Self-learning Multiscale Simulation for Achieving High Accuracy and High Efficiency Simultaneously

摘要

We propose a new multi-scale molecular dynamics simulation method which canachieve high accuracy and high sampling efficiency simultaneously withoutaforehand knowledge of the coarse grained (CG) potential and test it for abiomolecular system. Based on the resolution exchange simulations betweenatomistic and CG replicas, a self-learning strategy is introduced toprogressively improve the CG potential by an iterative way. Two tests showthat, the new method can rapidly improve the CG potential and achieve efficientsampling even starting from an unrealistic CG potential. The resulting freeenergy agreed well with exact result and the convergence by the method was muchfaster than that by the replica exchange method. The method is generic and canbe applied to many biological as well as non-biological problems.