Professor Kieron J. Burke of the University of California, Irvine is supported by an award from the Chemical Theory, Models and Computational Methods program in the Division of Chemistry to improve the accuracy and applicability of density functional calculations of electronic structure. Density functional theory (DFT) is a popular approach to computationally studying molecules and materials, allowing the design of new pharmaceuticals and materials using computers. The method has been used in more than 30,000 scientific papers each year and has resulted some impressive successes in predicting useful molecules. For example, it has been used to find a better catalyst for making ammonia, an important starting material in chemical processes producing plastics, textiles, fertilizers, dyes and other chemicals. DFT was also used to find the world's hottest superconductor, hydrogen sulfide under pressure. DFT is also used in many machine-learning applications in the physical sciences. However, even the best approximate density functionals, the basic component of DFT, have limited accuracy and ranges of applicability. Dr. Burke and his group are developing DFT methods that use machine learning to create new approximations in order to improve the computational results. He is also studying the origins of the approximations that underly DFT. The resulting improvements in DFT may have significant technological and economic impacts. Dr. Burke broadens understanding of DFT by hosting schools around the world and by training both undergraduate and graduate students in this important cross-disciplinary area of research.
Dr. Burke plans to improve DFT calculations in three distinct ways: One approach that is very novel (machine-learning), one very pragmatic and simple (density-corrected DFT), and one very old and deep (semiclassical approximations to functionals). Dr. Burke has pioneered the creation of new functionals with machine learning (ML). These functionals depend on the density everywhere in space, i.e., are radically different for the local and semilocal approximations that form the heart of most modern DFT exchange-correlation functionals. Dr. Burke is investigating approaches to broadening the applicability of ML density functionals. Regarding density-corrected DFT, he is evaluating functionals on Hartree-Fock densities. He uses a new method for the so-called semiclassical approximations which may lead to much more accurate kinetic energy and exchange energy functionals for realistic systems than any in use today.
This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.
