All Puffed Up: Exploring Ultra-diffuse Galaxy Origins through Galaxy Interactions

KUG 0203-Dw1 and KDG 013 were observed in the VLA D-configuration during July 2022, as a part of project 22A-225 (PI: D. Sand). Each target had a total on-source integration time of approximately 4 h. Both targets were observed with a velocity resolution of $\sim$0.8 $\text{km}\,\text{s}^{-1}$ (averaged to 5 $\text{km}\,\text{s}^{-1}$ during imaging). A 4 MHz band (1024 channels) was used for KUG 0203-Dw1 and an 8 MHz band (2048 channels) for KDG 013. These bands were centered at 1885 km s^-1 and 4074 km s^-1 for KUG 0203-Dw1 and KDG 013, respectively.¹¹1In addition, the data were also recorded simultaneously at a lower spectral resolution (62.5 kHz per channel, or $\sim$13 km s^-1) over a bandwidth of 32 MHz, providing broad coverage in velocity ($\sim$7000 km s^-1). However, in practice no significant H i signal at the target locations was detected beyond the narrower, higher spectral resolution band described in the main text, and we do not consider the low resolution data further here. KUG 0203-Dw1 was also selected for follow-up under Director’s Discretionary Time in the VLA C-configuration during winter of 2022 (PI: C. Fielder), with the same spectral configuration and a total on source integration time of 4 h. The results of these observations are presented in Figure 2. For KUG 0203-Dw1 we imaged the C and D array data both separately and together, with the latter used for the contour plots (referred to as VLA C+D data). The properties of these observations are summarized in Table 2.

The data were reduced with the H i pipeline²²2https://github.com/AMIGA-IAA/hcg_hi_pipeline of Jones et al. (2023b). For a full description we refer the reader to that work. In brief the pipeline performs both manual and automated flagging before proceeding with gain and phase calibrations, $UV$ continuum subtraction, and imaging using standard CASA tasks (Common Astronomy Software Applications; McMullin et al. 2007).

The KUG 0203-Dw1 data suffered from moderate interference in C-configuration where $\sim$19% of visibilities were flagged, while $\sim$42% were flagged in D-configuration. The higher fraction of flagged data in D-configuration is mostly the result of four antenna failures. A Briggs robust parameter of 0.5 was used during imaging for the C+D data to offer a compromise between sensitivity and resolution. The synthesized beam size was determined to be $27.0\times 22.0$ arcsec² and the central velocity is 1885 km s^-1 (a similar measurement to the host).

The KDG 013 D-array data had moderate interference with $\sim$22% of visibilities flagged. Imaging was conducted with automated masking, applying a Briggs robust parameter of 2.0 (i.e. natural weighting) in order to maximize sensitivity to the extended features. This resulted in a synthesized beam size of $74\times 60$ arcsec².

For the generation of the VLA moment zero maps in Figure 2, we utilized the SoFiA software (Serra et al., 2015) to create a source mask. Employing the standard smooth and clip algorithm, we conducted the process both with no smoothing and with Gaussian smoothing kernels approximately 0.5 and 1.0 times the beam size. Additionally, we applied boxcar spectral smoothing over 0 and 3 channels. The clip threshold was set at $3.5\sigma$, and to ensure robustness, we imposed a stringent 95% reliability threshold.

KUG 0203-Dw1 and KDG 013 were observed in July and September of 2022, under HST program ID 16890 (Sand et al., 2021). Both targets were observed by the Advanced Camera for Surveys (ACS) with the Wide Field Channel (WFC) in the F555W and F814W filters. The total integration time for KUG 0203-Dw1 is 2046 s in F555W and 2098 s in F814W. Integration times for KDG 013 correspond to 2046 s and 2089 s in F555W and F814W respectively. In addition, Wide Field Camera 3 (WFC3) images were taken in parallel to utilize as reference background fields as needed.

We present RGB color composite images for KUG 0203-Dw1 and KDG 013 in Figure 3, constructed from the combined F555W and F814W exposures. Highlighted sources are discussed in detail in Section 5. After obtaining the standard data products from the STScI archive, the individual exposures were aligned and background subtracted using the Dolphot 2.0 software (Dolphin, 2000, 2016) with the standard ACS and WFC3 parameters provided in the user manual. Then, a combined point source catalog for each of the fields was generated. Dolphot used the HST to Johnson-Cousins magnitude conversion factors presented in Sirianni et al. (2005) to determine the $V$- and $I$-band magnitudes from F555W and F815W magnitudes. The photometry was then corrected for Galactic extinction using the NASA/IPAC³³3https://irsa.ipac.caltech.edu/applications/DUST/ online tool, with values derived from the Schlafly & Finkbeiner (2011) extinction coefficients. Unless otherwise indicated, all magnitudes presented in this work are Milky Way extinction-corrected Vega magnitudes.

Completeness limits are determined via artificial star tests with the tools provided by Dolphot. We place nearly $100,000$ artificial stars into the ACS field, which span a large color range from $-$1 to 2 in F555W-F814W (well beyond the range used to select GCs). We then measure the fraction of those stars that we recover as a function of apparent magnitude. For KUG 0203-Dw1 we find that we are $90\%$ complete to $m_{F814W}=26.0$ and $50\%$ complete to $m_{F814W}=26.8$. For KDG 013 we are $90\%$ complete to $m_{F814W}=26.3$ and $50\%$ complete to $m_{F814W}=27.1$. Note that using the Sirianni et al. (2005) conversions the completeness limits are the same in $I$-band for the $(V-I)$ color range of GCs.

The optical/UV properties of both UDGs are determined by GALFIT (Peng et al., 2002), using the same methodology presented in Bennet et al. (2017, 2018) (see also Merritt et al., 2016). The optical properties are derived from the CFHT data, but the HST data has been shown to provide near identical results (see J21). Both galaxies were modeled using a standard Sérsic profile (Sérsic, 1963), but KUG 0203-Dw1 was more of a challenge for GALFIT to constrain and ultimately required two different Sérsic profiles to fit. Otherwise all fit parameters are allowed to vary without restriction. The uncertainties on the parameters for each UDG are determined by injecting 100 simulated UDGs into the image, each with the best fitting properties of the given UDG as determined by GALFIT. Then GALFIT is used to recover the properties of the simulated UDGs. The scatter on the recovered properties of the simulated UDGs represents the uncertainty on the best fit properties of the real UDG. We present our parameter estimates in Table 3. $FUV$, $NUV$, $g$, and $r$-band magnitudes are presented in the AB system, while $V$ and $I$-band magnitudes remain in the Vega system. CFHT $g$ and $r$-band data is converted to $V$ and $I$ following Jester et al. (2005) (see also footnote $a$ of Table 1). The best-fit GALFIT models both have an ellipticity, consistent with the elongated morphology of the UDGs, as well as a circularized radius. Note that some of the CFHT filters suffered from over-subtraction. To mitigate this issue general properties like radius were measured in filters where this was not an issue and then fixed when determining other measurements from more problematic filters. Errors associated with the over-subtraction are significantly smaller than the errors reported.

Stellar masses are derived using the relations from Table 1 of Zhang et al. (2017) and our CFHT derived $(V-I)$ and $M_{V}$. We derive a stellar mass-to-light ratio $\Gamma_{*}=0.5$ for KUG 0203-Dw1 and $\Gamma_{*}=0.9$ for KDG 013 which we then convert to stellar mass. We caution the reader to see this is an estimate from a model and not a derived quantity.

The UV data of both UDGs originate from GALEX (Martin & GALEX Team, 2005). Foreground sources are first masked (see the left panel of Figure 4). Then UV fluxes are determined via aperture photometry, with an aperture radius equivalent to two half-light radii of the best fit GALFIT model to the optical data (see the dashed aperture in Figure 4). We derive star formation rates using the relation from Iglesias-Páramo et al. (2006).

GCs typically have radii of a few parsecs (Brodie & Strader, 2006). Assuming a nominal radius of 5 pc, at the distance of KUG 0203-Dw1 and KDG 013 this corresponds to a radius of 0.04” and 0.02” respectively. The HST FWHM of the PSF is $\sim 0.15^{\prime\prime}$ for ACS and $\lesssim 0.1^{\prime\prime}$ for WFC3, indicating that with either camera GCs will appear as point sources.

In this work, the globular cluster candidate (GCC) selection is identical to that described in J21 and F23. In brief, we perform aperture photometry on the HST images using the Dolphot v2.0 software. We then perform subsequent cuts on the Dolphot produced point source catalog such that we limit to sources classified as ‘stars’ (other source identifiers are ‘elongated object’, ‘object too sharp’, and ‘extended object’), with no photometric flags, and with S/N $>5$. For both HST filters we further constrain with fit quality diagnostics. The sharpness is enforced to fall within the bounds of $-$0.3 to 0.3 and roundness is required to be less than 0.3, limiting elongated or distorted sources and those that are overly compact. We set a maximum limit on crowding to 0.5 mag in both filters. The Dolphot magnitude uncertainties are required to be less than 0.3 mag. Next a color cut is used, to limit sources to those within the color range expected for GCs. We use $0.5<(V-I)<1.5$ from Brodie & Strader (2006), which should generally eliminate blue star-forming regions from our source selection.

Then we apply a concentration cut, which is determined by comparing the flux of the background subtracted $F814W$ image in concentric circular apertures of 4 and 8 pixels in diameter (see similar approaches in e.g., Peng et al. 2011; Beasley & Trujillo 2016; Saifollahi et al. 2021). We define our concentration index as $C_{4-8}=-2.5\log_{10}{(\frac{N_{4\rm{pix}}}{N_{8\rm{pix}}})}$ where $N_{4\rm{pix}}$ represents the sum of flux values in a 4 pixel aperture. To select GCCs we allow the concentration index parameter to span the range $0.2-0.8$ mag where we expect GCs to lie, which accounts for the extended locus of point sources (in magnitude-concentration space) and slightly extended sources. This is because larger GCs may not be perfect point sources, particularly at the distance of KUG 0203-Dw1. This selection is designed to eliminate diffraction spikes, compact background galaxies, and any other high concentration index sources.

Last, we employ a cut in magnitude in an effort to maximize the number of real GCs and eliminate contaminants. We do this by assuming the GCs follow the Gaussian dwarf elliptical GC luminosity function (GCLF) of Miller & Lotz (2007), similar to the approach of Peng et al. (2011). The Gaussian GCLF peaks at $mu_{M_{I},\rm{Vega}}=-8.12$ mag with $\sigma_{M_{I}}=1.42$. We use the peak of the GCLF at the distance of each UDG to serve as a brightness minimum in selecting our GCCs. This is used in lieu of the completeness limit in order to minimize stellar contaminants and sample parameter space where our data is effectively complete. We also employ a brightness maximum, above which we expect most point sources to be foreground stars. This value corresponds to $M_{I}>-12.38$, or $3\sigma$ from the mean of the GCLF. In apparent magnitude, we select all GCCs in the range of $19.8<m_{I}<24.0$ (where the 90% completeness limit is 26) for KUG 0203-Dw1, and $21.3<m_{I}<25.6$ for KDG 013 (where the 90% completeness limit is 26.3). Such a cut means we are only sampling half of the luminosity function, specifically the bright half of the GCLF. As a result, our final GC counts will be corrected by multiplying the initial resultant GC counts derived in this manner by 2.

We provide a color-magnitude diagram in Figure 8 of point sources identified in Dolphot. All plotted points pass the initial set of signal-to-noise, sharpness, roundness, crowding, and magnitude uncertainty cuts, and red points pass the subsequent concentration cut. Point sources that fall within the color and magnitude selection criteria and lie within $2\times r_{1/2}$ constitute the final GCC sample. These sources are highlighted in the HST images in Figure 3, encircled by teal apertures. Note that in the KUG 0203-Dw1 frame (Figure 3, lower left) a majority of the GC candidates are centered on the two galaxies with few additional contaminants in the field, while the KDG 013 frame has substantially more contaminants (Figure 3, lower right).

After the DOLPHOT photometric catalog has been constrained following the procedure described above, we elect to select GCCs within $2\times$ the best fit circularized aperture radius, one of the most common choices within the literature for selecting GCs (most recently e.g., Müller et al. 2021; Danieli et al. 2022; Janssens et al. 2022; Jones et al. 2023a). While expanding to larger radii adds additional GC candidates, it also introduces a larger number of contaminants. In the case of KUG 0203-Dw1 $3\times r_{1/2}$ falls close to the host galaxy, meaning GCCs within this region could be associated with either galaxy. In the case of KDG 013, $3\times r_{1/2}$ doubles the number of GCCs but this field has a rather high abundance of contaminants (see Figure 3 lower right). We therefore take the standard approach of using $2\times r_{1/2}$ to mitigate contaminants. In Section 4.2 note that the best GALFIT models had some ellipticity. Selecting GCCs within an elliptical aperture yielded similar counts to those selected within a circularized aperture. Likewise, due to the tidal features of these UDGs, defining a strict boundary for these galaxies is non-trivial. We therefore choose to adopt the simpler approach of counting GCCs within a general circularized radius and subsequently subtracting contaminants.

Contaminant counts for false GCCs were determined with a parallel WFC3 field taken in an empty patch of sky close to the host systems. This is particularly useful for KUG 0203-Dw1, where the host lies in the frame making it difficult to disentangle GCs associated with the host and actual contaminants. Additionally, the coordinates of the parallels are checked to ensure they are sampling similar environments to that of the UDG. To determine contaminant rates we employ the same cuts used to identify GCCs in the ACS field as described in Section 5.1. The WFC3 field-of-view (160”x160”) is smaller than the ACS field-of-view (202”x202”), so the number of contaminants identified must be scaled relative to the ACS detector sky area incorporating the pixel size difference before scaling to within the area of the actual GC searches. We note that contaminant counts were also computed for the ACS pointings, which yield comparable GC abundances.

The raw GCC counts within $2\times$ the effective radii of either UDG are small ($<10$ per UDG), where number counts are dominated by Poisson statistics. We thus rely on a Bayesian approach to determine the number of GCs within the UDG aperture analytically. The procedure for this method is outlined in Section 4.3 of J21, which we summarize here. The count of potential false GC candidates within the UDG’s aperture conforms to a Poisson distribution with an unknown mean, denoted as ‘b,’ representing the background rate of false positives arising from GCs linked to the host galaxy and other incidental sources such as Milky Way foreground and compact background galaxies. The probability of false GCCs within the aperture of the UDG can be described by a normal Poisson process. Multiplying this probability with the analytic Bayesian solution with a flat prior to estimate $b$ and $b$ marginalized yields the probability mass function for the number of GCs. This result is written as:

where $O$ is short for observations, $N_{\rm{GC}}$ is the number of genuine GCs in the UDG apertures, $N_{\rm{GCC}}$ is the number of GCCs in the UDG aperture, $N_{\rm{bg}}$ is the number of background sources that pass as false GCCs, and $A_{\rm{bg}}$ is the area of the background field. $A_{\rm{bg}}$ is dimensionless as it is expressed in terms of the area of the circular aperture used to count GCCs around the target. Because we use the WFC3 field to determine background counts, $A_{\rm{bg}}$ is scaled such that it is in units of the ACS area.

While the WFC3 parallels are close to the systems of interest, we may still suffer from small number statistics. Thus we determine an estimate of contaminants using the web-based tool TRILEGAL v1.6 ⁴⁴4http://stev.oapd.inaf.it/cgi-bin/trilegal_1.6 (Girardi et al., 2005, 2012). TRILEGAL (TRIdimensional modeL of thE GALaxy), a model of Milky Way star counts, constructs a three-dimensional representation of our galaxy, incorporating various stellar populations in its thin disk, thick disk, halo, and bulge components calibrated using extensive survey data. Employing TRILEGAL, we generate a simulated star catalog with default settings and a Chabrier log-normal initial mass function (Chabrier, 2001). We query a full square degree centered on each galaxy, and then randomly down-sample 10,000 times to both the area of the ACS field and the WFC3 field to ensure our contamination estimates are consistent. We find similar estimates to those obtained from the WFC3 fields.

The search radius of KUG 0203-Dw1 is equivalent to $23.6\pm 3.6$ arcsec or $3.1\pm 0.6$ kpc at its distance (see Table 3). A total of 5 GCCs were identified within this aperture (see the outer dashed radiis in Figure 3). We find a total contaminant abundance in the parallel field of 13, which when scaled to the ACS area covered by the UDG yields a raw abundance of 1 contaminant. For comparison TRILEGAL estimates $3\pm 2$ contaminants. With our probabilistic approach for determining GC abundance and multiplying our result by $2$ to account for only sampling the bright half of the GCLF, our final most probable GC abundance for KUG 0203-Dw1 is $N_{\rm{GC}}=8\pm 2$.

The search radius of KDG 013 is $25.0\pm$1.8 arcsec or $6.7\pm 0.5$ kpc . Within this aperture we identify a total of 4 GCCs, with a raw contaminant abundance of 2 in the UDG area. TRILEGAL estimates $2\pm 1$ contaminants. After performing our probabilistic analysis we find our most probable GC abundance for KDG 013 is $N_{\rm{GC}}=6\pm 2$.

We can extend the $I$-band magnitude cut in Section 5.1 from $M_{I}<-8.12$ down to our $90\%$ completeness limit, which corresponds to $M_{I}<-6.14$ for KUG 0203-Dw1 (sampling about 2/3 of the faint-end GCLF) and $M_{I}<-7.42$ for KDG 013 (sampling nearly the full GCLF). While a looser cut such as this may allow for an increase in contaminants, we find the same predicted abundance for both systems (within error) that we do when determining the GC abundance from the bright half of the GCLF alone. At this extended magnitude limit we find $9^{+1}_{-2}$ GCs for KUG 0203-Dw1 and $8\pm 2$ for KDG 013.

The GC abundances for these systems are plotted in Figure 5 (orange stars) alongside the rest of our sample of UDGs with tidal features (other orange points). The nearby galaxy sample of Harris et al. (2013) is plotted for comparison, which includes elliptical, spiral, lenticular, and irregular galaxies and extends to brighter galaxies. We also include a number of UDG samples from both clusters (blue points) and groups (pink points) for comparison: Coma cluster UDGs (Forbes et al., 2020), Virgo cluster UDGs (Lim et al., 2020), Hydra I cluster UDGs (Iodice et al., 2020; La Marca et al., 2022), and group UDGs compiled in Somalwar et al. (2020). There are also a number of individual UDGs studied in the literature that we highlight, which are considered to be GC-rich or contain over-luminous GCs (DF2, DF4, DGSAT 1): Dragonfly 2 (DF2) and Dragonfly 4 (DF4) (NGC 1052 UDGs; Shen et al. 2021), Dragonfly 44 (DF44, a Coma UDG; Saifollahi et al. 2021 updated from the Forbes et al. 2020 values), NGC 5846-UDG1/MATLAS19 (NGC 5846 group UDG; Müller et al. 2021; Danieli et al. 2022), DGSAT 1 (in a low density region of the Pisces-Perseus supercluster; Janssens et al. 2022).

A study of local volume dwarfs performed by Carlsten et al. (2022) finds that in the stellar mass range of $7<\log{M_{*}/M_{\odot}}<8$ (the approximate mass range of these UDGs) the typical GC abundance ranges from approximately $1-7$. Specifically, the Fornax dwarf ($\log{M_{*}/M_{\odot}}\sim 8$) hosts $6$ bright GCs (de Boer & Fraser, 2016; Pace et al., 2021) at $\sim 1.5$ magnitudes below the peak of the GCLF. These results for local dwarfs are very comparable to the values we derive for KUG 0203-Dw1 and KDG 013, although both UDGs are close to the expected upper limit.

A convenient way to describe GC richness relative to the galaxy’s luminosity is to calculate specific frequency. While this quantity is common to calculate even at dwarf galaxy masses, the relation between GC abundance and luminosity does not hold as well as it does at higher masses. This is speculated to be due to the stellar-to-halo mass relation becoming less linear at low mass, as GCs are found to trace total DM halo mass and luminosity traces stellar mass (hence why at higher masses a specific frequency of 1 is more common). The definition of specific frequency presented in Harris & van den Bergh (1981) is $S_{N}=N_{\rm{GC}}10^{0.4(M_{V}+15)}$. Using our derived galaxy $M_{V}$ from the CFHT data and the GC abundances this yields a specific frequency of $S_{N}=26.9_{-8.8}^{+8.2}$ for KUG 0203-Dw1 and $S_{N}=3.3_{-0.6}^{+0.8}$ for KDG 013.

In Figure 5 we include diagonal lines of constant specific frequency. KUG 0203-Dw1 and KDG 013 both have specific frequencies consistent with other UDGs and dwarf galaxies (generally between 1 and 50).

When dwarf galaxies undergo collisions, the result can manifest as an increase in the effective radius and angular momentum of the system, accompanied by a decrease in surface brightness and a redistribution of star formation towards the outer regions of the resulting (merged) galaxy. Dwarf major mergers give rise to distinctive tidal features such as tails, bridges, and shells, expected to last $2-5$ Gyrs (e.g., Johnston et al., 2001; Pearson et al., 2018). But distinguishing effects caused by tidal interactions with a massive companion from those caused by a dwarf merger is often challenging, aside from when the dwarfs are not completely merged (Paudel et al., 2018).

There are observed properties in these UDGs that directly challenge the dwarf merger theory. A hydrodynamical simulation study (ROMULUS25) conducted by Wright et al. (2021) suggests that decentralized star formation leads to a steep color gradient. However, the UV emission in KUG 0203-Dw1 and KDG 013 is evenly distributed throughout either galaxy meaning there is no de-centralized tracer of star formation, and we detect no notable color gradient in the optical.

Additionally, the H i morphology observed in KUG 0203-Dw1 is a direct result of the interaction with KUG 0203-100 (see Section 4.1). While it is possible a dwarf merger occurred before the system began interacting with the host, we choose not to invoke multiple formation mechanisms for this UDG to avoid over-complication. It is plausible that KDG 013 resulted from a more recent gas-poor merger. However, given that no current simulations exist for comparison to recent (z $<1$) dwarf mergers yielding UDGs, as nearly all UDGs formed via mergers in the ROMULUS25 simulations occur at earlier times, a recent merger is difficult to assess. While bullet cluster-like collisions have also been suggested to produce UDGs (e.e.g, Ogiya et al., 2022; van Dokkum et al., 2022), this formation mechanism is still not well understood and the observed properties of our UDGs are not consistent with those suggested to have formed from such a mechanism, such as an over-abundance of GCs. Therefore we suggest that the dwarf major merger formation mechanisms is unlikely for either UDG.

Tidal dwarf galaxies (TDGs) form as a result of tidal interactions between two or more massive galaxies, where the tidal forces generated during close encounters lead to the ejection of stellar and gaseous material into extended tidal tails or bridges. These ejected materials can then coalesce and form new, self-gravitating galaxies known as TDGs (e.g., Elmegreen et al., 1993; Duc & Mirabel, 1998; Kaviraj et al., 2012). Note that TDGs and UDGs with tidal features (sometimes called tidal UDGs) refer to different classes of objects, which in some instances may or may not overlap.

Bournaud & Duc (2006) find in their TDG formation simulations that tidal streams responsible for forming a TDG typically dissolves within $300-500$ Myr. Additionally, the H i streams are expected to persist longer than stellar streams (Jones et al., 2019; Lee-Waddell et al., 2019; Namumba et al., 2021). KUG 0203-Dw1 has been shown to have a clear H i stream while KDG 013 does not, meaning that we might expect if they are TDGs then KUG 0203-Dw1 is $<500$ Myr old and KDG 013 is $>500$ Myr old. While long-lived TDGs are anticipated to form at the tips of H i tails (Bournaud & Duc, 2006), KUG 0203-Dw1’s central location within the stream, with leading and trailing tails more indicative of dwarf stripping, suggests otherwise.

Because TDGs form very differently from typical galaxies, the expectation is that TDGs do not contain GCs (or DM), and at present, studies of low redshift TDGs do not find GCs (e.g., Fensch et al., 2019). From our Bayesian analysis discussed in Section 5.2 we can obtain probabilities that either UDG has 0 real GCs. In the case of KUG 0203-Dw1 there is only a $0.75\%$ probability that none of the GC candidates are real, with a $7.26\%$ probability for KDG 013. Hence, the abundances of GCs provide strong evidence against categorizing these UDGs as young TDGs.

It has been suggested that long lived GCs could form in an old TDG (Fensch et al., 2019), but current observational evidence is lacking on this front. Therefore we posit that given our current understanding of TDG formation and the supporting observational evidence, a TDG origin is unlikely for either UDG, especially given their GC abundance.

Strong gravitational effects induced by tidal interactions with massive companions can expand a galaxy’s stellar halo, thereby decreasing the central surface brightness and pushing a dwarf galaxy into UDG parameter space (Errani et al., 2015; Conselice, 2018; Carleton et al., 2019; Tremmel et al., 2020). While previous research has primarily focused on UDGs in cluster environments (e.g., Sales et al., 2020; Carleton et al., 2021, and sources mentioned above), there is mounting evidence indicating that this phenomenon may also occur in lower density group environments. Hydrodynamical simulations by Jiang et al. (2019) and Liao et al. (2019) suggest that approximately 50% of group UDGs form from dwarfs on eccentric orbits undergoing tidal stripping and heating, with the remainder being pre-existing UDGs later affected by group dynamics. After a pericentric passage the UDG progenitor is stripped of a substantial portion of its gas and star formation slows (Román & Trujillo, 2017; Benavides et al., 2021). Jiang et al. (2019) and Liao et al. (2019) find that within 1 to 2 pericentric passages a dwarf is transformed into a UDG and has lost its cold gas.

Both KUG 0203-Dw1 and KDG 013 have tidal plumes, disturbed morphologies, and KUG 0203-Dw1 has a clear stellar and H i stream connection to its host galaxy. It is without doubt that the group environment must have played a role in shaping the way the UDGs appear now. Additionally, we have demonstrated both KUG 0203-Dw1 and KDG 013 display GC abundances typical of galaxies given their $V$-band brightness (Figure 5), suggestive of a classical dwarf origin.

KDG 013 has no clear H i detection, implying that NGC 830 or other group members may have pulled its gas off after pericentric passage, instigating its UDG nature. The situation is less clear for KUG 0203-Dw1, which still contains an over-density of H i and is in close proximity in projection to the host (only 10.9 kpc). Thus the UDG interaction with the host is likely in an early phase, where the UDG is being tidally puffed but has not yet been fully stripped of its gas. This is further motivated by the rather small circularized radius of the UDG, which just barely passes the UDG criteria ($1.53\pm 0.3$ kpc compared to the limit of $>1.5$ kpc). Because KUG 0203-Dw1 has a bluer stellar population and high specific star formation rate than KDG 013 one can envision them at different stages in the tidal processing in the group, with KUG 0203-Dw1 just forming into a UDG and KDG 013 already having been processed.

This is the formation mechanism that favors our GC abundance most, and that explains both the observed tidal features and the UDG nature of the galaxies without having to invoke additional UDG formation mechanisms in addition to tidal processing within the galaxy group.

While half of the group UDGs in the simulations of Jiang et al. (2019) and Liao et al. (2019) originate as dwarf galaxies where subsequent tidal effects transformed them into UDGs, the other half originate as pre-existing UDGs in the field or group periphery that then experience environmental tidal effects as they fall in. This would mean the tidal features are coincidental and unrelated to the diffuse nature of the UDGs.

Differentiating between a pre-existing UDG and a UDG resulting from group processing is non-trivial in these simulations. While we cannot completely rule out that KUG 0203-Dw1 and KDG 013 were pre-existing UDGs, our observations provide hints that this is less likely. The observation that KUG 0203-Dw1 narrowly meets the UDG criteria suggests that it might not have originally qualified as a UDG before interacting with its host. Likewise, we may expect pre-existing UDGs to become much more extended after interacting with a group, but KUG 0203-Dw1 and KDG 013 are both approximately average in size for UDGs in terms of half light radius (refer to Figure 6 in the following section). For example the pre-existing UDG explored in Liao et al. (2019) has a half light radius more than double that of the UDG formed by tidal heating.

Currently the lack of specific predictions for GC abundances in hydrodynamical simulations complicates distinguishing between UDG formation via tidal heating and pre-existing UDGs that suffered tidal heating. Kong et al. (2022) compared gas-rich field UDGs with hydrodynamical simulations, finding discrepancies between simulated and observed features. This implies that current galaxy formation models may lack accuracy for gas-rich UDGs. In addition, the field UDGs later accreted in the simulations by Jiang et al. (2019) and Liao et al. (2019) differ in their formation mechanisms (star formation feedback versus high spin halos). Thus a more comprehensive understanding of group UDGs is warranted before conclusive arguments for the processing of pre-existing UDGs can be made.

We contend that the stellar populations in both KUG 0203-Dw1 and KDG 013 are distinctly influenced by their group environments. We favor the tidally “puffed-up” origin as it explains both the tidal features and the UDG nature of these galaxies without having to employ a separate UDG formation mechanism in addition to group processing. However, presently, origins as pre-existing UDGs cannot be conclusively dismissed. Perhaps velocity measurements, akin to those conducted for UDGs in the Coma cluster by Alabi et al. (2018), are necessary for such distinction, especially if current hydrodynamical simulations prove accurate.

In addition to the two UDGs presented here, we also have previously reported an additional 3 UDGs that were identified and observed in an identical manner, constituting a full sample of 5 UDGs. These UDGs were selected such that they are a) in group-like environments within proximity to a more massive ‘host’ galaxy and b) have telltale signs of tidal disturbances through elongated or amorphous morphology, tidal tails or plumes, and/or stellar streams. These are all of the UDGs in the wide portion of the CFHT Legacy survey covering 150 deg² that have been identified to contain clear evidence of interaction given their morphology and/or the presence of stellar streams. Of this set of five, two of the UDGs - NGC 2708-Dw1 and NGC 5631-Dw1 - contain little evidence of recent star formation or neutral gas and exhibit slightly redder optical colors. The other three - UGC 9050-Dw1, KUG 0203-Dw1, and KDG 013 - in contrast have GALEX UV detections which is suggestive of recent star formation ($<100$ Myr) and also exhibit bluer optical colors (although the optical color errors are large). We briefly summarize the previously reported results of the first three UDGs within our sample below before comparing the set of five. The properties of these UDGs are also presented in Table 3. This is the first consistently identified and measured sample of UDGs exhibiting tidal features.

NGC 2708-Dw1 is an elongated UDG companion to NGC 2708 (of the NGC 2698 group), with a faint linear stellar stream connecting the two. J21 identified $N_{\rm{GC}}=3^{+1}_{-2}$ and $S_{N}=3.3^{+5.6}_{-1.0}$ for NGC 2708-Dw1. Additionally there is no trace of H i in the UDG itself or of a H i tidal connection to the host NGC 2708, indicating that the stellar stream originated from a gas poor object (likely the UDG itself) and not the disk of NGC 2708.

NGC 5631-Dw1 is more notably elongated, connected to NGC 5631 with a faint curved stellar stream. J21 finds a slightly higher GC abundance, with $N_{\rm{GC}}=5^{+1}_{-2}$ and a much higher specific frequency of $S_{N}=54^{+49}_{-33}$. There is a plausible H i detection coincident with the UDG and connected to the disturbed H i content of NGC 5631. However, because the H i tails are significantly larger than the stellar tails, they may not be physically connected nor associated with the UDG.

UGC 9050-Dw1 is a highly amorphous UDG with a distinct tidal tail, in the proximity of UGC 9050, but with no clear stellar or H i streams between the two. It contains a brighter “core” region with overlapping UV emission while the remainder of the galaxy consists of very low surface brightness and redder stellar material. F23 reports an extraordinary GC population of $N_{\rm{GC}}=52\pm 12$ and $S_{N}=122\pm 38$. The colors of the GCs exhibit a narrow spread, indicating that the vast majority of them likely formed in a narrow window in time. This UDG has a clear H i detection, with an estimated mass of $\log{M_{\rm{H}\,\rm{\textsc{i}}\ }/M_{\odot}}=8.44\pm 0.04$.

Physical Properties The five UDGs roughly span the physical parameter space with diverse luminosities and stellar/H i masses. We illustrate the size, luminosity and central surface brightness in Figure 6, comparing to UDG samples from the Coma cluster (van Dokkum et al., 2015), group environments (Somalwar et al., 2020), and H i-bearing UDGs (Leisman et al., 2017). UGC 9050-Dw1 is a clear outlier in the size-luminosity relation ($\sim 6$kpc in radius, though this is not the half-light radius because of the complex morphology of the UDG), with the remaining four UDGs consistent with the Coma cluster and group UDGs. NGC 2708-Dw1, NGC 5631-Dw1, KUG 0203-Dw1, and KDG 013 have comparable dimensions, with half-light radii ranging from 1.5 to 3.4 kpc, although KUG 0203-Dw1 just meets the UDG criteria. In terms of central surface brightness-half light radius parameter space, UGC 9050-Dw1 somewhat resembles the H i UDG population (although still rather large) while NGC 5631-Dw1 has a very faint central surface brightness. However, previous studies limited by surface brightness may not have thoroughly explored this parameter space, rather than NGC 5631-Dw1 being a significant outlier. We also note that these other samples tend to assume $n=1$ for the Sérsic index, which complicates this comparison.

Optical Morphology Our sample exhibits a variety in its optical morphology, not atypical for dwarf galaxies. NGC 2708-Dw1, NGC 5631-Dw1, and KUG 0203-Dw1 have associated stellar streams connecting to a massive host. UGC 9050-Dw1 has a very clear tail feature extending to the north of the UDG with more subtle material extending southward, and KUG 0203-Dw1 and KDG 013 each have a tidal plume extended towards the southwest.

Stellar Populations While we cannot make detailed assessments of the stellar populations within the UDGs, we can comment on the ages of the stellar population given the presence of UV emission. The UV emission from short-lived, high-luminosity stars, with lifetimes $<100$ Myr, peaks at ultraviolet wavelengths, highlighting recent star formation activity (Salim, 2014; Tuttle & Tonnesen, 2020). KUG 0203-Dw1, UGC 9050-Dw1, and KDG 013 all have UV emission detected by GALEX (in order of SFR). Thus we expect that these UDGs have younger stars than NGC 5631-Dw1 or NGC 2708-Dw1. Notably only the very central region of UGC 9050-Dw1 contains UV emission, meaning that the center of the galaxy may have a younger stellar population that the outskirts.

Gas Content Only two of the UDGs have clear neutral gas detections: KUG 0203-Dw1 ($\sim 10^{7}~\rm{M}_{\odot}$) and UGC 9050-Dw1 ($2.75\times 10^{8}~\rm{M}_{\odot}$). There are detections coincident with NGC 5631-Dw1 and KDG 013 but because there is no statistically significant concentrations of H i at the location of the galaxies, it is not clear that the H i is associated with them.

The morphology of the H i of KUG 0203-Dw1 shows both a connection between the UDG and the host, in addition to a spur to the southeast of the UDG and a trailing/leading tail detection to the northeast of the host. This suggests that the system initially contained gas, which is currently undergoing tidal stripping by the host galaxy, rather than the UDG stripping gas from the host. The latter would not lead to a pronounced concentration of H i centered on the UDG. In contrast, while the H i in UGC 9050-Dw1 is also distorted, it is not apparent if this is due to its neighbor UGC 9050. UGC 9050 does have an H i spur extended in the direction of UGC 9050-Dw1, possibly indicative of a prior interaction.

GC Populations In comparison to the two UV bereft UDGs included in our sample (NGC 2708-Dw1 and NGC 5631-Dw1 from J21), our sample of UV bright UDGs KUG 0203-Dw1 and KDG 013 contain similar but slightly more abundant GC populations (see Figure 5). All four have abundances comparable to normal galaxies at their luminosity (see Section 5.2), although KUG 0203-Dw1 and KDG 013 are at the upper end. In general, cluster UDGs lie above group UDGs in parameter space (not unexpected based on theoretical work, e.g., Carleton et al. 2021) with more elevated GC abundances. However, group UDGs seem to display a rather wide variety in GC abundances overall, especially given that of UGC 9050-Dw1 and the more GC abundance UDGs identified in Somalwar et al. (2020) and the NGC 1052 group (Shen et al., 2021; van Dokkum et al., 2022).

We plot the colors of the GCs of our sample in Figure 7. Note that the histograms have been normalized to have an integrated area of 1 due to the high GC abundance of UGC 9050-Dw1. Three of the UDGs in our sample have GC colors exhibiting a broad spread within the allowed $(V-I)$ color, as expected for the general UDG and dwarf galaxy population (see e.g., Sharina et al., 2005; Georgiev et al., 2009). Results in F23 show that a vast majority of the GCs identified in UGC 9050-Dw1 exhibit a narrow color spread ($\sigma_{(V-I)}$ $0.08$ mag) and are blue, with a very small redder GC candidate population consistent with contaminants. KUG 0203-Dw1 also appears to have a GC population with a relatively uniform and blue color, with mean $(V-I)=0.83$ mag, median $(V-I)=0.85$ mag, and $\sigma_{(V-I)}$ $0.08$ mag. Given that the field KUG 0203-Dw1 resides in has a small number of contaminants it is not likely that any of the GC candidates are foreground stars or background galaxies. However, because the GC abundance is small in KUG 0203-Dw1, this could also be a chance effect and we hesitate to draw significant conclusions based on the GC colors here. In the literature the distinction between UDGs with uniform and non-uniform GC colors persists regardless of tidal influences (possibly tidal: NGC 1052-DF2, NGC 1052-DF4 see van Dokkum et al. 2022; not tidal: NGC 5846-UDG1, DGSAT 1 see Müller et al. 2021; Janssens et al. 2022 respectively).

Environment All of the UDGs studied in this work reside in group environments. Each of the groups have a confirmed membership of $n\gtrsim 4$ except for UGC 9050-Dw1, which currently has only two known members (note that the UGC 9050 pair lies at the fringe of a larger group, although association is unclear). This is also the only group where the host galaxy has a similar mass (only one order of magnitude different in stellar mass and half an order of magnitude different in H i) to the UDG. This environment may be connected to the more unusual properties exhibited by the UDG. KUG 0203-Dw1 and KDG 013, though smaller, exhibit sizes comparable to the other members of their respective systems. In contrast, NGC 2708-Dw1 and NGC 5631-Dw1 are significantly dwarfed by their companions. Possibly, the stronger influence of the host on these smaller systems led to an earlier period of gas stripping. This, in turn, resulted in a smaller GC abundance and a lack of UV emission.

Plausible Formation mechanisms J21 concludes that NGC 2708-Dw1 and NGC 5631-Dw1 are most consistent with a tidally “puffed up” dwarf formation origin, given that they have GC abundances comparable to typical dwarf galaxies and both have stellar streams connecting to their hosts. We draw similar conclusions for the UDGs studied in this paper. While we cannot conclusively dismiss that any of these UDGs were UDGs prior to interacting with their group environments, the GC abundances, half-light radii, and other properties favor a group processing origin. Additionally, this formation scenario explains both the UDG nature of these objects and the tidal features, without invoking multiple formation mechanisms simultaneously. In contrast, F23 concludes that UGC 9050-Dw1 is most consistent with a dwarf major merger formation mechanism given its large size, abundant and relatively uniform in color GC population, and other unusual characteristics. These varying findings demonstrate that multiple formation mechanisms for UDGs in group environments are viable, highlighting that UDGs form in different ways.