This method, which uses a combination of quantum mechanical calculations, data science, and machine learning, is called the Adsorbate Chemical-Environment-based Graph Convolution Neural Network (ACE-GCN) and was recently described in the journal Nature Communications.  The ACE-GCN approach addresses a fundamental challenge that arises in performing quantum simulations of heterogeneous catalysts: molecules that bind and react on the surfaces of these catalysts may be arranged in many thousands of possible configurations, and it is often necessary to calculate the properties of each of these molecular arrangements to determine which controls the catalyst’s behavior.  The extreme expense of these calculations, however, often exceeds the availability of computational resources in academic and industrial research groups.  ACE-GCN provides an efficient strategy to accelerate these studies, reducing the number of required calculations by a factor of approximately ten, and thus opening new opportunities to screen and design heterogeneous catalysts using accurate simulation methods.

ACE-GCN works by creating mathematical structures known as graphs from molecular geometries.  Each such graph is then associated with a particular energy that is calculated from quantum mechanics.  Next, using a technique known as a neural network, the relationship between the graphs and the energies can be determined (see figure).  This relationship can then be applied to many molecular geometries to rapidly estimate their properties. 

Jeffrey Greeley and his team have demonstrated how ACE-GCN can be applied to study multiple chemical reactions related to electrochemistry and energy science.  In one such reaction, they show how the method can be used to analyze electrocatalytic techniques for removing nitrogen contaminants from agricultural runoff streams, and in a second example, they explore how ACE-GCN can be used to accelerate the computational study of electrocatalysts for low temperature polymer electrolyte fuel cells. 

The work was funded and supported by United States Department of Energy through the Office of Science, Office of Basic Energy Sciences (BES), Chemical, Biological, and Geosciences Division, Data Science Initiative, grant DE-SC0020381.  Following completion of the work, Pushkar Ghanekar began a position as a research scientist for Eli Lilly and Company, and Siddharth Deshpande started post-doctoral research at the University of Delaware.