Neural Network Predictions of Oxygen Interactions on a Dynamic Pd Surface

Implementation of artificial neural networks (NNs) are becoming increasingly common for surface chemistry applications. Such models are trained to higher-level /ab-initio/ calculations and are capable of achieving arbitrary levels of accuracy. The most common applications thus far have been specialized for either bulk or surface structures of one or two chemical components. NNs designed for the robust prediction of adsorbate interactions with dynamic surface constructions still remain challenging to create. Furthermore, since the training methods for NNs are so flexible, they are also easily expanded upon as demonstrated in previous work.

The purpose of this work is two fold:

1. To determine methodologies for producing a neural network capable of accurately reproducing properties of oxygen interactions with a dynamic Pd fcc(111) surface.
2. To establish a database of high quality PdO DFT calculations to use as a basis for future work, such as the inclusion of a third chemical species.