Design of new entropy-stabilized functional oxides – DETOX

Numerical simulation tools have been used for several years to guide the discovery of new compositions. Generally speaking, the approach involves calculating the formation energy of a possible set of compositions for a given crystal structure, and comparing this formation energy with that of potentially competing phases; a lower energy generally indicates a stable phase, and one that should therefore be tested experimentally.

This ‘classical’ method cannot be applied to high-entropy oxides, and even less so to those stabilised by entropy:

* in a high-entropy oxide, several cations (at least 5) will occupy the same sublattice of the crystal structure with a random distribution. Taking this chemical disorder into account will involve very long computation times, which are difficult to reconcile with the practical implementation of a very large number of calculations (bearing in mind, for example, that over 4,000 calculations would be required to explore all combinations of 5 cations from an initial set limited to just 16)

* when the formation energy of a possible high-entropy oxide composition is obtained, it is not enough to compare it with those of ‘a few’ competing phases, but with all possible phases with 4 cations, 3 cations, (...), i.e. the entire convex hull of formation energies, which would again require computation times that are difficult to reconcile with practical implementation.

The method we have proposed to implement and test in this project therefore relies on coupling these numerical simulation tools for calculating formation energies with artificial intelligence tools (supervised learning and Bayesian optimisation) in order to drastically reduce computation times – and make them compatible with practical implementation – without sacrificing accuracy and thus predictive power.

The main results of our project are:

* the development of a rational method for designing and predicting compositions, enabling both seamless interaction between theory and experiment with regular feedback, and the use of more accurate (and therefore more computationally intensive) simulation tools.

Indeed, by combining formation energy calculation tools with supervised learning and Bayesian optimisation, we have been able to reduce the number of calculations required to predict the most favourable compositions by over 90%: starting from a small number of initial calculations, these AI tools enable us to converge preferentially towards these favourable compositions without having to explore all possible compositions AND to make regular back-and-forth iterations between simulation and experimental validation without having to wait until calculations for all compositions have been completed.

* the parallel development of a second method for discovering new compositions using data mining of crystallographic databases

* the discovery of several new compositions that may have application potential with promising properties, depending on the composition, such as thermoelectric materials, dielectric materials or thermal barriers.

As a fundamental research project aimed at developing a new method for the discovery and design of materials, DETOX was not intended to pave the way for new applications in the short term. Nevertheless, the use of the method we have developed should undoubtedly help to accelerate the discovery of new compositions, some of which should have potential applications.

Entropy stabilized oxides, which have been discovered recently, constitutes a new class of functional materials with unique properties appealing for various applications for example as solid state electrolytes, thermal barriers, catalysts, dielectric or magnetic materials. To date, the design of new compositions mostly relies on trial-and-error or educated guess based on phenomenological criteria, which slows down the development of the field. In that context, our project aims at implementing a new methodology coupling theoretical calculations and experimental studies towards the rational design of new entropy-stabilized functional oxides, and at evaluating the potential of these new materials for applications.