Alloy Informatics Using Charge Profiles For Energy and Hydrogen Storage Applications

Studies have confirmed the crucial role of solid-state based solutions in hydrogen storage, with respect to liquid or gas states which cannot ensure comparably high amounts of storage per unit volume [1]. The empirical approach is not an option when it comes to the complexity of the quest for hydrogen storage materials, as well as to the strict urgency of a transition to clean and sustainable energy solutions. Here, we pursue machine learning approaches to composition searches for hydrogen storage materials. The compositional space to be explored is extremely vast, and still its vastness is not defined due to the on-going explorations, not only spanning the chemical compositions, but also the dimensionalities of the possible candidate systems [2]. Therefore, there is the need of finding a general predictive approach based on first-principles.<br/>Electron charge profiles shall be illuminating for identifying and classifying local defect properties, thus characterizing alloying, and this is why our machine learning approach is based on the ab-initio analysis of the charge profiles of interstitial hydrogen in pure FCC bulk metallic crystals. Our descriptors are constructed by differential electron charge profiles for a large variety of crystals and also, after analysis of their characterizing local quantities, such as radii and extrema, as well as consider the full profile properties and their integrals. The integration of the profiles and the distinctive features of the effects of hydrogen are discussed, and can be exploited to infer hydrogen mobility in the surrounding matrix, as well as electron clouds motions around hydrogen. After the building of a density dataset, we perform unsupervised machine learning. Basing on preliminary tests we discuss interesting trends between the profile features and the properties of the embedding pure crystals (atomic number, valence electrons, lattice constant, bulk modulus), uncovering compositional relationships (among different crystals) and non-compositional ones (among different hydrogen interstitial equilibrium positions in the same crystals). The results also shed light over the influence that the compressibility of the systems have upon the radii of the hydrogen electron density profiles, which finds interesting connections with the complementary problem of systematically changing an impurity in a same crystal [3].<br/>The results of this study might pave the way to novel material informatics methods for prediction of hydrogen storage behaviours by interpretation of the differential charge density profiles around the defect. Our goals are to understand up to which extent such charge profiles method can be exploited to predict diffusive, as well as the adsorption and desorption properties, overcoming the limited predictive capacity of empirical and thermodynamic based methods.<br/><br/>[1] P. Jena, J. Phys. Chem. Lett., 2011, 2, 206–211.<br/>[2] Xuebin Yu, <i>et al.</i> Progress in Materials Science, 2017, 88, 1–48.<br/>[3] L. Messina, <i>et al.</i> PHYSICAL REVIEW B 2016, 93, 184302