Exact Equivariance from Ordinary Neural Networks for Lattice Many-Body DynamicsNew

This paper presents a method for building machine-learning surrogates for correlated electron system simulations using ordinary multilayer perceptrons. The key innovation is applying finite-group averaging to achieve exact equivariance without specialized network architectures. The approach targets the repeated evaluation of electronic forces and transition energies that drive collective dynamics in lattice quantum systems. This work contributes to quantum chemistry applications by providing a practical ML framework that naturally respects physical symmetries.