Actinide oxides, such as UO2, PuO2, and their mixed oxide alloys (MOX), are fundational fuel materials for existing and next-generation nuclear reactors where their thermomechanical properties dictate fuel safety and operational performance. These materials are polar ionic crystals, and their dielectric properties therefore play a central role in their physical behavior. For instance, long range electrostatic interactions and dynamical polarizabilities shape the phonon spectrum, control the thermodynamic properties and impact the heat transport of the solids. From a simulation point-of-view, neglecting these effects risks systematic error in the prediction of properties that matter for fuel performance.

This project will develop and apply novel machine-learning atomistic simulation methodologies that explicitly incorporate dielectric properties into the description of actinide oxides, bridging a critical gap between current frameworks and the physic of polar materials.

In the first phase of this work, you will train a machine-learning interatomic potential with explicit long-range electrostatic capability for UO2, PuO2 and MOX. The potential will then be applied in large-scale finite-temperature simulations to quantify the impact of electrostatic interactions on elastic properties, thermal expansion, thermal conductivity and key thermodynamic properties such as heat capacity. The same potential will be used to compute infrared spectra, which are directly sensitive to the dynamic charge response of the material. Comparison with available experimental measurement will serve as an assessment of the accuracy of the atomistic model for actinide oxides.
Building on this framework, the second phase will extend the MLIP architecture to couple explicitly to external electric fields, enabling the direct simulation of polarizability derivatives and Raman-active vibrational modes. This will allow the computation of Raman spectra directly from molecular dynamics, providing a link between spectroscopic signatures and atomic structure.

The results produced at each stage of this project will constitute original contributions to the field and will be disseminated through publications in international peer-reviewed journals and presentations at international scientific conferences.

Beyond its scientific outcomes, the project will provide you with a versatile skill set applicable to many areas of materials science: ab initio electronic-structure calculations, the development of machine-learning interatomic potentials, classical molecular dynamics, and the key concepts of statistical physics and condensed-matter physics. You will also receive training in high-performance computing, with direct access to CEA supercomputing facilities. The supervising team includes recognized experts in all of these fields.
You will join the Fuel Behavior Modeling Laboratory at the Research Institute for Nuclear Systems for Low-Carbon Energy Production (IRESNE, CEA Cadarache), a dynamic research team where you will have regular opportunities to interact with fellow PhD students and researchers. This environment also provides extensive opportunities for national and international collaboration, including with:
• Experimental researchers from the Nuclear Fuel Studies Department,
• Teams from other CEA centers (Saclay, CEA/DAM),
• International partners.

This rich and multidisciplinary context will enable you to fully engage with the scientific community focused on nuclear materials science.
[1] Dubois, E. T., Tranchida, J., Bouchet, J., & Maillet, J. B. (2024). Atomistic simulations of nuclear fuel UO2 with machine learning interatomic potentials. Physical Review Materials, 8(2), 025402.
[2] King, D. S., Kim, D. Zhong, P., & Cheng, B. (2025). Machine learning of charges and long range interactions from energies and forces. Nature Communications 16, 8763.