Preprint reports larger molecular simulations using adjustable Gaussian basis

Researchers report a density functional theory solver that simulated systems of up to 2,742 atoms and 10,406 electrons on a single four-GPU node at triple-zeta scale. Density functional theory estimates the electronic properties of molecules and materials; conventional solvers rely on fixed sets of mathematical functions centered on atoms. The proposed solver instead adjusts the positions, shapes and combinations of Gaussian functions to minimize energy. In the researchers' tests, it reached the accuracy of the largest conventional basis sets with fewer parameters. Its reported peak memory use grew quadratically with the number of Gaussian functions.