Gaussian limit for Pfaffian point processes

Kai Wang, Mei Xu · 2025

Read Gaussian limit for Pfaffian point processes by Kai Wang, Mei Xu (2025) 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-2504-14487.txt.

Introduction

Kai Wang and Mei Xu prove a limit theorem for a broad class of random point patterns in this 2025 paper. The patterns, called Pfaffian point processes, arise where points repel each other in a structured way. The theorem concerns sums over the points, weighted by a chosen function. As the region grows, those sums settle toward a bell curve shape. The authors apply the result to two named processes, Sine-one and Sine-four, for step shaped weights.

Catalog record

CREATORKai Wang, Mei Xu
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE71 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2504-14487

Browse the full library · About the Tower of Babel

Full text

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

Gaussian limit for Pfaffian point processes Thanks: K.Wang is supported by grants NSFC (12231005,12326376) and the Shanghai Technology Innovation Project (21JC1400800). Kai Wang, Mei Xu Abstract . We prove a central limit theorem for linear statistics of a broad class of Pfaffian point processes. As an application, we derive Gaussian limits for scaled linear statistics of step functions in the Pfaffian Sine 4 \mathrm{Sine_{4}} and Sine 1 \mathrm{Sine}_{1} processes. 2020 Mathematics Subject Classification Primary 60G55; Secondary 30B20, 30H20. 1. introduction A determinantal point process is a random collection of points whose probability distribution can be expressed in terms of determinants. Such processes arise naturally in the study of various mathematical objects, including eigenvalues of random matrices, zeros of random analytic functions, and spanning trees on networks. Due to their wide applicability, determinantal point processes have been extensively studied; we refer to [Soshnikov2000] for a comprehensive review. However, in several important applications—such as eigenvalue distributions in the Gaussian orthogonal ensemble (GOE), Gaussian symplectic ensemble (GSE), and the real Ginibre ensemble, as well as in random combinatorial structures like tilings—the correlation functions are instead described by Pfaffians of antisymmetric matrices. These systems are governed by Pfaffian point processes, a natural counterpart to determinantal processes. To provide a precise definition of Pfaffian point processes, we first recall some preliminary notation. Let X be a locally compact Polish space, and let \rm Conf(X) denote the space of all locally finite configurations on X , i.e., \displaystyle{\rm Conf}(X)=\big\{\xi=\textstyle\sum\limits_{i}\delta_{x_{i}}\penalty\ |x_{i}\in X\penalty\ {\rm and}\penalty\ \xi(B)<+\infty\penalty\ {\rm for\penalty\ any\penalty\ compact}\penalty\ B\subset X\big\}. This configuration space is a Polish subspace of \mathcal{M}(X) , the space of Radon measures on X , equipped with the vague topology. A point process on X is a Borel probability measure \mathbb{P} on {\rm Conf}(X) . For further background, see, e.g., [ 8 , 21 , 22 , 23 ] . Now, let \mu be a Radon measure on X . A locally integrable function \rho_{k}:X^{k}\rightarrow[0,+\infty) is called the k -th ( k\in\mathbb{N}_{+} ) correlation function of the point process \mathbb{P} (with respect to the reference measure \mu ) if for any m\in\mathbb{N}_{+}, disjoint measurable compact subsets A_{1},\cdots,A_{m} of X , and positive integers k_{1},\cdots,k_{m} with \sum\limits_{i=1}^{m}k_{i}=k , the following identity holds: \displaystyle\mathbb{E}_{\mathbb{P}}\Big[\prod\limits_{i=1}^{m}\#_{A_{i}}(\#_{A_{i}}-1)\cdots(\#_{A_{i}}-k_{i}+1)\Big]=\int_{A_{1}^{k_{1}}\times\cdots\times A_{m}^{k_{m}}}\rho_{k}(x_{1},\cdots,x_{k})d\mu(x_{1})\cdots d\mu(x_{k}), where \#_{A}(\xi)=\xi(A) counts the points of \xi in A . For additional details, we refer to [ 8 , 13 , 16 , 29 ] . Let K:X\times X\rightarrow M_{2}(\mathbb{C}) be a matrix-valued kernel of the form \displaystyle K(x,y)=\left(\begin{array}[]{cc}K_{1,1}(x,y)&K_{1,2}(x,y)\\ K_{2,1}(x,y)&K_{2,2}(x,y)\end{array}\right), satisfying the following symmetry conditions: \displaystyle K_{1,1}(x,y)=K_{2,2}(y,x),\,K_{1,2}(x,y)=-K_{1,2}(y,x),\quad\text{ and }K_{2,1}(x,y)=-K_{2,1}(y,x). Define the matrix \displaystyle Z=\left(\begin{array}[]{cc}0&1\\ -1&0,\end{array}\right). and consider the antisymmetric kernel (1.1) \displaystyle\mathbb{K}(x,y)=ZK(x,y). Then, \mathbb{K}(x,y)^{T}=-\mathbb{K}(y,x) for any x,y\in X. A point process \mathbb{P} is called a Pfaffian point process if its k -th correlation function \rho_{k} (for every k\in{\mathbb{N}}_{+} ) admits the representation \displaystyle\rho_{k}(x_{1},\cdots,x_{k})=\mathrm{Pf}\big[\mathbb{K}(x_{i},x_{j})\big]_{i,j=1}^{k}. For further details, see [ 3 , 5 , 6 , 14 , 17 , 18 , 25 , 26 , 31 ] . A prominent example of Pfaffian point processes arises as the bulk scaling limits of Pfaffian Gaussian ensembles, known as the Pfaffian sine processes [ 13 ] . Denote the functions S(x):=\frac{sin(\pi x)}{\pi x},\,IS(x)=\int_{0}^{x}S(x)dx. The orthogonal sine process (or Pfaffian \mathrm{Sine}_{1} -process), is the Pfaffian point process on \mathbb{R} with the matrix kernel \displaystyle\mathbb{K}_{\mathrm{Sine}_{1}}(x,y)=ZK_{\mathrm{Sine}_{1}}(x,y),\quad K_{\mathrm{Sine}_{1}}(x,y)=\left(\begin{array}[]{cc}S(x-y)&S^{\prime}(x-y)-\frac{1}{2}sgn(x-y)\\ IS(x-y)&S(x-y),\end{array}\right) where Z=\left(\begin{array}[]{cc}0&1\\ -1&0,\end{array}\right) as before. Similarly, the symplectic sine process (or Pfaffian \mathrm{Sine}_{4} process), is the Pfaffian point process on \mathbb{R} with the matrix kernel \displaystyle\mathbb{K}_{\mathrm{Sine}_{4}}(x,y)=ZK_{\mathrm{Sine}_{4}}(x,y),\quad K_{\mathrm{Sine}_{4}}(x,y)=\frac{1}{2}\left(\begin{array}[]{cc}S(x-y)&S^{\prime}(x-y)\\ IS(x-y)&S(x-y)\end{array}\right). In the paper we will study the asymptotic behavior of linear statistics \displaystyle S_{f}(\xi)=\int_{X}fd\xi=\sum\limits_{x\in\xi}f(x),\quad\xi\in{\rm Conf}(X), for test functions f in a scaling limit. For the Pfaffian point process \mathbb{P} with the matrix kernel \mathbb{K} of the form ( 1.1 ), the expectation and variance of S_{f} (1.2) \displaystyle\mathbb{E}_{\mathbb{P}}[S_{f}]=\int_{X}f(x)A(x,x)d\mu(x), (1.3) \displaystyle\mathrm{Var}_{\mathbb{P}}(S_{f})=\int_{X}|f(x)|^{2}A(x,x)d\mu(x)-\int_{X^{2}}f(x)\overline{f(y)}\,\mathrm{det}K(x,y)d\mu(x)d\mu(y). Central limit theorems (CLTs) for point processes have been established in many cases. For example, Leblé [ 20 ] proved a central limit theorem for the fluctuations of linear statistics in the \mathrm{Sine}_{\beta} process, requiring test functions to be of class C^{4} with compact support. His approach combined the DLR equations, Laplace transform techniques, and transportation methods. Subsequently, Lambert [ 19 ] eveloped a simpler proof of a CLT for linear statistics of circular \beta -ensembles, , which holds at nearly microscopic scales for C^{3} functions and consequently yields a central limit theorem for \mathrm{Sine}_{\beta} processes. Related results can be found in [ 2 , 32 ] , among other works. The methods for determinantal point processes differ significantly from those used. The foundation was laid by Costin and Lebowitz [ 7 ] , who established a CLT for particle counts in the determinantal point process on \mathbb{R} with the kernel \sin(\pi(x-y))/\pi(x-y). Soshnikov [ 27 , 30 ] later extended this work, developing a comprehensive central limit theorem for general determinantal point processes with Hermitian kernels through the moment method. Breuer and Duits [ 4 ] further applied the moment method to obtain CLTs for orthogonal polynomial ensembles and biorthogonal ensembles where the underlying biorthogonal family satisfies a recurrence relation. For Pfaffian point processes, progress has been more limited. To date, the most general result is due to Kargin [ 17 ] , who proved a central limit theorem for Pfaffian point processes with finite-rank kernels. In this work, we establish a central limit theorem for a broad class of Pfaffian point processes. Our approach introduces a new tool called the finite-rank commutator property (FRCP), which quantifies how closely the kernel admits a tractable symmetry structure. Our main result extends Soshnikov’s framework for determinantal processes to the Pfaffian setting under FRCP. Key examples include the Pfaffian \mathrm{Sine}_{4} and \mathrm{Sine}_{1} processes, which satisfy this property. To formalize FRCP, we first recall essential operator constructions. Let A,B,D,A^{{\dagger}} denote the integral operator with kernel functions a(x,y),b(x,y),d(x,y),a(y,x) . Throughout the paper, when A is trace class, we assume a(x,y) is sufficiently smooth so that \int_{X^{k}}a(x_{1},x_{2})a(x_{2},x_{3})\cdots a(x_{k},x_{1})\,d\mu(x_{1})\cdots d\mu(x_{k})=\mathrm{Tr}(A^{k})=\mathrm{Tr}(A^{{{\dagger}}k}),\quad k\in\mathbb{N}_{+}. For f,g\in L^{2}(X,\mu) , we define the rank-1 operator f\otimes g by \displaystyle(f\otimes g)(h)=\langle h,g\rangle_{L^{2}(X,\mu)}f,\quad\mathrm{for}\quad h\in L^{2}(X,\mu). Then the definition of FRCP is as follows. Definition 1.1 . Let \mathbb{P} be a Pfaffian point process with the matrix-valued kernel \mathbb{K}(x,y)=ZK(x,y),K(x,y)=\lambda\left({\begin{array}[]{cc}a(x,y)&d(x,y)\\ b(x,y)&a(y,x)\end{array}}\right)(\lambda=\frac{1}{2}\,\mathrm{or}\,1) defined on a domain X with respect to the measure \mu . We say \mathbb{P} has the finite-rank commutator property (FRCP) if there exist a constant N and functions f^{(i)},g^{(i)},h^{(i)},e^{(i)}\in L^{2}(X,\mu)(i=1,\cdots,N), such that \displaystyle A^{{\dagger}}B-BA=\sum\limits_{i=1}^{N}f^{(i)}\otimes g^{(i)}, \displaystyle DB-(\alpha A^{2}+\beta A)=\sum\limits_{i=1}^{N}h^{(i)}\otimes e^{(i)}, where \alpha,\beta\in\mathbb{C} satisfy \alpha+\beta=\begin{cases}1,&\mathrm{if}\,\lambda=\frac{1}{2};\\ 0,&\mathrm{if}\,\lambda=1,\end{cases} The tuple \{N,f^{(i)},g^{(i)},h^{(i)},e^{(i)},\alpha,\beta\} parametrizes the FRCP for \mathbb{P} . As demonstrated in Section 3 , both the both the Paffian \mathrm{Sine}_{4} -process and the Paffian \mathrm{Sine}_{1} -process have FRCP. Building on this framework, we now present our central limit theorem for Pfaffian point processes with FRCP. Theorem 1.2 . Fixed an integer N . Consider a family \{\mathbb{P}_{L}\}_{L\geqslant 0} be a family of Pfaffian point processes having FRCP characterized by the data \{N,f^{(i)}_{L},g^{(i)}_{L},h^{(i)}_{L},e^{(i)}_{L},\alpha_{L},\beta_{L}\} with the matrix-valued kernel \mathbb{K}_{L}(x,y)=ZK_{L}(x,y),K_{L}(x,y)=\lambda_{L}\left({\begin{array}[]{cc}a_{L}(x,y)&d_{L}(x,y)\\ b_{L}(x,y)&a_{L}(y,x)\end{array}}\right)(\lambda_{L}=\frac{1}{2}\,\,\text{or}\,\,1), defined on the domain X_{L} with respect to the measure \mu_{L} . Under the following conditions: (i) \mathrm{Var}_{\mathbb{P}_{L}}(\#_{X_{L}})\rightarrow+\infty as L\rightarrow+\infty; (ii) Operators A_{L},B_{L},D_{L},A^{{\dagger}}_{L} are bounded with \sup\limits_{L}(\|A_{L}\|+\|A^{{\dagger}}_{L}\|+|\alpha_{L}|+|\beta_{L}|)<+\infty; (iii) A_{L} is trace-class and satisfies \|A_{L}-A_{L}^{2}\|_{1}=o\big(\mathrm{Var}_{\mathbb{P}_{L}}(\#_{X_{L}})\big)^{\delta} for some \delta>0; (iv) for all i,j=1,\cdots,N, and m,n\in\mathbb{Z}_{\geqslant 0}, the inner products \langle D_{L}A_{L}^{{{\dagger}}\,m}f^{(i)}_{L},{A_{L}}^{n}g^{(j)}_{L}\rangle,\langle D_{L}A_{L}^{{{\dagger}}\,m}f^{(i)}_{L},{A_{L}}^{n}e^{(j)}_{L}\rangle,\langle h^{(i)}_{L},{A_{t}}^{n}g^{(j)}_{L}\rangle,\langle h^{(i)}_{L},{A_{t}}^{n}e^{(j)}_{L}\rangle=o\big(\mathrm{Var}_{\mathbb{P}_{L}}(\#_{X_{L}})\big), then the normalized counting statistic \displaystyle\frac{\#_{X_{L}}-\mathbb{E}_{\mathbb{P}_{L}}[\#_{X_{L}}]}{\sqrt{\mathrm{Var}_{\mathbb{P}_{L}}(\#_{X_{L}})}} converges

Read the complete text