A sparse regression approach to modelling the relation between galaxy stellar masses and their host haloes

Abstract

Sparse regression algorithms have been proposed as the appropriate framework to model the governing equations of a system from data, without needing prior knowledge of the underlying physics. In this work, we use sparse regression to build an accurate and explainable model of the stellar mass of central galaxies given properties of their host dark matter (DM) halo. Our data set comprises 9521 central galaxies from the EAGLE hydrodynamic simulation. By matching the host haloes to a DM-only simulation, we collect the halo mass and specific angular momentum at present time and for their main progenitors in 10 redshift bins from z = 0 to z = 4. The principal component of our governing equation is a third-order polynomial of the host halo mass, which models the stellar-mass–halo-mass relation. The scatter about this relation is driven by the halo mass evolution and is captured by second- and third-order correlations of the halo mass evolution with the present halo mass. An advantage of sparse regression approaches is that unnecessary terms are removed. Although we include information on halo specific angular momentum, these parameters are discarded by our methodology. This suggests that halo angular momentum has little connection to galaxy formation efficiency. Our model has a root mean square error (RMSE) of 0.167log10(M*/M⊙), and accurately reproduces both the stellar mass function and central galaxy correlation function of EAGLE. The methodology appears to be an encouraging approach for populating the haloes of DM-only simulations with galaxies, and we discuss the next steps that are required.

Clustering of red-sequence galaxies in the fourth data release ofthe Kilo-Degree Survey

“…In order to account for the impact of data quality, we extend the works of Bautista et al (2018) and Icaza-Lizaola et al (2020) to allow for a more flexible relation between the systematic-induced variations of observed galaxy densities and the survey properties, while making use of heavy regularization to avoid overfitting. In comparison to Ross et al (2012); Crocce et al (2019b), our adopted framework for removing the impact of survey properties does not make any assumption regarding the lack of correlation between the survey properties.…”

“…By necessity, the data quality of large galaxy surveys such as KiDS is not homogeneous. The variable survey conditions can potentially affect the observed galaxy density and consequently can bias the cosmological inferences with these galaxy samples (Ross et al 2012;Leistedt & Peiris 2014;Leistedt et al 2016;Zhai et al 2017;Elvin-Poole et al 2018;Bautista et al 2018;Crocce et al 2019b;Kitanidis et al 2019;Rezaie et al 2019;Heydenreich et al 2020;Icaza-Lizaola et al 2020). In this section, first we describe the imaging systematics considered in our analysis, and then we discuss our mitigation strategy.…”

“…Therefore, we mitigate the impact of systematics by introducing a set of photometric weights. This approach has been widely utilized in galaxy clustering analyses: the clustering of LRGs in SDSS BOSS (Ross et al 2012(Ross et al , 2017b, galaxy clustering in the Dark Energy Survey (Elvin-Poole et al 2018;Crocce et al 2019b), DESI legacy survey (Kitanidis et al 2019), and finally clustering of LRGs in SDSS-eBOSS (Bautista et al 2018;Icaza-Lizaola et al 2020).…”

“…By up-weighting (downweighting) areas of the survey where the galaxy density is down-graded (enhanced) due to survey properties, this scheme mitigates the systematic modes present in the sample. In our method, similar to that of Bautista et al (2018); Icaza-Lizaola et al (2020), the photometric weights are estimated such that possible correlations between the survey properties are accounted for.…”

The Completed SDSS-IV extended Baryon Oscillation Spectroscopic Survey: measurement of the BAO and growth rate of structure of the luminous red galaxy sample from the anisotropic correlation function between redshifts 0.6 and 1

“…With respect to BOSS, it explores galaxies at higher redshifts, covering the range 0.6 < z < 2.2. Using the first two years of data from Data Release 14 (DR14), BAO and RSD measurements have been performed using different tracers and methods: LRG BAO (Bautista et al 2018), LRG RSD (Icaza-Lizaola et al 2020), quasar BAO (Ata et al 2018), quasar BAO with redshift weights (Zhu et al 2018), quasar BAO Fourier-space (Wang et al 2018), quasar RSD Fourierspace (Gil-Marín et al 2018), quasar RSD Fourier-space with redshift weights (Ruggeri et al 2017(Ruggeri et al , 2019, quasar RSD in configuration space (Hou et al 2018;Zarrouk et al 2018), and quasar tomographic RSD in Fourier space with redshift weights (Zhao et al 2019).…”