Abstract:Artificial neural networks (NN) are universal function approximators and have shown great ability in computing the ground state energy of the electronic Schrödinger equation, yet NN has not established itself as a practical and accurate approach to solve the vibrational Schrödinger equation for realistic polyatomic molecules to obtain vibrational energies and wave functions for the excited states. Here we purpose an efficient approach to use NN to solve the vibrational Schrödinger equation and demonstrate the new method on CH4, a five atom molecule with 9 degree of freedom. By using a NN with < 3,000 parameters, we are able to achieve vibration energies for the ground and excited states with an accuracy of less than 1 cm-1 as compared to the reference values obtained by using more than 109 basis functions. It is anticipated the new method is capable of providing highly accurate vibrational energies and wave functions for molecules with more than 10 atoms, beyond the limit for all the existing computational approaches.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to use artificial neural networks (NN) to solve the vibrational Schrödinger equation (SE) of polyatomic molecules efficiently and accurately, in order to obtain the vibrational energies and wave functions of excited states and ground states. Specifically, this research aims to overcome the curse - of - dimensionality problem encountered by existing methods when dealing with complex polyatomic molecules and to provide a high - precision calculation method for vibrational energies and wave functions of molecules with more than 10 atoms. ### Main Challenges 1. **Curse of Dimensionality**: Traditional methods solve SE by expanding the wave function with basis functions. However, as the dimension of the system increases, the number of basis functions grows exponentially, leading to a sharp rise in computational cost. 2. **Limitations of Existing NN Methods**: Although NN has made significant progress in solving the electronic SE, it has not yet established an effective application in solving the vibrational SE of actual polyatomic molecules, especially in solving the excited states, which is more challenging. 3. **Complexity of the Kinetic Energy Operator (KEO)**: For large molecules, it is difficult to calculate KEO accurately, which limits the application range of NN methods. ### Solutions This paper proposes a new strategy to use NN to solve the vibrational SE of polyatomic molecules. Specific methods include: - Using bond lengths as input coordinates to describe molecular configurations. - Selecting local energy configuration points with large training errors for optimization instead of relying on sampling points required for wave function integration. - Using the variance of local energy as a loss function to optimize NN parameters. - Gradually improving the training accuracy through an iterative process to ensure the accuracy of the final results. ### Experimental Verification The author takes methane (CH₄) as an example to demonstrate the effectiveness of the new method. By using an NN with about 3,000 parameters and about 20,800 working points, the vibrational energy can be obtained with an accuracy of less than 1 cm⁻¹, which is far better than the more than 10⁹ basis functions required by traditional methods. ### Potential Impact This method not only provides new ideas for solving the vibrational SE of polyatomic molecules but also may be extended to the solution of other quantum - mechanical problems, such as the electronic SE, especially after introducing appropriate symmetry conditions. This provides strong support for future research on the vibrational and electronic properties of complex molecular systems. ### Summary This paper successfully solves the key problems in solving the vibrational SE of polyatomic molecules by introducing a new NN - based method, opening up a new way for high - precision calculation of the vibrational energies and wave functions of complex molecules.

Solving the vibrational Schrödinger equation with artificial neural networks

Neural canonical transformations for vibrational spectra of molecules

Solving many-electron Schrödinger equation using deep neural networks

SchrödingerNet: A Universal Neural Network Solver for The Schrödinger Equation

Ab initio quantum chemistry with neural-network wavefunctions

Solution to the Schrödinger equation for bound states of polar molecules using shallow neural networks.

Deep-neural-network solution of the electronic Schrödinger equation

Accurate Computation of Quantum Excited States with Neural Networks

Machine learning for the solution of the Schrödinger equation

Achieving vibrational energies of diatomic systems with high quality by machine learning improved DFT method

Solving Schrödinger equation within arbitrary spherical quantum dots with neural network

Accurate Calculation of Interatomic Forces with Neural Networks Based on a Generative Transformer Architecture

Neutron scattering and neural-network quantum molecular dynamics investigation of the vibrations of ammonia along the solid-to-liquid transition

Deep Neural Network Computes Electron Densities and Energies of a Large Set of Organic Molecules Faster than Density Functional Theory (DFT)

Efficient solutions of fermionic systems using artificial neural networks

Hermite Neural Network Simulation for Solving the 2D Schrodinger Equation

Calculate electronic excited states using neural networks with effective core potential

Solving Schrödinger Wave Equation for the Charmonium Spectrum Using Artificial Neural Networks