Abstract: By admixing a fraction of exact exchange (EXX), hybrid functionals alleviate the self-interaction error in semi-local density functional theory (DFT) and furnish a more accurate description of the electronic structure in systems throughout chemistry, physics, and materials science. However, the conventional reciprocal-space evaluation of the EXX interaction is cubic scaling and computationally demanding (typically 10x–1000x more expensive than semi-local DFT), thereby limiting the applicability of hybrid DFT in the study of complex condensed-phase systems. To overcome this limitation, we have developed an accurate, efficient, and robust linear-scaling (order-N) EXX approach that exploits the sparsity of the quantum mechanical exchange interaction in real space using a localized representation of the occupied orbitals. The resulting exxl method achieves linear scaling efficiency by leveraging two levels of computational savings (domain truncation and pair selection), accounts for system heterogeneity from the ground up via on-the-fly construction of bounded domains for each orbital, and allows for a priori error control using an automated (black-box) error self-alignment algorithm. In doing so, exxl enables large-scale hybrid DFT calculations with controllable accuracy and robust transferability across a wide range of condensed-phase systems with varying degrees of heterogeneity—including liquids and aqueous ionic solutions, solid-liquid interfaces, porous materials, bipolaronic molten salts, and doped semi-conductors—using a single, physically motivated, and system-independent convergence parameter. Recently, we have also combined the exxl algorithm with SCDM (Selected Columns of the Density Matrix), a robust non-iterative orbital localization scheme, and ACE (Adaptively Compressed Exchange), a low-rank operator approximation that reduces unnecessary full-rank EXX evaluations, to form the SeA approach (SeA = SCDM + exxl + ACE). Implemented in the PWSCF module of Quantum ESPRESSO, SeA exploits three levels of computational savings (domain truncation, pair selection, and reduced full-rank EXX evaluations) and enables robust hybrid DFT calculations with an overall time-to-solution comparable to semi-local DFT. With these developments, this work brings us one step closer to routinely performing high-throughput hybrid DFT calculations on large-scale and heterogeneous condensed-phase systems, as needed for materials screening and discovery, sampling large swaths of potential energy surfaces, and generating high-quality quantum mechanical data for machine learning applications.
About the speaker
Born and raised in Brooklyn, Robert A. DiStasio Jr. was the first member of his family to attend college. He earned a B.S. summa cum laude in Biology and Chemistry (double major) from Portland State University while working with Carl C. Wamser and the late George S. Hammond on the synthesis of porphyrins for use in organic solar cells. In 2009, he was awarded a Ph.D. in Theoretical Chemistry from the University of California, Berkeley for his work on local and canonical approximations in Møller-Plesset perturbation theory with applications to dispersion interactions under the guidance of Martin Head-Gordon. This was followed by Postdoctoral Research in Condensed Matter Physics at Princeton University, where he worked with Roberto Car, Salvatore Torquato, and Frank H. Stillinger, as well as Alexandre Tkatchenko and Matthias Scheffler (at the Fritz Haber Institute of the Max Planck Society in Berlin).
In 2015, DiStasio joined the faculty at Cornell University as an Assistant Professor in the Department of Chemistry and Chemical Biology, and was promoted to Associate Professor in 2022. His research group at Cornell focuses on the development, implementation, and application of novel methodologies that extend the frontiers of electronic structure theory in complex condensed-phase environments. DiStasio has given more than 100 seminars and colloquia worldwide, published more than 80 articles in peer-reviewed academic journals (Erdős number of 3), and is an active contributor to the Q-Chem and Quantum ESPRESSO software packages. He is the proud recipient of the Faculty Early Career Development (CAREER) Award from the National Science Foundation (NSF), a 2020 Sloan Research Fellowship from the Alfred P. Sloan Foundation, the 2020 American Chemical Society (ACS) OpenEye Outstanding Junior Faculty Award in Computational Chemistry, and a 2022 Machine Learning in the Chemical Sciences & Engineering Award from the Camille and Henry Dreyfus Foundation.
When not doing research, DiStasio enjoys teaching, studying chess theory, woodworking, and spending time with his wife (Laura), recently born son (Roman), and their puggle (Margot).
