Issue 33 of Science@ifpen - Computational Code Performance

Special issue: Computational Performance 

Numerical simulations and scientific calculations are widely used in the energy field: to manage oil and gas reserves, exploit geothermal energy and design wind farms, for example. However, these physical models are increasingly complex, involving hydraulic, mechanical, chemical and thermal couplings. Numerical methods must also be able to handle phase changes as well as heterogeneous and multiscale media, that may contain evolving discontinuities. Accordingly, the algorithms to be developed require a large amount of operations.

Thanks to parallel algorithms, combined with specific programming models, data and operations are distributed between several computing resources — processors, cores and graphics cards — making the effective communication between these entities a real challenge. To improve code performance, IFPEN has developed algorithms based on domain decomposition techniques, particularly for particle simulations and sparse linear solvers. IFPEN is also developing adaptive methods, based on multiscale approaches,
as well as statistical treatments and a posteriori error estimates.

I hope you enjoy reading this issue.

Jocelyne Erhel, Member of IFPEN's Scientific Board
 

  Download the PDF of the letter