Expertise — Science by Computing

Building with scientific communities

The Science by Computing (SbC) team helps scientific communities develop simulation codes and analysis tools. Working alongside researchers, it adapts applications to supercomputer architectures, particularly exascale systems, and shares this expertise across disciplines.

Team leader: Charles Prouveur.

Life science

Contacts: Yves Vandenbrouck and Jean-Michel Batto.

Maison de la Simulation contributes to the computing infrastructure for large-scale genomic data analysis being developed through the CEA–CAD collaboration. Planned for deployment at the TGCC, it will support genomic research and data reanalysis for healthcare.

The project combines data-workflow orchestration, analysis software and HPC resources, with secure computing environments, controlled data access and a preference for open-source components.

Astrophysics — ARK2-MHD

Contact: Pascal Tremblin.

ARK2-MHD is an open-source magnetohydrodynamics code for large-scale simulation, including turbulent flows, magnetic fields and convective dynamos in stellar atmospheres.

Its second-order MUSCL–Hancock finite volume scheme combines Kokkos for portability across OpenMP, CUDA and HIP with MPI for distributed execution. I/O and in situ analysis tools support scaling to thousands of GPUs.

The work presented covers resolutions up to 4096³ and tests on AMD MI300, NVIDIA H100 and Intel Max 1550 GPUs.

Fusion plasmas — GYSELA-X++

Contact: Thomas Padioleau.

This work develops reusable gyrokinetic operators in modern C++, based on Kokkos, DDC and PDI. Separating physical and computational concerns makes the code easier to maintain and extend.

Support includes Python integration for in situ analysis, packaging from laptops to supercomputers, performance analysis and optimisation.

Physical chemistry — deMon2k and MDFT

deMon2k · Contact: Karim Hasnaoui.

With the TheoSim group at ICP Orsay, Maison de la Simulation supports this density functional theory (DFT) package through debugging, optimisation, parallelisation and GPU porting. The work uses MPI, OpenMP and ScaLAPACK for scalability, and MPI, CUDA/HIP and MAGMA for multi-GPU execution. It also includes an OpenACC implementation of the Kohn–Sham matrix.

MDFT · Contacts: Riccardo Spezia and Jean-Marc Lude.

MDFT uses molecular density functional theory to calculate solvation properties of complex solutes while retaining information on solvent structure, at a lower cost than QM/MM approaches.

Physical developments address different solvents, thermodynamic properties and coupling with an electronic structure code to optimise molecules in their solvent. Computational developments rewrite the spherical harmonics and FFT kernel with Kokkos FFT for heterogeneous computing.

Light–matter interaction — Smilei

Contact: Charles Prouveur.

Smilei is an open-source, collaborative Particle-In-Cell code written in C++. It supports studies ranging from relativistic laser–plasma interactions to astrophysics, with physical modules and geometries evolving alongside community needs.

Engineers at Maison de la Simulation contribute to the core code, GPU porting of its modules and their use on recent supercomputers.

Energy — AMITEX and canoP

AMITEX_FFTP · Contact: Yushan Wang.

This distributed Fortran solver uses FFTs to simulate nonlinear mechanical behaviour in heterogeneous materials represented by regular 3D images. It runs on workstations, clusters and large HPC infrastructures.

The collaboration addresses GPU porting of constitutive laws and of the solver, particularly 3D FFTs with pencil domain decomposition.

canoP · Contact: Pascal Tremblin.

This open-source computational fluid dynamics library uses p4est to adapt the mesh to the solution. Refinement and coarsening concentrate resources where gradients are steep.

Models cover single-phase and two-phase flows: monofluid, the five-equation bifluid5eq model and the seven-equation Baer–Nunziato bifluid7eq model. They address different flow regimes for energy applications.