Particle view of many-body electronic structure

Neural network wavefunctions can now provide near-exact ground-state solutions for molecules and solids, but their high-dimensional nature makes direct human interpretation intractable. We develop periodic dynamic Voronoi Metropolis sampling (PDVMS) to extract classical electron configurations from neural network wavefunctions and connect them with intuitive valence-bond pictures. Applying this framework, we quantitatively resolve the competing valence-bond structures of benzene and extend the particle view to periodic solids, revealing a spin-staggered valence-bond structure in graphene rather than the conventional mixed single-double bond network. The particle view further reveals how the spin-staggered structure emerges during wavefunction optimization and offers an intuitive explanation for emergent magnetism in graphene-based nanostructures. PDVMS thus provides a complementary framework for understanding many-body electronic structure.

• Z. Wang et al. Phys. Rev. X 16, 31048 (2026)
• [News] 成果丨粒子视角理解多体电子结构