Statistical physics, molecular modeling and simulation, software engineering

David A. Kofke, Professor

Molecular simulation is a method of physical inquiry in which the properties of model materials—defined in terms of their intermolecular interactions only—are studied to “measure” the bulk physical behaviors they exhibit. Our research aims to improve the ability of molecular simulation to do this, while also applying molecular simulation to understanding the behavior of systems that are interesting and of practical importance. Specific research areas targeted by our group are as follows.

Understanding free-energy calculations

The free-energy perturbation (FEP) method is a core technique in computational chemistry. Despite its importance and wide use, it is frequently applied in ways that are inefficient at best, and incorrect at worst. FEP calculations can be performed in either of two directions, depending on which system is taken as the reference. It is well known that these two calculations lead to results that differ systematically. One mistake commonly made is to assume that both calculations are equally wrong, and the best result is obtained by splitting the difference. By applying modeling concepts to the simulation process itself, we have shown that this heuristic is incorrect, and we are formulating a perspective on these calculations that enables them to be applied much more reliably and efficiently. Because free energy is the key to understanding any thermophysical behavior, this work can impact a very broad range of applications. Examples include pharmaceutical formulation and manufacturing, catalysis design, nanotechnology, biochemistry, and many more such fields in which applications of molecular modeling are at the leading edge of research and development.

Evaluation of cluster integrals by molecular simulation

An interesting and exciting extension of our work on free energy methods considers application of molecular simulation to evaluation of so-called “cluster integrals”. These quantities capture the elementary contributions of interactions between two molecules taken alone, three molecules, four, etc . and through a well established theoretical formalism permits them to be combined to yield properties of the bulk phase of 10²³ molecules! A limiting element of this formalism has lied in the inability to evaluate the basic integrals arising in the development. Using ideas from our free-energy work, we have made large advances toward the resolution of this problem. Our larger aim in this work is to permit simulation to yield useful results while focusing on the behaviors of just a few molecules at once. This could be viewed as a powerful way to parallelize the calculations involved in evaluating bulk properties from molecular models.

The above describe our activities to improve the methodology of simulation. In addition we have activities applying molecular simulation to understand complex phenomena.

Effect of entropy on polymorphism

Prediction of stable crystalline polymorphs should be completed by examining the relative free energies of candidate structures. In practice this is almost always done by examining the lattice energy alone, ignoring entropic contributions to the free energy. The error incurred with this procedure is generally considered to be of the order of the difference between candidate structures, but nevertheless the practice persists because it is inconvenient to consider the entropy rigorously, and approximate methods that introduce entropy tend not to lead to greater success in predicting the observed polymorphs. We are performing rigorous calculations of the free energies of crystalline polymorphs of a variety of molecules. We use these results to ascertain the true importance of entropy in polymorph stability, determine if widely used approximations are justified, and examine if new methods can be applied instead.

Object-oriented programming in molecular simulation

This is less a research than a development project driven by an interest in applying molecular simulation as a teaching tool. We are investigating the feasibility of using molecular simulation teaching tools as a basis for constructing extensible, object-oriented research-quality molecular simulation codes. Etomica is the name of the API and development environment that we are constructing. We now have a page devoted to describing and disseminating it: http://www.etomica.org/ In our latest work we are attempting to extend the design developed in Etomica to formulate a general-purpose API for molecular simulation, which can facilitate interfacing with a variety of simulation code bases.