Data mining density of states spectra for crystal structure classification: An inverse problem approach

Abstract

Abstract:The ability to model the density of states has been a long-standing problem in condensed matter physics. The classical methods that have been used are based on a variety of approaches, ranging from maximum entropy methods to recursion methods involving high dimensional data. In this work, we classify the crystal structure of an alloy based on the electronic structure, the inverse process of first principles calculations which calculate the electronic structure from crystal structure-based inputs. Here the electronic structure is represented by the density of states, and the classification of crystal structures is achieved through data mining. In addition to classifying alloys by crystal structure, we can classify alloys based on the degree of tetragonality and stoichiometry using solely the density of states spectra. This work seeks to describe the relationship between crystal and electronic structure based on a quantitative interpretation of the density of states, while discussing how capturing the principal contributions of the density of states suggests future work in modeling the density of states using less computationally expensive data mining techniques. 

Classification of Oxide Compounds through Data‐Mining Density of States Spectra

“…Because the curves were normalized for this analysis, the differences related with atomic size, chemistry, and orbital structure are due to a change in curve shape and is not reflective of the higher intensity of peaks related with larger atoms. We also conclude that stoichiometry has a bigger impact in defining the electronic structure for the oxide systems considered here as compared with the changing chemistries, interestingly coinciding with our previous findings on intermetallic systems 7 …”

“…We introduce here a methodology that addresses this problem by decomposing the DOS curve to its basic contributions so that the relationships between crystal structures and chemistries are identified based only on an electronic structure input, effectively providing a screening tool for assessing the electronic structure. This methodology was originally developed for analyzing intermetallic systems 7 but by analyzing ceramic systems, the robustness is demonstrated via application to different bonding types. For this paper, we consider only systems having perfect translational symmetry, and do not consider defects or local atomic configurations.…”

Electronic structure prediction via data-mining the empirical pseudopotential method

“…The classification analysis is performed with principal component analysis (PCA) and the prediction is done with partial least squares (PLS). Both are powerful tools for the analysis of materials and have been used to address materials science issues for a variety of reasons and materials 13–23.…”

Descriptors for dielectric constants of perovskite-type oxides by materials informatics with first-principles density functional theory

“…Broderick et al conducted successive studies on mining descriptors [20][21][22][23]. They clarified the relationship between the electronic states and physical properties of target materials through a data mining analysis, with the density of states (DOS) as spectral data.…”