Development of Hybrid Potentials to Suppress Locality Errors in the Fixed-Node Diffusion Monte Carlo Method

Locality errors are intrinsic to fixed-node diffusion Monte Carlo (FNDMC) calculations with semi-local pseudopotentials. Although methods such as the T-moves and determinant localization approximation can be used to evaluate the pseudopotentials in FNDMC at the cost of introducing locality error. The magnitude of these errors depends on the quality of the trial wave functions; which disrupts error cancelation between different systems. For such cases, the use of pseudo-Hamiltonians (PH) is expected to provide a reliable locality-error-free framework. However, due to their stringent constraints, constructing high-quality PH for every element in the periodic table is challenging. For example, in our previous study on 3d transition metals [J. Chem. Phys. 159, 164114 (2023)], we found that developing reliable PHs for Sc-CV was infeasible. To overcome these limitations, we have developed hybrid potentials [J. Chem. Phys. 153, 104111 (2020)], primarily composed by PH, along with a residual non-local potential to increase expressivity over the strict PH. In this talk, we present the theoretical framework and numerical optimization of hybrid potentials for Sc-V. We also and discuss the extent to which these hybrid potentials suppress locality errors in FNDMC calculations.