High-Throughput Computational Search for Stable Compositions and Configurations in High-Entropy Perovskite SrTiO3

Perovskite oxides exhibit remarkable versatility across various applications due to their structural and chemical adaptability.<br/>SrTiO3 (STO) stands out as a potential water-splitting photocatalyst with a band gap of 3.4 eV, while for efficient solar energy conversion of the photocatalyst, the band gap should be narrowed to less than 2.0 eV. To address this, we have investigated multi-doped STO candidates, aiming to narrow the band gap and enhance photocatalytic performance. For efficient solar energy conversion, there is still room for performance improvement such as band gap control by changing the composition of the constituent metals. Controlling the valence band structure through combinatorial doping of multiple atomic species with different ionic radii and valences is expected to play a critical role in improving catalytic performance. On the other hand, doping with a different atomic size or valence leads to several difficulties in synthesis. Our approach involves the design of water-splitting photocatalytic materials through heterogeneous composite doping, which introduces high configurational entropy into STO. This strategy relies on the "elemental cocktail effect" arising from substitution with multiple elements.<br/>Adopting the "PreFerred Potential" (PFP), which is a neural network potential [1], we have created a substantial dataset of metastable optimized structures for hetero-metallic substitution at Sr-sites within STO. The PFP is based on over 42 millions of density functional theory (DFT) datasets and achieves high accuracy and generality for nearly all elements of the periodic table, excluding radioactive ones. Its speed and precision surpass traditional DFT solvers, expediting the optimization process.<br/>Using PFP, we have calculated over 10,000 multi-stable optimized structures. Our extensive computational dataset has enabled predictions of the relationship between metal-site composition ratios and the formation free energy of substitutional solid solution phases, taking into account configurational entropy. Our detailed mapping of performance over composition and stability not only identifies optimal metal combinations for synthesis but also enables the prediction of the limit to solid solution formation. Indeed, our calculations for multi-element doping in STOs have successfully replicated experimental results.<br/>In our presentation, we will showcase how our simulations provide novel insights for material design within the relevant material families. Our calculation results, which consist of substitution composition simulation models created by PFP, are invaluable for guiding material design and advancing data science applications, such as supervised machine learning. These large amounts of stable structure data, categorized by composition, structural stability (formation free energy), configurations, and unit cell geometry (including space group), do not contain electronic structure information (since PFPs are quasi-atomic potentials).<br/>To further our research, we have begun collecting accurate band gap data using DFT on "Fugaku," Japan's preeminent supercomputer. Our presentation will discuss these band gap simulation results and structural insights, contributing to the progression of photocatalytic material design. We strongly believe that these results help to predict feasible metal composition rates for future perovskite synthesis.<br/><br/>[1] S. Takamoto, et al. Nat Commun 13, 2991 (2022).<br/>[2] Y. Harashima et al., Appl. Phys. Lett. 122, 262903 (2023)