| Contents |
| The simulation of large molecular systems can take advantage of their inherent multi-scalar nature. The different components of the interaction potentials used in molecular dynamics vary greatly both in the spatial and temporal range. Force splitting or multiple-time-stepping methods improve the computational efficiency by using different integration step-sizes for updating each component, i.e. longer step-sizes for the slower, long-range interactions. This allows for longer simulated times at the expense of introducing numerical instabilities due to resonances in the slow forces.
Here we present a multi-time-stepping method adapted to hybrid Monte Carlo, MTS-GSHMC. The new method combines the high sampling efficiency and stability of GSHMC with a force splitting method that includes the mollification of the slow forces. For benchmarking it has been implemented on the open-source software ProtoMol and tested on a water droplet and protein systems. Initial results show improved performance compared to the similar best performed multi-step method through achieving larger time-steps and enhanced sampling efficiency without compromising the accuracy of dynamical measurements as well as the numerical stability. |
|