Eigenvalue spectral properties of sparse random matrices obeying Dale's law

Understanding the dynamics of large networks of neurons with heterogeneous connectivity architectures is a complex physics problem that demands novel mathematical techniques. Biological neural networks are inherently spatially heterogeneous, making them difficult to mathematically model. Random recurrent neural networks capture complex network connectivity structures and enable mathematically tractability. Our paper generalises previous classical results to sparse connectivity matrices which have distinct excitatory (E) or inhibitory (I) neural populations. By investigating sparse networks we construct our analysis to examine the impacts of all levels of network sparseness, and discover a novel nonlinear interaction between the connectivity matrix and resulting network dynamics, in both the balanced and unbalanced cases. Specifically, we deduce new mathematical dependencies describing the influence of sparsity and distinct E/I distributions on the distribution of eigenvalues (eigenspectrum) of the networked Jacobian. Furthermore, we illustrate that the previous classical results are special cases of the more general results we have described here. Understanding the impacts of sparse connectivities on network dynamics is of particular importance for both theoretical neuroscience and mathematical physics as it pertains to the structure-function relationship of networked systems and their dynamics. Our results are an important step towards developing analysis techniques that are essential to studying the impacts of larger scale network connectivity on network function, and furthering our understanding of brain function and dysfunction.