Recommendation System to Predict the d-band Center of Core-Shell Bimetallic Nanoparticles Catalysts

Abstract

Core-shell nanoparticles are an important class of catalytic materials due to the presence of unsaturated bonds, phase-separated metals, and synergistic effect of more than one elements. Exploring the electronic properties such as the $d$-band center (${\epsilon }_d$) and the related catalytic properties of these materials are quite challenging due to resource consuming experiments and computationally expensive modeling of the nanoparticles. Therefore, a density functional theory (DFT) coupled recommendation system based approach is developed to predict the ${\epsilon }_d$ of the core-shell nanoparticles. Here, the recommendation system involves completion of a core-shell interaction matrix, where each matrix element represents the ${\epsilon }_d$ of a unique core-shell pair. Matrix completion by predicting the interaction matrix elements is carried out using neural network. The neural network generated a low dimensional representations for each core and shell metals, which are learned iteratively to predict corresponding ${\epsilon }_d$ of the pair. The learned representations brilliantly captures the similarities and dissimilarities between metals present in shell or core. A very low train/test RMSE of 0.009/0.009 is obtained with elemental features based model. The same model is further employed to predict the $d$-band width (${\epsilon }_w$), suggesting its transferability. This approach recommends the optimum range of ${\epsilon }_d$/ ${\epsilon }_w$ along with the combination of core-shell metals having most efficient adsorption of species for any reaction. The results presented in this study can help experimentally design the core-shell nanoparticles having a desired catalytic activity.