Abstract
Molecular dynamics simulations of ionic solutions depend sensitively on the force fields employed for the ions. To resolve the fine differences between ions of the same valence and roughly similar size and in particular to correctly describe ion-specific effects, it is clear that accurate force fields are necessary. In the past, optimization strategies for ionic force fields either considered ...
Abstract
Molecular dynamics simulations of ionic solutions depend sensitively on the force fields employed for the ions. To resolve the fine differences between ions of the same valence and roughly similar size and in particular to correctly describe ion-specific effects, it is clear that accurate force fields are necessary. In the past, optimization strategies for ionic force fields either considered single-ion properties (such as the solvation free energy at infinite dilution or the ion-water structure) or ion-pair properties (in the form of ion-ion distribution functions). In this paper we investigate strategies to optimize ionic force fields based on single-ion and ion-pair properties simultaneously. To that end, we simulate five different salt solutions, namely, CsCl, KCl, NaI, KF, and CsI, at finite ion concentration. The force fields of these ions are systematically varied under the constraint that the single-ion solvation free energy matches the experimental value, which reduces the two-dimensional {sigma,epsilon} parameter space of the Lennard-Jones interaction to a one dimensional line for each ion. From the finite-concentration simulations, the pair potential is extracted and the osmotic coefficient is calculated, which is compared to experimental data. We find a strong dependence of the osmotic coefficient on the force field, which is remarkable as the single-ion solvation free energy and the ion-water structure remain invariant under the parameter variation. Optimization of the force field is achieved for the cations Cs(+) and K(+), while for the anions I(-) and F(-) the experimental osmotic coefficient cannot be reached. This suggests that in the long run, additional parameters might have to be introduced into the modeling, for example by modified mixing rules.
Altmetric