Reproduce “Data-driven construction of machine-learning-based interatomic potentials for gas-surface scattering dynamics: the case of NO on graphite” on OpenResearch

About this paper

Samuel Del Fré, Gilberto A. Alou Angulo, Maurice Monnerville, Alejandro Rivero Santamaría

Researchers at Univ. Lille developed a data-driven workflow for constructing machine-learning interatomic potentials (MLIPs) for gas-surface scattering dynamics, applying it to NO on graphite. This method achieves near ab initio accuracy in molecular dynamics simulations, efficiently reproducing experimental trends and providing detailed atomistic insights into energy transfer processes.

Key insights

• Combining descriptor-guided sampling with active learning effectively and efficiently generates high-fidelity MLIPs that can represent complex gas-surface interactions over broad parameter spaces.
• The developed MLIPs enable large-scale molecular dynamics simulations to achieve statistically converged results with near ab initio accuracy at a significantly reduced computational cost.
• The methodology provides a transferable framework for studying diverse gas-surface systems, bridging the gap between high-fidelity electronic structure methods and the statistical requirements of comprehensive dynamics studies.

Summary by alphaXiv.