Kinetic energy in random recurrent neural networks

Abstract: The relationship between unstable fixed points and chaotic dynamics in high-dimensional neural dynamics remains elusive. In this work, we investigate the kinetic energy distribution of random recurrent neural networks by combining dynamical mean-field theory with extensive numerical simulations. We find that the average kinetic energy shifts continuously from zero to a positive value at a critical value of coupling variance (synaptic gain), with a power-law behavior close to the critical point. The steady-state activity distribution is further calculated by the theory and compared with simulations on finite-size systems. This study provides a first step toward understanding the landscape of kinetic energy, which may reflect the structure of phase space for the non-equilibrium dynamics.