G PU A ccelerated IN tensities MPI (GAIN-MPI): A new method of computing Einstein-A coefficients

• Published in: Computer physics communications 2017-05, Vol.214, p.216-224
• Main Authors: Al-Refaie, Ahmed F., Yurchenko, Sergei N., Tennyson, Jonathan
• Format: Article
• Language: English
• Subjects: Intensities; Rotation-vibration spectra; Transition probabilities
• ISSN: 0010-4655; 1879-2944
• DOI: 10.1016/j.cpc.2017.01.013

Calculating dipole transition intensities or the related Einstein A coefficients can dominate the computer usage for large line lists of transitions such as those being computed to model radiative transport through hot atmospheres. An algorithm for the efficient computation of line strengths is presented based on the use of the half-linestrength. This is implemented on GPUs that are shown to give up to a thousandfold speed-up compared to calculations on conventional computers. This algorithm is implemented in the program GAIN which was developed as part of the TROVE nuclear motion program, but can be adapted for use by other similar programs in a straightforward fashion. Program summary Program title:GAIN-MPI Program Files doi:http://dx.doi.org/10.17632/4x75jsphc6.1 Licensing provisions: MIT licence. Programming language: C++ 99, CUDA C and Fortran 95. Nature of problem: Computation of linestrengths using GPU hardware Solution method : Split the linestrength into smaller blocks and compute them in the GPU in parallel Restrictions: The current version is restricted to separable rovibrational basis sets Unusual features: Can be extended by user supplied concrete C++ classes for the MPI version