Molecular observation in astrophysics (radio telescopes and infrared telescopes) is the best source of physical and
chemical information on cold and / or diluted environments, 3K <T <100K. Molecules (like CO, H2O, NH3 or organic
molecules) are observed through their rotational or ro-vibrational spectra. The molecules are present in trace amounts
in a majority gas, H2 in general. These astrophysical spectra require a long chain of calculations in order to be understood
quantitatively and to determine the density of molecules observed, the temperatures and the turbulence of the gasses.
An important phenomenon is the excitation of molecular rotational and vibrational levels by collisions. If the kinetic
temperature of the gas is different from the internal temperature of the molecules, it means that the molecular collisions
To compute excitation rates (via the cross sections of ro-vibrational excitation), we make use of a series of codes : codes
of quantum chemistry,codes of fitting potential energy surfaces and the more computer intensive ones, the quantum
Current quantum dynamics codes are not fully optimized for massively parallel computers and / or computers with
CPUs and GPUs. Computions are heavy enough to deserve a close examination of the algorithmic and the coding. A
full calculation takes a few weeks or months, which are likely to be reduced by a factor of 10 or better with good coding.
The aim of the internship is to bring an existing code, reputed to be faultless, to a parallel architecture (OpenMP
and / or MPI) and opening the possibility of compiling with GPUs the associated linear algebra: full symmetric
matrices NXN, N> 5000, multiplications and linear algebra.
The student should be familiar with the OpenMP language and/or the GPU architecture, and be ready to invest in a
computer adventure. She / he will work on the computers of the MESOLUM / Fusion Mesocenters, and possibly on the
Jean-Zay supercomputer of the CNRS, for which calculation time is already available.
Equipe MFC: Daniel Comparat / LKB
Equipe MFC: Daniel Comparat, Yan Picard