CAREER: Tensor Factorization Methods for High-Level Electronic Structure Theory

With support from the Chemical Theory, Models and Computational Methods program in the Division of Chemistry, Devin Matthews of Southern Methodist University will endeavor to advance the development of complex quantum chemical Couple Cluster methods to enable practitioners of theory to model with very high accuracy much larger molecular systems than has been possible heretofore. Novel state-of-the-art algorithms and GPU (Graphics Processing Unit)-implementations will enable these high-accuracy predictions of parameters such as molecular heats of formation, reaction barrier heights, bond dissociation, ionization, and electron attachment energies.

These developments will have a widespread impact on chemical modelling and reliable benchmarks, e.g. through hierarchical reaction models and the Active Thermochemical Tables. The new “model chemistry” developed by Dr. Matthews and his group has the potential to impact a wide range of problems, including the development of novel renewable fuels, a better understanding of our effects on the atmosphere and our environment, and the formation of complex biological molecules in space and other harsh environments which led to the origins of life.

The Matthews group will also endeavor to improve awareness of and diversity in theoretical and computational chemistry by directly involving undergraduates and high school students in cutting-edge research and incorporating computational chemistry into the undergraduate physical chemistry curriculum. The Matthews group will also implement an interactive “knowledge web” spanning diverse topics in quantum chemistry and related fields.

This “Wikipedia for Quantum Chemistry” will help to improve and accelerate training of graduate students in the field as well as interested non-scientists.

Dr. Matthews and his group will be working on enabling highly accurate coupled cluster calculations on larger molecules by developing reduced-scaling approximations of CCSDT (Coupled Cluster Single-Double-Triple) theory methods and CCSDT(Q), as well as analytic gradients thereof. Graph-based techniques and knowledge-based algorithmic search through the Design-by-Transformation methodology will be used to produce optimal working equations, and a high-quality implementation will be produced by utilizing automated code generation that targets high-level tensor contraction interfaces for CPUs (Central Processing Units) and potentially GPUs.

To this will be added further optimizations such as coarse-grained parallelism, blocking, and loop fusion. The efficacy of this implementation in terms of computational efficiency, scaling, absolute and relative error characteristics, and error cancellation will be leveraged to create a new model chemistry suitable for accurate thermochemistry (at the ~1kJ/mol scale) of molecules with as many as 12 first- or second-row atoms.

The Matthews group will developing this new protocol to study substituted Criegee intermediates and possible water-catalysis of Criegee decomposition as well as oxygenated fuel combustion intermediates relevant to renewable fuel development. Dr. Matthews’s group will also work to achieve a more diverse student base in theoretical and computational chemistry by directly involving undergraduates and high school students from diverse backgrounds in cutting-edge research and incorporating computational chemistry into the undergraduate physical chemistry curriculum.

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.