Galaxy properties in low X-ray luminosity clusters at z = 0.25

Abstract

We present the first spectroscopic survey of intrinsically low X-ray luminosity clusters at z≫ 0, with Hubble Space Telescope (HST) WFPC2 imaging and spectroscopy from Calar Alto and WHT-LDSS2. We study 172 confirmed cluster members in a sample of ten clusters at 0.23 < z < 0.3, with L_X≲ 4 × 10⁴³h⁻² erg s⁻¹[0.1–2.4 keV] (Ω_m = 0.3, Λ= 0.7). The core of each cluster is imaged with WFPC2 in the F702W filter, and the spectroscopic sample is statistically complete to M_r∼−19.0 + 5 log h, within an 11 arcmin (∼1.8 h⁻¹ Mpc) field. The clusters are dynamically well-separated from the surrounding field and most have velocity distributions consistent with Gaussians. The velocity dispersions range from ∼350–850 km s⁻¹, consistent with the local L_X–σ correlation. All 10 clusters host a bright, giant elliptical galaxy without emission lines, near the centre of the X-ray emission. We measure the equivalent width of two nebular emission lines, [O ii] and Hα, and the Hδ absorption line to classify the cluster members spectrally. Galaxy morphologies are measured from the HST images, using the two-dimensional surface-brightness fitting software gim2d. Emission-line galaxies in these clusters are relatively rare, comprising only 22 ± 4 per cent of the sample. There is no evidence that these emission-line galaxies are dynamically distinct from the majority of the cluster population, though our sample is too small to rule out the ∼30 per cent difference that has been observed in more massive clusters. We find 11 galaxies, comprising 6 per cent of the cluster members, that are disc-dominated but show no sign of emission in their spectrum. Most of these are relatively isolated, spiral galaxies with smooth discs. We find no cluster members with a starburst or post-starburst spectrum. The striking similarity between the spectral and morphological properties of galaxies in these clusters and those of galaxies in more massive systems at similar redshifts implies that the physical processes responsible for truncating star formation in galaxies are not restricted to the rare, rich cluster environment, but are viable in much more common environments. In particular, we conclude that ram pressure stripping or cluster-induced starbursts cannot be solely responsible for the low star formation rates in these systems.

galaxies, clusters, general

1 Introduction

The evolutionary history of galaxies depends both on cosmic time and on the type of environment in which they exist. For example, recent studies have shown that the universal average star formation rate (SFR) has been declining steadily since at least z∼ 1 (Lilly et al. 1996; Madau et al. 1996; Cowie, Songalia & Barger 1999; Wilson et al. 2002). In a somewhat analogous way, star formation rates are known to monotonically decrease with increasing density at a given epoch (e.g. Balogh et al. 1997; Poggianti et al. 1999; Lewis et al. 2002). In both cases, the reason for the decrease in star formation is unknown. In particular, there is plenty of gas still available for star formation at the present day, so the sharp decline in activity over the past ∼5 Gyr is a critical issue. It is an intriguing possibility that the processes which influence the evolution of galaxies that end up in dense clusters may be more generally important to galaxies in other, more common, environments. If this is so, it may be possible to link the decline in the universal average star formation rate to environmental effects in a Universe which is growing hierarchically.

We now have a good empirical description for the galaxy populations of massive clusters. Galaxies within the virial radius have, on average, lower star formation rates, and less recent (<1 Gyr) star formation, than galaxies in the surrounding field; this is true both locally (Lewis et al. 2002; Gomez et al. 2002) and at higher redshifts (e.g. Couch & Sharples 1987; Balogh et al. 1997, 1999; Poggianti et al. 1999; Postman, Lubin & Oke 2001; Couch et al. 2001; Balogh et al. 2002a). Evidence is mounting that this deficiency in star formation activity is at least partially independent of the morphology—density relation (Dressler et al. 1997; Balogh et al. 1998; Poggianti et al. 1999; Couch et al. 2001; Lewis et al. 2002; Gomez et al. 2002). However, an explanation for this difference between cluster and field galaxies is still lacking. Ram pressure stripping of cold gas in the disc of a galaxy (Gunn & Gott 1972; Fujita 1998; Quilis, Moore & Bower 2000) is only likely to take place in the dense cores of rich clusters, and it seems unlikely that it can explain the suppression of star formation as far as several Mpc from the centre (Balogh et al. 1997; Kodama et al. 2001). Galaxy harassment (Moore et al. 1999) may be effective at destroying small galaxy discs, but the effect that this will have on the star formation rate of the galaxy is not clear. On the other hand, the observed radial and density dependences of galaxy stellar populations and morphologies are reproduced quite well by hierarchical models in which only the diffuse, hot halo gas expected to surround isolated galaxies is stripped (Larson, Tinsley & Caldwell 1980; Balogh, Navarro & Moris 2000; Diaferio et al. 2001; Okamoto & Nagashima 2001; Bekki, Couch & Shioya 2002). The observed increase in activity with redshift is most likely a consequence of the higher infall rate (Bower 1991; Ellingson et al. 2001; Kodama & Bower 2001), though projection effects may still play a role (Diaferio et al. 2001).

For most cluster galaxies, the last episode of star formation occurred many billions of years ago. Thus, if some physical mechanism is responsible for the transition from a more active state, it must have occurred in the distant past, and will be difficult to uncover by observing galaxies in their present state (e.g. Trager et al. 1998; de Jong & Davies 1997; Kewley et al. 2001). It is therefore necessary to consider how the galaxy populations have evolved over time (Butcher & Oemler 1984; Dressler et al. 1999; Kodama & Smail 2001; Kodama & Bower 2001). However, in hierarchical models of galaxy formation, the progenitors of today's most massive clusters are expected to be numerous smaller structures at higher redshifts (e.g. Kauffmann 1996). Thus, galaxies must be observed not only over a range of redshifts, but for a range of cluster masses as well.

For this purpose we have carried out an extensive observational campaign to obtain Hubble Space Telescope (HST) imaging and ground-based imaging and spectroscopy for ten clusters at z≈ 0.25, selected to have low X-ray luminosities (hereafter referred to as the low-L_X sample). This sample can be directly compared with studies of more massive clusters, both locally (Lewis et al. 2002; Gomez et al. 2002) and at higher redshift (Balogh et al. 1997; Smith et al. 2002). For example, in Balogh et al. (2002b, hereafter Paper I) we presented an analysis of the HST data for the present cluster sample, and compared it with a similar HST sample of high X-ray luminosity cluster cores. We found marginal (∼2σ) evidence that the low-L_X clusters have more disc-dominated galaxies at a fixed local density. This suggests that at least galaxy morphology is sensitive to the large-scale structure. In the present paper we will revisit this, and other issues, in light of the spectroscopic data.

The paper is organized as follows. In Section 2 we present the cluster selection, and the data acquisition, reduction and analysis. Our results are presented in Section 3, where we consider the dynamics and spectral properties of the galaxy population. In Section 4 we compare our results with those found for more massive clusters, and consider the implications of these results for models of galaxy evolution. We summarize our findings in Section 5. We use a cosmology with Λ= 0.7, Ω_m = 0.3, and parametrize the Hubble constant as H₀ = 100 h km s⁻¹ Mpc⁻¹.

2 Observations, reduction and analysis

2.1 Cluster selection

The cluster sample is the same as the low-L_X sample analysed in Paper I, with the addition of Cl 1633+57, comprising 10 X-ray faint clusters in the northern hemisphere.¹ The clusters are selected from the sample identified by Vikhlinin et al. (1998) in serendipitous, pointed ROSAT PSPC observations, restricted to a relatively narrow redshift range, z = 0.22–0.29 (σz/z∼ 0.1) and a mean redshift of z = 0.25, to reduce the effects of differential distance modulus and k-correction effects on the comparison between the systems. We compute L_X in the 0.1–2.4 keV band from the observed fluxes in the 0.5–2.0 keV band, corrected for galactic H i absorption and assuming a k-correction appropriate for an intracluster gas temperature equal to that expected from the local L_X–kT relation (Allen & Fabian 1998; Markevitch 1998), using the software package xspec. These luminosities are listed in Table 1, and range from 0.40 to 4.0 × 10⁴³h⁻² erg s⁻¹[0.1–2.4 keV] (Table 1). Although Cl 1444 + 63 is treated as two separate clusters, the X-ray luminosity in Table 1 is that of the combined clusters.

Table 1.

Properties of the 10 clusters.

2.2 Imaging and spectroscopy

HST imaging with WFPC2 is available for all clusters in the sample. The observations of all clusters but Cl 1633+57 are described more completely in Paper I. To summarize, each cluster was observed with three single orbit exposures in the F702W filter during Cycle 8. The photometry is calibrated on the Vega system, with updated zero-points taken from the current instrument manual. The final images reach a 3σ point source sensitivity of R₇₀₂∼ 25.5, and cover a field of 2.5 × 2.5 arcmin²(or 0.4 h⁻¹ Mpcat z = 0.25) with an angular resolution of 0.17 arcsec (∼0.5 h⁻¹ kpc). The cluster Cl 1633 + 57 was observed in Cycle 8 (Proposal ID 7374), and the data were retrieved from the CADC ²HST archive. These data are F702W WFPC2 observations, with four exposures of 1200 s. The total exposure time of 4800 s is therefore less than that of the other nine clusters (typically 7000 s; see Paper I). The calibrated images obtained from the CADC were combined in the same way as described in Paper I.

The spectroscopic sample is selected from ground-based imaging from the Palomar 200-inch telescope and the Isaac Newton telescope (INT). The single INT Wide-Field Camera (WFC) chip from which the galaxies were selected covers 11.4 arcmin at 0.33 arcsec pixel⁻¹, while the Palomar COSMIC images have a field-of-view of 13.7 arcmin with a pixel scale of 0.4 arcsec pixel⁻¹ (Kells et al. 1998). For all but two clusters, galaxies were selected for spectroscopic follow-up from the R-band images. In Cl 1309 + 32 and Cl 1444 + 63 the sample was selected from I-band images, because R was not available. Note that the two clusters Cl 1701+64 and Cl 1702+64 are sufficiently close together that spectroscopic targets for both could be selected from a single WFC chip. The conditions during the imaging observations were not photometric, so we have calibrated our images by comparing aperture magnitudes of several (usually 2–3) relatively isolated, early-type galaxies with the F702W photometry of the WFPC2 images, and converted this to standard R magnitudes, assuming R_F702W−R_c = −0.2 (Fukugita, Shimasaku & Ichikawa 1995). Because of uncertainties in the colour term, and the small number of calibration galaxies used for each cluster, the photometric calibration is likely to be accurate to only ∼0.2 mag.

The spectra were obtained over four observing runs, and we give a log of these in Table 2. Three sets of data were taken with MOSCA on the 3.5-m telescope at Calar Alto Observatory, using the g500 grism. The spectra have a dispersion of ∼2.7 Å pix⁻¹ and cover 4000 Å–8000 Å, with a resolution of ∼10–15 Å. The fourth observing run was with LDSS-2 on the William Herschel Telescope. Using the medium-blue grism, we obtained a dispersion of ∼4.5 Å pix⁻¹, covering 3500–9000 Å. The resolution of these spectra is ∼15–20 Å. Typically, two masks were observed for each cluster; in some cases a third mask was also obtained. Galaxies were selected for spectroscopic follow-up based solely on their instrumental R- or I-band magnitude, with preference given to brighter galaxies. The fraction of galaxies observed spectroscopically therefore declines toward fainter magnitudes. For each mask we obtain between 20 and 35 spectra through 1.5 arcsec wide slits, over an 11 arcmin field of view. In total we obtained 581 spectra, in variable conditions. Some galaxies observed in poor conditions were later reobserved in a subsequent run. We obtained reliable redshifts for a total of 317 galaxies, of which 172 are cluster members. A summary of the photometric and spectroscopic observations is given in Table 2.

Table 2.

Log of observations.

2.3 Data reduction and analysis

The spectroscopic data were reduced using iraf³ software. The images were bias-subtracted and median-combined to remove cosmic rays. Spectra were optimally extracted and the sky was subtracted by fitting a one- or two-degree polynomial to the counts on either side of the object. The spectrum was traced along the dispersion direction of the CCD to account for distortion. Wavelength calibration was based on either HgNeAr arc lamps or emission lines in the night-sky spectrum; the latter method was generally used to improve the calibration at the red end of the spectrum. The r.m.s. of the wavelength solution is typically ≲ 7000 Å, corresponding to Δz≲ 2 × 10⁻⁴. The spectra were not flat-fielded or flux-calibrated, which is inconsequential for this analysis since we restrict our attention to spectral features that are defined relative to the continuum, over a narrow wavelength range (∼150 Å).

Galaxy redshifts were determined by three of us (MLB, BLZ, AF), each with an independent method. MLB used the routine fxcor within the iraf environment to cross-correlate the spectra with high signal-to-noise ratio z = 0 galaxy spectral templates of similar resolution. BLZ used the Fourier Correlation Quotient (Bender 1990) within midas, which compares absorption lines with those of template stars, while AF measured the centroids of several prominent absorption lines. The agreement between the three measurements is good, and generally within the uncertainty (typically ∼100 km s⁻¹). The independent estimates allowed us to identify galaxies where one method failed (because of low signal-to-noise ratio around a critical line, for example). Spectra for which we could not resolve the discrepancy between different redshift estimates (≲5 per cent) were always of low signal-to-noise ratio, and were removed from further analysis.

In Fig. 1 we show the distribution of signal-to-noise ratio per resolution element, including only those spectra for which a redshift was obtained. The S/N is computed in the rest-frame continuum region 4050–4250 Å, redward of the 4000 Å break, and has a median of ∼10 per resolution element. It can easily be shown that, for spectra with a resolution of 15 Å, an emission line can be measured with an uncertainty <5 Å if S/N = , where W is the equivalent width of the line, assuming the uncertainty is dominated by the continuum flux (true for weak lines). Thus, lines as weak as W∼ 5 Å can be reliably measured if S/N = 4.3 Å, which is satisfied by ∼90 per cent of our spectra.

Figure 1.

The cumulative distribution of signal-to-noise ratio per resolution element, for the 315 galaxies in our sample with redshifts. The resolution element is ∼15 Å for the MOSCA spectra, and ∼20 Å for the LDSS-2 spectra.

2.4 Selection function and magnitude limits

For each cluster, we determine the fraction of galaxies in the photometric catalogue for which a redshift was obtained, as a function of instrumental magnitude. Galaxies in our spectroscopic sample are then weighted by the inverse of this selection function to obtain a sample that is statistically magnitude-limited. The selection function is determined separately for each cluster, as a function of instrumental magnitude. In Fig. 2 we show the combined selection function for the full sample, as a function of R-magnitude. We define the completeness as the fraction of galaxies with a redshift in each cluster field, considering only galaxies brighter than 1 magnitude above the faintest galaxy for which a redshift was obtained. This quantity is tabulated in Table 1.

Figure 2.

The average selection function for the whole sample. The top axis shows the k-corrected luminosity for a cluster at z = 0.25. In practice, the data are weighted by the selection function, as a function of instrumental magnitude, for each cluster separately.

Our sample is ≳20 per cent complete at R∼ 20.5, which corresponds to M_R∼−19.0+5 log h at z = 0.25 (including a k-correction of 0.2 mag). Unless otherwise stated, we limit our sample to galaxies brighter than this magnitude, which corresponds to ∼2 mag fainter than M*.

2.5 Line index measurements

We have shown that the signal-to-noise ratio and resolution of our spectra are sufficient to allow us to measure reliably spectral features with equivalent widths W≳ 5 Å. In particular we focus on the rest-frame equivalent width of the [O ii]λ3727 emission line, which is a signature of gas ionized by hot stars and hence the best indicator of current star formation in the blue region of the spectrum (e.g. Kennicutt 1983). However, it is less than perfect for several reasons. Perhaps most importantly, it is a weak line, usually with an equivalent width <30 Å. In moderate signal-to-noise ratio spectra like ours, therefore, it is difficult to detect low levels of star formation. Furthermore, the equivalent width of [O ii] is sensitive to metallicity and the ionization level of the gas (Charlot & Longhetti 2001; Charlot et al. 2002), which can introduce a scatter in the [O ii]–SFR relation of a factor ≳ 5. Finally, it seems likely that the H ii regions where [O ii] is produced will be more heavily obscured than the longer-lived stars which give rise to the continuum (e.g. Silva et al. 1998; Charlot & Fall 2000). In this case, the equivalent width will underestimate the star formation rate, even if a global extinction correction is applied. However, despite these difficulties, it has been shown that [O ii] correlates with better star formation indicators like Hα emission (Kennicutt 1992; Jansen et al. 2000, also see Fig. 3). Although there is a large scatter in this correlation, the average star-formation properties of a population can be well-described.

Figure 3.

The correlation between W₀(Hα+[N ii])and W₀([O ii])for galaxies in which a reliable measurement of both lines exists. Error bars are 1 −σ. The three galaxies represented by filled squareshave strong, broad [N ii] emission characteristic of non-thermal emission. The solid line is the average local relation measured by Kennicutt (1992). Three points are off the scale, at W₀(Hα+[N ii]) = 80 Å part of the error bars are just visible.

We also measure the Hδ absorption line, which is a strong feature in A stars and thus represents recent star formation. As this is an absorption feature, and affected by emission-filling, its precise measurement is more difficult than that of the emission lines, and little can be said about most of the galaxy population. However, galaxies with very strong Hδ absorption are easily detected, and these galaxies may play an important role in galaxy evolution (Dressler & Gunn 1983; Couch & Sharples 1987; Poggianti et al. 1999; Balogh et al. 1999).

These rest-frame equivalent widths are measured by fitting a line to the continuum on either side of the feature, and summing the flux above this continuum over a well-defined wavelength range. The definitions of the W₀([O ii]) and W₀(H δ) indices are the same as in Balogh et al. (1999)^,4. We will adopt the convention that W₀([O ii]) is positive when in emission, and W₀(H δ) is positive in absorption. Each line measured is checked to ensure that the continuum fit or spectral line pixels are not adversely affected by bad sky subtraction, bad pixels, or poor wavelength calibration. This removes 7.8 per cent of the cluster member [O ii] measurements from our analysis. Uncertainties are computed as described in Balogh et al. (1999), and include the Poisson noise contributed from the sky subtraction and, therefore, the degradation effected by the spectral resolution.

There are 167 cluster members with reliable measurements of [O ii], and 107 of these also have reliable measurements (not necessarily detections) of Hα+[N ii]. Thus, we can use measurements of this line to check the reliability of [O ii] as a star formation indicator in this sample. We measure W₀(Hα+[N ii]) in a manner analogous to that for W₀([O ii]), by computing the flux in the range 6555 Å< λ < 6575 Å, compared with the continuum in the regions 6490 Å < λ < 6537 Å and 6594 Å < λ < 6640 Å. We do not deblend the two adjacent [N ii] lines, but include them both in the equivalent width measurement. The [N ii]λ6548 line always contributes only a small amount of flux (<5 per cent), while the stronger [N ii]λ6583 line contributes ∼30 per cent of the flux. The observed correlation between W₀(Hα+[N ii]) and W₀([O ii]) is in excellent agreement with the local field correlation of Kennicutt (1992), as shown in Fig. 3. Galaxies for which W₀([O ii]) = 5 Å are almost always detected with W₀(Hα+[N ii]) = 20 Å. Only three (4 per cent) of these galaxies have strong, broad [N ii]λ6583 indicating that their emission is dominated by non-thermal processes (Veilleux & Osterbrock 1987; Kewley et al. 2001).

2.6 Morphologies

There are 78 galaxies for which we have both spectroscopy and HST WFPC2 imaging; 62 of these are confirmed cluster members (see Section 3.1 for the membership definition). As described in Paper I, we have measured fractional bulge luminosities, B/T, from these images using the two-dimensional surface brightness-fitting software package gim2d (Simard et al. 2002). The HST imaging only covers the central ∼0.6 Mpc in these clusters, and the number of disc-dominated, confirmed cluster members is small: only eight galaxies (18±5 per cent) brighter than R = 20 have B/T < 0.4. This is much less than the ∼40 per cent fraction of disc-dominated galaxies found in Paper I. This difference can be attributed to the different limiting magnitudes of the two analyses. The present spectroscopic sample is limited to R∼20, while in Paper I we considered a sample three magnitudes fainter, using only the HST imaging data. In Table 3 we list as F_stat the fraction of disc galaxies (B/T <0.4) for each cluster, with a statistical background subtraction based on the Medium Deep Survey (Ostrander et al. 1998; Ratnatunga, Griffiths & Ostrander 1999, see Paper I for details). We show both the value computed to R∼23, as published in Paper I, and the value considering only galaxies with R <20. These numbers are compared with F_spec, the fraction of disc-dominated galaxies determined from the spectroscopic sample of cluster members, complete to R≲20 (we combine the two clusters Cl 1444a and Cl 1444b to be consistent with the measurement of F_stat). The errors are all computed from bootstrap resampling, except in the case where no disc galaxies are found, where we assume a Poisson distribution. Adopting the brighter limit appropriate for our spectroscopic sample, the fraction of discs determined by photometric field correction is 26±17 per cent. This number is consistent with the spectroscopically-determined number (18±5 per cent), within the statistical uncertainties.

Table 3.

Fraction of disc galaxies.

2.7 The catalogue

A catalogue of the relevant derived quantities for all cluster members (see definition in Section 3.1) is given in Table 4. Galaxies are identified by an identification number from the photometric catalogue in column 2; serendipitous observations not in the original catalogue have an identification number of −99. The r magnitude, redshift and sky coordinates are given in columns 3–6. For galaxies observed with HST, column 6 lists the GIM2DB/T measurement. Note that the formal errors on B/T given by gim2d are typically ≲0.02; however these errors do not account for uncertainties in the sky level or point spread function, which likely dominate the measurement for faint galaxies. We have not, therefore, listed these uncertainties in Table 4. The spectral signal-to-noise ratio per resolution element is given in column 8, and the equivalent widths of W₀([O ii]), W₀(Hδ), and W₀(Hα+[N ii]) are listed in columns 9–11.

Table 4.

Properties of cluster members.

In Fig. 4 we show the images and spectra for all the cluster members with HST data. The brightest, central cluster members are shown separately in Fig. 5, as are the sample of disc-dominated galaxies without emission lines (see Section 4.1), which are shown in Fig. 6. Note that the spectra in the vicinity of Hα are usually dominated by residuals from night-sky emission lines, not real features or Poisson noise. Also, the wavelength solution beyond λ∼7000 Å is occasionally inaccurate, resulting in a small misalignment with Hα. This has a negligible effect on our measurement, since the Hα line still always lies within the bandpass we use to measure the flux.

Figure 4.

Images and spectra for all cluster members observed with HST. The F702W images are shown with logarithmically spaced contours over-plotted, and the B/T measurements from GIM2Dare printed on the images. All images are 3×3 arcsec², oriented north to the top and east to the left. To the left of the image, the galaxy identification and redshift are shown, together with the signal-to-noise ratio per resolution element of the spectrum, over 4050–4250 Å (rest frame). The spectra are smoothed to the instrumental resolution of ∼15 Å. For every galaxy we show the region around the [O ii] and Hδ lines, with the location of those lines marked with a vertical, dotted line. When the spectrum near Hα is sufficiently clear of night-sky lines to permit a reasonable measurement of the Hα line strength, this region of the spectrum is also shown, with the position of Hα indicated. The y-axis of the spectra gives the number of CCD counts, after sky subtraction; the x-axis is the wavelength, in Å.

Figure 5.

The central, bright galaxies of each of the clusters. The format of the figure is the same as for Fig. 4. The central galaxy of Cl0818 lies behind a bright, foreground spiral which dominates the observed spectrum, and is not shown. Ground-based images are shown where HST images are not available; these are also 3 arcsec on a side.

Figure 6.

Disc-dominated galaxies without emission lines. The format of the figure is the same as for Fig. 4.

3 Results

3.1 Dynamics

In Fig. 7 we show the redshift histogram of each cluster. For most clusters, the median redshift and its dispersion were determined using the biweight estimator (Beers, Flynn & Gebhardt 1990) on galaxies within Δz = 0.02 of the cluster. The exception is the double cluster Cl 1444, which has a bimodal redshift distribution, with a rest-frame velocity difference of 2080 km s⁻¹. In this case, the biweight estimator results in an unreasonably high dispersion for each peak in the distribution. Instead, we use a 2σ clipping method to better characterize the distribution of each structure. We recognize this treatment as arbitrary, and do not consider the dispersions of these two clusters to be well determined. A small correction is made for the estimated uncertainty of 100 km s⁻¹ on each redshift, by subtracting these in quadrature from the measured dispersion. For most of the clusters the velocity distribution includes ∼20 galaxies, and is well fitted by a Gaussian, shown as the smooth curves overplotted in Fig. 7. There appears to be a structure in redshift space between Cl 1701 and Cl 1702; with the small number of redshifts available, it is not possible to determine whether or not this substructure is associated with one of the clusters. The remaining clusters are well isolated from the surrounding field. Cluster members are taken to be all those within 3σ of the cluster velocity dispersion.

Figure 7.

Redshift distributions for each of the clusters in our sample. The redshift range plotted represents 9000 km s⁻¹ about the mean redshift, in the rest frame of the cluster. The smooth curves overlayed are Gaussian functions with the computed centre and observed-frame velocity dispersion for each cluster. Note the dispersions tabulated in Table 1are in the rest frame of the cluster and include a small correction for the 100 km s⁻¹ uncertainty on the redshifts.

The velocity dispersion, its uncertainty, and the number of galaxies within 3σ of each cluster are listed in Table 1. The velocity dispersions of all the clusters are fairly similar, with a mean σ= 548 km s⁻¹ and a standard deviation of 172 km s⁻¹. The uncertainty in velocity dispersion, determined by jackknife resampling, is generally quite large, =200 km s⁻¹. In particular, it is much larger for the clusters Cl 1309, Cl 1701 and Cl 1702, which appear to have substructure in the wings of the distribution. We expect the virial radii to be R_v[Mpc]∼0.0035(1+z)^−1.5σ[km s⁻¹]∼1.4 Mpc (Girardi et al. 1998), or ∼5.9 arcmin at z = 0.25. Because of the large uncertainties on the velocity dispersions, however, the virial radii of individual clusters are not well determined.

The measured velocity dispersions are in good agreement with those expected from the local correlation with X-ray luminosity, as shown in Fig. 8. Here, the X-ray luminosities are estimated bolometric luminosities for a Universe with Λ=0.7, Ω_m = 0.3, h = 0.7, assuming gas temperatures of 3 keV. The L_X–σ correlation is consistent with the local relation, over scales ranging from the richest clusters (David et al. 1993; Markevitch 1998) to groups (Xue & Wu 2000). It is now well known that this scaling is inconsistent with a purely self-similar model of the intracluster medium, but can be successfully matched by models with a substantial entropy floor due, perhaps, to the injection of energy from supernovae and AGN (Babul et al. 2002). However, our uncertainties on σ are too large to improve the existing constraints on the slope of this relation.

Figure 8.

The correlation between velocity dispersion σ and bolometric X-ray luminosity (Λ=0.7, Ω_m = 0.3, h = 0.7)for our sample is shown as solid circles with 1σ error bars on the velocity dispersion. This is compared with three local cluster samples, as indicated; error bars are omitted on these points for clarity.

Fig. 9 shows the normalized velocity-radius correlation for the 10 clusters. The cluster centres are taken to be the position of the brightest galaxy (Section 3.2), which are always near the X-ray centres from Vikhlinin et al. (1998). The radius is the distance to this centre, normalized to the cluster virial radius, while the velocity is the velocity difference from the cluster mean, normalized to the 1σ velocity dispersion. The cluster members are well-separated from the surrounding field. Emission-line galaxies and disc-dominated galaxies both avoid the central regions of the cluster. However, there is no measurable difference between the dynamics of the emission-line galaxies and the rest of the sample, as shown by the comparison of the normalized velocity histograms, in the right panel of Fig. 9. Both velocity distributions are consistent with a Gaussian distribution of unity variance.

Figure 9.

The normalized velocity-radius relation for galaxies in the 10 clusters. The velocities are measured relative to the cluster redshift, normalized to the cluster velocity dispersion. The radii are measured from the position of the central, bright galaxy (shown at R/R_vir = 0.01 for display purposes), and normalized to the cluster virial radius. Filled symbolsrepresent galaxies with W₀([O ii]) = 5 Å. Only galaxies within six times the cluster velocity dispersion σ are shown; cluster members are selected to be those within 3σ. The symbol shape corresponds to the galaxy morphology, as indicated in the legend. The normalized velocity distribution for cluster members is shown in the right-hand panel, for the full sample (open histogram) and the emission-line galaxies ( W₀([O ii]) = 5 Å, filled histogram ). Both are consistent with a Gaussian distribution of unit variance, shown as the smooth, solid curves.

3.2 Brightest cluster galaxies

The central galaxies of each cluster are shown in Fig. 5. The central galaxy of Cl0818 lies directly behind a bright, foreground spiral galaxy. The spectrum shows features from both galaxies, but is dominated by the foreground galaxy so is omitted from the rest of the analysis. All the other central galaxies are giant elliptical galaxies, none of which show emission lines, nor the prominent Balmer absorption lines that would indicate the presence of recent star formation activity. Star formation in central galaxies (e.g. see Crawford et al. 1999) is likely to be associated with cooling flow activity, which is now known to produce a reservoir of cold molecular gas (Edge 2001). Because only extended sources are included in the catalogue of Vikhlinin et al. (1998), at the faint flux limit the catalogue is biased against clusters with strong cooling flows (if they even exist at these low luminosities), which may therefore be related to the lack of emission in the central galaxies.

3.3 Spectral properties

The cumulative distribution of W₀([O ii]) for the 167 cluster galaxies, weighted by the spectroscopic selection function, is shown in Fig. 10. This sample is shown limited to M_r≤−18.5+5 log h, corresponding to R∼20 at z = 0.25, to allow a fair comparison with the field and cluster samples of Balogh et al. (1997, see Section 4). The mean (weighted by the selection function) is W₀([O ii]) =3.2 Å, and the median is W₀([O ii]) =0.7 Å. A total of 36 galaxies have W₀([O ii]) = 5 Å; accounting for the spectroscopic selection function, this corresponds to ∼22 per cent. None of the galaxies have W₀([O ii]) = 45 Å, or otherwise have spectra characteristic of a strong starburst.

Figure 10.

The cumulative distribution of W₀([O ii]), for the 167 cluster members with reliable measurements, brighter than M_R ∼−18.5+5 log h. The distribution is weighted by the spectroscopic selection function. This is compared with the distribution in the field and in clusters of high X-ray luminosity at z∼0.3, from Balogh et al. (1997). Both our survey and that of Balogh et al. sample the clusters out to the virial radius, and are statistically complete at this luminosity.

The cumulative distribution of W₀([O ii]), for the 167 cluster members with reliable measurements, brighter than M_R ∼−18.5+5 log h. The distribution is weighted by the spectroscopic selection function. This is compared with the distribution in the field and in clusters of high X-ray luminosity at z∼0.3, from Balogh et al. (1997). Both our survey and that of Balogh et al. sample the clusters out to the virial radius, and are statistically complete at this luminosity.

The fraction of emission-line galaxies, therefore, is comparable to the fraction of disc-dominated galaxies at our spectroscopic magnitude limit (see Section 2.6). In Fig. 11 we show these fractions as a function of luminosity. The emission-line fraction increases strongly with decreasing luminosity, which is a well-known result (e.g. Ellis et al. 1996; Lin et al. 1996; Lewis et al. 1999; Christlein 2000; Balogh et al. 2001). This trend is also seen, with less significance, in the fraction of disc-dominated galaxies, though the HST sample is smaller and morphologies are not measured for the faintest galaxies. The two fractions are comparable at all luminosities. However, the excess of disc galaxies relative to emission-line galaxies around L*, although statistically insignificant, corresponds to a bona fide population of ‘anaemic’ disc galaxies, which we discuss in Section 4.1.

Figure 11.

The fraction of emission-line galaxies ( W₀([O ii]) = 5 Å, solid circles) and disc-dominated galaxies ( B/T<0.4, open circles ) as a function of galaxy luminosity. Only bins with at least three galaxies are shown. Error bars are jackknife estimates or, in the case where the fraction is zero, estimates assuming Poisson statistics.

In Fig. 12 we show the dependence of emission-line equivalent width on B/T. Only two bulge-dominated galaxies (B/T = 0.6) show W₀([O ii])= 5 Å. One of these (Cl0841#38) is a double-nucleated galaxy with strong, broad [N ii] indicative of an active galactic nucleus (AGN). The other, Cl0849#20 is an Sa galaxy with a clear disc component, close to another bright galaxy which complicates the surface-brightness fitting procedure. Similarly, of the eight disc-dominated galaxies with B/T < 0.4, only three have W₀([O ii]) =5 Å, and one of these has broad Hα and strong [N ii] emission indicative of an AGN. This fraction of emission-line disc galaxies is consistent with the values seen in the inner regions (<0.1R_vir) of the high X-ray luminosity CNOC1 clusters (Balogh et al. 1998), and Abell 1689 at z = 0.18 (Balogh et al. 2002a). All of these galaxies have luminosities within ∼0.5 mag of M*_r∼−20.3+5 log h.

Figure 12.

The equivalent width of [O ii] as a function of fractional bulge luminosity, B/T. The two galaxies represented by open squareshave strong, broad [N ii] emission characteristic of non-thermal emission.

While the [O ii] line is present where star-formation is ongoing, the Hδ absorption line is expected to be strong in galaxies in which star formation has occurred sometime within the last ∼500 Myr or so (e.g. Couch & Sharples 1987; Poggianti et al. 1999; Balogh et al. 1999; Poggianti & Wu 2000). Measurements of W₀(Hδ) are shown as a function of W₀([O ii]) in Fig. 13. A population of galaxies with strong Hδ but no detectable emission are strikingly absent from this sample. Using the spectral classifications of Dressler et al. (1999), a k+a galaxy is one with W₀(Hδ) = 3Å and no detectable emission. This definition is somewhat arbitrary, and any physical interpretation of the Hδ strength needs to account for differences in the way in which the line is measured. Although 14 of the galaxies in our sample formally satisfy Dressler et al.'s definition of a k+a galaxy, most of these have W₀(Hδ) very close to the limit of 3 Å (Fig. 13). Given the large uncertainties (systematic and random) in the measurements and the model sensitivity of the interpretation, we cannot claim that the spectral properties of this population are strikingly unusual (see Section 4.1 for more discussion). We are only able to identify four galaxies which show W₀(Hδ) = 4 Å with at least 1σ confidence. All of these galaxies show nebular emission, and thus are e(a) galaxies in the classification scheme of Dressler et al. (1999). Both of these populations will be discussed further in Section 4.1.

Figure 13.

The rest-frame equivalent width of Hδ is shown as a function of W₀([O ii]), for all cluster members in which both lines could be measured. For galaxies with HSTimaging, the symbols correspond to the B/T ratio, as indicated in the legend. The sample error bars show the median 1-σ uncertainty in each index.

3.4 Population gradients

In Fig. 14 we show W₀([O ii]) as a function of clustercentric radius. The cluster centres are taken to be the position of the brightest galaxy. There is little trend in the mean or median equivalent width of the sample; both are always ≲5 Å out to 1 h⁻¹ Mpc. Galaxies with strong emission lines are only found outside the cores of these clusters, beyond ∼0.05 h⁻¹ Mpc, and the fraction of galaxies with W₀([O ii]) = 5 Å increases from <10 per cent within this radius to ∼30 per cent at 1 h⁻¹ Mpc, approximately the virial radius estimated for these systems. However, this may just be due to the fact that there are fewer galaxies in the core so the wings of the highly skewed distribution are insufficiently sampled. A Kolmogorov—Smirnov test cannot reject the hypothesis that the W₀([O ii]) distributions within and beyond 0.1 h⁻¹ Mpc are drawn from the same population.

Figure 14.

Left: [O ii] emission-line strengths as a function of cluster-centric distance. The long-dashedand short-dashedlines show the median and mean value, respectively, in bins of varying width, each containing 20 points. The solidline represents the fraction of galaxies with W₀([O ii]) = 5 Å, according to the scale on the right side of the figure. These quantities are weighted by the spectroscopic selection function, for a sample with R<20.5. Right: The same as the left panel, but as a function of local projected density. The density is computed from the distance to the fifth nearest neighbour, and corrected for the background using the number counts of Lin et al. (1999).

Galaxy morphologies are known to correlate strongly with the local density of galaxies (Dressler 1980; Postman & Geller 1984; Domínguez, Muriel & Lambas 2001). More recently, a similar density dependence has been found for the average star formation rate of galaxies (Couch et al. 2001; Kodama et al. 2001; Pimbblet et al. 2002; Lewis et al. 2002; Gomez et al. 2002). To compute the local, projected galaxy density, we measure the area enclosing the fifth nearest neighbour to each cluster member. We do this using the full photometric catalogue (i.e. not just galaxies with redshifts) and statistically correct for the average background density. This is similar to the definition of Dressler (1980), but we use the fifth nearest neighbour rather than the tenth because we found the latter method tends to wash out the densest regions of the clusters. Our results are unchanged if we adopt the tenth nearest-neighbour definition, but the resolution of the dense cores is poorer. It is important to compute the density to a fixed luminosity limit in all clusters, and we adopt M* + 1.5, which corresponds to M_r∼−19.5+5 log h, or R = 20 at z = 0.25. This limit is consistent with that of Dressler (1980), and is 3 mag brighter than that used in Paper I. We do not include the double cluster Cl 1444 in this analysis, because the projected surface density cannot be reliably determined from the photometric properties alone. For the background correction, we use the field number counts in R_c of Lin et al. (1999). The number of galaxies brighter than R_c = 20 is 1460±40 per square degree, where the error does not include cosmic variance. This is adjusted as necessary for the magnitude limit of each cluster. Finally, we invert the area containing the fifth nearest neighbour to obtain the density in units of galaxies per Mpc².

In the right panel of Fig. 14 we show the W₀([O ii]) as a function of local projected galaxy density. In the very densest regions, Σ = 500 h² Mpc⁻², there are no galaxies with strong emission lines. However, below this limit there is no evidence for a trend with density. The fraction of galaxies with W₀([O ii]) = 5 Å remains ≲20 per cent, and the median is <5 Å.

The lack of a correlation of W₀([O ii]) with density is surprising, especially given the local result of Lewis et al. (2002), that the correlation holds in all clusters, independent of mass. This is possibly a consequence of our small sample size, since there are very few galaxies with emission lines in the full sample. In particular, in the lowest density regions we have very few galaxies (13 with Σ<20), and thus cannot precisely determine the fraction of emission-line galaxies, especially when that fraction is ≪1.

4 Discussion

4.1 Anaemic spirals, starburst and post-starburst galaxies

Despite the similarity in the fraction of disc-dominated and emission-line galaxies (∼20 per cent), there is a significant population of disc-dominated galaxies in our sample that do not have detected emission lines. This may be analogous to the population seen in more massive clusters (Poggianti et al. 1999), and is a very cluster-specific population; in local field samples, almost all disc galaxies show strong emission lines (Kennicutt 1983; Jansen et al. 2000). In Fig. 12 we showed that there were six galaxies in our sample with B/T <0.4 and W₀([O ii]) <5 Å. In one of these galaxies, Cl 1701#149, there is Hα emission, W₀(Hα+[N ii]) =23±3 Å, which is weak enough to be consistent with the low observed W₀([O ii]). The remaining five galaxies are all convincing disc galaxies with no detectable emission. If we relax our definition of a disc-dominated galaxy to include those with larger B/T ratios (B/T < 0.5), we find five other clear examples of galaxies with a disc-like morphology but no evidence of nebular emission. We also include a galaxy (Cl 1309#119) without Hα emission ([O ii] is undetermined) to bring the total sample of such galaxies to 11. This population therefore comprises 6.5 ± 2 per cent of the cluster members, and 57 ± 17 per cent of the disc-dominated (B/T < 0.5) population. These are consistent with the fractions of late-type (later than Sa) galaxies without emission found in the z∼0.4 cluster sample of Poggianti et al. (1999). The images and spectra of these galaxies in our clusters are shown in Fig. 6. Most appear to be early or mid-type spirals, but have very smooth discs, without strong spiral structure or prominent H ii regions, and thus resemble anaemic galaxies (van den Bergh 1976, 1991). Particularly interesting is Cl0818#58, a smooth spiral galaxy that is elongated and asymmetric, and has the appearance expected of a galaxy in the process of being stripped (Quilis et al. 2000). We note that the slit width is 1.5 arcsec, half the width of the postage-stamp images shown in Fig. 6 and comparable to the size of the galaxies; therefore we do not expect the lack of emission to be owing to an aperture bias.

To improve upon the signal-to-noise ratio of individual spectra, we have coadded four classes of spectra, in Fig. 15. In particular, we show the coadded spectra of the anaemic spiral sample, including all galaxies with B/T < 0.5 and no detectable emission lines. Before coadding, the individual spectra are shifted to zero redshift and the shape of the continuum in the range 3500–5100 Å is removed with a spline fit. The spectra are then averaged pixel by pixel, weighted by the median flux in the rest-frame wavelength range 4050–4250 Å, to give more weight to better quality data. For presentation purposes, we fit the continuum of the Sb template spectrum from Kinney et al. (1996), and rescale our spectra to this continuum. Finally, the spectra are smoothed to the instrumental resolution of ∼15 Å.

Figure 15.

Coadded spectra of 11 anaemic spiral galaxies ( B/T < 0.5 and no emission lines), 13 normal spiral galaxies, 34 elliptical galaxies (B/T = 0.6)), 14 k+a galaxies ( W₀(Hδ) = 3 Å and W₀([O ii]) < 5 Å), and 8 e(a) galaxies. The spectra are renormalized to template continua from Kinney et al. (1996) as described in the text, and smoothed to the instrumental resolution of ∼15 Å. The positions of the Balmer absorption lines are marked with dashed, vertical lines.

The coadded anaemic spiral spectrum can be directly compared with the coadded spectra of 13 normal spiral galaxies (disc galaxies with emission lines, including clear examples of spiral galaxies in the absence of HST imaging) and 34 early-type galaxies (with HST imaging and B/T = 0.6), also shown in Fig. 15. The early-type galaxies are renormalized to the continuum of the elliptical template of Kinney et al. (1996). Apart from the lack of emission lines, the anaemic spiral spectra look quite similar to those of the normal spiral galaxies. In particular, the Balmer series is stronger than seen in the early-type spectra, comparable to that of normal spirals (W₀(Hδ) ∼ 2.5 Å). On the other hand, some absorption lines in the anaemic galaxies, most notably the G band at ∼4300 Å, are more similar to the strengths seen in elliptical galaxies.

It is remarkable that none of the galaxies in our sample show the very strong emission lines characteristic of a strong starburst. The strongest emission-line galaxies are almost all normal spiral galaxies (with the exception of the merging/interacting galaxies Cl0841#38 and Cl0849#20), and the fraction of strong emission-line galaxies is small relative to the fraction observed in the field at this redshift. Furthermore, as shown in Section 3.3, we find no convincing examples of the Balmer-strong galaxies without emission lines which may be in a post-starburst phase (e.g. Couch & Sharples 1987; Dressler & Gunn 1983; Balogh et al. 1999; Poggianti et al. 1999). Formally, there are fourteen galaxies which are classified k + a according to Dressler et al. (1999), though all but three of these have W₀(Hδ) ∼ 3 Å, close to the arbitrarily defined cut-off strength. The coadded spectrum of these galaxies is shown in Fig. 15. No evidence for Hβ or [Oiii]λ5007 emission is seen, confirming that the lack of [O ii] emission is real. Most of the spectrum appears similar to that of the elliptical population although, by selection, the Hδ absorption line is relatively strong. Most notably, the other Balmer lines are not especially prominent, which suggests that many of the high Hδ measurements are cases where the substantial measurement uncertainty results in an overestimate of the line strength, as suggested by Balogh et al. (1999). We do find eight galaxies with W₀(Hδ) = 4 Å, though the errors are such that only four of these exceed 4 Å with =1σ confidence. All of these galaxies have emission lines, and are thus e(a) galaxies (Poggianti et al. 1999). The coadded spectrum of these eight galaxies is also shown in Fig. 15. The Balmer series is clear even blueward of Hε, and is enhanced relative to that seen in the normal spiral, anaemic spiral, or k+a population. Emission lines at Hβ and [Oiii]λ5007 are seen, in addition to [O ii].

The rarity of galaxies with strong Balmer absorption lines is in good agreement with the results seen in X-ray luminous clusters at z∼ 0.3, from the work of Balogh et al. (1999). The discrepancy with the relatively high fraction of Hδ-strong galaxies in clusters at z∼ 0.4 (Poggianti et al. 1999) is still not understood. One suggestion that has been put forward is that the clusters of Poggianti et al. are dynamically younger, with more galaxy–galaxy interactions than in the more relaxed CNOC1 clusters (Balogh et al. 1999). However, we expect such interactions to be even more important in our sample of clusters, owing to their low velocity dispersions, and yet no large population of starburst or post-starburst galaxies is found. The fact that these clusters are X-ray selected and have central, giant elliptical galaxies, may suggest that they are dynamically old systems, in which all merger-induced star formation activity took place several Gyr ago. However, we still cannot rule out the possibility that the difference is due in part to evolution between z∼0.25 and z∼0.4 (corresponding to a difference of 1.3 Gyr in our assumed cosmology), or to spectroscopic sample selection effects, as discussed in Balogh et al. (1999).

4.2 Comparison with X-ray luminous clusters

The 10 clusters analysed in this work are analogous to more massive clusters in several ways. Dynamically, they appear to be relaxed systems in which the X-ray centre coincides closely with the position of the giant elliptical galaxy. Interestingly, none of these central galaxies show any sign of star formation. In contrast, in the ROSAT Brightest Cluster sample of clusters, ∼27 per cent have central galaxies which show emission lines, approximately independent of X-ray luminosity (Crawford et al. 1999). The clusters in our sample have luminosities which place them all in the lowest luminosity bin of Crawford et al.'s Fig. 4, where ∼22±10 per cent have central galaxies with Hα emission. Thus, our results for a sample of nine clusters (omitting Cl0818, for which no spectrum of the central galaxy was obtained) are not strongly inconsistent with this fraction.

In massive clusters, Carlberg et al. (1997) found that the velocity dispersion of blue galaxies is ∼30 per cent larger than that of the red galaxies, and suggested this was a sign that the star-forming galaxies are not yet in virial equilibrium with the cluster potential. A similar conclusion was reached by Dressler et al. (1999), who found that recently star-forming galaxies have a velocity dispersion ∼40 per cent larger than that of the passive, elliptical population. We find no statistically significant difference in the dynamics of the emission-line galaxy population, relative to that of the whole population. However, our sample of emission-line galaxies is too small to claim a significant difference from the more massive clusters. Using a Kolmogorov—Smirnov test, we can only rule out, at the 99 per cent or greater confidence level, velocity distributions (for the emission-line population) that are more than 2.8 times broader than the cluster velocity dispersion.

The distribution of W₀([O ii]) in our cluster sample (Fig. 10) is similar to that of Balogh et al. (1997), measured to the same luminosity limit, for the CNOC1 sample of X-ray-luminous clusters at z∼0.3 (Yee, Ellingson & Carlberg 1996; Carlberg et al. 1996). Both our sample and the high X-ray luminosity clusters show W₀([O ii]) distributions that are greatly suppressed relative to the field near that redshift, also taken from Balogh et al. (1997). Note that, since it is generally easier to get redshifts for emission-line galaxies, any incompleteness in our sample for this reason is likely to lead to our overestimation of the number of emission-line galaxies in our sample, thus strengthening our conclusions. Therefore, we conclude that over ∼2 orders of magnitude in cluster X-ray luminosity, there is little difference in the mean age of the stellar population for galaxies more luminous than ∼M*+2.5. It must, therefore, be a local process, rather than a global one associated with the large-scale mass distribution, which affects the star formation rates of galaxies.

Both the present study and the CNOC1 cluster sample present data out to approximately the virial radius of the clusters. However, the CNOC1 clusters are more massive, and there may be a difference in the range of local densities sampled. To test this, we have evaluated the local density for every galaxy in the CNOC1 sample, using the same method, and the same luminosity limit, as for our low-L_X cluster sample. In Fig. 16 we show the fraction of galaxies with W₀([O ii])= 5 Å as a function of local density, for the present sample and that of Balogh et al. (1997). The two functions are indistinguishable within the uncertainties, and thus we conclude that the level of star formation in the low-L_X clusters is comparable to that in the CNOC1 clusters at all densities probed. In particular, we see little trend with local density, and the fraction of emission-line galaxies is always <30 per cent.

Figure 16.

Top: The fraction of galaxies with W₀([O ii]) = 5 Å in the present sample (solid line) and in the CNOC1 sample (Balogh et al. 1997), as a function of local, projected galaxy density. Error bars are 1σ jackknife estimates. The fractions corresponding to the present sample are computed in equally populated bins containing 20 galaxies. The CNOC1 data are presented in bins each with 40 galaxies. Bottom: The fraction of disc galaxies (B/T < 0.4)from our HST sample as a function of local density (solid line) is compared with the fraction of spiral and irregular galaxies from (Dressler 1980dashed line).

In Paper I we showed that the low-L_X clusters have a significantly larger fraction of disc-dominated galaxies brighter than R∼23 than more massive clusters (Paper I). In particular, the data showed that, at a given local density, the bulges in massive clusters are systematically more luminous than the bulges in the low-L_X clusters, while the disc luminosity function is independent of cluster mass. In the present work we do not find a large difference between the fraction of emission-line galaxies at R <20. This may suggest that galaxy morphology (in particular, bulge size) is partially sensitive to large-scale structure, while star formation properties are not. Alternatively, this may just be reflecting the difference in the luminosity ranges considered in the two studies; unfortunately, the HST sample is too small to determine the disc fraction at R < 20 with enough precision to determine whether or not the small difference found in Paper I holds at this brighter magnitude.

Very recently, analysis of the correlation between star formation rate and local projected density in nearby clusters from the 2dF Galaxy Redshift Survey (Lewis et al. 2002) and the Sloan Digital Sky Survey (Gomez et al. 2002) has shown that star formation is reduced below the global average in all environments where the local density exceeds ∼1 galaxy (brighter than M*+1) per Mpc². This was shown to hold in systems of low velocity dispersion, and well outside the virialized cluster regions. Although our sample does not extend to such low densities, we confirm that, even at z∼0.25, where the Butcher–Oemler effect (Butcher & Oemler 1984; Ellingson et al. 2001) is beginning to appear, dense regions in low-mass structures have very low star formation rates.⁵ It will be of enormous interest to trace the star-formation rate correlation with density out to comparably low densities at this redshift and beyond, to compare with the low-redshift data.

4.3 Comparison with the morphology-density relation

It is interesting to compare the W₀([O ii]) dependence (or lack thereof) on density with the morphology—density relation (Dressler 1980). We show the measured disc fraction for our sample in this region as the heavy, dashed line in Fig. 16. Since our HST imaging is restricted to the central regions of the clusters, reliable morphologies are only available for the densest regions. To make a comparison at lower densities, we show the fraction of spiral and irregular galaxies in the sample of Dressler (1980), as a function of local density, converted to h = 1 for consistency with the results shown here. At the high density end, Dressler's data are consistent with the disc fractions we measure from the HST data.

There is a much stronger trend compared with the W₀([O ii]) data, and the fraction of spiral galaxies increases to ∼60 per cent at the lowest densities probed by our spectroscopic sample. At densities ≲50 Mpc⁻², the fraction of spiral galaxies expected from Dressler's data is at least a factor of two larger than the fraction of galaxies with W₀([O ii]) = 5 Å, significant at the =2σ level. In agreement with the results of Balogh et al. (1998), Couch et al. (2001) and Lewis et al. (2002), this suggests that the morphology—density relation is at least partially independent of the star formation rate dependence on density.

4.4 Physical mechanisms

We can use these results to draw some conclusions about the physical mechanisms which may be responsible for the low star formation rates among galaxies in dense environments. The clusters in this sample have velocity dispersions which are typically a factor ∼2 less than those of the most massive clusters in the Universe at z∼0.25. Since the ram-pressure force on a galaxy travelling through the intracluster medium is proportional to v² (Gunn & Gott 1972), we expect ram-pressure stripping of disc gas to be ∼4 times less effective in our cluster sample than in more massive clusters. In particular, models suggest that ram-pressure effects should be negligible in clusters with virial temperatures kT∼2 keV (Fujita & Nagashima 1999), corresponding approximately to the expected temperatures of our low-L_X cluster sample. However, this difference is not reflected in the current star formation rates of the observed galaxy population. Therefore, we conclude that ram-pressure stripping within clusters is not primarily responsible for the low star formation rates. This is in good agreement with the results of other studies, which have shown little or no trend in blue galaxy fraction with cluster X-ray luminosity (Fairley et al. 2002; Pimbblet et al. 2002).

Secondly, we have not found a population of starburst galaxies, nor evidence for a large population of galaxies which have had a recent burst. This is in agreement with more complete Hα studies of more massive clusters at z∼0.3(Couch et al. 2001; Balogh et al. 2002a). Thus, starbursts induced by the cluster environment (whether by interactions with the intracluster gas or with other galaxies) do not appear to be an important process, at least at the epoch at which the clusters are observed. There remain two appealing scenarios which are consistent with our data. The population gradients observed in rich clusters are consistent with models of ‘strangulation,’ in which satellite galaxies in haloes of any mass are stripped of their hot gas and consequently consume their available fuel supply fairly gradually (Larson et al. 1980; Balogh et al. 2000; Diaferio et al. 2001; Okamoto & Nagashima 2001; Bekki et al. 2002). The appearance of the disc-dominated galaxies which show no sign of star formation (Fig. 6) seems to support this hypothesis. With the possible exceptions of Cl 0818#58, which has a highly asymmetric appearance one might expect of a galaxy interacting strongly with the ICM, and Cl 1309#119, which may be interacting with a larger galaxy, all of these galaxies are fairly isolated, spiral galaxies with smooth discs and no sign of strong disturbance or bright H ii regions. The other possibility is that the galaxy transformation occurs in even smaller systems – galaxy groups – in the cluster infall region (Zabludoff & Mulchaey 1998; Kodama et al. 2001). In particular, at higher redshift galaxy groups are denser systems, with larger velocity dispersions than their local counterparts, and even processes like ram-pressure stripping may be able to take place (Fujita 2001). Furthermore, this may be the environment in which ‘strangulation’ itself is most effective, and subsequent evolution within clusters may make little difference to the observable properties of the population (Okamoto & Nagashima 2001). The next step is therefore to focus on galaxy groups at a series of redshifts; in particular, galaxies in environments close to the ‘critical density,’ at which environmental effects first become observable (Kodama et al. 2001; Lewis et al. 2002; Gomez et al. 2002).

5 Conclusions

We have analysed ground-based spectroscopy and HST-based morphologies of galaxies in 10 clusters at z≈0.25 with low X-ray luminosities. The sample includes 165 galaxies brighter than R = 20.5, which corresponds to M_r = −19+5 log h at z = 0.25. We have measured morphologies using the gim2d surface-brightness fitting software of Simard et al. (2002), and the strengths of important spectral features, in particular the [O ii] emission line. The properties of the 10 clusters can be summarized as follows.

• (i)

  All 10 clusters host a giant elliptical galaxy near the centre of the X-ray emission. None of the nine central galaxies for which we have a reliable spectrum show emission lines.

• (ii)

  Apart from the double cluster Cl 1444, and the close pair of clusters Cl 1701 and Cl 1702, all cluster velocity dispersions are consistent with a single Gaussian, and they appear dynamically well separated from the surrounding field. Thus, they appear to be evolved systems in approximate dynamical equilibrium.

• (iii)

  The measured velocity dispersions range from ∼350–850 km s⁻¹, and are consistent with the local L_X−σ relation observed in larger samples (e.g. Xue & Wu 2000).

• (iv)

  The fraction of cluster galaxies with W₀([O ii]) = 5 Å (2σ confidence limit) is 22±4 per cent. The mean is W₀([O ii]) =3.2 Å, and the median is W₀([O ii]) =0.7 Å. There is no evidence for a significant correlation between W₀([O ii]) and either radius or density, apart from the lack of strong emission-line galaxies in the densest, central regions (≲0.1 h⁻¹ Mpc). Also, we do not measure a significant difference in the dynamics of the emission-line galaxies, relative to the rest of the population.

• (v)

  Disc-dominated galaxies (B/T <0.4) comprise 18±5 per cent of the sample within the central 0.4 h⁻¹ Mpc covered by our WFPC2 images. Less than 25 per cent (2/8) of these galaxies show significant emission. The remainder, a population of ‘anaemic’ disc galaxies, are relatively isolated, regular spiral galaxies near L*, with smooth discs. Such galaxies are rarely found in local, field samples, but are also seen, in similar abundance, in more massive clusters (Poggianti et al. 1999).

• (vi)

  No galaxies in our sample have a spectrum characteristic of a post-starburst or of truncated star formation. Only four galaxies have Hδ = 4 Å with at least 1σ significance, and all of these show nebular emission lines. Thus there is no evidence that these cluster environments act to enhance star formation activity, even temporarily.

The distribution of W₀([O ii]) in these clusters is similar to that measured in the sample of Balogh et al. (1997), which is comprised of clusters approximately an order of magnitude more massive. Galaxies in both systems show low star formation rates even at projected surface densities as low as ∼10 h² Mpc⁻², where the fraction of spiral and irregular galaxies expected from the morphology-density relation is ∼60 per cent. The fact that star formation rates are so low even in these low-mass structures has important implications for understanding galaxy evolution in general. The phenomenon is not likely to be driven by extreme processes, such as ram-pressure stripping, or interaction-induced starbursts, which are expected to be important only in the richest clusters. Rather it is something that operates in more commonplace environments, possibly groups in the infall regions of clusters (Kodama et al. 2001; Lewis et al. 2002). Tracing the evolution of galaxies in these groups, therefore, may shed light on the processes responsible for the observed decline in the globally averaged star formation rate of the Universe.

Acknowledgments

We thank the referee, Chris Collins, for his expeditious report and useful suggestions which improved this paper. We are also grateful to the CNOC1 collaboration for allowing us to use their unpublished data. We acknowledge financial support from PPARC (MLB, RGB, RLD), the Royal Society (IRS), Leverhulme Trust (IRS, RLD), the Deutsche Forschungsgemeinschaft and the Volkswagen foundation (BLZ,AF). The data in this paper includes observations made with the NASA/ESA Hubble Space Telescope obtained at the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy Inc., under NASA contract NAS 5-26555.

1

The cluster Cl 1444+63 turns out to be two clusters aligned along the line of sight (see Section 2.3 ), and we consider the two clusters separately.

2

Canadian Astronomy Data Centre, which is operated by the Herzberg Institute of Astrophysics, National Research Council of Canada.

3

iraf is distributed by the National Optical Astronomy Observatories which is operated by AURA Inc. under contract with NSF.

4

Note that the W₀(Hδ)feature wavelength range given in Table 1 of Balogh et al. (1999) is incorrect. In both that work and the present analysis, the feature is taken to cover 4088–4116 Å.

5

Deriving a star formation rate from the W₀([O ii])requires knowledge of the galaxy B-band luminosity, metallicity and dust content. If these quantities in our clusters are similar to those in the z = 0 clusters, than the similarity in W₀([O ii]) distributions corresponds to a similarity in star formation rate distributions.