Projected Sobolev Natural Gradient Descent for Efficient Neural Network Solution of the Gross-Pitaevskii Equation

TL;DR

This paper presents a novel projected Sobolev natural gradient descent method for efficiently computing ground states of the Gross-Pitaevskii equation using neural networks, improving convergence speed and scalability.

Contribution

It introduces a projected Sobolev NGD algorithm with a hybrid sampling strategy and matrix-free solver, advancing neural network solutions for nonlinear PDEs.

Findings

Faster convergence than existing physics-informed neural networks

Linear scalability with spatial dimensions

High-quality initial guesses for traditional solvers

Abstract

This paper introduces a projected Sobolev natural gradient descent (NGD) method for computing ground states of the Gross-Pitaevskii equation. By projecting a continuous Riemannian Sobolev gradient flow onto the normalized neural network tangent space, we derive a discrete NGD algorithm that preserves the normalization constraint. The numerical implementation employs variational Monte Carlo with a hybrid sampling strategy to accurately account for the normalization constant arising from nonlinear interaction terms. To enhance computational efficiency, a matrix-free Nystr\"om-preconditioned conjugate gradient solver is adopted to approximate the NGD operator without explicit matrix assembly. Numerical experiments demonstrate that the proposed method converges significantly faster than physics-informed neural network approaches and exhibits linear scalability with respect to spatial dimensions. Moreover, the resulting neural-network solutions provide high-quality initial guesses that substantially accelerate subsequent refinement by traditional high-precision solvers.