UC Berkeley

Broadly speaking, molecular simulations are used for two different purposes. First, zooming out, it allows for high-throughput screening of a much larger chemical space than is feasible by experimental work. Second, zooming in, it serves as a tool similar to a microscope, allowing scientists to understand atomic-level phenomena that underlie chemical properties. These two uses serve a variety of applications, from drug discovery to catalyst development to semiconductor processing. In this work, we illustrate the steps necessary to apply molecular simulations for its two purposes, using nanoporous materials as the example application.

First, we consider the selection of the molecular simulation method itself. Common methods are Monte Carlo and molecular dynamics (MD) simulations. Though the fundamentals of these methods have existed for well over half a century, techniques are still being developed to overcome the small system sizes and timescales to which we are limited. Thermostats are one such technique, frequently used in MD simulations to allow sampling of the canonical ensemble without requiring the large computational expense of simulating a heat bath. Many thermostats have been proposed, and several are in common use. The computational chemist’s choice and parameterization of thermostat is not trivial. We have shown that some of the most common thermostats in use today do not sample their intended ensembles, and that this can bring about large errors in a simulation. Fortunately, we have also found that alternative thermostats exist which do not exhibit these errors, and we advise molecular simulation practitioners to use them.

Having selected the molecular simulation technique, a computational chemist that seeks to perform a high-throughput screening must also find a library of materials on which the technique can be performed. Some databases of synthesized nanoporous materials exist, such as the International Zeolite Association (IZA) database and the Computation-Ready Experimental Metal-Organic Framework database, and multiple databases of hypothetical materials are also available. However, if a novel material class is desirable for exploration, the library must be developed. We became interested in zeolite-templated carbons (ZTCs), a material class in which two-dimensional graphene sheets are assembled in a three-dimensional scaffold, but we found no computational library of ZTCs existed. We developed a Monte Carlo technique which generates a ZTC for a given zeolite template, and we performed this in silico synthetic procedure using zeolites taken from the IZA database and the hypothetical zeolite databases. We then found that we could use the mathematical concept of minimal surfaces to describe the ZTCs, and in so doing, we established a link between experimentally-known ZTCs and schwarzites, which had so far been purely hypothetical materials. Schwarzites are negatively-curved carbons, and with their establishment as experimentally-known materials, the triumvirate of two-dimensional nanocarbons (along with positively-curved fullerenes and nanotubes and flat graphene sheets) is completed.

With the simulation method and material library at hand, the computational chemist is ready to perform the high-throughput screening. One important application of nanoporous materials is for adsorptive separations, which can be more energy efficient than distillation. When designing adsorbents for particular separations, understanding how the molecular structure affects gas adsorption is important. We screened tens of thousands of hypothetical zeolites to fill in the gaps that remain in our understanding of the natural gas purification process. Through this screening, we were able to find which adsorbent properties were most correlated with the adsorbent’s separation performance. Furthermore, we were able to test the validity of the commonly-used Ideal Adsorbed Solution Theory (IAST) to predict mixture isotherms from pure-component data.

Finally, we examine a particular material in-depth, illustrating the second broad purpose of molecular simulations. Continuing with the theme of using nanoporous materials for separations, we studied the behavior of benzene and xylenes adsorbed in MOF-5, a prototypical metal-organic framework (MOF). We found that the adsorbates separated into liquid and vapor phases that extended over multiple unit cells of MOF-5. This result was surprising because condensation is not generally found in materials with pore size below 2 nm, as the confinement decreases the number of neighbors an adsorbate can interact with, suppressing the energetic benefit of the liquid phase over the more entropically-favorable vapor phase. However, the limiting pore size that had been found in prior studies assumed a one-dimensional capillary-like pore structure, whereas MOF-5 has a three-dimensional pore structure that less restricts the number of neighboring adsorbate molecules. Using NMR, our collaborators were able to find experimental evidence that further attested to our phase separation hypothesis.