Answers are grounded in this paper and cite their sources.
The neural network quantum state (NNQS) approach provides an artificial intelligence-based representation of many-electron wave functions for quantum chemistry. Although a full configuration-space NNQS can achieve high accuracy, its direct use for force evaluation and geometry optimization remains computationally demanding, motivating a reduced-space treatment. Here, we implement analytic interatomic force calculations at the complete active space configuration interaction (CASCI) level based on the highly expressive QiankunNet architecture. By constructing an effective second-quantized Hamiltonian within a prescribed active orbital space and solving the coupled-perturbed Hartree–Fock equations, we obtain analytic forces including orbital response contributions. Benchmark calculations demonstrate the compact representational capability of QiankunNet and its close agreement with conventional CASCI references for both energies and forces, even in strongly correlated systems. We further apply the resulting forces to construct reaction potential energy surfaces for both ethylene torsion and cyclobutadiene isomerization, and to optimize the anthracene geometry for vertical singlet–triplet energy gap calculation. This work offers a practical route toward efficient force-based simulations of multireference chemical systems.
No takes yet. Share an insight, caveat, or question.
Mingxin et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: