Transferable next-generation force fields from simple liquids to complex materials.

Abstract

Molecular simulations have had a transformative impact on chemists' understanding of the structure and dynamics of molecular systems. Simulations can both explain and predict chemical phenomena, and they provide a unique bridge between the microscopic and macroscopic regimes. The input for such simulations is the intermolecular interactions, which then determine the forces on the constituent atoms and therefore the time evolution and equilibrium properties of the system. However, in practice, accuracy and reliability are often limited by the fidelity of the description of those very same interactions, most typically embodied approximately in mathematical form in what are known as force fields. Force fields most often utilize conceptually simple functional forms that have been parametrized to reproduce existing experimental gas phase or bulk data. Yet, reliance on empirical parametrization can sometimes introduce limitations with respect to novel chemical systems or uncontrolled errors when moving to temperatures, pressures, or environments that differ from those for which they were developed. Alternatively, it is possible to develop force fields entirely from first principles, using accurate electronic structure calculations to determine the intermolecular interactions. This introduces a new set of challenges, including the transferability of the resulting force field to related chemical systems. In response, we recently developed an alternative approach to develop force fields entirely from first-principles electronic structure calculations based on intermolecular perturbation theory. Making use of an energy decomposition analysis ensures, by construction, that the resulting force fields contain the correct balance of the various components of intermolecular interaction (exchange repulsion, electrostatics, induction, and dispersion), each treated by a functional form that reflects the underlying physics. We therefore refer to the resulting force fields as physically motivated. We find that these physically motivated force fields exhibit both high accuracy and transferability, with the latter deriving from the universality of the fundamental physical laws governing intermolecular interactions. This basic methodology has been applied to a diverse set of systems, ranging from simple liquids to nanoporous metal-organic framework materials. A key conclusion is that, in many cases, it is feasible to account for nearly all of the relevant physics of intermolecular interactions within the context of the force field. In such cases, the structural, thermodynamic, and dynamic properties of the system become naturally emergent, even in the absence of explicit parameterization to bulk properties. We also find that, quite generally, the three-body contributions to the dispersion and exchange energies in bulk liquids are crucial for quantitative accuracy in a first-principles force field, although these contributions are almost universally neglected in existing empirical force fields.