Scientific context: Aluminum alloys represent, after steels, extremely attractive metallic materials for many applications. Due to the light weight, high ductility, and high strength they enter a broad range of use in strong, lightweight structures with high engineering added value for automotive, aeronautic and aerospace industry. Challenges still faces researchers from a fundamental point of view in the understanding of the structure and dynamic properties and their interplay at the atomic scale for a better control and efficient industrial casting methods. There are now some evidences that structural features not only at the local atomic scale but also in the medium range might govern the early stage of crystal nucleation. This is particularly important for rapid (additive) solidification where it is desirable to lower the dynamics, i.e. diffusion phenomena, in the liquid while taking advantage of equiaxed nucleation.

Although investigation of the local order is routinely performed even with ab initio simulation techniques, this is not the case for the intermediate range-order. The latter requires in addition large-scale molecular dynamics simulations with empirical potentials with the goal of investigating systematically the medium range-order and treating more complex alloys.

Thesis work: The purpose of this Thesis enter the research scope dedicated to the construction of empirical potentials with a systematic and automated search of the optimal set of parameters. This will be carried out by means of machine learning techniques inspired from artificial intelligence concepts, using large scale ab initio databases tuned on experimental data of the DLR. More precisely, variable selection on structural descriptors of the Behler-Parrinello form of the database will be used to train and test an Artificial Neural Network to fit empirical potentials with ab initio accuracy.

This work will be done under the auspices of the Chair of Materials in the Multidisciplinary Institute in Artificial intelligence (MIAI) through SIMaP and LIG laboratories belonging to Université Grenoble-Alpes as well as  the Institute of Materials Physics in Space of the German Aerospace Centre (DLR) in Cologne, Germany.

Candidate profile: We look for highly motivated candidates with a Master degree in Physics (or equivalent) and prior experience with computer simulations and a strong interest in computer science and machine learning. Students with some experience in machine learning and/or ab initio simulations are encouraged to apply. A good knowledge of written and spoken English is essential to communicate within this joint French-German project. The candidate should have some skills in programming languages (Fortran, C/C++, Python) and Linux. Basic knowledge of parallel computing will be appreciated. 

How to apply: The deadline for sending your application is August, 31th 2021 and interviews will be conducted before the end of september. Please send your application as soon as possible by email to: 

Noel Jakse ([email protected])

Emilie Devijver ([email protected])

Applications include: a concise but informative cover letter, CV, Master 1 and Master 2 (or equivalent) marks, names and contact of at least two references that can be joined for recommendation letters.