Materials modeling is used extensively by researchers worldwide in order to analyze, predict, and modify the behavior of materials and chemicals for energy applications. The pace of research in materials modeling is rapidly growing. These efforts create large, complex, unstructured collections of data that require a specialized digital infrastructure to enable scientists and engineers to utilize these data sets for specific aspects of materials science and facilitate the inverse materials design approaches. Exabyte Inc. and collaborators propose a digital ecosystem realizing what is now called the FAIR (Findable, Accessible, Interoperable, and Reusable) data principles. The ecosystem improves the speed and efficiency of the R&D of new advanced materials and chemicals by allowing the creation, storage, exchange, and analysis of materials data utilizing state-of-the-art artificial intelligence (AI), including AI techniques to find measurable qualifiers or physical descriptors that reduce the complexity of multidimensional databases and enable inverse design concepts for new materials functionality. We plan to implement a repeatable and reusable procedure able to: (i) construct property maps of materials, and (ii) identify statistically exceptional regions in the data space. Two AI methods have been developed for this by one of the authors: SISSO (Sure Independence Screening and Sparsifying Operator) and the subgroup discovery approach. During Phase I, we aim to (1) elaborate and apply the aforementioned methods on data generated by the earlier DOE-supported computations (MaterialsProject), (2) run calculations to augment any missing parts, (3) group materials within the resulting dataset with respect to descriptors and target properties, (4) characterize the data with respect to the new types of applications and properties, and (5) demonstrate the validity of the resulting inverse design approaches. We plan to focus on functional materials and first consider electronic properties relevant to semiconductor technologies in the energy sector. We plan to make the resulting model(s) accessible to new users, who will contribute their insights, run additional simulations to produce data, and thus help further improve the model(s). During Phase II, we plan to (1) include other material types, (2) expand the set of application areas and properties, (3) incorporate additional sources of computational data, and introduce experimental data to calibrate and validate the models and improve veracity. The proposed ecosystem improves the speed and efficiency of the materials research and development (R&D) and enables the development of new kinds of products in energy-relevant industrial sectors. Computational materials design drives the transformation of the industrial materials R&D and will become a multi-billion market by 2025. Additionally, the platform facilitates (a) dissemination of the flagship DOE initiatives, (b) creation of the collective intelligence, and (c) education of a new generation of scientists accelerating future scientific progress.
