Simulations of state-of-the-art fermionic neural network wave functions with diffusion Monte Carlo

Abstract

Recently developed neural network-based \emph{ab-initio} solutions (Pfau et. al arxiv:1909.02487v2) for finding ground states of fermionic systems can generate state-of-the-art results on a broad class of systems. In this work, we improve the results for this Ansatz with Diffusion Monte Carlo. Additionally, we introduce several modifications to the network (Fermi Net) and optimization method (Kronecker Factored Approximate Curvature) that reduce the number of required resources while maintaining or improving the modelling performance. In terms of the model, we remove redundant computations and alter the way data is handled in the permutation equivariant function. The Diffusion Monte Carlo results exceed or match state-of-the-art performance for all systems investigated: atomic systems Be-Ne, and the carbon cation C${}^{+}$.

References in corpus (1)

• Computational complexity and fundamental limitations to fermionic quantum Monte Carlo simulations

Cited by in corpus (5)

• Ab-initio study of interacting fermions at finite temperature with neural canonical transformation
• PyQMC: an all-Python real-space quantum Monte Carlo module in PySCF
• Assessing and Advancing the Potential of Quantum Computing: A NASA Case Study
• Second-order optimisation strategies for neural network quantum states
• Ab-Initio Potential Energy Surfaces by Pairing GNNs with Neural Wave Functions