Chemical Abundance Analysis of Tucana III, the Second r-process Enhanced Ultra-faint Dwarf Galaxy^*

Radial velocities for each star were measured by comparing the program star with a radial velocity standard star (HD 136202) observed on the first night of the run. Radial velocities were derived via cross-correlation of each order of the program star spectrum with the corresponding order of the standard star spectrum. The blue and the red arms of the spectrograph were considered independently. The mean values of the resulting relative velocities from each arm, for 28 orders in the blue and 6 orders in the red, with 3σ outliers rejected, e.g., ${\mathrm{RV}}_{\mathrm{blue}}=\sum ({v}_{n,\mathrm{blue}})/N$, where v_n,blue is the velocity derived from each order of the blue arm and N is the total number of blue orders. These velocities were then averaged to form the final reported velocity for each star: RV_final = (RV_blue + RV_red)/2. Errors were derived using the standard deviation of the individual velocities determined from each order, again considering the blue and red arms separately. The reported error on each velocity is calculated as ${\sigma }_{{\mathrm{RV}}_{\mathrm{final}}}={({\sigma }_{{\mathrm{RV}}_{\mathrm{blue}}}^{2}/2+{\sigma }_{{\mathrm{RV}}_{\mathrm{red}}}^{2}/2)}^{1/2}$. Measured radial velocities for all four stars are presented in Table 2. The radial velocity measurements were used to place each program star spectrum on a wavelength scale associated with rest wavelengths. These velocities can also be used to investigate whether the stars are in fact binaries, as discussed in Section 5.3.

Stellar parameters for the four stars were derived spectroscopically, following the method described by Hansen et al. (2017). In brief, we used the 2017 version of the MOOG spectral synthesis program (Sneden 1973), making the assumption of local thermodynamic equilibrium and including Rayleigh scattering treatment as described by Sobeck et al. (2011).⁴¹ Initial effective temperatures were determined from excitation equilibrium of Fe i lines and thereafter placed on a photometric scale using the relation from Frebel et al. (2013). Following this the surface gravities (log g) were determined from ionization equilibrium between the Fe i and Fe ii lines. Finally microturbulent velocities (ξ) were determined by removing any trend in line abundances with reduced equivalent widths for the Fe i lines. For the four stars studied here, J235738, J235550, J000549, and J234351, we were able to use 103, 122, 157, and 90 Fe i lines and 15, 13, 20, and 16 Fe ii lines, respectively, for this analysis. Final stellar parameters are presented in Table 2 and the lines used for the analysis of each star are listed in Table 3.

Table 3.  Fe i and Fe ii Lines Used for Parameter Determination

Stellar ID	Species	λ	χ	log gf	EW	log
		(Å)	(eV)		(mÅ)
J000549-593406	Fe i	3765.54	3.240	0.480	78.63	4.638
J000549-593406	Fe i	3790.09	0.989	−1.740	106.00	4.988
J000549-593406	Fe i	3805.34	3.300	0.310	79.81	4.913
J000549-593406	Fe i	3807.54	2.221	−0.990	80.07	4.934
J000549-593406	Fe i	3876.04	1.010	−2.890	74.29	5.158
J000549-593406	Fe i	3917.18	0.989	−2.150	101.40	5.122
J000549-593406	Fe i	4067.98	3.209	−0.530	41.48	4.630
⋮	⋮	⋮	⋮	⋮	⋮	⋮
J000549-593406	Fe ii	4233.17	2.580	−1.810	81.79	4.746
J000549-593406	Fe ii	4416.83	2.780	−2.410	57.10	4.979
J000549-593406	Fe ii	4489.18	2.828	−2.971	22.98	4.869
J000549-593406	Fe ii	4491.41	2.850	−2.640	28.78	4.702

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

Abundances were derived from equivalent width measurements and spectral synthesis using MOOG. We used α-enhanced ([α/Fe] = +0.4) 1D LTE ATLAS9 model atmospheres (Castelli & Kurucz 2003) and the solar photosphere abundances from Asplund et al. (2009). Line lists were generated using the linemake package⁴² (C. Sneden 2019, private communication), including molecular lines for CH, C₂, and CN and isotopic shift and hyperfine structure information. Measured stellar abundances are presented in Table 4.

Table 4.  Measured Abundances

	J000549				J234351				J235550				J235738
Element	[X/Fe]	log (X)	σ	N	[X/Fe]	log (X)	σ	N	[X/Fe]	log (X)	σ	N	[X/Fe]	log (X)	σ	N
C	−0.43	5.39	⋯		−0.28	5.46	⋯		−0.19	5.55	⋯		−0.05	5.80	⋯
N	<2.00	<7.22	⋯		<1.5	<6.64	⋯		<1.0	<6.14	⋯		<0.5	<5.75	⋯
NaI	⋯	⋯	⋯		0.32	3.87	0.35	2	⋯	⋯	⋯		0.53	4.19	0.33	2
MgI	0.45	5.44	0.29	9	0.50	5.41	0.37	7	0.40	5.31	0.27	8	0.64	5.66	0.31	5
AlI	−0.36	3.48	0.39	2	−0.99	2.78	0.36	2	⋯	⋯	⋯	2	−0.35	3.52	⋯	1
SiI	0.39	5.29	0.33	2	0.18	5.00	0.33	2	0.39	5.21	0.31	2	0.00	4.93	⋯	1
KI	0.71	3.13	0.25	2	0.41	2.75	⋯	1	0.71	3.05	⋯	1	0.56	3.01	0.31	2
CaI	0.40	4.13	0.28	19	0.35	4.00	0.28	11	0.44	4.09	0.28	13	0.46	4.22	0.28	13
ScII	0.15	0.69	0.28	5	0.09	0.55	0.27	4	−0.01	0.45	0.28	6	0.05	0.62	0.30	6
TiI	0.07	2.41	0.31	18	0.17	2.43	0.34	7	0.15	2.41	0.36	12	0.28	2.65	0.32	8
TiII	0.46	2.80	0.29	39	0.31	2.57	0.29	11	0.50	2.76	0.31	32	0.41	2.78	0.32	26
VI	−0.16	1.16	⋯	1	0.06	1.30	⋯	1	−0.18	6.44	⋯	1	−0.13	1.22	⋯	1
VII	0.07	1.39	0.23	2	0.13	1.37	0.21	2	0.10	6.72	0.25	2	0.00	1.35	⋯	1
CrI	−0.38	2.65	0.32	14	−0.34	2.61	0.33	7	−0.26	2.69	0.34	9	−0.17	2.89	0.34	6
CrII	0.08	3.11	0.23	2	0.29	3.24	0.30	2	0.35	3.30	0.20	2	0.24	3.30	⋯	1
MnI	−0.50	2.32	0.24	6	−0.42	2.32	0.25	5	−0.50	2.25	0.23	6	−0.39	2.46	0.23	4
FeI	−2.61	4.89	0.17	157	−2.69	4.81	0.20	90	−2.69	4.81	0.21	122	−2.58	4.92	0.20	103
FeII	−2.62	4.88	0.18	20	−2.70	4.80	0.17	16	−2.69	4.81	0.08	13	−2.59	4.91	0.14	15
CoI	0.03	2.41	0.30	3	0.14	2.44	0.29	3	−0.04	2.27	0.29	2	⋯	⋯	⋯
NiI	0.00	3.61	0.31	11	0.13	3.66	0.32	5	0.22	3.75	0.43	5	0.04	3.68	0.33	6
CuI	<−0.5	<1.28	⋯	1	<0.5	<2.00	⋯	1	<0.0	<1.50	⋯	1	−0.20	1.41	⋯	1
ZnI	−0.08	1.87	⋯	1	0.33	2.20	0.18	2	0.19	2.06	⋯	1	0.22	2.20	0.22	2
SrII	0.09	0.35	0.33	2	−0.26	−0.08	0.32	2	0.04	0.22	0.31	2	−0.03	0.27	0.31	2
YII	−0.20	−0.60	0.26	5	<−0.3	<−0.78	⋯		−0.28	−0.76	0.22	3	−0.05	−0.42	0.25	2
ZrII	0.05	0.02	0.26	3	0.00	−0.11	⋯	1	−0.04	−0.15	0.27	2	0.29	0.29	0.26	2
BaII	0.11	−0.32	0.32	4	−0.20	−0.71	0.38	4	−0.05	−0.56	0.24	4	0.14	−0.26	0.34	4
LaII	0.21	−1.30	0.23	2	0.25	−1.34	⋯	1	0.13	−1.46	0.23	2	0.16	−1.32	0.26	2
CeII	0.32	−0.71	0.29	4	<1.0	<−0.11	⋯		<0.30	<−0.81	⋯		<0.3	<−0.70	⋯
PrII	0.61	−1.28	⋯		<1.0	<−0.97	⋯		<0.70	<−1.27	⋯		0.50	−1.36	0.23	3
NdII	0.27	−0.92	0.24	8	0.69	−0.58	0.23	3	0.39	−0.88	0.33	4	0.33	−0.83	0.29	6
SmII	<0.50	<−1.15	⋯		<0.8	<−0.93	⋯		<0.70	<−1.03	⋯		<0.5	<−1.12	⋯
EuII	0.49	−1.61	0.21	2	0.44	−1.73	0.29	2	0.56	−1.61	0.22	3	0.22	−1.84	0.21	3
GdII	<0.70	<−0.84	⋯		<1.3	<−0.32	⋯		<1.00	<−0.62	⋯		<0.8	<−0.71	⋯
DyII	0.59	−0.92	0.24	3	<0.7	<−0.89	⋯		<1.00	<−0.59	⋯		0.70	−0.78	0.29	3
ErII	0.50	−1.19	⋯	1	<1.0	<−0.77	⋯		<0.60	<−1.17	⋯		⋯	⋯	⋯

Download table as:  ASCIITypeset image

We note here that the effective temperature derived for three of these stars (including J235532) are all equal, which is unexpected given that J235532 is somewhat redder than the other two stars. We have carefully reconsidered the derived effective temperatures for all five stars, including J235532, and find no errors in the analysis. Expected values from the Dartmouth isochrone (shown in Figure 1) suggest a temperature difference of ΔT ∼ 300 K between stars having the colors of J235532 and J235550, which agrees with our derived spectroscopic temperatures within errors.

Uncertainties on the derived abundance for J000549 arising from stellar parameter uncertainties are listed in Table 5. These uncertainties were computed by deriving abundances with different atmospheric models, each with one parameter varied by its uncertainty as given in Table 2 and added in quadrature. As all stars have similar stellar parameters and spectral quality we consider these uncertainties to be applicable to all four stars studied here.

Sample synthetic spectra for absorption lines of Sr, Ba, and Eu can be found in Figure 3, overlaid onto the observed spectra.

Zoom In Zoom Out Reset image size

We derive abundances for 18 non-neutron-capture elements from C to Zn, including a range of α-elements (Mg, Si, Ca, Ti) and iron peak elements (Cr, Mn, Fe, Co, Ni). Figure 4 shows the abundances of 15 of these elements as a function of [Fe/H] compared to stars in other ultra-faint dwarf galaxies and stars in the Milky Way halo. When considering the sample of all five stars, a range of metallicity ([Fe/H]) is observed as well as the expected trend in α-elements, i.e., that lower metallicity stars in the stellar population have higher α abundances because they were formed at a time at which stellar nucleosynthesis was dominated by SNe II, compared to the more metal-rich stars that are formed later after the stellar population has been polluted by SN Ia explosions (Tinsley 1979). This α "knee" is observed in the elements Ca and Ti; see Figure 4.

Zoom In Zoom Out Reset image size

None of the stars can be classified as carbon-enhanced metal-poor stars (CEMP; [C/Fe] > 0.7).

We derive abundances or upper limits for 11 neutron-capture elements from Sr to Er. Figure 5 shows abundances of the neutron-capture elements Sr, Ba, and Eu compared to stars in other ultra-faint dwarf galaxies and stars in the Milky Way halo.

Zoom In Zoom Out Reset image size

Figure 6 shows the neutron-capture element abundance pattern for each of the stars compared to the solar system s- and r-process abundance pattern from Simmerer et al. (2004). For each star the solar pattern has been scaled to the average residual between the star and the Sun for elements with abundances (not upper limits) from Ba to Er. It is clear from Figure 6 that the neutron-capture abundance pattern in four of the stars is better matched by the solar system r-process abundance pattern than the s-process pattern. J234351, with abundances measured for only six neutron-capture elements, is the exception. Neither the solar system r- nor s-process pattern matches the derived abundances for this star well. Note, however, that this star's spectrum has the lowest signal-to-noise of those studied here.

Zoom In Zoom Out Reset image size

We adopt the Hansen et al. (2017) definition of r-process enhanced stars: r-I stars are defined to have 0.3 < [Eu/Fe] < 1 and [Eu/Ba] > 0.4, while r-II stars have [Eu/Fe] > 1 and [Eu/Ba] > 0.4, with the additional constraint on [Eu/Ba] added to the traditional definition of r-process enhancement in order to ensure that the enhancement is entirely due to elemental production via the r-process, and is not confused by contributions from the s-process. Since Eu is nearly entirely produced in the r-process (94%, Koch & Edvardsson 2002) but Ba is produced primarily in the s-process (85% according to Burris et al. 2000), the ratio of Eu/Ba is often used to gauge whether the neutron-capture elements in a given star were produced primarily via the s- or r-process.

Our measurements show that overall Tuc III is moderately enhanced in r-process elements: three of the four stars studied here are r-I stars having 0.3 < [Eu/Fe] < 1, as is J235532, the Tuc III star previously studied by Hansen et al. (2017). Figure 7 shows the [Eu/H] ratios as a function of metallicity for the five Tuc III stars along with r-process enhanced stars in other dwarf galaxies and in the Milky Way halo.

Zoom In Zoom Out Reset image size

Measured radial velocities of the stars studied here are consistent with Simon et al. (2017) and Li et al. (2018a) as shown in Figure 8. We collect all measured radial velocities for the five Tuc III stars in Table 6. We confirm the results of Li et al. (2018a), i.e., that there is a significant velocity gradient across the Tuc III system. We also see some evidence for velocity variations in individual stars, discussed in more detail in Section 5.3.

Zoom In Zoom Out Reset image size

Tuc III is the second ultra-faint dwarf galaxy containing multiple stars enhanced in r-process elements. As discussed in Section 1, the first such galaxy, Ret II, is even more highly enhanced than Tuc III, i.e., many of the Ret II stars are r-II stars (see Figure 7). The fact that so many stars in a galaxy as small as Ret II share a common chemical pattern suggests that a single nucleosynthetic event must have occurred early in the history of the galaxy, polluting future generations of stars, and that most of the stars in the galaxy were impacted by the event.

The r-II stars in Ret II have an average enhancement in Eu of [Eu/H] ∼ −1. Ji et al. (2016a) used this level and an estimated dilution gas mass, i.e., the mass of hydrogen gas that the r-process material is diluted into, of Ret II of 10⁶ M_⊙ to argue that a binary neutron star merger was the most likely source of the excess of Eu detected in Ret II as other sources would not have produced enough Eu to enhance the galaxy to the level detected. In Tuc III we find an average enhancement in Eu of [Eu/H] ∼ −2. Assuming the same binary neutron star merger Eu ejecta mass as in Ji et al. (2016a), 10^−4.5 M_⊙, leads to a dilution gas mass of ∼2 × 10⁶ M_⊙, twice that of Ret II. If the r-process elements were produced via the same mechanism in Ret II and Tuc III, then either Tuc III had twice the gas mass to pollute or half the amount of r-process material. Current data do not distinguish between these possibilities, but we note that the ejecta mass from a neutron star merger could well vary from event to event (Côté et al. 2018). It is also worth noting that the stellar mass of Ret II is (2.6 ± 0.2) × 10³ M_⊙ (Bechtol et al. 2015), that of Tuc III is (0.8 ± 0.1) × 10³ M_⊙ (Drlica-Wagner et al. 2015), and that of the Tuc III stream is 3.8 × 10³ M_⊙ (Shipp et al. 2018), so many possible dilution scenarios are plausible.

Sakari et al. (2018b) show that r-process enhanced stars in the Milky Way halo at a range of metallicities have nearly identical r-process patterns, matching the solar system r-process pattern, regardless of the level of r-process enhancement. Hansen et al. (2017) also found this to be true for the r-process enhanced stars detected in classical and ultra-faint dwarf galaxies. The implication of this result is that there must either be a single mechanism for r-process element production, or else every r-process production mechanism must produce identical abundance patterns. If the former, the most likely site of r-process enhancement is binary neutron star mergers, based on observational evidence to date. Furthermore, in galaxies with masses as low as the ultra-faint dwarfs discussed here, the star formation history must have proceeded in such a way that there was likely only one enrichment event, early in the history of the galaxy (e.g., Ojima et al. 2018). The r-process enhancement of Tuc III is consistent with this result, although with a different r-process element or dilution gas mass than in the case of Ret II.

Hansen et al. (2017) noted that with a sample of one star it is difficult to determine whether the source of enhancement in this galaxy is inside or outside the galaxy. With this larger sample of stars in both the core and the tidal tails, we can now claim that the abundance pattern of Tuc III, like that of Ret II, must be due to an enhancement event inside the galaxy. Since it appears that stars throughout the galaxy show similar levels of r-process enhancement, we can further state that this enhancement event must have occurred early in the history of the galaxy, thereby polluting the galaxy on large scales.

Safarzadeh & Scannapieco (2017) used an adaptive mesh refinement cosmological simulation to show that the exact location of an enrichment event in small ultra-faint dwarf galaxies such as Ret II can have a large impact on the enrichment of all stars in the galaxy. In the case of Ret II, they compared the results of a binary neutron star merger located at the center of the galaxy compared to on the outskirts of the galaxy to show that the high levels of r-process enhancement seen in Ret II can only be explained if the event occurred very close to the center of the galaxy, and at a time at which the stars were still being formed. Another implication of the work of Safarzadeh & Scannapieco (2017) is that a galaxy with lower levels of r-process enhancement, such as Tuc III, may have experienced a similar nucleosynthetic event, but that it was located at the edges of the galaxy. Such events occurring on the edges of the galaxy are not particularly unexpected, given the "kicks" binary neutron stars experience. Now that multiple stars in Tuc III have been shown to be moderately enhanced in r-process elements, a more detailed comparison can be made to this theoretical work.

Both Simon et al. (2017) and Li et al. (2018a) considered in some detail the nature of Tuc III, because the low measured velocity dispersion (${0.1}_{-0.1}^{+0.7}$ km s⁻¹) leads to speculation as to whether Tuc III was truly an ultra-faint dwarf galaxy or rather a globular cluster at birth. A possible reason for the low measured velocity dispersion, as discussed by Simon et al. (2017), is that stripping of the stars has lowered the velocity dispersion as Tuc III merges with the Galaxy. Since the velocity dispersion may not clearly determine the nature of Tuc III, we consider here several chemical aspects of the stellar population that may shed light on its origin.

The four stars studied here have a mean metallicity of [Fe/H] ∼ −2.64 ± 0.15, significantly lower than the star studied by Hansen et al. (2017) ([Fe/H] ∼ −2.25). Figure 9 compares two of these spectra and demonstrates that Tuc III is not a monometallic system. Confirmation of Tuc III's metallicity dispersion using the five stars studied here provides further evidence that Tuc III is in fact a galaxy, because only galaxies, and not globular clusters, have gravitational wells deep enough to retain supernova ejecta, enabling the production of multiple generations of stars. The resulting range of metallicities observed in galaxies is the natural result of this extended formation (Tinsley 1979; Willman & Strader 2012). We note that if we were to use the photometric temperature for DES J235532 we would derive a ∼0.2 dex lower metallicity for this star. Consequently, using the photometric temperatures to consider the metallicities of these stars would result in no statistically significant metallicity range between the stars, weakening the evidence that Tuc III is a galaxy and not a globular cluster.

Zoom In Zoom Out Reset image size

We further investigate the nature of Tuc III by considering the abundances of elements involved in proton-capture reactions, specifically Mg and Al. The Mg–Al anticorrelation that is observed in globular clusters is thought to be produced via pollution of second generation stars in the cluster by massive asymptotic giant branch stars at the end of their lives, particularly in massive or very metal-poor clusters (e.g., Carretta et al. 2009). Figure 10 compares the Mg and Al abundances of the Tuc III stars to red giant stars in four globular clusters that have been shown to have a strong Mg–Al anticorrelation. The Tuc III stars studied here do not exhibit the proton burning trend.

Zoom In Zoom Out Reset image size

Finally, while it is true that globular clusters generally have neutron-capture enhancement similar to that observed here (and very different from other ultra-faint dwarf galaxies, see Ji et al. 2018, for example), we do not feel that this commonality is enough to claim that Tuc III is more likely to be a globular cluster. Furthermore, Tuc III's average metallicity, [Fe/H] ∼ −2.49 (Li et al. 2018a), would place it on the extreme low end of the distribution of globular cluster metallicities: the lowest metallicity globular cluster in the Harris (1996, 2010) catalog is NGC 7078 with [Fe/H] = −2.37. We therefore conclude that Tuc III is most likely an ultra-faint dwarf galaxy and not a globular cluster.

Since all five of our stars were studied using multiple observations over a span of two years, we can use multiepoch radial velocities to search for reflex motion due to the stars being in an undetected binary system. In Figure 8 we compare the radial velocities measured in this work with those measured by Simon et al. (2017) and Li et al. (2018a). We see very good agreement between the measurements of three of the stars: DES J235532, DES J235550, and DES J000549 do not appear have variable radial velocities according to these measurements and are therefore unlikely to be binaries with periods ≲1 yr. Two other stars may have unseen binary companions: DES J234351 shows large radial velocity variation, and DES J235738 may also show small velocity variation.

DES J234351 is identified as a binary by Li et al. (2018a) as well, who note a Δv ∼ 20 km s⁻¹ and exclude the star from further analysis of Tuc III's kinematics. Our higher precision velocities confirm the velocity shift between observations made in 2016 and 2017 and indicate that DES J234351 is very likely a binary system.

DES J235738 shows potential velocity variation with amplitude of ≥2.79 km s⁻¹. This (weakly) suggests that this star may in fact be a binary and warrants further kinematic measurements. If DES J235738 is shown to be in a binary system, mass transfer from its companion could potentially explain its chemistry as well, because this star appears to have some s-process enhancement in addition to the r-process enhancement shared with the other stars, although a higher S/N spectrum is needed to confirm this suggestion.

These are not the first binary stars discovered in an ultra-faint Milky Way satellite galaxy. Koch et al. (2014) measured velocities of one star in the Hercules dwarf galaxy over a two year baseline and concluded that it was in fact a binary system having a 135 day period, composed of a giant with a low-mass companion, likely a white dwarf. Despite the fact that mass transfer binaries can explain peculiar abundance patterns in some cases, no such binary scenario could be described by Koch et al. (2014) in the case of Hercules. Conversely, a binary star in Segue 1 does show signs of mass transfer (Frebel et al. 2014) through its high carbon abundance. Binary stars have also been detected via variable radial velocity signatures in Boo II (Ji et al. 2016d), Tri II (Kirby et al. 2017; Venn et al. 2017), Carina II (Li et al. 2018b), and Ret II (Simon et al. 2015; Minor et al. 2019).