Reproducibility of fixed-node diffusion Monte Carlo across diverse community codes: The case of water-methane dimer

Area of Science:

• Computational Chemistry
• Quantum Mechanics
• Materials Science

Background:

• Fixed-node diffusion quantum Monte Carlo (FN-DMC) is a robust method for solving the Schrödinger equation, widely used for predicting material and molecular properties.
• Its scalability and computational efficiency position it to tackle complex scientific challenges.
• However, the reproducibility of FN-DMC across various code implementations has been a concern due to algorithmic diversity and stochasticity.

Purpose of the Study:

• To assess the reproducibility of Fixed-node diffusion quantum Monte Carlo (FN-DMC) across different community codes.
• To identify the primary sources of discrepancies in FN-DMC results.
• To validate the reliability of FN-DMC for computational science.

Main Methods:

• A community-wide effort involving eleven different FN-DMC codes was undertaken.
• The water-methane dimer served as a canonical test case for comparison.
• Various approximations for treating pseudopotential non-locality were analyzed.

Main Results:

• The study confirmed that FN-DMC is reproducible across diverse codes when appropriate algorithmic details are employed.
• Approximations used to handle pseudopotential non-locality were identified as the main source of discrepancies.
• The T-move, determinant locality approximation, and determinant T-move schemes yielded reproducible results, unlike the older locality approximation.

Conclusions:

• Fixed-node diffusion quantum Monte Carlo (FN-DMC) is a mature and robust method for computational science.
• Careful selection of algorithmic details, particularly for pseudopotential treatment, is crucial for achieving reproducible results.
• This work highlights the potential for open and reliable computational science using FN-DMC.