Development of Transferable Force Fields

Over the next decade, it will become possible to gain microscopic-level insight into the behavior of complex chemical systems making use of theoretical advances and employing high-speed computational resources. Only with this molecular-based understanding will the research enterprise be able to develop chemicals, materials and processes that meet the increasing needs of society. Predicting phase equilibria and other thermophysical properties of multicomponent mixtures, given only the architecture of the molecules (types of atoms and their connectivity) and the experimental conditions, remains one of the grand challenges for the field of molecular simulation. The success of molecular simulation in this endeavor depends on the availability of efficient simulation algorithms and accurate force fields. For a long time, the available simulation techniques and the computer power were limiting progress. Over the last 15 years, however, enormous advances have been made in simulation methods and it is now becoming evident that more attention should be shifted to developing sufficiently accurate force fields. Already in 1982, Rowlinson and Swinton [Liquids and Liquid Mixtures (Butterworth Scientific: London, 1982)] wrote: "Both experiment and statistical theory have reached so high a degree of accuracy and sophistication that they put too great a strain on the weak link between them-our knowledge of intermolecular energies. It is improvement in this area that is needed next."

Force FieldsThe Siepmann research group is a pioneer in the development of transferable force fields and our TraPPE (transferable potentials for phase equilibria) force field is widely used in academia and industry. The term 'transferable' implies that the force field parameters for a given interaction site should be transferable between different molecules (e.g., identical parameters should be used for the methyl group in, say, n-hexane, 1-hexene, or 1-hexanol) and that the force field should be transferable to different state points (e.g., pressure, temperature, or composition) and to different properties (e.g., thermodynamic, structural, or transport). Thus, while a force field that uses special types of interaction parameters for specific molecules or special combining rules for specific unlike interactions might be very accurate for specific applications, it would not be considered transferable and would most likely have limited predictive power for unrelated applications. The figure shows a comparison of critical temperatures for some common organic molecules predicted using the TraPPE (circles) and OPLS (squares) force fields with the experimental data.

Beyond enabling accurate predictions of phase equilibria, the TraPPE force field philosophy has also resulted in force fields that excel in the prediction of structural quantities, providing important insights into experimentally observed phenomena. Pair distribution functions that depict the probability to find a specific pair of atoms at a given distance (normalized to the uniform distribution of pairs), play a central role in the description of liquids, but are difficult to determine experimentally. In a recent paper examining the structural properties of liquid water, the following statement was found, illustrating the experimental challenges: "... the proper extraction of the real-space pair correlation function from the experimental scattering is very difficult due to the uncertainty introduced in the experimental corrections, the proper weighting of OO, OH, and HH contributions, and numerical problems of Fourier transforming truncated data in Q-space...", and later, examining computational results: "...the resulting (pair) distribution functions from [TraPPE-pol] provide the current best benchmarks for real-space water structure over the biologically relevant temperature range studied here (275 to 350 K)" [G. Hura, et al., Phys. Chem. Chem. Phys. 5, 1981 (2003)].