Section: New Results

Other works


Participants: Laura Mendoza Virtually all magnetic fusion devices resort to tomography diagnostics for a variety of plasma emissions. All those diagnosis have a lot in common: the plasma is transparent to the observed quantity, such that the signal on a detector is derived from a spatial integration of the local emission. Solving the direct problem (i.e. from simulated emissivity to signals) requires modeling the diagnostic geometry and is used for physics code validation or diagnostic design. Solving the inverse problem (i.e. from experimental signals to reconstruct 2D emissivity) is useful for data interpretation and requires not only geometry modeling but also decomposing the unknown emissivity into basis functions and inversion-regularization routines. In this context, a python library, ToFu, solves the direct and inverse problems for synthetic diagnostics. The project objective for the second part of 2018 is to develop and optimize the existing geometry module in ToFu, with a special focus on the ray-tracing algorithms.

Discontinuous Galerkin solver

Participants: Philippe Helluy, Bruno Weber We have implemented and validated new optimizations in our Discontinuous Galerkin (DG) codes CLAC and SCHNAPS. In CLAC, Bruno Weber, our CIFRE PhD in the AxesSim company, has implemented a local time-step method and optimizations in order to run efficiently the OpenCL kernels both on CPU and GPU. This allows to run a huge electromagnetic simulation of a Bluetooth antenna in interaction with a full volumic human model. The simulation was run on the supercomputer Piz Daint (3rd at the "top 500" ranking in 2017). The computing hours were awarded through a PRACE call dedicated to small companies. In SCHNAPS we were able to assess the efficiency of the StarPU runtime for distributing the computational tasks efficiently on hybrid computers.

Finite volume methods for complex hyperbolic system

Participants: Philippe Helluy, Lucie Quibel In the thesis of Lucie Quibel (started in November 2017), we study numerical methods for solving compressible fluids with complex equation of states. The objective is to simulate liquid-vapor flows that occur in nuclear plants. The pressure behavior of the liquid-vapor mixture is very complex and obtained through measurements and tabulated laws. This sometimes prevent the system from being hyperbolic and leads to instabilities. We are trying to construct simpler but realistic laws that preserve the convexity structure and the scheme robustness.