Randomly twisted transfer operators and singular values statistics.

Frédéric Naud · 2026

Read Randomly twisted transfer operators and singular values statistics. by Frédéric Naud (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-23530.txt.

Introduction

Naud studies singular values of transfer operators twisted by large random permutation matrices. In the large N limit he obtains a Weyl law for those singular values, valid asymptotically almost surely with rapid decay. He also extends the polynomial method to an infinite-dimensional setting, which yields a smooth probabilistic Weyl law for singular values.

Catalog record

CREATORFrédéric Naud
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE97147
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.23530

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Randomly twisted transfer operators and singular values statistics. Frédéric Naud Abstract . In this paper, we investigate the singular values of a natural family of transfer operators twisted by large random permutation matrices. In the large N N limit, we obtain a Weyl law for its singular values, valid asymptotically almost surely with rapid decay. We also extend the so-called polynomial method to an infinite dimensional setting which implies a ”smooth” probabilistic Weyl law for singular values. 2000 Mathematics Subject Classification 37C30, 37D20 1. Setting and introduction Transfer operators are ubiquitous objects in the study of hyperbolic dynamical systems and their asymptotic statistics. In the early works of the pioneers, they arised through Markov codings and where used as a basic tool to construct invariant measures and establish decay of correlations, see the classics [ 11 , 2 , 27 ] . For analytic dynamical systems, general transfer operators can be described as weighted sums of composition operators acting on suitable holomorphic function spaces. They were pioneered by David Ruelle in [ 32 ] . More recently, we mention the fairly general papers of Bandtlow-Jenkinson [ 4 , 5 ] , see also [ 6 ] , for sharp estimates on the eigenvalues. Let us be more specific on the models we want to study here. Let \Omega_{0} be a bounded, simply connected, bounded open set in \mathbb{C} . We assume that the boundary \partial\Omega_{0} is at least C^{2} regular. Let \gamma_{1},\ldots,\gamma_{d}:\Omega_{0}\rightarrow\Omega_{0} be holomorphic mappings, and G:\Omega_{0}\rightarrow\mathbb{C} be a weight function. We assume that the following holds. (1) Each map \gamma_{j} extends continuously to \overline{\Omega_{0}} . (2) For all j=1,\ldots,d , we have \overline{\gamma_{j}(\Omega_{0})}\subset\Omega_{0} . (3) The weight G extends continuously to \overline{\Omega_{0}} and is holomorphic on \Omega_{0} . In addition, we pick U_{1},\ldots,U_{d}\in U_{N}(\mathbb{C}) to be permutation matrices , chosen uniformly and independently in the symetric group \mathcal{S}_{N} . The twisted transfer operator {\mathscr{L}}_{N} acts on holomorphic, vector valued functions F:\Omega_{0}\rightarrow\mathbb{C}^{N} by {\mathscr{L}}_{N}(F)(z):=\sum_{j=1}^{d}e^{G(\gamma_{j}z)}U_{j}F(\gamma_{j}z). These type of operators appear naturally in the geometric context of covers of hyperbolic surfaces, see for example in [ 20 , 12 , 29 ] , where their spectral properties are closely related to Laplace spectrum. We will let {\mathscr{L}}_{N} act on the Bergman space \mathcal{H}_{N}:=H^{2}(\Omega_{0},\mathbb{C}^{N}), defined by \mathcal{H}_{N}:=\left\{F:\Omega_{0}\rightarrow\mathbb{C}^{N}\ :\ F\ \mathrm{is\ holomorphic\ and}\ \int_{\Omega_{0}}\|F(z)\|^{2}dm(z)<\infty\right\}, where m stands for Lebesgue measure on \mathbb{C} and \|.\| is the usual Hermitian norm on \mathbb{C}^{N} . The operator {\mathscr{L}}_{N} is then compact and trace class, as composition operators by strict holomorphic contractions have this property, see [ 4 ] . The spectrum of {\mathscr{L}}_{N} is therefore discrete outside \{0\} and we denote by (\lambda_{j}({\mathscr{L}}_{N}))_{j\in\mathbb{N}} this sequence of eigenvalues with \lim_{j\rightarrow\infty}\lambda_{j}({\mathscr{L}}_{N})=0 . A typical example . Let \Omega_{0} be the open disc given by \Omega_{0}=D\left(1,\frac{3}{2}\right)=\left\{z\in\mathbb{C}\ :\ |z-1|<\frac{3}{2}\right\}. For all j=1,\ldots,d\leq\infty , set \gamma_{j}(z)=\frac{1}{j+z} . Then it is easy to see that each Möbius transformation \gamma_{j} satisfies \gamma_{j}(\Omega_{0})\subset D(1,1). The associated transfer operator with weights e^{G_{j}(z)}=\frac{1}{(j+z)^{2\sigma}} is often called the Mayer-Perron-Frobenius operator, see [ 21 ] , and acts on scalar functions by the popular formula {\mathscr{L}}_{\sigma}(f)(x):=\sum_{j=1}^{d}\frac{1}{(j+z)^{2\sigma}}f\circ\gamma_{j}(x). When d is finite, which is the case we only consider in this paper, it is often associated to the dynamics of continued fractions with bounded entries and the ”open” Gauss map. The limit set generated by \gamma_{1},\ldots,\gamma_{d} is a Cantor subset of [0,1] with positive Hausdorff dimension. These transfer operators are standard models of ”open chaotic systems” and are useful toy-models in quantum chaos, see for example the paper of Arnoldi-Faure-Weich [ 1 ] . In this paper we will focus on the behaviour of spectra and singular values of general operators {\mathscr{L}}_{N} in the large N regime, when permutation matrices U_{1},\ldots,U_{d} are chosen uniformly independently, d being fixed. For all probabilistic statements, {\mathbb{P}}_{N} will denote the probability of an event while {\mathbb{E}}_{N} will denote the expectation. An event \mathcal{E}_{N} is said to occur asymptotically almost surely (a.a.s.) if we have \lim_{N\rightarrow+\infty}{\mathbb{P}}_{N}(\mathcal{E}_{N})=1 . A first basic result that we will prove is a deterministic upper bound on the number of eigenvalues as N goes to infinity. Theorem 1.1 . Using the above notations, there exists a constant C_{0}>0 such that for all r>0 , \#\{j\in\mathbb{N}:\ |\lambda_{j}({\mathscr{L}}_{N})|\geq r^{-1}\}\leq C_{0}rN. This upper bound follows straightforwardly from an estimate of the trace norm of {\mathscr{L}}_{N} and Weyl inequalities, see \lx@sectionsign 2 . A natural and difficult problem is the following conjecture. Conjecture 1.2 . Assume that d\geq 2 . There exists an absolutely continuous measure \nu , compactly supported in \mathbb{C} , locally finite in \mathbb{C}^{*} , such that for all \varphi\in C_{0}^{\infty}(\mathbb{C}^{*}) , we have \lim_{N\rightarrow\infty}\frac{1}{N}{\mathbb{E}}_{N}\left(\sum_{\lambda\in\mathrm{Sp}({\mathscr{L}}_{N})}\varphi(\lambda)\right)=\int_{\mathbb{C}}\varphi d\nu. In the case when d=1 and G=0 , this is not true because spectra of single composition operators T_{\gamma} can be explicitely computed and involve powers of the derivative at the unique fixed point of \gamma in \Omega_{0} . The spectrum of {\mathscr{L}}_{N} is therefore the spectrum of the tensor product T_{\gamma}\otimes U where U is a random permutation matrix, and the limit distribution is not absolutely continous. We will assume that d\geq 2 in the whole paper. Should this conjecture be proved for d\geq 2 , it could be considered as an infinite dimensional variant of the ”single ring theorem”, see [ 17 , 8 ] . To understand why this is a hard problem (due to the non-self adjointness), one has to be aware that for the finite dimensional problem of sums of d i.i.d random permutation matrices, the limit law is currently still not known when d is fixed, see in [ 7 ] for more insights. The problem is likely to be slightly easier if U_{1},\ldots,U_{d} are chosen as independent, Haar distributed, unitary matrices, see [ 8 ] , but the techniques used in [ 8 ] should be replaced by a softer approach to allow an extension to infinite dimensional operators. A fist step toward the analysis of non-self adjoint spectra is to study singular values instead. We recall that if T:\mathcal{H}\rightarrow\mathcal{H} is a compact operator acting on a separable Hilbert space \mathcal{H} , the singular values sequence \mu_{0}(T)=\|T\|\geq\mu_{1}(T)\geq\ldots\geq\mu_{j}(T)\geq\ldots>0 is by definition the eigenvalue sequence of the positive compact self-adjoint operator \sqrt{T^{*}T} . We prove the following fact. Theorem 1.3 . Set for all r>0 , \mathcal{N}(r):=\#\left\{j\in\mathbb{N}\ :\ \mu_{j}({\mathscr{L}}_{N})\geq\frac{1}{r}\right\}. Using the above notations, there exist constants C_{2}>0 and r_{0}>0 with the following property. For all k\in\mathbb{N} , one can find A_{k}>0 such that as N\rightarrow+\infty , we have {\mathbb{P}}_{N}\left(C_{2}^{-1}N\leq\mathcal{N}(r_{0})\leq C_{2}N\right)\geq 1-\frac{A_{k}}{N^{k}}. In other words, with rapid convergence as N\rightarrow+\infty , we have an a.a.s Weyl law for the singular values of {\mathscr{L}}_{N} . The constant r_{0} can be made effective, for example if \Omega_{0} is a disc. The proof of Theorem 1.3 actually follows readily from the following moments result, which is of independent interest. Proposition 1.4 . Set \|{\mathscr{L}}_{N}\|_{2}^{2}=\sum_{j=0}^{\infty}\mu_{j}^{2}({\mathscr{L}}_{N}) . (1) There exists an explicit L_{1}=L_{1}(G,\gamma_{1},\ldots,\gamma_{d})>0 such that we have \lim_{N\rightarrow+\infty}\frac{{\mathbb{E}}_{N}(\|{\mathscr{L}}_{N}\|_{2}^{2})}{N}=L_{1}. (2) Consider the moments \mathbb{V}_{N}^{(k)}:={\mathbb{E}}_{N}\left(\left|\|{\mathscr{L}}_{N}\|_{2}^{2}-{\mathbb{E}}_{N}(\|{\mathscr{L}}_{N}\|_{2}^{2})\right|^{2k}\right), then for all k\geq 1 , L_{2k}:=\lim_{N\rightarrow\infty}\mathbb{V}_{N}^{(k)} exists and is finite. (3) Moreover, set for all i,j\in\{1,\ldots,d\} , H_{i,j}:=\int_{\Omega_{0}}e^{G\circ\gamma_{i}(z)}e^{\overline{G\circ\gamma_{j}(z)}}B_{\Omega_{0}}(\gamma_{i}z,\gamma_{j}z)dm(z), where B_{\Omega_{0}}(z,w) denotes the Bergman kernel of the space H^{2}(\Omega_{0},\mathbb{C}) . Consider the random variable X_{N}:=\|{\mathscr{L}}_{N}\|_{2}^{2}-{\mathbb{E}}_{N}(\|{\mathscr{L}}_{N}\|_{2}^{2}). Then as N\rightarrow+\infty , we have convergence in distribution X_{N}\overset{\mathrm{dist}}{\longrightarrow}2\sum_{1\leq i<j\leq d}(Z_{i,j}-1){\rm Re}(H_{i,j}), where for all i<j , Z_{i,j} are

Read the complete text