M2P2_HEA: Multi-scale methodology for predicting the properties of high entropy alloys

The implementation of the multi-scale methodology will contribute to optimizing already existing catalysts, but it will also accelerate the search for new materials with specific properties targeted towards a predefined application. One of the innovative aspects of the project lies in the fact that, to our knowledge, there is not yet a predictive methodology using neural networks constructed from experimental data, making it possible to directly define the structure and composition of a material based on a targeted property.

To reach the project’s goal, we will use a method that combines artificial intelligence (AI) techniques based on neural networks with multi-scale simulation and modeling tools. This will help us figure out the structure and thermodynamic properties of the system under study and connect them to the catalytic properties that are important.

The project is structured in two stages.

In the first, we will develop a fitting procedure to obtain the 3D atomic structure of materials from transmission electron microscopy (TEM) images and X-ray absorption spectra (XAS). The fitting algorithm will be developed using TEM images and reference X- ray absorption spectra simulated using quantum physics methods from in silico materials generated using classical methods (force fields, Monte Carlo, Molecular Dynamics). Then the procedure will be validated by comparing the data calculated on materials whose development we have mastered with experimental microscopy and spectroscopy data in order to obtain an atomistic model of the materials studied.

In a second step, the atomistic model fitting procedure will be used to determine the atomistic structure of a collection of real samples from their XAS and TEM experimental characterization. The catalytic properties measured experimentally, associated with the atomistic structure of the samples, will be used to instruct a neural network ultimately intended to search for a priori ideal compositions and structures of materials based on a targeted application. HEAs are particularly suitable for the development of the envisaged methodology because, due to their multi-component nature, they offer a large number of configurations that can be used to validate the atomistic model. For the catalytic properties, we chose the ethanol oxidation reaction (EOR). It is an important molecule, used both as a fuel and as a reagent for the production of higher-value compounds. In terms of foresight, this methodology, once developed, should be able to be adapted to other materials science systems such as multi-functional materials (electrical, magnetic, and optical properties) and materials for health (mechanical properties).

The consortium is made up of five teams. The IPCMS (Strasbourg) is responsible for developing the atomistic model. The HEAs will be synthesized at LCMCP (Paris), characterized ex-situ and under operando conditions by quick-EXAFS absorption spectroscopy (SOLEIL) and electron microscopy (IPCMS), and tested in catalysis at LEM (Paris). IFPEN (Lyon) will implement the validated 3D atomistic model to train the neural networks, which will then be used to guide the search for specific configurations corresponding to the desired catalytic property.