Toward an ab Initio Description of Adsorbate Surface Dynamics.

Area of Science:

• Computational materials science
• Surface science
• Machine learning in chemistry

Background:

• Density functional theory (DFT) calculations are computationally expensive for long molecular dynamics (MD) simulations.
• Machine learning potentials (MLPs) offer a way to achieve DFT-level accuracy at a lower computational cost.
• Understanding surface diffusion is crucial for catalysis and materials design.

Purpose of the Study:

• To train a generalizable machine learning potential (MLP) for simulating surface dynamics.
• To investigate the diffusion of surface-bound adsorbates on a Ag(111) facet using MLP-based MD simulations.
• To compare MLP/MD results with traditional DFT methods for estimating diffusion barriers.

Main Methods:

• An active learning curriculum was employed to train a generalizable MLP using the DeepMD-kit.
• Long molecular dynamics (MD) simulations were performed using the trained MLP.
• Diffusion coefficients of surface adsorbates on Ag(111) were calculated and analyzed.

Main Results:

• MLP-based MD simulations achieved DFT-level accuracy for adsorbate diffusion.
• The study identified potential limitations of using DFT-based nudged elastic band methods to determine surface diffusion barriers.
• Accurate diffusivities were computed for key surface-bound adsorbates on Ag(111).

Conclusions:

• MLPs combined with MD simulations provide a powerful approach for studying surface diffusion with high accuracy.
• This method offers insights into the shortcomings of conventional techniques for calculating surface diffusion barriers.
• The developed workflows and models show promise for broader applications in materials science and surface chemistry.