Abstract:Neural network-based variational Monte Carlo (NN-VMC) has emerged as a promising cutting-edge technique of ab initio quantum chemistry. However, the high computational cost of existing approaches hinders their applications in realistic chemistry problems. Here we report a development of NN-VMC that achieves a remarkable speed-up rate, thereby greatly extending the applicability of NN-VMC to larger systems. Our key design is a computational framework named Forward Laplacian, which computes the Laplacian associated with neural networks, the bottleneck of NN-VMC, through an efficient forward propagation process. We then demonstrate that Forward Laplacian can further facilitate more developments of acceleration methods across various aspects, including optimization for sparse derivative matrix and efficient network design. Empirically, our approach enables NN-VMC to investigate a broader range of systems, providing valuable references to other ab initio methods. The results demonstrate a great potential in applying deep learning methods to solve general quantum mechanical problems.

What problem does this paper attempt to address?

The paper primarily addresses the issue of high computational costs associated with the Neural Network Variational Monte Carlo (NN-VMC) method when applied to practical chemical problems. Specifically, the paper introduces a computational framework named "Forward Laplacian," which calculates the Laplacian operator related to the neural network inputs through an efficient forward propagation process, a bottleneck step in the NN-VMC method. By eliminating unnecessary computations and propagations, this approach significantly improves computational efficiency. The paper also designs an effective neural network architecture called LapNet, which leverages meticulously designed attention blocks and sparse derivatives to better utilize the advantages of the Forward Laplacian method. These developments enable researchers to explore a broader range of systems, including atoms, molecules, and chemical reactions, surpassing the capabilities of existing NN-VMC software packages. Furthermore, the paper demonstrates the method's accuracy in absolute energy calculations and reduces the computational cost of model training. Additionally, the researchers explore the method's potential in relative energy estimation, including the barrier heights of chemical reactions, ionization energies of transition metals, and the energy of non-covalent interactions between molecules. Experimental results indicate that the relative energies obtained using the NN-VMC method are consistent with gold-standard computational methods and experimental outcomes, highlighting the immense potential of using deep learning approaches to solve quantum mechanical problems. In summary, by proposing the Forward Laplacian framework and LapNet architecture, the paper resolves the challenges of computational efficiency in large system calculations for the NN-VMC method, offering a more efficient and accurate solution for the field of quantum chemistry.

A computational framework for neural network-based variational Monte Carlo with Forward Laplacian

Forward Laplacian: A New Computational Framework for Neural Network-based Variational Monte Carlo

Interatomic Force from Neural Network Based Variational Quantum Monte Carlo

Variance extrapolation method for neural-network variational Monte Carlo

Penalty and auxiliary wave function methods for electronic Excitation in neural network variational Monte Carlo

Transferable Neural Wavefunctions for Solids

Variational Monte Carlo on a Budget -- Fine-tuning pre-trained Neural Wavefunctions

Provable Convergence of Variational Monte Carlo Methods

Neural-network Variational Quantum Algorithm for Simulating Many-Body Dynamics

Wasserstein Quantum Monte Carlo: A Novel Approach for Solving the Quantum Many-Body Schrödinger Equation

DeepQMC: an open-source software suite for variational optimization of deep-learning molecular wave functions

Neural Projected Quantum Dynamics: a systematic study

Neural network backflow for ab-initio quantum chemistry

Highly Accurate Real-space Electron Densities with Neural Networks

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

Accelerating lattice quantum Monte Carlo simulations using artificial neural networks: Application to the Holstein model

A Blocked Linear Method for Optimizing Large Parameter Sets in Variational Monte Carlo

Efficient solutions of fermionic systems using artificial neural networks

Towards the ground state of molecules via diffusion Monte Carlo on neural networks

Improved Optimization for the Neural-network Quantum States and Tests on the Chromium Dimer

Enhancing the Expressivity of Variational Neural, and Hardware-Efficient Quantum States Through Orbital Rotations