Multilayer directed random networks: Scaling of spectral properties

G. Tapia-Labra,M. Hernández-Sánchez,J. A. Méndez-Bermúdez

2024-09-26

Abstract:Motivated by the wide presence of multilayer networks in both natural and human-made systems, within a random matrix theory (RMT) approach, in this study we compute eigenfunction and spectral properties of multilayer directed random networks (MDRNs) in two setups composed by $M$ layers of size $N$: A line and a complete graph (node-aligned multiplex network). First, we numerically demonstrate that the normalized localization length $\beta$ of the eigenfunctions of MDRNs follows a simple scaling law given by $\beta=x^*/(1+x^*)$, with $x^*\propto (b_{\rm eff}^2/L)^\delta$, $\delta\sim 1$ and $b_{\rm eff}$ being the effective bandwidth of the adjacency matrix of the network of size $L=M\times N$. Here, $b_{\rm eff}$ incorporates both intra- and inter-layer edges. Then, we show that other eigenfunction and spectral RMT measures (the inverse participation ratio of eigenfunctions, the ratio between nearest- and next-to-nearest-neighbor eigenvalue distances, and the ratio between consecutive singular-value spacings) of MDRNs also scale with $x^*$.

Disordered Systems and Neural Networks,Data Analysis, Statistics and Probability

What problem does this paper attempt to address?

The problem that this paper attempts to address is the scaling behavior of the characteristic function and spectral properties of Multilayer Directed Random Networks (MDRNs). Specifically, the authors investigate the scaling behavior of the localization length of the characteristic function of multilayer directed random networks using the Random Matrix Theory (RMT) approach. They find that this length follows a simple scaling law \( \beta = \frac{x^*}{1 + x^*} \), where \( x^* \propto \left(\frac{b_{\text{eff}}^2}{L}\right)^\delta \), \( \delta \approx 1 \), and \( b_{\text{eff}} \) is the effective bandwidth of the network's adjacency matrix.

In addition, the paper also demonstrates that several other characteristic functions and spectral RMT metrics (such as the inverse participation ratio, the ratio of the nearest neighbor to the next-nearest neighbor eigenvalue distances, and the ratio of consecutive singular value spacings) also follow similar \( x^* \) scaling behavior. These results are applicable not only to multilayer network structures but also to multiplex network structures with node alignment. These findings contribute to a better understanding of the localization phenomenon of characteristic functions in complex systems and their related transport properties.