Convergence of variational Monte Carlo simulation and scale-invariant pre-training
This work addresses convergence issues in quantum chemistry simulations, offering incremental improvements for researchers in computational physics and chemistry.
The paper tackled the convergence of variational Monte Carlo methods for optimizing neural network wave functions in electronic structure problems, providing theoretical bounds for energy minimization and proposing a scale-invariant loss for pre-training that empirically leads to faster training.
We provide theoretical convergence bounds for the variational Monte Carlo (VMC) method as applied to optimize neural network wave functions for the electronic structure problem. We study both the energy minimization phase and the supervised pre-training phase that is commonly used prior to energy minimization. For the energy minimization phase, the standard algorithm is scale-invariant by design, and we provide a proof of convergence for this algorithm without modifications. The pre-training stage typically does not feature such scale-invariance. We propose using a scale-invariant loss for the pretraining phase and demonstrate empirically that it leads to faster pre-training.