Chemical Enrichment in the Faintest Galaxies: the Carbon and Iron Abundance Spreads in the Bo\"otes I Dwarf Spheroidal Galaxy and the Segue 1 System
ABSTRACT We present an AAOmega spectroscopic study of red giant stars in Bootes I, which is an ultra-faint dwarf galaxy, and Segue 1, suggested to be either an extremely low-luminosity dwarf galaxy or a star cluster. Our focus is quantifying the mean abundance and abundance dispersion in iron and carbon, and searching for distant radial-velocity members, in these systems. The primary conclusion of our investigation is that the spread of carbon abundance in both Bootes I and Segue 1 is large. For Bootes I, 4 of our 16 velocity members have [C/H] < ~-3.1, while 2 have [C/H] > ~-2.3, suggesting a range of Delta[C/H] ~ 0.8. For Segue 1 there exists a range Delta[C/H] ~ 1.0, including our discovery of a star with [Fe/H] = -3.5 and [C/Fe] = +2.3, which is a radial velocity member at a distance of 4 half-light radii from the system center. The accompanying ranges in iron abundance are Delta[Fe/H] ~ 1.6 for both Bootes I and Segue 1. For [Fe/H] < -3.0, the Galaxy's dwarf galaxy satellites exhibit a dependence of [C/Fe] on [Fe/H] which is very similar to that observed in its halo populations. We find [C/Fe] ~ 0.3 for stars in the dwarf systems that we believe are the counterpart of the Spite et al. (2005) ``unmixed'' giants of the Galactic halo and for which they report [C/Fe] ~ 0.2, and which presumably represents the natal relative abundance of carbon for material with [Fe/H] = -3.0 to -4.0. We confirm the correlation between luminosity and both mean metallicity and abundance dispersion in the Galaxy's dwarf satellites, which extends to at least as faint as Mv = -5. The very low mean metallicity of Segue 1, and the high carbon dispersion in Bootes I, consistent with inhomogeneous chemical evolution in near zero-abundance gas, suggest these ultra-faint systems could be surviving examples of the very first bound systems. Comment: 62 pages, 18 figures, submitted to Astrophysical Journal 9 June 2010
arXiv:1008.0137v1 [astro-ph.GA] 1 Aug 2010
CHEMICAL ENRICHMENT IN THE FAINTEST GALAXIES:
THE CARBON AND IRON ABUNDANCE SPREADS IN THE
BO¨OTES I DWARF SPHEROIDAL GALAXY AND THE
SEGUE 1 SYSTEM
JOHN E. NORRIS1, ROSEMARY F.G. WYSE2,6, GERARD GILMORE3, DAVID
YONG1, ANNA FREBEL4, MARK I. WILKINSON5, V. BELOKUROV3, AND DANIEL
We present an AAOmega spectroscopic study of red giant stars in Bo¨ otes I,
which is an ultra-faint dwarf galaxy, and Segue 1, suggested to be either an
extremely low-luminosity dwarf galaxy or a star cluster. Our focus is quantifying
the mean abundance and abundance dispersion in iron and carbon, and searching
for distant radial-velocity members, in these systems.
The primary conclusion of our investigation is that the spread of carbon
abundance in both Bo¨ otes I and Segue 1 is large. For Bo¨ otes I, four of our 16
velocity members have [C/H] ? –3.1, while two have [C/H] ? –2.3, suggesting a
range of ∆[C/H] ∼ 0.8. For Segue 1 there exists a range ∆[C/H] ∼ 1.0, including
our discovery of a star with [Fe/H] = –3.5 and [C/Fe] = +2.3, which is a radial
velocity member at a distance of 4 half-light radii from the system center. The
accompanying ranges in iron abundance are ∆[Fe/H] ∼ 1.6 for both Bo¨ otes I
1Research School of Astronomy & Astrophysics, The Australian National University, Mount Stromlo
Observatory, Cotter Road, Weston, ACT 2611, Australia; email: email@example.com
2The Johns Hopkins University, Department of Physics & Astronomy, 3900 N. Charles Street, Baltimore,
MD 21218, USA
3Institute of Astronomy, University of Cambridge, Madingley Road, Cambridge CB3 0HA, UK
4Harvard-Smithsonian Center for Astrophysics, Cambridge, MA 02138, USA
5Department of Physics and Astronomy, University of Leicester, University Road, Leicester, LE1 7RH,
6Institute for Astronomy, University of Edinburgh, Royal Observatory, Blackford Hill, Edinburgh EH9
7Department of Physics, Macquarie University, North Ryde, NSW 2109, Australia
8Anglo-Australian Observatory, PO Box 296, Epping, NSW 1710, Australia
– 2 –
and Segue 1. For [Fe/H] < –3.0, the Galaxy’s dwarf galaxy satellites exhibit a
dependence of [C/Fe] on [Fe/H] which is very similar to that observed in its halo
populations. We find [C/Fe] ∼ 0.3 for stars in the dwarf systems that we believe
are the counterpart of the Spite et al. (2005) “unmixed” giants of the Galactic
halo and for which they report [C/Fe] ∼ 0.2, and which presumably represents
the natal relative abundance of carbon for material with [Fe/H] = –3.0 to –4.0.
Our second conclusion is confirmation of the correlation between (decreasing)
luminosity and both (decreasing) mean metallicity and (increasing) abundance
dispersion in the Galaxy’s dwarf satellites. This correlation extends to at least
as faint as MV = –5, and may continue to even lower luminosities. The very
low mean metallicity of Segue 1, and the high carbon dispersion in Bo¨ otes I,
consistent with inhomogeneous chemical evolution in near zero-abundance gas,
suggest these ultra-faint systems could be surviving examples of the very first
Subject headings: Galaxy: abundances − galaxies: dwarf − galaxies: individual
(Bo¨ otes I, Segue 1) − galaxies: abundances − stars: abundances
The dwarf spheroidal (dSph) galaxies associated with the Milky Way provide important
potential insight into the ΛCDM paradigm, the manner in which the baryonic material in
low luminosity systems is chemically enriched, and the formation of the halo populations
of the Galaxy (see Klypin et al. 1999; Moore et al. 1999; Tolstoy, Hill, & Tosi 2009, and
references therein). Recent studies of the newly discovered ultra-faint, high mass-to-light
ratio, dwarf systems (e.g. Belokurov et al. 2006, 2007) place intriguing constraints not only
on the minimal baryonic masses with which a galaxy can form, but also on the dark matter
they contain and the interplay between dark and luminous material in the production of the
chemical elements at the earliest times. A crucial aspect of the ongoing discussion is the
nature of the lowest luminosity ultra-faint systems. For example, while Geha et al. (2009)
identify Segue 1 as an ultra-faint dwarf galaxy, Niederste-Ostholt et al. (2009) have suggested
instead “it is a star cluster, originally from the Sagittarius galaxy”.
The chemistry of the ultra-faint systems is providing critical constraints on their masses
and their evolutionary histories, particularly by focusing on the most metal-poor stars. Kirby
– 3 –
et al. (2008) first reported stars with [Fe/H]1as low as –3.3, together with large abundance
spreads, in several ultra-faint dwarfs. Norris et al. (2008), in a study of Bo¨ otes I, found a
similar result, with their most metal-poor star having [Fe/H] = –3.4. Kirby et al. (2008)
also established that the mean metallicity of the dSphs continues to decrease with declining
luminosity, down to the faint limit of the ultra-faint systems.
Frebel et al. (2010a) obtained the first relative abundances of extremely metal-poor stars
in the ultra-faint systems, reporting not only a range in Fe within each galaxy, but also in
carbon, where two of three stars in UMa II were found to have [C/Fe] = 0.5 and 0.8 at [Fe/H]
= –3.2. Their work has been followed by further high-resolution, moderate S/N analyses
of additional giants with [Fe/H] < –3.0 in the dwarf galaxies (Segue 1: Geha et al. 2009;
Draco: Cohen & Huang 2009; Bo¨ otes I: Feltzing et al. 2009, Norris et al. 2010; Sculptor:
Frebel, Kirby, & Simon 2010b). Perhaps the most interesting result of these recent studies
is that at [Fe/H] ∼ –3.7, the relative abundances of a large number of elements are quite
similar to those found in the majority of Galactic halo giants (Frebel, Kirby, & Simon 2010b
(8 elements); Norris et al. 2010 (15 elements)). A second result is the report of surprisingly
large values for the ratio of abundances of two α-elements, specifically [Mg/Ca], for one star
in Draco (Fulbright et al. 2004), two stars in Hercules (Koch et al. 2008), and one star in
Bo¨ otes I (Feltzing et al. (2009), most simply interpreted in terms of inhomogeneous mixing
of supernova ejecta.
The present paper reports further results on the abundance ranges in the ultra-faint
dwarf Bo¨ otes I (MV,total∼ –6.3; Belokurov et al. 2006, Martin, de Jong, & Rix 2008) – in
particular, evidence for a large range in the abundance of carbon – together with evidence
for abundance spreads of both carbon and iron2in Segue 1 (MV,total∼ –1.5; Belokurov et
al. 2007, Martin et al. 2008). Section 2 presents observational material, while Sections 3
and 4 present radial velocities of stars in the fields of Bo¨ otes I and Segue 1 and address
the question of galaxy membership. In Section 5, we present atmospheric parameters Teff,
log g, and [Fe/H], while in Sections 6 and 7 we reconsider the question of membership of
two apparently C-rich, extremely metal-poor stars in Segue 1, and present relative carbon
abundances, [C/Fe], for 16 radial-velocity candidate members in Bo¨ otes I and three radial-
velocity candidate members of Segue 1. The available data show that of four radial-velocity
candidate members of Segue 1 for which data of sufficient quality are available, one is carbon-
rich and extremely metal-poor ([Fe/H] = –3.5, [C/Fe] = +2.3), similar to the extremely rare
1[Fe/H] = log(N(Fe)/N(H))star– log(N(Fe)/N(H)⊙)
2Based on analysis of the Ca II K line, and the assumption that [Ca/Fe] follows the basic Galactic
relationship. See Section 5.
– 4 –
carbon enriched metal-poor (CEMP) stars having [Fe/H] ? –3.5 in the Galactic halo. In
Section 8 we discuss our results for abundance spreads and dispersions and their implications
for the formation, chemical enrichment, and evolution of ultra-faint galaxies. We show how
the comparably massive globular cluster ω Cen is consistent with the inferred self-enrichment.
We continue the discussion and summarize our results in Section 9.
2. OBSERVATIONAL MATERIAL
Candidate red giant members of Bo¨ otes I and Segue 1 were observed with the Anglo-
Australian Telescope/AAOmega fiber-fed spectrograph3combination during 2007 April 18–
20 and 2006 May 23–29 (Bo¨ otes I only; the 2006 run was the first major visitor use of the
new AAOmega facility and these data sets were used to optimize and enhance the data
reduction system, 2dfdr; final data calibration and reduction used what is now the public
2dfdr system). This instrument provides simultaneous spectra of 400 targets (science targets
plus dedicated sky fibers) over a field of 2 degrees in diameter. The light is split by a dichroic
into blue and red regions and sent to two separate spectrographs; only the blue spectra from
the 2007 dataset will be discussed in this paper. These spectra have resolution R = 5000
and cover the wavelength range 3850–4540˚ A.
Stellar targets for observation were selected from the SDSS DR4 data set, based on
their position in the relevant color-magnitude diagram, using the selection masks from the
discovery papers (Bo¨ otes I: Belokurov et al. 2006; Segue 1: Belokurov et al. 2007).
2.1. Bo¨ otes I
We obtained useful spectra for stars in the magnitude range 17.5 ? g ? 21 (–1.6 ? Mg
? 1.9 for a distance of 65 kpc (Martin et al. 2008)). The input target list included stars up
to one degree from the galaxy center, equivalent to four half-light radii (rh) (Belokurov et
al. 2006; Martin et al. 2008), and contained blue-horizontal branch candidates (one of which
proved to be a quasar). Only the stars lying on the red giant locus will be discussed and
analyzed in this paper. The observed color-magnitude diagram of such stars with velocity
information is shown in Figure 1, while the distribution on the sky is shown in Figure 2.
3See http://www.aao.gov.au/local/ www/aaomega/
– 5 –
2.2. Segue 1
We obtained useful spectra for stars in the magnitude range 17 ? g ? 21.5 (0.4 ? Mg?
4.7 for a distance of 23 kpc (Martin et al. 2008)). We again selected targets many times the
nominal half-light radius (4.4′, Martin et al. 2008) from the center of this system, exploiting
the wide-field capability of the AAOmega system. Our target sample of 323 stars extended
to 40′from the galaxy center, corresponding to ∼ 9rh. As for the Bo¨ otes I field, the observed
color-magnitude diagram of those stars for which we obtained velocity information is shown
in Figure 3, while the distribution on the sky is shown in Figure 4. Note the wide color
range, due to the poorly known location of the red giant branch of this system. Segue 1 is
close enough that a direct comparison may be made with the fiducial locus of the metal-poor
globular cluster M92, derived from the SDSS imaging data (An et al. 2008) and adjusted4
to the same reddening and distance of Segue 1; this is the smooth curve in Figure 3.
3. RADIAL VELOCITIES AND MEMBERSHIP
Heliocentric radial velocities were determined using the HCROSS routine within the
FIGARO package (see http://www.starlink.rl.ac.uk/star/docs/sun86.htx/node425.html). This
performs a cross-correlation between the program stellar spectrum and a template, and de-
termines the relative radial velocity. An associated confidence level and formal error are
estimated (see Heavens 1993); we accept only velocities with confidence = 1 or 1.0000
since experience shows those to have the cleanest cross-correlation function and negligible
rate of spurious results. We excised the strong Ca II H & K lines from the correlation
analysis, although this made an insignificant difference in most cases. Further, there were
defects on the CCD that affected a small wavelength range in a subset of spectra and we also
calculated cross-correlations with that wavelength region (typically around 4380˚ A) excised.
The G-giant standard star HD171391 (heliocentric radial velocity of +6.9 km s−1), for which
spectra were obtained during the same observing run as the ultra-faint system candidates,
was used as the template since it provided more reliable cross-correlation than the alterna-
tive twilight-sky template for lower signal-to-noise spectra5. We calculated a weighted mean
(using the formal errors from the cross-correlation package as weights) of the velocities from
4We adopted the distance modulus of 14.75 for M92, taken from Kraft & Ivans (2003), E(B−V) of
0.02 mag, and the extinction and reddening in the SDSS filters calculated following An et al. (2008).
5We earlier reported, in Norris et al. (2008) heliocentric velocities for 16 high signal-to-noise candidate
members, using the twilight sky as a template. The velocities reported here differ in the mean by only
1.25 km s−1, with a dispersion of 3 km s−1.
– 6 –
the two different wavelength ranges, when both existed. We have also removed a handful
of stars which turned out to be velocity-variable (binaries?), or variable stars, or to have
photometry inconsistent with being a single star. This resulted in a sample of 122 stars in
the Bo¨ otes I field with reliable velocities, and 134 in the Segue 1 field. The observational
data and derived heliocentric velocities are given in Table 1 (Bo¨ otes I) and Table 2 (Segue 1).
The open symbols in Figure 1 and in Figure 2 indicate all those stars with reliable velocities.
The data were taken over several nights, with a wide variety of sky conditions, with the
resultant total integration times in 2007, for example, being 9.5 hr and 7.5 hr, for Bo¨ otes I
and Segue 1, respectively. The limitations of field acquisition, fiber placement and weather
mean that apparent magnitude is not a perfect predictor of signal-to-noise.
Our internal accuracy on one observation, from repeat observations of 6 stars in the
Bo¨ otes I field from 2006 and 2007 is 10 km s−1with a mean offset of 6 km s−1if stars
with confidence = 1.0000 are included, and 7 km s−1with a mean offset of −1.5 km s−1
if only the 4 stars with confidence = 1 are included6. As noted in Norris et al. (2008),
external errors on our radial velocities may be estimated from comparison with Martin et
al. (2007). Using only the velocities relative to the G-star template, HD171391, the six stars
from our 2007 dataset in common with Martin et al. have a mean offset of −2.7 km s−1and
a dispersion of 13 km s−1; this is dominated by one star, Martin et al.’s ID = 58, for which
they report an unusually large velocity error of 7.2 km s−1, compared to less than 2 km s−1
for the remaining 5 stars in common. Removing that star from the comparison gives a mean
offset of 3 km s−1and a dispersion of 2.3 km s−1. This particular subsample has, on average,
higher signal-to-noise than is typical for our observations. We estimate from our repeats and
standard-star observations that at typical S/N our velocities have combined internal and
external errors ∼ 10 km s−1.
3.1.1.Bo¨ otes I
Mu˜ noz et al. (2006) and Martin et al. (2007) reported a systemic heliocentric radial
velocity of ∼ 100 km s−1, with an internal dispersion of ∼ 7 km s−1, for Bo¨ otes I. The
histogram of our radial velocities is given in Figure 5, with the local maximum at around
6The difference in the cross-correlation analysis between a confidence value of 1 and one of 1.0000 is rather
subtle. Our experience with spectra of a range of signal-to-noise has shown that the resulting velocities are
such that a value of 1.0000 tends to give higher formal errors, while maintaining the same best estimate of
the velocity, when compared to a value of 1.
– 7 –
100 km s−1being due to members of Bo¨ otes I. With random errors of the radial velocities of
∼ 10 km s−1, our data are clearly incapable of resolving the internal kinematics of Bo¨ otes I.
(Our wide-field observations were designed to identify candidate radial-velocity members
out to large distances on the sky for abundance study.) Bo¨ otes I, at 65 kpc, is sufficiently
distant that interloper giant stars from the smooth stellar halo are unexpected, but field main
sequence stars could contaminate our candidate members of Bo¨ otes I. The line-of-sight is
towards Galactic latitude ∼ +70◦, so that all Galactic components will have a mean velocity
close to zero. Star-count models, including the Besan¸ con model of the Galaxy (Robin et
al. 2003), predict that stars in the Milky Way will have a distribution of heliocentric radial
velocities that peaks at ∼ −5 km s−1; matching our selection criteria (as well as we can)
to the Besan¸ con model interface predicts that of Galactic stars, ∼ 6% will be observed to
have heliocentric radial velocities at values higher than 75 km s−1. Thus with our sample
size of 122 stars, we expect of order eight Galactic stars at high positive velocities. The
Besan¸ con model predictions are indicated in Figure 5 and provide a satisfactory match to
the distribution at lower velocities. The extra peak due to the presence of Bo¨ otes I is quite
pronounced; the thin vertical dotted lines indicate the bin edges that contain our velocity
range for candidate radial-velocity members, 85 ≤ Vhelio(km s−1) < 130 (the bin edges are
80 km s−1and 140 km s−1and there are no other stars with velocities in these bins). There
are 36 stars in this range, with a mean velocity of 105 km s−1and a dispersion of 20 km s−1.
There is a total of 40 stars with velocities above 75 km s−1, so one might expect that of
order four of the those in the range we have selected as candidate radial-velocity members
of Bo¨ otes I are in fact Galactic contaminants, based on the Besan¸ con model predictions.
The filled symbols in Figures 1 and 2 indicate our candidate radial-velocity members.
These candidates are also flagged in the final column in Table 1.
Figure 6 shows the distribution of velocities as a function of projected radial distance
from the galaxy’s center. It is apparent from this figure that we identify radial-velocity
members well beyond the nominal half-light radius. The outer parts of Bo¨ otes I will likely be
more susceptible to field contamination, and we can investigate this by testing the observed
distribution of velocities at distant projected locations against a Gaussian, representing the
field Galaxy with no superposed dwarf galaxy. There are 56 stars that are more distant
than 35′(2.7 half-light radii, or 4.5 exponential scale-lengths) from the center of Bo¨ otes I,
where there is a gap in the distribution of candidate members in Figure 6. These stars have
a velocity distribution that is in fact well-represented by a Gaussian, with mean velocity
of −11 km s−1and dispersion of 75 km s−1(not dissimilar to the Besan¸ con predictions).
Adopting this Gaussian model, we calculate the fraction of these stars which would lie by
chance in the velocity interval 85 − 130 km s−1(our selection for members of Bo¨ otes I) to
be 7%, or four stars out of our 56. The number of candidate radial-velocity members of
– 8 –
Bo¨ otes I in this range of parameter space is six. That is, we have only moderate confidence
in our detection of member stars of Bo¨ otes I beyond 35′, based on velocity alone. That said,
high-resolution, relatively high-S/N data recently obtained for Boo–980, which lies at 3.9rh
and which we shall present elsewhere (Frebel et al. 2010c, in prep.), lead to a (preliminary)
radial velocity of 99.0 ± 0.5 km s−1(internal error) and abundance [Fe/H] = –2.9 ± 0.2 that
support its membership.
For completeness, repeating the fit to a Gaussian for the stars projected within the inner
35′, and fitting only to stars outside the nominal velocity membership range, predicts three
contaminating field stars in the list of 30 candidate members. Again this predicted total of
seven contaminants agrees with the Besan¸ con model predictions.
3.1.2. Segue 1
We noted in Section 1 the competing suggestions of Geha et al. (2009) and Niederste-
Ostholt et al. (2009)7that the Segue 1 system comprises either an ultra-faint dwarf or
material originating in the Sgr dSph. Given the possibility that it might actually comprise
two distinct components, together with its extremely small baryonic mass (∼ 1000M⊙, Mar-
tin et al. 2008), one might not be surprised to find that the establishment of membership is
problematic. We shall see that this is indeed the case.
The histogram of radial velocities for the Segue 1 field is given in Figure 7. Geha et
al. (2009) reported a value of 206 km s−1for the systemic radial velocity of Segue 1, with
an internal velocity dispersion of 4.3 km s−1. Again our data cannot resolve the internal
kinematics, and we may expect true members to be scattered into ±2σ of the systemic
velocity, where here σ = 10 km/s, our random error. Our velocity distribution shows a
reasonably well-defined local enhancement, of nine stars, between 185 ≤ vhelio(km s−1) <
230; these stars occupy the three bins indicated in the histogram of Figure 7. We use this
range to define our candidate members. Extending the range to 170 < vhelio(km s−1) < 250
would add one star at each end, for a total of 11 candidate radial-velocity members; the
plots here show only the nine candidates with 185 ≤ vhelio(km s−1) < 230. Figure 8 shows
the distribution of velocities as a function of projected radial distance from the system’s
center. It is apparent from these figures that we identify candidate radial-velocity members
well beyond the nominal half-light radius.
7Niederste-Ostholt et al. (2009) emphasize that Segue 1 lies in a very complex part of the outer Galaxy,
not only projected onto the tidal stream of the Sagittarius dSph, but also plausibly at the same distance
(∼ 25 kpc) with velocity similar to that of Sgr stream members.
– 9 –
The Galactic coordinates of Segue 1 are (ℓ, b) ∼ (220◦, +50◦) and, as discussed in
Geha et al. (2009), the velocity distribution of Galactic stars is expected to peak well below
the systemic velocity of Segue 1. The Besan¸ con model predictions discussed in Geha et
al. (2009) lead to the expectation that only 2.5% of the total sample of Galactic stars should
have velocities in the range 190 < vhelio(km s−1) < 220. However, the Besan¸ con model,
which assumes smooth standard kinematics for the field stars, is of limited usefulness in this
line-of-sight, given the known presence of the Sagittarius stream.
We may make a crude estimate of possible contamination directly from our own velocity
distribution function, since this presumably includes some of the Sgr stream and other local
halo structures which are not included in the Besan¸ con model. Our velocity distribution
declines roughly linearly in number, from the peak around 0 km s−1to close to zero objects
at ∼ 200 km s−1. Simply extrapolating that decline under the velocity range of Segue 1
is an uncertain process, but suggests that up to four of the nine candidates could be field
contaminants. Given the complexity of the local field, and the possible similarity of kine-
matics between Segue 1, some part of the Sgr streams, and possibly the nearby Orphan
stream (Belokurov et al. 2007), no robust ab initio spatial distribution model is available to
motivate a joint position-velocity membership criterion (see Niederste-Ostholt et al. 2009,
for an extended discussion).
Whatever Segue 1 is or was associated with, it has the color-magnitude diagram (CMD)
of an old metal-poor population, and the derived luminosity function above the turnoff and on
the RGB must be consistent with stellar evolution. Application of such consistency checks
is very uncertain. Geha et al. (2009) identify two (blue) horizontal branch members and
one might consider scaling from this to the expected number of RGB members. However,
as shown by Niederste-Ostholt et al. (2009), there may be a significant number of BHB
interlopers from the Sgr stream, so the status of the two HB stars ascribed by Geha et al. to
Segue 1 cannot be taken as assured. The total luminosity of Segue 1 could in principle be
used to predict the number of members on the RGB, for example by assuming a stellar
population identical to that of the metal-poor globular cluster M3 (for which Renzini & Fusi
Pecci (1988) have tabulated the relative memberships of different evolutionary stages). The
spatial distribution of our candidate members does not match well that used in the estimate of
the total luminosity (Belokurov et al. 2006), and indeed the general membership uncertainties
of stars in this line of sight mean that the estimated luminosity itself is uncertain. Keeping
this complication in mind, Table 2 of Renzini & Fusi Pecci shows that M3 contains 342
RGB stars for every 3 × 104L⊙, leading to the expectation of 4 RGB stars for Segue 1,
adopting a value of 350L⊙. This estimate includes stars all the way down to the base of
the RGB, for which Mg ∼ +3.0, or g ∼ 19.8 for stars at the distance of Segue 1. Seven
of our candidate radial-velocity members are brighter than this limit, suggesting indeed
– 10 –
that a significant fraction, around half, are non-members. It is not possible to say which
stars are the contaminants and which are bona fide members. Although, as may be seen
from Table 2, only 4/7 of our candidates are within 3 half-light radii, using this information
involves assumptions about the (unknown) dynamical state of the system. As we note below,
our chemical abundance data do not allow resolution of this uncertainty, but offer several
possible interpretations of the nature of Segue 1.
We complete this section by noting that there is also evident in Figure 8 a significant
sample of stars with velocity 300 km s−1, which are distributed broadly across the field, and
which are not predicted by models with standard Galactic kinematics. Such a local peak in
the velocity distribution was also found by Geha at al. (2009)8, who tentatively identify it
with tidal debris from the Sagittarius dSph. We discuss this stream further using additional
spectra in a separate paper (Frebel et al. 2010d, in prep.).
4.STARS WITH RELATIVELY HIGH-S/N AAOMEGA SPECTRA
Of some 98 Bo¨ otes I candidates having spectra with more than 200 counts per 0.34˚ A
pixel at 4150˚ A, and which hence had sufficient S/N for a determination of metal abundance
based on the Ca II K line strength (see Norris et al. 2008), 16 fell in the radial-velocity
range 90 < Vr< 115 km s−1, consistent with a high probability of being Bo¨ otes I members.
Table 3 presents basic data for these objects, where columns (1)–(3) contain the star name,
radial distance from the center of the galaxy, and radial velocity, respectively. These stars
have a mean velocity of 105 km s−1, and a dispersion of 6.5 km s−1.
We note here for completeness that had we increased the radial-velocity limits for can-
didate membership to the range 85–130 km s−1adopted above in Section 3.1.1, two further
objects would have been admitted as putative members. These are Boo–2 and Boo–71 in
Table 1, which have velocities 123 and 91 km s−1, and distances from system center of 15.4′
and 19.6′, respectively. For Boo–71 we obtain [Fe/H] = –2.2, using the technique described
in Norris et al. (2008). For Boo–2, however, our spectrum is of rather poor quality and
we hesitate to report an abundance – very roughly we estimate [Fe/H] ∼ –2.5. While these
objects are clearly worthy of further consideration, we shall not discuss them further here.
For the Segue 1 candidates, we shall consider here only those stars having spectra with
more than 200 counts per 0.34˚ A pixel at 4150˚ A and velocities in the range 185–230 km s−1
8There is no overlap between our sample and that of Geha et al. (2009)
– 11 –
following the discussion in Section 220.127.116.11The data for the five Segue 1 candidates that meet
these criteria are given in the first five rows of Table 4 (which has the same column structure
as Table 3). Their spectra are presented in Figure 9 over the wavelength range 3900–4400
˚ A. One sees immediately that these spectra are not what one might have expected for a
sample of objects taken from a stellar system with a monomodal abundance distribution.
Two things are obvious. First, there is a large range in the strength of the G-band at 4300
˚ A; and second, the Ca II H & K lines (at 3968˚ A and 3933˚ A) differ widely from star to star
within the group. In particular, Seg 1–7 and Seg 1–98 have weak and somewhat ill-defined
Ca II lines and very strong G-bands. Such behavior is not unprecedented in extremely
metal-poor stars of the Galactic halo. Figure 10 compares the spectra of these two stars
with those of the halo, extremely metal-poor, C-rich giants CS22949–037, CS29498–043, and
HE0107–5240 (obtained with the Double Beam Spectrograph on ANU’s 2.3m telescope on
Siding Spring Mountain), which collectively have –5.4 < [Fe/H] < –3.8 and 0.9 < [C/Fe] <
3.8 (McWilliam et al. 1995; Aoki et al. 2002; Christlieb et al. 2004). One consequence of
the apparent carbon-richness of the two Segue 1 objects is that the contamination by CH
lines in the region of the Ca K line may lead to erroneously overestimated iron abundances
derived from the K line on low resolution spectra (see e.g. Christlieb et al. 2002; Frebel et
al. 2005; and Beers & Christlieb 2005).
There are also two mundane explanations of the weak Ca II lines in Seg 1–7 and Seg 1–
98 that one should consider. The first is that the effect is due to the relatively low S/N
of our spectra. The second is that both stars are high-velocity (Vr ∼ 200 km s−1) stars
with Ca II H & K lines that have emission cores10. We searched our collection of some 3000
medium resolution (FWHM ∼ 2.5˚ A) spectra of Hamburg ESO Survey (HES) metal-poor
candidates (see Norris et al. 2007) for stars that have weak Ca II H & K lines as the result
of core emission, and found some 10 objects with relatively weak H & K emission leading to
weak overall H & K line absorption. In Figure 11 we compare the spectra of Seg 1–7 and
Seg 1–98 with four of these objects. There is clearly not a good match between the spectra
of the HES stars and those of the candidate Segue 1 stars. However, one might imagine that
if in the candidate Segue 1 stars there were weaker emission features than those seen in the
HES stars in Figure 11 the abundances we derived for the candidate Segue 1 stars could be
9For the larger velocity range 170–250 km s−1considered in Section 3.1.2 one further putative Segue 1
member is admitted. Seg 1–117 in Table 2 has radial velocity 173 km s−1, distance from galaxy center 28.0′,
and [Fe/H] = –1.1. We shall not consider this object further.
10Given the decrease of Ca II H & K lines emission with increasing age (at least in dwarfs; see Barry 1988),
the high velocities (and presumably large ages) of the Segue 1 objects might suggest less contamination from
– 12 –
5. ATMOSPHERIC PARAMETERS
To determine relative carbon abundances ([C/Fe]) in what follows, we shall also need
the atmospheric parameters effective temperature (Teff), surface gravity (log g), and metal
abundance ([M/H]), where for simplicity we shall assume [M/H] = [Fe/H]. In Tables 3 (for
Bo¨ otes I) and 4 (for Segue 1) we present data that we have used for this purpose. Columns
(4)–(6) contain ugriz photometry for g, (g − r)0, and (r − z)0from Data Release 7 of the
Sloan Digital Sky Survey (Abazajian et al. 200911), where the colors have been corrected for
reddenings corresponding to E(B − V ) = 0.02 (Belokurov et al. 2006) and 0.032 (Geha et
al. 2009), for Bo¨ otes I and Segue 1, respectively. From our spectra of the Segue 1 objects in
Table 4, we measured values of the Ca II K line-strength index, K′, defined by Beers et al.
(1999), which are presented in column (7) of Table 4. For Bo¨ otes I, K′values from Norris et
al. (2008) are included in column (7) of Table 3.
As described in Norris et al. (2010), Teffand log g can be estimated for metal-poor red
giants in dSph systems by using ugriz photometry together with the synthetic ugriz colors
of Castelli12and the Yale–Yonsei (YY) Isochrones (Demarque at al. 200413) with an age of
12 Gyr, and the assumption that the stars lie on the red giant branch of the system.
For the Bo¨ otes I and Segue 1 stars investigated here, values of Teff and log g were
obtained for each of (g − r)0 and (r − z)0, and also (B − V )0 (see below). A check on
these atmospheric parameters was obtained by using a technique similar to that described
by Norris et al. in which the absolute magnitude MV – derived from the apparent magnitude
and distance modulus of the parent system – replaces color, together with the adoption of
the Yale-Yonsei Isochrones and the assumption that the stars lie on the red giant branch of
the parent dSph. To determine the estimated values of absolute visual magnitude MV, we
used the Lupton14(V, g)– transformation, together with our adopted reddening, the distance
moduli of Martin et al. (2008), and the assumption AV = 3 × E(B − V ). Then, for the YY
isochrone of assumed abundance [Fe/H] (and the above age of 12 Gyr), by interpolation in
the (Teff, MV)– and (log g, MV)– relationships defined by the isochrone we determine the
– 13 –
values of Teff and log g corresponding to the derived value of MV. Agreement between the
atmospheric parameters obtained from ugriz colors with those from absolute magnitude for
Bo¨ otes I was excellent: for the 16 stars in Table 3 we obtained average differences ?∆Teff? =
70K and ?∆log g? = 0.2. For Segue 1, on the other hand, the agreement was considerably
poorer, with ?∆Teff? = 230K and ?∆log g? = 0.5. If, however, we exclude Seg 1–42, the
hottest star in the sample, and for which the two Teffand log g estimates differ by 440K and
0.9 dex respectively, we obtain ?∆Teff? = 140K and ?∆log g? = 0.4. We shall return to this
point below, once we have discussed the abundances of the putative Segue 1 members.
Metal abundances were determined in two ways. First, we used the calibration of Beers
et al. (1999), which permits one to determine [Fe/H], given observed values of K′and (B–
V)015. In order to do this, we determined (B–V)0from the values of (g − r)0in Tables 3
and 4. For stars with (g − r)0> 0.545 (corresponding to (B − V )0> 0.7), we adopted the
transformation (B − V )0= 1.197×(g − r)0+ 0.049, appropriate for metal-poor red giants,
following Norris et al. (2008)16. This applied for all but two objects (Seg 1–31 and Seg 1–42),
for which we adopted (B − V )0= 0.916×(g − r)0+ 0.187, from Zhao & Newberg (2006),
valid for metal poor-stars over the range –0.15 < (g − r)0< 0.55.
The second method of abundance determination involves model atmosphere analysis
of high-resolution, high S/N, spectra obtained with VLT/UVES (Norris et al. 2010, for
Boo−1137), and with VLT/UVES/Flames for a further seven of the Bo¨ otes I stars, which
will be reported elsewhere (Gilmore et al. 2010 in prep.).
The atmospheric parameters for Bo¨ otes I and Segue 1 are presented in Table 3 and
4, where columns (8)–(9) contain Teff and log g obtained from colors as described above,
while column (10) presents [Fe/H] (except for Seg 1–7 and Seg 1–98), based on the above
two methods, where the tabulated abundance gives precedence to the UVES based value if
available, failing which the AAOmega K′-based result is used.
Our abundances for Segue 1 deserve comment. For Seg 1–31 and Seg 1–71, we derive
relatively high values of [Fe/H] = –1.9 and –2.2, respectively. To give the reader a feeling
for the case for the high abundances of these objects, we compare their spectra in Figure 12
with those of Galactic halo giants having similar colors, and abundances in the range –4.8 <
15The reader should be aware the Beers et al. (1999) calibration assumes that the same [Ca/Fe] vs [Fe/H]
relationship applies within both the metal-poor Galactic calibration objects and the dSph satellites. While
this is not true at higher abundances, it appears to hold for [Fe/H] < –2.0 (e.g. Scl: Tolstoy et al. 2009;
UMa II and Com: Frebel et al. 2010a; Bo¨ otes I: Norris et al. 2010).
16Our quoted [Fe/H] values for Bo¨ otes I differ trivially from those of Norris et al. (2008) because of the
small differences between SDSS DR4 and DR7.
– 14 –
[Fe/H] < –2.2. (The colors of the Galactic stars are taken from Cayrel et al. (2004), Norris,
Bessell, & Pickles (1985), and Norris et al. (2007), while their abundances come from Cayrel
et al. (2004), Chiba & Yoshii (1998), and Norris et al. (2007).) Examination of the spectra,
in the region of the Ca II H & K lines (3900–4000˚ A), clearly supports our relatively high
abundances. That said, we recall that Geha et al. (2009) have reported a red giant in Segue 1
(their star 3451364) with [Fe/H] = –3.3. We note then that the Segue 1 spectra in Figure 12
have much stronger Ca II H & K lines than seen in CS22897–008 (in the third panel from
the top), which has [Fe/H] = –3.4, similar to the abundance of the Geha et al. star. Said
differently, there appears to exist a large abundance range (∆[Fe/H] ∼ 1 dex) within the
sample of candidate members of Segue 1.
The abundances of the more-enriched candidate members of Segue 1 are, however, close
to that of a typical field halo star at the distance of Segue 1 (the outer halo of Carollo et
al. 2007). Further, there are clear indications of an abundance gradient from the core of the
Sgr dSph to its tidal streams, with a mean metallicity [Fe/H] ∼ –1 derived from M giants
(at heliocentric distances greater than ∼ 10 kpc; Chou et al. 2007) and –1.8 from RR Lyrae
stars in the leading arm, at heliocentric distances of ∼ 50 kpc (Vivas, Zinn, & Gallart 2005).
These results suggest that at [Fe/H] ∼ –1.5 – –2, both field halo stars and Sgr debris would
be hard to distinguish from members of Segue 1, given that some models of the Sgr streams
predict similar velocities (e.g. Niederest-Ostholt et al. 2009).
For the two putative C-rich objects in Segue 1, we also derived [Fe/H] using the Beers
et al. (1999) formalism, which leads to [Fe/H] = −3.6 and –4.0 for Seg 1–7 and Seg 1–98,
respectively. At this stage we regard these values as uncertain for the reasons discussed in
Section 4. We shall defer further discussion of them until Section 6.
It is somewhat difficult to estimate realistic errors for Teffand log g that include both
internal and potential external errors. The cited errors in the DR7 ugriz colors propagate
to relatively small errors in Teff and log g, and it is probably more realistic to concentrate
on systematic effects. One measure of this could be the internal spread in the results of the
three methods based on calibrations of (g − r)0, (r − z)0, and (B − V )0described above.
We find that the averages of the standard error of the mean for the 16 Bo¨ otes I objects in
Table 3 are ?∆Teff? = 40K and ?∆log g? = 0.1. These are somewhat smaller that the average
differences of ?∆Teff? = 70K and ?∆log g? = 0.2 obtained above from colors and absolute
magnitude (i.e. assuming that the stars are indeed red giants at the distance of Segue 1).
We adopt the larger differences as error estimates in what follows. For the five Segue 1 stars
we find average differences of ?∆Teff? = 90K and ?∆log g? = 0.2; and if we exclude Seg 1–42,
the problematic star discussed above, these values become ?∆Teff? = 100K and ?∆log g? =
– 15 –
We now return to the issue of the discrepant values of Teffand log g obtained above for
Seg 1–42 from estimates based on color and absolute magnitude. While this object appears to
be the hottest of the stars in Table 4, it is also the most metal-rich, with [Fe/H] = –1.5. Given
that this is fundamentally at odds with basic concepts of evolution on the red giant branch
(more metal-rich stars should be cooler), and our assumption that Seg 1–42 lies on the giant
branch of Segue 1, we shall exclude Seg 1–42 from further consideration, on the assumption
it is not a red giant member of Segue 1. In what follows we shall adopt mean errors of
σ(Teff) = 140K and σ(log g) = 0.4 for the four remaining candidate Segue 1 giants, the
larger of our two estimates obtained when we exclude Seg 1–42 from consideration. We note
parenthetically here that the differences between our two methods of deriving atmospheric
parameters are commensurate, to within these errors, with the assumption that the other
stars in Table 4 lie on the RGB of Segue 1.
Abundance errors for Bo¨ otes I are σ[Fe/H] = 0.35 for data obtained with AAOmega
(Norris et al. 2008) and σ[Fe/H] = 0.15 for the results from VLT/UVES and VLT/Flames/UVES
(Norris et al. 2010; Gilmore et al. 2010 in prep.)17. For Seg 1–31 and Seg 1–71 we adopt
σ[Fe/H] = 0.4.
6. FURTHER ASSESSMENT OF THE PUTATIVE SEGUE 1 C-RICH,
EXTREMELY METAL-POOR STARS
It is difficult to over-emphasize the implications of Segue 1 membership of the two C-
rich, extremely metal-poor candidates Seg 1–7 and Seg 1–98. Put most simply, C-rich stars,
having [Fe/H] ? –3.3 are extremely rare: few are known, and only some seven have been
analyzed in detail18. The existence of such objects in the ultra-faint dwarfs would have
strong implications for these systems being building blocks of the Galaxy’s outer halo.
After the analysis reported here was complete, we sought to obtain high-resolution, high
S/N data for both Seg 1–7 and Seg 1–98. There were two significant developments. First,
17We note for completeness that for [Fe/H] the mean difference and the dispersion of differences for the
seven stars in our Table 3 having both AAOmega and VLT/Flames abundances are –0.06 and 0.44 dex,
respectively. For the four stars in common between our high-resolution VLT/Flames abundances in Table 3,
and those from the Keck/HIRES spectra of Feltzing et al. (2009) the corresponding values are –0.01 and
18We refer to CS22949–037 (McWilliam et al. 1995), CS22957–027 (Norris et al. 1997), CS29498–043 (Aoki
et al. 2002), HE0107–5240 (Christlieb et al. 2004), HE0557–4840 (Norris et al. 2007), HE1327–2326 (Frebel
et al. 2005), and G77–61 (Plez & Cohen 2005).