The Origin of Faint Tidal Features around Galaxies in the RESOLVE Survey

Our main goal is to detect faint tidal features around galaxies in the RESOLVE survey, a volume-limited census of stellar, gas, and dynamical mass encompassing more than 50,000 cubic Mpc of the nearby universe (Kannappan & Wei 2008). A more thorough description of the survey design will be given in S.J. Kannappan et al. (in preparation), but we briefly outline the key aspects of the survey below.

2.1.1. Survey Definition

2.1.2. Custom Photometry and Stellar Masses

2.1.3. H i Masses

2.1.4. Environment Metrics

This project utilizes r-band images from both Data Release 3 of the DECam Legacy Survey⁹ (DECaLS; Blum et al. 2016) and the IAC Stripe 82 Legacy Project (Fliri & Trujillo 2016). The DECaLS survey covers both RESOLVE subvolumes, with deep r-band image stacks of variable depths with a median 5σ point source depth of 23.6 mag. Assuming Poisson noise, this depth yields a median surface brightness limit of ∼27.9 mag arcsec⁻² for a 3σ detection of a 100 arcsec² low surface brightness feature (Schlegel et al. 2015), allowing us to probe moderately faint tidal features around RESOLVE galaxies. However, scattered light and other image artifacts may contribute to functionally shallower surface brightness limits. DECaLS also provides depth maps for each stacked image in units of the canonical galaxy flux inverse-variance for each pixel, where the canonical galaxy is defined as an exponential profile with effective radius 045 (much smaller than our features of interest). For the specific DECaLS images used in this study, the distribution of 5σ depths per pixel has a standard deviation of 0.6 mag around a median of 23.5 mag (similar to the point-source limit due to the small size of the "canonical" galaxy).

In contrast, the IAC Stripe 82 Legacy Project provides deep r-band co-adds for only RESOLVE-B, but reaches a surface brightness limit of 28.3 mag arcsec⁻² for a 3σ detection of a 100 arcsec² feature (Fliri & Trujillo 2016). See Figure 1 for a side-by-side comparison of the galaxy rf0358 in SDSS, DECaLS, and IAC. Though our analysis mainly focuses on results from the DECaLS images, the IAC co-adds allow for study of the dependence of our classifications on the surface brightness limits of our data.

Zoom In Zoom Out Reset image size

Although automating the detection of faint tidal features is an appealing goal, no such current methods can surpass visual inspection for detecting faint and subtle substructures as well as avoiding false detections (see Adams et al. 2012). While human detection is the main driver of this project, various software packages were used in order to facilitate and improve the identification process. First, cutouts of 512 × 512 pixels (134 × 134 arcsec) in the r-band were extracted of each galaxy from DECaLS DR3 image stacks. Due to an imperfect overlap of the RESOLVE-A sample and the DECaLS r-band coverage, an initial 331 out of 1501 galaxies were not inspected with the DECaLS r-band images. In addition, another 122 galaxies from both semesters were later removed from the detection sample during visual inspection due to various problems such as only part of the galaxy falling in the image, bright image defects, or fringing that would prevent any confident identification of tidal features. Thus, at the beginning of the detection process the sample of galaxies decreased from 1501 to 1048.

To improve detection of particularly faint features, each galaxy thumbnail was run through a modification of a previously developed masking and smoothing code (I. Dell'Antonio). For each image, SExtractor (Bertin & Arnouts 1996) was used to detect sources within the image, and the resulting catalog was used to create binary masks. These masks were then applied to the original image using IRAF, nulling out these pixel locations and allowing for only faint and small-scale features to remain nonzero. The masked thumbnail was then smoothed with a Gaussian filter with a sigma of 7 pixels (2 arcsec), chosen to enhance these features without smoothing them out entirely. One such final image is shown in Figure 2, with the faint tidal stream coming off the smaller galaxy on the left much clearer in the smoothed image. However, while this masking and smoothing process greatly enhanced some of the features, some other substructure was possibly masked out in this process as confirmed in the test case shown in Figure 3. As a result, both the original and the smoothed masked images needed to be inspected for faint tidal features.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

For the visual inspection, a code previously used for by-eye morphological classifications of RESOLVE-A galaxies (see Moffett et al. 2015) was adapted with the APLpy Python package (Robitaille & Bressert 2012) to display images of galaxies as well as images with overlaid contours revealing larger-scale galactic structure. In addition, the masked and smoothed image of each galaxy, as well as the original, were displayed in a DS9 (Joye & Mandel 2003) window to allow for interactive adjustment of the contrast and scaling for each thumbnail. For each of the galaxies in the identification sample, a single classifier (Hood) recorded a flag after visual inspection indicating a positive or negative detection of faint tidal features, as well as any comments on particularly interesting substructures. A similar process was used to classify tidal features around RESOLVE galaxies in the IAC images for comparison with the DECaLS results (see Section 3.1.1).

All DECaLS images of our sample were then reclassified according to the detection classes defined by Atkinson et al. (2013), hereafter A13, in order to aid comparison between the two studies. Each galaxy was put into a category from 0−4, with 4 denoting the "certain" detection of a tidal feature, 3 denoting a "probable" (∼75% certain) detection, 2 denoting a "possible" (∼50% certain) detection, 1 denoting a "hint" (∼25% certain) of a detection, and 0 denoting no evidence for tidal features. In addition, each galaxy in classes 1−4 was also assigned a flag to indicate the type of tidal feature seen. We categorized our features as either "narrow" or "broad." Our "narrow" category would encompass the "streams," "arms," and "linear features" categories from A13, while "broad" includes their "shells," "fans," and "miscellaneous diffuse structure" labels.

The results of our tidal feature classifications are summarized in Table 1. We find that 17^±2% of the 1048 RESOLVE galaxies inspected with DECaLS have tidal features detected with high confidence (classes 3 and 4). The uncertainty given is the range of possible percentages within a two-sided 1σ confidence interval given small number binomial statistics. As seen in Figure 4, most of the features detected in all classes are narrow; however, some broader faint features are detected as well. Though possibly visually indistinguishable, our sample presumably includes both tidal tails extending from the primary galaxy as well as tidal streams indicating a companion being stripped. We did not attempt to distinguish between different types of tidal features within our two categories, but future simulations of interactions causing these features would be useful to determine criteria for this level of sub-classification. A selection of features from each category found around RESOLVE galaxies is shown in the Appendix. For simplicity, we will refer to galaxies with confident detections of tidal features as "TF" galaxies, while the rest (classes 0−2) will be "NTF" galaxies. However, we will also discuss the effect on our results of treating class 2 galaxies as "TF" rather than "NTF" galaxies.

Zoom In Zoom Out Reset image size

Table 1.  Tidal Feature Detections

Confidence	Number	Percentage
4	46	${4.4}_{-1.2}^{+1.4}$
3	134	${12.8}_{-2.0}^{+2.2}$
2	65	${6.2}_{-1.4}^{+1.6}$
1	59	${5.6}_{-1.3}^{+1.6}$
0	744	${71.0}_{-2.9}^{+2.7}$

Download table as:  ASCIITypeset image

Although streams and other substructures are predicted to be common in the current cosmological framework, they are not easy to detect. Depending on many factors, such as the relative masses of the interacting galaxies or the geometry of the interaction, a majority of substructure resulting from accretions can be expected to have surface brightnesses of 30 mag arcsec⁻² or fainter (Bullock & Johnston 2005). This range is much lower than the limits of most imaging surveys, which in turn limits our ability to explore the statistical properties of these features. With the surface brightness limits of our DECaLS images being ∼27.9 mag arcsec⁻², we are able to probe the brightest of these features, but ultimately only put a lower limit on their frequency. In addition, the lifetime of a tidal feature resulting from a merger depends heavily on its surface brightness limit. Ji et al. (2014) find that the major-merger feature lifetime goes up by a factor of two for a surface brightness limit of 28 mag arcsec⁻² when compared to a shallower limit of 25 mag arcsec⁻². Unfortunately, their models also provide evidence that tidal features from relatively minor mergers (mass ratios less than about 1/6) may be difficult to detect even in surveys that reach below 28 mag arcsec⁻².

3.1.1. Comparison to Classifications with IAC Stripe 82 Co-adds

3.1.2. Noise Degradation Experiment

In order to further test the dependence of our classifications on the surface brightness limit of our data, we have artificially degraded the IAC images to the same surface brightness limit as our DECaLS cutouts. When we re-classify these noisy IAC images, we find that ∼18^±2% of the inspected galaxies host tidal features. We find that 21 of the TF galaxies in the DECaLS images are not similarly classified in the degraded IAC images; conversely, nine of the TF galaxies in the degraded images are not flagged with the DECaLS images. The exact source of these discrepancies is unclear; some of these galaxies have barely discernible tidal features whose classification may be unreliable. We quantify the differences in these classifications with Cohen's kappa statistic, which is traditionally used to measure inter-classifier agreement while taking into account the probability of classifiers agreeing by chance. Cohen's kappa is calculated according to the following formula, where p_o is the relative observed agreement between classifiers and p_e is the hypothetical probability of chance agreement:

$$\begin{eqnarray}&&\kappa \equiv \displaystyle \frac{{p}_{o}-{p}_{e}}{1-{p}_{e}}.\end{eqnarray} \tag{ 1 }$$

Our DECaLS and noisy IAC classifications have a Cohen's kappa value of 0.79 on a scale from −1 to 1, where 1 indicates perfect agreement. According to the hierarchy established by Landis & Koch (1977), our value corresponds to a "substantial" strength of agreement. As in Section 3.1.1, we do not find any statistically significant differences between the distributions of various parameters (stellar mass, color, etc.) for the TF galaxies from the two imaging sources. Therefore, any marginally detectable tidal features or false detections should have a minimal effect on the results below.

We have examined a multitude of galaxy properties in the RESOLVE catalog to compare their distributions for TF and NTF galaxies. A table containing our classifications and other information required to reproduce the analysis below (in combination with RESOLVE Data Releases 1 and 2; see online database¹⁰ ) is available in machine readable format in the supplementary data of this paper. Columns 4–7 are provided for RESOLVE galaxies not visually inspected in this work for completion. A summary of information included in the catalog is given in Table 2.

The distributions of G/S for TF and NTF galaxies show the most statistically significant difference between the two samples (Figure 5). A K-S test comparing the G/S distributions in Figure 5 yields p_KS ∼10⁻⁹, meaning the gas fraction distributions of TF and NTF galaxies have a negligible probability of coming from the same parent distribution at ∼6.0σ significance. We can see from the figure that TF galaxies in our sample tend to have higher gas fractions. We see the same result if including "possible" detections of tidal features (those in class 2) in the TF sample, at a higher significance of ∼7.0σ. Lotz et al. (2010) found in simulations that galaxy mergers with high gas fractions exhibit disturbed morphologies for longer periods of time than their gas-poor counterparts. Assuming the high surface brightness asymmetries studies by Lotz et al. are accompanied by low surface brightness features, it is not necessarily surprising that we tend to find more tidal features in gas-rich galaxies (but see Section 4.2). Since our strong upper limits cluster near M_gas < 0.05–0.1M_* as a matter of observational strategy, both galaxies with and without tidal features exhibit a false bimodality in gas fraction, with the upper limits causing a second peak slightly below the M_gas ∼ 0.1M_*, or log(G/S) = −1, threshold.

Zoom In Zoom Out Reset image size

Other galaxy properties that produced slightly weaker results were u − r color, SSFR, and the morphology metric μ_Δ introduced in Kannappan et al. (2013) to distinguish quasi-bulgeless (μ_Δ < 8.6), bulged disk (8.6 < μ_Δ < 9.5), and spheroid-dominated (μ_Δ > 9.5) galaxies. This list is somewhat unsurprising due to the tight correlation of colors and star formation histories with G/S, and to a lesser degree with galaxy structure (Kannappan 2004; Franx et al. 2008; Kannappan et al. 2013; Lilly & Carollo 2016; Saintonge et al. 2016). Additional parameters (nearest neighbor distance, stellar mass, group halo mass, axial ratio b/a, g − r color gradient, group virial radius, axial ratio of the inner disk region of the galaxy, H i asymmetry, group redshift, number of group members, and central status) inspected for correlations with tidal features did not yield statistically significant results, although we will see that some of these prove important when investigated by means other than direct global correlation.

Because of the possibility that gas-rich and gas-poor galaxies with tidal features might reflect different origins, we decided to look for parameters that might be important despite not showing a direct correspondence with tidal features, for example due to differing trends from distinct formation mechanisms in the gas-rich and gas-poor subpopulations. We applied an automated method to determine which parameters are most important for predicting tidal features. The Random Forest algorithm (Breiman 2001), implemented in this work with the public Python library scikit-learn (Pedregosa et al. 2011), generates classifications based on a random ensemble of decision trees based on all provided input parameters. From this process, the algorithm also determines the score of the importance of each input feature in the resulting classifications. We trained the algorithm using our by-eye identification of tidal features discussed in Section 2.3. We judged the performance of our classifier with the F1 metric, which is the harmonic mean of precision (the number of correct positive results divided by the number of all positive results) and recall (the number of correct positive results divided by the number of positive results that should have been returned). An F1 score reaches its best value at 1 and worst at 0. The best hyperparameters for our classifier, such as the number of generated decision trees, were optimized to maximize this F1 metric with ten-fold cross validation (Geisser 1975).

The relative contribution of each parameter to our Random Forest classification is shown in Figure 6. Our best-performing classifier only achieved an F1 score of 0.5; however, from the feature selection results of Figure 6, we were able to identify additional parameters to be further investigated. The top five most important parameters in descending order are as follows: the atomic G/S, the SSFR, the morphology metric μ_Δ, the distance to the galaxy's nearest neighbor in RESOLVE, and the stellar mass of the galaxy. Below we analyze these parameters in relation to our TF and NTF samples. Although group environment metrics (i.e., halo mass and number of group members) are deemed relatively unimportant by the Random Forest results, we analyze them as well to put our results in interpretive context.

Zoom In Zoom Out Reset image size

The strong correlation of increased G/S with tidal features indicates that separating our samples into gas-poor and gas-rich galaxies may yield differing results in relation to the parameters identified in Section 3.3. We make this divide at G/S = 0.1 in order to include all strong upper limits in the gas-poor population, as marked by the dashed line in Figure 5. In our resulting gas-rich sample, 142 out of 747 galaxies (∼19^±2%) have tidal features, ∼79% of which are classified as "narrow." In contrast, 38 out of 292 gas-poor galaxies (∼13^±3%) have tidal features, ∼74% of which are classified as "narrow." Below we analyze the roles of galaxy properties in relation to our gas-rich and gas-poor TF and NTF samples.

3.4.1. Star Formation

To investigate the possible distinct origins of tidal features around gas-rich and gas-poor galaxies, we consider the distributions of the specific star formation rates (SSFRs). A comparison of the SSFR versus G/S distributions for galaxies with and without tidal features is shown in Figure 7. We separate the gas-rich galaxies with H i masses (blue) from the gas-poor galaxies (red). As the SSFR is closely related to the galaxy's color, which is used to calibrate the photometric gas fractions technique we use for confused galaxies and galaxies with weak upper limits, we do not include these galaxies in this plot (i.e., we only plot the gas-rich galaxies with 21 cm detections). We have applied the Fasano & Franceschini (1987) variation of the Peacock (1983) test (hereafter referred to as the PFF test), an extension of the K-S test to two dimensions, to determine the statistical significance of the difference between the distributions of the TF and NTF galaxies for the gas-rich sample. Within this sample, galaxies with tidal features appear to be have both larger G/S (as seen in Section 3.3) as well as higher SSFR at ∼5.5σ significance. This trend remains the same when including "possible" detections of tidal features (those in class 2) in the TF sample, with a slightly higher significance of ∼6.1σ.

Zoom In Zoom Out Reset image size

However, we want to test whether this difference is driven mostly by the correlation between G/S and SSFR, or if there is an additional tendency of gas-rich TF galaxies to have higher SSFRs independent of G/S. We fit a linear relationship to the total SSFR versus G/S sample (shown in red) for gas-rich galaxies with 21 cm detections in the region in which the vertical scatter from the line appeared symmetrical (marked by the dashed lines). The SSFR residuals from this line are shown in Figure 8; we do not find a statistically significant difference between the distributions of residuals for TF and NTF galaxies, which also holds if class 2 confidence TF detections are counted as TF rather than NTF galaxies. Thus, we conclude the tendency of gas-rich TF galaxies to have higher SSFRs seen in Figure 7 is indeed driven primarily by G/S.

Zoom In Zoom Out Reset image size

We have similarly compared the SSFR distributions for TF and NTF galaxies in the gas-poor sample as shown in Figure 9. We do not find a statistically significant difference between the two distributions, which also holds if class 2 confidence TF detections are counted as TF rather than NTF galaxies.

Zoom In Zoom Out Reset image size

3.4.2. Morphology

To examine the relationship of morphology to our TF and NTF samples, we use the structure metric μ_Δ introduced in Kannappan et al. (2013) to distinguish quasi-bulgeless (μ_Δ < 8.6), bulged disk (8.6 < μ_Δ < 9.5), and spheroid-dominated (μ_Δ > 9.5) galaxies. Kannappan et al. define μ_Δ as

$$\begin{eqnarray}&&{\mu }_{{\rm{\Delta }}}={\mu }_{90}+1.7{\rm{\Delta }}\mu \end{eqnarray} \tag{ 2 }$$

combining an overall surface mass density

$$\begin{eqnarray}&&{\mu }_{90}=\mathrm{log}\displaystyle \frac{0.9{M}_{* }}{\pi {r}_{90,r}^{2}}\end{eqnarray} \tag{ 3 }$$

with a surface mass density contrast

$$\begin{eqnarray}&&{\rm{\Delta }}\mu =\mathrm{log}\displaystyle \frac{0.5{M}_{* }}{\pi {r}_{50,r}^{2}}-\mathrm{log}\displaystyle \frac{0.4{M}_{* }}{\pi {r}_{90,r}^{2}-\pi {r}_{50,r}^{2}}\end{eqnarray} \tag{ 4 }$$

where all radii are converted to physical kpc units.

A comparison of the μ_Δ versus G/S distributions for TF and NTF galaxies is shown in Figure 10. Similarly to Section 3.4.1, we separate the gas-rich galaxies (blue) from the gas-poor galaxies (red), although here we include confused galaxies and galaxies with weak upper limits for which G/S has been calculated with the photometric gas fractions technique. We calculate the significance of the difference between the distributions for gas-rich TF and NTF galaxies to be ∼5.8σ, with TF galaxies tending to be more disky than their NTF counterparts. We see a similar trend when including "possible" detections of tidal features (those in class 2) in the TF sample, but with a slightly higher significance of ∼6.2σ. The distributions of μ_Δ residuals from the linear fit to the μ_Δ versus G/S data (shown in red in Figure 10) are plotted in Figure 11. As for SSFR, we do not find a statistically significant difference between the distributions of μ_Δ residuals for gas-rich TF and NTF galaxies, regardless of whether we include class 2 confidence detections in the TF sample. Thus, we conclude the tendency of gas-rich TF galaxies to have diskier morphologies seen in Figure 10 is driven primarily by G/S.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

We also compared the μ_Δ distributions for gas-poor TF and NTF galaxies as shown in Figure 12. The TF galaxies in this sample do seem to be slightly concentrated at higher μ_Δ (more bulged morphologies) than their NTF counterparts, but the significance of the difference is only ∼2.0σ. We see similar trends if we include class 2 confidence TF detections as TF galaxies rather than NTF galaxies with a higher significance of ∼2.7σ.

Zoom In Zoom Out Reset image size

3.4.3. Stellar Mass

The stellar mass distributions for our gas-rich and gas-poor galaxies are shown in Figures 13 and 14, respectively. We do not find a statistically significant difference between the stellar mass distributions of gas-rich TF and NTF galaxies, regardless of whether we include "possible" detections (class 2) in the TF sample. In contrast, gas-poor galaxies show a much more prominent relationship between tidal features and stellar mass. Gas-poor galaxies with tidal features are more massive than their NTF counterparts at ∼5.2σ significance. The same trend is seen (at a slightly lower significance of ∼4.9σ) when counting class 2 confidence TF detections as TF rather than NTF galaxies.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

To analyze the environments of our galaxies, we consider both nearest neighbor distance and group mass and richness to assess environmental influences on local and halo scales.

As a statistical detail, in this section the Mann–Whitney–U (MWU) test has been applied to each pair of distributions in addition to the K-S test. Both the MWU and K-S tests return p-values that test the null hypothesis that the two samples have the same distribution. One difference is that the MWU test is mainly sensitive to discrepancies between the two medians while the K-S test is more sensitive to general differences between the two distributions (shape, spread, median, etc.). Another difference is that the K-S test is not suited for samples with repeated values. Since we expect pairs of duplicate nearest neighbor distances, as well as multiple groups of galaxies with the same group parameters, the MWU test may provide a more accurate p-value.

3.5.1. Nearest Neighbor Distance

For each galaxy in RESOLVE, we have calculated the projected distance to its nearest neighbor using the kd-tree algorithm described in Section 2.1.4 and suppressing peculiar velocities within groups (i.e., assuming a single group cz for galaxies in the same group). We compare the distributions of projected distances for TF and NTF galaxies within our gas-rich and gas-poor subsamples in Figures 15(a) and 16(a), respectively. In panels (b) and (c), we further divide each subsample based on whether the nearest neighbor is a member of the same group (it may not be if the galaxy is isolated, i.e., in an N = 1 group).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

For gas-rich galaxies with neighbors in the same group, those with/without tidal features have median separation of 0.06/0.1 Mpc; the difference between the two gas-rich distributions is significant at ∼2.9σ. For gas-poor galaxies with neighbors in the same group, those with/without tidal features have median separation of 0.03/0.09 Mpc; the difference between the two gas-poor distributions is significant at ∼4.5σ (a notably stronger result than for gas-rich galaxies). However, regardless of gas content, galaxies with nearest neighbors outside of their own group do not show a correlation between the presence of tidal features and nearest neighbor distance. We obtain results similar to these kd-tree results if we instead calculate nearest neighbor distances as projected distances to the nearest neighbor in a cylindrical volume within cz = 500 km s⁻¹ of the main object.

Thus, regardless of gas content, we find that galaxies with neighbors in the same group are closer to their nearest neighbors if they have tidal features than if they do not, with a stronger result observed for gas-poor galaxies than gas-rich ones. This result persists if we include "possible" detections of tidal features (those in class 2) in the TF sample, though at slightly higher significances of ∼3.2σ and ∼4.7σ for gas-rich and gas-poor galaxies, respectively. We infer that a substantial fraction of detected tidal features are the result of ongoing interactions, or early-stage mergers, especially in gas-poor galaxies.

3.5.2. Group Mass and Richness

Although group halo mass and number of group members appeared to be relatively insignificant indicators of tidal features according to the Random Forest results in Section 3.3, the relationship between tidal features and nearest neighbor distance shown in Figure 16 implies an environmental dependence, especially for our gas-poor sample. Group halo mass distributions for our gas-rich and gas-poor subsamples are shown in Figures 17 and 18, respectively, with each subdivided into TF and NTF galaxies . Though we do not find a statistically significant difference between the halo masses of gas-rich TF and NTF galaxies, the halo masses of gas-poor TF galaxies are more massive than those of their NTF counterparts at ∼4.1σ significance. The same trend is seen (at a slightly lower significance of ∼3.7σ) when including less confident detections of tidal features (those in class 2) in the TF sample.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Moreover, we find a difference in the relation of group number versus group halo mass for gas-poor TF and NTF galaxies (Figure 19). A PFF test (Section 3.4.1) shows ∼3.6σ significance for the differences between the two distributions. We see the same trend (at a lower significance of ∼3.3σ) when including less confident detections of tidal features (those in class 2) in the gas-poor TF sample. This result may be consistent with gas-poor TF galaxies residing in groups with fewer members at fixed group halo mass than gas-poor NTF galaxies, which could point to an enhanced fraction of merger remnants among gas-poor galaxies with tidal features.¹¹ Merger remnants would remain in the same halo but have fewer companions than before the merger. These completed mergers would complement the galaxy interactions and ongoing mergers evidenced by the closer nearest neighbor distances for TF galaxies in Figure 16. Overall, gas-poor galaxies with tidal features do appear to live in different environments than those without.

Zoom In Zoom Out Reset image size

A similar analysis of gas-rich galaxies does not yield any significant differences for TF and NTF environments. We lack any compelling evidence for a difference in group halo masses or number of members between gas-rich galaxies that do and do not have tidal features.

As previously mentioned, past surveys of faint substructure do not agree on the frequency of these features; see Table 1 in A13 for an overview of results from published surveys. It is difficult to compare these surveys with each other and with the present work due to differences in surface brightness limits and sample selection. For example, Martínez-Delgado et al. (2010) find many tidal features around a small sample (∼8) of nearby galaxies but their sample was not intended to be statistical and deliberately included galaxies already known to have tidal features.

Both Tal et al. (2009) and van Dokkum (2005) find frequencies of tidal features factors of 2−3 times higher than we find in this paper, but they consider only early-type galaxies. If we limit our sample to galaxies with μ_Δ > 9.5 (bulged galaxies, as defined in Section 3.4.2), we find that ${18.0}_{-6}^{+7} \%$ have tidal features, which is still much lower than the rates reported in these studies. van Dokkum (2005) uses images with limiting surface brightnesses of 29.5 mag arcsec ⁻² in the μ band and may therefore be able to see many faint features that are not recoverable in the DECaLS images. Furthermore, the luminosity cut on their sample disfavors low-mass, early-type galaxies, which may be less likely to host tidal features as suggested by Figure 14. Similarly, although Tal et al. (2009) use slightly shallower images than in the present work (V_vega < 27 mag arcsec⁻²), like van Dokkum et al. they consider only luminous ellipticals so their focus is on relatively massive systems, which we have found to be more likely to host tidal features (Figure 14.)

Duc et al. (2015) look for tidal features around early-type galaxies with K-band magnitudes brighter than −21.5 (i.e., stellar masses above ∼10^9.8 M_☉) in 35% of the ATLAS^3D volume-limited sample. They find that 66% of their sample galaxies have streams, tails, or shells. Their high tidal feature fraction could additionally be due to differences in image depth, as their images reach a limiting surface brightness of 28.5 mag arcsec⁻² in the g band, as well as a bias toward higher mass galaxies from the magnitude cut. Additionally, Duc et al. (2015) inspect residual images obtained by subtracting model galaxies as well as the original thumbnail images, which may reveal more substructure than the method presented here.

In contrast with the previous three studies, Miskolczi et al. (2011) examine face-on disk galaxies (>2' in diameter) and find strong evidence for stream-like features around ∼6% of their sample and less distinct features around a total of ∼19% of their sample. They detect individual features down to 28 mag arcsec⁻², slightly deeper than the limit of our own images. Their rate of tidal features is similar to our own, but the authors do not include broader features such as shells in their detections. If we restrict our sample to disk galaxies (μ_Δ < 9.5) and only include "narrow" features, we find that 13^±2% show tidal features. Their slightly higher detection rate could be due to a difference in detection strength; in fact, if we include galaxies with a confidence level of 2, our fraction of disk galaxies with narrow tidal features rises to ${17}_{-2}^{+3} \%$, in agreement with the fraction from Miskolczi et al. (2011).

We can also compare our results with theoretical predictions for spiral galaxies from Byrd & Howard (1992). They find that a galaxy in their studied sample has a 80%–90% chance of having tidal arms from close passages of small companions, taking into account companion decay time and arm lifetime. However, they do not take feature visibility into account, so even though we only find 13^±2% of the disk galaxies in our sample have "narrow" tidal features, this rate can be taken as a lower limit still possibly consistent with the much higher occurrence rate calculated by Byrd & Howard. See Section 4.2 for a more thorough discussion on the effect of feature observability on our results.

The best comparison to the present work is possibly provided by A13. As previously mentioned, our confidence classes are based on the same scheme as used in A13. As shown in Table 1, we find many more features in the "probable" confidence class (3) than in the "certain" class (4), in contrast to A13, who label many more features as class 4 than 3; see their Figure 8. They were able to detect tidal features around 12% of their 1781 galaxy sample with the strictest confidence (4); the fraction increased to 17.6^±1% when including probable but slightly less convincing features (including class 3). Atkinson et al. measure a mean limiting surface brightness of 27.7 mag arcsec⁻² for their g'-band images, but possibly reach a depth similar to that of our DECaLS images with their stacked g', r', and i' images. Their more inclusive 17.6% rate agrees with our own fraction; however, our results differ when looking at subsamples of galaxies. We find that gas-rich galaxies are more likely to be tidally disturbed (Section 3.4), while they find that red galaxies, presumably gas-poor, are two times more likely to host tidal features than their blue counterparts (22.4^±1.5% versus 11.6^±1.7%).

To investigate the source of this discrepancy, we first tested for any systematic differences in classifications. We inspected 120 galaxies from the sample studied in A13 in stacked g', r', and i' images from the CHFTLS-Wide fields (30 galaxies from each field). Our classifications agreed in confidence exactly with those of Atkinson et al. for 91 galaxies, with most of the disagreements about the remaining 29 galaxies occurring between "probable" and "certain" features (classes 3 and 4). Converting into a binary TF and NTF classification as used in this paper (where TF includes galaxies with features in the 3 and 4 confidence classes), we compared our classifications with those in A13 and found a Cohen's kappa score of 0.899, indicating a "substantial strength of agreement" (Landis & Koch 1977).

In addition, we inspected 107 galaxies from the A13 sample in the r-band DECaLS DR3 coadds used to inspect RESOLVE galaxies in this work. Our classifications agreed in confidence for 85 of these galaxies. Again converting these confidences into TF/NTF flags, we find a Cohen's kappa score of 0.811 when comparing the two classifications. In particular, we flag three galaxies as TF that were not detected with much confidence by A13, all of which were blue. We may be slightly more likely to detect tidal features around blue galaxies than they are, although the effect is slight.

As with the other surveys discussed, differences in sample selection must be taken into account. Atkinson et al. study 1781 galaxies in set apparent magnitude (15.5 mag < r < 17 mag) and redshift (0.04 < z < 0.2) ranges; their sample thus prioritizes bright red systems like the other magnitude-limited samples discussed previously. In contrast, RESOLVE is a volume-limited survey and reaches r-band completeness limits of M_r < −17 for RESOLVE-B and M_r < −17.33 for RESOLVE-A; as a result, this work is naturally weighted toward smaller and more gas-rich galaxies. If we limit our sample in absolute magnitude near the faint-galaxy limit of A13, i.e., apply M_r < −19.3 mag, our sample decreases to only 370 galaxies. However, due to RESOLVE's volume-limited nature, this subsample still represents a generally fainter luminosity distribution than the luminosity-biased A13 sample, as shown in Figure 20. For this subset of brighter RESOLVE galaxies, we find that ${22}_{-4}^{+5} \%$ of these images show tidal features, which agrees with the A13 rate within uncertainties. Splitting this sample by gas content yields measured TF fractions of ${20}_{-6}^{+7} \%$ and ${23}_{-6}^{+6} \%$ for gas-poor and gas-rich galaxies, respectively. We converted the SDSS magnitudes of these galaxies into MegaCam filters using the equations from Gwyn (2008) and then applied the same color cut given in Equation (1) of A13. 98% of our gas-poor galaxies were labeled "red" as expected, but only 61% of our gas-rich galaxies were labeled "blue" under Atkinson et al.'s system. Using these color labels, we find that ${23}_{-5}^{+6} \%$ and ${19}_{-6}^{+8} \%$ of our red and blue galaxies, respectively, have tidal features when looking at RESOLVE galaxies with M_r < −19.3 mag. Thus, although our fraction of blue galaxies with tidal features is slightly higher, our results still agree with those in A13 within our uncertainties.

Zoom In Zoom Out Reset image size

Although we discussed the effects of surface brightness limitations in Section 3.1, other detection biases affect our ability to identify tidal features. Principally, the timescale over which a feature is detectable depends on the internal properties of the progenitor (Darg et al. 2010). Simulations of equal-mass gas-rich mergers show timescales depend strongly on geometric parameters like the pericentric distance and relative orientation of the galaxies (Lotz et al. 2008). In addition, gas fraction correlates strongly with tidal feature longevity; Lotz et al. (2010) found that asymmetry was detectable for ≤300 Myr for f_gas ∼ 20% or log(G/S) = −0.6 and to ≥1 Gyr for f_gas ∼ 50% or log(G/S) = 0.0. Thus, gas-poor galaxies we classified as lacking tidal features may have had recent mergers or interactions >300 Myr ago but appear relaxed. Conversely, the higher rate of tidal features in our gas-rich sample could be just due to this observational effect.

To test whether our results that gas-rich TF galaxies have higher G/S, increased SSFR, and more disky morphologies are mainly just a result of this relationship between G/S and TF observability, we have created mock TF and NTF samples from our inspected RESOLVE galaxies based on an estimate of the merger rate and visibility timescales as a function of gas fraction. We assume the minor merger rate at z = 0 for mass ratios of 1:30 and higher is 0.2 mergers per halo per Gyr (Fakhouri & Ma 2008). In addition, we use estimates of merger visibility timescales as a function of gas fraction from Lotz et al. (2010). Each stellar mass bin, ranging from log(M_*) = 8.5–11.5 in steps of 0.5 dex, is assigned a merger visibility timescale using the median gas fraction of our gas-rich galaxies in that bin. We limit our sample of galaxies in the same ways as in Sections 3.4.1 and 3.4.2 for comparison. Thus, in Figure 21, only gas-rich galaxies with 21 cm detections are used, while the whole gas-rich sample is shown in Figure 23. Using this timescale, we calculate the expected number of visible mergers in each stellar mass bin, then randomly assign RESOLVE galaxies in that bin to our TF sample accordingly. We repeat this process 5000 times, comparing the SSFR versus G/S and μ_Δ versus G/S distributions for each iteration and recording the p-value that the TF and NTF distributions come from the same parent distribution as in Sections 3.4.1 and 3.4.2. Histograms of these p_PFF-values for SSFR versus G/S and μ_Δ versus G/S are shown in Figures 22 and 24, respectively. The median 2D kernel density estimations of these distributions are shown in Figures 21 and 23, respectively. From these figures, one can see that the dependence of visibility on gas fraction probed by these mock samples can contribute to the trends we see for the real TF and NTF samples in Figures 7 and 10. However, the median p_PFF when comparing TF and NTF galaxies for these mock SSFR versus G/S and μ_Δ versus G/S distributions is ∼10⁻⁴ for each parameter, much larger than the p_pff ∼ 10⁻⁸ and ∼10⁻⁹ seen for our TF and NTF classifications. Thus, we conclude that there is higher G/S, along with higher SSFR and diskier morphologies, for our gas-rich TF galaxies when compared to the NTF sample, independent of the visibility bias toward more tidal features with higher gas fractions.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Due to the difference in observability discussed above, studying gas-rich and gas-poor galaxies separately helps restrict our analyses to galaxies with similar feature timescales which could correspond to different formation scenarios. In Section 3, we studied the relationship between the presence of tidal features and other progenitor properties such as morphology, SSFR, stellar mass, and environment. Here, we will argue that different types of tidal features may dominate in the two populations. The tidal features around our gas-poor galaxies are possibly connected to the cannibalism of small satellites as well as mergers at various stages. In contrast, while our gas-rich TF galaxies may also be in various stages of merging, we will argue that accretion of gas or a gas-rich satellite could also explain the features seen in this population.

To recap, only our gas-rich galaxies showed strong differences between TF and NTF galaxies when looking at the morphology metric μ_Δ as well as SSFR (Figures 7 and 10) and these trends are derivative of the even stronger difference in G/S (Figure 5), such that TF galaxies have higher G/S, diskier morphology, and higher SSFR. In contrast, only our gas-poor galaxies show significantly different stellar masses and group halo masses between the TF and NTF galaxies (Figures 14 and 18), with TF galaxies having higher masses and a possible tendency to reside in groups with fewer members at a fixed halo mass (Figure 19). Both gas-rich and gas-poor TF galaxies with nearest neighbors in the same group were closer to their neighbors than their NTF counterparts (Figure 15 and 16), although this result was stronger for gas-poor galaxies.

4.3.1. Interpretation of Tidal Features around Gas-poor Galaxies

4.3.2. Interpretation of Tidal Features around Gas-rich Galaxies