PHY: Accelerated Always-On Fully-Coherent Network Analysis for Gravitational Wave Searches

2015年，LIGO探测器直接探测到引力波（GWs）标志着天文学新时代的到来。现在，科学家们可以用一种全新的方式来观察宇宙，尤其是像黑洞合并这样的极端现象，这将为天体物理学和基础物理学带来革命性的见解。探测引力波是由先进的仪器和用于分析引力波产生的大量数据的计算机算法实现的。一种数据分析方法能够在多大程度上挖掘高噪声数据，从而可靠地提取天体物理信号——它的灵敏度——突显了用GW探测器可以完成的科学研究。目前最具挑战性的数据分析问题之一是实现全球GW探测器网络数据分析的最优方法，即全相干全天（FCAS）搜索。虽然统计理论表明这种方法具有最高的灵敏度，但其高昂的计算成本限制了它目前的零星使用模式，而不是始终在线的使用模式。在该项目下，将实现快速FCAS搜索，以克服这一计算障碍。这将使来自探测器网络的数据能够以最高的灵敏度进行整体分析。在德克萨斯大学格兰德谷分校（UTRGV）这个少数族裔服务机构的前沿科学领域实施这个项目，将对吸引来自弱势群体的学生进入STEM领域产生深远而广泛的影响。通过这个项目，学生将获得先进的数据分析和计算技能，这些技能可以在各种各样的职业中转移，并与国家需求相关。加速FCAS搜索有两个主要因素：一种受自然启发的优化算法，称为粒子群优化（PSO），以及使用图形处理单元（gpu）的大规模并行实现。对于后者，PI将使用由NSF-MRI拨款部分支持的GPU集群。PSO是模仿鸟群寻找最佳食物来源的行为。在这里，它被用来寻找最适合的GW信号的数据。PSO本身可以将FCAS搜索的计算成本降低10倍，而GPU加速提供了额外的类似加速因素。该项目的一个关键方面将是开发一种新的否决机制，用于拒绝主导误报背景率的仪器非gw信号（“小故障”）。这些否决是通过相干网络数据分析和PSO对信号参数空间的非物理扇区的探索实现的。此外，新的数据调节方法将在搜索中使用和测试，例如基于自适应样条拟合的故障估计和减法算法。结合本质上更敏感的方法和更好的否决，可能会在开放的GW数据中发现新的信号，并确认边缘事件。这将创建一个更完整的CBC源群样本，这将有助于天体物理学家更好地理解正在发现的不寻常双星系统的形成渠道。使用加速码可以有效地探索信号参数空间的新扇区，例如具有次太阳质量成分的双星，并且在没有检测的情况下将设置更严格的速率约束。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

The direct detection of Gravitational Waves (GWs) by the LIGO detectors in 2015 marks the dawn of a new era in astronomy. Now, a completely new way of observing the Universe, especially extreme phenomena such as the merger of black holes, is available to scientists that will bring revolutionary insights in astrophysics and fundamental Physics. The detection of gravitational waves is enabled by advances in instrumentation as well as the computer algorithms used to analyze the copious amounts of data produced by them. The degree to which a data analysis method can dig into the highly noisy data to reliably extract an astrophysical signal -- its sensitivity -- underlines the science that can be done with GW detectors. One of the most challenging data analysis problem at present is the implementation of the optimal method, namely, fully-coherent all-sky (FCAS) search, for analyzing data from the worldwide network of GW detectors. While statistical theory suggests that this method has the highest sensitivity, its high computational cost limits it at present to a sporadic rather than always-on mode of use. Under this project, a fast FCAS search will be implemented that will overcome this computational barrier. This will allow data from a detector network to be analyzed in its entirety with the highest achievable sensitivity. Implementing this project in a frontier area of science at the University of Texas Rio Grande Valley (UTRGV), a minority serving institution, will have far-reaching broader impacts in terms of attracting students from under-represented groups to STEM. Through this project, students will acquire advanced data analysis and computing skills that are transferable across a wide variety of careers and relevant to national needs.There are two main ingredients involved in the acceleration of the FCAS search: a nature-inspired optimization algorithm called Particle Swarm Optimization (PSO), and a massively parallel implementation using Graphics Processing Units (GPUs). For the latter, the PI will use a GPU cluster under acquisition that is partially supported by an NSF-MRI grant. PSO is modeled on the behavior of a bird flock trying to find the best source of food. Here, it is being used to find the best-fit GW signal to the data. PSO by itself leads to a 10-fold reduction in the computational cost of an FCAS search while GPU acceleration provides an additional similar factor in speedup. A key aspect of the project will be the development of novel vetoes for rejecting instrumental non-GW signals ("glitches") that dominate the background rate of false alarms. These vetoes are enabled by coherent network data analysis and the exploration of unphysical sectors of signal parameter space by PSO. In addition, novel data conditioning approaches will be used and tested in the search, such as an adaptive spline fit based glitch estimation and subtraction algorithm. The combination of an intrinsically more sensitive method and better vetoes could potentially uncover new signals in open GW data and confirm marginal events. This will create a more complete sample of the CBC source population that will help astrophysicists achieve a better understanding of the formation channels for the unusual binary systems that are being discovered. New sectors of signal parameter space, such as binaries with sub-solar mass components, can be efficiently explored with the accelerated code and tighter rate constraints will be set in the absence of detections.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.