Robust galaxy image decompositions with Differential Evolution optimisation and the problem of classical bulges in and beyond the nearby Universe

TL;DR

This paper assesses the reliability of 2D galaxy image decompositions when distinguishing nuclear discs from classical bulges, by comparing supervised Nelder-Mead, unsupervised Differential Evolution (DE), and unsupervised MCMC fits on a nearby, nuclear-disc-dominated sample. Using 16 TIMER galaxies with Spitzer/ S4G images and artificially degraded resolutions, the study shows that DE and MCMC can recover structural parameters without subjective initial guesses and yield $n<2$, consistent with nuclear discs, whereas limited spatial resolution artificially inflates the bulge Sérsic index $n$, biasing classifications toward classical bulges. The results imply that higher-z studies may overestimate the classical bulge fraction due to resolution effects, a bias that persists even for facilities like Euclid, HST, and JWST at certain redshifts. The work advocates adopting unsupervised DE/MCMC decompositions, improved PSF handling (oversampling), and more sophisticated models (e.g., disc breaks, substructures) to obtain robust galaxy central structure, particularly for large surveys where subjective inspection is impractical.

Abstract

Deconstructing galaxies through two-dimensional decompositions has been shown to be a powerful technique to derive the physical properties of stellar structures in galaxies. However, most studies employ fitting algorithms that are prone to be trapped in local minima, or involve subjective choices. Furthermore, when applied on samples beyond the nearby Universe, results on the fraction of classical bulges in disc galaxies do not agree with studies on nearby galaxies. The latter studies point to a small fraction of classical bulges, possibly challenging our merger-driven picture of galaxy formation. Therefore, understanding the discrepancy between observations in and beyond the nearby Universe is of paramount importance. In this paper, I use a sample of 16 nearby galaxies drawn from the TIMER project, which previously have been shown to not host classical bulges, and perform decompositions applying different methodologies and employing the original images as well as artificially redshifted images. I show that the Differential Evolution algorithm is able to provide accurate measurements of structural properties with little subjective intervention, correctly indicating the presence of nuclear discs (not classical bulges). However, I also show that when the physical spatial resolution is not adequate, a systematic overestimation of the photometric bulge Sérsic index leads to the false conclusion of the presence of classical bulges. I discuss how this may be the root cause of the discrepancy mentioned above, and point out how this issue may be a problem even with data from facilities such as Euclid, HST and JWST.

Figures (37)

• Figure 1: Images used as input to the decompositions. For each galaxy, the left panel shows the original, unaltered S$^4$G image (with an average PSF FWHM of $\approx170$ pc), whereas the central and right panels show the artificially redshifted images, in which the PSF FWHM is $\approx1.7$ kpc and $\approx3.4$ kpc, respectively.
• Figure 1: continued.
• Figure 2: Photometric bulge Sérsic index $n$ obtained for the galaxies in our sample in three previous studies plotted against the values derived in this work via the supervised Nelder-Mead fits. Also plotted on the bottom right corner are fiducial errors bars derived by KimGadShe14 using budda. The relative average error on $n$ found by Kim et al. corresponds to 13%, which, for the galaxies studied here, implies a mean 1$\sigma$ error of 0.26.
• Figure 3: Distribution of the difference between the photometric bulge Sérsic indices obtained with and without employing an oversampled PSF. While the number of galaxies is small -- and a more statistically robust analysis is required to confirm the observed systematic offset -- the results shown here indicate that the bulge Sérsic index can be significantly overestimated if the PSF treatment does not include an oversampled PSF model.
• Figure 4: Photometric bulge Sérsic index $n$ obtained employing the Differential Evolution (DE) algorithm in unsupervised fits plotted against the values derived via the supervised Nelder-Mead (NM) fits. Also plotted on the bottom right corner are fiducial errors bars derived by KimGadShe14 using budda. The relative average error on $n$ found by Kim et al. corresponds to 13%, which, for the galaxies studied here, implies a mean 1$\sigma$ error of 0.26. This figure concerns only the fits to the original S$^4$G images and not the redshifted images.