A multi-modal transformer for predicting global minimum adsorption energy.

Area of Science:

• Materials Science
• Computational Chemistry
• Catalysis

Background:

• Accurate prediction of global minimum adsorption energy (GMAE) is vital for catalyst screening.
• Density functional theory (DFT) calculations are computationally expensive for determining GMAE due to numerous adsorption sites and configurations.

Purpose of the Study:

• To develop a rapid and accurate method for predicting GMAE.
• To overcome the computational limitations of traditional DFT methods for large-scale catalyst screening.

Main Methods:

• Designed AdsMT, a multi-modal transformer model.
• Utilized surface graphs and adsorbate feature vectors for prediction.
• Employed a cross-attention mechanism to capture adsorbate-surface interactions without site-binding information.

Main Results:

• Achieved low mean absolute errors of 0.09, 0.14, and 0.39 eV on three benchmark datasets.
• AdsMT effectively predicts GMAE, significantly reducing computational cost.
• Demonstrated interpretable potential of cross-attention scores for identifying favorable adsorption sites.

Conclusions:

• AdsMT offers a computationally efficient and accurate approach for GMAE prediction.
• The model facilitates large-scale catalyst screening.
• Integrated uncertainty quantification enhances prediction reliability.