arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 1704.00706astro-ph.IM
收藏

基于TDC1大样本光变曲线的碰撞测试差分平滑算法

Crash testing difference-smoothing algorithm on a large sample of simulated light curves from TDC1

S. Rathna Kumar

摘要:

在这项工作中,我们提出了改进的差分平滑算法测量的时间延迟从光变曲线的图像的引力透镜类星体。的改进主要包括一个更务实的方法来选择平滑的时间尺度自由参数,更现实的合成光变曲线的产生时间延迟不确定性的估计和使用的归一化$χ^2$的曲线图计算在很宽的范围内的试验时间延迟值,以评估测量的时间延迟的可靠性,也用于识别灾难性故障的情况。我们严格地测试了差异平滑算法的大样本上千对模拟光变曲线,它们之间具有已知的真实时间延迟,来自第一版强透镜时间延迟挑战(TDC 1)的两个最困难的“梯级”-梯级3和梯级4。并发现该算法测量的时间延迟的幅度高于时间延迟的真实值的固有趋势。然而,我们发现,这种系统的偏差是消除应用校正到每个测量的时间延迟的大小和符号的系统误差推断通过应用时间延迟估计器上的合成光曲线模拟测量的时间延迟。经过这些改进,TDC的差异平滑算法的性能指标被认为是有竞争力的最佳表现提交的TDC 1的两个测试的“横档”。在这项工作中使用的MATLAB代码和详细的结果可以在https://github.com/rathnakumars/difference-smoothing上公开获得

英文摘要:

In this work, we propose refinements to the difference-smoothing algorithm for measurement of time delay from the light curves of the images of a gravitationally lensed quasar. The refinements mainly consist of a more pragmatic approach to choose the smoothing time-scale free parameter, generation of more realistic synthetic light curves for estimation of time delay uncertainty and using a plot of normalized $χ^2$ computed over a wide range of trial time delay values to assess the reliability of a measured time delay and also for identifying instances of catastrophic failure. We rigorously tested the difference-smoothing algorithm on a large sample of more than thousand pairs of simulated light curves having known true time delays between them from the two most difficult `rungs' -- rung3 and rung4 -- of the first edition of Strong Lens Time Delay Challenge (TDC1) and found an inherent tendency of the algorithm to measure the magnitude of time delay to be higher than the true value of time delay. However, we find that this systematic bias is eliminated by applying a correction to each measured time delay according to the magnitude and sign of the systematic error inferred by applying the time delay estimator on synthetic light curves simulating the measured time delay. Following these refinements, the TDC performance metrics for the difference-smoothing algorithm are found to be competitive with those of the best performing submissions of TDC1 for both the tested `rungs'. The MATLAB codes used in this work and the detailed results are made publicly available at https://github.com/rathnakumars/difference-smoothing

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。

↑