SEGUE 1: AN UNEVOLVED FOSSIL GALAXY FROM THE EARLY UNIVERSE^*

In the Milky Way, it is well known that low-metallicity halo stars ([Fe/H] < −1.0) have enhanced abundances of α-elements (e.g., Edvardsson et al. 1993), while at higher metallicities the [α/Fe] ratios smoothly approach the solar ratio (e.g., Tinsley 1979; McWilliam 1997; Venn et al. 2004). The canonical interpretation of this behavior is that it reflects the timescales of nucleosynthesis from different kinds of supernovae. Massive stars produce large amounts of α-elements both during their stellar evolution and in their explosions as core collapse supernovae. These elements are quickly returned to the interstellar medium because the lifetimes of such stars are short, less than 10 Myr. Type Ia supernovae, which primarily produce iron, do not begin exploding until a poorly known delay time (typically assumed to be of the order of 10⁸ yr; Maoz & Mannucci 2012) has elapsed since an episode of star formation. The [α/Fe] plateau at [Fe/H] < −1.0 then corresponds to the epoch when only core-collapse supernovae contributed significantly to the overall nucleosynthesis, and the decline to [α/Fe] = 0 occurs when Type Ia supernovae begin occurring in significant numbers as well.

The turnover from the high [α/Fe] plateau in the classical dSph galaxies occurs at lower metallicity than in the Milky Way, [Fe/H] ∼−2.5 (e.g., Tolstoy et al. 2003; Venn et al. 2004), revealing that they have been enriched on slower timescales than what is observed in the halo of the Galaxy (see, e.g., Tolstoy et al. 2009 for a review). Note that, at least in the case of Sagittarius, McWilliam et al. (2013) have questioned whether this explanation for the decline in [α/Fe] at high metallicity is correct, and they suggest a top-light initial mass function (IMF) as an alternative.

More recently, Kirby et al. (2008, 2011b) measured metallicities for dozens of individual stars in ultra-faint dwarfs using the medium-resolution spectroscopy from Simon & Geha (2007). Stars with metallicities of [Fe/H] < −3.0 were uncovered in surprisingly large relative numbers, whereas essentially no stars with [Fe/H] > −1.0 were found. This general characteristic of overall metal-deficiency correlates with the low luminosities of these galaxies. They extend the metallicity–luminosity relationship for dSph galaxies by several orders of magnitude in luminosity (Kirby et al. 2008, 2011c, 2013). Vargas et al. (2013) then used the same spectra to measure [α/Fe] abundance ratios. They found that among their sample of eight ultra-faint dwarfs, only Segue 1 does not show declining [α/Fe] ratios with increasing metallicity.⁵ Thus, this galaxy is the only dwarf galaxy known to have minimal chemical enrichment from Type Ia supernovae.

However, medium-resolution spectroscopy is limited in its ability to detect and measure the abundances of some elements, such as the neutron-capture elements. High-resolution spectroscopy can provide highly detailed abundance information for stars that are bright enough. High-resolution spectra with large wavelength coverage of stars down to magnitude of V ∼ 19.2 have been obtained for about a dozen stars in ultra-faint dwarfs by now. Three stars each in Ursa Major II and Coma Berenices were observed by Frebel et al. (2010b), Koch et al. (2008) observed two stars in Hercules, and other studies reported on one star each in Leo IV (Simon et al. 2010), Boötes I (Norris et al. 2010c), Segue 1 (Norris et al. 2010a), and the stream passing just in front of Segue 1 that may have originated in an ultra-faint dwarf (Frebel et al. 2013b).

Detailed studies of these stars, nearly all of which are at [Fe/H] ≲ −2.0, revealed close-to-identical chemical abundance patterns compared with halo stars, both in terms of the abundance ratios as well as more global population signatures such as a significant fraction of metal-poor stars being strongly enhanced in carbon relative to iron, and showing low neutron-capture element abundances. It has thus been suggested that the chemical similarity of halo and the ultra-faint dwarf galaxy stars could be due to the stars having formed from early gas that was enriched in the same fashion, i.e., exclusively by massive stars (Frebel et al. 2010b; Simon et al. 2010; Norris et al. 2010c, 2010a). Moreover, if the surviving ultra-faint dwarf galaxies had earlier analogs that were accreted by the Milky Way in its early assembly phases, the chemical similarity of halo and dwarf galaxy stars could also be interpreted as indicating that the early enrichment history of the destroyed dwarfs closely resembled that of the surviving dwarfs observed today, despite the different environments they formed in. In this picture, very low luminosity primordial dwarfs may have provided the now-observed metal-poor "halo" stars to the halo.

In this context, it is interesting to note that the analysis of deep Hubble Space Telescope (HST) color–magnitude diagrams (CMDs) of three ultra-faint dwarf galaxies, Hercules, Leo IV, and Ursa Major I, by Brown et al. (2012) shows that they are at least as old as the oldest globular clusters and likely nearly as old as the universe itself (their age is consistent with that of the globular cluster M92, which on the same scale is measured at 13.7 Gyr). Preliminary results for three additional ultra-faint dwarfs suggest that the stellar populations of all six galaxies are indistinguishable (Brown et al. 2013). This demonstrates that the metal-poor stars in these galaxies are as old as their chemical composition (galaxy average metallicities range from [Fe/H] ∼ −2.0 to −2.6; Kirby et al. 2008) suggests. Because of the apparently small age spreads in these systems, their more metal-rich stars also have to be similarly old, suggesting rapid enrichment. This could be due to the low-mass nature of these systems which would lead to a fast, significant build up of metals after only a few supernova explosions. Considering a plausible chemical composition of a first galaxy that may have survived to the present day, Frebel & Bromm (2012) thus argued that Segue 1 (together with Ursa Major II, Coma Berenices, Boötes I, and Leo IV) are candidate systems for such surviving first galaxies. Bovill & Ricotti (2011) agree that most of the ultra-faint dwarfs are consistent with expectations for reionization fossils, although they do not place Segue 1 in this category as a result of earlier estimates of its metallicity lying above the luminosity-metallicity relation established by brighter dwarfs. The metallicities derived in this paper and improved measurements of the L–Z relation by Kirby et al. (2013) demonstrate that Segue 1 is in fact consistent with the extrapolated metallicity–luminosity relationship of more luminous systems. Only additional chemical abundance data for more stars in as many of the ultra-faint dwarfs as possible will allow detailed tests of the hypothesis that these objects are fossils of the first galaxies by establishing a detailed account of the chemical composition of each galaxy.

Thus, in this study, we present chemical abundance measurements for six stars in Segue 1 that are just bright enough to be observable with high-resolution spectroscopy. Segue 1 is the faintest galaxy yet detected, and it has an average metallicity of [Fe/H] ∼ −2.5 to −2.7 (Norris et al. 2010b; Simon et al. 2011). It was one of five new ultra-faint galaxies discovered by Belokurov et al. (2007) using a matched filter search of SDSS DR5 and SEGUE photometry. Although Segue 1 contains very few stars, its position 50° out of the Galactic plane and away from the Galactic center aids in separating member stars from the foreground Milky Way population. It is also close enough (23 kpc) to permit spectroscopy of stars down to ∼1 mag below the main sequence turnoff (e.g., Geha et al. 2009; Simon et al. 2011). Segue 1 was initially presumed to be a globular cluster because of its small half-light radius (30 pc), but Geha et al. (2009) presented a strong case based on internal stellar kinematics that Segue 1 is highly dark-matter-dominated and therefore a galaxy. Geha et al. (2009) also demonstrated that Segue 1 lies on or near standard dwarf galaxy scaling relations. Niederste-Ostholt et al. (2009) found photometric evidence for tidal debris near Segue 1 and proposed that the velocity dispersion of Segue 1 was inflated by contamination from these disrupted structures, but Simon et al. (2011) showed that contamination is unlikely and that the measured velocity dispersion is robust. The extensive spectroscopy by both Simon et al. (2011) and Norris et al. (2010b) also established that Segue 1 has metallicity spreads of 0.7 to 0.8 dex in [Fe/H] and 1.2 dex in [C/H]. Along with being extremely underluminous and the most dark-matter-dominated and lowest-metallicity object currently known, Segue 1 is not only a galaxy, but perhaps the most extreme galaxy known.

With our new observations, our aim is to quantify the chemical evolution of this galaxy by constraining its enrichment processes. This way, we can learn about the limited star formation that occurred in this early system. In Section  2, we describe the observations and in Section  3 our analysis techniques. We interpret our chemical abundance results (Section  4) within the context of early galaxy formation and chemical evolution in Section  5. We conclude in Section  6.

The first spectroscopy of Segue 1 was obtained by Geha et al. (2009), who used medium resolution Keck/DEIMOS spectra to identify 24 member stars of the galaxy, including a single red giant for which they estimated [Fe/H] = −3.3. Simon et al. (2011) followed up this study by observing an essentially complete spectroscopic sample (again with Keck/DEIMOS) out to 2.3 half-light radii around Segue 1 and down to a magnitude limit of r = 21.7 (∼1 mag below the main sequence turnoff). The Simon et al. (2011) survey identified five additional members on the Segue 1 red giant branch (RGB), and a total of 71 member stars. Independently, Norris et al. (2010b) used medium resolution blue spectra from AAOmega on the Australian Astronomical Telescope over a much wider field to search for bright Segue 1 members. They found four of the same RGB stars as Simon et al. (2011), as well as another extremely metal-poor (EMP) giant, Segue 1-7, nearly four half-light radii from the center of the galaxy. One other candidate RGB member from the Norris et al. (2010b) sample, Segue 1-42, was found by Simon et al. (2011) to have a radial velocity inconsistent with membership in Segue 1, while the last (and faintest) RGB candidate identified by Norris et al., Segue 1-270, is unlikely to be a genuine member given its location ∼10 half-light radii from Segue 1 and its velocity offset of ∼20 km s⁻¹ from the systemic velocity.

As a result of these extensive observations and the remarkably puny stellar population of the galaxy, the seven known RGB stars are likely to represent a complete inventory of all stars in Segue 1 currently in the red giant phase of evolution (it also has two horizontal branch (HB) stars; Simon et al. 2011). For a Plummer radial profile, 10% of member stars should be located at projected radii beyond three half-light radii, within which the Simon et al. (2011) sample is more than 90% complete. In fact, two of the seven giants (29%) are beyond this distance, suggesting that the present sample includes most (if not all) of the RGB stars in Segue 1. Norris et al. (2010a) obtained a high-resolution Very Large Telescope spectrum of the brightest giant star, Segue 1-7, and analyzed its unusual chemical inventory, confirming its extremely low metallicity and carbon enhancement. As described below, we have now obtained high-resolution spectra of the remaining six Segue 1 giants, including two stars with [Fe/H] < −3.5 and two others at [Fe/H] > −1.8 that are the most metal-rich stars in an ultra-faint dwarf galaxy to be studied in detail.

We observed five of our six target stars with the MIKE spectrograph (Bernstein et al. 2003) on the Magellan-Clay telescope in 2010 March and May, and 2011 March. Observing conditions during these runs were mostly clear, with an average seeing of 08 to 10. Additional details of the MIKE observations are given in Table 1. MIKE spectra have nearly full optical wavelength coverage from ∼3500–9000 Å. A 10 × 5'' slit yields a spectral resolution of ∼22, 000 in the red and ∼28, 000 in the blue wavelength regime. We used 2 × 2 on-chip binning and the 10 slit for all stars except SDSS J100639+160008. The seeing conditions were better when we began observing this star, so we opted to employ a 07 × 5'' slit. It yields a resolution of ∼28, 000 and ∼35, 000, respectively.

Table 1. Observing Details

Star	α	δ	UT Dates	Slit	texp	g	E(B − V)	S/N	S/N
	(J2000)	(J2000)			(hr)	(mag)	(mag)	(5300 Å)	(6000 Å)
SDSS J100714+160154	10 07 14.6	+16 01 54.5	2011 Mar 10	10	5.5	18.86	0.027	35	45
SDSS J100710+160623	10 07 10.1	+16 06 23.9	2011 Mar 13/22	10	7.3	19.20	0.026	28	45
SDSS J100702+155055	10 07 02.5	+15 50 55.3	2011 Mar 11/12	10	9.2	18.50	0.033	35	46
SDSS J100742+160106	10 07 42.7	+16 01 06.9	2010 Apr 01	115	3.6	18.60	0.027	20	30
SDSS J100652+160235	10 06 52.3	+16 02 35.8	2010 Mar 07/08/18/19/23/24	10	15	18.89	0.029	42	50
SDSS J100639+160008	10 06 39.3	+16 00 08.9	2010 Mar 18/19/22, 2010 May 08/09	07	8	19.48	0.031	26	33

Notes. The S/N is measured per ∼33 mÅ pixel (MIKE spectra) and ∼20 mÅ pixel (HIRES spectrum).

Download table as:  ASCIITypeset image

Integration times ranged from ∼6 to 15 hr. The observations were typically broken up in 55 minute exposures to avoid significant degradation of the spectra by cosmic rays. These individual spectra generally had low counts given the faintness of the objects, with little or no flux detected below 4000 Å.

We observed the final star in the sample, SDSS J100742+160106, with the HIRES spectrograph (Vogt et al. 1994) on the Keck I telescope on 2010 April 1. The observations were obtained with a 115 × 7'' slit (providing a spectral resolution of 37,500), the kv389 blocking filter, and a total integration time of 3.6 h. The autoguider had difficulty guiding accurately during much of this time, initially because of unusually good seeing conditions and later as the star transited within a few degrees of the zenith, so many of the exposures were either unguided or employed manual guiding by the telescope operator. These guiding issues decreased the signal-to-noise ratio (S/N) of some exposures, but should not have any effect on the measurement of equivalent widths we are interested in for the purposes of this paper. We reduced the HIRES spectra using the IDL pipeline developed by J. X. Prochaska and collaborators.⁶

Reductions of the individual MIKE spectra were carried out using the MIKE Carnegie Python pipeline initially described by Kelson (2003).⁷ The orders of the combined spectrum were normalized and merged to produce final one-dimensional blue and red spectra for further analysis. The S/N of the spectra is modest and ranges from 20 to 40 at ∼5300 Å and 30 to 50 at ∼6000 Å. Radial velocity measurements yield values between 200 and 208 km s⁻¹, which are consistent with previous measurements (Simon et al. 2011).

In Figure 1, we show representative portions of the spectra of the program stars around the Ba line at 4554 Å and the Mg b lines at 5170 Å. The large range of metallicities found in Segue 1 is easily visible. For comparison we also add CD −38 245 with [Fe/H] ∼ −4.2 as well as Arcturus ([Fe/H] = −0.5), which bracket the metallicities of our Segue 1 stars.

Zoom In Zoom Out Reset image size

We measured the equivalent widths of metal absorption lines throughout the spectra by fitting Gaussian profiles to them. Continuum placement was challenging at times given the modest S/N of the data. Generally, lines between 4000 Å and 7000 Å were measured, based on a line list described in Roederer et al. (2008). In Table 2, we list the lines used and their measured equivalent widths for all elements, and 3σ upper limits for selected elements. For blended lines, lines with hyperfine-structure, and molecular features such as CH, we used the spectral synthesis approach in which the abundance of a given species is obtained by matching the observed spectrum to a synthetic spectrum of known abundance.

Table 2. Equivalent Width Measurements of the Segue 1 Stars

El.	λ	χ	log gf	EW (mÅ)	log  (dex)	EW (mÅ)	log  (dex)	EW (mÅ)	log  (dex)	EW (mÅ)	log  (dex)	EW (mÅ)	log  (dex)	EW (mÅ)	log  (dex)
	(Å)	(eV)	(dex)	SDSS J100714+160154		SDSS J100710+160623		SDSS J100702+155055		SDSS J100742+160106		SDSS J100652+160235		SDSS J100639+160008
CH	4313	...	...	syn	...	syn	6.86	syn	6.19	syn	5.83	syn	5.98	syn	5.45
CH	4325	...	...	syn	8.45	syn	6.96	syn	6.19	syn	6.03	syn	6.08	syn	5.67
Na i	5889.950	0.00	0.108	240.13	4.67	190.82	4.26	148.77	3.86	118.78	3.59	59.86	2.71	78.14	2.79
Na i	5895.924	0.00	−0.194	228.99	4.90	171.68	4.36	146.43	4.13	104.38	3.63	49.00	2.80	57.83	2.69
Mg i	3829.355	2.71	−0.208	...	...	...	...	...	...	...	...	105.97	4.49	...	...
Mg i	3832.304	2.71	0.270	...	...	...	...	...	...	...	...	149.84	4.53	136.37	4.37
Mg i	3986.753	4.35	−1.030	136.44	6.79	...	...	41.40	5.57	45.98	5.71	...	...	...	...
Mg i	4057.505	4.35	−0.890	115.68	6.52	102.39	6.39	52.37	5.61	58.80	5.79	...	...	...	...
Mg i	4167.271	4.35	−0.710	147.70	6.72	113.51	6.41	66.83	5.66	...	...	...	...	...	...
Mg i	4351.906	4.34	−0.525	...	...	...	...	...	...	...	...	16.32	4.66	...	...

Only a portion of this table is shown here to demonstrate its form and content. A machine-readable version of the full table is available.

Download table as:  DataTypeset image

In many instances, the element abundances were so low that no absorption lines could be detected. The modest S/N of the data also made detections of weak lines difficult. This was the case for all neutron-capture elements in all our stars except the most metal-rich object and the Ba measurements in SDSS J100742+160106 (see Section 4.4). We determined upper limits through spectrum synthesis by matching synthetic spectra to the noise level in the region of the non-detected absorption line. These generally agree well with calculations for 3σ upper limits.

We use one-dimensional plane-parallel model atmospheres with α enhancement from Castelli & Kurucz (2004) and the latest version of the MOOG analysis code (Sneden 1973; Sobeck et al. 2011). The abundances are computed under the assumption of local thermodynamic equilibrium (LTE). We calculated final abundance ratios [X/Fe] using the solar abundances of Asplund et al. (2009). The elemental abundances for our sample are given below.

To derive stellar parameters spectroscopically we follow the procedure described in Frebel et al. (2013a). For completeness we repeat essential details below. We derive spectroscopic effective temperatures by demanding that there be no trend of Fe i line abundances with excitation potential. We use initial temperature guesses based on the shape of the Balmer lines. Using the ionization balance, i.e., demanding that Fe i lines yield the same abundance as Fe ii lines, we derive the surface gravity, log g, for all sample stars. The microturbulence, v_micr, is obtained iteratively in this process by demanding no trend of abundances with reduced equivalent widths.

However, this spectroscopic technique is known to deliver cooler temperatures (and lower surface gravities) than those derived from photometry, especially for cool giants (see, e.g., Hollek et al. 2011 for a discussion). To alleviate this effect, we thus applied temperature adjustments of T_{eff, corrected} = T_{eff, initial} − 0.1 × T_{eff, initial} + 670 as presented in Frebel et al. (2013a) after obtaining the spectroscopic parameters. This relation was derived from the analysis of seven well-studied stars in the literature for which photometric temperatures were available. We then repeated determining the stellar parameters spectroscopically but with the adjusted temperature fixed while finding the new, corresponding log g and v_micr. We estimate our temperature uncertainties to be ∼100–150 K (see also Frebel et al. 2010b as well as Section 4.5 for more details on uncertainties). Uncertainties in log g and v_micr are estimated to be 0.3 dex and 0.3 km s⁻¹. Finally, we visually inspect the shape of the Balmer lines to confirm the temperatures. Compared to other stars with known temperatures, our derived values are in qualitative agreement with the shapes of the Balmer lines for those temperatures. Having available stellar temperatures that are not as discrepant from photometric temperatures makes subsequent abundance comparisons with other stars more straightforward. Another advantage of the spectroscopic approach over photometric temperatures is that it is reddening-independent (although Segue 1 has only moderate reddening of E(B − V) ∼ 0.03 mag).

Figure 2 shows our final stellar parameters in comparison with α-enhanced ([α/Fe] = 0.3) 12 Gyr isochrones (Green et al. 1984; Kim et al. 2002) covering a range of metallicities. Our values agree very well with those of the isochrone. Table 3 lists the individual stellar parameters of all stars. For completeness we added Segue 1-7, the other star in Segue 1 that was observed at high spectral resolution by Norris et al. (2010a).

Zoom In Zoom Out Reset image size

Given the modest S/N of the spectra, we note that relatively few iron lines are measured, especially for Fe ii. In the case of the two most metal-poor stars, no Fe ii lines for the gravity determination were confidently detected. We thus chose gravity values from the isochrone appropriate for their effective temperatures. We note, however, that upper limits on the Fe ii lines (taken at face value) would yield gravities within 0.5 dex of the adopted values. This suggests that our final surface gravities, which place them on the RGB rather than the HB, are reasonable choices (see also below).

We also used published equivalent width measurements from Norris et al. (2010a) to obtain stellar parameters and abundances for Segue 1-7 in the same way as for our other Segue 1 stars. The agreement with the values they determined is very good. We find T_eff = 4990 K, log g = 2.05, v_micr = 1.35 km s⁻¹, and [Fe/H] = −3.55, compared to 4960 K, 1.9, 1.3 km s⁻¹, and −3.57 (Norris et al. 2010a).

We show the available SDSS photometry in Figure 3 in the form of a CMD overlaid with an M92 fiducial sequence and a HB track from M13. g magnitudes and reddening are also listed in Table 1. Overall, the location of the stars on the giant branch in the photometry is confirmed with the stellar spectroscopic parameters. We note that the g−i color of SDSS J100639+160008 from the SDSS is slightly offset from the best-fitting Segue 1 isochrone by ∼0.1 mag to the blue; since this offset disappears in other colors (e.g., g−r and r−z), it could indicate an error in the SDSS i magnitude. Since this star has the lowest metallicity in the sample we were not able to determine its surface gravity spectroscopically because no Fe ii lines were detected. However, we explicitly checked whether we could have obtained a slightly lower than adopted gravity, but the upper limit of Fe ii lines suggest a gravity close to the isochrone or perhaps slightly higher. The spectrum is not consistent with a gravity high enough to place the star on the main sequence. Even with a higher gravity the metallicity does not increase significantly, ruling out the possibility that SDSS J100639+160008 is a more metal-rich foreground star. We furthermore show the spatial distribution of the full sample of Segue 1 members from Simon et al. (2011) in Figure 3, along with the locations of the stars studied in this paper, and also Segue 1-7 from Norris et al. (2010a) which is located near 4 half-light radii, and hence was not covered in the Simon et al. (2011) study.

Zoom In Zoom Out Reset image size

To investigate the possibility of contamination of the Segue 1 giant sample by foreground stars, we consulted the Besançon model of the Milky Way (Robin et al. 2003). The model predicts that the surface density of Milky Way stars that meet the color and velocity cuts used by Simon et al. (2011) to identify Segue 1 members and have magnitudes consistent with the Segue 1 giants (19.5 > r > 17) is 0.8 stars deg⁻², or 0.07 stars in the area over which the Simon et al. (2011) survey is essentially complete. This estimate is in agreement with the empirical sample of non-Segue 1 members from Simon et al. (2011). Moreover, ∼90% of these potential contaminants are actually on the main sequence, and so would not be confused with Segue 1 giants given the surface gravity measurements from our high-resolution spectra. Thus, we expect only ∼0.01 Milky Way halo giants with the same color, magnitude, and velocity as Segue 1 to be spatially coincident with the galaxy; since there are only seven Segue 1 giants, it is very unlikely that our sample contains a Milky Way star masquerading as a Segue 1 member.

Carbon is an important element for tracing early star formation as well as enrichment and nucleosynthesis processes. We measured the carbon abundances in all the stars from two CH features at ∼4313 Å and 4323 Å. SDSS J100710+160623, SDSS J100702+155055, and SDSS J100742+160106 (with [Fe/H] = −1.7, −2.3, and −2.4, respectively) have [C/Fe] abundances close to the solar ratio. This is typical for many metal-poor stars, as can be seen in Figure 4. On the contrary, SDSS J100714+160154 is found to possess a very large overabundance of carbon. Given another critical abundance clue—enhanced neutron-capture element abundances associated with the s-process (see below)—this star appears to be a mildly metal-poor CH star that received its carbon from a binary companion.

Zoom In Zoom Out Reset image size

At [Fe/H] = −3.6 and −3.7, SDSS J100652+160235 and SDSS J100639+160008 have large [C/Fe] values of 0.9 and 1.2, respectively. Together with Segue 1-7, having [C/Fe] = 2.3 at [Fe/H] = −3.6, they can be classified as carbon-enhanced metal-poor (CEMP) stars, as their carbon exceeds a threshold value of [C/Fe] = 0.7 (Aoki et al. 2007). With the three most metal-poor stars in Segue 1 being CEMP stars, and keeping in mind that our Segue 1 sample is small, ∼50% of Segue 1's brightest stars presumably formed from gas that was enriched in carbon. (Here we are excluding SDSS J100714+160154 since its carbon abundance is an obvious result of a mass transfer event.) Moreover, this also implies that the CEMP fraction is ≳ 50% for stars with [Fe/H] < −3.5.

The fraction of CEMP stars among metal-poor halo stars is known to increase with decreasing metallicity (e.g., Rossi et al. 1999; Frebel et al. 2006b; Lucatello et al. 2006; Cohen et al. 2006), pointing to the importance of carbon in the early universe. Beginning with 15% of stars with [Fe/H] < −2.0 and 18% at [Fe/H] < −2.5, the fraction becomes 25% among stars with [Fe/H] < −3.0, 45% at [Fe/H] ⩽ −3.5, and 100% below [Fe/H] < −5.0 (based on literature data collected by Frebel 2010). A new study by Lee et al. (2013b) based on hundreds of thousands of SDSS stars with medium-resolution spectra finds similar results. The CEMP fraction for red giants in the halo is 31% ± 4% at [Fe/H] ⩽ −3.0 and 33% ± 11% at [Fe/H] ⩽ −3.5. The Segue 1 measurements suggest that the tendency of the most metal-poor stars to show carbon enhancement is not limited to the halo, but may also be a general feature of dwarf galaxies.

A possible explanation for this early presence of large amounts of carbon could be the existence of rotating massive stars. They could have provided large amounts of CNO elements during their evolution and/or supernova explosion (e.g., Meynet et al. 2006). Moreover, low-mass star formation may be facilitated in carbon-rich environments because carbon (together with oxygen⁸) may have provided a cooling channel for the primordial gas to sufficiently fragment (Bromm & Loeb 2003; Frebel et al. 2007a). If so, the carbon abundance of the entire system would need to be above the critical metallicity for the observed stars to have formed. Indeed, [C/H] > −3.0 for all stars, which is well above D_trans = log (10^[C/H] + 0.9 × 10^[O/H]) > −3.5 (Frebel et al. 2007a).

An alternative explanation for the carbon enhancement would be mass transfer from an unseen intermediate-mass binary companion that went through the asymptotic giant branch (AGB) phase. However, at least those CEMP stars that have been shown to be in binary systems also show significant amounts of s-process material and only occur at metallicities of [Fe/H] ≳ −3.0 (e.g., Masseron et al. 2010). The three carbon-enhanced EMP stars in Segue 1 are different (i.e., they can be classified as CEMP-no stars) and have [Fe/H] < −3.5. For those stars, the binary fraction is currently unknown (Masseron et al. 2010). Future radial velocity monitoring of CEMP stars will reveal whether the binary scenario could explain carbon enhancement at [Fe/H] < −3.5.

Following Frebel & Bromm (2012), an early galaxy should show large abundance spreads in terms of [X/H], as a result of inhomogeneous mixing of supernova metal yields (e.g., Greif et al. 2010). For [Fe/H], Segue 1 shows a >2 dex spread. For [C/H] this is also nearly 2 dex (again, excluding SDSS J100714+160154). The stars with higher [Fe/H], however, are not CEMP stars by the current definition because their high [Fe/H] keeps the [C/Fe] ratio down. In considering this criterion for an entire galaxy, it also becomes clear that the definition for stellar carbon enhancement, being based on the [C/Fe] ratio, may be an insufficient description. Regardless, assuming inhomogeneous mixing throughout Segue 1 resulting in [X/H] spreads, and keeping in mind that the carbon enrichment processes were likely completely different and/or decoupled from that of iron, the important conclusion is that all stars show a [C/H] abundance in excess of the critical amount, independent of their [Fe/H]. Coming back to the large fraction of CEMP stars observed in the halo, it is plausible that the halo CEMP stars (and some more metal-rich stars, although they would be very difficult to identify) formed in environments similar to that of Segue 1.

Abundances of the α-elements magnesium, calcium, and titanium were obtained from equivalent width measurements. We used spectrum synthesis to determine the silicon abundance. All four α-element abundances are enhanced at the [α/Fe] ∼ 0.5 dex level, even in the more metal-rich stars including SDSS J100714+160154. Figure 5 shows the abundance trends of all four α-elements, in comparison with various halo and dwarf galaxy stars. In Figure 6, we also show the comparison of all our elemental abundances specifically with those of the EMP star sample from Cayrel et al. (2004). Generally, the agreement between the samples is excellent.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Despite the broad agreement between the abundance patterns in Segue 1 and halo stars, careful examination of Figure 5 reveals that the Segue 1 stars have ∼0.1 dex higher abundances in each of the α-elements. However, this could be due to differences in surface gravity, log gf values, and stellar parameter determinations. Similar behavior was found by Hollek et al. (2011), who traced higher Mg abundances to using systematically lower gravities than other studies. We tested this possibility by independently deriving abundances for five stars in the Cayrel et al. sample (HD 2796, HD 122563,BD $-18\deg$ 5550, CS22892-052, CS31082-001), using their published equivalent widths and our log gf values and methods of determining stellar parameters. A stellar parameter comparison between studies can be found in (Frebel et al. 2013a). We find that using our technique produces no significant differences between our Segue 1 log (X) abundances and the respective published Cayrel et al. abundances. The mean difference between the two studies for each element (O, Na, Mg, Al, Si, Ca, Sc, Ti i, Ti ii, Cr, Mn, Fe i, Fe ii, Co, Ni, Zn) ranges from ∼0.00 to ∼0.1 dex with respective standard errors of the mean about as large or larger than the mean values. There are only two exceptions where the differences are more significant; there is an abundance difference for Na of 0.19 ± 0.04 and 0.11 ± 0.015 for Mg. Indeed, our Mg abundance are systematically higher by 0.1 dex owing to slightly lower surface gravities. The other α-elements Si, Ca, and Ti, however, are not gravity sensitive and yield essentially identical abundances. Given the good agreement with the Cayrel et al. stars, we conclude that the Segue 1 α-elements show the same behavior as what is found in typical metal-poor halo stars with [Fe/H] ≲ −1.5, which is consistent with massive stars having enriched Segue 1 at the earliest times.

Another noteworthy issue is that the EMP star Segue 1-7 has somewhat higher abundance ratios than the other Segue 1 stars. Our re-analysis of the published equivalent widths from Norris et al. (2010a) yielded slightly lower values, but Segue 1-7 remains with marginally enhanced α-abundances relative to both the halo sample and the rest of Segue 1. We also find that two of the three Segue 1 EMP stars, SDSS J100639+160008 and Segue 1-7, show an enhancement in silicon of [Si/Fe] = 1.2 and 0.8, respectively, which is well above the halo-typical α enhancement. The SDSS J100639+160008 Si abundance was derived from the line at 3905 Å, which is nearly saturated. The line is known to be blended with a CH feature, but given the strength of the line, the contribution of CH to the observed absorption is a minor issue. The Si abundance of Segue 1-7 is based on the 4102 Å line (Norris et al. 2010a) and somewhat uncertain. The similarly high Ca and Mg abundances of this star lend support to the Si measurement.

Interestingly, though, there are other metal-poor halo stars with [Fe/H] < −2.5 that exhibit similar behavior to these two stars. We find 12 stars in the Frebel (2010) database of metal-poor stars that have [Si/Fe] > 0.6 and [Si/Mg] > 0.4 (four stars from Cayrel et al. 2004, two stars each from McWilliam et al. 1995, Aoki et al. 2002b and Lai et al. 2008, and one star each from Preston et al. 2006 and Frebel et al. 2007b). We use the [Si/Mg] ratio to select stars that are not enhanced both in Si and Mg, as those exist as well. In addition to these 12 stars, S1020549, a star with [Fe/H] = −3.8 in Sculptor (Frebel et al. 2010a) and one in Ursa Major II (Frebel et al. 2010b) also show Si abundances of [Si/Fe] ∼ 1.0 (although the Ursa Major II star also has [Mg/Fe] ∼ 0.7).

These results demonstrate that high Si abundances (with and without significant Mg overabundances) are a rather ubiquitous feature indicating massive star enrichment driven by supernova explosions that are perhaps sensitive to progenitor properties (Aoki et al. 2002a). Indeed, Si is made in both hydrostatic and explosive oxygen burning processes. A different fraction of the Si produced during stellar evolution may survive to the supernova explosion (Woosley & Weaver 1995) depending on the stellar mass as well as on the uncertain values for the ¹²C(α,γ)¹⁶O reaction rate. Hence, this could explain the relatively large spread in the observed abundances as simply resulting from different stellar progenitor properties.

Considering all of the Segue 1 stars, what appears most striking is that overall, the [α/Fe] ratios are all enhanced and nearly identical. Moreover, there is no evolution evident with metallicity, even above [Fe/H] > −2. The same behavior was found by Vargas et al. (2013), who used medium-resolution spectra to determine α-abundances for stars in many ultra-faint dwarf galaxies, including Segue 1. The absence of a decline in [α/Fe] with [Fe/H] indicates no significant contribution of Fe by Type Ia supernovae to the gas clouds from which the observed stars formed. Instead, the evidence suggests that Segue 1 was enriched by massive stars alone. It is particularly notable that the highest metallicity star in the galaxy (with [Fe/H] = −1.4) has an average [α/Fe] = 0.43, consistent with massive star enrichment. For comparison, this halo-like, massive-progenitor-star α enhancement is not found among metal-poor stars with −2.5 ≲ [Fe/H] ≲ −1.5 in the classical dwarf galaxies. Those stars show lower [α/Fe] values close to the solar value, indicating a slower enrichment timescale of their respective hosts and a chemically "earlier" contribution (i.e., at lower metallicity of the system) of iron by Type Ia supernovae. We therefore conclude that chemical evolution in Segue 1 proceeded differently, or was truncated, in Segue 1 compared to both the classical dSphs and most other ultra-faint dwarfs.

Various abundances of other lighter and iron-peak elements were also determined for our Segue 1 stars. Overall, there is very good agreement with the respective abundances of halo stars of similar metallicities, as can be seen in Figure 6. Below, we briefly comment on each element.

The abundances derived from the Na D lines are known to be significantly different in a non-LTE analysis, resulting in large abundance corrections of 0.5 dex for giants (Baumueller et al. 1998). We note that we have not corrected our abundances since the literature abundances are all in LTE, and because our main focus is on an abundance comparison between star samples rather than absolute values. The sodium abundance ratios in our Segue 1 stars are about that of the Sun, within ±0.3 dex. However, Segue 1-7 has an even higher abundance, increasing the range from [Na/Fe] = 0.54 dex to 0.87 dex. In halo stars, aluminum is often deficient by almost 1 dex with respect to iron and the Sun. Our stars are no exception, with a small scatter of [Al/Fe] = 0.3 dex. However, as with Na, Segue 1-7's abundance is unusually high, although some halo stars have similar abundances.

A [Na/Fe] spread larger than that of the [Na/Al] ratio might be a sign of AGB nucleosynthesis and associated chemical enrichment as is found in globular clusters. We thus discuss Na and Al together to assess this possibility. Not considering Segue 1-7, the spread of [Na/Fe] and [Na/Al] are both about 0.5 dex, which is typical for halo stars. When including Segue 1-7, the [Al/Fe] spread does indeed become ∼0.2 dex larger than the [Na/Fe] spread. However, globular clusters typically have high Al abundances of [Al/Fe] > 0.4 (e.g., Yong et al. 2005; Carretta et al. 2012). None of the seven Segue 1 stars shares this trait. An enrichment similar to what is occurring within globular clusters can thus be ruled out. Instead, the signature of Segue 1-7 of high C, Na, Mg, and Al abundances—a pattern that has been found for some other metal-poor halo stars also (Aoki et al. 2002a)—perhaps rather signals a particular kind of supernova explosion as the source of early metals in Segue 1.

Scandium abundances show a tight trend in halo stars, and all Segue 1 abundances agree extremely well with them. Chromium, manganese, cobalt, and nickel are all very similar in this regard, and again there is excellent agreement of all Segue 1 abundances with the respective halo ratios. Since Mn was difficult to obtain for SDSS J100652+160235, we can only present an upper limit. Segue 1-7 has an unusually low Ni abundance. However, it was only determined from one line (Norris et al. 2010a). Finally, for completeness, we note that a zinc abundance could only be determined for the most metal-rich Segue 1 star. We show upper limits in Figure 6 for all other stars.

An emerging abundance characteristic of stars in the ultra-faint dwarf galaxies that have been observed at high spectral resolution is that they all display extremely low levels ([n − cap/Fe] ≪ 0.0) of neutron-capture elements (e.g., Koch et al. 2008; Frebel et al. 2010b; Simon et al. 2010; Norris et al. 2010a, 2010c; François et al. 2012; Koch et al. 2013). Segue 1 is no exception, having the lowest overall neutron-capture abundance level of any known system.

We attempted to measure abundances from the Sr ii line at 4077 Å and the Ba ii line at 4554 Å. There is a weak Ba detection in SDSS J100742+160106 and clearly visible Sr and Ba lines in SDSS J100714+160154 (which is the most metal-rich star and discussed separately below). Segue 1-7 also has a Sr detection (Norris et al. 2010a). For all other stars we could only derive upper limits on these two elements. Portions of the spectra around the Ba line can be seen in Figure 1 (left panel). This line is generally easily detected in metal-poor stars, even in noisy spectra, and especially in the stars with [Fe/H] > −2.5. Hence, it is rather unusual that it is not detected in most stars in Segue 1. For example, SDSS J100710+160623 has very low upper limits of [Sr/Fe] < −3.2 and [Ba/Fe] < −2.6. These are the lowest [Sr/Fe] and [Ba/Fe] values ever determined (see top panels in Figure 7). However, when normalizing to hydrogen rather than iron to assess the overall level of enrichment in Segue 1, the upper limits of SDSS J100710+160623 are at the same level as those of the other stars (bottom panel in Figure 7), [Sr/H] ≲ −5 and [Ba/H] ≲ −4.4.

Zoom In Zoom Out Reset image size

More specific clues regarding the level at which neutron-capture elements are present in this galaxy come from Segue 1-7 (Norris et al. 2010a) which has a measured Sr abundance of [Sr/H] = −4.9, and from SDSS J100742+160106 which has a weak Ba measurement of [Ba/H] = −4.5. However, the Ba line in Segue 1-7 is not detected, resulting in a limit of [Ba/H] < −4.5. Similarly, SDSS J100742+160106 has an upper limit for Sr of [Sr/H] < −5.2. The two detections, together with the upper limits of the other stars confirm that Segue 1, as a system, is extremely deficient in neutron-capture elements, at or below the 10⁻⁵ level in [Sr/H] and 10^−4.5 level in [Ba/H]. Nevertheless, it is apparent from the handful of Sr and Ba detections that there is neutron-capture material present in Segue 1, as is the case in every other known stellar system (Roederer 2013).

In Figure 7, the yellow shaded area indicates the upper enrichment level of Sr, bounded by Segue 1-7 at the high end. The same is shown for [Ba/H] with the measurement of SDSS J100742+160106 as the upper bound. It is interesting to note, though, that a handful of EMP halo stars have Sr and Ba abundances that are up to 1 dex lower than our upper enrichment bounds, showing that detections of these elements at the [Sr/H] ∼ −6 or [Ba/H] ∼ −5.5 level are possible, likely owing to much higher data quality. No halo stars above [Fe/H] = −2.5 that are so deficient in the heaviest elements have yet been identified, demonstrating that systems like the ultra-faint dwarfs cannot have contributed appreciably to the halo in this metallicity range (see Section 5).

According to Simon et al. (2011), Segue 1 has a mass within its half light radius of ∼6 × 10⁵ M_☉. Its stellar mass, however, is only ∼10³ M_☉ (Martin et al. 2008). Since a typical star formation efficiency is 1% per free fall time (Krumholz & Tan 2007), even if star formation in Segue 1 only lasted for a single free fall time, the initial gas cloud out of which the Segue 1 stars formed was likely ∼10⁵ M_☉. Larger gas masses would exceed the cosmic baryon fraction at the center of Segue 1, although if one considers the halo mass of the galaxy larger amounts of baryons could be allowed. Assuming that the gas comprising Segue 1 was (instantaneously and homogeneously) enriched by a first supernova at the earliest times, an average abundance of [Sr/H] = −5.0 (roughly that of Segue 1-7) would imply a total Sr mass of only ∼10⁻⁷ M_☉. However, the average Sr abundance was likely less than that of Segue 1-7 (as indicated by our lowest upper limit of [Sr/H] > −5.2), which would decrease the required total Sr mass present at Segue 1 at early times. Hence, we regard ∼10⁻⁷ M_☉ of Sr as a reasonable estimate. In the same way, less than ∼10⁻⁷ M_☉ of Ba appears to be present in Segue 1, assuming [Ba/H] < −4.5.

These extremely small amounts of neutron-capture elements could have come from just a single neutron-capture element production event. Possibilities include the r-process occurring in a core-collapse supernova, yielding of the order of 10⁻⁴ M_☉ of neutron-capture elements (Farouqi et al. 2010) or the s-process that may have operated in a massive rotating star, perhaps a Population III star (e.g., Chiappini et al. 2011; Pignatari et al. 2008, and see Jacobson & Frebel 2014 for a review on neutron-capture element sources). An AGB enrichment event is unlikely as a source for this small amount of neutron-capture material. Models of AGB nucleosynthesis generally suggest much larger amounts to be produced (Karakas 2010; Lugaro et al. 2012). For example, the low-metallicity stellar model described in Placco et al. (2013) loses 0.5 M_☉ during its AGB phase. This is orders of magnitude more than what is implied by the observed low Sr and Ba abundances in the Segue 1 stars.

More detailed yield calculations, paired with simulations of metal-mixing in such small systems, are required to fully characterize the neutron-capture enrichment process in systems like Segue 1. However, the fact that Segue 1 has such a uniformly low overall neutron-capture abundance suggests that the halo stars with equally low [Sr/H] and [Ba/H] may have originated in similar systems before being accreted by the Milky Way.

To complete the discussion on the neutron-capture elements, we also have to revisit SDSS J100714+160154. At [Fe/H] = −1.4 it has [Sr/Fe] = +0.90 and [Ba/Fe] = +1.85. These high values are difficult to reconcile with those of the other Segue 1 stars unless SDSS J100714+160154 has a different enrichment history. We suggest that SDSS J100714+160154 is likely a CH star in a binary system that underwent mass transfer. General CH star characteristics are moderate metallicity around [Fe/H] ∼ −1.5, high carbon abundance in combination with s-process enrichment, and radial velocity variation. The radial velocity measured from our MIKE spectrum agrees within the uncertainties with those determined by Simon et al. (2011), so binary amplitudes larger than ∼10 km s⁻¹ are unlikely. Despite the lack of evidence for binary orbital motion, SDSS J100714+160154 does have [Fe/H] ∼ −1.4, [C/Fe] = 1.4, and also [Pb/Fe] ∼ 2.0. High lead abundances are a signature of the s-process, and together with the overabundance in carbon, SDSS J100714+160154's abundance pattern is clearly consistent with that of a CH star. In Figure 8, we show the few measurable neutron-capture element abundances in SDSS J100714+160154 compared to the solar r- and s-process patterns (Burris et al. 2000). While the number of neutron-capture element abundance is limited, the pattern matches that of the solar s-process. Hence, the star's neutron-capture material is not a reflection of the birth gas cloud composition but is due to a later-time external enrichment event following mass transfer from the binary companion. While our sample is too small for a statistical analysis and also lacks long-term radial velocity monitoring, finding one binary in a sample of seven stars suggests that the binary fraction is substantial, consistent with the results of Martinez et al. (2011). Binary stars, and thus mass transfer events, are very common in general, so it is perhaps not too surprising that a star in our sample shows these characteristics.

Zoom In Zoom Out Reset image size

Given that we use our abundances to constrain a number of processes related to early star formation, chemical enrichment, and metal mixing, it is important to carefully consider any abundance uncertainties. Using a formula in Frebel et al. (2006a) to estimate the uncertainties on our equivalent width measurements (taking into account the S/N of the spectrum, pixel width and number of pixels across a spectral line), we estimate our measurements to have 3σ uncertainties of 10%–15%. This corresponds to up to 0.15 to 0.2 dex (and in the case of SDSS J100742+160106 to up to ∼0.3 dex) changes in abundances. These uncertainties are reflected in the standard deviations of our line abundances for each element (see Table 2) which range from 0.10 to 0.25 dex, and are listed in Table 5. We also derived systematic uncertainties of our abundances for two example stars by changing one stellar parameter at a time by an amount approximately equal to its random uncertainty. Uncertainties from different sources of random errors (standard deviations of the line abundances for a given element), systematic errors, and the total uncertainty for each element are given in Table 5. Overall, typical total uncertainties range from 0.15 to 0.30 dex.

Many abundances are based on only a few lines so that the associated standard errors are unrealistically small (<0.05 dex) compared with typical measurement uncertainties. We thus adopt a minimum uncertainty of 0.10 dex in such cases. Furthermore, for abundances of elements derived from only one line, we adopt a formal uncertainty of 0.15 dex, which should reasonably reflect our largest source of error, continuum placement uncertainties. Despite having two measurements for C, we adopt 0.20 dex for this element since the continuum placement was difficult for spectral regions with molecular bands. Finally, uncertainties on iron abundances are also somewhat larger. This is due to our method of correcting the spectroscopic temperatures, which introduces a slope of line abundance as a function of excitation potential and thus inflates the standard deviation.