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

2025/12/12 by Bao, Chenglong, Cui, Chen, Jiang, Kai +1
Physics and Astronomy · Materials Science · #Model Reduction and Neural Networks #Machine Learning in Materials Science #Quantum many-body systems

paper · doi:10.48550/arxiv.2512.11339

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 Nyström-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.

Citations

• MAD-NG: Meta-Auto-Decoder Neural Galerkin Method for Solving Parametric Partial Differential Equations
• Metric-driven numerical methods
• Toward fast, accurate and robust AI prediction of ground states in rotating BEC
• On preconditioned Riemannian gradient methods for minimizing the Gross-Pitaevskii energy functional: algorithms, global convergence and optimal local convergence rate
• AMStraMGRAM: Adaptive Multi-cutoff Strategy Modification for ANaGRAM
• Low-Rank Evolutionary Deep Neural Networks via Adaptive Tangent-Space Reduction
• Neural Network Acceleration of Iterative Methods for Nonlinear Schrödinger Eigenvalue Problems
• Functional Neural Wavefunction Optimization
• Improving Energy Natural Gradient Descent through Woodbury, Momentum, and Randomization
• Energy-Adaptive Riemannian Conjugate Gradient Method for Density Functional Theory
• Computing ground states of Bose-Einstein condensation by normalized deep neural network
• Multi evolutional deep neural networks (Multi-EDNN)
• Accurate neural quantum states for interacting lattice bosons
• Irrational-window-filter projection method and application to quasiperiodic Schrödinger eigenproblems
• pETNNs: Partial Evolutionary Tensor Neural Networks for Solving Time-dependent Partial Differential Equations
• The Challenges of the Nonlinear Regime for Physics-Informed Neural Networks
• A Unified Framework for the Error Analysis of Physics-Informed Neural Networks
• Randomized Sparse Neural Galerkin Schemes for Solving Evolution Equations with Deep Networks
• On the convergence of Sobolev gradient flow for the Gross-Pitaevskii eigenvalue problem
• Fourier Features Let Networks Learn High Frequency Functions in Low Dimensional Domains
• Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations
• Sobolev gradient flow for the Gross-Pitaevskii eigenvalue problem:\n global convergence and computational efficiency
• Adam: A Method for Stochastic Optimization
• New insights and perspectives on the natural gradient method
• An inverse iteration method for eigenvalue problems with eigenvector\n nonlinearities
• A new Sobolev gradient method for direct minimization of the Gross-Pitaevskii energy with rotation
• Note on Exchange Phenomena in the Thomas Atom

Related

• Variational Physics-informed Neural Operator (VINO) for Solving Partial Differential Equations
• Deep Learning via Neural Energy Descent
• Energy Manifold Natural Gradient Descent: Riemannian Optimization for Neural PDE Solvers
• Achieving High Accuracy with PINNs via Energy Natural Gradients
• Variational Physics-Informed Neural Networks For Solving Partial Differential Equations