Open positions

PhD position : Numerical methods for optimal transportation
Contact: E. Maitre, Emmanuel.Maitre@imag.fr

The UJF-MSTIC project MENTOL (Numerical Methods for OptimaL Transportation) aims at developping innovative methods and algorithms to handle generalized energies in optimal transportation, with application to image assimilation and registration. A first task is to build a benchmark set of existing numerical methods to solve classical optimal transportation algorithms. The next task will be devoted to design efficient algorithms for generalized energies.
The position requires a doctoral degree in Applied Mathematics, or in Computer Science with clear numerical aspects. Also required is a comprehensive expertise in a scientific programming language (C, C++, Fortran 90+) and in one or more of the following: numerical optimization, computational geometry, image analysis, wavelets, optimal transportation, numerical algorithms for partial differential equations.
The appointments will be at Joseph Fourier University, in Grenoble, France.

PhD position : Model coupling and hybrid computing for multi-scale CFD
Contact: G.-H. Cottet, Georges-Henri.Cottet@imag.fr

The coexistence of multi-scale phenomena are a prominent feature in fluid dynamics. The idea of model coupling is to combine different models for different scales for which they are best fitted.
The goal of hybrid computing is to further adapt the numerical strategy to take advantage of the best possible hardware for the different numerical models.
This thesis will focus on several prototypal problems in CFD where small scale turbulent transport is important. Its goal will be to combine spectral or finite volume on the one hand and particle method on the other hand, and to implement this coupling on hybrid CPU/GPU computers.
The developed software will be, as much as possible, non intrusive for the existing algorithms.
Physical insight into turbulent transport, numerical validations against DNS or LES, and computational efficiency analysis will be sought in this work.
This work will be done in the context of the ANR project HAMM.

References:
D. Rossinelli, M. Bergdorf, G.-H. Cottet and P. Koumoutsakos, GPU accelerated simulations of bluff body flows using vortex methods, J. Comput. Phys., 229 (9), 3316–3333, 2010. pdf file
G.-H. Cottet, G. Balarac and M. Coquerelle, Sub-grid particle resolution for the turbulent transport of a passive scalar, 12th Euromech Conference, Marburg, September 7-10, 2009. pdf file


Post-doc position : Turbulent transport and model coupling.
Contact: G.-H. Cottet, Georges-Henri.Cottet@imag.fr

The goal of this project is to develop hybrid codes adapted to various flow configurations where scalar transport is important, allowing to couple grid-based (spectral or finite-volume) and particle methods, and to use these codes for various Reynolds and Schmidt numbers.
The idea is to understand to which point a DNS of the scalar transport can be used, in combination with a LES of the Navier-Stokes equation, to investigate the scalar behavior in the scales that are dynamically meaningful.
Thanks to the parallel library PPM developed at ETH Zurich and the stability of particle methods for large time steps, particle methods will be used at very high resolution.
The project will first investigate, in the case of periodic homogeneous turbulence, the ability of the coupling strategy to predict the presence of a k-1 power law, the viscous-convective range, after the Corrsin-Oboukhov cascade, for high Schmidt numbers.
The second part of the project will be to couple particle methods with finite-volume methods in more complex geometries. In ths case the main objective will  be to simulate heat transfer for various cooling fluids used innuclear plants.
This work will be done as part of the ANR project SCALES.

References:
G.-H. Cottet, G. Balarac and M. Coquerelle, Sub-grid particle resolution for the turbulent transport of a passive scalar, 12th Euromech Conference, Marburg, September 7-10, 2009. pdf file
I.F. Sbalzarini, J.H. Walther, M. Bergdorf, S.E. Hieber, E.M. Kotsalis and P. Koumoutsakos, PPM - A highly
efficient parallel particle-mesh library for the simulation of continuum systems, J. Comput. Phys. 215
(566-588), 2006.


Post doc position: Hybrid computing of Richtmyer-Meshkov instabilities
Contact: J. Griffond, jerome.griffond@cea.fr

Location: CEA Paris / Bruyeres-le-Chatel (30 km south of Paris);
Duration: 12 or 18 months, supported by the french A.N.R. (starting summer 2011)
Citizenship: European Union
Important : Time required for administrative treatment : 6 months.

High performance computing is a continuously progressing field due to advances in numerical and algorithmic methods but even more in computer science. A recent hardware improvement seems promising : it consists in associating GPU units to classical CPU cores. Efficient use of this technology is the goal of the collaborative HAMM project (Hybrid Architecture and Multiscale Methods) centered around multiscale physical applications and involving CEA and the Laboratoire Jean Kuntzmann in Grenoble; the macro scales are dealt on the CPU while the micro scales are on the accelerating GPU cores. Among applications aimed in that framework, our interest focuses on the socalled  Richtmyer- Meshkov instability and the resulting turbulent mixing. Indeed, instabilities and mixing induced by shock waves interacting with interfaces between different materials are important phenomena encountered in many fields (astrophysics, inertial confinement fusion simulation,...). These phenomena can be described by the multimaterial Euler equations or even Navier-Stokes equations depending on the observation scale. Several numerical methods can be used to solve it among which finite volume methods have a central place. They are robust and stable and allow treatment of strong shocks but the price to pay is a large numerical dissipation and diffusion. On the opposite, particle methods [1] have a much less artificial diffusion but they remain difficult to use in highly compressible flows. The benchmark for the post-doctoral work is the mixing occurring after Richmtyer-Meshkov instabilities due to a weak shock wave. This configuration can be treated with the compressible in- house Triclade code [3] based on finite volume methods but also with the quasi-compressible PPM library [2] based on particle methods. Both tools are dedicated to 3D parallel simulations on supercomputers with distributed memory and MPI communications.
We routinely use the Triclade code for such applications but we hope that using PPM may also be of interest for that case. The main goal of the proposed post-doctoral position is to take advantage from hybrid architecture. A first step consists in applying a particle method code (on a classical architecture) to the benchmark configuration to check for its capabilities. Then, the code has to be carried on hybrid architecture by taking the structure of the particle method into account ; optimizations accounting for the latter architecture have also to be made in order to get the most efficient runs of the benchmark. The requested work requires good knowledge in the intensive parallel computing field and interest for programmation on GPU [4]. The computers to be used are those of the CCRT, one of the top ten computing centers in the world.

References:
[1] G.H Cottet and P. Koumoutsakos, Vortex methods, Cambridge Univ. Press, 2000
[2] I.F. Sbalzarini, J.H. Walther, M. Bergdorf, S.E. Hieber, E.M. Kotsalis and P. Koumoutsakos, J. Comput. Phys. 215 (566-588), 2006
[3] M. Boulet, J. Griffond Proc. of Int. Workshop Physics of Compressible Turbulent Mixing, Paris, 17- 21/07/2006.
[4] D. Rossinelli, M. Bergdorf, G.-H. Cottet and P. Koumoutsakos, GPU accelerated simulations of bluff body flows using vortex methods, J. Comput. Phys., 229 (9), 3316–3333, 2010.

Master Recherche en Mathématiques Appliquées : sujets 2009-2010

Simulation des performances du pneumatique sur sol meuble

Sensibilité et fiabilité pour la simulation des performances du pneumatique

Modélisation par éléments discrets du transport solide par charriage

Modélisation numérique de l’effet de la rugosité du sol sur la direction du ruissellement

Modélisation des propriétés optiques de cristaux photoniques graduels

Schémas d'intégration numérique d'ordre élevé pour la dynamique non regulière

Couplages de modèles hydrologiques et océanographiques

Méthodes particulaires pour la simulation multi-échelles d’écoulements turbulents

Autour de la modélisation level-set du couplage d'un solide élastique immergé avec le fluide qui l'entoure

Méthodes mathématiques pour les écoulements sur des surfaces

Résolution par éléments finis des équations des mousses liquides

Simulation numérique des coulées de lave de volcans

Un modèle de densité pour l’interaction laser–boîtes quantiques

Dynamique des chaines de billes

Méthode Élement Fini/Base Réduite Multi-Échelles -  Design de Nouvelles Fonctions de Base

Sujets de Thèse

NB :Nous n'acceptons pas de candidat n'ayant aucun financement.