We shall describe the development of Quantitative Materials Structure-Property Relationships (QMSPR) models (both binary and continuous) to compute superconducting and other properties of materials from their structural characteristics. We have compiled two superconductivity datasets consisting of 165 materials with continuous Tc values ranging from 2 K to 133 K and 466 materials with binary Tc values. To the best of our knowledge, these collections represent the largest publicly available superconductivity datasets. To characterize the structure of materials, we have developed novel descriptors that can reflect constitutional, topological, and spatial (stereochemical) characteristics of materials. These descriptors were derived from those used to characterize small organic molecules for cheminformatics applications.  More specifically, we have modified the SiRMS descriptors employed previously to characterize mixtures of organic compounds to make this method suitable for computing descriptors for materials. To calculate SiRMS descriptors, each material is represented by as many parts as needed according to its crystal structure. Thus, the structure of every material is characterized by both bounded and unbounded simplex as well as by 'mixture' descriptors. Atoms were differentiated according to their symbol, partial charge, and their group in the Periodic Table. Using these descriptors in combination with the Random Forest technique, we obtained the balanced accuracy (BA) of 0.9 for binary model and Q2=0.7 for continuous model. Interpretation of these models also allowed us to analyze how structural features of different materials influence the critical temperature. Current efforts are under way to extend approach for other properties such as band gap.  As additional computed and experimentally determined properties of materials become available we expect rapid proliferations of QMSPR approaches as significant part of materials informatics.

Quantitative materials structure-property relationships (QMSPR) modeling using novel electronic and structural descriptors

248th National Meeting (2014)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.