Communication Dans Un Congrès Année : 2024

Screening new Entropy Stabilized Oxides by DFT calculations and machine learning

Résumé

The Entropy Stabilized Oxides (ESO) are the MOx oxides where the metallic site M is a mixtureing of at least 5 elements with a composition between 5 and 35%. At high temperatures, these materials are stabilized by the configuration entropy. The ESO can have some special properties, such as a high dielectric constant, and can be used as catalysts, supra-conductors and electrolytes in Li-batteries [1]. Considering a base of 16 cationic elements, this leads to 4368 of possible quinary equimolar ESO for each new structure. With the addition of a 1% step variation in composition, it's up to 2.4 billion possibilities. Our work aims to predict, through DFT calculations and machine learning, which ESO could be a meta-stable candidate and what would be the temperature of stabilization. We use the Special Quasi-random Structure (SQS) to model the multi-component mixing on at the cation sites [2]. Then, some DFT calculations are used to calculate the formation enthalpy of oxides to build a learning database. Thanks to a convex hull home-made code, we can determine which phases are the most stable one. The entropy, considered only as configuration entropy, is approximated from the Boltzmann's formula, with the equation: S = – ∑ xi ln(xi). With the 4368 of possible equimolar quinary ESO in the NaCl structure, we built a learning database from high throughput DFT calculations, comprising approximately 10% of the entire combination set. Ustilizing supervised machine learning, we aimed to predict the formation enthalpy of all configurations with to an accuracy of a few kJ/mol, where with a selection of potential candidates was proposed for experimental verification. Despite the use of this model, DFT calculations can remain time-consuming, and many compositions may not be feasible to synthesize. Therefore, we propose the use of Bayesian optimization (BO) [3], a machine learning model that quickly identifies the optimal inputs for optimizing a criterion C. Thanks to the multi-dimensional convex hull code that we developed, in addition to the approximation of entropy from Boltzmann’s formula, we can determine the stabilization temperature of a given ESO. Thus, the BO can be applied to design ESO based on a new crystal family to establish the set of elements with the lowest stabilization temperature. REFERENCES : [1] N. Dragoe and D. Bérardan, Science, 2019, 366, 573 [2] A. Zunger et al, Physical Review Letters, 1990, 65, 353 [3] J. Mockus, Journal of Global Optimization, 1994, 4, 347

Domaines

Fichier non déposé

Dates et versions

hal-05582951 , version 1 (07-04-2026)

Identifiants

  • HAL Id : hal-05582951 , version 1

Citer

Sébastien Junier, Céline Barreteau, Jean-Claude Crivello. Screening new Entropy Stabilized Oxides by DFT calculations and machine learning. Journées Plénières du GDR IAMAT, Jul 2024, Toulouse, France. ⟨hal-05582951⟩
5 Consultations
0 Téléchargements

Partager

  • More