THE ASTRONOMICAL JOURNAL, 124:1452-1463, 2002 September
© 2002. The American Astronomical Society. All rights reserved. Printed in U.S.A.

We report the discovery of significant stellar substructure in the halo and outer disk of our nearest large galactic neighbor, M31. Our deep panoramic survey with the Isaac Newton Telescope Wide Field Camera currently maps out an area of 25 deg² around M31, extending along the semimajor axis to 55 kpc and is the first to allow an uninterrupted study of the density and color distribution of individual red giant branch stars across a large fraction of the halo of an external spiral galaxy. We find evidence for both spatial density and metallicity (as inferred from color information) variations, which are often, but not always, correlated. In addition to the previously reported giant stellar stream, the data reveal the presence of significant stellar overdensities at large radii close to the southwestern major axis, in the proximity of the very luminous globular cluster G1, and near the northeastern major axis, coinciding with and extending beyond the previously known northern spur. The most prominent metallicity variations are found in the southern half of the halo, where two large structures with above average metallicites are apparent; one of these coincides with the giant stellar stream while the other corresponds to a much lower level stellar enhancement. Our findings contrast with, but do not conflict with, past studies of the M31 halo and outer disk that have suggested a rather homogeneous stellar population at large radius: the bulk of our newly detected substructure lies in the previously uncharted far outer regions of the galaxy. We discuss the possible origin of the substructure observed and the implications it has for constraining the galaxy assembly process.

Key words: galaxies: evolutiongalaxies: halosgalaxies: individual (M31)galaxies: structureLocal Group

On-line material: color figure

     ¹ Based on observations made with the Isaac Newton Telescope, operated on the Island of La Palma by the Isaac Newton Group in the Spanish Observatorio del Roque de los Muchachos of the Instituto de Astrofísica de Canarias.

     Within the context of the hierarchical clustering theory for structure formation, large disk galaxies like the Milky Way and M31 arise from the merger and accretion of many smaller subsystems and from the smooth accretion of intergalactic gas (e.g., White & Rees 1978; Steinmetz & Navarro 2002). While the former process governs the growth of the dark matter halo (by far the dominant mass component in a disk galaxy) and much (if not all) of the stellar spheroid, angular momentumconserving collapse of the latter is the currently favored mechanism for forming the thin disk. It has been argued that the relative importance of these processesi.e., smooth versus discrete acquisition of massin the assembly of a galaxy is the primary factor that determines the final morphology (e.g., Baugh, Cole, & Frenk 1996; Steinmetz & Navarro 2002).

     Cosmological simulations of galaxy formation have become increasingly sophisticated in recent years and have now attained sufficient resolution to begin to address the internal structures of galaxies. A key result to emerge from state-of-the-art simulations within the popular cold dark matter (CDM) framework is that the dark matter subhalos that merge to form massive galaxies are much more resilient than previously thought (Klypin et al. 1999; Moore et al. 1999). Although dynamical friction and galactic tidal forces continually act to disrupt subhalos once they fall within the potential of the massive system, the dense central cores of the satellites appear to survive as distinct entities for at least several orbital timescales (e.g., Hayashi et al. 2002). The total mass in undisrupted cores is expected to be quite small (10% of the total mass of the system), with the bulk of the mass distributed much more smoothly; however, several hundred cores are expected to reside within the virial radius of a typical galaxy like the Milky Way.

     Signatures of the hierarchical nature of galaxy assembly are expected to be most obvious in the properties of the halo stellar populations. At least some of the accreted satellites will have experienced significant star formation prior to and during incorporation into the final system. As these systems will span a range of masses and have experienced different dynamical histories within the massive galaxy potential, it seems highly likely that they will also be characterized by distinct star formation and chemical enrichment histories, the details of which will be imprinted on the ages and metallicities of their constituent stars. Indeed, studies of Milky Way satellite stellar populations reveal a surprising variety of complex star formation histories, with no two dwarfs sharing a similar evolution (Mateo 1998). During the accretion of a satellite galaxy onto a massive host, tidal forces first shock the dark matter, then the stars, of the satellite, making them gravitationally unbound. Tidal debris will be deposited in both a leading and a trailing stream which, depending on the satellite's orbit and initial mass and the sphericity of the host potential, can maintain spatial and kinematic coherence within the halo for many gigayears (Johnston, Hernquist, & Bolte 1996; Helmi & White 1999; White & Springel 2000; Bullock, Kravtsov, & Weinberg 2001). It is therefore expected that the stellar halos of galaxies that form hierarchically should posses significant spatial and metallicity substructure in the form of disrupted satellites (e.g., dwarf galaxies and globular clusters) and their stellar detritus. This structure will be particularly apparent in the outer regions of the halo, where the dynamical mixing timescales are long. Furthermore, as there is no reason to believe that the accretion history will be exactly the same in different galaxies (for example, the number and masses of accreted satellites and their mix of stellar populations), one might expect to see noticeable differences in the properties of galactic stellar halos, even between systems of similar morphological type.

     It has also been postulated that some fraction of the accreted satellites lack a luminous stellar component, either due to feedback from early star formation, that expels the gas before significant star formation can occur (e.g., White & Rees 1978; Dekel & Silk 1986), or from the inhibition of gas cooling due to an ionizing UV background (e.g., Bullock, Kravtsov, & Weinberg 2000). The main reason for invoking these scenarios is the striking discrepancy between the numbers of satellites predicted to lie within the dark halo of a massive disk galaxy and the numbers of dwarf companions actually observed around the Milky Way and M31 (Klypin et al. 1999; Moore et al. 1999). If correct, detecting these CDM substructures through stellar density enhancements becomes very difficult, if not impossible. On the other hand, such clumpy structures will still exert an influence on the host galaxy by causing tidal heating and/or distortion of the thin stellar disk, particularly in the fragile outer regions (Moore et al. 1999; Font et al. 2001).

     Searching for and studying stellar substructure in the outskirts of disk galaxies provides an important test of CDM models of structure formation. Studies of the Milky Way stellar halo have so far provided encouraging support for hierarchical galaxy assembly. Evidence includes the discovery of the tidally disrupted Sagittarius dwarf galaxy and associated stellar streams (e.g., Ibata, Gilmore, & Irwin 1995; Mateo, Olszewski, & Morrison 1998; Majewski et al. 1999; Yanny et al. 2000; Ibata et al. 2001b; Ibata et al. 2001c), the phase-space clumping of halo stars in the solar neighborhood (Helmi et al. 1999), and the tidal tails emanating from halo globular clusters (e.g., Grillmair et al. 1995; Odenkirchen et al. 2001; Leon, Meylan, & Combes 2000) and dwarf spheroidal galaxies (e.g., Irwin & Hatzidimitriou 1995; Majewski et al. 2000). Unfortunately, our location within the disk sometimes renders interpretation of structures revealed through star counts rather difficult (e.g., Newberg et al. 2002) and underscores the need for an extragalactic perspective. It is thus clearly desirable to test whether the halos of other disk galaxies also exhibit substructure and to investigate how the nature of this substructure (as well as the nature of the field population resulting from totally disrupted satellites) varies with the properties of the host galaxy.

     Searches for low surface brightness tidal and/or extraplanar features have been carried out around several external edge-on disk systems but to date only one unambiguous detection has been made, in NGC 5907 (Shang et al. 1998). On the other hand, such studies are technically very difficult, and inaccurate flat-fielding and/or low-level scattered light can easily mask real signal at these levels (_V 28 mag arcsec^-2). A more preferable search technique involves using resolved star counts to map out the spatial density structure in nearby galaxy halos; however, with ground-based telescopes, this method is currently limited to galaxies within and around the Local Group. As these galaxies also subtend the largest angular sizes on the sky, quantitative study requires wide-field CCD mosaic cameras, as well as sophisticated processing techniques capable of dealing in an optimal way with large amounts of data, both of which have become available only in recent years.

     Using the Wide-Field Camera on the Issac Newton Telescope (INT WFC), we are carrying out a panoramic imaging survey of our nearest large neighbor, M31. The contiguous nature of our survey allows us to distinguish local density enhancements in the halo of M31 and outer disk from fluctuations in background galaxy counts and the foreground distribution of Galactic stars. The present paper reports interim results from this survey, namely, the discovery of significant spatial density and metallicity substructure in the red giant population of the halo and outer disk. The stellar halo of M31 has long been known to have rather different properties from the Milky Way halo, despite the overall similarity in the global properties of the galaxies. In particular, ground-based and Hubble Space Telescope (HST) studies indicate the M31 field halo is roughly an order of magnitude more metal-rich (e.g., Morris et al. 1994; Rich et al. 1996; Holland, Fahlman, & Richer 1996; Durrell, Harris, & Pritchet 2001; Reitzel & Guhathakurta 2002) and denser and/or larger (Reitzel, Guhathakurta, & Gould 1998) than that of the Milky Way. Our INT WFC survey reveals significant stellar substructure in the outskirts of M31 and provides the tantalizing suggestion that these halo differences may be due to a more active accretion or merger history within the M31 subsystem. Preliminary results, drawn from the first phase of our survey, were reported in Ibata et al. (2001a).

     The Wide Field Camera on the 2.5 m Isaac Newton Telescope is a four-chip EEV 4 K × 2 K CCD mosaic camera that images 0.29 deg² per exposure (Walton et al. 2001). On the nights of 2000 September 39, 2001 October 916, and 2001 November 13, we used this camera to image 91 contiguous fields (corresponding to 25 deg²) in the outer disk and halo of M31. Coverage currently extends to 40 (54 kpc) and 25 (34 kpc in projection) along the major and minor axes, respectively.

     Images were taken in the equivalent of Johnson V and Gunn i bands under mainly good atmospheric conditions, with 85% of the fields taken in photometric conditions with seeing better than 12. The exposure time of 8001000 s per passband per field allowed us to reach i = 23.5 and V = 24.5 (S/N 5), and is sufficient to detect individual red giant branch (RGB) stars to M_V 0 and main-sequence stars to M_V -1 at the distance of M31. Several fields taken in poorer conditions were reobserved and co-added as necessary to give an approximately uniform overall survey depth.

     All the on-target data plus calibration frames were processed using the standard INT Wide Field Survey (WFS) pipeline provided by the Cambridge Astronomical Survey Unit (Irwin & Lewis 2001). This package provides the usual facilities for instrumental signature removal, including in this case defringing of the i-band data, plus tools for object catalog creation, astrometric and photometric calibration, morphological classification, and cross-matching catalogs from different observations. The pipeline processing provides internal cross-calibration for the four CCDs at a level better than 1% within each pointing. Field-to-field variations in photometric zero points were calibrated and cross-checked using a combination of multiple nightly photometric standard sequence observations and the overlap regions between adjacent WFC pointings. The overall derived photometric zero points for the whole survey are good to the level of ±2% in both bands. The full details of the survey strategy, data processing, and calibration will be presented elsewhere.

     Objects were classified as noise artifacts, galaxies, or stars according to their morphological structure on all the images. As a sanity check, we compare our extended source counts in the outer halo fields with the deep V- and I-band galaxy counts of Smail et al. (1995). Over the magnitude ranges 22.5 < V < 23.5 and 20.5 < i < 21.5, we detect 11,500 and 7300 galaxies per deg², respectively, which can be compared with the 10,000 deg^-2 in each band predicted using Smail et al. (1995), indicating that we are not misclassifying large numbers of faint galaxies. In the outer halo fields, we typically detect equal numbers of stars and galaxies within the magnitude and color ranges of interest (see § 3.1 and Fig. 1). Of the extended sources, approximately 20% are compact in the sense of being within the 35 range of the stellar boundary in the classification statistic and having an ellipticity of less than 0.4. Making the plausible assumption that half these are genuine galaxies (the other half being genuinely stellar) leads us to expect that the contamination due to misclassified, barely resolved field galaxies is small and generally considerably less than 10% of the total number of detected sources. Indeed, the overdense regions we will discuss in this paper have stellar densities at least a factor of 2 higher, and the average star counts are a factor of 3 higher, than those in the outer halo fields.

FIG. 1.Selection criteria adopted to isolate stars in different regions of the color-magnitude diagram. The blue RGB (20.5 < i < 22.5; 8.7-0.4i < V-i < 23.5-i) and red RGB (20.75 < i < 21.75; V-i > 2.0) cuts are designed to select stars from the metal-poor and metal-rich sides of the red giant branch, respectively. The AGB cut (19.75 < i < 21.00; V-i > 2.5) is designed to select intermediate-age, moderate-metallicity stars in the thermally pulsing regime of the asymptotic giant branch. Field-to-field corrections for the varying Galactic foreground population (see § 3.2) are calculated using star counts in the region defined by 18.0 < i < 19.5 and 0.5 < V-i < 2.5. Also overlaid are the giant branch fiducial sequences of several Galactic globular clusters: left to right, NGC 6397 ([Fe/H] = -1.9), NGC 1851 ([Fe/H] = -1.3), 47 Tuc ([Fe/H] = -0.7), and NGC 6553 ([Fe/H] = -0.3), taken from Da Costa & Armandroff (1990) and Sagar et al. (1999). These sequences have been adjusted to an adopted M31 distance modulus of 24.47 (Holland 1998; Durrell et al. 2001) and a reddening of E(B-V) = 0.07. A distance modulus of 13.7 and intrinsic reddening of E(V-I) = 0.95 was assumed for NGC 6553 (Sagar et al. 1999). In placing the fiducials on the (V, i) plane, we have made use of the color equations derived for the WFS data.

     Stellar contamination from the Galactic foreground steadily increases to the northeast, because of the proximity of the Galactic plane, rising smoothly from an average contamination of 13,000 stars deg^-2 at the southwest extremity of the survey to 20,000 stars deg^-2 at the northeast extremity (integrated over all magnitudes). This foreground variation, coupled with our current lack of suitable comparison fields uncontaminated by M31, constrains to some extent detailed quantitative analysis of the spatial and metallicity distribution of the outer parts of the halo.

     Our INT WFC survey of M31 provides the first opportunity to make an uninterrupted study of the properties of the resolved stars across a large fraction of an external disk galaxy. "Stellar" sources in our catalog consist of M31 red giants, M31 upper main-sequence stars, Galactic foreground stars, and unresolved background galaxies. To examine the spatial distribution of various stellar populations within M31, we apply a series of magnitude and color cuts that are designed to isolate stars in different regions of the color-magnitude diagram (CMD). Figure 1 illustrates the main selection criteria adopted in our analysis. We define the blue RGB as stars with 20.5 < i < 22.5 and 8.7-0.4i < V-i < 23.5-i, and the red RGB as stars with 20.75 < i < 21.75 and V-i > 2.0. These magnitude and color ranges are consistent with those expected for red giant stars at the distance of M31. We also overlay several Galactic globular cluster fiducials spanning a range in metallicity. Comparison between our selection criteria and these fiducials demonstrates that our RGB cuts isolate the metal-poor (-2.0 [Fe/H] -0.6) and metal-rich (-0.6 [Fe/H] -0.3) sections of the giant branch, respectively.

     Figure 2 shows the standard coordinate projection of the surface density distribution of blue and red RGB stars across our current 25 deg² survey area. Each detected source classified as stellar on the i band and falling within the aforementioned magnitude and color limits is encoded as a point on these diagrams. The lower magnitude limits are conservatively set at roughly 1 mag brighter than the survey 5 detection threshold to mitigate the effects of varying completeness. The immediate impression from Figure 2 is the striking nonuniformity of the stellar distribution at large radii. The large-scale morphology of stars in the blue RGB map is that of a significantly flattened inner halo structure, while the bulk of the stars in the red RGB map appear associated with the outskirts of the stellar disk. Numerous stellar enhancements are present in both maps, with some features appearing more conspicuous in one color cut than another. The giant stellar stream discovered during the first phase of our survey (Ibata et al. 2001a) is apparent as an enhancement close to, but distinct from, the southern minor axis and extending out to 40 kpc. While visible in both maps, the feature appears considerably sharper in the distribution of red RGB stars. In addition, significant stellar overdensities are seen at large radii lying close to both major axes (at -15, -18 and +08, +18). A lower intensity enhancement is visible in the form of a diffuse extension northeast of the center (toward +17, +07). Very faint structure, some of which is at the limit of detectability, is seen emanating from the northern side of the disk. No correction has been applied to the maps for Galactic foreground contamination; however, we have verified that the distribution of stars with properties expected to belong to this population (selected from the box indicated in Fig. 1) is smooth across the entire survey area, and thus it is unlikely that Galactic foreground variations are causing any of the structure seen in Figure 2. Likewise, the Galactic extinction toward the outskirts of M31 is relatively uniform, ranging from E(V-i) = 0.06 to 0.11 (Schlegel, Finkbeiner, & Davis 1998). There is no evidence for significant clumping in the distribution of extended sources (i.e., background galaxies) across our survey area, and in any case galaxy clusters and other large-scale structure would be unlikely to dominate the counts at the magnitudes, colors, and angular scales of the observed features. Indeed, Couch, Jurcevic, & Boyle (1993) show that the galaxy-galaxy correlation function is very low on scales larger than 01 at the magnitudes of relevance for our survey. We thus conclude that the spatial density enhancements reflect real substructure associated with the halo and outer disk of M31.

FIG. 2.Left: Standard coordinate projection of the surface density of blue RGB stars across our current 25 deg2 survey area. The inner and outer ellipses are drawn assuming a position angle of 381, derived from analysis of the light distribution on a scanned Palomar Sky Survey plate. The outer ellipse denotes a flattened ellipsoid (aspect ratio 3 : 5) of semimajor axis length 55 kpc and indicates the current spatial extent of the survey. The inner ellipse has a semimajor axis of 2°(27 kpc) and represents an inclined disk with i = 77.5; the optical disk of M31 lies well within this boundary. The few white blotches indicate regions contaminated by saturated stars. The dwarf companions M32 and NGC 205 lie at (0°, -04) and (-05, 04), respectively. Much substructure is seen at large radii, including the giant stellar stream and stellar overdensities at both extremes of the major axis. No corrections have been made for foreground or background contamination. Right: Same as (left), except showing the surface density of red RGB stars. Note the lower Galactic foreground contamination on this map. Comparison with the projection on the left clearly indicates that the morphology of the substructure varies as a function of color.

     The substructure located near the southwestern major axis is of particular interest because it is located in the proximity of the very luminous globular cluster G1 (Meylan et al. 2001). This clump, which we will hereafter refer to as the G1 clump although the nature of the association with G1, if any, is presently unclear, is located at 35 kpc in projected radius and has a physical size of 07 × 05, or 10 × 7 kpc at the distance of M31. Compared with the mean density of RGB stars at similar radii around M31, we calculate that the G1 clump is overdense by a factor of 4. Integrating the excess RGB population over this region relative to nearby comparison fields and assuming a luminosity function similar to the field population studied by Rich et al. (1996) gives estimates for the total apparent magnitude of this feature of m_V = 12.1 and m_i = 10.5, with average surface brightnesses of around 28.5 and 27.0 mag arcsec^-2, respectively. At an average extinction of E(B-V) = 0.07 this is equivalent to absolute magnitudes of M_V = -12.6 and M_i = -14.1. While the magnitude and color of the clump are comparable within the errors to those of dwarf satellites in the M31 and Milky Way subgroups, the measured surface brightness is several magnitudes fainter than any of the currently known systems; on the other hand, it is similar to that measured for the low surface brightness ring in NGC 5907 (Shang et al. 1998).

     The density enhancement lying near the northeastern axis coincides with and extends beyond the "northern spur" first remarked on by Walterbos & Kennicutt (1988, hereafter WK88) as faint light bending away from the major axis (see also the deep images of Innanen et al. 1982). WK88 questioned whether this feature was actually associated with M31 or was due to a Galactic reflection nebula: our color-magnitude diagrams of this region unambiguously place the excess stars at the same distance as M31. The feature extends to a projected radius of 30 kpc at angles up to 20° off the major axis. The northern spur region is about a factor of 1.52 times more overdense than the G1 clump and thus has a mean surface brightness 0.50.8 mag arcsec^-2 higher. The northern spur region also appears as a low-level enhancement on a spatial density plot of stars selected to lie just above and redward of the RGB tip (see Figs. 1 and 3). Such stars are likely to be intermediate-age, moderate-metallicity ([Fe/H] -0.7; ages 38 Gyr) thermally pulsing asymptotic giant branch (AGB) stars (Girardi et al. 2000), and their presence is taken as evidence for an extended epoch of star formation in this region. There is also a very tentative detection of AGB stars in the vicinity of the G1 clump; however, all other substructure appears devoid of this younger population. While the effects of crowding will lead to some spurious structure in the AGB map (e.g., the strong detection of NGC 205), this is unlikely to be important in the diffuse outer regions of the galaxy under study here.

FIG. 3.Standard coordinate projection of the surface density of stars lying redward of and above the RGB tip. These stars are likely to be intermediate-age, moderate-metallicity thermally pulsing asymptotic giant branch stars. Of the substructure visible in the RGB map, only the northern spur is clearly detected on the AGB map; there is also marginal enhancement in the vicinity of the G1 clump. No corrections have been made for foreground or background contamination.

     Perhaps surprisingly, our current map of the RGB density around M31 reveals no obvious northern counterpart to the giant stellar stream discovered near the southern minor axis (Ibata et al. 2001a). We will return to this issue in the discussion below.

     Further insight into the nature of the stellar density variations in the halo and outer disk of M31 comes from analysis of the spatial variation of mean RGB color. Figure 2 already reveals that the morphology of the M31 substructure has a color dependence. This is further illustrated in Figure 4, which shows color-magnitude diagrams for fields in the giant stellar stream (right) and the G1 clump (left). Comparison with Galactic globular cluster fiducials reveals that the mean RGB color varies between these pointings, being slightly more metal-rich than 47 Tuc in the stellar stream and approximately equal to 47 Tuc in the G1 clump. The width of the G1 clump RGB also appears somewhat narrower than that of the stellar stream, indicating a small intrinsic metallicity and/or age dispersion. To quantify the RGB color variation as a function of position over the survey area, it is first necessary to make a statistical correction for the foreground Galactic contamination. Three outer halo fields were used to define a reference foreground population (albeit with a small halo contamination still present in these fields) and the scale factor required to match the target field i-band luminosity function brighter than i = 20 was computed using the foreground box in Figure 1. For each field, we then calculate the foreground-corrected median color distribution in the magnitude range 21 < i < 22 projected in a coordinate system orthogonal to the locus of the ensemble RGB for M31. This color measure quantifies the shift in the mean RGB locus as a function of field pointing. Figure 5 shows a color-coded plot of this variation. The "average" halo RGB color is denoted here as green; assuming an old stellar population, this corresponds to a metallicity of [Fe/H] -0.7, similar to 47 Tuc. Progressively bluer (or lighter gray) shades indicate bluer mean RGB colors while progressively redder shades (or darker gray) indicate redder mean RGB colors.

FIG. 4.Examples of two color-magnitude diagrams from our WFC survey. Left, field in the giant stellar stream; right, field in the G1 clump. Overplotted are the same globular cluster fiducial sequences as shown in Fig. 1. Many of the stars more luminous than the RGB tip are foreground contaminants. Stellar density varies between the CMDs as a result of the different overdensities of the features. We argue that the different mean colors and widths of the RGB primarily reflect intrinsic metallicity variations.

FIG. 5.Map of the variation in mean RGB color across M31 (see text for details). Each WFC pointing is represented as a color-coded polygon. The full metallicity range spanned by the plot is 0.75 dex. Lighter gray shades represent regions with progressively bluer mean RGB colors, hence lower metallicities. Darker gray shades represent regions with redder mean RGB colors, hence higher metallicities. The lightest gray polygons seen mostly around the edges of the survey are pointings where the stellar density was too low to make a definitive measurement of the mean stellar color. The "average" halo RGB color is denoted as green; assuming an old stellar population, this corresponds to a metallicity of [Fe/H] -0.7, i.e., similar to 47 Tuc. Yellow, orange and red regions indicate progressively redder mean colors, and hence higher metallicities. Turquoise and light blue regions indicate progressively bluer mean colors, and hence lower metallicities. The full metallicity range spanned by the plot (red through light blue) is 0.5 dex. Dark blue regions represent pointings for which the mean stellar density was too low to make a definitive measurement of stellar color.

     We adopt the view here that metallicity variations are the dominant driver of these color variations. While age variations may contribute somewhat to the observed behavior, there is yet to emerge any compelling evidence for a luminous AGB component in the extended halo of M31 that would accompany a youngtointermediate-age population (e.g., Rich et al. 1996; Holland et al. 1996; also see Fig. 3). For reference, the isochrones of Girardi et al. (2002) suggest an age variation of 15 Gyr at [Fe/H] = -0.7 would be required to explain the entire color variation seen over the survey area. Based on the assumption of a uniformly old age for the M31 halo, comparison with Galactic globular cluster tracks and theoretical isochrones indicates that color index variations of ± 0.2 in V-i (the full range of colors used in Fig. 5) correspond to metallicity variations ranging from -0.5 dex below the mean to 0.25 dex above the mean. As the relationship between RGB color and metallicity is highly nonlinear across the range of interest (see Fig. 4), the quoted spread in metallicities is unfortunately rather uncertain. The observed color variations are well in excess of our photometric errors at these magnitudes and also significantly larger than the zero-point uncertainties. Although our strategy of bootstrapping the calibration of nonphotometric pointings from adjacent fields could introduce some correlated color variations, the size of this effect is also estimated to be rather small, amounting to no more than a few hundredths of a magnitude.

     Inspection of Figure 5 reveals that while there are no obvious radial metallicity gradients present in the halo, there are significant chemical inhomogeneities on large (i.e., several kiloparsecs) scales. Further, comparison of Figures 5 and 2 indicates that while substructure in the metallicity map often corresponds to overdensities on the spatial density map, this is not universally true. The most prominent chemical substructure is found on the southern side of the galaxy, in the region of the stellar stream and farther to the north on the same side, where only a diffuse extension of the halo is present on the spatial density map. Both these regions appear significantly more metal-rich than the average halo, and it is tempting to speculate that the two might be related, possibly tracing out the projected orbit of the object producing the stellar stream. We note that line-of-sight depth through the halo may cause intrinsic metallicity variations to be somewhat washed out, depending on how localized the substructure is in space and how large the overdensity is. The true metallicity variations in the halo are thus likely to be somewhat underestimated in Figure 5.

     Another apparently metal-rich region is the northern spur; at first glance this might appear to be even more metal-rich than the outer stellar disk (the edge of which is approximately delineated by the inner ellipse); however, the disk fields contain OB stars, which make them appear bluer, leading to spuriously low metallicity estimates. Similarly, age effects also complicate the interpretation of the spur, where we have evidence for the presence of an intermediate-age population (Fig. 3). The metal-rich nature of this feature appears rather robust though, given that the young population will cause the intrinsic stellar colors to appear bluer, rather than redder.

     The G1 clump does not appear particularly distinct on the metallicity map, contrary to its appearance on the spatial density map, with the RGB stars in this region appearing just slightly more metal-poor than the background halo.

     Many previous studies have addressed the stellar population characteristics of the halo and outer disk of M31; however, these have almost exclusively sampled only a few discrete locations or have taken a panoramic but much shallower view. Early quantitative wide-area survey work of M31 was based on photographic plates (e.g., Hodge 1973; Innanen et al. 1982; WK88). Although these studies clearly revealed, among other things, the antisymmetric warping of the outer disk and the tidally induced twisting of the outer isophotes of the dwarf companions (see also Choi, Guhathakurta, & Johnston 2002), they did not go deep enough to directly resolve the stellar populations.

     More recent work has focused on detailed studies of resolved stars in several (generally small) fields sampling the outer disk and halo (see, for example, Mould & Kristian 1986; Morris et al. 1994; Rich et al. 1996; Holland et al. 1996; Richer 1996; Reitzel et al. 1998; Durrell et al. 2001; Ferguson & Johnson 2001; Sarajedini & Van Duyne 2001; Reitzel & Guhathakurta 2002). Both ground-based and HST studies have found a dominant population, which is significantly more metal-rich than that of the Milky Way halo and which is characterized by a considerable intrinsic dispersion. Specifically, measurements indicate [Fe/H] -0.7, comparable to 47 Tuc, with a spread of nearly 2 dex for M31, compared with [Fe/H] -1.5 for the Milky Way. Detailed metallicity distributions have now been calculated for stars in several fields, all of which are found to exhibit a similar shape, consisting of a peak at the metal-rich end and an extended tail toward lower metallicities (Holland et al. 1996; Durrell et al. 2001; Sarajedini & Van Duyne 2001; Reitzel & Guhathakurta 2002). The fraction of stars in the metal-poor component is estimated to be 25%50%, with no compelling evidence for a strong metallicity gradient and/or inhomogeneities in the halo. Additionally, Reitzel et al. (1998) find that the halo of M31 is roughly an order of magnitude larger and/or denser than the Galactic halo, though this conclusion is drawn from star counts in only three fields.

     Analysis of individual color-magnitude diagrams from our WFC survey indicates a mean metallicity in broad agreement with the findings of previous studies (see Fig. 4; also Ibata et al. 2001a) and reveals no obvious metallicity gradient in the halo out to at least 50 kpc. We do, however, find evidence for significant spatial and metallicity substructure in the halo. Figure 6 shows the location of published HST fields and a few recent ground-based fields overlaid on our survey area. It can be seen that these fields generally either sample the inner halo of M31 or sample at large radius along the southern minor axis, neither of which exhibits significant substructure in our survey. The chemical and spatial homogeneity of the halo that has been inferred from these studies is therefore not surprising. Most of the substructure our survey has uncovered lies in the previously uncharted outer regions of M31.

FIG. 6.Locations of some recent field stellar population studies in the outskirts of M31, including fields (squares) targeted by the ground-based studies of Morris et al. (1994), Reitzel, Guhathakurta, & Gould (1998), and Durrell, Harris, & Pritchet (2001), the WFPC2 studies (circles) of Holland, Fahlman, & Richter (1996), Rich et al. (1996), Sarajedini & Van Duyne (2001), Ferguson, Gallagher, & Wyse (2000), and Ferguson & Johnson (2001). Only the Morris et al. spur field and the Rich et al. G1 field lie close to halo substructures discovered in the present survey.

     An exception to this is the study by Morris et al. (1994), which resolved the stars in a small section of the northern spur and provided the first evidence that the stellar overdensity in these parts consisted of the fairly characteristic metal-rich population seen elsewhere in the outskirts of M31. Like us, these authors also found evidence for a significant AGB component lying above the RGB tip. In addition, Rich et al. (1996) obtained an HST Wide Field Planetary Camera 2 (WFPC2) CMD of the field population around the globular cluster G1, corresponding to a location near the very edge of the stellar overdensity discovered in our WFC survey, and found the RGB to be well matched by the 47 Tuc fiducial. They also noted that the luminosity function of the field population in this region could not be fitted by template globular cluster luminosity functions. While the results from these studies are in good qualitative agreement with those presented here, their lack of depth and/or areal coverage adds little further insight into the nature of the stellar populations associated with the substructure.

     The most extensive published wide-field survey of H I in M31 is still that of Newton & Emerson (1977), who mapped regions along the major axes out to 35 kpc. Apart from the warp in the disk at large radius, the outermost H I contour looks remarkably smooth and uniform, with no sign of perturbation in the direction of any of the halo substructure (see their Fig. 12a). Furthermore, while the optical and H I disks are warped in the same direction, the optical warp appears to begin at smaller radii than that of the H I and to exhibit a greater deviation from the plane (see also Innanen et al. 1982; WK88). This holds true along both major axes but seems particularly apparent in the northeast.

     An interesting high-velocity H I cloud lying near the northern minor axis of M31 was discovered by Davies (1975). The feature lies approximately 15 (20 kpc in projection) northwest of the center of M31 at (-09, +11) and has an extent of 1° and a heliocentric velocity of -450 km s^-1 (see Figs. 8 and 9). It lies just above NGC 205 and, rather curiously, along the projected extension of the giant stellar stream discovered in the southern half of M31. Based on the velocity of the cloud, it would appear unlikely that it has any connection to the northwestern disk, nor for that matter to either M32 (V_hel = -197 km s^-1) or NGC 205 (V_hel = -242 km s^-1). The cloud has a peak column density of 4 × 10¹⁹ atoms cm^-2, a gaseous velocity dispersion of 9 km s^-1 and, if at the distance of M31, a gas mass of 5 × 10⁶ M; it could thus represent a fairly massive compact high-velocity cloud. Our survey images reach several magnitudes fainter than the Palomar Sky Survey plates. Inspection of CMDs and detailed spiral density and metallicity maps centered on the cloud position reveals nothing unusual at this location. An alternative explanation for the cloud is that it is very local and related to the Magellanic Stream (Davies 1975); the nondetection of an associated stellar counterpart to the cloud in our deep images may be more consistent with this interpretation.

     Newton & Emerson (1977) also discovered an apparently detached H I cloud at large radius along the northeastern major axis of M31. Located at a projected distance of 3° or 34 kpc, this cloud has a velocity that differs by 100 km s^-1 from that expected for local disk rotation. The global properties of the cloud are similar to those inferred for the Davies cloud. The cloud lies at (+15, +2°) from the center of M31 and near our current survey limit in this quadrant; our maps reveal no obvious spatial substructure here, but the mean metallicity of the RGB stars is somewhat above average (see Fig. 5). The H I feature may be a signature of a major disk disturbance in the northeast quadrant, possibly related to the northern spur feature, which is located roughly 05 farther in.

     Substructure in the form of accreted and/or disrupted satellites is a generic prediction of hierarchical galaxy assembly. Even if the satellites are accreted before significant star formation has occurred, they will still interact gravitationally with the stellar disk of the host, causing tidal heating, distortion, and possibly even warping. Our WFC survey has led to the discovery of significant spatial and chemical substructure in the outskirts of M31, the most prominent of which is characterized in Figure 7. We first address whether any of this substructure is directly associated with the luminous satellite companions of M31.

FIG. 7.Cartoon illustrating the most prominent spatial and chemical substructure discovered in our WFC survey. A possible projected orbit of the giant stellar stream is indicated; the stream may connect to the enhancement seen in the spatial and chemical maps of the northern half of the disk. The contours delineate the approximate extent and orientation of the northern spur and G1 clump; the location of the G1 globular cluster is also indicated.

     M31 is known to possess approximately 15 satellites, of which the dwarf ellipticals (dE's) M32 and NGC 205 are the most proximate, lying at projected distances of 5 and 9 kpc, respectively, and among the most luminous (Mateo 1998). Figure 8 illustrates where the closest satellite companions lie in projection with respect to M31. Apart from the clump of stars close to the luminous cluster G1 and the northern spur, the nature of which is not yet clear, our survey has failed to reveal any obvious (i.e., still intact) additional satellites out to 3° from the center of M31. For reference, all known Milky Way satellites (including the Sagittarius dwarf) have surface brightnesses and angular extents that would have rendered them easily detectable in our survey, if lying within a projected radius of 55 kpc of M31.

FIG. 8.Projected distribution of the closest satellite companions around M31. Also shown is the nearest massive companion to M31, M33, and the position of the neutral hydrogen cloud detected by Davies (1975). As previously, the outer ellipse denotes the approximate limit of our current WFC survey.

     M32 and NGC 205 are known to exhibit anomalous surface brightness profiles and/or outer isophote twists suggestive of tidal interaction and disruption (e.g., Hodge 1973; Choi et al. 2002; see also Fig. 9). In addition, both display a variety of puzzling properties for galaxies of the dE class. M32 exhibits a strong intermediate-age stellar component, in addition to a classical old stellar component, and shows evidence for a large metallicity spread with a mean just below solar (Grillmair et al. 1996; Davidge et al. 2000). The signatures of an extended and complex star formation epoch are even more obvious in NGC 205, which possesses cold gas (both atomic and molecular) and dust, as well as young stars (Lee 1996; Young & Lo 1997). Furthermore, the H I shows a well-defined velocity gradient (unlike the stars) and is confined to the very inner regions of the optical galaxy; this inconsistency between gas and stars suggests that the gas may have been recently captured, perhaps because of a passage through the disk of M31. The geometrical alignment of the two satellites with the southern stream and the broadly similar metal abundances initially suggested that either or both of them could be responsible for the tidal feature (Ibata et al. 2001a). With a significant fraction of the northern half of the halo of M31 now mapped, we have failed to identify any obvious continuation of the stream on the northern side of NGC 205. Given that tidal interactions give rise to both leading and trailing streams from the disrupted satellite, such a feature might be expected if NGC 205 were the origin of the stream. M32 and a third system (either more distant, or now completely cannibalized?) now appear as more promising candidates for the origin of the stream. We have previously remarked on the possible connection between the stream and the similarly metal-rich, but lower level, stellar overdensity near the northeastern side of the disk (see Fig. 7). If this association is correct, it implies the projected orbit of the disrupting satellite wraps tightly around the M31 nucleus and does not cross into the northern half of the halo of M31. On the other hand, the lack of a northern counterpart to the stream might simply reflect the fact that the current orbit of the satellite has yet to traverse into this part of the halo or else that the stream in this region has already dispersed and is no longer visible as a spatial or chemical density enhancement.

FIG. 9.Isopleth maps of M31 from an APM scan of a 75 minute exposure Palomar Schmidt IIIaJ plate taken by Sydney van den Bergh in 1970. The lowest isophote contoured is at a level equivalent to B = 27 mag arcsec-1. Right: Distorted outer isophotes of M32 and NGC 205 after subtracting an elliptically averaged M31 profile. Cepa & Beckman (1988) quote tidal radii for M32 and NGC 205 of 0.84 kpc (006) and 1.93 kpc (014), respectively, which correspond to the relatively unperturbed inner parts of the profiles. Left: Location of the Davies (1975) H I cloud. In the southern half of the galaxy, the M31 outer isophotes are distorted in the direction of the stream. Although the lowest contour level is at the limit of reliability of the photographic data, this particular feature is present in other digitized wide-area photographic data (e.g., WK88).

     Other satellites of possible relevance for the interpretation of the M31 substructure are the dwarf spheroidal systems Andromeda I and Andromeda III. Both lie just beyond the boundary of our current survey to the southwest of M31, consistent with the general direction of the stream. The stellar populations in these systems are predominantly metal-poor, however, a fact that would seem to argue against them being the origin of the more metal-rich stream. Specifically, And I and And III have mean RGB metallicities of [Fe/H] -1.5 and -1.9, respectively (Da Costa et al. 1996; Da Costa, Armandroff, & Caldwell 2002), compared with the value of [Fe/H] -0.7 inferred for the stream. M33, the largest satellite of M31, lies at a projected distance of 15° toward the southeast. There is no evidence at present to connect M33 with any of the substructure seen around M31; however, the potential role this system could play within the M31 subsystem should be borne in mind.

     The extreme compactness of M32, coupled with the very high central surface brightness and the dwarflike luminosity, has led many authors to suggest that M32 is the leftover core of a larger galaxy, evolved to its current state as a result of prolonged tidal harassment within the halo of M31 (e.g., Faber 1973; Bekki et al. 2001; Graham 2002). Such a hypothesis may also explain the puzzling observation that M32 has no globular clusters; a system this luminous would be expected to possess at least 1520 of them (Harris 1991). The broad agreement between the stellar metallicity and dispersion in M32 and in the M31 halo (Grillmair et al. 1996; Durrell et al. 2001) leads one to speculate whether the former could be the origin of not only the stellar stream, but also much of the field halo. If the M31 halo was significantly polluted by stars from a fairly massive companion, this could also account for the higher density or size of the stellar halo (Reitzel et al. 1998) compared with that of the Milky Way. For such a scenario to be viable, M32 must have already made several revolutions around M31 to account for the high covering factor of the metal-rich population. Dynamical friction due to stars and dark matter in the halo of M31 will cause the orbit of M32 to spiral inward with time, while any nonsphericity in the potential and/or passages close to the disk will cause the orbit to precess. The significant velocity dispersion of M32 stars (van der Marel et al. 1994) implies that once pulled off, these stars will merge rather quickly into the halo population, and only the most recently stripped stars will still appear as coherent structures. Penarrubia, Kroupa, & Boily (2002) have shown that tidal debris from a disrupting satellite will spread out even faster if the host halo is significantly flattened, with a timescale of 34 Gyr before total disruption of the companion. Detailed dynamical modeling of satellite orbits within the potential of M31 will be presented in a future paper.

     Few constraints exist on the origin of the major-axis substructure at present. It would appear unlikely that the G1 clump and the northern spur are related to each other (for example, resulting from an object orbiting close to the plane of the disk) because of the different mean colors of the RGB stars in these regions and the presence of a prominent intermediate-age component in the spur but not in the clump. The G1 clump is particularly intriguing given its proximity to the anomalous globular cluster G1. G1 is among the most luminous and most massive globular clusters in M31 and is rather unique in having an intrinsic metallicity spread suggestive of self-enrichment (Meylan et al. 2001). The outer isophotes of the cluster are distinctly elongated, and the direction of this elongation appears to be in the same sense as that of the G1 clump. Meylan et al. have argued that G1 might be the core of a disrupted dE and such a scenario is supported by the fact that the metallicity reported for G1 ([Fe/H] = -0.95) agrees quite well with that inferred for the stellar overdensity, i.e., slightly below that of the average halo [Fe/H] -0.7. On the other hand, the significant spatial offset (05 or 7 kpc if at the distance of M31) seen between G1 and the peak of the stellar overdensity is difficult to understand in this picture, as is the fact that the luminosity and mean metallicity of the clump place it well off the luminosity-metallicity relation defined by Local Group dE and dSph galaxies (e.g., Mateo 1998). Given the metallicity of the G1 clump as determined from mean RGB color, the luminosity-metallicity relation predicts a luminosity more than 2 mag brighter than that observed. An alternative explanation for the feature is that it is a highly warped section of the far outer disk, perhaps perturbed or torn off by a previous tidal interaction. The clump lies just beyond the outermost H I contour of Newton & Emerson (1977), which warps southward in this quadrant, compared with the northern offset displayed by the clump. If the disk interpretation is correct, the fact that the gas and stars currently exhibit different behaviors here could imply the interaction happened a long time ago. Future photometric and kinematic observations will aid significantly in discriminating between these two possibilities.

     The location of the northern spur near the northeastern major axis of M31 and in the same direction of the gaseous warp provides strong support for the association of this feature with a severe warp in the outer stellar disk. This interpretation would naturally explain the metal-rich nature of the population in the spur, as well as the presence of a youngtointermediate-age component, as these are properties that are also known to characterize stars in the outer parts of the disk (Morris et al. 1994; Ferguson & Johnson 2001). WK88 noted that if the northern spur was an extension of the disk, it would be the most extreme example of a warped stellar disk ever found. It would also be highly asymmetric. Tidal forces due to an infalling satellite could possibly excite such a strong warp (Binney 1992; Huang & Carlberg 1997); however, the identity and location of the putative perturber is not obvious at present.

     Our INT WFC survey currently maps an area of 25 deg² around M31, extending to a semimajor axis of 55 kpc, and allows the first uninterrupted study of the density and color distribution of individual red giant stars across a large fraction of an external disk galaxy. We have found evidence for both spatial and metallicity (as inferred from color information) substructure, which are often, but not always, correlated. In addition to the giant stellar stream reported by Ibata et al. (2001a), the data reveal the presence of significant stellar overdensities at large radii close to the southwestern major axis, in the proximity of the very luminous globular cluster G1, and near the northeastern major axis, coinciding with and extending beyond the northern spur. The most prominent metallicity variations are found in the southern half of the halo, where two structures with above-average metallicities are apparent. One of these coincides with the aforementioned giant stellar stream, whereas the other corresponds to a much lower level stellar overdensity.

     Our results contrast with, but do not conflict with, the findings of previous M31 stellar population studies that have found a rather homogeneous population in the halo and outer disk. These studies have mostly focused on inner halo fields or fields at large radius along the southern minor axis, neither of which are found to exhibit significant substructure within the limits of our current survey. Of the newly detected substructure, only the northern spur appears to have a counterpart in published H I maps (Newton & Emerson 1977). Further, our data reveal no spatial or chemical overdensity coincident with the high-velocity cloud discovered by Davies (1975), which lies just north of the satellite companion NGC 205, along the projected northern extension of the giant stellar stream.

     We have argued that the stellar substructure seen in the outskirts of M31 is likely due to a variety of past and ongoing interaction and accretion events. The lack of any obvious northern counterpart to the southern stellar stream suggests that either the disrupting satellite has yet to traverse into the northern half of the halo or that the projected orbit wraps tightly around the center of M31. Indeed, there is tentative evidence to connect the stellar stream to the stellar and chemical enhancement seen toward the north end of the disk. Several pieces of evidence suggest that M32 could be the origin of the stream; these include the geometrical alignment of the two objects, the similarity in the mean metallicity and dispersion of their constituent stars and the wide range of peculiar properties exhibited by M32, all of which are consistent with it having once been a considerably more massive and luminous galaxy. One may even speculate that stars stripped from M32 have polluted a large fraction of the M31 halo, providing a viable explanation for the long-standing puzzle of why the field halo of M31 is more metal-rich and denser/larger than that of the Milky Way. Alternatively, the stream may be due to another satellite, perhaps now completely cannabilized or else lying beyond our current survey limits.

     The stellar substructure located close to the major axesthe northern spur and the G1 clumpcould plausibly result from tidal distortion and disruption of the outer disk due to, for example, the close passage of a massive satellite. If this explanation is correct, the interaction responsible for producing the G1 clump must have occurred some time ago, given the lack of evidence for a strong intermediate-age component and the slightly lower than average metallicity in this structure. On the other hand, the remarkable proximity and overall alignment between the clump and the very luminous globular cluster, G1, supports the alternative hypothesis that the two have a common origin. Meylan et al. (2001) have proposed that G1 may be the core of a stripped dwarf elliptical galaxy; this hypothesis clearly warrants further consideration. However, a satisfactory model must also account for the 05 projected offset seen between the peak of the stellar overdensity and the globular cluster.

     The Milky Way and M31 are the only two massive galaxies for which we have detailed information about how the properties of individual stars vary across a large fraction of the halo. The mean metallicity and size/density of the stellar halos in these systems differ substantially, and this has sometimes been cited as evidence for different formation mechanisms (e.g., Durrell et al. 2001). Our INT WFC survey reveals that, like the Milky Way, the stellar populations at large radii in M31 exhibit significant substructure. This substructure is expected in hierarchical CDM models of galaxy formation, either as a direct result of disrupting satellites or as a by-product of tidal interactions between infalling satellites and the fragile outer disk. The fact that luminous and clearly distorted satellites lie close to the center of M31 suggests that the differences in stellar halo between that galaxy and the Milky Way may be the result of a more active accretion history. Essential to the interpretation of our newly detected substructure will be deep photometric and spectroscopic observations of stars on the outskirts of M31. An approved Cycle 11 HST program with the newly installed Advanced Camera for Surveys is aimed at obtaining deep color-magnitude diagrams, reaching several magnitudes below the horizontal branch, in several fields where we have detected halo substructure and in two locations in the far outer disk. These data will provide more quantitative information about the ages and metallicities of stars in the various substructures and will help to constrain the accretion or merger history of the galaxy. Spectroscopic observations with 8 m telescopes will be essential for providing kinematical constraints to be used in modeling the satellite interactions and for probing the shape and potential of the massive dark halo of M31.

     Finally, it is worth noting that M31 is being used as a target in several large experiments that search for microlensing events due to massive compact halo objects. Without further detailed modeling, it is impossible to determine where the giant stellar stream lies with respect to the center of M31. If it lies in front of the galaxy, it will produce a population of lenses that are more probable than the average M31 bulge star to lens M31 bulge stars, whereas if it lies behind the galaxy, it will produce an extra population of sources that are more likely to be lensed by M31 bulge stars. A careful appraisal of the stream, as well as other M31 stellar substructure, will therefore be required to interpret the microlensing event rates toward M31.

     We thank the Institute of Astronomy, Cambridge, and the Anglo-Australian Observatory for hospitality during various collaborative visits.

• Baugh, C. M., Cole, S., & Frenk, C. S. 1996, MNRAS, 283, 1361 First citation in article | Crossref | ADS
• Bekki, K., Couch, W. J., Drinkwater, M. J., & Gregg, M. D. 2001, ApJ, 557, L39 First citation in article | IOPscience | ADS
• Binney, J. 1992, ARA&A, 30, 51 First citation in article | Crossref | ADS
• Bullock, J. S., Kravtsov, A. V., & Weinberg, D. H. 2000, ApJ, 539, 517 First citation in article | IOPscience | ADS
• . 2001, ApJ, 548, 33 First citation in article | IOPscience | ADS
• Cepa, J., & Beckman, J. E. 1988, A&A, 200, 21 First citation in article | ADS
• Choi, P. I., Guhathakurta, P., & Johnston, K. V. 2002, AJ, 124, 310 First citation in article | IOPscience | ADS
• Couch, W. J., Jurcevic, J. S., & Boyle, B. J. 1993, MNRAS, 260, 241 First citation in article | Crossref | ADS
• Da Costa, G. S., & Armandroff, T. E. 1990, AJ, 100, 162 First citation in article | Crossref | ADS
• Da Costa, G. S., Armandroff, T. E., & Caldwell, N. 2002, AJ, 124, 332 First citation in article | IOPscience | ADS
• Da Costa, G. S., Armandroff, T. E., Caldwell, N., & Seitzer, P. 1996, AJ, 112, 2576 First citation in article | Crossref | ADS
• Davidge, T. J., Rigaut, F., Chun, M., Brandner, W., Potter, D., Northcott, M., & Graves, J. E. 2000, ApJ, 545, L89 First citation in article | IOPscience | ADS
• Davies, R. D. 1975, MNRAS, 170, P45 First citation in article | Crossref | ADS
• Dekel, A., & Silk, J. 1986, ApJ, 303, 39 First citation in article | Crossref | ADS
• Durrell, P. R., Harris, W. E., & Pritchet, C. J. 2001, AJ, 121, 2557 First citation in article | IOPscience | ADS
• Faber, S. M. 1973, ApJ, 179, 423 First citation in article | Crossref | ADS
• Ferguson, A. M. N., Gallagher, J. S., & Wyse, R. F. G. 2000, AJ, 120, 821 First citation in article | IOPscience | ADS
• Ferguson, A. M. N., & Johnson, R. A. 2001, ApJ, 559, L13 First citation in article | IOPscience | ADS
• Font, A. S., Navarro, J. F., Stadel, J., & Quinn, T. 2001, ApJ, 563, L1 First citation in article | IOPscience | ADS
• Girardi, L., Bressan, A., Bertelli, G., & Chiosi, C. 2000, A&AS, 141, 371 First citation in article | Crossref | ADS
• Graham, A. W. 2002, ApJ, 568, L13 First citation in article | IOPscience | ADS
• Grillmair, C. J., Freeman, K. C., Irwin, M., & Quinn, P. J. 1995, AJ, 109, 2553 First citation in article | Crossref | ADS
• Grillmair, C. J., et al. 1996, AJ, 112, 1975 First citation in article | Crossref | ADS
• Harris, W. E. 1991, ARA&A, 29, 543 First citation in article | Crossref | ADS
• Hayashi, E., Navarro, J. F., Taylor, J. E., Stadel, J., & Quinn, T. 2002, ApJ, submitted (astro-ph/0203004) First citation in article | Preprint
• Helmi, A., & White, S. D. M. 1999, MNRAS, 307, 495 First citation in article | Crossref | ADS
• Helmi, A., White, S. D. M., de Zeeuw, P. T., & Zhao, H. 1999, Nature, 402, 53 First citation in article | Crossref | ADS
• Hodge, P. W. 1973, ApJ, 182, 671 First citation in article | Crossref | ADS
• Holland, S. 1998, AJ, 115, 1916 First citation in article | IOPscience | ADS
• Holland, S., Fahlman, G. G., & Richer, H. B. 1996, AJ, 112, 1035 First citation in article | Crossref | ADS
• Huang, S., & Carlberg, R. G. 1997, ApJ, 480, 503 First citation in article | IOPscience | ADS
• Ibata, R., Irwin, M., Lewis, G., Ferguson, A. M. N., & Tanvir, N. 2001a, Nature, 412, 49 First citation in article | Crossref | ADS | PubMed
• Ibata, R., Irwin, M., Lewis, G. F., & Stolte, A. 2001b, ApJ, 547, L133 First citation in article | IOPscience | ADS
• Ibata, R., Lewis, G. F., Irwin, M., Totten, E., & Quinn, T. 2001c, ApJ, 551, 294 First citation in article | IOPscience | ADS
• Ibata, R. A., Gilmore, G., & Irwin, M. J. 1995, MNRAS, 277, 781 First citation in article | Crossref | ADS
• Innanen, K. A., Kamper, K. W., van den Bergh, S., & Papp, K. A. 1982, ApJ, 254, 515 First citation in article | Crossref | ADS
• Irwin, M., & Hatzidimitriou, D. 1995, MNRAS, 277, 1354 First citation in article | Crossref | ADS
• Irwin, M., & Lewis, J. 2001, NewA Rev., 45, 105 First citation in article | Crossref | ADS
• Johnston, K. V., Hernquist, L., & Bolte, M. 1996, ApJ, 465, 278 First citation in article | Crossref | ADS
• Klypin, A., Kravtsov, A. V., Valenzuela, O., & Prada, F. 1999, ApJ, 522, 82 First citation in article | IOPscience | ADS
• Lee, M. G. 1996, AJ, 112, 1438 First citation in article | Crossref | ADS
• Leon, S., Meylan, G., & Combes, F. 2000, A&A, 359, 907 First citation in article | ADS
• Majewski, S. R., Ostheimer, J. C., Patterson, R. J., Kunkel, W. E., Johnston, K. V., & Geisler, D. 2000, AJ, 119, 760 First citation in article | IOPscience | ADS
• Majewski, S. R., Siegel, M. H., Kunkel, W. E., Reid, I. N., Johnston, K. V., Thompson, I. B., Landolt, A. U., & Palma, C. 1999, AJ, 118, 1709 First citation in article | IOPscience | ADS
• Mateo, M. L. 1998, ARA&A, 36, 435 First citation in article | Crossref | ADS
• Mateo, M., Olszewski, E. W., & Morrison, H. L. 1998, ApJ, 508, L55 First citation in article | IOPscience | ADS
• Meylan, G., Sarajedini, A., Jablonka, P., Djorgovski, S. G., Bridges, T., & Rich, R. M. 2001, AJ, 122, 830 First citation in article | IOPscience | ADS
• Moore, B., Ghigna, S., Governato, F., Lake, G., Quinn, T., Stadel, J., & Tozzi, P. 1999, ApJ, 524, L19 First citation in article | IOPscience | ADS
• Morris, P. W., Reid, I. N., Griffiths, W. K., & Penny, A. J. 1994, MNRAS, 271, 852 First citation in article | Crossref | ADS
• Mould, J., & Kristian, J. 1986, ApJ, 305, 591 First citation in article | Crossref | ADS
• Newberg, H. J., et al. 2002, ApJ, 569, 245 First citation in article | IOPscience | ADS
• Newton, K., & Emerson, D. T. 1977, MNRAS, 181, 573 First citation in article | Crossref | ADS
• Odenkirchen, M., et al. 2001, ApJ, 548, L165 First citation in article | IOPscience | ADS
• Penarrubia, J., Kroupa, P., & Boily, C. M. 2002, MNRAS, 338, 779 First citation in article | Crossref
• Reitzel, D. B., & Guhathakurta, P. 2002, AJ, 124, 234 First citation in article | IOPscience | ADS
• Reitzel, D. B., Guhathakurta, P., & Gould, A. 1998, AJ, 116, 707 First citation in article | IOPscience | ADS
• Rich, R. M., Mighell, K. J., Freedman, W. L., & Neill, J. D. 1996, AJ, 111, 768 First citation in article | Crossref | ADS
• Sagar, R., Subramaniam, A., Richtler, T., & Grebel, E. K. 1999, A&AS, 135, 391 First citation in article | Crossref | ADS
• Sarajedini, A., & Van Duyne, J. 2001, AJ, 122, 2444 First citation in article | IOPscience | ADS
• Schlegel, D. J., Finkbeiner, D. P., & Davis, M. 1998, ApJ, 500, 525 First citation in article | IOPscience | ADS
• Shang, Z., et al. 1998, ApJ, 504, L23 First citation in article | IOPscience | ADS
• Smail, I., Hogg, D. W., Yan, L., & Cohen, J. G. 1995, ApJ, 449, L105 First citation in article | IOPscience | ADS
• Steinmetz, M., & Navarro, J. F. 2002, NewA, 7, 155 First citation in article | Crossref | ADS
• van der Marel, R. P., Rix, H. W., Carter, D., Franx, M., White, S. D. M., & de Zeeuw, T. 1994, MNRAS, 268, 521 First citation in article | Crossref | ADS
• Walterbos, R. A. M., & Kennicutt, R. C. 1988, A&A, 198, 61 First citation in article | ADS
• Walton, N. A., Lennon, D. J., Greimel, R., Irwin, M. J., Lewis, J. R., & Rixon, G. T. 2001, ING Newsl., No. 4, 7 First citation in article
• White, S. D. M., & Rees, M. J. 1978, MNRAS, 183, 341 First citation in article | Crossref | ADS
• White, S. D. M., & Springel, V. 2000, in Proc. of the Second MPA/ESO Workshop, The First Stars, ed. A. Weiss, T. G. Abel, & V. Hill (New York: Springer), 327 First citation in article
• Yanny, B., et al. 2000, ApJ, 540, 825 First citation in article | IOPscience | ADS
• Young, L. M., & Lo, K. Y. 1997, ApJ, 476, 127 First citation in article | IOPscience | ADS