Merging Cluster Collaboration: Optical and Spectroscopic Survey of a Radio-selected Sample of 29 Merging Galaxy Clusters

The goal of the optical imaging survey is to obtain lensing-quality, wide-field imaging in at least two photometric bands. The two filters are chosen to straddle the 4000 Å break in order to select cluster members photometrically via red sequence relations. Furthermore, our weak lensing method makes use of these red sequence relations in order to select background galaxies for lensing studies (Jee et al. 2015, 2016; Golovich et al. 2017). Additionally, for clusters for whom our SuprimeCam observations came before our DEIMOS observations, we made use of the SuprimeCam images for spectroscopic target selection (see Section 3.3.1). Many clusters have archival imaging that we have obtained. We observed 18 systems with Subaru/SuprimeCam to complete the photometric survey.

The spectroscopic survey has a goal of obtaining ∼200 member galaxy velocities in each system. We used redshifts from the literature when available in order to reduce the amount of new observations required. When obtaining new spectra, we designed observations to also meet the goal of enabling studies of recent star formation and ultimately the link between mergers and star formation. We achieve this by adjusting the observed wavelength range for each cluster to the emitted wavelength range from Hβ to Hα for clusters with z ≲ 0.3 and [O ii] to [O iii] for clusters with z ≳ 0.3. The data available for star formation studies therefore varies from cluster to cluster, depending on the number of previously published redshifts and the redshift of the cluster. Additional observations were required for 18 systems, with many having no more than a handful of previously published member redshifts. In the following subsections we will detail the targeting, observing, and data reductions of our optical and spectroscopic surveys.

We observed 18 clusters over four nights using the 80 Megapixel SuprimeCam (Miyazaki et al. 2002) camera on the Subaru Telescope on Maunakea. Table 2 summarizes these observations. The basic strategy is to achieve weak-lensing quality in one filter and obtain a second filter to define the color of detected objects by straddling the 4000 Å break. For the lensing-quality image, the exposure time was 2880 s (8 × 360 s) and we rotated the field between each exposure by 15° in order to distribute the bleeding trails and diffraction spikes from bright stars azimuthally to be later removed by median-stacking. This scheme enabled us to maximize the number of detected galaxies, especially for background source galaxies for weak lensing near stellar halos or diffraction spikes. In the second and third filters (g and/or i), the exposure time was 720 s (4 × 180 s). These exposures were rotated by 30° from exposure to exposure for the same reason as above. In order to efficiently fill the time of each observing night, we added a third band to several clusters. The actual observing times may vary due to real-time changes to the observational plan due to unexpected lost time.

Archival Subaru/SuprimeCam imaging was downloaded from the SMOKA data archive (Baba et al. 2002), and is detailed in Table 3. We note that the observational strategy for the archival data did not prescribe rotating between exposures, so diffraction spikes and bleeding trails are present. Also, we did not make use of the full set of archival images for these clusters because we only required two bands of imaging in order to define the color and complete a color–magnitude selection. We utilized the deepest images available that satisfy this requirement, ensuring good seeing conditions.

3.2.1. Subaru/SuprimeCam: Data Reduction

The CCD processing (overscan subtraction, flat-fielding, bias correction, initial geometric distortion rectification, etc) was carried out with the SDFRED2 package (Ouchi et al. 2004). Much of the archival data required the first version of this pipeline (SDFRED1: Yagi et al. 2002). We refine the geometric distortion and World Coordinate System information using the SCAMP software (Bertin 2006). The Two Micron All Sky Survey (2MASS; Skrutskie et al. 2006) catalog was selected as a reference when the SCAMP software was run except for clusters covered by the Sloan Digital Sky Survey (SDSS; Adelman-McCarthy et al. 2007), for which the Data Release 5 catalogs were used. We also rely on SCAMP to calibrate out the sensitivity variations across different frames. For image stacking, we ran the SWARP software (Bertin et al. 2002) using the SCAMP result as input. We first created median mosaic images and then used it to mask out pixels (3σ outliers) in individual frames. These masked frames were weight-averaged to generate the final mosaic, which is used for the scientific analyses hereafter. Two example images are presented in Figure 2.

Zoom In Zoom Out Reset image size

3.2.2. Subaru/SuprimeCam: Photometric Catalog Generation

We conducted a spectroscopic survey utilizing the DEIMOS multi-object spectrograph (Faber et al. 2003) on the Keck II telescope at the W. M. Keck Observatory on Maunakea over the following nights: 2013 January 26, 2014 July 14, 2014 September 5, 2013 December 3–5 (half nights), 2014 June 22–23, 2015 February 15 , and 2015 December 13. In total, 54 slit-masks were observed. Each was milled with 1'' wide slits and utilized the 1200 line mm⁻¹ grating, which results in a pixel scale of 0.33 Å pixel⁻¹ and a resolution of ∼1 Å (50 km s⁻¹). For clusters with a redshift below 0.3, the grating was tilted to observe the following spectral features: Hβ, [O iii], Mg i (b), Fe i, Na i (D), [O i], Hα, and the [N ii] doublet. A typical wavelength coverage of 5400 to 8000 Å is shown in Figure 3 for a galaxy observed in CIZAJ2242. The actual wavelength coverage may be shifted by ∼±400Å depending where the slit is located along the width of the slit-mask. This spectral setup enables us to also study the star formation properties of the cluster galaxies; see related work by Sobral et al. (2015). For higher-redshift clusters (above 0.3), the grating was tilted to instead cover the following spectra features: [O ii], Ca(H), Ca(K), Hδ, G-band, Hγ, Hβ, and [O iii]. The position angle (PA) of each slit was chosen to lie between ±5° and 30° of the slit-mask PA to achieve optimal sky subtraction during reduction with the DEEP2 version of the spec2d package (Newman et al. 2013b). In general, for each mask we took three ∼900 s exposures except for a few cases where a few extra minutes at the end of the night were spent on an individual mask or when weather altered our observation plans in the middle of the night. In total, 54 slit-masks were observed with a total of ∼7000 slits over the course of the spectroscopic survey.

Zoom In Zoom Out Reset image size

3.3.1. Keck/DEIMOS: Target Selection

3.3.2. Keck/DEIMOS: Data Reduction

3.3.3. Archival Spectroscopy

We cut each spectroscopic catalog to only include objects within R₂₀₀ in projected space and to within $\bar{v}\pm 3{\sigma }_{v}$, where $\bar{v}$ is the average line-of-sight velocity and σ_v is the cluster velocity dispersion. This is accomplished with an iterative process starting with 5 Mpc and 10,000 km s⁻¹ and shrinking the radius and velocity window until an equilibrium catalog is achieved. The radius is determined in each step by translating the velocity dispersion into a mass using the Evrard et al. (2008) scaling relations followed by estimating R₂₀₀ based on the mass.

This reduces the chance of inclusion of galaxies that are uninvolved in the merger. An instructive example is A2061, where A2067 is ∼2.7 Mpc (30') to the northeast and at a similar redshift, but uninvolved in the merger. The iterative shrinking aperture was able to eliminate galaxies from A2067 from the redshift catalog despite being at a similar redshift because it is outside of R₂₀₀. A second example is A523 (z ∼ 0.1), which has two background groups at z ∼ 0.14 within R₂₀₀ in projection (Girardi et al. 2016).

After this process, 5413 spectroscopic galaxies remain across the 29 systems. The breakdown for each cluster is presented in Table 5.

We display the one-dimensional redshift distribution for 1RXSJ0603 in Figure 4. An analogous figure for the remaining 28 systems is available in the online version as a figure set. The corresponding normalized Gaussian distribution is overlaid with the cluster redshift and velocity dispersion given by the biweight and bias-corrected 68% confidence intervals. We implement the biweight statistic based on 10,000 bootstrap samples of the member galaxies and calculate the bias-corrected 68% confidence limits for the redshift and velocity dispersion from the bootstrap sample. This method is more robust to outliers than the dispersion of the Gaussians generated by our statistical model (Beers et al. 1990).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

View figure set (29 panels)

Tarball of figure set images

We test the goodness of fit of the corresponding Gaussian distribution using a Kolmogorov–Smirnov (KS) test. The results of this analysis are displayed in Figure 4 and in the online figure set for the other systems. We generally find good agreement between the spectroscopic data and single Gaussian distributions, which implies that the merging subclusters have line-of-sight velocity differences that are small compared to the velocity dispersion. The lowest p-value for the KS test is 0.007 for A781, which is known to be composed of several subclusters with large velocity differences (Geller et al. 2005). In Figure 5, the 29 resulting Gaussians are presented to demonstrate the sample distribution of spectroscopic cluster members. The area of a given Gaussian is proportional to the population of galaxies in the respective cluster catalogs.

Zoom In Zoom Out Reset image size

We also fit increasing numbers of Gaussians to the one-dimensional redshift distributions of each cluster utilizing an expectation-maximization Gaussian mixture model method from the Sci-Kit Learn python module. We varied the number of Gaussians from one to seven for each cluster. A single Gaussian model was strongly preferred for 27 of the 29 clusters according to the Bayesian Information Criterion. For A3365, the one Gaussian model was only slightly favored over a two Gaussian model, and for A781, a two-halo model was strongly preferred.

In a second paper (Golovich et al. 2018), we studied the three-dimensional distribution of galaxies and determine substructure and merger scenarios using a panchromatic data set.

In Section 3.3.1 we discuss our methods of selecting targets for our spectroscopic survey. Because our photometric survey was ongoing during this process, we utilized the best available photometry for spectroscopic targeting (see Table 4). Here, we analyze the success of the various targeting methods. Broadly, two distinct methods for selecting potential targets were implemented. For 21 of 54 slit-masks, potential targets were identified via a photometric redshift selection based on SDSS photometric redshifts. For the remaining 33 slit-masks, a red sequence selection was implemented; however, the quality of the imaging (seeing and depth) varied substantially depending on the source. In Table 4, the spectra are broken down by individual slit-mask, targeting method, imaging used for targeting, and redshift.

The biggest indication of the effect of the target imaging quality on the spectroscopic survey is with the fraction of targeted objects that yielded a secure redshift of a cluster member, which was our primary goal. In Table 6, the ∼7000 targeted objects are broken down by the type of object detected. Across the survey, 77% of all targeted objects yielded a secure redshift estimate. Of these, 49% were cluster galaxies. The largest sources of contaminants were background galaxies (26%) and stars (18%).

Background galaxies were detected at higher frequency with photometric redshift targeting. Detection of these objects also decreased with the redshift of the cluster. A substantial fraction of stars were detected in a few fields that had either sub-par imaging for target selection, or had low galactic latitude. The effect of imaging quality with regards to stellar contamination is evident in the five A3411 slit-masks. The first four were observed using INT/WFC targeting, and the fraction of stars increased for successive slit-masks as the best member candidates were depleted by earlier masks. For the fifth slit-mask, Subaru/SuprimeCam was utilized for targeting, and the fraction of stars decreased substantially. The trade-off was a larger fraction of background galaxies, which is explained by the increased depth of the imaging and also by the fact that the four previous slit-masks had identified many of the brightest cluster members. One benefit of the background galaxy redshift determination is for developing training sets for weak-lensing source selection. Furthermore, gravitational lensing is complicated by massive structures along the line of sight, and overdensities of background galaxies may help reveal these types of massive background structures; however, the spectroscopic survey was not designed to detect such systems, so any such detection is serendipitous. Foreground galaxies accounted for only 6% of secure redshifts, and these were predominantly detected in higher-redshift cluster fields. Finally, 305 objects were detected serendipitously; i.e., a single slit had one or more traces in addition to the targeted object. These were predominantly detected in low-galactic -latitude fields and were composed of stars, although, ∼50 cluster galaxies were detected in this manner across the survey.

The presence of radio relics in merging galaxy clusters constitutes a strong prior for ongoing merging activity. Given this, these 29 merging clusters are expected to be composed of 2 or more subclusters. However, 28 of the 29 systems are well fit by a single Gaussian (p > 0.05).

There are two potential explanations for this, which are not mutually exclusive: 1) radio relics indicate a merger occurring within the plane of the sky (transverse to the line of sight), and/or 2) radio relics indicate a merger observed near-apocenter. Based on the redshift results alone, both scenarios are plausible. First, most of our relics were detected in shallow surveys, and the surface brightness is higher when the line of sight intersects a large fraction of the emission in three dimensions (Skillman et al. 2013). Furthermore, detected radio relics have been shown to be highly polarized (e.g Govoni & Feretti 2004; Ferrari et al. 2008), which correlates with a transverse viewing angle of the merger (Ensslin et al. 1998). Second, radio relics occur for only a small fraction of the full merger phase, and it takes time for the radio relic to develop (see Figure 5 of Skillman et al. 2013). This may explain why the Bullet Cluster's bow shock is not coincident with a bright radio relic. Meanwhile, El Gordo contains radio relics, and it was shown to be returning from apocenter (Ng et al. 2015). Thus, it is likely a combination of the two scenarios that explain the unimodal redshift distributions in 28 of 29 systems in the sample. Recent magnetohydrodynamical simulations suggest that explanation (1) is more likely for radio-relic systems (Vazza et al. 2012; Wittor et al. 2017)

The one outlier, A781, is known to be composed of multiple clusters at various redshifts (Geller et al. 2005). The system is composed of two clusters in projection at z ∼ 0.3 and z ∼ 0.42. Here, we studied the z ∼ 0.3 system, which is further split into two redshift peaks (see the online figure analogous to Figure 4. The radio relic is associated with the slightly higher redshift peak on the western side of the cluster. The lower-redshift peak is associated with an infalling subcluster, which is yet to merge, based on the undisturbed X-ray surface brightness distribution (see Figure 1 of Sehgal et al. 2008, where this subcluster is referred to as the Middle subcluster).

The photometric data presented in this paper are sufficiently deep for detailed, wide-field, weak gravitational lensing analyses of each cluster (see Bartelmann & Schneider 2001; Hoekstra 2013, for a review). This will allow for the mapping of the total mass distribution of each system, as well as allow for mass estimation of each subcluster in a manner unbiased by the dynamical interaction of the merger (e.g., Jee et al. 2015, 2016). Both galaxy velocity dispersion based mass estimates assuming the system is virialized (Takizawa et al. 2010) and X-ray temperature or luminosity scaling relation based mass estimates assuming the cluster is in hydrostatic equilibrium (Zhang et al. 2010) overestimate the mass in merging systems. Takizawa et al. (2010), however, showed that mergers along the line of sight are more strongly affected by this, and most strongly near core passage. Radio-relic systems are typically observed ∼1 Gyr after pericentric passage (Ng et al. 2015; Golovich et al. 2016, 2017; van Weeren et al. 2017).

Because we took images with two photometric filters that straddle the 4000 Å break, colors may be assigned to objects, allowing color–magnitude selection. This allows for selection of background galaxies for weak lensing as well as cluster members based on the red sequence technique (see Figure 6).

Zoom In Zoom Out Reset image size

The spectroscopic data contain the added information of line-of-sight motion, which allows for a pure catalog of cluster members. From this catalog, dynamical modeling of the mergers may be achieved. Merging clusters are efficient astrophysical laboratories for studying several phenomena including particle acceleration, cool-core disruption, and potential self-interacting DM signals. Many of these are time-dependent and velocity-dependent, which requires accurate dynamical models. Furthermore, these dynamical models are invaluable for simulators in the form of constrained initial conditions. Finally, the spectral quality from DEIMOS allows for analyses of merging-induced star formation, galaxy evolution, and AGN activity (see, e.g., Sobral et al. 2015).

In this paper, we presented our observational strategy, reduction, and analysis of ∼20 hr of Subaru/SuprimeCam imaging of 29 merging galaxy clusters alongside our spectroscopic follow-up of 7000 objects (54 slit-masks) with Keck/DEIMOS. We presented ∼4340 new high-quality galaxy redshifts and spectral features from our spectroscopic survey matched to the photometry from our Subaru/SuprimeCam survey in Table 7. These data are combined with literature spectroscopy and SuprimeCam imaging, which resulted in 5297 cluster members in total across the 29 systems. A one-dimensional redshift analysis showed that 28 of 29 of the systems are well fit by a single Gaussian distribution. This suggests the ongoing mergers are occurring either within the plane of the sky or are observed near-apocenter (or a combination of the two factors). We analyzed the effect of different imaging sources and selection methods for targeting slits in our spectroscopic survey, and we discussed possible uses for this large data set of photometric and spectroscopic observations of galaxies within merging galaxy clusters.