Neural network wavefunctions optimized using the variational Monte Carlo method have been shown to produce highly accurate results for the electronic structure of atoms and small molecules, but the high cost of optimizing such wavefunctions prevents their application to larger systems. We propose the Subsampled Projected-Increment Natural Gradient Descent (SPRING) optimizer to reduce this bottleneck. SPRING combines ideas from the recently introduced minimum-step stochastic reconfiguration optimizer (MinSR) and the classical randomized Kaczmarz method for solving linear least-squares problems. We demonstrate that SPRING outperforms both MinSR and the popular Kronecker-Factored Approximate Curvature method (KFAC) across a number of small atoms and molecules. We will also discuss recent progress in understanding anti-symmetric neural architectures for fermionic wave functions.
Biography:
Lin Lin is currently a Professor in the Department of Mathematics at UC Berkeley, and a Senior Faculty Scientist at Lawrence Berkeley National Laboratory. His research centers on solving quantum many-body problems by employing both classical and contemporary methods with applications across various domains, including quantum chemistry, quantum physics, materials science, and quantum information theory. He is a recipient of the Sloan Research Fellowship, the National Science Foundation CAREER award, the Department of Energy Early Career award, the SIAM Computational Science and Engineering (CSE) early career award, and the Presidential Early Career Awards for Scientists and Engineers (PECASE), the ACM Gordon Bell prize (Team), and the Simons Investigator in Mathematics award.
Cory 373