Latent Spaces for Langevin Dynamics

Abstract: In the field of machine learning coarse-grained potentials in molecular dynamics, propagators require that the Hamiltonian is quadratic in momentum, thus limiting the family of coarse-graining functions. In this paper, we derive a general family of coarse-graining embedding functions for which Langevin dynamics can be applied. This has significant implications in molecular simulations, and it paves the way for Langevin dynamics to be run on non-geometric coarse-graining representations such as those given by principal components of time-lagged independent component analysis (TICA) or latent embeddings of molecules obtained from neural networks.