Localized and delocalized modes on random geometric graphs in 1D

Luca Schaefer, Barbara Drossel · 2025

Read Localized and delocalized modes on random geometric graphs in 1D by Luca Schaefer, Barbara Drossel (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-2508-18936.txt.

Introduction

Luca Schaefer and Barbara Drossel investigated in 2025 how waves spread, or fail to spread, on random one-dimensional networks. In such geometric graphs, points are scattered along a line and linked by proximity. Whether a vibration stays trapped near its start or spreads across the whole graph is a classic question with ties to disordered materials.

They studied two standard matrices for these graphs and mapped the density of states and a measure of how spread out each mode is. By separating the roles of system size, cluster sizes, link counts, small motifs, and repeated values, they build a full picture. They also compare the results with orderly graphs and with a standard model of electrons in disordered chains.

Catalog record

CREATORLuca Schaefer, Barbara Drossel
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE42 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2508-18936

Browse the full library · About the Tower of Babel

Full text

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

Localized and delocalized modes on random geometric graphs in 1D Luca Schaefer, Barbara Drossel Abstract We perform an extensive investigation of the localization properties of the eigenmodes of the Laplace and adjacency matrix for one-dimensional random geometric graphs. We evaluate the density of states, the probability distribution of the participation ratio and its relation to the eigenvalue. By disentangling the influence of system size, graph component size distribution, mean degree of nodes, network motifs, and degeneracy, we provide a comprehensive understanding of this system. We compare our findings to ordered graphs with the same mean degree and to one-dimensional tight-binding models. 1 Introduction The transport of charges in a semiconductor, the propagation of light in a crystal, and the dispersal of species in a patchy landscape are all affected by the presence of disorder. Starting from the seminal work by Anderson in 1958 [ 1 ] , it has become clear that a sufficiently strong disorder leads to a localization of eigenvectors of such systems. Anderson considered a tight-binding model for electrons with random on-site potential. Subsequent work generalized these findings to tight-binding models with disorder also in the hopping terms, and to other wave equations that can be mapped on the same type of eigenvalue problem. Generally, all states are localized in one dimension [ 2 ] , with two being the critical dimension for a disorder-induced metal-insulator phase transition [ 3 ] . When site and bond disorder are correlated so that the Hamiltonian of the Anderson model becomes a Laplace matrix where the sum of all entries in a row vanishes, there is a conserved quantity, leading to system-spanning eigenvectors in finite systems even in the presence of strong disorder and even in one dimension [ 4 , 5 ] . Such models apply, for instance, to classical harmonic chains and to the diffusion of a substance or a biological population through a set of coupled sites. For one-dimensional systems, it was shown that the number of system-spanning eigenvectors scales as 1/\sqrt{N} with N being the number of sites [ 4 , 6 , 7 , 8 , 5 ] . In the limit of infinitely large systems, this is a vanishing fraction of all modes, so that even in this case (almost) all modes are localized in 1D. More recently, research on the localization of eigenvectors was extended to complex networks. Complex networks incorporate disorder in their connection pattern and have ”localized” modes that are concentrated on a small number of network nodes, in addition to delocalized modes. The structure of a complex network is characterized by its adjacency matrix A_{ij} , with A_{ij} being 1 if the nodes i and j are connected and 0 otherwise. Evaluating the spectrum of the adjacency matrix of a network yields information on its structure [ 9 , 10 ] . The spectrum of the Laplace matrix is relevant when studying the diffusion of a substance on the network or when studying the synchronizability of a set of coupled oscillators, the couplings being represented by the edges of the network [ 11 , 12 ] . Laplacian spectra are also relevant for the linear stability analysis of dynamical processes on networks (see, for example, [ 13 ] ). Often, such studies are based on Erdős-Rényi and Barabási-Albert networks [ 14 , 15 ] . However, these two types of network do not include the concept of spatial proximity. In contrast, many realistic networks, such as transportation networks, wireless networks, social networks, or ecological networks, are embedded in space (see [ 16 ] for a review of spatial networks). A generic model for networks embedded in space is given by random geometric graphs (RGGs) [ 17 , 18 , 19 , 20 ] . Here, nodes are randomly distributed in space, and pairs of nodes with a distance smaller than a cut-off (connectivity radius) are connected. In one dimension, RGGs are a special case of interval graphs if the length of the intervals is constant, or circular-arc graphs for periodic boundary conditions and constant arc lengths. The local connection rules of RGGs lead to motifs, that is, recurrent substructures of the network [ 21 ] . A subset of those motifs are orbits that result from two nodes sharing the same neighborhood. Orbits lead to eigenvectors that are fully localized on the nodes of the orbits and have the value zero elsewhere [ 22 ] . The associated eigenvalues of the Laplacian are positive integers and the associated eigenvalues of the adjacency matrix are -1 and 0, depending on whether the nodes sharing the same neighborhood are connected or not (type-I and type-II orbits, respectively). Nyberg et al. have found a closed mathematical expression for the number of type-I orbits in 1D RGGs with periodic boundary conditions [ 22 , 21 ] . From this they conclude that approximately one third of the eigenvalues stem from type-I orbits in the limit of large N and fixed r . The same holds in the thermodynamic limit when the mean degree is fixed and much larger than two. Hamidouche et al. [ 23 ] compared the spectra of normalized Laplacians (which are rescaled such that the diagonal elements are 1) with those of a lattice model where all nodes have the same number of neighbors, and they found upper bounds for the deviation between the two in the limit of large node numbers. These bounds decrease with increasing mean number of neighbors per node. Similar results were obtained when the adjacency matrices were used instead of the normalized Laplacians [ 24 ] . The moments of the eigenvalue distribution of the adjacency matrix can be expressed in terms of the node density and connectivity radius, which in turn affects spreading processes, for instance of viruses [ 25 ] . In this work, we complement the cited studies of one-dimensional RGGs by an extensive investigation of the spectra of their adjacency and Laplace matrix. We focus on the participation ratio as a measure of localization and investigate its probability distribution and the relation between the participation ratios and the eigenvalues. Although the Laplace and adjacency matrix encode the same information about the structure of the RGG, the localization properties of their eigenmodes are qualitatively different due to the conservation law encoded in the Laplace matrix. By disentangling the influence of system size, mean degree, graph components, and orbits, we are able to provide an intuitive understanding of many observed features of the data. Comparison with ordered graphs where each node has the same number of neighbors, and with the 1D tight-binding model yields additional insights. 2 Models We generate one-dimensional random geometric graphs (RGG) of N nodes by drawing N random values from a uniform distribution on the interval [0,1) , which serve as coordinates of the nodes. Two nodes are connected if their distance is smaller than the connectivity radius, r . In the following, we assume periodic boundary conditions, that is, we distribute the nodes on a circle with circumference equal to 1. There are N! ways to label the N nodes of the graph. We choose the labeling such that it preserves the ordering of the positions of the nodes.

Read the complete text