Artificial neural network prediction indicators of density functional theory metal hydride models

• We modeled examples of a well-studied section and general section of the Sandia National Lab metal hydride database.
• An artificial neural network found correlations for several properties found in the database.
• The percent weight hydrogen was predicted from only DFT parameters.
• Statistical models give support to the prediction of the artificial neural network.
• Several unknown compounds were predicted and verified with the literature source.

The metal hydride is a capable candidate for mobile storage for hydrogen-powered vehicles. An artificial neural network (ANN) has proved useful for many applications, and capable of much more in discovery of new materials. Because of its ability to generalize from examples presented to it, an ANN is a powerful tool for discovering new metal hydride combinations. An ANN can deduce quantitative structure property relationships for metal hydrides. The ANN found correlations between fundamental electronic and energy values modeled ab initio and several experimental parameters. Some of the properties successfully predicted with good correlation are: entropy, enthalpy, temperature at 1 atm of pressure, pressure at 25 °C, and the percent weight of hydrogen stored. The marriage of ANN to computational modeling produces good predictions for many important properties of metal hydrides.