Fast probe of hydrogen molecules binding on alpha-iron surface via a machine-learning method and high-throughput molecular dynamics simulations

X Li and M Xu and TS Zheng and C Sun and XY Lan and J Hu and ZC Yu, MATERIALS TODAY COMMUNICATIONS, 47, 113019 (2025).

DOI: 10.1016/j.mtcomm.2025.113019

Adsorption is the first step for hydrogen to enter iron-based materials and lead to embrittlement. Identifying potential attacking sites of hydrogen on surface is key to enhancing materials resistance against hydrogen embrittlement. Molecular dynamics can simulate adsorption of hydrogen, but the calculation is too computationally demanding. Here, a machine-learning method for fast evaluation of hydrogen adsorption is demonstrated. Trained from high-throughput molecular dynamics simulations, the method shows good physical consistency, validity, and generalization ability. Given an atomistic model of iron surface, the method can predict relative binding energy at a speed 10,000 times faster than the metadynamics calculation. The developed machine-learning method and the computational framework might be used consistently with advanced characterization techniques for fast evaluation of hydrogen embrittlement risks on infrastructures like pipes, containers, etc. Moreover, the developed computational framework finds a broader range of applications with a variety of adsorption molecules and target material surface.

Return to Publications page