A new ray-tracing scheme for 3D diffuse radiation transfer on highly parallel architectures
Center for Computational Sciences, University of Tsukuba, 1-1-1 Tennodai, Tsukuba Ibaraki 305-8577
arXiv:1410.0763 [astro-ph.IM], (3 Oct 2014)
@article{2014arXiv1410.0763T,
author={Tanaka}, S. and {Yoshikawa}, K. and {Okamoto}, T. and {Hasegawa}, K.},
title={"{A new ray-tracing scheme for 3D diffuse radiation transfer on highly parallel architectures}"},
journal={ArXiv e-prints},
archivePrefix={"arXiv"},
eprint={1410.0763},
primaryClass={"astro-ph.IM"},
keywords={Astrophysics – Instrumentation and Methods for Astrophysics, Astrophysics – Astrophysics of Galaxies},
year={2014},
month={oct},
adsurl={http://adsabs.harvard.edu/abs/2014arXiv1410.0763T},
adsnote={Provided by the SAO/NASA Astrophysics Data System}
}
We present a new numerical scheme to solve the transfer of diffuse radiation on three-dimensional mesh grids which is efficient on processors with highly parallel architecture such as recently popular GPUs and CPUs with multi- and many-core architectures. The scheme is based on the ray-tracing method and the computational cost is proportional to N^5/3_m where Nm is the number of mesh grids, and is devised to compute the radiation transfer along each light-ray completely in parallel with appropriate grouping of the light-rays. We find that the performance of our scheme scales well with the number of adopted CPU cores and GPUs, and also that our scheme is nicely parallelized on a multi-node system by adopting the multiple wave front scheme, and the performance scales well with the amount of the computational resources. As numerical tests to validate our scheme and to give a physical criterion for the angular resolution of our ray-tracing scheme, we perform several numerical simulations of the photo-ionization of neutral hydrogen gas by ionizing radiation sources without the "on-the-spot" approximation, in which the transfer of diffuse radiation by radiative recombination is incorporated in a self-consistent manner.
October 8, 2014 by hgpu
