STAR-FORMING BRIGHTEST CLUSTER GALAXIES AT 0.25 < z < 1.25: A TRANSITIONING FUEL SUPPLY

The sample is initially defined to include all 83 clusters from McDonald et al. (2013b), which were selected from the SPT 2500 deg² survey (Bleem et al. 2015) and subsequently observed in the X-ray with the Chandra X-ray Observatory. To this sample, we add an additional eight clusters from Bleem et al. (2015) that have archival Chandra data from other sources (SPT-CLJ0106-5943, SPT-CLJ0232-4421, SPT-CLJ0235-5121, SPT-CLJ0516-5430, SPT-CLJ0522-4818, SPT-CLJ0658-5556, SPT-CLJ2011-5725, SPT-CLJ2332-5053). X-ray data products, such as central entropy (K₀), are presented for this full sample of 91 clusters in McDonald et al. (2013b)—we direct the reader there for a full description of our X-ray methodology.

For each cluster, we obtained some combination of ground-based g, r, i, z optical imaging; ground-based J, H, K near-IR imaging; and Spitzer IRAC 3.6 μm, 4.5 μm imaging, as described in detail in Bleem et al. (2015). These data have been used to provide optical confirmation of the SZ-selected clusters, and to estimate a photometric redshift based on the member galaxy colors. We remove from our sample all clusters for which the optical-IR follow-up was performed in $\leqslant 2$ filters, preventing a reliable fit to the stellar continuum. After this cut, we were left with a sample of 82 clusters. From this follow-up imaging, we initially select the BCG²⁷ as the brightest red sequence galaxy within an aperture of R₂₀₀ centered on the SZ peak, following Song et al. (2012). We then visually inspect each cluster and select a new BCG if one, or both, of two conditions are met: (i) there is a similarly bright galaxy that is significantly closer to the X-ray peak (29% of systems); (ii) there is a bright blue galaxy on or near the X-ray peak that was not selected due to our initial preference for red galaxies (3% of systems). Wherever possible, we use Hubble Space Telescope (HST) imaging from the SPT weak-lensing follow-up programs with IDs 12246, 12477, 13412 (PIs: Stubbs, High, Schrabback) to aid in the visual identification of the BCG.

In all cases where it was not clear which galaxy was the BCG, we retain the original red BCG, allowing for two BCGs in these clusters. In this case, we give each BCG a 50% weight when performing statistical analysis, and include the limiting cases in our error estimates (i.e., upper limit contains star-forming BCG, lower limit contains passive BCG). This procedure resulted in a sample of 90 BCGs in 82 clusters. The net effect of allowing multiple BCGs in cases where the identification of a single object was challenging is to increase our uncertainties, increasing the likelihood that the "true" answer lies without our error bars.

Several studies have found correlations between the amount of star formation in the BCG and the core entropy (e.g., Cavagnolo et al. 2008; O'Dea et al. 2008; Voit et al. 2015) and luminosity-derived cooling rate (e.g., Crawford et al. 1999; O'Dea et al. 2008; McDonald et al. 2010, 2011b). This has led to the conclusion that star formation is fueled in the BCG by residual cooling flows (Voigt & Fabian 2004; Tremblay et al. 2012; McDonald et al. 2014b). In order to test whether this trend was established at high-z, we require estimates of the core entropy and cooling rate for each cluster. Given that we only have ∼2000 X-ray counts per cluster, modeling the central entropy (e.g., Cavagnolo et al. 2009) or estimating the spectroscopically derived cooling rate (e.g., Voigt & Fabian 2004) is not feasible. Instead, we compute spectroscopic quantities (bolometric luminosity, temperature) from a circular aperture with a radius of 0.075 R₅₀₀ (where R₅₀₀ was derived based on the Y_X–M₅₀₀ relation of Vikhlinin et al. 2009), which should roughly correspond to the deprojected core temperature (see e.g., McDonald et al. 2014a). X-ray spectra extracted from this aperture are modeled with a photometric absorption (phabs) and plasma (apec) model, allowing the temperature, metallicity, and normalization of the plasma model to vary. This choice of aperture is meant to capture the "core" properties, reflecting the realistic fuel reservoir that the BCG may have access to.

The "central" density is computed based on the deprojected X-ray surface brightness profile, following Vikhlinin et al. (2006) and McDonald et al. (2013b). Because the measurement of electron density requires far fewer X-ray counts than the measurement of spectroscopic temperature, the central density is measured at r = 0.01 R₅₀₀, or roughly 10 kpc for a typical cluster in this sample. We combine the projected core temperature and deprojected core density to arrive at a pseudo-deprojected core entropy (${K}_{0}={{kT}}_{0.075{R}_{500}}\;{n}_{e,0.01{R}_{500}}^{-2/3}$). By measuring the "central" entropy within a fixed aperture (relative to the overall cluster size) at all redshifts, we avoid any issues of clusters appearing "coarser" at ever-increasing redshifts.

Unlike our previous work (McDonald et al. 2013b), here we measure "central" quantities at the X-ray peak, rather than the large-scale centroid. In general, the BCG is located closer to the X-ray peak than the centroid in cases of merging clusters, which motivated this choice. We discuss the effects of this choice in later sections. For a more detailed description of our X-ray analysis techniques, we direct the reader to McDonald et al. (2013b, 2014a).

Ground-based optical and/or near-infrared imaging for all clusters in this sample have been obtained at a variety of wavelengths as part of a confirmation and photometric redshift follow-up campaign. The acquisition, reduction, and calibration of these data are presented in detail in Song et al. (2012) and Bleem et al. (2015). Aperture photometry for the BCG was obtained from SExtractor (Bertin & Arnouts 1996), following Song et al. (2012) and Bleem et al. (2015).

In addition to these existing data, we acquired new ground-based u-band imaging with Megacam (McLeod et al. 2015) on the Magellan Clay telescope for 49 clusters in this sample. These 49 clusters were those that had the least restrictive upper limits on the BCG SFR at other wavelengths, generally lacking in spectroscopic or deep GALEX coverage. The exposure time for these observations was chosen to provide an overall sensitivity of our survey to obscured SFRs of ∼10 M_⊙ yr⁻¹; without these additional data, our sensitivity limit would vary significantly with redshift. These data were reduced using the standard photometric pipeline described in Bleem et al. (2015).

The position of each BCG was cross-referenced with the GALEX²⁸ (Morrissey et al. 2007) and WISE²⁹ (Wright et al. 2010) archives, assuming a maximum offset of 2'', from which we obtained near-UV (NUV) and near—mid IR (NIR, MIR) photometry for each BCG. If the BCG was undetected by either of these surveys, we instead obtained upper limits. Spitzer 3.6 μm, 4.5 μm, and WISE 4-band photometry was converted from Vega to AB magnitudes following Price et al. (2004). The resulting UV–optical–IR SEDs, which span 2000 Å to 22 μm in the observed frame, are shown in Figure 11 for our sample of 90 BCGs. As mentioned above, we discarded all BCGs for which we have $\leqslant 2$ photometric measurements (excluding upper limits).

The observed SEDs are fit in two stages. First, we model the full SED with a single-age population with formation redshift z_f, solar metallicity, and a IMF (Salpeter 1955). We assume a uniformly distributed range of z_f from 2 to 5 in our models, which leads to some uncertainty in our resulting stellar masses and SFRs. The spectrum for this old stellar population was generated using Starburst99 (Leitherer et al. 1999) and was convolved with our broadband filter set. The model spectrum was fit to the data using mpfitfun, ³⁰ which minimizes χ² with respect to the two free parameters (normalization, redshift). The redshift was allowed to vary within the measured uncertainty from Bleem et al. (2015), while the normalization, which corresponds to the stellar mass of the old population, was left free.

Beyond this single-component model, we also consider additional components in the UV and IR, representing contributions from a young stellar population and warm dust, respectively. At short wavelengths, we model excess emission using Starburst99, assuming a constant SFR over the past 30 Myr (roughly the AGN duty cycle). We note that adjusting this timescale down to 10 Myr or up to 100 Myr results in deviations in the derived SFR of ∼20%. We assume that the emission from young stars is obscured by dust, incorporating the Calzetti et al. (2000) extinction law and an intrinsic reddening of E(B − V) = 0.3 ± 0.1—this range is based on observations of nearby star-forming BCGs (Crawford et al. 1999; McDonald et al. 2012b). The young stellar component has a single free parameter, corresponding to the ongoing, extinction-corrected SFR. At long wavelengths, we mimic a dusty component with a mid-infrared power law (${F}_{\nu }\propto {\lambda }^{2.0\pm 0.5}$), following Casey (2012). This dust component is artificially truncated at shorter wavelengths, so that it will not add UV flux. We only include the young and dusty components if their inclusion improves the ${\chi }_{\mathrm{dof}}^{2}$, which is the case for 34% (31/90) of the BCGs in our sample. For the remaining 66% (59/90), the "passive-evolution" model yields a suitable fit to the data. The results of this SED fitting are shown in Figure 11.

For 36 of the 95 BCGs in this sample, we have optical spectroscopy from a combination of the IMACS (Dressler et al. 2011) and LDSS3 spectrographs on Magellan. These spectra were initially obtained as part of a spectroscopic redshift campaign that targeted, primarily, red sequence galaxies. The full details of this spectroscopic follow-up are provided in Ruel et al. (2014) and Bleem et al. (2015). For each spectrum, we measure the [O ii] and Hδ equivalent widths and the 4000 Å break strength (D₄₀₀₀), via indices defined by Balogh et al. (1999). [O ii] equivalent widths are converted to emission line fluxes using the continuum level, as determined by the SED modeling described in Section 2.2, at rest-frame 3727Å. From these 36 spectra we find significant (>3σ) [O ii] emission in five systems, one of which is the Phoenix cluster (McDonald et al. 2012a). This system allows us an opportunity to compare the index-based measurement of [O ii] to the total flux from deep integral field spectroscopy (McDonald et al. 2014c). Using the long-slit data, the index-based technique of Balogh et al. (1999) combined with our best-fit optical SED yields a line flux of ${f}_{[{\rm{O}}{\rm{II}}]}=2.2\times {10}^{-14}$ erg s⁻¹ cm⁻² for the central galaxy in the Phoenix cluster, while the total measured flux from McDonald et al. (2013a) is ${f}_{[{\rm{O}}{\rm{II}}]}=1.6\times {10}^{-14}$ erg s⁻¹ cm⁻². Thus, for this single comparison, the index-based approach does an adequate job of reproducing the measurement made via emission line modeling of significantly higher-quality data.

When correcting the [O ii] emission line flux for extinction, we assume that the ionized gas has a factor of two higher reddening than the stellar continuum, following Calzetti et al. (2000) and Calzetti (2001). This assumption is motivated by the idea that the warm ionized gas traces the highly ionizing O and B stars, which tend to be embedded in dusty, star-forming regions.

2.5.1. UV-derived SFRs

2.5.2. [O ii]-derived SFRs

2.5.3. IR-derived SFRs

2.5.4. Comparison of SFR Estimates

In Figure 1, we compare SFRs derived from the three indicators described above. We have limited overlap between the three subsamples, with only three BCGs having both IR- and UV-derived SFRs and an additional four BCGs with both UV- and [O ii]-derived SFRs. Only a single BCG (SPT-CLJ2344-4243) has SFRs inferred from all three methods. Nonetheless, we proceed to compare how well these estimates agree for systems in common. In general, SFR estimates from different indicators agree within the systematic errors. Assuming a one-to-one relation, we measure a scatter of 0.37 dex between UV- and IR-derived SFRs, and 0.15 dex between UV- and [O ii]-derived SFRs. The better agreement between the UV- and [O ii]-derived SFRs is most likely due to the fact that the dominant systematic uncertainties (extinction, old stellar population age) are correlated between these SFR estimates. SFRs for each BCG are listed in Table 2, along with relevant information about the host cluster.

Zoom In Zoom Out Reset image size

Before presenting results from this survey, we would like to draw attention to one potential issue with our approach: All three indicators of star formation employed here—UV continuum, [O ii] line emission, and IR continuum—are also indicators of active nuclei. The relative amounts of contamination at each of these wavelengths depends on the type of AGN considered. For example, in the Phoenix cluster, more than half of the total IR continuum comes from a dusty QSO. On the other hand, this AGN contributes <5% to the UV and [O ii] emission. Several of the SEDs shown in Figure 11 exhibit a power-law shape that is consistent with both a dusty starburst and an AGN.

In an attempt to quantify the contamination due to AGN in this sample, we consider in Figure 2 the 3.5–4.5 μm color for 49 BCGs with Spitzer data and 82 BCGs with WISE data. Color corrections from Stern et al. (2012) have been applied to match data from these telescopes to a common photometric system. Following Stern et al. (2005, 2012), we classify systems with mid-IR colors $[3.5]\mbox{--}[4.5]\gt 0$ as AGN, while those with redder colors are either passive or star-forming galaxies. In this color space, the Phoenix cluster (McDonald et al. 2012a) is the only BCG harboring a strong AGN. At most, we estimate that ∼four clusters in this sample may harbor strong AGN at their centers, based on this mid-IR color selection. This lack of AGN contamination is further confirmed by visual inspection of the Chandra data for each cluster, which show a general lack of strong X-ray point sources coincident with the BCG in all clusters, with the exception of Phoenix.

Zoom In Zoom Out Reset image size

We proceed with this work assuming that all systems identified as star forming are, indeed, star forming, but remain cognizant of the fact that some fraction of these may host an AGN. We return to this dilemma in the discussion section.

In Figure 3 we show the SFRs, derived via three different methods, for the sample of BCGs described in Section 2.1. This plot demonstrates that the WISE mid-IR data typically provide the best upper limits at low-redshift, but are unable to provide meaningful upper limits on the SFR for BCGs at z ≳ 0.5. At z > 0.5, the most restrictive limit generally comes from our deep u-band follow-up program, which was designed to achieve a redshift-independent sensitivity. For the most part, we are sensitive to SFRs higher than 10 M_⊙ yr⁻¹, with <10% of systems having limits higher than this threshold.

Zoom In Zoom Out Reset image size

In the right panel of Figure 3, we combine the constraints from the three different SF indicators. This plot shows a significant number of BCGs, with SFRs from 10 to 300 M_⊙ yr⁻¹ at z > 0.4. One quarter (7/31) of the star-forming BCGs have SFR > 100 M_⊙ yr⁻¹—of these, five are at z > 0.9. For comparison, prior to this work there were only four confirmed clusters with extinction-corrected SFR > 100 M_⊙ yr⁻¹ in their BCG: Abell 1835 (McNamara et al. 2006), RX J1504.1-0248 (Ogrean et al. 2010), MACS J1931.8-2634 (Ehlert et al. 2011), and the Phoenix cluster (McDonald et al. 2012a). The median SFR for the 31 star-forming BCGs identified over the full redshift range is 50 M_⊙ yr⁻¹. It is worth noting that five of the seven most star-forming BCGs (SFR > 100 M_⊙ yr⁻¹) are at z ≥ 0.9, despite the fact that only 15% of the clusters in this sample are at such high redshift. Likewise, three of the  most star-forming BCGs are at z > 1.1, despite this representing 6% of the sample.

We also show the average SFR as a function of redshift in five redshift bins, in the right panel of Figure 3. These averages, provided in Table 1, have uncertainties that are derived by bootstrapping errors on individual data points. For non-detections, we assume zero SFR for the lower-bound of the uncertainty and the upper limit for the upper bound. Thus, the ranges shown encompass both the statistical uncertainty in the mean from detections and the uncertainty in the true value of the non-detections. This analysis reveals a strong evolution in the average SFR in BCGs over the past nine Gyr. This increase in the average SFR is driven largely by a decrease in the number of non-detections at high-z, which we discuss next.

Table 1.  Average BCG Properties

z	Ncl	$\langle {M}_{*}\rangle$	$\langle \mathrm{SFR}\rangle$	$\langle \mathrm{sSFR}\rangle$
		(1012 M⊙)	(M⊙ yr−1)	(Gyr−1)
0.25−0.45	19	2.09	4.0−15	0.002−0.007
0.45−0.65	28	1.97	7.0−24	0.004−0.012
0.65−0.85	22	1.47	20−41	0.013−0.028
0.85−1.05	8	1.39	42−84	0.030−0.061
1.05−1.25	6	1.54	89−320	0.058−0.205

Note. Average values for BCGs in five redshift bins, as shown in Figures 3 and 5. The ranges quoted on $\langle \mathrm{SFR}\rangle$ and $\langle \mathrm{sSFR}\rangle$ correspond to the combined uncertainty in the measurements (bootstrapping errors), the uncertainty in the BCG choice, and the added uncertainty associated with stacking non-detections. The total number of clusters is less than the number of BCGs, due to the fact that we consider multiple potential BCGs for several clusters.

Download table as:  ASCIITypeset image

In Figure 4 we show how the fraction of BCGs with SFR > 10 M_⊙ yr⁻¹ (hereafter f_SF) has evolved over time. Within the sample presented here, this fraction evolves from ∼20% at 0.25 < z < 0.65, to ∼50% at 0.65 < z < 1.05, to ∼90% at z > 1.05. This evolution is visible in Figure 3 as a relative lack of non-detections at the highest redshifts compared to the lowest redshifts. This result indicates that the fraction of clusters harboring a starburst in their central, most massive galaxy grows by a factor of ∼3 between z ∼ 0.6 and z ∼ 1.1.

Zoom In Zoom Out Reset image size

Even more intriguing is that, at z ∼ 0, f_SF ∼ 0%–10% (Donahue et al. 2010; Fraser-McKelvie et al. 2014), suggesting an even more dramatic decline since z > 1. In Figure 4, we compile estimates of f_SF from various surveys, spanning 0 < z < 0.6. These surveys include both optically selected (McDonald 2011; M. McDonald et al., in preparation) and X-ray-selected (Donahue et al. 1992, 2010; Hoffer et al. 2012; Rawle et al. 2012; Fraser-McKelvie et al. 2014; Donahue et al. 2015) clusters, and are being compared here to an SZ-selected sample. For the ACCEPT (Cavagnolo et al. 2009) and CLASH (Donahue et al. 2015) samples, we apply a small correction because both of these cluster samples are biased toward cool core clusters. This bias correction assumes that the true, underlying fraction of cool core clusters is 30% (Haarsma et al. 2010; Hudson et al. 2010; McDonald et al. 2013b) and that non-cool cores do not have star-forming BCGs at z ≲ 0.5 (Cavagnolo et al. 2008; O'Dea et al. 2008; McDonald et al. 2010). This well-motivated correction brings the observed value of f_SF in line with other, more representative samples. Encouragingly, at z ∼ 0.4 there are five distinct measurements of f_SF based on three different selection methods, which all agree that the fraction of star-forming BCGs is 20 ± 5%. The agreement between our measurement of f_SF from SPT at 0.25 < z < 0.5 and these earlier works suggests that our methodology is sound.

We show, in Figure 4, a linear fit to the star-forming BCG fraction, f_SF, as a function of redshift. The best-fitting line passes through the origin and has a slope of 0.57, implying a rapid evolution in f_SF over the past ∼9 Gyr. Such a fit is not physically well-motivated, and does not capture the rapid growth at late times. However, regardless of the choice of parametrization, it is clear from Figure 4 that the fraction of clusters harboring a strongly star-forming BCG at z ∼ 1 is significantly (>3σ) higher than at z ∼ 0. This evolution is perhaps unsurprising, because galaxies in general were more star forming at early times. To address this comparison, we next compare the BCG evolution to that observed in field galaxy and other cluster members over similar redshift intervals.

When comparing BCGs to other galaxies, it is necessary to normalize by the stellar mass of the individual galaxy. For a typical BCG, with M_* ∼ 10¹²M_⊙, an SFR of 10 M_⊙ yr⁻¹ is negligible in terms of contributing to the overall mass of the cluster, requiring 100 Gyr to double the stellar mass. However, this same SFR in a low-mass galaxy like M82 is enough to power massive outflows and modify the galaxy morphology and stellar content on short (Myr) timescales (Förster Schreiber et al. 2003). Thus, if we want to compare BCG evolution to field galaxy evolution, we must consider instead the specific SFR, or sSFR, defined as sSFR $\equiv$ SFR/M_*.

In Figure 5 we show the sSFR for BCGs in this work, as well as BCGs in similar-mass, low-z clusters from Haarsma et al. (2010) and Fraser-McKelvie et al. (2014). In calculating the sSFR, we use the stellar mass of the old population obtained in the SED-fitting process (see Section 2.3). For comparison, we show the average sSFR for field and cluster galaxies from Alberts et al. (2014). This earlier work showed that, at present, cluster galaxies have suppressed star formation compared to the field. However, at z ∼ 1.2, galaxies in clusters are as star forming as their field counterparts, and are evolving more rapidly. To match the analysis of Alberts et al. (2014), who calculate sSFR(z) by stacking far-IR data on galaxy positions, we estimate average sSFR values for our data by separately summing the total SFR and total stellar mass for galaxies in five redshift bins. For BCGs with upper limits on their SFR, we assume two limiting cases: the case where the SFR is equal to the upper limit, and the case where the SFR is zero. The net result of this stacking is shown in Figure 5 and Table 1, where the height of the black boxes represents the combined uncertainty in the measurements (bootstrapping errors), the uncertainty in the BCG choice, and the added uncertainty associated with stacking non-detections.

Zoom In Zoom Out Reset image size

Figure 5 demonstrates that, at 0.5 ≲ z ≲ 1, BCGs are evolving similarly to the cluster member galaxies, suggesting a common fueling mechanism. At low-z (z ≲ 0.5), the evolution of BCGs is less rapid than in the cluster environment, suggesting that the quenching processes acting on member galaxies (e.g., ram-pressure stripping, strangulation) may not be affecting the central galaxy. This change in slope may be due to the gas reservoir being replenished in BCGs by cooling of the ICM, which is likely fueling star formation in the lowest redshift BCGs (e.g., McDonald et al. 2011b). We will return to this idea in the discussion below. At z ≳ 1 there is marginal evidence that BCGs may be evolving more rapidly than the member galaxies, suggesting a preferential quenching in the cluster core.

At low-z, star-forming BCGs are, with very few exceptions, found in low-entropy, relaxed "cool core" clusters (e.g., Cavagnolo et al. 2008). This appears to still be the case for CLASH clusters out to z ∼ 0.4 (Donahue et al. 2015), and for individual clusters at higher redshift (e.g., McDonald et al. 2012a). However, it has not yet been established whether the presence of a cool core at high-z correlates with a star-forming BCG.

In Figure 6, we show the distribution of central entropy (K₀) as a function of redshift for clusters hosting the most star-forming BCGs compared to those hosting the most passive BCGs. Specifically, in each redshift bin we separately compute the average central entropy for the clusters harboring the 20% most star-forming BCGs (e.g., Perseus, Abell 1835, Phoenix, etc.) and for those harboring the 80% most passive BCGs. At z < 0.3, we show clusters from the ACCEPT sample (Cavagnolo et al. 2009), where the SFRs are inferred from archival infrared data (z < 0.1; Fraser-McKelvie et al. 2014) or the Hα line luminosity (z > 0.1; Crawford et al. 1999). We recomputed the central entropy in a 15 kpc aperture (∼0.015R₅₀₀) for clusters at z < 0.3 to match our more coarse, high-z aperture, and to avoid resolution bias (e.g., Panagoulia et al. 2014). We only include clusters with M₅₀₀ > 3 × 10¹⁴ M_⊙ (based on L_X–M relation from Pratt et al. 2009) to mimic the SPT selection. We find, in agreement with previous works (Cavagnolo et al. 2008), that the clusters hosting the most star-forming BCGs have a typical central entropy of K₀ ∼ 30 keV cm², while those hosting the least star-forming BCGs have K₀ ∼ 100 keV cm². As the redshift increases, the average core entropy in clusters hosting the most star-forming BCGs increases to ∼40 keV cm² (z ∼ 0.4), and then to ∼100 keV cm² (z ∼ 0.7). In the two highest redshift bins (z ≳ 0.6) there is no statistical difference between the distribution of core entropy in clusters with star-forming and passive BCGs. We find no strong evolution in the core entropy of the full sample, or in the subsample of clusters with passively evolving BCGs. This seems to suggest that, while ICM cooling is likely responsible for providing the fuel for star formation in low-z clusters, a different mechanism is responsible for star formation in the high-z BCGs.

Zoom In Zoom Out Reset image size

Another mechanism for forming stars in BCGs is via mergers with gas-rich galaxies. Under the assumption that such mergers happen most often shortly after the infall of a group or other massive halo, which provides an influx of new galaxies along orbits that may not be stable, we would expect BCGs with star formation being fueled by mergers to reside in clusters with disturbed X-ray morphology. In Figure 7, we test this scenario. For clusters in the SPT sample, we define X-ray morphology using the "a_phot" parameter, following Nurgaliev et al. (2013), with a relaxed critereon of a_phot < 0.1. This quantity is less biased to signal to noise than other indicators, such as power ratios and centroid shift, but is consistent with these in the limit of high signal to noise, as is demonstrated by Nurgaliev et al. (2013). At low-z, we use the recently compiled list of "symmetry" measurements from Mantz et al. (2015), using overlapping clusters to determine a common "relaxed" criterion. As before, for clusters at z < 0.1 and 0.1 < z < 0.3, we use BCG SFRs from Fraser-McKelvie et al. (2014) and Crawford et al. (1999), respectively, cutting on mass (M₅₀₀ ≥ 3 × 10¹⁴M_⊙) in order to ensure uniformity in the samples.

Zoom In Zoom Out Reset image size

Figure 7 confirms that, at z < 0.3, the most star-forming BCGs tend to reside in relaxed, cool core clusters. The star formation in these BCGs is most likely being fueled by the cooling ICM, where cooling flows are most commonly found in relaxed clusters. On the contrary, at z ≳ 0.6, the most star-forming BCGs are found in clusters with disturbed X-ray morphology, with ∼90% of the most star-forming BCGs at z ≳ 0.6 being found in such systems. This implies that, at early times, star formation in the BCG is more strongly correlated with the dynamic state, rather than the cooling state, of the ICM—the inverse to what is observed in the nearby universe.

Figures 6 and 7 suggest that there is a transition from star formation in low-z BCGs being linked to low-entropy, cool core clusters, to star formation in high-z BCGs being related more to a disturbed cluster morphology than to the cooling properties of the ICM. Below we will discuss possible interpretations of this result, and others presented thus far, while also addressing any potential systematic biases in this study.

Based on Figures 6 and 7, there appears to be an evolving connection between star formation in the BCG and the host cluster morphology. In order to investigate the fuel source for this star formation (i.e., cooling, gas-rich mergers, etc.), and determine whether it is linked to the dynamical state of the cluster, we require rest-frame UV imaging at significantly higher angular resolution. Fortunately, many of the high-z clusters in this sample have been observed by the HST as part of various weak-lensing programs (PIs: Stubbs, High, Schrabback). These observations are all in the F606W filter, which corresponds to rest-frame u-band at z ∼ 0.7 and <3000 Å at z > 1. These data provide a detailed view of the young stellar populations in a subsample (10/14) of high-z star-forming BCGs.

In Figures 8 and 9, we highlight the X-ray and UV morphology of the cluster and BCG, respectively, for all 10 clusters with available rest-frame u-band (or bluer) HST imaging at z > 0.7. For each cluster, we also show a near-IR image of the core region to demonstrate that these star-forming galaxies do indeed have the highest stellar mass of all the galaxies in the core, which is consistent with being the BCG. We divided these 10 clusters by X-ray morphology into two subsamples: relaxed (a_phot < 0.3 and K₀ < 95 keV cm²) and unrelaxed (a_phot > 0.3 or K₀ > 95 keV cm²). The five most relaxed clusters (Figure 8), in general, have relatively smooth, symmetric UV morphologies that resemble the underlying old stellar (near-IR) distribution. We caution that much of this near-UV emission may originate in the old stellar populations (see e.g., Hicks et al. 2010)—we require either far-UV imaging or equally high angular resolution imaging in a near-IR band (allowing subtraction of the old population) to determine the morphology of the excess UV emission due exclusively to young stars. Two of the BCGs shown in Figure 8 (SPT-CLJ0000-5748, SPT-CLJ2043-5035) appear to be in the midst of major mergers, based on both the near-IR and near-UV imaging, despite the fact that these two BCGs reside in the two most relaxed high-z clusters in our sample. None of the star-forming BCGs in this high-z, relaxed cluster subsample show evidence for extended, asymmetric filaments of star formation, which are commonly found in analogous low-z systems (e.g., O'Dea et al. 2010; McDonald et al. 2011b; Donahue et al. 2015, Tremblay et al. 2015). Instead, the UV emission is concentrated in the BCG center for four out of five BCGs, perhaps indicating that these systems are experiencing either nuclear starbursts or are AGN misidentified as star-forming galaxies.

Zoom In Zoom Out Reset image size

On the other hand, in the five least-relaxed clusters (Figure 9), the UV emission in and around the BCG is clumpy and filamentary. In particular, SPT-CLJ0102-4915 and SPT-CLJ2106-5844 exhibit clumpy UV emission extending 61 and 57 kpc from the BCG center, respectively. This is comparable to the most extended star-forming filaments found in Abell 1795 (50 kpc; McDonald & Veilleux 2009) and Perseus (60 kpc; Conselice et al. 2001; Canning et al. 2014). In all five of these BCGs, there are a minimum of two distinct UV peaks—in several systems, the brightest UV peak is offset from the BCG nucleus. Unlike the five BCGs in the more relaxed clusters, none of these systems are consistent with being purely AGN, given their complex morphology.

Zoom In Zoom Out Reset image size

While incomplete, these HST data suggest a qualitative difference in UV morphology between BCGs in relaxed, cool core clusters and those in unrelaxed systems. In relaxed, cool core clusters, where low-z surveys tend to find star-forming BCGs, we find (in four out of five cases) smooth, centrally concentrated UV morphologies. In the more morphologically disturbed systems (based on X-ray imaging), we find (in five out of five cases) clumpy or filamentary UV emission, with several (≳3) distinct emission peaks. Follow-up studies involving deep, high spatial resolution imaging at both far-UV and near-IR of a larger sample of high-z BCGs are necessary to determine whether this emerging trend is merely coincidence or evidence for a link between central star formation and the dynamical state of the cluster core.

The results presented in Section 3 suggest that, at high-z, the potential for star formation in the BCG is maximized when that BCG belongs to a dynamically active cluster. This is contrary to the established wisdom—based on numerous studies of low-z clusters—that the most relaxed, cool core clusters tend to harbor the most star-forming BCGs (e.g., Crawford et al. 1999; Edwards et al. 2007; Cavagnolo et al. 2008; Donahue et al. 2010). We proceed with a discussion of these results under the assumption that the X-ray morphology traces the dynamical state of the cluster, or, more specifically, that morphologically disturbed clusters in the X-ray are undergoing (or have recently undergone) a major merger.

At early times, the cores of galaxy clusters contained a higher fraction of star-forming galaxies than they do today—this is known as the "Butcher-Oemler Effect" (Butcher & Oemler 1984). This trend continues with increasing redshift, such that at z ∼ 1.4, the SFR in the field and in the cluster environment are indistinguishable (Brodwin et al. 2013; Alberts et al. 2014; Wagner et al. 2015, see also Figure 5). Thus, we would expect there to be a significantly higher fraction of gas-rich galaxies in our high-z subsample than in the low-z systems, particularly in the inner cores where gas depletion is most effective at low-z. Regardless of the impact angle of a cluster–cluster merger (as long as it is a bound system), the cores will typically pass through one another within ∼2 crossings (e.g., Ricker & Sarazin 2001; Poole et al. 2006). The crossing of these two cores will result in both an increase in the effectiveness of ram-pressure stripping, due to the high relative velocities of the cores, and an increased rate of galaxy–galaxy mergers and harassment. Each of these elevated processes result in the removal of cool gas from member galaxy halos. Assuming that this gas remains cool, it would lead to an increase in the availability of fuel for star formation in the cores of merging clusters. This is not the case in low-z systems, however, because galaxies in the cores of low-z clusters tend to be gas-poor (hence the low SFRs), so very little gas is available to be removed from their halos and contributed to the core.

With this scenario in mind, we re-examine Figure 5, modeling the observed evolution in BCG star formation with two distinct evolutionary components. These models, shown in Figure 10, assume that the evolution can be described as the sum of two power laws in time ($\langle \mathrm{sSFR}\rangle \propto {e}^{t/\tau }$). The best-fitting models have decay times of 4 Gyr (low-z) and 0.7 Gyr (high-z), compared to 1.5 and 2.2 Gyr for high-z cluster and field galaxies, respectively, (Alberts et al. 2014). This model is meant to provide a qualitative assessment of what may be driving star formation in BCGs over time—there is no shortage of models with an equal number of free parameters that would provide an equivalently good fit to these data.

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

The models shown in Figure 5 were chosen to illustrate two epochs of declining star formation. From z ∼ 1.4 to z ∼ 0.8, the sSFR declines more rapidly (τ = 0.7 Gyr) than both the field (τ = 2.2 Gyr) and cluster (τ = 1.5 Gyr) environment, implying that the BCG is quenched more rapidly than the other cluster members. This is consistent with an extrapolation from the field to the cluster environment from Alberts et al. (2014), indicating that (unsurprisingly) quenching of star formation is a strong function of local galaxy density. At late time, from z ∼ 0.6 to z ∼ 0, the sSFR of BCGs is evolving more slowly (τ = 4 Gyr) than both the field and cluster environment. Considering that field galaxy evolution is approximately passive, this implies that BCGs have an additional source of fuel—presumably the cooling ICM in relaxed systems. The non-zero slope of this evolution may be due to an ever-improving balance between ICM cooling and AGN feedback in the cores of galaxy clusters, leading to a reduction in the efficiency of ICM cooling over time.

In summary, there appears to be a transition in the source of fuel for star formation in BCGs over the past ∼10 Gyr. In nearby clusters, star formation is likely to be fueled by the cooling ICM, and regulated by feedback from the central AGN. At early times, the most star-forming BCGs are found in dynamically unrelaxed clusters, suggesting that star formation may be predominantly fueled by interactions with other galaxies, similar to other, non-BCG members.

Several recent studies have suggested that the condition for thermal instability in the hot ICM can be described as t_cool/t_ff = 10, where t_cool is the cooling time and t_ff is the local free-fall time (Gaspari et al. 2012; McCourt et al. 2012; Sharma et al. 2012; Gaspari et al. 2015; Voit & Donahue 2015; Voit et al. 2015). This threshold can be interpreted as the ratio of the cooling time to the mixing time. Assuming that AGN feedback is anisotropic, preventing local thermal instabilities requires the heated gas to mix with the cooling gas on timescales shorter than the local cooling time.

In order to cross this threshold, one of two conditions can be met. The gas can cool, reducing the local cooling time, t_cool. In the precipitation-driven feedback scenario of Voit et al. (2015), this leads to rapid precipitation of cold clouds, which fuel AGN feedback, leading to an increase of t_cool. This cycle can repeat indefinitely, until a more energetic process drives t_cool to much larger values (≫1 Gyr). Alternatively, if the free-fall time is increased, gas at a fixed density and temperature will more readily condense out of the hot phase. This criteria is, in principle, met if the dense core of a galaxy cluster is dislodged from the minimum of the dark matter potential, as is the case during an interaction with another massive group or cluster. It remains unclear, however, whether increasing the free-fall time of the gas in this way would, in fact, lead to more favorable cooling. Strictly speaking, a dislodged cool core would have a lower value of t_cool/t_ff, but the simulations that arrived at the threshold of t_cool/t_ff = 10 for thermal instability assumed relaxed clusters. In the case of a relaxed cluster, t_ff is a proxy for the convective timescale of the hot gas. However, in a merging cluster, bulk flows will likely drive mixing on faster timescales, perhaps leading to less efficient cooling.

This represents an alternate scenario that would lead to a link between enhanced star formation in the BCG and an unrelaxed dynamical state of the cluster core at early times. We lean toward the merger-induced SF scenario to explain the uptick in star-forming BCGs at early times, but stress that follow-up studies, including spatially resolved far-UV (e.g., HST) and far-IR (e.g., ALMA) imaging and optical spectroscopy, of these BCGs are necessary to provide additional insights into the mechanism for enhancing star formation in these high-z BCGs.

As we discussed in Section 2.6, it is possible that many of the BCGs we have identified as star forming may, instead, be AGN. With the inhomogeneous dataset in hand, it is challenging to differentiate between starburst and AGN, or composite systems, in a uniform way for the full sample. Thus, it may be that both the estimated fraction of star-forming BCGs and the absolute SFRs of these BCGs is biased high.

Considering only clusters at z > 0.75 (the two highest redshift bins in Figure 4), we detect star formation in 9–10 out of 22 (41%–45%) BCGs, where the uncertainty represents cases where multiple BCGs were identified in a given cluster. This is a factor of two higher than the fraction of star-forming BCGs at z ∼ 0.4. If we consider only the subsample of clusters with HST follow-up and label systems with symmetric, nuclear UV emission as non-star-forming (see Figures 8 and 9), we find evidence of extended, asymmetric star formation in 4 of 10 (40%) of BCGs. Given the reduction in sample size by requiring HST follow-up, the uncertainty on this fraction is larger, with the 95% confidence interval being 16.7%–69.2%. While this is consistent with the observed star-forming fraction of ∼18% at z ∼ 0.4 (McDonald 2011; Donahue et al. 2015), it is inconsistent with the lower value of ∼5% measured at z ∼ 0 (Donahue et al. 2010; Fraser-McKelvie et al. 2014). Thus, while we can say with high confidence that the fraction of star-forming BCGs has evolved significantly from z ∼ 0 to z ≳ 0.4, our inability to differentiate between central AGN and nuclear starbursts limits our ability to say with certainty whether this fraction continues to grow.

It is important to note, however, that the fraction of nearby BCGs harboring AGN that are bright in the UV or IR continuum is low—the vast majority of cluster-centric AGN are radio galaxies that appear quiescent at most other wavelengths. If the high fraction of systems identified here as star forming are instead AGN, it implies a high accretion rate onto the supermassive black hole at early times (e.g., Russell et al. 2013). Of the hundreds of known BCGs at low-z, only a handful of systems appear to harbor rapidly accreting, radiatively efficient AGN (e.g., Russell et al. 2010; O'Sullivan et al. 2012; Ueda et al. 2013; Kirk et al. 2015; Reynolds et al. 2014; Walker et al. 2014). Thus, we can say with confidence that there is a dramatic increase in the amount of "activity" in BCGs from z ∼ 0 to z ∼ 1—whether that activity refers to massive bursts of star formation or the rapid growth of central supermassive black holes remains an open question.