Capturing symmetric correlation for Fermionic systems with a Jastrow factor based on Spectral Neighbor Analysis Potentials

Quantum Monte Carlo methods offer a route to efficiently and accurately account for electron correlation. QMC methods rely on a trial wave function to guide the importance sampling to allow for efficient sampling yielding low statistical variance. A typically utilized form of the trial wave function is the Slater-Jastrow wave function, where the Jastrow factor accounts for symmetric, dynamic correlations between particles. By capturing these correlations, the Jastrow factor reduces the mixed estimator bias, reduces errors introduced by the use of effective core potentials, and increases the efficiency of diffusion Monte Carlo simulations due to the reduction in variance. Motivated by recent works demonstrating the flexibility of machine learned wave functions, we propose a form of the Jastrow factor inspired by machine-learned interatomic potentials (ML-IAP). In the ML-IAP field, the Spectral Neighbor Analysis Potential (SNAP) utilizes a descriptor bispectrum components to capture the local atomic environment. As a Jastrow factor, the SNAP representation is flexible, enables rapid evaluation, and capable of capturing the dynamic instantaneous interactions in a system that manifest as higher-order many-body correlations. We present an initial assessment of the performance of the SNAP Jastrow. SNL is managed and operated by NTESS under DOE NNSA contract DE-NA0003525