使用GPU加速银道面尘埃辐射图像的高分辨率模拟与多参数反演

In order to use the large-scale high-resolution data of the Milky Way surveys performed recently both in China and abroad, and to properly understand the emission properties of dust grains in large areas of the Galactic plane, we plan to apply the GPU parallel computing technologies to optimize the existing three-dimensional radiative transfer model (3D RT model). The 3D RT model will be optimized for using the parallel computing technologies and processing the large-scale high-resolution data. The calculation of the 3D RT model will be accelerated significantly, thus we can efficiently perform the simulation of high resolution images of dust far-infrared emission in large areas of the Galactic plane, and efficiently invert free parameters of the 3D RT model. Our study will improve and innovate technologies for processing and mining massive data, and solve not only the contradiction between the large-scale data and the local region simulation of numerical models, but also the contradiction between the high-resolution data and the low-efficiency calculation of numerical models. Our study will effectively increase the number of inverted parameters to ensure the analysis of statistic properties of dust 3D distribution in both small and large scales, and build the foundation of efficiently and sufficiently using the high-quality survey data from observational facilities like LAMOST and FAST.

为利用国内外最新的大尺度高分辨率银河系巡天数据，准确把握银道面大天区尘埃辐射特性，本项目拟运用GPU并行运算技术，优化现有三维辐射转移模型，形成既适合并行运算、又切合巡天数据特征的模型代码和算法。从而显著提高模型运算速度，高效实现对银道面大天区尘埃远红外辐射图像的高分辨率模拟和多参数反演。通过本项目研究，将改进并创新处理海量巡天数据的技术手段，解决当前巡天观测的高空间覆盖率与数值模型的局部区域模拟之间的矛盾、观测数据的高空间分辨率与数值模型运算低效率之间的矛盾，低成本低耗时地实现大天区图像的高分辨率模拟，有效增加可反演参数数目。同时，为进一步分析不同尺度下尘埃密度三维分布的统计特性、充分挖掘运用LAMOST和FAST等观测设备提供的高质量巡天数据奠定基础。

为利用国内外最新的大尺度高分辨率银河系巡天数据，准确把握银道面大天区尘埃辐射特性，本项目运用 GPU 并行运算技术，优化了现有三维辐射转移模型的模拟反演过程，加速比达 1.0E3 量级，显著提高了模型拟合反演速度，从而实现了大尺度空间尘埃辐射的高分辨率模拟和反演。对 Hi-GAL 项目 4 平方度天区的尘埃远红外辐射图像进行了模拟和反演，其中三维网格分辨率达 10 pc × 10 pc × 10 pc。模拟拟合获得了 12 组模型参数的反演结果，通过分析反演结果获得了银道面尘埃的分布情况和气体尘埃比等物理参数。通过本项目研究，改进并创新了处理海量巡天数据的技术手段，解决了当前巡天观测的高空间覆盖率与数值模型的局部区域模拟之间的矛盾、观测数据的高空间分辨率与数值模型运算低效率之间的矛盾，低成本低耗时地实现了对大天区图像的高分辨率模拟的反演，为进一步分析不同尺度下尘埃密度三维分布的统计特性、充分挖掘运用 LAMOST 和 FAST 等观测设备提供的高质量巡天数据奠定了基础。