Performance prediction of high-entropy perovskites La0.8Sr0.2MnxCoyFezO3 with automated high-throughput characterization of combinatorial libraries and machine learning

Carlota Bozal-Ginesta,Sarah Fearn,Sergio Pablo-García,Giulio Cordaro,Juande Sirvent,Francesco Chiabrera,Changhyeok Choi,Lisa Laa,Marc Núñez,Andrea Cavallaro,Fjorelo Buzi,Ainara Aguadero,Guilhem Dezanneau,John Kilner,Alex Morata,Federico Baiutti,Alán Aspuru-Guzik,Albert Tarancón

DOI: https://doi.org/10.26434/chemrxiv-2024-4bn5w-v2

2024-05-24

Abstract:Perovskite oxides (ABO3) represent a large family of materials with wide application in many fields due to their celebrated structural and chemical flexibility. Such a vast space of compositions requires efficient exploration strategies now possible with automated high-throughput experiments combined with machine learning prediction algorithms. In this study, we investigate the compositionperformance relationships of high-entropy La0.8Sr0.2MnxCoyFezO3±𝞭 perovskite oxides (0 < x, y, z <1; x+y+z≈1) for application as oxygen electrodes in Solid Oxide Cells. After deposition of a continuous compositional map using thin film combinatorial pulsed laser deposition, we obtain experimental data of structural, composition and functional propertiesfor the whole material family under study through a combination of six advanced characterization methodologies with mapping capabilities. We prove that supervised machine learning methods, particularly random forests, effectively capture the complex relationships between composition, structural features, and electrochemical performance including oxygen transport properties. Using these predictive methods, we create an accurate continuous map of performance for the whole compositional space under study and we open it to the community. Moreover, our model yields an unambiguousstatistical correlation between the distortion of the oxygen sublattice (obtained from spectral analysis of their Raman-active modes) and the highest performances. Finally, the study consistently identifies Fe-rich high-entropy oxides as the optimal compounds with the lowest area-specific resistance values for oxygen electrodes at 700°C. Overall, the work proves the potential of a detailed exploration of relevant chemical spaces by coupling highthroughput experiments and machine learning models to gain new insights and optimize relevant families of materials.

Chemistry

What problem does this paper attempt to address?

The paper aims to address the following issues:

1. **Material Performance Prediction**: Investigate the composition-performance relationship of high-entropy perovskite oxide La0.8Sr0.2MnxCoyFezO3 (referred to as LSMCF) for solid oxide cell (SOC) oxygen electrode applications. Predict the material's performance through automated high-throughput characterization techniques and machine learning algorithms.

2. **Combination of High-Throughput Experiments and Machine Learning**: Utilize combinatorial thin film deposition technology to prepare continuous composition LSMCF films, and obtain their structural, compositional, and functional characteristic data through various advanced characterization methods. These data are used to train machine learning models, particularly random forest models, to capture the complex relationships between composition, structural features, and electrochemical performance.

3. **Optimization of Material Design**: Through machine learning modeling, it was determined that high-entropy oxides with higher iron content have the lowest area-specific resistance (ASR) values, providing new insights for optimizing oxygen electrode materials. Additionally, the study revealed a statistical correlation between oxygen sublattice distortion and its highest performance.

In summary, the paper demonstrates that by combining high-throughput experiments and machine learning models, it is possible to effectively explore the relevant chemical space, obtain guiding principles for new material design, and optimize the performance of specific material families.