One-year position, which may be renewed, starting preferably in September, 1st 2025

We are recruiting a research engineer in computational mechanics to work in the ERC RhEoVOLUTION project team https://erc-rheovolution.gm.univ-montp2.fr/) in developing machine learning tools for modelling the evolution of elastic and viscoplastic anisotropy in Earth materials. This work is essential for advancing our understanding of geodynamic processes and for improving the resolution and realism of large-scale geophysical simulations. Indeed, the propagation of seismic waves and the deformation of Earth’s mantle are fundamentally controlled by the anisotropic properties of rocks, which arise from the development of a crystallographic preferred orientation (CPO) of olivine—the dominant mantle mineral—in response to strain. Modeling the evolution of this texture-induced anisotropy is critical for: (1) mapping the deformation in the deep Earth using seismic observations and (2) Accurately simulating deformation and stress patterns in mantle convection and plate tectonics. Current numerical methods for simulating CPO evolution are either too simplified to capture the relevant physics or too computationally intensive for direct implementation in geodynamical models. To address this, we have initiated the development of neural network surrogate models, trained on synthetic datasets generated using polycrystal plasticity models. These models aim to provide fast and memory-efficient predictions of how elastic and viscoplastic tensors evolve under arbitrary deformation histories. Initial work has shown promising results for 2D deformation scenarios using feed-forward neural networks. However, recursive applications of the model currently suffer from error accumulation at large strains, limiting their practical use in long-term geodynamic simulations.

The research engineer will play a key role in improving and extending this machine learning framework. Main objectives include:
– Developing robust recursive modeling strategies to mitigate error accumulation in long-strain predictions, including: (i) Comprehensive analysis and characterization of the training database to ensure adequate coverage of deformation scenarios relevant to mantle dynamics, and (ii) Testing and implementing advanced neural network architectures, including physics-informed networks that respect tensorial symmetries
– Extending the surrogate models to predict elastic anisotropy evolution under 3D deformation fields and to predict viscoplastic anisotropy, leveraging both the shared physical origin of elastic and viscous anisotropy and previous work within the group on parameterizing viscous behavior.
– Integrating these developments into geodynamic models, in collaboration with the broader RhEoVOLUTION research team
– Exploring interdisciplinary applications, such as the modeling of anisotropic flow in glaciology or metal forming processes in materials science
– Ensuring the documentation, maintenance and publication of the methods and tools developed

To apply  : https://emploi.cnrs.fr/Offres/CDD/UMR5243-HELOUR-066/Default.aspx