Using Machine Learning to Predict Surface Adsorption
Tuskegee University, 1200 W. Montgomery Rd. Tuskegee, AL 36088
Abstract: The interaction of molecules on metallic surfaces plays an important role in a wide array of technologies from catalysts that remove harmful gases from the atmosphere to light-harvesting devices and devices to store hydrogen. Modeling these types of interactions, between molecules and a metallic substrate, can cost a large amount of computational time, limiting both the amount of systems one can study and the potential to improve device performance. To remedy this problem and cut down on computational cost one can employ machine learning techniques. In this talk I present the results of utilizing the Hierarchically Interacting Particle Neural Network (HIP-NN), a deep learning neural network, to predict the adsorption of hydrogen on various single crystal surfaces. Overall, a trained HIP-NN model predicts both the adsorption heights and energies of hydrogen on these surfaces close to the accuracy of the reference calculations, representing a large step forward in streamlining novel material discovery.
Host: Harsh Mathur.