Materials Modeling Optimization

Dallas Trinkle, University of Illinois at Urbana-Champaign

Usage Details

To develop new collaborations between materials modeling and NCSA, I propose to work with NCSA as an NCSA Fellow to move our proof-of-principle empirical potential optimization algorithm up to large scale parallelization, and integrate with the Materials Data Facility. Empirical potentials are a basic element of predictive multiscale modeling from electronic structure up through the atomic scale and to the mesoscale; the development of accurate, automated, robust empirical models of atomic interaction for materials simulation will have significant impact in Materials Science, Physics, and Chemistry, and potentially even Biology for the simulation of large biomolecules. By scaling up our new algorithm, we can provide a valuable tool to a wide swath of the research community, while leveraging a novel use of Materials Data Facility data.

This effort is designed to strengthen connections between materials modeling across campus and NCSA for future center proposals, as multiscale modeling has been a focus of several center proposals that I have led or co-led from University of Illinois with NCSA. This work connects with my recently awarded Strategic Research Initiative (SRI) grant from the College of Engineering (co-PI with Harley Johnson and David Ceperley) on “Building the Next UIUC Center for Computational Materials,” of which NCSA will be a key partner. The central computational infrastructure piece necessary for my proposed work is the development of a massively parallel evaluation engine for empirical potentials that is optimized to work with a large set of parameters and evaluate energies and forces for a set of structures; this would transform the optimization of empirical potentials into a “Big Data” problem, and opens up a wide array of robust optimization techniques. Moreover, we could consider ternary, quarternary, even quinternary chemical systems: impacting everything from advanced structural materials, high-entropy alloys, complex nanostructured systems, and more.