Development of Hybrid Potential for the Fixed-Node Diffusion Monte Carlo Method for Al to Cl

Diffusion Monte Carlo (DMC) is one of the most reliable ab initio methods, capable of accurately describing systems that challenge conventional density functional theory, such as noncovalent materials and magnetic compounds. Since the computational cost of DMC increases steeply with the number of valence electrons, the use of pseudopotentials is indispensable for practical applications. However, conventional semilocal pseudopotentials introduce locality errors when applied to DMC. To address this problem, the pseudo-Hamiltonian (PH) approach, in which the nonlocal component is incorporated into the walker diffusion process, has been proposed and successfully developed for several 3d transition-metal elements [J. Chem. Theory Comput. 18, 828 (2022), J. Chem. Phys. 159, 164114 (2023)]. Nevertheless, strong internal constraints among its parameters make it difficult to construct transferable PHs across elements. Therefore, we focus on developing a hybrid potential, in which most of the effective core potential is described by a PH, while the residual contribution is treated by nonlocal pseudopotentials. This approach is expected to significantly suppress the locality error compared to conventional semilocal pseudopotentials. In the talk, we will present the construction of PHs for elements from Al to Cl, describe their mathematical formulation and numerical optimization, and discuss the effectiveness of the resulting PHs in terms of transferability and reduction of locality errors.