NFD: Toward Real-time Mining of Short-timescale Gravitational Microlensing Events. (14th September 2018)

• Record Type: Journal Article
• Authors: Qiu, Jiaming; Sun, Yankui; Wu, Chao; Du, Zhihui; Wei, Jianyan
• Is Part Of: Publications of the Astronomical Society of the Pacific. Volume 130:Number 992(2018)
• Publication Date: 2018-09-14
• Subjects: gravitational lensing: micro; Astronomy -- Periodicals; Astronomy; Periodicals; Periodicals; 520.5
• DOI: 10.1088/1538-3873/aada6f
• Languages: English
• ISSNs: 0004-6280

Abstract

Abstract: To search short-timescale microlensing (ML) events ( T E < 1 day ) from high-cadence, wide-field survey in real time, we present an algorithm called NFD (normalized feature deviation) to monitor all the observed light curves and to alert abnormal deviation in the data stream of light curves. The NFD algorithm framework consists of three main modules: feature stream extraction, statistical analysis of the feature stream, and discrimination. This algorithm is designed to be effective for ML event search from variable baselines, as well as constant baselines. To thoroughly evaluate the performance of our algorithm, a large-scale simulative light curve data set of variable stars based on the physical model of gravitational microlensing effect was built, yielding 139, 968 valid sample light curves in total; then, a more challenging simulative data set for evaluation under GWAC-like environment was generated, including 12, 960 constant sample light curves and 3240 variable sample light curves. Sample precision rate (SPR) and average detecting position (ADP) as two metrics are proposed, where SPR measures the accuracy of our algorithm, and ADP reflects alarm sensitivity. Golang implementations are done to evaluate the computational efficiency of our algorithm. Results of simulative experiments show that our algorithm is fast (176, 000 stars in 3 seconds), accurate (SPR is 89% on constant baselines, and 73% on variable baselines in the second data set), and sensitive (in average, it can detect all kinds of ML event before the peak on both constant and variable baselines), making it fairly competent for the real-time mining mission of short-timescale gravitational microlensing events.