Different Origins of Nucleated and Non-nucleated Dwarf Elliptical Galaxies: Identified by the Deep-learning

TL;DR

This study presents a deep-learning detection framework to identify and classify Virgo cluster dwarf ellipticals (dEs) and their central nuclei, enabling a robust census of nucleated versus non-nucleated dEs. By combining a region-based CNN with an External Attention Network, the authors assemble 2,123 dEs (m_g < 20) over a wide Virgo region and validate detection against the NGVS catalog, achieving substantial recovery. The results show nucleated dEs preferentially reside near massive galaxies and cluster centers, while non-nucleated dEs are more diffusely distributed and aligned with the cluster's global potential, implying in-situ formation for nuclei and ex-situ accretion for non-nucleated systems. These findings link NSC formation to environmental processing and suggest a co-evolution of dEs and ultra-compact dwarfs within cluster environments, with broad implications for dwarf galaxy evolution in dense regions.

Abstract

Dwarf elliptical galaxies (dEs) are the dominant population in galaxy clusters and serve as ideal probes for studying the environmental impact on galactic evolution. A substantial fraction of dEs are known to harbor central nuclei, which are among the densest stellar systems in the Universe. The large-scale distribution and the underlying origin of nucleated and non-nucleated dEs remain unresolved. Using a state-of-the-art machine learning framework, we systematically scan the Virgo cluster region ($15\arcdeg \times 20\arcdeg$ centered at $R.A. = 187.2\arcdeg$ and $Dec. = 9.6\arcdeg$) and construct the largest homogeneous sample of dEs (of total 2,123) with robust nucleus classifications. We find that nucleated dEs are more spatially clustered and exhibit a stronger association with massive galaxies than their non-nucleated counterparts. This suggests that most nucleated dEs likely formed alongside massive galaxies within the cluster (i.e, the in-situ formation). In contrast, non-nucleated dEs are more widely distributed across the cluster and align more closely with Virgo's global potential well, as traced by the cluster's hot gas. This indicates that most non-nucleated dEs originated outside the cluster (i.e, the ex-situ formation) and were later accreted and redistributed within it. Our findings shed new light on how dEs and their central nuclei form and evolve.

Figures (10)

• Figure 1: The neural network architecture of our deep learning framework for dE galaxy detection and localization. Our framework extends the R-CNN to predict both segmentation masks and class probabilities. (a) The input training RGB image is processed by a custom layer CNN to produce multi-scale feature maps. (b) A region proposal network (RPN) identifies candidate regions likely to contain galaxies. (c) An example of the input inference image. (d) The probability map produced after processing through CNN and ROI alignment. (e) The potential dE candidates are identified. (f) An External Attention Network classifier, trained on a dataset of visually identified dE, assigns each region to a class (dE or non-dE). The final output shows "Accepted" (blue) and "Removed" (red) regions, effectively distinguishing true detections from false positives.
• Figure 2: All sky distribution of identified dEs in our sample. A subset of the full sample which have a line-of-sight radial velocity measurement is shown in black. We also highlight the six main subgroups of galaxies in the Virgo cluster area with large blue dot symbols and their names.
• Figure 3: The $g$-band magnitude distributions of our dE sample (gray) and a subset having radial velocity information (black).
• Figure 4: The surface number density map of dEs is compared with the spatial distribution of massive galaxies and X-ray emission in the Virgo cluster core region. The surface number density map is constructed using 2D Gaussian Kernel Density Estimation (KDE) with a kernel width of 0.$^{\circ}$175 (50 kpc). The color bar at the top represents the normalized density. Red contours indicate the surface flux density of hard X-ray emission (0.4–2.4 keV) in the Virgo cluster region, based on the ROSAT all-sky survey Bohringer94. Small gray dots denote individual dEs (from which the surface number density map is generated), while large black dots represent massive galaxies (M$_{*}$$>$ 10$^{10}$ M$_{\odot}$). Large blue dots with their names are the six main subgroups of galaxies.
• Figure 5: Top: Fraction of nucleated dEs as a function of magnitude. The black line shows the nucleated fraction measured by Sanchez19 in the Virgo cluster core, while the red line represents our measurement for the full dE sample. The error bar represents the fraction of dEs that we were not able to classify into nucleated and non-nucleated. Bottom: $g$-band magnitude distributions of nucleated (red histogram) and non-nucleated (blue histogram) dEs, which we have calculated assuming all dEs are located at the average distance of the Virgo cluster, i.e., 16.5 Mpc.