The Arecibo Galaxy Environment Survey – V. The Virgo cluster (I) Free

Abstract

We present 21-cm observations of a 10 × 2 deg² region in the Virgo cluster, obtained as part of the Arecibo Galaxy Environment Survey. 289 sources are detected over the full redshift range (−2000 < v_hel < +20 000 km s⁻¹) with 95 belonging to the cluster (v_hel < 3000 km s⁻¹). We combine our observations with data from the optically selected Virgo Cluster Catalogue (VCC) and the Sloan Digital Sky Survey. Most of our detections can be clearly associated with a unique optical counterpart, and 30 per cent of the cluster detections are new objects fainter than the VCC optical completeness limit. Seven detections may have no optical counterpart and we discuss the possible origins of these objects. Seven detections appear associated with early-type galaxies. We perform H i stacking on the H i-undetected galaxies listed in the VCC in this region and show that they must have significantly less gas than those actually detected in H i. Galaxies undetected in H i in the cluster appear to be really devoid of gas, in contrast to a sample of field galaxies from the Arecibo Legacy Fast ALFA (ALFALFA).

1 INTRODUCTION

As one of the nearest rich galaxy clusters, the Virgo cluster represents an important region for studies of galaxy evolution. While the optically selected Virgo Cluster Catalogue (VCC) of Binggeli, Sandage & Tammann (1985) is complete to an apparent photographic magnitude of +18.0 and is now over 25 years old, it is only much more recently that correspondingly sensitive studies of the H i gas have become possible. The H i Parkes All-Sky Survey (Staveley-Smith et al. 2000) was only sensitive to an H i mass of ∼2 × 10⁸ M_⊙ at the distance of the cluster (generally assumed to be 17.0 Mpc following Gavazzi et al. 1999), so can offer few insights into dwarf galaxies except at very close distances. The Arecibo Legacy Fast ALFA (ALFALFA) survey (Giovanelli et al. 2005; Haynes et al. 2011) offers a great improvement, with a sensitivity of ∼3 × 10⁷ M_⊙ at the distance of Virgo.

Very deep H i surveys, with the capability of detecting dwarfs and very gas-poor objects (with gas masses ≲×10⁷ M_⊙), have until now been restricted largely to pointed surveys. These have been limited to small numbers of objects (e.g. Gavazzi, Boselli & O’Neil 2005) and suffer an inherent optical bias. Conversely, previous and ongoing blind H i surveys, while not subject to an optical bias, have been more limited in depth [e.g. ALFALFA and the Jodrell survey of Davies et al. (2004), which was sensitive to about ∼6 × 10⁷ M_⊙].

The Arecibo Galaxy Environment Survey, AGES, is a complimentary survey to ALFALFA with significantly greater sensitivity (∼8 × 10⁶ M_⊙ at the Virgo distance) at the expense of the area of coverage. Of the survey’s target of 200 deg², 25 deg² is in the Virgo cluster region. Many different processes are known to act within clusters to remove gas – ram pressure stripping (Vollmer et al. 2001) and harassment (Moore et al. 1996), for example – but their relative importance is still not well understood. By studying the properties of the objects with the lowest gas masses – the most sensitive tracers of environmental influences – we hope to better understand how the gas is affected by its environment. The other targeted regions of AGES cover a wide range of galaxy densities from the local void and isolated galaxies, to galaxy pairs, groups and other clusters (for a more detailed summary, see Auld et al. 2006).

Although it does not normally dominate even the baryonic mass component of large galaxies, as the fuel for future star formation neutral hydrogen is nonetheless important in understanding galaxy evolution. It is also easily affected by environment, occasionally drawn into spectacular streams on the 100-kpc scale (e.g. Koopmann et al. 2008; Scott et al. 2011), and relatively easy to detect compared to other tracers of cold gas. Furthermore, it may also offer a way to detect optically dark galaxies, proposed as a solution to the missing satellites problem (Davies et al. 2006). In this context the importance of a deep, fully sampled H i survey in a region of high galaxy density becomes obvious.

The Virgo cluster is known to have a complex structure, and is thought to be comprised of at least three distinct clouds. The dominant component is the so-called subcluster A, with the giant elliptical M87 at its centre. It is also where the main bulk of the X-ray gas is centred (Böhringer et al. 1994) with other concentrations around M49 and M86, suggesting that these are the centres of other, smaller subclusters.

Gavazzi et al. (1999) describe clouds at 17, 23 and 32 Mpc distances, with numerous other separate regions of the cluster whose distances are less obviously distinct. Their findings were based on studies of the Tully–Fisher and Fundamental Plane relations. These velocity-independent distance estimators indicate that the high-velocity dispersion of the cluster (a total range of over 4000 km s⁻¹) means that a galaxy’s redshift gives little distance information – apart from verifying its membership of the cluster, it has little or no bearing on whether the object is likely to belong to one particular cloud or another.

More recently, Mei et al. (2007) studied the cluster using the method of surface brightness fluctuations. Although this survey was geared towards understanding the main body of the cluster at 17 Mpc distance, it also confirmed the presence of an infalling cloud at 23 Mpc. The main cloud, or subcluster A, was found to have a depth of 2.4 ± 0.4 Mpc, with a velocity dispersion of around 590 km s⁻¹. The key point is that there is no straightforward way to assign distances. Here we use the distances assignments given in the Galaxy On-Line Milano Network (GOLDMine) data base, which are based on measurements of Gavazzi et al. (1999)– see Section 4.2 for more details.

The remainder of this paper is organized as follows. In Section 2 we describe the observations and data reduction procedures. In Section 3 we describe the analysis of the data, while in Section 4 we describe the results. Finally in Section 5 we suggest some possible implications of these results. We assume a value of H₀ of 71 km s⁻¹ Mpc⁻¹ throughout. Our analysis here is largely restricted to cluster members; we discuss some of the background objects in more detail in Paper II.

2 OBSERVATIONS AND DATA DEDUCTION

Two areas of the Virgo cluster have been selected for study with AGES, VC1 and VC2, shown in Fig. 1. This paper examines the VC1 area while Paper II will consider VC2 and compare the results of the two areas. The areas were selected for study as they span very different regions of the cluster. VC1 spans the interior of the cluster and is centred on M49, with galaxies believed to be in three different populations at different distances (Gavazzi et al. 1999). Previous pointed H i observations (Gavazzi et al. 2003, 2005) indicate that this region is denser than VC2, with VC1 and VC2 respectively containing 4.2 and 2.2 detections per square degree. VC2 extends from the cluster interior to beyond the VCC survey region, though cluster members in that region are believed to belong to a single population at a uniform distance (Gavazzi et al. 1999). In this study the focus is on cluster members, though background objects are used where appropriate as a comparison sample.

The observations, data reduction and analysis procedures have been extensively described in Auld et al. (2006), as well as in Cortese et al. (2008), Minchin et al. (2010) and Davies et al. (2011). They are here only summarized. Observations were taken in 2008 January–June, 2009 February–June, 2010 January–June and 2011 January, using the Arecibo L-band Feed Array (ALFA) instrument on the Arecibo telescope in spectral line mode. The field observed to full depth spans 10° of RA by 2° of Dec. (in contrast, VC2 spans only 5 × 1 deg²) centred on M49. Due to ALFA’s hexagonal beam arrangement, a small area outside this range is also included. The full spatial range of the data considered here is from 12:08:36 to 12:49:36 in RA, and from +06:52:55 to +09:06:55 in Dec.

Observations are performed in drift scan mode, where the telescope is driven to the start of the scan position and the sky allowed to drift overhead. Each drift lasts for 20 min (covering 5° of RA), so two sets of drifts are needed to cover the full RA range. Shorter drifts ensure less loss of data in the event of a malfunction, and allow us to reach full depth over successive areas well before the observation campaign is complete. The ALFA instrument is rotated before the start of each scan to correct for the change in parallactic angle. At the start of each scan a signal from a high-temperature noise diode is injected for 5 s to provide flux calibration.

At Arecibo a point takes 12 s to cross the beam. To ensure Nyquist sampling scans are staggered by half the beam width (∼1 arcmin). With the seven beams of ALFA each point is scanned 25 times, giving a total integration time of 300 s per point. Each beam records data from two polarizations every second over 4096 channels, which span a velocity range from approximately −2000 to +20 000 km s⁻¹. The velocity resolution is equivalent to ∼5 km s⁻¹, or 10 km s⁻¹ after Hanning smoothing to remove Gibbs ringing from the Milky Way and other bright extended sources such as radio frequency interference (RFI) (see below).

The data are reduced using the aips++ packages livedata and gridzilla, described in Barnes et al. (2001), developed by the Australia Telescope National Facility (ATNF). livedata is used to estimate and remove the bandpass and calibrate the resultant spectra; gridzilla co-adds the data to produce the final 3D data cubes. The bandpass across the scan is estimated by the median of the data values, which is robust to bright point sources but can give incorrect values for bright extended sources. The data are analysed using the miriad task mbspect. From measurements of individual sources, the median noise level – rms – reached is approximately 0.6 mJy beam⁻¹ (after Hanning smoothing).

As a sensitivity limit, for an rms of 0.6 mJy, a 4σ detection with a top-hat profile, 50 km s⁻¹ velocity width, would have an H i mass at 17 Mpc distance of 8.2 × 10⁶ M_⊙. However, this is only approximate. Galaxies of any given mass will be detected with lower signal-to-noise ratio (S/N) levels if their velocity width is greater (or equally, if they have the same intrinsic velocity width but higher inclination angles) since their flux is spread out over more channels.

Follow-up observations were performed using the L-wide receiver. This uses the position-switching method, with the ON and OFF source times each of 5 min. The data were quickly reduced and additional observations taken if necessary. We use the Wideband Arecibo Pulsar Processors (WAPPs) with nine-level sampling and one polarization per board, giving us 4096 channels. We use two bandwidth settings, 50 MHz (giving a velocity resolution of 1.25 km s⁻¹) and 25 MHz (0.63 km s⁻¹ resolution). The criteria for performing follow-up observations are described in Section 2.2.

2.1 Data contamination: the Milky Way and radio frequency interference

Although the bandwidth observed spans over 20 000 km s⁻¹ of redshift, in certain regions the survey is effectively blinded due to the presence of strong contaminating sources. The signal from the Milky Way is extremely strong and fills the whole sky, and livedata is unable to correctly process the signal using our standard techniques. Consequently, we are unable to obtain meaningful data over the velocity range −50 < v < +50 km s⁻¹. Similarly, the Punta del Este Federal Aviation Administration radar, operating at 1350 MHz, causes strong false signals over the approximate velocity range +15 400 < v < +16 000 km s⁻¹, as well as the intermod at approximately 7000 km s⁻¹. Harmonics of the radar are present throughout the cube but are much weaker and do not significantly detract from the available bandwidth. Some scans were also affected by the signal from the L3 GPS satellite, generating false signals at approximately 8500 km s⁻¹. The survey is not blinded within these regions affected by RFI, but sensitivity is greatly reduced.

2.2 Source extraction

There is no universally accepted technique for H i source extraction from 3D data cubes. We therefore employ a variety of methods in order to recover as many real sources as possible, whilst attempting to minimize the number of spurious detections. First, we inspect the cube by eye, visually scanning the cube in different projections and recording any possible sources. The cube is scanned at least three times in this way and candidate detections are masked using the miriad task immask to avoid confusion. A source only detected once is rejected; sources detected two or three times are subject to further inspection.

The second technique is an automatic extractor which we developed specifically for AGES called glados (Galaxy Line Analysis for Detection Of Sources). It is partly based on the existing programs polyfind (described in Davies et al. 2001) and duchamp (see Whiting 2012). polyfind examines spectra and searches for peaks above a S/N threshold, whereas duchamp examines images and searches for a specified number of contiguous pixels above a fixed flux threshold. Both approaches have merits and problems. polyfind only requires a single input threshold, the S/N, which does not need to be altered if the actual rms varies. However, since there are many thousands of individual pixels which happen to be above (say) a peak S/N of 4, polyfind’s generated catalogues have a reliability estimated at 3 per cent (at the 4σ level), and are correspondingly laborious to search.

duchamp’s approach, in contrast, can generate catalogues with extremely high levels of reliability, but with poor completeness. There are many problems with using a fixed flux threshold for AGES data. First, the noise can vary across the cube, especially at the edges (both spatially and in velocity). To use duchamp to search for faint sources requires these regions to be truncated, otherwise many thousands of spurious detections result. Secondly, it is impossible to determine a sensible flux threshold without having searched the cube previously, negating the main benefit of an automatic extractor.

To overcome these problems, glados employs a combination of many different techniques. First, both of the individual polarizations are gridded separately, in addition to the averaged cube which is used for the final measurements. glados initially searches spectra along each point in the averaged cube, searching for peaks above a specified S/N threshold (we use 4σ) and minimum velocity width (we use 25 km s⁻¹). The use of a minimum velocity width criteria helps reduce the thousands of noise spikes which plague polyfind. If a candidate signal is found, the individual polarizations are checked at the same point for a corresponding detection.

This step of examining the separate polarizations helps reduce spurious detections, since the noise in each is uncorrelated. The penalty for this is that their noise level is higher, and so they must be smoothed so that weaker detections are not missed. Since this reduces velocity resolution, the end result is that it can fail to detect the narrowest sources. However, since the number of sources found even by a visual search below 25 km s⁻¹ velocity width is extremely small, this does not seem to be a major sacrifice.

Finally, many of the close bright sources are resolved by AGES and so span many pixels. glados uses the starlink‘tmatch’ package to combine these repeat detections. As this can result in merging close sources, the original output is retained. Using topcat a map of the original glados detections are plotted, and the post-merging detections are overlaid. An observer then checks by eye for any cases where separate sources may have been unfairly merged.

Our third source extraction technique is to inspect spectra extracted with mbspect at the positions of known galaxies, listed in VCC, which were not detected by the other two methods. The idea here is to search for very weak sources that the other techniques might not be sensitive enough to pick up without allowing many additional spurious detections. Although 217 spectra were examined, this only resulted in three additional detections (these are very similar to ALFALFA ‘code 2’ detections – see Giovanelli et al. 2007). This demonstrates that our extraction techniques are generally detecting even the weakest sources.

Once a source is accepted as a candidate, it is first inspected with miriad and/or a visual inspection of the standard averaged cube and both individual polarizations. These additional checks ensure that any polarized signals, which could be recorded in a visual search (since it is impractical to search the entirety of all three cubes by eye), and any unusual artefacts such as may result from RFI, are not accepted as real objects. The remaining sources are then subject to follow-up observations where required. This is performed if a source was only detected by one method, has no clear optical counterpart (see Section 3), or is in any way anomalous (such as being unclear if it is polarized emission or not, or if its measurements do not agree with catalogue data).

3 SOURCE CATALOGUES

The combination of source extraction methods described above produced a combined candidate list of 322 sources for the entire volume of the cube. 75 of these were considered uncertain, either because they were detected only by one method, had no obvious optical counterpart, or are of low S/N. Follow-up observations were performed on all of these, of which 33 were not detected, reducing our final sample to 289 objects.

Tables 1 and 2 give the measured parameters for sources within the cluster (cz < 3000 km s⁻¹; Giovanelli et al. 2007) and behind it (cz > 3000 km s⁻¹), respectively. The columns in each table are identical and are as follows: (1) source number in this catalogue; (2) right ascension J2000, error in seconds of time; (3) declination J2000, error in seconds of arc; (4) heliocentric velocity in km s⁻¹; (5) maximum velocity width at 50 per cent and (6) 20 per cent of the peak flux; (7) total flux in Jy; (8) distance in Mpc; (9) estimated H i mass, log M_⊙; (10) peak S/N and (11) rms in mJy.

3.1 Identification of optical counterparts

This region has been fully observed with the Sloan Digital Sky Survey (SDSS). We visually search the SDSS images (DR8) within a 3.5-arcmin radius of the H i coordinates as determined by mbspect. When a candidate is found, we check its measured redshift from the SDSS, NASA/IPAC Extragalactic Database (NED) and GOLDMine (Gavazzi et al. 2003) – having at least two optical redshift measurements is preferable, since there is a chance that a single measurement might be erroneous and cause problems in identifying the true optical counterpart. If the candidate’s optical redshift(s) agrees with the H i redshift to within 200 km s⁻¹, the object is defined to have a ‘sure’ optical counterpart. If the optical and H i redshifts differ by more than this then the optical candidate is rejected. The distribution of velocity offsets is shown in Fig. 2.

In many cases there is a lack of optical redshift data available. For those candidates, we accept them as associated with the H i if they are the only unique candidate within 3.5 arcmin (in most cases their optical position differs from the H i by much less than this, as shown in Fig. 3). We deem them to be ‘probable’ optical counterparts. Although we treat them as being associated with the H i throughout the analysis, their optical association is flagged as uncertain in the tables.

The optical and H i measurements of position and redshift are generally in very good agreement, with a mean difference of just 4 km s⁻¹ for the velocities and 24 arcsec for the positions. The few outliers which are present are low S/N H i sources for which it is difficult to precisely define the location of the H i (both in space and velocity).

3.2 Comparison with other H i surveys

67 galaxies in this region have also been detected by ALFALFA. We will comment in more detail on the detection rates of the two surveys in a future paper. Here we restrict the analysis to a comparison of our measurements of H i masses and velocity widths. We use the W50 estimate since that is the quantity given in the ALFALFA catalogue. We do not use the H i mass estimate as given in the ALFALFA catalogue, since that assumes a different distance estimate (see Gavazzi et al. 2005); instead we use their measured total flux and convert to an H i mass assuming the distance as described in Section 4.2.

76 galaxies are listed as having an H i detection in GOLDMine (after discounting those listed as unreliable). For these objects we use the H i mass listed in GOLDMine directly, since the distance assumption is the same. However, GOLDMine only lists the average of the W50 and W20, so for the velocity comparison we use the average of these quantities as determined for the AGES measurements. The velocity width and H i mass estimates are compared in Figs 4 and 5, respectively.

As with the comparisons with the optical data, AGES measurements are generally in good agreement with existing observations, though several outliers are seen. These are low S/N sources, where less sensitive surveys may not detect H i over as wide a velocity range, or may confuse part of the noise with the real detection. We use the AGES measurements in all cases.

3.3 Analysis of optical data

Optical data available for this region includes the SDSS and the Isaac Newton Telescope Wide Field Survey (INT WFS). However, the INT WFS data are too variable in quality to be of much use in this region. Instead, optical photometry is performed on the SDSS data which are uniformly available in this region, using the aperture photometry tools in the ds9 package funtools. We restrict photometric measurements to those objects where there is a single sure or probable counterpart, where the H i detection is not in doubt, and where the galaxy’s optical emission is uncorrupted by any foreground stars. For instance, in the case of the two bright galaxies VCC 905 and VCC 939, the two objects are so close that the AGES observations cannot resolve the separate objects, so we cannot accurately determine how much H i each galaxy contains (the H i profile, source 250, is nonetheless listed in Table 1).

This gives a sample size of 233 objects, of which 79 are members of the Virgo cluster (i.e. at a redshift of ≤3000 km s⁻¹). For completeness, the H i parameters of all 269 sure detections are shown in Tables 1 and 2, which detail the Virgo cluster and background objects, respectively. The optical properties of all detections are given in Tables 3 and 4. Note that if photometry is not relevant (i.e. when studying the velocity distribution of H i detections), then we use all sure H i detections, not just those with good optical photometry data.

4 RESULTS

4.1 Overall statistics

A total of 95 sure H i detections were made within the Virgo cluster. Of these, 26 are not members of the VCC. The VCC is considered to be complete to a photographic magnitude of 18. Taking this to be approximately equal the g-band magnitude, 14 of these detections are actually brighter than the completeness limit (four of these are marginal, with g > 17.5). This is, however, less than 5 per cent of the total of 282 VCC objects thought to be cluster members (see below) in this region, so the completeness magnitude of the VCC seems reasonable.

The VCC lists 342 objects within this region. 60 of these are excluded as cluster members by previous redshift measurements. The remaining 282 are classed as members, 95 of which are spectroscopically confirmed. AGES provides redshift measurements for 10 objects for which none was previously available, and finds that three are cluster members and seven are background objects.

Though AGES does detect a population of objects missed by the VCC, this population is small in comparison with the number of VCC cluster members. Some basic assumptions show that this is at least partly due to the sensitivity effects of each survey. As described earlier, the sensitivity limit of AGES is approximately 8 × 10⁶ M_⊙ at a distance of 17 Mpc. The completeness limit of the VCC is a photographic magnitude of 18.0, but it does contain some objects (in the VC1 region) as faint as 20.0 or 19.6 in the g band (an absolute magnitude of −11.6). Using the relation between H i and stellar mass of Gavazzi et al. (2008), AGES H i mass sensitivity is equivalent to an absolute magnitude of −9.8 (or to put it another way, the mass sensitivities of the two surveys are within a factor of 2 of each other).

The relatively low number of detections from AGES that are not part of the VCC immediately implies that the H i is usually associated with optical sources. This is indeed the case, but there are some exceptions which we discuss in Sections 4.7 and 5.

4.2 Spatial and velocity distribution

This study is focused on the H i detections that are clearly cluster members. For objects outside the cluster, the distance used is determined simply from the Hubble velocity, assuming a Hubble constant of 71 km s⁻¹ Mpc⁻¹. For objects within Virgo, the distance used is that given in the GOLDMine data base, which, as mentioned earlier, is based on upon Tully–Fisher and Fundamental Plane determinations. However, this was only possible for a limited sample of 134 objects, as described in Gavazzi et al. (1999). They then defined spatial regions within which all galaxies are assigned to the same distance, based on their sample with actual distance estimates.

We apply this same procedure to all our detections within the cluster, including those not listed in the VCC. We note, however, that there remains considerable distance uncertainty, as evidenced by Fig. 6 (we show our detections in Fig. 7). Although it appears that galaxy populations are quite distinctly separate in position–velocity space, the boundaries chosen by Gavazzi et al. (1999) do not respect this. The effects of this are, fortunately, not dramatic. H i mass scales as distance squared, so a change from 17 to 23 Mpc, or from 23 to 32 Mpc, increases the H i mass by a factor 1.8 and 1.9, respectively. Equivalently, this would result in a decrease of absolute magnitude by 0.66 and 0.72 magnitudes, respectively.

The spatial distribution of the AGES detections and non-detected VCC galaxies is shown in Fig. 8. The distributions are broadly similar, with the western region being richer in both H i-detected and non-detected galaxies. The western half contains approximately 22 non-detected and nine detected objects per square degree, the eastern half just 10 non-detected and three detected objects per square degree. Qualitatively, the structures defined by the objects are very similar. Gas-rich galaxies appear to trace roughly the same population of objects as the non-detected galaxies.

More quantitatively, however, 30 per cent of the H i detections in the cluster are found at velocities greater than 2000 km s⁻¹ (Fig. 9). In contrast, only 10 per cent of the non-detections have similar velocities. A Kolmogorov–Smirnov test reveals that the distributions are different to a 99.9 per cent significance level. The influence of the cluster on galaxy gas content is discussed further in Sections 4.5 and 4.6.

4.3 Morphology

4.3.1 Late-type galaxies

For morphological classification, we use the scheme adopted by the GOLDMine data base, where types −3 to +1 are early-type galaxies (ETGs), all others are late-type galaxies (LTGs) with +2 to +10 being spirals, +11 to +19 being irregulars and blue compact dwarfs (BCDs), whilst those of type 20 have not been classified. The distribution of the morphologies of H i detections and non-detections is shown in Fig. 10.

At first glance Fig. 10 suggests that there is a significant population of undetected LTGs as well as a population of H i detected early-types. We discuss the latter in Section 4.3.2; however, the former can be dismissed. None of the undetected type 20 (unclassified) objects resembles late-types – they are all small, faint and probably dwarf ellipticals (dEs). Of the other supposed late-types, those classified as spirals and irregulars, only two appear to be genuine LTGs, and those are too close to other H i detections to be resolved as separate sources by AGES. The others appear to be outright misclassifications, and visually do not at all resemble their late-type assignments.

Conversely, the unclassified objects detected in H i are definitely late-type objects, most of them probably dwarf irregulars (dIs) or BCDs (although three appear to be flat, edge-on discs). None of the unclassified objects is likely to be early-type objects, but for more discussion on this see the next section. Those undetected unclassified objects all appear to be ETGs, mostly very faint dEs. The type 12 objects undetected by AGES are either very faint (and so possibly misclassified) or have poor quality claimed H i detections which may be spurious. In short, it appears that virtually all LTGs are detected in H i in this region by AGES. Note that throughout the rest of the analysis we use the classifications given in GOLDMine, except where these appear to be obviously in error.

4.3.2 Early-type galaxies

Given the above definition of early-type morphology, seven H i detections are associated with ETGs; five of these are new H i detections. The H i association with VCC 1394 is ambiguous in that the H i coordinates are 2.2 arcmin from the optical position (no optical redshift data exist). It is the only obvious galaxy visible within the Arecibo beam and classed as a cluster member in the VCC, so is by far the most likely candidate.

Additionally, there have been several previous claims of H i detected in the bright S0 VCC 1535, initially by Bottinelli et al. (1990), as well as in GOLDMine (which records an of 2 × 10⁹ M_⊙, though the data used are unpublished). More recently ALFALFA (Haynes et al. 2011) find a much more modest of 1.4 × 10⁷ M_⊙, with a S/N of 3.2. However, within the reported velocity range of the source AGES finds no structure that could be considered as a detection. We disregard this galaxy in the remainder of the analysis.

It is important to confirm that those objects really are ETGs and not misclassified objects. di Serego Alighieri et al. (2007) argue that several of their detected ETGs show peculiar morphological characteristics and are only borderline ETGs. This does not appear to be the case for the AGES detections, which, with the exception of VCC 758 with its clear dust lane/disc, appear smooth and featureless at optical wavelengths.

Lisker, Grebel & Binggeli (2006) (hereafter L06) describe two methods employed to enhance any possible structures, which we employ here. The first is to create a residual optical image. A symmetrical model of each galaxy is created using the iraf tasks ellipse and bmodel, which is subtracted using imarit to reveal any asymmetrical features. The position angle and ellipticity are fixed, otherwise, as shown in L06, there is a risk the model image will also incorporate asymmetrical features. This technique is only possible for three of the early-type objects detected by AGES. Three more are too small and too faint for ellipse to create a successful model. For the remaining galaxy, VCC 758, there seems little point attempting such a procedure given the presence of a prominent central dust lane or disc.

The second method is to create an unsharp mask. This is done by Gaussian smoothing the image with various kernel sizes (using the gauss task in iraf) in order to search for features of different sizes. Kernel sizes of 3, 5, 10 and 20 pixels are used. The original image is then divided by the smoothed image using imarith. This has the advantage of being possible for all objects (but there is still no benefit in attempting this for VCC 758). Adopting the procedure in L06, both circular and elliptical unsharp mask images are created. This is because both can give rise to artificial structures and serve as a complimentary check on the other. Residual images and unsharp masks are shown in Figs 11 and 12.

Most of the detections show little evidence of structure. As a comparison, the type 2 galaxy VCC 94 is also shown. With clear spiral structure evident in both the unsharp mask and the residual image, this demonstrates that the techniques used here are able to reveal extended features not visible in the SDSS red giant branch (RGB) images. A caveat is the low luminosity of the dEs, but this is not an issue for the lenticulars.

VCC 180 does show a bar-like feature in its centre, which is seen most clearly in the unsharp mask but is also visible in the residual image. Only VCC 180 and VCC 758, out of the seven detected ETGs, seem to have any kind of unusual morphological feature, and that within VCC 180 is relatively small. There is certainly nothing approaching that seen in L06 fig. 7, or VCC 94, which shows clear spiral structure extending over the whole diameter of the galaxy. However, two of the four detected S0s show some signs of structure, while it is very difficult to properly assess the dEs VCC 190 and VCC 1964 since they are very faint.

Further differences between the detected ETGs become apparent when we consider the detection rates. While 25 per cent of the S0s listed in the VCC in this region are detected in H i, the fraction of detected dEs is less than 3 per cent. Arguably the latter fraction could be even lower, given the various problems presented by VCC 190, 1964 and 1394. In summary, while the numbers are small, the S0s are detected in H i relatively often and show some signs of structure, whereas the morphologically smooth dEs are almost never detected.

More generally, the VCC lists a total of 187 ETGs in this region, giving an H i-detected fraction of 3.7 per cent, broadly consistent with 2.3 per cent reported by di Serego Alighieri et al. (2007) who used ALFALFA data. This latter figure corrects for the VCC brightness completeness limit, the contamination of background galaxies in those VCC members without redshift data and those non-detections too close to the strong continuum source M87 (where they could not be detected). If the same corrections are applied to the VC1 region, the detected fraction rises to 8.8 per cent. This significantly higher detection rate is at least partly the result of the increased sensitivity of AGES relative to ALFALFA. It should be stressed that these are small number statistics, and the effects of cosmic variance are also important.

To attempt to improve our statistics we employ stacking. In this region there are 56 undetected early-type VCC galaxies whose spectra can be averaged to give an increase in sensitivity. The suitability of these objects rests on their having optical spectra (so that the spectra are all centred on the velocity of each galaxy, ensuring that the galaxies are averaged with each other when combined) and no H i-detected galaxies (including the Milky Way) or strong continuum sources nearby.

The lowest rms reached by stacking is 0.0876 mJy, equivalent to a mass sensitivity of 1.2 × 10⁶ M_⊙ at 17 Mpc (assuming a 4σ top-hat detection with a width of 50 km s⁻¹). There is a slight complication that not all of the objects stacked in this sample are actually at 17 Mpc distance, but when we restrict the stacking to only the closest objects the mass sensitivity decreases only slightly to 1.7 × 10⁶ M_⊙. The lack of any detections by stacking to such high sensitivities implies that deep direct observations (of Virgo cluster galaxies) will also fail to detect more ETGs in H i, which is to some extent borne out by the observations of Conselice et al. (2003).

It seems then that galaxies which are detected in H i are fundamentally different from those which are not. Our results, together with those of Conselice et al. (2003), imply that those galaxies which are not detected in H i are really devoid of gas. In contrast, Fabello et al. (2010) found that field galaxies not detected directly in H i nonetheless do possess some gas content, with stacking of ALFALFA data (approximately four times less sensitive than AGES) for 50 galaxies producing a detection. While field galaxies may lose gas, those in the cluster can become entirely depleted. We have already seen that the H i in the Virgo cluster is also almost always associated with LTGs, while the ETGs are almost always absent from the H i detections. In the following sections, we explore in more detail the properties of H i detected and undetected galaxies.

4.4 Colours

Colour–magnitude diagrams for the H i detections are shown in Fig. 14. As a comparison, photometry was also performed for the undetected galaxies with optical redshift measurements (76 objects) as well as the background H i-detected objects (154 objects). The requirement for an optical redshift ensures that the objects can be very reliably determined to be cluster members or background objects. In these plots only sure H i detections with a unique optical counterpart are shown. The majority of the undetected cluster members are, as we have seen, unambiguously early-type objects.

Objects in Virgo appear systematically redder than those in the background of the same luminosity, which could suggest an environmental influence of the cluster. It is difficult to disentangle this possibility from selection effects – objects in the background are selected by H i, and being more distant, must be richer in H i and therefore more likely to be bluer in comparison to any nearby detections, not just those in the cluster (see next section).

As expected, the non-detections (being ETGs) define a red sequence, while the detections trace a blue sequence. The non-VCC galaxies detected in H i in the cluster are almost exclusively faint, blue objects – small, gas-rich objects (see Section 4.5) are easier to detect from their H i content than in an optical survey.

Only one of the non-VCC detections appears to lie well on the red sequence. In this case, the optical candidate counterpart is small and faint, and may be a misidentified background object (although there are no obvious alternative counterparts). Conversely, one of the detected dEs is seen to lie on the blue sequence. This very faint object (VCC 1964) appears to be irregular in the SDSS images, so this may be a morphological misclassification.

The other detected early-types are all clearly members of the red sequence or in the transition region between red and blue. One other H i-detection is also seen in the red sequence, VCC 867. This is classed as a dI, but the assignation is questionable given its low optical luminosity and red colour.

4.5 H i mass-to-light ratio

The H i mass-to-light ratio varies as a function of absolute magnitude, as shown in Fig. 15. Fainter galaxies tend to be more gas-rich than brighter galaxies. This is at least partially a sensitivity effect – fainter galaxies require a higher gas fraction for the gas mass to be above the sensitivity limit. This does not explain why brighter galaxies do not have equally high mass-to-light ratios, which may perhaps relate to a ‘saturation’ density of H i above which all gas is molecular (Leroy et al. 2008).

Galaxies within the Virgo cluster generally have lower /L_g ratios than background galaxies of similar luminosities. This is an effect of both Malmquist bias and also related to environment, with cluster galaxies being subject to more gas loss than field objects (see Section 4.6).

It is apparent that with the exception of VCC 611, the detected dEs are relatively faint and gas-rich, of comparable ratio to the late-types of similar magnitude (i.e. dIs). The detected S0s are brighter and also rather poorer in gas content compared to most other detections of similar magnitude (i.e. spirals).

The effect of the variation in gas fraction at the same magnitude is perhaps reflected in the colours of the galaxies. Fig. 16 plots the ratio against g−i colour. Again there is much scatter but a clear trend – the most gas-rich galaxies are also the bluest.

It is difficult to assess whether there is any trend at all in terms of colour and gas fraction in terms of the ETGs. While the three detected S0s are among the reddest and most gas-poor objects, the detected dE VCC 611 is of comparable colour and gas fraction. The other three detected dEs are all of a very similar gas fraction to each other, but with a huge variation in colour – VCC 1964 is much bluer than the others, and as we saw in Section 4.4 it appears likely that this object is actually a dI.

4.6 H i deficiency

The distribution of H i deficiency is shown in Fig. 17. The deficiency for an isolated object is typically (our background sample has more scatter than this, probably because most background objects are more distant and subject to greater measurement errors). The mean deficiency for this cluster sample is 0.7, with 76 per cent greater than 0.3. In comparison, the mean deficiency in the background objects is 0.1.

The deficiencies confirm that there is indeed an environmental influence of the cluster acting to remove gas. Cluster galaxies typically show greater deficiencies than background objects. Although some cluster galaxies show very low deficiencies, 35 per cent of background objects have negative deficiencies.

There is a hint of variation in deficiency with distance from M49, the dominant galaxy in this part of the cluster. Fig. 18 shows the deficiency as a function of projected distance from M49 and declination for comparison. Although the trend is weak, 81 per cent of the 21 galaxies with deficiency >1.0 are within 2°.5 of M49. This bias in the distribution is not reproduced in the galaxies of lower deficiency, which are distributed evenly with distance from M49 – of the 57 objects width deficiency <1.0, 58 per cent are within 2°.5 of M49.

M49 lies in the centre of an X-ray emitting gas cloud, as shown in Fig. 1, and is believed to lie at 17 Mpc distance. As shown in Fig. 18, some of the strongly deficient objects within 2°.5 of M49 are purported to be at 23 Mpc. However, even if gas cloud associated with M49 is the cause of the high deficiencies, it does not necessarily imply that the cloud is 6 Mpc deep given the distance uncertainties discussed in Section 4.2. It is also worth briefly noting that deficiency is seen to remain high even as far as 4° from M49 (as well as in the cloud at 32 Mpc distance). This indicates that the environmental influence of the cluster probably continues beyond the edge of this region.

4.7 Optically undetected gas clouds and other exotica

In this survey, 199 objects have been assigned sure optical counterparts. An additional 72 have only a single likely optical counterpart. Of the remainder, the majority are either unsure H i detections, have multiple possible optical counterparts, or a bright foreground star is present that could be obscuring the corresponding galaxy. But in a few instances, the H i appears to be unassociated with any optical galaxies. We caution that determining which objects have merely an unusually faint optical counterpart, and which genuinely have none, is a highly subjective process. Within an area the size of the Arecibo beam there are invariably optical smudges that could potentially be associated with the H i. However, in the detections discussed below, the only possible optical counterparts are so extreme that their association with the H i must be questioned.

Only those cases where the H i detection is within the Virgo cluster are considered here, since more distant galaxies (if of particularly high ratios) might simply be too faint to detect optically. We show finding charts and spectra in Figs 19 and 20. The gas masses for these objects are amongst the lowest in the sample, with the highest at only 4 × 10⁷ M_⊙ (all have been confirmed by follow-up observations). They share few other similarities with each other – two detections have extremely low velocity widths (W20 ∼ 35 km s⁻¹) while the others are more typical of the rest of the sample (W20 ∼ 150 km s⁻¹).

The lack of normal optical counterparts for these objects suggests three possibilities: (1) the H i is associated with faint, optically unusual galaxies; (2) the H i was once associated with more typical optical galaxies but has since been removed and (3) the H i was never associated with any optical emission – these are the so-called ‘dark galaxies’. Of course it is possible that these detections are the result of all of these effects at work and there is no common explanation for all seven of these objects.

We cannot entirely discount any of these possibilities. The first option requires optical redshifts to determine if any optical emission is genuinely associated with the H i. Without this, we can only say that the optical counterpart candidates would be extremely unusual. For instance, the only remotely plausible candidate for AGESVC1_231 would be the bluest object in the entire sample, and also very compact (optical diameter of 2.3 kpc) despite the relatively high H i velocity width (W20 of 152 km s⁻¹).

It is also difficult to explain these detections as tidal debris – however, the formation of gas clouds by gravitational interactions is not well understood. The purported dark galaxy VIRGOHI21 (Minchin et al. 2010) was found to be part of a much larger extended structure (Haynes, Giovanelli & Kent 2007), greatly adding to the complexity of the situation. Duc & Bournaud (2008) demonstrated that such features may, potentially, be formed from high-speed as well as low-speed interactions, so that the original interacting galaxies may become widely separated over time. While the above detections do not appear to be embedded in large-scale extended H i features, the survival of such streams in the intracluster medium depends upon many complex factors (see Kapferer et al. 2009).

The third possibility, that these detections represent optically dark galaxies, presents its own difficulties. Ideally, of course, we would prefer a dark galaxy candidate to show all the features of more normal H i detections: a reasonably wide H i line with a clear double horn, the signature of a rotating disc. While five of these detections have reasonable velocity widths, none is double-horned. On the other hand, many dIs show both similar velocity widths and profile shapes. To prove both that these gas clouds do not have any optical counterparts and did not form via tidal interactions is a formidable challenge, requiring both optical redshift measurements as well as deeper (ideally higher resolution) H i observations. Even then, accepting that the gas clouds are embedded in gravitationally bound dark matter haloes requires something of a leap of faith.

VCC 1249 (AGESVC1_281) is also worth mentioning, as it is the only clear example of a detection showing gas stripping actually occurring. We show a renzogram of this detection in Fig. 21– contour maps of each channel are shown with a different colour for each velocity, overlaid on an SDSS g-band image. As the previous, higher resolution observations of Henning, Sancisi & McNamara (1993) also demonstrated, the H i is separated from the late-type VCC 1249 and lies mid-way between that galaxy and M49 (the same is true in velocity: VCC 1249 is at 276 km s⁻¹, the H i is at 499 km s⁻¹ and M49 at 997 km s⁻¹). Since this is the only example where we can claim confidently to have detected gas removed from a LTG (which is itself devoid of H i), it may be that in general the gas is rapidly ionized once removed from its host galaxy.

5 SUMMARY AND DISCUSSION

AGES detects 89 objects within the Virgo cluster in a 10 deg² area. 26 of these are new H i detections, while 21 of these are not members of the VCC. Of those 21 non-VCC galaxies detected in H i, 14 are optically brighter than the VCC completeness limit. The fact that AGES does not uncover many more new detections is likely due to its mass sensitivity, which is comparable to that of the VCC, and the fact that H i is almost always associated with an optical galaxy.

The basic structure of this region is of three clouds at 17, 23 and 32 Mpc distance. Both the H i and optical studies (i.e. the VCC) support this view, with all of the new H i detections being interspersed with the optical detections (both spatially and in position–velocity space). However, a greater fraction of the H i detections are found at higher velocities than the non-detections. This suggests that the gas removal mechanism(s) acting within the cluster proper are less significant within the infalling groups.

Although 76 per cent of H i detections in this region are more deficient than 0.3, there are still 24 per cent that are effectively non-deficient. This includes many of the new, non-VCC detections, which are generally small, blue and gas-rich (M_g fainter than −16, g−i bluer than 0.7 and with ratios greater than 0.3). Highly deficient (deficiency >1.0) giant galaxies apparently coexist with non-deficient dwarfs in the same population of objects. Simulations by Vollmer et al. (2001) and Smith (2009) have shown that dwarf galaxies will lose all their gas through ram-pressure stripping on their first pass through the cluster centre. Observationally some giant galaxies have lost more gas than some non-deficient dwarfs currently contain. Both of these statements imply that the non-deficient dwarfs cannot possibly have experienced the same environmental effects as the highly deficient giants. More likely the dwarfs are recent arrivals to the cluster, which have not yet had time to experience its full environmental effects.

Moreover there is a broad trend in H i deficiency. Most of the strongly deficient (greater than 1.0) galaxies are found close to M49, suggesting that the most efficient gas removal is only happening in this region. It also implies that the cluster is still assembling, otherwise galaxies that lose gas in the cluster centre would have had time to move to the outer regions (so we would see highly deficient galaxies everywhere).

It is significant that we detect every (or very nearly) LTG in H i. This is a surprising result, as we should not be able to detect deficient late-type dwarfs. In fact, we detect 20 galaxies which if they were more deficient than 1.5 would be undetectable. The question arises: where are all the undetected late-type objects? It seems very unlikely that only the giant galaxies have been subjected to the full force of the gas removal mechanism acting within the cluster, so deficient giants imply the presence of deficient dwarfs. Yet apparently there are none. It seems that dwarf galaxies are either gas-rich late-types or, as demonstrated by the lack of a detection through stacking, extremely gas-poor early-types.

Unless all dIs are recent arrivals to the cluster, this could suggest that galaxies which lose all of their gas content morphologically evolve into ETGs. This scenario has been previously described and modelled by Boselli et al. (2008), who found that ram-pressure stripping can produce a rapid gas depletion and morphological transformation. The rapidity (<150 Myr) of the process would explain why no LTGs are undetected in H i. Thus, any H i present in ETGs is either the last trace of their original gas or has been replenished by dying stars returning some of their gas to the interstellar medium. Certainly our results cannot rule out the Boselli et al. (2008) scenario merely by detecting three ETGs in H i– rather, the fact that all late-types are detected (whereas so few dEs seem to have any H i at all) is consistent with their scenario.

The S0s are, as we have seen, detected much more often than the dEs, have low ratios and two-third show signs of structure. Observationally, the detected S0s are broadly similar to spiral galaxies, but with particularly low gas fractions and a wide range of velocity widths (as can be seen directly in Fig. 22). The natural implication is that these objects have evolved from spirals whose gas content has been depleted. Although Boselli & Gavazzi (2006) find that massive S0s are unlikely to be formed via gas removal from spirals, they leave open the possibility that this may explain the origin of low-mass S0s.

In complete contrast, the dEs are very rarely detected in H i, but those that are have high ratios and show no obvious signs of structure. One of them appears to lie on the blue sequence, and the added complication of misidentification is more acute for these objects because of their low luminosities. Discounting this exception, the detected dEs are generally similar to dIs, except that they are much redder (as can be seen in Fig. 22). Arguably this lends further credence to the model of Boselli et al. (2008). Since their gas fractions are comparable to the dIs, however, it is also possible that these particular detections have only recently entered the cluster. ALFALFA have found that H i is far more common in ETGs in the field, with around 44 per cent of morphologically ETGs being detected in H i (as opposed to 9 per cent in the cluster) – see Grossi et al. (2008). If the dEs detected here are indeed undergoing morphological evolution, then this raises the question of why they are red and structureless but with comparable gas fractions to many dIs.

In short, while it is perfectly possible that dEs in general have evolved from dIs within the cluster, the particular detections of this survey do not substantiate this view. In contrast, the S0s we have detected appear much more likely candidates for examples of ongoing, gas-loss-driven morphological evolution.

This region of the cluster is rich in H i-detected galaxies of all morphological types, many of which appear to be entering the cluster for the first time. In some ways it is an ideal location to search for dark galaxies. Seven candidates are found, about 7 per cent of the H i detections within the cluster in this region. This is well below earlier predictions of 23 per cent for AGES and there are a huge number of additional caveats. We cannot exclude the possibility that these detections do have optical counterparts, albeit ones with unusual properties. Nor can we disregard the alternative option that the H i detections were formed by gas removal from some unidentified ordinary galaxies. Even if neither of these scenarios is true, it would still be a substantial leap to say that these gas detections are really gravitationally bound in dark matter haloes.

Yet it would also be premature to reject entirely the notion that these are indeed dark galaxies. Though the detection rate is below predictions, this may be due to the influence of the cluster environment. The alternative explanations for these objects also suffer from difficulties – the possible optical counterparts which are visible are too compact, and the unresolved nature of the H i detections makes it difficult to ascribe their formation to gas removal from any nearby galaxies. While the dark galaxy scenario remains a controversial hypothesis, current observations cannot determine if this explanation is any more or less valid than the others.

This work is based on observations collected at Arecibo Observatory. The Arecibo Observatory is operated by SRI International under a cooperative agreement with the National Science Foundation (AST-1100968), and in alliance with Ana G. Méndez-Universidad Metropolitana and the Universities Space Research Association.

This research has made use of the GOLDMine Database.

This research has made use of the NASA/IPAC Extragalactic Database (NED) which is operated by the Jet Propulsion Laboratory, California Institute of Technology, under contract with the National Aeronautics and Space Administration.

This work has made use of the SDSS. Funding for the SDSS and SDSS-II has been provided by the Alfred P. Sloan Foundation, the Participating Institutions, the National Science Foundation, the US Department of Energy, the National Aeronautics and Space Administration, the Japanese Monbukagakusho, the Max Planck Society and the Higher Education Funding Council for England. The SDSS website is http://www.sdss.org/.

The SDSS is managed by the Astrophysical Research Consortium for the Participating Institutions. The Participating Institutions are the American Museum of Natural History, Astrophysical Institute Potsdam, University of Basel, University of Cambridge, Case Western Reserve University, University of Chicago, Drexel University, Fermilab, the Institute for Advanced Study, the Japan Participation Group, Johns Hopkins University, the Joint Institute for Nuclear Astrophysics, the Kavli Institute for Particle Astrophysics and Cosmology, the Korean Scientist Group, the Chinese Academy of Sciences (LAMOST), Los Alamos National Laboratory, the Max-Planck-Institute for Astronomy (MPIA), the Max-Planck-Institute for Astrophysics (MPA), New Mexico State University, Ohio State University, University of Pittsburgh, University of Portsmouth, Princeton University, the United States Naval Observatory and the University of Washington.

REFERENCES