Pseudo-Hamiltonian for NNQMC study of large quantum many-body systems

Computing the quantum behavior of interacting electrons is central to chemistry and physics, but scaling wavefunction simulations to large, strongly correlated systems remains computationally prohibitive. While neural network-based quantum Monte Carlo (NNQMC) shows great promise, conventional semi-local pseudopotentials introduce expensive nonlocal terms that severely bottleneck efficiency. We integrate fully local pseudopotentials (pseudo-Hamiltonians) into NNQMC and apply them to main-group elements and complex iron–sulfur clusters. By eliminating costly nonlocal integrations, our approach accelerates NNQMC by over an order of magnitude while preserving high accuracy, unlocking the simulation of previously inaccessible strongly correlated systems.

• W. Fu et al. Nat Comput Sci (2026)
• Research Briefing
• [News] 成果丨赝哈密顿量赋能神经网络量子蒙特卡洛方法