While molecular dynamics (MD) and Monte Carlo (MC) simulations have become an established method in the study of systems such as proteins, drug targets or membranes, the quality of their predictions do strongly depend on the validity of underlying physical assumptions. Lack of physical validity in simulations occurs from two main sources, the user and the simulation program. Users may choose wrong or incompatible parameters or algorithms, while coding errors within the program may render otherwise valid simulations unphysical. Testing of fundamental physical assumptions, such as correct integration, sampling of Boltzmann ensembles, or ergodicity between degrees of freedom, allows to reduce the risk of errors from both sources, thereby significantly increasing the robustness of molecular simulations.<br/><br/>We have published an open-source python package (https://physical-validation.readthedocs.io) allowing users to read in simulation results from a number of different MD programs (GROMACS, LAMMPS, HOOMD, AMBER, GROMOS) and test these results for physical validity. These tests allow users to verify their respective systems and setups. Based on these tests, we will discuss the validity of frequently used MD simulation techniques, such as thermostat and barostat algorithms, thermostat coupling schemes, hydrogen-mass repartitioning, and various MD and MC/MD-hybrid integrators.<br/><br/>We also propose to include physical validation in code correctness tests of simulation packages. We have implemented such physical sanity checks for the GROMACS simulation package, and every major release since GROMACS 2018 has passed these tests on selected systems before shipping. We will discuss this approach, and showcase some errors caught with this new testing approach.

Physical validity in molecular simulations

2019 Fall National Meeting

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.