Neural Network Quantum State for Light Nuclei with Chiral Interactions

Constructing the formula of high-quality trial wave functions for the ground state of light nuclei with realistic nuclear interactions requires strong intuition to model the interparticle correlations. Accounting for many-body effects when three-body interactions are present makes the correlations more complicated. Moreover, exploring and optimizing trial-wave-function parameters demands large computational resources. In this presentation, we introduce an ansatz in which the correlation functions are generated directly by neural networks, without relying on explicit analytic trial-function formulas. This ansatz incorporates both two-body and three-body correlations. Optimized by variational Monte Carlo, it can capture the overwhelming majority of the ground-state energy for A=3 nuclei as benchmarked against Green's function Monte Carlo. A 91% improvement over standard Variational Monte Carlo is achieved for Triton with the softest chiral interaction. The result indicates the potential of neural networks to build expressive wave functions for Quantum Monte Carlo calculations.