Dimensional reduction of dynamical systems by machine learning: automatic generation of the optimum extensive variables and their time-evolution map

Abstract

A framework is proposed to generate a phenomenological model that extracts the essence of a dynamical system (DS) with large degrees of freedom using machine learning. For a given microscopic DS, the optimum transformation to a small number of macroscopic variables, which is expected to be extensive, and the rule of time evolution that the variables obey are simultaneously identified. The utility of this method is demonstrated through its application to the nonequilibrium relaxation of the three-state Potts model.