HIGH-THROUGHPUT SCREENING FOR NOVEL PROTON CONDUCTORS
Embargo until
2020-05-01
Date
2019-02-22
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Johns Hopkins University
Abstract
Solid Oxide Fuel Cells (SOFCs) convert chemical energy to electrical energy very efficiently compared to systems that require mechanical processes. However, SOFCs require high operating temperature in the range of 650 C – 1,000 C which limits the environments where they can be used. Switching the diffusing ion from oxygen to hydrogen could potentially halve the operating temperature into the range of 400 C – 800 C. This system is what is known as Proton Conducting Solid Oxide Fuel Cell (PC-SOFC). Despite this prospect, three decades since the discovery of proton conduction in solids, we still do not have a mainstream system that takes advantage of proton conduction to generate electricity. The main obstacle to engineer this system lies with the material used as the solid electrolyte. With the rise of computing power and number of publicly available materials dataset, we aim to find an alternative material to solve this problem.
Along the way, we acknowledged that a significant number of quantum mechanical calculations would need to be done to train and verify our methods. We took this opportunity to enhance the grids used to evaluate Brillouin zone integration which would directly impact the calculation cost. This was done by creating a database of pre-calculated grids which was made through an exhaustive search for every specified grid density. This database was then benchmarked on 102 randomly-selected materials. We found that for well-converged calculations, the grids generated through the database have less than half as many symmetrically irreducible k-points as the conventionally generated Monkhorst-Pack grids. This means these calculations would require less than half the time to complete for the same level of accuracy.
On searching for alternative proton conductors, we began by limiting our criteria to proton mobility and identifying the activation energy for proton diffusion as the descriptor. The entire ICSD database was filtered for oxides and then categorized by space group and structure type. We then selected a total of 51 samples from the cubic perovskite, hexagonal perovskite, spinel, and elpasolite structure types along with several other randomly selected oxides as training data. These training data was generated using a combination of Density Functional Theory (DFT) calculation and Nudged Elastic Band (NEB) method.
An energy model based on the bond-valence method and the screening Coulomb interaction was made and fitted to the training data. In analyzing the data, we found that each structure type has noise which would be minimized when the average is taken over the structure type. The difference between the training data and energy results for the structure types mentioned above were below 0.05 eV, where 0.1 eV is the accuracy for activation energy approximation using the NEB method.
Finally, we screened all the oxide structures from the ICSD database that has more than 20 members. Based on the ranking of proton mobility, we identified CrVO4 orthorhombic structure type as a novel alternative to the cubic perovskite. We verified this result by running DFT calculations and NEB method on 6 members of the CrVO4 structure type and found that the average difference for this structure type was 0.07 eV.
Description
Keywords
k-points, brillouin zone, reciprocal space, high-throughput, screening, proton conductor, diffusion, database, computational