93 Modeling Transition Metal Nanoclusters for Hydrogen Storage Capacity Using Artificial Neural Networks

Chemistry has often sought to explain bulk properties by looking at the molecular features that drive an overall system. This leads to chemical relationships, which predict whether such a property can be predicted based on the features of the smaller molecules. By correctly modeling these small molecules, the orbital energies, bond angles, lengths, and correct crystal structure are discovered using highly accurate ab initio molecular modeling techniques. The molecular orbital energies give an accurate representation of the bond the hydrogen and metal form in a metal hydride complex. Artificial neural networks using feed-forward back-propagation architecture reveal the relationship in this data to the percent hydrogen storage capacity of the metals in the bulk system. They do this by a self-consistent procedure adjusting the weight of the input according to the error of the preceding calculation. We are studying numerous transition element clusters to represent the bulk property of hydrogen adsorption. From these studies, the most promising potential materials will be further studied experimentally.