Scalable Fast Algorithms for Computational Quantum Chemistry using STRUMPACK

Development of novel molecular materials for catalysis and energy storage is enhanced by leveraging facts discovered through computer simulations. However, the errors inherent in even the best density functional theory (DFT) calculations can prevent making accurate predictions of molecular properties. Computational methods based on local coupled cluster (LCC) theory and developed in this project will improve the accuracy of computer simulation at reduced computational cost. During the Phase I proof-of-principle stage a Q-Chem-lead team will explore compressed representations of physical quantities involved in finding the solution to LCC equations, develop basic algorithms for evaluating LCC terms, and prototype computer software that implements the algorithms. In order to implement efficient LCC software, the team will adopt the high performance STRUMPACK numerical library developed with the support of the Department of Energy’s Office of Scientific Computing Research (ASCR). Specific areas of chemistry research that will benefit from the new tools include rational catalyst design, development of molecular materials, and applications of data science in chemistry. First, in the context of rational catalyst design, there will be interest in the greater predictive power that LCC can provide, which will reduce the errors inherent in even the best density functional theory (DFT) calculations. This will enable quantum chemistry simulations to better complement experiments in pursuit of rational catalyst design. Second, regarding molecular materials, the increased accuracy and feasibility of the LCC approach will permit validation of predictions of properties from DFT, and in some cases even replace them. Third, and more broadly, data science groups interested in reducing the cost of computational chemistry calculations using machine learning can only do as well as the data they learn on: hence such groups are likely to harness LCC methods to enhance the fidelity of ML-driven approaches to simulation in fields such as gas adsorption for clean energy, and biophysical simulations.