Position-momentum conditioning, relative entropy decomposition and convergence to equilibrium in stochastic Hamiltonian systems (2312.09475v1)

Abstract: This paper is concerned with a class of multivariable stochastic Hamiltonian systems whose generalised position is related by an ordinary differential equation to the momentum governed by an Ito stochastic differential equation. The latter is driven by a standard Wiener process and involves both conservative and viscous damping forces. With the mass, diffusion and damping matrices being position-dependent, the resulting nonlinear model of Langevin dynamics describes dissipative mechanical systems (possibly with rotational degrees of freedom) or their electromechanical analogues subject to external random forcing. We study the time evolution of the joint position-momentum probability distribution for the system and its convergence to equilibrium by decomposing the Fokker-Planck-Kolmogorov equation (FPKE) and the Kullback-Leibler relative entropy with respect to the invariant measure into those for the position distribution and the momentum distribution conditioned on the position. This decomposition reveals a manifestation of the Barbashin-Krasovskii-LaSalle principle and higher-order dissipation inequalities for the relative entropy as a Lyapunov functional for the FPKE.