Exit versus escape in a stochastic dynamical system of neuronal networks explains heterogenous bursting intervals

TL;DR

This paper investigates the stochastic dynamics of neuronal networks, revealing that interburst intervals are better characterized as escape times from a basin of attraction, with implications for understanding burst duration variability.

Contribution

It introduces a mean-field model linking burst intervals to basin escape dynamics and analyzes how noise influences the distribution of escape times in neuronal systems.

Findings

Interburst durations do not match the time to reach the boundary.

Maximum of the probability density function is not at the attractor.

Escape times can involve multiple returns inside the basin before final escape.

Abstract

Neuronal networks can generate burst events. It remains unclear how to analyse interburst periods and their statistics. We study here the phase-space of a mean-field model, based on synaptic short-term changes, that exhibit burst and interburst dynamics and we identify that interburst corresponds to the escape from a basin of attraction. Using stochastic simulations, we report here that the distribution of the these durations do not match with the time to reach the boundary. We further analyse this phenomenon by studying a generic class of two-dimensional dynamical systems perturbed by small noise that exhibits two peculiar behaviors: 1- the maximum associated to the probability density function is not located at the point attractor, which came as a surprise. The distance between the maximum and the attractor increases with the noise amplitude $\sigma$, as we show using WKB approximation and numerical simulations. 2- For such systems, exiting from the basin of attraction is not sufficient to characterize the entire escape time, due to trajectories that can return several times inside the basin of attraction after crossing the boundary, before eventually escaping far away. To conclude, long-interburst durations are inherent properties of the dynamics and sould be expected in empirical time series.