Machine learning methods for mixed oxydes at the atomic scale

Envoyé par JBouchet 
Machine learning methods for mixed oxydes at the atomic scale
mercredi 31 mars 2021 10:07:51
A fully funded PhD position on multi-scale modeling with machine learning of nuclear fuels is open at CEA CADARACHE (Aix en Provence), France. The PhD is expected to start autumn 2021 and lasts three years.

The description at the atomic scale of the effects of irradiation, in particular the defects, on the properties of the nuclear fuel (UO2 and MOX) is a major issue for the CEA. Numerous works have shown the relevance of DFT + U to describe 5f electrons and characterizing these defects, but also its limits in terms of studying extended defects. At the higher scale, classical molecular dynamics uses semi-empirical potentials, well suited to calculate thermodynamic properties, but poorly predictive on the defect and migration energies.
The ambition of this thesis is to maintain DFT precision while processing simulation boxes of several thousand atoms, this by developing a digital potential based on a DFT database.
The first objective is therefore the development of a CEA database from DFT simulations in order to characterize the different materials (elasticity, temperature properties) as well as the identified defects (vacancies, interstitials, Frenkel pairs, etc.). This database, shared and archived on the CEA network, will be enriched during the thesis as needed.
The second objective is the development of a digital potential based on Machine Learning tools. For this, neural networks associated with descriptors of the local environment of atoms (symmetry functions) will be used.
The role of the inhomogeneous spatial distribution of defects will be studied and compared to statistical methods in link with another PhD in the laboratory. The PhD thesis is proposed within the Framework of the CEA FOCUS project dedicated to numerical experiments and numerical twins. Strong interactions with local and international experimental teams At the end of the thesis, the candidate will have acquired highly valuable skills in multi-scale materials modeling, data bases and machine learning.

The thesis will take place at the CEA CADARACHE, in the Provence area close to Aix en Provence.

The position is funded by a grant from CEA (net grant: ~1600 €/month).

The candidates must have a Master degree in quantum or solid-state physics and have strong computational skills. They shall send a CV, a letter of motivation, a transcript of academic results, and two contacts for references to:

Johann Bouchet (
+33 4 42 25 61 80

Bâtiment 151
13108 Saint-Paul-lez-Durance