Abstract:Recent advances in machine learning have facilitated numerically accurate solution of the electronic SchrödingerNet equation (SE) by integrating various neural network (NN)-based wavefunction ansatzes with variational Monte Carlo methods. Nevertheless, such NN-based methods are all based on the Born-Oppenheimer approximation (BOA) and require a separate and computationally expensive training for each nuclear configuration. In this work, we propose a novel NN architecture, SchrödingerNet, to solve the full electronic-nuclear SE by defining a loss function designed to equalize local energies across the system. This approach is based on a rotationally invariant total wavefunction ansatz that includes both nuclear and electronic coordinates. This strategy allows for an efficient and accurate generation of a continuous potential energy surface at any geometry within the well-sampled nuclear configuration space, and also incorporates non-BO corrections, through a single training process. Comparison with benchmarks of atomic and molecular systems demonstrates its accuracy and efficiency

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: how to directly solve the full Schrödinger Equation (SE) containing electron and nuclear degrees of freedom through a neural network without relying on the Born - Oppenheimer Approximation (BOA), so as to efficiently and accurately generate a continuous potential energy surface and incorporate non - BOA corrections at the same time. Specifically, traditional quantum chemistry methods usually assume that electron and nuclear motions are decoupled, that is, they use BOA to simplify the problem. However, this approximation will introduce significant errors in some cases, especially when approaching electron - degenerate states. In addition, traditional methods require separate and computationally expensive training for each nuclear configuration to construct a global potential energy surface. To solve these problems, this paper proposes a new neural network architecture named SchrödingerNet, whose main features include: 1. **Directly solving the full Schrödinger equation**: SchrödingerNet directly solves the total wave function containing electron and nuclear coordinates by defining a loss function to equalize the local energy everywhere in the system. 2. **Rotation - invariant total wave function assumption**: This model adopts a rotation - invariant total wave function assumption that covers electron and nuclear coordinates, ensuring physical correctness. 3. **Generating a continuous potential energy surface in a single training**: Through one training, SchrödingerNet can generate a continuous potential energy surface for any geometric structure and can incorporate non - BOA corrections without requiring separate training for each nuclear configuration. 4. **Efficient energy prediction**: Different from the traditional variational Monte Carlo method, SchrödingerNet is based on a local - energy loss function, avoiding the need for integration, making energy prediction more efficient. Through these innovations, SchrödingerNet can not only describe the entire nuclear configuration space of the system more accurately, but also effectively handle complex situations such as bond breaking, providing a new and efficient method for solving the high - dimensional nuclear - electron Schrödinger equation.

SchrödingerNet: A Universal Neural Network Solver for The Schrödinger Equation

Deep-neural-network solution of the electronic Schrödinger equation

Solving many-electron Schrödinger equation using deep neural networks

Solving Schrödinger equation within arbitrary spherical quantum dots with neural network

Ab-Initio Potential Energy Surfaces by Pairing GNNs with Neural Wave Functions

Solving Schrodinger equations using physically constrained neural network

Ab initio quantum chemistry with neural-network wavefunctions

Deep learning neural network for approaching Schrödinger problems with arbitrary two-dimensional confinement

Simulating Quantum Many-Body States with Neural-Network Exponential Ansatz

Neural Schrödinger Equation: Physical Law as Deep Neural Network

Gold-standard solutions to the Schrödinger equation using deep learning: How much physics do we need?

Solving Schrödinger Equation Using Tensor Neural Network

Ab initio solution of the many-electron Schrödinger equation with deep neural networks

Solving the vibrational Schrödinger equation with artificial neural networks

Towards a Foundation Model for Neural Network Wavefunctions

Deep-neural-network approach to solving the ab initio nuclear structure problem

Deep-neural-network approach to solving the<i> ab</i><i> initio</i> nuclear structure problem

Solving Schrödinger Equation with a Language Model

NNQS-Transformer: an Efficient and Scalable Neural Network Quantum States Approach for Ab initio Quantum Chemistry

Spin-symmetry-enforced Solution of the Many-Body Schrödinger Equation with a Deep Neural Network