OBSCURED STARBURST ACTIVITY IN HIGH-REDSHIFT CLUSTERS AND GROUPS

The entire Cl1604 supercluster was imaged at 3.6–8 μm by Infrared Array Camera (IRAC) and at 24 μm by Multiband Imaging Spectrometer (MIPS) as part of the Spitzer program GO-30455 (PI: L. M. Lubin). The area imaged by MIPS covers a region ∼20' × 60' in size and was observed in slow-scan mode with eight scan legs and half-array (148'') cross-scan steps. We repeated this pattern six times to obtain a total exposure time of 1200 s pixel⁻¹ for all but the very edges of the resulting mosaic. The IRAC observations cover a ∼20' × 30' subset of this area and consist of a grid of 5 × 6 pointings with 260'' separations. At each pointing we performed 36 medium-dithered, 30 s exposures, resulting in a total exposure time of 1080 s pixel⁻¹ throughout.

The MIPS and IRAC data were processed using the standard Spitzer Science Center (SSC) reduction pipeline into individual basic calibrated data images. The IRAC images were further processed using a modified version of the SWIRE survey pipeline (Surace et al. 2005) to remove various instrumental artifacts. The resulting images were then co-added using the MOPEX software and the final IRAC and MIPS mosaics have pixel scales of 06 pixel⁻¹ and 12 pixel⁻¹, respectively.

Object detection on the IRAC imaging was carried out using SExtractor primarily on the 3.6 μm mosaic, but also on the 4.5 μm mosaic where the two do not overlap. Sources were selected if they had more than five contiguous pixels above 1.8σ of the background fluctuations. Aperture photometry was carried out in all four IRAC bands using dual-image mode in SExtractor with a fixed 19 aperture. We applied an aperture correction of 1.2 to account for the flux outside our measurement aperture.

Source detection in the 24 μm mosaic was performed following the methodology of the COSMOS survey described in Le Floc'h et al. (2009). Briefly, we use SExtractor to detect all sources at a significance of >1.8σ over an area of at least three pixels. These positions are then used as input to the DAOPHOT package (Stetson 1987) implemented within IRAF (Tody 1986). Bright, isolated sources are used to construct a single empirical point-spread function (PSF) for the entire mosaic. Using DAOPHOT, PSF-fitting photometry is performed at each detected source position and the flux enclosed within a 6'' aperture is computed. We then subtract the best-fit, scaled PSFs from the original mosaic, and a second iteration of SExtractor and DAOPHOT is run to detect and characterize any remaining sources. The catalog from this second pass is joined with the original catalog to produce the final source catalog. Finally, we apply an aperture correction of 1.69. With an exposure time of 1200 s and a moderate background level of 22.7 MJy sr⁻¹, we find our 3σ point-source sensitivity to be roughly 40 μJy, which is in good agreement with estimates from the SSC's SENS-PET online tool.

The optical imaging of the supercluster used in this study consists of 17 pointings of the Advanced Camera for Surveys (ACS) on the Hubble Space Telescope (HST). The ACS camera consists of two 2048 × 4096 CCDs with a pixel scale of 005 pixel⁻¹, resulting in a ∼3' × 3' field of view (FOV). The 17 pointing mosaic was designed to image nine of the ten galaxy density peaks observed in the field of the supercluster by Gal et al. (2005, 2008). Observations were taken in both the F606W and F814W bands and consist of 15 pointings from GO-11003 (PI: L. M. Lubin) with effective exposure times of 1998 s⁴ and 2 GTO pointings from G0-9919 (PI: H. C. Ford) centered on clusters A and D with effective exposure time of 4840 s. Our average integration time of 1998 s resulted in completeness depths of ∼26.5 mag in each band.

The ACS data were reduced using the pipeline developed by the HST Archive Galaxy Gravitational Lens Survey (HAGGLES) team. The pipeline processing consists of calibrating the raw data using the best reference files provided by the HST archive, subtracting the background on each chip, iteratively determining the best shifts between the dithered exposures using multiple calls to SExtractor (Bertin & Arnouts 1996) and multidrizzle (Koekemoer et al. 2002), aligning the final drizzled image to the USNO-B1 catalog (Monet et al. 2002) and resampling the images to a pixel scale of 003 pixel⁻¹. Object detection and photometry was carried out using SExtractor in dual-image mode, with the detection image being a weighted average of the F606W and F814W images. Detection parameters DETECT_MINAREA and DETECT_THRESH were set to three pixels above 3σ, while the deblending parameters DEBLEND_NTHRESH and DEBLEND_MINCONT were set to 32 and 0.03, respectively. An outline of the region covered by the ACS mosaic relative to the MIPS observations is shown in Figure 1. Full details of the HAGGLES pipeline can be found in P. J. Marshall et al. (2011, in preparation).

The Cl1604 supercluster has been extensively mapped using the Low-Resolution Imaging Spectrograph (LRIS; Oke et al. 1995) and the Deep Imaging Multi-object Spectrograph (DEIMOS; Faber et al. 2003) on the Keck 10 m telescopes (Oke et al. 1998; Postman et al. 1998; Lubin et al. 1998; Gal & Lubin 2004; Gal et al. 2008). The target selection, spectral reduction, and redshift measurements are described in detail in Section 3 of Gal et al. (2008). The resulting DEIMOS spectra have a resolution of ∼1.7 Å (68 km s⁻¹) and typical spectral coverage from 6385 Å to 9015 Å, while the LRIS resolution is ∼7.8 Å (330 km s⁻¹) with spectral coverage from 5500 Å to 9500 Å.

Our final spectroscopic catalog contains 1638 unique objects. Redshifts derived for these objects are given a spectroscopic quality, Q_spect, between 1 and 4, where 1 indicates that a secure redshift could not be determined, 2 is a redshift obtained from a single feature, 3 is a redshift derived from at least one secure and one marginal feature, and 4 is assigned to spectra with redshifts obtained from several high signal-to-noise features. Q_spect = −1 is used for sources securely identified as stars. In this sample, we find 140 stars and 1138 extragalactic objects with Q_spect ⩾ 3. A total of 414 galaxies are in the nominal redshift range of the supercluster, 0.84 ⩽ z ⩽ 0.96. The spatial distribution of these galaxies is shown in Figure 3.

Zoom In Zoom Out Reset image size

In addition to redshifts, we also use our optical spectra to measure the equivalent widths (EWs) of the [O ii] and Hδ features for spectral classifications and to calculate SFRs from the [O ii] line luminosity. Due to the relatively low signal-to-noise ratio (S/N) of a single DEIMOS spectrum, bandpass techniques are used for the EW measurements of individual galaxies. For consistency, we also adopt bandpass techniques to measure the absolute luminosity of the [O ii] emission line, making no corrections for the internal extinction of each galaxy. Measurements are made adopting the standard bandpasses of Fisher et al. (1998) and calculated using the methodology described in Lemaux et al. (2010). For flux measurements, absolute spectrophotometric calibration is achieved by using the methods described in Lemaux et al. (2010), adopting slit throughput of ω_slit = 0.37 for all supercluster members. SFRs are derived from [O ii] line luminosities following the prescription of Kennicutt (1998).

To determine the host cluster or group of each 24 μm detected galaxy within the complex structure of the supercluster, we use the galaxy's position and velocity offset relative to the systemic position and redshift of each cluster/group. Creating cluster and group specific subsamples in this manner will allow us to examine the large-scale environments of the starburst population in terms of the properties of their host systems. We consider a galaxy associated with a specific cluster or group if (1) its comoving velocity relative to the median cluster/group redshift is less than three times the system's velocity dispersion, σ_v and (2) it is located within two projected virial radii, R_vir, of the cluster/group center. This latter constraint is motivated not only by a desire to sample a full range of environments around each system, but also due to a large population of infalling galaxies observed near multiple Cl1604 clusters that extend beyond R_vir; this is discussed further below.

We determined R_vir for each cluster and group using the system's measured σ_v and the relation R_vir = R₂₀₀/1.14 (Biviano et al. 2006; Poggianti et al. 2009), where R₂₀₀ is the radius at which the density of the cluster is 200 times that of the critical density of the universe. We measured σ_v for each system using an iterative clipping procedure that is described in detail in Gal et al. (2008), while R₂₀₀ was in turn calculated using the equation

$\begin{equation} R_{200} = \frac{\sqrt{3}\sigma _{\rm v}}{10 H(z)}. \end{equation} \tag{ 1 }$

The derived values of σ_v and R_vir for the Cl1604 clusters and groups are listed in Table 1.

The result of applying the two conditions |dv| < 3σ and R < 2R_vir for both 24 μm detected and undetected galaxies near clusters A, B, and D and groups C, F, and G are shown on the top and bottom panels of Figure 4, respectively. The differing dynamical states of these clusters are readily apparent. In the case of the relaxed cluster A, a large majority of the system's spectroscopically confirmed members fall within R_vir, while in the dynamically unrelaxed clusters B and D, a substantial population are found at greater radii. We interpret this as evidence that both systems are actively accreting a considerable fraction of their galaxy populations.

Zoom In Zoom Out Reset image size

Using the above selection criteria, we have associated a total of 71 of the 126 24 μm detected supercluster members with a specific cluster or group in the Cl1604 complex. The remaining galaxies that fall within the redshift boundaries of the supercluster are found in loose groups, filamentary or sheet-like structures near these systems. For the remainder of the paper, we will refer to galaxies associated with clusters A, B, and D as cluster members and those associated with groups C, F, and G as group members, while the remaining galaxies will be referred to as the supercluster field sample (or simply the superfield population for short).

To determine the level of obscured activity occurring in each 24 μm source, we calculate the galaxy's total IR luminosity (L_TIR; 8–1000 μm) from its monochromatic 24 μm flux density using the recipe of Chary & Elbaz (2001) and the synthetic spectra of Chary & Elbaz (2001) and Dale & Helou (2002). These spectral templates model the 8–1000 μm spectral energy distribution (SED) of galaxies with a wide range of IR luminosities and can be used to derive bolometric corrections converting mid-IR luminosity to total IR luminosity. This correction is template and luminosity dependent as the mid-IR shape of the model SEDs varies monotonically with increasing IR luminosity. For each galaxy, we calculated the bolometric correction from the template SED which best reproduced the observed 24 μm flux at the galaxy's measured redshift. This was done by redshifting the templates to the observed frame, converting them from luminosity to flux per unit wavelength, and convolving the SED with the MIPS 24 μm response curve. The template SED that best reproduced the observed 24 μm flux of the galaxy was then normalized to this flux and integrated to obtain L_TIR.

We used the derived L_TIR values to obtain the level of star formation occurring in each galaxy by way of the L_TIR–SFR relation from Kennicutt (1998):

$\begin{equation} {\rm SFR} \hspace{3.61371pt} (M_{\odot } \hspace{3.61371pt} {\rm yr}^{-1}) = 4.5\times 10^{44} L_{\rm TIR} \hspace{3.61371pt} ({\rm erg \hspace{3.61371pt} s}^{-1}). \end{equation} \tag{ 2 }$

The distribution of L_TIR as a function of redshift for supercluster members is shown in Figure 5. Our 3σ flux limit of 40 μJy corresponds to a luminosity of 3 × 10¹⁰ L_☉ at z = 0.9, the median redshift of the supercluster. This luminosity in turn corresponds to an SFR of 5.2 M_☉ yr⁻¹. Therefore our MIPS observations allow us to sample the luminous end of the IR normal (L_TIR  <  10¹¹ L_☉), LIRG (L_TIR  =  10¹¹–10¹²L_☉), and Ultra LIRG (ULIRG; L_TIR > 10¹² L_☉) populations in the Cl1604 systems. Only two supercluster members have luminosities consistent with ULIRGs, while the majority (64.3%) have LIRG-like luminosities. The remaining 34.1% fall in the IR normal regime. For simplicity, we will often refer to galaxies in all three luminosity classes collectively as the starburst population. In Section 6, we justify this by demonstrating that a significant fraction of these galaxies are experiencing bursty episodes of star formation activity, especially those within the cluster and group environments.

Zoom In Zoom Out Reset image size

The mid-IR flux of galaxies can originate from both obscured star formation and the reprocessing of active galactic nucleus (AGN) related emission by dust. In this section, we discuss the identification of galaxies whose mid-IR flux is dominated by the latter in order to exclude them from any analysis of the supercluster's starburst population. To do this, we have fit starburst and AGN spectral templates to our ground-based optical, IRAC, and MIPS photometry using the Hyperz photometric redshift software (Bolzonella et al. 2000). The templates we use for these fits are a subset of the SWIRE template library (e.g., Polletta et al. 2007) and include five spiral (S0, Sa–Sd), three starburst, and four AGN templates. The starburst SEDs are those of NGC 6090, M82, and Arp200, while the AGN templates are a combination of models, broadband photometry, and composite spectra of local AGNs and correspond to Seyfert 1.8, Seyfert 2, QSO1, and QSO2 SEDs. We adopt a Calzetti reddening prescription (Calzetti et al. 2000) with up to a magnitude of optical extinction to account for heavily obscured sources.

We carried out the Hyperz SEDs fitting on 98 supercluster members detected at 24 μm and all four IRAC bands (out of a total 126 detected at 24 μm). Of this sample, three were best fit by QSO SEDs and another eight were well matched to Seyfert SEDs. The location of these galaxies in IRAC color–color space is shown in Figure 6. All of the Hyperz-selected AGNs fall within the color selection wedge of Lacy et al. (2004) and they generally follow the color trend expected for sources with a negative power-law spectral slope (i.e., Donley et al. 2008). Also shown in Figure 6 are the colors of X-ray luminous AGNs found in the supercluster through our Chandra observations of the system (Kocevski et al. 2009a, 2009b). There are a total of nine X-ray-detected AGNs in the supercluster with Seyfert-like luminosities. Seven of these are detected at 24 μm and in all four IRAC bands, of which five were selected as AGNs by our Hyperz fitting. The outstanding two were found to have a significant starburst component and are most likely starburst/AGN blends. In order to be conservative in our AGN selection, we have not employed hybrid starburst-AGN templates which would have selected these sources as AGNs, but plan to in a future study of the obscured AGN population within the supercluster.

Zoom In Zoom Out Reset image size

The final sample of AGN excluded from our analysis consists of 13 sources (five IR/X-ray selected, six IR selected, and two X-ray selected) out of 98 potential hosts, resulting in a roughly 13% contamination. An additional 28 sources were not screened using Hyperz because they lacked a detection in one or more of the IRAC bands. Given our contamination rate, we expect four of these galaxies to host an AGN. Since these make up less than 4% of our final sample, we do not believe their presence will significantly alter our results.

Of the 13 AGNs found using Hyperz, seven galaxies are part of our cluster/group sample, while the remaining six are part of the supercluster field. The only two cluster/group members with ULIRG-level luminosities are among the three galaxies best fit by the QSO SEDs and are located near the center of cluster D. With these AGNs excluded, our final sample of 24 μm sources consists of 50 cluster, 13 group, and 50 field galaxies. In the following sections, we take a detailed look at the various properties of these galaxies.

First, we find increased activity in the dynamically unrelaxed clusters B and D relative to the more massive and relaxed cluster A. The fraction of cluster members detected at 24 μm in the latter is 22.5%, while in clusters B and D this increases to 36.2% and 34.5%, respectively. These increased fractions are inconsistent with the cluster A fraction at roughly the 93% level, assuming a binomial error distribution. This finding suggests that the dynamical state of these clusters is directly related to the level of dust-enshrouded star formation occurring in their member galaxies.

In the case of cluster D, the 24 μm detected galaxies are preferentially located either within the filament connected to the system or aligned with its infall direction. This is one of the clearest indications from our data that enhanced star formation within these systems is associated with actively infalling or recently accreted galaxies. This finding is similar to the results of Fadda et al. (2008), who detected an overdensity of starburst galaxies in two filaments feeding A1763. The authors suggest that the filament environment is a favorable setting for the onset of starburst activity, a conclusion that our observations support.

We have performed an extensive test to confirm that the increased activity we find among the unrelaxed clusters is not due to variations in our spectroscopic sampling. After carefully correcting for our spectroscopic incompleteness, we find that the density of 24 μm sources in the redshift range 0.84 ⩽ z ⩽ 0.96 is significantly greater near clusters B and D relative to both cluster A and the general field population. We performed this calculation in the following manner. To correct for our spectroscopic incompleteness, we binned our spectroscopic sample of 1278 objects in observed color (F606W–F814W) and magnitude (F814W) space and calculated the ratio of galaxies confirmed to lie within 0.84 ⩽ z ⩽ 0.96 to those outside this range for each color–magnitude bin. We then used these ratios to determine the probability that a given 24 μm source without a redshift would fall within the supercluster redshift range had it been targeted for spectroscopic observations. These probabilities are then used as weights, which are summed to estimate the number of 24 μm detected members missed in any given region within the supercluster. In carrying out this analysis, we chose our color and magnitude bins to be 0.5 and 1.0 mag wide, respectively, in order to ensure adequate number statistics in each bin and we have adopted a magnitude limit of F814W < 23.5.

With an estimate of our spectroscopic incompleteness, we chose three cluster environments (R < R_vir in systems A, B, and D) and two field environments in which to compare the number density of 24 μm sources. The field regions have been chosen to avoid the cluster and group cores, but to remain within our ACS imaging and the regions well sampled by our spectroscopy. Although these regions will undoubtedly be denser than the nominal field well outside the Cl1604 supercluster, we have decided to compare against them as opposed to field surveys since the 24 μm flux limit and our spectroscopic and ACS coverage in these regions are identical to that of the cluster cores. This allows for a fairly straightforward comparison of the source densities in these different environments.

The corrected number density of 24 μm sources in each region, integrated over the redshift range of 0.84 ⩽ z ⩽ 0.96, is listed in Table 2. We find an excess of 24 μm sources within the virial radius of each cluster relative to the supercluster field. In the three regions encompassing the cluster centers, the average projected density of dusty, star-forming galaxies is nearly twice that of the adjacent field. This overdensity is detected with a statistical significance of just over 3σ. Furthermore, the excess of 24 μm detected galaxies in each cluster appears well correlated with the system's dynamical state. We find a nearly two-fold increase in the density of sources in the dynamically unrelaxed cluster D relative to the more relaxed cluster A. This is consistent with our earlier finding that the two dynamically disturbed clusters have a greater fraction of 24 μm detected members.

It is important to note that the overdensity of 24 μm sources found in clusters B and D cannot simply be a result of the higher overall density of galaxies in these systems as a similar excess is not observed in cluster A. In fact, cluster A is more massive and optically richer than either of its counterparts. We would have, therefore, expected to find a greater excess in that system if the density of 24 μm sources simply scaled with the general galaxy population and this is not the case.

Our findings are similar to those of Geach et al. (2006), who detected a substantial population of 24 μm detected galaxies in Cl0024+16 and relatively few in the more massive and relaxed cluster MS0451-03. Given that dynamically unrelaxed clusters are typically undergoing active assembly, the increased activity observed in these systems further suggests that the starburst population consists of recently accreted field galaxies whose activity is either triggered directly prior to cluster infall or by cluster-specific processes during infall.

The second insight provided by the spatial distribution of the 24 μm population is that they are less centrally concentrated than their undetected counterparts when summed over all of the Cl1604 clusters (systems A through D). This can be seen on the top panel of Figure 8, which shows the cumulative distribution of 24 μm detected and undetected member galaxies as a function of cluster-centric distance. Employing a Kolmogorov–Smirnov (K-S) test on the normalized distribution, we calculate only a 3.5% probability that the 24 μm detected cluster sample is drawn from the same parent distribution as the undetected population. Undoubtedly, the radial distribution of the cluster members is heavily influenced by the large number of 24 μm bright galaxies in the filament extending outward from cluster D. On the other hand, the group sample does not show a similar segregation. The radial distribution of the group members is shown on the bottom panel of Figure 8. A K-S test on the distribution of the detected and undetected group galaxies results in a 12.6% probability that the two subsamples are not drawn from the same distribution, a statistically insignificant difference.

Zoom In Zoom Out Reset image size

This difference in the spatial concentration of group and cluster 24 μm galaxies was also recently observed by Krick et al. (2009), who measured a less concentrated distribution of star-forming galaxies in a massive, evolved cluster at z = 1 than in two dynamically younger systems at the same redshift. These results imply that while starburst activity is found throughout the group population, it is predominantly found among infalling galaxies in the cluster environments.

At the redshift of the Cl1604 systems the spectral window of our DEIMOS observations encompasses the [O ii] doublet at 3727 Å and the Hδ absorption line at 4101 Å, which we can use to probe the dominant mode of star formation activity (i.e., bursty versus continuous) occurring in the 24 μm detected galaxies. Much has been written in the literature regarding the ability of these features to provide insight into the recent star formation histories of galaxies (see Oemler et al. 2009 for a recent review). We briefly discuss the utility of these lines below, before presenting our spectral classification of the 24 μm bright population.

Due to the lifetime of the stars that give rise to each feature, namely O- and B-type stars for [O ii] and longer-lived A-type stars for Hδ, these lines provide a measure of the average SFR of a galaxy over different timescales, with [O ii] measuring current activity (τ ∼ 10⁷ yr) and Hδ reflecting more extended activity (τ ∼ 10⁹ yr). As a result, the strengths of these lines, in the absence of significant dust extinction, are well understood for different types of activity. In galaxies undergoing normal, continuous star formation, the EW of both [O ii] and Hδ increases with increasing activity until the SFR is high enough that Hδ emission from new stars begins to infill the Hδ absorption line, leading to a overall decrease in the line's EW. On the other hand, in bursty conditions, when an epoch of high star formation activity is followed by one with significantly less, the EW of Hδ increases dramatically as there is insufficient Hδ emission from the current generation of stars to infill the Hδ absorption produced by the previous generation. Galaxies that have experienced a rapid reduction in their activity, i.e., post-quenched systems, show strong Hδ absorption for the same reasons, except that they also have little or no [O ii] emission.

Using these characteristics, Dressler et al. (1999) devised a set of spectral classes based on the EW of [O ii] emission and Hδ absorption that reflect different modes of star formation activity. In this classification scheme, e(a) and e(b) galaxies are ongoing starbursts with varying levels of dust obscuration. While both classes exhibit strong Balmer absorption (EW(Hδ) >4 Å), only moderate [O ii] emission (EW[O ii] >−40  Å) is visible in e(a)'s, while e(b)'s show strong [O ii] emission (EW[O ii] <−40  Å). On the other hand, the spectra of e(c) galaxies are consistent with normal star formation, having moderate levels of both [O ii] emission (EW[O ii] >−40  Å) and Balmer absorption (EW(Hδ) <4 Å). Finally k+a, or post-starburst systems, are characterized as having strong Balmer absorption features superimposed on an older K star spectrum and lacking [O ii] emission.

Using this spectral classification scheme, we have classified the 24 μm detected cluster, group, and field galaxies based on the EW of [O ii] emission and Hδ absorption measured in their spectra. Out of the 113 non-AGN, 24 μm bright supercluster members, 101 have spectra with adequate S/N for this analysis. The EWs were measured as described in Section 3.3 and a line was considered detected if its measured EW is greater than three times the error associated with that measurement. The results of our analysis are listed in Table 3. In general, we find that the 24 μm detected cluster and group members predominantly have e(a)- and e(c)-type spectra, which indicates they are comprised of a mix of obscured starbursts and galaxies undergoing normal, continuous star formation. We find similar results in the superfield sample, with the notable difference being a reversal in the e(a) to e(c) ratio. A greater fraction of the cluster and group galaxies appear to be undergoing dusty starburst activity compared to the field sample (47.5% versus 32.5%), where a majority have spectra consistent with normal star formation (57.5%). The reversal in the e(a) to e(c) ratio indicates that star-forming galaxies in these denser environments are, on average, experiencing burstier activity than their counterparts in the field.

Our spectral classification of the supercluster members suggests an excess of galaxies with strong Balmer absorption features, which could possibly be due to increased starburst activity. To investigate this further, we have stacked the spectra of the 24 μm detected galaxies as a function of environment to produce high-S/N composite spectra. These co-added spectra can be used to gauge the average spectral properties of the stacked subsamples to greater accuracy than is possible using single galaxy spectra (e.g., Dressler et al. 2004). We will use these properties to compare the dominant mode of star formation (i.e., bursty versus continuous) occurring in the cluster, group, and superfield populations.

To produce the composites, the spectra of galaxies in each region are normalized and averaged using an inverse variance weighting. We find that the resulting spectra do not depend significantly on whether we use an (optical) luminosity-weighted average or a unit-weighted average, especially for the most luminous 24 μm sources.

In total the group, cluster, and superfield subsamples include 8, 30, and 47 24 μm detected galaxies, respectively. These totals decrease to 7, 20, and 28, respectively, when we only consider galaxies with LIRG-level luminosities (L_TIR > 10¹¹ L_☉). For each stack, the cluster and group samples were restricted to galaxies within the virial radius of each system. The resulting composite spectra for the 24 μm population of each region are shown in Figure 9.

Zoom In Zoom Out Reset image size

For the higher S/N composite spectra, fitting techniques were used to determine the EW of the [O ii] and Hδ features. The [O ii] doublet was fit using a double-Gaussian model at fixed wavelength separation (2.8  Å) plus a linear continuum. The Hδ line was also fit with a two-Gaussian model plus a linear continuum, the two Gaussians characterizing the absorption and emission component of the feature. We also measure the strength of the 4000  Å break (D_n4000) using the ratio of the average flux in the bandpasses of Balogh et al. (1997).

In Figure 10, we plot the EW of [O ii] against that of Hδ measured in the composite spectra of the 24 μm detected cluster, group, and superfield galaxies. The region of parameter space that Dressler et al. (2009) calculated, which should encompass 95% of all normal star-forming galaxies, is denoted by two dashed lines. Dressler et al. (2009) derive this from the mean relationship between EW([O ii]) and EW(Hδ) observed in the Sloan Digital Sky Survey (SDSS) and from the analysis of Goto (2003).⁵ We find that the detected superfield galaxies have on average moderate [O ii] and Hδ line strengths, placing them well within the region where we expect to find galaxies undergoing continuous star formation. This is not the case for the 24 μm detected group galaxies, which show an excess of Balmer absorption indicating a greater incidence of bursty activity. The detected cluster members exhibit a similar excess, but only among the more luminous galaxies which have at least LIRG-level luminosities (L_TIR > 10¹¹ L_☉). When considering all detected cluster members (with all IR luminosities), the average EW(Hδ) is elevated compared with the superfield sample, but consistent with normal star formation. When we average only the more luminous members, we find a substantial increase in EW(Hδ) that is not seen in the field population with equivalent luminosities.

Zoom In Zoom Out Reset image size

The different spectral properties we find as a function of IR luminosity suggest that the lower luminosity cluster members are undergoing high levels of continuous star formation, while the cluster LIRGs are experiencing more bursty activity. The reason we do not find a substantial change in the EW(Hδ) in the group sample when we cut on IR luminosity is because almost all of the detected group galaxies (7/8) are LIRGs and show strong Balmer absorption. It is important to note that we do not find a similar increase in Balmer absorption among the LIRGs in the field. Both the superfield LIRG sample and the entire superfield sample have average spectral properties consistent with normal, continuous star formation. This is largely due to Hδ emission infill that is absent in the spectra of galaxies in the denser environments. This indicates that the cluster and group galaxies are either affected by more dust obscuration, experiencing burstier activity, or both, as the two are known to be correlated (Dopita et al. 2002; Kewley et al. 2004).

This increased starburst activity among the more luminous cluster and group galaxies is also evident in a D_n4000–EW(Hδ) diagram. The D_n4000 strength is an indicator of mean stellar age and this diagram has been effectively used by Kauffmann et al. (2003) to provide insight on the star formation histories of SDSS galaxies. In Figure 11, we plot the D_n4000 strength versus EW(Hδ) measured from the cluster, group, and superfield composite spectra. Also shown are models tracking the evolution of EW(Hδ) and D_n4000 as a function of age for galaxies with different star formation histories. The model tracks are derived from the synthetic templates of Bruzual & Charlot (Bruzual 2007), assuming a Salpeter initial mass function (Salpeter 1955) and solar metallicity. The tracks shown include a single, exponentially declining burst (with τ = 0.01 Gyr) and a continuous star formation model (τ = 1.5 Gyr). There are also three tracks for galaxies experiencing secondary bursts of star formation 1 Gyr after an initial burst which produce 5%, 10%, and 20% of additional stellar mass. Since the Bruzual & Charlot templates do not model nebular emission lines, we have corrected for Hδ emission infill in the composite spectra using the correlation between [O ii] and Balmer line strengths as described in Schiavon et al. (2006).⁶

Zoom In Zoom Out Reset image size

In Figure 11, we see that the group and cluster LIRG subsamples have elevated levels of Hδ absorption relative to the continuous star formation model. In both cases, the measured Hδ and D_n4000 strengths are more consistent with the burst tracks than the continuous track. The group galaxies show the greatest deviation from the continuous track, having the highest measured EW(Hδ) values in the supercluster. We also find that they have the lowest D_n4000 values, suggesting this environment hosts galaxies with extremely young stellar populations. On the other hand, the average spectral properties of the superfield galaxies (both the entire 24 μm detected population and the LIRG subsample) are more consistent with an older stellar population that has been produced by a more continuous rather than bursty star formation history.

Our interpretation of the D_n4000–Hδ diagram is consistent with the results from our [O ii]-Hδ analysis: the 24 μm bright galaxies in the denser environments appear to be undergoing a recent burst of star formation not observed in the superfield population. In both Figures 10 and 11, this assessment is largely based to the increased Balmer absorption found among the cluster and group galaxies. Although a Balmer excess can be produced through the rapid truncation of normal star formation, all of the galaxies considered here have high IR luminosities and inferred SFRs. It is more likely that this excess is the result of a temporary increase in activity experienced within the cluster and group environments. This finding has important ramifications as it indicates that the star-forming cluster and group galaxies are not simply analogs of the field population that have yet to be quenched. A greater fraction of the galaxies in the denser environments are experiencing a burst of star formation compared to field galaxies at the same redshift.

A final noteworthy point regarding the spectra of the 24 μm detected sample is the small fraction of galaxies that have [O ii] emission lines strong enough to be classified as e(b) galaxies. This lack of strong [O ii] emission has previously been noted for IR-luminous cluster galaxies (Marcillac et al. 2007; Dressler et al. 2009) and is thought to be the result of selective dust extinction, where younger stellar populations that produce the [O ii] doublet are more heavily obscured than the older A-type stars that give rise to Balmer absorption features (Poggianti & Wu 2000). Poggianti et al. (1999) previously proposed that due to such extinction, e(a) galaxies in higher redshift clusters could be the dusty, starbursting progenitors of the k+a population observed in these systems, a scenario that our observations support.

We can gauge the level of dust obscuration affecting the 24 μm bright members by comparing their optically derived SFRs (determined from their [O ii] line luminosities) to the more dust-insensitive rates obtained from their IR emission. To estimate these rates, we employed the $L([ \rm{O\,{\mathsc{ii}}}])\hbox{--}{\rm SFR}$ and L_TIR–SFR relationships of Kennicutt (1998) as described in Sections 3.3 and 4.2. The results are shown in Figure 12, where we plot the ratio of optical to IR-derived SFRs as a function of IR luminosity (and optical color, see Section 8). For any given galaxy several factors may offset the optically derived SFR from its IR counterpart in addition to extinction, including variations in metallicity and ionization parameter, which likely contribute to the significant scatter seen in Figure 12. Nonetheless, the overall trend is apparent: an increasing disparity between the [O ii] and 24 μm derived SFRs with increasing IR luminosity. This reflects the well-known result that the severity of dust extinction scales with star formation activity in IR-luminous systems (Dopita et al. 2002; Kewley et al. 2004). For the most IR-luminous galaxies (L_TIR > 10^11.5), the discrepancy between the optical- and IR-derived rates is roughly one order of magnitude. Although the star formation activity of these galaxies is not completely hidden by dust at optical wavelengths, the use of optical-line diagnostics would severely underestimate the level of ongoing activity in these systems.

Zoom In Zoom Out Reset image size

One of the central issues debated since the initial detection of the Butcher–Oemler effect is whether galaxies experience a period of enhanced star formation during cluster assembly. Despite spectroscopic evidence for starburst activity in distant clusters, the answer to this question has remained elusive since the evolution of the star-forming properties of cluster galaxies often mimics the evolution found in the field (e.g., Bai et al. 2007). In our study of the Cl1604 clusters, we have found evidence that the elevated star formation activity in these systems is not simply a reflection of the increased field activity at z ∼ 0.9. In summary, we find an overdensity of IR-bright galaxies in the Cl1604 clusters relative to the coeval field and that this excess scales with a system's dynamical state. These galaxies exhibit elevated levels of Balmer absorption and increased levels of obscuration which suggests that their star formation activity is burstier than that of field galaxies with similar IR luminosities.

Our findings regarding the excess of 24 μm bright galaxies in dynamically active systems are consistent with several previous studies that have reported source overdensities in unrelaxed or merging clusters (Geach et al. 2006; Marcillac et al. 2007). Recently Saintonge et al. (2008), using MIPS observations of eight clusters, found that while the fraction of galaxies undergoing dust-obscured star formation steadily increases with cluster redshift, dynamically unrelaxed systems exhibit an excess of sources beyond the nominal redshift trend. The authors attribute this increase to a higher galaxy accretion rate in unrelaxed clusters. Our results generally agree with this assessment given that the 24 μm detected galaxies appear to be an infalling population. This is especially apparent in cluster D, where we find dust-obscured activity associated with galaxies streaming along a large-scale filament connected to the system.

That being said, the differences we observe in the IR-bright cluster members relative to the field indicate that their star formation activity cannot simply be residual field activity that has yet to be quenched. Their optical colors are generally redder (which we have shown correlates well with extinction in this sample), their morphologies are more centrally concentrated, and their spectra show signs of elevated starburst activity relative to the field. Dressler et al. (2004) previously demonstrated that the average spectral properties of moderate-redshift clusters could not be accounted for through a mix of passive and continuously star-forming galaxies and that starburst activity must play a larger role at higher redshifts. However, they found this to be true of the field population as well. In our study, we find that the composite spectra of the cluster and group members show a greater contribution from starburst galaxies than we see in the field population. When one further considers that the lifetime of a burst (∼500 Myr) is far shorter than the time required for an infalling galaxy to reach the cluster core (1260 Myr in cluster D for a galaxy infalling from the virial radius at 582 km s⁻¹), it becomes evident that the elevated starburst activity we observe must be triggered, to some extent, by processes related to the cluster environment. This finding is at odds with scenarios that simply rely on an increased rate of passive accretion to explain the greater incidence of star-forming galaxies in higher redshift clusters (e.g., Ellingson et al. 2001; Loh et al. 2008).

We have presented three observational clues that can help shed light on the mechanism responsible for initiating the elevated starburst activity. We observe (1) the fraction of 24 μm detected galaxies showing signs of a recent merger or tidal interaction is nearly twice as high within the cluster/group virial radius (65.8%) than it is outside this distance (35.7%), (2) interaction signatures are three times as common among the detected members than the undetected members, and (3) the 24 μm detected sample within the virial radius exhibits a higher central concentration of rest-frame B-band light than coeval field galaxies with equivalent IR luminosities.

Galaxy mergers and weak tidal interactions (i.e., harassment) are two mechanisms that can both produce the disturbed morphologies we observe and trigger episodes of enhanced star formation (Barnes & Hernquist 1991; Moore et al. 1996; Barton et al. 2007). Given the increased frequency of disturbed morphologies among the 24 μm detected members, it is likely interactions play a role in triggering the increased activity we observe. This cannot be said of the IR-luminous field population at similar redshifts. While LIRG-level activity at low redshifts is predominantly associated with galaxy interactions (Ishida 2004), this is not the case at the redshift of the Cl1604 systems. For example, Melbourne et al. (2005) report that a majority of LIRGs in the field at z ∼ 1 appear to be undisturbed late-type systems rather than galaxies with peculiar or irregular morphologies. We also observe this for galaxies on the cluster/group outskirts and in the supercluster field, but find the opposite result within the cluster/group centers, where a majority of the 24 μm detected galaxies exhibit morphological disturbances. If continuous, LIRG-level activity is the normal mode of star formation for gas-rich field galaxies at z ∼ 0.9 and triggered starburst activity more prevalent in denser environments, then this could explain many of the physical differences we observe in our 24 μm detected sample as a function of cluster/group-centric distance.

Can interactions explain our third point: higher central concentrations for the 24 μm detected galaxies in the cluster and group centers? Simulations have shown that both numerous tidal interactions and mergers can funnel material to the center of a galaxy and trigger a circumnuclear starburst (e.g., Noguchi 1988; Hernquist 1989; Mihos & Hernquist 1996). Observationally, this effect has been detected in several low-redshift clusters. In a survey of Hα emission from late-type galaxies in eight Abell clusters, Moss & Whittle (2000) found that circumnuclear starburst activity is preferentially found in regions of high galaxy densities and associated with galaxies that have peculiar or distorted morphologies. They conclude that tidal interactions, working more efficiently near the cluster centers, are responsible for both the disturbed morphologies and the compact star formation. This is in excellent agreement with our findings as we observe both an increase in starburst activity and interaction signatures at low cluster-centric radii.

Since both mergers and harassment can give rise to many of the morphological disturbances we find, our observations cannot rule out either mechanism as the starburst catalyst. Although major mergers are not expected to be common within the cluster environment given the high relative velocities of cluster galaxies, Oemler et al. (2009) recently found evidence for merger activity associated with 24 μm detected galaxies near the core of A851. Citing Struck (2006), they proposed that the compression of recently accreted groups and bound galaxy pairs could lead to an increased merger rate and elevated SFRs among infalling galaxies. In the Cl1604 clusters, we do observe signatures of galaxy–galaxy mergers that could not originate from harassment alone, such as interacting pairs and ring galaxies, indicating that merger activity is present to some extent in these high density environments. That being said, the increase in disturbed morphologies we find within the cluster centers is largely driven by an increase in galaxies exhibiting tidal features and not those involved in obvious mergers. The fraction of 24 μm detected cluster members that have either an irregular morphology or M interaction classification is the same within the cluster cores as it is on the cluster outskirts.⁹ On the other hand, the fraction of detected galaxies showing tidal features (both with and without a nearby companion) is substantially higher within the cluster centers. This is not to say harassment is the dominate mechanism affecting these galaxies since bone fide mergers may also be responsible for generating many of the tidal features we observe. Considering further that mergers are not always obvious in optical imaging, it is difficult to judge the relative importance of the two mechanisms other than to say signatures of both are common among our 24 μm detected cluster sample.

In the case of the Cl1604 group sample, the picture is a bit clearer. Among the IR-bright group galaxies, we find that irregular morphologies are as common as late-type morphologies and that almost all detected galaxies within R_vir (7/8) show clear signs of ongoing mergers or interactions. These interactions appear to be having a substantial effect on the star formation activity of the detected group galaxies, as they have among the highest IR luminosities in the supercluster sample and exhibit a large Balmer excess relative to both IR-bright field galaxies and the detected cluster population. Therefore, while a mix of harassment and mergers are likely driving the enhanced star formation found in the Cl1604 clusters, mergers appear to be the dominate triggering mechanism in the group environment, where the lower relative velocity of galaxies is much more conducive to such activity.