(Invited) Rapid Computational Screening of Materials for Water Splitting Using Ab Initio and Machine Learned Models: Thermodynamic and Kinetics of Solar Thermal H2 generation

Abstract

To continue to meet global energy demands, efficient methods of utilizing renewable energy must be developed. Converting solar energy into chemical fuels is a promising approach, but efficient and cost effective methods for producing solar fuels have not yet been developed. Solar thermal water splitting is a particularly promising possibility because it has a high theoretical hydrogen production efficiency. However, achieving this efficiency requires finding the proper redox material. Currently, the most promising materials are metal oxides, including spinels and perovskites. These materials split water via a high temperature cycle in which the material is reduced, forming oxygen vacancies, in one step, then oxidized by stripping oxygen from water in the next. An efficient STWS process is extremely demanding on materials and requires that they be thermodynamically capable of being reduced, withstand the high temperatures of solar thermal water splitting, and form oxygen vacancies with enough energy to reduce water. Furthermore, materials must also be kinetically viable, able to complete the water splitting cycle quickly enough to allow for large scale production of hydrogen. The initial reduction step uses concentrated solar energy to heat the material to temperatures exceeding 1350oC and creating oxygen vacancies in the process. This reduced material can subsequently be oxidized by splitting water and forming hydrogen. A variety of materials have been found that can undergo this process, CeO2, FeAl2O4, and SLMA, but they all fall short in some aspect. To discover new STWS materials and optimize these compounds for their STWS abilities, it is important to have a detailed understanding of the electronic structure which governs both the thermodynamics and kinetics of these materials. Using atomistic modeling, we identify the intrinsic material properties that enable high performance for the oxidation and reduction steps of the two-step cycle. This understanding can be used to guide the rational doping of these compounds as well as establish design principles for the design of new high performance materials. While the thermodynamic and kinetics of STWS chemistry can be directly modeled using quantum chemical methods, these calculations are too computationally intensive to examine large numbers of materials to identify promising candidates. This is especially the case for calculations that consider the disordered spin structure and high temperature atomic structure of these materials – which we explicitly account for in our approach. We have developed efficient computational methods to identify STWS materials with desirable thermodynamic and kinetic properties with these critical considerations. We have used high level calculations on a subset of materials to inform a machine learning approach that identifies descriptors for the activation barriers for the rate limiting step for water splitting. These descriptors can be determined from simpler calculations and will allow for rapid calculation and determination of water splitting materials. This work focuses on the use of density functional theory to develop general descriptors of the water splitting reaction and in turn a more fundamental understanding of the reaction mechanism and efficient approach to screening materials for their STWS kinetics.