arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2009.12609astro-ph.IM
收藏

线辐射转移的LOC程序

LOC program for line radiative transfer

M. Juvela

摘要:

辐射转移建模是许多天体物理模拟的一部分,用于进行综合观测和协助分析观测结果。我们集中在星际介质发射的无线电线路的建模。对于高分辨率模型,这可能是一个重大的计算挑战。我们的目标是一个线辐射转移(RT)程序,充分利用多核CPU和GPU。计算机化对于加快计算速度和使个人计算机能够处理大型建模任务至关重要。该程序基于光线跟踪,并使用标准的加速lambda迭代(ALI)方法来加快收敛速度。该程序适用于1D和3D网格。1D版本利用对称性来加速RT计算。3D版本使用八叉树网格,并针对低内存使用进行了优化,以支持大型模型的计算。测试表明,该方法与其他RT代码的结果在~ 2%范围内一致。这是典型的代码间差异,通常与模型设置的不同解释有关。运行时间与蒙特卡罗代码相比毫不逊色。在1D测试中,GPU上的GPU运行速度比单个CPU内核快20倍。尽管路径计算复杂,但使用八叉树离散化的3D模型也观察到高达10的加速。GPU能够计算具有数亿个单元的模型,如在星际云的大规模模拟中遇到的。该软件具有良好的性能和精度,能够在个人计算机上处理许多RT建模任务。它是用Python编写的,计算密集型部分作为编译的OpenCL内核实现,它也可以作为进一步实验替代RT实现的平台。

英文摘要:

Radiative transfer modelling is part of many astrophysical simulations and is used to make synthetic observations and to assist analysis of observations. We concentrate on the modelling of the radio lines emitted by the interstellar medium. In connection with high-resolution models, this can be significant computationally challenge. Our goal is a line radiative transfer (RT) program that makes good use of multi-core CPUs and GPUs. Parallelisation is essential to speed up computations and to enable the tackling of large modelling tasks with personal computers. The program LOC is based on ray-tracing and uses standard accelerated lambda iteration (ALI) methods for faster convergence. The program works on 1D and 3D grids. The 1D version makes use of symmetries to speed up the RT calculations. The 3D version works with octree grids and, to enable calculations with large models, is optimised for low memory usage. Tests show that LOC gives results that are in agreement with other RT codes to within ~2%. This is typical of code-to-code differences, which often are related to different interpretations of the model set-up. LOC run times compare favourably with those of Monte Carlo codes. In 1D tests, LOC runs were by up to a factor ~20 faster on a GPU than on a single CPU core. In spite of the complex path calculations, up to ~10 speed-up was observed also for 3D models using octree discretisation. GPUs enable calculations of models with hundreds of millions of cells, as encountered in the context of large-scale simulations of interstellar clouds. LOC shows good performance and accuracy and and is able to handle many RT modelling tasks on personal computers. Being written in Python, with the computing-intensive parts implemented as compiled OpenCL kernels, it can also a serve as a platform for further experimentation with alternative RT implementations.

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。

↑