Optimizing Lennard-Jones parameters by coupling single molecule and ensemble target data
This contribution is a proof-of-concept that a diverse set of training observables leads to a meaningful force field even if a very limited number of thermodynamic state points (i.e. temperatures) is used. This approach generates optimized force-field parameters, enabling the user to extract additional information from MD simulations. The ultimate goal is to extend this approach and enable an increased amount of observables to be reproduced using a single force field for a series of chemically similar molecules (e.g. oligomer hydrocarbons). Specifically, we present a new optimization strategy to expand the limits of existing force fields and investigate how much added error is introduced to already reproducible observables. For this purpose, we optimized the Lennard-Jones parameters of n-octane using an isolated molecule's relative conformational energies and the liquid-phase density (293.15 K) for a molecular ensemble as optimization objectives. To test the impact on other observables, additional substances and temperatures that were not part of the training set were evaluated. This evaluation includes the surface tension, viscosity and density for n-hexane, n-heptane, n-octane and n-nonane at 293.15, 315.15 and 338.15 K. We show that it is possible to expand the limits of a force field, improving its overall accuracy at a small cost to its previously well-reproduced observables. Additionally, we propose approaches for further developments of the optimization strategy to increase the observables accuracies that suffer a loss in exchange for the capability of reproducing additional properties.