Nodal Count for Orthogonally Invariant Ensembles
Lior Alon, Dan Mikulincer, John Urschel · 2025
Read Nodal Count for Orthogonally Invariant Ensembles by Lior Alon, Dan Mikulincer, John Urschel (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-2511-02784.txt.
Introduction
Lior Alon, Dan Mikulincer, and John Urschel count sign changes along eigenvectors of random matrices. Read as operators on signed complete graphs, these vectors have a nodal count, the number of sign flips between neighbors. Their 2025 paper focuses on orthogonally invariant ensembles, with the Gaussian Orthogonal Ensemble as the central case.
As matrix size grows, the spread of nodal counts settles onto the same limiting law as the eigenvalues themselves. For the Gaussian case, that limit is the semicircle law. The outcome overturns a conjecture from quantum chaos and quantum graphs, which had predicted a Gaussian shape for the nodal count. The finding links two counts that were not expected to share one law.
Catalog record
| CREATOR | Lior Alon, Dan Mikulincer, John Urschel |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 61 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2511-02784
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Nodal Count for Orthogonally Invariant Ensembles Lior Alon, Dan Mikulincer, John Urschel Abstract . We investigate the nodal count of eigenvectors of random matrices interpreted as operators on signed complete graphs. Our focus is on orthogonally invariant ensembles, with particular attention to the Gaussian Orthogonal Ensemble (GOE). We establish that, as the matrix size tends to infinity, the distribution of nodal counts converges to the same limiting law as the eigenvalue distribution. In the GOE case, this limit is the semicircle law. This result refutes a conjecture, motivated by quantum chaos and quantum graphs, which predicted Gaussian behavior of the nodal count. 2020 Mathematics Subject Classification Primary 05C50, 15A18, 15B52. 1. Introduction The purpose of this paper is to investigate the asymptotic nodal count sequence (size of the nodal sets of eigenvectors) of a random real symmetric n\times n matrix as n\to\infty . The nodal count sequence of a matrix A is \phi(A)=\left(\phi(A,k)\right)_{k=1}^{n} , with \phi(A,k):=\left|\{i<j,\ :\ A_{ij}\bm{\varphi}^{(k)}_{i}\bm{\varphi}^{(k)}_{j}>0\}\right|, where \bm{\varphi}^{(k)} is the eigenvector of the k -th smallest eigenvalue of A . We consider ensembles of matrices that are invariant under orthogonal transformations, most notably, the Gaussian orthogonal ensemble (GOE). Our result for GOE matrices disproves a recent conjecture about the nodal count for randomly signed graphs [ 3 ] . While our focus is on the discrete setting—namely, nodal counts for matrices—our motivation stems from foundational results on the measure of nodal sets of eigenfunctions on manifolds, as well as the significant progress over the past two decades on the study of nodal sets of random waves. The study of nodal patterns dates back to the 17th century when it was popularized by Ernst Chladni, who revealed intricate sand patterns on vibrating plates. Over time, mathematicians developed a rigorous theory to understand the interplay between order and chaos in these patterns, paving the way for spectral geometry. The first rigorous study of nodal sets was by Sturm in 1836 [ 28 ] , who showed that the k -th eigenfunction of a Sturm–Liouville operator on an interval has exactly k-1 zeros. In higher dimensions, nodal sets—the zero sets of eigenfunctions—exhibit much greater complexity. This naturally leads to questions about their geometric properties and how these depend on the underlying domain or manifold. A fundamental question in spectral geometry regarding the size of nodal sets was asked by S.T. Yau [ 29 ] . For a smooth compact manifold M , let \lambda_{k} denote the k -th smallest eigenvalue of the Laplacian \Delta_{M} and \phi(\Delta_{M},k) denote the measure of the nodal set of its eigenfunction. Yau’s conjecture states that c_{1}\sqrt{\lambda_{k}}<\phi(\Delta_{M},k)<c_{2}\sqrt{\lambda_{k}} for some positive constants c_{1},c_{2} . The proof for real analytic manifolds was given by Donnelly and Fefferman [ 14 ] . For smooth manifolds, Logunov and Malinnikova proved the lower bound and gave a polynomial upper bound [ 21 , 20 , 22 ] . To go beyond asymptotic bounds, additional structure is required. For instance, much more can be computed explicitly in the presence of Gaussian randomness. Berry [ 11 ] proposed that if a compact planar domain \Omega\subset\mathbb{R}^{2} is chaotic (in terms of billiard dynamics), then the local behavior of a Dirichlet eigenfunction with eigenvalue \lambda can be modeled by a Gaussian random field F_{\lambda} satisfying -\Delta F_{\lambda}=\lambda F_{\lambda} almost surely. He showed that the nodal set of F_{\lambda} in \Omega has average length C\cdot\mathrm{area}(\Omega)\sqrt{\lambda} and variance of order \mathrm{area}(\Omega)\log(\lambda) [ 11 ] . Since then, nodal sets of random wave models and random eigenfunctions have been extensively studied, with notable results for Laplace eigenfunctions on the two-dimensional sphere [ 24 , 25 ] and torus [ 27 , 19 ] . It was long believed that in all such cases, the properly normalized nodal length has a limiting Gaussian distribution. However, recent remarkable non-universality results identified sequences of eigenvalues for which the limiting nodal length distribution exists but is non-Gaussian [ 23 , 26 ] . Our results reveal a comparable breakdown of universality. We now turn to our main objects of interest, the nodal count of a matrix, or equivalently, of a finite graph. Given a real symmetric n\times n matrix A , we associate a weighted signed graph G with n vertices and an edge (i,j) for each pair i<j with non-zero off-diagonal entry A_{ij}\neq 0 . Every edge has an edge weight |A_{ij}| , and the edge sign is -\mathrm{sgn}(A_{ij}) . Thus, for example, if A is a graph Laplacian, then G is unweighted and unsigned. Sort the eigenvalues of a A in increasing order, \lambda_{1}\leq\lambda_{2}\leq\ldots\leq\lambda_{n} , and denote the k -th eigenvector by \bm{\varphi}^{(k)} . Definition 1.1 . The nodal count \phi(A,k) is the number of edges (i,j) of G for which \varphi^{(k)} changes sign with respect to the sign of the edge, namely A_{ij}\bm{\varphi}^{(k)}_{i}\bm{\varphi}^{(k)}_{j}>0 . Under the generic assumption that \bm{\varphi}^{(k)}_{j}\neq 0 for all k,j , we write \phi(A,k):=\left|\{i<j,\ :\ A_{ij}\bm{\varphi}^{(k)}_{i}\bm{\varphi}^{(k)}_{j}>0\}\right|=\frac{1}{2}\sum_{i<j}(1+\mathrm{sgn}(A_{ij}\bm{\varphi}^{(k)}_{i}\bm{\varphi}^{(k)}_{j})). Fiedler was the first to introduce and study this count [ 15 ] , who showed that an analogue of Sturm’s theorem holds: if G is a tree, then \phi(A,k)=k-1 for all k . Conversely, Band [ 8 ] showed that if \phi(A,k)=(k-1) for all k , then G must be a tree. As in the continuous setting, there are no such uniform estimates for general graphs. Instead, Berkolaiko [ 9 ] generalized Fiedler’s result and showed that for general graphs the following inequality always holds: 0\leq\phi(A,k)-(k-1)\leq\beta, where \beta=\beta(G)\geq 0 is the first Betti number of the graph, and \beta(G)=0 if and only if G is a tree. In the presence of randomness, if G is unweighted, and \bm{\varphi} is replaced by a random Gaussian vector, then the nodal count is binomial and converges to a Gaussian as n\to\infty [ 7 ] . Thes heuristics, together with numerical simulations and analogous results for quantum graphs [ 16 , 1 ] , led to the common belief that the nodal surplus sequence \sigma(A,k):=\phi(A,k)-(k-1) for k=1,\ldots,n should behave like a Gaussian centered at \beta/2 as \beta\to\infty . We introduce some notation for this matter: Notations 1.2 . For \bm{x}\in\mathbb{R}^{n} , we define \mathrm{avg}(\bm{x})=\frac{1}{n}\sum_{i=1}^{n}\bm{x}_{i} , \mathrm{std}(\bm{x})=\left(\frac{1}{n}\sum_{i=1}^{n}(\bm{x}_{i}-\mathrm{avg}(\bm{x}))^{2}\right)^{1/2} , the empirical distribution of its coordinates \mathrm{emp}(\bm{x}):=\frac{1}{n}\sum_{k=1}^{n}\delta_{\bm{x}_{k}}, and the normalized empirical measure \mathrm{emp}_{\mathcal{N}}(\bm{x}):=\mathrm{emp}\left(\frac{\bm{x}-\mathrm{avg}(\bm{x})\bm{1}}{\mathrm{std}(\bm{x})}\right). Let \phi(A)=(\phi(A,k))_{k=1}^{n} and \sigma(A)=(\sigma(A,k))_{k=1}^{n} . The common belief was that \mathrm{emp}_{\mathcal{N}}(\sigma(A)) is very close to a Gaussian centered at \beta/2 with variance of order \beta , for large n and \beta . This was disproved by two of the authors [ 5 ] , as they proved the following sharp bounds \tfrac{\beta}{n}\leq\mathrm{avg}(\sigma(A))\leq\beta-\tfrac{\beta}{n}. Examples of matrices saturating the upper and lower bounds were given for any possible choice of n and \beta [ 5 ] . These bounds were based on Berkolaiko’s nodal-magnetic theorem [ 10 , 12 ] , which states that the nodal surplus is a Morse index of the eigenvalue. More precisely, by assigning to each edge (i,j) an angle \theta_{ij}=-\theta_{ji} , the eigenvalues of the Hermitian matrix (A_{\theta})_{rs}=e^{i\theta_{rs}}A_{rs} are functions of \theta , that have critical points when \theta_{ij}\in\{0,\pi\} for all edges. At such a critical point, the Morse index of \lambda_{k} turns out to precisely equal the nodal surplus \sigma(A_{\theta},k) . Note that such a critical point corresponds to a signing e^{-i\theta_{ij}}\in\{-1,1\} of the edges of the graphs. Moreover,