Accelerating Variable Cell Shape Molecular Dynamics with a Position-Dependent Mass Matrix

In molecular dynamics (MD), the accessible time scales are limited by the necessity to choose sufficiently small time steps so that the fastest vibrations of the system can still be modeled. Mass tensor molecular dynamics (MTMD) aims to increase the time step by augmenting the Hamiltonian with a position-dependent mass matrix. Higher masses are assigned to modes with fast vibrations. These modes are identified by using an approximate Hessian matrix. The approximate Hessian matrix presented in this paper is applicable to the simulation of molecular systems, where no changes in the bonding pattern occur. We have adapted the MTMD method to variable cell shape systems and present a suitable symplectic integrator. The efficiency of the method is demonstrated for a system of molecular crystals consisting of <i>N</i>-(4-Methylbenzylidene)-4-methylaniline, where we could sample transitions between two polymorphs and thereby increase the time step by a factor of 4.4 to speed up the simulation. We have also simulated liquid water at the density function theory level, where we have achieved an acceleration by a factor of 2.8.