Machine Learning for Intermolecular Interactions


Machine learning promises to provide high-quality predictions at tremendously reduced computational cost compared to standard quantum chemistry methods. While most applications of machine learning to chemistry involve single molecules, we are creating machine learning models for intermolecular interactions. This application is much more challenging because

(a) low-level computational chemistry methods are not necessarily suitable for training such a model, as they can exhibit large errors until high levels of theory are reached;

(b) most single-molecule machine learning descriptors use a short cutoff distance to determine an atom’s local environment, whereas in intermolecular interactions, the relevant distances can be long-range;

(c) it is not necessarily obvious what kinds of van der Waals complexes are most suitable for development of a training set of reference data. We have uncovered and solved a subtle problem in the most obvious approaches to training set generation [1] and also an apparent limitation in the standard atomic decomposition inherent in neural network potentials, [2] These issues, our solutions, and a status report on our creation of general intermolecular interaction models will be presented.

David Sherrill

David Sherrill

Professor - Chemistry Department, University of Georgia

David Sherrill he graduated from MIT with a B.S. in Chemistry in 1992. He held an NSF Graduate Research Fellowship working with Professor Fritz Schaefer at the University of Georgia, and he graduated with his Ph.D. in Chemistry in 1996. After working with Martin Head-Gordon as an NSF Postdoctoral Fellow at UC Berkeley, Dr. Sherrill joined the faculty of Georgia Tech in 1999. He was promoted to Associate Professor in 2005 and full Professor in 2008.

He is a joint faculty member between the School of Chemistry and Biochemistry and the School of Computational Science and Engineering. His research focuses on the development of new methods in electronic structure theory, their implementation as efficient computer programs, and their application to challenging chemical problems.

He is one of the lead principal investigators behind the Psi electronic structure program. He is one of the leading experts on theoretical studies of pi-pi, CH/pi, and other types of intermolecular interactions.

Wednesday, October 27, 2020
8:00 PM (KSA Time)
Organized by:
Chemistry Department - KFUPM