Machine-Learning Interatomic Potentials for NO Scattering on Graphite

A data-driven workflow for constructing machine-learning interatomic potentials (MLIPs) tailored to gas-surface scattering dynamics. The benchmark system is nitric oxide (NO) scattering from highly oriented pyrolytic graphite (HOPG). The final model enables large-scale molecular dynamics simulations of NO scattering at a computational cost far below ab initio molecular dynamics.

Use Cases

• Training machine-learned interatomic potentials based on ab initio molecular dynamics data.
• Simulating adsorption energetics and scattering probabilities based on molecular dynamics trajectories.
• Analyzing translational energy loss and rotational excitation in gas-surface collisions.
• Applying active learning strategies to refine potentials for extended energy and temperature ranges.

Strengths

• The final MLIP reproduces reference energies and forces with high fidelity.
• The workflow uses a query-by-committee active-learning strategy for refinement.
• The dataset is openly available under a CC-BY-NC-4.0 license.

Limitations

• The dataset size is 13.9 KB, indicating a very small or highly compressed data scope.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

Samuel Del Fré via figshare

Collection Method

Data-driven workflow starting from ab initio molecular dynamics, using SOAP descriptors, PCA, farthest point sampling, and active learning.

Freshness

Last updated 2026-05-13 04:03:21