临界曲线附近的微引力透镜效应模拟

在光滑物质分布模型下,临界曲线是强引力透镜系统中像平面上一条放大率为无穷的线,而考虑少量离散质量的微透镜效应后,源平面上的放大率分布会出现复杂的结构,为暗物质成分的探测提供了一种有效途径.模拟临界曲线附近微透镜效应存在临界曲线上放大率无穷大和计算量巨大的困难.要达到所需的模拟精度,直接使用传统的光线追踪算法需要巨大的计算资源.为此发展了一个能实现海量计算的Graphics Processing Unit (GPU)并行方法来模拟临界曲线附近的微引力透镜效应.在型号为NVIDIA Tesla V100S PCIe 32 GB的GPU上,对于需要处理13000多个微透镜天体、发射1013量级光线的模拟,耗时在7000 s左右.在GPU并行的基础上,与直接的光线追踪算法相比,插值近似的引入使计算速度提升约两个数量级.利用该方法生成80个放大率分布图,并从中抽取800条光变曲线,进行了微焦散线数密度和峰值放大率的统计.

Microlensing Simulation near the Critical Curve

In the smooth mass distribution model, the critical curve is a line with magnification divergence on the image plane in the strong gravitational lens system. After considering the microlensing of a very small amount of discrete mass, the magnification map in the source plane has a complex structure, which provides an effective way to detect the dark matter. There are difficulties including the magnification divergence at critical curve and the huge amount of computation in microlensing simulation. To achieve the required simulation accuracy, using inverse ray-shooting directly will require huge computational resources. Therefore we developed a Graphics Processing Unit (GPU)-based parallel method to achieve the goal for massive computing. It takes about 7000 seconds to calculate $\sim13000$ microlenses for $10^{13