Functional central limit theorem for subgraph counts in a dynamic random connection model

Rajat Subhra Hazra, Nikolai Kriukov, Michel Mandjes, Moritz Otto · 2025

Read Functional central limit theorem for subgraph counts in a dynamic random connection model by Rajat Subhra Hazra, Nikolai Kriukov, Michel Mandjes, Moritz Otto (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-2511-18003.txt.

Introduction

Rajat Subhra Hazra, Nikolai Kriukov, Michel Mandjes, and Moritz Otto prove a limit theorem for counting small patterns, called subgraphs, in a network that changes over time. Their network is a dynamic random connection model, where points appear in space and connect at random with a chance that depends on distance.

The theorem is a functional central limit theorem, meaning the whole path of the count over time, once rescaled, behaves like a standard random process. To prove the needed tightness, they extend a cumulant method to the dynamic setting. The result gives a firm basis for judging fluctuations in evolving spatial networks.

Catalog record

CREATORRajat Subhra Hazra, Nikolai Kriukov, Michel Mandjes, Moritz Otto
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE45 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2511-18003

Browse the full library · About the Tower of Babel

Full text

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

Functional central limit theorem for subgraph counts in a dynamic random connection model Rajat Subhra Hazra, Nikolai Kriukov, Michel Mandjes, Moritz Otto Abstract . We prove a functional central limit theorem for subgraph counts in a dynamic version of the random connection model. To establish tightness, we develop a dynamic extension of the cumulant method. Affiliations. Rajat Subhra Hazra is with Mathematical Institute, Leiden University, The Netherlands. [email protected] Nikolai Kriukov is with Korteweg-de Vries Institute for Mathematics, University of Amsterdam, The Netherlands. [email protected] Michel Mandjes is with Mathematical Institute, Leiden University, The Netherlands, and Korteweg-de Vries Institute for Mathematics, University of Amsterdam, The Netherlands. [email protected] Moritz Otto is with Mathematical Institute, Leiden University, The Netherlands. [email protected] Keywords. Random graphs, stochastic geometry, functional limit laws MSC2020 subject classifications. 05C80, 60D05, 60G55. Acknowledgments. The research of Kriukov, Mandjes and Otto was supported by the European Union’s Horizon 2020 research and innovation programme under the Marie Sklodowska-Curie grant agreement no. 945045, and by the NWO Gravitation project N etworks under grant agreement no. 024.002.003. 1. Introduction The random connection model (RCM) is a fundamental stochastic geometric model used to describe spatial networks in which nodes are connected with a probability that depends on their mutual distance. More formally, given a homogeneous Poisson point process in \mathbb{R}^{d} , two points x and y are connected independently with probability that depends on the distance between x and y . The RCM generalizes the classical random geometric graph (or Gilbert graph) by allowing soft (probabilistic) connections instead of deterministic ones. It has been widely studied in the context of percolation theory, continuum random graphs, and wireless networks (see e.g., Meester & Roy [ 12 ] , Penrose [ 13 ] , and Roy [ 15 ] ). In the study of the RCM, an important branch of the literature concerns subgraph counts where one investigates the number of occurrences of a fixed finite graph H (such as a triangle or a star) as an induced or embedded subgraph of the RCM. These counts not only capture local connectivity patterns but also play a key role in understanding the higher-order structure and limiting behavior of the model. Subgraph counts have been studied using tools from Poisson point process theory, cumulants, Stein’s method, and U-statistics (see, e.g., Penrose [ 14 ] , Schulte & Thäle [ 16 ] , Liu & Privault [ 11 ] and Heerten et al. [ 7 ] ). In particular, asymptotic normality for these counts provide insights into scaling limits. A major trend in the random graph literature is the increasing focus on dynamically evolving networks , in which edges and/or vertices change stochastically over time to better reflect the behavior of real-world systems such as communication, social, or biological networks. These so-called temporal random graphs , as described by e.g. Holme & Saramäki [ 8 ] , capture both the structural and dynamic complexity of such systems and give rise to new mathematical challenges and insights. This shift reflects a growing interest in understanding not only static connectivity properties, but also how network behavior evolves over time. In principle, any property studied in the static setting can be extended to the dynamic context. For instance, the sample-path large deviation principle developed in Braunsteins et al. [ 2 ] generalizes the foundational result for the static Erdős–Rényi graph by Chatterjee & Varadhan [ 3 ] , while the maximum eigenvalue process studied by Hazra et al. [ 5 ] generalizes the static result from Erdős et al. [ 4 ] . In this paper, we consider a dynamic variant of the RCM, in which point locations and (potential) edges are sampled once and remain fixed, but where each point alternates between being active and inactive over time. Only edges between pairs of simultaneously active points are included into the graph. This setting naturally gives rise to a multivariate subgraph count process, tracking the number of occurrences of various subgraphs (such as edges, triangles, or stars) over time. The main result of the paper is a functional central limit theorem (CLT) for a centered and normalized version of this multivariate subgraph count process. As an application of this result, we present a functional CLT for the clustering coefficient process in the dynamic RCM. In Section 2 we formally introduce our model and state our main result. Our proofs build on the recently developed cumulant method, as introduced in a series of recent works (see e.g. Last et al. [ 10 ] , Schulte & Thäle [ 17 , 16 ] ). This method establishes a connection between Poisson U-statistics and a specific class of partitions, allowing their cumulants to be expressed in terms of certain tensor products (see Section 3 ). After having determined the candidate mean and covariances to feature in the functional CLT (Section 4 ), the fundamental challenge is to extend the use of cumulants beyond establishing finite-dimensional convergence (Section 5 ) to the functional setting. We do so by effectively applying them to prove tightness (Section 6 ), thereby illustrating the broad applicability of this approach. 2. Model and main results 2.1. Model description Denote W:=[-\tfrac{1}{2},\tfrac{1}{2}]^{d}, and let \eta_{n} be a Poisson point process on W\times\mathbb{D}([0,T],\{0,1\}), with intensity measure n\,\mathrm{d}x\otimes\mathbb{Q} . Here T>0 is the time horizon, and \mathbb{Q} denotes the law of a two-state Markov jump process on \{0,1\} with exponential waiting times, which jumps from 0 to 1 at rate \mu and from 1 to 0 at rate \lambda , started at its stationary distribution. In order to avoid potential boundary effects, throughout this paper we work with the torus metric on W , i.e., we identify opposite faces of W (i.e., the space ‘wraps around’ like a torus). We write P=(X,(A(t))_{t\in[0,T]}) for a canonical element in \eta_{n} , where X\in W is the location of P and A(t) is its status process at time t\in[0,T] . Our model is a random connection model (RCM) in which the underlying graph’s vertices alternate between being present and absent as follows. Let \mu>0,\,\lambda>0 be model parameters. At time t=0 , each of the points in \eta_{n} is active independently with probability \varrho:={\mu}/{(\mu+\lambda)} , and inactive else. As time goes on, the points in \eta_{n} independently alternate between being active and inactive: inactive (active) points become active (inactive) after an exponentially distributed time with rate \mu ( \lambda , respectively). Note that the expected number of active Poisson points is \varrho\,n for all t\geq 0 . Let \mathcal{X}\subset W be locally finite. In the RCM on \mathcal{X} , between any pair of distinct points X_{i},X_{j}\in\mathcal{X} we place a (potential) edge independently with probability \varphi_{n}(X_{i}-X_{j}) , where (1) \displaystyle\varphi_{n}({x}):=\varphi\bigg(\frac{\|{x}\|^{d}}{\nu_{n}}\bigg). Here, (\nu_{n})_{n\geq 1} is a positive real-valued sequence with \nu_{n}\to 0 as n\to\infty , and \varphi:[0,\infty)\to[0,1] (typically referred to as the profile function ) is assumed to be non-increasing and to satisfy the normalization \int\varphi(t)\,\mathrm{d}t=1 . The edges are sampled once and for all, but they appear in the RCM only when both adjacent nodes are simultaneously active (see Figure 1 ). The resulting dynamic random graph process we denote by \mathrm{RCM}(\mathcal{X},\varphi_{n}) . Figure 1 . The dynamic RCM at different times. Nodes alternate between active (full dot) and inactive (empty dot). Potential

Read the complete text