Scalable neural quantum states architecture for quantum chemistry

Abstract Variational optimization of neural-network representations quantum states has been successfully applied to solve interacting fermionic problems. Despite rapid developments, significant scalability challenges arise when considering molecules large scale, which correspond non-locally spin Hamiltonians consisting sums thousands or even millions Pauli operators. In this work, we introduce scalable parallelization strategies improve neural-network-based variational Monte Carlo calculations for ab-initio chemistry applications. We establish GPU-supported local energy parallelism compute the objective potentially complex molecules. Using autoregressive sampling techniques, demonstrate systematic improvement in wall-clock timings required achieve coupled cluster with up double excitations baseline target energies. The performance is further enhanced by accommodating structure resultant into ordering. algorithm achieves promising comparison classical approximate methods and exhibits both running time advantages over existing based methods.