Molecular simulation continues to drive scientific progress. In all simulations, one of the main challenges is how to get good computational models to describe the chemistry and physics of the system. For decades we have relied on physical insight to build these models, but even these often have computationally practical approximations. A new approach using models derived by machine learning has developed in the past decade. In this approach, we use large databases of calculations from density functional theory to build neural network models of atomistic systems.

You can see our work in these papers.

boes-2016-neural-networ

A comparison of a neural network potential with ReaxFF for gold

boes-2017-neural-networ

Using a neural network potential to model coverage dependent adsorption

boes-2017-model-segreg

Using a neural network potential to model segregation in an alloy

This is an exciting new area we will be publishing in for the next few years!

We are huge proponents of using org-mode to write scientific documents. This has enabled our group to develop novel data sharing strategies kitchin-2015-examp,kitchin-2015-data-surfac-scien and increase our productivity significantly.

We recently published this work on automated data sharing kitchin-2017-autom-data.

Alloy catalysts frequently do not have the same reactivity as the parent metals. We use density functional theory calculations to understand how the electronic structure of the alloy differs from the parent metals, and how those differences lead to different reactivities. We are especially interested in modeling heterogeneous site distributions boes-2015-estim-bulk, modeling XPS spectra of alloys boes-2015-core-cu, and the limit of single atom alloys tierney-2009-hydrog-dissoc.

We have used density functional theory to identify S-tolerant alloys inoglu-2011-ident-sulfur. Our current work will develop a new understanding of selectivity in acetylene hydrogenation.

PhD Jacob Boes: boes-2017-multis-model

We are generally interested in understanding the reactivity of metal oxides towards oxygen and fuels. We use computational tools to model the electronic structure and reactivity of metal oxides for applications in water splitting and chemical looping.

We derived a number of heuristic rules that describe the reactivity of perovskites akhade-2011-effec,akhade-2012-effec. These principles were expanded into a concept of outer electrons that was used to describe the reactivity of a broad range of oxides calle-vallejo-2013-number. These observations have been found to be robust across a broad range of assumptions in the models used, although some care must be taken in a few cases curnan-2014-effec-concen. We have found novel relationships between oxide electronic structure and reactivity for oxides with rocksalt structures xu-2014-relat.

We did some work in infiltrating solid oxide fuel cell electrodes with electrocatalysts to improve their activity chao-2011-prepar-mesop,chao-2012-struc-relat.

An outstanding challenge in modeling oxides is the highly correlated electrons especially in the 3d orbitals of transition metals. We have expanded an approach to use a linear response method to compute U, making a nearly predictive, first-principles method for computing oxide formation energies with pretty good accuracy xu-2015-accur-u.

One new direction we have taken is to identify oxide polymorphs that may be grown as epitaxial thin films mehta-2015-ident-poten. Subsequent work xu-2017-first-princ has shown some design principles for predicting epitaxial stabilization, including the successful synthesis of columbite SnO₂ wittkamper-2017-compet-growt.

PhD Robin Chao: chao-2012-improv-solid

We have worked in these research areas in the past, but are not currently actively working on them.