Uncertainty quantification for classical molecular dynamics
Molecular dynamics simulations represent the veritable workhorse in atomistic simulations of the widest range of problems imaginable. Dassault Systemes BIOVIA’s customers use MD daily in both industrial and academic research. We see it applied with great success in life science projects (protein modeling, drug discovery, cell signaling, membrane permeability, etc.) and in advanced materials simulations (polymer network formation in composites, mass and charge transport in battery electrolytes, etc.). Classical MD is capable of simulating huge systems over long trajectories; ab initio MD is significantly more limited in both system size and the length of the simulation. In either case, however, one is left with the task of estimating the accuracy of the final prediction, be it a physical property (e.g., transport coefficients) or an observable process (ligand binding or SEI formation on Li-ion battery electrode). The question becomes increasingly relevant as scientists who perform the calculations interact more with engineers – in the context of for example battery cell design or mechanical testing and aging of polymers. Engineering applications in the virtual+real world require accurate values with quantified uncertainties.
There are numerous approaches to the MD uncertainty quantification, as explained e.g. in a recent review. The challenge is to find the method most appropriate for a given problem. For example, classical MD estimation of the diffusion coefficient in an electrolyte could pose different issues to that of the fracture test of a polymer. Another angle to the challenge is to investigate the size and time length effect on the conclusions. This comes from the desire to perform, at least for some applications, more expensive ab initio MD simulations. A recipe learned from the classical MD would save a lot of computational effort in the AIMD case.
The challenge then is to recommend the most appropriate strategy for UQ and validation for two problems: (i) classical MD modeling of transport in electrolyte, with estimates of the potential system size and simulation length impact on the suggested strategy, (ii) static and dynamic mechanical properties of rubber materials. The simulations can be performed either using BIOVIA Materials Studio Forcite engine, or any other classical MD engine of your choice.a