DIES:采用即时重发射方法的并行尘埃辐射传输程序
DIES: Parallel dust radiative transfer program with the immediate re-emission method
AI总结:
本文提出了并行尘埃辐射传输程序DIES,采用即时重发射(IRE)方法实现快速一维球状云模型计算,支持GPU加速,运行时间约1秒,适用于大规模参数研究。
摘要:
辐射传输(RT)建模是解释星际尘埃热辐射观测的必要工具。它也经常是多重物理建模的一部分。在此背景下,辐射传输计算的效率非常重要,即使对于一维模型也是如此。我们研究了使用所谓的即时重发射(IRE)方法来快速计算一维球状云模型。我们希望确定与传统蒙特卡洛模拟中使用的类似加权方法是否能加快尘埃温度的估计。我们提出了程序DIES,这是IRE方法的并行实现,使得在图形处理器(GPUs)上进行计算成为可能。我们用外部和内部加热的云模型测试了该程序,并检查了使用不同加权采样方案的潜在改进。该程序的执行时间与以前的程序相比具有优势,尤其是在GPU上运行时。另一方面,加权方案只产生有限的改进。在内部辐射源的情况下,基本的IRE方法能很好地采样重发射,而传统的蒙特卡洛需要使用空间重要性采样。对于外部加热的模型,通过加权初始光子方向可以实现一些噪声降低。只有在光学非常薄的模型中,加权——例如所提出的强制首次相互作用方法——才能将噪声降低几倍。IRE方法对于内部和外部加热的模型都表现良好,通常不需要任何额外的加权方案。对于我们的测试模型,DIES程序的运行时间约为1秒,甚至适用于更大的参数研究。
英文摘要:
Radiative transfer (RT) modelling is a necessary tool in the interpretation of observations of the thermal emission of interstellar dust. It is also often part of multi-physics modelling. In this context, the efficiency of radiative transfer calculations is important, even for one-dimensional models. We investigate the use of the so-called immediate re-emission (IRE) method for fast calculation of one-dimensional spherical cloud models. We wish to determine whether weighting methods similar to those used in traditional Monte Carlo simulations can speed up the estimation of dust temperature. We present the program DIES, a parallel implementation of the IRE method, which makes it possible to do the calculations also on graphics processing units (GPUs). We tested the program with externally and internally heated cloud models, and examined the potential improvements from the use of different weighted sampling schemes. The execution times of the program compare favourably with previous programs, especially when run on GPUs. On the other hand, weighting schemes produce only limited improvements. In the case of an internal radiation source, the basic IRE method samples the re-emission well, while traditional Monte Carlo requires the use of spatial importance sampling. Some noise reduction could be achieved for externally heated models by weighting the initial photon directions. Only in optically very thin models does weighting - such as the proposed method of forced first interaction - result in noise reduction by a factor of several. The IRE method performs well for both internally and externally heated models, typically without the need for any additional weighting schemes. With run times of the order of one second for our test models, the DIES program is suitable even for larger parameter studies.
机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。