The performance of a Jastrow factor utilizing the Spectral Neighbor Analysis Potential descriptor for atoms, molecules, and solids.

Motivated by recent works demonstrating the flexibility of machine learned wave functions for Quantum Monte Carlo, we have implemented a Jastrow factor inspired by the Spectral Neighbor Analysis Potential (SNAP), J_SN. In general, a Jastrow factor offers a description of dynamic instantaneous many-body correlations between particles, the inclusion of which reduces errors introduced by the use of effective core potentials and increases the efficiency of diffusion Monte Carlo simulations. J_SN utilizes the form of SNAP which itself is a machine-learned interatomic potentials that utilizes a basis of bispectrum components to describe the local atomic environment. As a Jastrow factor, J_SN, the bispectrum descriptors offer flexibility, enable rapid evaluation, and can describe high-order correlations. We demonstrate enhanced performance in atomic carbon, molecular H$_2$O, solid hexagonal boron nitride, though the improvements are not uniform. Additionally, new implementation details will be highlighted.

SNL is managed and operated by NTESS under DOE NNSA contract DE-NA0003525.