From VIPERS to SDSS: Unveiling galaxy spectra evolution over 9 Gyr through unsupervised machine learning^⋆

¹ Univ. Grenoble Alpes, CNRS, IPAG, Grenoble, France
e-mail: julienhm.dubois@gmail.com; didier.fraix-burnet@univ-grenoble-alpes.fr
² Institute of Space Sciences (ICE, CSIC), Campus UAB, Carrerde Can Magrans, s/n, 08193 Barcelona, Spain
e-mail: siudek@ice.csic.es
³ IRAP, Université de Toulouse, CNRS, CNES, UPS, 14, avenue Edouard Belin, 31400 Toulouse, France

Received: 19 December 2023
Accepted: 11 April 2024

Abstract

Aims. This study aims to trace the chronological evolution of galaxy spectra over cosmic time. Focusing on the VIPERS dataset, we seek to understand the diverse population of galaxies within narrow redshift bins, comparing our findings with the previously mapped diversity of SDSS galaxies.

Methods. We used Fisher-EM, an unsupervised sub-space model-based classification algorithm to classify a dataset of 79 224 galaxy spectra from the VIMOS Public Extragalactic Redshift Survey (VIPERS). The dataset was divided into 26 samples by bins of redshift ranging from z ∼ 0.4 to z ∼ 1.2, which were classified independently. Classes of subsequent bins were linked through the k-nearest neighbours method to create a chronological tree of classes at different epochs.

Results. Based on the optical spectra, three main chronological galaxy branches emerged: (i) red passive, (ii) blue star forming, and (iii) very blue, possibly associated with AGN activity. Each of the branches differentiates into sub-branches, discriminating finer properties such as D4000 break, colour, star-formation rate, and stellar masses, and/or disappear with cosmic time. Notably, these classes align remarkably well with the branches identified in a previous SDSS analyses, indicating a robust and consistent classification across datasets. The chronological ‘tree’ constructed from VIPERS data provides valuable insights into the temporal evolution of these spectral classes.

Conclusions. The synergy between VIPERS and SDSS datasets enhances our understanding of the evolutionary pathways of galaxy spectra. The remarkable correspondence between independently derived branches in both datasets underscores the reliability of our unsupervised machine-learning approach. The three sub-trees show complex branching structures that highlight different physical and evolutionary behaviours. This study contributes to the broader comprehension of galaxy evolution by providing a chronologically organised framework for interpreting optical spectra within specific redshift ranges.