CMN is a cross-cutting working group on numerical methods, simulation codes and supercomputers, whose mission is to harmonise the laboratory's numerical developments and computing practices.

Project 1. Development of a Lattice-Boltzmann code and GPU porting

The laboratory's unsteady simulation activities include a research line on “ Lattice Boltzmann ” methods (LBM). This work, led by Simon Marié, uses the research code PyLB developed by the laboratory since 2016. It focuses on 3 main aspects:

1. LBM modelling of miscible multi-species flows to study Saffman-Taylor instabilities, the subject of Lucien VIENNE's PhD thesis defended in 2019. This work validated the models for simulating three-species viscous fingering.
2. Numerical analysis of LBM and comparison of the scheme with other approaches. This involved several collaborations: with the M2N laboratory at Cnam, comparing with remeshed vortex methods, and with ONERA, through Alexandre SUSS's PhD thesis comparing LBM and Navier-Stokes solvers and coupling the two methods zonally. The latter showed the benefit of coupling both methods for unsteady wall-bounded simulations.
3. Unsteady flows through porous media at low Darcy numbers. This work, in collaboration with the M2N laboratory, studies wake transitions downstream of porous walls and the boundary layer near such walls.

All of the laboratory's LBM work has been supported by computing hours at national HPC centres through projects A5, A7, A9, A11 and A13. PyLB has also recently been ported to the GPU hardware the laboratory acquired in 2022.

Project 2. Development of high-performance tools

HPC is essential to CFD. CFD simulations rely on complex mathematical models of fluid behaviour, which require intensive computations to solve systems of nonlinear partial differential equations. HPC lets researchers carry out these computations quickly and efficiently, so they can test many scenarios in a short time. This matters most for designing and optimising systems such as aircraft wings or turbomachines, where every small change can significantly affect performance.

Numerical methods are equally essential to CFD. The equations governing fluid motion are usually too complex to solve analytically, so numerical methods are needed to approximate their solutions. Modern numerical methods are highly sophisticated and can solve complex CFD problems with great accuracy.

DynFluid follows technological advances and emerging programming paradigms in order to make efficient use of the resources available in supercomputing centres, so as to study highly complex flow phenomena that require considerable computing resources.

A common practice in this theme is to optimise the laboratory's codes for the compilers, computational methods and programming languages suited to the target architecture. This contributes to the development of high-performance numerical tools in the laboratory and to scientific publications in this field, such as:

• Carlos Junqueira-Junior, João Luiz F. Azevedo, Jairo Panetta, William R. Wolf, Sami Yamouni, On the scalability of CFD tool for supersonic jet flow configurations, Parallel Computing, Volume 93, 2020, 102620;
• Nicolas Alferez, Emile Touber, dNami: a framework for solving systems of balance laws using explicit numerical schemes on structured meshes, https://doi.org/10.5281/zenodo.6720082;
• Paola Cinnella, Xavier Gloerfelt, Francesco Grasso, Donatella Passiatore, Luca Sciacovelli, Dynamique d’une couche limite hypersonique de gaz dense (In English: Dynamics of hypersonic boundary layer of a dense gas), In: Grand Challenges Joliot-Curie, 2019, pg 18-26.
• Luca Sciacovelli, Paola Cinnella, Giuseppe Pascazio, Écoulements turbulents hypersoniques hors d’équilibre thermochimique, In: Rapport d'activité GENCI 2020, pg 31.