Quantitative predictions from molecular simulations using explicit or implicit interactions

Abstract

AbstractEquilibrium simulations of molecular systems allow to extract many physicochemical properties. Given an “accurate enough” model, a “large enough” simulation system and “long enough” simulations, such calculations should yield accurate predictions of properties that can be tested by experimental measurements. Non‐equilibrium simulations can be used as a tool to obtain specific properties like viscosity or conductivity, but they have the drawback that in general only one property per simulation is produced. In addition, a range of methods is available for computing free energy differences. We here review the state of the art of using classical simulation models for generating quantitative predictions. Popular force fields have significant predictive power already but there is room for improvement. Bonded force potentials may need to be replaced by more accurate ones to better reproduce vibrational frequencies. Simplification of non‐bonded force terms, such as cut‐offs for electrostatic or dispersion interactions, should be avoided. Routine usage of force field methods will therefore require some tuning of parameters. Despite the extensive toolbox that is available for producing quantitative results, the computational cost of explicit solvent simulation is significant and therefore, approximate methods like implicit solvent models remain popular and are still being developed. Based on fundamental arguments as well as on examples of solvation free energies, host–guest complexation and non‐covalent association of molecules in solution, we conclude that implicit solvents as well as algorithmic simplifications are most useful when validation using experimental data or rigorous theoretical treatments is possible.This article is categorized under:Molecular and Statistical Mechanics > Molecular Dynamics and Monte‐Carlo MethodsTheoretical and Physical Chemistry > Statistical MechanicsSoftware > Simulation Methods