Surface structure and configuration is very important in chemistry, especially for catalyst materials. In order to predict the structure of materials, a surface Pourbaix diagram is a very powerful tool. Fortunately, we can build a surface Pourbaix diagram by DFT-calculated adsorption energies using computational hydrogen scheme. However, adsorption energy of diverse configuration is needed to build the diagram and it tooks too much computational cost to calculate by common DFT method. To overcome this problem, we applied crystal graph convolutional neural network (CGCNN) model to predict adsorption energies of adsrobates on metal NPs, and build a surface Pourbaix diagram of the NPs. By our machine learning model, the adsorption energies of O and OH adsorbates on Pt NPs could be predicted with the MAE less than 0.3 eV, not only for a single adsorbate but also configuration with more than two adsorbates. In addition, a surface Pourbaix diagram of Pt NPs could be built by the predicted adsorption energy, and it is very similar to DFT calculated ones. Finally, we could predict the adsorption energies and built surface Pourbaix diagram of big NPs up to 7 nm, which is almost impossible to be calculated by common DFT method.

Modified crystal graph convolutional neural network to predict adsorption energy and surface Pourbaix diagram of metal nanoparticles

ACS Spring 2022

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.