arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 1006.4637astro-ph.CO
收藏

模拟星系的光谱能量分布:引入人工神经网络

Modelling the spectral energy distribution of galaxies: introducing the artificial neural network

L. Silva, A. Schurer, G. L. Granato, C. Almeida, C. M. Baugh, C. S. Frenk, C. G. Lacey, L. Paoletti, A. Petrella, D. Selvestrel

摘要:

星系的光谱能量分布是恒星形成历史和星系中恒星和气体几何排列的复杂函数。恒星辐射通过尘埃分布的辐射转移的计算非常耗时。这方面变得不可接受,特别是当处理的半分析星系形成模型填充宇宙学卷的预测,然后与多波长的调查。主要是为了这个目的,我们已经实现了一个人工神经网络算法到分光光度和辐射转移代码GRASIL,以计算光谱能量分布的星系在很短的计算时间。这允许避免采用可能与模型输出的模拟星系无关的经验模板。人工神经网络已经实现了计算尘埃发射光谱(计算的瓶颈),并分别为恒星形成的分子云和弥漫的尘埃(由于其不同的属性和依赖关系)。我们已经定义了输入神经元,有效地确定它们的发射,这意味着这种实现具有普遍适用性,并且与特定的星系形成模型无关。我们已经训练了网络的光盘和球形几何形状,并测试其性能,以再现盘和星暴星系的SED,以及球状星系的半分析模型。我们已经检查了,对于这个模型的SED和星系计数的Herschel带与ANN近似得到的几乎是叠加到相同的数量与完整的GRASIL。我们的结论是,这种方法似乎是强大的和有利的,并将在另一篇论文中提出的应用程序,一个更复杂的SAM。

英文摘要:

The spectral energy distribution of galaxies is a complex function of the star formation history and geometrical arrangement of stars and gas in galaxies. The computation of the radiative transfer of stellar radiation through the dust distribution is time-consuming. This aspect becomes unacceptable in particular when dealing with the predictions by semi-analytical galaxy formation models populating cosmological volumes, to be then compared with multi-wavelength surveys. Mainly for this aim, we have implemented an artificial neural network algorithm into the spectro-photometric and radiative transfer code GRASIL in order to compute the spectral energy distribution of galaxies in a short computing time. This allows to avoid the adoption of empirical templates that may have nothing to do with the mock galaxies output by models. The ANN has been implemented to compute the dust emission spectrum (the bottleneck of the computation), and separately for the star-forming molecular clouds and the diffuse dust (due to their different properties and dependencies). We have defined the input neurons effectively determining their emission, which means this implementation has a general applicability and is not linked to a particular galaxy formation model. We have trained the net for the disc and spherical geometries, and tested its performance to reproduce the SED of disc and starburst galaxies, as well as for a semi-analytical model for spheroidal galaxies. We have checked that for this model both the SEDs and the galaxy counts in the Herschel bands obtained with the ANN approximation are almost superimposed to the same quantities obtained with the full GRASIL. We conclude that this method appears robust and advantageous, and will present the application to a more complex SAM in another paper.

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。

↑