Probing the influence of topological and geometric disorder on the spectrum of the differential Laplacian operator on networks Thanks: Submitted to the editors August 10, 2026. Charles Emmett Maher, Jeremy L. Marzuola, Katherine A. Newhall Abstract Metric networks are network-shaped, one-dimensional structures on which one can solve differential equations to simulate a wide range of physical systems including conjugated molecules, photonic crystals, quantum mechanics in waveguide networks, and acoustic metamaterials. More concretely, a metric network is a network whose edges are each assigned a notion of length and a coordinate describing position. One can then define function spaces and differential operators on these objects to model the aforementioned systems. Recent software advancements have made it feasible to analyze partial differential equations on large, compact metric networks with a vast array of structures. Here, we generate compact metric network structures using the spatial tessellations of two-dimensional hyperuniform point patterns, which have suppressed large-scale density fluctuations relative to typical disordered point patterns. This choice of structure is inspired by the exotic physical properties of network materials with these structures in other contexts. Then, we characterize the eigenvalue spectrum structure of the differential Laplace operator on these networks. In particular, we find that gaps can form in the eigenvalue spectra of these networks whose widths increase when the distribution of edge lengths is narrow and as the number of triangular faces increases. Importantly, many of the structures we consider are realizable in Euclidean space, meaning they are well-suited for practical applications in, e.g., metamaterial design. This work can thus be used to inform the design of metric network-based systems with spectral gaps with tunable widths and locations. keywords metric networks, disordered structures, eigenvalue spectra, hyperuniformity Funding. CEM is supported by the collaborative NSF DMREF Grant No. CMMI-2323342. KAN is supported by the collaborative NSF DMREF Grant No. CMMI-2323342 and NSF Grant No. DMS-2307297. JLM is supported by NSF Applied Math Grant DMS-2307384. † † runningheads: Influence of topological and geometric disorder on the spectrum of the differential Laplacian on networks / C. E. Maher, J. L. Marzuola, and K. A. Newhall MSC 82D30, 34B45, 58C40 1 Introduction Network structures are of fundamental importance in a plethora of contexts including biological systems [ 64 , 84 ] , industrial applications [ 23 , 61 ] , and pure mathematics [ 7 ] . Of present interest is the application of networks to materials science [ 53 , 54 , 60 , 62 , 69 , 72 ] . Disordered network metamaterials are a material class of particular interest due to their wide variety of exotic physical properties including negative Poisson’s ratio [ 67 , 68 , 71 ] , exotic mechanical responses [ 77 , 89 ] , and the ability to mimic the mechanical properties of biological tissues [ 55 , 66 , 75 , 86 ] . Recent advancements in additive manufacturing methods have made realizing computationally-generated network structures feasible, allowing for the direct experimental assessment of the theoretically-predicted properties of such structures [ 56 ] . With these experimental developments, it is increasingly important to understand the relationship between the structure of a network and its properties so we can manufacture materials with tailored physical properties. Such materials have applications in diverse fields including aerospace [ 1 ] and biomedical devices [ 14 ] . Importantly, in each of the areas noted so far, it is important to consider the entire spatial extent of the network structure. Metric networks 1 1 1 We note that these objects are also often referred to as metric graphs , but we opt to use the term metric network to be consistent with other mentions of network objects in this paper. , which can be thought of as network-shaped spaces, are a natural way to describe these spatially extended network systems where the dynamics of interest take place on the edges [ 9 ] . More specifically, a metric network consists of a set of vertices and edges where each edge is associated with a positive length and a coordinate describing a position on the edge. One can then define function spaces and differential operators on the metric network (a combination sometimes called a quantum graph) and compute the solutions to differential equations on these networks. A wide range of physical systems can be modeled with this framework, including molecules [ 4 , 10 , 19 , 44 ] , photonic crystals [ 5 , 41 , 42 ] , quantum mechanics in waveguide networks [ 18 ] , and acoustic metamaterials [ 46 ] . By contrast, the common notion of combinatorial networks, which consist of a set of vertices that are connected by a set of edges that describe the relationships between discrete entities in the network [ 59 ] , are used to model, e.g., ecological systems [ 20 ] , metabolic [ 33 ] and gene regulatory networks [ 13 ] , and opinion dynamics [ 76 ] . Because there is no notion of space on combinatorial networks, one can only model dynamics on them using ordinary differential equations. The dynamics on combinatorial networks are often described using the graph Laplacian matrix [ 52 ] , which is a discrete analog of the differential Laplacian that can be used to model wave and heat transport on metric networks. A problem of interest for metric networks is the characterization of the eigenvalue spectrum of the differential (i.e., not graph) Laplacian operator (see, e.g., Chapter 3 of Ref. [ 9 ] and references therein for an overview). Of particular interest are gaps in this spectrum, which are important to the study of solid-state physics [ 43 ] , photonics [ 42 ] , and waveguide systems [ 47 , 87 ] . A large amount of the work on gaps in the spectra of metric networks focuses on the spectral gap, which is the gap between the 0 eigenvalue and the next largest one [ 6 , 8 , 11 , 35 , 36 , 45 ] . In periodic metric networks, one can create gaps elsewhere in the eigenvalue spectrum by applying “decorations” to the vertices, following the method from Ref. [ 9 ] which adapts a method used for combinatorial networks [ 70 ] . The study of differential equations on metric networks has focused primarily on small periodic and compact networks [ 63 ] , which have a finite number of edges that have finite lengths. Relatively little work has been done on large compact metric networks due to the inaccessibility of methods to accurately compute solutions to differential equations on such structures. Recent advances in computing have made it feasible to study the spectra of and simulate dynamics on large, finite metric network structures. For example, Holden and Vasil developed code to compute the spectrum of the Laplacian on large finite and periodic metric networks and showed that the solutions of partial differential equations (PDEs) on metric networks embedded in a manifold converge to the solution of its corresponding PDE posed on the manifold itself as the density of vertices increases [ 32 ] . Böttcher and Porter have developed a Python package that treats a large class of PDEs on metric networks with up to \sim\!10^{4} edges and \sim\!10^{4} vertices to probe the relationship between structure and dynamics on metric networks [ 12 ] . Weller has developed a similar Julia package for elliptic and parabolic PDEs on metric networks [ 85 ] . Here, we use QGLAB, a MATLAB package developed by Goodman, Conte, and Marzuola that is used for computing the spectra of elliptic operators and simulating the time evolution of PDEs on metric networks [ 30 ] . We choose to use QGLAB due to its adaptability in constructing large random metric networks and building the corresponding Laplacian operator [ 30 ] . To narrow down the space of possible network structures to examine, we consider networks generated from the spatial tessellations of hyperuniform point patterns, which are of interest due to their broad range of desirable physical properties [ 16 , 17 , 21 , 34 , 51 , 72 ] . Hyperuniform point patterns have anomalously suppressed density fluctuations compared to those in typical disordered systems [ 80 ] , resulting in a structure factor S(\mathbf{k}) that tends to 0 as the wavenumber k\equiv|\mathbf{k}| tends to 0. Of particular interest are disordered stealthy hyperuniform systems, whose structure factors are identically zero for some set of wavenumbers around the origin [ 79 ] . Network materials generated from the tessellations of these point patterns have desirable physical properties, for example, nearly optimal elastic moduli and thermal and electrical conductivity [ 74 , 78 ] . In addition, these network materials can possess complete, isotropic photonic bandgaps [ 21 , 51 , 72 ] . Inspired by these exotic properties, here we generate metric networks from the tessellations of hyperuniform point patterns and characterize their eigenvalue spectra. In this work, we generate compact metric network structures from the tessellations of disordered two-dimensional (2D) hyperuniform, disordered stealthy hyperuniform, and nonhyperuniform point patterns with different degrees of short-scale translational disorder. In particular, we apply the Delaunay tessellation [ 7 ] , Gabriel tessellation [ 24 ] , and Delaunay-centroidal tessellation [ 21 ] schemes to these point patterns respecting periodic boundary conditions, then break the periodicity in different ways, which are discussed in detail in Sec. 3 . We focus primarily on compact metric network systems due to the existing capabilities to manufacture such structures in a laboratory setting [ 56 ] . Then, using QGLAB, we compute the eigenvalues and eigenmodes of the combinations of the point patterns, tessellation schemes, and boundary types above and examine how they affect the eigensystem. We find that 2D metric networks with a broader distribution of edge lengths and fewer triangular cells tend to have flatter eigenvalue distributions. By contrast, metric networks with a narrow distribution of edge lengths and a large fraction of triangular cells exhibit deep troughs in their eigenvalue distributions. These results offer a method to tune the position and size of gaps in the eigenvalue spectrum of disordered compact metric networks. Moreover, because eigenmodes can be used to construct solutions of linear PDEs, this study of the eigensystems of metric networks can be used to inform the design of disordered network metamaterials with, e.g., targeted heat and wave transport characteristics. Spectral properties of disordered metric networks have been studied in a variety of ways previously, but not regularly at the scale of the models generated here and with comparable disorder. For instance, in Ref. [ 28 ] , the authors consider the large-network limit of spectral average functions and spectral two-point correlation functions. They consider both a periodic-orbit approach and a supersymmetry approach, with the disorder generated by a fixed choice of rationally independent bond lengths. The authors give many details on star networks, but discuss how to apply spectral statistics more broadly in their setting. Similarly, in Ref. [ 27 ] they consider the autocorrelation statistics of high-energy eigenfunctions in large metric networks. In Refs. [ 2 ] and [ 3 ] , the authors consider a family of disordered networks and study the statistics of the distribution of the oscillation of eigenvectors relative to the maximal amount of oscillation possible (as related to the energy of the eigenfunction and the number of cycles in the network). On trees for instance, as the eigenvalue increases in energy, the eigenfunction increases in oscillation. However, in more topologically interesting examples, this is a much more challenging question, and the authors prove that the oscillations distribute in a Gaussian fashion along the allowed values for a variety of networks in a large network limit. In Ref. [ 73 ] , the authors consider the related problem of randomized transmission/reflection coefficients for disordered networks attached to incoming about outgoing leads. The total number of vertices in the networks studied however were relatively small. The authors in Ref. [ 39 ] consider the scattering matrix in metric networks with both randomly chosen edges and randomly determined vertex conditions. The rest of this paper is organized as follows. In Sec. 2 , we present the pertinent mathematical background and definitions. In Sec. 3 , we describe how we generate the metric networks examined in this work and how we characterize their eigensystems. In Sec. 4 , we characterize the structure and eigensystems of metric networks and examine the effects of disorder on their eigenvalue spectra. Finally, in Sec. 5 we offer concluding remarks and outlook for future studies. 2 Mathematical Background In this section, we first review the construction of metric networks and how one can define differential operators on them, following closely the setup in Ref. [ 30 ] . Then, we present the basic definitions associated with hyperuniform point patterns and networks. 2.1 Metric networks A metric network can be considered as a complex of edges on which one can define function spaces and differential operators. First, consider a directed combinatorial network \Gamma=(\mathcal{V},\mathcal{E}) with a set of vertices \mathcal{V}=\{\mathtt{v}_{n},n=1,\dots,|\mathcal{V}|\} and set of directed edges \mathcal{E}=\{\mathtt{e}_{m}=(\mathtt{v}_{i}\rightarrow\mathtt{v}_{j}),m=1,\dots,|\mathcal{E}|\} . The degree of each vertex d_{i} is the number of edges that either originate or terminate at vertex \mathtt{v}_{i} . To obtain a metric network, we associate a real interval [0,\ell_{m}] of length \ell_{m} to each edge and a coordinate x that increases from 0 to \ell_{m} as the edge is traversed from \mathtt{v}_{i} to \mathtt{v}_{j} . Here, we only consider compact metric networks, so |\mathcal{V}| , |\mathcal{E}| , and all \ell_{m} are positive and finite. A function \mathbf{\Psi}(x) defined on \Gamma can be interpreted as a collection of functions defined on each of the edges of \Gamma , \mathbf{\Psi}|_{\mathtt{e}_{m}}=\Psi_{m}(x) . A Laplace operator on \Gamma is defined by \bigtriangleup|_{\mathtt{e}_{m}}=\frac{\textrm{d}^{2}}{\textrm{d}x^{2}},\quad\textrm{for}\quad 00 indicates a point pattern is hyperuniform. Equivalently, a point pattern is hyperuniform if the number variance grows more slowly than the volume of the observation window, i.e., \sigma_{N}^{2}(R)/v_{1}(R)\rightarrow 0 as R\rightarrow\infty . Hyperuniform point patterns can be split into three distinct classes based on the scaling behaviors of their structure factors and number variances [ 80 ] : \sigma_{N}^{2}(R)\sim\begin{cases}R^{d-1},\quad\alpha>1&\textrm{(Class I)},\\ R^{d-1}\textrm{ln}(R),\quad\alpha=1&\textrm{(Class II)},\\ R^{d-\alpha},\quad 0<\alpha<1&\textrm{(Class III)},\end{cases} (10) where class I and III are the strongest and weakest types of hyperuniformity, respectively. Recently, the notion of hyperuniformity was extended to treat spatially embedded network structures via the definition of a variance-based method to characterize the large-scale density fluctuations in spatially-embedded networks analogous to number variance [ 50 ] . In particular, Maher and Newhall [ 50 ] found that networks generated from the tessellations of hyperuniform point patterns did not completely inherit the hyperuniformity of their progenitor point patterns. However, networks derived from hyperuniform point patterns with large degrees of short-scale translational order were effectively hyperuniform [ 15 ] , suggesting that they should still possess the desirable physical properties hyperuniform systems are known to possess [ 81 ] . Indeed, these desirable properties of the networks have been verified both experimentally and computationally (see e.g., Refs. [ 81 ] and [ 82 ] and references therein for examples). Additionally, we note that other methods have been employed to characterize the density fluctuations in networks derived from Delaunay [ 58 ] and Voronoi [ 57 ] tessellations of hyperuniform and nonhyperuniform point patterns and characterize their physical and mathematical properties [ 65 ] . 3 Methods In this section, we first describe the point patterns, tessellation schemes, and structural boundary conditions imposed to generate the networks examined herein. We show representative networks of each ensemble considered herein in the Supplementary Material [ 49 ] , Figures SM 1 -SM 7 . Then, we present the methods used to characterize the eigensystems of these metric networks with varying degrees of geometrical and topological disorder. 3.1 Point Patterns We generate networks from three different point patterns in this work: two hyperuniform models and one nonhyperuniform model. The first hyperuniform model we consider is the uniformly randomized lattice (URL) [ 37 ] . Here, we make the particular choice of displacing the points of the \mathbb{Z}^{2} lattice by a random 2 -dimensional vector uniformly distributed on a scaled version of the unit cell \mathcal{F} of \mathbb{Z}^{2} , i.e., a\mathcal{F}\equiv[-a/2,a/2)^{2} , where a controls the perturbation strength. Regardless of the exact value of a>0 , the URL is a class I hyperuniform system with \alpha=2 . One can tune the degree of short-scale translational disorder in the point pattern with the specific choice of a , in particular, larger values of a result in a greater degree of short-scale translational disorder. In this work we use \mathbb{Z}^{2} lattices with N=400 points in a square box under periodic boundary conditions as our initial conditions and values of a up to 0.3. We choose this system size because it is similar to the sizes used in the 3D-printed samples used in previous work [ 60 ] and choose this maximum value of a to avoid very short edges which leads to numerical discretization issues. Analogous considerations will apply to the rest of the point patterns described below. Additionally, we consider a modified URL procedure where we use the A_{2} lattice as an initial condition. In this work, we are not interested in examining the cloaking properties of this particular scheme [ 37 ] , so we still use a perturbation of the form [-a/2,a/2)^{2} , where a is now the nearest-neighbor distance between points in the lattice. Since we generate this point pattern in a periodic square box, which A_{2} is not commensurate with, we choose N=418 points to mitigate deformation of the A_{2} structure in a square box [ 83 ] . Hereafter, we differentiate between these two types of URL point patterns by calling them \mathbb{Z}^{2} URLs and A_{2} URLs. Figure 1 depicts an eigenmode of the Laplacian on an example metric network generated from the Delaunay tessellation (described in Sec. 3.2 ) of the A_{2} URLs described here. The second type of hyperuniform point pattern considered here is a disordered stealthy hyperuniform point pattern. These point patterns have S(k)=0 for 0