Missions

The Geosciences Montpellier laboratory is recruiting a research engineer as part of the RhEoVOLUTION ERC project, which aims to develop innovative approaches to modeling deformation in the Earth's interior. The engineer's mission will be to develop supervised machine learning models to predict how rocks change their structure and mechanical properties - in particular develop anisotropy in their elastic and viscoplastic behaviors - when they deform. This is essential to :
- Image the Earth's interior by analyzing the propagation of seismic waves (elastic anisotropy).
- Accurately model convection in the Earth's mantle and plate tectonics (viscoplastic anisotropy).
At the heart of the project is olivine, the most abundant mineral in the mantle. Both anisotropies result from a preferential orientation of olivine crystals (“texture”), which is acquired and modified by deformation. Physical models simulating this process are either too simplified or too computationally expensive for large-scale geodynamic simulations. This is where machine learning comes in. We have already set up a framework and trained neural networks on synthetic data generated from crystal plasticity models. These surrogate models effectively predict, for short deformation histories, how olivine textures and elastic anisotropy evolve in 2D flows. However, challenges remain, particularly when it comes to chaining predictions over longer deformation histories.

Activités

The primary objective of IR will be to make the machine learning models robust for long-term recursive use, which is essential for simulating realistic geodynamic scenarios. Tasks include:
- analyzing and expanding the training database to better represent the diversity of flow conditions observed in the mantle.
- exploring new machine learning architectures, including physically informed networks that respect tensor symmetries.
- improving generalization to prevent the accumulation of errors during iterative predictions.
Next, he/she will focus on:
- extend the models to 3D deformation models.
- model viscoplastic anisotropy, based on the links between texture and elastic and viscoplastic anisotropies.
- work on integrating models into geodynamic finite element codes, in collaboration with the RhEoVOLUTION team.
Applications beyond the geosciences, such as glacier dynamics and material shaping in metallurgy, may also be explored.

Compétences

The candidate should have a solid background in mechanics and applied mathematics, proven skills in scientific programming/numerical simulation, and be keen to apply these to understanding the dynamics of the Earth (and other planets). Knowledge of geophysics and geology is a plus, not a prerequisite. Experience in AI and deep learning, in particular for solving regression problems in physics, and good knowledge of solid mechanics, in particular crystal plasticity, are major assets. Experience in using national and regional computing facilities (HCP) will also be appreciated.

Contexte de travail

This position is part of the ERC RhEoVOLUTION project, which aims to unravel how the evolution of rock rheology controls strain localization at different scales in the Earth. To do so, we will develop a framework for modeling self-consistently strain localization in rocks deforming by ductile processes. The experimentation plays a key role in the projet by providing essential physical constraints for the development of models. The ERC RhEoVOLUTION project team is composed of researchers in Earth Sciences, Glaciology, Materials Sciences and Applied Mathematics working in Montpellier, Grenoble, Nice and Argentina. The IR will be based at Géosciences Montpellier, a joint CNRS & University of Montpellier research unit (UMR 5243), attached to the Observatory of Sciences of the Universe - Mediterranean Research Observatory of the Environment (OSU OREME) - http://www.gm.univ-montp2.fr.

Contraintes et risques

N/A

Informations complémentaires

Further information : https://erc-rheovolution.gm.univ-montp2.fr/