(493e) Pioneering a New Paradigm in Exchange-Correlation Functional Design

Density functional theory (DFT) is the workhorse method of computational catalysis owing to its balance of speed and accuracy: it is fast enough to study systems relevant to catalysis (100+ atoms), and accurate enough (~0.2 eV error) to make quantitative conclusions about trends in reactivity. The core approximation in DFT is the exchange correlation functional, which maps the exact many-electron system to an approximate system of non-interacting electrons. A plethora of exchange-correlation functionals have been developed over the past five decades, many of which provide excellent accuracy for specific types of systems (e.g. metals, molecules, etc.). However, there are tradeoffs between speed and accuracy for different levels of approximation, and there is no consensus on the "best overall" functional, especially for the solid-gas/liquid interfaces relevant to heterogeneous catalysis. Moreover, nearly all functionals developed thus far follow Pardew's paradigm of "Jacob's Ladder", where the inputs to the functional are based on a limited selection of possible inputs (e.g. density, density gradient, kinetic energy density, Fock exchange energy). This talk introduces a fundamentally different paradigm of exchange-correlation functional design, where the input space is expanded to include spatially-localized multipole expansions of the electron density. The basic idea of multipole functionals will be introduced, along with some strategies that have been employed in constructing specific proof-of-concept functionals, including machine learning and data-driven approaches. The current status of self-consistent multipole functionals will be presented, along with a brief history of how a textile/chemical engineer ended up developing exchange-correlation functionals, and a perspective on the future challenges and prospects for this new class of density functionals.