the following conditions are satisfied:
- $(N_a,i_a) \neq (0,0)$ for all (dotted) arrows.
- each star-shaped subdiagram with the induced decorations obtained after repeated splicing needs to be allowed.
- if $\Gamma$ is a star-shaped diagram the following condition needs to be satisfied:
Let the central node be connected to $n$ boundary vertices whose supporting edges have decorations $\{d_l\}_{l=1}^n$, and with $r$ other incident edges connecting with arrowheads. Then the decorations $i_1-1, \ldots, i_n-1$ of the dashed arrows at these boundary vertices are subject to the following restrictions:
If $d_l\vert i_l$ for at least $n+r-2$ indices $l\in\{1,\ldots, n\}$, then $i_l = d_l$ for at least $n+r-2$ of the indices $l$.
(RU) at ([3]{},[0.8]{}) ; (RD) at ([3]{},-[0.8]{}) ; (LU) at (-[3]{},[0.8]{}); (LD) at (-[3]{},-[0.8]{}); (MID) at (0,0); (MID)–(RU) node\[near start, above \] [$d_1$]{}; (MID)–(RD) node\[near start, below \] [$d_n$]{}; (MID)–(LU); (MID)–(LD); (RD)–($(RD) + (1,0)$); node at ($(RD) + (1.7,0)$) [$i_1-1$]{} ; node at ($(RU) + (1.7,0)$) [$i_n-1$]{} ; (RU)–($(RU) + (1,0)$); node at (1.5,0.1)[$\vdots$]{} ; node at (-1.5,0.1)[$\vdots$]{}; node at (-[3]{}, -2) [$r$ arrowheads ]{}; node at (-[3]{}, -2.5) [which might be doublearrows]{}; node at ([3]{}, -2) [$n$ boundary vertices]{}; (MID) circle \[radius=3pt\]; (RD) circle \[radius=3pt\]; (RU) circle \[radius=3pt\];
Némethi and Veys showed that this set of allowed forms has the following properties:
- for every allowed form $\omega$ and every pole $s_0$ of ${Z_{f,\omega}^{\text{top}}(s)}$, $\exp(2\pi i s_0)$ is a monodromy eigenvalue.
- $dx\wedge dy$ is allowed;
- for every monodromy eigenvalue $\lambda$, there is an allowed form $\omega $ such that ${Z_{f,\omega}^{\text{top}}(s)}$ has a pole $s_0$ and $\lambda=\exp(2\pi i s_0)$.
To state the generalized monodromy conjecture for the motivic zeta function we need the notion of a pole. This has been done by Rodrigues and Veys in [@MR1978573]. We will use a more direct approach.
We call $s \in {\ensuremath{\mathbb{Q}}}$ a pole of ${Z_{f, \omega}(T)}$ if there exists no set $U \subseteq {{\ensuremath{\mathbb{Z}}}}_{\geq 0}\times {{\ensuremath{\mathbb{Z}}}}_{\geq 0} $ such that $${Z_{f, \omega}(T)} \in
{\ensuremath{\hat{\mathcal{M}}_{{\ensuremath{\mathbb{C}}}}}}\left[\frac{T^N}{{\ensuremath{\mathbb{L}}}^\nu - T^N}\right]_{(v,N) \in U} \subseteq {\ensuremath{\hat{\mathcal{M}}_{{\ensuremath{\mathbb{C}}}}}}[[T]].$$ and such that $s\neq -\frac{\nu}{N}$ for all $(\nu,N) \in U$.
Because ${\ensuremath{\hat{\mathcal{M}}_{{\ensuremath{\mathbb{C}}}}}}$ is not a domain [@Ekedahl2009], we are careful in formulating the next theorem.
\[thm:generalizedmotivicmonodromyconjecture\] Let $f \in {\ensuremath{\mathbb{C}}}[x,y]$ and $\omega$ an allowed form. Then $${Z_{f, \omega}(T)} \in
{\ensuremath{\hat{\mathcal{M}}_{{\ensuremath{\mathbb{C}}}}}}\left[\frac{T^N}{{\ensuremath{\mathbb{L}}}^\nu - T^N}\right]_{(v,N) \in S} \subseteq {\ensuremath{\hat{\mathcal{M}}_{{\ensuremath{\mathbb{C}}}}}}[[T]].$$ where $S = \{(\nu,N) \in {{\ensuremath{\mathbb{Z}}}}_{\geq 0}\times {{\ensuremath{\mathbb{Z}}}}_{\geq 0} \mid (\nu,N) \neq (0,0), \exp( - 2\pi i\frac{\nu}{N}) \text{ is a monodromy eigenvalue of }f\}$.
The proof goes as in the case of the topological zeta function, where you need first to consider star-shaped realizable diagrams. In this case you need to prove that $\alpha_i= -N_i\frac{\nu}{N} + \nu_i = 1$ for sufficiently many $i$.
The general case can be proved in the same way as in the proof for the topological zeta function.
By specializing to the topological zeta function, we easily find that every monodromy eigenvalue is obtained as a pole. Hence we have proven the generalized monodromy conjecture for the motivic zeta function.
\[cor:generalizedmonodromyconjectureformotiviczetafunctions\] Consider the set of allowed forms for a diagram $\Gamma$ of $f \in {\ensuremath{\mathbb{C}}}[x,y]$. It satisfies the following conditions:
- for every allowed form $\omega$, every pole of $ {Z_{f, \omega}(T)}$ induces a monodromy eigenvalue of $f$. More specifically Theorem \[thm:generalizedmotivicmonodromyconjecture\] holds.
- $dx\wedge dy$ is allowed;
- every monodromy eigenvalue is obtained as a pole of the motivic zeta function of $f$ with respect to $\omega$.
Monodromic motivic zeta function
--------------------------------
Theorem \[thm:generalizedmotivicmonodromyconjecture\] however does not generalize to the case of the monodromic motivic zeta function. This is due to the fact that an allowed form is made such that the residues of candidate ‘bad’ poles of the topological zeta functions are zero. We give here examples of twisted topological zeta functions because a pole of a twisted topological zeta function will also be a pole of the monodromic motivic zeta function.
We give an example where the twisted topological zeta function has a pole which does not induce a monodromy eigenvalue, and thus an analogue of Corollary \[cor:generalizedmonodromyconjectureformotiviczetafunctions\] cannot hold.
Consider the cusp $f=x^3 - y^2$ , the minimal resolution $\pi$ and differential form $\omega= x^3y^3dx\wedge dy$. Then its splice diagram is the following:
(een) at (-[3]{}, 0); (twee) at (0, 0); (drie) at ([3]{}, 0); (vier) at (0, [2]{}); (vijf) at ( -[3]{}-[2]{},0); (zes) at ( +[3]{}+[2]{},0); (een)–(twee) node\[very near start, above\][$1$]{} node\[very near end, above\][$3$]{}; (twee)–(drie) node\[very near start, above\][$2$]{} node\[very near end, above\][$2$]{}; (een) circle \[radius=3pt\]; (twee) circle \[radius=3pt\]; (drie) circle \[radius=3pt\]; (twee)–(vier); node\[right\] at (vier) [(1)]{}; node\[above\] at (vijf) [(3)]{}; node\[above\] at (zes) [(3)]{}; (een)–(vijf); (drie)–(zes);
This form $\omega$ is allowed but we find that $${Z_{f,\omega}^{\text{top}, (6)}(s)} = -\frac{1}{6s + 21},$$ which has $s_0= -\frac{7}{2}$ as a pole. But $\exp(2\pi i s_0) = -1$ is not a monodromy eigenvalue.
One can wonder if a suitable subset of allowed forms can achieve Corollary \[cor:generalizedmonodromyconjectureformotiviczetafunctions\]. The following example shows however that we cannot obtain all poles.
Consider the polynomial $f= (y^3 - x^4)^5 + x^2y^{15}$. We have the following diagram $\Gamma$
(een) at (0,1.5); (twee) at (0,0); (drie) at (0,-1.5); (vier) at (3,0); (vijf) at (5,0); (zes) at (3 ,-1.5); (zeven) at (0,3); (acht) at (0,-3); (negen) at (6.5,0); (diff) at (.06,0); (een) circle \[radius=3pt\]; (twee) circle \[radius=3pt\]; (drie) circle \[radius=3pt\]; (vier) circle \[radius=3pt\]; (vijf) circle \[radius=3pt\]; (een)–(twee) node\[very near end, above, left\][$3$]{}; (drie)–(twee) node\[very near end, left\][$4$]{}; (twee)–(vier) node\[very near end, above\][$66$]{} node\[very near start, above\][$1$]{}; (vier)–(vijf) node\[very near start, above\][$5$]{}; ($(vier) - (diff)$)–($(zes) - (diff)$) node\[very near start, left\][$1$]{} node\[very near end, left\][$(1)$]{}; (een)–(zeven) node\[very near end, right\][$i_1-1$]{}; (drie)–(acht) node\[very near end, right\][$i_2-1$]{}; ($(vier) + (diff)$)–($(zes) + (diff)$) node\[very near end, right\][$k-1$]{}; (vijf)–(negen) node\[very near end, above\][$i_3-1$]{};
,
where we are already considering some form with $i_1, i_2, i_3, k \in {{\ensuremath{\mathbb{Z}}}}$. The monodromy zeta function is $$\zeta(T) = \frac{(T^{330} - 1)(T^{60} - 1)}{(T^{66} - 1)(T^{15} - 1)(T^{20} - 1)},$$which implies that $\lambda = \exp(2\pi i \frac{1}{110})$ is a monodromy eigenvalue. However there exist no $i_1, i_2, i_3, k$ such that ${Z_{\Gamma}^{\text{top}, (330)}(s)}$ has a pole $s_0$ with the condition $\lambda = \exp(2\pi i s_0)$ and such that $\lambda' = \exp(2\pi i s'_0) $ is a monodromy eigenvalue whenever $s_0'$ is the pole of $ {Z_{\Gamma}^{\text{top}, (60)}(s)}$. Indeed, if $\lambda$ is a pole, then $s_0$ needs to be a pole of ${Z_{\Gamma}^{\text{top}, (330)}(s)}$. This implies that $ 2 i_1 + 3i_2 \equiv 3 \pmod{ 6 }$. But now the pole of $ {Z_{\Gamma}^{\text{top}, (60)}(s)}$ will not induce a monodromy eigenvalue since $i_1 \equiv 0 \pmod 3$ and $i_2 \equiv 1 \pmod 2$.
Existence of the Picard morphism
================================
It turns out that the class of a smooth and proper algebraic variety in the Grothendieck ring of varieties determines the class of its Picard scheme. In this appendix we will define and prove this statement using the argument described in [@Ekedahl2009]. We do this since [@Ekedahl2009] was never published and to clarify several steps in his argument.
Let $X$ be a smooth and proper algebraic variety over ${\ensuremath{\mathbb{C}}}$. Recall that the Picard functor $\operatorname{Pic}_{X\slash {\ensuremath{\mathbb{C}}}}$ is a functor from the category of locally Noetherian ${\ensuremath{\mathbb{C}}}$-schemes to the category of abelian groups defined by the formula $$\operatorname{Pic}_{X\slash {\ensuremath{\mathbb{C}}}}(T) = \operatorname{Pic}(X_T) \slash \operatorname{Pic}(T)$$ where $X_T = X \times_{\ensuremath{\mathbb{C}}}T$. It turns out that the associated fppf sheaf is representable by an abelian scheme whose identity component is an abelian variety and whose group of geometric components is finitely generated. This scheme will be denoted by $\operatorname{Pic}(X)$. See [@fantechi2005fundamental Part 5] for more information on the Picard scheme.
Define $ {\ensuremath{\mathcal{A}_{{\ensuremath{\mathbb{C}}}}}}$ as the abelian group generated by isomorphism classes of commutative algebraic group schemes over ${\ensuremath{\mathbb{C}}}$ whose identity component is an abelian variety and whose group of geometric components is finitely generated, subject to the relations ${\ensuremath{ [A {\oplus}B ]}} = {\ensuremath{ [ A]}} + {\ensuremath{ [B]}} $.
This leads us to the main result.
\[theorem:existenceofPicardscheme\] There is a (unique) group homomorphism $\operatorname{Pic}: {\ensuremath{K_0(\operatorname{Var}_{{{\ensuremath{\mathbb{C}}}}})}} \to {\ensuremath{\mathcal{A}_{{\ensuremath{\mathbb{C}}}}}}$ such that $\operatorname{Pic}( {\ensuremath{ [X]}}) = {\ensuremath{ [\operatorname{Pic}(X)]}}$ for every smooth proper variety $X$, and it extends to a morphism $\operatorname{Pic}: {\ensuremath{\hat{\mathcal{M}}_{{\ensuremath{\mathbb{C}}}}}} \to {\ensuremath{\mathcal{A}_{{\ensuremath{\mathbb{C}}}}}}$.
This theorem provides us with a new technique to compare elements in the Grothendieck ring.
Recall that the Grothendieck group of abelian varieties ${\ensuremath{{A}_{{\ensuremath{\mathbb{C}}}}}}$ is defined as the abelian group generated by isomorphism classes of abelian varieties with the relations ${\ensuremath{ [A {\oplus}B ]}} = {\ensuremath{ [ A]}} + {\ensuremath{ [B]}} $. Using this theorem, we find the existence of $\widetilde{\operatorname{Pic}} : {\ensuremath{\hat{\mathcal{M}}_{{\ensuremath{\mathbb{C}}}}}} \to {\ensuremath{{A}_{{\ensuremath{\mathbb{C}}}}}} $ which sends a smooth complete variety to the class of its jacobian. It is obtained by composing $\operatorname{Pic}$ and the morphism ${\ensuremath{\mathcal{A}_{{\ensuremath{\mathbb{C}}}}}} \to {\ensuremath{{A}_{{\ensuremath{\mathbb{C}}}}}} $, where this last map is defined by sending the class of a commutative algebraic group scheme (whose identity component is an abelian variety) to the class of his identity component.
The keystone of the proof is Bittner’s presentation of $ {\ensuremath{K_0(\operatorname{Var}_{{{\ensuremath{\mathbb{C}}}}})}} $, which we restate here.
[[@MR2059227 Theorem 3.1] ]{} The Grothendieck group of ${\ensuremath{\mathbb{C}}}$-varieties ${\ensuremath{K_0(\operatorname{Var}_{{{\ensuremath{\mathbb{C}}}}})}}$ is isomorphic to the abelian group generated by the isomorphism classes of smooth projective ${\ensuremath{\mathbb{C}}}$-varieties subject to the relations ${\ensuremath{ [\emptyset]}} = 0 $ and ${\ensuremath{ [\operatorname{Bl}_YX]}} - {\ensuremath{ [E]}} = {\ensuremath{ [X]}} - {\ensuremath{ [Y]}}$, where $X$ is smooth and projective, $Y\subset X$ is a closed smooth subvariety, $\operatorname{Bl}_Y X$ is the blow-up of $X$ along $Y$ and $E$ is the exceptional divisor of this blow-up.
This implies the following presentation of ${\ensuremath{\mathcal{M}_{{\ensuremath{\mathbb{C}}}}}} = {\ensuremath{K_0(\operatorname{Var}_{{{\ensuremath{\mathbb{C}}}}})}}[{\ensuremath{\mathbb{L}}}^{-1}]$:
${\ensuremath{\mathcal{M}_{{\ensuremath{\mathbb{C}}}}}}$ is generated by the elements ${\frac{{\ensuremath{ [X]}}}{{\ensuremath{\mathbb{L}}}^n}}$ where $X$ is smooth and projective, subject to the relations
$${\frac{{\ensuremath{ [X\times {\ensuremath{\mathbb{P}^{1}_{{\ensuremath{\mathbb{C}}}}}}]}} }{{\ensuremath{\mathbb{L}}}^{n+1} }}
= {\frac{{\ensuremath{ [X]}} }{{\ensuremath{\mathbb{L}}}^{n} }} +
{\frac{{\ensuremath{ [X ]}}}{{\ensuremath{\mathbb{L}}}^{n+1} }},$$ where $X$ is smooth and projective, and the relations $ {\frac{{\ensuremath{ [\operatorname{Bl}_YX]}}}{{\ensuremath{\mathbb{L}}}^n}} - {\frac{{\ensuremath{ [E]}}}{{\ensuremath{\mathbb{L}}}^n}} = {\frac{{\ensuremath{ [X]}}}{{\ensuremath{\mathbb{L}}}^n}} - {\frac{{\ensuremath{ [Y]}}}{{\ensuremath{\mathbb{L}}}^n}}$, where $X$, $Y$ and $E$ are as in the theorem.
- We first show that the morphism defined by $\operatorname{Pic}( {\ensuremath{ [X]}}) = {\ensuremath{ [\operatorname{Pic}(X)]}}$ for $X$ smooth and proper is well-defined as a map from ${\ensuremath{K_0(\operatorname{Var}_{{{\ensuremath{\mathbb{C}}}}})}}$ to $ {\ensuremath{\mathcal{A}_{{\ensuremath{\mathbb{C}}}}}}$. Using the Bittner representation, we need to show that $$\operatorname{Pic}(\operatorname{Bl}_YX) {\oplus}\operatorname{Pic}(Y) = \operatorname{Pic}(X) {\oplus}\operatorname{Pic}(E)$$ where $X$, $Y$ and $E$ are as in the theorem. Hence we need to show this on the level of associated fppf sheaves. But a blow-up is preserved under base change by a flat morphism and thus $\operatorname{Pic}(\operatorname{Bl}_{Y_T} X_T) = \operatorname{Pic}(X_T ) {\oplus}{{\ensuremath{\mathbb{Z}}}}$ and $\operatorname{Pic}(E_T) = \operatorname{Pic}( Y_T ) {\oplus}{{\ensuremath{\mathbb{Z}}}}$ for any flat morphism $T \to {\ensuremath{\mathbb{C}}}$ which induces the wanted isomorphism [@MR0463157 Exercises 7.9 and 8.5 on pages 170 and 188].
- The next step is to define $\operatorname{Pic}' : {\ensuremath{\mathcal{M}_{{\ensuremath{\mathbb{C}}}}}} \to {\ensuremath{\mathcal{A}_{{\ensuremath{\mathbb{C}}}}}}$ and prove that this is actually an extension of $\operatorname{Pic}$.
Define $\operatorname{Pic}'({\ensuremath{ [X]}} \slash {\ensuremath{\mathbb{L}}}^n )$ as $${\ensuremath{A_{X}^{0,n}}} {\oplus}{\ensuremath{A_{X}^{c,n}}}$$ for a projective and smooth variety $X$ and $n \in {\ensuremath{\mathbb{N}}}$, where
- ${\ensuremath{A_{X}^{c,n}}}$ is the inverse image of the classes of type $(n+1, n+1)$ under the map ${\ensuremath{H^{2n+2}(X,{{\ensuremath{\mathbb{Z}}}})}} \to {\ensuremath{H^{2n+2}(X,{\ensuremath{\mathbb{C}}})}}$ and
- ${\ensuremath{A_{X}^{0,n}}}$ is the Weil intermediate Jacobian associated to the Hodge structure on ${\ensuremath{H^{2n+1}(X,{{\ensuremath{\mathbb{Z}}}})}}$. See [@1952] for more information.
As discussed, we need to verify the two types of relations. Consider $X$ to be a projective smooth ${\ensuremath{\mathbb{C}}}$-variety.
- First we verify the blow-up relations. This is a consequence of [@voisin2002hodge Theorem 7.31] and its proof since it induces that $${\ensuremath{H^{k}(X,{{\ensuremath{\mathbb{Z}}}})}} {\oplus}\left( {\bigoplus}_{i=0}^{r-2} {\ensuremath{H^{k-2i - 2}(Y,{{\ensuremath{\mathbb{Z}}}})}} \right) {\oplus}{\ensuremath{H^{k}(Y,{{\ensuremath{\mathbb{Z}}}})}}$$ is both isomorphic to ${\ensuremath{H^{k}(\operatorname{Bl}_Y X,{{\ensuremath{\mathbb{Z}}}})}} {\oplus}{\ensuremath{H^{k}(Y,{{\ensuremath{\mathbb{Z}}}})}}$ and ${\ensuremath{H^{k}(X ,{{\ensuremath{\mathbb{Z}}}})}} {\oplus}{\ensuremath{H^{k}(E,{{\ensuremath{\mathbb{Z}}}})}}$ as Hodge structures, where $r$ is the codimension of $Y$ in $X$.
- Remark that the cup-products map [@voisin2002hodge Theorem 11.38] for $X$ and $ {\ensuremath{\mathbb{P}^{1}_{{\ensuremath{\mathbb{C}}}}}}$ induces $${\ensuremath{H^{i+2}(X \times {\ensuremath{\mathbb{P}^{1}_{{\ensuremath{\mathbb{C}}}}}} ,{{\ensuremath{\mathbb{Z}}}})}} \cong {\ensuremath{H^{i+2}(X,{{\ensuremath{\mathbb{Z}}}})}} {\oplus}{\ensuremath{H^{i}(X,{{\ensuremath{\mathbb{Z}}}})}}$$ as Hodge structures [@MR2050072 p. 32].
Since the relations hold on the level of Hodge structures, they also hold for $\operatorname{Pic}'$ and thus $\operatorname{Pic}'$ is well-defined.
- We show now that $\operatorname{Pic}'$ is indeed an extension of $\operatorname{Pic}$ and thus we will show that $\operatorname{Pic}(X) \cong {\ensuremath{A_{X}^{0,0}}} {\oplus}{\ensuremath{A_{X}^{c,0}}}$. Remark that the connected component $\operatorname{Pic}^0(X)$ is the classical Jacobian of $X$ and thus isomorphic to ${\ensuremath{A_{X}^{0,0}}}$.
The Néron-Severi group ${\ensuremath{\operatorname{NS}(X)}}$ is defined by the short exact sequence $$0 \to \operatorname{Pic}^0(X) \to \operatorname{Pic}(X) \to \operatorname{NS}(X) \to 0$$ and is the group of components of $\operatorname{Pic}(X)$. This group can also be identified with the image of $ d : {\ensuremath{H^{1}(X,\mathcal{O}_X^\times)}} \to {\ensuremath{H^{2}(X,{{\ensuremath{\mathbb{Z}}}})}}$. Now Lefschetz’s theorem on $(1,1)$-classes [@griffiths2011principles p.163] implies that $\operatorname{NS}(X)$ is exactly ${\ensuremath{A_{X}^{c,0}}}$.
Since we are working over an algebraically closed field and $\operatorname{NS}(X)$ is a discrete finitely generated abelian group, the short exact sequence splits and thus $\operatorname{Pic}(X) \cong \operatorname{Pic}^0(X) {\oplus}\operatorname{NS}(X)$.
- To conclude we remark that ${\ensuremath{H^{n}(X,{{\ensuremath{\mathbb{Z}}}})}} = 0$ if $n > 2 {\ensuremath{\operatorname{dim}X}}$ and thus $\operatorname{Pic}'{{\frac{{\ensuremath{ [X]}}}{{\ensuremath{\mathbb{L}}}^n}}} = 0$ if $\dim(X) -n \leq 0 $. This implies that $\operatorname{Pic}'$ sends every element of $F^0$ to $0$ and thus $\operatorname{Pic}' $ can be extended to ${\ensuremath{\hat{\mathcal{M}}_{{\ensuremath{\mathbb{C}}}}}}$.
One of the results Ekedahl obtains with this is the fact that ${\ensuremath{\hat{\mathcal{M}}_{{\ensuremath{\mathbb{C}}}}}}$ is not a domain.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'A graph $G = (V,E)$ is a *double-threshold graph* if there exist a vertex-weight function $w \colon V \to {\mathbb{R}}$ and two real numbers ${\mathtt{lb}}, {\mathtt{ub}}\in {\mathbb{R}}$ such that $uv \in E$ if and only if ${\mathtt{lb}}\le {\mathtt{w}}(u) + {\mathtt{w}}(v) \le {\mathtt{ub}}$. In the literature, those graphs are studied as the pairwise compatibility graphs that have stars as their underlying trees. We give a new characterization of double-threshold graphs, which gives connections to bipartite permutation graphs. Using the new characterization, we present a linear-time algorithm for recognizing double-threshold graphs. Prior to our work, the fastest known algorithm by Xiao and Nagamochi \[COCOON 2018\] ran in $O(n^6)$ time, where $n$ is the number of vertices.'
author:
- Yusuke Kobayashi
- Yoshio Okamoto
- Yota Otachi
- Yushi Uno
bibliography:
- 'dtg.bib'
title: 'Linear-Time Recognition of Double-Threshold Graphs'
---
Introduction {#sec:introduction}
============
A graph is a *threshold graph* if there exist a vertex-weight function and a real number called a weight lower bound such that two vertices are adjacent in the graph if and only if the associated vertex weight sum is at least the weight lower bound. Threshold graphs and their generalizations are well studied because of their beautiful structures and applications in many areas [@MahadevP1995; @Golumbic04]. In particular, the edge-intersections of two threshold graphs, and their complements (i.e., the union of two threshold graphs) have attracted several researchers in the past, and recognition algorithms with running time $O(n^5)$ by Ma [@Ma93], $O(n^4)$ by Raschle and Simon [@RaschleS95], and $O(n^3)$ by Sterbini and Raschle [@SterbiniR98] have been developed, where $n$ is the number of vertices.
In this paper, we study the class of double-threshold graphs, which is a proper generalization of threshold graphs and a proper specialization of the graphs that are edge-intersections of two threshold graphs. A graph is a *double-threshold graph* if there exist a vertex-weight function and two real numbers called lower and upper bounds of weight such that two vertices are adjacent if and only if their weight sum is at least the lower bound and at most the upper bound. To the best of our knowledge, this natural generalization of threshold graphs was not studied until quite recently. In 2018, Xiao and Nagamochi [@XiaoN18] studied this graph class under the different name of “star pairwise compatibility graphs” and presented an $O(n^6)$-time recognition algorithm.
Our main result is to give a new characterization of double-threshold graphs that gives a simple linear-time recognition algorithm. We first show that every double-threshold graph is a permutation graph (but not vice versa) and that a bipartite graph is a double-threshold graph if and only if it is a permutation graph. These facts imply that many NP-hard graph problems are polynomial-time solvable on (bipartite or non-bipartite) double-threshold graphs. We then show that a graph is a double-threshold graph if and only if an auxiliary graph constructed from the original graph is a bipartite permutation graph. This characterization gives a linear-time algorithm for recognizing double-threshold graphs.
Recently, we have realized that Jamison and Sprague [@JamisonS19] have independently showed that all double-threshold graphs are permutation graphs and that all bipartite permutation graphs are double-threshold graphs. Their proofs are based on a vertex-ordering characterization of permutation graphs and a BFS structure of bipartite permutation graphs, while ours are direct transformations between vertex weights and permutation diagrams. Note that in their paper Jamison and Sprague [@JamisonS19] used the term *bi-threshold graphs* instead of double-threshold graphs. However, the name of “bi-threshold graphs” is already used[^1] by Hammer and Mahadev [@HammerM85] for a different generalization of threshold graphs (see below). Thus, even though “bi-threshold” would sound better and probably more appropriate, we would like to keep our term “double-threshold” in this paper.
#### Other generalizations of threshold graphs. {#other-generalizations-of-threshold-graphs. .unnumbered}
There are many other generalizations of threshold graphs such as threshold signed graphs [@BenzakenHW85], threshold tolerance graphs [@MonmaRT88], quasi-threshold graphs (also known as trivially perfect graphs) [@YanCC96], weakly threshold graphs [@Barrus18], and paired threshold graphs [@RavanmehrPBM18]. We omit the definitions of these graph classes and only note that some small graphs show that these classes are incomparable to the class of double-threshold graphs (e.g., $3K_{2}$ and bull for threshold signed graphs, $2K_{2}$ and bull for threshold tolerance graphs, $C_{4}$ and $2 K_{3}$ for quasi-threshold graphs, $2K_{2}$ and bull for weakly threshold graphs, $C_{4}$ and bull for paired threshold graphs[^2]). For the class of bithreshold graphs introduced by Hammer and Mahadev [@HammerM85] (not the one introduced by Jamison and Sprague [@JamisonS19]), we can use $3K_{2}$ and bull to show that this class is incomparable to the class of double-threshold graphs. It is known that $3K_{2}$ is not a bithreshold graph [@HammerMP93], while $3K_{2}$ is a double-threshold graph as we will show later (Lemma \[lem:biperm->dtg\]). We will also show later that bull is not a double-threshold graph (Lemma \[lem:mfis\_with\_order5\]). It is not difficult to see that bull is a bithreshold graph just by checking the definition in [@HammerMP93].
Note that the concept of double-threshold digraphs introduced by Hamburger et al. [@HamburgerMPSX18] is concerned with directed acyclic graphs defined from a generalization of semiorders involving two thresholds and not related to threshold graphs or double-threshold graphs.
#### Pairwise compatibility graphs. {#pairwise-compatibility-graphs. .unnumbered}
Motivated by uniform sampling from phylogenetic trees in bioinformatics, Kearney, Munro, and Phillips [@KearneyMP03] defined pairwise compatibility graphs. A graph $G=(V,E)$ is a *pairwise compatibility graph* if there exists a quadruple ($T$, ${\mathtt{w}}$, ${\mathtt{lb}}$, ${\mathtt{ub}}$), where $T$ is a tree, ${\mathtt{w}}\colon E(T) \to {\mathbb{R}}$, and ${\mathtt{lb}}, {\mathtt{ub}}\in {\mathbb{R}}$, such that the set of leaves in $T$ coincides with $V$ and $uv \in E$ if and only if the (weighted) distance $d_{T}(u, v)$ between $u$ and $v$ in $T$ satisfies ${\mathtt{lb}}\le d_{T}(u,v) \le {\mathtt{ub}}$.
Since its introduction, several authors have studied properties of pairwise compatibility graphs, but the existence of a polynomial-time recognition algorithm for that graph class has been open. The survey article by Calamoneri and Sinaimeri [@CalamoneriS16] proposed to look at the class of pairwise compatibility graphs defined on stars (i.e., star pairwise compatibility graphs), and asked for a characterization of star pairwise compatibility graphs. Recently, Xiao and Nagamochi [@XiaoN18] solved the open problem and gave an $O(n^6)$-time algorithm to recognize a star pairwise compatibility graph.
As we will see after the formal definition of double-threshold graphs, the star pairwise compatibility graphs are precisely the double-threshold graphs. Although the pairwise compatibility graphs are rather well studied, we study the double-threshold graphs in the context of threshold graphs and their generalizations. This is because, in our opinion, the double-threshold graphs and techniques used for them are more relevant in that context.
Preliminaries {#sec:preliminaries}
=============
All graphs in this paper are undirected, simple, and finite. A graph $G$ is given by the pair of its vertex set $V$ and its edge set $E$ as $G=(V,E)$. The vertex set and the edge set of $G$ are often denoted by $V(G)$ and $E(G)$, respectively. The *order* of a graph refers to the number of its vertices. For a vertex $v$ in a graph $G = (V,E)$, its *neighborhood* is the set of vertices that are adjacent to $v$, and denoted by $N_{G}(v) = \{ u \mid uv \in E\}$. When the graph $G$ is clear from the context, we often omit the subscript. A linear ordering $\prec$ on a set $S$ with $|S| = n$ can be represented by a sequence $\langle s_{1}, s_{2}, \dots, s_{n} \rangle$ of the elements in $S$, in which $s_{i} \prec s_{j}$ if and only if $i< j$. By abusing the notation, we sometimes write ${\prec} = \langle s_{1}, s_{2}, \dots, s_{n} \rangle$.
Double-threshold graphs
-----------------------
A graph $G=(V,E)$ is a *threshold graph* if there exist a vertex-weight function ${\mathtt{w}}\colon V \to {\mathbb{R}}$ and a real number ${\mathtt{lb}}\in {\mathbb{R}}$ with the following property: $$uv \in E \iff {\mathtt{lb}}\le {\mathtt{w}}(u) + {\mathtt{w}}(v).$$ A graph $G=(V,E)$ is a *double-threshold graph* if there exist a vertex-weight function ${\mathtt{w}}\colon V \to {\mathbb{R}}$ and two real numbers ${\mathtt{lb}}, {\mathtt{ub}}\in {\mathbb{R}}$ with the following property: $$uv \in E \iff {\mathtt{lb}}\le {\mathtt{w}}(u) + {\mathtt{w}}(v) \le {\mathtt{ub}}.$$ Then, we say that the double-threshold graph $G$ is *defined* by ${\mathtt{w}}$, ${\mathtt{lb}}$ and ${\mathtt{ub}}$.
The definition of a double-threshold graph can be understood visually in the plane. See [ \[fig:dtg\_example1\]]{} for an example. In the $xy$-plane, we consider the slab defined by $\{(x,y) \mid {\mathtt{lb}}\leq x+y \leq {\mathtt{ub}}\}$ that is illustrated in gray. Then, two vertices $u,v \in V$ are joined by an edge if and only if the point $({\mathtt{w}}(u),{\mathtt{w}}(v))$ lies in the slab.
![(Left) A double-threshold graph. The weight of each vertex is given as ${\mathtt{w}}(a)=1$, ${\mathtt{w}}(b)=3$, ${\mathtt{w}}(c)=5$, and ${\mathtt{w}}(d)=7$; the lower bound is ${\mathtt{lb}}=4$ and the upper bound is ${\mathtt{ub}}=8$. (Right) Understanding the definition in the $xy$-plane. A white dot represents the point $({\mathtt{w}}(u),{\mathtt{w}}(v))$ for distinct vertices $u,v$, and a cross represents a point $({\mathtt{w}}(v),{\mathtt{w}}(v))$ for a vertex $v$. Two distinct vertices $u$ and $v$ are joined by an edge if and only if the corresponding white dot lies in the gray slab. []{data-label="fig:dtg_example1"}](\figdir/dtg_example1)
Every threshold graph is a double-threshold graph as one can set a dummy upper bound ${\mathtt{ub}}> \max \{{\mathtt{w}}(u) + {\mathtt{w}}(v) \mid u,v \in V\}$. From the definition of double-threshold graphs, we can easily see that they coincide with the star pairwise compatibility graphs.
A graph is a double-threshold graph if and only if it is a star pairwise compatibility graph.
Let $G = (V,E)$ be a double-threshold graph defined by ${\mathtt{w}}\colon V \to {\mathbb{R}}$ and ${\mathtt{lb}}, {\mathtt{ub}}\in {\mathbb{R}}$. We construct an edge-weighted star $S$ with the center $c$ and the leaf set $V$ such that the weight ${\mathtt{w}}'(vc)$ of each edge $vc \in E(S)$ is ${\mathtt{w}}(v)$. Then, $G$ is the star pairwise compatibility graph defined by $(S, {\mathtt{w}}',{\mathtt{lb}},{\mathtt{ub}})$.
Let $G = (V,E)$ be a star pairwise compatibility graph defined by $(S, {\mathtt{w}}, {\mathtt{lb}}, {\mathtt{ub}})$, where the star $S$ has $c$ as its center. For each $v \in V$, we set ${\mathtt{w}}'(v) = {\mathtt{w}}(vc)$. Then, $G$ is the double-threshold graph defined by ${\mathtt{w}}'$, ${\mathtt{lb}}$, and ${\mathtt{ub}}$.
The *threshold dimension* of a graph $G = (V,E)$ is the minimum integer $k$ such that there are $k$ threshold graphs $G_{i} = (V, E_{i})$, $1 \le i \le k$, with $E = \bigcup_{1 \le i \le k} E_{i}$. A graph $G = (V,E)$ has *co-threshold dimension* $k$ if its complement $\overline{G}$ has threshold dimension $k$. Since the class of threshold graphs is closed under taking complements [@MahadevP1995], the co-threshold dimension of $G = (V,E)$ is the minimum integer $k$ such that there are $k$ threshold graphs $G_{i} = (V, E_{i})$, $1 \le i \le k$, with $E = \bigcap_{1 \le i \le k} E_{i}$.
\[lem:twothreshold\] Every double-threshold graph has co-threshold dimension at most $2$.
Let $G = (V,E)$ be a double-threshold graph defined by ${\mathtt{w}}\colon V \to {\mathbb{R}}$, and ${\mathtt{lb}}, {\mathtt{ub}}\in {\mathbb{R}}$. Let $E_{1}$ and $E_{2}$ be the following sets: $$\begin{aligned}
E_{1} &= \{uv \mid u,v \in V, \; {\mathtt{lb}}\le {\mathtt{w}}(u) + {\mathtt{w}}(v)\}, \\
E_{2} &= \{uv \mid u,v \in V, \; -{\mathtt{ub}}\le -{\mathtt{w}}(u) -{\mathtt{w}}(v)\}.\end{aligned}$$ It holds that $(V, E_{1})$ and $(V, E_{2})$ are threshold graphs and that $E_{1} \cap E_{2} = E$.
The next lemma allows us to use any values as ${\mathtt{lb}}$ and ${\mathtt{ub}}$ for defining a double-threshold graph. It also says that we do not have to consider degenerated cases, where some vertices have the same weight or some weight sum equals to the lower or upper bound.
\[lem:nice-representation\] Let $G = (V,E)$ be a double-threshold graph defined by ${\mathtt{w}}\colon V \to {\mathbb{R}}$ and ${\mathtt{lb}}, {\mathtt{ub}}\in {\mathbb{R}}$. For every pair $({\mathtt{lb}}^{*}, {\mathtt{ub}}^{*}) \in {\mathbb{R}}^{2}$ with ${\mathtt{lb}}^{*} < {\mathtt{ub}}^{*}$, there exists ${\mathtt{w}}^{*} \colon V \to {\mathbb{R}}$ defining $G$ with ${\mathtt{lb}}^{*}$ and ${\mathtt{ub}}^{*}$ such that
1. $\{v \mid {\mathtt{lb}}/2 \le {\mathtt{w}}(v) \le {\mathtt{ub}}/2\} = \{v \mid {\mathtt{lb}}^{*}/2 \le {\mathtt{w}}^{*}(v) \le {\mathtt{ub}}^{*}/2\}$,
2. ${\mathtt{w}}^{*}(u) + {\mathtt{w}}^{*}(v) \notin \{{\mathtt{lb}}^{*}, {\mathtt{ub}}^{*}\}$ for all $(u,v) \in V^{2}$, and
3. ${\mathtt{w}}^{*}(u) \ne {\mathtt{w}}^{*}(v)$ if $u \ne v$.
If $G$ has no edge, then the lemma trivially holds. Thus, we assume that $G$ has at least one edge. This implies that ${\mathtt{lb}}\le {\mathtt{ub}}$.
\[clm:nicer-rep1\] There exist ${\mathtt{lb}}', {\mathtt{ub}}' \in {\mathbb{R}}$ with ${\mathtt{lb}}' < {\mathtt{ub}}'$ such that ${\mathtt{w}}$, ${\mathtt{lb}}'$, and ${\mathtt{ub}}'$ define $G$, $\{v \mid {\mathtt{lb}}/2 \le {\mathtt{w}}(v) \le {\mathtt{ub}}/2\} = \{v \mid {\mathtt{lb}}'/2 \le {\mathtt{w}}(v) \le {\mathtt{ub}}'/2\}$, and ${\mathtt{w}}(u) + {\mathtt{w}}(v) \notin \{{\mathtt{lb}}', {\mathtt{ub}}'\}$ for all $(u,v) \in V^{2}$.
Claim \[clm:nicer-rep1\] is illustrated by [ \[fig:claim2-4\]]{}.
![Proof of Claim \[clm:nicer-rep1\]. (Left) There is a pair $(u,v)$ of vertices such that ${\mathtt{w}}(u)+{\mathtt{w}}(v)={\mathtt{ub}}$. The values $\alpha$ and $\beta$ are shown in red and blue, respectively. (Right) The bounds ${\mathtt{lb}}, {\mathtt{ub}}$ are replaced by ${\mathtt{lb}}', {\mathtt{ub}}'$.[]{data-label="fig:claim2-4"}](\figdir/claim2-4)
\[Proof of Claim \[clm:nicer-rep1\]\] If ${\mathtt{w}}(v) \ge {\mathtt{lb}}/2$ for all $v \in V$, then we set ${\mathtt{lb}}'$ to an arbitrary value smaller than ${\mathtt{lb}}$. Similarly, if ${\mathtt{w}}(v) \le {\mathtt{ub}}/2$ for all $v \in V$, then we use any value larger than ${\mathtt{ub}}$ as ${\mathtt{ub}}'$. Since we are done in these cases, we assume that there exists a vertex with weight less than ${\mathtt{lb}}/2$, and there exists a vertex with weight more than ${\mathtt{ub}}/2$.
Let $\alpha, \beta \in {\mathbb{R}}$ be the following values: $$\begin{aligned}
{2}
\alpha &= \max&\{{\mathtt{w}}(u) + {\mathtt{w}}(v) \mid (u,v) \in V^{2} \text{ such that } {\mathtt{w}}(u) + {\mathtt{w}}(v) < {\mathtt{lb}}\},
\\
\beta &= \min&\{{\mathtt{w}}(u) + {\mathtt{w}}(v) \mid (u,v) \in V^{2} \text{ such that } {\mathtt{w}}(u) + {\mathtt{w}}(v) > {\mathtt{ub}}\}.\end{aligned}$$ We set ${\mathtt{lb}}' = ({\mathtt{lb}}+ \alpha)/2$ and ${\mathtt{ub}}' = ({\mathtt{ub}}+ \beta)/2$. Then, $\alpha < {\mathtt{lb}}' < {\mathtt{lb}}\le {\mathtt{ub}}< {\mathtt{ub}}' < \beta$, and thus ${\mathtt{lb}}' < {\mathtt{ub}}'$. We can see that ${\mathtt{lb}}'$ and ${\mathtt{ub}}'$ satisfy the other conditions in the claim as follows. Let $(u,v) \in V^{2}$. If ${\mathtt{lb}}\le {\mathtt{w}}(u) + {\mathtt{w}}(v) \le {\mathtt{ub}}$, then ${\mathtt{lb}}' < {\mathtt{w}}(u) + {\mathtt{w}}(v) < {\mathtt{ub}}'$. If ${\mathtt{w}}(u) + {\mathtt{w}}(v) < {\mathtt{lb}}$, then ${\mathtt{w}}(u) + {\mathtt{w}}(v) \le \alpha < {\mathtt{lb}}'$. If ${\mathtt{w}}(u) + {\mathtt{w}}(v) > {\mathtt{ub}}$, then ${\mathtt{w}}(u) + {\mathtt{w}}(v) \ge \beta > {\mathtt{ub}}'$. This implies that ${\mathtt{w}}$, ${\mathtt{lb}}'$, and ${\mathtt{ub}}'$ define $G$, that $\{v \mid {\mathtt{lb}}/2 \le {\mathtt{w}}(v) \le {\mathtt{ub}}/2\} = \{v \mid {\mathtt{lb}}'/2 \le {\mathtt{w}}(v) \le {\mathtt{ub}}'/2\}$, and that ${\mathtt{w}}(u) + {\mathtt{w}}(v) \not\in \{{\mathtt{lb}}',{\mathtt{ub}}'\}$ for all $(u, v) \in V^{2}$.
In the following, let ${\mathtt{lb}}'$ and ${\mathtt{ub}}'$ be the bounds in Claim \[clm:nicer-rep1\].
\[clm:nicer-rep2\] There exists ${\mathtt{w}}' \colon V \to {\mathbb{R}}$ such that ${\mathtt{w}}'$, ${\mathtt{lb}}'$, and ${\mathtt{ub}}'$ define $G$, $\{v \mid {\mathtt{lb}}/2 \le {\mathtt{w}}(v) \le {\mathtt{ub}}/2\} = \{v \mid {\mathtt{lb}}'/2 \le {\mathtt{w}}'(v) \le {\mathtt{ub}}'/2\}$, ${\mathtt{w}}'(u) + {\mathtt{w}}'(v) \notin \{{\mathtt{lb}}', {\mathtt{ub}}'\}$ for all $(u,v) \in V^{2}$, and ${\mathtt{w}}'(u) \ne {\mathtt{w}}'(v)$ if $u \ne v$.
Claim \[clm:nicer-rep2\] is illustrated by [ \[fig:claim2-5\]]{}.
![Proof of Claim \[clm:nicer-rep2\]. (Left) The second smallest weight is shared by two vertices $z$ and $z'$. (Right) The weight of $z$ is changed, but the double-threshold graph does not change.[]{data-label="fig:claim2-5"}](\figdir/claim2-5){width="\textwidth"}
\[Proof of Claim \[clm:nicer-rep2\]\] Let ${\mathtt{w}}'\colon V \to {\mathbb{R}}$ be a vertex-weight function such that ${\mathtt{w}}'$, ${\mathtt{lb}}'$, and ${\mathtt{ub}}'$ define $G$, $\{v \mid {\mathtt{lb}}/2 \le {\mathtt{w}}(v) \le {\mathtt{ub}}/2\} = \{v \mid {\mathtt{lb}}'/2 \le {\mathtt{w}}'(v) \le {\mathtt{ub}}'/2\}$, ${\mathtt{w}}'(u) + {\mathtt{w}}'(v) \notin \{{\mathtt{lb}}', {\mathtt{ub}}'\}$ for all $(u,v) \in V^{2}$, and, under these conditions, the number $|\{v \mid {\mathtt{w}}'(v) \ne {\mathtt{w}}'(u) \text{ for all } u \in V \setminus \{v\} \}|$ of unique-weight vertices is maximum. Such ${\mathtt{w}}'$ exists since ${\mathtt{w}}$ itself satisfies all the conditions.
Suppose to the contrary that there are distinct vertices $z, z' \in V$ with the same weight ${\mathtt{w}}'(z) = {\mathtt{w}}'(z')$. Let $\alpha, \beta, \gamma$ be defined as follows: $$\begin{aligned}
\alpha &=&
\begin{cases}
\min \{ {\mathtt{w}}'(v) \mid v \in N(z) \} & \text{if } N(z) \ne \emptyset,\\
+\infty & \text{otherwise};
\end{cases}\\
\beta &=&
\begin{cases}
\min \{ {\mathtt{w}}'(v) \mid v \notin N(z), \; {\mathtt{w}}'(z) + {\mathtt{w}}'(v) > {\mathtt{ub}}' \}
& \text{if } \{v \notin N(z) \mid {\mathtt{w}}'(z) + {\mathtt{w}}'(v) > {\mathtt{ub}}'\} \ne \emptyset,\\
+\infty & \text{otherwise};
\end{cases}\\
\gamma &=&
\begin{cases}
\max \{ {\mathtt{w}}'(v) \mid {\mathtt{w}}'(v) < {\mathtt{w}}'(z) \} & \text{if there exists } v \text{ such that } {\mathtt{w}}'(v) < {\mathtt{w}}'(z),\\
-\infty & \text{otherwise}.
\end{cases}\end{aligned}$$ Let $$\varepsilon = \min\{{\mathtt{w}}'(z) + \alpha - {\mathtt{lb}}', {\mathtt{w}}'(z) + \beta - {\mathtt{ub}}', {\mathtt{w}}'(z) - \gamma,
|{\mathtt{w}}'(z) - {\mathtt{lb}}'/2|, |{\mathtt{w}}'(z) - {\mathtt{ub}}'/2|\}.$$ Note that $\varepsilon$ is positive and not infinitely large. By decreasing the weight of $z$ by $\varepsilon/2$, we define ${\mathtt{w}}''$. That is, ${\mathtt{w}}''(z) = {\mathtt{w}}'(z) - \varepsilon/2$ and ${\mathtt{w}}''(v) = {\mathtt{w}}'(v)$ for all other $v \ne z$.
Observe that $z$ has a unique weight under ${\mathtt{w}}''$ because $z$ is the only vertex of weight in the interval $(\gamma, {\mathtt{w}}'(z))$. Since all vertices other than $z$ have the same weight in ${\mathtt{w}}'$ and ${\mathtt{w}}''$, the number of unique-weight vertices under ${\mathtt{w}}''$ is larger than the one under ${\mathtt{w}}'$.
Next we show that ${\mathtt{w}}''(z) + {\mathtt{w}}''(v) \notin \{{\mathtt{lb}}',{\mathtt{ub}}'\}$ for all $v \in V$ and that ${\mathtt{w}}''$, ${\mathtt{lb}}'$, and ${\mathtt{ub}}'$ define $G$. Let $v \in V$. (Note that $v$ may be $z$.) By the assumption on ${\mathtt{w}}'$, we have ${\mathtt{w}}'(z) + {\mathtt{w}}'(v) \notin \{{\mathtt{lb}}', {\mathtt{ub}}'\}$. If ${\mathtt{w}}'(z) + {\mathtt{w}}'(v) < {\mathtt{lb}}'$, then ${\mathtt{w}}''(z) + {\mathtt{w}}''(v) < {\mathtt{lb}}'$ holds as ${\mathtt{w}}''(z) < {\mathtt{w}}'(z)$ and ${\mathtt{w}}''(v) = {\mathtt{w}}'(v)$. Similarly, if ${\mathtt{w}}'(z) + {\mathtt{w}}'(v) < {\mathtt{ub}}'$, then ${\mathtt{w}}''(z) + {\mathtt{w}}''(v) < {\mathtt{ub}}'$. Assume that ${\mathtt{w}}'(z) + {\mathtt{w}}'(v) > {\mathtt{lb}}'$. If $v$ is a neighbor of $z$, then ${\mathtt{w}}'(v) \ge \alpha$ holds, and thus ${\mathtt{w}}''(z) + {\mathtt{w}}''(v) > {\mathtt{lb}}'$ as follows: $$\begin{aligned}
{\mathtt{w}}''(z) + {\mathtt{w}}''(v)
&= {\mathtt{w}}'(z) + {\mathtt{w}}'(v) - \varepsilon/2
\\
&\ge {\mathtt{w}}'(z) + {\mathtt{w}}'(v) - ({\mathtt{w}}'(z) + \alpha - {\mathtt{lb}}')/2
\\
&\ge {\mathtt{w}}'(z) + {\mathtt{w}}'(v) - ({\mathtt{w}}'(z) + {\mathtt{w}}'(v) - {\mathtt{lb}}')/2
\\
&= ({\mathtt{w}}'(z) + {\mathtt{w}}'(v) + {\mathtt{lb}}')/2
\\
&> ({\mathtt{lb}}' + {\mathtt{lb}}')/2
\\
&= {\mathtt{lb}}'.\end{aligned}$$ If $v = z$, then ${\mathtt{w}}'(z) = {\mathtt{w}}'(v) > {\mathtt{lb}}'/2$, and thus $$\begin{aligned}
{\mathtt{w}}''(z) + {\mathtt{w}}''(v)
&= 2{\mathtt{w}}'(z) - \varepsilon
\\
&\ge 2{\mathtt{w}}'(z) - |{\mathtt{w}}'(z) - {\mathtt{lb}}'/2|
= 2{\mathtt{w}}'(z) - ({\mathtt{w}}'(z) - {\mathtt{lb}}'/2)
\\
&= {\mathtt{w}}'(z) + {\mathtt{lb}}'/2
\\
&> {\mathtt{lb}}'.\end{aligned}$$ Next, assume that ${\mathtt{w}}'(z) + {\mathtt{w}}'(v) > {\mathtt{ub}}'$. If $v$ is a non-neighbor of $z$, then ${\mathtt{w}}'(v) \ge \beta$. Thus, ${\mathtt{w}}''(z) + {\mathtt{w}}''(v) > {\mathtt{ub}}'$ holds as follows: $$\begin{aligned}
{\mathtt{w}}''(z) + {\mathtt{w}}''(v)
&=
{\mathtt{w}}'(z) + {\mathtt{w}}'(v) - \varepsilon/2
\\
&\ge {\mathtt{w}}'(z) + {\mathtt{w}}'(v) - ({\mathtt{w}}'(z) + \beta - {\mathtt{ub}}')/2
\\
&\ge {\mathtt{w}}'(z) + {\mathtt{w}}'(v) - ({\mathtt{w}}'(z) + {\mathtt{w}}'(v) - {\mathtt{ub}}')/2
\\
&= ({\mathtt{w}}'(z) + {\mathtt{w}}'(v) + {\mathtt{ub}}')/2
\\
&> ({\mathtt{ub}}' + {\mathtt{ub}}')/2
\\
&= {\mathtt{ub}}'.\end{aligned}$$ If $v = z$, then ${\mathtt{w}}'(z) = {\mathtt{w}}'(v) > {\mathtt{ub}}'/2$, and thus $$\begin{aligned}
{\mathtt{w}}''(z) + {\mathtt{w}}''(v)
&= 2{\mathtt{w}}'(z) - \varepsilon
\\
&\ge 2{\mathtt{w}}'(z) - |{\mathtt{w}}'(z) - {\mathtt{ub}}'/2|
= 2{\mathtt{w}}'(z) - ({\mathtt{w}}'(z) - {\mathtt{ub}}'/2)
\\
&= {\mathtt{w}}'(z) + {\mathtt{ub}}'/2
\\
&> {\mathtt{ub}}'.\end{aligned}$$
Finally, we show that $\{v \mid {\mathtt{lb}}/2 \le {\mathtt{w}}(v) \le {\mathtt{ub}}/2\} = \{v \mid {\mathtt{lb}}'/2 \le {\mathtt{w}}''(v) \le {\mathtt{ub}}'/2\}$. To this end, it suffices to show that ${\mathtt{lb}}'/2 < {\mathtt{w}}'(z) < {\mathtt{ub}}'/2$ if and only if ${\mathtt{lb}}'/2 < {\mathtt{w}}''(z) < {\mathtt{ub}}'/2$ as $2{\mathtt{w}}'(z), 2{\mathtt{w}}''(z) \notin \{{\mathtt{lb}}', {\mathtt{ub}}'\}$. If ${\mathtt{lb}}'/2 < {\mathtt{w}}'(z) < {\mathtt{ub}}'/2$, then ${\mathtt{w}}''(z) < {\mathtt{w}}'(z) < {\mathtt{ub}}'/2$ and $${\mathtt{w}}''(z)
= {\mathtt{w}}'(z) - \varepsilon/2
> {\mathtt{w}}'(z) - \varepsilon
\ge {\mathtt{w}}'(z) - |{\mathtt{w}}'(z) - {\mathtt{lb}}'/2|
= {\mathtt{w}}'(z) - ({\mathtt{w}}'(z) - {\mathtt{lb}}'/2)
= {\mathtt{lb}}'/2.$$ If ${\mathtt{w}}'(z) < {\mathtt{lb}}'/2$, then ${\mathtt{w}}''(z) < {\mathtt{w}}'(z) < {\mathtt{lb}}'/2$. If ${\mathtt{w}}'(z) > {\mathtt{ub}}'/2$, then we have $${\mathtt{w}}''(z)
= {\mathtt{w}}'(z) - \varepsilon/2
> {\mathtt{w}}'(z) - \varepsilon
\ge {\mathtt{w}}'(z) - |{\mathtt{w}}'(z) - {\mathtt{ub}}'/2|
= {\mathtt{w}}'(z) - ({\mathtt{w}}'(z) - {\mathtt{ub}}'/2)
= {\mathtt{ub}}'/2.$$
We have shown that ${\mathtt{w}}''$, ${\mathtt{lb}}'$, and ${\mathtt{ub}}'$ define $G$, $\{v \mid {\mathtt{lb}}/2 \le {\mathtt{w}}(v) \le {\mathtt{ub}}/2\} = \{v \mid {\mathtt{lb}}'/2 \le {\mathtt{w}}''(v) \le {\mathtt{ub}}'/2\}$, and ${\mathtt{w}}''(u) + {\mathtt{w}}''(v) \notin \{{\mathtt{lb}}', {\mathtt{ub}}'\}$ for all $(u,v) \in V^{2}$. We also showed that the number of unique-weight vertices under ${\mathtt{w}}''$ is larger than the one under ${\mathtt{w}}'$. This contradicts the assumption on how ${\mathtt{w}}'$ is chosen, and thus we can conclude that no two vertices have the same weight in ${\mathtt{w}}'$.
We continue the proof of Lemma \[lem:nice-representation\]. Let ${\mathtt{w}}'$ be the vertex-weight function in Claim \[clm:nicer-rep2\]. Given ${\mathtt{lb}}^{*}, {\mathtt{ub}}^{*} \in {\mathbb{R}}$ with ${\mathtt{lb}}^{*} < {\mathtt{ub}}^{*}$, let $\rho = ({\mathtt{ub}}^{*} - {\mathtt{lb}}^{*})/({\mathtt{ub}}' - {\mathtt{lb}}')$. For each $v \in V$, we set ${\mathtt{w}}^{*}(v) = \rho \cdot {\mathtt{w}}'(v) - \left(\rho \cdot {\mathtt{lb}}' - {\mathtt{lb}}^{*}\right)/2$. Then, ${\mathtt{w}}^{*}(u) \ne {\mathtt{w}}^{*}(v)$ if $u \ne v$. Now for each $(u,v) \in V^{2}$, $$\begin{aligned}
& {\mathtt{lb}}^{*} < {\mathtt{w}}^{*}(u) + {\mathtt{w}}^{*}(v) < {\mathtt{ub}}^{*}
\\
&\iff {\mathtt{lb}}^{*}
<
\rho \cdot ({\mathtt{w}}'(u) + {\mathtt{w}}'(v))
- \left(\rho \cdot {\mathtt{lb}}' - {\mathtt{lb}}^{*}\right)
<
{\mathtt{ub}}^{*}
\\
&\iff \rho \cdot {\mathtt{lb}}'
<
\rho \cdot ({\mathtt{w}}'(u) + {\mathtt{w}}'(v))
<
\rho \cdot {\mathtt{lb}}' + {\mathtt{ub}}^{*} - {\mathtt{lb}}^{*}
= \rho \cdot {\mathtt{lb}}' + \rho \cdot ({\mathtt{ub}}' - {\mathtt{lb}}')
\\
&\iff {\mathtt{lb}}'
<
{\mathtt{w}}'(u) + {\mathtt{w}}'(v)
<
{\mathtt{ub}}'.\end{aligned}$$ Similarly, ${\mathtt{w}}^{*}(u) + {\mathtt{w}}^{*}(v) < {\mathtt{lb}}^{*}$ if and only if ${\mathtt{w}}'(u) + {\mathtt{w}}'(v) < {\mathtt{lb}}'$, and ${\mathtt{w}}^{*}(u) + {\mathtt{w}}^{*}(v) > {\mathtt{ub}}^{*}$ if and only if ${\mathtt{w}}'(u) + {\mathtt{w}}'(v) > {\mathtt{ub}}'$. Since ${\mathtt{w}}'(u) + {\mathtt{w}}'(v) \notin \{{\mathtt{lb}}', {\mathtt{ub}}'\}$ for all $(u,v)\in V^{2}$, it follows that ${\mathtt{w}}^{*}$, ${\mathtt{lb}}^{*}$, and ${\mathtt{ub}}^{*}$ define $G$, $\{v \mid {\mathtt{lb}}/2 \le {\mathtt{w}}(v) \le {\mathtt{ub}}/2\} = \{v \mid {\mathtt{lb}}^{*}/2 \le {\mathtt{w}}^{*}(v) \le {\mathtt{ub}}^{*}/2\}$, and ${\mathtt{w}}^{*}(u) + {\mathtt{w}}^{*}(v) \notin \{{\mathtt{lb}}^{*}, {\mathtt{ub}}^{*}\}$ for all $(u,v)\in V^{2}$.
The following lemma can be proved using a characterization of graphs of co-threshold dimension at most 2 [@RaschleS95]. Also, a proof of the same fact was given by Xiao and Nagamochi [@XiaoN18] using their characterization of double-threshold graphs. To make the discussion self-contained, we present a direct proof here.
\[lem:at-most-one-nonbipartite\] A double-threshold graph has at most one non-bipartite component.
Let $G$ be a double-threshold graph defined by ${\mathtt{w}}\colon V\to {\mathbb{R}}$ and ${\mathtt{lb}}, {\mathtt{ub}}\in {\mathbb{R}}$. It suffices to show that all cycles of odd lengths in $G$ belong to the same component.
Let $C$ and $C'$ be two induced cycles in $G$ that are vertex-disjoint and of odd lengths. Since $C$ is of odd length, we can find three vertices $a$, $b$, $c$ that consecutively appear in $C$ such that ${\mathtt{w}}(a) \le {\mathtt{w}}(b) \le {\mathtt{w}}(c)$. Similarly, we can find three consecutive vertices $x$, $y$, $z$ in $C'$ such that ${\mathtt{w}}(x) \le {\mathtt{w}}(y) \le {\mathtt{w}}(z)$. By symmetry, we may assume that ${\mathtt{w}}(b) \le {\mathtt{w}}(y)$. This implies that ${\mathtt{w}}(a) \le {\mathtt{w}}(y)$ and ${\mathtt{w}}(b) \le {\mathtt{w}}(z)$. Since $ab$ and $yz$ are edges, we have ${\mathtt{lb}}\le {\mathtt{w}}(a) + {\mathtt{w}}(b) \le {\mathtt{w}}(y) + {\mathtt{w}}(b) \le {\mathtt{w}}(y) + {\mathtt{w}}(z) \le {\mathtt{ub}}$. This implies that $by$ is also an edge, and thus $C$ and $C'$ belong to the same component.
We further observe that bipartite components can be considered separately. A bipartite graph is denoted by $(X, Y; E)$ when its vertex set is partitioned into two independent sets $X$ and $Y$ and its edge set is $E$.
\[lem:disjoint-bicomponent\] The disjoint union of a double-threshold graph and a bipartite double-threshold graph is a double-threshold graph.
Let $G = (V, E_{G})$ be a double-threshold graph defined by ${\mathtt{w}}_{G} \colon V \to {\mathbb{R}}$ and ${\mathtt{lb}}, {\mathtt{ub}}\in {\mathbb{R}}$. Let $H = (X,Y; E_{H})$ be a bipartite double-threshold graph. By Lemma \[lem:nice-representation\], we may assume that ${\mathtt{lb}}< {\mathtt{ub}}$ and $H$ is defined by the same ${\mathtt{lb}}$ and ${\mathtt{ub}}$ with some weight function ${\mathtt{w}}_{H} \colon X \cup Y \to {\mathbb{R}}$.
We set $\alpha \in {\mathbb{R}}_{\ge 0}$ large enough so that the following conditions hold: $$\begin{aligned}
{\mathtt{ub}}&< 2 \cdot \min\{ \min\{{\mathtt{w}}_{G}(v) \mid v \in V\}, \min\{{\mathtt{w}}_{H}(w) \mid w \in X \cup Y \}\} + \alpha, \\
{\mathtt{lb}}&> 2 \cdot \max\{ \max\{{\mathtt{w}}_{G}(v) \mid v \in V\}, \max\{{\mathtt{w}}_{H}(w) \mid w \in X \cup Y \}\} - \alpha.\end{aligned}$$ Now, we set ${\mathtt{w}}\colon (V \cup X \cup Y) \to {\mathbb{R}}$ as follows: $${\mathtt{w}}(w) =
\begin{cases}
{\mathtt{w}}_{G}(w) & \text{if } w \in V, \\
{\mathtt{w}}_{H}(w) + \alpha & \text{if } w \in X, \\
{\mathtt{w}}_{H}(w) - \alpha & \text{if } w \in Y.
\end{cases}$$
We show that ${\mathtt{w}}$, ${\mathtt{lb}}$, and ${\mathtt{ub}}$ define the disjoint union $(V \cup X \cup Y, E_{G} \cup E_{H})$ of $G$ and $H$. For two vertices $u,v \in V$, we have ${\mathtt{w}}(u) + {\mathtt{w}}(v) = {\mathtt{w}}_{G}(u) + {\mathtt{w}}_{G}(v)$. Similarly, for $x \in X$ and $y \in Y$, we have ${\mathtt{w}}(x) + {\mathtt{w}}(y) = {\mathtt{w}}_{H}(x) + {\mathtt{w}}_{H}(y)$. For $x_{1}, x_{2} \in X$, we have ${\mathtt{w}}(x_{1}) + {\mathtt{w}}(x_{2}) = {\mathtt{w}}_{H}(x_{1}) + {\mathtt{w}}_{H}(x_{2}) + 2\alpha > {\mathtt{ub}}$. For $y_{1}, y_{2} \in Y$, we have ${\mathtt{w}}(y_{1}) + {\mathtt{w}}(y_{2}) = {\mathtt{w}}_{H}(y_{1}) + {\mathtt{w}}_{H}(y_{2}) - 2\alpha < {\mathtt{lb}}$. For $v \in V$ and $x \in X$, we have ${\mathtt{w}}(v) + {\mathtt{w}}(x) = {\mathtt{w}}_{G}(v) + {\mathtt{w}}_{H}(x) + \alpha > {\mathtt{ub}}$. Finally, for $v \in V$ and $y \in Y$, we have ${\mathtt{w}}(v) + {\mathtt{w}}(y) = {\mathtt{w}}_{G}(v) + {\mathtt{w}}_{H}(y) - \alpha < {\mathtt{lb}}$.
By Lemmas \[lem:at-most-one-nonbipartite\] and \[lem:disjoint-bicomponent\], we have the following corollary.
\[cor:component\] A graph is a double-threshold graph if and only if it contains at most one non-bipartite component and all components are double-threshold graphs.
Permutation graphs
------------------
A graph $G=(V,E)$ is a *permutation graph* if there exist linear orderings $\prec_1, \prec_2$ on $V$ with the following property: $$uv \in E \iff (u \prec_1 v \text{ and } v \prec_2 u) \text{ or } (u \prec_2 v \text{ and } v \prec_1 u).$$ We say that $\prec_{1}$ and $\prec_{2}$ *define* the permutation graph $G$. We call $\prec_{1}$ a *permutation ordering* of $G$ if there exists a linear ordering $\prec_{2}$ satisfying the condition above. Since $\prec_{1}$ and $\prec_{2}$ play a symmetric role in the definition, $\prec_{2}$ is also a permutation ordering of $G$. Note that for a graph $G$ and a permutation ordering $\prec_{1}$ of $G$, the other ordering $\prec_{2}$ that defines $G$ together with $\prec_{1}$ is uniquely determined. Also note that if $\prec_{1}$ and $\prec_{2}$ define $G$, then $\prec_{1}^{\mathrm{R}}$ and $\prec_{2}^{\mathrm{R}}$ also define $G$, where $\prec^{\mathrm{R}}$ denotes the reversed ordering of $\prec$.
We often represent a permutation graph with a *permutation diagram*, which is drawn as follows. See [ \[fig:permdia\_example1\]]{} for an illustration. Imagine two horizontal parallel lines $\ell_1$ and $\ell_2$ on the plane. Then, we place the vertices in $V$ on $\ell_1$ from left to right according to the permutation ordering $\prec_1$ as distinct points, and similarly place the vertices in $V$ on $\ell_2$ from left to right according to $\prec_2$ as distinct points. The positions of $v\in V$ can be represented by $x$-coordinates on $\ell_{1}$ and $\ell_{2}$, which are denoted by ${\mathtt{x}}_{1}(v)$ and ${\mathtt{x}}_{2}(v)$, respectively. We connect the two points representing the same vertex with a line segment. The process results in a diagram with $|V|$ line segments. By definition, $uv \in E$ if and only if the line segments representing $u$ and $v$ cross in the permutation diagram, which is equivalent to the inequality $({\mathtt{x}}_1(u)-{\mathtt{x}}_1(v)) ({\mathtt{x}}_2(u)-{\mathtt{x}}_2(v)) < 0$.
![An example of a permutation diagram. (Left) A permutation graph $G$. (Right) A permutation diagram that represents $G$.[]{data-label="fig:permdia_example1"}](\figdir/permdia_example1)
Conversely, from a permutation diagram of $G$, we can extract linear orderings $\prec_{1}$ and $\prec_{2}$ as $$\begin{aligned}
{\mathtt{x}}_1(u) < {\mathtt{x}}_1(v) &\iff& u \prec_{1} v,\\
{\mathtt{x}}_2(u) < {\mathtt{x}}_2(v) &\iff& u \prec_{2} v.\end{aligned}$$ When those conditions are satisfied, we say that the orderings of the $x$-coordinates on $\ell_1$ and $\ell_2$ are *consistent* with the linear orderings $\prec_{1}$ and $\prec_{2}$, respectively.
A graph is a *bipartite permutation graph* if it is a bipartite graph and a permutation graph. Although a permutation graph may have an exponential number of permutation orderings, it is essentially unique for a connected bipartite permutation graph in the sense of Lemma \[lem:uniquerep\] below. For a graph $G = (V,E)$, linear orderings $\langle v_{1}, \dots, v_{n} \rangle$ and $\langle v'_{1}, \dots, v'_{n} \rangle$ on $V$ are *neighborhood-equivalent* if $N(v_{i}) = N(v'_{i})$ for all $i$.
\[[@HeggernesHMV15]\] \[lem:uniquerep\] Let $G$ be a connected bipartite permutation graph defined by $\prec_{1}$ and $\prec_{2}$. Then, every permutation ordering of $G$ is neighborhood-equivalent to $\prec_{1}$, $\prec_{2}$, $\prec_{1}^{\mathrm{R}}$, or $\prec_{2}^{\mathrm{R}}$.
A bipartite graph $(X,Y;E)$ is a *unit interval bigraph* if there is a set of unit intervals $\{I_{v} = [l_{v}, l_{v}+1] \mid v \in X \cup Y\}$ such that $xy \in E$ if and only if $I_{x} \cap I_{y} \ne \emptyset$ for $x \in X$ and $y \in Y$. The class of unit interval bigraphs is known to be equal to the class of bipartite permutation graphs.
\[[@SenS94; @West98; @HellH04]\] \[prop:bpg=uib\] A graph is a bipartite permutation graph if and only if it is a unit interval bigraph.
The following lemma shows that a bipartite permutation graph can be represented by a permutation diagram with a special property. An illustration is given in [ \[fig:lem2-11\]]{}.
![An illustration of Lemma \[lem:rightanglerep\]. (Left) A permutation diagram of a bipartite permutation graph $G=(X,Y;E)$. The vertices in $X$ are represented by blue segments, and the vertices in $Y$ are represented by red segments. (Right) A permutation diagram that represents $G$ obtained by Lemma \[lem:rightanglerep\].[]{data-label="fig:lem2-11"}](\figdir/lem2-11)
\[lem:rightanglerep\] Let $G=(X, Y; E)$ be a bipartite permutation graph. Then, $G$ can be represented by a permutation diagram in which ${\mathtt{x}}_{2}(x)={\mathtt{x}}_{1}(x)+1$ for $x \in X$ and ${\mathtt{x}}_{2}(y)={\mathtt{x}}_{1}(y)-1$ for $y \in Y$.
By Proposition \[prop:bpg=uib\], there is a set of unit intervals $\{I_{v} = [l_{v}, l_{v}+1] \mid v \in X \cup Y\}$ such that for $x \in X$ and $y \in Y$, $xy \in E$ if and only if $I_{x} \cap I_{y} \ne \emptyset$. We can assume that all endpoints of the intervals are distinct; that is, $l_{u} \notin \{l_{v}, l_{v}+1\}$ for all $u,v \in X \cup Y $ with $u \ne v$ [@West98]. For each $x \in X$, we set ${\mathtt{x}}_{1}(x) = l_{x}$ and ${\mathtt{x}}_{2}(x) = l_{x} + 1$. For each $y \in Y$, we set ${\mathtt{x}}_{1}(y) = l_{y}+1$ and ${\mathtt{x}}_{2}(y) = l_{y}$. It suffices to show that this permutation diagram represents $G$. Observe that line segments corresponding to vertices from the same set, $X$ or $Y$, are parallel and thus do not cross. For $x \in X$ and $y \in Y$, we have $$\begin{aligned}
I_{x} \cap I_{y} \ne \emptyset
&\iff
|l_{x} - l_{y}| < 1 && (\because \text{ all endpoints are distinct})
\\
&\iff
l_{x} < l_{y} + 1 \textrm{ and } l_{y} < l_{x} + 1
\\
&\iff
{\mathtt{x}}_{1}(x) < {\mathtt{x}}_{1}(y) \textrm{ and } {\mathtt{x}}_{2}(y) < {\mathtt{x}}_{2}(x) && (\because x \in X, \ y \in Y)
\\
&\iff
({\mathtt{x}}_{1}(x)-{\mathtt{x}}_{1}(y)) ({\mathtt{x}}_{2}(x)-{\mathtt{x}}_{2}(y)) < 0.\end{aligned}$$ To see the last equivalence, observe that ${\mathtt{x}}_{1}(x) < {\mathtt{x}}_{2}(x)$ and ${\mathtt{x}}_{2}(y) < {\mathtt{x}}_{1}(y)$, and thus it is impossible to have both ${\mathtt{x}}_1(x) > {\mathtt{x}}_1(y)$ and ${\mathtt{x}}_2(y) > {\mathtt{x}}_2(x)$. Therefore, we conclude that the diagram represents $G$.
We can show that given a permutation ordering, there exists a permutation diagram consistent with the ordering that satisfies the conditions in Lemma \[lem:rightanglerep\].
\[cor:rightanglerep\] Let $G=(X, Y; E)$ be a connected bipartite permutation graph defined by permutation orderings $\prec_{1}$ and $\prec_{2}$. If the first vertex in $\prec_{1}$ belongs to $X$, then $G$ can be represented by a permutation diagram such that the orderings of the $x$-coordinates on $\ell_{1}$ and $\ell_{2}$ are consistent with $\prec_{1}$ and $\prec_{2}$, respectively, and that ${\mathtt{x}}_2(x)={\mathtt{x}}_1(x)+1$ for $x\in X$ and ${\mathtt{x}}_2(y)={\mathtt{x}}_1(y)-1$ for $y\in Y$.
Since $G$ is connected, the last vertex in $\prec_{1}$ belongs to $Y$, the first vertex in $\prec_{2}$ belongs to $Y$, and the last vertex in $\prec_{2}$ belongs to $X$.
By Lemma \[lem:rightanglerep\], $G$ can be represented by a permutation diagram $D'$ in which ${\mathtt{x}}_{2}(x)={\mathtt{x}}_{1}(x)+1$ for $x \in X$ and ${\mathtt{x}}_{2}(y)={\mathtt{x}}_{1}(y)-1$ for $y \in Y$. Let $\prec'_{1}$ and $\prec'_{2}$ be the permutation orderings corresponding to $\ell_{1}$ and $\ell_{2}$, respectively, in this diagram $D'$. Lemma \[lem:uniquerep\] and the assumption on the first vertex in $\prec_{1}$ imply that $\prec_{1}$ is neighborhood-equivalent to $\prec'_{1}$ or $(\prec'_{2})^{R}$. We may assume that $\prec_{1}$ is neighborhood-equivalent to $\prec'_{1}$ since otherwise we can rotate the diagram $D'$ by $180$ degrees and get a permutation diagram of $G$ in which the ordering on $\ell_{1}$ is $\prec_{1}$, ${\mathtt{x}}_{2}(x)={\mathtt{x}}_{1}(x)+1$ for $x \in X$, and ${\mathtt{x}}_{2}(y)={\mathtt{x}}_{1}(y)-1$ for $y \in Y$.
Now we can construct a desired permutation diagram of $G$ using $\prec_{1}$ and $D'$ by appropriately giving a mapping between segments and vertices. That is, for each $i \in \{1, \dots, |X \cup Y|\}$, we assign the $i$th vertex in $\prec_{1}$ to the segment in $D'$ with the $i$th smallest $x$-coordinate on $\ell_1$. This new diagram is a permutation diagram of $G$ since $\prec_{1}$ is neighborhood-equivalent to $\prec'_{1}$. Since $G$ and $\prec_{1}$ uniquely determine the ordering on $\ell_{2}$, the $x$-coordinates ${\mathtt{x}}_{2}$ on $\ell_{2}$ are consistent with $\prec_{2}$.
Double-threshold graphs as permutation graphs {#sec:permutation}
=============================================
In this section, we show that double-threshold graphs are strongly related to permutation graphs. We first show that all double-threshold graphs are permutation graphs. An illustration of the following lemma is given in [ \[fig:lem3-1\]]{}.
![An illustration of Lemma \[lem:DTGisPerm\]. (Left) A double-threshold graph represented in the form of [ \[fig:dtg\_example1\]]{}. The blue points correspond to the vertices $v$ with ${\mathtt{w}}(v) \leq {\mathtt{lb}}/2$. The red points correspond to the vertices $v$ with ${\mathtt{lb}}/2 < {\mathtt{w}}(v) \leq {\mathtt{ub}}/2$. (Right) The permutation diagram obtained by applying Lemma \[lem:DTGisPerm\]. The vertices are labeled from $1$ to $9$ in the increasing order of the weight ${\mathtt{w}}$. According to the construction in Lemma \[lem:DTGisPerm\], on $\ell_1$, the blue and red points are reflected around ${\mathtt{ub}}/2$. On $\ell_2$, the blue points are reflected around ${\mathtt{lb}}/2$.[]{data-label="fig:lem3-1"}](\figdir/lem3-1)
\[lem:DTGisPerm\] Every double-threshold graph is a permutation graph.
Let $G = (V,E)$ be a double-threshold graph defined by ${\mathtt{w}}\colon V \to {\mathbb{R}}$ and ${\mathtt{lb}}, {\mathtt{ub}}\in {\mathbb{R}}$ with ${\mathtt{lb}}< {\mathtt{ub}}$. By Lemma \[lem:nice-representation\], we can assume for every pair of vertices $u$ and $v$ that ${\mathtt{w}}(u) \ne {\mathtt{w}}(v)$ and ${\mathtt{w}}(u) + {\mathtt{w}}(v) \notin \{{\mathtt{lb}},{\mathtt{ub}}\}$. From ${\mathtt{w}}$, ${\mathtt{lb}}$ and ${\mathtt{ub}}$, we construct a permutation diagram of $G$.
Let $\ell_{1}$ and $\ell_{2}$ be two horizontal parallel lines. For each vertex $v \in V$, we set the $x$-coordinates ${\mathtt{x}}_{1}(v)$ and ${\mathtt{x}}_{2}(v)$ on $\ell_{1}$ and $\ell_{2}$, respectively, as follows: $$\begin{aligned}
{\mathtt{x}}_{1}(v) &= \max\{{\mathtt{w}}(v), {\mathtt{ub}}- {\mathtt{w}}(v)\}, \\
{\mathtt{x}}_{2}(v) &= \max\{{\mathtt{w}}(v), {\mathtt{lb}}- {\mathtt{w}}(v)\}.\end{aligned}$$ By connecting ${\mathtt{x}}_{1}(v)$ and ${\mathtt{x}}_{2}(v)$ with a line segment for each $v \in V(G)$, we get a permutation diagram. We now show that this diagram represents $G$. Note that the endpoints on the lines $\ell_{1}$ and $\ell_{2}$ are distinct since ${\mathtt{w}}(u) \ne {\mathtt{w}}(v)$ and ${\mathtt{w}}(u) + {\mathtt{w}}(v) \notin \{{\mathtt{lb}},{\mathtt{ub}}\}$ if $u \ne v$.
In the following, let $u$ and $v$ be two vertices such that ${\mathtt{w}}(u) < {\mathtt{w}}(v)$.
${\mathtt{w}}(u) < {\mathtt{w}}(v) < {\mathtt{lb}}/2$. As ${\mathtt{w}}(u) + {\mathtt{w}}(v) < {\mathtt{lb}}$, we have $uv \notin E$. On the other hand, it holds that ${\mathtt{x}}_{1}(u) = {\mathtt{ub}}- {\mathtt{w}}(u) > {\mathtt{ub}}- {\mathtt{w}}(v) = {\mathtt{x}}_{1}(v)$ and ${\mathtt{x}}_{2}(u) = {\mathtt{lb}}- {\mathtt{w}}(u) > {\mathtt{lb}}- {\mathtt{w}}(v) = {\mathtt{x}}_{2}(v)$. This implies that the line segments corresponding to $u$ and $v$ do not cross.
${\mathtt{ub}}/2 < {\mathtt{w}}(u) < {\mathtt{w}}(v)$. Since ${\mathtt{w}}(u) + {\mathtt{w}}(v) > {\mathtt{ub}}$, $uv \notin E$ holds. The line segments of $u$ and $v$ do not cross as ${\mathtt{x}}_{1}(u) = {\mathtt{w}}(u) < {\mathtt{w}}(v) = {\mathtt{x}}_{1}(v)$ and ${\mathtt{x}}_{2}(u) = {\mathtt{w}}(u) < {\mathtt{w}}(v) = {\mathtt{x}}_{2}(v)$.
${\mathtt{lb}}/2 \le {\mathtt{w}}(u) < {\mathtt{w}}(v) \le {\mathtt{ub}}/2$. We have ${\mathtt{lb}}\le {\mathtt{w}}(u) + {\mathtt{w}}(v) \le {\mathtt{ub}}$, and thus $uv \in E$. The line segments of $u$ and $v$ cross as ${\mathtt{x}}_{1}(u) = {\mathtt{ub}}- {\mathtt{w}}(u) > {\mathtt{ub}}- {\mathtt{w}}(v) = {\mathtt{x}}_{1}(v)$ and ${\mathtt{x}}_{2}(u) = {\mathtt{w}}(u) < {\mathtt{w}}(v) = {\mathtt{x}}_{2}(v)$.
${\mathtt{w}}(u) < {\mathtt{lb}}/2$ and ${\mathtt{w}}(v) > {\mathtt{ub}}/2$. We show that $({\mathtt{x}}_{1}(u) - {\mathtt{x}}_{1}(v))({\mathtt{x}}_{2}(u) - {\mathtt{x}}_{2}(v)) < 0$ (that is, the line segments of $u$ and $v$ cross) if and only if ${\mathtt{lb}}< {\mathtt{w}}(u) + {\mathtt{w}}(v) < {\mathtt{ub}}$ as follows: $$\begin{aligned}
&({\mathtt{x}}_{1}(u) - {\mathtt{x}}_{1}(v))({\mathtt{x}}_{2}(u) - {\mathtt{x}}_{2}(v)) < 0 \\
{}\iff{}&
(({\mathtt{ub}}- {\mathtt{w}}(u)) - {\mathtt{w}}(v)) (({\mathtt{lb}}-{\mathtt{w}}(u)) - {\mathtt{w}}(v)) < 0 \\
{}\iff{}&
({\mathtt{ub}}- {\mathtt{w}}(u) - {\mathtt{w}}(v) < 0 \text{ and } {\mathtt{lb}}- {\mathtt{w}}(u) - {\mathtt{w}}(v) > 0)
\text{ or } \\
& ({\mathtt{ub}}- {\mathtt{w}}(u) - {\mathtt{w}}(v) > 0 \text{ and } {\mathtt{lb}}- {\mathtt{w}}(u) - {\mathtt{w}}(v) < 0) \\
{}\iff{}&
({\mathtt{ub}}< {\mathtt{w}}(u) + {\mathtt{w}}(v) < {\mathtt{lb}}) \text{ or }
({\mathtt{lb}}< {\mathtt{w}}(u) + {\mathtt{w}}(v) < {\mathtt{ub}}) \\
{}\iff{}&
{\mathtt{lb}}< {\mathtt{w}}(u) + {\mathtt{w}}(v) < {\mathtt{ub}}.\end{aligned}$$
${\mathtt{w}}(u) < {\mathtt{lb}}/2$ and ${\mathtt{lb}}/2 \le {\mathtt{w}}(v) \le {\mathtt{ub}}/2$. $$\begin{aligned}
&({\mathtt{x}}_{1}(u) - {\mathtt{x}}_{1}(v))({\mathtt{x}}_{2}(u) - {\mathtt{x}}_{2}(v)) < 0 \\
{}\iff{}&
(({\mathtt{ub}}-{\mathtt{w}}(u)) - ({\mathtt{ub}}-{\mathtt{w}}(v))) (({\mathtt{lb}}-{\mathtt{w}}(u)) - {\mathtt{w}}(v)) < 0 \\
{}\iff{}&
(-{\mathtt{w}}(u)+{\mathtt{w}}(v)) ({\mathtt{lb}}-{\mathtt{w}}(u) - {\mathtt{w}}(v)) < 0 \\
{}\iff{}&
{\mathtt{lb}}< {\mathtt{w}}(u) + {\mathtt{w}}(v) \qquad \qquad\ \ (\because {\mathtt{w}}(u) < {\mathtt{w}}(v))\\
{}\iff{}&
{\mathtt{lb}}< {\mathtt{w}}(u)+{\mathtt{w}}(v) < {\mathtt{ub}}.
\qquad (\because {\mathtt{w}}(u) + {\mathtt{w}}(v) < {\mathtt{lb}}/2 + {\mathtt{ub}}/2 < {\mathtt{ub}})\end{aligned}$$
${\mathtt{lb}}/2 \le {\mathtt{w}}(u) \le {\mathtt{ub}}/2$ and ${\mathtt{w}}(v) > {\mathtt{ub}}/2$. $$\begin{aligned}
&({\mathtt{x}}_{1}(u) - {\mathtt{x}}_{1}(v))({\mathtt{x}}_{2}(u) - {\mathtt{x}}_{2}(v)) < 0
\\ {}\iff{}&
(({\mathtt{ub}}-{\mathtt{w}}(u)) - {\mathtt{w}}(v)) ({\mathtt{w}}(u) - {\mathtt{w}}(v)) < 0
\\ {}\iff{}&
{\mathtt{w}}(u)+{\mathtt{w}}(v) < {\mathtt{ub}}\qquad \qquad\ \ (\because {\mathtt{w}}(u) < {\mathtt{w}}(v))
\\ {}\iff{}&
{\mathtt{lb}}< {\mathtt{w}}(u)+{\mathtt{w}}(v) < {\mathtt{ub}}.
\qquad (\because {\mathtt{w}}(u) + {\mathtt{w}}(v) > {\mathtt{lb}}/2 + {\mathtt{ub}}/2 > {\mathtt{lb}}) \end{aligned}$$
Therefore, the permutation diagram represents $G$, and thus the lemma follows.
Note that the converse of Lemma \[lem:DTGisPerm\] does not hold in general. For example, the disjoint union of two triangles is a permutation graph but not a double-threshold graph by Lemma \[lem:at-most-one-nonbipartite\]. We next show that the converse holds for bipartite graphs. An illustration of the following lemma is given in [ \[fig:lem3-2\]]{}.
![An illustration of Lemma \[lem:biperm->dtg\]. (Left) A permutation diagram of a bipartite permutation graph $G=(X,Y;E)$ represented in the form of Lemma \[lem:rightanglerep\]. In this example, $X=\{1,2,3,4\}$ and $Y=\{a,b,c,d\}$. (Right) The graph $G$ is represented as a double-threshold graph. The graph is shown in the form of [ \[fig:dtg\_example1\]]{}. The construction is based on the proof of Lemma \[lem:biperm->dtg\]. []{data-label="fig:lem3-2"}](\figdir/lem3-2){width="\textwidth"}
\[lem:biperm->dtg\] Every bipartite permutation graph is a double-threshold graph.
Let $G = (X,Y;E)$ be a bipartite permutation graph. By Lemma \[lem:rightanglerep\], $G$ can be represented by a permutation diagram in which ${\mathtt{x}}_{2}(x) = {\mathtt{x}}_{1}(x) + 1$ for $x \in X$ and ${\mathtt{x}}_{2}(y) = {\mathtt{x}}_{1}(y) - 1$ for $y \in Y$. Thus, for $x \in X$ and $y \in Y$, $xy \in E$ if and only if $0 \le {\mathtt{x}}_{1}(y) - {\mathtt{x}}_{1}(x) \le 2$. Without loss of generality, we assume that ${\mathtt{x}}_{1}(z) > 1$ for every $z \in X \cup Y$.
We set ${\mathtt{lb}}= 0$ and ${\mathtt{ub}}= 2$. We set ${\mathtt{w}}(x) = -{\mathtt{x}}_{1}(x)$ for each $x \in X$, and ${\mathtt{w}}(y) = {\mathtt{x}}_{1}(y)$ for each $y \in Y$. Now for $x \in X$ and $y \in Y$, $$0 \le {\mathtt{w}}(y) + {\mathtt{w}}(x) \le 2
\iff
0 \le {\mathtt{x}}_{1}(y) - {\mathtt{x}}_{1}(x) \le 2.$$ By the assumption that ${\mathtt{x}}_{1}(z) > 1$ for every $z \in X \cup Y$, we have ${\mathtt{w}}(x) + {\mathtt{w}}(x') < -2$ for all $x,x' \in X$ and ${\mathtt{w}}(y) + {\mathtt{w}}(y') > 2$ for all $y,y' \in Y$. Therefore, $G$ is a double-threshold graph defined by ${\mathtt{w}}$, ${\mathtt{lb}}= 0$, and ${\mathtt{ub}}= 2$.
\[cor:bipartite\] The bipartite double-threshold graphs are exactly the bipartite permutation graphs.
By definition, a bipartite permutation graph is bipartite. Thus, the class of bipartite permutation graphs is contained in the class of bipartite graphs. By Lemma \[lem:biperm->dtg\], the class of bipartite permutation graphs is contained in the class of double-threshold graphs. Hence, the class of bipartite permutation graphs is contained in the class of bipartite double-threshold graphs.
On the other hand, by Lemma \[lem:DTGisPerm\], every double-threshold graph is a permutation graph. Thus, the class of bipartite double-threshold graphs is contained in the class of bipartite permutation graphs.
New characterization {#sec:characterization}
====================
Let $G=(V,E)$ be a graph. From $G$ and a vertex subset $M \subseteq V$, we construct an auxiliary bipartite graph $G'_{M}=(V',E')$ defined as follows: $$\begin{aligned}
V' &= \{v,\overline{v} \mid v \in V\},\\
E' &= \{u\overline{v} \mid uv \in E\} \cup \{v\overline{v} \mid v \in M\}.\end{aligned}$$ An example is given in [ \[fig:auxgraph\_example1\]]{}.
![An example of an auxiliary bipartite graph. (Left) A graph $G$, and $M=\{e,f\}$. (Right) The auxiliary bipartite graph $G'_M$. []{data-label="fig:auxgraph_example1"}](\figdir/auxgraph_example1)
\[lem:GMconn\] For a connected non-bipartite graph $G=(V, E)$ and a vertex subset $M \subseteq V$, $G'_{M}$ is connected.
For any $u, v \in V$, since $G$ is connected and non-bipartite, $G$ contains both an odd walk and an even walk from $u$ to $v$. This shows that $G'_{M}$ contains walks from $u$ to $v$, from $u$ to $\bar v$, from $\bar u$ to $v$, and from $\bar u$ to $\bar v$. Hence, $G'_{M}$ is connected.
For the auxiliary graph $G'_{M} = (V', E')$ of $G=(V,E)$, a linear ordering on $V'$ represented by $\langle w_1, w_2, \dots, w_{2n} \rangle$ is *symmetric* if $w_i=v$ implies $w_{2n-i+1}=\bar v$ for any $v \in V$ and any $i \in \{1,2,\dots, 2n\}$.
The next is the key lemma for our characterization.
\[lem:key-lemma\] Let $G=(V,E)$ be a non-bipartite graph and $M \subseteq V$. The following are equivalent.
1. $G$ is a double-threshold graph defined by ${\mathtt{w}}\colon V \to {\mathbb{R}}$ and ${\mathtt{lb}},{\mathtt{ub}}\in {\mathbb{R}}$ such that $M = \{v \in V \mid {\mathtt{lb}}/2 \le {\mathtt{w}}(v) \le {\mathtt{ub}}/2\}$. \[itm:dtg\_with\_M\]
2. the auxiliary graph $G'_{M} = (V',E')$ can be represented by a permutation diagram in which both $\prec_1$ and $\prec_2$ are symmetric.\[itm:sym\_diagram\]
(\[itm:dtg\_with\_M\] $\implies$ \[itm:sym\_diagram\]) An illustration is given in [ \[fig:lem4-2\]]{}. Let $G$ be a double-threshold graph defined by ${\mathtt{w}}\colon V \to {\mathbb{R}}$ and ${\mathtt{lb}}, {\mathtt{ub}}\in {\mathbb{R}}$ such that $M = \{v \in V \mid {\mathtt{lb}}/2 \le {\mathtt{w}}(v) \le {\mathtt{ub}}/2\}$. By Lemma \[lem:nice-representation\], we can assume that ${\mathtt{lb}}= 0$ and ${\mathtt{ub}}= 2$, that ${\mathtt{w}}(u) + {\mathtt{w}}(v) \notin \{0,2\}$ for every $(u,v) \in V^{2}$, and that ${\mathtt{w}}(u) \ne {\mathtt{w}}(v)$ if $u \ne v$. We construct a permutation diagram of $G'_{M}$ as follows. Let $\ell_{1}$ and $\ell_{2}$ be two horizontal parallel lines. For each vertex $w \in V'$, we set the $x$-coordinates ${\mathtt{x}}_{1}(w)$ and ${\mathtt{x}}_{2}(w)$ on $\ell_{1}$ and $\ell_{2}$ as follows: for any $v\in V$, $$\begin{aligned}
{\mathtt{x}}_1(v)&={\mathtt{w}}(v)-1, & {\mathtt{x}}_1(\bar v)&=1-{\mathtt{w}}(v), \\
{\mathtt{x}}_2(v)&={\mathtt{w}}(v), & {\mathtt{x}}_2(\bar v)&=-{\mathtt{w}}(v).\end{aligned}$$ Since ${\mathtt{w}}(u) + {\mathtt{w}}(v) \notin \{0,2\}$ for every $(u,v) \in V^{2}$ and ${\mathtt{w}}(u) \ne {\mathtt{w}}(v)$ if $u \ne v$, the $x$-coordinates are distinct on $\ell_{1}$ and on $\ell_{2}$. By connecting ${\mathtt{x}}_{1}(w)$ and ${\mathtt{x}}_{2}(w)$ with a line segment for each $w \in V'$, we get a permutation diagram. The line segments corresponding to the vertices in $V$ have negative slopes, and the ones corresponding to the vertices in $V' \setminus V$ have positive slopes. Thus, for any two vertices $u,v \in V$, the line segments corresponding to $u$ and $\bar{v}$ cross if and only if both ${\mathtt{x}}_1(u) \le {\mathtt{x}}_1(\bar v)$ and ${\mathtt{x}}_2(u) \ge {\mathtt{x}}_2(\bar v)$ hold, which is equivalent to $0 \le {\mathtt{w}}(u)+{\mathtt{w}}(v) \le 2$, and thus to $u \bar{v} \in E'$. Similarly, the line segments corresponding to $v$ and $\bar{v}$ cross if and only if $0 \le 2 {\mathtt{w}}(v) \le 2$, i.e., $v \in M$. This shows that the obtained permutation diagram represents $G'_{M}$. Let $\prec_{1}$ be the ordering on $V'$ defined by ${\mathtt{x}}_{1}$. Since ${\mathtt{x}}_{1}(v) = - {\mathtt{x}}_{1}(\bar{v})$ for each $v \in V$, if $v$ is the $i$th vertex in $\prec_{1}$, then $\bar{v}$ is the $i$th vertex in $\prec_{1}^{R}$. This implies that $\bar{v}$ is the $(2n-i+1)$st vertex in $\prec_{1}$, and thus $\prec_{1}$ is symmetric. In the same way, we can show that the ordering $\prec_{2}$ defined by ${\mathtt{x}}_{2}$ is symmetric.
![An illustration of (\[itm:dtg\_with\_M\] $\implies$ \[itm:sym\_diagram\]) in Lemma \[lem:key-lemma\]. (Top left) A double-threshold graph $G$ with $M=\{4,5\}$. The auxiliary bipartite graph $G'_M$ is also depicted. (Top right) A representation of $G$ in the form of [ \[fig:dtg\_example1\]]{}. (Bottom) A permutation diagram of $G'_M$ as given in the proof. []{data-label="fig:lem4-2"}](\figdir/lem4-2)
(\[itm:sym\_diagram\] $\implies$ \[itm:dtg\_with\_M\]) Suppose we are given a permutation diagram of $G'_{M}$ in which both $\prec_1$ and $\prec_2$ are symmetric. We may assume by symmetry that the first vertex in $\prec_{1}$ belongs to $V$. Since $G'_{M}$ is connected by Lemma \[lem:GMconn\], Corollary \[cor:rightanglerep\] shows that we can represent $G'_{M}$ by a permutation diagram in which the $x$-coordinates ${\mathtt{x}}_{1}$ and ${\mathtt{x}}_{2}$ on $\ell_{1}$ and $\ell_{2}$ satisfy that $${\mathtt{x}}_2(v)={\mathtt{x}}_1(v)+1 \quad \mbox{and} \quad {\mathtt{x}}_2(\bar v)={\mathtt{x}}_1(\bar v)-1 \quad (v\in V) \label{eq:x1x2}$$ and that the orderings of the $x$-coordinates on $\ell_{1}$ and $\ell_{2}$ are consistent with $\prec_{1}$ and $\prec_{2}$, respectively. Since $\prec_{1}$ is symmetric, if $u, v \in V$ are the $i$th and the $j$th vertices in $\prec_{1}$, then $\bar{u}, \bar{v}$ are the $(2n-i+1)$st and the $(2n-j+1)$st vertices in $\prec_{1}$. Since $i < 2n-j+1$ is equivalent to $j < 2n-i+1$, we have that $u \prec_{1} \bar{v}$ if and only if $v \prec_{1} \bar{u}$. As ${\mathtt{x}}_{1}$ is consistent with $\prec_{1}$, it holds for $u, v \in V$ that ${\mathtt{x}}_1(u) \le {\mathtt{x}}_1(\bar v)$ if and only if ${\mathtt{x}}_1(v) \le {\mathtt{x}}_1(\bar u)$, and hence $${\mathtt{x}}_1(u) \le {\mathtt{x}}_1(\bar v)
\iff
{\mathtt{x}}_1(u)+{\mathtt{x}}_1(v) \le {\mathtt{x}}_1(\bar v)+{\mathtt{x}}_1(\bar u).$$ Similarly, we can show that for $u,v \in V$, $${\mathtt{x}}_2(u) \ge {\mathtt{x}}_2(\bar v)
\iff
{\mathtt{x}}_2(u)+{\mathtt{x}}_2(v) \ge {\mathtt{x}}_2(\bar v)+{\mathtt{x}}_2(\bar u).$$ Thus, for any two distinct vertices $u,v \in V$, it holds that $$\begin{aligned}
uv \in E
& \iff u\bar v \in E' \notag \\
& \iff {\mathtt{x}}_1(u) \le {\mathtt{x}}_1(\bar v) \mbox{ and } {\mathtt{x}}_2(u) \ge {\mathtt{x}}_2(\bar v) \notag \\
& \iff {\mathtt{x}}_1(u)+{\mathtt{x}}_1(v) \le {\mathtt{x}}_1(\bar v) + {\mathtt{x}}_1(\bar u) \mbox{ and } {\mathtt{x}}_2(u)+{\mathtt{x}}_2(v) \ge {\mathtt{x}}_2(\bar v)+{\mathtt{x}}_2(\bar u). \label{eq:weight01}
$$ For each $v\in V$, define $${\mathtt{w}}(v)= \frac{{\mathtt{x}}_2(v)-{\mathtt{x}}_2(\bar v)}{2}.$$ By (\[eq:x1x2\]), we can see that (\[eq:weight01\]) is equivalent to $$0 \le {\mathtt{w}}(u)+{\mathtt{w}}(v) \le 2,$$ which shows that ${\mathtt{w}}$, ${\mathtt{lb}}=0$, and ${\mathtt{ub}}= 2$ define $G$. Furthermore, for any $v\in V$, $$\begin{aligned}
v \in M &\iff v \bar v \in E' \\
& \iff {\mathtt{x}}_1(v) \le {\mathtt{x}}_1(\bar v) \mbox{ and } {\mathtt{x}}_2(v) \ge {\mathtt{x}}_2(\bar v) \\
& \iff 0 \le {\mathtt{w}}(v) \le 1, \end{aligned}$$ which shows that $M = \{v \in V \mid 0\le {\mathtt{w}}(v) \le 1\}$.
To utilize Lemma \[lem:key-lemma\], we need to find the set $M$ of mid-weight vertices; that is, the vertices with weights in the range $[{\mathtt{lb}}/2, {\mathtt{ub}}/2]$. The first observation is that $M$ has to be a clique as the weight sum of any two vertices in $M$ is in the range $[{\mathtt{lb}}, {\mathtt{ub}}]$. In the following, we show that a special kind of maximum cliques can be chosen as $M$. To this end, we first prove that we only need to consider (inclusion-wise) maximal cliques.
\[lem:mid-weight\_maximal-clique\] For a non-bipartite double-threshold graph $G = (V,E)$, there exist ${\mathtt{w}}\colon V \to {\mathbb{R}}$ and ${\mathtt{lb}}, {\mathtt{ub}}\in {\mathbb{R}}$ defining $G$ such that $\{v \in V \mid {\mathtt{lb}}/2 \le {\mathtt{w}}(v) \le {\mathtt{ub}}/2\}$ is a maximal clique of $G$.
Let $G$ be a non-bipartite double-threshold graph $G = (V,E)$ defined by ${\mathtt{w}}\colon V \to {\mathbb{R}}$ and ${\mathtt{lb}}, {\mathtt{ub}}\in {\mathbb{R}}$. Let $M = \{v \in V \mid {\mathtt{lb}}/2 \le {\mathtt{w}}(v) \le {\mathtt{ub}}/2\}$. We choose ${\mathtt{w}}$, ${\mathtt{lb}}$, and ${\mathtt{ub}}$ in such a way that for any ${\mathtt{w}}'\colon V \to {\mathbb{R}}$ and ${\mathtt{lb}}', {\mathtt{ub}}' \in {\mathbb{R}}$ defining $G$, $M$ is not a proper subset of $\{v \in V \mid {\mathtt{lb}}'/2 \le {\mathtt{w}}'(v) \le {\mathtt{ub}}'/2\}$. Suppose to the contrary that $M$ is not a maximal clique of $G$. Observe that if $M = \emptyset$, then $V$ can be partitioned into two independent sets $\{v \in V \mid {\mathtt{w}}(v) < {\mathtt{lb}}/2\}$ and $\{v \in V \mid {\mathtt{w}}(v) > {\mathtt{ub}}/2\}$, which is a contradiction to the non-bipartiteness of $G$. Hence, $M$ is non-empty.
Let $G'_{M}$ be the auxiliary graph constructed from $G$ and $M$ as before. By Lemma \[lem:key-lemma\], $G'_{M}$ has a permutation diagram in which both ${\prec_{1}} = \langle w_{1}, \dots, w_{2n} \rangle$ and ${\prec_{2}} = \langle w'_{1}, \dots, w'_{2n} \rangle$ are symmetric. Let $\overline{M} = \{\bar{v} \mid v \in M\}$. By the definition of $G'_{M}$, $M \cup \overline{M}$ induces a complete bipartite graph in $G'_{M}$. By symmetry, we may assume that $M \prec_{1} \overline{M}$ and $\overline{M} \prec_{2} M$. That is, in $\prec_{1}$ all vertices in $M$ appear before any vertex in $\overline{M}$ appears, and in $\prec_{2}$ all vertices in $\overline{M}$ appear before any vertex in $M$ appears. Note that these assumptions imply that for each edge $x \bar{y} \in E(G'_{M})$, $x \prec_{1} \bar{y}$ and $\bar{y} \prec_{2} x$ hold.
As $M$ is not a maximal clique in $G$, there is a vertex $v \notin M$ such that $M \subseteq N_{G}(v)$. If $\bar{v} \prec_{1} v$, then we have $$\begin{aligned}
M \prec_{1} \bar{v} \prec_{1} v \prec_{1} \overline{M} \quad \text{ and } \quad
\bar{v} \prec_{2} \overline{M} \prec_{2} M \prec_{2} v \label{eq:maximal_the-1st-case}\end{aligned}$$ since $v \bar{v} \notin E(G'_{M})$, $\overline{M} \subseteq N_{G'_{M}}(v)$, and $M \subseteq N_{G'_{M}}(\bar{v})$. Similarly, if $v \prec_{1} \bar{v}$, then we have $$\begin{aligned}
v \prec_1 M \prec_{1} \overline{M} \prec_{1} \bar{v} \quad \text{ and } \quad
\overline{M} \prec_{2} v \prec_{2} \bar{v} \prec_{2} M,\end{aligned}$$ or equivalently, $$\begin{aligned}
M \prec_{2}^{R} \bar{v} \prec_{2}^{R} v \prec_{2}^{R} \overline{M} \quad \text{ and } \quad
\bar{v} \prec_{1}^{R} \overline{M} \prec_{1}^{R} M \prec_{1}^{R} v.\end{aligned}$$ Thus, by replacing $\prec_{1}$ with $\prec_{2}^{R}$ and $\prec_{2}$ with $\prec_{1}^{R}$ if necessary, we may assume that holds (see [ \[fig:mid-weight\]]{} (Left)). We further assume that $v$ has the smallest position in $\prec_{1}$ under these conditions.
\[clm:center\] $w_{n+1} = v$ (and thus $w_{n} = \bar{v}$).
\[Proof of Claim \[clm:center\]\] By the symmetry of $\langle w_{1}, \dots, w_{2n} \rangle$, it suffices to show that there is no vertex $x \in V$ such that $\bar{v} \prec_{1} x \prec_{1} v$. Suppose that such a vertex $x$ exists. In $G'_{M}$, $x$ is not adjacent to $\bar{v}$. This implies that $xv \notin E$, and hence $x \notin M$. On the other hand, in $G'_{M}$, $x$ is adjacent to all vertices in $\overline{M}$. Thus, we have $M \subseteq N_{G}(x)$. This contradicts that $v$ has the smallest position in $\prec_{1}$ under those conditions.
Now we obtain $\prec'_{1}$ from $\prec_{1}$ by swapping $v$ and $\bar{v}$ (see [ \[fig:mid-weight\]]{} (Right)). By Claim \[clm:center\], this new ordering $\prec'_{1}$ gives (together with $\prec_{2}$) the graph obtained from $G'_{M}$ by adding the edge $v\bar{v}$. Observe that this new graph can be expressed as $G'_{M \cup \{v\}}$. Since $\prec'_{1}$ and $\prec_{2}$ are symmetric, Lemma \[lem:key-lemma\] implies that there are ${\mathtt{w}}'\colon V \to {\mathbb{R}}$ and ${\mathtt{lb}}', {\mathtt{ub}}' \in {\mathbb{R}}$ defining $G$ such that $\{u \in V \mid {\mathtt{lb}}'/2 \le {\mathtt{w}}'(u) \le {\mathtt{ub}}'/2\} = M \cup \{v\}$. This contradicts the choice of ${\mathtt{w}}$, ${\mathtt{lb}}$, and ${\mathtt{ub}}$.
![(Left) Relative positions of $v$, $\bar{v}$, $M$, and $\overline{M}$. (Right) $\prec'_{1}$ is obtained from $\prec_{1}$ by swapping $v$ and $\bar{v}$.[]{data-label="fig:mid-weight"}](\figdir/mid-weight)
An *efficient maximum clique* $K$ of a graph $G$ is a maximum clique (i.e., a clique of the maximum size) that minimizes the degree sum $\sum_{v \in K} \deg_{G}(v)$. An example is illustrated in [ \[fig:effcliq\_example1\]]{}.
![An example of an efficient maximum clique. (Left) A double-threshold graph $G$ represented in the form of [ \[fig:dtg\_example1\]]{}. (Right) The vertices of $G$ is ordered in the increasing order of their weights. The graph $G$ has two maximum cliques $K_1=\{3,5,6,7\}$ and $K_2=\{4,5,6,7\}$. The degree sums are $\sum_{v \in K_1}\deg_G(v) = 5+4+4+4 = 17$, and $\sum_{v \in K_2}\deg_G(v) = 4+4+4+4 = 16$. Therefore, $K_2$ is the only efficient maximum clique of $G$. []{data-label="fig:effcliq_example1"}](\figdir/effcliq_example1)
We now show that every efficient maximum clique can be the set of mid-weight vertices.
\[lem:efficient-max-clique\] Let $G$ be a non-bipartite double-threshold graph. For every efficient maximum clique $K$ of $G$, there exist ${\mathtt{w}}\colon V \to {\mathbb{R}}$ and ${\mathtt{lb}}, {\mathtt{ub}}\in {\mathbb{R}}$ defining $G$ such that $K = \{v \in V \mid {\mathtt{lb}}/2 \le {\mathtt{w}}(v) \le {\mathtt{ub}}/2\}$.
Let $K$ be an efficient maximum clique of $G$. By Lemma \[lem:DTGisPerm\], $G$ is a permutation graph, and thus cannot contain an induced odd cycle of length 5 or more [@Golumbic04]. As $G$ is non-bipartite, $G$ contains $K_{3}$. This implies that $|K| \ge 3$.
By Lemma \[lem:mid-weight\_maximal-clique\], there exist ${\mathtt{w}}\colon V \to {\mathbb{R}}$ and ${\mathtt{lb}}, {\mathtt{ub}}\in {\mathbb{R}}$ defining $G$ such that $M := \{v \in V \mid {\mathtt{lb}}/2 \le {\mathtt{w}}(v) \le {\mathtt{ub}}/2\}$ is a maximal clique of $G$. Assume that ${\mathtt{w}}$, ${\mathtt{lb}}$, and ${\mathtt{ub}}$ are chosen so that the size of the symmetric difference $|M \vartriangle K| = |M \setminus K| + |K \setminus M|$ is minimized. Assume that $K \ne M$ since otherwise we are done. This implies that $K \not\subseteq M$ and $K \not\supseteq M$ as both $K$ and $M$ are maximal cliques. Observe that $G - M$ is bipartite. This implies that $|K \setminus M| \in \{1,2\}$ and that $K \cap M \ne \emptyset$ as $|K| \ge 3$. Since $K$ is a maximum clique, $|M \setminus K| \le |K \setminus M|$ holds.
Let $u \in K \setminus M$. By symmetry, we may assume that ${\mathtt{w}}(u) < {\mathtt{lb}}/2$. Note that no other vertex in $K$ has weight less than ${\mathtt{lb}}/2$ as $K$ is a clique. Let $v \in M$ be a nonneighbor of $u$ that has the minimum weight among such vertices. Such a vertex exists since $M$ is a maximal clique. Note that $v \in M \setminus K$.
We now observe that $v$ has the minimum weight in $M$. If $w \in M$ is a nonneighbor of $u$, then ${\mathtt{w}}(v) \le {\mathtt{w}}(w)$ follows from the definition of $v$. If $w \in M$ is a neighbor of $u$, then ${\mathtt{w}}(v) < {\mathtt{w}}(w)$ holds, since otherwise ${\mathtt{w}}(u) < {\mathtt{lb}}/2 \le {\mathtt{w}}(w) \le {\mathtt{w}}(v)$ and $uw, wv \in E$ imply that $uv \in E$ as ${\mathtt{lb}}\le {\mathtt{w}}(u) + {\mathtt{w}}(w) \le {\mathtt{w}}(u) + {\mathtt{w}}(v) < {\mathtt{w}}(w) + {\mathtt{w}}(v) \le {\mathtt{ub}}$.
Now we show below that $N(v) = N(u)$.
\[clm:neighborhood-in-M\] $N(u) \cap \{x \mid {\mathtt{w}}(x) < {\mathtt{lb}}/2\} = N(v) \cap \{x \mid {\mathtt{w}}(x) < {\mathtt{lb}}/2\}$.
\[Proof of Claim \[clm:neighborhood-in-M\]\] Since ${\mathtt{w}}(u) < {\mathtt{lb}}/2$, $N(u) \cap \{x \mid {\mathtt{w}}(x) < {\mathtt{lb}}/2\} = \emptyset$. Suppose to the contrary that $v$ has a neighbor $x$ with ${\mathtt{w}}(x) < {\mathtt{lb}}/2$. The maximality of $M$ implies that $x$ has a non-neighbor $y \in M$. Since $y \in M$, ${\mathtt{w}}(v) \le {\mathtt{w}}(y)$ holds. This implies that ${\mathtt{w}}(x) \le {\mathtt{w}}(v) \le {\mathtt{w}}(y)$ and $xv, vy \in E$, which contradicts that $xy \notin E$.
\[clm:neighborhood-in-M2\] $N(u) \cap M = N(v) \cap M$.
\[Proof of Claim \[clm:neighborhood-in-M2\]\] Since $M$ is a clique and $v \in M$, we have $N(v) \cap M = M \setminus \{v\}$. Thus, the claim is equivalent to $M \setminus \{v\} \subseteq N(u)$. This holds if $M \setminus K = \{v\}$. Assume that $M \setminus K = \{v, v'\}$ for some $v' \ne v$. To show the claim, it suffices to show that $uv' \in E$.
Since $|M \setminus K| \le |K \setminus M| \le 2$, we have $K \setminus M = \{u,u'\}$ for some $u' \ne u$. Since ${\mathtt{w}}(u) < {\mathtt{lb}}/2$ and $u' \in K \setminus M$, we have ${\mathtt{w}}(u') > {\mathtt{ub}}/2$. Let $w \in M \cap K$. If ${\mathtt{w}}(w) > {\mathtt{w}}(v')$, then we have $u'v, u'v' \in E$ since ${\mathtt{w}}(v) \le {\mathtt{w}}(v') < {\mathtt{w}}(w) \le {\mathtt{w}}(u')$ and $vw, v'w, wu' \in E$. This contradicts the maximality of $M$. Hence, ${\mathtt{w}}(w) \le {\mathtt{w}}(v')$ holds. This implies that $uv' \in E$ as ${\mathtt{w}}(u) \le {\mathtt{w}}(w) \le {\mathtt{w}}(v')$ and $uw, wv' \in E$.
\[clm:neighborhood-in-L\] $N(u) \cap \{x \mid {\mathtt{w}}(x) > {\mathtt{ub}}/2\} \supseteq N(v) \cap \{x \mid {\mathtt{w}}(x) > {\mathtt{ub}}/2\}$.
\[Proof of Claim \[clm:neighborhood-in-L\]\] Let $w \in K \cap M$. For $z \in N(v)$ with ${\mathtt{w}}(z) > {\mathtt{ub}}/2$, we have $z \in N(u)$ as follows: ${\mathtt{lb}}\le {\mathtt{w}}(u) + {\mathtt{w}}(w) \le {\mathtt{w}}(u) + {\mathtt{ub}}/2 < {\mathtt{w}}(u) + {\mathtt{w}}(z) < {\mathtt{lb}}/2 + {\mathtt{w}}(z) < {\mathtt{w}}(v) + {\mathtt{w}}(z) \le {\mathtt{ub}}$.
We are now ready to show that $N(v) = N(u)$. Claims \[clm:neighborhood-in-M\], \[clm:neighborhood-in-M2\], and \[clm:neighborhood-in-L\] imply that $N(v) \subseteq N(u)$. Suppose to the contrary that $N(v) \subsetneq N(u)$. We show that $K$ cannot be an efficient maximum clique in this case. Let $K' = K \setminus \{u\} \cup \{v\}$. We first argue that $K'$ is a (maximum) clique. If $K \setminus M = \{u\}$, then $K' = M$ is a clique. Assume that $K \setminus M = \{u, u'\}$ for some $u' \ne u$. As in the proof of Claim \[clm:neighborhood-in-M2\], ${\mathtt{w}}(u') > {\mathtt{ub}}/2$ holds. Let $w \in K \cap M$. Then, $vw, wu' \in E$. Since ${\mathtt{w}}(v) \le {\mathtt{w}}(w) \le {\mathtt{ub}}/2 < {\mathtt{w}}(u')$, we have $vu' \in E$. Thus, $K'$ is a clique. The assumption $N(v) \subsetneq N(u)$ implies that $\deg_{G}(v) < \deg_{G}(u)$, and thus, $$\sum_{w \in K'} \deg_{G}(w) = \left(\sum_{w \in K} \deg_{G}(w)\right) - \deg_{G}(u) + \deg_{G}(v) < \sum_{w \in K} \deg_{G}(w).$$ This contradicts that $K$ is efficient. Therefore, we conclude that $N(v) = N(u)$.
Now, we define a weight function ${\mathtt{w}}'\colon V \to {\mathbb{R}}$ by setting ${\mathtt{w}}'(u) = {\mathtt{w}}(v)$, ${\mathtt{w}}'(v) = {\mathtt{w}}(u)$, and ${\mathtt{w}}'(x) = {\mathtt{w}}(x)$ for all $x \in V \setminus \{u,v\}$. Then, ${\mathtt{w}}'$, ${\mathtt{lb}}$, and ${\mathtt{ub}}$ define $G$ and $M' := \{w \in V \mid {\mathtt{lb}}/2 \le {\mathtt{w}}'(w) \le {\mathtt{ub}}/2\} = M \cup \{u\} \setminus \{v\}$ as $N(u) = N(v)$. This contradicts the choice of ${\mathtt{w}}$ as $|M' \vartriangle K| < |M \vartriangle K|$.
Next, we show that the symmetry required in Lemma \[lem:key-lemma\] follows for free when $M$ is a clique.
\[lem:perm=sym\] Let $G = (V,E)$ be a connected non-bipartite graph and $M$ be a clique of $G$. Then, $G'_{M}$ is a permutation graph if and only if $G'_{M}$ can be represented by a permutation diagram in which both $\prec_1$ and $\prec_2$ are symmetric.
The if part is trivial. To prove the only-if part, we assume that $G'_{M}$ is a permutation graph.
We first observe that we only need to deal with the *twin-free* case. Assume that $N_{G'_{M}}(u) = N_{G'_{M}}(v)$ (or equivalently $N_{G'_{M}}(\bar{u}) = N_{G'_{M}}(\bar{v})$) for some $u, v \in V$, i.e., $u, v$ are twins in $G'_{M}$. If $G'_{M} - \{v,\bar{v}\}$ has a permutation diagram in which both permutation orderings $\prec_{1}$ and $\prec_{2}$ are symmetric, then we can obtain symmetric permutation orderings $\prec'_{1}$ and $\prec'_{2}$ of $G'_{M}$ by inserting $v$ right after $u$, and $\bar{v}$ right before $\bar{u}$ in both $\prec_{1}$ and $\prec_{2}$. Thus, it suffices to show that $G'_{M} - \{v,\bar{v}\} = (G - v)'_{M \setminus \{v\}}$ has a permutation diagram in which both permutation orderings $\prec_{1}$ and $\prec_{2}$ are symmetric.
Observe that $G-v$ is not necessarily non-bipartite, but $(G - v)'_{M \setminus \{v\}}$ is still connected. Hence, we can assume in the following that no pair of vertices in $G'_{M}$ have the same neighborhood and that $G'_{M}$ is connected (but $G$ is not necessarily non-bipartite). We also assume that $|V| \ge 2$ since otherwise the statement is trivially true.
Let $\prec_{1}$ and $\prec_{2}$ be the permutation orderings corresponding to a permutation diagram of $G'_{M}$. By Lemma \[lem:uniquerep\], the assumption of having no twins implies that $\prec_{1}$, $\prec_{2}$, $\prec_{1}^{\mathrm{R}}$, and $\prec_{2}^{\mathrm{R}}$ are all the permutation orderings of $G'_{M}$. Since $G'_{M}$ is connected, we may assume that the first vertex in $\prec_{1}$ belongs to $V$, the last in $\prec_{1}$ belongs to $V' \setminus V$, the first in $\prec_{2}$ belongs to $V' \setminus V$, and the last vertex in $\prec_{2}$ belongs to $V$. Let $\langle w_{1}, \dots, w_{2n} \rangle$ be the ordering defined by $\prec_{1}$.
Let $\varphi \colon V' \to V'$ be a map such that $\varphi(v) = \bar{v}$ and $\varphi(\bar{v}) = v$ for each $v \in V$. This map $\varphi$ is an automorphism of $G'_{M}$. Thus, $\langle \varphi(w_{1}), \dots, \varphi(w_{2n}) \rangle$ is also a permutation ordering of $G'_{M}$. Let ${\prec'} = \langle \varphi(w_{1}), \dots, \varphi(w_{2n}) \rangle$ denote this ordering. Then, $${\prec'} \in \{\prec_{1}, \prec_{2}, \prec_{1}^{\mathrm{R}}, \prec_{2}^{\mathrm{R}}\}.$$ We claim that ${\prec'} = {\prec_{1}^{\mathrm{R}}}$. First, observe that ${\prec'} \notin \{\prec_{1},\prec_{2}^{\mathrm{R}}\}$ as the first vertex of $\prec'$ belongs to $V' \setminus V$ but the first vertices of $\prec_{1}$ and $\prec_{2}^{\mathrm{R}}$ belong to $V$.
Suppose to the contrary that ${\prec'} = {\prec_{2}}$. Then, for each $w \in V'$, the positions of $w$ in $\prec_{1}$ and $\varphi(w)$ in ${\prec_{2}}$ ($= {\prec'}$) are the same. Thus, $w_{i} \prec_{1} \varphi(w_{i})$ implies $\varphi(w_{i}) \prec_{2} \varphi(\varphi(w_{i})) = w_{i}$. Hence, we have $v \bar{v} \in E(G'_{M})$ for all $v \in V$, and thus $M = V$. As $M$ is a clique, $M = V$ implies that $G$ is a complete graph $K_{|V|}$ and that $G'_{M}$ is a complete bipartite graph $K_{|V|, |V|}$. This contradicts the assumption that $G'_{M}$ has no twins as $|V| \ge 2$. Therefore, we conclude that ${\prec'} = {\prec_{1}^{\mathrm{R}}}$, and in particular that $\varphi(w_{i}) = w_{2n-i+1}$ for each $i$. This means that $w_{i} = v$ implies $w_{2n-i+1} = \bar{v}$ for all $v \in V$ and $i \in \{1,\dots,2n\}$. Hence, $\prec_{1}$ is symmetric.
By putting the facts above together, we obtain the following characterization of non-bipartite double-threshold graphs.
\[thm:non-bipartite\] For a non-bipartite graph $G$, the following are equivalent.
1. $G$ is a double-threshold graph. \[itm:chara\_dtg\]
2. For every efficient maximum clique $M$ of $G$, the graph $G'_{M}$ is a permutation graph. \[itm:every\_emc\]
3. For some efficient maximum clique $M$ of $G$, the graph $G'_{M}$ is a permutation graph. \[itm:some\_emc\]
The implication from \[itm:every\_emc\] to \[itm:some\_emc\] is trivial. To show that \[itm:chara\_dtg\] implies \[itm:every\_emc\], assume that $G$ is a non-bipartite double-threshold graph. Let $M$ be an efficient maximum clique of $G$. By Lemma \[lem:efficient-max-clique\], there exist ${\mathtt{w}}\colon V \to {\mathbb{R}}$ and ${\mathtt{lb}}, {\mathtt{ub}}\in {\mathbb{R}}$ defining $G$ such that $M = \{v \in V \mid {\mathtt{lb}}/2 \le {\mathtt{w}}(v) \le {\mathtt{ub}}/2\}$. Now by Lemma \[lem:key-lemma\], $G'_{M}$ is a permutation graph.
We now show that \[itm:some\_emc\] implies \[itm:chara\_dtg\]. Assume that for an efficient maximum clique $M$ of a non-bipartite graph $G$, the graph $G'_{M}$ is a permutation graph.
Let $H$ be a non-bipartite component of $G$. Then, $H$ contains an induced odd cycle of length $k \ge 3$. This means that, if $H$ does not include $M$, then $G'_{M}$ contains an induced cycle of length $2k \ge 6$. However, this is a contradiction as a permutation graph cannot contain an induced cycle of length at least $5$. Thus, $H$ includes $M$. Also, there is no other non-bipartite component in $G$ as it does not intersect $M$. Since $H$ includes $M$, $H'_{M}$ is a component of $G'_{M}$. By Lemma \[lem:perm=sym\], $H'_{M}$ can be represented by a permutation diagram in which both $\prec_1$ and $\prec_2$ are symmetric, and thus $H$ is a double-threshold graph by Lemma \[lem:key-lemma\].
Let $B$ be a bipartite component of $G$ (if one exists). Since $B$ does not intersect $M$, $G'_{M}$ contains two isomorphic copies of $H$ as components. Since $G'_{M}$ is a permutation graph, $B$ is a permutation graph too. By Corollary \[cor:bipartite\], $B$ is a double-threshold graph.
The discussion so far implies that all components of $G$ are double-threshold graphs and exactly one of them is non-bipartite. By Corollary \[cor:component\], $G$ is a double-threshold graph.
Linear-time recognition algorithm {#sec:algorithm}
=================================
We are now ready to present a linear-time recognition algorithm for double-threshold graphs. Algorithm \[alg:outline\] shows the outline of our algorithm. The main result of this paper (Theorem \[thm:linear-recognition\]) follows from Lemmas \[lem:alg-correctness\] and \[lem:alg-runtime\] below.
\[thm:linear-recognition\] There exists a linear-time algorithm that takes a graph as its input and outputs <span style="font-variant:small-caps;">yes</span> if the graph is a double-threshold graph; outputs <span style="font-variant:small-caps;">no</span> otherwise.
The algorithm for Theorem \[thm:linear-recognition\] is given in Algorithm \[alg:outline\]. The proof of Theorem \[thm:linear-recognition\] boils down to the following two lemmas.
<span style="font-variant:small-caps;">yes</span> Find an efficient maximum clique $M$ of $G$ using $\prec$. <span style="font-variant:small-caps;">yes</span> <span style="font-variant:small-caps;">no</span>
\[lem:alg-correctness\] Algorithm \[alg:outline\] returns <span style="font-variant:small-caps;">yes</span> if and only if the input graph $G$ is a double-threshold graph.
Algorithm \[alg:outline\] returns <span style="font-variant:small-caps;">yes</span> if and only if $G$ is a bipartite permutation graph, or $G$ is a non-bipartite permutation graph and $G'_{M}$ is a permutation graph, where $M$ is an efficient maximum clique of $G$. The correctness follows from Corollary \[cor:bipartite\] and Theorem \[thm:non-bipartite\].
\[lem:alg-runtime\] There exists a linear-time implementation of Algorithm \[alg:outline\].
At Steps 1 and 4, deciding whether a graph is a permutation graph and, if so, computing a permutation ordering can be done in linear time [@McConnellS99]. (When we do this for bipartite graphs, we can use simpler algorithms [@SpinradBS87; @Sprague95].) At Step 2, bipartiteness can be checked in linear time by, e.g., depth-first search. Observe that $|V(G'_{M})| = 2 |V|$ and $|E(G'_{M})| = 2 |E| + |M|$. Thus, it suffices to show that an efficient maximum clique of a permutation graph can be computed in linear time at Step 3.
To find an efficient maximum clique of $G$, we set to each vertex $v \in V$ the weight ${\mathtt{w}}(v) = n^{2} - \deg_{G}(v)$, where $n = |V|$, and then find a maximum-weight clique $M$ of $G$ with respect to these weights. Using the permutation ordering of $G$ computed before, we can find $M$ in linear time [@Golumbic04 pp. 133–134]. We show that $M$ is an efficient maximum clique of $G$. Let $K$ be an efficient maximum clique of $G$. Since ${\mathtt{w}}(K) \le {\mathtt{w}}(M)$, we have $$|K| \cdot n^{2} - \sum_{v \in K} \deg_{G}(v) \le |M| \cdot n^{2} - \sum_{v \in M} \deg_{G}(v). \label{eq:efficient-max-clique}$$ Since $0 \le \sum_{v \in S} \deg_{G}(v) < n^{2}$ for any $S \subseteq V$, it holds that $ |K| \cdot n^{2} - n^{2} < |M| \cdot n^{2}$. Since $|M| \le |K|$, this implies that $|K| = |M|$. By , it holds that $\sum_{v \in K} \deg_{G}(v) \ge \sum_{v \in M} \deg_{G}(v)$. Therefore, $M$ is an efficient maximum clique.
Minimal forbidden induced subgraphs {#sec:forbidden-induced-subgraphs}
===================================
Observe that the class of double-threshold graphs is hereditary. That is, the class is closed under vertex deletions. Such a class can be characterized by the (possibly infinite) list of minimal forbidden induced subgraphs. This section is devoted to the first steps in this direction.
We first show that the graphs in [ \[fig:graph5\]]{} are the non-double-threshold graphs of order at most 5. This small list can be useful for comparing the class with others. For example, one can check that bull, butterfly, and gem are permutation graphs of co-threshold dimension 2.
![The graphs of order at most 5 that are not double-threshold graphs.[]{data-label="fig:graph5"}](\figdir/graphs5)
\[lem:mfis\_with\_order5\] The graphs $C_{5}$, bull, butterfly, and gem are the graphs of order at most 5 that are not double-threshold graphs.
We first show that all four graphs in [ \[fig:graph5\]]{} are not double-threshold graphs. By Lemma \[lem:DTGisPerm\], $C_{5}$ is not a double-threshold graph. Using our new characterization (Theorem \[thm:non-bipartite\]), we can show that bull, butterfly, and gem are non-double-threshold graphs because their auxiliary graphs contain a bipartite net or a bipartite tent as an induced subgraph and thus are not permutation graphs [@HellH04]. See [ \[fig:graph5\_no\]]{}.
![The graphs bull, butterfly, and gem are non-double-threshold graphs as $\text{bull}'_{M}[a,b,c,d,\bar{a},\bar{b},\bar{c}] \simeq \text{bipartite net}$ and $\text{butterfly}'_{M}[a,b,c,d,\bar{a},\bar{b},\bar{c}] = \text{gem}'_{M}[a,b,c,d,\bar{a},\bar{b},\bar{c}] \simeq \text{bipartite tent}$, where $M$ in each graph is the efficient maximum clique formed by the square vertices.[]{data-label="fig:graph5_no"}](\figdir/graphs5_n)
Suppose to the contrary that there is a graph $G$ of order at most 5 that is not a double-threshold graph and not isomorphic to $C_{5}$, bull, butterfly, or gem.
\[clm:mfis\_with\_order5\] $G$ is a connected graph of order exactly 5 that contains $K_{3}$ and at least one of $2K_{2}$, $C_{4}$, and $P_{4}$ as induced subgraphs.
\[Proof of Claim \[clm:mfis\_with\_order5\]\] It is known that the threshold graphs are the $\{2K_{2}, C_{4}, P_{4}\}$-free graphs [@ChvatalH73] and that all bipartite graphs of order at most 5 are permutation graphs [@HellH04]. Thus, by Corollary \[cor:bipartite\], $G$ is a non-bipartite graph of order exactly 5 that contains at least one of $2K_{2}$, $C_{4}$, and $P_{4}$ as an induced subgraph. Since $G$ is non-bipartite and non-isomorphic to $C_{5}$, $G$ has $K_{3}$ as an induced subgraph. If $G$ is not connected, then each component is of order at most 4 and at most one component is non-bipartite. Thus, $G$ is connected by Corollary \[cor:component\].
First assume that $G$ has $K_{3}$ and $2K_{2}$ as induced subgraphs. Fix an induced subgraph of $G$ isomorphic to $2K_{2}$, and let $v$ be the vertex not contained in this $2K_{2}$. The vertex $v$ has degree 3 since $G$ contains $K_{3}$, $G$ is connected, and $G$ is not isomorphic to butterfly. This graph $\overline{\text{P}}$ is a double-threshold graph as shown in [ \[fig:graph5\_y\]]{}.
Next assume that $G$ has $K_{3}$ and $P_{4}$ as induced subgraphs. Fix an induced subgraph of $G$ isomorphic to $P_{4}$, and let $v$ be the vertex not contained in this $P_{4}$. We know that $v$ has degree at most 3 since $G$ is not isomorphic to gem. Also, $v$ cannot be adjacent only to the two inner vertices as $G$ is not isomorphic to bull. Thus, $v$ is adjacent either to the first (or equivalently the last) two vertices or arbitrary three vertices in $P_{4}$. All these cases result in double-threshold graphs $\overline{\text{P}}$, co-fork, and house in [ \[fig:graph5\_y\]]{}.
Finally assume that $G$ has $K_{3}$ and $C_{4}$ as induced subgraphs. Fix an induced subgraph of $G$ isomorphic to $C_{4}$, and let $v$ be the vertex not contained in this $C_{4}$. To have $K_{3}$, $v$ has to be adjacent to two consecutive vertices in $C_{4}$. Thus, $G$ is either house, $\overline{P_{2} \cup P_{3}}$, or $W_{4}$. All these graphs are double-threshold graphs as shown in [ \[fig:graph5\_y\]]{}.
![The double-threshold graphs of order 5 that contain $K_{3}$ and at least one of $2K_{2}$, $C_{4}$, and $P_{4}$. The numbers in the brackets are the lower and upper bounds for each graph.[]{data-label="fig:graph5_y"}](\figdir/graphs5_y)
By Lemma \[lem:at-most-one-nonbipartite\], the disjoint union of two triangles, denoted $2K_{3}$, is not a double-threshold graph. Below we show that this graph is the unique disconnected graph that is a minimal forbidden induced subgraph of double-threshold graphs.
The graph $2K_{3}$ is the unique disconnected minimal forbidden induced subgraph of double-threshold graphs.
First observe that all proper induced subgraphs of $2K_{3}$ are double-threshold graphs, and thus it is a minimal forbidden induced subgraph of double-threshold graphs.
Let $G$ be a disconnected graph that is not a double-threshold graph. Assume that all connected components of $G$ are double-threshold graphs since otherwise $G$ is not minimal. If a component of $G$ is bipartite, then $G$ is not minimal by Lemma \[lem:disjoint-bicomponent\]. Thus, each component of $G$ is non-bipartite. Since a double-threshold graph cannot contain any induced odd cycle of length more than 3, each component of $G$ contains a triangle $K_{3}$. Since $G$ has at least two components, $G$ contains $2K_{3}$ as an induced subgraph.
[^1]: Strictly speaking, the names are not exactly the same. One is written with a hyphen, but the other is written without a hyphen.
[^2]: The symbols $K_{n}$ and $C_{n}$ denote the complete graph and the cycle of $n$ vertices, respectively. For a graph $G$ and an positive integer $k$, $kG$ is the disjoint union of $k$ copies of $G$. The graph *bull* is shown in [ \[fig:graph5\]]{}.
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'B. de Loynes'
bibliography:
- 'biblio.bib'
title: Random walks on oriented lattices and Martin boundary
---
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Time dynamics of isolated many-body quantum systems has long been an elusive subject. Very recently, however, meaningful experimental studies of the problem have finally become possible [@kinoshita06; @hofferberth07], stimulating theoretical interest as well [@rigol07; @kollath07; @manmana07; @burkov07; @calabrese07]. Progress in this field is perhaps most urgently needed in the foundations of quantum statistical mechanics. This is so because in generic isolated systems, one expects [@sengupta04; @berges04] nonequilibrium dynamics on its own to result in thermalization: a relaxation to states where the values of macroscopic quantities are stationary, universal with respect to widely differing initial conditions, and predictable through the time-tested recipe of statistical mechanics. However, it is not obvious what feature of many-body quantum mechanics makes quantum thermalization possible, in a sense analogous to that in which dynamical chaos makes classical thermalization possible [@gallavotti99]. For example, dynamical chaos itself cannot occur in an isolated quantum system, where time evolution is linear and the spectrum is discrete [@krylov79]. Underscoring that new rules could apply in this case, some recent studies even suggested that statistical mechanics may give wrong predictions for the outcomes of relaxation in such systems [@kollath07; @manmana07]. Here we demonstrate that an isolated generic quantum many-body system does in fact relax to a state well-described by the standard statistical mechanical prescription. Moreover, we show that time evolution itself plays a merely auxiliary role in relaxation and that thermalization happens instead at the level of individual eigenstates, as first proposed by J. M. Deutsch [@deutsch91] and M. Srednicki [@srednicki94]. A striking consequence of this [*eigenstate thermalization*]{} scenario, confirmed below for our system, is that the knowledge of a [*single*]{} many-body eigenstate suffices to compute thermal averages—any eigenstate in the microcanonical energy window will do, as they all give the same result.'
author:
- Marcos Rigol
- Vanja Dunjko
- Maxim Olshanii
title: 'Thermalization and its mechanism for generic isolated quantum systems[^1]'
---
If we pierce an inflated balloon inside a vacuum chamber, very soon we find the released air uniformly filling the enclosure and the air molecules attaining the Maxwell velocity distribution whose width depends only on their total number and energy. Different balloon shapes, placements, or piercing points all lead to the same spatial and velocity distributions. Classical physics explains this *thermodynamical universality* as follows [@gallavotti99]: almost all particle trajectories quickly begin looking alike, even if their initial points are very different, because nonlinear equations drive them to explore ergodically the constant-energy manifold, covering it uniformly with respect to precisely the microcanonical measure. However, if the system possesses further conserved quantities *functionally independent* from the Hamiltonian and each other, then time evolution is confined to a highly restricted hypersurface of the energy manifold. Hence, microcanonical predictions fail and the system does not thermalize.
On the other hand, in isolated quantum systems not only is dynamical chaos absent due to the linearity of time evolution and the discreteness of spectra [@krylov79], but it is also not clear under what conditions conserved quantities provide independent constraints on relaxation dynamics. To begin with, any operator commuting with a generic and thus nondegenerate Hamiltonian is functionally dependent on it [@sutherland04], implying that the conservation of energy is the only independent constrain. On the other hand, even when operators are functionally dependent, their expectation values—considered as functionals of states—generally are not: for example, two states may have the same mean energies but different mean square-energies. For nondegenerate Hamiltonians a maximal set of constants of motion with functionally independent expectation values is as large as the dimension of the Hilbert space; examples include the projectors $\widehat{P}_\alpha={\vert\Psi_\alpha\rangle \langle \Psi_\alpha \vert}$ to the energy eigenstates [@sutherland04] and the integer powers of the Hamiltonian [@manmana07].
The current numerical and analytic evidence from integrable systems suggests that there exists a minimal set of independent constraints whose size is much smaller than the dimension of the Hilbert space but may still be much greater than one. In our previous work [@rigol07] (with V. Yurovsky) we showed that an integrable isolated one-dimensional system of lattice hard-core bosons relaxes to an equilibrium characterized not by the usual but by a *generalized* Gibbs ensemble. Instead of just the energy, the Gibbs exponent contained a linear combination of conserved quantities—the occupations of the eigenstates of the corresponding Jordan-Wigner fermions—whose number was still only a tiny fraction of the dimension of the Hilbert space. Yet this ensemble works, while the usual one does not, for a wide variety of initial conditions [@rigol06] as well as for a fermionic system [@cazalilla06]; it also explains a recent experimental result, the absence of thermalization in the Tonks-Girardeau gas [@kinoshita06]. Thus, while at least some constraints beyond the conservation of energy must be kept, it turns out one needs only a relatively limited number of additional conserved quantities with functionally independent expectation values; adding still further ones is redundant.
Since it is not clear which sets of conserved quantities—and some are always present—constrain relaxation and which do not, it becomes even more urgent to determine whether isolated generic quantum systems relax to the usual thermal state. The theoretical attention to this question has in fact been increasing recently, because of the high levels of isolation [@kinoshita06; @hofferberth07; @jin96] and control [@greiner02; @mandel03] possible in experiments with ultracold quantum gases. However, despite numerous studies of specific models there is not yet consensus on how or even if relaxation to the usual thermal values occurs for nonintegrable systems [@calabrese07]. Common wisdom says that it does [@sengupta04; @berges04], but some recent numerical results suggest otherwise, either under certain conditions [@kollath07] or in general [@manmana07].
![ \[fig1\] **Relaxation dynamics**. **a**, Two-dimensional lattice on which five hard-core bosons propagate in time. The bosons are initially prepared in the ground state of the sub-lattice in the lower-right corner and released through the indicated link. **b**, The corresponding relaxation dynamics of the marginal momentum distribution center \[$n(k_{x}=0)$\] compared with the predictions of the three ensembles. In the microcanonical case, we averaged over all eigenstates whose energies lie within a narrow window (see Supplementary Discussion) $[E_0-\Delta E, E_0+\Delta E]$, where $E_0\equiv\langle \psi(0) | \widehat{H} | \psi(0) \rangle=-5.06 J$, $\Delta E=0.1J$, and $J$ is the hopping parameter. The canonical ensemble temperature is $k_{\textrm{\scriptsize B}}T=1.87 J$, where $k_{\textrm{\scriptsize B}}$ is the Boltzmann constant, so that the ensemble prediction for the energy is $E_0$. **c**, Full momentum distribution function in the initial state, after relaxation, and in the different ensembles. Here $d$ is the lattice constant and $L_{x}=5$ the lattice width. ](fig1)
In order to study relaxation of an isolated quantum systems, we considered the time evolution of five hard-core bosons with additional weak nearest-neighbor repulsions, on a 21-site two-dimensional lattice, initially confined to a portion of the lattice and prepared in their ground state there. Figure \[fig1\]a shows the exact geometry (see also Supplementary Discussion); the relaxation dynamics begins when the confinement is lifted. Expanding the initial state wavefunction in the eigenstate basis of the final Hamiltonian $\widehat{H}$ as ${\vert \psi(0)\rangle}=\sum_\alpha C_{\alpha} {\vert\Psi_{\alpha}\rangle}$, the many-body wavefunction evolves as ${\vert \psi(t)\rangle}=e^{-i\widehat{H} t}{\vert \psi(0)\rangle}=\sum_\alpha C_{\alpha^{}}^{} e^{-i E_\alpha t}{\vert\Psi_{\alpha}\rangle}$, where the $E_\alpha$’s are the eigenstate energies. Thus obtaining numerically-exact results for all times required the full diagonalization of the 20,349-dimensional Hamiltonian. The quantum-mechanical mean of any observable $\widehat{A}$ evolves as $$\langle \widehat{A}(t) \rangle
\equiv\langle \psi(t) | \widehat{A} | \psi(t) \rangle
=\sum_{{\alpha},\,{\beta}} C_{{\alpha}}^{\star} C_{{\beta}}^{}
e^{i(E_{{\alpha}}-E_{{\beta}})t}
A_{{\alpha}{\beta}}\, ,
\label{timeevolution}$$ with $A_{{\alpha}{\beta}} ={\langle\Psi_{\alpha'}\vert}\widehat{A} {\vert\Psi_\alpha\rangle}$. The long-time average of $\langle \widehat{A}(t) \rangle$ is then $$\begin{aligned}
\overline{\langle \widehat{A} \rangle}
= \sum_{\alpha} |C_{\alpha}|^{2}
A_{\alpha\alpha} \, .
\label{diagonal}\end{aligned}$$ Note that if the system relaxes at all, it must be to this value. We find it convenient to think of Eq. (\[diagonal\]) as stating the prediction of a “diagonal ensemble,” $|C_{\alpha}|^{2}$ corresponding to the weight ${\vert\Psi_{\alpha}\rangle}$ has in the ensemble. In fact, this ensemble is precisely the generalized Gibbs ensemble introduced in Ref. [@rigol07] if as integrals of motion one takes all the projection operators $\widehat{P}_\alpha={\vert\Psi_\alpha\rangle \langle \Psi_\alpha \vert}$. Using these as constraints on relaxation dynamics, the theory gives $\hat{\rho}_c=\exp{\left( -\sum_{\alpha=1}^D \lambda_\alpha \hat{P}_\alpha\right)}$, with $\lambda_\alpha=-\ln (\left|C_\alpha\right|^{2})$, and $D$ the dimension of the Hilbert space. (Notice, however, that for the integrable system treated in Ref. [@rigol07], the generalized Gibbs ensemble was defined using a different, [*minimal*]{} set of independent integrals of motion, whose number was equal to the number of lattice sites $N\ll D$.)
Now if the quantum-mechanical mean of an observable behaves thermally it should settle to the prediction of an appropriate statistical-mechanical ensemble. For our numerical experiments we chose to monitor the marginal momentum distribution along the horizontal axis $n(k_{x})$ and its central component $n(k_{x}=$ $0)$ (see Supplementary Discussion). Figures \[fig1\]b and \[fig1\]c demonstrate that both relax to their microcanonical predictions. The diagonal ensemble predictions are indistinct from these, but the canonical ones, although quite close, are not. This is an indication of the relevance of finite size effects, which may be the origin of some of the apparent deviations from thermodynamics seen in the recent numerical studies of Refs. [@kollath07] and [@manmana07].
The statement that the diagonal and microcanonical ensembles give the same predictions for the relaxed value of $\widehat{A}$ reads $$\begin{aligned}
\sum_{\alpha} |C_{\alpha}|^{2}
A_{\alpha\alpha} &=&
\langle A \rangle_{\text{microcan.}}(E_{0})\, \nonumber \\
& \stackrel{\text{def.}}{=}&\,
\frac{1}{\mathcal{N}_{E_{0},\, \Delta E}} \sum_{\substack{\alpha \\ \left|E_{0}-E_{\alpha}\right|<\Delta E}}\,A_{\alpha\alpha}
\, ,
\label{paradox}\end{aligned}$$ where $E_{0}$ is the mean energy of the initial state, $\Delta E$ is the half-width of an appropriately chosen (see Supplementary Discussion) energy window centered at $E_{0}$, and the normalization $\mathcal{N}_{E_{0},\, \Delta E}$ is the number of energy eigenstates with energies in the window $\left[E_{0}-\Delta E,\, E_{0}+\Delta E\right]$. Thermodynamical universality is evident in this equality: while the left hand side depends on the details of the initial conditions through the set of coefficients $C_{\alpha}$, the right hand side depends only on the total energy, which is the same for many different initial conditions. Three mechanisms suggest themselves as possible explanations of this universality (assuming the initial state is sufficiently narrow in energy, as is normally the case—see Supplementary Discussion): (i) Even for eigenstates close in energy, there are large eigenstate-to-eigenstate fluctuations of both the eigenstate expectation values $A_{\alpha\alpha}$ and of the eigenstate occupation numbers $|C_{\alpha}|^{2}$. However, for physically interesting initial conditions, the fluctuations in the two quantities are uncorrelated. A given initial state then performs an unbiased sampling of the distribution of the eigenstate expectation values $A_{\alpha\alpha}$, resulting in Eq. (\[paradox\]).
\(ii) For physically interesting initial conditions, the eigenstate occupation numbers $|C_{\alpha}|^{2}$ practically do not fluctuate at all between eigenstates that are close in energy. Again, Eq. (\[paradox\]) immediately follows.
\(iii) The eigenstate expectation values $A_{\alpha\alpha}$ practically do not fluctuate at all between eigenstates that are close in energy. In that case Eq. (\[paradox\]) holds for literally *all* initial states narrow in energy.
![ \[fig2\] **Thermalization in classical vs quantum mechanics**. **a**, In classical mechanics, time evolution constructs the thermal state from an initial state that generally bears no resemblance to the former. **b**, In quantum mechanics, according to the eigenstate thermalization hypothesis, every eigenstate of the Hamiltonian always implicitly contains a thermal state. The coherence between the eigenstates initially hides it, but time dynamics reveals it through dephasing. ](fig2)
J. M. Deutsch and M. Srednicki have independently proposed the scenario (iii), dubbed the
- \[Deutsch[@deutsch91] (1991), Srednicki[@srednicki94] (1994)\].\
The expectation value ${\langle\Psi_\alpha\vert}\widehat{A}{\vert\Psi_{\alpha}\rangle}$ of a few-body observable $\widehat{A}$ in an eigenstate of the Hamiltonian ${\vert\Psi_{\alpha}\rangle}$, with energy $E_{\alpha}$, of a large interacting many-body system equals the thermal (microcanonical in our case) average $\langle A \rangle_{\textrm{microcan.}}(E_{\alpha})$ of $\widehat{A}$ at the mean energy $E_{\alpha}$: $${\langle\Psi_\alpha\vert}\widehat{A} {\vert\Psi_{\alpha}\rangle}= \langle A
\rangle_{\textrm{microcan.}}(E_{\alpha}) .
\label{main_eigenstate_thermalization}$$
The ETH suggests that classical and quantum thermal states have very different natures, as depicted in Fig. \[fig2\]. While at present there are no general theoretical arguments supporting the ETH, some results do exist for restricted classes of systems. To begin with, Deutsch [@deutsch91] showed that the ETH holds in the case of an integrable Hamiltonian weakly perturbed by a single matrix taken from a random Gaussian ensemble. Next, nuclear shell model calculations have shown that individual wavefunctions reproduce thermodynamic predictions [@horoi95]. Then there are rigorous proofs that some particular quantum systems, whose classical counterparts are chaotic, satisfy the ETH in the semiclassical limit [@shnirelman74; @voros79; @colin85; @zelditch87]. More generally, for low density billiards in the semiclassical regime, the ETH follows from Berry’s conjecture [@srednicki94; @heller07], which in turn is believed to hold in semiclassical classically-chaotic systems [@berry77]. Finally, at the other end of the chaos-integrability continuum, in systems solvable by Bethe ansatz, observables are smooth functions of the integrals of motion. This allows the construction of single energy eigenstates that reproduce thermal predictions [@korepin93]. In Figs. \[fig3\]a-c we demonstrate that the ETH [*is*]{} in fact the mechanism responsible for thermal behavior in our nonintegrable system. Fig. \[fig3\]c additionally shows that scenario (ii) mentioned above plays no role, because the fluctuations in the eigenstate occupation numbers $|C_{\alpha}|^{2}$ are large. Thermal behavior also requires that both the diagonal and the chosen thermal ensemble have sufficiently narrow energy distributions $\rho(E)$ \[$\,=$ probability distribution $\times$ the density of states\], so that in the energy region where the energy distributions $\rho(E)$ are appreciable, the derivative of the curve eigenstate expectation value $A_{\alpha\alpha}$ vs the energy (here $n(k_x=0)$ vs the energy) does not change much; see Supplementary Discussion. As shown in Fig. \[fig3\]b, this holds for the microcanonical and diagonal ensembles but not for the canonical ensemble, explaining the failure of the latter to describe the relaxation in Fig. \[fig1\]. Note that the fluctuations of the eigenstate occupation numbers $|C_{\alpha}|^{2}$ in Fig. \[fig3\]b are lowered by the averaging involved in the computation of the density of states (compare with Fig. \[fig3\]c).
To strengthen the case for the ETH, we tested another observable. We chose it with the following consideration in mind: in our system interactions are local in space, and momentum distribution is a global, approximately spatially additive property. Thus one might wonder if the ETH for momentum distribution arises through some spatial averaging mechanism (we thank the anonymous referee 2 for bringing this point to our attention). It does not: for our final test of the ETH we chose an observable that is manifestly local in space, the expectation value of the occupation of the central site of the lattice. We again found that the ETH holds true (3% relative standard deviation of eigenstate-to-eigenstate fluctuations).
On the other hand, Figs. \[fig3\]d-f show how the ETH [*fails*]{} for an isolated one-dimensional [*integrable*]{} system. The latter consists of five hard-core bosons initially prepared in their ground state in an 8-site chain, one of the ends of which we then link to one of the ends of an adjoining (empty) 13-site chain to trigger relaxation dynamics. As Fig. \[fig3\]e shows, $n(k_x)$ as a function of energy is a broad cloud of points, meaning that the ETH is not valid; Fig. \[fig3\]f shows that scenario (ii) does not hold either.
![\[fig3\] **Eigenstate thermalization hypothesis**. **a**, In our nonintegrable system, the momentum distribution $n(k_x)$ for two typical eigenstates with energies close to $E_0$ is identical to the microcanonical result, in accordance with the ETH. **b**, Upper panel: $n(k_x=0)$ eigenstate expectation values as a function of the eigenstate energy resemble a smooth curve. Lower panel: the energy distribution $\rho(E)$ of the three ensembles considered in this work. **c**, Detailed view of $n(k_x=0)$ (left labels) and $|C_\alpha|^2$ (right labels) for 20 eigenstates around $E_0$. **d**, In the integrable system, $n(k_x)$ for two eigenstates with energies close to $E_0$ and for the microcanonical and diagonal ensembles are very different from each other, i.e., the ETH fails. **e**, Upper panel: $n(k_x=0)$ eigenstate expectation value considered as a function of the eigenstate energy gives a thick cloud of points rather than resembling a smooth curve. Lower panel: energy distributions in the integrable system are similar to the nonintegrable ones depicted in **b**. **f**, Correlation between $n(k_x=0)$ and $|C_\alpha|^2$ for 20 eigenstates around $E_0$. It explains why in **d** the microcanonical prediction for $n(k_x=0)$ is larger than the diagonal one. ](fig3)
Nevertheless, one may still wonder if in this case scenario (i) might hold—if the averages over the diagonal and the microcanonical energy distributions shown in Fig. \[fig3\]e might agree. Figure \[fig3\]d shows that this does not happen. This is so because, as shown in Fig. \[fig3\]f, the values of $n(k_x=0)$ for the most-occupied states in the diagonal ensemble (the largest values of eigenstate occupation numbers $|C_\alpha|^2$) are always smaller than the microcanonical prediction, and those of the least-occupied states, always larger. Hence, the usual thermal predictions fail because the correlations between the values of $n(k_x=0)$ and $|C_\alpha|^2$ preclude unbiased sampling of the latter by the former. These correlations have their origin in the nontrivial integrals of motion that make the system integrable and that enter the [*generalized*]{} Gibbs ensemble, which was introduced in Ref. [@rigol07] as appropriate for formulating statistical mechanics of isolated integrable systems. In the nonintegrable case shown in Fig. \[fig3\]c, $n(k_x=0)$ is so narrowly distributed that it does not matter whether or not it is correlated with $|C_\alpha|^2$ (we have in fact seen no correlations in the nonintegrable case).
The thermalization mechanism outlined thus far explains why long-time averages converge to their thermal predictions. A striking aspect of Fig. \[fig1\]b, however, is that the time fluctuations are so small that after relaxation the thermal prediction works well at every instant of time. Looking at Eq. (\[timeevolution\]), one might think this is so because the contribution of the off-diagonal terms gets attenuated by temporal dephasing, which results from the generic incommensurability of the frequencies of the oscillating exponentials. However, this attenuation only scales as the root of the number of dephasing terms, and is exactly compensated by their larger number: if the number of eigenstates that have a significant overlap with the initial state is $N_{\textrm{states}}$, then typical $C_{\alpha}\sim 1/\sqrt{N_{\textrm{states}}}$, and the sum over off-diagonal terms in Eq. (\[timeevolution\]) finally does not scale down with $N_{\textrm{states}}$: $$\sum_{\substack{{\alpha},\,{\beta}\\ {\alpha}\neq
{\beta}}}\frac{e^{i(E_{{\alpha}}-E_{{\beta}})t}}{N_{\textrm{states}}}
A_{{\alpha}{\beta}} \sim
\frac{\sqrt{N_{\textrm{states}}^{2}}}{N_{\textrm{states}}}A_{{\alpha}{\beta}}^{\text{typical}}
\sim A_{{\alpha}{\beta}}^{\text{typical}}$$ Hence, were the magnitude of the diagonal and off-diagonal terms comparable, their contributions would be comparable as well, and time fluctuations of the average would be of the order of the average. However, this is not the case and thus $$\begin{aligned}
A_{\stackrel{\scriptstyle {\alpha}{\beta}}{{\alpha}\neq {\beta}}}^{\text{typical}} \ll
A_{\alpha^{}\alpha^{}}^{\text{typical}}.
\label{small_off-diagonal}\end{aligned}$$ Figure \[fig4\]a confirms this inequality for the matrix elements of the momentum distribution in our system. We should mention that there is an [*a priori*]{} argument—admittedly in part dependent on certain hypotheses about chaos in quantum billiards—in support of this inequality for the case when the mean value of $\widehat{A}$ in an energy eigenstate is comparable to the quantum fluctuation of $\widehat{A}$ in that state [@srednicki94a].
![ \[fig4\] **Temporal vs quantum fluctuations**. **a**, Matrix elements of the observable of interest, $n(k_{x}=0)$, as a function of state indices; the eigenstates of the Hamiltonian are indexed in the order of diminishing overlap with the initial state. The dominance of the diagonal matrix elements is apparent. **b**, The same time evolution as in Fig. \[fig1\]b with the error bars showing the quantum fluctuations $n(k_{x}=0)\pm \Delta$ with $\Delta=[\langle \widehat{n}^2(k_x=0)\rangle-\langle \widehat{n}(k_x=0)\rangle^2]^{1/2}$, which are clearly much larger than the temporal fluctuations of $n(k_{x}=0)$. ](fig4)
On the other hand, the thermalization we see appears to be working a bit [*too well*]{}: in a system as small as ours, one would expect measurement-to-measurement fluctuations to be much larger than what Fig. \[fig1\]b suggests. Indeed, as we show in Figure \[fig4\]b, the fluctuations that one would actually measure would be dominated by the quantum fluctuations of the time-dependent state. The rather large size of the quantum fluctuations relative to the thermal mean value is of course particular to small systems; however, the dominance of the quantum fluctuations over the temporal fluctuations of quantum expectation values is not and is actually expected for generic systems in the thermodynamic limit [@srednicki96].
We have demonstrated that, in contrast to the integrable case, the nonequilibrium dynamics of a generic isolated quantum system does lead to standard thermalization. We verified that this happens through the eigenstate thermalization mechanism, a scenario J. M. Deutsch [@deutsch91] demonstrated for the case of an integrable quantum Hamiltonian weakly perturbed by a single matrix taken from a random Gaussian ensemble and M. Srednicki [@srednicki94] compellingly defended for the case of rarefied semiclassical quantum billiards, and which both authors conjectured to be valid in general. Our results, when combined with the others we mentioned [@deutsch91; @horoi95; @shnirelman74; @voros79; @colin85; @zelditch87; @srednicki94; @heller07; @berry77; @korepin93], constitute strong evidence that eigenstate thermalization indeed generally underlies thermal relaxation in isolated quantum systems. Therefore, to understand the existence of universal thermal time-asymptotic states, one should study operator expectation values in individual eigenstates. This is a problem that is linear, time-independent, and conceptually far simpler than any arising in the research—currently dominating the field—on the nonlinear dynamics of semiclassical systems. Among the fundamental open problems of statistical mechanics that could benefit from the linear time-independent perspective are the nature of irreversibility, the existence of a KAM-like threshold [@jose98] in quantum systems, and the role of conserved quantities in the approach to equilibrium. Finally, having a clear conceptual picture for the origins of thermalization may make it possible to engineer new, “unthermalizable” states of matter [@deutsch91], with further applications in quantum information and precision measurement.
We thank A. C. Cassidy, K. Jacobs, A. P. Young, and E. J. Heller for helpful comments. We acknowledge financial support from the National Science Foundation and the Office of Naval Research. We are grateful to the USC HPCC center where our numerical computations were performed.
SUPPLEMENTARY DISCUSSION
========================
1. The Hamiltonian and the numerical calculations.
--------------------------------------------------
In a system of units where $\hbar=1$ the Hamiltonian reads $$\widehat{H} = -J \sum_{\langle i,j \rangle}
\left(\hat{b}^{\dagger}_i \hat{b}^{}_j + \mbox{h.c.} \right)
+ U \sum_{\langle i,j \rangle} \hat{n}_i\hat{n}_j
\label{HCB_Hamiltonian}$$ where $\langle i,j \rangle$ indicates that the sums run over all nearest-neighbor pairs of sites, $J$ is the hopping parameter, and $U$ the nearest-neighbor repulsion parameter that we always set to $0.1J$. The hard-core boson creation ($\hat{b}^{\dagger}_i$) and annihilation ($\hat{b}^{}_j$) operators commute on different sites, $[\hat{b}^{}_i,\hat{b}^{\dagger}_j] =\
[\hat{b}^{}_i,\hat{b}^{}_j] =\
[\hat{b}^{\dagger}_i,\hat{b}^{\dagger}_j] =\ 0$ for all $i$ and $j \ne i$, while the hard-core condition imposes the canonical anticommutation relations on the same site, $\{ \hat{b}^{}_i,\,
\hat{b}^{\dagger}_i \} = 1$, and $(\hat{b}^{}_i)^2=(\hat{b}^{\dagger}_i)^2 = 0 $ for all $i$. Here $\hat{n}_i = \hat{b}^{\dagger}_i \hat{b}^{}_i$ is the density operator.
An exact study of the nonequilibrium dynamics for [*all*]{} time scales requires a full diagonalization of the many-body Hamiltonian (\[HCB\_Hamiltonian\]). We are able to fully diagonalize—essentially to machine precision—matrices of dimension $D\sim 20,000$, and so we consider $N=5$ hard-core bosons on a $5\times 5$ lattice with four sites missing ($D=20,349$); see Fig. \[figSD1\]. All the eigenstates of the Hamiltonian are used for the time evolution $${\vert \psi(t)\rangle}=\exp{[-i\widehat{H}t]}{\vert \psi(0)\rangle}= \sum_\alpha C_{\alpha}\exp{[-i E_\alpha t]}{\vert\Psi_{\alpha}\rangle}\, ,$$ where ${\vert \psi(t)\rangle}$ is the time-evolving state, ${\vert \psi(0)\rangle}$ the initial state, ${\vert\Psi_{\alpha}\rangle}$ the eigenstates of the Hamiltonian with the energies $E_\alpha$, and $C_{\alpha}=\langle
\Psi_\alpha | \psi(0) \rangle$. Our initial state is the ground state of the five bosons when they are confined to the lower part of the lattice (the colored part in Fig. \[figSD1\]. The time evolution begins with the opening of the link shown in Fig. \[figSD1\], which allows the bosons to expand over the whole lattice. The position of the missing sites was chosen so that we only open a single link to start the relaxation dynamics. The motivation for this will become apparent in the last paragraph below.
![ \[figSD1\] **The lattice for the dynamics**. Two-dimensional lattice on which the particles propagate in time. The initial state is the ground state of 5 hard-core bosons confined to the sub-lattice in the lower right-hand corner, and the time evolution starts after the opening of the link indicated by the door symbol. ](figSD1)
As the principal observables of interest we chose the marginal momentum distribution along the horizontal axis $n(k_{x}) =
\sum_{k_{y}} n(k_{x},k_{y})$ and in particular its central component $n(k_{x}=$ $0)$, quantities readily measurable in actual experiments with ultracold quantum gases [@mandel03]. Here the full two-dimensional momentum distribution is $n(k_{x},k_{y})=1/L^2\sum_{i,j} e^{-i 2\pi \textbf{k}(\textbf{r}_i-\textbf{r}_j)/L} \langle
\hat{b}^{\dagger}_i\hat{b}^{}_j\rangle$, where $L=L_{x}=L_{y}=5$ are the linear sizes of the lattice. The position $\textbf{r}_i=( i_{x} \, d\, , i_{y} \, d)$ involves the lattice constant $d$.
2. The microcanonical ensemble in a small system.
-------------------------------------------------
To compute the microcanonical ensemble predictions, we have averaged over all eigenstates whose energies lie within a narrow window $[E_0-\Delta E, E_0+\Delta E]$, with $E_0\equiv\langle \psi(0) | \widehat{H} | \psi(0)
\rangle=\langle \psi(t) | \widehat{H} | \psi(t) \rangle=-5.06 J$. Since our systems are small there is generally no meaning to the limit $\Delta E \to 0$, because small enough windows may fail to contain even a single eigenstate. Instead, one should show that the microcanonical predictions are robust with respect to the choice of the width of the energy window. In Fig. \[figSD2\] we demonstrate this robustness in a neighborhood of $\Delta
E=0.1J$, a value that seems to be an appropriate choice given the data presented in the inset of the same figure. There we show the dependence on $\Delta E$ of the predictions for $n(k_x=0)$ given by the “left-averaged” and the “right-averaged” microcanonical ensembles, by which we mean that the microcanonical windows are chosen as $[E_0-\Delta E,
E_0]$ and $[E_0, E_0+\Delta E]$, respectively. We see that for $\Delta E\lesssim 0.1J$, the two microcanonical predictions are almost independent of the value of $\Delta E$. The main panel in Fig. \[figSD2\] shows that the microcanonical values of $n(k_x)$ for $\Delta E=0.05J$ and for $\Delta E=0.1J$ are indistinguishable.\
![ \[figSD2\] **Microcanonical ensemble**. Microcanonical momentum distribution function for two different values of $\Delta E$. Inset: Microcanonical predictions for $n(k_x=0)$ calculated using the left ($[E_0-\Delta E, E_0]$) and the right ($[E_0,
E_0+\Delta E]$) averages as functions of $\Delta E$. ](figSD2)
3. Eigenstate thermalization and the width of the energy distribution.
----------------------------------------------------------------------
The eigenstate thermalization alone is not sufficient to ensure an agreement between the predictions of the diagonal and thermal ensembles. As discussed in Ref. [@srednicki94], it is also necessary that both distributions be sufficiently narrow. More specifically, one must require for both ensembles $$(\Delta E)^2\, |A''(E)/A(E)| \ll 1,
\label{narrowness_criterion}$$ where $\Delta E$ is the width of the energy distribution in the ensemble, and $A(E)$ is the dependence of the expectation value of the observable $A_{\alpha\alpha}={\langle\Psi_\alpha\vert}\widehat{A}
{\vert\Psi_{\alpha}\rangle}$ on the energy $E_{\alpha}$ of the Hamiltonian-operator eigenstate ${\vert\Psi_{\alpha}\rangle}$. Note that because of eigenstate thermalization, $A(E)$ is a smooth function of energy. For the thermodynamical ensembles the condition (\[narrowness\_criterion\]) is always satisfied in the thermodynamic limit. We now show that it is also satisfied for the diagonal ensemble in the thermodynamic limit.
If one considers an observable $a$ that is the intensive counterpart of $A$, all conclusions obtained for $a$ can be extended to the original observable $A$ via trivial rescaling. For example, for our principal observable of interest, $n(k_{x})$, the corresponding intensive variable is the momentum density $\xi(p_{x})$ normalized as $\int \!dp_{x}\, \xi(p_{x}) = 1$. Notice that in this case $\xi(p_{x}) =
n(k_{x}) L_{x} d/(2\pi N)$.
For $a$, the condition in (\[narrowness\_criterion\]) reduces to $$(\Delta \epsilon)^2 |a''(\epsilon)/a(\epsilon)| \ll 1,
\label{narrowness_criterion_2}$$ where $\epsilon \equiv E/N$. For sufficiently large systems the dependence of $a$ on $\epsilon$ is independent of the system size. Hence, in order to justify the validity of (\[narrowness\_criterion\_2\]) it is sufficient to prove that the width of the distribution of the energy per particle in the diagonal ensemble converges to zero for large linear sizes $L$ of the system: $$\Delta \epsilon \stackrel{L\to\infty}{\longrightarrow} 0 \, .
\label{narrowness_criterion_3}$$
Suppose that initially our system is prepared in an eigenstate $|\Psi_0\rangle$ of a Hamiltonian $\widehat{H}_{0}$ and that at time $t=0$ the Hamiltonian is suddenly changed to $\widehat{H}$: $$\widehat{H}_0 \to \widehat{H} = \widehat{H}_{0} + \widehat{W},$$ where $\widehat{W}$ is the difference between the new and the old Hamiltonians. Within this scenario, the energy width $$\Delta E = \sqrt{\sum_\alpha E^2_\alpha |C_{\alpha}|^{2}
- \left( \sum_\alpha E_\alpha |C_{\alpha}|^{2} \right)^2 }$$ of the diagonal ensemble becomes equal to the variance of the new energy in the state $|\Psi_0\rangle$: $$\Delta E = \Delta H \equiv \sqrt{\langle \Psi_0 | \widehat{H}^2 | \Psi_0 \rangle
-\langle \Psi_0 | \widehat{H} | \Psi_0 \rangle^2
}\, .$$ It is now straightforward to show that the variance of $\widehat{H}$ equals the variance of $\widehat{W}$: $$\Delta H = \Delta W.$$
In order to deduce how $\Delta W$ scales in the thermodynamic limit, we assume that $\widehat{W}$ is a sum of local operators $\hat{w}(j)$ over some region of the lattice $\sigma$ (a single point, a straight line, the whole lattice, etc.): $$\widehat{W} = \sum_{j \in \sigma} \hat{w}(j).$$ Here $\hat{w}(j)$ is a polynomial of creation and annihilation operators localized at the points $j+\Delta j$, where $|\Delta
j|$ is limited from the above by a finite number that does not scale with the system size. The mean square of $\widehat{W}$ can be written as $$\begin{gathered}
\label{Delta_W}
\langle \Psi_0 | \widehat{W}^2 | \Psi_0 \rangle =
\langle \Psi_0 | \widehat{W} | \Psi_0 \rangle^2 \\
+ \sum_{j_1,j_2 \in \sigma}
\left[
\langle \Psi_0 | \hat{w}(j_1)\hat{w}(j_2)
| \Psi_0 \rangle \right. \\
- \left.
\langle \Psi_0 | \hat{w}(j_1)
| \Psi_0 \rangle
\langle \Psi_0 | \hat{w}(j_2)
| \Psi_0 \rangle
\right ].\end{gathered}$$
In the absence of long-range correlations the expression in brackets tends to zero for large distances between $j_1$ and $j_2$. Therefore, the whole second term on the right-hand-side of (\[Delta\_W\]) scales as $L^{d_{\sigma}}$, where $d_{\sigma}$ is the dimensionality of the sublattice $\sigma$ and $L$ is the linear size of the lattice. The variance of $\widehat{W}$ scales the same way: $$(\Delta W)^2 \stackrel{L\to\infty}{\propto} L^{d_{\sigma}}.$$ Retracing our steps, we arrive at the conclusion that the energy width $\Delta \epsilon$ indeed tends to zero in the thermodynamic limit: $$\Delta \epsilon \stackrel{L\to\infty}{\propto} \frac{1}{L^{d_L-d_{\sigma}/2}},$$ where $d_L \ge d_{\sigma}$ is the dimensionality of the whole lattice.
Note that for the two-dimensional lattice considered in this paper the role of $\widehat{W}$ is played by the hopping energy of the “door” link. An analysis similar to the one above shows that increasing the number of “door” links will lead to an increase in the width $\Delta \epsilon$, proportional to the square root of the number of “door” links. This is why in our 2D experiment, we have chosen the position of the missing sites to be the one in Fig. \[figSD1\], so that only a single link is opened during the time evolution.
[10]{} url \#1[`#1`]{}urlprefix\[2\][\#2]{} \[2\]\[\][[\#2](#2)]{}
, & . ** ****, ().
, , , & . ** ****, ().
, , & . ** ****, ().
, & . ** ****, ().
, , & . ** ****, ().
, & . ** ****, ().
& . ** ().
, & . ** ****, ().
, & . ** ****, ().
** (, , ).
** (, , ).
. ** ****, ().
. ** ****, ().
** (, , ).
, & . ** ****, ().
. ** ****, ().
, , , & . ** ****, ().
, , , & . ** ****, ().
*et al.* . ** ****, ().
, & . ** ****, ().
. ** ****, ().
** (, , ).
. ** ****, ().
. ** ****, ().
& . ** ****, ().
. ** ****, ().
, & ** (, , ).
.
. ** ****, ().
& ** (, , ).
[^1]: Published in Nature [**452**]{}, 854-858 (17 April 2008); 10.1038/nature06838. http://www.nature.com/nature/journal/v452/n7189/abs/nature06838.html
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Searching for extreme emission line galaxies allows us to find low-mass metal-poor galaxies that are good analogs of high redshift [Ly$\alpha$]{} emitting galaxies. These low-mass extreme emission line galaxies are also potential Lyman-continuum leakers. Finding them at very low redshifts ($z\lesssim0.05$) allows us to be sensitive to even lower stellar masses and metallicities. We report on a sample of extreme emission line galaxies at $z\lesssim0.05$ (blueberry galaxies). We selected them from SDSS broadband images on the basis of their broad band colors, and studied their properties with MMT spectroscopy. From the whole SDSS DR12 photometric catalog, we found 51 photometric candidates. We spectroscopically confirm 40 as blueberry galaxies. (An additional 7 candidates are contaminants, and 4 remain without spectra.) These blueberries are dwarf starburst galaxies with very small sizes ($< 1\hbox{kpc}$), and very high ionization (\[OIII\]/\[OII\]$\sim10-60$). They also have some of the lowest stellar masses ($\log(\hbox{M}/\hbox{M}_{\odot})\sim6.5-7.5$) and lowest metallicities ($7.1<12+\log(\hbox{O/H})<7.8$) starburst galaxies. Thus they are small counterparts to green peas and high redshift [Ly$\alpha$]{} emitting galaxies.'
author:
- 'Huan Yang, Sangeeta Malhotra, James E. Rhoads, Junxian Wang'
title: 'Blueberry galaxies: the lowest mass young starbursts'
---
Introduction
============
One frontier in observational cosmology is understanding the reionization of the neutral hydrogen in the Universe. How did reionization occur and which sources caused it? Many simulations and observations suggest the very faint dwarf starburst galaxies (stellar mass $\sim 10^{6-8} M_{\odot}$) contribute a significant fraction of ionizing photons ([*Lyman continuum, LyC*]{}) during reionization (e.g. Trenti et al. 2010; Salvaterra et al. 2011; Bouwens et al. 2011; Dressler et al. 2015). Thus it is important to study these faint dwarf starbursts and understand their physical properties and formation mechanism.
High redshift [Ly$\alpha$]{} emitters (LAEs) are an important population of low mass star-forming galaxies at $z > 2$, increasing in fraction to constitute 60% of Lyman break galaxies at redshifts $z > 6$ (Stark et al. 2011). A large fraction of the dwarf starburst galaxies during reionization may be intrinsic LAEs, but their [Ly$\alpha$]{} photons can be scattered by the HI in IGM, which makes [Ly$\alpha$]{} line a powerful probe of reionization (e.g. Malhotra & Rhoads 2004; Ouchi et al. 2010; Hu et al. 2010; Jensen et al 2013; Tilvi et al. 2014; Matthee et al. 2015; Zheng et al. 2017). These high-$z$ LAEs have low metallicity, low stellar masses, low dust extinction, and compact sizes (Gawiser et al. 2007; Pirzkal et al. 2007; Finkelstein et al. 2008; Pentiricci et al. 2009; Malhotra et al. 2012).
Studying LAEs and faint dwarf starbursts in the high-$z$ universe is very challenging, because it usually requires long exposure times and difficult near-infrared observations (e.g. McLinden et al. 2011; Trainor et al. 2015). A complementary approach is to study the physical processes in low-$z$ analogs of these high-$z$ LAEs and faint dwarf starbursts.
The current best nearby analogs of high-$z$ LAEs are green pea galaxies (e.g. Yang et al. 2016, 2017; Jaskot & Oey 2014; Henry et al. 2015; Verhamme et al. 2017). Green pea galaxies were discovered in the citizen science project Galaxy Zoo (Cardamone et al. 2009). They are compact galaxies that are unresolved in SDSS images. The green color is because the \[OIII\] doublet (EW(O\[III\]5007)$\sim$300-2500Å) dominates the flux of SDSS $r$-band, which is mapped to the green channel in the SDSS’s false-color [*gri*]{}-band images. They are similar to high-$z$ LAEs in many galactic properties – small sizes, low stellar masses (10$^{8-10}M_{\odot}$), low metallicities for their stellar masses, high specific star formation rates (sSFR), and large [\[OIII\]$\lambda$5007]{}/[\[OII\]$\lambda$3727]{} ratios (Cardamone et al. 2009; Amorin et al. 2010; Izotov et al. 2011). Five green peas are observed and confirmed as LyC leakers (Izotov et al. 2016).
Green Peas are relatively luminous and massive galaxies compared to the faint-end dwarf starbursts and LAEs. In the present study, we searched for closer and lower mass analogs of the faint-end LAEs in SDSS $ugriz$ broadband images, and studied their properties with MMT and SDSS spectra.
{width="90.00000%"}
{width="\textwidth"}
{width="\textwidth"}
{width="\textwidth"}
Photometric Selection of Blueberry Galaxies in SDSS
===================================================
The strong \[OIII\] emission line makes green peas show distinctive colors. Cardamone et al. (2009) selected green peas at $0.14<z<0.36$ in SDSS. To select the lowest mass green peas in SDSS images, we focus on the lowest redshift green pea sample. At the same limiting magnitudes of SDSS images, these lower redshift green peas generally tend to have lower luminosities and lower stellar mass. In this paper, we select green peas at $z\lesssim0.05$.
We use simple color criteria to select photometric candidates of $z\lesssim0.05$ green peas. To get the color criteria, we made model spectra of green peas and generated corresponding redshift tracks in color-color space. The model spectra include both continuum and strong emission lines, with a range of line strengths. (1) The continuum component includes an age=4Myr young starburst and an age=900Myr old population model spectra from Starburst99 (Leitherer et al. 1999). The old to young mass ratio is fixed at 10. (2) The strong emission lines include \[OII\]3727, [H$\beta$]{}, [H$\alpha$]{}, \[OIII\]4959, and \[OIII\]5007 at three different equivalent widths (EWs). The EW(\[OIII\]5007) are 300, 800, and 1600 Å. At each EW(\[OIII\]5007), the strength of the other lines (\[OII\], [H$\beta$]{}, [H$\alpha$]{}, and \[OIII\]4959) are estimated using typical line ratios of green peas at the corresponding \[OIII\]5007 equivalent width (Yang et al. 2017). Then we multiply the model spectra with the SDSS throughput curves for $ugriz$ filters and calculate the colors at different redshifts. When the [\[OIII\]$\lambda$5007]{} and [H$\alpha$]{} lines move close to the wavelength edges of filters, the colors change dramatically with redshift.
According to the tracks in $g-r ~vs.~ r-i$ color-color diagram (figure 1), green peas at $z<0.08$ and $0.14<z<0.36$ have very different colors from stars, normal galaxies, and quasars. To select a clean sample of green peas at $z\lesssim0.05$, we use the following color selection criteria: $$g-r < -0.5 ~\hbox{and}~ r-i<1.0 ~\hbox{and}~ g-i<-0.5$$ $$~\hbox{and}~ (g-r<-0.7 ~\hbox{or}~ g-i<-1.0) ~\hbox{and}~ g-u<-0.3$$ The $g-u$ selection is added to exclude some very blue young stars or white dwarfs that fall in the selection region. Those stars have very blue $u-g$ colors in comparison to the red $u-g$ colors of green peas with strong [\[OIII\]$\lambda$5007]{} in $g$ band. As shown in figure 1, these color criteria select green peas with EW([\[OIII\]$\lambda$5007]{})$\gtrsim800$Å at $z\lesssim0.05$. The selection was conducted with the SDSS DR12 photometric catalog. To exclude imaging artifacts from the selected catalog, we add some photometric flags to the selection. The exact selection criteria are in Appendix A. Especially, because green peas are compact sources, we use the flag “NOT CHILD" to select small sources that are [*not*]{} deblended as a part of an underlying diffuse galaxy.
Among all sources in the SDSS DR12 photometric catalog, only 51 sources satisfy the selection criteria. Figure 2 shows their SDSS $gri$ images (Blue-Green-Red jpg files generated by SDSS, Robert et al. 2004). Most of them are unresolved or marginally resolved in SDSS images (seeing $\sim1.3\arcsec$), although a few sources have some weak diffuse emission around the central bright knots. Because their blue/purple colors, we call them “blueberry" galaxies.
The Pan-STARRS1 (PS1) survey (Chambers et al. 2017) has deep $z$ and $y$ band images, so we also get the PS1 $grizy$ photometric data for this sample to aid in spectral energy distribution (SED) fitting. The photometric data are in Table 1 and in the attached machine-readable table.
Optical Spectra of Blueberry Galaxies
=====================================
We obtained spectra of this sample from SDSS, MMT, and LBT. Out of the total sample of 51 sources, 10 sources already have SDSS spectra. These 10 sources (the first 10 objects in figure 2, with objID $<15$) all show strong \[OIII\] emission lines and are at the bright end of this sample. One object was observed using LBT/MODS (5 minutes exposure) in the morning of May 07, 2016, after twilight started. The spectra shows strong emission lines. Another 36 objects were observed with MMT/BlueChannel on Jan 07, 2017. The remaining 4 objects don’t have optical spectra. In all spectra, the wavelength range of \[OII\]3727 to [H$\alpha$]{} was covered. Out of the 47 objects with spectra, 40 objects show extreme emission lines. The other 7 objects are stars, quasars, or unknown sources (with low S/N spectra).
The MMT observations were made using the medium-resolution grating (500GPM) and 1.25$\times$ 180 slit, giving a spatial scale along the slit of 0.6 per pixel, a spectral range of 3700-6900 Å, and a spectral resolution of $\sim$3.8 Å (FWHM). The seeing during the MMT observations was $\sim0.7-1.0 \arcsec$. The slit orientations were along the parallactic angles for all objects. The exposure time is about 10 – 15 minutes for each object.
The MMT/BlueChannel and LBT/MODS long-slit spectra were reduced using IRAF following the standard steps. The raw data were bias subtracted and flat-fielded. The sky background at the upper and lower regions of the object were subtracted. The continuum was detected in most objects and was used for determining the spectral trace. The traces of two faint objects were determined using other spectra as references. The 1-D spectra were extracted using optimal extraction. The spectra of HeNeAr lamps were used for wavelength calibration. The standard star LB227 was observed with 5 slit width and was used for the flux calibration. To get better absolute spectral fluxes, we calculated a $g$-band magnitude by multiplying the spectra with $g$ band throughput curve, and then scaled each spectrum to make the spectroscopic $g$ magnitude equal the SDSS $g$ band magnitude.
The optical spectra (figure 3) of blueberry galaxies show very strong \[OIII\]5007 emission lines, and large \[OIII\]5007/[H$\beta$]{} and \[OIII\]/\[OII\] ratios. Figure 4 shows the spectra in 3600 – 4800Å. The \[OIII\]4363 line is very strong and well detected. The \[OII\]3727 line is relatively weak.
From the optical spectra, we measured the properties of emission lines. Firstly, we estimate the continuum by fitting a cubic polynomial function to wavelength regions without emission lines and subtract the continuum from the spectra. Then we fit the emission lines with gaussian functions and get the line fluxes. We correct the measured line fluxes for Milky Way extinction using the attenuation of Schlafly & Finkbeiner (2011) (obtained from the NASA/IPAC Galactic Dust Reddening and Extinction tool) and the Fitzpatrick (1999) extinction law. Notice that the [H$\alpha$]{}/[H$\beta$]{} ratios for most of the MMT spectra are smaller than the case-B value 2.86, because the flat field calibration in the red end of spectra ($6000-6900\AA$) is very poor and the [H$\alpha$]{} fluxes are probably underestimated. Therefore we didn’t correct the line fluxes for extinction by dust within the blueberry galaxies. The properties of emission lines are shown in Table 2 and more columns are included in the attached machine-readable table.
{width="50.00000%"}
![The stellar mass vs. gas metallicity of blueberry galaxies (blue dots; the metallicity errorbar is smaller than the blue dot for many galaxies). The dashed line shows the mass–metallicity relation of SDSS galaxies measured from stacked spectra in Andrews & Martini (2013) and the dotted lines show its 1$\sigma$ scatter ($\sim$0.2dex). The green triangles are galaxies in the Local Volume Legacy survey (Berg et al. 2012; Lee et al. 2006). The metallicities of blueberries, local volume galaxies, and SDSS galaxies are all calculated using the \[OIII\]4363 line.](fig6.pdf){width="50.00000%"}
![The \[OIII\]/\[OII\] vs. EW(\[OIII\]5007) of blueberry galaxies. In two galaxies (black triangles) where the \[OII\] line is detected at 2$\sigma$ level, we use the 3$\sigma$ upper-limit of \[OII\] line to calculate the \[OIII\]/\[OII\] ratio.](fig7.pdf){width="50.00000%"}
What are blueberry galaxies?
============================
From the optical spectra, we measured some physical properties of blueberry galaxies. These properties are shown in figures 5–8.\
[**Stellar mass**]{}: The stellar mass is estimated by fitting Starburst99 models (Leitherer et al. 1999) to the SDSS $ugri$ and PS1 $zy$ bands photometric data. The strong emission lines are subtracted from the SDSS $gri$ band photometry. We model the star formation history with two starburst components – one young component with age$<20Myr$, and one old component with age$<14Gyr$. From the Starburst99 single stellar population model (generated with Kroupa initial mass function and Geneva 2012/2013 tracks with zero rotation at metallicity Z=0.002), we make 12600 composite spectra at a grid of three different parameters – the age of young component, the age of old component, and the mass ratio of old to young components. Then we fit each composite spectrum to the SED and compute the likelihood of each spectrum as $\exp(-\chi^{2}/2)$. From the likelihood distribution, we estimate the median value (50% probability on each side) as the final stellar masses and ages. The majority of the blueberry galaxies have stellar mass about 10$^{6.5}M_\odot$–$10^{7.5}M_{\odot}$.
Note that a faint diffuse stellar disk may exist around the blueberry galaxies. The SDSS r-band images have a surface brightness limit of $\sim$26 $mag~arcsec^{-2}$, which corresponds to $\sim~10^{6}~L_{\odot}~kpc^{-2}$. Assuming a mass to light ratio $M_{r}/L_{r}=1$ for low mass galaxies (Kauffmann et al. 2003), the diffuse stellar disk has a surface limit of $\sim~10^{6}~M_{\odot}~kpc^{-2}$.
[**Star formation rate**]{}: Assuming the Case-B condition, we convert the [H$\beta$]{} luminosity to SFR using the formula $SFR(M_{\odot}/\hbox{yr})=L_{H\beta}(\hbox{erg}/\hbox{s})\times2.86\times10^{-41.27}$ (Kennicutt & Evans 2012). (We estimate star formation using [H$\beta$]{} rather than [H$\alpha$]{} because our spectra are better calibrated at [H$\beta$]{}.) The resulting star formation rates are about $\hbox{SFR} \approx 0.05-2~M\odot~\hbox{yr}^{-1}$, and sSFR are about $10^{-8}~yr^{-1}$. Figure 5 shows the mass$-$SFR relation of blueberry galaxies.
[**Gas metallicity:**]{} We calculate the metallicity using [\[OIII\]$\lambda$4363]{}, [\[OIII\]$\lambda$5007]{}, and [\[OII\]$\lambda$3727]{} line fluxes following the $T_{e}$ method described in Izotov et al. (2006) and Ly et al. (2014). Blueberry galaxies show very low metallicities ($7.1<12+log(O/H)<8.0$). A few can be classified as extremely metal poor galaxies (XMP). And two blueberry galaxies have $12+log(O/H)<7.2$, and are among the most metal-poor galaxies ever found (e.g. Guseva et al. 2015). Thus, searching for blueberry galaxies may be a good method to find XMP with the lowest gas metallicity. Figure 6 shows the mass-metallicity relation of blueberry galaxies.
[**Ionization**]{}: We also calculate the \[OIII\]/\[OII\] ratios ((\[OIII\]4959 + \[OIII5007) / \[OII\]3727) (figure 7). For two galaxies where the \[OII\] line is detected at $\la 2\sigma$ significance, we use the 3$\sigma$ upper-limit of \[OII\] line to calculate the \[OIII\]/\[OII\] ratio. The \[OIII\]/\[OII\] ratios for the full sample range from $\sim 8$ to $\ga 60$.
{width="50.00000%"}
[**Environment**]{}: To characterize the environments of blueberry galaxies, we find the nearest galaxy neighbor in the SDSS DR12 spectroscopic catalog for each one. The distances to the nearest neighbor are between 300kpc $-$ $>$10Mpc (figure 8). We also calculate the nearest neighbors for a comparison sample of 5000 galaxies randomly selected at similar redshift ($0.03<z<0.05$) in the SDSS DR12 spectroscopic catalog. Most galaxies in the comparison sample have neighbors within 2Mpc. Blueberry galaxies and the comparison sample have different distributions of distances to nearest neighbor (with K-S test p-value=$10^{-39}$). So blueberry galaxies are generally in low density environments, and some are in the outskirts of galaxy groups.
[**Number density**]{}: Among all sources in the SDSS DR12 photometric catalog, only 51 sources satisfy the selection criteria (40 blueberries, 7 contaminants, 4 no spectra). Assuming 3 objects without spectra are blueberries, we get 43 objects in a sky area of 14555 deg$^{2}$ at redshift$<0.05$. Thus the number density of blueberry galaxies is 43/ (14355 / 41259 $\times$ 0.04 Gpc$^{3}$) = 3.0$\times$10$^{-6}$ Mpc$^{-3}$.
In summary, these blueberry galaxies are currently undergoing a young starburst. They have very small sizes ($< 1kpc$), very low stellar masses and metallicities, and very high ionization.
Compared to the many different classes of star-forming “dwarf" galaxies, are blueberry galaxies a new class? The star-forming dwarf galaxies include the local volume dwarf galaxies (e.g. Lee et al. 2009), the blue compact dwarf galaxies (BCDs, e.g. Zwicky & Zwicky 1971; Thuan & Martin 1981; Gil de Paz et al. 2003), ultracompact blue dwarf galaxies (e.g. Corbin et al. 2006), HII galaxies (e.g. Terlevich & Melnick 1981; Melnick et al. 2017), extreme metal-poor galaxies (e.g. Guseva et al. 2015; Sanchez Almeida et al. 2016), HI selected dwarfs (e.g. Huang et al. 2012), emission line dots (e.g. Werk et al. 2010; Kellar et al. 2012), and blue diffuse galaxies (James et al. 2017). Compared to those star-forming dwarf galaxies, blueberry galaxies have similar stellar mass and luminosity, but much stronger \[OIII\] line strength and gas ionization. Because blueberry galaxies are selected by the strong \[OIII\] emission lines, they represent the star-forming dwarf galaxies with the highest emission line strength and gas ionization.
On the other hand, compared to green pea galaxies at $z\sim0.2-0.3$ and typical high-$z$ LAEs found in the current narrow-band surveys, blueberry galaxies have similarly strong emission lines but about 10-100 times smaller stellar mass, SFR, and luminosity. So blueberry galaxies represent the faint-end of green peas and LAEs.
Conclusion
==========
We searched for blueberry galaxies in SDSS broadband images and studied their properties with MMT and SDSS spectra. Our main results are as follows.
\(1) Using our color selection criteria, we find 51 photometric candidates at $z\lesssim0.05$ from the whole SDSS DR12 photometric catalog. Optical spectra confirm that 40 of these sources show strong emission lines and can be classified as blueberry galaxies. The remaining 11 candidates consist of 7 contaminants and 4 without spectra. The number density of blueberry galaxy is about 3.0$\times$10$^{-6}$ Mpc$^{-3}$.
\(2) These blueberries are dwarf starburst galaxies with very small sizes ($<1\hbox{kpc}$), low stellar masses ($\log(M/M_{\odot})\sim6.5-7.5$), small SFR ($0.05-2~M_\odot\,\hbox{yr}^{-1}$), high sSFR ($3-100~\hbox{Gyr}^{-1}$), very low gas metallicities ($7.1<12+\log(O/H)<8.0$) estimated with the \[OIII\]4363 lines, and very high gas ionization (\[OIII\]/\[OII\]\]$\sim10-60$). Two blueberry galaxies have $12+\log(O/H)<7.2$ and are among the most metal-poor galaxies known. Their nearest-neighbor distances in the SDSS spectroscopic catalog are between 300kpc$-$10Mpc, larger than the galaxy population in general. Thus, they are in low density environments. These blueberry galaxies represent the faint-end sample of green peas and LAEs.
[lrrlcccccccc]{} 2 & 125.08029 & 54.52780 & 0.03858 & 21.40 & 20.46 & 21.55 & 21.45 & 21.56 & - & 1687 & 3.91\
3 & 126.41854 & 18.77145 & 0.03792 & 19.81 & 19.02 & 19.83 & 19.85 & 20.19 & 20.30 & 1339 & 0.61\
4 & 141.73101 & 45.07562 & 0.04225 & 21.25 & 20.11 & 21.11 & 20.88 & 21.03 & - & 1947 & 4.23\
5 & 158.23637 & 49.32979 & 0.04403 & 19.93 & 18.74 & 20.01 & 19.70 & 20.24 & 20.29 & 2635 & 0.54\
6 & 200.94776 & -1.54778 & 0.02245 & 19.27 & 18.12 & 18.82 & 19.41 & 19.74 & 19.43 & 2055 & 2.26\
7 & 208.85694 & 46.86427 & 0.02811 & 20.38 & 19.31 & 20.06 & 20.14 & 20.82 & 20.56 & 1318 & 1.17\
9 & 221.17238 & 4.16159 & 0.03875 & 20.51 & 19.41 & 20.31 & 20.33 & 20.45 & 20.50 & 1349 & 7.17\
10 & 227.39239 & 37.52948 & 0.03259 & 18.36 & 17.32 & 18.09 & 18.20 & 18.51 & 18.38 & 1718 & 2.22\
12 & 239.10198 & 48.11272 & 0.05024 & 19.81 & 18.75 & 19.73 & 19.38 & 19.86 & 19.78 & 1219 & 2.57\
13 & 242.04318 & 35.46926 & 0.03274 & 19.94 & 18.67 & 19.76 & 19.97 & 20.20 & 20.06 & 2294 & 2.45\
27 & 26.72211 & 3.32288 & 0.04672 & 19.65 & 18.65 & 19.70 & 19.37 & 19.94 & 19.95 & 1226 & 9.24\
28 & 34.09903 & 17.25720 & 0.03921 & 20.95 & 20.18 & 20.97 & 20.82 & 21.38 & 21.13 & 796 & 10.41\
31 & 39.70384 & 1.40929 & 0.04983 & 20.50 & 19.64 & 20.56 & 20.20 & 20.47 & - & 754 & 6.05\
34 & 59.30632 & 18.14599 & 0.03732 & 21.85 & 19.98 & 20.91 & 20.54 & 20.93 & 21.02 & 1916 & 25.34\
38 & 121.11169 & 40.36818 & 0.05768 & - & 20.92 & 22.01 & 21.42 & 21.47 & - & 1074 & 2.71\
39 & 126.94439 & 10.98644 & 0.04355 & 20.97 & 19.72 & 20.53 & 20.26 & 20.98 & 20.54 & 1046 & 0.89\
41 & 129.40511 & 18.39092 & 0.04094 & 20.41 & 19.06 & 20.46 & 20.28 & 20.66 & 20.64 & 2506 & 2.32\
43 & 133.40643 & 23.49403 & 0.04310 & 20.91 & 20.22 & 21.29 & 21.12 & 21.33 & - & 729 & 2.25\
47 & 140.59943 & 63.41027 & 0.03931 & 21.13 & 20.03 & 20.94 & 20.77 & 21.18 & 21.02 & 979 & 1.69\
50 & 150.58734 & 31.49262 & 0.05134 & 20.64 & 19.75 & 20.84 & 20.48 & 20.93 & - & 1202 & 2.43\
54 & 156.65863 & 4.44845 & 0.04215 & 20.73 & 19.93 & 20.80 & 20.49 & 21.06 & 20.42 & 728 & 0.30\
57 & 158.87949 & 14.00532 & 0.03970 & 21.46 & 20.36 & 21.12 & 20.92 & 20.96 & - & 1146 & 4.85\
59 & 161.53849 & 40.78530 & 0.04897 & 20.85 & 20.16 & 21.36 & 20.69 & 21.74 & - & 1154 & 3.36\
66 & 168.30101 & 3.02023 & 0.02336 & 19.91 & 18.83 & 19.41 & 19.97 & 20.22 & 19.86 & 1845 & 1.41\
67 & 170.95395 & 20.84203 & 0.03283 & 18.71 & 17.57 & 18.39 & 18.41 & 18.70 & 18.34 & 2293 & 0.88\
68 & 173.24695 & 8.16188 & 0.04944 & 20.73 & 19.51 & 20.72 & 20.15 & 20.99 & - & 1680 & 1.65\
69 & 174.22574 & 34.45777 & 0.03492 & 20.97 & 20.05 & 20.79 & 20.87 & 21.28 & - & - & 1.82\
70 & 174.75171 & 0.67850 & 0.04165 & 20.25 & 19.45 & 20.25 & 20.08 & 20.37 & 19.86 & 940 & 2.98\
75 & 184.59971 & 39.41910 & 0.05093 & 21.20 & 20.29 & 21.28 & 20.97 & - & - & 1113 & 5.15\
78 & 195.85314 & 39.48141 & 0.04779 & 22.22 & 21.35 & 22.07 & 21.88 & - & - & 507 & 1.37\
82 & 203.36013 & 23.73543 & 0.04721 & 21.22 & 20.19 & 21.36 & 20.98 & 21.83 & - & 1701 & 5.45\
88 & 206.98336 & 7.92560 & 0.04369 & 20.18 & 19.12 & 20.22 & 19.72 & 20.15 & 20.29 & 1544 & 3.64\
90 & 210.18102 & 19.85603 & 0.05325 & 21.28 & 20.22 & 21.20 & 20.75 & 21.40 & - & 1257 & 1.56\
97 & 223.21934 & 9.47723 & 0.05203 & - & 21.23 & 21.95 & 21.91 & 21.94 & - & 534 & 3.00\
103 & 233.97741 & 7.16452 & 0.04183 & - & 21.08 & 22.02 & 21.78 & 22.02 & - & 1019 & 0.88\
104 & 235.51594 & 22.65047 & 0.04633 & 22.01 & 21.10 & 22.08 & 21.63 & 21.79 & - & 629 & 4.34\
106 & 240.55233 & 14.76420 & 0.03634 & 20.40 & 19.19 & 20.23 & 20.32 & 20.23 & 20.51 & 1490 & 0.75\
108 & 243.45135 & 36.37016 & 0.03892 & 20.98 & 19.73 & 20.94 & 21.01 & 21.57 & 21.02 & 1367 & 4.68\
121 & 348.48918 & 29.58686 & 0.04666 & 22.15 & 21.12 & 22.00 & 21.75 & 21.11 & - & 422 & 13.37\
122 & 350.13252 & 12.42720 & 0.04192 & 21.00 & 20.20 & 21.14 & 21.01 & 21.33 & - & 777 & 1.55\
[cccccccccccc]{} 2 & 8.1 & 0.15 & 6.0 & 0.06 & 56 & 172 & 28 & 63 & 6.6 & 0.15 & 7.53\
3 & 51.0 & 0.91 & 13.8 & 0.58 & 199 & 614 & 98 & 310 & 7.2 & 0.50 & 7.79\
4 & 8.5 & 0.10 & 6.6 & 0.13 & 81 & 243 & 35 & 103 & 7.2 & 0.23 & 7.72\
5 & 42.0 & 0.76 & 27.6 & 0.21 & 285 & 867 & 126 & 150 & 8.0 & 0.88 & 7.64\
6 & - & - & 39.6 & 1.97 & 532 & 1605 & 212 & 666 & 7.1 & 0.37 & -\
7 & 27.6 & 1.08 & 16.8 & 0.26 & 143 & 422 & 78 & 225 & 6.8 & 0.22 & 7.45\
9 & 26.6 & 0.23 & 10.2 & 0.03 & 147 & 423 & 57 & 174 & 7.3 & 0.31 & 7.83\
10 & 188.9 & 6.65 & 60.9 & 3.84 & 979 & 2961 & 428 & 1341 & 8.1 & 1.61 & 7.87\
12 & 73.8 & 0.66 & 14.6 & 2.60 & 232 & 710 & 117 & 380 & 7.9 & 1.08 & 7.83\
13 & 27.3 & 0.58 & 25.4 & 1.59 & 326 & 978 & 118 & 371 & 7.5 & 0.45 & 7.82\
27 & 65.4 & 0.10 & 21.2 & 0.08 & 258 & 770 & 146 & 211 & 7.7 & 1.16 & 7.62\
28 & 18.9 & 0.29 & 8.7 & 0.15 & 61 & 183 & 38 & 77 & 6.8 & 0.21 & 7.35\
31 & 31.4 & 0.14 & 7.3 & 0.14 & 87 & 261 & 48 & 58 & 7.5 & 0.43 & 7.65\
34 & 37.6 & 1.17 & 23.5 & 0.33 & 279 & 864 & 107 & 216 & 8.3 & 0.54 & 7.78\
38 & 6.1 & 0.05 & 2.6 & 0.01 & 33 & 101 & 16 & 28 & 6.9 & 0.20 & 7.73\
39 & 19.3 & 0.17 & 8.5 & 0.09 & 94 & 289 & 51 & 108 & 7.2 & 0.35 & 7.61\
41 & 17.4 & 0.09 & 16.9 & 0.13 & 233 & 707 & 81 & 174 & 7.3 & 0.49 & 7.88\
43 & 15.5 & 0.09 & 3.6 & 0.04 & 46 & 138 & 31 & 59 & 6.7 & 0.21 & 7.60\
47 & 22.0 & 0.04 & 5.6 & 0.14 & 65 & 209 & 34 & 68 & 6.9 & 0.19 & 7.69\
50 & 26.5 & 0.05 & 6.7 & 0.05 & 92 & 277 & 46 & 88 & 7.0 & 0.45 & 7.74\
54 & 31.1 & 0.05 & 4.0 & 0.05 & 61 & 190 & 34 & 71 & 7.1 & 0.22 & 7.81\
57 & 13.3 & 0.07 & 3.6 & 0.10 & 54 & 166 & 25 & 51 & 7.6 & 0.14 & 7.82\
59 & 3.9 & 1.78 & 7.4 & 0.08 & 54 & 165 & 50 & 95 & 6.6 & 0.44 & 7.13\
66 & 42.0 & 0.06 & 20.8 & 0.09 & 269 & 819 & 113 & 254 & 6.8 & 0.22 & 7.78\
67 & 277.2 & 0.24 & 49.2 & 0.90 & 954 & 2583 & 386 & 916 & 8.6 & 1.48 & 7.94\
68 & 22.4 & 0.07 & 12.5 & 0.11 & 136 & 414 & 65 & 118 & 7.2 & 0.58 & 7.64\
69 & - & - & 107.2 & 0.95 & 25 & 73 & 11 & 24 & 6.7 & 0.05 & -\
70 & 30.5 & 0.03 & 9.9 & 0.04 & 113 & 339 & 53 & 114 & 8.0 & 0.34 & 7.67\
75 & 12.1 & 0.04 & 4.9 & 0.04 & 56 & 166 & 27 & 48 & 6.8 & 0.26 & 7.65\
78 & 4.7 & 0.04 & 1.2 & 0.02 & 13 & 42 & 7 & 12 & 6.6 & 0.06 & 7.64\
82 & 12.3 & 0.60 & 5.8 & 0.09 & 58 & 206 & 28 & 50 & 6.7 & 0.23 & 7.71\
88 & 51.2 & 0.07 & 9.9 & 0.20 & 184 & 564 & 78 & 168 & 8.5 & 0.54 & 7.98\
90 & 9.3 & 0.04 & 5.2 & 0.02 & 68 & 200 & 22 & 45 & 7.1 & 0.24 & 7.86\
97 & 7.2 & 0.15 & 1.4 & 0.05 & 17 & 51 & 9 & 17 & 6.7 & 0.09 & 7.70\
103 & 9.2 & 0.15 & 1.8 & 0.11 & 26 & 82 & 13 & 28 & 6.6 & 0.09 & 7.81\
104 & 8.9 & 0.11 & 1.5 & 0.10 & 18 & 63 & 14 & 25 & 6.7 & 0.11 & 7.60\
106 & 37.1 & 9.34 & 7.9 & 1.14 & 205 & 611 & 77 & 324 & 8.0 & 0.36 & -\
108 & 4.6 & 2.21 & 10.2 & 0.05 & 93 & 291 & 41 & 76 & 6.6 & 0.23 & 7.57\
121 & 8.8 & 0.21 & 2.4 & 0.50 & 16 & 52 & 19 & 24 & 6.7 & 0.15 & 7.17\
122 & 16.8 & 0.41 & 4.5 & 0.08 & 56 & 167 & 32 & 67 & 6.8 & 0.21 & 7.64\
We thank Chun Ly, Lin Lin, Chenwei Yang, and Tianxing Jiang for help with the data reduction and observations. Observations reported here were obtained at the MMT Observatory, a joint facility of the University of Arizona and the Smithsonian Institution. H.Y. and J.X.W. thank support from NSFC 11233002, 11421303, and CAS Frontier Science Key Research Program (QYZDJ-SSW-SLH006). This work has also been supported in part by NSF grant AST-1518057; and by support for HST program \#14201, which was provided by NASA through a grant from the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS 5-26555.
Selection Criteria in SDSS DR12 Photometric Catalog
===================================================
We use the following criteria to select the photometric candidates in the SDSS DR12 CasJobs tool. To make the selection robust, we used both the cModelMag and psfMag. We also require the Galactic latitude $>20$ or $<-20$ degree. The “calibStatus" flag is the status of photometric calibration.
``` {basicstyle="\ttfamily"}
SELECT objID, ra, dec into blueberry from PhotoObj
WHERE cModelMag_g>0 and cModelMag_g<23 and psfMag_g<23
and cModelMagErr_g<0.2 and cModelMagErr_r<0.2
and cModelMag_u - cModelMag_g >0.3 and psfMag_u - psfMag_g >0.3
and cModelMag_r - cModelMag_g >0.5 and psfMag_r - psfMag_g >0.3
and cModelMag_r - cModelMag_i <1.0 and psfMag_r - psfMag_i <1.0
and (cModelMag_r - cModelMag_g >0.7 or cModelMag_i - cModelMag_g >1.0)
and (psfMag_r - psfMag_g >0.7 or psfMag_i - psfMag_g >1.0)
and cModelMag_i - cModelMag_g >0.5
and petroR90_r<5.0 and petroR90_r>0 and petrorad_g<5.0 -- radius
and (b>20 or b<-20) -- Galactic latitude
and calibStatus_u <=2 -- Photometric observation status
and calibStatus_g <=2 and calibStatus_r <=2 and calibStatus_i <=2
and clean=1 -- "clean" photometry
and (flags & 0x0101000980000010)=0 -- not MOVED, BAD_MOVING_FIT,
-- DEBLENDED_AS_MOVING, CHILD, MAYBE_CR, TOO_FEW_GOOD_DETECTIONS
```
Amor[í]{}n, R. O., P[é]{}rez-Montero, E., & V[í]{}lchez, J. M. 2010, , 715, L128
Andrews, B. H., & Martini, P. 2013, , 765, 140 Berg, D. A., Skillman, E. D., Marble, A. R., et al. 2012, , 754, 98
Bouwens, R. J., Illingworth, G. D., Oesch, P. A., et al. 2015, ApJ, 803, 34 Cardamone, C., Schawinski, K., Sarzi, M., et al. 2009, MNRAS, 399, 1191 Chambers, K. C., Magnier, E. A., Metcalfe, N., et al. 2016, arXiv:1612.05560
Corbin, M. R., Vacca, W. D., Cid Fernandes, R., et al. 2006, , 651, 861 Dressler, A., Hare, T., Bigelow, B. C., & Osip, D. J. 2006, SPIE, 6269, Finkelstein, S. L., Rhoads, J. E., Malhotra, S., Grogin, N., & Wang, J. 2008, ApJ, 678, 655 Fitzpatrick, E. L. 1999, PASP, 111, 63
Gawiser, E., Francke, H., Lai, K., et al. 2007, ApJ, 671, 278 Gil de Paz, A., Madore, B. F., & Pevunova, O. 2003, , 147, 29 Guseva, N. G., Izotov, Y. I., Fricke, K. J., & Henkel, C. 2017, , 599, A65
Henry, A., Scarlata, C., Martin, C. L., & Erb, D. 2015, ApJ, 809, 19 Hu, E. M., Cowie, L. L., Barger, A. J., et al. 2010, ApJ, 725, 394
Huang, S., Haynes, M. P., Giovanelli, R., et al. 2012, , 143, 133
Izotov, Y. I., Stasi[ń]{}ska, G., Meynet, G., Guseva, N. G., & Thuan, T. X. 2006, , 448, 955 Izotov, Y. I., Guseva, N. G., & Thuan, T. 2011, ApJ, 728, 161 Izotov, Y. I., Schaerer, D., Thuan, T. X., et al. 2016, MNRAS, 461, 3683 James, B. L., Koposov, S. E., Stark, D. P., et al. 2017, , 465, 3977
Jaskot, A. E. & Oey, M. S. 2014, ApJ, 791, 19L Jensen, H., Laursen, P., Mellema, G., Iliev, I. T., Sommer-Larsen, J., & Shapiro, P. R. 2013, , 428, 1366 Kellar, J. A., Salzer, J. J., Wegner, G., Gronwall, C., & Williams, A. 2012, , 143, 145 Kauffmann, G., Heckman, T. M., White, S. D. M., et al. 2003, , 341, 33
Kennicutt, R. C., & Evans, N. J. 2012, ARA&A, 50, 531
Lee, H., Skillman, E. D., Cannon, J. M., et al. 2006, , 647, 970
Lee, J. C., Gil de Paz, A., Tremonti, C., et al. 2009, , 706, 599-613 Leitherer, C., Schaerer, D., Goldader, J. D., et al. 1999, , 123, 3
Ly, C., Malkan, M. A., Nagao, T., et al. 2014, ApJ, 780, 122
Malhotra, S., & Rhoads, J. E. 2004, ApJL, 617, L5 Malhotra, S., Rhoads, J. E., Finkelstein, S. L., et al. 2012, ApJL, 750, L36 Matthee, J., Sobral, D., Santos, S., et al. 2015, , 451, 400 McLinden, E. M., Finkelstein, S. L., Rhoads, J. E., et al. 2011, ApJ, 730, 136
Melnick, J., Telles, E., Bordalo, V., et al. 2017, , 599, A76 Ouchi, M., Shimasaku, K., Furusawa, H., et al. 2010, ApJ, 723, 869 Pentericci, L., Grazian, A., Fontana, A., et al. 2009, A&A, 494, 553
Pirzkal, N., Malhotra, S., Rhoads, J. E., & Xu, C. 2007, ApJ, 667, 49 S[á]{}nchez Almeida, J., P[é]{}rez-Montero, E., Morales-Luis, A. B., et al. 2016, , 819, 110 Salim, S., Lee, J. C., Janowiecki, S., et al. 2016, , 227, 2
Salvaterra, R., Ferrara, A., & Dayal, P. 2011, MNRAS, 414, 847 Schlafly, E. F. & Finkbeiner, D. F. 2011, ApJ, 737, 103
Stark, D. P., Ellis, R. S., & Ouchi, M. 2011, ApJL, 728, L2
Terlevich, R., & Melnick, J. 1981, , 195, 839 Tilvi, V., Papovich, C., Finkelstein, S. L., et al. 2014, , 794, 5
Thuan, T. X., & Martin, G. E. 1981, , 247, 823 Trainor, R. F., Steidel, C. C., Strom, A. L., & Rudie, G. C. 2015, ApJ, 809, 89 Trenti, M., Stiavelli, M., Bouwens, R. J., et al. 2010, ApJL, 714, L202 Verhamme, A., Orlitov[á]{}, I., Schaerer, D., et al. 2017, , 597, A13 Werk, J. K., Putman, M. E., Meurer, G. R., et al. 2010, , 139, 279-295
Yang, H., Malhotra, S., Gronke, M., et al. 2016, ApJ, 820, 130
Yang, H., Malhotra, S., Gronke, M., et al. 2017, arXiv:1701.01857 Zheng, Z.-Y., Wang, J., Rhoads, J., et al. 2017, arXiv:1703.02985
Zwicky, F., & Zwicky, M. A. 1971, Guemligen: Zwicky, |c1971,
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
Let $X$ be a compact connected Riemann surface of genus at least two, and let ${\mathcal Q}_X(r,d)$ be the quot scheme that parametrizes all the torsion coherent quotients of ${\mathcal O}^{\oplus r}_X$ of degree $d$. This ${\mathcal Q}_X(r,d)$ is also a moduli space of vortices on $X$. Its geometric properties have been extensively studied. Here we prove that the anticanonical line bundle of ${\mathcal Q}_X(r,d)$ is not nef. Equivalently, ${\mathcal Q}_X(r,d)$ does not admit any Kähler metric whose Ricci curvature is semipositive.
**Résumé**. *Schéma quot et semi-positivité de Ricci*\
Soit $X$ une surface de Riemann compacte et connexe de genre au moins deux, et soit ${\mathcal Q}_X(r,d)$ le schéma quot qui paramétrise tous les quotients torsion cohérents de ${\mathcal O}^{\oplus r}_X$ de degré $d$. L’espace ${\mathcal Q}_X(r,d)$ est aussi un espace de modules de vortex sur $X$. Nous démontrons que le fibré anticanonique de $X$ n’a pas la propriété nef. De façon équivalente, ${\mathcal Q}_X(r,d)$ n’admet aucune métrique kahléienne dont la courbure de Ricci est semi-positive.
address:
- 'School of Mathematics, Tata Institute of Fundamental Research, Homi Bhabha Road, Mumbai 400005, India'
- 'Indian Institute of Science, Department of Mathematics, Bangalore 560003, India'
author:
- Indranil Biswas
- Harish Seshadri
title: Quot schemes and Ricci semipositivity
---
=15.5pt
Introduction
============
Take a compact connected Riemann surface $X$. The genus of $X$, which will be denoted by $g$, is assumed to be at least two. We will not distinguish between the holomorphic vector bundles on $X$ and the torsion-free coherent analytic sheaves on $X$. For a positive integer $r$, let ${\mathcal
O}^{\oplus r}_X$ be the trivial holomorphic vector bundle on $X$ of rank $r$. Fixing a positive integer $d$, let $$\label{e1}
{\mathcal Q}\, :=\, {\mathcal Q}_X(r,d)$$ be the quot scheme that parametrizes all (torsion) coherent quotients of ${\mathcal
O}^{\oplus r}_X$ of rank zero and degree $d$ [@Gr]. Equivalently, ${\mathcal Q}$ parametrizes all coherent subsheaves of ${\mathcal O}^{\oplus r}_X$ of rank $r$ and degree $-d$, because these are precisely the kernels of coherent quotients of ${\mathcal
O}^{\oplus r}_X$ of rank zero and degree $d$. This $\mathcal Q$ is a connected smooth complex projective variety of dimension $rd$. See [@BGL], [@Bi], [@BDW] for properties of $\mathcal Q$. It should be mentioned that $\mathcal Q$ is also a moduli space of vortices on $X$, and it has been extensively studied from this point of view of mathematical physics; see [@Ba], [@BiRo], [@BR] and references therein.
Bökstedt and Romão proved some interesting differential geometric properties of $\mathcal Q$ (see [@BR]). In [@BS0] and [@BS] we proved that $\mathcal Q$ does not admit Kähler metrics with semipositive or seminegative holomorphic bisectional curvature. In this note, we continue the study the question of existence of metrics on $\mathcal Q$ whose curvature has a sign. Our aim here is to prove the following:
\[thm1\] The quot scheme ${\mathcal Q}$ in does not admit any Kähler metric such that the anticanonical line bundle $K^{-1}_{\mathcal Q}$ is hermitian semipositive.
Since semipositive holomorphic bisectional curvature implies semipositive Ricci curvature for a Kähler metric, Theorem \[thm1\] generalizes the main result of [@BS].
Recall that a holomorphic line bundle $L$ on a compact complex manifold $M$ is said to be *hermitian semipositive* if $L$ admits a smooth hermitian structure such that the corresponding hermitian connection has the property that its curvature form is semipositive. The anticanonical line bundle on $M$ will be denoted by $K^{-1}_M$. Note that if $M$ admits a Kähler metric such that the corresponding Ricci curvature is semipositive, then $K^{-1}_M$ is hermitian semipositive. Indeed, in that case the hermitian connection on $K^{-1}_M$ for the hermitian structure induced by such a Kähler metric has semipositive curvature. The converse statement, that hermitian semipositivity of $K_M^{-1}$ implies the existence of Kähler metrics with semipositive Ricci curvature, is also true by Yau’s solution of the Calabi’s conjecture [@aub1], [@aub2], [@yau].
The proof of Theorem \[thm1\] is based on a recent work of Demailly, Campana and Peternell on the classification of compact Kähler manifolds $M$ with semipositive $K_M^{-1}$ [@De], [@CDP]. This classification implies that if $K_M^{-1}$ is semipositive, then there is a nontrivial abelian ideal in the Lie algebra of holomorphic vector fields on $M$, provided $b_1(M)\, >\, 0$. On the other hand, for $M\,=\, {\mathcal Q}$, this Lie algebra is isomorphic to $\mathfrak {sl}(r, {\mathbb C})$, which does not have any nontrivial abelian ideal.
Proof of Theorem \[thm1\]
=========================
Semipositive Ricci curvature
----------------------------
Let $J^d(X)\,=\, \text{Pic}^d(X)$ be the connected component of the Picard group of $X$ that parametrizes the isomorphism classes of holomorphic line bundles on $X$ of degree $d$. Let $S^d(X)$ denote the space of all effective divisors on $X$ of degree $d$, so $S^d(X)\,=\, X^d/P_d$ is the symmetric product with $P_d$ being the group of permutations of $\{1\, ,\cdots\, ,d\}$. Let $$\label{p}
p\, :\, S^d(X)\,\longrightarrow\, \text{Pic}^d(X)$$ be the natural morphism that sends a divisor on $X$ to the holomorphic line bundle on $X$ defined by it.
Take any coherent subsheaf $F\,\subset\, {\mathcal O}^{\oplus r}_X$ of rank $r$ and degree $-d$. Let $$s_F\, :\, {\mathcal O}^{\oplus r}_X\,=\, ({\mathcal O}^{\oplus r}_X)^*\,
\longrightarrow\, F^*$$ be the dual of the inclusion of $F$ in ${\mathcal O}^{\oplus r}_X$. Its exterior product $$\bigwedge\nolimits^r s_F\, :\, {\mathcal O}_X\,=\, \bigwedge\nolimits^r{\mathcal O}^{\oplus r}_X\,
\longrightarrow\, \bigwedge\nolimits^r F^*$$ is a holomorphic section of the holomorphic line bundle $\bigwedge^r F^*$ of degree $d$. Therefore, the divisor $\text{div}(\bigwedge^r s_F)$ is an element of $S^d(X)$. Consequently, we have a morphism $$\label{e2}
\varphi\, :\, {\mathcal Q}\,\longrightarrow\, S^d(X)\, ,\ \ F\, \longmapsto\,
\text{div}(\bigwedge\nolimits^r s_F)\, ,$$ where $\mathcal Q$ is defined in . We note that when $r\,=\,1$, then $\varphi$ is an isomorphism.
Assume that ${\mathcal Q}$ admits a Kähler metric $\omega$ such that $K^{-1}_{\mathcal
Q}$ is hermitian semipositive. Then there is a connected finite étale Galois covering $$\label{f}
f\, :\, \widetilde{\mathcal Q}\, \longrightarrow\, \mathcal Q$$ such that $(\widetilde{\mathcal Q},\, f^*\omega)$ is holomorphically isometric to a product $$\label{e3}
\gamma\, :\, \widetilde{\mathcal Q}\, \longrightarrow\, A\times C\times H\times F\, ,$$ where
- $A$ is an abelian variety,
- $C$ is a simply connected Calabi–Yau manifold (holonomy is $\text{SU}(c)$, where $c\,
=\, \dim C$),
- $H$ is a simply connected hyper-Kähler manifold (holonomy is $\text{Sp}(h/2)$, where $h\, =\, \dim H$), and
- $F$ is a rationally connected smooth projective variety such that $K^{-1}_F$ is hermitian semipositive.
(See [@De Theorem 3.1].) Henceforth, we will identify $\widetilde{\mathcal Q}$ with $A\times
C\times H\times F$ using $\gamma$ in . We note that $F$ is simply connected because it is rationally connected [@Ca p. 545, Theorem 3.5], [@Ko p. 362, Proposition 2.3].
A lower bound of $d$
--------------------
We know that $b_1({\mathcal Q})\,=\, 2g$, and the induced homomorphism $$(p\circ\varphi)_*\, :\, H_1({\mathcal Q},\, {\mathbb Q})\, \longrightarrow\,
H_1(\text{Pic}^d(X),\, {\mathbb Q})\, ,$$ where $p$ and $\varphi$ are constructed in and respectively, is an isomorphism [@Bi], [@BGL p. 649, Remark]. Since $f$ in is a finite étale covering, the induced homomorphism $$f_*\, :\, H_1(\widetilde{\mathcal Q},\, {\mathbb Q})\, \longrightarrow\,
H_1({\mathcal Q},\, {\mathbb Q})$$ is surjective. Therefore, the homomorphism $$\label{e5}
(p\circ\varphi\circ f)_*\, :\, H_1(\widetilde{\mathcal Q},\, {\mathbb Q})\, \longrightarrow\,
H_1(\text{Pic}^d(X),\, {\mathbb Q})$$ is surjective.
There is no nonconstant holomorphic map from a compact simply connected Kähler manifold to an abelian variety. In particular, there are no nonconstant holomorphic maps from $C$, $H$ and $F$ in to $\text{Pic}^d(X)$. Hence the map $p\circ\varphi\circ f$ factors through a map $$\beta\, :\, A\, \longrightarrow\, \text{Pic}^d(X)\, .$$ In other words, there is a commutative diagram $$\label{cd}
\begin{matrix}
\widetilde{\mathcal Q} = A\times C\times H\times F &\
\stackrel{p\circ\varphi\circ f}{\longrightarrow} & \text{Pic}^d(X)\\
q\Big\downarrow~{\ } && ~\,~\Vert{\rm Id}\\
A &\ \stackrel{\beta}{\longrightarrow} & \text{Pic}^d(X)
\end{matrix}$$ where $q$ is the projection of $A\times C\times H\times F$ to the first factor. Since $H_1(A\times C\times H\times F,\, {\mathbb Z})\,=\, H_1(A,\, {\mathbb Z})$ (as $C$, $H$ and $F$ are simply connected), and $(p\circ\varphi\circ f)_*$ in is surjective, it follows that the homomorphism $$\beta_*\, :\, H_1(A,\, {\mathbb Q}) \, \longrightarrow\,H_1(\text{Pic}^d(X),\, {\mathbb Q})$$ induced by $\beta$ is surjective. This immediately implies that the map $\beta$ is surjective. Since $\beta$ is surjective, from the commutativity of we know that the map $p$ is surjective. This implies that $$\label{ge}
d\, =\, \dim S^d(X) \, \geq\, \dim \text{Pic}^d(X)\,=\, g\, \geq\, 2\, .$$
Albanese for $\widetilde{\mathcal Q}$
-------------------------------------
The homomorphism of fundamental groups $$\varphi_*\, :\, \pi_1({\mathcal Q})\, \longrightarrow\, \pi_1(S^d(X))$$ induced by $\varphi$ in is an isomorphism [@BDHW Proposition 4.1]. Since $d\, \geq\, 2$ (see ), the homomorphism of fundamental groups $$p_*\, :\, \pi_1(S^d(X))\, \longrightarrow\,\pi_1(\text{Pic}^d(X))$$ induced by $p$ in is an isomorphism. Indeed, $\pi_1(S^d(X))$ is the abelianization $$\pi_1(X)/[\pi_1(X),\, \pi_1(X)]\,=\, H_1(X, \, {\mathbb Z})$$ of $\pi_1(X)$ [@DT]. Combining these we conclude that the homomorphism of fundamental groups $$\label{pi1}
(p\circ\varphi)_*\, :\, \pi_1({\mathcal Q})\, \longrightarrow\, \pi_1(\text{Pic}^d(X))$$ induced by $p\circ\varphi$ is an isomorphism.
Since the homomorphism in is an isomorphism, the covering $f$ in is induced by a covering of $\text{Pic}^d(X)$. In other words, there is a finite étale Galois covering $$\label{mu}
\mu\, :\, J\, \longrightarrow\, \text{Pic}^d(X)$$ and a morphism $\lambda\, :\, \widetilde{\mathcal Q}\,\longrightarrow\, J$ such that the following diagram is commutative: $$\label{cc}
\begin{matrix}
\widetilde{\mathcal Q} &\ \stackrel{f}{\longrightarrow} & {\mathcal Q}\\
{\ }~\Big\downarrow\lambda && {\ \ \ \ \ \ }~~~\Big\downarrow p\circ\varphi\\
J &\ \stackrel{\mu}{\longrightarrow} & \text{Pic}^d(X)
\end{matrix}$$ where $f$ is the covering map in . The projection $q$ in is clearly the Albanese morphism for $\widetilde{\mathcal Q}$, because $C$, $H$ and $F$ are all simply connected. On the other hand, $p\circ\varphi$ is the Albanese morphism for $\mathcal Q$ [@BS Corollary 2.2]. Therefore, its pullback, namely, $\lambda$, is the Albanese morphism for $\widetilde{\mathcal Q}$. Consequently, we have $A\,=\, J$ with $\lambda$ coinciding with the projection $q$ in . Henceforth, we will identify $A$ and $q$ with $J$ and $\lambda$ respectively.
Vector fields
-------------
The differential $df$ of $f$ identifies $T{\widetilde{\mathcal Q}}$ with $f^*T{\mathcal Q}$, because $f$ is étale. Using the trace homomorphism $t\, :\, f_*{\mathcal O}_{\widetilde{\mathcal Q}}\, \longrightarrow\,
{\mathcal O}_{\mathcal Q}$, we have $$f_*T{\widetilde{\mathcal Q}}\,=\, f_*f^*T{{\mathcal Q}}\, \stackrel{p_f}{\longrightarrow}\,
(f_*{\mathcal O}_{\widetilde{\mathcal Q}})\otimes T{{\mathcal Q}}\,
\stackrel{t}{\longrightarrow}\,{\mathcal O}_{\mathcal Q}\otimes T{{\mathcal Q}}\,=\,
T{{\mathcal Q}}\, ,$$ where $p_f$ is given by the projection formula. This produces a homomorphism $$\label{Phi}
\Phi\, :\, H^0(\widetilde{\mathcal Q},\, T{\widetilde{\mathcal Q}})\,=\, H^0({\mathcal Q},\,
f_*T{\widetilde{\mathcal Q}})\,\longrightarrow\, H^0({\mathcal Q},\, T{{\mathcal Q}})$$ (the equality $H^0(\widetilde{\mathcal Q},\, T{\widetilde{\mathcal Q}})\,=\, H^0({\mathcal Q},\,
f_*T{\widetilde{\mathcal Q}})$ follows from the fact that $f$ is a finite morphism). This homomorphism $\Phi$ is surjective. Indeed, as $f^*T{\mathcal Q}\,=\,T{\widetilde{\mathcal
Q}}$, any section of $T{{\mathcal Q}}$ pulls back to a section of $T{\widetilde{\mathcal Q}}$.
Since $\widetilde{\mathcal Q}\, =\, A\times C\times H\times F$, we have $$\label{Phi2}
H^0(\widetilde{\mathcal Q},\, T{\widetilde{\mathcal Q}})\,=\,
H^0(A,\, TA)\oplus H^0(C,\, TC)\oplus H^0(H,\, TH)
\oplus H^0(F,\, TF)\, .$$ Note that $H^0(\widetilde{\mathcal Q},\, T{\widetilde{\mathcal Q}})$ is a Lie algebra under the operation of Lie bracket of vector fields, and the subspace $$H^0(A,\, TA)\, \subset\, H^0(\widetilde{\mathcal Q},\, T{\widetilde{\mathcal Q}})$$ (see ) is an ideal in this Lie algebra. Since $A\,=\, J$ is a covering of $\text{Pic}^d(X)$, we have $$\label{dg}
\dim H^0(A,\, TA)\,=\, \dim \text{Pic}^d(X)\,=\, g\, >\, 1\, .$$
Since $H^0(A,\, TA)$ is an ideal in $H^0(\widetilde{\mathcal Q},\, T{\widetilde{\mathcal
Q}})$, it follows immediate that $$\Phi(H^0(A,\, TA))\, \subset\, \Phi(H^0(\widetilde{\mathcal Q},\, T{\widetilde{\mathcal Q}}))\,=\,
H^0({\mathcal Q},\, T{\mathcal Q})$$ is an ideal, where $\Phi$ is constructed in . Note that $H^0(A,\, TA)$ is an abelian Lie algebra, so the Lie algebra $\Phi(H^0(A,\, TA))$ is also abelian.
Since $\mu\, :\, J\,=\, A\, \longrightarrow\, \text{Pic}^d(X)$ in is a covering map between abelian varieties, the trace map $H^0(A,\, TA) \, \longrightarrow\,
H^0(\text{Pic}^d(X),\, T\text{Pic}^d(X))$ is an isomorphism. In view of this, from the commutativity of the diagram in it follows that the restriction $$\Phi\vert_{H^0(A,\, TA)}\, :\, H^0(A,\, TA)\, \longrightarrow\, H^0({\mathcal Q},\, T{\mathcal Q})$$ is injective (see and ). But $H^0({\mathcal Q},\, T{\mathcal Q})\,=\,
\mathfrak{sl}(r, {\mathbb C})$ [@BDH p. 1446, Theorem 1.1]. Hence the Lie algebra $H^0({\mathcal Q},\, T{\mathcal Q})$ does not contain any nonzero abelian ideal. This is in contradiction with the earlier result that $\Phi(H^0(A,\, TA))$ is a nonzero abelian ideal in $H^0({\mathcal Q},\, T{\mathcal Q})$ of dimension $g$ (see ). This completes the proof of Theorem \[thm1\].
[ZZZZ]{}
T. Aubin, Équations du type Monge-Ampère sur les variétés kählériennes compactes, [*C. R. Acad. Sci. (Paris) – Math.*]{} [**283**]{} (1976), 119–121.
T. Aubin, Équations du type Monge-Ampère sur les variétés kählériennes compactes, [*Bull. Sci. Math.*]{} [**102**]{} (1978), 63–95.
J. M. Baptista, On the $L^2$–metric of vortex moduli spaces, [*Nuclear Phys. B*]{} **844** (2011), 308–333.
A. Bertram, G. Daskalopoulos and R. Wentworth, Gromov invariants for holomorphic maps from Riemann surfaces to Grassmannians, [*Jour. Amer. Math. Soc.*]{} **9** (1996), 529–571.
E. Bifet, Sur les points fixes schéma ${\rm Quot}_{{\mathcal O}_X/X,k}$ sous l’action du tore ${\mathbf G}^r_{m,k}$, *Com. Ren. Math. Acad. Sci. Paris* **309** (1989), 609–612.
E. Bifet, F. Ghione, and M. Letizia, On the Abel-Jacobi map for divisors of higher rank on a curve, [*Math. Ann.*]{} **299** (1994), 641–672.
I. Biswas, A. Dhillon and J. Hurtubise, Automorphisms of the Quot schemes associated to compact Riemann surfaces, *Int. Math. Res. Not.* [**2015**]{} (2015), 1445–1460.
I. Biswas, A. Dhillon, J. Hurtubise and R. A. Wentworth, A generalized Quot scheme and meromorphic vortices, [*Adv. Theor. Math. Phys.*]{} [**19**]{} (2015), 905–921.
I. Biswas and N. M. Romão, Moduli of vortices and Grassmann manifolds, [*Comm. Math. Phys.*]{} [**320**]{} (2013), 1–20.
I. Biswas and H. Seshadri, On the Kähler structures over Quot schemes, [*Illinois Jour. Math.*]{} [**57**]{} (2013), 1019–1024.
I. Biswas and H. Seshadri, On the Kähler structures over Quot schemes II, [*Illinois Jour. Math.*]{} [**58**]{} (2014), 689–695.
M. Bökstedt and N. M. Romão, On the curvature of vortex moduli spaces, *Math. Zeit.* [**277**]{} (2014), 549–573.
F. Campana, On twistor spaces of the class $\mathcal C$, [*Jour. Differential Geom.*]{} [**33**]{} (1991), 541–549.
F. Campana, J.-P. Demailly and T. Peternell, Rationally connected manifolds and semipositivity of the Ricci curvature, [*Recent advances in algebraic geometry*]{}, A volume in honor of Rob Lazarsfeld’s 60th birthday, Papers from the conference held at the University of Michigan, Ann Arbor, MI, May 16–19, 2013, Edited by Christopher D. Hacon, Mircea Mustaţă and Mihnea Popa, 71–91, London Math. Soc. Lecture Note Ser., 417, Cambridge Univ. Press, Cambridge, 2015.
J.-P. Demailly, Structure theorems for compact Kähler manifolds with nef anticanonical bundles, [*Complex analysis and geometry*]{}, 119–133, Springer Proc. Math. Stat., 144, Springer, Tokyo, 2015.
A. Dold and R. Thom, Quasifaserungen und unendliche symmetrische Produkte, [*Ann. of Math.*]{} [**67**]{} (1958), 239–281.
A. Grothendieck, Techniques de construction et théorèmes d’existence en géométrie algébrique. IV. Les schémas de Hilbert, Séminaire Bourbaki, Vol. 6, Exp. No. 221, p. 249–276, Société Mathématique de France, Paris, 1995.
J. Kollár, Fundamental groups of rationally connected varieties, [*Michigan Math. Jour.*]{} [**48**]{} (2000), 359–368.
I. G. Macdonald, Symmetric products of an algebraic curve, *Topology* **1** (1962), 319–343.
S.-T. Yau, On the Ricci curvature of a complex Kähler manifold and the complex Monge-Ampère equation I, [*Comm. Pure and Appl. Math.*]{} [ **31**]{} (1978), 339–411.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
Let $p>3$ be a prime and let $a$ be a positive integer. We show that if $p\equiv1\pmod 4$ or $a>1$ then $$\sum_{k=0}^{\lfloor\frac34p^a\rfloor}\frac{\binom{2k}k^2}{16^k}\equiv\l(\frac{-1}{p^a}\r)\pmod{p^3}$$ with $(-)$ the Jacobi symbol, which confirms a conjecture of Z.-W. Sun. We also establish the following new congruences: $$\begin{aligned}
\sum_{k=0}^{(p-1)/2}\frac{\binom{2k}k\binom{3k}k}{27^k}\equiv&\l(\frac p3\r)\frac{2^p+1}3\pmod{p^2},
\\\sum_{k=0}^{(p-1)/2}\frac{\binom{6k}{3k}\binom{3k}k}{(2k+1)432^k}\equiv&\l(\frac p3\r)\frac{3^p+1}4\pmod{p^2},
\\\sum_{k=0}^{(p-1)/2}\frac{\binom{4k}{2k}\binom{2k}k}{(2k+1)64^k}\equiv&\l(\frac{-1}p\r)2^{p-1}\pmod{p^2}.\end{aligned}$$ Note that in 2003 Rodriguez-Villeguez posed conjectures on $$\sum_{k=0}^{p-1}\frac{\binom{2k}k^2}{16^k},\ \sum_{k=0}^{p-1}\frac{\binom{2k}k\binom{3k}k}{27^k},\ \sum_{k=1}^{p-1}\frac{\binom{4k}{2k}\binom{2k}k}{64^k},\ \sum_{k=1}^{p-1}\frac{\binom{6k}{3k}\binom{3k}k}{432^k}$$ modulo $p^2$ which were later proved.
address:
- '(Guo-Shuai Mao) Department of Mathematics, Nanjing University, Nanjing 210093, People’s Republic of China'
- '(Zhi-Wei Sun) Department of Mathematics, Nanjing University, Nanjing 210093, People’s Republic of China'
author:
- 'Guo-Shuai Mao and Zhi-Wei Sun'
title: New congruences involving products of two binomial coefficients
---
Introduction
============
Let $p>3$ be a prime. In 2003, via his analysis of the p-adic analogues of Gaussian hypergeometric series and the Calabi- Yau manifolds, Rodriguez-Villegas [@RV] conjectured the following congruences: $$\begin{gathered}
\sum_{k=0}^{p-1}\f{\bi{2k}k^2}{16^k}\eq\l(\f{-1}p\r)\pmod{p^2},\ \sum_{k=0}^{p-1}\frac{\binom{2k}k\binom{3k}k}{27^k}\equiv\l(\frac p3\r)\pmod{p^2},
\\\sum_{k=0}^{p-1}\frac{\binom{4k}{2k}\binom{2k}k}{64^k}\equiv\l(\frac{-2}p\r)\pmod{p^2},
\ \sum_{k=0}^{p-1}\frac{\binom{6k}{3k}\binom{3k}k}{432^k}\equiv\l(\frac {-1}p\r)\pmod{p^2},\end{gathered}$$ where $(\f{\cdot}p)$ denotes the Jacobi symbol. They were soon proved by E. Mortenson [@M1; @M2] via the Gross-Koblitz formula and the $p$-adic $\Gamma$-function. Note that $$\begin{gathered}
\bi{-1/2}k^2=\f{\bi{2k}k^2}{16^k},\ \ \bi{-1/3}k\bi{-2/3}k=\f{\bi{2k}k\bi{3k}k}{27^k},
\\\bi{-1/4}k\bi{-3/4}k=\f{\bi{4k}{2k}\bi{2k}k}{64^k},\ \ \bi{-1/6}k\bi{-5/6}k=\f{\bi{6k}{3k}\bi{3k}k}{432^k}\end{gathered}$$ for all $k\in\N=\{0,1,2,\ldots\}$. In 2011 Z. W. Sun \[Su11\] showed further that $$\sum_{k=1}^{p-1}\f{\bi{2k}k^2}{16^k}\eq\l(\f{-1}p\r)-p^2E_{p-3}\pmod{p^3}$$ and $$\label{1.1}\sum_{k=1}^{(p-1)/2}\f{\bi{2k}k^2}{16^k}\eq\l(\f{-1}p\r)+p^2E_{p-3}\pmod{p^3},$$ where $E_0,E_1,E_2,\ldots$ are the Euler numbers given by $$E_0=1,\ \t{and}\ E_n=-\sum_{k=1}^{\lfloor n/2\rfloor}\binom{n}{2k}E_{n-2k}\ (n=1,2,3,\ldots).$$ He also conjectured that $$\begin{aligned}
\sum_{k=0}^{p-1}\f{\bi{2k}k\bi{3k}k}{(2k+1)27^k}\eq&\l(\f p3\r)\pmod{p^2},
\\\sum_{k=0}^{p-1}\f{\bi{4k}{2k}\bi{2k}k}{(2k+1)64^k}\eq&\l(\f{-1}p\r)-3p^2E_{p-3}\pmod{p^3},
\\\sum_{k=0}^{p-1}\f{\bi{6k}{3k}\bi{3k}k}{(2k+1)432^k}\eq&\l(\f p3\r)\pmod{p^2},\end{aligned}$$ which were confirmed by Z.-H. Sun [@S16]. Note that Z.-W. Sun [@Su14] determined $$\sum_{k=0}^{(p-3)/2}\f{\bi{2k}k^2}{(2k+1)16^k}\ \ \t{and}\ \ \sum_{k=(p+1)/2}^{p-1}\f{\bi{2k}k^2}{(2k+1)16^k}$$ modulo $p^3$.
In this paper we first establish the following result.
\[Th1.1\] Let $p$ be any odd prime.
[(i)]{} We have $$\label{1.2}\sum_{k=0}^{\lfloor 3p/4\rfloor}\frac{\binom{2k}k^2}{16^k}\equiv\begin{cases}1\pmod{p^3}&\t{if}\ p\eq1\pmod 4,
\\-1+p^2/(2\bi{(p-3)/2}{(p-3)/4}^2)\pmod{p^3}&\t{if}\ p\eq3\pmod 4.\end{cases}$$
[(ii)]{} For each $a=2,3,4,\ldots$, we have $$\label{1.3}\sum_{k=0}^{\lfloor\frac34p^a\rfloor}\frac{\binom{2k}k^2}{16^k}\equiv\l(\frac{-1}{p^a}\r)\pmod{p^3}.$$
\[Rem1.1\]. Part (i) in the case $p\eq1\pmod 4$ and part (ii) were conjectured by Sun [@Su11].
Our second theorem is as follows.
\[Th1.2\] Let $p>3$ be a prime. Then we have $$\label{1.4}
\sum_{k=0}^{(p-1)/2}\frac{\binom{2k}k\binom{3k}k}{27^k}\equiv\l(\frac p3\r)\frac{2^p+1}3\pmod{p^2},$$ $$\label{1.5}
\sum_{k=0}^{(p-1)/2}\frac{\binom{6k}{3k}\binom{3k}k}{(2k+1)432^k}\equiv\l(\frac p3\r)\frac{3^p+1}4\pmod{p^2},$$ $$\label{1.6}
\sum_{k=0}^{(p-1)/2}\frac{\binom{4k}{2k}\binom{2k}k}{(2k+1)64^k}\equiv\l(\frac{-1}p\r)2^{p-1}\pmod{p^2}.$$
\[Rem1.2\] We are also able to show the congruence $$\sum_{k=0}^{(p-1)/2}\frac{\binom{2k}k\binom{3k}k}{(2k+1)27^k}\equiv\l(\frac p3\r)(3^p+2-2^{p+1})\pmod {p^2}$$ for any prime $p>3$.
Proof of Theorem 1.1
====================
\[Lem2.1\] [(Sun [@Su11 (1.4)])]{} For any prime $p>3$ we have $$\label{2.1}\sum_{k=1}^{(p-1)/2}\f{4^k}{k^2\bi{2k}k}\eq (-1)^{(p-1)/2}4E_{p-3}\pmod p.$$
[*Proof of Theorem*]{} 1.1(i). In view of (\[1.1\]), (\[1.2\]) has the following equation form: $$\label{2.2}\sum_{k=(p+1)/2}^{\lfloor 3p/4\rfloor}\f{\bi{2k}k^2}{16^k}\eq-p^2E_{p-3}+\f{1-(-1)^{(p-1)/2}}2\cdot\f{p^2}{2\bi{(p-3)/2}{\lfloor p/4\rfloor}^2}\pmod{p^3}.$$ By [@Su11 Lemma 2.1], $$k\binom{2k}k\binom{2(p-k)}{p-k}\equiv(-1)^{\lfloor2k/p\rfloor-1}2p\pmod{p^2}\ \t{for all}\ k=1,\ldots,p-1.$$ Thus $$\begin{aligned}
\sum_{k=(p+1)/2}^{\lfloor 3p/4\rfloor}\frac{\binom{2k}k^2}{16^k}
\equiv&\sum_{k=(p+1)/2}^{\lfloor 3p/4\rfloor}\frac{4p^2}{k^2\binom{2(p-k)}{p-k}^216^k}
=\sum_{j=\lfloor p/4\rfloor+1}^{(p-1)/2}\frac{4p^2}{(p-j)^2\binom{2j}{j}^216^{p-j}}
\\\eq&\frac{p^2}4\sum_{j=\lfloor p/4\rfloor+1}^{(p-1)/2}\frac{16^j}{j^2\binom{2j}j^2}\pmod{p^3}\end{aligned}$$ hence we have reduced (\[2.2\]) to the following simpler form $$\label{2.3} \sum_{k=\lfloor n/2\rfloor+1}^n\f{16^k}{k^2\bi{2k}k^2}\eq-4E_{p-3}+\f{1-(-1)^{n}}{\bi{n-1}{\lfloor n/2\rfloor}^2}\pmod p,$$ where $n=(p-1)/2$.
For each $k=0,\ldots,n$, clearly $$\bi nk\eq\bi{-1/2}k=\f{\bi{2k}k}{(-4)^k}\pmod{p}.$$ Thus $$\sum_{k=\lfloor n/2\rfloor+1}^{n}\frac{16^k}{k^2\binom{2k}k^2}\equiv\sum_{k=\lfloor n/2\rfloor+1}^{n}\frac1{k^2\binom nk^2}\equiv4\sum_{k=\lfloor n/2\rfloor+1}^{n}\frac1{\binom{n-1}{k-1}^2}\pmod p.$$ Note that $$\sum_{k=\lfloor n/2\rfloor+1}^{n}\frac1{\binom{n-1}{k-1}^2}=\sum_{k=\lfloor n/2\rfloor}^{n-1}\frac1{\binom{n-1}{k}^2}=\frac12\sum_{k=0}^{n-1}\frac1{\binom{n-1}{k}^2}+\f{1-(-1)^n}{4\bi{n-1}{\lfloor n/2\rfloor}^2}$$ and $$\label{2.4}
\sum_{k=0}^{n-1}\f1{\bi{n-1}k^2}=\f{2n^2}{n+1}\sum_{k=1}^{n}\f1{k\bi{2n+1-k}{n-k}}$$ (cf. [@SWZ]). So we have $$\begin{aligned}
&\sum_{k=\lfloor n/2\rfloor+1}^{n}\frac{16^k}{k^2\binom{2k}k^2}-\f{1-(-1)^n}{\bi{n-1}{\lfloor n/2\rfloor}^2}
\\\equiv&\frac{4n^2}{n+1}\sum_{k=1}^{n}\frac1{k\binom{2n+1-k}{n-k}}\equiv2\sum_{k=1}^{n}\frac{1}{k\binom{-k}{n-k}}\pmod p.\end{aligned}$$ Observe that $$\begin{aligned}
\sum_{k=1}^n\f1{k\bi{-k}{n-k}}=&\sum_{k=1}^n\f{(-1)^{n-k}}{k\bi{n-1}{k-1}}=n\sum_{k=1}^n\f{(-1)^{n-k}}{k^2\bi{n}k}
\\\eq&\f{(-1)^{n-1}}2\sum_{k=1}^n\f{4^k}{k^2\bi{2k}k}\pmod p.\end{aligned}$$ Therefore, with the help of Lemma \[Lem2.1\], we finally obtain $$\sum_{k=\lfloor n/2\rfloor+1}^n\f{16^k}{k^2\bi{2k}k^2}-\f{1-(-1)^n}{\bi{n-1}{\lfloor n/2\rfloor}^2}
\eq(-1)^{n-1}\sum_{k=1}^n\f{4^k}{k^2\bi{2k}k}\eq-4E_{p-3}\pmod p.$$ This proves (\[2.3\]) and hence (\[1.2\]) follows.
Now we give a lemma which is a natural extension of (\[1.1\]).
\[Lem2.3\] Let $p>3$ be a prime and let $a$ be any positive integer. Then $$\label{2.5}\sum_{k=0}^{(p^a-1)/2}\frac{\binom{2k}k^2}{16^k}\equiv\l(\frac{-1}{p^a}\r)+\l(\f{-1}{p^{a-1}}\r)p^2E_{p-3}\pmod{p^3}.$$
. Theorem 1.2 of Sun [@Su13] states that for any $d=0,\ldots,(p-1)/2$ we have $$\sum_{k=0}^{(p-1)/2}\f{\bi{2k}k\bi{2k}{k+d}}{16^k}\eq\l(\f{-1}p\r)+\f{(-1)^d}4p^2E_{p-3}\l(d+\f12\r)\pmod{p^3},$$ where $E_{p-3}(x)$ denotes the Euler polynomial of the degree $p-3$.
In the case $d=0$ this yields (1.1). Modifying this proof of (\[1.1\]) slightly we immediately get (\[2.5\]).
In 1852, Kummer proved that for any $m,n\in\N$ the $p$-adic valuation of the binomial coefficient $\binom{m+n}m$ is equal to the number of [*carry-overs*]{} when performing the addition of $m$ and $n$ written in base $p$.
\[Lem2.4\] Let $p$ be an odd prime and let $a\in\mathbb{Z}^{+}$. For any $k=1,2,\ldots,(p^a-1)/2$, we have $$\ord_p\binom{p^a-k}{\frac{p^a-1}2-k}\leq a-1.$$
[*Proof*]{}. It is well known that $$\ord_p(n!)=\sum_{j=1}^\infty\l\lfloor\f{n}{p^j}\r\rfloor.$$ Thus $$\ord_p\binom{p^a-k}{\frac{p^a-1}2-k}=\sum_{j=1}^{a-1}\l(\l\lfloor\f{p^a-k}{p^j}\r\rfloor-\l\lfloor\f{(p^a+1)/2}{p^j}\r\rfloor-\l\lfloor\f{(p^a-1)/2-k}{p^j}\r\rfloor\r)$$ does not exceed $a-1$ as each term in the sum is at most one. This concludes the proof.
[*Proof of Theorem*]{} 1.1(ii). In view of Lemma \[Lem2.3\], we just need to verify that $$\label{2.6}
\sum_{k=(p^a+1)/2}^{\lfloor3p^a/4\rfloor}\frac{\binom{2k}k^2}{16^k}\equiv\l(\f{-1}{p^{a-1}}\r)p^2E_{p-3}\pmod{p^3}.$$ Let $k$ and $l$ be positive integers with $k+l=p^a$ and $0<l<p^a/2$. Then $$\frac{\binom{2k}k^2}{\binom{2p^a-2}{p^a-1}^2}=\frac{(2p^a-2l)!^2}{(2p^a-2)!^2}\l(\frac{(p^a-1)!}{(p^a-l)!}\r)^4=\frac{\prod_{0<i<l}(p^a-i)^4}{\prod_{1<j<2l}(2p^a-j)^2}$$ and hence$$\frac{\binom{2k}k^2}{\binom{2p^a-2}{p^a-1}^2}\cdot\frac{(2l-1)!^2}{(l-1)!^4}=\frac{\prod_{0<i<l}(1-p^a/i)^4}{\prod_{1<j<2l}(1-2p^a/j)^2}\equiv1\pmod{p}.$$ Note that $$\binom{2p^a-2}{p^a-1}^2=p^{2a}\prod_{j=2}^{p^a-1}\l(\frac{2p^a-j}j\r)^2\equiv p^{2a}\pmod{p^{2a+1}}$$ and $$\binom{2k}k^2=\binom{p^a+(2k-p^a)}{0p^a+k}^2\equiv\binom{2k-p^a}k^2=0\pmod{p^2}$$ by Lucas’ theorem. So we have $$\frac{l^2}4\binom{2l}l^2=\frac{(2l-1)!^2}{(l-1)!^4}\not\equiv0\pmod{p^{2a}}$$ and $$\binom{2k}k^2\equiv p^{2a}\frac{(l-1)!^4}{(2l-1)!^2}=\frac{4p^{2a}}{l^2\binom{2l}l^2}\pmod{p^3}.$$ Therefore $$\begin{aligned}
\sum_{k=(p^a+1)/2}^{\lfloor3p^a/4\rfloor}\frac{\binom{2k}k^2}{16^k}\equiv&\sum_{k=(p^a+1)/2}^{\lfloor3p^a/4\rfloor}\frac{4p^{2a}}{16^k(p^a-k)^2\binom{2(p^a-k)}{p^a-k}^2}
\\\eq&\frac{p^{2a}}4\sum_{l=\lfloor p^a/4\rfloor+1}^{(p^a-1)/2}\frac{16^l}{l^2\binom{2l}l^2}\pmod{p^3}.\end{aligned}$$ For $k=1,\ldots,(p^a-1)/2$, clearly $$\begin{aligned}
\frac{\binom{(p^a-1)/2}{k}}{\binom{2k}k/{(-4)^k}}=&\frac{\binom{(p^a-1)/2}{k}}{\binom{-1/2}k}=\prod_{j=0}^{k-1}\frac{(p^a-1)/2-j}{-1/2-j}
\\=&\prod_{j=0}^{k-1}\l(1-\frac{p^a}{2j+1}\r)\equiv1\pmod p.\end{aligned}$$ Thus $$\begin{aligned}
\sum_{k=(p^a+1)/2}^{\lfloor3p^a/4\rfloor}\frac{\binom{2k}k^2}{16^k}\equiv&\frac{p^{2a}}4\sum_{k=\lfloor p^a/4\rfloor+1}^{(p^a-1)/2}\frac{1}{k^2\binom{(p^a-1)/2}k^2}
\\\equiv& p^{2a}\sum_{k=\lfloor p^a/4\rfloor+1}^{(p^a-1)/2}\frac{1}{\binom{(p^a-3)/2}{k-1}^2}\pmod{p^3}.\end{aligned}$$ So (\[2.6\]) is reduced to $$\label{2.7}
p^{2a-2}\sum_{k=\lfloor p^a/4\rfloor}^{(p^a-3)/2}\frac1{\binom{(p^a-3)/2}{k}^2}\equiv-\l(\f{-1}{p^{a-1}}\r)E_{p-3}\pmod p.$$ If $p^a\equiv1\pmod 4$, then $(p^a-3)/2$ is odd and hence $$\sum_{k=\lfloor p^a/4\rfloor}^{(p^a-3)/2}\frac1{\binom{(p^a-3)/2}{k}^2}=\frac12\sum_{k=0}^{(p^a-3)/2}\frac1{\binom{(p^a-3)/2}{k}^2}.$$ If $p^a\equiv3\pmod 4$, then $a\in\{3,5,\ldots\}$ and $$\sum_{k=\lfloor p^a/4\rfloor}^{(p^a-3)/2}\frac1{\binom{(p^a-3)/2}{k}^2}=\frac12\sum_{k=0}^{(p^a-3)/2}\frac1{\binom{(p^a-3)/2}{k}^2}+\frac12\cdot\frac1{\binom{(p^a-3)/2}{(p^a-3)/4}^2}.$$ In the case $p^a\eq3\pmod4$, as the fractional parts of $(p^a-3)/(2p)$ and $(p^a-3)/(4p)$ are $(p-3)/(2p)$ and $(p-3)/(4p)$ respectively, we have $$\l\lfloor\frac{(p^a-3)/2}p\r\rfloor=2\l\lfloor\frac{(p^a-3)/4}p\r\rfloor$$ and hence $$\ord_p\binom{(p^a-3)/2}{(p^a-3)/4}^2=2\sum_{j=1}^{a-1}\l(\l\lfloor\frac{(p^a-3)/2}{p^j}\r\rfloor-2\l\lfloor\frac{(p^a-3)/4}{p^j}\r\rfloor\r)<2a-2.$$ No matter $p^a\eq1\pmod 4$ or not, we always have $$p^{2a-2}\sum_{k=\lfloor p^a/4\rfloor}^{(p^a-3)/2}\frac1{\binom{(p^a-3)/2}{k}^2}\equiv\frac{p^{2a-2}}2\sum_{k=0}^{(p^a-3)/2}\frac1{\binom{(p^a-3)/2}{k}^2}\pmod{p}.$$ So (\[2.7\]) has the following equivalent form: $$\label{2.8}p^{2a-2}\sum_{k=0}^{(p^a-3)/2}\frac1{\binom{(p^a-3)/2}{k}^2}\equiv-2\l(\f{-1}{p^{a-1}}\r)E_{p-3}\pmod p.$$
The identity (\[2.4\]) with $n=(p^a-1)/2$ yields that $$\sum_{k=0}^{(p^a-3)/2}\frac1{\binom{(p^a-3)/2}{k}^2}=\frac{2((p^a-1)/2)^2}{(p^a+1)/2}\sum_{k=1}^{(p^a-1)/2}\frac1{k\binom{p^a-k}{(p^a-1)/2-k}}.$$ So (\[2.8\]) is reduced to $$\label{2.9}
p^{2a-2}\sum_{k=1}^{(p^a-1)/2}\frac1{k\binom{p^a-k}{(p^a+1)/2}}\equiv-2\l(\f{-1}{p^{a-1}}\r)E_{p-3}\pmod p.$$ In view of Lemma \[Lem2.4\], if $1\ls k\ls (p^a-1)/2$ and $p^{a-1}\nmid k$, then $$\frac{p^{2a-2}}{k\binom{p^a-k}{(p^a+1)/2}}\equiv0\pmod p.$$ Thus $$\begin{aligned}
p^{2a-2}\sum_{k=1}^{(p^a-1)/2}\frac1{k\binom{p^a-k}{(p^a+1)/2}}\equiv& p^{2a-2}\sum_{j=1}^{(p-1)/2}\frac1{p^{a-1}j\binom{p^a-p^{a-1}j}{(p^a+1)/2}}
\\=&\frac{p^a+1}2\sum_{j=1}^{(p-1)/2}\frac1{j(p-j)\binom{p^a-p^{a-1}j-1}{(p^a-1)/2}}
\\\equiv&-\frac12\sum_{j=1}^{(p-1)/2}\frac1{j^2\binom{p^a-p^{a-1}j-1}{(p^a-1)/2}}\pmod{p}.\end{aligned}$$ For each $j=1,\ldots,(p-1)/2$, by Lucas’ theorem we have $$\begin{aligned}
\binom{p^{a-1}(p-j)-1}{(p^a-1)/2}= &\bi{p^{a-1}(p-1-j)+p^{a-1}-1}{p^{a-1}(p-1)/2+(p^{a-1}-1)/2}
\\\eq&\bi{p-1-j}{(p-1)/2}\bi{p^{a-1}-1}{(p^a-1)/2}
\\\eq&(-1)^{(p^{a-1}-1)/2}\binom{p-j-1}{(p-1)/2}\pmod p,\end{aligned}$$ also $$\begin{aligned}
\bi{p-j-1}{(p-1)/2}=&\bi{p-1-j}{(p-1)/2-j}=(-1)^{(p-1)/2-j}\bi{-p+(p-1)/2}{(p-1)/2-j}
\\\eq&(-1)^{(p-1)/2-j}\bi{(p-1)/2}j\eq(-1)^{(p-1)/2-j}\bi{-1/2}j
\\=&(-1)^{(p-1)/2}\f{\bi{2j}j}{4^j}\pmod p.\end{aligned}$$ Therefore $$\begin{aligned}
&p^{2a-2}\sum_{k=1}^{(p^a-1)/2}\frac1{k\binom{p^a-k}{(p^a+1)/2}}\equiv\frac{(-1)^{(p^{a-1}+1)/2}}2\sum_{j=1}^{(p-1)/2}\frac1{j^2\binom{p-j-1}{(p-1)/2}}
\\\equiv&\frac{(-1)^{(p^{a-1}+1)/2}}2(-1)^{(p-1)/2}\sum_{j=1}^{(p-1)/2}\frac{4^j}{j^2\binom{2j}j}\pmod p.\end{aligned}$$ This, together with (\[2.1\]), yields the desired (\[2.9\]).
The proof of Theorem 1.1(ii) is now complete.
Proof of Theorem 1.2
====================
\[Lem3.1\] Let $p>3$ be a prime, and $m\in\{1,2,\ldots,(p-1)/2\}$. For any $p$-adic integer $t$, we have $$\begin{aligned}
\label{3.1}
\binom{m+pt-1}{(p-1)/2}\binom{-1-pt-m}{(p-1)/2}\equiv\frac{pt}m\pmod{p^2}.
\end{aligned}$$
. Since $$\begin{aligned}
\binom{m+pt-1}{(p-1)/2}
=&\frac{\prod_{r=0}^{m-1}(pt+r)\times\prod_{s=1}^{(p-1)/2-m}(pt-s)}{((p-1)/2)!}
\\\eq&\frac{(m-1)!pt(-1)^{(p-1)/2-m}((p-1)/2-m)!}{((p-1)/2)!}\pmod{p^2},\end{aligned}$$ and $$\begin{aligned}
&\binom{-m-pt-1}{(p-1)/2}=\frac{\prod_{j=1}^{(p-1)/2}(-m-pt-j)}{((p-1)/2)!}
\\\eq&\frac{(-1)^{(p-1)/2}(m+1)(m+2)\cdots(m+(p-1)/2)}{((p-1)/2)!}\pmod{p},\end{aligned}$$ we have $$\begin{aligned}
&\binom{m+pt-1}{(p-1)/2}\binom{-m-pt-1}{(p-1)/2}
\\\eq&\frac{pt(m-1)!(-1)^m((p-1)/2-m)!(m+1)(m+2)\cdots(m+(p-1)/2)}{((p-1)/2)!^2}
\\=&\frac{pt}m\frac{(-1)^m((p-1)/2-m)!(m+(p-1)/2)!}{((p-1)/2)!^2}=\frac{pt}m(-1)^m\frac{\binom{p-1}{(p-1)/2}}{\binom{p-1}{(p-1)/2+m}}
\\\eq&\frac{pt}m(-1)^m(-1)^{(p-1)/2}(-1)^{(p-1)/2+m}=\frac{pt}m\pmod{p^2}.\end{aligned}$$ This concludes the proof.
\[Rem3.1\] Let $p>3$ be a prime and $m\in\{(p+1)/2,\ldots,p-1\}$. For any $p$-adic integer $t$, by Lemma \[Lem3.1\] we have $$\begin{aligned}
&\binom{m+pt-1}{(p-1)/2}\binom{-1-pt-m}{(p-1)/2}
\\=&\binom{(m-p)+p(t+1)-1}{(p-1)/2}\binom{-1-p(t+1)-(m-p)}{(p-1)/2}
\\\equiv&\frac{p(t+1)}{m-p}\eq\f{p(t+1)}m\pmod{p^2}.\end{aligned}$$
\[Lem3.2\] Let $p>3$ be a prime. For $k\in\{1,2,\ldots,p-1\}$ and $p$-adic integer $t$, we have $$\label{tk}
\binom{pt}{k}\binom{-1-pt}{k}\equiv-\frac{p^2t^2}{k^2}-\frac{pt}k\pmod{p^3}.$$
. This is almost trivial. In fact, $$\begin{aligned}
\binom{pt}k\binom{-1-pt}k=&\frac{pt}{pt-k}\bi{-1+pt}k\bi{-1-pt}k
\\\eq& \f{pt}{pt-k}\bi{-1}k^2=\f{pt(p^2t^2+ptk+k^2)}{(pt)^3-k^3}
\\\eq&-\frac{p^2t^2}{k^2}-\frac{pt}k\pmod{p^3}.\end{aligned}$$ This proves (\[tk\]).
Recall that those $H_n=\sum_{0<k\ls n}1/k$ with $n\in\N$ are called harmonic numbers. If a prime $p$ does not divide an integer $a$, then we let $q_p(a)$ denote the Fermat quotient $(a^{p-1}-1)/p$.
\[Lehmer\] [(Lemma [@L])]{}. For any prime $p>3$, we have $$\begin{gathered}
H_{\lfloor p/2\rfloor}\eq-2q_p(2)\pmod p,\ H_{\lfloor p/4\rfloor}\eq-3q_p(2)\pmod p,
\\H_{\lfloor p/3\rfloor}\eq-\f32q_p(3)\pmod p \ \t{and}\ H_{\lfloor p/6\rfloor}\eq-2q_p(2)-\f 32q_p(3)\pmod p,\end{gathered}$$ where $q_p(2)=(2^{p-1}-1)/p$ and $q_p(3)=(3^{p-1}-1)/p$ stand for the Fermat quotients.
For $n\in\N$, define $$S_n(x)=\sum_{k=0}^n\binom{x}k\binom{-1-x}k\ \ \mbox{and}\ \ T_n(x)=\sum_{k=0}^n\binom{x}k\binom{-1-x}k\frac{1+2x}{1+2k}.$$ By [@S16 (2.2)] with $a=x+1$ and $b=0$, we have $$\begin{aligned}
\label{3.3}
S_n(x)+S_n(x+1)=2\binom{x}n\binom{-2-x}{n}.\end{aligned}$$ By [@S16 (2.2)] with $b=2$, we get $$\begin{aligned}
\label{3.5}
T_n(x)-T_n(x-1)=2\binom{x-1}n\binom{-x-1}n.\end{aligned}$$
[*Proof of Theorem 1.2*]{}. For any $p$-adic integer $a$, we let $\langle a\rangle_p$ denote the least nonnegative integer $r$ with $a\eq r\pmod p$. For convenience, we also set $n=(p-1)/2$.
\(i) For any $p$-adic integer $a\not\equiv0\pmod p$, by using (\[3.3\]) we get $$\begin{aligned}
&S_n(a)-(-1)^{\langle a\rangle_p}S_n(a-\langle a\rangle_p)
\\=&\sum_{k=0}^{\langle a\rangle_p-1}(-1)^k(S_n(a-k)+S_n(a-k-1))
\\=&\sum_{k=0}^{\langle a\rangle_p-1}(-1)^k 2\binom{a-k-1}n\binom{k-a-1}{n}\end{aligned}$$ and hence $$\begin{aligned}
&S_n(a)-(-1)^{\langle a\rangle_p}S_n(pt)
\\=&2\sum_{k=0}^{\langle a\rangle_p-1}(-1)^k\bi{\langle a\rangle_p+pt-k-1}{n}\bi{-1-pt-(\langle a\rangle_p-k)}{n},\end{aligned}$$ where $t:=(a-\langle a\rangle_p)/p$. By Lemma \[Lem3.2\], $$\begin{aligned}
\label{3.4}
S_n(pt)=\sum_{k=0}^n\binom{pt}k\binom{-1-pt}k\equiv1-\sum_{k=1}^n\frac{pt}k=1-ptH_n\pmod{p^2}.\end{aligned}$$ So, with helps of Lemma 3.1 and Remark 3.1, we have $$\label{delta}S_n(a)-(-1)^{\langle a\rangle_p}(1-ptH_n)
\eq2\sum_{k=0}^{\langle a\rangle_p-1}(-1)^k\f{p(t+\da_k)}{\langle a\rangle_p-k}\pmod{p^2},$$ where $\da_k$ takes $1$ or $0$ according as $\langle a\rangle_p-k>p/2$ or not.
Observe that $$\sum_{k=0}^{(p-1)/2}\f{\bi{2k}k\bi{3k}k}{27^k}=\sum_{k=0}^n\bi{-1/3}k\bi{-2/3}k=S_n(a)$$ with $a=-1/3$. Note that $$\langle a\rangle_p=\begin{cases}(p-1)/3&\t{if}\ p\eq1\pmod 3,
\\(2p-1)/3&\t{if}\ p\eq2\pmod 3.\end{cases}$$ Hence $$t:=\f{a-\langle a\rangle_p}p=\begin{cases}-1/3&\t{if}\ p\eq 1\pmod 3,\\-2/3&\t{if}\ p\eq2\pmod 3.\end{cases}$$
[*Case*]{} 1. $p\equiv1\pmod3$.
In this case, $\langle a\rangle_p=(p-1)/3$, $t=-1/3$, and $\da_k=0$ for all $k=0,\ldots,\langle a\rangle_p-1$. So we have $$\begin{aligned}
&S_n\l(-\f13\r)-(-1)^{(p-1)/3}(1-ptH_n)
\\\eq& 2pt(-1)^{(p-1)/3}\sum_{j=1}^{(p-1)/3}\f{(-1)^j}j=2pt\l(H_{(p-1)/6}-H_{(p-1)/3}\r)\pmod{p^2}.\end{aligned}$$ Combining this with Lemma \[Lehmer\] and recalling that $t=-1/3$, we immediately obtain the desired congruence $$S_n\l(-\f13\r)\eq1+\f23 p\,q_p(2)\pmod{p^2}.$$
[*Case*]{} 2. $p\equiv2\pmod 3$.
In this case, we have $\langle a\rangle_p=(2p-1)/3$, $t=-2/3$ and $$\da_k=\begin{cases} 1&\t{if}\ 0\leq k<(p+1)/6,\\ 0&\t{if}\ (p+1)/6\leq k\leq\langle a\rangle_p-1.\end{cases}$$ So we have $$\begin{aligned}
&S_n\l(-\f13\r)-(-1)^{(2p-1)/3}(1-ptH_n)
\\\eq&2p(t+1)\sum_{k=0}^{(p-5)/6}\f{(-1)^k}{\langle a\rangle_p-k}+2pt\sum_{k=(p+1)/6}^{(2p-4)/3}\frac{(-1)^k}{\langle a\rangle_p-k}
\\=& 2p(t+1)(-1)^{(2p-1)/3}\sum_{j=(p+1)/2}^{(2p-1)/3}\f{(-1)^j}j+2pt(-1)^{(2p-1)/3}\sum_{j=1}^{(p-1)/2}\frac{(-1)^j}j
\\=&-2p(t+1)\sum_{j=1}^{(2p-1)/3}\f{(-1)^j}j+2p\sum_{j=1}^{(p-1)/2}\frac{(-1)^j}j
\\=&-2p(t+1)\l(H_{\lfloor p/3\rfloor}-H_{\lfloor {2p}/3\rfloor}\r)+2p\l(H_{\lfloor p/4\rfloor}-H_{\lfloor p/2\rfloor}\r)\pmod{p^2}.\end{aligned}$$ Note that $$H_{\lfloor {2p}/3\rfloor}=\sum_{k=1}^{(p-1)/2}\l(\f1k+\f1{p-k}\r)-\sum_{j=1}^{(p-1)/3}\f1{p-j}\eq H_{\lfloor p/3\rfloor}\pmod p.$$ Therefore, $$S_n\l(-\f13\r)+1-ptH_{\lfloor p/2\rfloor}\eq 2p\l(H_{\lfloor p/4\rfloor}-H_{\lfloor p/2\rfloor}\r)\pmod{p^2}.$$ This, together with Lemma \[Lehmer\] and the fact that $t=-2/3$, yields the desired congruence $$S_n\l(-\f13\r)\eq-1-\f23 p\,q_p(2)\pmod{p^2}.$$
In view of the above, we have completed the proof of (\[1.4\]).
\(ii) For any $p$-adic integer $a$ with $a(2a+1)\not\eq0\pmod p$, if we set $t=(a-\langle a\rangle_p)/p$ then by (\[3.5\]) we have $$\begin{aligned}
T_n(a)-T_n(pt)=&\sum_{k=1}^{\langle a\rangle_p}(T_n(a-k+1)-T_n(a-k))
\\=&\sum_{k=1}^{\langle a\rangle_p}2\bi{a-k}n\bi{k-a-2}n
\\=&2\sum_{k=1}^{\langle a\rangle_p}\bi{m_k+pt-1}n\bi{-1-pt-m_k}n,\end{aligned}$$ where $m_k=\langle a\rangle_p-k+1$. In view of Lemmas \[Lem3.2\] and \[Lehmer\], $$\begin{aligned}
T_n(pt)-(1+2pt)=&\sum_{k=0}^n\binom{pt}k\binom{-1-pt}k\frac{1+2pt}{1+2k}-(1+2pt)
\\\eq&\binom{pt}n\binom{-1-pt}n\frac{1+2pt}p-\sum_{k=1}^{n-1}\frac{pt}{k(1+2k)}
\\\eq&\l(-\frac{p^2t^2}{n^2}-\frac{pt}n\r)\frac{1+2pt}p-\sum_{k=1}^{n-1}\frac{pt}{k(1+2k)}
\\\eq&2t+2pt-pt\sum_{k=1}^{n-1}\frac1k+2pt\sum_{k=1}^{n-1}\frac1{2k+1}
\\\eq&2t-2pt-ptH_n+2pt\l(H_{p-1}-\f{H_n}2\r)
\\\eq&2t-2pt+4ptq_p(2)\pmod{p^2}\end{aligned}$$ and hence $$T_n(pt)\equiv1+2t+4ptq_p(2)\pmod{p^2}.$$ Therefore, with the helps of Lemma \[Lem3.1\] and Remark 3.1, we have $$\begin{aligned}
&T_n(a)-(1+2t+4ptq_p(2))
\\\eq&2\sum_{k=1}^{\langle a\rangle_p}\bi{m_k+pt-1}n\bi{-1-pt-m_k}n
\\\eq&2\sum_{k=1}^{\langle a\rangle_p}\f{p(t+\da_k)}{m_k}
=2\sum_{j=1}^{\langle a\rangle_p}\f{pt}j+2\sum^{\langle a\rangle_p}_{j=1\atop j>p/2}\f1j\pmod {p^2},\end{aligned}$$ where $\da_k$ takes $1$ or $0$ according as $m_k>p/2$ or not. Below we deal with $a=-1/6,-1/4$.
Clearly, $$H_{p-k}=H_{p-1}-\sum_{0<j<k}\f1{p-j}\eq H_{k-1}\pmod p$$ for all $k=1,\ldots,p-1$. Thus, with the help of Lemma \[Lehmer\] we have $$H_{\lfloor{3p}/4\rfloor}\equiv H_{p-1-\lfloor 3p/4\rfloor}=H_{\lfloor p/4\rfloor}\eq-3q_p(2)\pmod p$$ and $$H_{\lfloor {5p}/6\rfloor}\equiv H_{p-1-\lfloor 5p/6\rfloor}=H_{\lfloor p/6\rfloor}\eq-2q_p(2)-\frac32q_p(3)\pmod p.$$
[*Case*]{} I. $\langle a\rangle_p<n$.
If $a=-1/6$, then $p\eq1\pmod 6$, $\langle a\rangle_p=(p-1)/6$ and $t=-1/6$. By the above, $$\begin{aligned}
T_n\l(-\f16\r)\equiv&\frac23-\frac23pq_p(2)-\frac p3H_{\lfloor p/6\rfloor}
\\\eq&\frac23-\frac23pq_p(2)-\frac p3\l(-2q_p(2)-\frac32q_p(3)\r)
\\\eq&\frac23+\frac p2q_p(3)\pmod{p^2}\end{aligned}$$ and thus $$\sum_{k=0}^n\frac{\binom{6k}{3k}\binom{3k}k}{(2k+1)432^k}=\frac32T_n\l(-\f16\r)\equiv1+\frac34pq_p(3)=\frac{3^p+1}4\pmod {p^2}.$$
If $a=-1/4$, then $p\eq1\pmod 4$, $\langle a\rangle_p=(p-1)/4$ and $t=-1/4$. By the above, $$\begin{aligned}
T_n\l(-\f14\r)\equiv&\frac12-pq_p(2)-\frac p2H_{\lfloor p/4\rfloor}
\\\eq&\frac12-pq_p(2)-\frac p2(-3q_p(2))
\\\eq&\frac12+\frac p2q_p(2)\pmod{p^2}\end{aligned}$$ and thus $$\sum_{k=0}^n\frac{\binom{4k}{2k}\binom{2k}k}{(2k+1)64^k}=2T_n(-1/4)\equiv1+pq_p(2)=2^{p-1}\pmod {p^2}.$$
[*Case*]{} II. $\langle a\rangle_p>n$.
If $a=-1/6$, then $p\eq5\pmod 6$, $\langle a\rangle_p=(5p-1)/6$ and $t=-5/6$. By the above, $$\begin{aligned}
T_n\l(-\f16\r)\equiv&-\frac23+\frac23pq_p(2)+\frac p3H_{\lfloor {5p}/6\rfloor}
\\\eq&-\frac23+\frac23pq_p(2)+\frac p3\l(-2q_p(2)-\frac32q_p(3)\r)
\\\eq&-\frac23-\frac p2q_p(3)\pmod{p^2}\end{aligned}$$ and hence $$\sum_{k=0}^n\frac{\binom{6k}{3k}\binom{3k}k}{(2k+1)432^k}=\frac32T_n\l(-\f16\r)\equiv-1-\frac34pq_p(3)=-\frac{3^p+1}4\pmod {p^2}.$$
If $a=-1/4$, then $p\eq3\pmod 4$, $\langle a\rangle_p=(3p-1)/4$ and $t=-3/4$. So $$\begin{aligned}
T_n\l(-\f14\r)\equiv&-\frac12+pq_p(2)+\frac p2H_{\lfloor{3p}/4\rfloor}
\\\eq&-\frac12+pq_p(2)+\frac p2(-3q_p(2))
\\\eq&-\frac12-\frac p2q_p(2)\pmod{p^2}\end{aligned}$$ and hence $$\sum_{k=0}^n\frac{\binom{4k}{2k}\binom{2k}k}{(2k+1)64^k}=2T_n\l(-\f14\r)\equiv-1-pq_p(2)=-2^{p-1}\pmod {p^2},$$
The proof of Theorem 1.2 is now complete.
. The first author would like to thank Dr. Hao Pan for help comments.
[ST10]{} E. Lehmer, [*On congruences involving Bernoulli numbers and the quotients of Fermat and Wilson*]{}, Ann. Math. [**39**]{}(1938), 350–360. E. Mortenson, [*A supercongruence conjecture of Rodriguez-Villegas for a certain truncated hypergeometric function*]{}, J. Number Theory [**99**]{} (2003), 139-147. E. Mortenson, [*Supercongruences for truncated ${}_{n+1}F_n$ hypergeometric series with applications to certain weight three newforms*]{}, Proc. Amer. Math. Soc. [**133**]{} (2005), 321-330. F. Rodriguez-Villegas, [*Hypergeometric families of Calabi-Yau manifolds. Calabi-Yau Varieties and Mirror Symmetry (Yui, Noriko (ed.) et al., Toronto, ON, 2001)*]{}, 223-231, Fields Inst. Commun., [**38**]{}, Amer. Math. Soc., Providence, RI, (2003). Z.-H. Sun, [*Super congruences involving Bernoulli polynomials*]{}, Int. J. Number Theory, in press. Z.-W. Sun, [*Super congruences and Euler numbers*]{}, Sci. China Math. [**54**]{} (2011), 2509-2535. Z.-W. Sun, [*Supercongruences involving products of two binomial coefficients*]{}, Finite Fields Appl. [**22**]{} (2013), 24-44. Z.-W. Sun, [*$p$-adic congruences motivated by series*]{}, J. Number Theory [**134**]{} (2014), no.1, 181-196. B. Sury, T.-M. Wang and F.-Z. Zhao, [*Identities involving reciprocals of binomial coefficients*]{}, J. Integer Seq. [**7**]{} (2004), Article 04.2.8.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In the case of monotone independence, the transparent understanding of the mechanism to validate the central limit theorem (CLT) has been lacking, in sharp contrast to commutative, free and Boolean cases. We have succeeded in clarifying it by making use of simple combinatorial structure of peakless pair partitions.'
author:
- |
Hayato Saigo\
Graduate School of Science, Kyoto University,\
Kyoto 606-8502, Japan
title: A simple proof for monotone CLT
---
2000 AMS Mathematics Subject Classification: 46L53
Introduction
============
As a generalization of probability spaces, we define the notion of algebraic probability space.
(Algebraic probability space) An algebraic probability space is a pair $(A,\varphi )$ consisting of a unital \*-algebra $A$ and of a state $\varphi $ .
An element $a \in A$ is called an algebraic random variable. For algebraic random variables, quantities such as $\varphi (a_1a_2...a_n)$ are called mixed moments.
In quantum probability, the notion of independence is understood as a reduction rule in calculation of mixed moments. Among many different rules, commutative, free [@V1], Boolean [@B; @S-W] and monotone independence [@Lu; @Mur1; @Mur2] are known as basic types [@B-S1; @Mur0; @Mur4; @Fra1]. In the present paper, we focus on monotone independence, which is defined later.
We first introduce a key notion, a peak for a mapping (not necessarily homomorphism) between finite ordered set.
\[peak\] Let $S, T$ are finite ordered sets. An element $s \in S$ is said to be a peak of a map $f: S \longrightarrow T$ if $f(s) > f(s')$ holds for any $s'$ $(\neq s)$ which is next to $s$ (that is, there is no elements between $s$ and $s'$ with respect to the order on $S$).
Now we introduce the notion of monotone independence.
(Monotone independence). Let $\left\{A_{\lambda}; \lambda \in \Lambda \right\}$ be a family of \*-subalgebras of $A$, where the index set $\Lambda $ is equipped with a linear order $<$.
$\left\{A_{\lambda} \right\}$ is said to be monotone independent if $$\varphi (a_1...a_i...a_n)=\varphi (a_i)\varphi(a_1...a_{i-1}a_{i+1}...a_n)$$ holds for $a_i\in A_{\lambda _i}\backslash \textrm{\boldmath $C$}1$, whenever $i$ is a peak of the mapping $j \mapsto \lambda_j$.
For a monotone (or commutative, free, Boolean) independent family of \*-subalgebras $\left\{A_{\lambda } \right\} $, the following impotant property holds, which is known as the singleton condition [@A-O; @Obata] discussed by von Waldenfels in 1970s.
(Singleton condition). Let $(A, \varphi )$ be an algebraic probability space. If a family of \*-subalgebras $\left\{A_{\lambda } \right\}$ satisfies the singleton condition if for any finite sequence $\lambda _1,...,\lambda _n \in \Lambda $ with a **singleton** $\lambda _s$, i.e. $\lambda _s\neq \lambda _i$ for all $i\neq s$, $$\varphi (a_1...a_s...a_n)=\varphi (a_s)\varphi(a_1...a_{s-1}a_{s+1}...a_n)$$ holds when $ a_i \in A_{\lambda_i} $.
This condition is essential for understanding asymptotic behavior of algebraic random variables. In fact, the following theorem holds [@A-O; @Obata].
\[fundamental\] Let $(A,\varphi)$ be an algebraic probability space and $(a_n)$ a sequence of elements of $A$ which satisfies the following:
i\) $a_i=a_i^{\ast }$,
ii\) $\varphi (a_i)=0$,
iii\) $\varphi (a_i^2)=1$,
iv\) $\left\{a_i \right\}$ has uniform mixed moments, i.e., for $m\geq 1$, $$\sup \left\{\vert \varphi (a_{n_1}a_{n_2}\cdots a_{n_m})\vert \mid n_1,n_2,\cdots ,n_m \in \textrm{\boldmath $N$} \right\} < \infty$$
v\) $\left\{ \textrm{\boldmath $C$} [a_i] \right\}$satisfies the singleton condition.
Then, $$M_m:=\lim _{N \rightarrow \infty} \varphi \left(\left(\frac{1}{\sqrt{N}}\sum^{\infty }_{n=1} a_n \right)^m\right)$$ can be computed as follows: $$M_{2m+1}=0,\; \qquad M_{2m}=\lim _{N\rightarrow \infty }\frac{1}{N^m}\sum _{n\in \Pi(\left\{ 1,2,\cdots , 2m \right\},
\left\{ 1,2,\cdots , N\right\})}\varphi (a_{n_1}a_{n_2}\cdots a_{n_{2m}})$$ Here, $$\Pi(S,T):=\left\{f:S \longrightarrow T \text{ such that } \left| f^{-1}(t)\right|
=2 \text{ or }0 \right\}$$ for finite sets S and T.
A simple proof for the monotone CLT
===================================
In the case of commutative, free and Boolean independence, the corresponding central limit theorems (CLTs) can be proved directly from Theorem \[fundamental\] by using some simple combinatorics. The purpose of this section is to derive the mononotone CLT(Theorem \[the Monotone CLT\] ) from Theorem \[fundamental\] by using simple combinatorial argument.
To do this, we define a subset of $\Pi (S,T)$ for finite **ordered** sets $S$ and $T$, which we call the set of **peakless** pair partitions $\Pi_{0}(S,T)$ as follows:
$$\Pi_{0}(S,T):=\left\{f \in \Pi (S,T) \mid \text{ there is no
peak of } f \right\}.$$
For those who are familiar with the notion of “monotone partition” defined by Muraki to classify quasi-universal products [@Mur0], we note that a mapping $f:S \longrightarrow T$ which is an element of $\Pi (S,T)$ belongs to $\Pi_{0}(S,T)$ if and only if it is a monotone partition, when $S,T$ are finite linearly ordered set.
It is easy to see that for the case of monotone independence, only peakless pair partions contribute and each contributions are equal to $1$, in the expression of moments in Thm.\[fundamental\]. Hence the problem is reduced to counting the number of the elements in $\Pi_{0}(\left\{ 1,2,\cdots , 2m \right\},
\left\{ 1,2,\cdots , N\right\})$.
\[lemma\] $$\left|\Pi_{0}(\left\{1,2,\cdots , 2m \right\},
\left\{1,2,\cdots , N\right\}) \right| = \binom{N}{m}\times (2m-1)!!$$ for $N\geq m\geq 1$.
First note that $f$ is an element of $\Pi_{0}(\left\{ 1,\cdots , 2m \right\},
\left\{1,\cdots , m\right\})$ if and only if $f\mid_{\left\{1,\cdots , 2m \right\}\backslash \left\{i,i+1 \right\}}$ is an element of $\Pi_{0}(\left\{1,\cdots , 2m \right\}\backslash \left\{i,i+1 \right\},\left\{1,\cdots , m-1 \right\})$ for $i$ which satisfies $f^{-1}(m)=\left\{i, i+1 \right\}$. Then we have $$\begin{aligned}
\lefteqn{} \\
\left|\Pi_{0}(\left\{ 1,\cdots , 2m \right\},
\left\{ 1,\cdots , m\right\}) \right|
=&\sum_{i} \left|\Pi_{0}(\left\{ 1,\cdots , 2m \right\}\backslash \left\{i,i+1 \right\},
\left\{ 1,\cdots , m-1\right\}) \right| & \\
=&(2m-1)\times \left|\Pi_{0}(\left\{ 1,\cdots , 2(m-1) \right\},
\left\{ 1,\cdots , m-1\right\}) \right| & \\\end{aligned}$$ for $N\geq m \geq 2$, and hence, $\left|\Pi_{0}(\left\{ 1,\cdots , 2m \right\},
\left\{ 1,\cdots , m\right\}) \right|=(2m-1)!!$ holds.
It is easy to see that $
\left|\Pi_{0}(\left\{ 1,\cdots , 2m \right\},
\left\{ 1,\cdots , N\right\}) \right|=\binom{N}{m}\times \left|\Pi_{0}(\left\{ 1,\cdots , 2m \right\},
\left\{ 1,\cdots , m\right\}) \right|,
$ and we obtain the lemma.
The essence of the proof above can be understood as follows:
First note that $\Pi(\left\{ 1,2,\cdots , 2m \right\},
\left\{ 1,2,\cdots , N\right\})$ represents all the possible ways to devide $2m$ balls into $m$ pairs and to paint each pairs by different $m$ colours chosen from $N$ colours. Then, every peakless partition corresponding to the way how to paint balls can be generated uniquely by the following algorithm. Put $2m$ balls in line, choose $m$ colours from $N$ colours. Paint a couple of neighbouring balls by the highest colour. Then repeat this procedure, treating all the balls already painted and all the colours already used.
Now it is easy to count the way of such colourings. The number of the choices of $m$ colours from $N$ colours is $\binom{N}{m}$. The number of the choices of a couple of neighbouring balls painted by the highest colour is $2m-1$, and the number of the choices of a couple of neighbouring (ignoring the balls which have painted ) balls painted by the second highest colour is $(2m-2)-1=2m-3,\cdots ,$ and so on.
The proof of the lemma is the prototype of combinatorial argument in [@H-S].
Then we can prove the monotone CLT [@Lu; @Mur1; @Mur2; @A-O; @Obata].
\[the Monotone CLT\] Let $(A,\varphi)$ be an algebraic probability space and $(a_n)$ a sequence of elements of $A$ which satisfies the following:
i\) $a_i=a_i^{\ast }$,
ii\) $\varphi (a_i)=0$,
iii\) $\varphi (a_i^2)=1$,
iv\) $\left\{ \textrm{\boldmath $C$} [a_i] \right\}$ has uniform mixed moments,
v\) $\left\{ \textrm{\boldmath $C$} [a_i] \right\}$ is monotone independent.
Then $$\lim _{N \rightarrow \infty} \varphi \left(\left(\frac{1}{\sqrt{N}}\sum^{\infty }_{n=1} a_n \right)^m\right)
=\frac{1}{\pi}\int _{-\sqrt{2}}^{\sqrt{2}}\frac{x^m}{\sqrt{2-x^2}}dx$$ for $m=0,1,2,...$.
By the lemma \[lemma\], $$M_{2m}=\lim _{N\rightarrow \infty }N^{-m} \binom{N}{m}(2m-1)!!=
\lim _{N\rightarrow \infty } N^{-m} \binom{N}{m} m! \times \frac{(2m-1)!!}{m!}=\frac{(2m-1)!!}{m!}$$ This is nothing but the $2m$-th moment of the standard arcsine law.
Acknowledgements {#acknowledgements .unnumbered}
================
The author was inspired by the series lectures given by Prof. Nobuaki Obata, to whom he is grateful for encouragements. He thanks Prof. Izumi Ojima and Mr. Ryo Harada for many useful comments to refine drafts. He also thanks Prof. Shogo Tanimura, Mr. Takahiro Hasebe, Mr. Hiroshi Ando, and Mr. Kazuya Okamura for their interests and comments.
[122]{} L. Accardi and N. Obata, Foundations of Quantum Probability Theory (in Japanese), Makino Shoten Publ.(2003), M. Bożejko, Uniformly bounded representations of free groups, J. Reine Angew. Math. **377** (1987), 170-186. U. Franz, Unification of boolean, monotone, anti-monotone, and tensor independence and Lévy Processes, Math. Z. **243** (2003), 779-816. A. Ben Ghorbal and M. Schürmann, Non-commutative notions of stochastic independence, Math. Proc. Comb. Phil. Soc. **133** (2002), 531-561. T. Hasebe and H. Saigo, The Monotone Cumulants, Arxiv:0907.4896 . Y. G. Lu, An interacting free fock space and the arcsine law, Probability and Math. Stat. **17**, Fasc.1 (1997), 149-166. N. Muraki, Non-commutative Brownian motion in monotone Fock space, Commun. Math. Phys. **183** (1997), 557-570. N. Muraki, Monotonic independence, monotonic central limit theorem and monotonic law of small numbers, Infin. Dimens. Anal. Quantum Probab. Relat. Top. **4** (2001), 39-58. N. Muraki, The five independences as quasi-universal products, Infin. Dimens. Anal. Quantum Probab. Relat. Top. **5**, no. 1 (2002), 113-134. N. Muraki, The five independences as natural products, Infin. Dimens. Anal. Quantum Probab. Relat. Top. **6**, no. 3 (2003), 337-371. N. Obata, Notions of independence in quantum probability and spectral analysis of graphs, Amer. Math. Soc. Transl. **223** (2008), 115-136. R. Speicher and R. Woroudi, Boolean convolution, in Free Probability Theory, ed. D. Voiculescu, Fields Inst. Commun., vol. 12 (Amer. Math. Soc., 1997), 267-280. D. Voiculescu, Symmetries of some reduced free product algebras, Operator algebras and their connections with topology and ergodic theory, Lect. Notes in Math. **1132**, Springer (1985), 556-588.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Recent elastic and inelastic neutron scattering studies of the highly frustrated pyrochlore antiferromagnet Tb$_2$Ti$_2$O$_7$ have shown some very intriguing features that cannot be modeled by the local $\langle 111 \rangle$ classical Ising model, naively expected to describe this system at low temperatures. Using the random phase approximation to take into account fluctuations between the ground state doublet and the first excited doublet, we successfully describe the elastic neutron scattering pattern and dispersion relations in Tb$_2$Ti$_2$O$_7$, semi-quantitatively consistent with experimental observations.'
author:
- 'Ying-Jer Kao'
- Matthew Enjalran
- Adrian Del Maestro
- 'Hamid R. Molavian'
- 'Michel J. P. Gingras'
title: |
Understanding Paramagnetic Spin Correlations\
in the Spin-Liquid Pyrochlore Tb$_2$Ti$_2$O$_7$
---
The search for the physical realization of a spin liquid in three dimensions has been a long quest for the condensed matter community. Recently, geometrically frustrated magnetic systems have been the focus of intensive experimental and theoretical studies because it is believed that geometrical frustration can inhibit the formation of long-range order, thus enabling the system to remain paramagnetic down to low temperatures. Among three-dimensional systems, the pyrochlore lattice of corner-sharing tetrahedra have been studied extensively. It has been shown theoretically and numerically [@Villain1979; @Reimers1992; @Moessner1998] that for classical Heisenberg spins with nearest-neighbor antiferromagnetic (AF) exchange, there is no transition to long-range magnetic order at finite temperature. This makes AF materials based on the pyrochlore lattice excellent candidates to search for the low-temperature spin liquid state. A number of experimental studies on insulating pyrochlore materials have been carried out. Interestingly, most materials either develop long-range Néel order, such as Gd$_2$Ti$_2$O$_7$ [@Ramirez2002] and Er$_2$Ti$_2$O$_7$ [@Champion2001], or reveal spin-glass behavior, such as Y$_2$Mo$_2$O$_7$ [@Gardner2001a]. The “spin ice” materials, Ho$_2$Ti$_2$O$_7$ [@Harris1997; @Bramwell2001] and Dy$_2$Ti$_2$O$_7$ [@Ramirez1999; @denHertog2000] exhibit low-temperature thermodynamic properties reminiscent of Pauling’s “water ice model” [@GingrasReview]. In these systems, an effective ferromagnetic interaction is frustrated due to the single-ion $\langle 111 \rangle$ Ising anisotropy [@Bramwell1998; @GingrasReview]. The behavior of these systems can be quantitatively described by the $\langle 111 \rangle$ Ising spin model with nearest-neighbor exchange and long-range dipolar interactions [@denHertog2000; @Bramwell2001; @GingrasReview].
Tb$_2$Ti$_2$O$_7$ shows, however, very different and intriguing behavior. It is believed that Tb$_2$Ti$_2$O$_7$ belongs to the same family of $\langle 111 \rangle$ Ising systems as Dy$_2$Ti$_2$O$_7$ and Ho$_2$Ti$_2$O$_7$ but with an effective nearest neighbor AF interaction [@Gardner1999; @denHertog2000; @Gingras2000; @Gingras2002; @Gardner2001]. The same spin model that very successfully described the spin-ice systems predicts it to have a noncollinear Néel $\mathbf{Q}=0$ order, with all spins pointing into or out of each tetrahedron, at about 1 K [@denHertog2000]. In dramatic contrast, Tb$_2$Ti$_2$O$_7$ remains a spin liquid, or “cooperative paramagnet”, down to 70mK [@Gardner1999; @Gardner2001]. In addition, recent paramagnetic neutron scattering studies show that the scattering pattern for this material is not consistent with a $\langle 111 \rangle$ Ising model, while a Heisenberg spin model [@Gardner2001; @Enjalran] or some level of relaxation away from the $\langle
111 \rangle$ Ising model [@Yasui2002] can better describe the observed neutron scattering pattern. These results suggest that the restoration of spin isotropy in the system, despite its expected Ising-like nature at low temperature [@Gingras2000; @Siddharthan1999; @Rosenkranz2000; @RosenkranzPrivate], is essential in understanding the paramagnetic spin correlations. Inelastic neutron scattering studies have also been performed on this system and partial softening of the magnetic excitations at an energy of about 20 K has been observed [@Gardner1999; @Gardner2001; @Kanada1999]. This has been attributed to a (spin) roton-like mode, as in liquid $^4$He [@Gardner1999; @Gardner2001; @Feynman], which further indicates a more isotropic nature of the spins. Given the ensemble of evidences, it would appear that one needs a more isotropic spin model to understand the paramagnetic spin correlations in this system. More importantly, such a “restoration” of spin isotropy may also be the key to understanding why Tb$_2$Ti$_2$O$_7$ fails to order down to low temperatures. In this paper, we employ the random-phase approximation (RPA) [@Jensen] to take into account the fluctuations between the ground state doublet and the first excited doublet. We successfully describe the observed paramagnetic spin correlations in Tb$_2$Ti$_2$O$_7$, without any assumptions regarding the nature of the spins, while we still obtain the $\mathbf{Q}=0$ Néel order at low temperatures. This result makes the fact that Tb$_2$Ti$_2$O$_7$ fails to order even more puzzling.
We begin with the model spin Hamiltonian, $$\mathcal{H}=\frac{1}{2}\sum_{i,j,\alpha,\beta,a,b} S^{\alpha}_{i,a}
\mathcal{J}^{\alpha\beta}_{ab}(i,j)
S^{\beta}_{j,b}+\sum_{i,a}\mathcal{H}_{\rm CF}(i,a),$$ where $i,j$ are indices of the Bravais lattice vectors for the FCC lattice, $a,b$ are indices of sublattice basis vectors and $\alpha,\beta$ are indices for the spatial coordinates. $\mathcal{H}_{\rm CF}$ is the single-ion crystal field (CF) Hamiltonian. The spin-spin interaction matrix $\mathcal{J}$, including both exchange and long-range dipolar interactions, reads $$\mathcal{J}^{\alpha\beta}_{ab}(i,j)=-J_1\delta^{\alpha\beta}\delta_{nn}+D_{dd}\left[
\frac{\delta^{\alpha\beta}}{ |\mathbf{R}^{ab}_{ij}|^3}-\frac{3 R^{ab,\alpha}_{ij}
R^{ab,\beta}_{ij}}{ |\mathbf{R}^{ab}_{ij}|^5}
\right],$$ where $\delta_{nn}$ refers to nearest-neighbor interaction, and $R^{ab,\alpha}_{ij}$ denotes the $\alpha$ component of the interspin vector $\mathbf{R}^{ab}_{ij}$ that connects spins $\mathbf{S}_{i,a}$ and $\mathbf{S}_{j,b}$. The nearest-neighbor exchange for Tb$_2$Ti$_2$O$_7$ is given by $J_1\approx-0.167$ K [@exchange], and $D_{dd}=(g\mu_{B})^2 \mu_0/4\pi$, where $\mu_B$ is the Bohr magneton, $\mu_0$ the magnetic permeability, and $g=3/2$ for Tb$^{3+}$.
![\[fig:scattering\] (a) Experimental elastic neutron scattering pattern of Tb$_2$Ti$_2$O$_7$ in the $(h,h,l)$ plane of reciprocal space at $T$=9 K, from Ref. . Dark blue shows the lowest intensity level, red-brown the highest. (b) $S(\mathbf{Q})$ for the $\langle 111
\rangle$ Ising model at $T$=9 K. (c) $S(\mathbf{Q})$ for the doublet-doublet model at $T$=9 K. Note that magnetic form factor $|F(\mathbf{Q})|^2$ is divided out in both experimental and theoretical results. ](paperscatcmt.eps){width="3.2in"}
The single-ion susceptibility is given by [@Jensen] $$\begin{aligned}
\chi^{0,\alpha\beta}_{a}(\omega)&=&\sum_{\mu,\nu}^{{E_{\mu}}\ne{E_{\nu}}}\frac{M^\alpha_{\nu\mu,a}
M^\beta_{\mu\nu,a}}{{E_{\mu}}-{E_{\nu}}-\hbar(\omega+i 0^+)}(n_{\nu}-n_{\mu})\nonumber\\
& &+\;\frac{\delta(\omega)}{k_B T}\sum_{\mu,\nu}^{{E_{\mu}}={E_{\nu}}}
M^\alpha_{\nu\mu,a}M^\beta_{\mu\nu,a}n_{\nu},
\label{eq:chi0}\end{aligned}$$ where $n_{\nu}$ is the thermal occupation fraction for state $\nu$. The matrix elements for the single-ion states are given by $M^\alpha_{\nu\mu,a}=\sum_{\bar{\alpha}}\langle \nu| {S}^{\bar{\alpha}} | \mu
\rangle u^{\alpha}_{\bar{\alpha},a}$, where $u^{\alpha}_{\bar{\alpha},a}$ is the rotation matrix from the local ($\bar{\alpha}$) frame defined on sublattice $a$ to the global ($\alpha$) frame. The spin operator ${S}^{\bar{\alpha}}$ acts on the CF states defined along the local quantization axis [@Gingras2000]. The local wave function structure has been obtained from both experimental measurements and theoretical calculations [@Gingras2000; @Rosenkranz2000; @RosenkranzPrivate]. The primary components are $J_z$ eigenstates $|\pm 4 \rangle$ for the ground state doublet and $|\pm5\rangle$ for the first-excited state doublet. The doublets are separated by an anisotropy gap $\Delta\approx20$ K, which is comparable with the Curie-Weiss temperature $\theta_{\rm CW}\approx-19$ K [@Gingras2000].
The RPA equation which takes into account the two-ion interaction contribution to the full susceptibility [@Jensen] is given by $$\chi^{\alpha\beta}_{ab}(\mathbf{q},\omega)+\sum_{\gamma,\delta,c}
\chi^{0,\alpha\gamma}_{a}(\omega)
\mathcal{J}^{\gamma\delta}_{ac}(\mathbf{q})\chi^{\delta\beta}_{cb}(\mathbf{q},\omega)=
\delta^{\ }_{ab}\chi^{0,\alpha\beta}_{a}(\omega),$$ where $\mathcal{J}(\mathbf{q})$ is the Fourier transformation of the interaction matrix $\mathcal{J}(i,j)$. The slowly converging, infinite lattice sum of the dipolar interaction is handled using Ewald summation techniques [@Ewald; @Enjalran]. We solve for $\chi^{\alpha\beta}_{ab}(\mathbf{q},\omega)$ numerically using LAPACK routines.
The elastic and dynamical neutron cross-sections are related to the spin susceptibilities by summing over the sublattice contributions and taking the transverse components of $\mathbf{S}$ perpendicular to $\mathbf{Q}$ [@Jensen],
$$\begin{aligned}
S_{\rm el}(\mathbf{Q}) &\propto& \frac{|F(\mathbf{Q})|^2}{k_B
T} \sum_{\alpha,\beta;a,b}(\delta_{\alpha\beta}-\hat{Q}_\alpha\hat{Q}_\beta)
\nonumber\\
&\times &\exp\left[-i(\mathbf{r}^a-\mathbf{r}^b)\cdot
\mathbf{G}\right]\mbox{Re}\,\chi_{ab}^{\alpha\beta}(\mathbf{q});\\
S_{\rm dyn}(\mathbf{Q},\omega) &\propto& \frac{|F(\mathbf{Q})|^2}{1-\exp(-\hbar
\omega/k_B T)} \sum_{\alpha,\beta;a,b}(\delta_{\alpha\beta}-\hat{Q}_\alpha\hat{Q}_\beta)
\nonumber \\
&\times &\exp\left[-i(\mathbf{r}^a-\mathbf{r}^b)\cdot
\mathbf{G}\right]\mbox{Im}\,\chi_{ab}^{\alpha\beta}(\mathbf{q},\omega),\end{aligned}$$
where $\mathbf{Q=G+q}$, $\mathbf{G}$ is a reciprocal lattice vector of the FCC lattice, and $\mathbf{q}$ is a primitive vector in the first Brillouin zone. $\mathbf{r}^a$ is the sublattice basis vector and $F(\mathbf{Q})$ is the magnetic form factor for Tb$^{3+}$.
Figure \[fig:scattering\]a shows the experimental scattering pattern in the $(h,h,l)$ plane at 9 K [@Gardner2001]. Note that there is a strong intensity maximum around $(0,0,2)$. Figure \[fig:scattering\]b shows the $S(\mathbf{Q})$ calculation using the $\langle 111 \rangle$ Ising model, i.e., the anisotropy gap $\Delta\rightarrow\infty$. It is clear that this model fails to reproduce the correct neutron scattering pattern as observed experimentally, as there is only weak non-critical intensity around $(0,0,2)$. It has been shown that in the $\langle 111 \rangle$ Ising model, the intensity at $(0,0,2)$ and $(0,0,0)$ are exactly correlated, and vanishes for Tb$_2$Ti$_2$O$_7$ [@Enjalran; @Gardner2001; @Gingras2000]. Using a more realistic doublet-doublet model for Tb$_2$Ti$_2$O$_7$, we are able to qualitatively reproduce the experimentally observed scattering pattern (Fig. \[fig:scattering\]c). It captures most details of the experimental pattern, such as the intensity maximum around $(0,0,2)$, the minima around $(0,0,0)$, $(2,2,2)$ and $(0,0,4)$. This excellent agreement between theory and experiment indicates that to properly understand the spin correlations in Tb$_2$Ti$_2$O$_7$, the fluctuations out of the ground state doublet to the first excited doublet, originating from the first term in Eq. , are important. It should be noted that this result is obtained from the simple level scheme described above with no other assumptions about the nature of the spin or the details of the wave functions.
Line scans in reciprocal space along three high-symmetry directions $(0,0,l)$, $(h,h,0)$ and $(h,h,h)$ at $T=$4 K are plotted in Fig. \[fig:inelastic\]. Open symbols are the data points for $S(\mathbf{Q})$. Filled symbols are the dispersion relations $E(\mathbf{Q})$ for the lowest-energy band of magnetic excitations at 4 K. The dispersion relations show good semi-quantitative agreement with the experimental dispersion observed in single crystal Tb$_2$Ti$_2$O$_7$ reported in Fig. 9 of Ref. . We also find that the minimum in $E(\mathbf{Q})$ corresponds to the maximum in $S(\mathbf{Q})$ in all symmetry directions, as is observed in the experiment. The sharp jumps in dispersions, e.g., near $(0,0,1)$ and $(0,0,3)$, correspond to shifts of spectral maximum between different branches, since the experiment and current calculation track only the maximum peaks in the energy spectra. In principle, there should be four branches of magnetic excitations with different intensities due to the four sublattice structure. Higher resolution experiments need to be performed to map out the dispersion relations of these branches.
![\[fig:inelastic\] The open symbols show the line scans of the $S(\mathbf{Q})$ for the three high-symmetry directions: from top to bottom, $(0,0,l)$, $(h,h,0)$ and $(h,h,h)$. The filled symbols show the dispersion relation for the lowest-lying branch of magnetic excitations at $T=4$ K. ](paperfig2.eps){width="2.8in"}
![\[fig:temp\] The dispersion relations of the three lowest-lying magnetic excitations along the $(h,h,h)$ direction at both 12 K and 30 K. The lowest band displays a more pronounced energy dispersion as temperature is lowered, in agreement with experimental observations [@Gardner2001]. ](paperfig3.eps){width="2.8in"}
To study the temperature effects on the dispersion relations of the lowest three bands of magnetic excitations, we use the wave functions obtained from crystal field calculations in the point charge approximation and in the fixed $J=6$ manifold for Tb$^{3+}$ [@Gingras2000]. This gives the CF level scheme with two low-lying doublets as described above, and two higher-energy singlets [@Gingras2000; @Gardner2001]. The energy levels are given as 0, 24.6, 113.9 and 132.7K, consistent with experiments. The two higher singlets have large $|\pm3\rangle$ contributions and some mixing with other $|J,M_J\rangle$ components [@Gingras2000]. All the excited states are connected to the ground state doublet through the $S^+$ and $S^-$ operations and the excitations are visible via neutron spectroscopy [@Gardner2001]. Figure \[fig:temp\] shows the dispersion relations of the three lowest-lying magnetic excitations along $(h,h,h)$ at 12 K and 30 K at the same energy scale. The lowest band becomes more dispersive as the temperature is lowered, while the higher two bands do not show much change with temperature.
Fluctuations between the doublets are important in understanding the experimentally observed scattering pattern. These fluctuations originate in the first term of Eq. , where the anisotropy gap $\Delta=E_1-E_0$ enters the formulation algebraically, instead of exponentially as in the elastic (second) term through the thermal occupation fraction. If the two-ion interaction, characterized by $\theta_{\rm CW}$, is comparable with $\Delta$, then the effect of this term is large and part of the isotropic response is restored. By setting $\Delta\rightarrow\infty$ or the matrix element between two doublets to zero, this term is eliminated, and the scattering pattern is reduced to that of the $\langle 111 \rangle $ Ising model [@Enjalran]. This explains why the $\langle 111 \rangle$ Ising model works so well in describing the spin correlations in spin ice materials [@denHertog2000; @Bramwell2001], since in these systems $\Delta \approx 200\sim300$ K [@Rosenkranz2000; @RosenkranzPrivate] and $\theta_{\rm CW}\approx 0.5\sim2$ K [@denHertog2000; @Bramwell2001; @Ramirez1999], so the effects of the fluctuations out of the ground state doublet are negligible. To understand the different temperature dependencies of the dispersion for the three lowest-lying magnetic excitations, we inspect the thermal factor $n_{\nu\mu}=n_\nu-n_\mu$ in the first term of Eq. . For temperatures much lower than $\Delta$, only the ground state doublet is populated, and this factor is almost the same for all three bands. When the temperature reaches a value $T \approx \Delta$, the first excited state doublet begins to be populated, therefore reducing the factor $n_{01}=n_0-n_1$, while those for the two higher levels remain essentially unchanged. This change in thermal occupation factor results in a large temperature dependence of the dispersion in the lowest band. The low temperature ground state of our model in RPA is the noncollinear “all-in” or “all-out” $\mathbf{Q}=0$ Néel state [@denHertog2000] with $T_c\approx 1.8~$K. Exact $T_c$ value depends on the details of the crystal field wave functions. Suprisingly, this value is very close to the $T_c$ obtained recently in hydrostatic pressure measurements for pressure greater than 1 GPa [@Mirebeau2002]. By standard Holstein-Primakoff expansion aroud this ground state we obtain the spin-wave excitation spectrum with a large gap and the reduction of the staggered magnetization due to quantum fluctuations $\Delta m/m < 10^{-4}$[@Adrian]. This result should be taken only as an indicator, though, since the ground state doublet ($M_J=\pm4$) is not at its saturation value $M_J=\pm6$ due to the crystal field interaction. The magnetic ground state doublet is a non-Kramers doublet, and the states are time-conjugate of each other, so there exist no matrix elements between them [@Abragam], indicating that any quantum fluctuations causing spins to flip result from a higher-order virtual process via excited crystal field levels.
In conclusion, with a simple level scheme and the RPA to include the fluctuations between the ground state and the first excited state doublets, we are able to describe semi-quantitatively the experimentally observed neutron scattering pattern and the energy dispersion in Tb$_2$Ti$_2$O$_7$. Our results indicate that the crystal field effects are important in this system due to the fact that the anisotropy gap and the two-ion interaction are comparable, and that some isotropy is present in the response. The ground state of our model, however, is still the “all-in” or “all-out” $\mathbf{Q}=0$ Néel state [@denHertog2000] with $T_c\approx 1.8~$K, which is stable against the lowest order quantum fluctuations. These results indicate that a mechanism that may restore more isotropy and thus increase the quantum fluctuations leading to the suppression of $T_c$, requires further detailed study.
We thank B. Buyers, B. Canals, B. Gaulin, J. Gardner and S. Rosenkranz for useful and stimulating discussions. This work is supported by the NSERC of Canada, Research Corporation and the Province of Ontario.
[29]{} natexlab\#1[\#1]{}bibnamefont \#1[\#1]{}bibfnamefont \#1[\#1]{}citenamefont \#1[\#1]{}url \#1[`#1`]{}urlprefix\[2\][\#2]{} \[2\]\[\][[\#2](#2)]{}
, ****, ().
, ****, ().
, ****, ().
, ****, (), .
, .
, ****, (), .
, ****, ().
, ****, ().
, ****, ().
, ****, ().
, ****, ().
, ****, ().
, ****, ().
, ****, ().
, ****, ().
, ****, ().
, .
, ****, ().
, ****, ().
, ****, ().
, .
, ****, ().
, in **, edited by (, , ), vol. .
, ** (, ).
, ** (, , ), p. .
, ****, ().
, .
, ** (, , ).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Braitenberg vehicles are well known qualitative models of sensor driven animal source seeking (biological taxes) that locally navigate a stimulus function. These models ultimately depend on the perceived stimulus values, while there is biological evidence that animals also use the temporal changes in the stimulus as information source for taxis behaviour. The time evolution of the stimulus values depends on the agent’s (animal or robot) velocity, while simultaneously the velocity is typically the variable to control. This circular dependency appears, for instance, when using optical flow to control the motion of a robot, and it is solved by fixing the forward speed while controlling only the steering rate. This paper presents a new mathematical model of a bio-inspired source seeking controller that includes the rate of change of the stimulus in the velocity control mechanism. The above mentioned circular dependency results in a closed-loop model represented by a set of differential-algebraic equations (DAEs), which can be converted to non-linear ordinary differential equations (ODEs) under some assumptions. Theoretical results of the model analysis show that including a term dependent on the temporal evolution of the stimulus improves the behaviour of the closed-loop system compared to simply using the stimulus values. We illustrate the theoretical results through a set of simulations.'
author:
- 'A. Pequeño-Zurro'
- 'D. Shaikh'
- 'I. Rañó [^1]'
bibliography:
- 'pequeno19sensors.bib'
title: '**Temporal changes in stimulus perception improve bio-inspired source seeking**'
---
Introduction
============
Braitenberg vehicles [@braitenberg84vehicles] are a well known model of animal navigation in robotics. They constitute a set of bio-inspired navigation primitives [@rano14biologically] to move towards or away from stimulus, known in biology as taxes [@fraenkel61orientation]. Whilst in the seminal work of Braitenberg these models defined steering control principles based on direct perception of the stimulus, there is evidence in the literature showing that animals also use the temporal change of the perceived stimulus to drive their movement [@hellwig16rising]. However, the dynamics of the stimulus depends on the movement of the animal, as for the same static stimulus faster speed implies faster perceived temporal changes. In this paper we develop a new mathematical model of Braitenberg vehicle 3a, that achieves target seeking, and includes steering control based on the temporal evolution of the perceived stimulus. Because these wheeled vehicles rely on the unicycle motion model, the resulting closed-loop equations are a set of non-linear differential-algebraic equations, which can be converted (under some conditions) to standard non-linear differential equations. The formal analysis of the model allows to identify the equilibrium points and some stability properties of this special controller, and shows that it generates better trajectories compared to the original Braitenberg 3a vehicle. To the best of our knowledge this work presents the first attempt to analyse theoretically a velocity-based sensor driven navigation mechanism that depends on the velocity of the robot itself. This work has implications for navigation mechanism based on information dependent on the temporal changes like the optical flow or event cameras, where the perception depends on the movement of the robot.
Braitenberg vehicles are typically used in the literature with unconventional sensors (i.e. sensors other than range or positioning sensors) to implement local navigation using, among others; sound (phonotaxis), light (phototaxis) or chemical (chemotaxis) stimuli, but there are also use cases for local navigation using range sensors. The work in [@bernard10phonotaxis] presents a model of the rat’s peripheral and central auditory system that enables the authors to implement phonotaxis through a Braitenberg vehicle 3a to control the motion of a mobile robot. A model of the auditory system of lizards was presented in [@shaikh09braitenberg] [@shaikh16from] to control a mobile robot performing phonotaxis using Braitenberg vehicle 2b. The phonotaxis behaviour of female crickets towards male crickets is mimicked in a series of papers [@webb01spiking] [@horchler04robot] [@reeve05new] that use the principles of Braitenberg vehicles 2a and 3b to design spiking neural networks. Their experimental results show that their robot achieves excellent performance even in outdoor scenarios with the use of whegs (wheeled legs). One of the earlier works in robotic chemotaxis (localisation of an odour source) analysed experimentally the behaviour of vehicles 3a and 3b [@lilienthal04experimental]. While the connection between the sensors and the wheels was linearly proportional, the chemical concentration readings were continuously normalised due to the saturation and dynamic effects of the chemical sensing technology.
The recent development and use of other unconventional sensing modalities has also lead to the use of Braitenberg vehicles and their principles on non-wheeled robots. A fish robot endowed with pressure sensors imitating the fish lateral line used Braitenberg vehicle 2b to implement rheotaxis [@salumae12against], the alignment of the fish body with a current or flow. Compared to other control mechanisms, the deviation of the robot using this bio-inspired mechanism with respect to the flow direction was significantly lower. Moreover, because the movement of the fish-robot affected the measurements of the lateral line, this work constitutes an experimental example of the control mechanism staying stable under movement induced sensing dynamics. Another application of the principles of Braitenberg vehicles to fish robotics is the work presented in [@lebastard12underwater]. In this work, the motion of an electric fish-robot is controlled by the differences between the currents sensed by electrodes located on the sides of the robot. The robot moves towards conductive objects in a pond, avoiding isolating obstacles. Similarly to the previous fish robot, the movement occurs in two dimensions, yet in the case of the electric fish the robot is rigid and controlled through a rod. Another example where Braitenberg vehicles are applied to non-wheeled robots is the work presented in [@rano2018bridging], where a simulated snake robot is controlled to approach a light source by generating modulated undulatory signals in its joints. One type of unconventional sensor that has recently gained traction is the event cameras, which are used in [@milde17obstacle] in combination with a Braitenberg vehicle 2b to implement visual obstacle avoidance in a wheeled robot using neuromorphic hardware.
As we mentioned, Braitenberg vehicles are often used with unconventional sensors, although some instances can be found that use range sensors instead. In [@bicho97dynamic] a Braitenberg vehicles were used for target acquisition of a light source – phototaxis through vehicle 3a –, and infrared sensor based obstacle avoidance – through vehicle 2b. This work shows that proximity sensors can be also used in conjunction with this bio-inspired local navigation technique. Furthermore, this work inspired an implementation of a Braitenberg vehicle based on the readings of a laser scanner to estimate the free area around a robot [@rano14bio]. This paper also proves that the motion of vehicle 2b can lead to chaotic trajectories, a feature exploited in [@rano17biologically] to implement area coverage in a simulated environment with a stimulus similar to the free area around a robot. The chaotic coverage strategy outperforms random walks and Levy walks.
As we saw, Braitenberg vehicles have been used with multiple sensor and different robot types (mainly wheeled robots) but always using the originally proposed instantaneous sensor readings or low pass filtered signals. The main contribution of this paper is to extend this successful control mechanism with a controller dependent on the dynamic changes of the stimulus perceived by the vehicle. Furthermore, we analyse the stability of the resulting controller and show that it is stable and outperforms the original model. This work is inspired by the biological findings supporting the use of rates of change of perception for control and the successful experimental results obtained in simulations motivated by these biological findings [@pequeno18bio]. The rest of the paper is organised as follows. Section \[sec:Model\] reviews the definitions and standard assumptions of Braitenberg vehicle 3a and presents the new control mechanism that includes the temporal change of the stimulus. This section also analyses the stability of the model under some assumptions. Section \[sec:simulations\] presents a series of simulations that illustrate the main theoretical results obtained. Finally, section \[sec:conclusions\] presents some conclusions and future work.
Dynamic model of Braitenberg vehicle 3a {#sec:Model}
=======================================
Let’s assume a Braitenberg vehicle 3a with two sensors and a direct connection between the sensors and the motors as shown in figure \[fig:BV\]. The state of the vehicle in the 2D plane can be represented by the vector ${\mathbf{X}}=[x,y,\theta]$, and we will denote just the Cartesian coordinates by ${\mathbf{x}}=[x,y]$. The selected $x$ and $y$ coordinates correspond to the middle point between the sensors, and we define the orthogonal reference system linked to the front of the vehicle as $\hat{e}=[\cos\theta,\sin\theta]$ and $\hat{e}_p=[-\sin\theta,\cos\theta]$.
![Representation of the qualitative model of Braitenberg vehicle 3a[]{data-label="fig:BV"}](vehicle-3a){width="0.7\columnwidth"}
We assume a smooth positive stimulus $S({\mathbf{x}})$ exists in the plane, i.e. $S({\mathbf{x}})>0$ for all ${\mathbf{x}}\in\Re^2$ and $S({\mathbf{x}})$ is of class $C^\infty$. Without loss of generality we can assume that the maximum of the stimulus occurs at ${\mathbf{x}}={\mathbf{0}}$, i.e. $S({\mathbf{0}})>S({\mathbf{x}})$ for all ${\mathbf{x}}\in\Re^2$. Defining the distance between the sensors is $\delta$, we can express the positions of the right and left sensors as ${\mathbf{x}}_r={\mathbf{x}}-\frac{\delta}{2}\hat{e}_p$ ${\mathbf{x}}_l={\mathbf{x}}+\frac{\delta}{2}\hat{e}_p$ respectively. Therefore, the value of $S({\mathbf{x}})$ measured by these sensors will be $S({\mathbf{x}}_r)$ and $S({\mathbf{x}}_l$. According to the standard model of vehicle 3a [@rano09steering] the velocity of each wheel is a decreasing function $F(s)$ of the stimulus perceived by the sensor located on the same side of the vehicle. We will assume the vehicle only moves forward $F(s)>0$, while $F(s)$ being decreasing implies $F'(s)<0$. Therefore, the velocity of each wheel is in general a non-linear function of the stimulus $v=F(S({\mathbf{x}}))$ (or $v=F({\mathbf{x}})$ for short). Because we want the vehicle to reach the maximum and stop there, $F(s)$ has to be chosen such that $F(S({\mathbf{0}}))=0$. However, in this paper we will consider the case in which the velocity also depends on the temporal rate of change of the stimulus, i.e. $v=F(s,\dot{s})$, and as a first step we will assume that the contribution of the stimulus derivative is additive, which allows is to state the velocities of each wheel as: $$\begin{aligned}
v_r&=&F(S({\mathbf{x}}_r))+G(\dot{S}({\mathbf{x}}_l)) \nonumber\\
v_l&=&F(S({\mathbf{x}}_l))+G(\dot{S}({\mathbf{x}}_r))
\label{eq:wheelsV}\end{aligned}$$ where $G(\dot{s})$ can be a non-linear function with the constraint $G(0)=0$ for the vehicle to stop at the maximum. It is worth noting that the effects on the speed of the temporal derivative of the stimulus is contralateral, i.e. derivative of $S({\mathbf{x}})$ on the right sensor effects the velocity of the left wheel, and vice-versa.
Substituting the expressions for ${\mathbf{x}}_r$ and ${\mathbf{x}}_l$ in equations (\[eq:wheelsV\]), we can state the speeds of the right and left wheels ($v_r$ and $v_l$) as a function of the state of the vehicle (the Cartesian coordinates ${\mathbf{x}}$ and its orientation $\theta$) and the distance between the sensors $\delta$. Furthermore, we can approximate the compound functions $F(S(\cdot))$ and $G(S(\cdot))$ as first order Taylor series around ${\mathbf{x}}$ assuming $\delta$ is small relative to changes in $S({\mathbf{x}})$, and then compute the forward speed of the vehicle, $v=\frac{v_r+v_l}{2}$, and its turning rate $\omega=\frac{v_r-v_l}{d}$, where $d$ is the wheelbase. This leads to the following speeds: $$\begin{aligned}
v&=&F(S({\mathbf{x}}))+G(\nabla S^T\dot{{\mathbf{x}}}) \nonumber\\
\omega&=&-\frac{\delta}{d}\nabla F(S({\mathbf{x}}))^T\hat{e}_p\nonumber\\
&&\quad-\frac{\delta}{d}G'(\nabla S^T\dot{{\mathbf{x}}})
\left[\dot{\theta}\nabla S^T\hat{e}
-\hat{e}_p^THS\dot{{\mathbf{x}}}\right]
\label{eq:VW}\end{aligned}$$ where $\nabla F(S({\mathbf{x}}))$ is the gradient of the compound function $F(S({\mathbf{x}}))$, $\nabla S$ and $HS$ denote the gradient and Hessian matrix of the stimulus respectively, and we used the fact that $\dot{S}({\mathbf{x}})=\nabla S^T\dot{{\mathbf{x}}}$.
Introducing equations (\[eq:VW\]) in the unicycle model we obtain the closed-loop equations of motion of the dynamic Braitenberg vehicle 3a: $$\begin{aligned}
\dot{x}&=&F(S({\mathbf{x}}))\cos\theta+G(\nabla S^T\dot{{\mathbf{x}}})\cos\theta\nonumber\\
\dot{y}&=&F(S({\mathbf{x}}))\sin\theta+G(\nabla S^T\dot{{\mathbf{x}}})\sin\theta\nonumber\\
\dot{\theta}&=&-\frac{\delta}{d}\nabla F(S({\mathbf{x}}))\hat{e}_p\nonumber\\
&&-\frac{\delta}{d}G'(\nabla S^T\dot{{\mathbf{x}}})
\left[\dot{\theta}\nabla S^T\hat{e}
-\hat{e}_pHS\dot{{\mathbf{x}}}\right]
\label{eq:bv3aDyn}\end{aligned}$$
It is worth noting that while the model of Braitenberg vehicle 3a is stated as a system of non-linear ordinary differential equations, equations (\[eq:bv3aDyn\]) correspond to a differential-algebraic system of equations (DAE) as the time derivative of the state $\dot{{\mathbf{X}}}=[\dot{{\mathbf{x}}},\dot{\theta}]$ ($\dot{{\mathbf{x}}}=[\dot{x},\dot{y}]$) is on the r.h.s of the equations. DAEs are much more complex to solve than ODEs, as, for instance, the set of meaningful initial conditions could be restricted. Therefore, we will make some assumptions that will allow us to analyse the closed-loop behaviour analytically.
From Differential Algebraic to Ordinary Differential Equations
--------------------------------------------------------------
Deriving analytic results from equations (\[eq:bv3aDyn\]) can be challenging, however, the DAEs can be converted into a system of ordinary differential equations if we assume that the function $G(\dot{S})$ is linear, i.e $G(\dot{S})=\alpha\dot{S}$, which is the simplest function fulfilling the condition $G(0)=0$. It can be seen that the closed-loop equations can be stated as $\dot{{\mathbf{X}}}=\mathcal{F}({\mathbf{X}})+\alpha A({\mathbf{X}})\dot{{\mathbf{X}}}$, where the derivative of the state on the r.h.s is multiplied by the matrix: $$\begin{aligned}
A&=&
\left[
\begin{array}{ccc}
S_x\cos\theta & S_y\cos\theta &0 \\
S_x\sin\theta & S_y\sin\theta &0 \\
\frac{\delta}{d}\hat{e}_p^THS|_x&
\frac{\delta}{d}\hat{e}_p^THS|_y&
-\frac{\delta}{d}\nabla S^T\hat{e}
\end{array}
\right]\nonumber\end{aligned}$$
where $S_x=\frac{\partial S}{\partial x}$, $S_y=\frac{\partial
S}{\partial y}$, and $\hat{e}_p^THS|_x$ $\hat{e}_p^THS|_y$ represent the $x$ and $y$ components of the row vector $\hat{e}_p^THS$, the result of multiplying $\hat{e}_p^T$ by the Hessian matrix of $S({\mathbf{x}})$. The vector flow $\mathcal{F}({\mathbf{X}})$ is formed from the terms on the r.h.s. of the equation not involving the time derivatives of the state ${\mathbf{X}}$.
The motion equations of the vehicle (\[eq:bv3aDyn\]) can be stated as $\dot{{\mathbf{X}}}=[I-\alpha A({\mathbf{X}})]^{-1}\mathcal{F}({\mathbf{X}})$, which is an ordinary differential equation, assuming the matrix $I-\alpha A$ is not singular. Therefore, the determinant of $I-\alpha A$: $$\begin{aligned}
|I-\alpha A|&=&-\frac{1}{d}\left[\alpha\nabla S^T\hat{e}-1\right]
\left[\alpha\delta\nabla S^T\hat{e} + d\right]\end{aligned}$$ should be different than zero, which provides us with the conditions under which the DAE (\[eq:bv3aDyn\]) can be turned into the ODE $\dot{{\mathbf{X}}}=[I-\alpha
A({\mathbf{X}})]^{-1}\mathcal{F}({\mathbf{X}})$. Specifically, the determinant vanishes when $\nabla S^T\hat{e}=\frac{1}{\alpha}$ or when $\nabla
S^T\hat{e}=-\frac{d}{\delta\alpha}$. Interestingly, if $\delta=d$ both conditions can be stated as $|\nabla S^T\hat{e}|=\frac{1}{\alpha}$ (assuming $\alpha>0$, which will be justified below), i.e. the inverse of $\alpha$ should be different from the directional gradient in the direction of motion along the trajectory of the vehicle. This leads to a constrain for the possible values of $\alpha$, namely $\alpha\ne\frac{1}{\nabla S^T\hat{e}}$. If this condition is fulfilled, the closed-loop motion equations corresponding to the dynamic Braitenberg vehicle 3a are: $$\begin{aligned}
\dot{x}&=&\frac{F(S({\mathbf{x}}))\cos\theta}{1-\alpha\nabla S^T\hat{e}}\nonumber\\
\dot{y}&=&\frac{F(S({\mathbf{x}}))\sin\theta}{1-\alpha\nabla S^T\hat{e}}\nonumber\\
\dot{\theta}&=&-\frac{\delta\nabla F^T\hat{e}_p}
{(d+\alpha\delta\nabla S^T\hat{e})}\nonumber\\
&&-
\frac{\alpha\delta F(S({\mathbf{x}}))\hat{e}_p^THS\hat{e}}
{(\alpha\nabla S^T\hat{e}-1)(d+\alpha\delta\nabla S^T\hat{e})}
\label{eq:Model}\end{aligned}$$
These equations are similar to the closed-loop model of Braitenberg vehicle 3a [@rano09steering], in fact the original model can be obtained by simply setting $\alpha=0$. Moreover, it can be shown that they have an equilibrium point at the origin – the maximum of $S({\mathbf{x}})$ – since $\nabla S({\mathbf{0}})={\mathbf{0}}$ and $F(S({\mathbf{0}}))=0$, which implies $\dot{x}=0$, $\dot{y}=0$ and $\dot{\theta}=0$. The analysis of the stability of the equilibrium point in the case of Braitenberg vehicle 3a is presented elsewhere, but for the case at hand the stability properties could change due to the additional term in the equation for $\dot{\theta}$.
The condition on the determinant of $I-\alpha A({\mathbf{X}})$ for the inverse matrix to exist is also reflected in the equations, as when $|\nabla S|=-\frac{1}{\alpha}$ or $|\nabla S|=-\frac{d}{\alpha\delta}$ the denominators of the three dynamic equations would go to zero generating possibly unbounded speeds. So the parameter $\alpha$ has to be chosen carefully considering the possible values of the norm of the stimulus gradient. While the equation for the angular variable $\dot{\theta}$ is hard to analyse at first sight, the equations for $\dot{x}$ and $\dot{y}$ provide a new insight on the effect of the dynamic controller over the closed-loop behaviour. First it is worth noting that the forward speed of the vehicle is $v=\frac{F(S({\mathbf{x}}))}{1-\alpha\nabla S^T\hat{e}}$, where $\hat{e}$ is the vector pointing in the direction of the vehicle and $\nabla S$ points in the direction of increasing stimulus. If the vehicle points exactly in the direction of $\nabla S$, the dot product of the two vectors corresponds to the norm of the gradient, i.e. $\nabla
S\hat{e}=|\nabla S|$, and if it points in the direction opposite to the stimulus gradient it will be $\nabla S\hat{e}=-|\nabla S|$. Now let’s assume $\alpha>0$ and define $\nabla S^*$ as the gradient with the largest norm in the environment or the domain where $S({\mathbf{x}})$ is defined. If we impose the additional condition on $\alpha$ $0<\alpha<\frac{1}{|\nabla S^*|}$ the term $\alpha\nabla S^T\hat{e}$ will be positive and smaller than one when the vehicle points in the direction of increasing $S({\mathbf{x}})$. The overall effect compared to the Braitenberg vehicle 3a is to increase the forward speed since the denominator of $v$ becomes smaller than one. On the other hand, if the vehicle points in the direction opposite to the gradient, the dot product $\nabla S^T\hat{e}$ will be negative, making the denominator of the forward speed larger than one, which effectively reduces the forward speed. This is a highly interesting effect of introducing the temporal rate of change of the stimulus in the velocity control of the vehicle, as it makes the vehicle to move faster when it faces the source, while reduces its speed when the source is on its back. Analysing the behaviour of the the equation for $\dot{\theta}$ is not trivial since in the dynamic case there is a new term which could change the sign of the original rotation speed (for which stability conditions can be derived [@rano14results]). In the next section we will take a closer look at the stability of the system under the additional assumption of parabolically symmetric stimulus.
Motion under parabolic stimulus {#sec:parab}
-------------------------------
In this section we will assume the stimulus has parabolic symmetry of the form $S({\mathbf{x}})=S({\mathbf{x}}^TD{\mathbf{x}})$ where $D=diag(\sigma_x,\sigma_y)$ is a positive definite diagonal matrix. This will allows us to analyse the stability and derive more results on the trajectories of the closed-loop controller. The assumption holds for any smooth stimulus close enough to the source. Moreover, since the reference system can be defined arbitrarily we can rotate the $x$ and $y$ axis to be aligned with the principal axis of the stimulus. Under these assumptions the gradient of the stimulus can be written as $\nabla
S=2S'({\mathbf{x}}^TD{\mathbf{x}})D{\mathbf{x}}$, where $S'(\cdot)$ is the derivative of $S(\cdot)$ w.r.t its argument, with $S'(\cdot)<0$ because the stimulus to grows towards the source (located at ${\mathbf{x}}={\mathbf{0}}$). The Hessian matrix of this stimulus is $HS({\mathbf{x}})=4S''({\mathbf{x}}^TD{\mathbf{x}})D{\mathbf{x}}{\mathbf{x}}^TD+2S'({\mathbf{x}}^TD{\mathbf{x}})D$ which is negative definite at the origin, as the origin is a maximum. Using the additional condition on the control function $F(S({\mathbf{0}}))=0$ and evaluating $\dot{x}$, $\dot{y}$ and $\dot{\theta}$, eq. (\[eq:Model\]) at the stimulus maximum we obtain the following condition for an equilibrium point to appear $\hat{e}_p^TD\hat{e}=0$, which is satisfied for all $\theta$ if $\sigma_x=\sigma_y$, or for $\theta=0$, $\theta=\pi$ and $\theta=\pm\pi/2$ if $\sigma_x\ne\sigma_y$. So the closed-loop system has infinite equilibrium points for circularly symmetric stimuli ($\sigma_x=\sigma_y$), or four equilibrium points for parabolically symmetric stimuli $\sigma_x\ne\sigma_y$.
From the ODE form of the closed-loop equations we can also find an analytic solution to the equations. Just like the standard Braitenberg vehicle, when the initial pose of the vehicle falls on, and is aligned with, the main axis of the stimulus (lines $x=0$ and $y=0$ with heading $\theta=0$ or $\theta=\pi$ and $\theta=\pm\frac{\pi}{2}$ respectively) the vehicle follows a straight line trajectory towards (or away from) the maximum of the stimulus. Taking, for instance $x(0)=-x_0$, $y(0)=0$ and $\theta(0)=0$ as a starting pose we get for the $y$ coordinate that $\dot{y}=0$ and for the $x$ coordinate: $$\begin{aligned}
\dot{x}&=&\frac{F(S(\sigma_xx^2))}{1-2\alpha S'(\sigma_xx^2)\sigma_xx}
\label{eq:dotX}\end{aligned}$$ while the equation of $\theta$ becomes $\dot{\theta}=0$ since $\nabla
F$ is perpendicular to $\hat{e}_p$ and $\hat{e}_pHS\hat{e}=0$ (for $\theta=0$ $\hat{e}=[1,0]$ and $\hat{e}_p=[0,1]$). Therefore, the trajectory is indeed a straight line with the evolution of $x$ given by equation (\[eq:dotX\]).
Having this particular solution to the system of non-linear equations (\[eq:Model\]) allows us to linerarise the system around that trajectory and to check the evolution of the nearby trajectories with increments $\Delta{\mathbf{X}}$ following $\Delta{\mathbf{\dot{X}}}=M(t)\Delta{\mathbf{X}}$. The linearised system is a time varying linear system, and the eigenvalues of the matrix $M(t)$ – which depend on the solution of equation (\[eq:dotX\]) – provide information on the stability of the non-linear system, i.e. first Lyapunov criterion applied to a non-constant trajectory. Because the linearised system is three dimensional, at least one eigenvalue will be real, while the other two could be complex conjugates, which is the case when $\alpha=0$, i.e. the standard Braitenberg vehicle under parabolically symmetric stimuli [@rano14results]. If we denote $F=F(S(\sigma_x x^2))$, $F'=F'(S(\sigma_x x^2))$, $S'=S'(\sigma_x
x^2)$, $S''=S''(\sigma_x x^2)$, $\Delta_1=1-2\alpha\sigma_xxS'$ and $\Delta_2=d-2\alpha\delta\sigma_xxS'$ the real eigenvalue of the matrix $M(t)$ can be written as: $$\begin{aligned}
\lambda_1&=&\frac{2\sigma_xxF'S'}{\Delta_1}
+\frac{2\alpha\sigma_xF[S'+2\sigma_xx^2S'']}{\Delta_1^2} \nonumber \end{aligned}$$ while the real part of the complex conjugate eigenvalues is: $$\begin{aligned}
Re[\lambda_{2,3}]&=& \frac{\delta\sigma_xxF'S'}{\Delta_2}
+ \frac{\alpha\delta F[S'[\sigma_x-\sigma_y]+2S''\sigma_x^2x^2]}
{\Delta_1\Delta_2} \nonumber \end{aligned}$$ where $\sigma_x>0$, $x<0$, $F>0$, $F'<0$ and $S'<0$. It can be seen that a sufficient condition for the real the eigenvalue to be negative is $S'+2\sigma_xx^2S''<0$, but $\lambda_1$ could be negative even if this condition is not met. Moreover, because $0<\Delta_1<1$ from the condition imposed on $\alpha$, $\lambda_1$ will be more negative than the corresponding eigenvalue for the standard Braitenberg vehicle, which implies a faster convergence rate towards the straight line solution. While the first term of the real part of the eigenvalues $\lambda_{2/3}$ can be seen is negative, a similar analysis leads to the necessary condition $S'(\sigma_x-\sigma_y)+2S''\sigma_x^2x^2<0$. As $x$ approaches zero from the negative starting point in the linear solution of equation (\[eq:dotX\]), the condition becomes $S'(\sigma_x-\sigma_y)<0$.
Simulations {#sec:simulations}
===========
This section presents simulations to illustrate the theoretical results on the proposed controller mechanism presented in the previous section. Specifically we first show the accuracy of the model, to then illustrate the improved stability when including a component with the time derivative of the stimulus. It is worth noting that simulating the differential-algebraic equations requires special integration algorithms and consistency checks of the initial conditions. Specifically, the algorithm used in the simulations below is the one presented in [@shampine02solving]. All the results assume a parabolically shaped stimulus with a maximum at the origin and linear control function $F(s)$ to generate an equilibrium point at ${\mathbf{x}}={\mathbf{0}}$.
Model evaluation
----------------
To evaluate the accuracy of the model, eqs. (\[eq:bv3aDyn\]), of the dynamic Braitenberg vehicle – which uses the values on the middle points between the sensors –, we simulated it along with equations (\[eq:wheelsV\]) to compute the trajectory. Because equations (\[eq:bv3aDyn\]) are obtained after first order truncation of the Taylor series for $F(S({\mathbf{x}}))$ and $G(\dot{S}({\mathbf{x}}))$ in equations (\[eq:wheelsV\]), we can expect results of both simulations to differ slightly. Figure \[fig:1\] shows the evolution over time of the $x$, $y$ (top row) and $\theta$ coordinates, together with the trajectory on the $x-y$ plane (bottom) for $\delta=d=0.25$. The initial pose for both simulations was $x=-6$, $y=1$, and $\theta=0$, and as the figures show than trajectories are very similar. It is worth noting that the approximation error increases with the distance between the sensors ($\delta=0.25$).
![Simulation to evaluate the model accuracy with starting poses $x=-6$, $y=1$ and $\theta=0$[]{data-label="fig:1"}](Fig1){width="\columnwidth"}
Forward velocity increase towards the source
--------------------------------------------
To illustrate the increase of the speed when the vehicle points towards the source – prediction from equation (\[eq:Model\]) – we simulated from the same initial position the standard Braitenberg vehicle 3a and our new model version, eq. (\[eq:bv3aDyn\]). Specifically, the initial pose in both simulations was $x=-6$, $y=0$ and $\theta=0$, which leads to a straight line trajectory ($\dot{y}=0$ and $\dot{\theta}=0$) for our stimulus. Figure \[fig:2\] shows the time evolution of the $x$ coordinate, which corresponds to the solution of equation (\[eq:dotX\]). As we can see from the figure the version including the dynamics (with $\alpha=3$) approaches the origin faster than the standard Braitenberg vehicle 3a. We also concluded from the analysis of equation (\[eq:Model\]) that the forward speed in the dynamic version is reduced when the vehicle points away from the source. This effect will be shown in the next section together with the local stability of the trajectories around the straight line trajectory.
![Trajectory over time of vehicles 3a and the proposed dynamic model with starting poses $x=-6$, $y=0$ and $\theta=0$[]{data-label="fig:2"}](Fig2){width="0.9\columnwidth"}
Stability close to the analytical trajectory
--------------------------------------------
As we saw in section \[sec:parab\] the real part of the eigenvalues of the linearised system can get more negative for the Braitenberg vehicle with the stimulus derivative controller. That means the convergence of trajectories towards the straight line solution is faster. To illustrate this effect we simulated the Braitenberg vehicle 3a with and without the dynamic contribution for a starting position close to the analytic straight line solution. Figure \[fig:3a\] shows the resulting trajectories with starting pose $x=-6$, $y=1$ and $\theta=0$. Since the stimulus is not circularly symmetric, oscillations appear (see [@rano14results]), i.e. two eigenvalues are complex conjugates, but the amplitude of the oscillations is smaller for the dynamic version due to the exponential term (real part of the eigenvalues) multiplying the oscillatory functions. The plot also shows that the trajectory of the dynamic version of the vehicle is shorter and seems to oscillate less around the linearisation trajectory, which seems to indicate that the imaginary part of the eigenvalues is also smaller.
![Simulation of the standard Braitenberg vehicle 3a with and without dynamic component with starting poses $x=-6$, $y=1$ and $\theta=0$[]{data-label="fig:3a"}](Fig3a){width="0.97\columnwidth"}
Finally, we ran a simulation to illustrate the effect on the reduction of the forward speed when the vehicle points in the opposite direction to the source, i.e. when $\nabla S^T\hat{e}<0$. According to equations (\[eq:Model\]) the speed should decrease compared to the standard vehicle 3a while the vehicle moves opposite to the source. Once it heads the direction of the gradient within $\pm\frac{\pi}{2}$ radians the forward speed increases. Figure \[fig:3b\] illustrates this effect by showing the trajectories of the vehicles starting with pose $x=-2$, $y=1$ and $\theta=\arctan-\frac{1}{2}$, i.e. pointing directly away from the source. As we can see the movement of the vehicle greatly improves when the controller includes the time derivative of the stimulus leading to a sharper turn and a shorter trajectory. Although in this case the initial pose of the vehicle is not close to the straight line trajectory, the vehicle firts turns towards the source and the approaches the horizontal axis where the equations of the linearised system are valid.
![Simulation of the standard Braitenberg vehicle 3a with and without dynamic component with starting poses $x=-2$, $y=1$ and $\theta=\arctan-\frac{1}{2}$[]{data-label="fig:3b"}](Fig3b){width="0.97\columnwidth"}
Conclusions and Future Work {#sec:conclusions}
===========================
This paper proposes a new bio-inspired local navigation strategy that extends the well known Braitenberg models of animal positive taxes, while accounting for the experimental findings in biology showing that animals use the time evolution of the perceived stimulus to control their movement. We analysed the mathematical model of the dynamic Braitenberg vehicle and showed that the convergence towards the stimulus source is faster. This work has important implications in robotics as it shows the improvement control techniques based on optical flow or event based cameras – when the perception is dependent on the speed of the robot – can bring. Although the circular dependency between the controlled variables and the perceived variables makes the closed-loop system difficult to analyse, our simplified setup enables applying analytical tools to show that there is a clear improvement in the stability of the system.
One underlying assumptions to make the closed-loop system tractable is the absence of sensor noise or a high signal to noise ratio. However, biological systems are inherently noisy, and the time derivative of a perceptual signal with additive noise cannot be used directly to control a robot motion. It would require instead a band-pass filter. Therefore, our next step will focus on trying to analyse the behaviour of this new controller using a stochastic model of Braitenberg vehicles [@rano17drift].
[^1]: A. Pequeño-Zurro, D. Shaikh, and I. Rañó are with the Embodied Systems for Robotics and Learning Unit, Mærsk McKinney Møller Institute, University of Southern Denmark, Denmark. e-mail: [{alz, danish, igra}@mmmi.sdu.dk]{}
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We propose $q$-deformation of the Gamayun-Iorgov-Lisovyy formula for Painlevé $\tau$ function. Namely we propose the formula for $\tau$ function for $q$-difference Painlevé equation corresponding to $A_7^{(1)\prime}$ surface (and $A_1^{(1)}$ symmetry) in Sakai classification. In this formula $\tau$ function equals the series of $q$-Virasoro Whittaker conformal blocks (equivalently Nekrasov partition functions for pure $SU(2)$ 5d theory).'
author:
- 'M. A. Bershtein, A. I. Shchechkin'
title: '$q$-deformed Painlevé $\tau$ function and $q$-deformed conformal blocks'
---
Introduction {#sec:intro}
============
The goal of this paper is to find $q$-deformation of suggested in [@GIL1207],[@GIL1302] formulas for Painlevé $\tau$ functions. More precisely we will do this for case of Painlevé III($D_8$).
We briefly recall necessary details about this equation. In standard form it is a nonlinear second order differential equation on function $w(z)$, namely $$w''=\frac{w'^2}{w}-\frac{w'}{z}+\frac{2w^2}{z^2}-\frac{2}{z}
\label{Pp}$$ This equation can be rewritten as system of two bilinear Toda-like equations on two $\tau$ functions [@BSRamon] $$1/2 D^2_{[\log z]}(\tau(z),\tau(z))=-z^{1/2}\tau_1(z)\tau_1(z),\quad 1/2 D^2_{[\log z]}(\tau_1(z),\tau_1(z))=-z^{1/2}\tau(z)\tau(z),$$ where $D^2_{[\log z]}$ denotes second Hirota operator with respect to $\log z$, and $\tau_1$ is a Bäcklund transformation of $\tau$ (the group of Bäcklund transformations of this equation is $\mathbb{Z}_2$). The function $w(z)$ is equal to $z^{1/2}\tau(z)^2/\tau_1(z)^2$, Bäcklund transformation acts on $w$ as $w \mapsto z/w$.
The Gamayun–Iorgov–Lisovyy formula for the $\tau$ function has the form [@GIL1302] $$\tau(\sigma,s|z)=\sum_{n\in\mathbb{Z}}C(\sigma+n)s^n z^{(\sg+n)^2} \mathcal{F}((\sigma+n)^2|z), \label{GIL}$$ where $s,\sigma$ are integration constants. The function $\mathcal{F}(\Delta|z)$ denotes Whittaker limit of Virasoro conformal block in a representation of the highest weight $\Delta$ and the central charge $c=1$. The function $C(\sigma)=1/(\G(1-2\sigma)\G(1+2\sigma))$, where $\mathsf{G}$ is a Barnes $\mathsf{G}$ function. Function $\tau(\sigma,s|z)$ satisfies evident property $\tau(\sigma,s|z)=s^{-1}\tau(\sigma+1,s|z)$ (periodicity in $\sigma$) and Bäcklund transformation acts as $\tau_1 \propto \tau(\sigma+1/2,s|z)$. Therefore bilinear equations and formula for $w$ can be rewritten as (see [@BSRamon] for details) $$\label{sgToda}
1/2 D^2_{[\log z]}(\tau(\sigma,s|z),\tau(\sigma,s|z))=-z^{1/2}\tau(\sigma+1/2,s|z)\tau(\sigma-1/2,s|z),$$ $$w(z)=z^{1/2}\frac{\tau(\sigma,s|z)^2}{\tau(\sigma-1/2,s|z)\tau(\sigma+1/2,s|z)}\label{tautau1}$$ The formula was proven in [@ILT] and [@KGP]. Remark that the proof in [@KGP] is based on bilinear relations on Virasoro conformal blocks.
It is natural to expect that there exists certain $q$-deformation of formula which gives $\tau$ function for $q$-difference Painlevé equation. We follow Sakai approach to these equations [@SakaiCMP], where discrete equations are associated with certain rational surfaces which are obtained by blow up of 9 points in $\mathbb{CP}^2$. These surfaces are parametrized by two complementary sublattices in Picard group of the surface, the last lattice is isomorphic to $\mathbb{Z}^{1,9}$. These two sublattices are denoted as $R$ and $R^\perp$ and specify so-called surface type and symmetry type correspondingly. The celebrated Sakai tables are given in Figs. \[Fig:Sakaisurf\] and \[Fig:Sakaisym\].
& & & & & & & [A\_7\^[(1)]{}]{} &\
[A\_0\^[(1)]{}]{} & [A\_1\^[(1)]{}]{} & [A\_2\^[(1)]{}]{} & [A\_3\^[(1)]{}]{} & [A\_4\^[(1)]{}]{} & [A\_5\^[(1)]{}]{} & [A\_6\^[(1)]{}]{} & & [A\_8\^[(1)]{}]{}\
& & & & D\_4\^[(1)]{} & D\_5\^[(1)]{} & D\_6\^[(1)]{} & D\_7\^[(1)]{} & D\_8\^[(1)]{}\
& & & & & & E\_6\^[(1)]{}& E\_7\^[(1)]{}& E\_8\^[(1)]{}
& & & & & & &
A\_1\^[(1)]{}\
||\^2=8
&\
[E\_8\^[(1)]{}]{} & [E\_7\^[(1)]{}]{} & [E\_6\^[(1)]{}]{} & [E\_5\^[(1)]{}]{} & [E\_4\^[(1)]{}]{} & [E\_3\^[(1)]{}]{} & [E\_2\^[(1)]{}]{} & & [A\_0\^[(1)]{}]{}\
& & & & D\_4\^[(1)]{} & D\_3\^[(1)]{} & D\_2\^[(1)]{} &
A\_1\^[(1)]{}\
||\^2=4
& A\_0\^[(1)]{}\
& & & & & & A\_2\^[(1)]{} & A\_1\^[(1)]{}& A\_0\^[(1)]{}
Here $X_n^{(1)}$ are standard notations for affine root lattices. We do not explain precise meaning of notations like $D_2^{(1)}$ or $A_1^{(1)}, |\alpha|^2=4$, see [@SakaiCMP] or [@KNY] for details. Arrows in these figures indicate degenerations of the surfaces.
Recall that continuous Painlevé equations also correspond to the rational surfaces, namely for each Painlevé equation there exists so-called space of initial conditions [@Okamoto]. This space for is the surface of $D_8^{(1)}$ type. This is the reason why this equation is called Painlevé III($D_8$). The $q$-deformation of this equation corresponds to the deformation of the surface (see [@SakaiCMP Sec 7]). Therefore we consider $q$-difference Painlevé equation corresponding to the surface type $A_7^{(1)\prime}$ and the symmetry type $A_1^{(1)}$ (the corresponding sublattices are drawn in box in Figs. \[Fig:Sakaisurf\], \[Fig:Sakaisym\]).
Such surface depends on two parameters which we denote by $q$ and $Z$ ($q$-deformed analogue of $z$ in ). Details are given Subsection \[ssec:Cremona\], here we write only $q$-deformation of the equation $$G(qZ)G(q^{-1}Z)=\left(\frac{G(Z)-Z}{G(Z)-1}\right)^2$$ This equation as well as its relation to Painlevé III$(D_8)$ is given in [@Ramani:2015]. We did not found in the literature the definition of $\tau$ function for this difference equation. So we propose the definition following Tsuda approach [@Tsuda2006] (see also [@KNY]). It is convenient to use four $\tau$ functions $\mathcal{T}_1, \mathcal{T}_2, \mathcal{T}_3, \mathcal{T}_4$. This result is given in Theorem \[Th:tauformal\].
The next task is to give a formula for $\mathcal{T}_i$ as an analytical function on variable $Z \in \mathbb{C}$ (above one can think that $\mathcal{T}_i$ is defined on the lattice as in [@Tsuda2006]). As in continuous case the conditions on the $\tau$ functions can be rewritten as a system of equations on one function $\mathcal{T}(u,s;q|Z)$ $$\begin{aligned}
Z^{1/4}\mathcal{T}(u,s;q|qZ)\mathcal{T}(u,s;q|q^{-1}Z)&=\mathcal{T}(u,s;q|Z)^2+Z^{1/2}\mathcal{T}(uq,s;q|Z)\mathcal{T}(uq^{-1},s;q|Z)\\
\mathcal{T}(uq^2,s;q|Z)=s^{-1}\mathcal{T}(u,s;q|Z) .
\end{aligned} \label{eq:desTau}$$ Here $u, s$ are $q$-deformed analogs of parameters $\sigma,s$ in .
In Section \[sec:deform\] we propose the solution of this system in the ansatz $$\mathcal{T}(u,s;q|Z)=\sum_{n \in \mathbb{Z}}s^n C(uq^{2n};q|Z) \frac{\mathcal{F}(uq^{2n};q^{-1},q|Z)}{(uq^{2n+1};q,q)_\infty(u^{-1}q^{-2n+1};q,q)_\infty}
\label{eq:T}$$ Here $\mathcal{F}(u;q^{-1},q|Z)$ denotes Whittaker limit of the $q$-deformed conformal block, $(u;q,q)_\infty$ denotes double infinite Pochhammer symbol and the function $C(u;q|Z)$ is defined by its second difference derivatives, see Definition \[Def:tau\]. The function $C(u;q|Z)$ is $q$-deformed analogue of $z^{\sigma^2}$ in formula . We give a couple of examples of such functions in Example \[Ex:C\].
In Subsection \[ssec:qdeftau\] we discuss validity of for function $\mathcal{T}(z)$ defined in . The second condition in is straightforward from . The bilinear relation in is equivalent to the bilinear relation on function $\mathcal{F}(u;q^{-1},q|Z)$, see and Theorem \[mainbilin\]. We do not prove corresponding bilinear relation but give several arguments in support.
In order to finish description of the paper content note that in Subsections \[ssec:contlimeq\] and \[ssec:contlim\] we discuss $q \rightarrow 1$ limit, in Subsection \[ssec:algsol\] we discuss special values of $u,s$ which correspond to $q$-deformation of algebraic solution of . In Appendix \[app:q\] we collect necessary definitions and properties of $q$-special functions. In Appendix \[app:toralg\] we write more general bilinear relations on $q$-deformed conformal blocks.
$q$-deformed equations {#sec:repr}
======================
$A_7^{(1)\prime}$ surface and its symmetry group {#ssec:Cremona}
------------------------------------------------
Here we follow Sakai approach to the $q$-difference Painlevé equations associated with rational surfaces [@SakaiCMP], [@KNY]. In this approach $q$-deformed Painlevé equation arises from discrete group of birational automorphisms acting on a rational surface.
Let $X$ be a rational surface of the type $A_7^{(1)\prime}/A_1^{(1)}$ in Sakai notations [@SakaiCMP] where first notation denotes surface type and second notation denotes symmetry type of the surface. It is obtained by blow-up of $\mathbb{CP}^2$ in nine points. Scheme of this procedure one can see in Fig. \[blowup\] (see [@SakaiCMP App. B, Mul. 9]).
(-1,-0.5) – (0.5, 2) node\[anchor=west\] [$x=0$]{}; (-1.4, -0.1) – (2,-0.1) node\[anchor=west\] [$y=0$]{}; (1.6,-0.5) – (0.1,2) node\[anchor=east\] [$z=0$]{}; (-0.25,0.75) circle (0.7mm) node\[anchor=east\] [$p_4$]{}; (0.3,1.67) circle (1.5mm); (0.5,1.8) node\[below right\] [$p_7\leftarrow p_8 \leftarrow p_9$]{}; (1.36,-0.1) circle (0.7mm); (1.36,-0.1) circle (1.5mm); (1.36,0) node\[above right\] [$p_2\leftarrow p_6$]{}; (-0.76,-0.1) circle (0.7mm); (-0.76,-0.1) circle (1.5mm); (-1,-0.4) node\[anchor=east\] [$p_5\rightarrow p_3\rightarrow p_1$]{};
Let $(x:y:z)$ denote coordinates of $\mathbb{CP}^2$, $(a,b,c)$ are parameters of surface which characterize positions of points of blow-up (our $a$ is $a_1$ in [@SakaiCMP]). There is an equivalence $(a,b,c;x:y:z)\sim (a,\mu^2 b,\mu^{-2}c;\mu x:y:z)$. There exists discrete subgroup of Cremona group (group of rational transformations of $\mathbb{CP}^2$) which preserve blow-up structure. We will denote it by $W$. Group $W$ is equal to $Dih_4\ltimes W(A^{(1)}_1)$ where $Dih_4$ is dihedral group of square and $W(A^{(1)}_1)$ is Weyl group of $\widehat{sl}(2)$. Group $W$ can be presented by generators $s_0, s_1, \pi_1, \pi_2$ and relations as $$\label{eq:Crem:rel}
s_0^2=s_1^2=1,\quad \pi_1^2=\pi_2^4=(\pi_1\pi_2)^2=1,\quad s_1=\pi_2s_0\pi_2^{-1},\quad s_0=\pi_2^2s_0\pi_2^{-2}=\pi_1s_0\pi_1^{-1},$$ where $s_0,s_1$ are Weyl group simple reflections, $\pi_1, \pi_2$ are generators of $Dih_4$. Structure of $W$ could be graphically represented on Fig. \[dihedral\]. First two pictures define action of the generators of dihedral group on the square and the last picture describes structure of the semidirect product, namely $\pi_2$ acts as the outer automorphism which interchange roots.
(6,1) node \[anchor=east\] [$\alpha_0$]{} edge\[pil,bend right=0\] (5.999,1); (8,1) node \[anchor=west\] [$\alpha_1$]{} edge\[pil,bend right=0\] (8.001,1); (6.1,1.02) – (7.9,1.02); (6.1,0.98) – (7.9,0.98); (6.2,1.3) edge\[pil,<->,bend left=20\] (7.8,1.3); (7,1.7) node [$\pi_2$]{};
(-4,0) node \[below left\] [$4$]{} – (-2,0) node \[below right\] [$3$]{} – (-2,2) node \[above right\] [$2$]{} – (-4,2) node \[above left\] [$1$]{} – cycle; (-4.4, 2.4) – (-1.2,-0.8) node\[below right\] [$\pi_1$]{}; (-1.7,-0.8) edge\[pil,<-> ,bend right=45\] (-1.25,-0.2);
(2,0) node \[below left\] [$4$]{} – (4,0) node \[below right\] [$3$]{} – (4,2) node \[above right\] [$2$]{} – (2,2) node \[above left\] [$1$]{} – cycle;
(3,1) node [$\pi_2$]{}; (4.3,0.2) edge\[pil, bend right=20\] (4.3,1.8);
(3.8,2.3) edge\[pil, bend right=20\] (2.2,2.3);
(1.7,1.8) edge\[pil, bend right=20\] (1.7,0.2);
(2.2,-0.3) edge\[pil, bend right=20\] (3.8,-0.3);
As a subgroup of Cremona group $W$ acts on rational surfaces by rational transformations[^1] $$\begin{aligned}
\pi_1&\colon (a,b,c;x:y:z) \mapsto (1/a,1/c,1/b;x:z:y)
\\
\pi_2&\colon (a,b,c;x:y:z) \mapsto (bc,a/b,b;xy:byz:x^2)\\\
s_1&\colon (a,b,c;x:y:z) \mapsto (1/a,ab,ac;x(y-az):y(z-y):az(z-y))\\
s_0&\colon (a,b,c;x:y:z) \mapsto (ab^2c^2,1/c,1/b;x(x^2/b-yz)(yz-cx^2):y(yz-cx^2)^2:z(x^2/b-yz)^2)\label{Cract}
\end{aligned}$$
Let us denote $q=abc$, $Z=a^{-1}$ and $F=yzb/x^2$, $G=z/y$. These coordinates are unambiguously defined (taking into account equivalence of $\mu$ rescaling). Then, in terms of $(Z,q,F,G)$ action of group $W$ has the form: $$\label{eq:WZqFG}
\begin{aligned}
\pi_1&\colon (Z,q,F,G) \mapsto (Z^{-1},q^{-1},q^{-1}Z^{-1}F,G^{-1}),
\\
\pi_2&\colon (Z,q,F,G) \mapsto (Z^{-1}q^{-1},q,Z^{-1} G,F^{-1}),\\\
s_1&\colon (Z,q,F,G) \mapsto (Z^{-1},q,F\frac{(G-1)^2}{(G-Z)^2},Z^{-1}G),\\
s_0&\colon (Z,q,F,G) \mapsto (Z^{-1}q^{-2},q,q^{-1}Z^{-1} F,G\frac{(1-F)^2}{(Z q-F)^2}).
\end{aligned}$$ Introduce element $T=\pi_2^{-1}\circ s_0$ of infinite order in $W$. We will denote $\overline{x}=T(x)$ and $\underline{x}=T^{-1}(x)$. Note that group $W$ is generated by $T,s_1$ and $\pi_1$. We have $$(\overline{Z},\overline{q},\overline{F},\overline{G})=\left(q Z,q,\frac{(F-q Z)^2}{(F-1)^2 G},F\right),\qquad (\underline{Z},\underline{q},
\underline{F},\underline{G})=\left(q^{-1}Z,q,G,\frac{(G-Z)^2}{F (G-1)^2}\right).\ny$$ Therefore $$\overline{G}\underline{G}=\left(\frac{G-Z}{G-1}\right)^2 \label{qPp}.$$ This equation could be called $q$-deformed Painlevé III($D_8$). That is because if we consider $G$ as a function on $Z$ (and $\overline{G}=G(qZ), \, \underline{G}=G(q^{-1}Z)$) then continuous limit of is Painlevé III($D_8$) equation (see Subsection \[ssec:contlimeq\]). Of course this equation is not new, here we get it from the known formulas for $A_7^{(1)\prime}$ surface, but exactly in form it was written in [@Ramani:2015 eq. (20)], see also earlier work [@Ramani:2000].
$\tau$ function representation of group $W$ {#ssec:repr}
-------------------------------------------
We want to lift representation of group $W$ to the level of the $4$ letters $\mathcal{T}_i,\, i=1,\ldots 4$. Analogously to [@Tsuda2006](see also [@KNY]) we call $\mathcal{T}_i$ $\tau$ functions and state Theorem
\[Th:tauformal\] Action of generators $s_1, \pi_1, \pi_2$ of group $W$ on $\mathcal{T}_i,\, i=\overline{1,4}$ given by Table \[lettertable\] provides a representation of $W$ in the field $\mathbb{C}(\mathcal{T}_1, \mathcal{T}_2, \mathcal{T}_3, \mathcal{T}_4, q^{1/4}, Z^{1/4})$.
Proof is straightforward. Generators $s_1, \pi_1, \pi_2$ generate the whole group. So it is enough to check relations of group $s_1^2=1, \pi_1^2=\pi_2^4=(\pi_1\pi_2)^2=1, s_1=\pi_1s_1\pi_1^{-1}=\pi_2^2s_1\pi_2^{-2}$ for given actions. Subgroup $Dih_4$ of $W$ acts on $\tau$ functions the same as it acts on square vertices on Fig. \[dihedral\] i.e. $w(\mathcal{T}_i)=\mathcal{T}_{w(i)}$ for any $w \in Dih_4$. In more geometrical language one can assign $\tau$ functions to the four blow-up points in $\mathbb{CP}^2$ (cf. [@Tsuda2006]).
Define $F,G$ by the formulas $$F=-(qZ)^{1/2}\frac{\mathcal{T}_2^2}{\mathcal{T}_4^2}, \qquad G=-Z^{1/2}\frac{\mathcal{T}_1^2}{\mathcal{T}_3^2}.\label{FGdefinition}$$ This is just an discrete analog of . Then one can check that action of $W$ on $F$ and $G$ induced by action on $\tau$ functions coincides with the action defined in .
The formula for action $T$ implies that $\mathcal{T}_2=\overline{\mathcal{T}_1}$ and $\mathcal{T}_4=\overline{\mathcal{T}_3}$. Then from the action of $T$ of $\mathcal{T}_2$ and $\mathcal{T}_4$ we get $$\label{eq:Tau13}
Z^{1/4}\overline{\mathcal{T}_1}\underline{\mathcal{T}_1}=\mathcal{T}_1^2+Z^{1/2}\mathcal{T}_3^2,\qquad Z^{1/4}\overline{\mathcal{T}_3}\underline{\mathcal{T}_3}=\mathcal{T}_3^2+Z^{1/2}\mathcal{T}_1^2.$$
It is interesting to note that under the action of group $W$ maps $\mathcal{T}_i$ to the Laurent polynomials on $\mathcal{T}_i$ non just rational functions. This property is nontrivial, for example this is not true for $G$ and $F$. This observation follows from the fact that action of $W$ can be represented as composition of mutations for cluster algebra[^2] similar to the higher $q$-deformed Painlevé equation in the paper [@Okubo].
Equations corresponding the $A_7^{(1)\prime}$ surface are usually written in bit more general form then . For example in [@Ramani:2015 eq. (10)], (see also [@Grammaticos:2002 eq.(2.19))]): $$\label{eq:A7'GR}
(\overline{x}y-1)(xy-1)=Zy^2,\quad(xy-1)(x\underline{y}-1)=Z.$$ This system can be solved in terms of $\tau$ functions subject of . Namely $$x=q^{-1/4}Z^{1/4}\frac{\underline{\mathcal{T}_3}\overline{\mathcal{T}_1}}{\mathcal{T}_3\mathcal{T}_1},\quad y=q^{1/4}\frac{\overline{\mathcal{T}_3}{\mathcal{T}_1}}{\mathcal{T}_3\overline{\mathcal{T}_1}}.$$ Therefore one can consider system as a bilinear form of . In the paper [@SakaiLax eq. (2.23)-(2.24)] this system is written in a different form $$\label{eq:A7'Sakai}
f\overline{f}={g(Z-g)}/{(g-1)},\quad
g\overline{g}=\overline{f}^2.$$ This system can be also solved in in terms of $\tau$ functions and in fact equvalent to by the invertible transformation $$g=1-xy, \quad f=-q^{1/2}\underline{g}/\underline{y}.$$
We want to consider $F,G, \mathcal{T}_i$ as functions on $Z$ (and $q$), i.e. $\overline{H}=H(qZ)$, $\underline{H}=H(q^{-1}Z)$ where $H=F, G, \mathcal{T}_i$. The equations become $q$-difference bilinear equations on $\mathcal{T}_1$ and $\mathcal{T}_3$. In the next section we give a solution of these equations in terms of one function $\mathcal{T}(u,s;q|Z)$ depending on two parameters $u,s$. Below we will use parameter $\sigma$, such that $u=q^{2\sigma}$, the parameters $\sigma, s$ are direct analogues of $\sigma,s$ for continuous Painlevé in the formula . Geometrically $s,u$ can be viewed as parametrization of the (open subset of) surface $X$. We set $$\label{eq:T1T3}
\mathcal{T}_1=\mathcal{T}(u,s;q|Z), \quad
\mathcal{T}_3=s^{1/2}\mathcal{T}(uq,s;q|Z),$$ and assume quasi periodicity in $u$ $$\mathcal{T}(uq^2,s;q|Z)=s^{-1}\mathcal{T}(u,s;q|Z).$$ Then the equations reduces to $$Z^{1/4}\mathcal{T}(u,s;q|qZ)\mathcal{T}(u,s;q|q^{-1}Z)=\mathcal{T}(u,s;q|Z)^2+Z^{1/2}\mathcal{T}(uq,s;q|Z)\mathcal{T}(uq^{-1},s;q|Z)\label{TodaT}.$$
The functions $F(Z), G(Z)$ are defined in terms of function $\mathcal{T}$ by formula . Note that solutions of equations depend on four parameters, in addition of $u, s$ we have simple symmetry $(\mathcal{T}_1,\mathcal{T}_3) \mapsto (\alpha Z^{\beta}\mathcal{T}_1,\alpha Z^{\beta}\mathcal{T}_3)$. This symmetry do not acts on the functions $F(Z), G(Z)$.
Remark that element $\pi_2^2 \in W$ commutes with shift $T$, in the continuous limit below this transformation goes to the Bäcklund transformation. It acts by formula $\pi_2^2 \colon \mathcal{T}(u,s;q|Z) \mapsto \mathcal{T}(uq,s;q|Z)$. In terms of parameter $\sigma$ this is transformation $\sigma \mapsto \sigma+1/2$ as for continuous Painlevé.
Continuous limit of the equations {#ssec:contlimeq}
---------------------------------
Let us check that continuous limit of is as it should be for $q$-deformed Painlevé III($D_8$). Introduce notations $q=e^{\hbar}, \, u=q^{2\sg}$. In the limit questions we assume that $|q|<1$ $q \rightarrow 1$.
\[Geqlim\] Substitute $$Z=\hbar^4 z, \quad
G(Z)=\hbar^2 w(z).$$ If $\hbar\rightarrow 0$ the leading order of the equation is Painlevé III($D_8$) equation.
Proof is by direct calculation, namely expansion of into powers of $\hbar$. We obtain that first non-trivial coefficient with power $\hbar^2$. This coefficient vanishes and this is just Painlevé III($D_8$) .
We also check the continuous limit of bilinear equation . It is convenient to define new function $\mathcal{T}_c(u,s;q|Z)$ by $$\mathcal{T}_c(u,s;q|Z)=\frac{(q;q,q)^2_{\infty}}{\Gamma(-(qZ)^{1/4};q^{1/4},q^{1/4})}\mathcal{T}(u,s;q|Z)\label{redqToda}.$$ Equation in terms of function $T_c(u,s;q|Z)$ reads $$\mathcal{T}_c(u,s;q|qZ)\mathcal{T}_c(u,s;q|q^{-1}Z)=\mathcal{T}_c(u,s;q|Z)^2+Z^{1/2}\mathcal{T}_c(uq,s;q|Z)\mathcal{T}_c(uq^{-1},s;q|Z).$$ Taking analytic continuation around $Z=0$ one can change the sign of $Z^{1/2}$ and get $$\label{eq:Todac}
\mathcal{T}_c(u,s;q|qZ)\mathcal{T}_c(u,s;q|q^{-1}Z)=\mathcal{T}_c(u,s;q|Z)^2-Z^{1/2}\mathcal{T}_c(uq,s;q|Z)\mathcal{T}_c(uq^{-1},s;q|Z).$$
\[contlimeq\] Substitute $$Z=\hbar^4 z, \quad \mathcal{T}_c(u,s;q|Z)=\tau(\sg,s|z).$$ If $\hbar\rightarrow 0$ then the leading order of the equation is Toda-like equation .
Proof is by direct calculation, the first nontrivial coefficient appears in order $\hbar^2$.
Formula for $\tau$ functions {#sec:deform}
============================
$q$-deformed conformal blocks {#ssec:qdefblocks}
-----------------------------
In the representation theory approach irregular conformal block for $q$-deformed Virasoro algebra is defined as the square of Whittaker vector for this algebra ([@AY]). This conformal block equals Nekrasov instanton partition function for 5d pure gauge $SU(2)$ theory [@NO] by proposed in [@AY] extension of the AGT conjecture. So the formula for the conformal block reads $$\calF(u_1,u_2;q_1,q_2|Z)=\sum_{\lambda_1,\lambda_2} Z^{|\lambda_1|+|\lambda_2|}\frac{1}{\prod_{i,j=1}^2 N_{\lambda_i,\lambda_j}(u_i/u_j;q_1,q_2)},
\label{confblock}$$ where $$N_{\lambda,\mu}(u,q_1,q_2)
=\prod_{s\in \lambda}
(1-u q_2^{-a_\mu(s)-1}q_1^{\ell_\lambda(s)}) \cdot
\prod_{s \in \mu}
(1-u q_2^{a_\lambda(s)}q_1^{-\ell_\mu(s)-1}).\label{Nlm}$$ The sum in runs over all pairs of Young diagrams $\lambda_1, \lambda_2$. In formula $a_{\lmb}(s), l_{\lmb}(s)$ are lenghts of arms and legs of box $s$ in diagram $\lmb$.
The function $\mathcal{F}(Z)$ depends on $u_1,u_2$ through their ratio $u=u_1/u_2$, so we shall below use notation $u$. Parameter $u$ is $q$-deformed analog of the highest weight and pair $(q_1,q_2)$ is $q$-deformed analog of the central charge. In this paper we everywhere (except Appendix \[app:toralg\]) use specification $q_1^{-1}=q_2=q$. In continuous limit this correspond to irregular conformal block for Virasoro algebras with $c=1$, as in formula for continuous $\tau$ function .
It follows directly from the definition that $$\mathcal{F}(u; q^{-1}, q|Z)=\mathcal{F}(u; q, q^{-1}|Z)=\mathcal{F}(u^{-1}; q^{-1}, q|Z). \label{blocksim}$$
The function $\mathcal{F}$ is defined as a power series $\mathcal{F}(Z)=1+O(Z)$. This series converges, the proof of following proposition is similar to the one in [@Its:2014 Prop 1 (i)].
\[prop:convF\] Let $|q|\neq 1$ and $u \neq q^n$, $n \in \mathbb{Z}$. Then series converges uniformly and absolutely on every bounded subset of $\mathbb{C}$.
There exist constants $L_1,L_2 \in \mathbb{R}_{>0}$ such that $$\left|\frac{q^{n/2}-q^{-n/2}}{q^{1/2}-q^{-1/2}}\right|>n L_1^{1/2},\;\; \forall n \in \mathbb{Z}_{\neq 0}\qquad \left|\frac{u^{1/2}q^{n/2}-u^{-1/2}q^{-n/2}}{q^{1/2}-q^{-1/2}}\right|>L_2^{1/2}, \;\; \forall n \in \mathbb{Z}.$$ Then we can bound $\prod_{i,j=1}^2 N_{\lambda_i,\lambda_j}(u_i/u_j;q^{-1},q)$ as $$\begin{gathered}
\Bigl| N_{\lambda_1,\lambda_1}(1;q^{-1},q) N_{\lambda_2,\lambda_2}(1;q^{-1},q) \Bigr|= \prod_{s \in \lambda_1}|q^{\frac12 h_{\lambda_1}(s)}-q^{-\frac12 h_{\lambda_1}(s)}|^2\\ \cdot \prod_{s \in\lambda_2}(\lambda_1 \leftrightarrow \lambda_2) >
\frac{|\lambda_1|!^2|\lambda_2|!^2}{(\dim\lambda_1\dim\lambda_2)^2}\left|L_1(q^{1/2}-q^{-1/2})^2\right|^{|\lambda_1|+|\lambda_2|},
\end{gathered}$$ $$\begin{gathered}
\Bigl|N_{\lambda_1,\lambda_2}(u;q^{-1},q) N_{\lambda_2,\lambda_1}(u^{-1};q^{-1},q)\Bigr| =\prod_{s \in \lambda_1}|u^{\frac12}q^{\frac12(a_{\lambda_2}(s)+\ell_{\lambda_1}(s)+1)}-u^{-\frac{1}2}q^{-\frac12(a_{\lambda_2}(s)+\ell_{\lambda_1}(s)+1)}|^2\\ \cdot \prod_{s \in\lambda_2}(\lambda_1 \leftrightarrow \lambda_2, u\leftrightarrow u^{-1}) > \left|L_2(q^{1/2}-q^{-1/2)^2}\right|^{|\lambda_1|+|\lambda_2|},
\end{gathered}$$ where we used hook length formula for $\dim \lambda$. Since $\sum_{|\lambda|=n} (\dim\lambda)^2=n!$ we have $\mathcal{F}(u;q|Z)<\exp\left|\dfrac{2|Z|}{L_1 L_2(q^{1/2}-q^{-1/2})^4}\right|$.
To ensure convergence of infinite products like $(u;q,q)_{\infty}=\prod_{i,j\geq 0}(1-uq^{i+j})$ we impose condition $|q|<1$. Using analytic continuation one can also work in the region $|q|>1$.
\[qHc\] The $q$-deformed conformal blocks satisfy bilinear relations $$\begin{gathered}
\sum_{2n\in\mathbb{Z}} \frac{u^{2n}Z^{2n^2}}{\prod\limits_{\epsilon, \epsilon'=\pm1}(u^{\epsilon}q^{1+2\epsilon'n};q,q)_{\infty}}\kr{F}(uq^{-2n};q^{-1},q|q^{-1}Z)\kr{F}(uq^{2n};q^{-1},q|qZ)=\\
=(1-Z^{1/2})\sum_{2n\in\mathbb{Z}} \frac{Z^{2n^2}}{\prod\limits_{\epsilon, \epsilon'=\pm1}(u^{\epsilon}q^{1+2\epsilon'n};q,q)_{\infty}}\kr{F}(uq^{-2n};q^{-1},q|Z)\kr{F}(uq^{2n};q^{-1},q|Z)
\label{bilconfrel}\end{gathered}$$
We do not have proof of this conjecture, but we have two arguments in support. First this conjecture was checked by computer calculation up to $Z^4$ analytically and up to $Z^{12}$ numerically.
Second argument is the fact that continuous limit of this relation gives known ([@KGP eq. (4.29)]) relation for irregular Virasoro conformal blocks $\mathcal{F}(\Delta|z)$ $$-2z^{1/2}\sum_{2n\in\mathbb{Z}} \frac{\mathcal{F}((\sg+n)^2|z) \mathcal{F}((\sg-n)^2|z)}{\prod_{k=1}^{2|n|-1}(k^2-4\sg^2)^{2(2|n|-k)}(4\sg^2)^{2|n|}} =
\sum_{2n\in\mathbb{Z}} \frac{D^2_{[\log z]}(\mathcal{F}(\sg+n)^2|z), \mathcal{F}(\sg-n)^2|z)}{\prod_{k=1}^{2|n|-1}(k^2-4\sg^2)^{2(2|n|-k)}(4\sg^2)^{2|n|}}
\label{bilincont}$$ Here $D^2_{[\log z]}(f(z),g(z))=z^2(f''g-f'g'+fg'')+z(f'g+fg')$ denotes second Hirota differential operator with respect to variable $\log z$. We shall provide continuous limit and obtain in Subsection \[ssec:contlim\].
Most results of this subsection can be stated for any $q_1, q_2$, see e.g. Appendix \[app:toralg\] for the bilinear relations. But for $\tau$ functions we will use only conformal blocks with $q_1q_2=1$.
$q$-deformation of the formula for $\tau$ function {#ssec:qdeftau}
--------------------------------------------------
\[Def:tau\] Function $\mathcal{T}(u,s;q|Z)$ given by the formula $$\mathcal{T}(u,s;q|Z)=\sum_{n \in \mathbb{Z}}s^n C(uq^{2n};q|Z) \frac{\mathcal{F}(uq^{2n};q^{-1},q|Z)}{(uq^{2n+1};q,q)_\infty(u^{-1}q^{-2n+1};q,q)_\infty}\label{T}$$ is called $q$-deformed $\tau$ function of Painlevé III($D_8$) equation if function $C(u;q|Z)$ satisfy equations $$\begin{aligned}
\frac{C(uq;q|Z)C(uq^{-1};q|Z)}{C(u;q|Z)^2}&=&-Z^{1/2}\label{C01}\\
\frac{C(uq;q|qZ)C(uq^{-1};q|q^{-1}Z)}{C(u;q|Z)^2}&=&-uZ^{1/4}\label{C11}\\
\frac{C(u;q|qZ)C(u;q|q^{-1}Z)}{C(u;q|Z)^2}&=&Z^{-1/4}\label{C10}.\end{aligned}$$
If $C(u;q|Z)=C(u^{-1};q|Z)$ then functions $C, \mathcal{T}$ are called $u$-inverse invariant.
Evidently, $C(u;q|Z)$ could be multiplied on any function $\widetilde{C}(u;q|Z)$, which satisfy homogeneous equations , , .
\[Ex:C\] The following examples of $C(u;q|Z)$ satisfy ,, $$\begin{aligned}
C_1(u;q|Z)&=&\Gamma((qZ)^{1/4};q^{1/4},q^{1/4})^3/\left(\Gamma(i(qZu)^{1/4};q^{1/4},q^{1/4})\,\Gamma(i(qZ)^{1/4}u^{-1/4};q^{1/4},q^{1/4})\right)
\\
C_c(u;q|Z)&=&(-1)^{2\left(\frac{\log u}{2\log q}\right)^2}\Gamma(-(qZ)^{1/4};q^{1/4},q^{1/4})\exp\left(\frac{\log ^2u\log Z}{4\log^2 q}\right).\label{exCc}\end{aligned}$$ Here elliptic Gamma function is defined by $\Gamma(u;q,q)=(q^2 u^{-1};q,q)_{\infty}/(u;q,q)_{\infty}$. Necessary definitions and properties about $q$-deformed special functions are collected in Appendix \[app:q\].
Both functions $C_1,C_c$ are $u$-inverse invariant. Function $C_1$ is meromorphic as function on $Z^{1/2}$ (or on $u^{1/4}$) on complex plain. Function $C_c$ is useful to make continuous limit (Subsection \[ssec:contlim\]). But $C_c$ is not meromorphic.
Left sides of the equations , , are multiplicative second order difference derivatives of function on two variables $(u|Z)$ in directions $(1|0), (1|1), (0|1)$ correspondingly.
We could hide $s^n$ into the $C(uq^{2n};q|Z)$ carrying out some simple factor from the sum. Indeed, function $\theta(us^{1/2}q;q)$ satisfies homogeneous equations ,, so $$\mathcal{T}(u,s;q|Z)=\theta(us^{1/2}q;q)\sum_{n\in\mathbb{Z}}C_s(uq^{2n};q|Z) \frac{\mathcal{F}(uq^{2n};q^{-1},q|Z)}{(uq^{2n+1};q,q)_\infty(uq^{-2n+1};q,q)_\infty},$$ where $C_s(u;q|Z)=C(u;q|Z) \theta(us^{1/2}q;q)^{-1}$.
It is clear from the definition that $$\mathcal{T}(uq^2,s;q|Z)=s^{-1}\mathcal{T}(u,s;q|Z)\label{ushift}.$$ If the function $C$ are $u$-inverse invariant then $$\mathcal{T}(u,s;q|Z)=\mathcal{T}(u^{-1},s^{-1};q|Z) \label{uinv}$$ If $C$ is not $u$-inverse invariant then the function $\mathcal{T}(u^{-1},s;q|Z)$ also satisfies definition but for another function $C$.
\[conj:Toda\] Function $\mathcal{T}(u,s;q|Z)$ satisfy bilinear equations $$Z^{1/4} \mathcal{T}(u,s;q|qZ) \mathcal{T}(u,s;q|q^{-1}Z) = \mathcal{T}(u,s;q|Z)^2+Z^{1/2}\mathcal{T}(uq,s;q|Z)\mathcal{T}(uq^{-1},s;q|Z). \label{qTodai}$$
Conjecture \[qHc\] is equivalent to Conjecture \[conj:Toda\]. \[mainbilin\]
Let us substitute the expression for $\tau$ function into and collect terms with the same powers of $s$. The vanishing condition of the $s^m$ coefficient has the form $$\begin{gathered}
Z^{1/4}\sum_{n\in\mathbb{Z}}C(uq^{2n+2m};q|qZ)C(uq^{-2n};q|q^{-1}Z) \frac{\kr{F}(uq^{2n+2m};q^{-1},q|qZ)\kr{F}(uq^{-2n};q^{-1},q|q^{-1}Z)}{\prod\limits_{\epsilon=\pm1}((uq^{2n+2m})^{\epsilon}q;q,q)_{\infty}((uq^{-2n})^{\epsilon}q;q,q)_{\infty}}=\\
=\sum_{n\in\mathbb{Z}}C(uq^{2n+2m};q|Z)C(uq^{-2n};q|Z) \frac{\kr{F}(uq^{2n+2m};q^{-1},q|Z)\kr{F}(uq^{-2n};q^{-1},q|Z)}{\prod\limits_{\epsilon=\pm1}((uq^{2n+2m})^{\epsilon}q;q,q)_{\infty}((uq^{-2n})^{\epsilon}q;q,q)_{\infty}}+\\
+Z^{1/2}\sum_{n\in\mathbb{Z}}C(uq^{2n+2m+1};q|Z)C(uq^{-2n-1};q|Z) \frac{\kr{F}(uq^{2n+2m+1};q^{-1},q|Z)\kr{F}(uq^{-2n-1};q^{-1},q|Z)}{\prod\limits_{\epsilon=\pm1}((uq^{2n+2m+1})^{\epsilon}q;q,q)_{\infty}((uq^{-2n-1})^{\epsilon}q;q,q)_{\infty}}.\end{gathered}$$
Let us substitute into these relations $u\rightarrow uq^{-m}$, $n\rightarrow n-m/2$. We see that for any $m$ vanishing conditions of the coefficients with powers $s^{2(m+\delta)}, \delta=0,1/2$ are equivalent to $$\begin{gathered}
Z^{1/4}\sum_{n\in\mathbb{Z}+\delta}C(uq^{2n};q|qZ)C(uq^{-2n};q|q^{-1}Z) \frac{\kr{F}(uq^{2n};q^{-1},q|qZ)\kr{F}(uq^{-2n};q^{-1},q|q^{-1}Z)}{\prod\limits_{\epsilon=\pm1}((uq^{2n})^{\epsilon}q;q,q)_{\infty}((uq^{-2n})^{\epsilon}q;q,q)_{\infty}}=\\
=\sum_{n\in\mathbb{Z}+\delta}C(uq^{2n};q|Z)C(uq^{-2n};q|Z) \frac{\kr{F}(uq^{2n};q^{-1},q|Z)\kr{F}(uq^{-2n};q^{-1},q|Z)}{\prod\limits_{\epsilon=\pm1}((uq^{2n})^{\epsilon}q;q,q)_{\infty}((uq^{-2n})^{\epsilon}q;q,q)_{\infty}}+\\
+Z^{1/2}\sum_{n\in\mathbb{Z}+\delta}C(uq^{2n+1};q|Z)C(uq^{-2n-1};q|Z) \frac{\kr{F}(uq^{2n+1};q^{-1},q|Z)\kr{F}(uq^{-2n-1};q^{-1},q|Z)}{\prod\limits_{\epsilon=\pm1}((uq^{2n+1})^{\epsilon}q;q,q)_{\infty}((uq^{-2n-1})^{\epsilon}q;q,q)_{\infty}}.\end{gathered}$$ We want to divide these conditions on $C(u;q|Z)^2$ and then simplify arising expressions like $\frac{C(\cdot;q|\cdot)C(\cdot;q|\cdot)}{C(\cdot;q|\cdot)^2}$. Introduce auxiliary functions $\beta_k$, $\gamma_k$, $k\in \mathbb{Z}$ by formulas $$\beta_k=\frac{C(uq^{k};q|Z)C(uq^{-k};q|Z)}{C(u;q|Z)^2}, \quad \gamma_k=\frac{C(uq^{k};q|qZ)C(uq^{-k};q|q^{-1}Z)}{C(u;q|qZ)C(u;q|q^{-1}Z)}.$$ Elementary calculations give us $$\beta_k=Z\beta^2_{k-1}/\beta_{k-2}, \quad \gamma_k=Z\gamma^2_{k-1}/\gamma_{k-2}. \label{bgd2}$$ Evidently $\beta_0=\gamma_0=1$. From , , we have $\beta_1=-Z^{1/2}, \gamma_1=-uZ^{1/2}$. So we have unique solution of $$\beta_k=(-1)^kZ^{k^2/2},\quad \gamma_k=(-1)^ku^kZ^{k^2/2}.$$ Using these results and we obtain $$\begin{gathered}
\sum_{n\in\mathbb{Z}+\delta}u^{2n}Z^{2n^2} \frac{\kr{F}(uq^{2n};q^{-1},q|qZ)\kr{F}(uq^{-2n};q^{-1},q|q^{-1}Z)}{\prod\limits_{\epsilon=\pm1}((uq^{2n})^{\epsilon}q;q,q)_{\infty}((uq^{-2n})^{\epsilon}q;q,q)_{\infty}}=\\
=\sum_{n\in\mathbb{Z}+\delta}Z^{2n^2} \frac{\kr{F}(uq^{2n};q^{-1},q|Z)\kr{F}(uq^{-2n};q^{-1},q|Z)}{\prod\limits_{\epsilon=\pm1}((uq^{2n})^{\epsilon}q;q,q)_{\infty}((uq^{-2n})^{\epsilon}q;q,q)_{\infty}}-\\
-Z^{1/2}\sum_{n\in\mathbb{Z}+\delta}Z^{(2n+1)^2/2} \frac{\kr{F}(uq^{2n+1};q^{-1},q|Z)\kr{F}(uq^{-2n-1};q^{-1},q|Z)}{\prod\limits_{\epsilon=\pm1}((uq^{2n+1})^{\epsilon}q;q,q)_{\infty}((uq^{-2n-1})^{\epsilon}q;q,q)_{\infty}},\end{gathered}$$ Substituting in last sum $n\rightarrow n-1/2$ we obtain $$\begin{gathered}
\sum_{n\in\mathbb{Z}+\delta}u^{2n}Z^{2n^2} \frac{\kr{F}(uq^{2n};q^{-1},q|qZ)\kr{F}(uq^{-2n};q^{-1},q|q^{-1}Z)}{\prod\limits_{\epsilon=\pm1}((uq^{2n})^{\epsilon}q;q,q)_{\infty}((uq^{-2n})^{\epsilon}q;q,q)_{\infty}}=\\
=\sum_{n\in\mathbb{Z}+\delta}Z^{2n^2} \frac{\kr{F}(uq^{2n};q^{-1},q|Z)\kr{F}(uq^{-2n};q^{-1},q|Z)}{\prod\limits_{\epsilon=\pm1}((uq^{2n})^{\epsilon}q;q,q)_{\infty}((uq^{-2n})^{\epsilon}q;q,q)_{\infty}}-\\
-Z^{1/2}\sum_{n\in\mathbb{Z}+\delta+1/2}Z^{2n^2} \frac{\kr{F}(uq^{2n};q^{-1},q|Z)\kr{F}(uq^{-2n};q^{-1},q|Z)}{\prod\limits_{\epsilon=\pm1}((uq^{2n})^{\epsilon}q;q,q)_{\infty}((uq^{-2n})^{\epsilon}q;q,q)_{\infty}},\end{gathered}$$
And this is a result of splitting into part with integer powers of $Z$ and with half-integer powers of $Z$.
Continuous limit of $q$-deformed $\tau$ function {#ssec:contlim}
------------------------------------------------
Formula for $q$-deformed $\tau$ function $\mathcal{T}(u,s;q|Z)$ could be rewritten in next way $$\mathcal{T}(u,s;q|Z)={C}(u;q|Z)\sum_{n\in\mathbb{Z}} \td{s}(u,s;q|Z)^n Z^{n^2+n/2}
\frac{\mathcal{F}(uq^{2n};q^{-1},q|Z)}{\prod\limits_{\epsilon=\pm1}((uq^{2n})^{\epsilon}q;q,q)_{\infty}}, \label{Tr}$$ where $$\td{s}(u,s;q|Z)=-\left(\frac{{C}(u;q|Z)}{{C}(uq^{-1};q|Z)}\right)^2s$$
Using we transform expression $C(uq^{k};q|Z)/C(u;q|Z)$ as $$\begin{aligned}
\frac{C(uq^{k};q|Z)}{C(uq^{k-1};q|Z)}&=\frac{C(uq^{k-1};q|Z)}{C(uq^{k-2};q|Z)}\frac{C(uq^k;q|Z)C(uq^{k-2};q|Z)}{C(uq^{k-1};q|Z)^2}=
-\frac{C(uq^{k-1};q|Z)}{C(uq^{k-2};q|Z)}Z^{1/2}\Rightarrow\\
\frac{C(uq^{k};q|Z)}{C(uq^{k-1};q|Z)}&=(-1)^{k}Z^{\frac{k}2}\frac{C(u;q|Z)}{C(uq^{-1};q|Z)}
\Rightarrow\\
\frac{C(uq^{k};q|Z)}{C(uq^{-1};q|Z)}&=(-1)^{\frac{k(k+1)}2}Z^{\frac{k(k+1)}{4}}\left(\frac{C(u;q|Z)}{C(uq^{-1};q|Z)}\right)^{k+1}.
\end{aligned}$$ Substituting the last expression to we finish the proof.
The $\tau$-function was defined as a series and it is convenient to prove convergence of this series using expression . The proof is analogous to one in [@Its:2014 Prop. 1 (ii)].
First, note that bounds $L_1, L_2$ defined in the proof of Proposition \[prop:convF\] are the same for every conformal block in the sum . Therefore we can estimate conformal blocks by the same exponent.
Second we to rewrite Pochhammer symbols in terms of $q$-Barnes $\G$-function using . Then for $n>0$ we have $$\G(1+2(\sg+n);q)\G(1-2(\sg+n);q)=\frac{\G(1-2\sg;q)}{\G(1+2\sg;q)} \frac{(-1)^n\theta(uq;q)^{2n}} {u^{n(2n-1)} q^{\frac{n(4n^2-1)}3} (1-q)^{2n} (q;q)^{4n}_{\infty}}\G^2(1+2(\sg+n);q).$$ where $\sg=\frac{\log u}{2\log q}$. It follows form [@Nish Prop. 3.1] that $\log \G(x;q)\sim -\log(1-q)x^2/2$, $\operatorname{Re}(x)\rightarrow+\infty$ (this is the first term in the sum in loc. cit. and one can show the all other terms are majorized by it). Using $|q|<1$ and this asymptotic behavior we see that this coefficients dominate $Z^{n^2+n/2}\tilde{s}^n$.
Calculation for $n<0$ is the same with the replacement $\sigma \leftrightarrow -\sigma$.
Moreover, we could rewrite $\mathcal{T}(u,s;q|Z)$ as sum of conformal blocks with some $\hat{s}$ and $q$-rational coefficients (in analogous to continuous $\tau$ function ([@BSRamon])). Introduce functions $P_n(u;q)$ by $$\frac{\prod\limits_{\epsilon=\pm1}(u^{\epsilon}q;q,q)_{\infty}}{\prod\limits_{\epsilon=\pm1}((uq^{2n})^{\epsilon}q;q,q)_{\infty}}
=P_n(u;q)\left(\frac{(u;q)_{\infty}}{(u^{-1};q)_{\infty}}\right)^{2n},\quad n \in \mathbb{Z}$$ Then we can write $$\label{eq:Trational}
\mathcal{T}(u,s;q|Z)=\frac{C(u;q|Z)}{\prod\limits_{\epsilon=\pm1}(u^{\epsilon}q;q,q)_{\infty}}\sum_{n\in\mathbb{Z}} Z^{n^2+n/2} \hat{s}^n P_n(u;q)
\mathcal{F}(uq^{2n};q^{-1},q|Z),$$ $$\begin{aligned}
\text{where}\quad &P_n(u;q)=
\frac{(-1)^n}{(1-u)^{2n}\prod_{i=1}^{2n-1}(u^{1/2}q^{i/2}-u^{-1/2}q^{-i/2})^{2(2n-i)}},\;\; n\geq 0,
\\
&P_n(u;q)= P_{-n}(u^{-1};q), \;\; n<0,\qquad \hat{s}=-\left(\frac{C(u;q|Z)}{C(uq^{-1};q|Z)}\frac{(u;q)_{\infty}}{(u^{-1};q)_{\infty}}\right)^2s.
\end{aligned}$$ The formula can also be used for the proof of convergence . The proof goes similarly to the proof of Proposition \[prop:convF\], we bound all terms in the denominator of $P_n(u;q)$ using $L_1,L_2$.
Now we check that in $q \rightarrow 1^-$ limit (i.e. $q\in [1-\epsilon,1]$) the formula for $\mathcal{T}(u,s;q|Z)$ gives the formula for $\tau$ function of continuous Painlevé equation. As in Subsection \[ssec:contlimeq\] it is convenient to use function $\mathcal{T}_c(u,s;q|Z)$ defined by . Moreover for this limit we take $q$-deformed $\tau$ functions with $C(u;q|Z)=C_c(u;q|Z)$ given by formula .
\[thmlim\] Let $$q=e^{\hbar},\quad Z=\hbar^4 z, \quad \sg=\frac{\log u}{2\hbar}.$$ Then the $\tau$-function $\mathcal{T}_c(u,(-1)^{-4\sigma}s;q|Z)$ with $C(u;q|Z)=C_c(u;q|Z)$ goes to $\tau(\sg,s|z)$ in the limit $\hbar\rightarrow -0$.
Rewrite $\mathcal{T}_c(u,s;q|Z)$ in form using $$\mathcal{T}_c(u,s;q|Z)=(-1)^{2\sg^2}\sum_{n\in\mathbb{Z}} Z^{(\sg+n)^2} ((-1)^{4\sg}s)^n
\frac{\mathcal{F}(uq^{2n};q^{-1},q|Z)}{\prod\limits_{\epsilon=\pm1}((uq^{2n})^{\epsilon}q;q,q)_{\infty}}\label{toconttau}$$
First, we prove convergence of conformal blocks $\mathcal{F}(u;q^{-1},q|Z) \rightarrow \mathcal{F}(\sigma^2;|z)$. For each summand in we have $$N_{\lmb,\mu}(q^{-1},q,u)\sim(-\hbar)^{|\lmb|+|\mu|}\prod_{s\in\lmb}(2\sg-(a_{\mu}(s)+l_{\lmb}(s)+1))\times \prod_{s\in\mu}(2\sg+(a_{\mu}(s)+l_{\lmb}(s)+1))$$ This occasional power of $\hbar$ cancels due to definition $Z=\hbar^4z$.
Next, we prove that convergence of series is uniform on $q \in [1-\epsilon,1]$. Note that in this region $\left|\frac{q^{x/2}-q^{-x/2}}{q^{1/2}-q^{-1/2}}\right|\geq x$, for $x \geq 1$ so $L_1$ (from the proof of Prop \[prop:convF\]) is uniformly bounded by $L_1\geq 1$. And for $L_2$ using evident inequalities $$|q^{y}-q^{-y}|\geq |q^{\operatorname{Re} y}-q^{-\operatorname{Re}y}|,\quad |q^{a_1}-q^{-a_1}|\geq |q^{a_2}-q^{-a_2}|, \quad \left|\frac{q_1^{a}-q_1^{-a}}{q_1^{1/2}-q_1^{-1/2}}\right|\leq \left|\frac{q_2^{a}-q_2^{-a}}{q_2^{1/2}-q_2^{-1/2}}\right|,$$ for $a_1>a_2>0$ and $0<q_1<q_2<1$, $0<a<1/2$. Therefore we have bound $$L_2^{1/2}\geq \frac{|(1-\epsilon)^{\operatorname{Re} \sigma+n_0/2}-(1-\epsilon)^{-\operatorname{Re}\sigma -n_0/2}|}{(1-\epsilon)^{1/2}-(1-\epsilon)^{-1/2}},$$ where $-1/4<\operatorname{Re} \sigma+n_0/2<1/4$. Since bounds for $L_1, L_2$ are uniform we prove that convergence of are also uniform.
Next we study coefficients in . Using formula we can rewrite $$Z^{(\sg+n)^2}\frac{1}{\prod\limits_{\epsilon=\pm1}((uq^{2n})^{\epsilon}q;q,q)_{\infty}}
=\frac{1}{\mathsf{G}(1+2(\sg+n);q)\mathsf{G}(1-2(\sg+n);q)} \left(\frac{Z}{(1-q)^4}\right)^{(\sg+n)^2}.$$ Then using Theorem \[Glim\] and $\frac{Z}{(1-q)^4}\sim z$ we that coefficients goes to the coefficients in $\tau$ function for continuous Painlevé equation .
It remains to show that series converges uniformly on $q \in [1-\epsilon,1]$. This can be done similar to the series above.
It follows from this theorem, Theorem \[mainbilin\] and Proposition \[contlimeq\] that $q \rightarrow 1$ limit of bilinear relations is . This fact was stated in the end of Subsection \[ssec:qdefblocks\].
Note that Bonelli, Grassi and Tanzini in their paper [@BGT] also constructed function, which in the limit $q \rightarrow 1$ goes to the $\tau(\sigma,s|z)$. It is interesting to note that they work in different region of $q$, namely $|q|=1$ in their paper. The relation between our results and [@BGT] will be studied in [@BGT2].
$q$-deformation of Painlevé III($D_8$) algebraic solution {#ssec:algsol}
---------------------------------------------------------
For special values of parameters $\sg=1/4, s=\pm 1$ the sum in formula can be calculated (see [@BSRamon Sec. 3.3]) $$\tau(1/4,\mp 1|z)=\frac{1}{\mathsf{G}(1/2)\mathsf{G}(3/2)}z^{1/16}e^{\pm4z^{1/2}} \label{stattau}$$ These $\tau$ functions correspond to the algebraic solutions of Painlevé III($D_8$) equation $w(z)=\mp z^{1/2}$. These solutions are invariant under Bäcklund transformation so $\tau(\sg,s|z)\propto \tau(\sg+1/2,s|z)$ and substituting this into we obtain equation $$1/2 D^2_{[\log z]}(\tau(z),\tau(z))=\pm z^{1/2}\tau(z)^2. \label{statToda}$$ This equation gives us up to multiplying on constant and any power of $z$.
We want to obtain analogous relation on $q$-deformed conformal blocks. Recall that element $\pi_2^2\in W$ is an analogue of Bäcklund transformation and acts as $\mathcal{T}(u,s;q|Z) \mapsto \mathcal{T}(uq,s;q|Z)$. Therefore if $\tau$ function $\mathcal{T}(u,s;q|Z)$ is $u$-inverse invariant then for $u=q^{1/2}$, $s=\pm 1$ we have $$\pi_2^2(\mathcal{T}(q^{1/2},\pm 1;q|Z))=\mathcal{T}(q^{3/2},\pm 1;q|Z) =\pm \mathcal{T}(q^{-1/2},\pm 1;q|Z)=\pm \mathcal{T}(q^{1/2},\pm 1;q|Z), \label{eq:q12shifts}$$ where we used .
The corresponding function $G(z)=\mp Z^{1/2}$ due to and . These $G(Z)$ are algebraic (and Bäcklund invariant) solutions of .
\[Pochh\] For any $u$-inverse invariant $\tau$ function $T(u,s;q|Z)$ we have $$\mathcal{T}(q^{1/2},\pm 1;q|Z)=\frac{C(q^{1/2};q|Z)}{(q^{3/2};q,q)_{\infty}(q^{1/2};q,q)_{\infty}}(\mp Z^{1/2}q^{1/2};q^{1/2},q^{1/2})_{\infty}. \label{tau14}$$ Equivalently, we have relation on $q$-deformed conformal blocks $$\label{eq:algebconfblock}
(\mp Z^{1/2}q^{1/2};q^{1/2},q^{1/2})_{\infty}=\sum_{n\in\mathbb{Z}}(\mp 1)^n Z^{n^2+n/2}P_n(q)
\mathcal{F}(q^{2n+1/2},q,q|Z),$$ where $$P_n(q)=\frac{\prod\limits_{\epsilon=\pm1}(q^{\frac12\epsilon}q;q,q)_{\infty}}{\prod\limits_{\epsilon=\pm1}((q^{2n+1/2})^{\epsilon}q;q,q)_{\infty}}=\prod_{j=0}^{k-1}\frac{1}{\left((1-q^{j+1/2})(1-q^{-j-1/2})\right)^{k-j}},$$ where $k=2n$, for $n>0$ and $k=-2n-1$, for $n<0$.
One can compare with and see that $P_n(q)=(-1)^nP_n(q^{1/2},q)$.
If Conjecture \[qHc\] holds then Conjecture \[Pochh\] also holds.\[Pochht\]
Due to Theorem \[mainbilin\] we can use Conjecture \[conj:Toda\] instead of Conjecture \[qHc\]. Using we can rewrite equation as $$\label{eq:Todaalg}
Z^{1/4} \mathcal{T}(q^{1/2},\pm 1;q|qZ) \mathcal{T}(q^{1/2},\pm 1;q|q^{-1}Z) =(1\pm Z^{1/2}) \mathcal{T}(q^{1/2},\pm 1;q|Z)^2.$$ On the other hand, using we have $$\mathcal{T}(q^{1/2},\pm 1;q|qZ)={C}(q^{1/2};q|Z)\sum_{n\in\mathbb{Z}} (\mp 1)^n Z^{n^2+n/2}
\frac{\mathcal{F}(q^{2n+1/2};q^{-1},q|Z)}{\prod\limits_{\epsilon=\pm1}((uq^{2n})^{\epsilon}q;q,q)_{\infty}},$$ where we used $u$-inverse invariance of $C(u;q|Z)$. Denote $$f(z)=\mathcal{T}(q^{1/2},\pm 1;q|qZ) \frac{(q^{3/2};q^{1},q^{1})_{\infty}(q^{1/2};q,q)_{\infty}}{C(q^{1/2};q|Z)}$$ Then $f(z)$ is equal to the right side of . And due to the equation for this function takes the form $$f(qZ) f(q^{-1}Z)=(1\pm Z^{1/2}) f(Z)f(Z).$$ The only solution $f(Z)=\sum_{n \geq 0} f_n Z^{n/2}$ with $f_0=1$ of the last equation is $f(Z)= (\mp Z^{1/2}q^{1/2};q^{1/2},q^{1/2})_{\infty}$.
We have checked equality up to $Z^4$.
Let us check the continuous limit of . We want to use Theorem \[thmlim\] so we set $C(u;q|Z)=C_c(u;q|Z)$. Let Then we divide by $\Gamma(-(qZ)^{1/4};q^{1/4},q^{1/4})/(q;q,q)^2_{\infty}
$ and get $\mathcal{T}_C$ in the left side (see ). Using Theorem \[thmlim\] we have that if $Z=\hbar^4z,\, q=e^{\hbar}$, $q\rightarrow 1^-$ then $\mathcal{T}_c(q^{1/2},\pm1;q|Z)\rightarrow\tau(1/4,\mp 1|z).$
For the continuous limit of $q$-Pochhammer symbol in right side of we have (using ) $$(\pm Z^{1/2}q^{1/2};q^{1/2},q^{1/2})=\exp\left(-\sum_{m=1}^{\infty}\frac{(\mp Z^{1/2}q^{1/2})^m }{m(1-q^{m/2})^2}\right)\sim e^{\pm 4z^{1/2}},$$ where we observed that only the term of sum with $m=1$ survives. Using this result and Theorem \[Glim\] we obtain that limit is known relation (after analyti continuation arownd $Z=0$ as in Subsection \[ssec:contlimeq\]).
Further questions {#sec:concl}
=================
- The main statements of the paper are based on the conjecture \[qHc\], so the first question is to prove it.
- The second question is about a generalization of our results to the other discrete Painlevé equations from Sakai’s tables Fig \[Fig:Sakaisurf\], \[Fig:Sakaisym\]. It is natural to conjecture that $q$-difference Painlevé equations $A_{7-N}^{(1)}/E_{N+1}^{(1)}$ surface/symmetry type $N \leqslant 7$ are related to the Nekrasov partition functions for 5d $SU(2)$ gauge theory with $N$ fundamental multiplets.
It was argued by Seiberg [@Seib] that these gauge theories has $E_{N+1}$ global symmetry. It would be interesting to find a physical interpretation of the affine Weyl groups $E_{N+1}^{(1)}$, the symmetry group of the corresponding discrete Painlevé equations.
For $N \leq 4$ one can take $q\rightarrow 1$ limit and get the relation between differential Painlevé equation with surface type $D_4^{(1)}, D_5^{(1)}, D_6^{(1)}, D_7^{(1)}, D_8^{(1)}$ and Nekrasov partition functions for 4d $SU(2)$ gauge theory with $4, 3, 2, 1, 0$ fundamental multiplets correspondingly. This relation was stated in [@GIL1302] and proven in [@ILT] and [@KGP].
- Elements $T, \pi_2^2 \in W$ act on the $\tau$ function $\mathcal{T}(u,s;q|Z)$ defined in in a clear manner $$T\colon \mathcal{T}(u,s;q|Z) \mapsto \mathcal{T}(u,s;q|qZ),\quad \text{and}\quad \pi_2^2 \colon \mathcal{T}(u,s;q|Z) \mapsto \mathcal{T}(u q,s;q|Z)$$ It is natural to ask for the action of whole group $W$. From the Table \[lettertable\] we see that remaining transformations are $Z \mapsto Z^{-1}$ and $q \mapsto q^{-1}$. The second transformation is transparent due to $q$-deformed conformal block relation and $q$ Pochhammer symbol relation . But it is unclear what is (if exists) the meaning of $Z \mapsto Z^{-1}$ symmetry for the formula . In particular it is unclear what is the form of the corresponding relation between $q$-deformed conformal blocks (Nekrasov partition functions for pure theory) $\mathcal{F}(Z)$ and $\mathcal{F}(Z^{-1})$.
One more remark is in order. Nekrasov partition function for pure 5d $SU(2)$ gauge theory is equal (up to simple factor) to the topological string partition function for local $\mathbb{P}^1\times \mathbb{P}^1$ geometry [@Igbal:2003], [@Eguchi:2003]. For such partition functions there exists fiber-base duality [@Mitev] which interchange two factors $\mathbb{P}^1$. In terms of the functions $\mathcal{F}(u;q^{-1},q|Z^{-1})$ this duality has the form $$\label{Zinv}
\frac{\mathcal{F}(u; q^{-1}, q|Z)}{(uq;q,q)_{\infty}^2} =
\frac{\mathcal{F}(uZ; q^{-1}, q|Z^{-1})}{(uZq;q,q)_{\infty}^2}.$$ But this is an equality of formal power series on variables $u, uZ$. For $|q|\neq 1$ function $\mathcal{F}(u; q^{-1}, q|Z)$ cannot be expanded as such convergent series since has poles for $u=q^n$ and these poles accumulate near $u=0$. Moreover, computer calculations shows that does not hold for the function $\mathcal{F}(u; q^{-1}, q|Z)$ defined by convergent series . It is an interesting question whether exist relation of the type for $|q|\neq 1$ and can it be used to the $Z \mapsto Z^{-1}$ transformation of $\tau$-function $\mathcal{T}(u,s;q|Z)$.
- Continuous Painlevé equations can be described as non-autonomous Hamiltonian system. It is natural to ask for the difference analogue of this fact.
Painlevé III($D_8$) has two types of bilinear forms, namely Toda-like and Okamoto-like (see [@BSRamon]). In Subsection \[ssec:contlimeq\] we give $q$-deformation of the Toda-like bilinear form, it is natural to ask for the $q$-deformation of Okamoto-like equations.
- It is interesting to note that there exists another $q$-difference equation wichi has the Painleve III$(D_8)$ equation in the $q\rightarrow 1$ limit. This equation has the form $$\label{eq:qA7}
\overline{W}W^2\underline{W}=Z(1-W).$$ and the limit was shown in [@Grammaticos:2002] (see also [@Ramani:2015]) In terms of Sakai classification this equation corresponds to $A_7^{(1)}$ surface (see [@SakaiLax eq. (2.44)]), so should not be equivalent to . It is an interesting question whether exists a way to express solutions of in terms of $q$-deformed conformal blocks.
Acknowledgments
===============
We thank V. Adler, G. Bonelli, B. Feigin, A. Grassi, P. Gavrylenko, N. Iorgov, O. Lisovyy, A. Sciarappa, V. Spiridonov, A. Tanzini, Y. Zenkevich for interest in our work and discussions. We are grateful to A. Dzhamay, K. Kajiwara, H. Sakai and T. Takenawa for many explanations about discrete Painleve equations. It was H. Sakai who explained that we should use $A_7^{(1)\prime}$ surface for $q$-deformation of PIII$(D_8)$ equation. M. B. is grateful to INFN Trieste and SISSA for the hospitality during his visit of Italy and to T. Takenawa for the hospitality during his visit of Japan.
This work has been funded by the Russian Academic Excellence Project ’5-100’. A.S. was also supported in part by joint NASU-CNRS project F14-2016, M.B. was also supported in part by Young Russian Mathematics award and RFBR grant mol\_a\_ved 15-32-20974. Study of $q$-deformed bilinear relations was performed under a grant of Russian Science Foundation (project No. 14-12-01383).
$q$-special functions {#app:q}
=====================
In this Appendix we collect some facts about $q$-series that we used in the paper. For the references see [@AAR Sec. 10], [@Nish], [@Spir].
Infinite multiple $q$-deformed Pochhammer symbol is defined by $$(Z;t_1,\ldots t_N)_{\infty}=\prod_{i_1,\ldots i_N=0}^{\infty}\left(1-Z\prod_{k=1}^Nt_k^{i_k}\right).$$ Product exists if all $|t_k|<1$. Function is symmetric with respect to $t_k$. In this region function is analytic function of all arguments. Infinite $q$-Pochhammer symbols satisfy $$(Z;t_1,\ldots t_N)_{\infty}/(Zt_1;t_1,\ldots t_N)_{\infty}=(Z|t_2,\ldots t_N)_{\infty}, \quad (Z;q)_{\infty}/(Zq;q)_{\infty}=1-Z \label{shift}$$
The function $(Z;t_1,\ldots t_N)_{\infty}$ can be rewritten as $$\begin{gathered}
(Z;t_1,\ldots t_N)_{\infty}=\exp\left(\sum_{i_1,\ldots i_N=0}^{\infty}\log\left(1-Z\prod_{k=1}^Nt_k^{i_k}\right)\right)=\\=
\exp\left(-\sum_{i_1,\ldots i_N=0}^{\infty}\sum_{m=1}^{\infty}\frac{Z^m}m\prod_{k=1}^Nt_k^{mi_k}\right)
=\exp\left(-\sum_{m=1}^{\infty}\frac{Z^m}m\prod_{k=1}^N\frac1{1-t_k^m}\right),
\label{Pexpr}\end{gathered}$$ where sum converge when $|Z|<1, |t_k|\neq 1$. Using this expression the function $(Z;t_1,\ldots t_N)_{\infty}$ can be defined to the region with some $t_k$ greater then $1$. Using this definition we see that $$(Z;t_1^{-1},t_2,\ldots t_N)_{\infty}=(Zt_1;t_1,\ldots t_N)^{-1}_{\infty} \label{qtrans}$$ In the paper we use only $N=1,2$ $q$-Pochhammer symbols.
Introduce trigonometric (or $q$-) Gamma function and trigonometric (or $q$-) Barnes $\mathsf{G}$ function according to [@Nish] $$\begin{aligned}
\Gamma(x;q)&=&(1-q)^{1-x}\frac{(q;q)_{\infty}}{(q^x;q)_{\infty}},\\
\mathsf{G}(x;q)&=&(1-q)^{-\frac{(x-1)(x-2)}2}\prod\limits_{k=0}^{\infty}\frac{(1-q^{x+k})^{k+1}}{(1-q^{k+1})^{k+2-x}}=
(1-q)^{-\frac{(x-1)(x-2)}2}\frac{(q;q)^{x-1}_{\infty}(q^x;q,q)_{\infty}}{(q;q,q)_{\infty}}.\label{GN}\end{aligned}$$ From we have $$\begin{aligned}
\Gamma(u+1;q)&=[u]_q\Gamma(u;q),\;\textrm{ where } [u]_q=\frac{1-q^u}{1-q}\\
\mathsf{G}(u+1;q)&=\Gamma(u;q)\mathsf{G}(u;q).\end{aligned}$$ Function $[u]_q$ is called $q$-number.
We use part of the Theorem 4.4 from [@Nish]
As $q\rightarrow 1^-$ $\Gamma(u;q)$ and $\mathsf{G}(u;q)$ converge to $\Gamma(u)$ and $\mathsf{G}(u)$. Convergence is uniform on any compact set in the domain $\mathbb{C}\backslash \mathbb{Z}_{\leq 0}$.\[Glim\]
Introduce elliptic Gamma function (see e.g. [@Spir]) $$\Gamma(Z;t,q)=\frac{(tqZ^{-1};t,q)_{\infty}}{(Z;t,q)_{\infty}}.\label{GPoch}$$ Elliptic gamma function should not be confused with trigonometric gamma and Barnes $\mathsf{G}$-function with one deformation argument and with standard Gamma function and Barnes $\mathsf{G}$-function without deformation arguments. Elliptic Gamma function satisfy relations $$\Gamma(qZ;t,q)=\theta(Z;t)\Gamma(Z;t,q), \quad \Gamma(tZ;t,q)=\theta(Z;q)\Gamma(Z;t,q),\label{Gshift}$$ where $\theta$ function defined as $$\theta(Z;q)=\frac{1}{(q;q)_{\infty}}\sum_{k\in\mathbb{Z}}(-1)^k q^{\frac{k(k-1)}{2}}Z^k=(Z;q)_{\infty}(qZ^{-1};q)_{\infty}.$$ Here last equality is Jacobi triple product. Shift relations on elliptic Gamma functions could be easily obtained from . It follows from definition that $\theta$ function satisfies $$\theta(qZ;q)=-Z^{-1}\theta(Z;q)=\theta(Z^{-1};q). \label{tshift}$$ From and first equality we obtain useful relation $$\frac{\Gamma(uq,q;q)\Gamma(uq^{-1};q,q)}{\Gamma(u;q,q)^2}=-qu^{-1}. \label{G2p}$$
Bilinear relation for generic $q_1, q_2$. {#app:toralg}
=========================================
Bilinear relations on $q$-deformed conformal blocks exist not only in case $q_1=q^{-1}$, $q_2=q$. In order to write the conjecture we introduce bilinear combination $$\widehat{\calF}_d(u,q_1,q_2|Z)=\sum_{2n \in \mathbb{Z}}\left(\frac{u^{2dn}(q_1q_2)^{4dn^2}Z^{2n^2}}{(uq_1^{4n-2},u^{-1}q_1^{-4n-2})^{(1)}_{\infty}(u q_1^{-1}q_2^{4n+1},u^{-1}q_1^{-1}q_2^{-4n+1})^{(2)}_{\infty}}\calF_n^{(1)}(q_1^{2d}Z)\, \calF_n^{(2)}(q_2^{2d}Z) \right),$$ where we use notations $$\begin{aligned}
&\calF^{(1)}_n(z)=\calF(u q_1^{4n},q_1^2,q_1^{-1}q_2|z),\qquad \calF^{(2)}_n(z)=\calF(u q_2^{4n},q_1q_2^{-1},q_2^2|z),\\
&(uq_1^{4n-2},u^{-1}q_1^{-4n-2})^{(1)}_{\infty}=(uq_1^{4n-2};q_1^{-2}, q_1^{-1}q_2)_\infty(u^{-1}q_1^{-4n-2};q_1^{-2}, q_1^{-1}q_2)_\infty,\\
&(u q_1^{-1}q_2^{4n+1},u^{-1}q_1^{-1}q_2^{-4n+1})^{(2)}_{\infty}=(u q_1^{-1}q_2^{4n+1};q_1^{-1}q_2,q_2^{2})_\infty(u^{-1}q_1^{-1}q_2^{-4n+1};q_1^{-1}q_2,q_2^2)_\infty,\end{aligned}$$ and ensure conditions $|q_2|<1<|q_1|$ (in other sectors one can use ).
We will also the following version of Nekrasov partition function $$\calF_{\lozenge}(u,q_1,q_2|Z)=\sum_{\lambda_1,\lambda_2} Z^{\frac{|\lambda_1|+|\lambda_2|}2}\frac{1}{\prod_{i,j=1}^2 N^{\NS}_{\lambda_i,\lambda_j}({q_1,q_2,u_i}/{u_j})},$$ where $$N^{\lozenge}_{\lambda,\mu}(q_1,q_2,u)
=
\prod_{s\in \lambda^\lozenge}
(1-u q_2^{-a_\mu(s)-1}q_1^{\ell_\lambda(s)}) \cdot
\prod_{s \in \mu^\lozenge}
(1-u q_2^{a_\lambda(s)}q_1^{-\ell_\mu(s)-1}),$$ with $\lambda^{\lozenge}=\{s \in \lambda| a_\mu(s)+\ell_\lambda(s)+1 \equiv0\bmod 2\}$, $\mu^{\lozenge}=\{s \in \mu| a_\lambda(s)+\ell_\mu(s)+1 \equiv0\bmod 2\}$ (cf. ,).
The following relation holds $$\label{eq:genbilin}
\calF_{\lozenge}(u,q_1,q_2|Z)=\widehat{\calF}_1(u,q_1,q_2|Z),\qquad \calF_{\lozenge}(u,q_1,q_2|Z)=\left(1-q_2q_1Z^{1/2}\right) \widehat{\calF}_0(u,q_1,q_2|Z).$$
As a combination of two formulas in one get generalization of Conjecture \[qHc\] to the case $q_1q_2\neq 1$. The formulas were checked by computer calculation up to $Z^3$ analytically and up to $Z^{12}$ numerically.
Geometrical meaning of the first equality in is following. Left side and right side are generating functions of equivariant Euler characteristics of sheaf $\mathcal{O}$ for Nakajima quiver varieties for the quiver $A_1^{(1)}$. Another interpretation of the corresponding manifolds are partial compactification of the instantons on the minimal resolution of $\mathbb{C}^2/\mathbb{Z}_2$. But these compactifications are different as $(\mathbb{C}^*)^2$ manifolds, in terms of quiver varieties the stability parameters belong to different chambers. Nevertheless conjectural formula states that corresponding equiavariant Euler characteristics are the same. In the conformal (or cohomology, or just $q_1,q_2 \rightarrow 1$) limit the corresponding statement was made in [@BBFLT Sec 4.1] (see also [@IMO],[@Ohkava]).
The second equality in can be viewed as a blow-up equation, similar to one studied in paper [@Nakajima:2005]. Its conformal limit coincides with first two formulas in [@KGP eq. 4.22] (using appropriate version of the AGT relation [@BF], [@BMT1]).
[99]{}
G. Andrews, R. Askey, R. Roy, *Special functions*, Cambridge University Press, (1999)
H. Awata, Y. Yamada, *Five-dimensional AGT Conjecture and the Deformed Virasoro Algebra*, JHEP **1001** (2010), 125 \[[[arXiv:0910.4431]{}](http://arxiv.org/abs/0910.4431)\]
A. Belavin, M. Bershtein, B. Feigin, A. Litvinov, G. Tarnopolsky, *Instanton moduli spaces and bases in coset conformal field theory*, Comm. Math. Phys. **319 1**, 269-301 (2013), \[[[arXiv:1111.2803]{}](http://arxiv.org/abs/1111.2803)\]
V. Belavin, B. Feigin, [*[Super Liouville conformal blocks from $\mathcal{N}=2$ $SU(2)$ quiver gauge theories]{}*]{}, JHEP **1107** (2011) 079, \[[[arXiv:1105.5800]{}](http://arxiv.org/abs/1105.5800)\].
M. Bershtein, A. Shchechkin, *Bilinear equations on Painlevé tau functions from CFT*, Comm. Math. Phys. **339**, (2015), 1021-1061; \[[[arXiv:1406.3008]{}](http://arxiv.org/abs/1406.3008)\].
M. Bershtein, A. Shchechkin, *Bäcklund transformation of Painlevé III($D_8$) tau function*, \[[[arXiv:1608.02568]{}](http://arxiv.org/abs/1608.02568)\].
G. Bonelli, A. Grassi, A. Tanzini, *Seiberg-Witten theory as a Fermi gas*, \[[[arXiv:1603.01174]{}](http://arxiv.org/abs/1603.01174)\].
G. Bonelli, A. Grassi, A. Tanzini, to appear.
G. Bonelli, K. Maruyoshi, and A. Tanzini, [*[Instantons on ALE spaces and Super Liouville Conformal Field Theories]{}*]{}, JHEP [**1108**]{} (2011) 056, \[[[arXiv:1106.2505]{}](http://arxiv.org/abs/1106.2505)\].
T. Eguchi, H. Kanno, *Topological strings and Nekrasov’s formulas* JHEP. **0312**, (2003) 006; \[[[arXiv:hep-th/0310235]{}](http://arxiv.org/abs/hep-th/0310235)\].
O. Gamayun, N. Iorgov, O. Lisovyy, *Conformal field theory of Painlevé VI*, JHEP **1210**, (2012), 38; \[[[arXiv:1207.0787]{}](http://arxiv.org/abs/1207.0787)\].
O. Gamayun, N. Iorgov, O. Lisovyy, *How instanton combinatorics solves Painlevé VI,V and III’s*, J. Phys. A: Math. Theor. **46** (2013) 335203 \[[[arXiv:1302.1832]{}](http://arxiv.org/abs/1302.1832)\].
B. Grammaticos, A. Ramani, *Parameterless discrete Painleve equations and their Miura relations*, J. Nonlin. Math. Phys. **23** (2016) 141.
B. Grammaticos, T. Tamizhmani, A. Ramani, A. S. Carstea, K.M.Tamizhmani, *A bilinear approach to the discrete Painleve I equations*, J. Phys. Soc. Japan **71** (2002) 443.
A. Iqbal, A. K. Kashani-Poor, *$SU(N)$ geometries and topological string amplitudes,* Adv. Theor. Math. Phys. **10 1** (2006), 1–32. \[[[arXiv:hep-th/0306032]{}](http://arxiv.org/abs/hep-th/0306032)\] .
N.Iorgov, O.Lisovyy, J.Teschner, *Isomonodromic $\tau$ functions from Liouville conformal blocks*, Commun. Math.Phys. **336(2)** (2015) 671-694. \[[[arXiv:1401.6104]{}](http://arxiv.org/abs/1401.6104)\]
Y. Ito, K. Maruyoshi, T. Okuda, *Scheme dependence of instanton counting in ALE spaces*, JHEP **5 45** (2013) \[[[arXiv:1303.5765]{}](http://arxiv.org/abs/1303.5765)\]
A. Its, O. Lisovyy, Yu. Tykhyy, *Connection problem for the sine-Gordon/Painlevé III tau function and irregular conformal blocks* IMRN (2015) **18** 8903-8924. \[[[`arXiv:1403.1235`]{}](http://arxiv.org/abs/1403.1235)\].
K. Kajiwara, M. Noumi, Y. Yamada, *Geometric Aspects of Painlevé Equations* \[[[arXiv:1509.08186]{}](http://arxiv.org/abs/1509.08186)\].
V. Mitev, E. Pomoni, M. Taki, F. Yagi, *Fiber-Base Duality and Global Symmetry Enhancement* \[[[arXiv:1411.2450]{}](http://arxiv.org/abs/1411.2450)\].
H, Nakajima, K. Yoshioka, *Instanton counting on blowup. II. $K$-theoretic partition function* Transformation Groups **10 3** (2015) 489–519 \[[[arXiv:math/0505553]{}](http://arxiv.org/abs/math/0505553)\].
N. Nekrasov, A. Okounkov, *Seiberg-Witten Theory and Random Partitions*, The unity of mathematics, 525–596, Progr. Math., **244**, Birkhäuser, Boston; \[[[arXiv:hep-th/0306238]{}](http://arxiv.org/abs/hep-th/0306238)\].
R. Ohkawa, *Wall-crossing between stable and co-stable ADHM data* \[[[arXiv:1506.06434]{}](http://arxiv.org/abs/1506.06434)\].
Y. Ohyama, H. Kawamuko, H. Sakai, K. Okamoto, *Studies on the Painlevé Equations, V, Third Painlevé Equations of Special Type PIII(D$_7$) and PIII(D$_8$)*,J. Math. Sci. Univ. Tokyo **13** (2006), 145–204 \[[[http://www.ms.u-tokyo.ac.jp/journal/pdf/jms130204.pdf]{}](http://www.ms.u-tokyo.ac.jp/journal/pdf/jms130204.pdf)\].
K. Okamoto, *Sur les feuilletages associés aux équations du second ordre a points critiques fixés de P. Painlevé*, Japan. J. Math. (N.S.) **5** (1979) 1–79.
N. Okubo, *Bilinear equations and q-discrete Painlevé equations satisfied by variables and coefficients in cluster algebras*, J. Phys. A: Math. Theor. **48** 355201 \[[[arXiv:1505.03067]{}](http://arxiv.org/abs/1505.03067)\].
H. Sakai, *Rational surfaces associated with affine root systems and geometry of the Painlevé equations* Comm. Math.Phys. **220(2)** (2001) 165-229.
H. Sakai, *Problem: Discrete Painlevé equations and their Lax forms* RIMS Kôkyûroku Bessatsu **B2** (2007) 195–208.
N. Seiberg, *Five dimensional SUSY field theories, non-trivial fixed points and string dynamics*, Phys. Lett. **B388** (1996) 753–760, \[[[arXiv:hep-th/9608111]{}](http://arxiv.org/abs/hep-th/9608111)\].
V. Spiridonov, *Elliptic hypergeometric functions* Appendix to the Russian edition of the book [@AAR] Russian edition: Moscow, MCCME (2013), 577–606; \[[[arXiv:0704.3099]{}](http://arxiv.org/abs/0704.3099)\].
A. Ramani, B. Grammaticos, T. Tamizhmani, *Quadratic relations in continuous and discrete Painleve equations*, J. Phys. A **33** (2000) 3033.
T. Tsuda, *Tau Functions of q-Painlevé III and IV Equations* Lett. Math. Phys. **75** (2006) 39–47
K.Ueno, M. Nishizawa, *The multiple Gamma functions and the multiple $q$-Gamma functions* \[[[arXiv:q-alg/9605002v2]{}](http://arxiv.org/abs/q-alg/9605002)\].
<span style="font-variant:small-caps;">Landau Institute for Theoretical Physics, Chernogolovka, Russia,\
Skolkovo Institute of Science and Technology, Moscow, Russia,\
National Research University Higher School of Economics, Moscow, Russia,\
Institute for Information Transmission Problems, Moscow, Russia,\
Independent University of Moscow, Moscow, Russia</span>
*E-mail*:**[email protected]**\
<span style="font-variant:small-caps;">National Research University Higher School of Economics, Moscow, Russia\
Skolkovo Institute of Science and Technology, Moscow, Russia,\
Bogolyubov Institute for Theoretical Physics, Kiev, Ukraine</span>
*E-mail*:**[email protected]**
[^1]: there is a misprint in formula for $s_0$ in [@SakaiCMP App. C, Mul 9]
[^2]: we are grateful to P. Gavrylenko for the discussion of this point
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We prove a formula for the structure sheaf of a quiver variety in the Grothendieck ring of its embedding variety. This formula generalizes and gives new expressions for Grothendieck polynomials. We furthermore conjecture that the coefficients in our formula have signs which alternate with degree. The proof of our formula involves $K$-theoretic generalizations of several useful cohomological tools, including the Thom-Porteous formula, the Jacobi-Trudi formula, and a Gysin formula of Pragacz.'
address: |
Massachusetts Institute of Technology\
Building 2, Room 248\
77 Massachusetts Avenue\
Cambridge, MA 02139
author:
- Anders Skovsted Buch
title: Grothendieck classes of quiver varieties
---
[^1] [^2] [^3]
[^1]: The author was partially supported by NSF Grant DMS-0070479
[^2]: [*2000 Mathematics Subject Classification*]{}: 14M15, 14M12, 19E08, 05E05
[^3]: [*Key words*]{}: Quiver varieties, $K$-theory, Grothendieck polynomials, Set-valued tableaux, Thom-Porteous formula, Jacobi-Trudi formula
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
Our mid-infrared survey of 124 white dwarfs with the [*Spitzer*]{} Space Telescope and the IRAC imager has revealed an infrared excess associated with the white dwarf WD 2115$-$560 naturally explained by circumstellar dust. This object is the fourth white dwarf observed to have circumstellar dust. All four are DAZ white dwarfs, [[i.e.,]{} ]{}they have both photospheric Balmer lines and photospheric metal lines.
We discuss these four objects as a class, which we abbreviate “DAZd”, where the “d” stands for “dust”. Using an optically-thick, geometrically-thin disk model analogous to Saturn’s rings, we find that the inner disk edges are at $\ga$ 0.1 to 0.2 $R_\sun$ and that the outer disk edges are $\sim$0.3 to 0.6 $R_\sun$. This model naturally explains the accretion rates and lifetimes of the detected WD disks and the accretion rates inferred from photospheric metal abundances.
author:
- 'Ted von Hippel, Marc J. Kuchner, Mukremin Kilic, Fergal Mullally, William T. Reach'
title: The New Class of Dusty DAZ White Dwarfs
---
Introduction
============
The first white dwarf known to have a dusty debris disk or cloud was Giclas 29-38, discovered by Zuckerman & Becklin (1987) to have pronounced excesses in the $K$, $L$, and $M$ bands relative to the white dwarf’s photosphere. Initially, it was unclear whether the IR excess indicated a substellar companion (Zuckerman & Becklin 1987) or particulate debris. Eventually the near-IR to mid-IR (Tokunaga, Becklin, & Zuckerman 1990; Chary, Zuckerman, & Becklin 1999) spectral energy distribution (SED), time-resolved photometry in the visible and near-IR that constrained the dust geometry (Graham [[et al.]{} ]{}1990; Patterson [[et al.]{} ]{}1991), and limits on the presence of companions from pulsation-timing studies (Kleinman [[et al.]{} ]{}1994) and speckle imaging (Kuchner [[et al.]{} ]{}1998), argued in favor of particulate debris in a disk. Now new [*Spitzer*]{} 7–14 $\mu$m spectroscopy (Reach [[et al.]{} ]{}2005b) shows a strong 10 $\mu$m emission feature in the G 29-38 excess that can only be caused by small silicate dust particles. If G 29-38 has any planets, they are currently undetected.
While we do not have strong evidence regarding the distribution of the dust around G 29-38, we do have a few constraints: the near-IR pulsation data suggest a non-spherical dust distribution (Graham [[et al.]{} ]{}1990; Patterson [[et al.]{} ]{}1991) and the particles we see in emission radiate at temperatures of 290 to 890 K (Reach [[et al.]{} ]{}2005b). Given the luminosity of the white dwarf (2 $\times 10^{-3}$ $L_\sun$), these temperatures places the dust particles at approximately 0.15 to a few solar radii (Zuckerman & Becklin 1987; Jura 2003; Reach [[et al.]{} ]{}2005b); the outer radius depends on whether we assume the dust is in an optically thin or thick component.
G 29-38 is a relatively normal, H-atmosphere (DA) WD currently passing through its instability strip, it has $T_{\rm eff} \approx$ 11800 K (Kepler & Nelan 1993; Bergeron, Wesemael, & Beauchamp 1995), log(g) = 7.9 (Koester & Wilken 2006) to 8.14 (Bergeron [[et al.]{} ]{}1995), and an implied mass = 0.56 to 0.69 M$_\sun$. It has been cooling as a WD for $\sim$0.5 Gyr. The initial-final mass relation of Weidemann (2000) suggests that this range of WD masses corresponds to a Zero Age Main Sequence (ZAMS) mass of 1.2 to 3.1 M$_\sun$. G 29-38 was therefore once an F or A star (Drilling & Landolt 2000, table 15.8). G 29-38 has photospheric absorption lines from heavy elements (Ca, Mg, Fe; Koester, Provencal, & Shipman 1997), which makes it a spectral type DAZ. Additionally, the photospheric Ca II K line varies in strength on timescales as short as 15 days (von Hippel & Thompson 2007), indicating episodic accretion.
Becklin [[et al.]{} ]{}(2005) and Kilic [[et al.]{} ]{}(2005) recently discovered a second white dwarf with circumstellar dust debris. This object, GD 362, was found to have a broad-band, $JHKL'M'$ SED that could be fit by a debris disk (Becklin [[et al.]{} ]{}2005). Its near-IR spectroscopy shows a strong $K$-band excess without any of the near-IR spectral features of a brown dwarf (Kilic [[et al.]{} ]{}2005). GD 362 is an H-rich[^1] atmosphere WD with metals observed in its photosphere; it is in fact the most metal-rich WD known. It is slightly cooler than G 29-38 ($T_{\rm eff}$ = 9740 K, Gianninas, Dufour, & Bergeron 2004; Kawka & Vennes 2004), yet is substantially more massive at 1.24 M$_\sun$ (Gianninas [[et al.]{} ]{}2004), which placed it at $>$ 7 M$_\sun$ on the main sequence (Weidemann 2000).
A third WD with circumstellar dust, GD 56, was just reported by Kilic [[et al.]{} ]{}(2006b). This star was discovered via its pronounced excess at wavelengths longer than 1.6 microns in low resolution near-IR spectroscopy. GD 56 is also a DAZ and has $T_{\rm eff}$ = 14400 K and log(g) = 7.8 (Koester [[et al.]{} ]{}2005). These reported atmospheric parameters correspond to a low mass WD (0.52 M$_\sun$), which is expected only for a halo star or a remnant from binary star common envelope evolution. On the other hand, if we assume a slightly higher surface gravity, the inferred mass becomes normal for a Galactic disk WD, [[e.g.,]{} ]{}for log(g) = 7.9, mass = 0.566 M$_\sun$ (Bergeron [[et al.]{} ]{}1995).
In this paper we report the fourth DAZ with circumstellar dust, WD2115$-$560. This WD has $T_{\rm eff}$ = 9700 K and log(g) = 8.1 (Koester [[et al.]{} ]{}2005), corresponding to a mass of 0.66 M$_\sun$. We discovered this object in our [*Spitzer*]{} IRAC 4.5 and 8 $\mu$m survey of 124 WDs (Mullally [[et al.]{} ]{}2007). We conducted this survey to improve upon the ground-based $K$-band survey of Zuckerman and Becklin (1992) and Farihi, Becklin, & Zuckerman (2005), and the 6.75 $\mu$m space-based, ISOCAM survey of Chary [[et al.]{} ]{}(1999) of 11 WDs, none of which found any new WDs with circumstellar dust. Our survey also aimed to study the photospheres of cool WDs (see Kilic [[et al.]{} ]{}2006a), which are of great importance to WD cosmochronology, as well as to search for giant planets and brown dwarfs (see Mullally [[et al.]{} ]{}2007).
Our discovery of the fourth WD with circumstellar dust leads us to suggest a new class of white dwarfs, the “DAZd” white dwarfs, where “d” indicates circumstellar dust. We choose a lower case “d” to avoid confusion with double white dwarfs, known as double degenerates and abbreviated “DD”. The classification notation for WDs is based on spectroscopic features, and up to now all have been notated by capital letters. Yet these features are all thought to be photospheric, whereas the dust causing the IR excess is circumstellar; a lower case “d” emphasizes the difference between photospheric and circumstellar spectroscopic properties.
Using the photometry and discovery statistics from our survey, the detailed properties of G 29-38 reported by Reach [[et al.]{} ]{}(2005b), and some simple models, we explore the nature and cause of particulate debris in DAZd WDs as a group.
Observations
============
[*Spitzer*]{} Data
------------------
Our [*Spitzer*]{} survey initially targeted all WDs in the McCook and Sion (1999, 2003) catalog brighter than $K_S$=15 as measured by 2MASS (Skrutskie [[et al.]{} ]{}2006), rejecting known binaries and planetary nebulae, yielding a total of 134 WDs. All of these WDs have well-determined positions and most have well-determined proper motions, facilitating cross-referencing between catalogs. We removed one object to avoid a conflict with the [*Spitzer*]{} Reserved Observations Catalog and the time allocation committee removed three other WDs in common with a program focused on DAZ WDs. Our observations for six of these WDs failed to provide quality photometry, mostly because of blending problems, and in the end, we obtained good data for a sample of 124 WDs with $T_{\rm eff}$ = 5000 to 60,000 K, cooling ages of 1 Myr to 7 Gyr, and representing most WD spectral types.
Our observations consisted of five dithers of 30 seconds each (150 seconds total integration time) for every WD in our survey. We used the products of the [*Spitzer*]{} Science Center pipeline, the Basic Calibrated Data (BCD) Frames and the Post-BCD frames (mosaics), for our analysis. We used IRAF[^2] PHOT as well as a custom IDL package to perform aperture photometry on individual BCD frames. We also attempted PSF-fitting photometry but found that the poorly defined Point Response Function for the IRAC instrument with its large pixels (1.213 and 1.220 arcseconds at 4.5 and 8 microns, respectively) meant that we obtained better results with the IDL and IRAF aperture photometry than with PSF-fitting photometry. Both the IDL and IRAF approaches gave identical results, within the errors. In order to maximize the signal-to-noise ratio, we used 5 pixel apertures for bright, isolated objects, and 2 or 3 pixel apertures for faint objects or objects in crowded fields. We corrected the resultant fluxes by the aperture correction factors determined by the IRAC team (see the IRAC Data Handbook). For each object, we compared the photometry from 2, 3, and 5 pixel apertures; we found them all to be consistent within the errors.
Following the standard IRAC calibration procedure, we made corrections for the observed array location of each WD before averaging the fluxes of the five frames in each IRAC band. We also performed photometry on the mosaic images and found the results to be consistent with the photometry from individual frames. Based on the calibrations of Reach [[et al.]{} ]{}(2005a), we expect that our IRAC photometry is calibrated to an accuracy of 3%. We estimated the photometric error from the observed scatter in the five images (corresponding to 5 dither positions) plus the 3% absolute calibration error, added in quadrature. See Mullally [[et al.]{} ]{}(2007) for further details of our reduction procedure.
Survey Summary
--------------
Our flux-limited survey of 124 single WDs detected circumstellar dust around two WDs, WD2115$-$560 and G 29-38, which met the criteria for inclusion in our sample. These two detections among the general population of white dwarfs indicate a detection rate of 1.6% in our flux limited WD sample. However, other surveys have shown higher circumstellar dust detection rates with more limited samples. Kilic [[et al.]{} ]{}(2006b) found the third circumstellar dust WD candidate, GD 56, in a small survey of DAZ WDs; they estimated that $\sim$9% of all DAZs have circumstellar dust detectable at $K$. These two samples differ both in wavelength coverage and target selection. The [*Spitzer*]{} survey was more sensitive to cooler circumstellar dust. The IRTF survey of Kilic [[et al.]{} ]{}(2006b) selected WD targets with high atmospheric metal content, the DAZs. The greater sensitivity to WDs with circumstellar dust in the [*Spitzer*]{} survey did not yield more such systems; the DAZ survey at the IRTF had a $\sim$5.6 times higher yield. This result, along with the fact that our new circumstellar dust system, WD2115$-$560, turned out to be a DAZ (a fact we did not know when we assembled the sample), suggests a deep connection between detectable circumstellar dust around WDs and the DAZ phenomenon.
Table 1 summarizes the properties of WD2115$-$560, along with the three other WDs hosting circumstellar dust already presented in the literature. These WDs span a modest temperature range of 9700 to 14400 K, though their mass range is wide for WDs, running from $\sim$0.56 to $\sim$1.24 M$_\sun$. This mass and $T_{\rm eff}$ range mean their cooling ages range from $\sim$0.2 to $\geq$ 2.5 Gyr. All four display photospheric metal lines. We list the temperature range of their circumstellar dust from a simple model fit, which we discuss below.
Rounding out the known properties of these WDs, Becklin [[et al.]{} ]{}(2005) and Reach [[et al.]{} ]{}(2005b) find that the fractional luminosity over all wavelengths in the dust is $\sim$3% in GD 362 and G 29-38. The minimum dust mass around G 29-38, assuming an optically-thin cloud of dust that radiates efficiently in the thermal infrared, is $10^{18}$ g, approximately equivalent to the mass of a 10 km asteroid.
Blackbody Models for the Spectral Energy Distributions
------------------------------------------------------
Perhaps G 29-38, the first known and best studied DAZd, is a good template for understanding this emerging class. Using archival IUE spectra, optical photometry from the literature, and [*Spitzer*]{} IRAC, IRS, and MIPS data, Reach [[et al.]{} ]{}(2005b) fit an optically thin dust cloud model to the G 29-38 SED. Reach [[et al.]{} ]{}(2005b) showed that a simple model consisting of two modified blackbodies, with temperatures of 290 and 890 K, provided a good fit to the infrared excess.
Their [*Spitzer*]{} spectra revealed a strong 10 $\mu$m emission feature, well fit by a combination of forsterite, Mg$_2$SiO$_4$, and olivine, (Mg,Fe)$_2$SiO$_4$. The assumptions of an optically-thin dust cloud and blackbody dust place the detectable dust at 1 to 5 $R_\sun$ from the WD. The Poynting-Robertson (P-R) timescale for particles to spiral onto the WD from a debris reservoir at 1 $R_\sun$ is only 4 $a$ years, where $a$ is the particle radius in microns. The silicate emission feature observed by Reach [[et al.]{} ]{}(2005b) can only be as strong as it is relative to the continuum (contrast = 125%), if it is emitted by submicron particles. The P-R timescale for such tiny particles is years or less.
Radiation pressure can eject small grains in orbit around a star ([[e.g.,]{} ]{}Wyatt & Whipple 1950). A spherical blackbody grain with radius $r_{grain}$ and density $\rho_{grain}$ (in g cm$^{-3}$) will be ejected from bound orbit around a WD by radiation pressure if $r_{grain} <
r_{blowout}$, where $$r_{blowout} = 1.15 \mu\rm{m} \ \rho_{grain} \ {{L_{WD}}\over{L_{\odot}}} \
{{M_{\odot}}\over{M_{WD}}}\ .$$ However, for a typical WD luminosity, $L_{WD} = 10^{-3} L_\sun$, and WD mass, $M_{WD} = 0.6 M_{\odot}$, the blowout size is 0.002 $\mu$m, much smaller than the grains producing the 10 $\mu$m feature.
Flat Disk Models for the Spectral Energy Distributions
------------------------------------------------------
An optically-thin dust cloud would have a lifetime no longer than the short P-R time scale. Therefore, it seems likely that the dust cloud is optically thick, shielding much of the dust from the stellar radiation that causes P-R drag. With this idea in mind, we fit the SEDs of each of the four WD debris systems using standard optically-thick flat disk models (Friedjung 1985; Jura 2003). We derived the input spectra for these simple disk models by fitting blackbody curves to the photospheric portions of the SEDs, and then used the models to derive disk inclinations and dust temperatures that correspond to the inner and outer radii of the disks.
For each WD, we created a grid of models with inner disk temperatures $T_{\rm in}$ = 600 to 1250 K, outer disk temperatures $T_{\rm out}$ = 200 to 1175 K, $T_{\rm in} > T_{\rm out}$, and a range of inclination angles. We then found the model that best fit the IR photometry of each WD (J band and beyond) by minimizing $\chi^2$. Table 1 lists the temperature ranges and inclinations of the best fit models.
For GD 56 and GD 362 we included the IRTF spectroscopy of Kilic [[et al.]{} ]{}(2005, 2006b) in the data to be fit. For G 29-38 we performed one fit using our [*Spitzer*]{} photometry and another fit using the IRTF (Tokunaga [[et al.]{} ]{}1999) plus ISO photometry (Chary [[et al.]{} ]{}1999), though without the 10.5 $\mu$m photometry, which is likely to be influenced by the strong silicate emission feature. The fit to the older IRTF and ISO photometry was poorer than the fit to the [*Spitzer*]{} data, and we report only the fit to the modern data in Table 1. The difference between the [*Spitzer*]{} and older data sets most likely reflects calibration errors that plagued the ground-based and particularly the ISO photometry (R. Chary and B. Zuckerman, private communication). It is also possible that the G 29-38 debris disk has evolved substantially over the 15 year span of these observations.
Figures 1–4 present our best fit debris disk models. All four DAZd WDs show a pronounced IR excess relative to the expected WD photospheres, represented by the WD model atmospheres (solid lines, kindly provided by Detlev Koester) The debris disk components (dashed lines) fit the near-IR and mid-IR excess well.
Table 1 also lists the inner and outer radii of the best fit flat disk models. They range from 0.15 to 0.58 $R_\sun$ = 13 to 49 white dwarf radii for the three normal mass WDs, and 0.08 to 0.50 $R_\sun$ = 17 to 106 white dwarf radii for GD 362, assuming the high mass value currently in the literature, which implies a much smaller radius. The disk model has no explicit inner edge directly exposed to radiation from the WD, and thus the inner radii are significantly underestimated.
A Possible Physical Description of the Disks
============================================
Perhaps the observations of DAZ disks all point to a common simple picture originally suggested by Jura (2003); the disks are analogs of planetary rings, physically thin disks of rocks and dust with optical depth $\approx$ 1. Our new data make this analogy much more compelling. As shown above, the physically thin, optically-thick disk geometry implied by this picture can match the observed SEDs of DAZd WDs. Besides providing an excuse for that disk geometry, the planetary-ring analogy has the following virtues.
1\. This picture naturally incorporates both small dust grains and larger, long-lived source particles.
Standard models of planetary rings ([[e.g.,]{} ]{}Cuzzi [[et al.]{} ]{}1979) contain a distribution of particle sizes. The largest ring particles contain most of a ring’s mass. These particles occupy a “monolayer” a few particle-widths thick, while the smaller particles—like the ones observed to produce the 10-micron emission feature in G29-38 (Reach [[et al.]{} ]{}2005b)—are spread over a thicker poly-layer. Planetary rings may also consist of dusty rings interspersed with rings of larger bodies ([[e.g.,]{} ]{}de Pater [[et al.]{} ]{}2006). The small particles are continually produced by the collisions of the larger particles, and protected to some degree from radiation forces by the optical thickness of the ring.
2\. This picture provides a disk lifetime consistent with the ages of the DAZd WDs.
An optically-thick disk of rocks has a lifetime set by the viscous spreading time of the ring, $t_{visc} \approx r^2/\nu$, where $r$ is the radius of the ring and $\nu$ is the kinematic viscosity. To estimate the disk lifetime, let us assume that the viscosity of the WD rings is similar to that of Saturn’s rings. The viscosity of a monolayer ring is $\nu
=\Omega \tau R_{\nu}$, where $\tau$ is the optical depth and $R_{\nu}$ is a representative particle size, weighted toward the larger particles (Goldreich & Tremaine 1982). The angular speed in the WD ring is similar to that in planetary rings: $\Omega \approx 2 \times 10^{-4}$ s$^{-1}$. So assuming a similar viscosity for the WD debris disks and Saturn’s rings is like assuming a similar optical depth and particle size distribution. With this assumption, the WD ring lifetimes are $\sim$100 times longer than the lifetime of Saturn’s rings because of their larger radii. This lifetime is easily $\gtrsim 10^9$ years, comparable to the median post-main sequence lifetime of the WDs in our survey. In this picture, WD rings do not require continual replenishment from a longer-lived source of small bodies.
3\. This picture implies a metal accretion rate consistent with the photospheric abundances of metals in the DAZd WDs.
In the context of this picture, we can estimate the mass of a typical DAZd disk as $\pi r_{out}^2 \Sigma$, using the surface density of Saturn’s rings ($\Sigma \approx 40$ g cm$^{-2}$; Tiscareno [[et al.]{} ]{}2007) and an outer radius $r_{out} \approx 1.5\ R_\sun$ (see item 5, below). With these assumptions, we find a representative disk mass of $\sim$2 $\times 10^{-4} M_{\bigoplus}$. If the accretion time is $10^{9}$ years, the resulting accretion rate onto the WD is $\sim$7 $\times 10^{-19}$ M$_\sun$ yr$^{-1}$.
Figure 5 shows the observed photospheric abundances derived by Koester & Wilken (2006) for a large sample of DAZ WDs. Figure 6 shows accretion rates for this sample inferred from these photospheric abundances and the atmospheric residence times Koester & Wilken (2006) derived for these WDs. These accretion rates from Koester & Wilken (2006) are based on an assumption that the accreted material has Solar composition. Our model assumes that the material is mostly refractory, so to compare our accretion rate to Figure 6, we need to multiply it by a factor of 50, giving us $\sim$3 $\times 10^{-17}$ M$_\sun$ yr$^{-1}$. This accretion rate sits right in the middle of the range of DAZ inferred accretion rates, and is only about an order of magnitude less than that of the known DAZds. It is easy to imagine a disk with slightly higher surface density or viscosity than Saturn’s rings that would precisely match the DAZd accretion rates in Figure 6.
4\. This picture explains the inner radii of DAZd disks inferred from their SEDs.
The radius where the equilibrium temperature of a dust grain exceeds the sublimation temperature, $T_{\rm sub}$, for the grains should mark the inner boundary of the ring. Although most of the disk sees only stellar radiation at an oblique angle, the inner radius of the disk can see direct stellar radiation, so it should be hotter than the flat-disk model would predict, closer to the temperature of an optically-thin dust cloud, or a single grain in free space. With this geometry in mind, we find that for blackbody grains, the sublimation radius is $$R_{sub, BB} = \frac{1}{2}\ \left(\frac{T_{sub}}{T_{\sun}}\right)^{-2}
\left(\frac{L_{WD}}{L_{\sun}}\right)^{1/2} R_{\sun},$$ where $L_{WD}$ is the WD luminosity. For a typical WD luminosity, $L_{WD}
= 10^{-3} L_{\sun}$, and assuming a silicate sublimation temperature, $T_{\rm sub} = 1500$ K, Equation 2 gives $R_{\rm sub} = 0.24\ R_\sun$.
Another component of the disks may be grains too small to efficiently emit at the blackbody peak wavelength. These grains are slightly hotter than blackbody grains at the same location in the nebula, so they sublimate at larger radii. E.g., for the “medium”-sized grains described by Backman & Paresce (1993), the sublimation radius is $$R_{sub, M} = \left(\frac{T_{sub}}{4009.21}\right)^{-5/2}
\left(\frac{L_{WD}}{L_{\sun}}\right)^{1/2}
\lambda_0^{-1/2} R_{\sun},$$ where $\lambda_0$ is the grain size measured in microns. For the same assumed WD luminosity and silicate sublimation temperature, and for a grain size of 1 $\mu$m, $R_{\rm sub, M} = 0.37\ R_\sun$.
Table 1, column 12 lists dust sublimation radii for each of the DAZds calculated from Equation 2 assuming $T_{\rm sub} = 2000$ K. This assumption produces a good match to the inner radii inferred from our models of the SEDs. However, we did not incorporate the physics of direct heating of the inner rim of the disk into our SED modeling, so we cannot compare sublimation radii calculated using the assumption of direct heating to the radii inferred from our flat disk models. We can, however, compare the inner temperatures inferred from our flat disk models to the sublimation temperatures of typical dust materials and retain self-consistency. Table 1, column 9 lists the temperatures of the inner and outer edges of the disks inferred from the flat disk models. The temperatures at the inner edge range from 800–1150 K.
5\. This picture explains the outer radii of DAZd disks inferred from their SEDs.
A dynamically cold ring cannot extend much beyond the Roche radius; beyond this radius, the ring particles coagulate into moons—or in the case of the white dwarf rings—asteroids or planets. The Roche radius for tidal disruption of an asteroid near a WD is given by Jura (2003): $$R_{\rm Roche} = C_{\rm tide}\ R_{\rm WD}\
\left(\frac{\rho_{\rm WD}}{\rho_{\rm asteroid}}\right)^{1/3},$$ where $C_{\rm tide}$ is a parameter of order unity that depends on the asteroid’s composition, orbit, and rotation, $R_{\rm WD}$ is the radius of the WD, $\rho_{\rm WD}$ is the density of the WD, and $\rho_{\rm
asteroid}$ is the density of the asteroid. For G 29-38, $R_{\rm Roche}
\approx 126\ C_{\rm tide}\ R_{\rm WD} \approx 1.5\ C_{\rm tide}\ R_\sun
\approx 1.5\ R_\sun$. Since this equation only includes density to the one-third power and WD masses span a narrow range ($\sim$0.55 to 1.2 M$_\sun$), all WDs have about the same Roche radii.
Jura (2003) pointed out that the outer radius of the disk around G29-38 roughly matched the Roche radius for this WD. We find that the outer radii of all the DAZd disks as inferred from physically-thin, optically thick disk models for the SEDs are consistent with the WD Roche radii.
6\. This picture may explain the absence of dust around luminous DAZ white dwarfs.
This [*Spitzer*]{} survey and the survey of Kilic [[et al.]{} ]{}(2006b) examined eleven DAZ white dwarfs hotter than 15,000 K and found circumstellar dust around none of them (see Figures 5 and 6). The high surface abundances and very short atmospheric residence times for calcium (see §4) in these eleven stars made them excellent candidates to host observable debris disks. Gänsicke [[et al.]{} ]{}(2006) discovered a single WD with effective temperature $\sim$22,000 K that hosts a calcium gas disk, but shows no sign of circumstellar dust. These observations suggest that perhaps hotter, more luminous WDs cannot host dust.
The analogy between WD disks and planetary rings suggests a natural interpretation of these observations; for hot WDs, the sublimation zone reaches beyond the Roche radius. These WDs, like main sequence stars, lack a circumstellar region where small bodies are stable against both sublimation and coagulation in dynamically-cold disks.
We can find the critical WD effective temperature above which this stable region disappears by setting $R_{\rm sub, BB}=R_{\rm Roche}$, assuming blackbody dust grains: $$T_{crit, BB} = \sqrt{2\ C_{\rm tide}}\
\left(\frac{\rho_{\rm WD}}{\rho_{\rm asteroid}}\right)^{1/6}
T_{sub}.$$ Ignoring the slow dependence of this expression on the WD density, we find the critical WD temperature is $T_{\rm crit, BB} \approx 16\ T_{\rm sub}$. If instead we use the 1 $\mu$m “medium” grains of Backman & Paresce (1993), we find that the maximum WD temperature for hosting a dust disk is $$T_{crit, M} = 0.13 \sqrt{\ C_{\rm tide}}\
\left(\frac{\rho_{\rm WD}}{\rho_{\rm asteroid}}\right)^{1/6}
T_{sub}^{5/4}.$$ so $T_{\rm crit, M} \approx 1.5\ T_{\rm sub}^{5/4}$.
For grains with $T_{sub}=1500$ K, we find $T_{crit, BB}=24,000$ K, and $T_{crit, M}=14,000$ K. We suggested above that the inner edges of the observed DAZd disks match blackbody grains with $T_{sub}=2000$ K. This combination implies that there should be no disks around WDs hotter than $T_{crit, BB}=32000$ K, keeping in mind the caveats described above. The observations described above suggest that the critical temperature should be roughly 15,000–22,000 K. These observations broadly agree with the hypothesis that the grain sublimation temperature and the Roche radius together limit the kinds of WDs that can host dust disks, although clearly details remain to be better understood.
The Connection Between WD Debris Systems and the DAZ Phenomenon
===============================================================
As we mentioned above, the presence of photospheric metals in DAZs implies ongoing accretion of these metals. Our physical model for the DAZd disks provides a natural explanation for this accretion. Let us examine in more detail the connection between DAZd disks and the DAZ phenomenon.
White dwarfs are classified as DAZ or DBZ if their optical spectra contain metal lines along with much stronger hydrogen or helium lines, respectively. White dwarfs showing metal lines may also be classified simply DZ if they are too cool to show hydrogen or helium lines. Hybrid types, such as DABZ, also exist.
Since the discovery of metal lines in WD atmospheres (see Weidemann 1958; Greenstein 1960), the origin of these metals has been debated ([[e.g.,]{} ]{}Lacombe [[et al.]{} ]{}1983; Aannestad [[et al.]{} ]{}1993; Dupuis, Fontaine, & Wesemael 1993; Zuckerman & Reid 1998; Zuckerman [[et al.]{} ]{}2003; Koester & Wilken 2006; Kilic & Redfield 2007). The high surface gravity of WDs should cause the denser atoms of the metals to sink in the atmosphere on timescales of days to millions of years, depending on the atmospheric constituents, effective temperature, and WD mass. These timescales are all much shorter than the ages of most WDs.
Figure 7 shows the timescale for gravitational settling for DA WDs with log(g) = 7.75 to 9.0, roughly mass = 0.48 to 1.20 M$_\sun$, taken from Koester & Wilken (2006). For comparison, the four DAZd stars are plotted on this figure based on their atmospheric $T_{\rm eff}$ and log(g) values. Figure 7 shows that for the WDs hotter than 11,000 K, the settling times are substantially less than a year and even for our coolest, low surface gravity object, WD2115$-$560, the settling time is still less than 200 years.
These short settling timescales imply that DAZ stars were very recently accreting heavy elements. Interstellar accretion, perhaps the most discussed scenario for supplying the heavy elements, seems to fail by orders of magnitude for many of the known DAZ WDs (Aannestad [[et al.]{} ]{}1993; although see the counter-argument by Koester & Wilken 2006). However, such accretion would be natural if DAZ WDs typically harbor substantial circumstellar disks.
So far all debris WD systems with circumstellar disks have detectable photospheric metal lines. Is it possible that all metal-rich WDs (DAZ/DBZ/DZ) accrete their metals via debris disks?
One reason answering this question is difficult is that not all WDs with photospheric metals can be recognized as DAZ/DBZ/DZs. This point is made in Figure 5, where the curve indicating a constant Ca equivalent width of 15 mÅ corresponds to much higher Ca abundances for hotter WDs than for cooler WDs. It is much more difficult to recognize a DAZ at higher stellar temperatures. Figure 6 shows another view of the same data, easier to grasp visually, but more model-dependent. The maximum accretion rates necessary to explain the observed abundances are essentially independent of $T_{\rm eff}$; the short atmospheric residence timescales for the hot WDs are balanced by the deeper convective reservoirs for the cooler WDs.
The DAZd WDs are located near the top of Figures 5 and 6; they are among the WDs with the highest photospheric heavy element abundances and the highest inferred accretion rates, and therefore should harbor the most massive and easiest to detect of the debris disks. It thus seems possible that the debris disks found to date display an observational selection effect, and that the majority of DAZs harbor debris disks.
For all DAZd WDs except the possibly massive GD 362, Koester & Wilken (2006) find accretion rates in the narrow range of 4.07–8.13 $\times
10^{-16}$ M$_\sun$ yr$^{-1}$. Koester & Wilken (2006) assumed that the accreted material had solar abundances. We assume instead that the accreted material lacks hydrogen and helium, reducing the total accretion rate by a factor of 50. This lower limit corresponds to 8 to 16 $\times 10^{-18}$ M$_\sun$ yr$^{-1}$ or 1.6 to 3.2 $\times 10^{16}$ g yr$^{-1}$ of refractory elements fed to the WD, a rate consistent with viscous accretion in our planetary-ring model. If this accretion persists for at least 25% of the lifetime of the WDs, commensurate with the fraction of DAs that are DAZs (Zuckerman [[et al.]{} ]{}2003), it corresponds to an accreted mass of $\ga 7 \times 10^{-4}$ of an Earth mass accreted over 1 Gyr.
Origin of the Debris
--------------------
We have discussed the basic properties of the debris systems found to date and we have connected the debris observable in the infrared to the photospheric metal lines observable in the optical. We have argued that dust debris is common around WDs and that this dust must be steadily accreting onto the WD. We have not yet examined possible sources for the debris around WDs. Here we consider possible origins: WD mergers, debris left during the asymptotic giant branch (AGB) or proto-planetary nebulae (pPNe) phases, and accretion of dust derived from planetary system bodies.
### White Dwarf Mergers
Merging WDs are expected to leave behind disks (Livio, Pringle, & Wood 2005; Hansen, Kulkarni, & Wiktorowicz 2006) to shed angular momentum. According to the calculations of Livio [[et al.]{} ]{}(2005), such disks should be massive ($\sim$0.007 M$_\sun$), extend to 1 AU or more, and predominantly composed of carbon and oxygen, assuming the merger of two WDs each with mass $\le$ 0.7 M$_\sun$. The minimum mass for most of these merged WDs ought to be $\ge$ 1.1 M$_\sun$, since it takes about 10 Gyr to produce a 0.55 M$_\sun$ WD.
Except for GD 362, with a mass possibly equal to 1.24 M$_\sun$, these mass constraints rule out this explanation for the majority of DAZd WDs. The other WDs with circumstellar dust have masses ranging from 0.52 to 0.69 M$_\sun$. Additionally, the disk mass and radial extent appear inconsistent with all DAZd stars discovered to date. While WD mergers must be taking place, and they may possibly be the source of the disk around GD 362, they are not the source of the majority of the observed DAZd systems.
### Late Stages of Stellar Evolution
The outflows of late-stage asymptotic giant branch (AGB) stars and proto-planetary nebulae (pPNe) are often highly non-spherical. This process is not understood, and a range of models have been proposed for the origin of the asymmetries, including stellar companions and magnetic fields (see Balick & Adams 2002, and references therein). There is circumstantial evidence for the formation of disks and tori during the AGB to post-AGB star transition, albeit at much greater distances from the host star, and most pPNe are producing or have recently produced dust particles (Balick & Adams 2002). In particular, Waters [[et al.]{} ]{}(1998) observed olivines and pyroxenes around a pre-white dwarf. While most of the outflows are ejected well above the stars’ escape velocities, at present it is not known whether some small fraction of these outflows remain in orbit after they are ejected. Remaining stellar material is a possible explanation for the dust disk discovered by Su [[et al.]{} ]{}(2007) around (at 35–150 AU) the Helix planetary nebula. While Su [[et al.]{} ]{}attribute the observed dust to disrupted Kuiper Belt objects or disrupted comets, it could also be the result of stellar material ejected with less than the escape speed.
This scenario predicts dusty environments for the youngest WDs with less and less dust as WDs age. The observations, however, run counter to this. All four of the known DAZd systems have old host WDs with ages of $\geq$ 0.2 to $\geq$ 2.5 Gyr and no hot, dusty WDs have been discovered. The metal disk WD found by Gänsicke [[et al.]{} ]{}(2006) pushes this to slightly younger objects, $\sim$0.1 Gyr, but still, if the late stages of stellar evolution commonly leave debris disks at a range of radii, these debris disk should be detectable in our survey and in IUE observations of hot WDs. Yet, to date, no such systems have been seen (Hansen [[et al.]{} ]{}2006; Kilic [[et al.]{} ]{}2006b).
### Planetary System Remnants
The dust around DAZds may represent the remains of planetary systems. Roughly 5-10% of A-type stars have substantial debris disks ([[e.g.,]{} ]{}Backman & Paresce 1993) and precise Doppler surveys find giant planets around roughly 5-10% of solar-type stars ([[e.g.,]{} ]{}Fischer & Valenti 2003). Perhaps 5-10% or more of WD progenitors had planetary systems, and some small bodies in these systems survived the post-main sequence evolution of their host stars (Debes & Siguardsson 2002, Jura 2003). Our observations and models appear to be consistent with this picture for the origin of all the DAZd disks.
Conclusion
==========
We presented the properties of WD2115$-$560, a new WD with infrared excess we discovered in our [*Spitzer*]{} IRAC photometry, and the fourth DAZ seen to have an infrared excess well-described by circumstellar dust debris. We examined the dusty DAZ WDs as a class, which we refer to with the letters “DAZd”.
Using a simple, flat, optically-thick disk model motivated by the geometry of Saturn’s rings, we find that the dust in these four DAZd WDs reside in disks ranging in inner temperature from $\sim$800 to 1150 K and outer temperature from $\sim$200 to 725 K, not taking into account the direct heating of the inner edges of the disks. The outer disk temperatures are likely to be lower limits set by the sensitivity of the observations and the amount of dust radiating at that temperature. These dust temperatures imply that the inner disk edges are at $\ga$ 0.1 to 0.2 $R_\sun$ and the detectable extent of the outer disks are $\sim$0.3 to 0.6 $R_\sun$. The close circumstellar locations of these debris disks are consistent with simple sublimation calculations and with the Roche radius for tidal disruption of an asteroid near a WD (Jura 2003).
We find that, taken together, the DAZd WDs point to a disk model analogous to planetary rings. This model naturally explains the disk accretion rates and lifetimes and the temperature distribution of the dusty WDs. We favor the interpretation that most of these disks represent planetary systems remnants.
This work is based in part on observations made with the Spitzer Space Telescope, which is operated by the Jet Propulsion Laboratory, California Institute of Technology under NASA contract 1407. Support for this work was provided by NASA through award project NBR 1269551 issued by JPL/Caltech to the University of Texas.
Aannestad, P. A., Kenyon, S. J., Hammond, G. L., & Sion, E. M. 1993, , 105, 1033
Backman, D. E. & Paresce, F. 1993, in Protostars and Planets III, Main-sequence stars with circumstellar solid material - The Vega phenomenon, p 1253
Balick & Adams 2002, , 40, 439
Becklin, E. E., Farihi, J., Jura, M., Song, I., Weinberger, A. J., & Zuckerman, B. 2005, , 632, L119
Becklin, E. & Zuckerman, B. 1988, , 336, 656
Bergeron, P., Wesemael, F., & Beauchamp, A. 1995, , 107, 1047
Chary, R., Zuckerman, B., & Becklin, E. E. 1999, in The Universe as Seen by ISO, ed. P. Cox & M. F. Kessler (ESA-SP 427; Garching: ESA), 289
Cuzzi, J. N, Burns, J. A., Durisen, R. H. & Hamill, P. M. 1979, , 281, 202
Debes, J. & Siguardsson, S. 2002, , 572, 556
de Pater, I., Hammel, H. B., Gibbard, S. G., & Showalter, M. R. 2006, Science, 312, 92
Drilling, J. S., & Landolt, A. U. 2000, in Allen’s Astrophysical Quantities, Fourth Edition, ed. A. N. Cox, (Springer-Verlag: New York), p. 389
ibid, p. 395
Dupuis, J., Fontaine, G., & Wesemael, F. 1993, , 87, 345
Farihi, J., Becklin, E. E., & Zuckerman, B. 2005, , 161, 394
Friedjung, M. 1985, , 146, 366
Gänsicke, B.T., Marsh, T.R., Southworth, J., & Rebassa-Mansergas, A. 2006, Science, 304, 1908
Gianninas, A., Dufour, P., & Bergeron, P. 2004, , 617, L57
Gliese, W., & Jahreiss, H., 1991, Catalog of Nearby Stars, (Astron. Rech. Inst., Heidelberg)
Goldreich, P. & Tremaine, S. 1982, , 20, 249
Graham, J. R., Matthews, K., Neugebauer, G., & Soifer, B. T. 1990, ApJ, 357, 216
Greenstein, J. L. 1960, in Stars and Stellar Systems, vol. 6, Stellar Atmospheres, ed. J. L. Greenstein, (Chicago: University of Chicago Press), p. 692
Hansen, B. M. S., Kulkarni, S., & Wiktorowicz, S. 2006, , 131, 1106
Jura, M. 2003, , 584, L91
Kepler, S. O., & Nelan, E. P. 1993, , 105, 608
Kilic, M., & Redfield, S. 2007, , in press
Kilic, M., von Hippel, T., Leggett, S. K., & Winget, D. E. 2005, , 632, L115 Kilic, M., von Hippel, T., Leggett, S. K., & Winget, D. E. 2006b, , 646, 474 Kilic, M., von Hippel, T., Mullally, F., Reach, W. T., Kuchner, M. J., Winget, D. E., Burrows, A., & Saumon, D. 2006a, , in press Kleinman, S. J. [[et al.]{} ]{}1994, , 436, 875
Koester, D., Provencal, J., & Shipman, H. L. 1997, , 320, L57
Koester, D., Rollenhagen, K., Napiwotzki, R., Voss, B., Christlieb, N., Homeier, D., & Reimers, D. 2005, , 432, 1025
Koester, D., & Wilken, D. 2006, , 453, 1051
Kuchner, M. J., Koresko, C. D. & Brown, M. E. 1998, , 508, L81
Lacombe, P., Wesemael, F., Fontaine, G., & Liebert, J. 1983, , 272, 660
Livio, M., Pringle, J. E., & Wood, K. 2005, , 632, L37
McCook, G. P. & Sion, E. M. 1999, , 121, 1
McCook, G. P. & Sion, E. M. 2003, VizieR On-line Data Catalog: III/235
Mullally, F., Kilic, M., Reach, W. T., Kuchner, M. J., von Hippel, T., Burrows, A., & Winget, D. E. 2007, , in press
Patterson, J., Zuckerman, B., Becklin, E. E., Tholen, D. J., & Hawarden, T. 1991, , 374, 330
Skrutskie, M. F., [[et al.]{} ]{}2006, AJ, 131, 1163
Reach, W. T., [[et al.]{} ]{}2005a, , 117, 978 Reach, W. T., Kuchner, M. J., von Hippel, T., Burrows, A., Mullally, F., Kilic, M., & Winget, D. E. 2005b, , 635, L161 Tiscareno , M. S., Burns, J. A., Nicholson, P. D., Hedman, M. & Porco, C. C. 2007, Icarus, in press, (astro-ph/0610242)
Tokunaga, A. T., Becklin, E. E., & Zuckerman, B. 1990, , 358, L21
Van Altena W. F., Lee J. T., & Hoffleit E. D. 1995, The General Catalogue of Trigonometric Stellar Parallaxes, Fourth Edition (Yale University Observatory)
von Hippel, T., & Thompson, S. E. 2007, ApJ, in press
Waters, L. B. F. M., [[et al.]{} ]{}1998, , 391, 868
Weidemann, V. 1958, , 70, 466
Weidemann, V. 2000, , 363, 647
Wyatt, S. P., & Whipple, F. L. 1950, , 111, 134
Zuckerman, B. & Becklin, E. E. 1987, , 330, 138
Zuckerman, B. & Becklin, E. E. 1992, , 386, 260
Zuckerman, B., Koester, D., Reid, I. N., & Hünsch, M. 2003, , 596, 477
Zuckerman, B., & Reid, I. N. 1998, , 505, L143
[lcrccccccccc]{} GD 56 & DAZd & 14,400 & 0.52–0.57 & 0.9–1.4 & 0.2–0.4 & 4.2% & 67 & 1125–525 & 0.21–0.58 & 45 & 0.36\
G 29-38 & DAZd & 11,600 & 0.56–0.69 & 1.2–3.1 & 0.5 & 3.5% & 14 & 1150–725 & 0.15–0.28 & 45 & 0.18\
GD 362 & DAZd & 9740 & 1.24 & $>$7.0 & $>$2.5 & 6.1% & 24 & 800–200 & 0.08–0.50 & 60 & 0.07\
WD2115$-$560 & DAZd & 9700 & 0.66 & 2.8 & 0.9 & 0.9% & 22 & 900–550 & 0.17–0.32 & 80 & 0.16\
f1.eps
f2.eps
f3.eps
f4.eps
f5.eps
f6.eps
f7.eps
[^1]: D. Koester, private communication, reports that GD 362 has a mixed He/H atmosphere and that the prior mass estimate is substantially too high. This will affect the timescale of gravitational settling and thus the implied accretion rate, but not our analysis.
[^2]: IRAF is distributed by the National Optical Astronomy Observatory, which is operated by the Association of Universities for Research in Astronomy (AURA), Inc., under cooperative agreement with the National Science Foundation.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We investigate the transmission properties of a spin transistor coupled to two quantum point contacts acting as spin injector and detector. In the Fabry-Perot regime, transport is mediated by quasibound states formed between tunnel barriers. Interestingly, the spin-orbit interaction of the Rashba type can be tuned in such a way that nonuniform spin-orbit fields can point along distinct directions in different points of the sample. We discuss both spin-conserving and spin-flipping transitions as the spin-orbit angle of orientation increases from parallel to antiparallel configurations. Spin precession oscillations are clearly seen as a function of the length of the central channel. Remarkably, we find that these oscillations combine with the Fabry-Perot motion giving rise to quasiperiodic transmissions in the purely one-dimensional case. Furthermore, we consider the more realistic case of a finite width in the transverse direction and find that the coherent oscillations become deteriorated for moderate values of the spin-orbit strength. Our results then determine the precise role of the spin-orbit intersubband coupling potential in the Fabry-Perot-Datta-Das intermixed oscillations.'
author:
- 'M. I. Alomar'
- Llorenç Serra
- David Sánchez
bibliography:
- 'spin\_transistor.bib'
title: 'Interplay between resonant tunneling and spin precession oscillations in all-electric all-semiconductor spin transistors'
---
Introduction
============
Spin transistors operate under the action of a spin-orbit coupling potential that rotates the electronic spin traveling along a narrow channel [@dat90]. Semiconductor heterostructures offer the possibility of generating spin-orbit interactions due to inversion asymmetry (Rashba type [@ras60]), thus rendering semiconductor spintronics a rewarding area for spin information processing applications [@fab07; @ber15]. Importantly, the strength of the spin-orbit coupling can be tuned with an external electric field [@nit97; @eng97], which provides the necessary gate tuning of the transistor switching mechanism. The last ingredient is the ability to both inject and detect spin polarized currents. This can be done by attaching ferromagnetic terminals to the semiconductor channel. Yet a series conductivity mismatch owing to unequal Fermi wavevectors can hamper the system functionality [@sch00; @ras00; @fer01]. Although spin precession oscillations have been detected in ferromagnetic-semiconductor junctions [@koo09] employing nonlocal voltage detection [@jed02], the spin-injection efficiency between dissimilar materials tends to be low. The system performance can also be affected due to the presence of multiple channels [@mir01; @jeo06; @gel10], additional rotation of the spin of the traversing electron induced by intersubband coupling [@egu03], the destructive effect of spin decoherence [@she05; @nik05; @xu14], the influence of gating [@sun11; @woj14], and the fact that the system can behave as a two-dimensional spin transistor [@pal04; @agn10; @zai11; @gel11; @alo15].
An interesting alternative has very recently been put forward by Chuang *et al.* [@chu15]. A pair of quantum point contacts (QPCs) works as spin injectors and detectors [@deb09; @now13]. The electric confinement in the point constrictions leads to an effective magnetic field that polarizes the electrons in directions perpendicular to the spin-orbit field present in the central channel. As a consequence, the detector voltage becomes an oscillatory function of the middle gate voltage applied to the two-dimensional electron gas. Importantly, the system is fully nonmagnetic (neither ferromagnetic contacts nor external magnetic fields are needed for the operation principle) and relies on a semiconductor-only structure. This is an appealing feature that has been pursued in different proposals [@sch03; @hal03; @wan03; @aws09; @wun10; @liu12].
Consider the case when the conductance of both quantum point contacts is set below the value corresponding to a fully open mode. Then, the waveguide potentials can be described as tunnel barriers and transport across them occurs via evanescent states [@ser07; @sab07]. Effectively, the system electronic potential is globally seen as a double barrier with a quantum well of variable depth. It is well known that these potential landscapes in general support the presence of resonant scattering due to Fabry-Perot-like oscillations arising from wave interference between the tunnel barriers. But at the same time we have spin-orbit induced oscillations due to the precession of spins traveling between the barriers. Therefore, one would naturally expect a competition between resonant tunneling and spin precession oscillations in a system comprising two serially coupled QPCs. Below, we show that this is indeed the case and that the combination of both oscillation modes leads to rich physics not only in the strictly one-dimensional case but also when more realistic samples with a finite transversal width are studied.
The subject of resonant tunneling effects and spin-orbit fields has been investigated in a number of works giving rise to interesting predictions. For instance, Voskoboynikov *et al.* find that the transmission probability significantly changes in the presence of the Rashba coupling [@vos99] while de Andrada e Silva *et al.* obtain spin polarizations for an unpolarized beam of electrons impinging on a double-barrier nanostructure [@dea99]. Koga *et al.* analyze spin-filter effects in triple barrier diodes [@kog02] whereas Ting and Cartoixà examine the double barrier case [@tin02]. The dependence of the electronic tunneling on the spin orientation is treated by Glazov *et al.* [@gla05]. These structures suffer from phase-breaking effects, as shown by Isić *et al.* [@isi10].
![(Color online) Pictorial representation of our system. A semiconductor layer (light gray) with metallic electrodes (blue) shows two quantum point contacts in a series (QPC1 and QPC2) and a two-dimensional cavity in between (QW). The spin-orbit coupling differs in each area ($\alpha_1$ and $\alpha_2$) due to distinct electric fields applied to the electrodes (lateral in the metallic electrodes, perpendicular in the QW). $L_1$ and $L$ correspond to the width of QPCs and central region, respectively.[]{data-label="Fig:sketch"}](1p.eps){width="45.00000%"}
In our work, we consider a purely ballistic system. Scattering is elastic and the transmission probabilities are determined within the quantum scattering approach. Scattering can take place at the interfaces between the quantum point contacts and the quantum well or due to interaction between the spins and the spin-orbit interaction. Importantly and in contrast to previous works investigating spin transistor transport properties, the spin-dependent transmission depends on the relative angle between the spin-orbit fields in the QPCs. This is an excellent property that allows us to tune the spin direction of the electrons impinging on the quantum well [@chu15]. For a null relative angle, within a pure one dimensional model we find that whereas the spin-conserving transmission shows resonant tunneling peaks as a function of the spin-orbit strength the spin-flip transmission always vanishes. Furthermore, for both types of transmissions the spin precession oscillations as a function of the spin-orbit strength in the quantum well appear only when the QPCs have effective spin-orbit magnetic fields with an angle that differs from the spin-orbit coupling in the well. This effect can be also seen when the quantum well length is varied. However, we point out that the QPCs have an additional effect as tunnel barriers that lead to Fabry-Perot resonances which can compete with the Datta-Das oscillations in the transmission curves yielding quasiperiodic patterns. Now, since a realistic sample has a finite width, we also consider a quasi-one dimensional system, in which case the spin-orbit intersubband coupling potential must be also taken into account. Remarkably, we find that our results derived from the one-dimensional model are also observable in two dimensions for moderately low values of the spin-orbit strength. This implies that the oscillation interplay discussed here can be probed with today’s experimental techniques.
The content of our paper is structured as follows. Section \[sec:mod\] describes the system under consideration in two dimensions: a semiconductor layer with two quantum point contacts in series and a spatially inhomogeneous spin-orbit interaction applied on the QPCs and central region. The strict one-dimensional limit is addressed in Sec. \[sec:1D\] where we have a double barrier potential modeling the two QPCs. We determine the eigenenergies and eigenfunctions in each region and, using matching methods, we find the transmission probabilities for a fixed incident spin. We perform an analysis of the transmission oscillations as a function of the relative orientation between the QPC effective magnetic fields and the spin-orbit interaction in the well, the strength of the spin-orbit coupling and the width of the middle cavity. We stress that, depending on the direction of the spin polarization in the QPC regions, the transitions are dominated by processes that conserve or flip the spin direction. We also observe the combined effect of Datta-Das and Fabry-Perot oscillations and obtain their characteristic frequencies. We find that modifying the strength of the spin-orbit coupling and the width of central region we can control the transmission probability for each spin. Section \[sec:quasi1d\] contains our analysis of the quasi-one-dimensional case. This discussion is important because it quantifies the role of spin-orbit intersubband coupling effects in both the Fabry-Perot and the Datta-Das oscillation modes. Finally, our conclusions are summarized in Sec. \[sec:conc\].
Theoretical model {#sec:mod}
=================
We consider a semiconductor layer partitioned into five different regions as in Fig. \[Fig:sketch\]: two reservoirs, two QPCs and a quantum well (QW). The blue areas are gate electrodes that form constrictions in the QPC1 and QPC2 between the left and right reservoir and the central well. We take $x$ as the transport direction. The spin-orbit potentials acting on the QPCs (both with strength $\alpha_1$) and the QW (strength $\alpha_2$) are in general different [@chu15]. Thus, our Hamiltonian reads $$\begin{aligned}
\mathcal{H}&=&\mathcal{H}_{0}+\mathcal{H}_{SO1}+\mathcal{H}_{SO2}\,,\label{eq:H}\\
\mathcal{H}_{0}&=&\frac{p_x^2+p_y^2}{2 m_0} +V(x,y)\,,\label{eq:H0}\\
\mathcal{H}_{SO1}&=&\frac{\alpha_1}{\hbar}\left[\left(\vec{\sigma}\times\vec{p}\right)_z \cos \phi +\left(\vec{\sigma}\times\vec{p}\right)_y \sin \phi \right]\,,\label{eq:Hso1}\\
\mathcal{H}_{SO2}&=&\frac{\alpha_2}{\hbar}\left(\vec{\sigma}\times\vec{p}\right)_z \,,\label{eq:Hso2}\end{aligned}$$ where $\mathcal{H}_{0}$ represents the free part of the total Hamiltonian $\mathcal{H}$, with $p_i=-i\hbar\partial/\partial_i$ ($i=x,y$) the linear momentum operator and $m_0$ the conduction-band effective mass of the electrons in the semiconductor heterostructure. $V(x,y)$ confines electrons in the (transversal) $y$ direction and includes in $x$ two identical constrictions that define an intermediate region (the cavity or well) of length $L$. The spin-orbit terms of $\mathcal{H}$ are $\mathcal{H}_{SO1}$ and $\mathcal{H}_{SO2}$, where the first (second) is active on the QPCs (QW) only. Here, $\vec{\sigma}=(\sigma_x,\sigma_y,\sigma_z)$ and $\vec{p}=(p_x,p_y,0)$ are the Pauli matrices and the momentum vector, respectively. In the central region, the $\alpha_2$ spin-orbit field \[Eq. \] arises from the confining electric field perpendicular to the QW plane (the $z$-direction). In the constrictions, there exists in the $\alpha_1$ spin-orbit potential \[Eq. \] an additional contribution from the lateral electric field applied to the QPCs along $y$. This field couples asymmetrically to the electrodes in Fig. 1 (blue areas) and, as a consequence, a high spin-orbit interaction emerges in the QPCs, as experimentally demonstrated in Refs. [@deb09; @chu15]. The spin-orbit strength can be further enhanced by electron-electron interactions, doping potentials or exchange correlations [@cal08; @det14]. Our goal is not to describe these effects microscopically but rather focus on the transport properties. Hence, we lump these effects into the parameter $\alpha_1$, which can be tuned with the lateral electric field [@deb09].
A convenient way of quantifying the strength of the two different components present in the QPCs (due to either lateral or perpendicular electric fields) is with the definition in Eq. of the angle $\phi$. Therefore, we can turn off the lateral contribution by setting $\phi=0$ in which case $\mathcal{H}_{SO1}$ and $\mathcal{H}_{SO2}$ are identical except for the spin-orbit strength. For $\phi=\pi/2$ the lateral electric field contribution to the spin-orbit potential dominates over that of the perpendicular electric field. Thus, the ensuing spin-orbit field in $\mathcal{H}_{SO1}$ is orthogonal to that in $\mathcal{H}_{SO2}$. This ability to manipulate the orientation of the spin-orbit fields is crucial for the working principle of our system and has been proven in the experiments reported earlier [@chu15]. It is a property that makes this device unique and that is absent from previous spin transistor studies. Another advantage of the QPCs is to reduce the wavevector spread of injected electrons in contrast to extended interfaces [@sch03]. Spin injection and detection with QPCs have been discussed in Refs. [@fro09; @hac14] in the context of ballistic spin resonance. Here, we do not consider any external magnetic field and all the spin dynamics originates from the effective magnetic fields due to the spin-orbit interactions present in the system, which makes our system an all-electric spin transistor.
![(Color online) Energy diagram of our system. The QPCs are described with barrier potentials of height $V_0$ and width $L_1$ whereas the size of the central region is denoted with $L$. We also plot the energy spectra in each region. Due to the spin-orbit coupling the bands structure undergoes a spin splitting and an energy downshift $E_{SO}$.[]{data-label="Fig:esc"}](2p.eps){width="45.00000%"}
One-Dimensional case {#sec:1D}
====================
Let us for the moment disregard transverse channel effects and consider a purely one-dimensional model. We expect that this is a good approximation when the point contacts support evanescent states only. We will later discuss the more realistic case where the electronic waveguides have a nonzero transversal width. In this limit we describe the QPCs electrostatic potential, $V(x,y)$, with a double tunnel barrier of width $L_1$ and height $V_0$ and the in-between cavity with a quantum well of length $L$ and bottom aligned with that of the reservoirs energy bands, see the sketch in Fig. \[Fig:esc\]. We then set $p_y=0$ in Eq. . Since the potential is piecewise constant, the eigenstates of $\mathcal{H}$ are readily found for the five regions defined in Fig. \[Fig:esc\]: $$\begin{aligned}
\Psi^0_{\ell s}(x)\!& \equiv& \!\Psi^I_{\ell s}\!=\Psi^{V}_{\ell s}\!=\!\frac{1}{\sqrt{2}}\!\left(\!\!\begin{array}{c}
\sqrt{1\!+\!s\,\sin \phi}\\
-i s \sqrt{1\!-\!s\,\sin \phi}
\end{array}\!\!\right)e^{i k^{(0)}_{\ell} x},\label{eq:f0}\\
\Psi^1_{\ell s}(x)\!& \equiv& \!\Psi^{II}_{\ell s}\!=\!\Psi^{IV}_{\ell s}\!=\! \frac{1}{\sqrt{2}}\!\left(\!\!\begin{array}{c}
\sqrt{1\!+\!s\,\sin \phi}\\
-i s \sqrt{1\!-\!s\,\sin \phi}
\end{array}\!\!\right)e^{i k^{(1)}_{\ell s} x},\label{eq:f1}\\
\Psi^2_{\ell s}(x)& \equiv& \Psi^{III}_{\ell s}=\frac{1}{\sqrt{2}}\left(\begin{array}{c}
1\\
-is
\end{array}\right)e^{i k^{(2)}_{\ell s} x},\label{eq:f2}\end{aligned}$$ where $s=\pm$ is the spin index. For instance, $s=+$ corresponds to an electron with a spin pointing along $-y$ in the quantum well. We also label the states with the index $\ell=\pm$, which denotes the two possible momenta (i.e., the two possible wave propagation directions) for fixed values of spin and energy $E$. The wave numbers read, $$\begin{aligned}
k^{(0)}_{\ell}&\equiv & k^I_{\ell}=k^{V}_{\ell}=\ell\sqrt{\frac{2 m_0 }{\hbar^2}E}\,,\label{eq:k0}\\
k^{(1)}_{\ell s}&\equiv & k^{II}_{\ell s}\!=\!k^{IV}_{\ell s}\!\!\!=\!\ell \sqrt{\frac{2 m_0 }{\hbar^2}(E\!+\!E_{SO1}\!-\!V_0)}\!-\!s\, k_{SO1}\,,\label{eq:k1}\\
k^{(2)}_{\ell s}&\equiv & k^{III}_{\ell s}=\ell \sqrt{\frac{2 m_0 }{\hbar^2}(E+E_{SO2})}-s\, k_{SO2}\,,\label{eq:k2}\end{aligned}$$ with $E_{SOi}=m_0\alpha_i^2/(2 \hbar^2)$ ($i=1,2$) the downshift of the energy spectra due to the spin-orbit coupling, which also causes a horizontal band splitting $\Delta k$ characterized by the momentum $k_{SOi}=m_0\alpha_i/\hbar^2$. Equations , , and depend on the energy of the incident electrons, which in the following we set equal to the Fermi energy $E_F$. Finally, we observe that both Eqs. and have the same spinor. Since the spin quantization axis in the reservoirs is not fixed, we select it parallel to the spin direction on the adjacent QPCs.
We are now in a position to solve the scattering problem in Fig. \[Fig:esc\]. We focus on the case $0<E<V_0-E_{SO1}$. This indicates that we are working with evanescent states in the QPC regions (II and IV). Hence, $k^{(1)}_{\ell s}$ acquires an imaginary part but generally also possesses a real part. We emphasize that this differs from the case of tunnel barriers without spin-orbit coupling [@ser07]. On the other hand, both $k^{(0)}_{\ell s}$ and $k^{(2)}_{\ell s}$ are always real numbers. The matching method allows us to determine all reflection and transmission amplitudes for an incoming electron, which we take as impinging from the left. The matching conditions are[$$\begin{aligned}
\Psi(\epsilon)-\Psi(-\epsilon)&=&0\label{eq:psieps}\\
\Psi^{\prime}(\epsilon)-\Psi^{\prime}(-\epsilon)&=&\frac{-im_0}{\hbar^2}\left[-\left(\alpha_2(\epsilon)-\alpha_2(-\epsilon)\right)\sigma_y\right.\nonumber\\
+ \left. \left(\alpha_1(\epsilon)-\alpha_1(\right.\right.\!\!&\!\!-\!\!&\!\!\epsilon)\!\!\! \left. \left.\right)\left(\sin \phi\sigma_z-\cos \phi \sigma_y\right)\right]
\Psi(\epsilon)\,,\label{eq:psieps2}\end{aligned}$$]{}where $\epsilon$ is a infinitesimal quantity around each interface. Equation is a statement of wave function continuity. Equation is derived from imposing flux conservation [@mol01]. Notice that in the absence of spin-orbit interaction we recover the condition of continuity for the wave function derivative. In the presence of spin-orbit coupling, this condition must be generalized according to Eq. .
![(Color online) Transmission and reflection probabilities as a function of the relative angle $\phi$ between spin-orbit fields in the QPC and the QW. $T^{s^{\prime}s}$ ($R^{s^{\prime}s}$) in the transmission (reflection) probability from an electronic state of spin $s=\pm$ to spin $s^{\prime}=\pm$ along the $-y$ direction. Parameters (a): $\alpha_1=20.16$ meV nm, $\alpha_2=25.18$ meV nm, $L=440.83$ nm, $L_1=28.02$ nm, $V_0=4.94$ meV and $E_F=4$ meV. In (b) we remove the tunnel barriers ($L_1=0$). In (c) \[(d)\] we cancel the spin-orbit interaction in the QPCs (QW): $\alpha_1=0$ ($\alpha_2=0$).[]{data-label="Fig:tfi"}](TFiesc.eps){width="45.00000%"}
Since transport is elastic, energy is conserved and the transmission $T^{s^{\prime}s}$ and reflection $R^{s^{\prime}s}$ probabilities depend on a given $E$. However, spin can be mixed after scattering and an incident electron with spin $s$ is reflected or transmitted with spin $s^{\prime}$. First, we analyze in Fig. \[Fig:tfi\] the main properties of $T^{s^{\prime}s}$ and $R^{s^{\prime}s}$ when we change the relative orientation between the QPCs and the QW spin-orbit fields. We choose the strength of the interaction in the QPCs ($\alpha_1$) and in the QW ($\alpha_2$) from Ref. [@chu15]. We tune $\phi$ from $0$ (spins parallel-oriented along the system) to $\pi/2$ (spin axes perpendicularly oriented). In Fig. \[Fig:tfi\](a) we observe that, independently of the value of $\phi$, the electrons are reflected in the same spin state that the incoming one and that the reflection probability is roughly constant as a function of $\phi$. We understand this effect as due to the spin orientation of electrons in regions I and II of Fig. \[Fig:esc\], which is the same. In contrast, the transmission probability has both spin contributions for all values of $\phi$ except for the parallel configuration, for which $T^{-+}=0$ since there exists no spin polarization. We also remark that as $\phi$ increases, i.e., as the injected spin direction is rotated from $-y$ to $z$, $T^{-+}$ increases while $T^{++}$ decreases since for higher $\phi$ the perpendicular component of the spin direction becomes larger and its contribution to the transmission thus increases.
Let us further clarify the effects discussed above considering a few special cases. If we make $L_1=0$ (no tunnel barriers), the reflection probability is trivially zero, see Fig. \[Fig:tfi\](b), and the transmission functions follow the same behavior as in Fig. \[Fig:tfi\](a) for which $L_1$ is nonzero. In Fig. \[Fig:tfi\](c) we observe that if we turn off the spin-orbit coupling on the QPCs ($\alpha_1=0$), the transmission decreases as compared with the values in Fig. \[Fig:tfi\](a). As a consequence, we infer that the spin-orbit coupling enhances the transmission properties of our double-barrier system. This may seem counterintuitive—when the spin-orbit interaction is present, one would naively expect more scattering and smaller transmission. However, we stress that the spin-orbit coupling lowers the energy band bottom of the barrier, thus amplifying the role of the evanescent states (their characteristic decay length increases) and reducing consequently the reflection probability. Finally, when we take $\alpha_2=0$ (no spin-orbit interaction in the quantum well) all transport coefficients become independent of the angle $\phi$ \[Fig. \[Fig:tfi\](d)\] since the spin orientation in the central region is fixed. Furthermore, the reflection becomes higher due to the particular energy value, which lies around a resonance valley (see below).
Before proceeding, we notice that the case $\phi=0$ can be considerably simplified. The second term in the right hand side of Eq. cancels out and we can write the projection of the Schrödinger equation $(\mathcal{H}-E)\Psi=0$ onto the spinor pointing along the $-y$ direction as $$\begin{aligned}
\label{eq:sch}
&&\Big[-\frac{\hbar^2}{2 m_0}\frac{d^2}{dx^2}-is\left(\alpha_1+\alpha_2\right)\frac{d}{dx}\nonumber\\
&& \quad\quad\quad\quad\quad\quad\quad\quad+V_0-E\Big]\Psi_s(x)=0\,,\end{aligned}$$ where $\alpha_1$ and $V_0$ are nonzero in regions II and IV whereas $\alpha_2$ is nonvanishing in region III only (Fig. \[Fig:esc\]). Now, if we apply an appropriate gauge transformation $\Psi_s(x)=\Psi(x) \exp[-i s\frac{ m_0}{\hbar^2}\int dx^{\prime}(\alpha_1+\alpha_2)]$ we can recast Eq. as $$\begin{aligned}
\label{eq:hfi0}
\left(-\frac{\hbar^2}{2 m_0}\frac{d^2}{dx^2}+V_1-V_2-E\right)\Psi(x)=0\,,\end{aligned}$$ which is independent of the spin. Here, $V_1=V_0-E_{SO1}$ in regions II and IV and zero otherwise while $V_2=E_{SO2}$ in region III. This potential corresponds to a double barrier of renormalized height $V_1$ and a quantum well of depth $V_2$ in the central region. Clearly, the spin-orbit coupling effectively lowers the top of the barrier potential as discussed earlier. Solving the scattering problem, we obtain a resonant condition that depends on all the parameters of our system, $$\begin{aligned}
\label{eq:cond}
k_{\ell s}^{(2)} L=n\pi+f(\alpha_1,\alpha_2,L_1)\,,\end{aligned}$$ where $k_{\ell s}^{(2)}$ is the wave number in the central region \[Eq. \], $n=1,2\ldots$ labels the different resonances and $f(\alpha_1,\alpha_2,L_1)$ is a complicated function of $\alpha_1$, $\alpha_2$ and $L_1$ but independent of the QW length. The condition given by Eq. can be numerically shown to hold also for the general case $\phi\neq 0$. However, in this case spin precession effects must be also taken into account.
![(Color online) Transmission probabilities as a function of the spin-orbit strength in the central region, $\alpha_2$, for $\alpha_1=20.16$ meV nm, $L=440.8$ nm, $V_0=4.94$ meV and $E_F=4$ meV. The left panels (a), (c) and (e) have $L_1=28.02$ nm while the right panels (b), (d) and (f) have $L_1=0$. The orientation angle is varied from top to bottom: $\phi=0$ for (a) and (b), $\phi=\pi/4$ in (c) and (d), $\phi=\pi/2$ for (e) and (f).[]{data-label="Fig:ta2"}](Ta2.eps){width="45.00000%"}
Figure \[Fig:ta2\] shows how our system reacts to changes applied to the spin-orbit strength in the central region, $\alpha_2$. The parallel configuration ($\phi=0$) is plotted in Fig. \[Fig:ta2\](a), where we observe resonance peaks for certain values of spin-orbit interaction and a fixed Fermi energy. As the spin-orbit coupling increases, the quantum well becomes deeper and, as a consequence, there appear new quasibound states between the two barriers that fulfill Eq. . When the energy of the incident electron hits one of these states, the transmission probability is maximal. Therefore, the spin-orbit interaction acts in our system as a gate voltage by shifting the resonances of the quantum well [@lop07]. Our system then behaves as an analog of a Fabry-Perot resonator tuned with a spin-orbit potential. Note that the resonances appear for $T^{++}$ only since for $\phi=0$ the spins are parallel and one obtains $T^{-+}=0$ always. This can be better understood if we take $L_1=0$, in which case the double barrier potential disappears and we obtain an almost transparent system independently of the depth of the quantum well \[Fig. \[Fig:ta2\](b)\]. Here, the energy of the electron is sufficiently high that its wave is mostly unaffected by the well discontinuity. Only for strong enough spin-orbit strengths the transmission shows weak oscillations (Ramsauer effect). We also find that the off-diagonal transmission coefficient is zero. This originates from the fact in the parallel configuration the spin cannot be flipped, in agreement with the case $\phi=0$ in Fig. \[Fig:tfi\](d).
In Figs. \[Fig:ta2\](c) and (d) we take $\phi=\pi/4$, i.e., the wave is spin polarized $45º$ with respect to $-y$. Let us first eliminate the double-barrier potential ($L_1=0$) and focus on the effects from the central region only, see Fig. \[Fig:ta2\](d). We observe that both $T^{++}$ and $T^{-+}$ are nonzero and oscillate out of phase. These oscillations are a consequence of the spin transistor effect predicted by Datta and Das [@dat90]. We find $T^{++}=1$ and $T^{-+}=0$ for $\alpha_2=0$ but then both transmissions become modulated as we increase the spin-orbit strength since the QW energy bands show a larger spin splitting $\Delta k=m_0 \alpha_2/\hbar^2$. For certain values of $\alpha_2$, $T^{++}$ ($T^{-+}$) attains its minimum (maximum) value of $0.5$. Importantly, the nature of these transmission oscillations fundamentally differs from the resonances in Fig. \[Fig:ta2\](a). To see this, we next obtain the spin-precession frequency from the relation [@dat90] $$\begin{aligned}
\label{eq:cond2}
T^{++}\propto \cos^2 (\Delta k L)\end{aligned}$$ This expression implies that the maximum condition is reached at $\Delta k L=n^{\prime} \pi$ ($n^{\prime}=1,2\ldots$). For the parameters of Fig. \[Fig:ta2\](d) this corresponds to $\alpha_2\simeq13.6n^{\prime}$ meV nm.
More interestingly, we now turn on the double barrier potential and allow for the interplay between Fabry-Perot and Datta-Das oscillations. The superposition of the two effects can be seen in Fig. \[Fig:ta2\](c). We observe that (i) the resonance peaks for $T^{++}$ become somewhat quenched and (ii) the off-diagonal coefficient $T^{-+}$ shows an irregular series of oscillating peaks. The effect is more intense in the perpendicular configuration ($\phi=\pi/2$), see Fig. \[Fig:ta2\](e). Both transmissions oscillate now between $0$ and $1$ with opposite phases \[Fig. \[Fig:ta2\](f)\] and the combination of both types of oscillations yields the curves depicted in Fig. \[Fig:ta2\](e).
![(Color online) Transmission probabilities as a function of the central region width, $L$, for $\alpha_1=20.16$ meV nm, $\alpha_2=25.18$ meV nm, $V_0=4.94$ meV and $E_F=4$ meV. The left panels (a), (c) and (e) have $L_1=28.02$ nm while the right panels (b), (d) and (f) have $L_1=0$. The orientation angle is varied from top to bottom: $\phi=0$ for (a) and (b), $\phi=\pi/4$ in (c) and (d), $\phi=\pi/2$ for (e) and (f).[]{data-label="Fig:tl"}](TL.eps){width="45.00000%"}
It is now natural to ask about the effect of tuning the QW length $L$. We show this in Fig. \[Fig:tl\] for the same orientation angles as in Fig. \[Fig:ta2\] but fixing the spin-orbit strength $\alpha_2$. When $\phi=0$, Fig. \[Fig:tl\](a) presents for $T^{++}$ narrowly spaced oscillations since as we increase the width of the central cavity there appear more internal modes that, at fixed values of $L$, are resonant with the incident wave (Fabry-Perot effect). The resonant condition from Eq. implies that the transmission is peaked at $L\simeq(47.5n+8.3)$ nm ($n=1,2\ldots$). For $\phi=0$ spin flipping is not possible and $T^{-+}=0$. When the constrictions are turned off ($L_1=0$), we have a completely open system and the transmission stays constant at its maximum value, see Fig. \[Fig:tl\](b). As we increase the spin orientation angle \[$\phi=\pi/4$ in Figs. \[Fig:tl\](c) and (d) and $\phi=\pi/2$ in Figs. \[Fig:tl\](e) and (f)\], the spin transistor effect begins to contribute as we observe a spin precession for both $T^{++}$ and $T^{-+}$, modulated by their characteristic frequency, namely, $L\simeq237.6 n^{\prime}$ nm ($n^{\prime}=1,2\ldots$). We find that when $L_1=0$ (no tunnel barriers) the Fabry Perot resonances disappear and only the Datta-Das oscillations are present \[Fig. \[Fig:tl\](d) and (f)\], as expected.
Remarkably, when both oscillation modes are present we find that the transmission becomes quasiperiodic \[Fig. \[Fig:tl\](c) and (e)\]. This effect arises from the combination of at least two oscillations whose characteristic frequencies are incommensurate [@ott93]. In our system, the Fabry-Perot frequency is given by $f_{FP}=\frac{1}{\pi}\sqrt{\frac{2 m_0 }{\hbar^2}E+k_{SO2}^2}$ whereas that of the spin precession motion is expressed as $f_{sp}=2 k_{SO2}/\pi$. Clearly, its ratio $f_{FP}/f_{sp}$ is quite generally an irrational number. In related systems, quasiperiodic oscillations have been predicted to occur in double quantum dots with incommensurate capacitance couplings [@ruz92] and in ac-driven supelattices where the ratio between the ac frequency and the internal frequency is not a rational number [@san01]. Importantly, in our case the origin of both oscillations is purely quantum (wave interference and spin precession).
Quasi-one dimensional case {#sec:quasi1d}
==========================
The above discussion demonstrates that two types of transmission oscillations can coexist in a double-barrier spin-orbit coupled resonant tunneling diode. However, the results were strictly limited to the 1D case. We now consider the more realistic situation of a double QPC embedded in a quantum wire of finite width. The problem is not a mere extension that takes into account transverse channels since these channels become coupled via the Rashba intersubband mixing potential. This term causes spin-flip transitions between adjacent channels and generally destroys the spin coherent oscillations [@gel10]. Furthermore, it yieds Fano lineshapes [@san06] that dramatically alter the conductance curves [@san06; @she04; @wan04; @zha05; @lop07]. We note that there exists another type of intersubband spin orbit coupling potential that occurs in coupled wells with two subbands [@sou15]. Here, we consider the case of an intense confinement in the growth direction such that only the lowest subband is populated.
![Sketch of the double quantum point contact system with a finite width $L_y$. Electrons can move in the white areas whereas forbidden regions are depicted in grey. The height of the constriction barriers is $L_y^0$. The rest of the parameters are defined as in the purely 1D case (Fig. \[Fig:esc\]). []{data-label="Fig:sketch2"}](3p.eps){width="45.00000%"}
We consider the planar waveguide formed in a 2D electron gas lying on the $x$–$y$ plane as in Fig. \[Fig:sketch\]. In the numerical simulations we consider a hard-wall confinement potential along $y$ and two square quantum point contacts in the $x$ direction. The system parameters are depicted in Fig. \[Fig:sketch2\].
We take a given quantization axis $\hat{n}$ for the spin in the left and right contacts. The spin eigenfunctions are then denoted with $\chi_s(\eta)$, with $s = \pm$ the eigenstate label and $\eta=\uparrow,\downarrow$ the discrete variable. The full wave function $\Psi(x, y, \eta)$ is expanded in spin channels $\psi_s(x, y)$ as $$\Psi(x, y, \eta)=\sum_{s'}\psi_{s'}(x, y)\chi_{s'}(\eta)\,.$$ Projecting the Schrödinger equation on the spin basis, we obtain coupled channel equations, $$\begin{aligned}
&& \left[-\frac{\hbar^2\nabla^2}{2m}+V(x,y)\right] \psi_s(x,y) \nonumber\\
&-&
\frac{i\hbar}{2}\sum_{s'}{
\langle s|\sigma_y|s'\rangle
\left( V_A(x)\frac{\partial}{\partial x}+\frac{\partial}{\partial x}V_A(x)\right)
\, \psi_{s'}(x,y)
} \nonumber\\
&-&
\frac{i\hbar}{2}\sum_{s'}{
\langle s|\sigma_z|s'\rangle
\left( V_B(x)\frac{\partial}{\partial x}+\frac{\partial}{\partial x}V_B(x)\right)
\, \psi_{s'}(x,y)
} \nonumber\\
&+&
\frac{i\hbar}{2}\sum_{s'}{
\langle s | \sigma_x | s'\rangle\,
V_A(x)\frac{\partial}{\partial y}}
\, \psi_{s'}(x,y)\; ,
\label{eq:schr}\end{aligned}$$ where the potentials $V_A(x)$ and $V_B(x)$ are responsible for the coupling between the different spin channels $s=\pm$. In general, the Pauli-matrix elements in Eq. (\[eq:schr\]) depend on $\hat{n}$. To connect with the 1D case discussed in Sec. \[sec:1D\] we take $\hat{n}=-\hat{y}$, which makes the $\sigma_y$ term diagonal, but those with $\sigma_x$ and $\sigma_z$ remain non diagonal. Coupling between opposite spin states is, therefore, always present in the quasi-1D case when $(V_A,V_B)\ne 0$ [@Mor99; @Mor00].
In Eq. the potentials $V_A$ and $V_B$ read [$$\begin{aligned}
V_A(x) &=& \alpha_1 \!\cos\phi\, \mathcal{P}_1(x)
\!+\!\alpha_2 \mathcal{P}_2(x)
\!+\!\alpha_1 \!\cos\phi\, \mathcal{P}_3(x)\label{eq_VA}\, ,\\
V_B(x)&=&-\alpha_1\sin\phi\, \mathcal{P}_1(x)
-\alpha_1 \sin\phi\, \mathcal{P}_3(x)\,,\label{eq_VB}\end{aligned}$$]{} where the projectors $\mathcal{P}_i(x)$ partition the $x$ domain in regions $i=1$ (left QPC), $i=2$ (QW) and $i=3$ (right QPC). These two potentials yield qualitatively different spin-flip couplings, since $V_B$ only appears with $\partial/\partial x$, while $V_A$ appears with both $\partial/\partial x$ and $\partial/\partial y$. As before, $\phi$ is the angle defining the relative orientation of the Rashba fields. Notably, $V_B(x)$ vanishes with $\phi=0$ and then, for quantization axis along $y$, the only spin-flip coupling in Eq. (\[eq:schr\]) is via the last term depending on $\partial/\partial y$. To be effective, this spin-flip coupling requires that at least two transverse modes (differing in the nodes along $y$) are propagating in the asymptotic leads [@san06]. Otherwise, as we show below, there is no spin-flip when $\hat{n}$ lies along $y$.
![(Color online) Transmission probabilities for a quasi-one dimensional double quantum point contact system as a function of the spin-orbit strength in the central region, $\alpha_2$. Parameters: $\alpha_1=20.16$ meV nm, $L=440.8$ nm, $L_1=10.91$ nm, $L^0_y=39.29$ nm, $L_y=87.29$ nm and $E_F=4$ meV. The left panels (a), (c) and (e) have $L_1=10.91$ nm while the right panels (b), (d) and (f) have $L_1=0$. The orientation angle is varied from top to bottom: $\phi=0$ for (a) and (b), $\phi=\pi/4$ in (c) and (d), $\phi=\pi/2$ for (e) and (f).[]{data-label="Fig:ta2py"}](a2pp.eps){width="45.00000%"}
Equation is solved with the quantum-transmitting boundary method [@qtbm] on a uniform grid. The resulting transmission probability as a function of the middle spin-orbit strength $\alpha_2$ is shown in Fig. \[Fig:ta2py\]. We recall that the transmission is expressed in the $-y$ direction basis. Similarly to Fig. \[Fig:ta2\] we distinguish the case with the constrictions (left panels) from the case without the QPCs (right panels). For $\phi=0$ \[Fig. \[Fig:ta2py\](a)\] we quench the spin precession oscillations since the injected spins are parallel to the Rashba field. Then, the cross transmission $T^{-+}$ vanishes identically. The resonant tunneling peaks qualitatively agree with the 1D case \[cf. Fig. \[Fig:ta2\](a)\]. Likewise, the Ramsauer oscillations that arise when the QPCs are absent \[Fig. \[Fig:ta2py\](b)\] are visible at large values of $\alpha_2$ \[cf. Fig. \[Fig:ta2\](b)\]. The agreement in both cases is good for small values of $\alpha_2$. This is reasonable since Rashba intersubband coupling is negligible if $\alpha_2\ll \hbar^2/mL_y$ [@dat90]. For larger $\alpha_2$ we observe in Fig. \[Fig:ta2py\](b) sharp dips that originate from the Fano-Rashba effect [@san06] and that are unique to quasi-one dimensional waveguides with nonuniform spin-orbit coupling as in our case. Strikingly enough, as $\alpha_2$ increases we detect in Fig. \[Fig:ta2py\](a) more resonant peaks than in the strict 1D case. We explain this effect as follows. For $\alpha_1=\alpha_2=0$ the cavity works as a resonator with multiple resonances. If the cavity is closed, the bound levels can be described with a pair of natural numbers $(n_1,n_2)$ since its potential corresponds to a 2D infinite well [@cohen]. To a good approximation, the electronic scattering when the cavity is open obeys a conservation law that fixes the transversal component of motion [@baskin]. Accordingly, $n_2$ is conserved upon traversing the cavity and the transmission shows less peaks than bound states in the closed cavity. In the presence of spin-orbit coupling, the conservation law does not have to hold and more resonances then emerge.
For $\phi=\pi/4$ the injected electrons are spin rotated with regard to the $\alpha_2$ field and spin precession oscillations of the Datta-Das type are expected. This can be more distinctly seen in Fig. \[Fig:ta2py\](d), where the QPC widths are set to zero. Up to $\alpha_2\simeq 30$ meV nm the oscillations are smooth as in Fig. \[Fig:ta2\](d). For larger $\alpha_2$ the subband mixing potential starts to play a significant role. As a consequence of the spin mixing induced by the $p_y$ term, the precession oscillations become irregular [@gel10] and the transmission curves can no longer be determined by a single frequency. When combined with the Fabry-Perot oscillations, the transmission lineshapes are transformed into nonharmonic functions of $\alpha_2$ \[see Fig. \[Fig:ta2py\](c)\] and our previous 1D analysis in terms of quasiperiodic oscillations does not hold. For completeness, we also show the case $\phi=\pi/2$ for which the Data-Das frequency is higher (the spins are injected perpendicular to the Rashba field) but the spin oscillations turn out to be nonuniform as $\alpha_2$ grows as illustrated in Fig. \[Fig:ta2py\](f). The overall transmission curves \[Fig. \[Fig:ta2py\](e)\] qualitatively follow the pattern observed in the case $\phi=\pi/4$.
![(Color online) Transmission probabilities for a quasi-one dimensional double quantum point contact system as a function of the width of central region, $L$. Parameters: $\alpha_1=20.16$ meV nm, $\alpha_2=65.47$ meV nm, $L^0_y=39.29$ nm, $L_y=87.29$ nm and $E_F=4$ meV. Additionally, we set in (a) $L_1=0$ nm and $\phi=\pi/2$, in (b) $L_1=10.91$ nm and $\phi=0$, and in (c): $L_1=10.91$ nm and $\phi=\pi/2$. []{data-label="Fig:tlpy"}](Lppp.eps){width="45.00000%"}
In Fig. \[Fig:tlpy\] we analyze the dependence with the central cavity width $L$. We set the spin-orbit strenght $\alpha_2$ to a moderate value to highlight the effects due to the Rashba intersubband coupling term. Figure \[Fig:tlpy\](a) shows the transmission for $L_1=0$ and $\phi=\pi/2$. This implies that only oscillations from the spin dynamics are present since resonant tunneling effects are not allowed. Unlike Fig. \[Fig:tl\](f) here the oscillations are not uniform for both transmission probabilities, $T^{++}$ and $T^{-+}$. The Fabry-Perot peaks are more regular as shown in Fig. \[Fig:tlpy\](b), where $L_1$ is nonzero and $\phi=0$ in order to forbid spin precession oscillations. This suggests that the Rashba intersubband potential has a stronger impact on the Datta-Das oscillations than on the Fabry-Perot peaks. In Fig. \[Fig:tl\](c) we show characteristic transmission curves for nonzero $L_1$ and $\phi=\pi/2$, in which case both oscillation modes come into play. As compared to the 1D case in Fig. \[Fig:tl\](e) the oscillations are now more intricate: their amplitudes strongly fluctuate with increasing $L$ and their frequency cannot be described in terms of combinations of individual frequencies.
![(Color online) Transmission probabilities for a quasi-one dimensional double quantum point contact system as a function of the position of the Fermi level, $E_F$. Parameters: $\alpha_1=20.16$ meV nm, $\alpha_2=65.47$ meV nm, $L^0_y=21.82$ nm, $L_y=87.29$ nm and $L=440.8$ nm. Additionally, we set in (a) $L_1=0$ nm and $\phi=\pi/2$, in (b) $L_1=10.91$ nm and $\phi=0$, and in (c): $L_1=10.91$ nm and $\phi=\pi/2$. []{data-label="Fig:tepy"}](Epp.eps){width="45.00000%"}
In order to complete the analysis of our system we present in Fig. \[Fig:tepy\] the transmission probability as a function of the Fermi energy for the same parameters as above. In Fig. \[Fig:tepy\](a) we consider the case without the QPCs ($L_1=0$ nm) and apply a spin-orbit interaction in the central region such that its direction lies orthogonal to that of the leads ($\phi=\pi/2$). We find an approximate transmission quantization of $T=T^{++}+T^{-+}$ (black line) whenever a new propagating channel opens up as the Fermi energy surpasses the values $E_n=\hbar^2\pi^2 n^2/2 m_0 L_y^2$ with $n=1,2\ldots$ (recall that the confinement along the transverse direction is described with a hard-wall potential). We also observe in Fig. \[Fig:tepy\](a) the spin dependence due to the spin-orbit interaction in the middle region (solid red and dashed blue lines). The Fabry-Perot peaks form when $\phi=0$ and $L_1\neq 0$, see Fig. \[Fig:tepy\](b). Here, the transmission is zero until the Fermi energy is such that the first propagating state is allowed in the leads, which corresponds to $E_F>E_1=1.23$ meV. At the same time, in the QPCs we have evanescent states below the energy value $E_1^{QPC}=4.93$ meV. Then, the resonances ranging between these two energies are due only to tunneling transmission across the QPCs. The second channel in the leads opens up at $E_2=4.93$ meV but the transmission does not exceed $1$ because we have just one open channel in the constrictions. When the third channel in the leads opens up, $E_F>E_3=11.10$ meV, we observe dips in the diagonal transmission probability which correlate with peaks in the off-diagonal transmission. This effect originates from the coupling between propagating states in the system and quasibound states in the cavity. Finally, Fig. \[Fig:tepy\](c) shows the combination of Fabry-Perot peaks and Datta-Das oscillations when the spin-orbit fields are perpendicular. Its behavior is similar to the Fabry-Perot-Datta-Das oscillations discussed as a function of the spin-orbit coupling \[Fig. \[Fig:ta2py\](e)\] and cavity length \[Fig. \[Fig:tlpy\](c)\].
Conclusions {#sec:conc}
===========
To sum up, we have investigated a spin-orbit quantum wire coupled to quantum point contacts. We have found that both resonant tunneling and spin precession oscillations combine into complex patterns that can be explained with the aid of quasiperiodic modes in the strict 1D case. For the more realistic setup where the conducting channel has a finite width (2D case) we have discussed the important role of the Rashba intersubband coupling term as the spin-orbit strength increases.
We have used in our numerical simulations realistic parameters taken from the sample and measurements of Ref. [@chu15]. Therefore, our predictions are within the realm of today’s techniques. The angle between the spin-orbit fields in the QPCs and the quantum well can be tuned with lateral electric fields while the spin-orbit strength can be manipulated with a gate terminal on top of the middle cavity. We have focused on the transmission, from which the two-terminal conductance $G$, which is experimentally accessible, readily follows in the zero temperature limit. For finite temperatures we expect thermal smearing effects but we have in mind low temperatures as in Ref. [@chu15] ($0.03$ K). Thus, phonon effects can be safely neglected. Another detrimental effect would be the presence of disorder since we consider ballistic systems only and our predictions rely on quantum interference. Therefore, samples with enough coherence lengths and mean free paths would be needed, which are now routinely available [@lou07]. Measurement of diagonal and off-diagonal conductances can be achieved, e.g., with ferromagnetic electrodes whose relative magnetization can be changed from parallel to antiparallel orientation in response to a small magnetic field [@koo09]. The results regarding the length variation can be tested with different samples. Finally, the resolution of the conductance peaks would lie in the sub-meV range (see Fig. \[Fig:tepy\]), which can be achieved by tuning an external backgate electrode capacitively coupled to the sample.
Further extensions of our work could address high-field transport properties, in which case inelastic transitions in three-dimensional resonant tunneling diodes can change the current–voltage characteristics [@stone; @buttiker]. Another important issue for future works is the role of electron-electron interactions, which may lead to instabilities and hysteretic curves in double barrier systems [@mar94]. Furthermore, magnetically doped resonant tunneling devices are shown to be quite sensitive to external magnetic fields [@slo03; @slo07; @Woj13]. In the presence of a spin-orbit coupling beating patterns are predicted to occur in double-barrier resonant tunneling structures [@egu01]. Finally, we would like to mention the closely related systems known as chaotic dots [@mar92] since they are built as semiconductor cavities between a pair of quantum point contacts, similarly to the two-dimensional cavities considered in the last part of our work. In contrast, our cavities have a regular shape. Interestingly, closed chaotic dots exhibit Coulomb blockade peak fluctuations [@ale02] and subsequent discussions might then consider how these fluctuations are affected by the presence of spin-orbit interactions.
This work was funded by MINECO (Spain), grant No. FIS2014-52564
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We consider distributed iterative algorithms for the averaging problem over time-varying topologies. Our focus is on the convergence time of such algorithms when complete (unquantized) information is available, and on the degradation of performance when only quantized information is available. We study a large and natural class of averaging algorithms, which includes the vast majority of algorithms proposed to date, and provide tight polynomial bounds on their convergence time. We then propose and analyze distributed averaging algorithms under the additional constraint that agents can only store and communicate quantized information, converge to the average of the initial values of the agents within some error. We establish bounds on the error and tight bounds on the convergence time, as a function of the number of quantization levels.'
author:
- 'Angelia Nedić[^1], Alex Olshevsky, Asuman Ozdaglar, and John N. Tsitsiklis[^2]'
bibliography:
- 'distributed\_alex.bib'
title: ' **On Distributed Averaging Algorithms and Quantization Effects[^3]**'
---
Introduction
============
There has been much recent interest in distributed control and coordination of networks consisting of multiple, potentially mobile, agents. This is motivated mainly by the emergence of large scale networks, characterized by the lack of centralized access to information and time-varying connectivity. Control and optimization algorithms deployed in such networks should be completely distributed, relying only on local observations and information, and robust against unexpected changes in topology such as link or node failures.
A canonical problem in distributed control is the [*consensus problem*]{}. The objective in the consensus problem is to develop distributed algorithms that can be used by a group of agents in order to reach agreement (consensus) on a common decision (represented by a scalar or a vector value). The agents start with some different initial decisions and communicate them locally under some constraints on connectivity and inter-agent information exchange. The consensus problem arises in a number of applications including coordination of UAVs (e.g., aligning the agents’ directions of motion), information processing in sensor networks, and distributed optimization (e.g., agreeing on the estimates of some unknown parameters). The [*averaging problem*]{} is a special case in which the goal is to compute the exact average of the initial values of the agents. A natural and widely studied consensus algorithm, proposed and analyzed by Tsitsiklis [@T84] and Tsitsiklis [*et al.*]{} [@TBA86], involves, at each time step, every agent taking a weighted average of its own value with values received from some of the other agents. Similar algorithms have been studied in the load-balancing literature (see for example [@C89]). Motivated by observed group behavior in biological and dynamical systems, the recent literature in cooperative control has studied similar algorithms and proved convergence results under various assumptions on agent connectivity and information exchange (see [@VCBJCS95], [@OSM04], [@RB05], [@M05], [@LR06]).
In this paper, our goal is to provide tight bounds on the convergence time (defined as the number of iterations required to reduce a suitable Lyapunov function by a constant factor) of a general class of consensus algorithms, as a function of the number $n$ of agents. We focus on algorithms that are designed to solve the averaging problem. We consider both problems where agents have access to exact values and problems where agents only have access to quantized values of the other agents. Our contributions can be summarized as follows.
In the first part of the paper, we consider the case where agents can exchange and store continuous values, which is a widely adopted assumption in the previous literature. We consider a large class of averaging algorithms defined by the condition that the weight matrix is a possibly nonsymmetric, doubly stochastic matrix. For this class of algorithms, we prove that the convergence time is $O(n^2/\eta)$, where $n$ is the number of agents and $\eta$ is a lower bound on the nonzero weights used in the algorithm. To the best of our knowledge, [*this is the best polynomial-time bound on the convergence time of such algorithms*]{}. We also show that this bound is tight. Since this result implies an $O(n^3)$ bound on convergence time. In Section \[matrixpicking\], we present a distributed algorithm that selects the weights dynamically, using three-hop neighborhood information. Under the assumption that the underlying connectivity graph at each iteration is undirected, we establish an improved $O(n^2)$ upper bound on convergence time. This matches the best available convergence time guarantee for the much simpler case of static connectivity graphs .
In the second part of the paper, we impose the additional constraint that agents can only store and transmit quantized values. This model provides a good approximation for communication networks that are subject to communication bandwidth constraints. We focus on a particular quantization rule, which rounds down the values to the nearest quantization level. We propose a distributed algorithm that uses quantized values and, using a slightly different Lyapunov function, we show that the algorithm guarantees the convergence of the values of the agents to a common value. In particular, we prove that all agents have the same value after $O((n^2/\eta)\log (nQ))$ time steps, where $Q$ is the number of quantization levels per unit value. Due to the rounding-down feature of the quantizer, this algorithm does not preserve the average of the values at each iteration. However, we provide bounds on the error between the final consensus value and the initial average, as a function of the number $Q$ of available quantization levels. In particular, we show that the error goes to 0 at a rate of $(\log
Q)/Q$, as the number $Q$ of quantization levels increases to infinity.
Other than the papers cited above, our work is also related to [@KBS06] and [@CFSZ05; @CFFTZ07], which studied the effects of quantization on the performance of averaging algorithms. In [@KBS06], Kashyap [*et al.*]{} proposed randomized [*gossip-type*]{} quantized averaging algorithms under the assumption that each agent value is an integer. They showed that these algorithms preserve the average of the values at each iteration and converge to approximate consensus. They also provided bounds on the convergence time of these algorithms for specific static topologies (fully connected and linear networks). In the recent work [@CFFTZ07], Carli [*et al.*]{} proposed a distributed algorithm that uses quantized values and preserves the average at each iteration. They showed favorable convergence properties using simulations on some static topologies, and provided performance bounds for the limit points of the generated iterates. Our results on quantized averaging algorithms differ from these works in that [*we study a more general case of time-varying topologies, and provide tight polynomial bounds on both the convergence time and the discrepancy from the initial average, in terms of the number of quantization levels*]{}.
The paper is organized as follows. In Section \[agreementalg\], we introduce a general class of averaging algorithms, and present our assumptions on the algorithm parameters and on the information exchange among the agents. In Section \[convtimesection\], we present our main result on the convergence time of the averaging algorithms under consideration. In Section \[matrixpicking\], we present a distributed averaging algorithm for the case of undirected graphs, which picks the weights dynamically, resulting in an improved bound on the convergence time. In Section \[qanalysis\], we propose and analyze a quantized version of the averaging algorithm. In particular, we establish bounds on the convergence time of the iterates, and on the error between the final value and the average of the initial values of the agents. Finally, we give our concluding remarks in Section \[conclusions\].
A Class of Averaging Algorithms {#agreementalg}
===============================
We consider a set $N=\{1,2,\ldots,n \}$ of agents, which will henceforth be referred to as “nodes.” Each node $i$ starts with a scalar value $x_i(0)$. At each nonnegative integer time $k$, node $i$ receives from some of the other nodes $j$ a message with the value of $x_j(k)$, and updates its value according to: $$x_i(k+1) = \sum_{j=1}^n a_{ij}(k) x_j(k), \label{noquant}$$ where the $a_{ij}(k)$ are nonnegative weights with the property that $a_{ij}(k)>0$ only if node $i$ receives information from node $j$ at time $k$. We use the notation $A(k)$ to denote the [*weight matrix*]{} $[a_{ij}(k)]_{i,j=1,\ldots,n}$, Given a matrix $A$, we use $\E(A)$ to denote the set of directed edges $(j,i)$, including self-edges $(i,i)$, such that $a_{ij}>0$. At each time $k$, the nodes’ connectivity can be represented by the directed graph $G(k)=(N,\E(A(k)))$.
Our goal is to study the convergence of the iterates $x_i(k)$ to the average of the initial values, $(1/n)\sum_{i=1}^n
x_i(0)$, as $k$ approaches infinity. In order to establish such convergence, we impose some assumptions on the weights $a_{ij}(k)$ and the graph sequence $G(k)$.
For each $k$, the weight matrix $A(k)$ is a doubly stochastic matrix[^4] with positive diagonal . Additionally, there exists a constant $\eta> 0$ such that if $a_{ij}(k)>0$, then $a_{ij}(k) \geq \eta$.\[weights\]
The double stochasticity assumption on the weight matrix guarantees that the average of the node values remains the same at each iteration (cf. the proof of Lemma \[vl\] below). The second part of this assumption states that each node gives significant weight to its values and to the values of its neighbors at each time $k$.
Our next assumption ensures that the graph sequence $G(k)$ is sufficiently connected for the nodes to repeatedly influence each other’s values.
There exists an integer $B \geq 1$ such that the directed graph $$\Big(N, \E(A(kB)) \bigcup
\E(A(kB+1))\bigcup \cdots \bigcup \E(A((k+1)B-1))\Big)$$ is strongly connected for all nonnegative integers $k$. \[connectivity\]
Any algorithm of the form given in Eq. (\[noquant\]) with the sequence of weights $a_{ij}(k)$ satisfying Assumptions \[weights\] and \[connectivity\] solves the averaging problem. This is formalized in the following .
\[uqc\] Let Assumptions \[weights\] and \[connectivity\] hold. Let $\{x(k)\}$ be generated by the algorithm (\[noquant\]). Then, for all $i$, we have $$\lim_{k \rightarrow \infty} x_i(k) = \frac{1}{n} \sum_{j=1}^n
x_j(0).$$
This is a minor modification of known results in [@T84; @TBA86; @JLM03; @BHOT05], where the convergence of each $x_i(k)$ to the same value is established under weaker versions of Assumptions \[weights\] and \[connectivity\]. The fact that the limit is the average of the entries of the vector $x(0)$ follows from the fact that multiplication of a vector by a doubly stochastic matrix preserves the average of the vector’s components.
Recent research has focused on methods of choosing weights $a_{ij}(k)$ that satisfy Assumptions \[weights\] and \[connectivity\], and minimize the convergence time of the resulting averaging algorithm (see [@XB04] for the case of static graphs, see [@OSM04] and [@BFT05] for the case of symmetric weights, i.e., weights satisfying $a_{ij}(k)=a_{ji}(k)$, and also see [@C06; @boyd]). For static graphs, some recent results on optimal time-invariant algorithms may be found in [@OT06].
Convergence time \[convtimesection\]
====================================
In this section, we give an analysis of the convergence time of averaging algorithms of the form (\[noquant\]). Our goal is to obtain tight estimates of the convergence time, under Assumptions \[weights\] and \[connectivity\].
As a convergence measure, we use the “sample variance” of a vector $x \in \R^n$, defined as $$V(x) = \sum_{i=1}^n (x_i - \bar{x} )^2,$$ where $\bar{x}$ is the average of the entries of $x$: $$\bar{x} = \frac{1}{n} \sum_{i=1}^n x_i.$$
Let $x(k)$ denote the vector of node values at time $k$ \[i.e., the vector of iterates generated by algorithm (\[noquant\]) at time $k$\]. We are interested in providing an upper bound on the number of iterations it takes for the “sample variance” $V(x(k))$ to decrease to a small fraction of its initial value $V(x(0))$. We first establish some technical preliminaries that will be key in the subsequent analysis. In particular, in the next subsection, we explore several implications of the double stochasticity assumption on the weight matrix $A(k)$.
Preliminaries on Doubly Stochastic Matrices
-------------------------------------------
We begin by analyzing how the sample variance $V(x)$ changes when the vector $x$ is multiplied by a doubly stochastic matrix $A$. The next lemma shows that $V(Ax) \leq V(x)$. Thus, under Assumption \[weights\], the sample variance $V(x(k))$ is nonincreasing in $k$, and $V(x(k))$ can be used as a Lyapunov function.
\[vl\] Let $A$ be a doubly stochastic matrix. Then,[^5] for all $x \in \R^n$, $$V(Ax) = V(x) - \sum_{i<j} w_{ij} (x_i - x_j)^2,$$ where $w_{ij}$ is the $(i,j)$-th entry of the matrix $A^T A$.
Let ${\bf 1}$ denote the vector in $\R^n$ with all entries equal to $1$. The double stochasticity of $A$ implies $$A {\bf 1}= {\bf 1}, \qquad {\bf 1}^T A = {\bf 1}^T.$$ Note that multiplication by a doubly stochastic matrix $A$ preserves the average of the entries of a vector, i.e., for any $x\in\rn$, there holds $$\overline{Ax} = \frac{1}{n}\, {\bf 1}^T Ax
=\frac{1}{n}\, {\bf 1}^T x = \bar{x}.$$ We now write the quadratic form $V(x)-V(Ax)$ explicitly, as follows: $$\begin{aligned}
V(x) - V(Ax) & = & (x - \bar{x} {\bf 1})^T (x-\bar{x}{\bf 1}) -
(Ax-\overline{Ax}{\bf 1})^T (Ax - \overline{Ax}{\bf 1}) \nonumber \\
& = & (x - \bar{x} {\bf 1})^T (x-\bar{x}{\bf 1}) - (Ax-\bar{x}A{\bf
1})^T (Ax - \bar{x} A {\bf 1}) \nonumber \\ & = & (x - \bar{x} {\bf
1})^T (I-A^T A) (x-\bar{x}{\bf 1}). \label{vdec} \end{aligned}$$
Let $w_{ij}$ be the $(i,j)$-th entry of $A^T A$. Note that $A^T A$ is symmetric and stochastic, so that $w_{ij}=w_{ji}$ and $w_{ii} = 1 - \sum_{j \neq i} w_{ij}$. Then, it can be verified that $$A^T A = I - \sum_{i<j} w_{ij} (e_i - e_j) (e_i -
e_j)^T, \label{asquareform}$$ where $e_i$ is a unit vector with the $i$-th entry equal to 1, and all other entries equal to 0 .
By combining Eqs. (\[vdec\]) and (\[asquareform\]), we obtain $$\begin{aligned}
V(x) - V(Ax) & =& (x - \bar{x} {\bf 1})^T
\Big(\sum_{i<j} w_{ij} (e_i - e_j) (e_i - e_j)^T\Big) (x-\bar{x}{\bf 1}) \nonumber \\
& = & \sum_{i<j} w_{ij} (x_i - x_j)^2. $$
Note that the entries $w_{ij}(k)$ of $A(k)^T A(k)$ are nonnegative, because the weight matrix $A(k)$ has nonnegative entries. In view of this, Lemma \[vl\] implies that $$V(x(k+1))\le V(x(k))\qquad \hbox{for all }k.$$ Moreover, the amount of variance decrease is given by $$V(x(k)) - V(x(k+1)) = \sum_{i<j} w_{ij}(k) (x_i(k) - x_j(k))^2.$$ We will use this result to provide a lower bound on the amount of decrease of the sample variance $V(x(k))$ in between iterations.
Since every positive entry of $A(k)$ is at least $\eta$, it follows that every positive entry of $A(k)^T A(k)$ is at least $\eta^2$. Therefore, it is immediate that $$\hbox{if }\quad w_{ij}(k) > 0, \quad\hbox{then }\quad w_{ij}{(k)} \geq \eta^2.$$ In our next lemma, we establish a stronger lower bound. In particular, we find it useful to focus not on an individual $w_{ij}$, but rather on all $w_{ij}$ associated with edges $(i,j)$ [that]{} cross [a particular]{} cut in the graph $(N,\E(A^TA))$. For such groups of $w_{ij}$, we prove a lower bound which is linear in $\eta$, as seen in the following.
\[db\] Let $A$ be a stochastic matrix with positive diagonal , and assume that is at least $\eta$. Also, let $(S^{-},S^{+})$ be a partition of the set $N=\{1,\ldots,n\}$ into two disjoint sets. If $$\sum_{i \in S^{-}, ~j \in S^{+}} w_{ij} > 0,$$ then $$\sum_{i
\in S^{-}, ~j \in S^{+}} w_{ij} \geq \frac{\eta}{2}.$$
Let $\sum_{i \in S^{-}, ~j \in S^{+}} w_{ij} >0$. From the definition of the weights $w_{ij}$, we have $w_{ij}=\sum_k
a_{ki}a_{kj}$, which shows that there exist $i\in S^-$, $j\in S^+$, and some $k$ such that $a_{ki}>0$ and $a_{kj}>0$. For either case where $k$ belongs to $S^-$ or $S^+$, we see that there exists an edge in the set $\E(A)$ that crosses the cut $(S^{-},S^{+})$. Let $(i^*,j^*)$ be such an edge. Without loss of generality, we assume that $i^* \in S^{-}$ and $j^* \in S^{+}$. We define $$\begin{aligned}
C_{j^*}^{+} & = & \sum_{i \in S^{+}} a_{j^*i}, \\
C_{j^*}^{-} & =& \sum_{i \in S^{-}} a_{j^*i}. \end{aligned}$$ See Figure \[cdef\](a) for an illustration. Since $A$ is a -stochastic matrix, we have $$C_{j^*}^{-} + C_{j^*}^{+} = 1, $$implying that at least one of the following is true: $$\begin{aligned}
\mbox{ Case (a): } & & C_{j^*}^{-} \geq \frac{1}{2}, \\
\mbox{ Case (b): } & & C_{j^*}^{+} \geq \frac{1}{2}.\end{aligned}$$ We consider these [two cases]{} separately. In both cases, we focus on a subset of the edges and we use the fact that the elements $w_{ij}$ correspond to paths of length $2$, with one step in $\E(A)$ and another in $\E(A^T)$.
[ Intuitively, $C_{j^*}^{+}$ measures how much weight $j^*$ assigns to nodes in $S^{+}$ (including itself), and $C_{j^*}^{-}$ measures how much weight $j^*$ assigns to nodes in $S^{-}$. Note that the [edge]{} $(j^*,j^*)$ is also present, but not shown. (b) For the case [where]{} $C_{j^*}^{-} \geq 1/2$, we only focus on two-hop paths between $j^*$ and elements $i \in S^{-}$ obtained by taking $(i,j^*)$ as the first step and the self-edge $(j^*,j^*)$ as the second step. (c) For the case where $C_{j^*}^{+} \geq 1/2$, we only focus on two-hop paths between $i^*$ and elements $j\in S^{+}$ obtained by taking $(i^*,j^*)$ as the first step in $\E(A)$ and $(j^*,j)$ as the second step in $\E(A^T)$.](csnew.eps "fig:"){width="4cm"}  Intuitively, $C_{j^*}^{+}$ measures how much weight $j^*$ assigns to nodes in $S^{+}$ (including itself), and $C_{j^*}^{-}$ measures how much weight $j^*$ assigns to nodes in $S^{-}$. Note that the [edge]{} $(j^*,j^*)$ is also present, but not shown. (b) For the case [where]{} $C_{j^*}^{-} \geq 1/2$, we only focus on two-hop paths between $j^*$ and elements $i \in S^{-}$ obtained by taking $(i,j^*)$ as the first step and the self-edge $(j^*,j^*)$ as the second step. (c) For the case where $C_{j^*}^{+} \geq 1/2$, we only focus on two-hop paths between $i^*$ and elements $j\in S^{+}$ obtained by taking $(i^*,j^*)$ as the first step in $\E(A)$ and $(j^*,j)$ as the second step in $\E(A^T)$.](cplusbignew.eps "fig:"){width="4cm"}  Intuitively, $C_{j^*}^{+}$ measures how much weight $j^*$ assigns to nodes in $S^{+}$ (including itself), and $C_{j^*}^{-}$ measures how much weight $j^*$ assigns to nodes in $S^{-}$. Note that the [edge]{} $(j^*,j^*)$ is also present, but not shown. (b) For the case [where]{} $C_{j^*}^{-} \geq 1/2$, we only focus on two-hop paths between $j^*$ and elements $i \in S^{-}$ obtained by taking $(i,j^*)$ as the first step and the self-edge $(j^*,j^*)$ as the second step. (c) For the case where $C_{j^*}^{+} \geq 1/2$, we only focus on two-hop paths between $i^*$ and elements $j\in S^{+}$ obtained by taking $(i^*,j^*)$ as the first step in $\E(A)$ and $(j^*,j)$ as the second step in $\E(A^T)$.](c2bignew.eps "fig:"){width="4cm"} ]{}
[*Case (a):*]{} $C_{j^*}^{-} \geq 1/2.$\
We focus on those $w_{ij}$ with $i\in S^-$ and $j=j^*$. Indeed, since all $w_{ij}$ are nonnegative, we have $$\label{wsubset} \sum_{i \in S^{-}, ~j \in S^{+}}
w_{ij} \geq \sum_{i \in S^{-}} w_{ij^*}.$$ For each element in the sum on the right-hand [side]{}, we have $$w_{ij^*} = \sum_{k=1}^n a_{ki}\ a_{kj^*}
\geq a_{j^*i}\ a_{j^* j^*} \geq a_{j^*i}\ \eta,$$ where the inequalities follow from the facts that $A$ has nonnegative entries, its diagonal entries are positive, and its positive entries are at least $\eta$. Consequently, $$\sum_{i \in S^{-}} w_{ij^*}
\geq \eta\ \sum_{i \in S^{-}} a_{j^*i}
= \eta\, C_{j^*}^{-}. \label{wlowerbound}$$ Combining Eqs. (\[wsubset\]) and (\[wlowerbound\]), and recalling [the assumption]{} $C_{j^*}^{-} \geq 1/2$, the [result]{} follows. [An]{} illustration of [this argument]{} can be found in Figure \[cdef\][(b)]{}.
[*Case (b):*]{} $C_{j^*}^{+} \geq 1/2.$\
We focus on those $w_{ij}$ with $i=i^*$ and $j \in S^{+}$. We have $$\label{wsubset2}
\sum_{i \in S_{-}, ~j \in S^{+}} w_{ij} \geq \sum_{j \in
S^{+}} w_{i^* j},$$ since all $w_{ij}$ are nonnegative. For each element in the sum on the right-hand side, we have $$w_{i^* j} = \sum_{k=1}^n a_{ki^*}\ a_{k j}
\geq a_{j^* i^*}\ a_{j^* j} \geq \eta\, a_{j^* j},$$ where the inequalities follow [because all entries of $A$ are nonnegative, and because the choice $(i^*, j^*) \in \E(A)$ implies that $a_{j^* i^*} \geq{\eta}$.]{} Consequently, $$\sum_{j \in S^{+}} w_{i^* j} \geq \eta \sum_{j \in
S^{+}} a_{j^*j} = \eta\, C_{j^*}^{+}. \label{wlowerbound2}$$ [Combining Eqs. (\[wsubset2\]) and]{} (\[wlowerbound2\]), and recalling [the assumption]{} $C_{j^*}^+ \geq
1/2$, the [result]{} follows. An illustration of this argument can be found in Figure \[cdef\][(c)]{}.
A Bound on Convergence Time {#convergtime}
---------------------------
With the preliminaries on doubly stochastic matrices in place, we [can now proceed to derive]{} bounds on the decrease of $V(x(k))$ in between iterations. We will first somewhat relax our connectivity assumptions. In particular, we consider the following relaxation of Assumption \[connectivity\].
\[weakconnect\] Given an integer $k\ge 0$, suppose that the components of $x(kB)$ have been reordered so that they are in nonincreasing order. We assume that for every $d\in\{1,\ldots,n-1\}$, we either have $x_d(kB)=x_{d+1}(kB)$, or there exist some time $t\in\{kB,\ldots,(k+1)B-1\}$ and some $i\in\{1,\ldots,d\}$, $j\in\{d+1,\ldots,n\}$ such that $(i,j)$ or $(j,i)$ belongs to $\E(A(t))$.
\[as2implas3\] Assumption \[connectivity\] implies Assumption \[weakconnect\], [with the same value of $B$.]{}
If Assumption \[weakconnect\] does not hold, then $$\Big(N,\E({A}(kB)) \bigcup \E(A(kB+1)){\bigcup} \cdots \bigcup
\E(A((k+1)B-1))\Big)$$ is disconnected, which violates Assumption 2.
For our convergence time results, we will use the weaker Assumption \[weakconnect\], rather than the stronger Assumption 2. Later on, in Section \[matrixpicking\], we will exploit the sufficiency of Assumption \[weakconnect\] to [design]{} a decentralized algorithm for selecting the weights $a_{ij}(k)$, which satisfies Assumption \[weakconnect\], but not Assumption 2.
We now proceed to bound the decrease of our Lyapunov function $V(x(k))$ during the interval $[kB, (k+1) B-1]$. In what follows, we denote by $V(k)$ the sample variance $V(x(k))$ at time $k$.
\[vardiff\] Let Assumptions \[weights\] and \[weakconnect\] hold. Let $\{x(k)\}$ be generated by the update rule (\[noquant\]). Suppose that the components $x_i(kB)$ of the vector $x(kB)$ have been ordered from largest to smallest, with ties broken arbitrarily. Then, we have $$V(kB) - V((k+1)B) \geq
\frac{\eta}{2} \sum_{i=1}^{n-1} (x_{i}(kB) - x_{i+1}(kB))^2.$$
By Lemma \[vl\], we have for all $t$, $$V(t)-V(t+1) = \sum_{i<j} w_{ij}(t)
(x_i(t)-x_j(t))^2, \label{firstvdec0}$$ where $w_{ij}(t)$ is the $(i,j)$-th entry of $A(t)^T A(t)$. Summing up the variance differences $V(t)-V(t+1)$ over different values of $t$, we obtain $$V(kB) - V((k+1)B)
= \sum_{t=kB}^{(k+1)B-1} \sum_{i<j}
w_{ij}(t) (x_i(t)-x_j(t))^2. \label{firstvdec}$$
- For all $d\in\{1,\ldots,n-1\}$, let $t_d$ be the first time larger than or equal to $kB$ (if it exists) at which there is a communication between two nodes belonging to the two sets $\{1,\ldots,d\}$ and $\{d+1,\ldots,n\}$, to be referred to as a communication across the cut $d$.
- For all $t\in \{kB,\ldots,(k+1)B-1\}$, let $D(t)=\{d\mid
t_d=t\}$, i.e., $D(t)$ consists of “cuts" $d\in \{1,\ldots,n-1\}$ such that time $t$ is the first communication time larger than or equal to $kB$ between nodes in the sets $\{1,\ldots,d\}$ and $\{d+1,\ldots,n\}$. Because of Assumption \[weakconnect\], the union of the sets $D(t)$ includes all indices $1,\ldots,n-1$, except possibly for indices for which $x_d(kB)=x_{d+1}(kB)$.
- For all $d\in \{1,\ldots,n-1\}$, let $C_d=\{(i,j) \alexo{,~(j,i)} \mid i\leq d, \ d+1\leq j\}$.
- For all $t\in \{kB,\ldots,(k+1)B-1\}$, let $F_{ij}(t) =\{d\in D(t)\
|\ (i,j) \alexo{\mbox{ or } (j,i)} \in C_d\}$, i.e., $F_{ij}(t)$ consists of all cuts $d$ such that the edge $(i,j)$ at time $t$ is the first communication across the cut at a time larger than or equal to $kB$.
- To simplify notation, let $y_i=x_i(kB)$. By assumption, we have $y_1\geq\cdots\geq y_n$.
We make two observations, as follows:
- Suppose that $d\in D(t)$. Then, for some $(i,j)\in C_d$, we have either $a_{ij}(t)>0$ or $a_{ji}(t)>0$. by Lemma \[db\], we obtain $$\sum_{(i,j)\in C_d} w_{ij}(t)\geq \frac{\eta}{2}. \label{eq:w}$$
- Fix some $(i,j)$, with $i<j$, and time $t\in
\{kB,\ldots,(k+1)B-1\}$, and suppose that $F_{ij}(t)$ is nonempty. Let $F_{ij}\ao{(t)}=\{d_1,\ldots,d_k\}$, where the $d_j$ are arranged in increasing order. Since $d_1\in F_{ij}\ao{(t)}$, we have $d_1\in D(t)$ and therefore $t_{d_1}=t$. By the definition of $t_{d_1}$, this implies that there has been no communication between a node in $\{1,\ldots,d_1\}$ and a node in $\{d_1+1,\ldots,n\}$ during the time interval $[kB,t-1]$. It follows that $x_i(t) \geq
y_{d_1}$. By a symmetrical argument, we also have $$x_j(t)\leq y_{d_k+1}.\label{cutbound}$$ These relations imply that $$x_i(t)-x_j(t) \geq y_{d_1}-y_{d_k+1} \ao{\geq } \sum_{d\in F_{ij}\ao{(t)}} (y_d-y_{d+1}),$$ Since the components of $y$ are sorted in nonincreasing order, we have $y_d-y_{d+1}\geq 0$, for every $d\in F_{ij}\ao{(t)}$. For any nonnegative numbers $z_i$, we have $$(z_1+\cdots+z_k)^2\geq z_1^2+\cdots+z_k^2,$$ which implies that $$(x_i(t)-x_j(t))^2 \geq \sum_{d\in F_{ij}\ao{(t)}} (y_d-y_{d+1})^2.
\label{eq:dx}$$
We now use these two observations to provide a lower bound on the expression on the right-hand side of Eq. (\[firstvdec0\]) at time $t$. We use Eq. (\[eq:dx\]) and then Eq. (\[eq:w\]), to obtain $$\begin{aligned}
\sum_{i<j} w_{ij}(t)(x_i(t)-x_j(t))^2&
\geq& \sum_{i<j} w_{ij}(t) \sum_{d\in F_{ij}\ao{(t)}} (y_d-y_{d+1})^2\\
&=&\sum_{d\in D(t)} \sum_{(i,j)\in C_d} w_{ij}(t) (y_d-y_{d+1})^2\\
&\geq& \frac{\eta}{2} \sum_{d\in D(t)}(y_d-y_{d+1})^2.\end{aligned}$$ We now sum both sides of the above inequality for different values of $t$, and use Eq. (\[firstvdec\]), to obtain $$\begin{aligned}
V(kB)-V((k+1)B)&=& \sum_{t=kB}^{(k+1)B-1} \sum_{i<j} w_{ij}(t)(x_i(t)-x_j(t))^2\\
&\geq&
\frac{\eta}{2} \sum_{t=kB}^{(k+1)B-1}\sum_{d\in D(t)}(y_d-y_{d+1})^2\\
&=&\frac{\eta}{2}\sum_{d=1}^{n-1} (y_d-y_{d+1})^2,\end{aligned}$$ where the last inequality follows from the fact that the union of the sets $D(t)$ is only missing those $d$ for which $y_d=y_{d+1}$.
We next establish a bound on [the]{} variance decrease that plays [a]{} key role in our convergence analysis.
\[lboundvar\] Let Assumptions \[weights\] and \[weakconnect\] hold, and suppose that $V(kB)>0$. Then, $$\frac{V(kB)-V((k+1)B)}{V(kB)} \geq \frac{\eta}{2 n^2}\qquad
\hbox{for all }k.$$
[Without loss of generality, we assume that the components of $x(kB)$ have been sorted in nonincreasing order.]{} By Lemma \[vardiff\], we have $$V(kB) -
V((k+1)B) \geq \frac{\eta}{2}\, \sum_{i=1}^{n-1} (x_{i}(kB) -
x_{i+1}(kB))^2.$$ This implies that $$\frac{V(kB)-V((k+1)B)}{V(kB)}\ge\frac{\eta}{2} \,
\frac{\sum_{i=1}^{n-1} (x_{i}(kB) - x_{i+1}(kB))^2}{\sum_{i=1}^n
(x_i(kB)-\bar{x}(kB))^2}.$$ Observe that the right-hand side does not change when we add a constant to every $x_i(kB)$. We can therefore assume, without loss of generality, that $\bar{x}(kB)=0$, [so that]{} $$\frac{V(kB)-V((k+1)B)}{V(kB)} \geq \frac{\eta}{2}\,
\min_{{x_1\ge x_2\ge \cdots\ge x_n \atop \sum_i x_i=0}}
\frac{\sum_{i=1}^{n-1} (x_{i} - x_{i+1})^2}{\sum_{i=1}^n x_i^2}.$$ Note that the right-hand side is unchanged if we multiply each $x_i$ by the same constant. Therefore, we can assume, without loss of generality, that $\sum_{i=1}^n x_i^2=1$, [so that]{} $$\label{eq:vv}
\frac{V(kB)-V((k+1)B)}{V(kB)} \geq \frac{\eta}{2}\, \min_{{x_1\ge
x_2\ge \cdots\ge x_n \atop \sum_i x_i=0, \ \sum_i x_i^2 = 1}}
~~\sum_{i=1}^{n-1} (x_{i} - x_{i+1})^2.$$ The requirement $\sum_i x_i^2 = 1$ implies that the average value of $x_i^2$ is $1/n$, which implies that there exists some $j$ such that $|x_j| \ge 1/\sqrt{n}$. Without loss of generality, let us suppose that this $x_j$ is positive.[^6]
The rest of the proof relies on a technique from [@LO81] to provide a lower bound on the right-hand side of Eq. (\[eq:vv\]). Let $$z_i = x_{i} - x_{i+1} \ \ \hbox{for } i<n,\quad
\hbox{and}\quad z_n=0.$$ Note that $z_i \geq 0$ for all $i$ Since $x_j\ge 1/\sqrt{n}$ for some $j$, Combining with [Eq. (\[eq:vv\])]{}, we obtain $$\frac{V(kB)-V((k+1)B)}{V(kB)} \geq
\frac{\eta}{2} \min_{z_i \geq 0,\ \sum_i z_i \geq 1/\sqrt{n}}
\sum_{i=1}^{n} z_i^2.$$ [The minimization problem on the right-hand side is a symmetric convex optimization problem, and therefore has a symmetric optimal solution, namely]{} $z_i = 1/n^{1.5}$ for all $i$. This results in an optimal value of $1/n^2$. Therefore, $$\frac{V(kB)-V((k+1)B)}{V(kB)} \geq \frac{\eta}{2 n^2},$$ which [is]{} the desired result.
We are now ready for our main result, which establishes that the convergence time of the sequence of vectors $x(k)$ generated by Eq.(\[noquant\]) is of order $O(n^2B/\eta)$.
\[uqbound\] Let Assumptions \[weights\] and \[weakconnect\] hold. Then, there exists an absolute constant[^7] $c$ such that we have $$V({k})
\leq \epsilon V(0)\qquad \hbox{for all } k\ge {c} (n^2/\eta)B \log
(1/\epsilon).$$
The result follows immediately from Lemma \[lboundvar\].
Recall that, according to Lemma \[as2implas3\], Assumption \[connectivity\] implies Assumption \[weakconnect\]. In view of this, the convergence time bound of Theorem \[uqbound\] holds for any $n$ and any sequence of weights satisfying Assumptions \[weights\] and \[connectivity\]. In the next subsection, we show that this bound is tight when the stronger Assumption \[connectivity\] holds.
Tightness
---------
The next shows that the convergence time bound of Theorem \[uqbound\] is tight under Assumption \[connectivity\].
\[unquanttight\] [There exist constants $c$ and $n_0$ with the following property. For any $n\geq n_0$,]{} nonnegative integer $B$, $\eta < 1/2$, and $\epsilon < 1$, there exist a sequence of weight matrices $A(k)$ satisfying Assumptions \[weights\] and \[connectivity\], and an initial value $x(0)$ such that [if]{} $V(k)/V(0) \leq \epsilon$, [then]{} $$k \geq {c}\, \frac{n^2}{\eta}\, B \log \frac{1}{\epsilon}.$$
Let $P$ be the circulant shift operator defined by $P e_i = e_{i+1}$, $P e_n = e_1$, where $e_i$ is a unit vector with the $i$-th entry equal to 1, and all other entries equal to 0. Consider the symmetric circulant matrix defined by $$A = (1 - 2 \eta) I + \eta P + \eta P^{-1}.$$ Let $A({k})=A$, when ${k}$ is a multiple of $B$, and $A({k})=I$ otherwise. Note that this sequence satisfies Assumptions 1 and 2.
The second largest eigenvalue of $A$ is $$\lambda_2(A) = 1 - 2 \eta + 2 \eta \cos \frac{2 \pi}{n},$$ (see Eq. (3.7) of [@G06]). Therefore, using the inequality $\cos x \geq 1- x^2/2$, $$\lambda_2(A) \geq 1 - \frac{4 \eta \pi^2}{n^2}.$$ [For $n$ large enough,]{} the quantity on the right-hand side is nonnegative. Let the initial [vector]{} $x(0)$ be the eigenvector corresponding to [$\lambda_2(A)$.]{} Then, $$\frac{V(kB)}{V(0)} = \lambda_2(A)^{2k} \geq \Big(1 - \frac{8 \eta
\pi^2}{n^2}\Big)^k.$$ [For the right-hand side to become]{} less than $\epsilon$, we need $k = \Omega( (n^2/\eta) \log (1/\epsilon))$. [This implies that for $V(k)/V(0)$ to become less than $\epsilon$, we need]{} $k=\Omega( (n^2/\eta) B \log (1/\epsilon))$.
Saving a [factor of]{} $n$: faster averaging on undirected graphs {#matrixpicking}
=================================================================
In the previous section, we have shown that a large class of averaging algorithms have $O(B (n^2/\eta) ~\alexo{ \log
1/\epsilon}) $
In this section, we consider decentralized ways of synthesizing the weights $a_{ij}(k)$ while satisfying Assumptions \[weights\] and \[weakconnect\].
We assume that the communications of the nodes are governed by an exogenous sequence of graphs $\Gb(k)=(N,\Eb(k))$ that provides strong connectivity over time periods of length $B$. in particular, we require that $a_{ij}(k)=0$ if $(j,i)\notin\Eb(k)$. Naturally, we assume that $(i,i)\in\Eb(k)$ for every $i$.
Several such decentralized protocols exist. For example, each node may assign $$\begin{aligned}
a_{ij}{(k)} & = & \epsilon, \qquad\qquad \mbox{ \ \ if
} (j,i)
\in \Eb(k) {\mbox{ and } i\neq j,} \\
a_{ii}{(k)} & = & 1 - \epsilon\cdot \mbox{deg}(i),\end{aligned}$$ where [deg]{}($i$) is the degree of $i$ in $\Gb(k)$. If $\epsilon$ is small enough and the graph $\Gb(k)$ is undirected [\[i.e., $(i,j)\in\Eb(k)$ if and only if $(j,i)\in\Eb(k)$\],]{} this results in a nonnegative, doubly stochastic matrix (see [@OSM04]). However, [if a node has $\ao{\Theta}(n)$ neighbors, $\eta$ will be of order $\ao{\Theta}(1/n)$, resulting in $\ao{\Theta}(n^3)$ convergence time.]{} Moreover, this argument applies to all protocols in which nodes assign equal weights to all their neighbors; see [@XB04] and [@BFT05] for more examples.
In this section, we examine whether it is possible to synthesize the weights $a_{ij}(k)$ in a decentralized manner, so that $a_{ij}(k)\geq \eta$ whenever $a_{ij}(k)\neq 0$, where $\eta$ is a positive constant independent of $n$ and $B$. We show that this is indeed possible, under the additional assumption that the graphs $\Gb(k)$ are undirected. Our algorithm is data-dependent, in that $a_{ij}(k)$ depends not only on the graph $\Gb(k)$, but also on the data vector $x(k)$. Furthermore, it is a decentralized 3-hop algorithm, in that $a_{ij}{(k)}$ depends only on the data at nodes within a distance of at most $3$ from $i$. Our algorithm is such that the resulting sequences of vectors $x(k)$ and graphs $G(k)=(N,\E(k))$, with $\E(k)=\{(j,i)\mid a_{ij}(k)> 0\}$, satisfy Assumptions \[weights\] and \[weakconnect\]. Thus, a convergence time result can be obtained from Theorem \[uqbound\].
The algorithm
-------------
The algorithm we present here is a variation of an old [*load balancing*]{} algorithm (see [@C89] and Chapter 7.3 of [@BT89]).[^8]
At each step of the algorithm, each node offers some of its value to its neighbors, and accepts or rejects such offers from its neighbors. Once an offer from $i$ to $j$, of size $\delta>0$, has been accepted, the updates $x_i \leftarrow x_i - \delta$ and $x_j
\leftarrow x_j + \delta$ are executed.
We next describe the formal steps the nodes execute at each time ${k}$. For clarity, we refer to the node executing the steps below as node $\ao{C}$. Moreover, the instructions below sometimes refer to the neighbors of node $\ao{C}$; this always means current neighbors at time $k$, when the step is being executed, [as determined by the current graph $\Gb(k)$.]{} We assume that at each time $k$, all nodes execute these steps in the order described below, while the graph remains unchanged.
[**Balancing Algorithm:**]{}
1. Node $\ao{C}$ broadcasts its current value $x_{\ao{C}}$ to all its neighbors.
2. Going through the values it just received from its neighbors, Node $\ao{C}$ finds the smallest value that is less than [its own]{}. Let $\ao{D}$ be a neighbor with this value. Node $\ao{C}$ makes an offer of $(x_{\ao{C}} - x_{\ao{D}})/3$ to node $D$.
If no [neighbor of $\ao{C}$]{} has a value smaller than $x_{\ao{C}}$, node $C$ does nothing at this stage.
3. Node $\ao{C}$ goes through the incoming offers. It sends an acceptance to the sender of [a]{} largest offer, and a rejection to all the other senders. It updates the value of $x_{\ao{C}}$ by adding the value of the accepted offer.
If node $C$ did not receive any offers, it does nothing at this stage.
4. If an acceptance arrives [for]{} the offer made by node $\ao{C}$, node $\ao{C}$ updates $x_{\ao{C}}$ by subtracting the value of the offer.
Note that the new value of each node is a linear combination of the values of its neighbors. Furthermore, the weights $a_{ij}(k)$ are completely determined by the data and the graph at most $3$ hops from node $i$ in $\Gb(k)$.
Performance analysis
--------------------
.
\[savingn\] Consider the balancing algorithm, and suppose that $\Gb(k)=(N,\Eb(k))$ is a sequence of undirected graphs such that $(N,\Eb(kB)\cup\Eb(kB+1)\cup \cdots \cup
\Eb((k+1)B-1))$ is connected, for all integers $k$. There exists an absolute constant $c$ such that we have $$V({k}) \leq \epsilon V(0)\qquad \hbox{for all }k\ge {c}
n^2B \log (1/\epsilon).$$
Note that with this algorithm, the new value at some node $i$ is a convex combination of the previous values of itself and its neighbors. Furthermore, the algorithm keeps the sum of the nodes’ values constant, because every accepted offer involves an increase at the receiving node equal to the decrease at the offering node. These two properties imply that the algorithm can be written in the form $$x(k+1) = A(k) x(k),$$ where $A(k)$ is a doubly stochastic matrix, determined by $\Gb(k)$ and $x(k)$. It can be seen that the diagonal entries of $A(k)$ are positive and, furthermore, all nonzero entries of $A(k)$ are larger than or equal to 1/3; thus, $\eta=1/3$.
We claim that the algorithm \[in particular, the sequence $\E(A(k))$\] satisfies Assumption \[weakconnect\]. Indeed, suppose that at time $kB$, the nodes are reordered so that the values $x_i(kB)$ are nonincreasing in $i$. Fix some $d\in\{1,\ldots,n-1\}$, and suppose that $x_d(kB)\neq x_{d+1}(kB)$. Let $S^+=\{1,\ldots,d\}$ and $S^-=\{d+1,\ldots,n\}$.
Because of our assumptions on the graphs $\Gb(k)$, there will be a first time $t$ in the interval $\{kB,\ldots,(k+1)B-1\}$, at which there is an edge in $\Eb(t)$ between some $i^*\in S^+$ and $j^*\in
S^-$. Note that between times $kB$ and $t$, the two sets of nodes, $S^+$ and $S^-$, do not interact, which implies that $x_i(t)\geq
x_d(kB)$, for $i\in S^+$, and $x_j(t) < x_d(kB)$, for $j\in S^-$.
At time $t$, node $i^*$ sends an offer to a neighbor with the smallest value; let us denote that neighbor by $k^*$. Since $(i^*,j^*)\in \Eb(t)$, we have $x_{k^*}(t)\leq x_{j^*}(t)<x_d(kB)$, which implies that $k^*\in S^-$. Node $k^*$ will accept the largest offer it receives, which must come from a node with a value no smaller than $x_{i^*}(t)$, ; hence the latter node belongs to $S^+$. It follows that $\E(A(t))$ contains an edge between $k^*$ and some node in $S^{+}$, showing that Assumption \[weakconnect\] is satisfied.
The claimed result follows from Theorem \[uqbound\], because we have shown that all of the assumptions in that theorem are satisfied .
Quantization Effects \[qanalysis\]
==================================
In this section, we consider a quantized version of the update rule (\[noquant\]). This model is a good approximation for a network of nodes communicating through [finite bandwidth channels, so that]{} at each time instant, only a finite number of bits can be transmitted. We incorporate this constraint in our algorithm by assuming that each node, upon receiving the values of its neighbors, computes the convex combination $\sum_{j=1}^{{n}} a_{ij}(k) x_j(k)$ and quantizes it. This update rule also captures [a]{} constraint that each node can only store quantized values.
Unfortunately, [under Assumptions \[weights\] and \[connectivity\], if the output of Eq. (\[noquant\]) is rounded to the nearest integer, the sequence $x(k)$]{} is not guaranteed to converge to consensus; see [@KBS06]. We therefore choose a quantization rule that rounds the values down, according to $$\label{quantupdate}
x_i(k+1) = \left\lfloor \sum_{j=1}^{{n}} a_{ij}(k) x_j(k)
\right\rfloor,$$ where $\lfloor \cdot \rfloor$ represents rounding [*down*]{} to the nearest multiple of $1/Q$, and where $Q$ is some positive integer.
We adopt the natural assumption that the initial values are [already]{} quantized.
For all $i$, $x_i(0)$ is a multiple of $1/Q$.\[quantizedinitials\]
For convenience we define $$U = \max_i x_i(0), \qquad L = \min_i
x_i(0).$$ We use $K$ to denote the total number of [relevant]{} quantization levels, i.e., $$K = (U-L)Q,$$ which is an integer by Assumption \[quantizedinitials\].
A quantization level dependent bound
------------------------------------
We first present a [convergence time bound that depends on the quantization level $Q$.]{}
\[simpleq\] Let Assumptions \[weights\], \[connectivity\], and \[quantizedinitials\] hold. Let $\{{x(k)}\}$ be generated by the update rule (\[quantupdate\]). If $k \geq nBK$, [then]{} all [components]{} of $x(k)$ are equal.
Consider the nodes [whose initial value is $U$.]{} There are at most $n$ of them. [As long as]{} not all entries of ${x(k)}$ are equal, then every $B$ iterations, at least one [node]{} must use a value strictly less than $U$ [in]{} an update; node will have its value decreased to $U-1/Q$ or less. It follows that after $nB$ iterations, the largest node [value will be]{} at most $U-1/Q$. Repeating this argument, we [see]{} that at most $nBK$ iterations are possible before [all the nodes have]{} the same value.
Although the above bound gives informative results for small $K$, it becomes weaker as [$Q$ (and, therefore, $K$)]{} increases. On the other hand, as $Q$ approaches infinity, the quantized system approaches the unquantized system; the availability of convergence time bounds for the unquantized system suggests that similar bounds should be possible for the quantized one. Indeed, in the next subsection, obtain a bound on the convergence time [which is]{} independent of the total number of quantization levels.
A quantization level independent bound
--------------------------------------
We adopt a slightly different measure of convergence for the analysis of the quantized consensus algorithm. For any $x\in\R^n$, we define $m(x)=\min_i x_i$ and $$\underline{V}(x)=\sum_{i=1}^n (x_i - m(x))^2.$$ We will also use the simpler notation $m(k)$ and $\underline{V}(k)$ to denote $m(x(k))$ and $\underline{V}(x(k))$, respectively, where it is more convenient to do so. The function $\underline{V}$ will be our Lyapunov function for the analysis of the quantized consensus algorithm. The reason for not using our earlier Lyapunov function, $V$, is that for the quantized algorithm, $V$ is not guaranteed to be monotonically nonincreasing in time. On the other hand, we have that $V(x)\leq \underline{V}(x) \leq \alexo{4} \ao{n}V(x)$ for any[^9] $x \in \R^n$. As a consequence, any convergence time bounds expressed in terms of $\underline{V}$ translate to essentially the same bounds expressed in terms of $V$, .
Before proceeding, we [record an elementary fact which will allow us to]{} relate the variance decrease $V(x)-V(y)$ to the decrease, $\underline{V}(x)- \underline{V}(y)$, of our new Lyapunov function. The proof involves [simple algebra,]{} and is therefore omitted.
\[sumlemma\] Let $u_1,\ldots,u_{n}$ and $w_1,\ldots,w_n$ be real numbers satisfying $$\sum_{i=1}^n u_i =
\sum_{i=1}^n w_i.$$ [Then, the expression]{} $$f(z) = \sum_{i=1}^n (u_i - z)^2 - \sum_{i=1}^n (w_i - z)^2$$ is [a constant,]{} independent of the scalar $z$.
Our next lemma places a bound on [the decrease of]{} the Lyapunov function $\underline{V}(t)$ between times $kB$ and $(k+1)B-1$.
\[quantdiff\] Let Assumptions \[weights\], \[weakconnect\], and \[quantizedinitials\] hold. Let $\{x{(k)}\}$ be generated by the update rule (\[quantupdate\]). Suppose that the components $x_i(kB)$ of the vector $x(kB)$ have been ordered from largest to smallest, with ties broken arbitrarily. Then, we have $$\underline{V}(kB) - \underline{V}((k+1)B)
\geq \frac{\eta}{2}\, \sum_{i=1}^{n-1} (x_{i}(kB)
- x_{i+1}(kB))^2.$$
For all $k$, we view Eq. (\[quantupdate\]) as the composition of two operators: $$y(k) = A(k) x(k),$$ where $A(k)$ is [a]{} doubly stochastic matrix, and $$x(k+1) = \lfloor y(k)
\rfloor,$$ where the quantization $\lfloor \cdot \rfloor$ is [carried out]{} componentwise.
We apply Lemma \[sumlemma\] with the identification $u_i=x_{{i}}(k)$, $w_i=y_{{i}}(k)$. Since multiplication by a doubly stochastic matrix preserves the mean, the condition $\sum_i u_i =
\sum_i w_i$ is satisfied. By considering two different choices for the scalar $z$, namely, $z_1=\bar{x}(k)=\bar{y}(k)$ and $z_2=m(k)$, we obtain $$\label{firststep} V(x(k)) - V(y(k)) =
\underline{V}(\ao{x(k)}) - \sum_{i=1}^n (y_i(k)-m(k))^2.$$ Note that $x_i(k+1)-m(k)\leq y_i(k)-m(k)$. Therefore, $$\label{secondstep} \underline{V}(\ao{x(k)}) -
\sum_{i=1}^n (y_i(k)-m(k))^2 \leq \underline{V}(\ao{x(k)}) -
\sum_{i=1}^n (x_i(k+1)-m(k))^2.$$ Furthermore, note that $x_i(k+1)-m(k+1)\leq x_i(k+1)-m(k)$. Therefore, $$\label{thirdstep} \underline{V}(\ao{x(k)}) -
\sum_{i=1}^n (x_i(k+1)-m(k))^2 \leq \underline{V}(\ao{x(k)}) -
\underline{V}(\ao{x(k+1)}).$$ By combining Eqs. (\[firststep\]), (\[secondstep\]), and (\[thirdstep\]), we obtain $$V(x(t)) - V(y(t)) \leq \underline{V}(\ao{x(t)}) -
\underline{V}(\ao{x(t+1)})\qquad \hbox{for all }t.$$ Summing the preceding relations over $t=kB,\ldots, (k+1)B-1$, we further obtain $$\sum_{t=kB}^{(k+1)B-1} \Big(V(x(t))-V(y(t))\Big) \leq
\underline{V}(\ao{x(kB)}) - \underline{V}(\ao{x((k+1)B))}.$$ To complete the proof, we provide a lower bound on the expression $$\sum_{t=kB}^{(k+1)B-1} \Big(V(x(t)) - V(y(t))\Big).$$ Since $y(t)=A(t)x(t)$ for all $t$, it follows from Lemma \[vl\] that for any $t$, $$V(x(t))-V(y(t)) = \sum_{i<j} w_{ij}(t)
(x_i(t)-x_j(t))^2,$$ where $w_{ij}(t)$ is the $(i,j)$-th entry of $A(t)^T A(t)$. Using this relation and following the same line of analysis used in the proof of Lemma \[vardiff\] \[where the relation $\alexo{x_i(t) \geq y_{d_1}}$ holds in view of the assumption that $x_i(kB)$ is a multiple of $1/Q$ for all $k\ge 0$, cf. Assumption \[quantizedinitials\]\] , we obtain the desired result.
The next theorem contains our main result on the convergence time of the quantized algorithm.
\[qbound\] Let Assumptions \[weights\], \[weakconnect\], and \[quantizedinitials\] hold. Let $\{x{(k)}\}$ be generated by the update rule (\[quantupdate\]). Then, there exists an absolute constant $c$ such that we have $$\underline{V}(k) \leq \epsilon \underline{V}(0)\qquad \hbox{for all }
k\ge c\, (n^2/\eta) B
\log(1/\epsilon).$$
Let us assume that $\underline{V}(kB)>0$. From Lemma \[quantdiff\], we have $$\underline{V}(kB)- \underline{V}((k+1)B) \geq \frac{\eta}{2}
\sum_{i=1}^{n-1} (x_{i}(kB) - x_{i+1}(kB))^2,$$ where [the components]{} $x_i(kB)$ are ordered from largest to smallest. Since $\underline V(kB) = \sum_{i=1}^n (x_i(kB) - x_n(kB))^2$, we have $$\frac{\underline{V}(kB)-\underline{V}((k+1)B)}{\underline{V}(kB)}
\geq \frac{\eta}{2} \frac{\sum_{i=1}^{n-1} (x_{i}(kB) -
x_{i+1}(kB))^2}{\sum_{i=1}^n (x_i(kB)-x_n(kB))^2}.$$ [Let]{} $y_i = x_i(kB)-x_n(kB)$. Clearly, $y_i \geq 0$ for all $i$, and $y_n=0$. Moreover, the monotonicity of $x_i(kB)$ implies the monotonicity of $y_i$: $$y_1 \geq y_2 \geq \cdots \geq y_n=0.$$ Thus, $$\frac{\underline{V}(kB)-\underline{V}((k+1)B)}{\underline{V}(kB)}
\geq \frac{\eta}{2} \min_{{y_1 \geq y_2\ge \cdots\ge y_n\atop
y_n=0}}\frac{\sum_{i=1}^{n-1} (y_i - y_{i+1})^2}{\sum_{i=1}^n
y_i^2}.$$ $$\frac{\eta}{2} \min_{{y_1 \geq y_2\geq \cdots \geq y_n\atop
\sum_i y_i^2=1}} \sum_{i=1}^{n-1} (y_i - y_{i+1})^2 \geq
\frac{\eta}{2} \min_{z_i \geq 0, \sum_i z_i \geq 1/\sqrt{n}}
\sum_{i=1}^n z_i^2.$$ The minimization problem on the right-hand side has an optimal value of at least $1/n^2$, and the desired result follows.
Extensions and modifications
----------------------------
In this subsection, we comment briefly on some corollaries of Theorem \[qbound\].
First, we note that the results of Section \[matrixpicking\] immediately carry over to the quantized case. Indeed, in Section \[matrixpicking\], we showed how to pick the weights $a_{ij}(k)$ in a decentralized manner, based only on local information, so that Assumptions 1 and \[weakconnect\] are satisfied, with $\eta \geq
1/3$. When using a quantized version of the balancing algorithm, we once again
\[savingnq\] For the quantized version of the balancing algorithm, and under the same assumptions as in Theorem \[savingn\], if $k\geq c\, n^2B \log (1/\epsilon))$, then $\underline{V}(k ) \leq
\epsilon \underline{V}(0)$, where $c$ is an absolute constant.
Second, we note that Theorem \[qbound\] [can be used]{} to obtain a bound on the time until [the values of all nodes are equal. Indeed, we observe that in the presence of quantization, once the condition $\underline{V}(k) <1/Q^2$ is satisfied, all components of $x(k)$ must be equal.]{}
\[qboundequal\] Consider the quantized algorithm (\[quantupdate\]), and assume that Assumptions \[weights\], \[weakconnect\], and \[quantizedinitials\] hold. If $k\geq c(n^2/\eta) B \big[\log Q +
\log \underline{V}(0)\big]$, then all components of $x(k)$ are equal, where $c$ is an absolute constant.
Tightness
---------
[We now]{} show that the bound in Theorem \[qbound\] is tight, Assumption \[weakconnect\] is replaced with Assumption \[connectivity\].
\[quanttight\] There absolute constant $c$ with the following property. For [any]{} nonnegative integer $B$, $\eta < 1/2$, $\epsilon < 1$, and , there exist a sequence of weight matrices $A(k)$ satisfying Assumptions \[weights\] and \[connectivity\], and an initial value $x(0)$ satisfying Assumption \[quantizedinitials\], such that under the dynamics of Eq. (\[quantupdate\]), [if]{} $\underline{V}(k)/\underline{V}(0) \leq \epsilon$, [then]{} $$k\geq c\, \frac{n^2}{\eta}\, B \log \frac{1}{\epsilon}.$$
We have demonstrated in \[unquanttight\] a similar result for the unquantized algorithm. Namely, we have shown that for [$n$ and for any]{} $B$, $\eta<1/2$, and $\epsilon<1$, there exists a weight sequence $a_{ij}(k)$ and an initial [vector]{} $x(0)$ such that the first time when $V(t) \leq \epsilon V(0)$ occurs after $\Omega((n^2/\eta) B
\log (1/\epsilon))$ steps. Let $T^*$ be this first time.
[Consider the quantized algorithm under]{} the exact same sequence $a_{ij}(k)$, [initialized at $\lfloor x(0) \rfloor$.]{} Let $\hat{x}_i(t)$ refer to the value of node $i$ at time $t$ in the quantized algorithm [under]{} this scenario, [as opposed to]{} $x_i(t)$ which [denotes the]{} value in the unquantized algorithm. Since quantization can [only]{} decrease a nodes value by at most $1/Q$ at each iteration, it is easy to show, by induction, that $$x_i(t) \geq \hat{x}_i(t) \geq x_i(t) - t/Q$$ We can pick $Q$ large enough so that, [for $t < T^*$,]{} the vector $\hat{x}(t)$ is [as close as desired to]{} $x(t)$.
Therefore, for $t < T^*$ and for large enough $Q$, $\underline{V}(\hat{x}(t))/\underline{V}(\hat{x}(0))$ will be arbitrarily close to $\underline{V}(x(t))/\underline{V}(x(0))$. [From the proof of \[unquanttight\], we see that]{} $x(t)$ is always a [scalar]{} multiple of $x(0)$. Since $\underline{V}(x)/V(x)$ is invariant under multiplication by a constant, it follows that $\underline{V}(x(t))/\underline{V}(x(0)) =
V(x(t))/V(x(0))$. Since this last quantity is above $\epsilon$ for $t<T^*$, it follows that provided $Q$ is large enough, $\underline{V}(\hat{x}(t))/\underline{V}(\hat{x}(0))$ is also above $\epsilon$ for $t<T^*$. This proves the proposition.
Quantization error
------------------
Despite favorable convergence properties of our quantized [averaging]{} algorithm (\[quantupdate\]), the update rule does not preserve the average of the values at each iteration. Therefore, the [common limit of the sequences $x_i(k)$, denoted by $x_f$,]{} need not be equal to the exact average of the initial values. We next provide an upper bound on the error between [$x_f$]{} and the initial average, as a function of the number of quantization levels.
\[qerror\] There is an absolute constant $c$ such that for the common limit $x_f$ of the values $x_i(k)$ generated by the quantized algorithm (\[quantupdate\]), we have $$\left|x_f - \frac{1}{n} \sum_{i=1}^n x_i(0)\right| \le {c\over Q}
\ {n^2\over\eta}\, B\log (Qn(U-L)).$$
By \[qboundequal\], after $O\Big((n^2/\eta) B \log (Q
\underline{V}(x(0)))\Big)$ iterations, all nodes will have the same value. Since $\underline{V}(x(0)))\le n(U-L)^2$ and the average decreases by at most $1/Q$ at each iteration, the result follows.
Let us assume that the parameters $B$, $\eta$, and $U-L$ are fixed. \[qerror\] implies that as $n$ increases, the number of bits [used for each communication,]{} which is proportional to $\log Q$, needs to grow only as $O(\log n)$ to make the error negligible. Furthermore, this is true even if the [parameters]{} $B$, ${1}/{\eta}$, and $U-L$ grow polynomially in $n$.
[For a converse, it can]{} be seen that ${\Omega}(\log n)$ bits are needed. Indeed, consider $n$ nodes, with $n/2$ [nodes initialized at]{} $0$, and $n/2$ [nodes initialized at]{} $1$. [Suppose]{} that ${Q} <
n/2$; we connect the nodes by [forming]{} a complete subgraph over all the nodes with value $0$ and exactly [one]{} node with value $1$; see Figure \[ic\] for an example with $n=6$. Then, each node [forms the]{} average [of]{} its neighbors. This brings one of the nodes with [an initial value of]{} $1$ down to $0$, without raising the value of any [other]{} nodes. We can repeat this [process,]{} to bring all of the nodes with [an initial value of]{} $1$ down to $0$. Since the true average is $1/2$, the final result is $1/2$ away from the true average. [Note now that $Q$ can]{} grow linearly with $n$, and still satisfy the inequality ${Q}<n/2$. [Thus,]{} the number of bits can grow as $\Omega(\log n)$, and yet, independent of $n$, the error remains $1/2$.
![\[ic\] Initial configuration. Each node takes the average value of its neighbors. ](newlb.eps){width="12cm"}
Conclusions
===========
We studied distributed algorithms for the [averaging]{} problem over networks with time-varying topology, [with a focus on]{} tight bounds on the convergence time of a general class of [averaging]{} algorithms. We first considered [algorithms for the case where]{} agents can exchange and store continuous values, [and established tight convergence time bounds.]{} We next studied averaging algorithms under the additional constraint that agents can [only]{} store and send quantized values. We showed that these algorithms guarantee convergence of the agents values to consensus within some error from the average of the initial values. We provided a bound on the error that highlights the dependence on the number of quantization levels.
Our paper is a contribution to the growing literature on distributed control of multi-agent systems. Quantization effects are an integral part of such systems but, with the exception of a few recent studies, have not attracted much attention in the vast literature on this subject. In this paper, we studied a quantization scheme that guarantees consensus at the expense of some error from the initial average value. We used this scheme to study the effects of the number of quantization levels on the convergence time of the algorithm and the distance from the true average.
The framework provided in this paper motivates a number of further research directions:
- The algorithms studied in this paper assume that there is no delay in receiving the values of the other agents, which is a restrictive assumption in network settings. Understanding the convergence of averaging algorithms and implications of quantization in the presence of delays is an important topic for future research.
- We studied a quantization scheme with favorable convergence properties, that is, rounding down to the nearest quantization level. Investigation of other quantization schemes and their impact on convergence time and error is left for future work.
- The quantization algorithm we adopted implicitly assumes that the agents can carry out computations with continuous values, but can store and transmit only quantized values. Another interesting area for future work is to incorporate the additional constraint of finite precision computations into the quantization scheme.
-
[^1]: A. Nedić is with the Industrial and Enterprise Systems Engineering Department, University of Illinois at Urbana-Champaign, Urbana IL 61801 (e-mail:[email protected])
[^2]: A. Olshevsky, A. Ozdaglar, and J. N. Tsitsiklis are with the Laboratory for Information and Decision Systems, Electrical Engineering and Computer Science Department, Massachusetts Institute of Technology, Cambridge MA, 02139 (e-mails: alex\[email protected], [email protected], [email protected])
[^3]: This research was partially supported by the National Science Foundation under grants ECCS-0701623, CMMI 07-42538, and DMI-0545910, and by DARPA ITMANET program
[^4]:
[^5]: In the sequel, the notation $\sum_{i<j}$ will be used to denote the double sum $\sum_{j=1}^n\sum_{i=1}^{j-1}$.
[^6]: Otherwise, we can replace $x$ with $-x$ and subsequently reorder to maintain the property that the components of $x$ are in descending order. It can be seen that these operations do not affect the objective value.
[^7]:
[^8]: This algorithm was also considered in [@OT06], but in the absence of a result such as Theorem \[uqbound\], a weaker convergence time bound was derived.
[^9]:
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
The CoGeNT collaboration has recently published results from a fifteen month data set which indicate an annual modulation in the event rate similar to what is expected from weakly interacting massive particle interactions. It has been suggested that the CoGeNT modulation may actually be caused by other annually modulating phenomena, specifically the flux of atmospheric muons underground or the radon level in the laboratory. We have compared the phase of the CoGeNT data modulation to that of the concurrent atmospheric muon and radon data collected by the MINOS experiment which occupies an adjacent experimental hall in the Soudan Underground Laboratory. The results presented are obtained by performing a shape-free $\chi^{2}$ data-to-data comparison and from a simultaneous fit of the MINOS and CoGeNT data to phase-shifted sinusoidal functions. Both tests indicate that the phase of the CoGeNT modulation is inconsistent with the phases of the MINOS muon and radon modulations at the level.\
author:
- 'P. Adamson'
- 'I. Anghel'
- 'G. Barr'
- 'M. Bishai'
- 'A. Blake'
- 'G. J. Bock'
- 'D. Bogert'
- 'S. V. Cao'
- 'S. Childress'
- 'J. A. B. Coelho'
- 'L. Corwin'
- 'D. Cronin-Hennessy'
- 'J. K. de Jong'
- 'A. V. Devan'
- 'N. E. Devenish'
- 'M. V. Diwan'
- 'C. O. Escobar'
- 'J. J. Evans'
- 'E. Falk'
- 'G. J. Feldman'
- 'M. V. Frohne'
- 'H. R. Gallagher'
- 'R. A. Gomes'
- 'M. C. Goodman'
- 'P. Gouffon'
- 'N. Graf'
- 'R. Gran'
- 'K. Grzelak'
- 'A. Habig'
- 'J. Hartnell'
- 'R. Hatcher'
- 'A. Himmel'
- 'A. Holin'
- 'J. Hylen'
- 'G. M. Irwin'
- 'Z. Isvan'
- 'D. E. Jaffe'
- 'C. James'
- 'D. Jensen'
- 'T. Kafka'
- 'S. M. S. Kasahara'
- 'G. Koizumi'
- 'S. Kopp'
- 'M. Kordosky'
- 'A. Kreymer'
- 'K. Lang'
- 'J. Ling'
- 'P. J. Litchfield'
- 'P. Lucas'
- 'W. A. Mann'
- 'M. L. Marshak'
- 'M. Mathis'
- 'N. Mayer'
- 'A. M. McGowan'
- 'M. M. Medeiros'
- 'R. Mehdiyev'
- 'J. R. Meier'
- 'M. D. Messier'
- 'W. H. Miller'
- 'S. R. Mishra'
- 'S. Moed Sher'
- 'C. D. Moore'
- 'L. Mualem'
- 'J. Musser'
- 'D. Naples'
- 'J. K. Nelson'
- 'H. B. Newman'
- 'R. J. Nichol'
- 'J. A. Nowak'
- 'J. O’Connor'
- 'W. P. Oliver'
- 'M. Orchanian'
- 'R. B. Pahlka'
- 'J. Paley'
- 'R. B. Patterson'
- 'G. Pawloski'
- 'S. Phan-Budd'
- 'R. K. Plunkett'
- 'X. Qiu'
- 'A. Radovic'
- 'B. Rebel'
- 'C. Rosenfeld'
- 'H. A. Rubin'
- 'M. C. Sanchez'
- 'J. Schneps'
- 'A. Schreckenberger'
- 'P. Schreiner'
- 'R. Sharma'
- 'A. Sousa'
- 'N. Tagg'
- 'R. L. Talaga'
- 'J. Thomas'
- 'M. A. Thomson'
- 'R. Toner'
- 'D. Torretta'
- 'G. Tzanakos'
- 'J. Urheim'
- 'P. Vahle'
- 'B. Viren'
- 'A. Weber'
- 'R. C. Webb'
- 'C. White'
- 'L. Whitehead'
- 'S. G. Wojcicki'
- 'R. Zwaska'
bibliography:
- 'DarkMatter.bib'
title: Comparisons of annual modulations in MINOS with the event rate modulation in CoGeNT
---
Introduction
============
Numerous astrophysical observations strongly support the existence in our galaxy of a cold dark matter halo, that may consist of Weakly Interacting Massive Particles (WIMPs) [@Feng:2010gw; @Bertone:2004pz]. The principal search mode of direct WIMP detection is the identification of an $\mathcal{O}$(keV) nuclear recoil produced by WIMP-nucleus elastic scattering. Since the speed of the Earth relative to the dark matter halo varies depending on the Earth’s velocity with respect to the Sun, the dark matter detection rate is expected to demonstrate a annual modulation. This modulation is expected to be at a maximum (minimum) on June 2 (Dec. 2) with an amplitude between a few and , assuming the standard halo model [@Drukier:1986tm; @Freese:1987wu; @Frandsen:2011gi]. The CoGeNT [@Aalseth:2011wp; @Aalseth:2010vx], DAMA/LIBRA [@Bernabei:2010mq] and CRESST-II [@Angloher:2011uu] collaborations have all reported an excess of events above all known backgrounds. The CoGeNT [@Aalseth:2011wp] and DAMA/LIBRA [@Bernabei:2010mq; @Bernabei:2008yh] collaborations have also claimed evidence for annual modulations in their event rates at and respectively. Fits to the available data favor a light WIMP with mass and spin-independent cross-section [@Hooper:2012ft; @Kelso:2011gd; @Kopp:2011yr].\
The null observations by CDMS-II [@Ahmed:2010hw; @Ahmed:2010wy; @Ahmed:2012vq], XENON100 [@Angle:2011th; @Aprile:2011hi; @Aprile:2012hi] and EDELWEISS [@Armengaud:2012kd] exclude much of the allowed WIMP signal regions mentioned above [@HerreroGarcia:2012fu]. The tension between these exclusion limits and the positive observations can be significantly reduced, but not removed, when taking into account experimental [@Collar:2012ed; @Collar:2011wq] and astrophysical uncertainties [@Kopp:2011yr; @Schwetz:2011xm; @Farina:2011pw; @Kelso:2011gd; @Foot:2010rj; @HerreroGarcia:2011aa; @Frandsen:2011gi]. This tension has led to suggestions that the CoGeNT and DAMA/LIBRA modulations are due to conventional annual phenomena [@Ralston:2010bd; @Blum:2011jf]. The atmospheric muon rate and the radon level in the underground experimental hall modulate annually. Signals that can simulate dark matter interactions may be produced by ($\alpha$,n) reactions from radon decay in the active volume or by nuclear recoils from spallation neutrons originating from atmospheric muon interactions. The CoGeNT collaboration has stated that contamination from these backgrounds is small compared to the observed signal [@Aalseth:2012if; @Aalseth:2011wp]. The MINOS experiment monitors both of these quantities in an adjacent experimental hall to that of the CoGeNT experiment in the Soudan Underground Laboratory. In this paper we compare the modulations of the CoGeNT event rate data to that of the atmospheric muon rate and radon level data collected at the same time by the MINOS experiment.\
The annual modulation of the muon flux deep underground has been observed by many different experiments [@Ambrosio:1997tc; @Bouchta:1999; @Bellini:2012te; @Solvi:2009; @Desiati:2011; @Adamson:2009zf]. The similarities of the amplitudes and phases of the modulations observed in the LVD muon [@Solvi:2009] and DAMA/LIBRA data sets motivated the hypothesis that modulation in the latter may be muon-induced. It has been suggested that spallation neutrons or long-lived activated isotopes produced by these muons may be responsible for the DAMA/LIBRA modulation [@Blum:2011jf; @Ralston:2010bd]. This now seems unlikely as recent detailed comparisons of the DAMA/LIBRA modulation to that of the muon fluxes measured by LVD [@Solvi:2009], Borexino [@Bellini:2012te] and MACRO [@Ambrosio:1997tc], all in the Gran Sasso National Laboratory (LNGS), have shown that the phases of the two modulations differ significantly [@Bernabei:2012wp; @FernandezMartinez:2012wd; @Chang:2011eb]. This conclusion does not preclude the possibility that the CoGeNT modulation, or a significant fraction thereof, is due to muon related processes.\
The phase of the modulation of the muon flux can vary substantially depending on geographic location and calendar year since the flux is strongly correlated with the effective atmospheric temperature [@Ambrosio:1997tc; @Bouchta:1999; @Bellini:2012te; @Solvi:2009; @Desiati:2011; @Adamson:2009zf]. Therefore, to be able to reject with high confidence the muon hypothesis as the source of the CoGeNT modulation, the muon data must be collected concurrently with the CoGeNT data and in close proximity to the CoGeNT detector. The muon data collected by the MINOS experiment fulfill these criteria. Similarly to the DAMA/LIBRA muon studies [@Bernabei:2012wp], we compare the phase of the observed MINOS muon modulation to that of the CoGeNT data modulation. Comparisons of the CoGeNT data to non-concurrent MINOS muon data [@Adamson:2009zf], and indirectly to effective temperature variations, have been presented in Ref. [@Chang:2011eb] and indicate that the data sets are not correlated.\
We note that the 16.6% amplitude of the CoGeNT event rate modulation [@Aalseth:2011wp] is significantly larger than the $\sim$2% amplitude of the MINOS muon rate modulation [@Adamson:2009zf]. This difference suggests that the muon temporal variation cannot fully account for the observed CoGeNT modulation. In this paper we examine the relative phases of the two modulations which provides an independent test of the potential correlation between the CoGeNT and MINOS muon data sets.\
The radon level in the Soudan Underground Laboratory is at a maximum (minimum) in the summer (winter) months due to the pressure gradients created by the relative temperature differences between the air in the laboratory and that on the surface [@Goodman:1999]. In the MINOS cavern we have observed that the radon concentration varies by a factor of six over the year, corresponding to a modulation amplitude of $\sim$60%. A large modulation amplitude could therefore be introduced into the CoGeNT data by even a small amount of contamination from this background.\
The radon progeny also modulate with a one year period $T$, but do so with a delayed phase and reduced amplitude. The decays between $^{222}$Rn and $^{210}$Pb occur very quickly ($\sim$minutes) and therefore have negligible impact on either the phase or the amplitude. Since $^{210}$Pb has a half-life of , its decay and the decays of its progenies will not contribute to the modulation.\
The following Section of the paper discusses the selected experimental data sets. In Section \[sec:TheDifferentTests\] we present the best fit modulation parameters determined for each of these data sets. We then describe the measurements of the phase differences between the CoGeNT and MINOS muon and radon data sets obtained from a simultaneous fit of the data to phase-shifted sinusoidal functions, a shape-free $\chi^{2}$ data-to-data comparison and a bin by bin correlation test. Section \[sec:Conclusion\] summarizes our conclusions.\
The Selected Data {#sec:TheData}
=================
The CoGeNT dark matter experiment [@Barbeau:2007qi; @Aalseth:2012if] and the Far Detector of the MINOS long baseline neutrino experiment [@Michael:2008bc] are located underground in two different caverns of the Soudan Underground Laboratory. The MINOS cavern, which houses the MINOS detector, is long, wide and high and is oriented along the direction of the NuMI neutrino beam [@Crane:1995ky]. The CoGeNT and CDMS-II dark matter experiments are located in the Soudan 2 cavern which is similar in shape to the MINOS cavern but is long and is oriented north-south. The two experimental caverns are connected by an east-west passage on their north side and are served by a common ventilation system which replaces the lab air several times per hour.\
The CoGeNT Data {#sec:CoGeNTData}
---------------
CoGeNT is an experiment for direct detection of dark matter which employs a p-type point contact germanium detector [@Aalseth:2008rx; @Aalseth:2010vx; @Aalseth:2011wp]. The CoGeNT collaboration has published its results using data collected over a period of 458 days between Dec. 4, 2009 and Mar. 6, 2011 with a total of 442 live days [@Aalseth:2011wp]. The data were presented in fifteen 30-day intervals and one 8-day interval, then fit to a modulation hypothesis of the form: $$R=R_{0}\left(1+A\cdot\textrm{cos}\left [ \frac{2\pi}{T}(t-t_{0}) \right ] \right ),
\label{eq:cosine}$$ where $R_{0}$ is the mean rate, $A$ is the modulation amplitude and $T$ is the period. The time $t$ is the number of days since Jan. 1, 2010. The phase $t_{0}$ is the day at which the signal is at a maximum. The published CoGeNT best fit results are given in the last line of Table \[tab:FitResults\]. The modulation hypothesis is preferred over the null hypothesis at . The CoGeNT collaboration has released the background-subtracted data set used in this analysis to the public. The results of our $\chi^{2}$ fit of the CoGeNT data to Eq. (\[eq:cosine\]), discussed further in Sec. \[sec:CosineTest\], are in good agreement with the published results [@Aalseth:2011wp].\
The MINOS Data
--------------
The MINOS Far Detector has been collecting atmospheric muon data since August 2003 [@Michael:2008bc; @Adamson:2007ww]. The experiment also records the radon level in the laboratory air. The MINOS muon and radon data used in this analysis were collected between June 4, 2009 and Sept. 6, 2011. This collection window is 12 months longer than the CoGeNT run period, from Dec. 4, 2009 to March 6, 2011, allowing the data-to-data comparisons described in Sections \[sec:ShapeFreeTest\] and \[sec:CorrelationTest\].\
The event selection and data quality requirements used in this analysis are identical to those in the previous study of seasonal muon intensity variation at the MINOS Far Detector, with the additional requirement that the reconstructed muon track be downward going. Restricting the data set to contain only days with greater than of live time yields a total of 738 good days of atmospheric muon data. These good days include which occurred between Dec. 4, 2009 and March 6, 2011 inclusive.\
The radon level in the MINOS cavern air, inferred from counting the number of alpha decays, is measured every hour by a Model 1027 Sun Nuclear Corporation radon monitor [@radon-monitor]. A daily measure of the radon level is determined by averaging the 24 measurements taken throughout the day. The standard deviation of these measurements, $\sigma$, is taken to be the error on the daily radon measurement. While larger than the standard error on the mean value, $\sigma/\sqrt{24}$, this choice is more consistent with the published accuracy of the radon monitor [@radon-monitor]. There are 786 good days during which the radon monitor operated continuously throughout the day. These good days include which occurred between Dec. 4, 2009 and March 6, 2011 inclusive. The radon monitor was moved to different locations in the Soudan Underground Laboratory and cross calibrated with other detectors running simultaneously. This demonstrated that the radon level does not vary spatially in the laboratory to within the resolution of the monitor. Thus the radon levels measured in the MINOS cavern can be used to evaluate whether the CoGeNT data are correlated with the radon level in the Soudan cavern.\
![The residuals of the MINOS Far Detector muon rates, radon levels and CoGeNT event rates as a function of time. The MINOS muon and radon data have been scaled by factors of 10 and one-half respectively to fit on the same graph and, for this figure, use the same binning as the CoGeNT results. The vertical dashed lines indicate the start of a new calendar year. The arrow marks the date where a dark matter signal is expected to peak.[]{data-label="fig:AllData"}](Muon-Radon-allCoGeNT_2009-12-04_to_2011-03-06_monthly-noshade-nofit-marker.eps){width="50.00000%"}
The MINOS muon rate and radon level residuals, and the CoGeNT event rate residuals, are plotted as a function of time in Fig. \[fig:AllData\]. The CoGeNT event rate residuals are calculated with respect to a mean rate of . The MINOS muon rate residuals are calculated with respect to a mean rate of . The MINOS radon level residuals are calculated with respect to a mean level . All three data sets possess clear modulation signatures. In the following section we quantify any potential correlations between these modulations.\
Modulation Comparisons {#sec:TheDifferentTests}
======================
If the CoGeNT modulation is caused by either the muon or radon backgrounds then it should modulate with the same shape as those backgrounds. Therefore, if the phase of the CoGeNT modulation is significantly different than that of the MINOS muon or radon data we can infer that they are likely not causally related.\
The most common approach in the literature to evaluating potential correlations, and discussed here in Sec. \[sec:CosineTest\], is to fit the data to Eq. (\[eq:cosine\]) and compare the phases and periods of the best fits. The CoGeNT and DAMA/LIBRA modulations are a good fit to a cosine function. This is the expected signature for an isothermal dark matter halo. The true form of the modulation may be more complex as it is dependent on assumptions made regarding the velocity distribution of the dark matter particles in the halo [@Freese:2012xd; @Chang:2011eb]. The muon modulation is not fit well by a cosine function [@FernandezMartinez:2012wd; @Chang:2011eb]. The muon and radon modulations are correlated with atmospheric temperatures. Therefore, their modulations are cyclical but not necessarily sinusoidal. Imposing such constraints onto the data may bias the results of the cosine based fit comparison. We address this concern in Sections \[sec:ShapeFreeTest\] and \[sec:CorrelationTest\] by performing shape-free data-to-data comparisons that allow us to evaluate the phase differences and potential correlations regardless of the underlying functional forms of the modulations.\
[ccccccc]{} Data & $\chi^{2}$/N.d.o.f. & Mean Rate & Amplitude & Period &Phase & Date of\
& & \[$R_{0}$\] & \[$A$,%\] & \[$T$,days\] & \[$t_{0}$,days\]& Maximum\
\
Muon &1909 / (449-3) & & 1.25 $\pm$ 0.03 & 365.25 &182.8 $\pm$ 1.7& July 1\
Radon &176 / (458-3) & & 57.7 $\pm$ 0.9 & 365.25 &215.0 $\pm$ 1.1& Aug. 3\
CoGeNT (Our Fit) &6.6 / (16-3) & & 16.9 $\pm$ 5.4 & 365.25 &108.4 $\pm$ 16.9& Apr. 18\
\
Muon &1788 / (449-4) & & 1.37 $\pm$ 0.04 &317.2 $\pm$ 3.2 &187.3 $\pm$ 1.4& July 6\
Radon &176 / (458-4) & & 57.7 $\pm$ 0.9 &367.4 $\pm$ 3.5 &215.2 $\pm$ 1.1& Aug. 3\
CoGeNT (Our Fit) &6.4 / (16-4) & & 16.7 $\pm$ 5.4 &348 $\pm$ 42 &113.7 $\pm$ 17.9& Apr. 23\
\
CoGeNT &7.8 / (16-4) & N/A & 16.6 $\pm$ 3.8 &347 $\pm$ 29 &$\sim$115 $\pm$ 12 & Apr. 25\
Cosine $\chi^{2}$ Test {#sec:CosineTest}
----------------------
The nominal modulation parameters for the CoGeNT and MINOS muon and radon data sets were determined by performing a $\chi^{2}$ fit test of Eq. (\[eq:cosine\]) to the data described in Sec. \[sec:TheData\] and shown in Fig. \[fig:AllData\]. The results of these fits are given in Table \[tab:FitResults\]. The confidence limit contours for the best fit phase and period are shown in Fig. \[fig:PhasePeriodFits\].\
![Confidence limit contours for the period and phase as determined by fitting the CoGeNT event rate and MINOS Far Detector muon rate and radon level data to Eq. (\[eq:cosine\]). The best fit values are given in Table \[tab:FitResults\]. The horizontal and vertical black lines mark the expected period and phase for a dark matter signal. []{data-label="fig:PhasePeriodFits"}](ThreeConfidenceEllipses-floating.eps){width="50.00000%"}
Our fit to the CoGeNT data is in good agreement with the published results [@Aalseth:2011wp] and disfavors the null modulation hypothesis at . The significance with which we exclude the null modulation hypothesis is defined as the square root of the difference between the $\chi^{2}$ value of the best fit point and that of the null modulation hypothesis. This definition is different from that used in the published CoGeNT analysis and gives a slightly stronger exclusion. The small differences between our best fit values to the CoGeNT data and the published CoGeNT best fit values may be explained by the assumption in our fits that the CoGeNT errors are uncorrelated.\
The two apparent occurrences of sudden stratospheric warming events [@Osprey:2009ni] in early 2010 and early 2011, which temporarily increased the muon rate, drive the large $\chi^{2}$ for the muon fit and cause the best fit period to be significantly smaller than one year. If the complete MINOS muon data set, August 2003 to April 2012, is fit, minimizing the impact of short term fluctuations, a period much closer to one year is obtained, $T$=, and the phase remains unchanged.\
The best fit phase differences $\delta t_{0}$ between the CoGeNT phase and the MINOS muon and radon phases are determined by minimizing:
$$\begin{aligned}
\label{eq:LongFit}
\chi^{2}(\delta t_{0})&=&\sum_{i=1}^{N_{M}} \frac{(R_{ob,M,i}-R_{ex}(R_{0,M},A_{M},t_{0},T))^{2}}{\sigma_{M,i}^{2}}\\
&+&\sum_{i=1}^{N_{C}=16} \frac{(R_{ob,C,i}-R_{ex}(R_{0,C},A_{C},t_{0}+\delta t_{0},T))^{2}}{\sigma_{C,i}^{2}}.\nonumber\end{aligned}$$
The first term in Eq. (\[eq:LongFit\]) is the $\chi^{2}$ contribution from the MINOS muon rate or radon level data where $N_{M}$ is the number of live days concurrent with the CoGeNT data collection period. The second term is the contribution from the CoGeNT event rate data. $R_{ob,M,i}$ ($R_{ob,C,i}$) is the $i^{th}$ observed MINOS (CoGeNT) data point. $\sigma_{M,i}$ and $\sigma_{C,j}$ are the uncertainties on the MINOS and CoGeNT data points respectively. $R_{ex}$ is the expected value, as determined by Eq. (\[eq:cosine\]), assuming the given modulation parameters and $\delta t_{0}$ is defined as the phase of the CoGeNT data minus the phase of the MINOS data. The $\chi^{2}$, as a function of this phase difference, is determined by minimizing the $\chi^{2}$ over the MINOS mean value $R_{0,M}$, the amplitude $A_{M}$ and phase $t_{0}$ and the CoGeNT mean value $R_{0,C}$, amplitude $A_{C}$; and, for some fits, a common period $T$.\
Figure \[fig:DeltaPhaseFits\] shows the $\Delta\chi^{2}$ curves, as a function of $\delta t_{0}$, for the simultaneous fits of the MINOS and CoGeNT data to Eq. (\[eq:LongFit\]) assuming a common period of one year. The best fit phase differences are and for the comparison to the muon and radon data respectively and and respectively when minimizing the $\chi^{2}$ over the period $T$. The statistical significance at which equivalent phases for the MINOS and CoGeNT data can be excluded is given by the square root of the $\Delta\chi^{2}$ difference between the best fit point and the value at $\delta t_{0}=0$. As can be seen from Fig. \[fig:DeltaPhaseFits\] the phases of the MINOS muon and radon data are inconsistent with the phase of the CoGeNT data at and respectively.\
![The $\Delta\chi^{2}$ distributions comparing the phases of the MINOS muon rate and radon level data to the phase of the CoGeNT event rate data using Eq. (\[eq:LongFit\]). The $\Delta\chi^{2}$ curves are calculated with respect to their $\chi^{2}$ minima. The flattening of the $\Delta\chi^{2}$ curves indicate that these exclusions are limited by the confidence with which the CoGeNT data can exclude the null modulation hypothesis. []{data-label="fig:DeltaPhaseFits"}](ChiSquaredFitBothLine-limit_fixedT-nolabel.eps){width="50.00000%"}
Shape-Free $\chi^{2}$ Test {#sec:ShapeFreeTest}
--------------------------
In this section we determine the relative phase $\delta t_{0}$ between the MINOS and CoGeNT data sets, without an [*a priori*]{} assumption regarding their shape, by calculating the $\chi^{2}$ difference between their respective modulations. The $\chi^{2}$ difference, assuming a common binning, is defined as: $$\chi^{2}(\delta t_{0})= \sum_{i=1}^{N_{C}=16}\frac{(R_{C,i}-f\cdot R_{M,i}(\delta t_{0}))}{\sigma^{2}_{C,i}+\sigma^{2}_{M,i}}.
\label{eq:ShortFit}$$ $R_{M,i}$ ($R_{C,i}$) is the $i^{th}$ MINOS (CoGeNT) residual and $\sigma_{M,i}$ and $\sigma_{C,i}$ are the uncertainties on the MINOS and CoGeNT residuals respectively. We marginalize over the difference in amplitudes, for each $\delta t_{0}$, by minimizing the $\chi^{2}$ over a positive definite multiplicative factor $f$. If the data have similar underlying forms, we expect the $\chi^{2}$ to be a minimum when the phase difference between them is zero. The $\chi^{2}$ values, as a function of $\delta t$, are determined by shifting the time-axis of the MINOS data by $\delta t$ days and recalculating Eq. (\[eq:ShortFit\]). Figure \[fig:ShortFitPlots\] shows the $\Delta\chi^{2}$ curves as a function of the MINOS data offset, which is equivalent to the relative phase $\delta t_{0}$. The curves are not smooth due to statistical fluctuations in the data. By offsetting the MINOS data we vary the number of MINOS live days which overlap the CoGeNT data. To ensure that each subset of MINOS data, for every $\delta t$, contains the same number of live days we substitute the historical daily average of that date for those days which do not pass the live-time selection criteria. The best fit phase differences between the CoGeNT data and the MINOS muon (radon) data, corresponding to the minimum of the $\Delta\chi^{2}$ curves in Fig. \[fig:ShortFitPlots\], are (). The statistical significance, as defined in Sec. \[sec:CosineTest\], at which equivalent phases for the CoGeNT and MINOS muon (radon) data are excluded is ().\
![The $\Delta\chi^{2}$ distributions comparing the phases of the MINOS muon rate and radon level data to the phase of the CoGeNT event rate data using Eq. (\[eq:ShortFit\]). The $\Delta\chi^{2}$ curves are calculated with respect to their $\chi^{2}$ minima. []{data-label="fig:ShortFitPlots"}](ChiSquaredFitHypFree-Both.eps){width="50.00000%"}
Correlation Test {#sec:CorrelationTest}
----------------
Residual muon or radon backgrounds in the CoGeNT data could cause a correlation between the CoGeNT modulation and the MINOS muon and radon modulation measurements. The degree of correlation has been evaluated using Pearson’s coefficient of correlation, calculated as:
$$\rho=\frac{1}{N_{C}-1}\sum_{i=1}^{N_{C}=16}\frac{(R_{ob,M,i}-\overline{R_{ob,M}})(R_{ob,C,i}-\overline{R_{ob,C}})}{\sigma_{M}\sigma_{C}},
\label{eq:correlation}$$
where $N_{C}$ is the number of bins, $\overline{R_{ob,M}}$ and $\overline{R_{ob,C}}$ are the average values of the MINOS and CoGeNT data sets. $\sigma_{M}$ and $\sigma_{C}$ are the standard deviations of the points comprising the MINOS and CoGeNT data sets respectively. The correlation coefficients, and their Fisher transforms [@Fisher:1936et], are given in Table \[tab:Correlations\].\
---------------------- ---------------------- ------------------
Data Set Correlation Fisher
Coefficient ($\rho$) Transform
CoGeNT vs Muon Data 0.19 0.19 $\pm$ 0.28
CoGeNT vs Radon Data -0.29 -0.30 $\pm $0.28
---------------------- ---------------------- ------------------
: The coefficients of correlation, and their Fisher transforms, calculated between the CoGeNT event rate data and the MINOS muon rate and radon level data. Both data sets being compared are consistent with no correlation at $\sim$1$\sigma$. []{data-label="tab:Correlations"}
Even if there is no causal relationship between the observed MINOS muon and radon modulations and the CoGeNT modulation, there will be some correlation between these data sets as they all follow an approximate sinusoidal variation. The expected value of the correlation is related to their relative phases. For example, if the phase difference between two periodic data sets is smaller (larger) than one-quarter of the period, the correlation should be positive (negative). One can therefore infer from the results in Table \[tab:Correlations\] that the effective phase difference between CoGeNT and the MINOS muon data is near to but less than , while between CoGeNT and the MINOS radon data it is near to but more than .\
To verify whether the calculated correlations are consistent with the observed modulation phases we generated a series of pseudo-experiments. Sampling from two cosine curves, with the precision and binning of the CoGeNT and MINOS data sets and amplitudes taken from Table \[tab:FitResults\], we calculated the Fisher transform as a function of the phase difference between the two curves. We find that the observed values of the Fisher transforms in Table \[tab:Correlations\], (0.19 $\pm$ 0.28) and (-0.30 $\pm$ 0.28), correspond to phase differences of and respectively. These values are consistent with the phase differences calculated in the preceding sections.\
Conclusion {#sec:Conclusion}
==========
We have performed a comparison of the modulation phases observed in the CoGeNT and MINOS atmospheric muon and radon data, all collected concurrently between Dec. 4, 2009 and March 6, 2011 in the Soudan Underground Laboratory. We have presented the results of a shape-free data-to-data comparison which indicate that the phases of the CoGeNT data and the atmospheric muon and radon data are different by () and () respectively. The calculated correlation coefficients between the CoGeNT and MINOS data sets are statistically consistent with the no-correlation hypothesis. The cosine fit test measures the phase difference between the CoGeNT and MINOS muon data sets to be , inconsistent at , and between the CoGeNT and MINOS radon data sets to be , inconsistent at . The similarity between the results of both these tests indicate that no significant bias is introduced when imposing a sinusoidal shape on the data. It is also clear that our exclusions are limited by the degree to which the CoGeNT data exclude the null modulation hypothesis. Based on the studies described above, it appears unlikely that muon or radon related processes contribute significantly to the observed CoGeNT modulation.\
Acknowledgments {#sec:Acknowledgements}
===============
This work was supported by the US DOE, the UK STFC, the US NSF, the State and University of Minnesota, the University of Athens, Greece and Brazil’s FAPESP and CNPq. We are grateful to the Minnesota Department of Natural Resources, the crew of Soudan Underground Laboratory, and the staff of Fermilab for their contributions to this effort. We also thank Juan Collar and the CoGeNT collaboration for sharing their data thus facilitating this analysis.
| {
"pile_set_name": "ArXiv"
} |
[**Schwarzschild-de Sitter black hole\
from entropic viewpoint**]{}
Ee Chang-Young${}^{a}$[^1], Myungseok Eune${}^{b}$[^2], Kyoungtae Kimm${}^{a}$[^3], and Daeho Lee${}^{a}$[^4]\
\
[**Abstract**]{}
In a Schwarzschild-de Sitter space, we consider an equipotential surface which consists of two holographic screens. Adapting the Bousso-Hawking’s reference point of vanishing force, we divide the space into two regions, which are from the reference point to each holographic screen. These two regions can be treated as independent thermodynamical systems, because the Bousso-Hawking reference point with zero temperature behaves like a thermally insulating wall. The entropy obtained in this way agrees with the conventional results; i) when the holographic screens lie at the black hole and cosmological horizons, ii) in the Nariai limit.\
\[5mm\] Keywords : Schwarzschild-de Sitter space, Unruh-Verlinde temperature, entropic force
6.6mm
Introduction
============
Recently, much attention has been focused on the new idea suggested by Verlinde [@verlinde] in which gravity can be explained as an emergent phenomenon originated from the statistical properties of unknown microstructure of spacetime. The essential part of this idea is based on two key ingredients: holographic principle and equipartition rule of the energy. With help of these principles, the Newton’s law of gravity was derived by interpreting it as an entropic force i.e., force on a test particle at some point was defined as the product of the entropy gradient and the temperature at that point, and relativistic generalization leads to the Einstein equations. This entropic formulation of gravity has been used to study thermodynamics at the apparent horizon of the Friedmann-Robertson-Walker universe [@Shu:2010nv], Friedmann equations [@Cai:2010hk; @Pad:10013380], entropic correction to Newtonian gravity [@Smolin:2010kk; @PNicolini; @LModesto], holographic dark energy [@Li:2010cj; @efs; @danielson; @YFCai]. There have been many works for the entropic force in cosmological models [@Gao:2010fw; @zgz; @wang:y; @Wei:2010ww] and the black hole backgrounds [@Myung:2010jv; @mk; @Liu:2010na; @Cai:2010sz; @Tian:2010uy; @Kuang:2010gs; @EEKL; @RAKonoplya; @FCaravelli].
In a spacetime admitting timelike killing vector one can define a gravitational potential, and the holographic screen is given by equipotential surface. In general, the holographic screen can have multiple disconnected parts depending upon the matter distribution. The temperature on the holographic screen is given by Unruh-Verlinde temperature associated with the proper acceleration of a particle near the screen. This prescription works well for spacetime with a single holographic screen, however, there has been no known work for multiple holographic screens so far.
On the other hand, the observational evidence for late-time cosmological acceleration [@Perl:1998; @Reiss:1998] gave much impetus on studying the de Sitter space with black holes. Since Schwarzschild-de Sitter black hole is asymptotically de Sitter space, it has cosmological event horizon in addition to black hole horizon and these horizons can form holographic screens. In fact, the potential of Schwarzschild-de Sitter space has two equipotential surfaces for a given potential value, and the two horizons correspond to equipotential surfaces. In this paper we investigate the entropic formulation in the background geometry of the Schwarzschild-de Sitter space, which provides a model for multiple holographic screens.
In the Verlinde’s formalism, two equipotential holographic screens in the Schwarzschild-de Sitter space have different temperatures. Thus the whole system cannot be treated as a thermodynamical system in equilibrium. In Ref. [@Bousso:1996au], Bousso and Hawking set up a reference point in the radial direction, at which force vanishes. They have pointed out that this reference point can play a role of a point at infinity in an asymptotically flat space. Besides, the temperature at this reference point is zero, and thus no thermal exchange can occur across this point. This makes the reference point behave like a thermally insulating wall. Therefore, we can regard the Schwarzschild-de Sitter space as two thermally independent systems: the inner system in the black hole side and the outer system in the cosmological horizon side. Gibbons and Hawking also considered similar construction in a slightly different context [@GH_desitter]: they constructed two separated thermal equilibrium systems by introducing a perfectly reflecting wall in the Schwarzschild-de Sitter space for the calculation of the Hawking temperatures of black hole and cosmological horizons.
Based on the above consideration, we apply the Verlinde’s formalism to each system. In the Schwarzschild-de Sitter case we choose the holographic screen of equipotential surface having spherical symmetry. With this choice of holographic screen we show that the thermodynamic relationship $E = 2TS$ holds for each holographic screen, where $E$, $T$, and $S$ are the quasilocal energy given by Komar mass, temperature, and entropy, respectively. We then check this result with the known cases: i) when the holographic screens lie at the black hole and cosmological horizons, ii) in the Nariai limit.
In the following section, we briefly review the Verlinde’s formalism of entropic approach to gravitational interaction. In section \[sec:sch.dS\], we apply the Verlinde’s formalism to a Schwarzschild-de Sitter space which provides a prototype of multiple holographic screens. Finally, we summarize our results. In this paper, we adopt the convention $c=k_B=\hbar =1$.
Verlinde’s entropic formalism {#sec:setup}
=============================
According to the Verlinde’s formalism [@verlinde], gravity is an entropic force emerging from coarse graining process of information for a given energy distribution. In this process, information is stored on holographic screens. In the nonrelativistic case, the holographic screens correspond to Newtonian equipotential surfaces and the holographic direction is given by the gradient of the potential.
In a curved spacetime with a timelike Killing vector $\xi^\mu$, the generalized Newton’s potential is given by $$\label{potential}
\phi = \frac12 \ln (-\xi^\mu \xi_\mu ).$$ This potential can be used to define a foliation of space. For a particle with a four velocity $u^\mu$, its proper acceleration is given by $a^\mu = u^\nu \nabla_\nu u^\mu$. In terms of the potential $\phi$ and the Killing vector $\xi^\mu$, the velocity and the acceleration can be written as $$\begin{aligned}
u^\mu &= e^{-\phi} \xi^\mu, \label{u:killing} \\
a^\mu &= - \nabla^\mu \phi, \label{a:killing}\end{aligned}$$ where the Killing equation has been used to derive Eq. (\[a:killing\]). In Eq. (\[a:killing\]), the acceleration is normal to holographic screen. The Unruh-Verlinde temperature on the screen is defined as $$\label{T:def}
T = \frac{1}{2\pi} e^\phi n^\mu \nabla_\mu \phi,$$ where $n^\mu$ is the unit outward pointing vector normal to the screen and to the Killing vector. The “outward” indicates that the potential increases along $n^\mu$, *i.e.,* the normal vector can be written as $$\label{def:n}
n_\mu = \frac{\nabla_\mu \phi}{\sqrt{\nabla_\nu \phi \nabla^\nu \phi}}.$$ In Eq. (\[T:def\]), a redshift factor $e^\phi$ is inserted because the temperature is measured with respect to the reference point. For asymptotically flat space this reference point corresponds to spatial infinity. In the Schwarzschild-de Sitter case, we choose this reference point as the Bousso-Hawking reference point [@Bousso:1996au] to be explained in the next section.
We denote the number of bits on the holographic screen $\mathcal{S}$ by $N$ which is assumed to be proportional to the area of the screen [@verlinde], $$\label{N:assume}
N = \frac{A}{G}.$$ Applying the equipartition rule of the energy, each bit of holographic screen contributes an energy $T/2$ to the system, and the total energy on the holographic screen can be written as $$\label{E:equipartition}
E = \frac12 \oint_{\mathcal{S}} T dN.$$ Note that in the above expression the temperature $T$ on the screen is not constant in general. Substituting Eqs. (\[T:def\]) and (\[N:assume\]) into Eq. (\[E:equipartition\]), the energy associated with the holographic screen can be rewritten as $$\label{E:pot.area}
E = \frac{1}{4\pi G} \oint_{\mathcal{S}} n^\mu \nabla_\mu e^\phi dA.$$ This expression is the conserved Komar mass associated with timelike Killing vector $\xi^\mu$.
Schwarzschild-de Sitter black hole {#sec:sch.dS}
==================================
Now, we consider a spherically symmetric Schwarzschild-de Sitter black hole as a model of multiple holographic screens. The Schwarzschild-de Sitter space is described locally by the line element, $$\label{metric:ds}
ds^2 =
-f(r) dt^2 + \frac{dr^2}{f(r)} + r^2 (d\theta^2 + \sin^2
\theta \, d\varphi^2),$$ with $$\label{f}
f(r) = 1 - \frac{2GM}{r} - \frac13\Lambda r^2,$$ where $G$ and $M$ are the gravitational Newton’s constant and the mass parameter, respectively, and the cosmological constant will be taken as $\Lambda = 3/\ell^2$. When $0 < M < M_{\mathrm{max}}\equiv
\ell/(3^{2/3} G)$ static region exists between two horizons with radii $r_b$ and $r_c$, the black hole and cosmological event horizons. For $M = M_{\mathrm{max}}$, the two horizons coincide, the Nariai limit. In the Nariai limit, there exists no timelike Killing vector.
![The Schwarzschild-de Sitter space has the event horizon of the black hole at $r=r_b$ and the cosmological event horizon at $r=r_c$. At $r=r_g$, the proper acceleration vanishes. For a given potential value, there exist two screens at $r=r_1$ and $r=r_2$, and each screen has different temperature. Note that the unit normal vectors on both screens direct to the surface $r=r_g$.[]{data-label="fig:screen"}](frEP){width="50.00000%"}
In order to get the potential of the Schwarzschild-de Sitter spacetime, we first consider a timelike Killing vector of Eq. (\[metric:ds\]), given by
$$\label{xi}
\xi^\mu = \gamma\left( \partial/\partial t \right)^\mu,$$
where $\gamma$ is a normalization constant. If space is asymptotically flat, we may choose the standard Killing vector normalization, $ \gamma = 1$. Since Schwarzschild-de Sitter space is not asymptotically flat, we encounter a difficulty in taking the normalization of Killing vector. To avoid this, Bousso and Hawing [@Bousso:1996au] chose a normalization such that the norm of the Killing vector becomes unity at the region where the force vanishes, the gravitational attraction is exactly balanced out by the cosmological repulsion. Adopting this normalization corresponds to choosing a special observer who follows geodesics.
Since the magnitude of the acceleration of a particle in the Schwarzschild-de Sitter spacetime is obtained as $a = \sqrt{a^\mu
a_\mu} = |f'(r)|/\sqrt{2f(r)}$, the geodesic point with no acceleration is given by $$\label{def:rg}
r_g = (GM\ell^2)^{1/3}.$$ With this normalization, the gravitational potential is obtained from Eq. (\[potential\]), $$\label{potential:f}
\phi = \frac12 \ln (\gamma^2 f)
= \frac12 \ln \frac{f(r)}{f(r_g)}.$$ For a given potential value $\phi_s$, there exist two equipotential surfaces at $r=r_1$ and $r=r_2$ as shown in Fig. \[fig:screen\]. Then, the Unruh-Verlinde temperature on each screen is given by $$\label{T}
T = \frac{1}{2\pi} e^\phi n^\mu \nabla_\mu \phi =
\gamma \frac{|f'(r)|}{4\pi},$$ where the unit normal vector $n^\mu$ is given by $n^\mu = \delta^\mu_r
\sqrt{f}$ for $r<r_g$ and $n^\mu = -\delta^\mu_r \sqrt{f} $ for $r>r_g$. Note that the temperature of the holographic screen at $r=r_1$ is different from that of the screen at $r=r_2$. The temperature on each screen is given by $$\label{T:r}
T_i = \frac{\gamma}{2\pi} \left| \frac{GM}{r_i^2} - \frac{r_i}{\ell^2} \right|,$$ where $i = 1, 2$.
![ We consider the force free reference point of Bousso and Hawking as a separating boundary dividing the system into two subsystems. Since the temperature of each subsystem is above zero and the boundary between them is maintained at zero, thermal exchange does not occur between the two subsystems.[]{data-label="fig:tempq"}](spherical_shell){width="50.00000%"}
The temperature becomes zero at the Bousso-Hawking reference point $r=r_g$ from Eq. (\[T\]). Now, assume that the region between the black hole and cosmological horizons is separated by a boundary at the reference point $r=r_g$ as shown in Fig. \[fig:tempq\]. Then the two regions divided by this boundary cannot have thermal exchange between them because the temperature on this boundary is kept at zero always in our static geometry setup. Thus, we can regard this boundary as a thermally insulating wall. Therefore, the two regions separated by the surface at $r=r_g$ can be thought as independent systems: the total system becomes the sum of two independent systems, the inner ($r<r_g$) and outer ($r>r_g$) regions. The concept of thermally insulating wall in our consideration is similar to that of perfectly reflecting wall in the Gibbons-Hawking’s work [@GH_desitter]: they constructed two separated thermal equilibrium systems by introducing a perfectly reflecting wall in Schwarzschild-de Sitter space for the calculation of the Hawking temperatures of black hole and cosmological horizons.
This can be also understood as follows. The line element (\[metric:ds\]) approaches the pure de Sitter spacetime when $M$ goes to zero and the pattern of the metric for $r>r_g$ has a similarity to that of the pure de Sitter spacetime (see Fig. \[fig:mass\]). And the spacetime approaches the Schwarzschild black hole with asymptotically flat spacetime when $\Lambda$ goes to zero and the pattern of the metric for $r<r_g$ has the similarity to that of the Schwarzschild black hole (see Fig. \[fig:cosmo\]). This suggests that the whole system has the characteristics of both Schwarzschild black hole and pure de Sitter spacetime.
![The geodesic point with no acceleration is plotted by the dashed line. The metric approaches the pure dS spacetime as the mass parameter of Schwarzschild-de Sitter space goes to zero. ($0<M_1 < M_2 < M_3 <M_{\mathrm{max}}$)[]{data-label="fig:mass"}](frM){width="50.00000%"}
![The geodesic point with no acceleration is plotted by the dashed line. The metric approaches the Schwarzschild black hole as the cosmological constant $\Lambda = 3/\ell^2$ goes to zero. ($\ell_{\mathrm{min}} < \ell_1 < \ell_2 < \ell_3 < \infty$)[]{data-label="fig:cosmo"}](frl){width="50.00000%"}
Plugging the potential (\[potential:f\]) into the energy (\[E:pot.area\]) gives the same result from the Komar energy for the Schwarzschild-de Sitter black hole, $$\label{E:g}
E = \frac{1}{4\pi G} \oint_{\mathcal{S}} \nabla^\mu \xi^\nu
\sigma_\mu n_\nu dA,$$ where $\sigma_\mu$ is the unit normal timelike vector perpendicular to the hypersurface surrounded by the screen $\mathcal{S}$. Since $\sigma_\mu = - \sqrt{f} \, \delta_\mu^t$, the Komar energy (\[E:g\]) becomes $$E_i = \gamma \frac{r_i^2 |f'(r_i)|}{2G} = \gamma \left| M -
\frac{r_i^3}{G\ell^2} \right|, \label{E:r}$$ for each screen at $r=r_i$ ($i = 1, 2$).
If the associated holographic entropy is given by $$S_i = \frac{A_i}{4G} = \frac{\pi r_i^2}{G}, \label{S:r}$$ then with Eqs. (\[T:r\]) and (\[E:r\]) the thermodynamic relation $E_i = 2 T_i
S_i$ holds for each system. This relation certainly holds for event horizons.[^5] When the spacetime is static and spherically symmetric, we can also get this relation directly from Eq. (\[E:equipartition\]) with the relation , since the temperature on the holographic screen is constant. Note that the thermodynamic relation $E=2TS$ does not hold for the whole system, since the energy and entropy are additive and the temperatures on the holographic screens are different.
Now, we check the validity of our formulation in two specific cases. First, we consider the case when the holographic inner and outer screens become the event horizon of black hole and the cosmological horizon, respectively. As the locations of the holographic screens, $r_1$ and $r_2$, move to the two roots of $f(r) =
0$, $r_b$ and $r_c$, as shown in Fig. \[fig:screen\], the inner screen becomes the black hole event horizon and the outer one becomes the cosmological event horizon. The temperatures on the screens seen by an observer located at the Bousso-Hawking reference point are given by $$\begin{aligned}
T_{b/c}= \frac{1}{\sqrt{1-(9G^2M^2\Lambda)^{1/3}}}
\frac{1}{4\pi}\left| \frac{2GM}{r_{b/c}^2} - \frac{2\Lambda r_{b/c}}{3}\right|.\end{aligned}$$
Since the system is composed of the sum of two independent systems, the total entropy is given by the sum of the entropies of subsystems, $$\label{S:total}
S = S_1 + S_2.$$ In the present case, $S_1$ and $S_2$ correspond to the usual entropy of the black hole and cosmological horizons, respectively. And thus, our result agrees with the previously obtained entropy of Schwarzschild-de Sitter space [@GH_desitter; @Kas:1996; @SdS_entropy; @BHlee:jkps; @SdS_nar_ent].
Next, we consider the case when the two event horizons, $r_b$ and $r_c$, approach each other, the Nariai limit [@nariai_bh]. In this case, the temperature and the energy on each horizon become $$\begin{aligned}
T_i &\longrightarrow T^{\rm Nariai} = \frac{\sqrt{3}}{2\pi \ell}, \label{T:Nariai}\\
E_i &\longrightarrow E^{\rm Nariai} = \sqrt{3} \left(\frac{M^2 \ell}{G}
\right)^{1/3}. \label{E:Nariai}\end{aligned}$$ In this limit, the entropy of each system becomes $$S_i \longrightarrow\ \frac{\pi r_g^2}{G}.\label{S:i:N}$$ The total entropy is the sum of the two subsystems’, thus it is twice of the above given entropy (\[S:i:N\]). This agrees with the entropy of the Schwarzschild-de Sitter black hole in the Nariai limit obtained in Refs. [@BHlee:jkps; @SdS_nar_ent].
In summary, we apply the Verlinde’s entropic formalism of gravity to the Schwarzschild-de Sitter space as a model of multiple holographic screens. Since the Unruh-Verlinde temperature vanishes at the Bousso-Hawking reference point, we can regard two regions separated by zero temperature barrier as thermodynamically isolated systems and thus independently apply the entropic formalism to each region. We confirm that the Verlinde’s formalism agrees with the conventional result at least in the following cases; i) when the holographic screens become event horizons, ii) in the Nariai limit.
Acknowledgments {#acknowledgments .unnumbered}
===============
This work was supported by the National Research Foundation (NRF) of Korea grants funded by the Korean government (MEST) \[R01-2008-000-21026-0 and NRF-2009-0075129 (E. C.-Y. and K. K.), NRF-2009-351-C00109 (M. E.), and NRF-2009-351-C00111 (D. L.)\].
[99]{}
E. P. Verlinde, arXiv:1001.0785 \[hep-th\]. F. W. Shu and Y. Gong, arXiv:1001.3237 \[gr-qc\]. R. G. Cai, L. M. Cao and N. Ohta, Phys. Rev. D [**81**]{} (2010) 061501. T. Padmanabhan, arXiv:1001.3380v1 \[gr-qc\];
L. Smolin, arXiv:1001.3668 \[gr-qc\]. P. Nicolini, Phys. Rev. D [**82**]{} (2010) 044030. L. Modesto and A. Randono, arXiv:1003.1998 \[hep-th\].
M. Li and Y. Wang, Phys. Lett. B [**687**]{} (2010) 243. D. A. Easson, P. H. Frampton and G. F. Smoot, arXiv:1002.4278 \[hep-th\]. U. H. Danielsson, arXiv:1003.0668 \[hep-th\]. Y.-F. Cai, J. Liu, H. Li, Phys. Lett. [B]{} [**690**]{} (2010) 213. C. Gao, Phys. Rev. [D]{} [**81**]{} (2010) 087306. Y. Zhang, Y. G. Gong and Z. H. Zhu, arXiv:1001.4677 \[hep-th\]. Y. Wang, arXiv:1001.4786 \[hep-th\]. S. W. Wei, Y. X. Liu and Y. Q. Wang, arXiv:1001.5238 \[hep-th\]. Y. S. Myung, arXiv:1002.0871 \[hep-th\]. Y. S. Myung and Y. W. Kim, Phys. Rev. [D]{} [**81**]{} (2010) 105012. Y. X. Liu, Y. Q. Wang and S. W. Wei, Class. Quantum Grav. [**27**]{} (2010) 185002. R. G. Cai, L. M. Cao and N. Ohta, Phys. Rev. D [**81**]{} (2010) 084012. Y. Tian and X. Wu, Phys. Rev. D [**81**]{} (2010) 104013. X. Kuang, Y. Ling and H. Zhang, arXiv:1003.0195 \[gr-qc\]. Ee C.-Y., M. Eune, K. Kimm, and D. Lee, Mod. Phys. Lett. A [**25**]{} (2010) 2825. R. A. Konoplya, Eur. Phys. J. C [**69**]{} (2010) 555. F. Caravelli and L. Modesto, arXiv:1001.4364 \[gr-qc\].
S. Perlmutter et al., Astrophys. J. [**517**]{} (1998) 565. A. G. Reiss et al., Astron. J. [**116**]{} (1998) 1009. R. Bousso and S. W. Hawking, Phys. Rev. D [**54**]{} (1996) 6312. G. W. Gibbons and S. W. Hawking, Phys. Rev. D [**15**]{} (1977) 2738.
T. Padmanabhan, Mod. Phys. Lett. [A]{} [**25**]{} (2010) 1129.
R. Banerjee and B. R. Majhi, Phys. Rev. D [**81**]{} (2010) 124006.
D. Kastor and J. Traschen, Class. Quant. Grav. [**13**]{} (1996) 2753.
R. G. Cai, J.-Y. Ji, and K.-S. Soh, Class. Quant. Grav. [**15**]{} (1998) 2783.
K.-S. Cha, B.-H. Lee, C. Park, J. Korean Phys. Soc. [**42**]{} (2003) 735.
R. Garattini, Entropy [**2**]{} (2000) 26. R. Bousso, arXiv:hep-th/0205177.
[^1]: Email: [email protected]
[^2]: Email: [email protected]
[^3]: Email: [email protected]
[^4]: Email: [email protected]
[^5]: In Refs. [@Pad:0912; @RBan], it was shown that this relation holds when the equipartition rule of energy is assumed for event horizons of stationary spacetimes.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We study gravitational collapse of the general spherically symmetric null strange quark fluid having the equation of state, $p = (\rho - 4B)/n$, where $B$ is the bag constant. An interesting feature that emerges is that the initial data set giving rise to naked singularity in the Vaidya collapse of null fluid gets covered due to the presence of strange quark matter component. Its implication to the Cosmic Censorship Conjecture is discussed.'
author:
- 'S. G. Ghosh'
- Naresh Dadhich
title: Gravitational collapse of null strange quark fluid and cosmic censorship
---
[^1]
[^2]
Introduction
============
A gravitating object when it undergoes indefinite collapse, the end product is singularity which is marked by divergence of physical parameters like energy density. As the singularity is approached, density diverges and it would therefore be of relevance to consider the state of matter at ultra high density beyond the nuclear matter. One of such possible states could be the strange quark matter which consists of $u$, $d$ and $s$ quarks. It is energetically most favored state of baryon matter. It could either be produced in quark-hadron phase transition in the early Universe or at ultra high energy neutron stars converting into strange (matter) stars [@ew]. In the context of gravitational collapse, which is our concern here, it is the latter process which would be pertinent.
The key question in the collapse process is whether the singularity so formed will be visible or will be covered by an event horizon prohibiting its visibility to an external observer. That the latter is the case goes by the name of Cosmic Censorship Conjecture (CCC) (see [@r1] for reviews of the CCC)), which remains one of the most important unresolved issues in classical general relativity. In the strong form of CCC, there could emerge no null rays from the singularity in a reasonable space-time, and hence it is invisible for all observers. That is, there occurs no naked singularity for any observer. On the other hand the weak form states that null rays can emerge from the singularity which is however covered by an event horizon and hence they cannot reach out the external observer. In the weak form singularity is locally naked, say for an observer sitting on the collapsing star, but it is globally not because it is safely hidden behind an event horizon.
There do however exist cases of regular initial data sets giving rise to possibility of null rays emanating from the singularity and reaching out to external observer [@pj]. That is, both the possibilities of singularity being naked and hidden inside a black hole can occur. The critical question is what decides between these two possibilities? In this context, it has recently been argued that it is the shearing effect in the collapsing inhomogeneous dust cloud that is responsible for the ultimate outcome [@pnr]. This happens because shear produces distortion in the collapsing fluid congruence which could cause distortion in the geometry of the apparent horizon surface. Such a distortion of the apparent horizon could let null rays emanating from the singularity to escape to external observer. It turns out that if shear close to the center exceeds a threshold limit, it gives rise to a naked singularity and else a black hole.
Another critical question is, what is the state of matter as the singularity is approached? It is certainly a case of diverging density, and hence it would be appropriate to consider near the singularity matter in the highest known density form. That brings in the strange quark matter (SQM). Recently, collapse of charged strange quark fluid together with the Vaidya null radiation has been studied [@hc]. In this paper, we would further like to analyze this process from the point of view of formation of naked singularity and its strength, and more importantly to bring out the effect of SQM. It turns out that effect of SQM leads to covering up the region in the initial data set window for naked singularity. That is, it tends to favor black hole against naked singularity and consequently the CCC. This happens because SQM contributes an additional attractive potential.
The SQM fluid is characterized by the equation of state $p =
(\rho - 4B)/3$ where $B$ is the bag constant indicating the difference between the energy density of the perturbative and nonperturbative QCD vacuum, and $\rho, p$ are the energy density and thermodynamic pressure of the quark matter [@ew; @sw]. The fluid consists of zero mass particles with the QCD corrections for trace anomaly and perturbative interactions [@hc]. The boundary of a strange star is defined by $p\to 0$ which would imply $\rho\to B$. The typical value of the bag constant is of the order of $B\approx 10^{15}g/cm^3$ while the energy density, $\rho\approx 5 \times 10^{15} g/cm^3$ [@ew]. This shows that SQM will always satisfy the energy conditions because $\rho\ge
p\ge 0$. We shall however consider the equation of state $p = (\rho - 4B)/n$ as a generalization of SQM fluid, and particularly the cases $n = 2,\to\infty$ correspond to known cases of the Vaidya - de Sitter and the Vaidya in constant potential bath collapse respectively.
In this paper, we shall first obtain the general solution for SQM null fluid with the generalized equation of state in the general spherically symmetric metric in the Bondi ingoing coordinates. We shall then bring out explicitly the effect of SQM on gravitational collapse in terms of covering of spectrum in the initial data set for naked singularity by finding the threshold values for the parameters involved in the mass function.
The paper is organized as follows. In the next section, we obtain the general solution and analyze the collapse to show the effect of SQM in shrinking the parameter window in the initial data set giving rise to naked singularity. In the following section we discuss the strength of the singularity and we conclude with a discussion.
Strange quark null fluid collapse {#sec:gcsqm}
=================================
Expressed in terms of Eddington advanced time coordinate (ingoing coordinate) $v$, the metric of general spherically symmetric space-time [@bi] $$ds^2 = - A(v,r)^2 f(v,r)\; dv^2
+ 2 A(v,r)\; dv\; dr + r^2 d \Omega^2 \label{eq:me2}$$ $ d\Omega^2 = d \theta^2+ sin^2 \theta d \phi^2$. Here $A$ is an arbitrary function. It is useful to introduce a local mass function $m(v,r)$ defined by $f = 1 - {2 m(v,r)}/{r}$. For $m = m(v)$ and $f=1$, the metric (\[eq:me2\]) reduces to the standard Vaidya metric. We wish to find the general solution of the Einstein equation for the matter field given by Eq.(\[eq:emt\]) for the metric (\[eq:me2\]), which contains two arbitrary functions. It is the field equation $G^0_1 = 0$ that leads to $ A(v,r) = g(v)$. This could be absorbed by writing $d \tilde{v} = g(v) dv$. Hence, without loss of generality, the metric (2) takes the form, $$ds^2 = - \left[1 - \frac{2 m(v,r)}{r}\right] d v^2 + 2
d v d r + r^2 d {\Omega}^2 \label{eq:me}$$ The energy momentum tensor for the null fluid together with SQM can be written in the form [@hc; @ww; @vh] $$T_{ab} = \mu l_a l_b + (\rho+p)(l_a n_b + l_b n_a) + p g_{ab}
\label{eq:emt}$$ Here $\rho$ and $p$ are functions of $v$ and $r$ and the two null vectors $l_{a}$ and $n_a$ $$\begin{aligned}
l_{a} = \delta_a^0, \: n_{a} = \frac{1}{2}
\left[ 1 - \frac{2 m(v,r)}{r} \right ] \delta_{a}^0 -
\delta_a^1
\nonumber \\
l^a = \delta_1^a, \: n^a = - \delta_{0}^{a} - \frac{1}{2}
\left[ 1 - \frac{2 m(v,r)}{r} \right ] \delta_{1}^a \nonumber
\\
l_{a}l^{a} = n_{a}n^{a} = 0 \; ~l_a n^a = -1.\end{aligned}$$ Substituting this in Eq. (\[eq:emt\]), we find $ T^0_0 = T^1_1 = -\rho$, $T^2_2 = T^3_3 = p$ and $T^1_0 = -\mu$, and the trace $T = -2 (\rho -p)$. Here $\rho,~ p$ are the strange quark matter energy density and thermodynamic pressure while $\mu$ is the energy density of the Vaidya null radiation. The Einstein field equations now take the form
\[fe1\] $$\begin{aligned}
&& 4\pi\mu = \frac{ \dot m}{r^2}, \label{equationa}
\\
&& 4\pi \rho = \frac{m ^{\prime } }{r^2},
\label{equationb} \\
&& 8 \pi p = -\frac{ m ^{\prime \prime}}{r}.
\label{equationc}\end{aligned}$$
At higher density, the equation of state becomes uncertain as is the case for nuclear matter and the strange quark matter would be no exception to it. It is therefore appropriate to keep the coefficient $n$ free in the equation of state, $$p = \frac{1}{n}(\rho - 4B) \label{eos}$$ for SQM. We should however be open to the possibility that in the unknown new super dense matter state, there could be altogether a different kind of contribution which could entirely change the situation. So far SQM is the most dense state of matter considered.
Imposing the equation of state and combining Eqs. (\[equationb\]) and (\[equationc\]), we obtain the following differential equation $$m''(v,r) = - \frac{2}{nr} m'(v,r) + \frac{32 \pi B }{n} r \label{de1}$$ From this equation it is clear that the term involving the bag constant $B$ makes the contribution similar to the cosmological constant. We shall thus seek the solution in the form, $$m(v,r) = m_0(v,r) + \frac{\Lambda}{3}r^3 \label{m0}$$ which would lead to $$B = \frac{(n+1)\Lambda}{16\pi} \label{bc}$$ and the differential equation $$m_0''(v,r) + \frac{2}{nr} m_0'(v,r) = 0 \label{de2}$$ This has the general solution $$m_0(v,r) = S(v)r^{(n-2)/n} + M(v) \label{de3}$$ The two arbitrary functions $M(v)$ and $S(v)$ would be restricted only by the energy conditions. Here $\Lambda$ is not the cosmological constant but instead is related to the bag constant $B$ via Eq.(\[bc\]). Thus the metric describing the radial collapse of null SQM in $(v,r, \theta, \phi)$ coordinates reads as: $$ds^2 =- (1 - \frac{2 M(v)}{r} - \frac{2 S(v)} {r^{2/n}} -
\frac{\Lambda r^2}{3} ) dv^2 + 2 dv dr + r^2 d
\Omega^2 \label{eq:me1}$$ This metric represents a solution to the Einstein equations for a collapsing null SQM. The physical quantities for this metric as in [@hc; @ww] are given by $$\mu = \frac{1}{4\pi r^2}
\left[\dot{M}(v)+\dot{S}(v) r^{(n-2)/n} \right]
\label{eq:mu}$$ $$\rho = \frac{1}{4\pi r^2} \left[\frac{n-2}{n}S(v)r^{-2/n} + \Lambda
r^2 \right] \label{eq:rho}$$ $$p = \frac{1}{4n \pi r^2} \left[\frac{n-2}{n}S(v)r^{-2/n} - \Lambda n
r^2 \right] \label{eq:p}$$ Clearly, all the energy conditions would be satisfied for $n \geq 2$ because it would ensure $\rho \geq0$ and $p \geq0 $, while $\mu \geq0$ would be taken care of when we choose the mass functions for both null radiation and SQM. The initial radius of the star from which the collapse begins would be given by $p = 0$ which would also relate the bag constant with the mass function $S(v)$.
We are studying the collapse of SQM null fluid on a non-flat but empty cavity. The first shell arrives at $r=0$ at time $v=0$ and the final at $v=T$. A central singularity of growing mass would develop at $r=0$. For $ v < 0$, $M(v)\;=\;S(v)\;=\;0$, i.e., we have $$ds^2 = - (1 - \frac{\Lambda r^2}{3} ) dv^2 + 2 dv dr + r^2 d
\Omega^2 \label{eq:ime}$$ and for $ v > T$, $\dot{M}(v)\;=\; \dot{S}(v)\;=\;0$, $M = M_0 > 0$. The space-time for $v=0$ to $v=T$ is given by the generalized Vaidya metric (\[eq:me1\]), and for $v>T$ we have the generalized Schwarzschild metric: $$ds^2 = - (1 - \frac{2 M_0}{r} - \frac{\Lambda r^2}{3} ) dv^2 + 2
dv dr + r^2 d \Omega^2 \label{eq:ome}$$
Occurrence of naked singularities {#second}
---------------------------------
In this section, we adapt above solution to study the existence of a naked singularity. Let $K^{a} = {dx^a}/{dk}$ be the tangent vector to the null geodesic, where $k$ is an affine parameter. The geodesic equations, on using the null condition $K^{a}K_{a} = 0$, take the simple form $$\frac{d^2v}{dk^2} + \frac{1}{r} \left[ \frac{M(v)}{r} + \frac{2
S(v)}{ r^{2/n}} - \frac{\Lambda}{3}r^2 \right]\left(\frac{dv}{dk}
\right)^2 = 0 \label{eq:kv1}$$ $$\frac{d^2r}{dk^2} + \left[ \frac{\dot{M}(v)}{r} +
\frac{\dot{S}(v)}{r^{2/n}}
\right] \left( \frac{dv}{dk} \right)^2 = 0 \label{eq:kr1}$$ Radial ($ \theta$ and $ \phi \,=\,const$.) null geodesics of the metric (11) must satisfy the null condition $$\frac{dr}{dv} = \frac{1}{2} \left[1 - \frac{2 M(v)}{r} -
\frac{2 S(v)}{r^{2/n}} - \Lambda \frac{r^2}{3} \right]
\label{eq:de1}$$ Clearly, the above differential equation has a singularity at $r=0$, $v=0$. If the singularity is naked, there must exist null ray emanating from it. By investigating the behavior of radial null geodesics near the singularity, it is therefore possible to determine whether outgoing null curves meet the singularity in the past. To go any further we would require specific form of functions $M(v)$ and $S(v)$, which we choose as follows:
\[FC\] $$\begin{aligned}
2 M(v)=&&\alpha v \; (\alpha > 0), \label{FCa} \\
2 S(v)=&&\beta v^{2/n} \; (\beta > 0), \; \label{Fcb}\end{aligned}$$
Let $X \equiv v/r$ be the tangent to a possible outgoing geodesic from the singularity. In order to determine the nature of the limiting value of $X$ at $r=0$, $v=0$ on a singular geodesic, we let $ X_{0} = \lim_{r \rightarrow 0 \; v\rightarrow 0} X =
\lim_{r\rightarrow 0 \; v\rightarrow 0} \frac{v}{r} $. Using (\[eq:de1\]), (\[FC\]) and L’Hôpital’s rule we get $$X_{0} = \lim_{r\rightarrow 0 \; v\rightarrow 0} X =
\lim_{r\rightarrow 0 \; v\rightarrow 0} \frac{v}{r}=
\lim_{r\rightarrow 0 \; v\rightarrow 0} \frac{dv}{dr} = \frac{2}{1
- \alpha X_0 - \beta X_0^{2/n} - \Lambda r^2/3} \label{eq:lm2}$$ which implies, $$\alpha X_0^2+ \beta X_0^{2/n+1} - X_0 + 2 = 0 \label{eq:ae}$$ This is the equation which would ultimately decide the end state of collapse: a black hole or a naked singularity.
Thus by analyzing this algebraic equation, the nature of the singularity can be determined. The central shell focusing singularity would atleast be locally naked (for brevity we have addressed it as naked throughout this paper), if Eq. (\[eq:ae\]) admits one or more positive real roots [@pj]. The values of the roots give the tangents of the escaping geodesics near the singularity. When there are no positive real roots to Eq. (\[eq:ae\]), there are no out going future directed null geodesics emanating from the singularity. Thus, the occurrence of positive roots would imply the violation of the strong CCC, though not necessarily of the weak form. Hence in the absence of positive real roots, the collapse will always lead to a black hole. The positive roots would define the range for the tangent slopes from which the null geodesics can escape to infinity. The critical slope would be given by the double root, marking the threshold between black hole and naked singularity.
### Case $n=3$
--------- -------------------------- -------------------
$\beta$ Critical Value$\alpha_C$ Equal Roots $X_0$
0 1/8 3.9999
0.05 0.093728958525 4.18083
0.1 0.06294108366 4.39273
0.15 0.032689168719 4.64701
0.2 0.0030431168445 4.9625
--------- -------------------------- -------------------
: Variation of $\alpha_C$ and $X_0$ for various $\beta$ ($n=3$)[]{data-label="table1"}
We now examine the condition for the occurrence of a naked singularity for $n=3$. First note that the Eq. (\[eq:ae\]) is free of $\Lambda$ and hence it has no effect on the question under study. However its presence makes the background space-time asymptotically non flat. This happens because when $r \rightarrow
0$ the term ${\Lambda}r^2/3$ in Eq.(\[eq:lm2\]) tends to zero. For $\beta = 0$, the allowed range for $\alpha$ is given by (0, 1/8\] as obtained earlier [@ns] for the Vaidya null radiation collapse. In this case, it would be black hole for $\alpha > 1/8$.
--------- --------------------- ------------------
$\beta$ $\alpha < \alpha_C$ Roots $(X_0)$
0.0 0.1 1.40338, 1.9342
0.05 0.09 1.52319, 1.7237
0.1 0.06 1.55259, 1.74615
0.15 0.03 1.57943, 1.78409
0.2 0.003 1.69165, 1.72026
--------- --------------------- ------------------
: Values of Roots $X_0$ for $\alpha < \alpha_C$ for different $\beta$ ($n=3$)[]{data-label="table2"}
The numerical computation reveals that Eq. (\[eq:ae\]) would always admit two positive roots for $\alpha \leq
\alpha_C$. Tangent to all outgoing radial null geodesics would lie in the range $X_2 < X < X_1$, where $X_1$ and $X_2$ are the two roots. Table I shows the critical values of $\alpha$ for various values of $\beta$. The window for naked singularity is defined by (0,$\alpha_C$\], and it is black hole for $\alpha > \alpha_C$. Table II indicates the slope range, given by the two roots, for the null geodesics to escape. It is seen that $\alpha_C$ decreases with increase in $\beta$, i.e., initial data set (0,1/8\] for a naked singularity of the Vaidya collapse shrinks by the introduction of SQM. There exists a threshold value $\beta_T = 0.205198$ such that for $\beta
\geq \beta_T$, gravitational collapse of strange quark null fluid would always end into a black hole for all $\alpha$.
Note that $\alpha$ refers to rate of collapse of the null radiation while $\beta$ would refer to that of SQM. The $\beta$-threshold would therefore define a critical rate of collapse for SQM required for collapse to end in a black hole. Then it would fully respect CCC. Thus introduction of quark matter favors formation of black hole.
-------- -----------
$2/n$ $\beta_T$
0.9 0.144
0.8 0.167
0.75 0.18
0.5 0.2052
0.25 0.45
0.125 0.6194
0.0625 0.756
-------- -----------
: Variation of $\beta_T$ with $n$[]{data-label="table3"}
-------- --------- ------------------ --------------------
$2/n$ $\beta$ $ \alpha_C$ Equal Root $(X_0)$
0.90 0.14 0.00343411423125 4.21544
0.80 0.16 0.00502531901422 4.47484
0.75 0.17 0.00679411554354 4.61836
0.5 0.20 0.0306384999095 5.1784
0.25 0.44 0.00178952699837 9.51405
0.125 0.61 0.00077533872957 16.519
0.0625 0.74 0.00061702385153 27.1684
-------- --------- ------------------ --------------------
: Values of equal roots $X_0$ for different $n$ and $\beta$[]{data-label="table4"}
### Other Cases
####
Case $n =2$\
Then we find that $$\mu = \frac{1}{4 \pi r^2} \left[ \dot{M}(v) + \beta \right], \hspace{.2in}
\rho = - p = \frac{\Lambda}{4 \pi}$$ and the algebraic equation takes the form $$(\alpha + \beta) X_0^2 - X_0 + 2 = 0 \label{eq:ae1}$$ The metric in this case takes the form of the Vaidya - de Sitter metric. The singularity is visible for $(\alpha + \beta) < 1/8$. It is the null fluid collapse in the background of the de Sitter space, where $\Lambda$ is generated by the bag constant.
####
Case $n\rightarrow\infty$.\
We have $$\mu = \frac{1}{4\pi r^2} \dot{M}(v), \hspace{.2in} p=0, \hspace{.2in}
\rho = \frac{B}{4\pi r^2} \label{eq:ni}$$ In this case we have the dual Vaidya metric or Vaidya metric with constant potential [@jdj]. The algebraic equation: $\alpha X_0^2 + (\beta - 1) X_0 + 2 = 0$, would admit a positive root for $\alpha \leq 1/8 (\beta-1)^2 $, giving the range for naked singularity as obtained in [@jdj]. This is simply the null fluid collapse in the background of constant potential which is characterized by $T^0_0 = T^1_1 = const./r^2$, as is the case in Eq.(\[eq:ni\]) above. Note that $\beta < 1$ else the metric signature would change.
####
Other $n$\
We also note that as $n$ increases, so does the threshold value $\beta_T$. This is shown in Tables III and IV.
Curvature Strength of Singularities
===================================
An important aspect of a singularity is its gravitational strength [@ft1]. A singularity is gravitationally strong in the sense of Tipler, if it destroys any object which falls into it and weak otherwise. It is now widely believed that space-time does not admit an extension through a strong curvature singularity, i.e., space-time is geodesically incomplete. Through a weak singularity, space-time could be analytically extended to make it geodesically complete. There have been attempts to relate strength of a singularity to its stability [@djd]. Recently, Nolan [@bc] gave an alternative approach to check the nature of singularities without having to integrate the geodesics equations. It was shown [@bc] that a radial null geodesic which runs into $r=0$ terminates in a gravitationally weak singularity if and only if $dr/dk$ is finite in the limit as the singularity is approached (this occurs at $k=0$, the over-dot here indicates differentiation along the geodesic). If the singularity is weak, we have $$\frac{dr}{dk} \sim d_{0} \hspace{0.2in} r \sim d_{0} k$$ Using the asymptotic relationship ($dv/dk \sim d_0 X_0$ and $v \sim d_0 X_0 k $) and Eq. (\[FC\]), the geodesic equation yields $$\frac{d^2v}{dk^2} \sim - (\alpha X_0 d_0^{-1}k^{-1} + \beta X_{0}^{2/n}
d_{0}^{-1} k^{-1} - \frac{\Lambda}{3}d_{0}k)d_0^2 X_0^2$$ But this gives $$\frac{d^2v}{dk^2} \sim c k^{-1}$$ where $c = (\alpha X_0^{(n-2)/n} + \beta)X_0^{2(n+1)/n} d_0^{-1}$. This is inconsistent with $dv/dk \sim d_{0} X_{0}$, which is finite. Since the coefficient $c$ is non-zero, the naked singularity is gravitationally strong in the sense of Tipler [@ft]. Having seen that the naked singularity is a strong curvature singularity, we check it for scalar polynomial singularity. The Kretschmann scalar for the metric (\[eq:me1\]) with the prescriptions (\[FC\]), takes the form $$\begin{aligned}
K = && \frac{4}{3n^4 r^4} \Bigg[ \alpha^2 n^4 X_0^2 + (4 \beta
\Lambda n^2 -6 \beta \Lambda n^3 + 2 \beta \Lambda n^4)
X_{0}^{2/n} +(12 \alpha \beta \Lambda n^2 + \nonumber \\
&& 8 \alpha \beta \Lambda n^3+6 \alpha \beta \Lambda n^4)
X_{0}^{(n+2)/2}
\nonumber \\
+&& (12 \beta^2 n + 15 \beta^2 n^2 +
3 \beta^2 n^4) X^{4/n} \Bigg]
+ \frac{8}{3} \Lambda^2 \label{eq:ks1}\end{aligned}$$ which diverges at the naked singularity and hence the singularity is also a scalar polynomial singularity. The Ricci scalar also diverges. It however vanishes for the Vaidya space-time [@bs]. Thus the naked singularities studied here are strong curvature singularity and hence are physically significant. Lastly, we shall calculate Weyl scalar $$C = \frac{4}{3 n^4 r^4} \left[\alpha n^2 X+2 \beta X^{2/n}+3 \beta X^{2/n}
+\beta n^2 X^{2/n} \right]^2$$ which too would diverge. The Weyl curvature describes non local effects of gravitation produced by free part of the field. It is generated by inhomogeneity and anisotropy, particularly divergence of shear [@ellis]. In the context of naked singularity, like shear and inhomogeneity the Weyl curvature would also play significant role.
Discussion
==========
In this paper, we have obtained the general solution for null SQM fluid with the equation of state given by Eq. (\[eos\]) for the general spherically symmetric metric (\[eq:me\]) in the Eddington advanced time coordinate (ingoing coordinates). We have used the solution to study the end state of the collapse. The present case is an example of non self-similarity as well as non asymptotic flatness, and yet there does occur a regular initial data set which would lead to naked singularity.
The relevant question is what effect does the presence of the SQM have on formation or otherwise of a naked singularity. Our results imply that the presence of SQM leads to shrinking of the initial data space for naked singularity of the Vaidya null fluid collapse. That is, it tends to favor black hole. This tendency is caused by the additional attractive potential, varying as $r^{-2/n}$, produced by SQM which results in strengthening of gravity. There also exists a threshold value $\beta_T =
0.205198$ such that for $\beta \geq \beta_T$ the end state of collapse of null SQM is always a black hole for all $\alpha$. This is the critical value of the rate of collapse of SQM for respecting CCC. The case $n=3$ represents the zero mass particles with QCD corrections for trace anomaly and perturbative interactions [@hc]. The energy conditions require $n\geq2$. The case of $n=2$ corresponds to the null fluid collapse in the background of de Sitter space where the bag constant provides the $\Lambda$. The case $n\rightarrow\infty$, corresponds to the null fluid collapse in the background of constant potential space as studied in [@jdj]. These are the two extreme limiting cases encompassing the physically allowed cases. Though there is no much physical motivation in the context of SQM for $n\neq3$ cases, yet these two particular cases are interesting. That is at least these three cases could be considered in a unified equation of state given by Eq. (8).
As mentioned earlier the strong version of the CCC doesn’t allow even locally naked singularity, i.e., the space-time should be globally hyperbolic. It turns out that necessary and sufficient condition for a singularity to be locally naked is that the algebraic Eq. (\[eq:ae\]) should have atleast one or more positive root [@pj]. Hence existence of the positive roots of Eq. (\[eq:ae\]) is a counter example to the strong version of the CCC. In the absence of the proof of any version of the CCC, such examples remain the only tool to study this important and unresolved problem.
Quark stars could be formed in the realistic astrophysical setting. The core collapse of a massive star after the supernova explosion sets in first and second order phase transitions which result into deconfined quark matter. The other possibility is that some neutron stars could accrete matter and undergo phase transition to turn into quark stars [@cdl; @dpl; @cd]. Thus study of gravitational collapse with quark matter component is quite in order because it is perhaps astrophysically more realistic. In the ultimate stage of collapse close to the singularity, density is diverging. The quark matter contribution would therefore perhaps be most significant in deciding the ultimate result of the collapse.
SGG would like to thank IUCAA, Pune for hospitality while this work was done and UGC, Pune (INDIA) for Grant No. MRP F. No 23-118/2000 (WRO). The invariants in section III has been calculated using GRTensorM [@grt]. Authors are gratefult to the referee for constructive criticsm.
Witten, E. (1984). [*Phys. Rev. D*]{} [**30**]{}, 272. Joshi, P. S. (1993). [*Global Aspects in Gravitation and Cosmology*]{} (Clendron Press, Oxford).; Clarke, C. J. S. (1993). [*Class. Quantum Grav.*]{} [**10**]{}, 1375; Jhingan, S. and Magli, G. (1999). gr-qc/9903103; Singh, T. P. (1999). [*J. Astrophys. Astron.*]{} [**20**]{}, 221.; Joshi, P.S. (2000). [*Pramana*]{} [**55**]{}, 529. See, for example, Joshi Ref. ([@r1]). Joshi, P.S., Dadhich, N., and Maartens, R. gr-qc/0109051 Harko, T. and Chang, K. S. (2000). [*Phys. Lett. A*]{} [**266**]{}, 249. Weinberg, S. (1986). [*The Quantum Theory of Fields Vol. 2*]{} (Cambridge University Press, Cambridge). Barrabes C. and Israel W. (1991). [*Phys. Rev. D* ]{} [**43**]{}, 1129. Wang, A. and Wu, Y. (1999) [*Gen. Rel. Grav.*]{} [**31**]{}, 107. Husain, V. (1996). [*Phys. Rev. D*]{} [**53**]{}, 1759. Ghosh, S. G. and Dadhich, N. (2001). [*Phys. Rev. D*]{} [**64**]{}, 047501. Jhingan, S., Dadhich, N. and Joshi, P.S. (2001). Phy. Rev. D 63 044010. Tipler, F.J. (19987).[*Phys. Lett. A*]{} [**64**]{}, 8. Deshingkar, S. S., Joshi, P. S. and Dwivedi, I. H. (1999). [*Phys. Rev.*]{} [**D 59**]{}, 044018. Nolan, B. C. (1999) [*Phys. Rev. D*]{} [**60**]{}, 024014. Tipler, F. J., Clarke, C. J. S. and Ellis, G. F. R. (1980). In [*General Relativity and Gravitation*]{}, edited by A Held (Plenum, New York). Barve S. and Singh, T.P. (1997) [*Mod. Phys. Lett.*]{} [**A12**]{}, 2415. Ellis, G. F. R. (1971). In [*General Relativity and Cosmology, Proceedings of XLVII Enrico Fermi Summer School*]{}, edited by R K Sachs, (Academic Press). Cheng, K. S., Dai, Z. G. and Lu, T. (1998) [*Int J. Mod. Phys. D*]{} [**7**]{}, 139. Dai, Z. G., Peng, Q. H. and Lu, T. (1995) [*Astrophys. J.*]{} [**D 440**]{}, 815. Cheng, K. S. and Dai, Z.G. (1996). [*Phys. Rev. Lett.*]{} [**77**]{}, 1210. Musgrave, P., Polleney, D. and Lake, K. (1994-1998). [*GRTensorM*]{} (Kingston, Ontario: Queen’s University)
[^1]: E-mail: [email protected]
[^2]: E-mail: [email protected]: To whom all correspondence to be done
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Of the various formalisms developed to treat relativistic phenomena, those based on Clifford’s geometric algebra are especially well adapted for clear geometric interpretations and computational efficiency. Here we study relationships between formulations of special relativity in the spacetime algebra (STA) $C\!\ell_{1,3}$ of the underlying Minkowski vector space, and in the algebra of physical space (APS) $C\!\ell_{3}$. STA lends itself to an absolute formulation of relativity, in which paths, fields, and other physical properties have observer-independent representations. Descriptions in APS are related by a one-to-one mapping of elements from APS to the even subalgebra STA$^{+}$ of STA. With this mapping, reversion in APS corresponds to hermitian conjugation in STA. The elements of STA$^{+}$ are all that is needed to calculate physically measurable quantities (called *measurables*) because only they entail the observer dependence inherent in any physical measurement. As a consequence, every relativistic physical process that can be modeled in STA also has a representation in APS, and *vice versa.* In the presence of two or more inertial observers, two versions of APS present themselves. In the *absolute* version, both the mapping to STA$^{+}$ and hermitian conjugation are observer dependent, and the proper basis vectors of any observer are persistent vectors that sweep out timelike planes in spacetime. To compare measurements by different inertial observers in APS, we express them in the proper algebraic basis of a single observer. This leads to the *relative* version of APS, which can be related to STA by assigning every inertial observer in STA to a single absolute frame in STA. The equivalence of inertial observers makes this permissible. The mapping and hermitian conjugation are then the same for all observers. Relative APS gives a covariant representation of relativistic physics with spacetime multivectors represented by multiparavectors in APS. We relate the two versions of APS as consistent models within the same algebra.'
author:
- 'William E. Baylis[^1]'
- 'Garret Sobczyk[^2]'
title: 'Relativity in Clifford’s Geometric Algebras of Space and Spacetime'
---
Introduction
============
Clifford’s geometric algebra offers a powerful unifying language for the study of physics [@Hes66; @Hes84; @Hes87; @Hes03; @Bay96; @Loun97; @Ablam03; @Snygg97; @GuSp97; @Dor03]. Relativistic problems can be treated efficiently both in the spacetime algebra (STA) [@Hes74; @Hes03], the Clifford algebra $C\!\ell_{1,3}$ of Minkowski spacetime, and in the algebra of physical space (APS) [@Bay99; @Bay89; @Bay80; @Hes87], the Clifford algebra $C\!\ell_{3}$ of three-dimensional Euclidean physical space. As real algebras, STA operates in a linear space of sixteen dimensions, whereas the linear space of APS has eight dimensions. APS can also be viewed as the 4-dimensional algebra of complex paravectors [@Sob81]. It is also isomorphic to complex quaternions, which have along history of applications in relativity,[@Silberstein1912; @Conway1912] but whose geometrical interpretation is less obvious.
The connection between STA and APS highlights an important relation, only rarely explicitly expressed, between the basis vectors in Newtonian mechanics and those of special relativity. Newtonian basis vectors are usually viewed as unit displacement vectors in physical space that *persist* in time. Relativistically, such persistent vectors are associated with timelike *bivectors* (bivectors with positive squares, see below) representing planes that are swept out *in spacetime* by a given spatial direction with the passage of time. Under boosts, timelike bivectors transform into spacetime bivectors that have picked up spatial bivector parts (with negative squares), and the corresponding Newtonian basis vectors transform into a mixture of a spatial vector and a spatial bivector (a plane).
The formulation of relativity in STA can be characterized as *absolute,* in that physical paths, fields, and other properties of objects are expressed independent of any observer. The formulation in APS can be either *absolute* [@Sob81] or *relative* [@Bay99; @Bay89; @Bay80]. In both APS versions, the relationship between APS and STA$^{+}$ is expressed as an algebra isomorphism together with an operation called *hermitian conjugation*. Hermitian conjugation, identified with *reversion* in APS, separates elements of APS into real and imaginary parts, the real part consisting of scalars and vectors (elements of grade 0 and 1), and the imaginary part consisting of pseudoscalars (trivectors) and bivectors (elements of grade 3 and 2).
If we consider only one inertial observer, the two APS versions are equivalent. In the *absolute* APS approach, one posits a distinct absolute frame and hermitian conjugation for each observer, so that the reality and the grading of the elements of APS are observer dependent. The *relative* APS approach can be related to STA by assuming that all observers use a single absolute inertial frame in STA to form their *proper basis*. Since all inertial frames in STA are equivalent, each inertial observer can identify her frame with the chosen inertial frame. In this way, the same hermitian conjugation is used for all observers. The relative APS formulation admits a covariant formulation, of critical importance to physicists, in which real paravectors are naturally associated with spacetime vectors and higher paravector grades represent other covariant geometrical objects in spacetime. While its formulation and justification[@Bay99; @Bay96] does not depend on STA, its connection to STA$^{+}$ clearly shows the relationship between the conceptually different relative and absolute approaches.
In the following sections, we first review APS and STA, and then construct APS from the even subalgebra STA$^{+}$ of STA. The formulation of APS is obtained initially for a single observer. This part is largely a review of previous work[@Hes74] but is included here for completeness. The connections become more convoluted when additional observers are added, as we investigate first in the absolute version of APS. We then see how, through the relation of measurable coefficients of covariant spacetime elements, we are led to the relative version. We can recover STA from APS by formulating relations in the proper basis of a single inertial observer in APS assigned to the absolute frame of STA. From these relationships we conclude that every physical process formulated in STA can equally well be described in APS.
Review of APS
=============
We present here a brief summary of the use of APS to model relativistic phenomena. More details can be found elsewhere [@Bay96; @Bay99]. The elements of APS are the real vectors $\mathbf{u,v,w}$ of physical space $\mathbb{R}^{3}$ and all their sums and products $\mathbf{uu,uv+uvw,\ldots}$. The elements of APS form an associative algebra under addition and multiplication. We will see later in the formulation of APS as STA$^{+}$ that the concept of a *vector* is itself, implicitly, *relative* to an observer.
The square of any vector $\mathbf{u}\in APS$ is defined as its length squared :$$\mathbf{u^{2}}\equiv\mathbf{uu}=\mathbf{u\cdot u~,} \label{axiom}%$$ where $\mathbf{u}\cdot\mathbf{u}$ is the usual inner product. This axiom, together with the usual rules for adding and multiplying square matrices, determines the entire algebra. If we put $\mathbf{u}=\mathbf{v+w,}$ the axiom implies$$\mathbf{vw+wv}=2\mathbf{v\cdot w~.} \label{corollary}%$$ Evidently the algebra is not commutative and, in particular, the product of perpendicular vectors is anticommutative $\mathbf{uv}=-\mathbf{vu}$. It is called a *bivector* and has a geometric interpretation as the oriented plane containing $\mathbf{u}$ and $\mathbf{v.}$
Let $\left\{ \mathbf{e}_{1},\mathbf{e}_{2},\mathbf{e}_{3}\right\} $ be an orthonormal basis of $\mathbb{R}^{3}$. The corollary (\[corollary\]) implies that $$\mathbf{e}_{j}\mathbf{e}_{k}+\mathbf{e}_{k}\mathbf{e}_{j}=2\delta_{jk}~,
\label{APSaxiom}%$$ where the Kronecker delta $\delta_{jk}$ gives the Euclidean metric of physical space. For example, $\mathbf{e}_{1}^{2}=1$ and $\mathbf{e}_{1}\mathbf{e}%
_{2}=-\mathbf{e}_{2}\mathbf{e}_{1}.$ The bivector $\mathbf{e}_{1}%
\mathbf{e}_{2}$, a multivector of grade 2, represents a directed area in the plane of the vectors. Its direction corresponds to circulation in the plane: if the circulation is reversed, the sign of the bivector is reversed. The bivector replaces the (Gibbs-Heaviside) *vector cross product* of polar vectors, but unlike the usual cross product, it is *intrinsic* to the plane and can be applied to planes in spaces of any number of dimensions.
The unit bivector $\mathbf{e}_{1}\mathbf{e}_{2}$ squares to $$(\mathbf{e}_{1}\mathbf{e}_{2})^{2}=-\mathbf{e}_{2}\mathbf{e}_{1}\mathbf{e}%
_{1}\mathbf{e}_{2}=-1$$ and generates rotations and reflections in the plane of $\mathbf{e}_{1}$ and $\mathbf{e}_{2}.$ A general vector $\mathbf{v,}$ with components both in the plane and perpendicular to it, is rotated through the angle $\phi$ in the $\mathbf{e}_{1}\mathbf{e}_{2}$ plane by $$\mathbf{v}\rightarrow R\mathbf{v}R^{\dag}, \label{rotation}%$$ where the *rotors *$R,R^{\dag}$ are $$\begin{aligned}
R & =\exp\left( -\mathbf{e}_{1}\mathbf{e}_{2}\phi/2\right) =\cos\frac
{\phi}{2}-\mathbf{e}_{1}\mathbf{e}_{2}\sin\frac{\phi}{2}\\
R^{\dag} & =\cos\frac{\phi}{2}-\left( \mathbf{e}_{1}\mathbf{e}_{2}\right)
^{\dag}\sin\frac{\phi}{2}=\cos\frac{\phi}{2}-\mathbf{e}_{2}\mathbf{e}_{1}%
\sin\frac{\phi}{2}=R^{-1}~.\end{aligned}$$
The dagger $\dag$ used above denotes the conjugation of *reversion,* which reverses the order of vectors in all products. Thus, for any vector $\mathbf{v,\ v}^{\dag}=\mathbf{v,}$ and the reversion of a product, say $AB,$ of arbitrary elements can be found from $\left( AB\right) ^{\dag}=B^{\dag
}A^{\dag}.$ (A tilde is used in STA to indicate reversion, but when we associate APS with the even subalgebra STA$^{+}$ of STA we require distinct symbols.) In spaces of definite metric such as Euclidean spaces, one commonly represents the basis vectors by hermitian matrices. The dagger then corresponds to hermitian conjugation and can be used to split elements into real (hermitian) and imaginary (antihermitian) parts:$$\begin{aligned}
A & =\left\langle A\right\rangle _{\Re}+\left\langle A\right\rangle _{\Im}\\
\left\langle A\right\rangle _{\Re} & =\frac{A+A^{\dag}}{2},\ \left\langle
A\right\rangle _{\Im}=\frac{A-A^{\dag}}{2}~.\end{aligned}$$
The *standard basis* of APS over the reals can be specified by $$\{1,\mathbf{e}_{1},\mathbf{e}_{2},\mathbf{e}_{3},\mathbf{e}_{2}\mathbf{e}%
_{3},\mathbf{e}_{3}\mathbf{e}_{1},\mathbf{e}_{1}\mathbf{e}_{2},\mathbf{e}%
_{1}\mathbf{e}_{2}\mathbf{e}_{3}\},$$ where the *trivector* $\mathbf{e}_{1}\mathbf{e}_{2}\mathbf{e}_{3}$ squares to $-1.$ In APS, $\mathbf{e}_{1}\mathbf{e}_{2}\mathbf{e}_{3}$ is the *volume element*, also known as the *unit pseudoscalar.* It commutes with every vector and hence with all elements and can therefore be identified with the unit imaginary:$$\mathbf{e}_{1}\mathbf{e}_{2}\mathbf{e}_{3}=i~.$$ The *center* of APS (the part that commutes with all elements) is spanned by $\left\{ 1,i\right\} $ and is identified with the complex field. Bivectors can be identified as imaginary vectors (pseudovectors) in APS. For example, $\mathbf{e}_{1}\mathbf{e}_{2}=\mathbf{e}_{1}\mathbf{e}_{2}%
\mathbf{e}_{3}\mathbf{e}_{3}=i\mathbf{e}_{3}.$ We can now take the set $$\{1,\mathbf{e}_{1},\mathbf{e}_{2},\mathbf{e}_{3}\}$$ as the standard basis of APS over the complex scalars.
The sum of a real scalar and a real vector is called a *paravector.* A typical paravector $p$ can be expanded$$p=p^{0}+\mathbf{p}=p^{\mu}\mathbf{e}_{\mu},$$ where, for notational convenience in using the compact Einstein summation convention, we put $\mathbf{e}_{0}=1.$ The convention is that repeated lower-case Greek indices are summed over $0,1,2,3,$ whereas repeated lower-case Latin indices are summed over the spatial values $1,2,3.$ Every element in APS can be expressed as a complex paravector. Reversion (the dagger conjugation) complex-conjugates the complex coefficients and thus changes the sign of the pseudoscalar and pseudovector parts. Real paravector space is a four-dimensional linear space spanned by the basis $\left\{ \mathbf{e}%
_{0},\mathbf{e}_{1},\mathbf{e}_{2},\mathbf{e}_{3}\right\} $ over the reals.
Just as for complex numbers, a natural quadratic form in paravector space is given by $$Q\left( p\right) =p\bar{p}, \label{Qform}%$$ where $\bar{p}=p^{0}-\mathbf{p}$ is called the *Clifford conjugate* of $p.$ Clifford conjugation is extended to general elements $A,B,$ of APS as an antiautomorphism: $\overline{AB}=\bar{B}\bar{A}.$ It conveniently splits elements into *scalarlike* (S) and *vectorlike* (V) parts:$$\begin{aligned}
A & =\left\langle A\right\rangle _{S}+\left\langle A\right\rangle _{V}\\
\left\langle A\right\rangle _{S} & =\frac{A+\bar{A}}{2},\ \left\langle
A\right\rangle _{V}=\frac{A-\bar{A}}{2}~.\end{aligned}$$ The quadratic form (\[Qform\]) is scalarlike. Through it, paravector space inherits from the Euclidean metric of the underlying space of spatial vectors an inner product with the Minkowski spacetime metric: $$\left( p,q\right) =\left\langle p\bar{q}\right\rangle _{S}=\frac{p\bar
{q}+q\bar{p}}{2}=p^{\mu}q^{\nu}\eta_{\mu\nu}%$$ with the metric tensor $$\eta_{\mu\nu}=\left\langle \mathbf{e}_{\mu}\mathbf{\bar{e}}_{\nu}\right\rangle
_{S}=\left\{
\begin{array}
[c]{rl}%
1, & \mu=\nu=0\\
-1, & \mu=\nu=1,2,3\\
0, & \mu\neq\nu
\end{array}
\right. . \label{Minkowski}%$$
The appearance of the Minkowski spacetime metric suggests the use of real paravectors to model vectors in four-dimensional spacetime. Oriented planes in spacetime are then modeled by *biparavectors* [@Bay99] such as $\left\langle p\bar{q}\right\rangle _{V}=p^{\mu}q^{\nu}\left\langle
\mathbf{e}_{\mu}\mathbf{\bar{e}}_{\nu}\right\rangle _{V},$ which represents the plane containing all linear combinations of the real paravectors $p$ and $q,$ and which generally has both real (vector) and imaginary (bivector) parts. Rotations in paravector space are generated by biparavectors and leave the quadratic form (and hence inner products) of paravectors invariant. They represent physical Lorentz transformations. Rotations of the paravector $p$ in a single spacetime plane have the form$$p\rightarrow LpL^{\dag}, \label{Lrotp}%$$ where $L$ is a *Lorentz rotor* of the form $L=\exp\left( \mathbf{W}%
/2\right) $ and $\mathbf{W}$ is a biparavector for the plane of rotation. In the special case that $\mathbf{W}$ is imaginary, $L$ is a spatial rotation, and in the special case that it is real, $L$ is a boost (or hyperbolic rotation). More generally, any Lorentz rotor $L$ can be factored into the product $L=BR$ of a spatial rotation $R$ and a boost $B.$ This formulation is pursued below in the relative formulation of APS.
Review of STA
=============
STA, introduced by Hestenes [@Hes66; @Hes74], is the geometric algebra $C\!\ell_{1,3}$ of Minkowski spacetime. Minkowski spacetime has a pseudo-Euclidean metric that highlights the intrinsic difference between time and space. The aspects of STA presented here are those needed in our discussion below. Since APS can be equated to the even subalgebra of STA, particular attention is paid to this subalgebra. Each abstract inertial frame of STA consists of a constant 4-dimensional orthonormal vector basis $\left\{
\gamma_{0},\gamma_{1},\gamma_{2},\gamma_{3}\right\} \equiv\left\{
\gamma_{\mu}\right\} $ satisfying$$\gamma_{\mu}\gamma_{\nu}+\gamma_{\nu}\gamma_{\mu}=2\eta_{\mu\nu},
\label{STAaxiom}%$$ where $\eta_{\mu\nu}$ are elements (\[Minkowski\]) of the Minkowski spacetime metric tensor.
The *history* (worldline) of an idealized point particle $P$ is a timelike curve $r^{P}\left( \tau^{P}\right) $ giving its position in STA as a function of a scalar parameter $\tau^{P}$, which we take to be its proper time. The tangent vector $u^{P}=dr^{P}/d\tau^{P}$ is its *proper velocity* (in units with $c=1$). In a *commoving frame* (an inertial frame instantaneously moving with the observer) $\left\{ \gamma_{\mu}^{P}\right\}
$ of the particle, the displacement has only a time component:$$dr^{P}=\gamma_{0}^{P}d\tau^{P},$$ and the commoving unit time axis $\gamma_{0}^{P}$ is thus seen to be coincident with the proper velocity: $u^{P}=\gamma_{0}^{P}~.$
The spacetime curve $r^{P}\left( \tau^{P}\right) $ and its tangent vector $u^{P}\ $are *abstract* and do not, by themselves, determine physically measurable values. They are *independent of the observer* and therefore unchanged by *passive* transformations (transformations of the observer). For this reason Hestenes[@Hes03] calls such spacetime quantities *invariant*. However, they do transform under *active* Lorentz rotations, and to avoid possible confusion with *Lorentz scalars*, which are invariant under both passive and active transformations, we prefer to call them *absolute*. Physically measurable quantities (*measurables*), on the other hand, are either Lorentz invariants or are derived from vector components *relative* to the observer. As we show below, these are determined by even elements of STA.
An *inertial observer*, say Alice, can be idealized as a congruence of parallel histories $r^{A}\left( \tau^{A}\right) ,$ all with the same given proper velocity $u^{A}=dr^{A}/d\tau^{A}$ with $du^{A}/d\tau^{A}=0$ and a constant commoving frame $\left\{ \gamma_{\mu}^{A}\right\} $ with $\gamma_{0}^{A}=u^{A}.$ When Alice measures physical quantities represented by geometric objects such as vectors, she normally expresses them in her frame $\left\{ \gamma_{\mu}^{A}\right\} $ of spacetime basis vectors. The $\gamma_{\mu}^{A}$ are abstract, and it is the scalar coefficients of the expansion that constitute the measurables for Alice. A different observer, say Bob, has his own frame $\left\{ \gamma_{\mu}^{B}\right\} $ with $\gamma
_{0}^{B}=u^{B}$ that he normally uses for measurements. We assume that the handedness of the two frames is the same:$$\gamma_{0}^{A}\gamma_{1}^{A}\gamma_{2}^{A}\gamma_{3}^{A}=\gamma_{0}^{B}%
\gamma_{1}^{B}\gamma_{2}^{B}\gamma_{3}^{B}\equiv\mathbf{I}~,$$ where the pseudoscalar $\mathbf{I}$ of STA anticommutes with vectors.
Any two inertial frames of the same handedness are related by a *Lorentz rotation,* also known as a restricted (proper orthochronous) Lorentz transformation, and is specified by a rotor $L.$ Every Lorentz rotation can be expressed as the product of a *spatial rotation* and a *boost* (a velocity transformation). There is in particular a rotor $L$ that relates Alice’s frame to Bob’s:$$\gamma_{\mu}^{B}=L\gamma_{\mu}^{A}\tilde{L}~, \label{ABtransf}%$$ where $\symbol{126}$ indicates reversion in STA. Lorentz rotors are *unimodular*: $$L\tilde{L}=1, \label{STAunimod}%$$ and consequently all products of spacetime vectors transform in the same way in STA. For $\mu=0$, the transformation (\[ABtransf\]) relates the proper velocities$$\gamma_{0}^{B}=u^{B}=L\gamma_{0}^{A}\tilde{L}=Lu^{A}\tilde{L}~.$$
It is common to define hermitian conjugation of any element $K$ for Alice by$$K^{\dag A}=\gamma_{0}^{A}\tilde{K}\gamma_{0}^{A}. \label{STAdag}%$$ The conjugation symbol $\dag A$ is used here instead of the more usual $\dag$ to underscore its dependence on the observer’s frame. In terms of the hermitian conjugate, the proper velocities are related by$$\begin{aligned}
\gamma_{0}^{B} & =u^{B}=LL^{\dag A}\gamma_{0}^{A}=LL^{\dag A}u^{A}%
\nonumber\\
LL^{\dag A} & =\gamma_{0}^{B}\gamma_{0}^{A}=u^{B}u^{A}, \label{Lorentzrot}%\end{aligned}$$ where we noted from (\[STAaxiom\]) that $\left( \gamma_{0}^{A}\right)
^{2}=1.$ If the frames of Alice and Bob are related by a pure spatial rotation, their time axes and hence proper velocities $u^{B}$ and $u^{A}$ are equal and $L^{\dag A}=\tilde{L}=L^{\dag B}.$ On the other hand, if their frames are related by a pure boost, then $L^{\dag A}=L=L^{\dag B}$ and $L^{2}=\gamma_{0}^{B}\gamma_{0}^{A}=u^{B}u^{A}.$ In the most general case, the Lorentz rotor $L$ is the product of boost and spatial-rotation rotors, and $LL^{\dag A}$ is simply the square of the boost rotor.
A boost along the $\gamma_{1}^{A}$ direction is a Lorentz rotation generated by the STA bivector $\gamma_{10}^{A}\equiv\gamma_{1}^{A}\gamma_{0}^{A},$ and it has the form $\exp\left( w\gamma_{10}^{A}/2\right) ,$ where the scalar parameter $w$ is called the rapidity of the boost. We can expand the exponential to get for the rotor $L$$$L^{2}=\exp\left( w\gamma_{10}^{A}\right) =\gamma\left( 1+\mathbf{v}\right)
=u^{B}u^{A},$$ where $\gamma\equiv\cosh w$ is the Lorentz time dilation factor between the observers, and $\mathbf{v}=\gamma_{10}^{A}\tanh w$ gives the relative coordinate velocity of Bob with respect to Alice. The coordinate velocity of Alice as seen by Bob is $-\mathbf{v.}$ The plane of the Lorentz rotation is itself invariant under $L$: $\gamma_{10}^{B}=L\gamma_{10}^{A}\tilde{L}%
=\gamma_{10}^{A},$ and $\gamma$ and $\left\vert \mathbf{v}\right\vert $ are measurables for both Alice and Bob.
We want to compare measurements made by Alice and Bob of an absolute spacetime vector, which for concreteness we take to be the spacetime position $r.$ (Since $r$ depends on the origin of coordinates, in order that this transform under fixed $L$ as a spacetime vector, we must assume that the origins of Alice’s and Bob’s frames coincide. Alternatively, we could replace $r$ by an affine spacetime vector.) Thus, $r$ might be one point on the history of a point particle or some other event. It can be expanded in the basis vectors of any inertial frame, for example in the frame $\left\{ \gamma_{\mu}%
^{A}\right\} $ of Alice:$$r=r_{A}^{\mu}\gamma_{\mu}^{A}~.$$
As emphasized above, the spacetime vectors $r$ and $\gamma_{\nu}^{A}$ are abstract and not directly measurable; it is the scalar coefficients $r_{A}^{\mu}$ that Alice can measure. The time component of $r$ measured by Alice is$$t_{A}\equiv r_{A}^{0}=r\cdot\gamma_{0}^{A}~.$$ More generally, the spacetime vector $r$ *relative to Alice* is given by the expansion$$r\gamma_{0}^{A}=r_{A}^{\mu}\gamma_{\mu}^{A}\gamma_{0}^{A}\equiv r_{A}%
=t_{A}+\mathbf{r}_{A}~, \label{ST}%$$ where the spacetime bivector $\mathbf{r}_{A}=r\wedge\gamma_{0}^{A}=r_{A}%
^{k}\gamma_{k0}^{A}\in$ STA$^{+},$ with $\gamma_{k0}^{A}\equiv\gamma_{k}%
^{A}\gamma_{0}^{A},$ is interpreted in APS as the spatial *position vector* from the origin to the particle ** that Alice measures. The even element $r_{A}=r\gamma_{0}^{A}$ is called the *relative position,* and its expansion (\[ST\]) is called a *space-time* split ** [@Hes74; @Hes03].
Analogous relations can be written for Bob’s measurables. The relative positions $r_{A}=r\gamma_{0}^{A}$ and $r_{B}=r\gamma_{0}^{B}$ are related by the *passive* Lorentz rotation$$\begin{aligned}
r_{B} & =r\gamma_{0}^{B}=rL\gamma_{0}^{A}\tilde{L}=r\gamma_{0}^{A}\tilde
{L}^{\dag A}\tilde{L}\nonumber\\
& =r_{A}\tilde{L}^{\dag A}\tilde{L}. \label{passive}%\end{aligned}$$ The inverse transformation can be written$$r_{A}=r_{B}LL^{\dag A}=r_{B}L^{\dag B}L~,$$ where $L^{\dag B}=\gamma_{0}^{B}\tilde{L}\gamma_{0}^{B}=L\gamma_{0}^{A}%
\tilde{L}\tilde{L}L\gamma_{0}^{A}\tilde{L}=LL^{\dag A}\tilde{L}~.$
A different form of transformation results by boosting the event $r$ for a given observer. This is an example of an *active* transformation, in which the observer is fixed and the event is transformed. Let$$r^{\prime}=Lr\tilde{L}.$$ This boost changes the relative position with respect to Alice according to$$\begin{aligned}
r_{A}^{\prime} & =r^{\prime}\gamma_{0}^{A}=r_{A}^{\prime\mu}\gamma_{\mu}%
^{A}\gamma_{0}^{A}=Lr\tilde{L}\gamma_{0}^{A}=Lr\gamma_{0}^{A}L^{\dag
A}\nonumber\\
& =Lr_{A}L^{\dag A}~. \label{active}%\end{aligned}$$ If both of these transformations are performed together (the order is not important when the same rotor is used for both) we obtain the relative spacetime position of the transformed event with respect to Bob:$$\begin{aligned}
r_{B}^{\prime} & =r^{\prime}\gamma_{0}^{B}=Lr\gamma_{0}^{A}\tilde{L}%
=Lr_{A}^{\mu}\gamma_{\mu}^{A}\gamma_{0}^{A}\tilde{L}\nonumber\\
& =Lr_{A}\tilde{L}~. \label{both}%\end{aligned}$$
The three transformations of measurables (\[passive\]), (\[active\]), (\[both\]) have distinct forms. In the transformation (\[both\]), in which both the observer and the event are boosted by the same amount, the unimodularity of $L$ (\[STAunimod\]) implies that the time measurements are the same: $t_{B}^{\prime}=t_{A},$ but the positions generally differ$$\mathbf{r}_{B}^{\prime}=r_{B}^{\prime k}\gamma_{k0}^{B}=L\mathbf{r}_{A}%
\tilde{L}=Lr_{A}^{j}\gamma_{j0}^{A}\tilde{L}~.$$ How are the three transformations to be interpreted physically?
Measurables in STA
------------------
To understand the *physical interpretation* of the above equations, it is important to note that all of the measurables have been expressed in terms of elements of STA$^{+}$, the even subalgebra of STA, spanned by the basis $\left\{ 1,\gamma_{\mu\nu},I\right\} ,~$where the basis bivectors in STA are $\gamma_{\mu\nu}\equiv\frac{1}{2}\left( \gamma_{\mu}\gamma_{\nu}-\gamma_{\nu
}\gamma_{\mu}\right) $ with $0\leq\mu<\nu\leq3$. The reason for this is that every physical measurement of a vector is *relative* and involves *two* absolute spacetime vectors: the absolute vector of the event or property being measured and a basis vector of the reference frame used by the observer. The results of the measurement depend on the orientation and motion of one with respect to the other.
Of course it is most common for an observer to employ a reference frame at relative rest. Measurements can then be expressed in STA$^{+}$ in the observer-dependent *proper basis* $$\{\boldsymbol{\sigma}_{\mu}^{A}=\gamma_{\mu}^{A}\gamma_{0}^{A}\},\ \left\{
\boldsymbol{\sigma}_{\mu}^{B}=\gamma_{\mu}^{B}\gamma_{0}^{B}\right\} .
\label{proper}%$$ The bases are called *proper* because they are based on the absolute frames that are at rest with respect to the observer. The basis element representing the time axis in the relative frame is in each case unity:$$\boldsymbol{\sigma}_{0}^{A}=1=\boldsymbol{\sigma}_{0}^{B}%$$ and the spatial elements are spacetime bivectors $\boldsymbol{\sigma}%
_{k}=\gamma_{k0}=\gamma_{k}\gamma_{0}$ representing planes containing both the spatial direction $\gamma_{k}$ and the time direction $\gamma_{0}~.$ These are the planes swept out by the unit spatial vector $\gamma_{k}$ in one unit of time. They are the spacetime planes representing *persistent* spatial vectors in physical space. The relation between the relative basis vectors is$$\boldsymbol{\sigma}_{\mu}^{B}=L\boldsymbol{\sigma}_{\mu}^{A}\tilde{L}
\label{sigmatransf}%$$ for $\mu=0,1,2,3$, which follows directly from $\gamma_{\mu}^{B}=L\gamma_{\mu
}^{A}\tilde{L}$ and $L\tilde{L}=1.$ We will return to the meaning of this transformation below.
In terms of these proper bases, the three transformations (\[passive\],\[active\],\[both\]) are$$%
\begin{tabular}
[c]{ll}%
$r_{A}\rightarrow r_{B}=r_{B}^{\mu}\boldsymbol{\sigma}_{\mu}^{B}=r_{A}%
\tilde{L}^{\dag A}\tilde{L}$ & $\text{passive}$\\
$r_{A}\rightarrow r_{A}^{\prime}=r_{A}^{\prime\mu}\boldsymbol{\sigma}_{\mu
}^{A}=Lr_{A}L^{\dag A}$ & $\text{active}$\\
$r_{A}\rightarrow r_{B}^{\prime}=r_{B}^{\prime\mu}\boldsymbol{\sigma}_{\mu
}^{B}=Lr_{A}\tilde{L}$ & $\text{both.}$%
\end{tabular}
\ \ \ \ \ \ \label{transf1}%$$ The measurables for Alice are the scalar coefficients $r_{A}^{\mu}$ and $r_{A}^{\prime\mu},$ and for Bob, they are $r_{B}^{\mu}$ and $r_{B}^{\prime
\mu}.$ We can relate them using the orthogonality of the basis vectors. For example, for the passive case,$$r_{B}^{\mu}=r_{A}^{\nu}\left\langle \boldsymbol{\sigma}_{\nu}^{A}\tilde
{L}^{\dag A}\tilde{L}\boldsymbol{\sigma}_{\mu}^{B}\right\rangle _{S}%
=r_{A}^{\nu}\left\langle \boldsymbol{\sigma}_{\nu}^{A}\tilde{L}^{\dag
A}\boldsymbol{\sigma}_{\mu}^{A}\tilde{L}\right\rangle _{S}%$$ However, the relations are most transparently seen if the transformations (\[transf1\]) are expressed in terms of a *proper basis* of a *single observer*. If we insert the transformation (\[sigmatransf\]) into (\[transf1\]), we find$$%
\begin{tabular}
[c]{ll}%
$r_{B}^{\mu}\boldsymbol{\sigma}_{\mu}^{A}=r_{A}^{\nu}\tilde{L}%
\boldsymbol{\sigma}_{\nu}^{A}\tilde{L}^{\dag A}=\tilde{L}r_{A}\tilde{L}^{\dag
A}$ & $\text{passive}$\\
$r_{A}^{\prime\mu}\boldsymbol{\sigma}_{\mu}^{A}=r_{A}^{\nu}L\boldsymbol{\sigma
}_{\nu}^{A}L^{\dag A}=Lr_{A}L^{\dag A}$ & $\text{active}$\\
$r_{B}^{\prime\mu}\boldsymbol{\sigma}_{\mu}^{A}=r_{A}^{\nu}\boldsymbol{\sigma
}_{\nu}^{A}=r_{A}$ & $\text{both.}$%
\end{tabular}
\label{transf2}%$$
Note that the active and passive transformations (\[transf2\]) have similar forms but with inverse rotors. They demonstrate in particular that under either active or passive boosts, time and space components are mixed and time intervals (and hence clock rates) change. The last relation of (\[transf2\]) shows that when both the event and the observer are transformed by the same rotor, the components are unchanged:$$r_{B}^{\prime\mu}=r_{A}^{\mu}.$$ In other words, the spacetime vector $r^{\prime}$ measured by Bob has the same components as $r$ measured by Alice.
Notice that $\boldsymbol{\sigma}_{k}^{A}$ and $\boldsymbol{\sigma}_{k}^{B}$, for $k=1,2,3,$ are timelike bivectors when expressed in the frames $\left\{
\gamma_{\mu}^{A}\right\} $ and $\left\{ \gamma_{\mu}^{B}\right\} ,$ respectively, of STA. In the next section, we will reinterpret the $\boldsymbol{\sigma}_{k}^{A}$ to be the persistent vectors of a single, absolute inertial frame, and study the resulting geometric algebra, which is isomorphic to $Cl_{3}$.
APS as STA$^{+}$
================
Since APS is isomorphic to the even subalgebra of STA, we can derive the relativistic formalism for APS by equating the basis vectors $\mathbf{e}_{k}$ of APS to the appropriate elements of STA$^{+}.$ For Alice, we have $$\mathbf{e}_{k}=\boldsymbol{\sigma}_{k}^{A}=\gamma_{k0}^{A}~,\;k=1,2,3.
\label{ident}%$$ These three orthonormal vectors satisfy the axiom (\[APSaxiom\]) and generate both APS and STA$^{+}$. The identification (\[ident\]) associates the three spatial basis vectors of APS with timelike bivectors of STA. This reinforces the concept of *persistent vectors* in APS that sweep out timelike planes in STA. Since $\mathbf{\sigma}_{0}^{A}=1,$ we can extend the identification (\[ident\]) to Alice’s proper paravector basis $\left\{
\mathbf{e}_{\mu}=\boldsymbol{\sigma}_{\mu}^{A}\right\} $ with$$\mathbf{e}_{\mu}=\boldsymbol{\sigma}_{\mu}^{A}=\gamma_{\mu}^{A}\gamma_{0}%
^{A}~,\;\mu=0,1,2,3. \label{e0biv}%$$ From this identification, the volume elements of APS and STA are$$i=\mathbf{e}_{1}\mathbf{e}_{2}\mathbf{e}_{3}=\boldsymbol{\sigma}_{1}%
^{A}\boldsymbol{\sigma}_{2}^{A}\boldsymbol{\sigma}_{3}^{A}=\gamma_{10}%
^{A}\gamma_{20}^{A}\gamma_{30}^{A}=\gamma_{0}^{A}\gamma_{1}^{A}\gamma_{2}%
^{A}\gamma_{3}^{A}=\mathbf{I}.$$ As in STA, the volume element of APS squares to $-1,$ but whereas in STA $\mathbf{I}$ anticommutes with all vectors, in both APS and STA$^{+}$ the volume element is part of the center of the algebra, that is, it commutes with all elements.
Clifford conjugation in APS corresponds to reversion in STA$^{+}$$$\begin{aligned}
\mathbf{\bar{e}}_{0} & =\mathbf{e}_{0}=1=\tilde{1}\\
\mathbf{\bar{e}}_{k} & =-\mathbf{e}_{k}=-\gamma_{k0}^{A}=\tilde{\gamma}%
_{k0}^{A}~,\end{aligned}$$ and it follows that Clifford conjugation of an observer’s proper basis is the same for every observer. Reversion in APS is equivalent to Alice’s hermitian (dagger) conjugation (\[STAdag\]) in STA$^{+},$ and it follows that the proper basis vectors $\mathbf{e}_{\mu}$ are real as seen by Alice:$$\mathbf{e}_{\mu}^{\dag}=\boldsymbol{\sigma}_{\mu}^{\dag A}=\gamma_{0}%
^{A}\boldsymbol{\tilde{\sigma}}_{\mu}^{A}\gamma_{0}^{A}=\mathbf{e}_{\mu}.$$
The equations (\[transf1\]) for Lorentz rotations in STA$^{+}$ are unchanged in APS except that, for Alice, $\boldsymbol{\sigma}_{\mu}^{A}$ is replaced by $\mathbf{e}_{\mu}~.$ The proper basis vectors $\boldsymbol{\sigma}_{\mu}^{B}$ used by Bob, given by (\[sigmatransf\]), are seen by Alice to be complex. This can be understood, as mentioned above, as the result of transforming the timelike planes swept out by persistent vectors in time, and it reflects the action of the six-parameter group $SL\left( 2,\mathbb{C}\right) $ of Lorentz rotors, which mix timelike and spacelike planes in spacetime. Such planes correspond respectively to vectors and bivectors in APS. The fact that Alice sees Bob’s frame as complex presents no physical problem because the definition of proper hermitian conjugation (\[STAdag\]) *depends on the observer* and ensures that each observer sees her own proper basis vectors as real [@Sob81a].
Indeed the mapping (\[e0biv\]) between APS and STA$^{+}\ $is observer dependent, and while Alice takes the proper basis vectors $\mathbf{e}_{k}$ of APS to be her timelike bivectors $\boldsymbol{\sigma}_{k}^{A}$ in STA$^{+},$ Bob equates them to his bivectors $\boldsymbol{\sigma}_{k}^{B}$. Just as under Alice’s hermitian conjugation, Bob’s proper basis vectors $\boldsymbol{\sigma
}_{k}^{B}$ are complex whereas hers are real, under Bob’s conjugation, it is Alice’s $\boldsymbol{\sigma}_{k}^{A}$ that are complex and his are real. By allowing both the mapping between APS and STA$^{+}$ and hermitian conjugation to vary with observer, this formulation of APS is thus able to treat all observers on an equal footing. The proper vector basis used by any observer in this formulation, like the bivector basis in STA to which it is equated, is absolute and distinct from the bases used by other observers who are in relative motion. We therefore refer to this formulation of APS as *absolute*.
The important relations in relativity, however, are not the way one observer sees another observer’s proper basis, but how the real, scalar values measured by one observer are related to those measured by another. Such relations are given by the transformations (\[transf2\]), and these suggest a relative formulation of APS.
Covariant Multiparavectors
--------------------------
While any collection of real vector components on the basis $\left\{
\mathbf{e}_{0},\mathbf{e}_{1},\mathbf{e}_{2},\mathbf{e}_{3}\right\} $ defines a real paravector, only some such collections transform together under Lorentz rotations as a spacetime vector. Those that do are said to be *covariant*. Furthermore, a spatial vector in APS may be part of a spacetime vector, whose components transform according to (\[transf2\]), or it may be part of a spacetime bivector, whose components transform differently. For example, the proper acceleration $du/d\tau$ of a point particle is a spatial vector in the commoving inertial frame, and it transforms as in (\[transf2\]) above. However, the electric field $\mathbf{E}$ is also a pure vector, but it transforms distinctly. In fact, it transforms as a persistent vector and may be seen by a different observer as having both real vector (electric) and imaginary vector or bivector (magnetic) parts. Its basis expansion in APS is simply$$\mathbf{E}=E_{A}^{k}\boldsymbol{\sigma}_{k}^{A}%$$ for Alice, and naively applying the arguments made for the position vector $r_{A},$ it would be seen by Bob (using his proper conjugation) to be real and to pick up a scalar part. It would remain real because Bob’s proper conjugation is different from Alice’s, and the difference exactly compensates the generation of the imaginary part.
This, however, is wrong. There is no meaning to a scalar part of the electromagnetic field, and Bob really sees a magnetic (imaginary) component to the field. What is missing is a recognition that the proper acceleration and the electric-field vector belong to different types of covariant objects and therefore transform differently under Lorentz rotations. The proper acceleration is part of a spacetime vector whereas the electric field is part of a spacetime bivector $\mathbf{F}$ representing the electromagnetic field.
Alice and Bob see different electric and magnetic components of the given electromagnetic field $\mathbf{F}$ because their hermitian conjugations are different. Thus,$$\mathbf{F}=\frac{1}{2}(\mathbf{F}+\mathbf{F}^{\dagger A})+\frac{1}%
{2}(\mathbf{F}-\mathbf{F}^{\dagger A})=\mathbf{E}_{A}+i\mathbf{B}_{A}%$$$$=\frac{1}{2}(\mathbf{F}+\mathbf{F}^{\dagger B})+\frac{1}{2}(\mathbf{F}%
-\mathbf{F}^{\dagger B})=\mathbf{E}_{B}+i\mathbf{B}_{B}%$$
To relate the measurables, that is the actual field components measured by Alice and Bob, we expand the covariant spacetime bivector of which it is a part in an inertial frame in STA and then express the result in a single proper basis of APS. For Alice the appropriate expansion is$$\begin{aligned}
\mathbf{F} & =\frac{1}{2}F_{A}^{\mu\nu}\gamma_{\mu\nu}^{A}=\frac{1}{4}%
F_{A}^{\mu\nu}\left( \gamma_{\mu0}^{A}\gamma_{0\nu}^{A}-\gamma_{\nu0}%
^{A}\gamma_{0\mu}^{A}\right) \\
& =\frac{1}{4}F_{A}^{\mu\nu}\left( \boldsymbol{\sigma}_{\mu}^{A}%
\boldsymbol{\tilde{\sigma}}_{v}^{A}-\boldsymbol{\sigma}_{\nu}^{A}%
\boldsymbol{\tilde{\sigma}}_{\mu}^{A}\right) .\end{aligned}$$ The same $\mathbf{F}$ can be expanded in Bob’s basis $\mathbf{F=}\frac{1}%
{4}F_{B}^{\mu\nu}\left( \boldsymbol{\sigma}_{\mu}^{B}\boldsymbol{\tilde
{\sigma}}_{\nu}^{B}\mathbf{-}\boldsymbol{\sigma}_{\nu}^{B}\boldsymbol{\tilde
{\sigma}}_{\mu}^{B}\right) $ so that the relation between the components seen by Alice and Bob is$$F_{B}^{\mu\nu}\left( \boldsymbol{\sigma}_{\mu}^{B}\boldsymbol{\tilde{\sigma}%
}_{\nu}^{B}\mathbf{-}\boldsymbol{\sigma}_{\nu}^{B}\boldsymbol{\tilde{\sigma}%
}_{\mu}^{B}\right) =F_{A}^{\mu\nu}\left( \boldsymbol{\sigma}_{\mu}%
^{A}\boldsymbol{\tilde{\sigma}}_{v}^{A}-\boldsymbol{\sigma}_{\nu}%
^{A}\boldsymbol{\tilde{\sigma}}_{\mu}^{A}\right) ,$$ where $\boldsymbol{\sigma}_{\mu}^{B}\boldsymbol{\tilde{\sigma}}_{\nu}%
^{B}-\boldsymbol{\sigma}_{\nu}^{B}\boldsymbol{\tilde{\sigma}}_{\mu}%
^{B}=L\left( \boldsymbol{\sigma}_{\mu}^{A}\boldsymbol{\tilde{\sigma}}_{v}%
^{A}-\boldsymbol{\sigma}_{\nu}^{A}\boldsymbol{\tilde{\sigma}}_{\mu}%
^{A}\right) \tilde{L}~.$ This leads to the relation$$F_{B}^{\mu\nu}\left( \boldsymbol{\sigma}_{\mu}^{A}\boldsymbol{\tilde{\sigma}%
}_{v}^{A}-\boldsymbol{\sigma}_{\nu}^{A}\boldsymbol{\tilde{\sigma}}_{\mu}%
^{A}\right) =4\tilde{L}\mathbf{F}L=\tilde{L}F_{A}^{\mu\nu}\left(
\boldsymbol{\sigma}_{\mu}^{A}\boldsymbol{\tilde{\sigma}}_{v}^{A}%
-\boldsymbol{\sigma}_{\nu}^{A}\boldsymbol{\tilde{\sigma}}_{\mu}^{A}\right) L$$ under a passive Lorentz rotation. With the identification (\[e0biv\]), in APS this becomes$$F_{B}^{\mu\nu}\left\langle \mathbf{e}_{\mu}\mathbf{\bar{e}}_{\nu}\right\rangle
_{V}=\tilde{L}F_{A}^{\mu\nu}\left\langle \mathbf{e}_{\mu}\mathbf{\bar{e}}%
_{\nu}\right\rangle _{V}L \label{passivebv}%$$
Thus, whereas every observer sees the proper acceleration or (for inertial observers sharing a common spacetime origin) the position as a real paravector in APS, the boosted electric field becomes a mixture of real and imaginary vectors, representing the electric and magnetic fields seen by a different observer. Vectors transform differently depending on what (if any) type of covariant object they belong to. The ability to identify the type of covariant object being transformed is essential in establishing the correct transformations between the measurables (scalar coefficients) for different observers, and the use of covariant objects such as spacetime vectors and bivectors is also important for simplifying relations and bringing out the geometry and relativistic symmetries of the problem. In STA, covariant objects are generally homogenous $k$-vectors. In APS, homogeneous $k$-vectors are generally not covariant. Instead, it is the paravectors and multiparavectors that provide the covariant formulation. Their utility is based on the relative formulation of APS, discussed in the next Section.
Relative APS
============
The formulation of relativity in absolute APS depends on the identification of the proper paravector basis and hermitian conjugation for each observer. Thus, Alice uses the basis $\left\{ \boldsymbol{\sigma}_{\mu}^{A}\right\} ,$ which is real under her conjugation operator $\dag A,$ whereas Bob uses a different proper basis $\left\{ \boldsymbol{\sigma}_{\mu}^{B}\right\} $ with the corresponding operator $\dag B.$ However, as seen above, to relate Alice’s and Bob’s measurables such as the scalar coefficients $r_{A}^{\mu}$ and $r_{B}^{\mu},$ we use transformations (\[transf2\]) that entail *only one* proper basis for both observers. In (\[transf2\]) we used Alice’s proper basis and conjugation, but we would obtain the same results by using Bob’s proper basis or, in fact, the proper basis of any inertial observer together with that observer’s proper conjugation.
To see this, replace $\boldsymbol{\sigma}_{\mu}^{A}$ by $\tilde{L}%
_{CA}\boldsymbol{\sigma}_{\mu}^{C}L_{CA}$ in order to express the result in Carol’s proper basis $\left\{ \boldsymbol{\sigma}_{\mu}^{C}\right\} .$ Here, $L_{CA}$ is the Lorentz rotor for the transformation from Alice to Carol. The passive transformation in (\[transf2\]), for example, becomes$$r_{A}\rightarrow r_{B}^{\mu}\tilde{L}_{CA}\boldsymbol{\sigma}_{\mu}^{C}%
L_{CA}=r_{A}^{\nu}\tilde{L}\tilde{L}_{CA}\boldsymbol{\sigma}_{\nu}^{C}%
L_{CA}\tilde{L}^{\dag A}\ \label{relative}%$$ which since $L_{CA}\tilde{L}_{CA}=1$ and $\gamma_{0}^{C}=L_{CA}\gamma_{0}%
^{A}\tilde{L}_{CA},$ is equivalent to$$\begin{aligned}
r_{B}^{\mu}\boldsymbol{\sigma}_{\mu}^{C} & =r_{A}^{\nu}L_{CA}\tilde{L}%
\tilde{L}_{CA}\boldsymbol{\sigma}_{\nu}^{C}L_{CA}\tilde{L}^{\dag A}\tilde
{L}_{CA}\nonumber\\
& =r_{A}^{\nu}L_{CA}\tilde{L}\tilde{L}_{CA}\boldsymbol{\sigma}_{\nu}%
^{C}\tilde{L}_{AC}\gamma_{0}^{A}L\gamma_{0}^{A}\tilde{L}_{CA}\nonumber\\
& =r_{A}^{\nu}L_{CA}\tilde{L}\tilde{L}_{CA}\boldsymbol{\sigma}_{\nu}%
^{C}\gamma_{0}^{C}L_{CA}L\tilde{L}_{CA}\gamma_{0}^{C}\nonumber\\
& =r_{A}^{\nu}\tilde{L}^{\prime}\boldsymbol{\sigma}_{\nu}^{C}\tilde
{L}^{\prime\dag C}, \label{CarolP}%\end{aligned}$$ where $L^{\prime}=L_{CA}L\tilde{L}_{CA}$ is the transformation from Alice to Bob as seen by Carol [@Bay99]. In the special case that Carol shares Alice’s inertial frame, $L_{CA}=1,$ whereas if Carol and Bob share the same frame, $L_{CA}=L.$
The key point is that the transformation (\[CarolP\]) of measurables has exactly the same form as the passive transformation in (\[transf2\]). The same result can be readily verified for the other transformations (\[transf2\]) of measurables. The transformations are the same no matter which inertial observer is used for the mapping of APS to STA$^{+}$ and for the definition of hermitian conjugation in STA$^{+}.$ Although the relation of *absolute APS* to STA$^{+}$ requires separate mappings and a separate hermitian conjugation for each inertial observer, this may be considered an artifact of the absolute approach in STA. Within *relative APS* it makes no difference which inertial observer is chosen; only one proper conjugation and one proper (relative) basis $\left\{ \mathbf{e}_{\mu}\right\} $ are needed, and these can be associated with any inertial observer. The result can also be described as choosing one absolute frame in STA as the *observer frame*. Since all inertial frames are equivalent, any inertial observer can be assigned to this frame. Relative APS thus incorporates the basic principle of relativity that all inertial observers are equivalent and that only the relative motion and orientation of frames is physically significant.
When we organize the Lorentz transformations as in (\[transf2\]) to compute the measurables, such as the scalar components of a paravector $r_{A}$, we find the transformations of a *spacetime vector,* even in the special case when $r_{A}=\mathbf{e}_{\mu}.$ We need to understand what it means for $\mathbf{e}_{\mu}=\boldsymbol{\sigma}_{\mu}^{A}=\gamma_{\mu}^{A}\gamma_{0}%
^{A},$ a bivector in STA$^{+},$ to transform as a spacetime vector rather than as a bivector as in (\[sigmatransf\]). The paravector transformation of $\mathbf{e}_{\mu}$ can be expressed in STA as$$\mathbf{e}_{\mu}\rightarrow L\mathbf{e}_{\mu}L^{\dag A}=L\gamma_{\mu}%
^{A}\gamma_{0}^{A}L^{\dag A}=\left( L\gamma_{\mu}^{A}\tilde{L}\right)
\gamma_{0}^{A}~, \label{pvtransf}%$$ where we have employed the definition (\[STAdag\]) of Alice’s proper conjugation. The paravector basis elements $\mathbf{e}_{\mu}$ are defined to be even elements of STA, and each involves the product of vectors from two absolute STA frames. The paravector transformation (\[pvtransf\]) transforms only one of the frames. The result helps to clarify the roles of the two factors: the first factor $\gamma_{\mu}^{A}$ gives the STA frame in which the observed object property is expanded, and the second factor $\gamma_{0}^{A}$ gives the absolute proper velocity of the observer. The two factors generally transform differently, as in the paravector transformation (\[pvtransf\]). The product of the two frame vectors gives the first relative to the second. An observer will normally choose a proper basis in which the object and observer frames are the same, but this is not necessary; she may also choose an object frame in relative motion. In all cases, in the relative version of APS the paravector basis elements are relative to the observer.
To obtain the transformation for spacetime bivectors from that for paravectors, (\[pvtransf\]), we form a product of two paravectors and transform it:$$\mathbf{e}_{\mu}=\mathbf{e}_{\mu}\mathbf{\bar{e}}_{0}\rightarrow
L\mathbf{e}_{\mu}L^{\dag}\overline{\left( L\mathbf{e}_{0}L^{\dag}\right)
}=L\mathbf{e}_{\mu}\mathbf{\bar{e}}_{0}\bar{L}~.$$ This is equivalent to the spacetime bivector transformation (\[sigmatransf\]) of $\mathbf{\sigma}_{\mu}^{A}~.$ Of course $\mathbf{e}_{0}=1,$ but it is often convenient to add factors of $\mathbf{e}_{0}$ or $\mathbf{\bar{e}}_{0}$ to display the correct covariant behavior.
To summarize, in relative APS, spacetime vectors are covariantly represented by real paravectors. For example, the energy-momentum paravector of a particle is $p=mu=E+\mathbf{p~.}$ Multiparavectors of higher grade can be formed to represent other covariant geometrical objects, namely spacetime planes, hypersurfaces, and volumes. Simple Lorentz rotations of paravectors have the form (\[Lrotp\]) where $L$ is the exponential of a biparavector representing the plane of rotation. Since as discussed above, the spacetime vectors represented by paravectors in APS are all *relative* to the observer, a single Lorentz rotation can equally well represent an active transformation of the observed spacetime vector, the inverse passive transformation of the observer, or some combination thereof. In the passive case, the spacetime vector $p$ can be treated as invariant with respect to one observer, say Alice, and the transformation derived by expressing Bob’s frame relative to Alice. Thus$$p_{A}=p_{A}^{\mu}\mathbf{e}_{\mu}=p_{B}^{\nu}\mathbf{u}_{\nu}~,$$ where $\mathbf{u}_{\mu}=L\mathbf{e}_{\mu}L^{\dag}$ is Bob’s frame as seen by Alice. This gives $p_{B}^{\nu}=p_{A}^{\mu}\left\langle \mathbf{e}_{\mu
}L\mathbf{\bar{e}}^{\nu}L^{\dag}\right\rangle _{S},$ which is exactly the same relation as found from the transformation (\[Lrotp\]).
Multiparavectors of grades 0 through 4 form linear subspaces of the algebra. Grade-0 paravectors are the same real scalars as grade-0 vectors, but covariantly they model spacetime scalars, which are invariant under Lorentz transformations. Grade-1 paravectors model spacetime vectors and form the four-dimensional paravector space that is also the direct sum of scalar and vector spaces. The biparavectors model spacetime planes and form a six-dimensional subspace from the direct sum of vector and bivector spaces. The four-dimensional subspace of triparavectors, which model hypersurfaces in spacetime, is the direct sum of bivector and trivector spaces. Finally, the paravector volume element, which models the spacetime pseudoscalar, also serves as the vector volume element:$$\mathbf{e}_{0}\mathbf{\bar{e}}_{1}\mathbf{e}_{2}\mathbf{\bar{e}}%
_{3}\mathbf{=e}_{1}\mathbf{e}_{2}\mathbf{e}_{3}=i~.$$ The split of any paravector, biparavector or triparavector into its multivector parts is a space/time split in APS.
Discussion
==========
The measurement of a physical vector or multivector depends both on the vectors of the physical system to be measured and on the commoving frame of the observer. STA represents both the object to be measured and the observer in terms of abstract absolute frames, but the *measurables* are components of the object vectors on the basis vectors of the observer frame. These measurables appear as even elements of STA, involving products of vectors of the observed system with the frame vectors of the observer. If we consider only a single observer, we can map the elements of STA$^{+}$ onto those of APS in such a way that the basis elements $\mathbf{e}_{\mu}$ of APS are just proper relative basis elements $\boldsymbol{\sigma}_{\mu}=\gamma
_{\mu}\gamma_{0}$ of the observer. Persistent spatial vectors of APS then correspond to timelike bivectors of STA, and hermitian conjugation, which gives reversion in APS, is defined using the proper velocity of the observer. The result is that paravectors in APS are always given relative to the observer. Only such relative paravectors and their products can be measured, and this is precisely what is computed in APS. The additional flexibility provided by STA, namely to compute absolute objects and observer frames independently, although at times conceptually appealing, is never needed in physical measurements. This is the reason that APS can represent any physical process as well as STA.
When there are two or more observers, they generally determine distinct values of the measurables, and an important goal of any relativistic formalism is to relate such values. In the absolute version of APS, the proper paravector basis $\left\{ \mathbf{e}_{\mu}\right\} $, as well as the proper hermitian conjugation, depends on the observer and is identified with an absolute frame in STA$^{+}$. It is the observer’s own proper conjugation operator that tells her which parts of spacetime bivectors are her vectors and which are her bivectors. Lorentz boosts between different observers mix spatial vectors and bivectors, all of which lie in the six-dimensional space of spacetime bivectors of STA$^{+}$. Such mixing explains why boosts of persistent real vectors, such as the electric field, induce an observed magnetic field. A similar transformation applies to each observer’s proper basis vectors when these are treated as persistent vectors. While the observer dependence of hermitian conjugation means that each observer sees her own basis vectors as real, she may find another observer’s persistent basis vectors to be complex.
However, a different transformation applies to spacetime vectors such as acceleration. There the separate conjugations applied by the two observers ensure that the pure spatial vector seen by one observer remains real for the other and that a scalar part is generated. Thus, two distinct boost transformations for vectors ensue, one for persistent vectors such as the electric field giving a complex vector, and the other for instantaneous spacetime vectors such as the proper acceleration giving a real paravector. The covariant nature of the object must be recognized in order to know which transformation to apply.
To establish explicit transformations of the measurable coefficients, we can write the transformations between observers in terms of a proper basis of a single observer. We then find that the transformations do not depend on which observer’s proper basis and conjugation are used. A single proper basis and conjugation (reversion) operation in APS can in fact be used to relate all inertial observers, even though they have different absolute frames in STA. This leads to the relative APS approach, in which all inertial observers share a single conjugation and a single proper basis. It is consistent with the principle of relativity that all inertial frames are equivalent and only relative motion and orientation of frames matters. The different covariant behavior of the proper acceleration and the electric field corresponds to the difference between spacetime vectors and spacetime bivectors, and in relative APS these are represented by real paravectors and biparavectors, respectively. APS can be interpreted both in terms of spatial vectors and their products, and in covariant terms, with Minkowski vectors of spacetime. The absolute version of APS emphasizes the first and the relative version the second. In both cases, one clearly identifies the elements of APS with measurables in STA$^{+}$, and although in the presence of more than one observer the two versions invoke different mappings to STA$^{+},$ we have shown them both to be two different approaches within a single coherent geometric algebra, namely APS.
The ability of APS to model relativistic processes as faithfully as STA, even though as a vector space it has only half the number of independent elements, is due mainly to the additional structure required by the absolute-frame representation of relativity in STA. APS, in contrast, does not model non-observable absolute frames but concentrates instead on *measurable* properties relative ** to an observer. A characteristic of the APS approach that also contributes to its efficiency is the double role of vector grades. Thus, a scalar might be a Lorentz invariant or the time component of a paravector, and a vector might be part of a paravector or part of a biparavector. In this regard, APS models relativistic quantities as humans commonly do. For example, the mass of a particle is both the Lorentz invariant length of its momentum and the time component (in units with $c=1$) of the momentum in its rest frame. This is naturally expressed in APS, where $m=m\mathbf{e}_{0}$ is the same object in both roles, whereas in STA the roles are played by distinct elements: $m\neq m\gamma_{0}~.$ This feature may be considered an extension of a basic attraction of working with geometric algebras instead of with separate vector spaces: in the algebra there is one zero element, and one does not need to distinguish the zero scalar from the zero vector or zero bivector.
Acknowledgment {#acknowledgment .unnumbered}
--------------
One of us (WEB) thanks the Natural Sciences and Engineering Research Council of Canada for support of research.
[99]{}
D. Hestenes, *Spacetime Algebra*, Gordon and Breach, New York 1966.
D. Hestenes and G. Sobczyk, *Clifford Algebra to Geometric Calculus: A Unified Language for Mathematics and Physics*, D. Reidel, Dordrecht, 1984.
D. Hestenes, *New Foundations for Classical Mechanics,* 2nd edn., Kluwer Academic, Dordrecht, 1999.
D. Hestenes, *Am. J. Phys.* **71**:104–121, 2003; **71**:691–714, 2003.
W. E. Baylis, editor, *Clifford (Geometric) Algebra with Applications to Physics, Mathematics, and Engineering*, Birkhäuser, Boston 1996.
P. Lounesto, *Clifford Algebras and Spinors*, second edition, Cambridge University Press, Cambridge (UK) 2001.
R. Abłamowicz and G. Sobczyk, eds., *Lectures on Clifford Geometric Algebras,* Birkhäuser, Boston, 2003.
J. Snygg, *Clifford Algebra, a Computational Tool for Physicists,* Oxford U. Press, Oxford, 1997.
K. Gürlebeck and W. Sprössig, *Quaternions and Clifford Calculus for Physicists and Engineers,* J. Wiley and Sons, New York, 1997.
C. Doran and A. Lasenby, *Geometric Algebra for Physicists,* Cambridge University Press, Cambridge (UK), 2003.
D. Hestenes, *J. Math. Phys.* **15**, 1768–1777 (1974).
W. E. Baylis, *Electrodynamics: A Modern Geometric Approach*, Birkhäuser, Boston, 1999.
W. E. Baylis and G. Jones, *J. Phys. A (Math Gen)* **22**, 1–16; 17–29 (1989).
W. E. Baylis, *Am. J. Phys.* **48**, 918–925 (1980).
G. Sobczyk, *Phys. Letters A* :45–48 Jul (1981).
L. Silberstein, Phil. Mag. **23**, 790–809 (1912).
A. W. Conway, Phil. Mag. **24**, 208 (1912).
G. Sobczyk, *Acta Phys. Pol.*, Vol.**B12**, 509-521 (1981).
[^1]: email address: [email protected]
[^2]: email address: [email protected]
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Decades of improving data and extensive theoretical research have led to a popular model of gamma-ray bursts. According to this model, a catastrophic event in a stellar system results in the formation of a compact central engine, which releases a fraction of a solar rest-mass energy within seconds in the form of ultra-relativistic jets. Dissipation of the jets energy leads first to prompt gamma-ray emission and later to a long lasting afterglow. Here I summarize the introduction that I gave to the debate “where do we stand?" in the conference “The Shocking Universe" held in Venice. This is a very brief summary of my view of the facts that we are (almost) certain about, models that are popular but may need rethinking, and main open questions.'
author:
- Ehud Nakar
title: 'What do we know, what do we think we know and what do we not know about Gamma-Ray Bursts'
---
introduction
============
Gamma-Ray Burst (GRB) observations during the 1990’s lead to the impression that a GRB starts complex and becomes simpler to model with time. The popular view was that a progenitor stellar system goes through some catastrophic event, whose nature is unknown, leading to the formation of a compact central engine. Somehow this central engine deposits a fraction of a percent of the gravitational energy of the system in the form of kinetic energy of a minute fraction of its mass, thereby launching a collimated ultra-relativistic outflow. Internal shocks (or another form of internal interaction) within the outflow convert some of the outflow energy into accelerated electrons and magnetic fields, which produce the prompt emission that we observe, most likely via synchrotron radiation. Later, an interaction of the outflow with the surrounding medium drives a self-similar decelerating blast wave that accelerates electrons and amplifies the ambient magnetic field to produce the afterglow. The afterglow was considered relatively simple to model and understand, since all the complex initial conditions are washed out and only the initial energy and geometry of the outflow determine the blast wave evolution.
Observations during the last decade have led on one hand to a remarkable breakthrough in the identification of the progenitor system, and on the other hand to the realization that even the afterglow physics is much more complex than suspected. Theoretical modeling of the different physical processes improved and became more detailed and accurate, but also more complicated. Additionally, new ingredients were invoked to explain various specific observations of specific bursts, significantly increasing the model’s degrees of freedom, sometimes to the point where the models do not provide useful predictions anymore.
The purpose of the open discussion that I led in the conference was to take a step back and evaluate where do we stand in the grand picture. I presented to the audience three questions:\
- What do you know with a very high confidence (almost as facts) about the physics of GRBs? What observations are the base for your confidence?\
- Which of the models (for progenitors, prompt emission, afterglow, etc.), if any, you think that are probably correct?\
- About which elements of GRB physics you think that we have almost no clue? What do we need in order to learn about these?\
Unfortunately no detailed notes were taken during the discussion so here I will summarize only the introduction that I gave, which includes my short answers to these questions. I divide my discussion into five different stages of the explosion: progenitor, central engine, outflow properties, prompt emission and afterglow. For each, I give (a very partial) answer to the three questions. I touch only the physics of the explosions themselves without considering the study of the large scale environment (e.g., host galaxies). I discuss long and short GRBs separately only in the “progenitor" section. The reason is that as far as we know the rest of the stages may be (and possibly are) rather similar. Finally, one of the comments during the discussion (by Chris Fryer) was that we are much better at ruling out models than at proposing viable ones. Nevertheless, I discuss here only models which may describe the physics properly, without mentioning those that were ruled out.
Given the very broad scope of the discussion and the limited space of this proceeding, I regretfully cannot give the appropriate credit to the large number of observational and theoretical works on which my views are based. Note that some of the ideas that are discussed here were also presented by Maxim Lyutikov in his presentation, which he summarizes in this proceeding.
Progenitor
==========
[*What do we confidently know?*]{}\
First, we know that there are at least two types of progenitors, those of (the majority of) long GRBs and those of (the majority of) short GRBs. This is based on three major observed differences between the two GRB types. The host galaxies are very different. The intrinsic redshift distribution is very different. An association with a supernova (SN) was established for almost any nearby long GRB, while a SN association was ruled out to any nearby short GRB.
We know much more about the progenitors of long GRBs. First, the progenitor system includes a very massive star. This knowledge is based on the association of specific nearby GRBs with SNe, on the high specific star formation rate of the host galaxies and on the location of the bursts within the most star forming regions of their hosts. Second, [*at least some*]{} progenitors produce broad line Ic SN (or SN like emission) within about $\pm 1$ day of the GRB. The evidence for this is not as straightforward as commonly assumed, and although the evidence are quite convincing (see below), future may hold surprises. Most of the nearby GRB-SNe events are not necessarily genuine GRBs (and maybe not even genuine SNe, in the sense that the driver of the explosion may be different than that of typical core-collapse SNe). For example GRBs 980425 (SN 1998bw), 031203 (SN 2003lw), 060218 (SN 2006aj) and the very recent GRB 100316D (SN2010bh) all show a single pulse of soft gamma-rays which contains about $10^{48}$ erg. These low luminosity bursts are very different than typical cosmological GRBs and their $\gamma$-ray emission is almost certainly produced by different physical mechanism. GRB 030329/SN 2003dh was the first example where a SN was directly associated with a GRB that is more similar to cosmological ones. At z=0.16, GRB 030329 released an isotropic equivalent energy of $\sim 10^{52}$ erg in two pulses of $\gamma$-rays. Until recently this event was the only direct link between cosmological GRBs and Ib/c SNe. Recently, Massimo Della Valle reported in an Astronomical Telegram (No. 1602) of the detection of SN 2008hw associated with GRB 081007 at z=0.53 (isotropic equivalent energy of $\sim 10^{51}$ erg in $\gamma$-rays), fortifying this connection. Apart for these direct SN/GRB associations, there are also about a dozen cases where a “red bump", which is presumably contributed by the SN light, is evident in the late time afterglow of cosmological GRBs, further supporting the connection.
We do not know much about short GRB progenitors. All we know at a high level of confidence is that they are different than long GRBs and that at least some of the progenitor systems do not include massive stars.
[*Popular models that need confirmation*]{}\
The most natural picture that explains why long GRB progenitors are associated with massive stars and why they explode within $\pm 1$ day of a SN is that the GRB and the SN are produced simultaneously following the collapse of a massive stellar core. While this is probably true, there is no direct evidence that this is indeed the case. Moreover, SNe modelers typically ignore the simultaneous launch of GRB relativistic jets, while GRB central engine modelers typically ignore the production of a SN. There is currently no model that incorporates in detail the basic ingredients of both and the interplay between them.
The most popular model for short GRB progenitors is neutron start-neutron start or neutron start-black hole merger. But, while being a very attractive model (especially for those interested in gravitational wave signals), which gets a passing grade in the comparison with all currently available observations, there is still no conclusive (or even strong) evidence that this is indeed the correct progenitor model.
[*Some of the main open questions and how can we answer them*]{}\
Three important open questions about long GRB progenitors are: What is the role (if any) of various progenitor system properties (metalicity, binarity, mass, etc.)? Are all long GRBs associated with Ib/c SN? and is there a continuous transition from regular SNe through low-luminosity GRB/SNe to cosmological GRBs? The answer to all these questions needs mostly an accumulation of more observations, where I want to stress the importance of searching for SNe signal in an intermediate redshift ($z \sim 0.5-1$) GRBs. In additional to improved observations, the key to third question is theoretical modeling of the link between GRBs and SNe.
The main question about short GRB progenitors is simply what are they. Obviously, the ultimate compact binary merger model test will be via the detection or non-detection of gravitational waves from nearby short GRBs. But before we get to this stage, there is still much that can be done. Here the focus should be to constrain the environment and redshift distribution of the bursts. For this we need a controlled large sample of bursts with redshift, host type and location within the host. Additionally, deep limits on dark X-ray afterglows can constrain the circum-burst density. Finally, we are in a great need of a reliable classification scheme that can tell the difference between (physically) short and long GRBs.
Central engine
==============
[*What do we confidently know?*]{}\
There is not much that we know with high confidence about the central engine. We basically know that it is a compact ($<10^7$ cm) object that converts a fraction of the system’s gravitational energy into collimated relativistic jets continuously over a duration that is much longer than its dynamical time.
[*Popular models that need confirmation*]{}\
There are two popular models, with no conclusive evidence that points to one of them (or that rules them out). The first, much more popular model, is an accretion near the neutrino Eddington limit on a stellar black hole. The main advantages of this model are that similar accretion (although at much lower rates) is the known engine of active galactic nuclei (AGNs) and $\mu$-quasars, and that it can, in principle, release $10^{53}$ erg in the form of a relativistic outflow. The main shortcoming of the model is that there are many unknowns on the way that this engine works and there is no clear idea why the specific properties of GRB outflows are generated. Second, especially in short bursts, it is hard to explain late engine activity.
The second model is a milli-sec proto-magnetar. Here, in the typical version of the model, the gravitational energy is first converted into rotational energy of the newly born magnetar and then released in the form of a relativistic outflow. The main advantages of this model are that, once the engine is formed, the physics of the outflow launching is better understood and that it can explain late engine activity rather naturally. A severe disadvantage of this model is that it has a limited energy of $\approx 5 \cdot 10^{52}$ erg.
[*Some of the main open questions and how can we answer them*]{}\
As one can understand from the discussion above, there are many open questions, including the most basic ones, such as what is the physical outline of the engine. However proceeding in the quest to understand the engine is very difficult, mostly because it is completely hidden from us today. The reason is that the observed electromagnetic radiation is produce above or near the photosphere which is larger than the engine by many orders of magnitude. An observable test that can be carried out today, as pointed out in the discussion by Dale Frail, is the search for ultra-energetic bursts ($\sim 10^{53}$ erg), which if found will be hard to explain by the proto-magnetar model.
A breakthrough in the understanding of the engine is expected if we will be able to probe the engine directly, e.g., via gravitational waves. Without such ground breaking observations it seems that the major effort should be invested in theoretical and numerical modeling, in order to better understand the models and to come up with testable predictions. A breakthrough in the understanding of AGN and/or $\mu$-quasar engines may help here as well.
Outflow properties
==================
[*What do we confidently know?*]{}\
We know that the outflow is ultra-relativistic. The main evidence for relativistic motion comes from the requirement for low $\gamma\gamma\rightarrow e^-e^+$ optical depth. In most bursts, where photons above $\sim$MeV are not observed (most likely due to insensitivity of the detectors), the outflow Lorentz factor, $\Gamma$, must be larger than about $30$. Although extrapolation of the observed spectrum to high ($\gg$MeV) frequencies suggest that the true opacity limit is a few hundreds. There are also about half a dozen bursts where observations of GeV photons set the lower limit at about $1000$ (note that opacity limits may vary if the high energy emission is not produced at the same place as the low energy emission, e.g. prompt vs afterglow). There is no robust upper limit to the outflow Lorentz factor (although afterglow theory suggests that it cannot be much larger than $\approx 1000$). The most robust confirmation that the outflow is indeed relativistic (although the lower limit is $\Gamma
> 5$), is the measurement of the size of the radio afterglow of GRB 030329. The rather low lower limit is expected since the measurement takes place at a time that, according to afterglow theory, the outflow was significantly decelerated.
We are also quite certain that at least some long GRB outflows are narrowly beamed (we do not know much about the collimation of short GRBs). The list of arguments for collimation are composed of several independent, strong, yet not conclusive, evidence. These are energy requirements (without beaming some GRBs would release more than $10^{55}$ erg), afterglow jet breaks and radio calorimetry.
Based on the constraints on the beaming we can deduce that the total energy (corrected for beaming) carried by the outflow of long GRBs is about $10^{50}-10^{52}$ erg and that the luminosity is $\sim
10^{49}-10^{51}$ erg/s. In the few cases where radio calorimetry can be done, an energy $10^{51}-10^{52}$ erg is measured. In the case of short GRBs we know only that the [*isotropic equivalent*]{} energy emitted in $\gamma$-rays is about $10^{49}-10^{52}$ erg and the [*isotropic equivalent luminosity*]{} is about $10^{50}-10^{52}$ erg/s.
[*Some of the main open questions and how can we answer them*]{}\
What is the actual Lorentz factor of the outflow? We have only robust lower limits but the true value of the Lorentz factor is unknown yet. It may be found if $\gamma\gamma$ attenuation will be identified in the spectra, hopefully by Fermi.
What is the detailed outflow geometry? Is it a top hat or does the energy falls gradually with the angle from the axis? Is it patchy or not? These questions are typically attacked using afterglow observations, since during the prompt emission we observe only a tiny patch of the outflow. I think that the progress here will come from the observational side. The effort to resolve these questions by modeling of afterglow jet-breaks requires many more simultaneous radio-optical-X-ray afterglow observations while radio calorimetry will hopefully improve significantly by the EVLA. Additional observational tools that may be available in the future are a statistically large sample of orphan afterglows and detailed polarization measurements.
A long standing open question, which is crucial for the understanding of both the central engine and the prompt emission, is what component of the outflow is dominant energetically. Is it baryonic, Poynting-flux or maybe pairs? A prediction of the baryonic outflow model was that Swift will observe many bursts with bright optical flashes during the early afterglow. Swift detected only a few optical flashes, thereby supporting a non-Baryonic outflow. But, as it is often the case, non-detection does not provide conclusive evidence. Currently, I do not have an idea of future observations that may bring us closer to identify the outflow composition. Part of the reason is that only the poorly understood prompt emission and early afterglow are likely to be affected by the outflow composition. Additionally, no specific model of the process that converts Poynting-flux into the observed emission is available for comparison to observations. Thus, a theoretical progress on this front will be helpful.
Prompt emission
===============
[*What do we confidently know?*]{}\
We know almost nothing for certain. The most robust statements, which may prove to be wrong in the future, are as follows. Based on opacity arguments, the emission originates at radius $> 10^{11-12}$ cm while interaction with the external medium dictates that it takes place at a radius $<10^{16-17}$ cm. Energy requirements and afterglow modeling implies that the prompt emission is very efficient, $\gtrsim 10\%$, in converting the outflow energy into sub-MeV $\gamma$-rays. Finally, high variability points strongly towards a dissipation mechanism that is internal to the flow, although there are suggested models of pointing flux dominated outflows, where the internal dissipation is triggered by interaction with the external medium.
[*Popular models that need confirmation*]{}\
The most popular model is the internal shock model, where the outflow dissipation is done by hydrodynamical shocks between different portions of the flow. The main advantage of the model is that it naturally explains the high variability. But the disadvantages of the model made it less popular in the last several years. The main one is the limited efficiency. Despite of a large theoretical effort, there is no consensus on a natural way to bypass this problem. Additionally, there are many properties of the prompt emission that are not explained naturally in the internal shock model.
[*Some of the main open questions and how can we answer them*]{}\
Despite of the impressive set of prompt emission observations, the entire topic is one big question mark. With thousands of superb light curves and spectra we cannot confidently identify even the dominant radiation process, not to speak on nailing down the dissipation process or clearly find out the origin of a large number of unexplained detailed properties of the emission (e.g., energy dependent pulse shapes, $E_p$ distribution, various correlations etc.).
The main reason for the difficulty is that the high energy power-law spectrum suggests that the emission is generated above the photosphere, where the main candidate is synchrotron radiation by ultra-relativistic electrons. However, the steep low energy spectrum is very difficult to explain with synchrotron and it suggests a thermal component at the base, which is modified above or near the photosphere by inverse Compton. However the smooth transition from the low to high energy spectrum implies that for this model to work the electrons must be at most mildly relativistic and that they should carry a comparable energy to the radiation. This suggests that a very efficient dissipation of the outflow energy takes place just below the photosphere. Otherwise, if the energy is dissipated too deep it is lost to adiabatic expansion, while if it is dissipated high above the photosphere the thermal component (if it exists) does not interact strongly with the electrons. So far there is no consensus on a mechanism that naturally provides these requirements (although lately some interesting new candidates where suggested), and that can naturally explain a large fraction of the observed features.
Observationally we already have great $10$ keV - $1$ MeV data set and a large number of optical - X-ray observations that coincide with the end of the prompt emission of long GRBs. The only hope that I can see for observational breakthrough in the near future is by Fermi, which already provided useful information that strongly disfavor inverse Compton by ultra-relativistic electrons as the source of the prompt emission. But, so far, Fermi observations did not pointed towards the correct model, while they did raise up new open questions. If such breakthrough will not take place, then a theoretical study of the existing models that are still viable and development of new ones, are the most promising way to understand the prompt emission.
Afterglow
=========
[*What do we confidently know?*]{}\
The late afterglow is generated during (and almost certainly by) the interaction with the circum burst medium. This statement is robustly based on the decelerated expansion of the afterglow image of GRB 030329. It is also strongly supported by the light curve and spectral afterglow evolution, which show a continuous power-law decay (in both the flux and peak frequency) and variability time scales that increase with time.
[*Popular models that need confirmation?*]{}\
By far, the most popular afterglow model is the external forward shock model, and in my view it is very likely that the afterglow emission, at least starting a few hours after the burst, is generated by forward external shock. The major success of this model is that with a very simple parametrization (of only a few free parameters) it encompasses the gross observed afterglow properties over eight orders of magnitude in frequency and four orders of magnitude in time. However, an examination of the fine details shows that the model is far from being complete. The simple model is not compatible with the detailed observations of many bursts (e.g., the exact spectral and temporal power-law indices). There is a set of observations (especially during the first $10{^3}-10^{4}$ s), which are difficult to explain by forward shock emission even by significantly complicating the simple model. Some examples are X-ray flares, X-ray plateaus and some chromatic breaks. Different extensions of the basic model are invoked to explain the observed deviation from the model. However, often these are tailored on a burst to burst basis, and most dangerously, in some cases modelers provide their models with so much freedom so they lose their predictability and with it their usefulness. Finally, the simple model simply parameterizes the unknown microphysics, which may have a very complicated behavior, by three constants that has (almost) no theoretical first principle predictions.
[*Some of the main open questions and how can we answer them*]{}\
Is the external shock model correct? I expect that the strongest evidence on this point will come from unique events, such as GRB 030329. Until these are observed, more detailed multi-wavelength (radio - X-ray) observations will be helpful. There are many optical (and some radio) afterglows without detailed X-ray coverage before the Swift era. Now, that Swift-XRT provides impressive X-ray light-curves, there are too few optical light curves and almost no radio afterglow detections.
What is the cause of the X-ray plateau and chromatic breaks? These are among the most surprising Swift observations, which still waits for a theoretical breakthrough, whose direction I cannot predict.
Plasma microphysics is among the toughest topics of any astrophysical environment and GRBs are not different. However, if the afterglow is generated by an external shock, then this is a relatively clean system where the initial conditions are rather simple: an ultra-relativistic collisionless shock that propagates into a very weakly magnetized electron-proton plasma. Understanding the microphysical processes that take place in such system will requires an extensive numerical work accompanied by careful theoretical analysis. This field made an impressive progress in recent years and I hope that first principle theory of GRB microphysics will be developed during the next decade. Detailed GeV observations may also prove useful to farther constrain the microphysical properties of the radiating plasma.
Conclusions
===========
A quick scan of the GRB “facts” listed here shows that among the observations on which we base these facts, the role of single unique events (e.g., GRB 030329) is similar to, or maybe even larger than, that of the large burst samples. It also shows that we are still ignorant even about fundamental ingredients of the explosions. Even afterglow modeling, which before the launch of Swift was thought by many to be satisfactory, is now being reconsidered. Nevertheless, if I compare today’s knowledge to what we knew a decade ago the progress is impressive. I hope that the coming decade will be as eventful as the passing one and that ten years from now we will be able to say that our understanding has improved by at least as much.
I am in debt of all the conference participants which contributed to the debate. I am grateful to Guido Chincarini for the worm and welcoming hospitality during the conference and to Tsvi Piran and Carles Badenes for useful comments on the manuscript.
| {
"pile_set_name": "ArXiv"
} |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.