COSMOS 場域的後星爆星系：機械學習辨識以及快速淬滅路徑的演化分析

Abstract

後星暴星係是最近在其恆星形成過程中經歷了顯著截斷的天體。研究這些過渡物體可以為星系演化提供有價值的見解。正在進行和即將進行的調查提供了覆蓋廣闊天空區域的大量光度數據。對這些數據應用光譜能量分佈（SED）擬合技術將識別出大量的星暴後星系，但計算時間較長。機器學習 (ML) 技術可以提供更快、更有效率的替代方案。

我使用了 COSMOS2020 目錄，它收集了 1.27 deg^2 宇宙演化巡天 (COSMOS) 場的測量結果。本研究追求兩個目標：（1）建構獨立於 SED 擬合的 ML 分類模型來辨識星暴後星系。 (2) 研究具有廣泛紅移和質量的後星暴星系，以研究透過星系質量函數的快速猝滅。

ML方法可以利用光學+近紅外線資訊很好地辨識後星暴星系。使用XGBoost分類器，模型可以在犧牲召回率（0.1）的同時實現高精度（0.85）。透過未來的光學/近紅外線巡天，該模型可以有效地識別大量後星暴星系。

本研究提出了質量為 log(M_*/M_solarmass}) > 8、0.1 < z < 3.0 的恆星形成、靜止和後星暴星系的質量函數。透過研究後星暴星系質量函數的轉變並將其與靜止星系的質量函數進行比較，可以限制它們的壽命。靜止星系質量函數的累積可以透過後星暴星系的轉變來解釋。在 log(M_*/M_solarmass}) > 10$ 的高品質下，星暴後星系的壽命估計為 60-300 Myr。然而，在低質量 log(M_*/M_solarmass}) < 10 時，在 z < 1 處進行了更長的星暴後壽命估計 > 1 Gyr。後星暴星系轉變為靜止狀態的回收情景可以解決這個問題，顯示只有不到 60% 的後星暴星系直接轉變為靜止狀態。然而，還需要進一步的測試來證實這項結果。

Post-starburst galaxies are objects that have recently experienced a significant truncation in their star formation. Studying these transitional objects can offer valuable insights into galaxy evolution. Ongoing and forthcoming surveys provide vast photometric data covering extensive sky areas. Applying spectral energy distribution (SED) fitting techniques to these data will identify a large number of post-starburst galaxies, but the computation time is lengthy. Machine learning (ML) techniques can provide faster and more efficient alternatives.

I used the COSMOS2020 catalog, which gathers measurements in the 1.27 deg^2 Cosmic Evolution Survey (COSMOS) field. This research pursues two objectives: (1) Construct a SED-fitting-independent ML classification model to identify post-starburst galaxies. (2) Studying post-starburst galaxies across a wide range of redshifts and masses to investigate rapid quenching through galaxy mass function.

The ML methods can identify post-starburst galaxies well with the optical+NIR information. With the XGBoost classifier, the model can achieve high precision (0.85) while sacrificing the recall rate (0.1). With future optical/NIR surveys, the model can identify a large number of post-starburst galaxies efficiently.

The mass functions of star-forming, quiescent, and post-starburst galaxies with masses log(M_*/M_solarmass}) > 8 at 0.1 < z < 3.0 are presented in this study. By studying the transition of the mass functions of post-starburst galaxies and comparing them with those of quiescent galaxies, their lifetimes can be constrained. The buildup of the mass functions of quiescent galaxies can be explained by the transition of post-starburst galaxies. At high masses with log(M_*/M_solarmass}) > 10, the lifetime of post-starburst galaxies is estimated to be 60-300 Myr. However, a longer post-starburst lifetime estimate > 1 Gyr is carried out at z < 1 at low masses log(M_*/M_solarmass}) < 10. A recycling scenario for post-starburst galaxies transitioning into quiescence can address this issue, indicating that less than 60% of post-starburst galaxies transition directly into quiescence. However, further testing is required to confirm this result.