STAR FORMATION AND UV COLORS OF THE BRIGHTEST CLUSTER GALAXIES IN THE REPRESENTATIVE XMM-NEWTON CLUSTER STRUCTURE SURVEY

Optical spectra of the objects were obtained with the newly installed Goodman spectrograph (Clemens et al. 2004) on the 4.1 m SOAR Telescope in Cerro Pachón, Chile. The observations were taken between 2008 September and 2009 May. For all objects, a 168 wide slit and a 600 lines per mm grating was used, for a wavelength coverage of 640–915 nm. Beginning 2009 April 17, a blue blocking filter (GG-495) was used to prevent second-order scattering of blue light onto the CCD, which was not a serious issue for any of these data. All data were taken during photometric observing conditions. All observations were taken under excellent conditions, nearly photometric, and good seeing (<1''). Our strategy was to observe each object twice: once along the major axis and once along the minor axis. Due to hardware limitations for rotating the spectrograph, a fully perpendicular observation was not always possible. Information about the observations can be found in Table 4. Observations of spectrophotometric standard stars (Hamuy et al. 1992) were made for flux calibration (Table 4).

Data reduction consisted of CCD overscan subtraction, trimming the image. The Goodman spectrograph suffers from 20% peak-to-peak spectroscopic fringing redward of about 7000 Å. We used normalized quartz flats taken directly after the completion of each exposure to remove this fringing signature. However, the fringing pattern shifts and distorts, probably due to flexure of the instrument. The fringing redward of 8000 Å was the most difficult systematic to remove. Since the wavelength scale of a fringe feature was broader than a typical emission line, it was straightforward to determine which sources had emission lines brighter than ∼1–2 Å in equivalent width (EQW). Wavelength solutions were obtained based on night sky lines. We measured redshifts, line widths, and equivalent widths, and estimated the Hα line luminosity.

We extracted single spectra with widths of 20–40 pixels, corresponding to 3''–6'' in the spatial direction. We detected emission lines (Hα, [N ii]6584/6548 Å, [S ii]6717/6730 Å, and usually [O i]6300 Å) from seven BCGs out of 31. (The 32nd, the BCG in RXCJ1311.4−0121, also known as Abell 1689, is discussed in the following section.) We report upper limits and Hα equivalent width detections in Table 4 and emission-line properties in Table 5. Error estimates incorporate counting statistics and fit uncertainties based on bootstrap simulations of the spectra, accomplished with the task splot in IRAF (Tody 1993). The residual fringing modulates red Goodman spectrum continua on wavelength scales somewhat broader than typical line widths, and it represents the most serious source of systematic uncertainty. Upper limits are thus typically 1–2 Å. For Hα emission lines, underlying stellar absorption can affect the total flux estimate. We have added 1 Å of uncertainty in quadrature to our Hα equivalent width uncertainties. This uncertainty only matters for the faintest systems. The two most prominent emission-line systems, RXCJ1141.4−1216 and RXCJ2014.8−2430, were also the bluest in UVW1 − R and were the two best detections in the bluest OM UV filter, UVM2. RXCJ2014.8−2430 and RXCJ1141.4−1216 were also the brightest GALEX NUV sources, 19.84 ± 0.085 and 19.68 ± 0.09 in AB magnitudes. They represent the most luminous GALEX NUV sources in REXCESS as well.

To complete the spectroscopic coverage of a relevant comparison sample of X-ray bright clusters of galaxies, the Brightest 55 X-ray Cluster sample (B55; Piccinotti et al. 1982), we report here upper limits for BCG Hα emission lines, as obtained using the Modular Spectrograph¹⁰ on the 2.5 m DuPont Telescope on Las Campanas. These observations were taken and reduced as described in Donahue & Voit (1993). One of these clusters, Abell 1689 (RXCJ1311.4−0120), is also in the REXCESS sample. We provide here upper limits of Hα equivalent widths of <1 Å for four additional BCGs in clusters with unknown emission-line properties: Abell 1650, Abell 3571, Triangulum Australis, and Abell 4038. Abell 4038 is also known as Klemola 44. Our results are reported in Table 6. Other optical emission-line measurements for the B55 sample are provided in Peres et al. (1998), Crawford et al. (1999), and Cavagnolo et al. (2008). Non-detections of emission-line signatures from 3C129 and Abell 3532 (Klemola 22) are reported in Spinrad (1975) and Cristiani et al. (1987), respectively.

Because we have reliable X-ray measurements for every cluster and because the X-ray sample is selected specifically to calibrate scaling relationships, our sample is particularly well suited for studies of possible relationships between BCG star formation and BCG UV–optical colors, but also among BCG mass, luminosity, cluster mass, and cluster dynamic state. Our emission-line-based classification of inactive versus active is consistent with that used to classify elliptical galaxies for GALEX galaxy studies. Although our original intent for our BCG study was to examine the stellar mass (Haarsma et al. 2010) and star formation signatures (this paper) in these BCGs, the scope of this paper is expanded to include quantifying the contribution of the UV from the older population, since an excess UV attributed to star formation must be over and above that contributed by old stars. That we detected nearly all of the BCGs in the UVW1 band of the XMM-Newton OM means at least some of them are dominated by old stellar populations.

All BCGs seem to have at least some UV emission despite their "red and dead" reputation. Elliptical galaxies exhibit differing amounts of UV shortward of 2500 Å, sometimes called the UV upturn, because L_λ slopes upward to shorter wavelengths (see O'Connell 1999 for a review and references.) Brown et al. (2000, 2002), using Hubble Space Telescope (HST) observations, showed the culprits for this emission turn out to be eHBs. The UV continuum light from galaxies therefore can be produced by two different populations of stars: (1) the most massive, recently formed stars and (2) eHBs. In order to assess the amount of excess UV attributable to recent star formation, we need to establish the baseline contribution of the old population. This task is not as easy as it might sound. As long as these old UV stars are numerous, the UV–optical colors of an inactive BCG should be relatively constant for BCGs of similar metallicity, but observations show that the UV colors of ellipticals exhibit scatter. One explanation is that the UV–optical color of an elliptical galaxy is exquisitely sensitive to any star formation in the last 100 billion years. Studies of inactive elliptical galaxies show that tiny amounts of star formation can induce scatter in UV–optical or near-infrared colors as large as observed, even though their line emission is below detection thresholds similar to those of our study (e.g., Yi et al. 2005; Kaviraj et al. 2007; Rawle et al. 2008). The REXCESS sample includes BCGs with and without line emission, so we discuss the implications of our data for both populations.

Commonly used population synthesis codes, such as Bruzual & Charlot (1993) or Starburst99¹¹ (Leitherer et al. 1999), do not predict the UV spectrum of an old stellar population (e.g., Magris & Bruzual 1993). Therefore, in order to explore the baseline UV contributions of an old population, we turned to empirical UV spectral templates (M60, M49, and NGC 1399) to generate the inferred observed broadband colors as a function of redshift in Figures 2–4. Our results suggest that BCGs hosting old populations with little to no star formation have a limited range of UV–optical colors and that BCGs with UV–optical colors bluer than this range host the strongest emission-line systems, and presumably, recent star formation. Whether the latter is true or the Hα emission comes from a low-luminosity AGN, accretion of cold gas by the BCG has taken place.

The elliptical galaxies for which we have templates host old stellar populations, as confirmed by the presence of Hβ equivalent widths, 1.4–1.6 Å in absorption (Kuntschner 2000; Kuntschner et al. 2001). The templates demonstrate that over the redshift range of our sample, there is very little color change arising from shifting the baseline spectrum as a function of wavelength, so we do not k-correct our UV magnitudes, in order to stay as close to the data as possible. The templates also show that for a z = 0.1 galaxy, the UVW1 bandpass sits at the local minimum between the UV-upturn feature shortward of 250 nm and the near-UV (250 nm rest) emission from normal stars (Figure 1). Therefore, the sensitivity of the UVW1–optical color to the behavior of eHBs is minimized compared to UV–optical colors based on the GALEX filters. The templates also show variation in this color between them.

The inactive BCGs in our sample seem to represent a sample of galaxies with less dispersion in their UV–optical colors than that of ellipticals as a whole, and their colors are consistent with colors of inactive ellipticals. Whether this uniformity means that they have much less contamination from star formation or are more uniform in their metallicity and age than their lower-mass cousins is an open question.

On the other hand, the BCGs in our sample with emission lines show a consistent correlation between excess UV emission (over that expected from an old population) and the strength of the Hα emission line. This correlation suggests that star formation is occurring in the subset of BCGs where we expect star formation (e.g., Cavagnolo et al. 2008; Dunn & Fabian 2008): the CCs with the shortest central ICM cooling times. The UV–optical colors of these objects therefore appear to be sensitive to star formation.

BCGs with a UV excess over and above what is expected from an old population are more likely to exhibit Hα emission. Furthermore, the line strength of Hα is strongly correlated with the UV excess (Figure 5).

Zoom In Zoom Out Reset image size

Seven BCGs in our sample exhibited detectable Hα emission (also [N ii]6548 Å, 6584 Å, [S II]6717 Å, 6731 Å, and occasionally [O i]6300 Å). All of these BCGs inhabit clusters identified as CC in Pratt et al. (2009), and all of these are associated with a 1.4 GHz radio source. All three CCs in Pratt et al. (2009) that were not detected in Hα have central cooling times estimated from XMM-Newton observations (at 0.03 R₅₀₀) longer than 10⁹ yr, while only two out of seven in the detected sample have cooling times this long.

Six of these seven emission-line BCGs also showed a UV excess, as defined by having a UVW1 − R color significantly bluer (UVW1 − R <4 or NUV − R <5.7) than that expected from ellipticals with no star formation (UVW1 − R ∼5, NUV−R∼6). The brightest two of these were the only two galaxies in the sample with detected OM UVM2 flux at a statistical confidence >5σ. Five of the six emission-line galaxies with public GALEX NUV observations were detected in the NUV, whereas 11 of the 17 non-emission-line BCGs with such observations were detected, and none of them with m_NUV < 20.

Cluster ICM core properties of electron density (n_eh(z)⁻²) and cooling time are strongly correlated with Hα equivalent widths. In Figures 6–8, we show two strong correlations (with central electron density and cooling time) and, for comparison, a weak correlation (with cluster mass). We explored several tests of correlation, using tasks in the survival statistics package in IRAF/STSDAS (Isobe et al. 1986). Survival statistics utilize information in the upper limits as well as the detections. The measures of correlation (or lack thereof) were consistent from test to test. A generalized Kendall's τ test and Spearman's ρ test showed the probability that the equivalent width of Hα was not correlated with central electron density or cooling time was <0.0001. A similar set of tests showed that the probability that the equivalent width of Hα was not correlated with the UV–optical color was 1%–2%. Similarly, pairing UV–optical colors with central electron density or cooling time produced probabilities of ∼0.02 that the paired properties were not correlated. The UV correlations with Hα, central electron density, and cooling time are undoubtedly diluted by the fact that most of the galaxies in our sample have UV emission dominated by an old stellar population. For contrast, the same statistics for the correlation of X-ray mass and Hα equivalent width gave a probability ∼0.33–0.38 that it was not correlated, which confirms the visual impression.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

We computed the [N ii]6583/Hα ratios for the seven BCGs with Hα emission. We note that the [N ii]/Hα ratio is unusually high for H ii regions, but in the range observed for cooling flow BCGs (e.g., Heckman et al. 1989; Figure 9). This ratio, which is sensitive to the temperature and excitation of the emission-line gas, trends with the relative strength of the Hα line, in equivalent width here. The most luminous Hα systems have lower [N ii]/Hα ratios than the weak systems. The less luminous emission-line systems may be increasingly dominated by excitation processes such as cosmic-ray heating, conduction, and even weak shocks (e.g., Ferland et al. 2008), while the more luminous systems may have proportionally more heating from stars. Forbidden line emission can be elevated in photoionized systems with additional sources of heat (e.g., Voit & Donahue 1997). This interpretation is consistent with the correlation of UV–optical color and Hα equivalent width. For this work, we interpret Hα and forbidden line emission as signs of activity, at least some of which is associated with recent star formation, but we note the many alternate sources of Hα that may be present in these systems (Ferland et al. 2008; Heckman et al. 1989; Voit & Donahue 1990; Jafelice & Friaca 1996; Begelman & Fabian 1990).

Zoom In Zoom Out Reset image size

The SFRs based on a UV excess or an Hα luminosity in terms of a specific SFR are model dependent. Hα emission is sensitive to the most recent star formation, since only the hottest, shortest-lived stars ionize hydrogen. Hα rates are therefore closer to being an instantaneous measure of star formation. In contrast, the UV is sensitive to a broader range of hot stars, and therefore the age of the burst affects the magnitude of the SFR. The UV is easily obscured by gas and dust, and represents a lower limit, modulo the age assumption. Furthermore, the lack of certainty about the exact level of UV light contributed by eHBs affects the estimate of UV excess for the fainter systems (the level of the baseline is irrelevant for strong UV excesses.) Finally, any given amount of UV excess light can be associated with a nearly limitless amount of stars associated with a single burst, since a tiny amount of UV can represent the UV tail of a huge burst, say 10⁸ yr ago. The UV rates computed here are based on an assumption of continuous star formation, which provides SFRs at the low end of the scale (Kennicutt 1998).

The excess UV of the two bluest galaxies in UV–optical light is consistent with that emitted by an unobscured population continuously forming stars for timescales longer than 10⁸ yr, at about 2 solar masses per year for RXCJ1141.4−1216 and 8 solar masses per year for RXCJ2014.8−2430, based on a UV excess over a baseline color of UVW1 − R =  4.7 − 5 (Kennicutt 1998). Similar SFRs are inferred from relations produced by models in Hirashita et al. (2003), in which the onset of constant star formation ranges between 10⁷and10⁸ yr in the past (Table 5). The rates for the more recent times of onset are about 60% higher because the UV luminosity increases to a maximum at a population age of 10⁸ yr (Hirashita et al. 2003). These rates are underestimates, since we have not taken into account attenuation of the UV by the ISM of the host galaxy.

Hα is produced by ionization by the hottest, youngest stars, and therefore the SFR based on Hα emissions is an indicator of the nearly instantaneous SFR. Table 5 shows that the Hα rates are consistent with the range of derived UV rates. The approximate diameter of the brightest two emission-line systems is ∼10 kpc (of order twice the slit width), similar to that of the emission-line nebula in Abell 2597 (Voit & Donahue 1997). Given the unknowns in the estimation process (attenuation, star formation history, slit for an extended source Hα measurement), any differences in the rates can easily be explained. For example, somewhat higher UV rates can be explained by an older starburst and slit losses, and somewhat higher Hα rates can be explained by moderate UV attenuation. Therefore, to first order, the production of UV (in excess of that from an old population) and optical emission lines in these systems is not pathologically different from that seen in star-forming galaxies.

The main result is that the appearance of emission lines together with a UV excess is consistent with star formation being the source of both. None of these galaxies has an X-ray or optically bright AGN; however, since eight out of 10 (nine out of 11 if one includes RXCJ0211.4−4017 as a CC) CC clusters in this sample have radio-loud AGNs, some of the Hα may be associated with the AGN. For example, cosmic-ray heating from relativistic particles may be important in generating some of the Hα emission (Ferland et al. 2008), especially in the weaker systems, in which the larger [N ii]/Hα (>1) ratios suggest that the interstellar gas may be hotter than typical of regions photoionized by stars. We showed in the previous section that the Hα equivalent width is correlated with the UV excess, so at least some of the Hα luminosity is likely generated by UV-producing stars.

The incidence of emission-line sources in the REXCESS sample (22%) is slightly low compared to the incidence of Hα-emitting systems in other low-redshift cluster samples, but not remarkably so. The B55 sample (Piccinotti et al. 1982; Peres et al. 1998), including the spectroscopic observations reported in Table 6, has an incidence rate of 38% for the now complete sample. For the sample of X-ray clusters in Crawford et al. (1999), the incidence was 31.5% ± 4%. Bauer et al. (2005) show that in the ROSAT Brightest Cluster Sample/extended Brightest Cluster Sample (BCS/eBCS) (Ebeling et al. 1998) sample with z > 0.15, 13 of the 33 clusters with public Chandra data and available optical spectroscopy had Hα, an incidence rate of 39%. A study of luminous extended medium sensitivity (EMSS) X-ray clusters (Stocke et al. 1991; Gioia et al. 1990), at z = 0.07 − 0.4 from the Einstein Observatory (Donahue et al. 1992) showed rates of 40%–50% in a flux-limited, complete, but high X-ray luminosity, sample. They reported only weak evidence for lower incidence at the highest redshifts of this study, so there was very little evidence for evolution in the incidence rate for z < 0.4. In contrast, Vikhlinin et al. (2007) claim that there is a lack of X-ray evidence for CCs at high redshift (z > 0.5), which also implies that very few high-redshift BCGs will host emission-line systems.

BCGs of optically selected clusters have a much lower incidence rate of emission lines. von der Linden et al. (2007) found an incidence rate of ∼15% in a study of 625 Sloan Digital Sky Survey BCGs in optically selected clusters (Miller et al. 2005) with a control sample matched in absolute magnitude, redshift, and color. This rate was not much higher than that of the control sample. However, this study omitted the most massive BCGs because of the lack of truly massive galaxies in their control sample. Edwards et al. (2007) examine 93 clusters (selected somewhat heterogeneously, including both X-ray selected clusters and optically selected C4/DR3 clusters) between z = 0.010and0.067. Of these, 15% had emission lines, with no trend in galaxy magnitude or velocity dispersion. Of the "cooling flow" clusters, 71% ± ^9%_14% had optical emission; if those BCGs were within 50 kpc of the X-ray centroid, 85%–100% had optical emission. If the cooling flow clusters were omitted, only 10% of the sample had any line emission, similar to the control sample.

Statistically, the incidence rate in the REXCESS sample is somewhat lower than that of other X-ray samples (B55 and EMSS, for example) and similar to that of optically selected samples. The binomial probability of finding seven or fewer BCGs with emission lines given an expectation of 15% is about 9.8%, statistically consistent with the incidence rate in optically selected cluster surveys. If the expectation of detecting emission lines is 35%, the probability of finding seven or fewer is 4.5%, indicating a result about 2σ different from expectations for X-ray-selected surveys. REXCESS therefore has a somewhat lower number of BCGs with optical emission lines than the B55 (Crawford et al. 1999) or EMSS (Donahue et al. 1992) samples. The incidence of emission-line systems inside the CC sample (7 out of 10) is certainly consistent with the high incidence (∼70%) found in Edwards et al. (2007).

We note that the comparisons presented here are not of cluster samples matched in mass. The number density of clusters in the wide-area optical surveys is high, so their average masses must be lower and the range of masses sampled greater than the X-ray-selected surveys under discussion here. The B55 and EMSS samples are complete, flux-limited samples, but REXCESS is not. We do not see, to the limit of our small detection statistics, evidence for some dependence of the incidence rate of emission-line BCGs on the mass of the cluster, but the REXCESS sample does not include low-mass clusters or groups.

The explanation—if one is needed for a 2σ discrepancy—is not clear. X-ray flux-limited samples have on average higher-mass clusters than the larger optically selected samples. They may preferentially include objects with higher gas mass fractions for a given mass or objects with X-ray bright CCs (Pratt et al. 2009; Fabian et al. 1994; Edge et al. 1992). Analysis of the REXCESS sample itself shows no evidence whatsoever of any mass dependence of the presence of a CC (Böhringer et al. 2009). Certainly, the EMSS may have included a few more CC clusters because of its fixed-size sliding-box selection method, as discussed in Donahue et al. (1992). However, the B55 survey was selected by collimator flux, and therefore is unlikely to be biased toward CC clusters. The REXCESS selection had no specific morphological criterion, so one might expect the selection to be similar to that of the B55 sample. The relatively small sample size, however, does not allow hard conclusions to be drawn, since the similarity in REXCESS BCG emission-line incidence to that of an optically selected sample could be coincidental. Other X-ray samples are limited in their size, dynamic range in mass or redshift, and the degree to which the sample contents are representative of the full population of clusters of galaxies (Vikhlinin et al. 2007; Santos et al. 2008; Sun et al. 2007).

The discrepancy in the emission-line BCG incidence rates between optically and X-ray-selected samples does not appear to have a ready explanation in terms of incidence rates varying with cluster masses, X-ray temperatures, or BCG optical luminosity, at least to the limits probed by the REXCESS cluster sample. However, we do see a strong correlation with the presence of low-entropy gas, short central ICM cooling times, high central electron densities, and, as we will show below, relaxed dynamical state. If optical samples include a larger proportion of unrelaxed, disturbed clusters and a smaller proportion of clusters with CCs, then the lower incidence rates of emission-line BCGs in optical samplesare easier to understand. Larger and more inclusive studies, with attention to sample selection, are required for further progress in exploring the incidence of star formation in BCGs of clusters of galaxies of all types, in order to understand how and why the thermodynamic state of the intracluster gas is coupled to star formation in the BCG.

For reasons unknown, BCGs are surprisingly good standard candles, at least at low redshifts (Postman & Lauer 1995), and inactive BCGs may be an even more homogeneous set. In the following sections, we will explore the apparent similarity of the UV and optical properties in the inactive BCGs in our sample, as well as the tendency for outliers to host Hα emission-line systems. We will show that the UVW1 − R color, a quantity sensitive to metallicity and age of the stellar populations, is insensitive to BCG R-band luminosity and color in a metric aperture and to the redshift range spanned by our sample. Because the REXCESS clusters have homogeneous and high-quality X-ray measurements, we are also able to show that the UVW1–optical colors of inactive BCGs do not vary with halo gas temperature (equivalently cluster halo mass) or dynamical state.

6.4.1. Lack of Trends in UV–Optical and Optical–Infrared Colors

As we showed in Section 5, the UVW1 − R colors of the inactive galaxies are ∼4.7, with a typical reddest color of ∼5. The galaxies with emission lines comprise the outliers in the observed color relationships. Because this sample includes BCGs only, the dynamic range in stellar mass and optical luminosity is not large. Over the luminosity range spanned by our sample, we do not detect any trend in the observed UV–optical colors of the BCG sample with R magnitude or, if M/L is similar, stellar mass inside a 10 h⁻¹ kpc aperture (Figure 10). Notably, the emission-line systems identify nearly all of the systems that differ the most from the mean UV, optical, and near-IR relations for these BCGs.

Zoom In Zoom Out Reset image size

We note that the scatter in the corresponding rest-frame R − K colors for the 30 galaxies in our sample with 2MASS (Skrutskie et al. 2006) K-band measurements is only 0.15 (0.95 ± 0.15), from Haarsma et al. (2010). Excluding the two BCGs with red colors (R − K > 1.2), RXCJ1302.8−0230 and RXCJ2319.6−7313 (both of these are emission-line BCGs in CC clusters), the dispersion drops to 0.11 (0.92 ± 0.11), and is not much larger than the B − R scatter (∼0.06) in metric colors reported for a low-redshift BCG sample by Postman & Lauer (1995). A similar result is obtained if all REXCESS CC BCGs are omitted from the estimation of the mean color. The typical measurement uncertainty for R − K is about 0.10 mag, including systematics. Most of the statistical uncertainty arises from the 2MASS K-band photometry. But some ∼0.05 mag of this uncertainty is the systematic offset between SOAR and 2MASS photometry, and should not add to the scatter. Therefore, some of the scatter in R − K is intrinsic, and is related to the metallicity and age of the underlying populations. If R − K and UVW1 − R colors were sensitive to the same underlying condition (metallicity and age), there might be a correlation, with the UVW1 − R scatter being more sensitive. To test whether the scatter in R − K is correlated with the scatter in UVW1 − R, we plotted these colors in Figure 11. Since we see that UVW1 − R is not correlated with R − K for the inactive BCGs, we can conclude that whatever variations in the stellar population drive variations in the UV−R color (presumably metallicity and age), those variations are not detectable in R − K colors measured to 8% or better.

Zoom In Zoom Out Reset image size

The BCGs in the sample with the reddest outlier R − K colors (RXJ 1302.8-0230 and RXJ 2319.6-7313) exhibit somewhat bluer UV–optical colors (a UVW1 − R ∼  4, compared to a mean of 4.9). They show no obvious central point sources (indeed, no BCG in our sample does), and these two galaxies exhibit weak but detectable Hα emission. The Hα emission and a redder R − K color may be indicative of the contribution from a recently formed sub-population of stars. There are at least two counterarguments to this hypothesis. (1) The [N ii]/Hα ratios in these two systems are higher than that seen in H ii regions. As mentioned in Section 6.2, the high ratios in weak emission-line systems may indicate processes at work other than star formation. (2) The two most powerful emission-line BCGs (RXJ 2014.8-2430 and RXJ 1141.4-1216) exhibit R − K colors typical for the REXCESS sample, suggesting that excess flux from emission lines themselves is not sufficient to cause color anomalies. The equivalent widths of the emission lines in the two R − K outliers are also very small compared to the width of the R-band filter. Therefore, we have no ready explanation for the two outliers, aside from the fact they are notably BCGs with line emission, clearly not inactive.

6.4.2. Lack of Trend of UVW1–Optical Color with Redshift or Halo Properties

Our data show no trend of UV–optical colors with redshift, galaxy mass, or the mass of the cluster of galaxies. Figure 2 shows that inactive BCGs in the sample exhibit no correlation of UVW1–optical color with redshift. Within the range of absolute magnitudes probed by these BCGs, we do not see a trend with optical luminosity or R − K color. Since the optical mass-to-light ratios of these galaxies are likely to be similar, this lack of trend implies that the UVW1–optical colors of our inactive sample are also insensitive to the mass of the host galaxy. Figure 12 shows no correlation of UV–optical color with X-ray temperature for the inactive BCGs in our sample. Since T_X is strongly correlated with cluster gravitating mass, this result suggests the UVW1–optical color of the BCG is insensitive to the mass of the host cluster.

Zoom In Zoom Out Reset image size

In Figure 13, we show a plot of UV excess versus 〈w〉, where the latter is the standard deviation of the distance between the X-ray peak and emission centroid evaluated in increasing apertures, as detailed in Böhringer et al. (2009). This parameter has been shown to reflect quite closely the true dynamical state of a system in the simulations of Poole et al. (2006). As can be seen in Figure 13, there is no clear trend of UV excess with dynamical state.

Zoom In Zoom Out Reset image size

In summary, we can say that the lack of observed trends suggests that no matter what the level of undetected star formation in the inactive BCGs might be, the UVW1 − R colors of inactive BCGs are not affected by the degree of dynamical relaxation measured by the parameter 〈w〉, by the mass of the host cluster, or indeed by anything we have measured thus far.

The bluer UV photometry measurements, including UVM2 photometry from the XMM-Newton OM, and the NUV (with similar but broader than the UVM2 bandpass) and FUV photometry from GALEX, are more sensitive to the contributions of the UV-upturn population, the eHBs, than the UVW1 filter. However, for this analysis, the data were too shallow to detect most of the BCGs in the sample, and the colors even for the detected galaxies are not very accurate, especially for the inactive galaxies. Our primary interest in this paper is in the star formation properties of the BCGs and how they correlate with cluster and BCG properties, but we briefly discuss the potential of this sample for studying the properties of UV emission from older stars.

The mechanism for creating an eHBs is unknown, but a survey of the literature revealed two classes of possibilities for prematurely exposing the hot core of a horizontal branch star: (1) extreme winds, sensitive to metallicity (e.g., Yi 2008) and (2) mass transfer in binary stars (Han et al. 2007). A possibly relevant wrinkle to these mechanisms is based on the idea that helium preferentially settles in the cores of massive clusters, enhancing the UV produced by helium core-burning stars (Peng & Nagai 2009; Lee et al. 2005).

The wind-based explanations for the eHBs predict a trend of reddening UV–optical colors with redshift since z = 0.2 (e.g., Yi 2008). Whether variations in the relative UV component from galaxy to galaxy are indeed correlated with metallicity is still a matter of debate: Burstein et al. (1988) reported a correlation between Mg₂ index, considered a metallicity indicator, and UV excess, but Rich et al. (2005) see no correlation in a sample of 172 early-type galaxies. It is possible that star formation in lenticulars dilutes this relation in a sample of early types, as Donas et al. (2007) see a weak correlation in ellipticals, but none in lenticulars. However, Rawle et al. (2008) show that NUV−J colors of cluster red-sequence galaxies exhibit a strong correlation with metallicity and suggest that a relatively tiny amount of star formation could introduce the large scatter seen even in this color.

In their GALEX study of a dozen BCGs at z < 0.2, Ree et al (2007) report evolution of GALEX FUV−V colors. In contrast, the inactive BCGs (plotted without diamonds or squares in Figure 4) in our sample do not differ in their UV–optical or FUV–optical color in a significant way from each other or from the three low-redshift template galaxies, one of which (NGC 1399) is the bluest local galaxy in Ree et al. (2007). Assuming V − R ∼ 0.5, the 11 FUV − R measurements in our sample are somewhat bluer than the dozen galaxies discussed in Ree et al. (2007). The REXCESS non-detections must be redder than the detections, and therefore the colors of those galaxies likely overlap their measurements. However, the presence of inactive but blue BCGs at z ∼ 0.1 suggests that the evolution may not be so strong as suggested by the single star wind models. Alternatively, UV emission from recently formed stars, in galaxies without detectable Hα emission, would confound empirical tests of this model. The comparison to the Ree et al. (2007) data set is further compromised by several limitations. Ree et al. (2007) use estimated total optical magnitudes, while our colors are measured in fixed metric apertures. A total optical magnitude of a BCG is difficult to measure definitively, because they often have an extended halo of stars that fades into intracluster light, whose luminosity is comparable to that of the central portions of the BCG (Gonzalez et al. 2005). The UV emission seems more compact than the optical emission, so the choice of aperture size affects the inferred color. Furthermore, the shallowness of most of the archival FUV observations covering our sample means we can only detect the bluest examples of the brightest galaxies, with color uncertainties ranging from 0.1 to 0.4, and thus we may be detecting only the galaxies in our sample with recent star formation. Given those limitations of the archival data, the similarity of the colors may even be surprising. Deeper observations with more accurate FUV fluxes and compatible optical measurements are required for a comparison with the full REXCESS BCG sample, to probe the range of FUV colors in this sample. Accurate FUV−NUV colors would be a useful diagnostic in distinguishing the contribution of UV-upturn stars from non-ionizing but relatively young main-sequence stars.

The helium excess model, further developed by Peng & Nagai (2009), predicts that the amplitude of the UV upturn should correlate with the dynamical state of a cluster. A high helium abundance affects the production and the UV-upturn strength of eHBs (e.g., Dorman et al. 1995). In the model of Peng & Nagai (2009), the central helium abundance in a cluster of galaxies is enhanced by sedimentation in relaxed systems. If the UV emission in the inactive systems is dominated by these old stars (and not low levels of star formation), the lack of correlation with dynamical state suggests that helium sedimentation cannot explain the observed range of UV colors in these galaxies.

Intriguingly, a binary star model for origin of the UV upturn (Han et al. 2007) predicts an almost constant UV–optical color. In this model, binary interactions strip horizontal branch stars. This process is independent of metallicity and redshift.

All of the models are difficult to challenge with observations, because tiny amounts of star formation, impossible to detect with conventional emission-line diagnostics, may obscure or induce trends, particularly at 250 nm. It is notable, for example, that to reproduce the observed level of scatter in UV–optical or UV–infrared colors, Han et al. (2007) must add a second, younger stellar population to produce significant amounts of excess UV in an otherwise inactive BCG or elliptical. Disentangling the contributions of the various populations of these systems (commonly regarded as simple, single age, elderly populations in the past) requires more UV work, particularly spectroscopy, to obtain the needed diagnostics. Deeper GALEX photometry of the REXCESS BCG sample would be useful to determine whether there are empirical trends in these high-mass objects. More sophisticated simulations are required to determine whether trends predicted from the simpler models we have in hand could be detected under more realistic conditions.