AI 中文总结
利用主成分分析增强的Transformer TDE光谱分类器，在SDSS DR7目录中发现两个新的光学-紫外潮汐瓦解事件及一个可能事件，突出机器学习分类器挖掘能力，为发现此类事件提供新方法。
AI 中文摘要
光学光谱特征在当前光学-紫外潮汐瓦解事件（TDEs）的识别中起决定性作用。理论和观测方法估计TDE的发生率为$10^{-5}-10^{-4}$星系$^{-1}$年$^{-1}$，因此拥有超过10$^5$星系光谱的大型光学光谱目录可能包含具有TDE光谱特征的偶然光谱，这需要建立有效的选择方法才能发现。本文引入了一种主成分分析增强的Transformer TDE光谱分类器，在评估数据集上实现了0.88的精度和0.99的召回率，并报告了在广泛使用的SDSS DR7目录中的令人振奋的发现：两个新发现的TDE和一个报告的可能的TDE。对于SDSS J124225.39+642919.0，确认了拍摄光谱时GALEX目录中存在紫外瞬变，其发生时间早于光谱观测时间，MJD < 52316（2002年2月11日），使其成为目前发现的最早的光学-紫外TDE。对于SDSS J152459.70+045423.1，其光谱与TDE-H+He光谱的所有特征匹配，并且是在卡特琳娜实时瞬变调查记录的光学爆发期间拍摄的。这次爆发开始于54269 < MJD < 54476（2007年6月18日 - 2008年1月11日），使其成为报告的光学TDE中最早的之一。两个新TDE的发现突出了基于机器学习的分类器在挖掘大量目录中隐藏宝藏方面的能力，并标志着发现光学-紫外TDE的一种新方法。
英文摘要
Optical spectroscopic features are decisive in the current identification of optical-UV tidal disruption events (TDEs). Regarding that the TDE estimated occurrence rate is $10^{-5}-10^{-4}$ galaxy$^{-1}$ yr$^{-1}$ by both theoretical and observational methods, large optical spectroscopic catalogs with >10$^5$ galaxy spectra can include some serendipitous spectra with TDE spectroscopic features, which can be found after building a useful selection method. We hereby introduce a principal component analysis enhanced Transformer TDE spectrum classifier which achieves a precision of 0.88 and a recall of 0.99 on our evaluation dataset, and report its inspiring discoveries in the widely-used SDSS DR7 catalog: two newly discovered TDEs and one reported likely TDE. For SDSS J124225.39+642919.0, we confirm the presence of a UV transient in GALEX catalog when the spectrum was taken, and its occurrence time should be earlier than the spectrum observation time, MJD < 52316 (February 11, 2002), making it the earliest optical-UV TDE discovered by now. For SDSS J152459.70+045423.1, its spectrum matches all features of the TDE-H+He spectrum, and was taken during an optical outburst recorded by the Catalina Real-time Transient Survey. The start of this outburst lies in 54269 < MJD < 54476 (June 18, 2007 - January 11, 2008), making it one of the earliest among the reported optical TDEs. The discovery of two new TDEs highlights the power of machine-learning based classifiers in digging out buried treasures in large-volume catalogs, and marks a new method for discovering optical-UV TDEs.
CommentsAccepted by ApJ
