ML-GLE: A machine learning enhanced Generalized Langevin equation framework for transient anomalous diffusion in polymer dynamics

In this work we introduce ML-GLE, a machine learning framework to generate long-term single-polymer dynamics by exploiting short-term trajectories from molecular dynamics (MD) simulations of polymer melts. Even with current advances in machine learning for MD, these polymeric materials remain difficult to simulate and characterize due to prohibitive computational costs when long relaxation timescales are involved. Our method relies on a 3D neural auto-regressive model for single polymer lower dimensional collective variables, called normal modes. This enhances the Generalized Langevin Equation (GLE) capabilities in modelling diffusion phenomena. We exploit a particular GLE solution which is known to reproduce the mean square displacement curve relative to transient anomalous diffusion and connect it with the normal modes collective variables. ML-GLE is capable of emulating the single polymer statistical properties in the long-term, predicting the diffusion coefficient. As a consequence, this results in an enormous acceleration in terms of simulation time with respect to the full-size simulation. Moreover, this approach is also scalable with system size. •Long-term polymer dynamics in molecular dynamics simulations requires a substantial computational cost.•The Generalized Langevin equation can be used to model single polymer diffusion properties.•Neural autoregressive models trained on short trajectories emulate and extrapolate single polymer chain long-term dynamics.