On the pair correlation statistics for determinantal point processes on the sphere
Maryna Manskova · 2025
Read On the pair correlation statistics for determinantal point processes on the sphere by Maryna Manskova (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-2508-17952.txt.
Introduction
Maryna Manskova, a solo author, wrote a math paper in 2025 on points spread over a sphere. The paper studies how close pairs of points tend to be when the points are placed by random rules. Some rules push points apart at short range, while others do not.
Manskova works out the mean value of a pair-count score for three such rules. Two rules come from a class of processes known for spreading points out. The third rule jitters points on a fixed grid. She compares all three with the case of points placed with no link at all. At short range, the special rules show clear push-apart behavior. At larger range, all cases agree closely with the fully random case.
Catalog record
| CREATOR | Maryna Manskova |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 47 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2508-17952
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
On the pair correlation statistics for determinantal point processes on the sphere Maryna Manskova Abstract . In this paper, we study the expected value of the pair correlation statistics of randomized point configurations on the sphere, with the emphasis on point configurations generated by determinantal point processes. We study the cases of the spherical ensemble, the harmonic ensemble, and jittered sampling, and compare our results with those for the “truly random” (i.i.d.) case. Our results give evidence of the small-scale repulsion phenomenon which is characteristic for determinantal point processes, while on larger scales there is good agreement between all our studied cases and the i.i.d. case. 1. Introduction Let \{x_{i}\}_{i=1}^{\infty} be a sequence in [0,1] . The pair correlation function of \{x_{i}\}_{i=1}^{N} is defined as \displaystyle R(s,N)=\frac{1}{N}\#\left\{(i,j):1\leq i,j\leq N,\ i\neq j,\ \|x_{i}-x_{j}\|\leq\frac{s}{N}\right\}, where \|\cdot\| denotes the distance to the nearest integer and s\geq 0 . The limiting function of R(s,N) , if it exists, describes the asymptotic distribution of small-scale gaps between elements of the sequence. The normalisation of distances is 1/N , which corresponds to the average spacing between x_{1},\dots,x_{N} . If \{X_{i}\}_{i=1}^{\infty} is a sequence of independent, uniformly distributed random variables in [0,1] , the following convergence holds almost surely (1) \displaystyle\lim\limits_{N\rightarrow\infty}R(s,N)=2s for every s\geq 0 . We say that a sequence \{x_{i}\}_{i=1}^{\infty} has Poissonian pair correlations if it satisfies ( 1 ). This notion can be extended to deterministic point sets in unit cubes [0,1]^{d} [ 6 , 14 , 23 ] and on compact Riemannian manifolds [ 21 ] . Our particular interest in this paper are points on spheres. In the d -dimensional case, we study a local statistics of point sets defined by (2) \displaystyle\frac{1}{N}\#\left\{i\neq j\in\{1,\dots,N\}\colon d(x_{i},x_{j})\leq\frac{s}{N^{1/d}}\right\} for a distance function d(x,y) . For i.i.d. random points the limit for N\rightarrow\infty equals C_{d}s^{d} for a suitable constant C_{d} almost surely. Accordingly, deterministic point sets are said to have Poissonian pair correlations if they share this asymptotic behaviour. We study the notion of pair correlations for different samples of probabilistic point sets on the sphere. Let (x_{1},\dots,x_{N}) be a sample of points in \mathbb{S}^{d} equipped with a distance function \delta(x,y) (e.g. chordal or geodesic distance). We rewrite ( 2 ) as G_{s,N}(x_{1},\dots,x_{N})\coloneqq\frac{1}{N}\sum\limits_{1\leq i\neq j\leq N}F_{s,N}(x_{i},x_{j}), where F_{s,N}(x,y)=\mathbbm{1}_{[0,s]}\left(\delta(x,y)N^{1/d}\right). In the following chapters, we compute \mathbb{E}\left[G_{s,N}\right] as N\rightarrow\infty for different determinantal point processes on spheres (and projective spaces). A characteristic property of these processes is a repulsive correlation structure. For small values of s , the asymptotic behaviour of \lim\limits_{N\rightarrow\infty}\mathbb{E}\left[G_{s,N}\right] provides information on the degree of repulsion between the points, which turns out to be significantly different from C_{d}s^{d} as in the i.i.d. case. We recall the basic definitions of determinantal point processes, study the behaviour of \lim\limits_{N\rightarrow\infty}\mathbb{E}\left[G_{s,N}\right] for some particular examples (spherical ensemble, harmonic ensemble, and jittered sampling), and compare our findings to the case of i.i.d. points on the sphere. 2. Preliminaries and Notation This section contains some preparatory calculations concerning the geometry of hyperspheres. Here and throughout this paper we denote by \mathbb{S}^{d} the unit sphere in \mathbb{R}^{d+1} and by \sigma=\sigma_{d} the normalised surface area measure on \mathbb{S}^{d} . The geodesic distance \vartheta=\vartheta_{d} takes values in [0,\pi] . Note that \vartheta is equivalent to the Euclidean distance, since \|\mathbf{x}-\mathbf{y}\|=\sqrt{2(1-\cos\vartheta(\mathbf{x},\mathbf{y}))}=2\sin\frac{\vartheta(\mathbf{x},\mathbf{y})}{2}. We will use the notation C_{d}(\mathbf{x},\varphi)=\{\mathbf{y}\in\mathbb{S}^{d}\,|\,\vartheta(\mathbf{x},\mathbf{y})<\varphi\} for the hyperspherical cap with center \mathbf{x} and of angular radius \varphi . Recall that the surface area of \mathbb{S}^{d} is \omega_{d}=2\frac{\pi^{\frac{d+1}{2}}}{\Gamma\left(\frac{d+1}{2}\right)}, where \Gamma(x) is the gamma function. The normalised surface area of a cap C_{d}(\varphi) of angular radius \varphi is \sigma_{d}(C(\varphi))=\frac{\omega_{d-1}}{\omega_{d}}\int\limits_{0}^{\varphi}(\sin\theta)^{d-1}\mathrm{d}\theta. For d=2 , the formula is \sigma_{2}(C(\varphi))=\frac{1}{2}(1-\cos\varphi)=\left(\sin\frac{\varphi}{2}\right)^{2}. We introduce the notion \displaystyle S(\rho,\tau)=\sigma\left(C(\mathbf{x}_{1},\pi/2)\cap C(\mathbf{x}_{2},\rho)\right) for the normalised area of the intersection of a hemisphere of \mathbb{S}^{2} and a spherical cap of angular radius \rho , where \vartheta(\mathbf{x}_{1},\mathbf{x}_{2})=\pi/2+\tau . Note that \tau is the distance from the centre of the cap to the boundary of the hemisphere. One can use a concise formula for the surface area of the intersection of two spherical caps presented in [ 17 ] . In the notation of [ 17 ] , we have n=3 , \theta_{1}=\pi/2 , \theta_{2}=\rho , and \theta_{v}=\pi/2+\tau . This gives us S(\rho,\tau)=\int\limits_{\tau}^{\rho}\sin\varphi\cdot 2\arcsin\sqrt{1-\left(\frac{\tan\tau}{\tan\varphi}\right)^{2}}\,\mathrm{d}\varphi. We provide a geometric proof for the formula. Lemma 1 . For 0<\tau<\rho<\pi/2 we have S(\rho,\tau)=\frac{1}{4\pi}\left(\pi-\alpha(\tau)\cos\rho-2\beta(\tau)\right), where \displaystyle\beta(\tau)=\arcsin\left(\frac{\sin\tau}{\sin\rho}\right)\quad\text{and}\quad\alpha(\tau)=2\arctan\left(\frac{\cot\beta}{\cos\rho}\right). Proof. Let us fix the South hemisphere. We can take any point \mathbf{x}=\mathbf{x}(1,\theta,\pi/2-\tau) as the center of the second cap. For \tau>\rho , the cap C\left(\mathbf{x},\rho\right) does not intersect the equator. Hence S(\rho,\tau)=0 . For 0<\tau<\rho<\pi/2 , we get two points of intersection denoted by \mathbf{y} and \mathbf{z} . The area S(\rho,\tau) can be expressed as S(\rho,\tau)=S^{(1)}(\rho,\tau)-S^{(2)}(\rho,\tau), where S^{(1)}(\rho,\tau) is the area of the sector of C(\mathbf{x},\rho) between the points \mathbf{y} and \mathbf{z} , and S^{(2)}(\rho,\tau) is the area of the spherical triangle \triangle\mathbf{xyz} . We denote the plane angles of the triangle by \alpha=\angle\mathbf{yxz} and \beta=\angle\mathbf{xyz}=\angle\mathbf{xzy} . The normalised areas of the sector and the triangle are given by S^{(1)}(\rho,\tau)=\frac{\alpha}{4\pi}(1-\cos\rho)\quad\text{and}\quad S^{(2)}(\rho,\tau)=\frac{1}{4\pi}(\alpha+2\beta-\pi). This gives us S(\rho,\tau)=\frac{1}{4\pi}(\pi-\alpha\cos\rho-2\beta). In order to find \alpha and \beta , we denote by \mathbf{w} the middle point between \mathbf{y} and \mathbf{z} on the equator. By the symmetry w.r.t. the great circle passing through \mathbf{x} , \mathbf{w} and the South pole, we have \angle\mathbf{xwz}=\angle\mathbf{xwy}=\pi/2 . The angular length of the arc \mathbf{xw} is \tau . We obtain formulas for \alpha and \beta by applying the laws of cosine and sine to \triangle\mathbf{xwz} , \cos\frac{\pi}{2}=-\cos\frac{\alpha}{2}\cos\beta+\sin\frac{\alpha}{2}\sin\beta\cos\rho\quad\text{and}\quad\frac{\sin\beta}{\sin\tau}=\frac{\sin\frac{\pi}{2}}{\sin\rho}. ∎ Another result we will use in this paper is a particular case of the Funk-Hecke formula (see [ 22 ] ). Lemma 2 . For all bounded measurable f\colon[-1,1]\rightarrow\mathbb{R} and all \mathbf{y}\in\mathbb{S}^{d} , \displaystyle\int_{\mathbb{S}^{d}}f(\langle\mathbf{x},\mathbf{y}\rangle)\mathrm{d}\sigma_{d}(\mathbf{x})=\frac{\omega_{d-1}}{\omega_{d}}\int_{-1}^{1}f(t)(1-t^{2})^{d/2-1}\mathrm{d}t. In the case of i.i.d. random points in \mathbb{S}^{d} , we compute the value of the constant for the asymptotic of ( 2 ). Proposition 1 . For points \{\mathbf{x}_{1},\dots,\mathbf{x}_{N}\} taken independently from the uniform distribution on \mathbb{S}^{d} , we have (3) \lim\limits_{N\rightarrow\infty}\mathbb{E}[G_{s,N}(\mathbf{x}_{1},\dots,\mathbf{x}_{N})]=\frac{\omega_{d-1}}{\omega_{d}}\cdot\frac{s^{d}}{d}. Proof. We have \displaystyle\lim\limits_{N\rightarrow\infty}\mathbb{E}[G_{s,N}] \displaystyle=\lim\limits_{N\rightarrow\infty}\frac{1}{N}\cdot N(N-1)\cdot\mathbb{E}\left[\mathbbm{1}_{[0,s]}\left(\vartheta(\mathbf{x},\mathbf{y})N^{1/d}\right)\right] \displaystyle=\lim\limits_{N\rightarrow\infty}(N-1)\cdot\sigma_{d}(C(sN^{-1/d})) \displaystyle=\frac{\omega_{d-1}}{\omega_{d}}\cdot\frac{s^{d}}{d}. ∎ Note that for \mathbb{S}^{2} we obtain \lim\limits_{N\rightarrow\infty}\mathbb{E}[G_{s,N}]=\frac{s^{2}}{4} . Our goal is to compare the case of i.i.d. random points ( 3 ) to other probabilistic point sets obtained from determinantal point processes (such as spherical ensemble, harmonic ensemble, and jittered sampling), with a particular emphasis of level repulsion phenomena, which correspond to very small values of s . 3. Determinantal point process In this section, we recall some basic theory of determinantal point processes. The best general reference here is [ 16 ] . Let (\Lambda,\mathcal{B},\mu) be a measurable space, where \Lambda is a locally compact Polish space and \mu is a Radon measure on \Lambda . A sample of N distinct points \{x_{1},\dots,x_{N}\} in \Lambda can be seen as a simple counting measure \chi=\sum_{i}\delta_{x_{i}} . A simple random point process on \Lambda is a random simple counting measure on \Lambda . Let \mathcal{X} be a simple point process on \Lambda . A k -point correlation function \rho_{k}\colon\Lambda^{k}\rightarrow[0,+\infty) is a function such that for any family of disjoint sets A_{1},\dots A_{k}\in\mathcal{B} , we have \mathbb{E}\left[\prod\limits_{j=1}^{k}\mathcal{X}(A_{j})\right]=\int\limits_{A_{1}}\cdots\int\limits_{A_{k}}\rho_{k}(x_{1},\dots,x_{k})\mathrm{d}\mu(x_{1})\cdots\mathrm{d}\mu(x_{k}). The point process \mathcal{X} with correlation functions \rho_{k} is called determinantal if there exists a kernel K\colon\Lambda^{2}\rightarrow\mathbb{C} such that \rho_{k}(x_{1},\dots,x_{k})=\det\left[K(x_{i},x_{j})\right]_{1\leq i,j\leq k}, for all k\geq 1 and x_{1},\dots,x_{k}\in\Lambda . Moreover, the joint intensities \rho_{k} satisfy: \mathbb{E}\sum\limits_{x_{1},\dots,x_{k}\in\Lambda}f(x_{1},\dots,x_{k})=\int f(x_{1},\dots,x_{k})\rho_{k}(x_{1},\dots,x_{k}) for x=(x_{1},\dots,x_{k}) generated by the associated point process and for