Randomly coupled differential equations with elliptic correlations

DOI10.1214/22-AAP1886MaRDI QIDQ6165249zbMATH OpenOpenAlexFDO

Authors László Erdős, Torben Krüger, D. Renfrew

Publication date 31 July 2023

Published in The Annals of Applied Probability (Search for Journal in Brave)

Full work available at URL https://arxiv.org/abs/1908.05178, https://projecteuclid.org/journals/annals-of-applied-probability/volume-33/issue-4/Randomly-coupled-differential-equations-with-elliptic-correlations/10.1214/22-AAP1886.full

zbMATH Keywords

elliptic random matrixmatrix Dyson equationnon-Hermitian random matrixpartially symmetric correlationtime evolution of neural networks

Mathematics Subject Classification ID

Random matrices (algebraic aspects) (15B52) Random matrices (probabilistic aspects) (60B20)

Abstract: We consider the long time asymptotic behavior of a large system of $N$ linear differential equations with random coefficients. We allow for general elliptic correlation structures among the coefficients, thus we substantially generalize our previous work [14] that was restricted to the independent case. In particular, we analyze a recent model in the theory of neural networks [27] that specifically focused on the effect of the distributional asymmetry in the random connectivity matrix $X$. We rigorously prove and slightly correct the explicit formula from [28] on the time decay as a function of the asymmetry parameter. Our main tool is an asymptotically precise formula for the normalized trace of $f(X)g(X^{\ast })$, in the large $N$ limit, where $f$ and $g$ are analytic functions.

Review text

In order to describe the evolution of a network of \(N\) fully connected neurons, one can consider the following system of linear differential equations: \[ \partial_t u(t) = -u(t) + g X u(t), \] where \(u\) has values in \(\mathbb{C}^N\), \(X\in\mathbb{C}^{N\times N}\) is a random matrix, and \(g>0\) is a coupling parameter. In the simplest setting the entries of the matrix \(X\) would be independent or even i.i.d.\ random variables. However, recent experimental data suggests that in reality \(X_{ij}\) and \(X_{ij}\) are not independent. One could therefore assume that for every \(i\neq j\), \[ \mathbb{E} X_{ij} X_{ji} = \rho \mathbb{E} |X_{ij}|^2 = \frac{\rho}{N}, \] where the complex number \(\rho\) is the correlation coefficient (the pairs \((X_{ij}, X_{ji})\) are still assumed to be independent for different index pairs \(\{i,j\}\)). Note that \(\rho=0\) in the independent case and \(\rho = 1\) if \(X=X^*\) is Hermitian. Depending on the value of the coupling parameter \(g\) the solution \(u\) typically grows or decays exponentially, but for a critical value it exhibits a power law decay with an exponent which depends on the symmetry properties of \(X\). The extreme cases \(\rho=0\) and \(\rho = 1\) were studied in a previous article of the authors [SIAM J. Math. Anal. 50, No. 3, 3271--3290 (2018; Zbl 1392.60011)]. In the paper under review, they focus on the remaining case \(0<|\rho| <1\) (some mathematically non-rigorous analysis of this intermediate case appears in the literature; here the authors present complete proofs). The results about the long time asymptotic behaviour of the above system of \(N\) linear differential equations are presented under a general assumption that \(X\) is an elliptic or elliptic-type random matrix. One of the main tools in the proofs is an asymptotically precise formula for \(\frac{1}{N}\mathbb{E} \operatorname{Tr}(f(X) g(X^*))\), where \(f\) and \(g\) are analytic functions. The paper contains 4 figures.

Recommendations

• Power law decay for systems of randomly coupled differential equations
• Large deviations and mean-field theory for asymmetric random recurrent neural networks
• Infinite order systems of differential equations and large scale random neural networks
• Stability of linear dynamic systems with random time-dependent parameters
• Asymptotic behavior of differential equations driven by periodic and random processes with slowly decaying correlations

Cites work

• A note on mixed matrix moments for the complex Ginibre ensemble
• A sharpened condition for strict log-convexity of the spectral radius via the bipartite graph
• Around the circular law
• Eigenvalues of non-Hermitian random matrices and Brown measure of non-normal operators: Hermitian reduction and linearization method
• Elliptic Law
• scientific article; zbMATH DE number 3901742 (Why is no real title available?)
• Inhomogeneous circular law for correlated matrices
• Local elliptic law
• Local inhomogeneous circular law
• Location of the spectrum of Kronecker random matrices
• Low rank perturbations of large elliptic random matrices
• Non-Hermitian random matrices with a variance profile. I: Deterministic equivalents and limiting esds
• Non-negative matrices and Markov chains.
• On statistics of bi-orthogonal eigenvectors in real and complex Ginibre ensembles: combining partial Schur decomposition with supersymmetry
• Operator-valued Semicircular Elements: Solving A Quadratic Matrix Equation with Positivity Constraints
• Power law decay for systems of randomly coupled differential equations
• Quadratic vector equations on complex upper half-plane
• Random matrices with slow correlation decay
• Random matrices: universality of ESDs and the circular law
• Scattering in quantum dots via noncommutative rational functions
• Stability of the matrix Dyson equation and random matrices with correlations
• Statistical properties of eigenvectors in non-Hermitian Gaussian random matrix ensembles
• The distribution of overlaps between eigenvectors of Ginibre matrices
• The Dyson equation with linear self-energy: spectral bands, edges and cusps
• The elliptic law
• The isotropic semicircle law and deformation of Wigner matrices
• Universality for general Wigner-type matrices

Cited in

(3)

• On the rightmost eigenvalue of non-Hermitian random matrices
• Mean left-right eigenvector self-overlap in the real Ginibre ensemble
• Mesoscopic spectral CLT for block correlated random matrices

This page was built for publication: Randomly coupled differential equations with elliptic correlations

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6165249)