Bridging the Gap between Transformer-Based Neural Networks and Tensor Networks for Quantum Chemistry

• DOI: 10.1021/acs.jctc.4c01703
• Publication Date: 2025-03-02T14:51:12Z
• Authors: Bowen Kan; Yingqi Tian; Yangjun Wu; Yunquan Zhang; Honghui Shang

The neural network quantum state (NNQS) method has demonstrated promising results in ab initio chemistry, achieving remarkable accuracy molecular systems. However, efficient calculation of systems with large active spaces remains challenging. This study introduces a novel approach that bridges tensor states the transformer-based NNQS-Transformer (QiankunNet) to enhance and convergence for relatively spaces. By transforming into space configuration interaction type wave functions, QiankunNet achieves surpassing both pretraining density matrix renormalization group (DMRG) traditional coupled cluster methods, particularly strongly correlated regimes. We investigate two transformation methods: sweep-based direct conversion (Conv.) entanglement-driven genetic algorithm (EDGA) method, Conv. showing superior efficiency. effectiveness this is validated on H2O (10e, 24o) cc-pVDZ basis set, demonstrating an routine between DMRG also offering direction advancing representation complex