Stellar Masses of Giant Clumps in CANDELS and Simulated Galaxies Using Machine Learning

Abstract

A significant fraction of high redshift star-forming disc galaxies are known to host giant clumps, whose formation, nature and role in galaxy evolution are yet to be understood. In this work we first present a new automated method based on deep neural networks to detect clumps in galaxy images and show that it is more sensitive and faster than previous proposed methods. We then use this method to systematically detect clumps in the rest-frame optical and UV images of a complete sample of ∼ 1500 star forming galaxies at 1 < z < 3 in the CANDELS survey as well as in images from the VELA forward modeled zoom-in cosmological simulations. Clump stellar masses are estimated through Bayesian SED fitting. We show that observational effects have a dramatic impact on the derived clump properties leading to an overestimation of the clump mass up to a factor of 10, which highlights the importance of fair and direct comparisons between observations and simulations and the limitations of current HST data to study the resolved structure of distant galaxies. After correcting for these effects with a simple mixture density network, we estimate that the clump stellar mass function follows a power-law down to the completeness limit (10 7 solar masses) with the majority of the clumps being less massive than 10 9 solar masses. We find that between 20 − 40% of ∼ 10 10 solar mass star-forming galaxies contain at least one massive (M c > 10 7 M ) off-centered clump but only ∼ 2 − 5% of the total galaxy stellar mass is in those clumps. We also show indications that the contribution of clumps to the stellar mass is more important in extended and low mass galaxies. The simulations explored in this work overall reproduce the shape of the observed clump stellar mass function and clumpy fractions when confronted under the same conditions although tend to lie in the lower limit of the confidence intervals of the observations. This agreement suggests that most of the observed clumps are formed in-situ through violent disk instabilities.

Finite Resolution Deconvolution of Multi-Wavelength Imaging of 20,000 Galaxies in the COSMOS Field: The Evolution of Clumpy Galaxies Over Cosmic Time

“…Lower clumpy fractions are found in galaxies that have low sSFRs compared to the main sequence, with the clumpy fractions increasing by ∼10% at the highest ∆ log(sSFR) bin. This is different what was reported by Huertas-Company et al (2020), where they observed lower clumpy fractions that show no dependencies on ∆ log(sSFR).…”

“…The methodologies for detecting clumps vary from studies to studies, and no universal way have been defined. A number of methods have been used to detect clumpy structures, but the majority are based on identifying pixels with enhanced intensity through a computer algorithm (e.g., Wuyts et al 2012;Guo et al 2015;Zanella et al 2019;Huertas-Company et al 2020). Our normalized profiles naturally allow us to perform a similar analysis.…”

“…The error bar represents the Poisson count statistic. The white markers are measurements made by Huertas-Company et al (2020). They classified clumpy galaxies as those having clumps with stellar mass greater than 10 10 M , and measured ∆ log(sSFR) in relation to the M * -sSFR relation from Fang et al (2018) be systematic errors as various methodologies are used for detecting clumps.…”

The nature of giant clumps in high-z discs: a deep-learning comparison of simulations and observations

“…The method of skip links provides the machine additional information on how to reconstruct the image. The UNet has already been used in other astrophysical studies (Hausen & Robertson 2020;Huertas-Company et al 2020;Hong et al 2020;Boucaud et al 2020).…”

“…The images of high-redshift discs typically show giant clumps. The clumps are especially pronounced in rest-frame UV and H 𝛼 emission maps, which serve as star-formation tracers (Elmegreen & Elmegreen 2006;Genzel et al 2006Genzel et al , 2008Förster Schreiber et al 2009Wuyts et al 2012Wuyts et al , 2013Guo et al 2015;Shibuya et al 2016;Guo et al 2018;Huertas-Company et al 2020). Clumps are also detected in interferometric submilimeter observations.…”