Abstract:While the computation of interatomic forces has become a well-established practice within variational Monte Carlo (VMC), the use of the more accurate Fixed-Node Diffusion Monte Carlo (DMC) method is still largely limited to the computation of total energies on structures obtained at a lower level of theory. Algorithms to compute exact DMC forces have been proposed in the past, and one such scheme is also put forward in this work, but remain rather impractical due to their high computational cost. As a practical route to DMC forces, we therefore revisit here an approximate method, originally developed in the context of correlated sampling and named here the Variational Drift-Diffusion (VD) approach. We thoroughly investigate its accuracy by checking the consistency between the approximate VD force and the derivative of the DMC potential energy surface for the SiH and C2 molecules and employ a wide range of wave functions optimized in VMC to assess its robustness against the choice of trial function. We find that, for all but the poorest wave function, the discrepancy between force and energy is very small over all interatomic distances, affecting the equilibrium bond length obtained with the VD forces by less than 0.004 au. Furthermore, when the VMC forces are approximate due to the use of a partially optimized wave function, the DMC forces have smaller errors and always lead to an equilibrium distance in better agreement with the experimental value. We also show that the cost of computing the VD forces is only slightly larger than the cost of calculating the DMC energy. Therefore, the VD approximation represents a robust and efficient approach to compute accurate DMC forces, superior to the VMC counterparts.

Interatomic Force from Neural Network Based Variational Quantum Monte Carlo

Force and stress calculation with neural network wavefunction for solids

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

Quantum neural networks force fields generation

Learning the Quantum Centroid Force Correction in Molecular Systems: A Localized Approach

Highly Accurate Real-space Electron Densities with Neural Networks

Force and Stress Calculations with a Neural-Network Wave Function for Solids

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

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

Efficient calculation of unbiased atomic forces in ab initio Variational Monte Carlo

Variance extrapolation method for neural-network variational Monte Carlo

Ab initio quantum chemistry with neural-network wavefunctions

High Accuracy Uncertainty-Aware Interatomic Force Modeling with Equivariant Bayesian Neural Networks

Practical Schemes for Accurate Forces in Quantum Monte Carlo

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

Ab Initio Calculation of Real Solids Via Neural Network Ansatz

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

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

Efficient solutions of fermionic systems using artificial neural networks

Accurate and scalable graph neural network force field and molecular dynamics with direct force architecture

Polynomially efficient quantum enabled variational Monte Carlo for training neural-network quantum states for physico-chemical applications