Heterogeneous populations of quadratic integrate-and-fire neurons: on the generality of Lorentzian distributions

Abstract: Over the last decade, next-generation neural mass models have become increasingly prominent in mathematical neuroscience. These models link microscopic dynamics with low-dimensional systems of so-called firing rate equations that exactly capture the collective dynamics of large populations of heterogeneous quadratic integrate-and-fire (QIF) neurons. A particularly tractable type of heterogeneity is the distribution of the QIF neurons' excitability parameters, or inputs, according to a Lorentzian. While other distributions -- such as those approximating Gaussian or uniform distributions -- admit to exact mean-field reductions, they result in more complex firing rate equations that are challenging to analyze, and it remains unclear whether they produce comparable collective dynamics. Here, we first demonstrate why Lorentzian heterogeneity is analytically favorable and, second, identify when it leads to qualitatively different collective dynamics compared to other types of heterogeneity. A stationary mean-field approach enables us to derive explicit expressions for the distributions of the neurons' firing rates and voltages in macroscopic stationary states with arbitrary heterogeneities. We also explicate the exclusive relationship between Lorentzian distributed inputs and Lorentzian distributed voltages, whose width happens to coincide with the population firing rate. A dynamic mean-field approach for unimodal heterogeneities further allows us to comprehensively analyze and compare collective dynamics. We find that different types of heterogeneity typically yield qualitatively similar dynamics. However, when gap junction coupling is present, Lorentzian heterogeneity induces nonuniversal behavior, obscuring a diversity-induced transition to synchrony.