 |
Center for Nonlinear Studies |
The Langevin equation is a stochastic differential equation commonly used in Molecular Dynamics for simulating molecular systems in a canonical ensemble. However, the temporal discretization of this equation for coarse-grained (CG) models may necessitate parameter choices outside of the classical range for which many current time-stepping methods were developed. For this reason, it is of importance to understand how existing Langevin integrators perform on non-trivial CG molecular systems. In particular, we are concerned with how large of an integration time step can be used without introducing unacceptable amounts of error into quantities of interest. From the observations made here, we then develop a new class of numerical schemes that are designed to perform better in this extended parameter regime by reproducing key statistical physics quantities exactly in the case of a free particle and harmonic oscillator. One particular scheme from this class is chosen and then tested on atomistic and coarse-grained model systems representative of materials science and biological applications. The results show that this scheme yields at least comparable performance to existing methods, and for some test cases significant improvements in accuracy and/or numerical stability.
Host: Kipton Barros