Abstract：White dwarf main sequence binary star (WDMS) is a kind of binary star system, the main star is a white dwarf star, and the companion star is a small mass main-sequence star.WDMS has an important meaning to study the evolution of close binary stars, especially the evolution of the common envelope. Many physical parameters, such as effective temperature, metal abundance, and surface gravity acceleration can be obtained by studying the spectra of WDMS. The accurate measurement of these physical parameters can not only solve the classification problem of WDMS but also provide basic data for the study of binary star theoretical model.The spectra of WDMS is synthetic spectra, which consists of the spectra of the main star and companion star. WDMS spectra have two major limitations, one is noise interference, and the second is that the blue and red parts are suppressed by the spectral characteristics of the host and companion stars respectively.It is very meaningful to obtain the spectra of white dwarf and companion star by analyzing the spectra of WDMS. At present, the main decomposition method is to use a large number of the white dwarf and M-type star template spectra to fit the WDMS spectra and use the best combination of a group of spectra to represent the spectra of white dwarf and companion star, so as to obtain various physical parameters of stars. This method needs to traverse all spectral combinations and will consume many computing resources due to the matching by the template. Generative adversarial network (GAN) has a good effect and application in signal reconstruction.Based on GAN, this paper constructs a neural network to decompose the spectra of WDMS and directly generates the spectra of white dwarf and companion star through the network. The network model in this paper is an unsupervised learning model, and only three kinds of spectra including WDMS spectra, white dwarf and M-type template spectra, are required for model training, without the intermediate results of other decomposition methods.The model proposed in this paper is easy to be optimized, and some network models can be replaced by CNN and RNN etc. The improvement and optimization methods that can be used to the neural network are also applicable to the network model in this paper.The experimental model in this paper was built with PyTorch deep learning framework and use GPU for accelerated training.This method is used to decompose 1 746 WDMS spectra of SDSS. Compared with other methods, the results show that the trained network model can give similar results with less computational resources. It proves that this model has a good ability to decompose spectra. This method can also be applied to the decomposition of other binary star spectra.
姜 斌，曲美霞，李青苇，曹书画，钟云鹏. 基于生成对抗网络的白矮主序双星光谱分解研究[J]. 光谱学与光谱分析, 2020, 40(10): 3298-3302.
JIANG Bin, QU Mei-xia, LI Qing-wei, CAO Shu-hua, ZHONG Yun-peng. Study on Spectra Decomposition of White Dwarf Main-Sequence Binary Stars Based on Generation Antagonistic Network. SPECTROSCOPY AND SPECTRAL ANALYSIS, 2020, 40(10): 3298-3302.
[1] Zasowski G, Johnson Jennifer A, Frinchaboy P M, et al. The Astronomical Journal, 2013, 146(4): 81.
[2] Jiang Bin, Luo Ali, Zhao Yongheng, et al. Monthly Notices of the Royal Astronomical Society, 2013, 430(2): 986.
[3] Ren J J, Rebassa Mansergas A, Parsons S G, et al. Monthly Notices of the Royal Astronomical Society, 2018, 477: 4641.
[4] JIANG Bin, ZHAO Zi-liang, HUANG Hao, et al(姜 斌，赵梓良，黄 灏，等). Spectroscopy and Spectral Analysis（光谱学与光谱分析）, 2019, 39(6): 1829.
[5] Heller R, Homeier D, Dreizler S, et al. Astronomy & Astrophysics, 2009, 496: 191.
[6] Ian J Goodfellow, Jean Pouget Abadie, Mehdi Mirza, et al. Generative Adversarial Nets, arXiv: 1406. 2661.
[7] Dominic Stark, Barthelemy Launet, Kevin Schawinski, et al. PSFGAN: a Generative Adversarial Network System for Separating Quasar Point Sources and Host Galaxy Light, arXiv: 1803. 08925.
[8] Levi Fussell, Ben Moews. Forging New Worlds: High-Resolution Synthetic Galaxies with Chained Generative Adversarial Networks, arXiv: 1811. 03081.
[9] Martin Arjovsky, Soumith Chintala, Leon Bottou. Wasserstein Generative Adversarial Networks, 2018 the 24th International Conference on Pattern Recognition (ICPR). arXiv: 1701. 07875.
[10] Goodfellow I, McDaniel P, Papernot N. Communications of the ACM, 2018, 61(7): 56.
[11] Koester D. MmSAI, 2010, 81: 921.
[12] Gustafsson B, Edvardsson B, Eriksson K, et al. Astronomy & Astrophysics, 2008, 486：951.
