Phase-space Properties and Chemistry of the Sagittarius Stellar Stream Down to the Extremely Metal-poor ([Fe/H] ≲ −3) Regime

SEGUE's emphasis on the distant halo is suitable for studying Sgr (and other streams; Koposov et al. 2010; Newberg et al. 2010) and has, indeed, been extensively explored for this purpose (Yanny et al. 2009a; Belokurov et al. 2014; de Boer et al. 2014, 2015; Gibbons et al. 2017; Chandra et al. 2022; Thomas & Battaglia 2022). The novelty is the availability of complete phase-space information thanks to Gaia. Hence, we are in a position to construct a larger sample of confident Sgr stream members than previous efforts.

Stellar atmospheric parameters, namely, effective temperatures (T_eff), surface gravity ($\mathrm{log}g$), and metallicities (in the form of [Fe/H]), as well as α-element abundances ([α/Fe]), and v_los values for SEGUE stars were obtained via application of the SSPP ¹⁶ routines. The final run of the SSPP to SEGUE spectra was presented alongside the ninth data release (DR9) of SDSS (Ahn et al. 2012) and has been included, unchanged, in all subsequent DRs. Recently, Rockosi et al. (2022) reevaluated the internal precision of SSPP's atmospheric parameters for DR9, which are no worse than 80 K, 0.35 dex, and 0.25 dex for T_eff, $\mathrm{log}g$, and [Fe/H], respectively, across the entire metallicity and color ranges explored. Unless explicitly stated, we consider SEGUE's as our fiducial stellar parameters throughout the remainder of this paper. These also serve as input for the Bayesian isochrone-fitting code StarHorse (Santiago et al. 2016; Queiroz et al. 2018), which, in turn, provides ages and distances.

In this work, we only consider spectra with moderate signal-to-noise ratio (S/N > 20 pixel⁻¹). We keep only those stars within 4500 < T_eff/K < 6500, which is the optimal interval for the performance of SSPP. Moreover, we limit our sample to low-metallicity stars ([Fe/H] ≤ −0.5), which removes most of the contamination from the thin disk, but maintains the majority of Sgr stream members; out of 166 stars analyzed by Hayes et al. (2020), only five (3%) show [Fe/H] > −0.5.

We crossmatch the above-described SEGUE low-metallicity sample with Gaia's early DR3 (EDR3; Gaia Collaboration et al. 2021) using a 15 search radius. In order to ensure the good quality of the data at hand, we only retain those stars whose renormalized unit weight errors are within the recommended range (ruwe ≤ 1.4; Lindegren et al. 2021a). Parallax biases and error inflation are handled following Lindegren et al. (2021b) and Fabricius et al. (2021), taking into account magnitudes, colors, and on-sky positions. Stars with largely negative parallax values (parallax < −5 mas) are discarded. Also, we emphasize that only those stars with an available parallax measurement are considered to ensure good distance results. Those stars with potentially spurious astrometric solutions are also removed (fidelity_v2 < 0.5; Rybizki et al. 2022).

We applied StarHorse to this SEGUE+Gaia EDR3 sample in order to estimate precise distances that would allow us to study the Sgr stream. At ≥10 kpc from the Sun, our derived uncertainties are at the level of ∼13%. The medians of the derived posterior distributions are adopted as our nominal values, while 16th and 84th percentiles are taken as uncertainties. Further details regarding stellar-evolution models and geometric priors can be found in Queiroz et al. (2018, 2020) and Anders et al. (2019). See also Anders et al. (2022) for details regarding the compiled photometric data.

With the StarHorse output at hand, we restrict our sample to stars with moderate (<20%) fractional Gaussian uncertainties in their estimated distance values. We refer to this catalog as the "SEGUE/StarHorse low-metallicity sample" or close variations of that. We refer the interested reader to Perottoni et al. (2022) for an initial application of these data. Its coverage in a Cartesian Galactocentric projection can be appreciated in Figure 1. By color coding this plot with the mean [Fe/H] in spatial bins, the footprint of the Sgr stream is already perceptible as metal-rich trails at ∣Z∣ ≳ 20 kpc.

Zoom In Zoom Out Reset image size

Positions (α, δ) and proper motions (${\mu }_{\alpha }^{* }={\mu }_{\alpha }\cos \delta$, μ_δ ) on the sky, v_los values from SEGUE, and StarHorse heliocentric distances are converted to Galactocentric Cartesian phase-space coordinates using Astropy Python tools (Astropy Collaboration et al. 2013, 2018). The adopted position of the Sun with respect to the Galactic Center is (X, Y, Z)_⊙ = (−8.2, 0.0, 0.0) kpc (Bland-Hawthorn & Gerhard  2016; Bennett & Bovy 2019). The local circular velocity is V _circ = (0.0, 232.8, 0.0) km s⁻¹ (McMillan 2017), while the Sun's peculiar motion with respect to V _circ is (U, V, W)_⊙ = (11.10, 12.24, 7.25) km s⁻¹ (Schonrich et al. 2010).

For reference, we write, below, how each component of the total angular momentum ( L ) is calculated.

$$\begin{eqnarray}&&\begin{array}{l}{L}_{x}={{YV}}_{z}-{{ZV}}_{y}\\ {L}_{y}={{ZV}}_{x}-{{XV}}_{z}\\ {L}_{z}={{XV}}_{y}-{{YV}}_{x},\end{array}\end{eqnarray} \tag{ 1 }$$

where $L=\sqrt{{L}_{x}^{2}+{L}_{y}^{2}+{L}_{z}^{2}}$. We recall that, although L is not fully conserved in an axisymmetric potential, with the exception of the L_z component, it has been historically used for the identification of substructure in the Galaxy as it preserves a reasonable amount of clumping over time (see Helmi 2020 for a review).

For the entire SEGUE/StarHorse low-metallicity sample, we also compute other dynamical parameters, such as orbital energy (E) and actions ( J = (J_R , J_ϕ , J_z ) in cylindrical frame). The azimuthal action is equivalent to the vertical component of angular momentum (J_ϕ ≡ L_z ) and we use these nomenclatures interchangeably. In order to obtain these quantities, orbits are integrated for 10 Gyr forward with the AGAMA package (Vasiliev 2019) within the axisymmetric Galactic model of McMillan (2017). A total of 100 initial conditions were generated for each star with a Monte Carlo approach, accounting for uncertainties in proper motions, v_los, and distance. The final orbital parameters are taken as the medians of the resulting distributions, with 16th and 84th percentiles as uncertainties.

Given our goals, we seek to construct a sample of Sgr stream members that is both (i) larger in size and (ii) with greater purity than previously considered by J20, but (iii) with a similarly extended metallicity range, reaching the extremely metal-poor ([Fe/H] < −3) regime. J20 have shown that stars from the Sgr stream can be selected to exquisite completeness in the (L_z , L_y ) plane, which exploits the polar nature of their orbits. We reproduce their criterion in Figure 2 (left panels, dashed lines). However, PP21 have recently argued that the J20's criterion also includes ≈21% of interlopers.

Zoom In Zoom Out Reset image size

We inspect the aforementioned (L_z , L_y ) plane, splitting it into J_z > J_R (predominantly polar orbits; Figure 2, top row) and J_z < J_R (radial/eccentric orbits; bottom row) as was done in Naidu et al. (2020). In Figure 2, there exists an excess of stars toward negative values of L_y (top left panel), which is prograde (top middle), and with J_z ≳ 2000 kpc km s⁻¹ (top right; Thomas & Battaglia 2022), corresponding to the footprint of the Sgr stream (see Malhan et al. 2022). In the case where J_z < J_R , where the Sgr stream completely vanishes, the (L_z , E) space is dominated by the Gaia Sausage/Enceladus (GSE; Belokurov et al. 2018; Haywood et al. 2018, also Helmi et al. 2018).

Building on the above-described facts, we quantify how useful the J_z > J_R condition is for eliminating potential GSE contaminants within our Sgr stream members. We look at a suite of chemodynamical simulations of Milky Way-mass galaxies with stellar halos produced by a single GSE-like merger presented in Amarante et al. (2022). Within these models, the fraction of GSE debris that end up (at redshift z = 0) on orbits with J_z > J_R is always below 9%. Therefore, we incorporated this condition into our selection as it should remove >90% of potential GSE stars.

Lastly, we restrict the kinematic locus occupied by the Sgr stream in (L_z , L_y ) in comparison to J20 and consider only those stars at >6 kpc from the Sun, in conformity with the V21 model. In this work, the conditions that a star must fulfill in order to be considered a genuine member of the Sgr stream are listed below:

• 1.  
  J_z > J_R ;
• 2.  
  heliocentric distance > 6 kpc;
• 3.  
  −10 < L_y /(10³ kpc km s⁻¹) < −3;
• 4.  
  −4 < L_z /(10³ kpc km s⁻¹) < +1.

This selection is delineated by the yellow box in the top left panel of Figure 2. It is clear that our criteria are more conservative than J20's. Nevertheless, the raw size of our final Sgr stream sample (∼1600 stars) is twice as large as the one presented by these authors despite the sharp cut at [Fe/H] < −0.5. Moreover, the metallicity range covered reaches [Fe/H] ∼ −3, with >200 VMP stars in the sample (top left in Figure 3). Finally, although these Sgr stream candidates were identified from their locus in (L_z , L_y ), we found them to be spatially cohesive and in agreement with the V21 model in configuration space (bottom row of Figure 3). Even so, the potential contamination by other known polar streams (see Malhan et al. 2021) is explored in the Appendix.

Zoom In Zoom Out Reset image size

With this new set of selection criteria at hand, we verified which known Galactic globular clusters (GCs) would be connected to the Sgr system. Consequently, we examined the orbital properties of 170 GCs from the Gaia EDR3-based catalog of Vasiliev & Baumgardt (2021). We found that a total of seven GCs can be linked to this group, including NGC 6715/M54, Whiting1, Koposov1, Terzan7, Arp2, Terzan8, and Pal12. We note that M54 has long been recognized to be the nuclear star cluster of Sgr dSph (e.g., Bellazzini et al. 2008). Furthermore, most of these other GCs had already been attributed to Sgr by several authors (Massari et al. 2019; Bellazzini et al. 2020; Forbes 2020; Kruijssen et al. 2020; Callingham et al. 2022; Malhan et al. 2022).

We begin our study of the Sgr stream's stellar populations by looking at the metallicity distributions obtained for the leading and trailing arms and the differences between them. In Figure 3, the immediately perceptible feature is the excess of VMP stars in the leading arm. On the contrary, the trailing arm presents a significant contribution of metal-rich ([Fe/H] ≳ −1) stars. This property had already been noticed by several authors (e.g., Carlin et al. 2018; Hayes et al. 2020) and is recovered despite the intentional bias of the SEGUE catalog to low-metallicity stars (note the excess at [Fe/H] ≲ −1 in the black/all-sample histogram; see Bonifacio et al. 2021 and Whitten et al. 2021 for discussions).

The final median [Fe/H] values we obtained for the leading and trailing arms are $-{1.46}_{-0.03}^{+0.02}$ and $-{1.28}_{-0.05}^{+0.03}$, respectively, where upper and lower limits represent bootstrapped (10⁴ times) 95% confidence intervals. These metallicity values derived from SEGUE are ∼0.3–0.4 dex lower than the ones obtained from APOGEE data (Hayes et al. 2020; Limberg et al.2022), but we recall that this is due to SEGUE's target selection function (Rockosi et al. 2022).

We further notice that the locations of these metallicity peaks for both arms of the stream are well aligned with the secondary, more metal-poor, [Fe/H] peak for stars in the core of Sgr (grayish histograms in Figure 3; Hayes et al. 2020/APOGEE ¹⁷ and Minelli et al. 2023). Although differences in metallicity scales might be at play, this observation could be relevant for the evolution of the Sgr system in the presence of the Milky Way and LMC.

From StarHorse's output, we should also, in principle, be able to access information regarding ages for individual stars as this parameter is a byproduct of the isochrone-fitting procedure (e.g., Edvardsson et al. 1993; Jørgensen & Lindegren 2005, and Sanders & Das 2018). However, there are some caveats in this approach. First, it becomes increasingly difficult to distinguish between isochrones of different ages toward both the cooler regions of the main sequence as well as the upper portions of the red giant branch (see Figure 2 of Souza et al. 2020 for a didactic visualization). However, it is still possible to circumvent this issue by looking at the turnoff and subgiant areas where isochrones tend to be better segregated (see discussion in Vickers et al. 2021). Second, even at these evolutionary stages, variations in ages and metallicities have similar effects on the color–magnitude diagram (e.g., Yi et al. 2001; Pietrinferni et al. 2004; Demarque et al. 2004; Pietrinferni et al. 2006, and Dotter et al. 2008). Hence, spectroscopic [Fe/H] values can be leveraged as informative priors to break this age–metallicity degeneracy. Third, distant non-giant stars are quite faint, which is the case for our Sgr stream sample. This is where SEGUE's exquisite depth, with targets as faint as g = 19.5, where g is SDSS broad band centered at 4800 Å (Fukugita et al. 1996), comes in handy.

In this spirit, we attempt to provide a first estimate of the typical ages for stars in the Sgr stream. Similar to recent efforts (Bonaca et al. 2020; Buder et al. 2022; Xiang & Rix 2022), we selected stars in the SEGUE/StarHorse low-metallicity sample near the turnoff and subgiant stages. For the sake of consistency, for this task, we utilized stellar parameters derived by StarHorse itself during the isochrone-fitting process as these will be directly correlated with the ages at hand. These turnoff and subgiant stars are mostly contained within $4.5\lt \mathrm{log}{g}_{\mathrm{SH}}\lesssim 3.6$ and T_{eff, SH} ≳ 5250 K, where the subscript "SH" indicates values from StarHorse instead of SSPP. A parallel paper describes in detail this (sub)sample with reliable ages (Queiroz et al. 2023). In any case, for the purpose of this work, we highlight that typical differences between SEGUE's atmospheric parameters and those obtained with StarHorse are at the level of SSPP's internal precision.

We found a total of 56 turnoff or subgiant stars in the Sgr stream (31 in the leading arm plus 25 in the trailing one) for which ages are most reliable (top right panel of Figure 3). As expected, these are quite faint (17.5 < g < 19.5), which reinforces the value of a deep spectroscopic survey such as SEGUE. Members of the Sgr stream (blue and red histograms representing leading and trailing arms, respectively) appear to be older (11–12 Gyr) than the bulk of our sample, which is mostly composed of thick-disk stars. It is reassuring that the age distribution for the entire SEGUE/StarHorse low-metallicity sample (black) peaks at 10–11 Gyr, which is, indeed, in agreement with ages derived from asteroseismic data for the chemically defined, i.e., high-α, thick-disk population (Silva Aguirre et al. 2018; Miglio et al. 2021). We quantify this visual interpretation with a kinematically selected thick-disk sample, following 100 < ∣ V − V _circ∣/(km s⁻¹) < 180 (check Venn et al. 2004; Bensby et al. 2014; Li & Zhao 2017; Posti et al. 2018; Koppelman et al. 2021), where V = (V_x , V_y , V_z ) is the total velocity vector of a given star, i.e., $V=\sqrt{{V}_{x}^{2}+{V}_{y}^{2}+{V}_{z}^{2}}$. Within the SEGUE/StarHorse low-metallicity data (∼7800 stars), we found a median age of 10.6 Gyr for this population.

For the Sgr stream specifically, the bootstrapped median age for the leading arm is ${11.6}_{-0.2}^{+0.4}\,\mathrm{Gyr}$. For the trailing arm, we found ${11.8}_{-0.2}^{+0.3}\,\mathrm{Gyr}$. This translates to ${11.7}_{-0.2}^{+0.3}\,\mathrm{Gyr}$ considering all Sgr stream stars. Of course, uncertainties for individual stars are still substantial, usually at the level of >25% (∼3 Gyr). Therefore, we hope that it will be possible to test this scenario, that the Sgr stream is dominated by stars older (by ∼1 Gyr) than those from the Galactic thick disk, with data provided by the upcoming generation of spectroscopic surveys, such as 4MOST (de Jong et al. 2019), SDSS-V (Kollmeier et al. 2017), and WEAVE (Dalton et al. 2016), and building on the statistical isochrone-fitting framework of StarHorse.

The original motivation for us to identify Sgr stream members in the SEGUE/StarHorse catalog was to analyze the evolution of its kinematics extending deeply into the VMP regime, similar to Gibbons et al. (2017) and J20. The former was the first to propose the existence of two populations in the Sgr stream. Its main limitation was the lack of complete phase-space information, which are now available thanks to Gaia. Regarding the latter, the caveats were the small amount of (∼50) VMP stars in their sample (from the H3 survey) and potential contamination by Milky Way foreground stars (PP21). Here, instead of splitting the Sgr stream into two components, our approach is to model its velocity distribution across different [Fe/H] intervals. The results of this exercise can provide constraints to future chemodynamical simulations attempting to reproduce the Sgr system as was recently done for GSE (Amarante et al. 2022).

Figure 4 displays the distributions of total velocity (V) across different metallicity ranges, from VMP (left) to metal-rich (right). The color scheme is blue/red for the leading/trailing arm as in Figure 3. From visual inspection, one can notice that both histograms become broader at lower [Fe/H] values. In order to quantify this effect of increasing velocity dispersion (σ_V ) with decreasing [Fe/H], we model these distributions, while also accounting for uncertainties, using a Markov Chain Monte Carlo (MCMC) method implemented with the emcee Python package (Foreman-Mackey et al. 2013). As in Li et al. (2017, 2018), the Gaussian log-likelihood function is written as

$$\begin{eqnarray}&&\mathrm{log}{ \mathcal L }=-\displaystyle \frac{1}{2}\displaystyle \sum _{i=1}^{N}\left[\mathrm{log}\left({\sigma }_{V}^{2}+{\sigma }_{V,i}^{2}\right)+\displaystyle \frac{{\left({V}_{i}-\langle V\rangle \right)}^{2}}{\left({\sigma }_{V}^{2}+{\sigma }_{V,i}^{2}\right)}\right],\end{eqnarray} \tag{ 2 }$$

where V_i and σ_V,i are the total velocity and its respective uncertainty for the ith star within a given [Fe/H] bin. We adopt only the following uniform priors: 0 < 〈V〉/(km s⁻¹) < 500 and σ_V > 0. Lastly, we run the MCMC sampler for 500 steps with 50 walkers, including a burn-in stage of 100. Although some of the V histograms in Figure 4 show non-Gaussian tails, this exercise is sufficient for the present purpose.

Zoom In Zoom Out Reset image size

The results of our MCMC calculations are presented in Table 1. Upper and lower limits are 16th and 84th percentiles, respectively, from the posterior distributions. Between −1.5 < [Fe/H] ≤ −0.5, we found no statistically significant (<1σ) evidence for σ_V variations. However, at [Fe/H] ≤ −1.5, the σ_V increases substantially for both arms. According to present data, the VMP component of the Sgr stream (left panel of Figure 4) is dynamically hotter than its metal-rich counterpart at the ≳2σ level. We also verified that this effect is less prominent (∼1σ) for GSE (green histograms in Figure 4; Table 1) even with a not-so-pure (at least 18% contamination; Limberg et al. 2022) selection (Feuillet et al. 2020), which is to be expected given the advanced stage of phase mixing of this substructure.

Table 1. Velocity Dispersion for the Leading and Trailing Arms of Sgr, as Well as GSE, in Bins of [Fe/H]

Substructure	σV	σV	σV	σV
	(km s−1)	(km s−1)	(km s−1)	(km s−1)
	[Fe/H] ≤ −2.0	−2.0 < [Fe/H] ≤ −1.5	−1.5 < [Fe/H] ≤ −1.0	−1.0 < [Fe/H] ≤ −0.5
Leading	${70}_{-4}^{+4} \%$ (194)	${62}_{-3}^{+3} \%$ (284)	${51}_{-2}^{+2} \%$ (405)	${53}_{-4}^{+5} \%$ (94)
Trailing	${44}_{-5}^{+6} \%$ (62)	${38}_{-3}^{+3} \%$ (185)	${32}_{-5}^{+2} \%$ (319)	${28}_{-2}^{+2} \%$ (203)
GSE	${87}_{-2}^{+2} \%$ (1153)	${86}_{-1}^{+1} \%$ (4314)	${83}_{-1}^{+1} \%$ (8020)	${82}_{-2}^{+2} \%$ (1322)

Download table as:  ASCIITypeset image

Now, we put our results in context with those in the literature. With the understanding that the Sgr stream is comprised of two kinematically distinct populations (V21), the increasing σ_V as a function of decreasing metallicity can be interpreted as larger fractions of the diffuse (J20) component contributing to the low-[Fe/H] (dynamically hotter) bins. On the contrary, the main component, which contains most of the stars of the substructure, is preferentially associated with the high-[Fe/H] (dynamically colder) intervals.

PP21 recently argued that the broad velocity distribution for metal-poor stars in the Sgr stream could be an artifact of Milky Way contamination in the J20 Sgr stream data. However, this effect is still clearly present in our 2× larger sample with more rigorous selection criteria. To summarize, in the low-metallicity regime, there appears to be a considerable contribution from both ancient and recently formed wraps of the stream. On the other hand, at high metallicities ([Fe/H] > −1.0), only the newest wrap is represented.

V21's is a tailored N-body model of the Sgr system designed to match several properties of its tidal tails. In particular, in order to mimic the aforementioned misalignment between the stream track and its proper motions in the leading arm, the authors invoke the presence of an LMC with a total mass of 1.5 × 10¹¹ M_⊙, compatible with the findings in Erkal et al. (2019), Shipp et al. (2021), and Koposov et al. (2022). The initial conditions are set to reproduce the present-day positions and velocities of both Sgr and LMC building on earlier results (Vasiliev & Belokurov 2020). However, unlike the LMC and Sgr, the Milky Way is not modeled in a live N-body scheme. Hence, it comes with the limitation that V21 depend on the Chandrasekhar analytical prescription for dynamical friction (e.g., Mo et al. 2010). See Ramos et al. (2022) for a discussion on how this approximation might influence the stripping history of the stream.

In the fiducial model, the initial stellar mass of the Sgr dSph is 2 × 10⁸ M_⊙ and follows a spherical King density profile (King 1962). Moreover, the system is embedded in an, also spherical, extended dark matter halo of 3.6 × 10⁹ M_⊙. Other key features of V21's work is the capability of recovering crucial kinematic and structural features of Sgr's remnant (as in Vasiliev & Belokurov 2020), accounting for perturbations introduced by the gravitational field of the LMC (Garavito-Camargo et al. 2019; Cunningham et al. 2020; Petersen & Penarrubia 2020; Erkal et al. 2021; Garavito-Camargo et al. 2021; Petersen & Penarrubia 2021), and properly following mass loss suffered by the system.

Despite the close match between observations and the V21 model, there are a few limitations that could affect their results. For instance, the model does not account for the gaseous component, which may be relevant for the distribution of the debris as discussed in Wang et al. (2022) and references therein. An additional caveat is the lack of bifurcations in the modeled stream, as originally observed by Belokurov et al. (2006) and Koposov et al. (2012, see discussions by Oria et al. 2022). Finally, Sgr likely experienced at least one pericentric passage ≳6 Gyr ago as can be inferred from dynamical perturbations in the Galactic disk (Binney & Schonrich 2018; Laporte et al. 2018, 2019; Bland-Hawthorn & Tepper-Garcia 2021; McMillan et al. 2022, and see Antoja et al. 2018) as well as the star formation histories of both the Milky Way (Ruiz-Lara et al. 2020) and Sgr itself (Siegel et al. 2007; de Boer et al. 2015). Hence, the V21 simulation, which starts only 3 Gyr in the past, is unable to cover this earlier interaction.

Figure 5 shows observational and simulation data in (Λ_Sgr, v_los), where Λ_Sgr is the stream longitude coordinate as defined by Majewski et al. (2003) based on Sgr's orbital plane. Leading/trailing arm stars are represented by blue/red dots. These are overlaid on the V21 model, where gray and colored points denote dark matter and stellar particles, respectively. We split these simulated particles according to their stripping time (t_strip ¹⁸ ). For the remainder of this paper, we refer to the portion of the (simulated) stream formed more recently (t_strip < 2 Gyr) as the new wrap (green). The more ancient (t_strip > 2 Gyr) component is henceforth the old wrap (orange). Stellar particles that are still bound to the progenitor by the end of the simulation (redshift z = 0) are colored black.

Zoom In Zoom Out Reset image size

We divide our Sgr stream data into the same metallicity intervals as in Figure 4 and Table 1. Essentially, our selected members of the Sgr stream share all regions of phase space with the V21 model particles. Notwithstanding, the bottom-most panel of Figure 5 reveals a first interesting feature. Metal-rich stars in the sample are almost exclusively associated with the new wrap, though this is more difficult to immediately assert for the leading arm because of the overlap between new and old portions within 180° ≲ Λ_Sgr < 300°.

As we move toward the lower metallicity (upper) panels of Figure 5, we see larger fractions of observed Sgr stream stars coinciding with the old wrap in phase space. At the same time, the dense groups of stars overlapping with the new wrap fade away as we reach the VMP regime (top panel ). As a direct consequence, stream members are more spread along the v_los axis in Figure 5, which, then, translates into the higher σ_V discussed in Section 3.3 for metal-poor/VMP stars. In general, the new wrap is preferentially associated with metal-rich stars, but also extends into the VMP realm. Conversely, the old component contains exclusively metal-poor ([Fe/H] ≲ −1) stars. Therefore, these suggest that, at low metallicities, the Sgr stream is composed of a mixture between old and new wraps and this phenomenon drives the increasing σ_V quantified in Table 1.

The dichotomy between metal-rich/cold and metal-poor/hot portions of the Sgr stream has been suggested, by J20, to be linked to the existence of a stellar halo-like structure in the Sgr dSph prior to its infall. This stellar halo would have larger velocity dispersion, be spatially more extended, and have lower metallicity than the rest of the Sgr galaxy. As a consequence of its kinematics, this component would be stripped at earlier times. Indeed, we verified that the old wrap of the Sgr stream stripped >2 Gyr ago in V21's model, is mainly associated with metal-poor stars (Figure 5), in conformity with J20's hypothesis. Meanwhile, the majority of the most metal-rich stars can be attributed to the new wrap. In order to check how the present-day properties of the Sgr stream are connected to those of its dSph progenitor, hence testing other conjectures of J20, we now look at the initial snapshot of V21's simulation, including the satellite's orbit and disruption history.

Simply to comprehend the assembly of the stream over time according to the V21 model, we plot the distribution of t_strip in the left panel of Figure 6. The excess at t_strip = 0 is due to N-body particles that remain bound to the progenitor. On top of these histograms, we add the trajectory of the Sgr dSph in the simulation (blue line and dots) in terms of its Galactocentric distance. With this visualization, it is clear how intense episodes of material being stripped (at both 2.0 < t_strip/Gyr ≲ 2.7 and 0.5 ≲ t_strip/Gyr ≲ 1.5) are intrinsically related to close encounters between Sgr and the Milky Way (at ∼2.5 and ∼1.2 Gyr ago), which originates the new and old wraps discussed in Section 4.2. Also, note how most of the material is associated with the recently formed component (new wrap) of the stream.

Zoom In Zoom Out Reset image size

In order to test J20's conjecture that the stripped portion of the Sgr dSph associated with the formation of the old wrap was already dynamically hotter than the new one prior to the galaxy's disruption, we check the σ_V , with respect to Sgr, of these components in the initial snapshot of V21's simulation, which starts 3 Gyr in the past (redshift z ∼ 0.25 in Planck Collaboration et al. 2020 cosmology). Indeed, the σ_V of stars that end up forming the old wrap, i.e., stripped at earlier times, is higher (∼18 km s⁻¹) in comparison to the σ_V of stars from the new component (∼14 km s⁻¹).

In the middle panel of Figure 6, the initial snapshot is presented in configuration space as (X, Y)_Sgr,0, an Sgr-centered frame. The orange dots (t_strip > 2 Gyr/old wrap) in this plot are less centrally concentrated (90% of stellar particles within ∼4 kpc) than the green ones (t_strip < 2 Gyr/new wrap; 90% within 3 kpc). This behavior is clear from the right panel of the same figure that shows the cumulative distributions of galactocentric radii (r_Sgr,0) in the same system. Also, stars that remain bound until redshift z = 0 have even lower σ_V (∼11 km s⁻¹) and are spatially more concentrated (90% within < 2 kpc) than the other components.

From the above-described properties of the V21 model, we can infer that the periphery of the simulated Sgr dSph contains a larger fraction of stars that end up as the old wrap (stripped earlier) in comparison to its central regions. Therefore, with the understanding that the old wrap is essentially composed of low-metallicity stars), we reach the conclusion that the core regions of the Sgr dSph were more metal-rich than its outskirts prior to its accretion. We recall that, indeed, previous works reported evidence for a metallicity gradient in the Sgr remnant (Bellazzini et al. 1999; Layden & Sarajedini 2000; Siegel et al. 2007; McDonald et al. 2013; Mucciarelli et al. 2017; Vitali et al. 2022). Nevertheless, fully understanding how these stellar-population variations in the Sgr system relate to its interaction with the Milky Way remains to be seen (for example, via induced star formation bursts; Hasselquist et al.2021).

Although our interpretation favors a scenario where the Sgr dSph had enough time to develop a metallicity gradient before its disruption, quantifying this effect is difficult. One way to approach this would be by painting the model with ad hoc metallicity gradients, and, then comparing with, for instance, the present-day [Fe/H] variations observed across the Sgr stream (Hayes et al. 2020 and references therein). We defer this exploration to a forthcoming paper.

Apart from T_eff, $\mathrm{log}g$, and [Fe/H], the SSPP also estimates α-element abundances based on the wavelength range of 4500 ≤ λ/Å ≤ 5500 (Lee et al. 2011), which contains several Ti i and Ti ii lines as well as the Mg i triplet (∼5200 Å). de Boer et al. (2014) utilized [α/Fe] values made available by SEGUE for stars in the Sgr stream to argue that a knee existed at [Fe/H] ≲ −1.3 in the [α/Fe]–[Fe/H] diagram (Wallerstein 1962; Tinsley 1979) for this substructure. However, this result is not supported by contemporaneous high-resolution spectroscopic data, especially from APOGEE (Hayes et al. 2020; Limberg et al. 2022; Horta et al. 2023), also H3 (Johnson et al. 2020; Naidu et al. 2020). If the position of Sgr's α knee was truly located at such high [Fe/H], it would imply that it should be even more massive than GSE under standard chemical-evolution prescriptions (e.g., Matteucci & Brocato1990).

Here, we revisit the α abundances for the Sgr stream using SEGUE, but with a larger sample with lower contamination. In the left panel of Figure 7, we can see the continuous decrease of [α/Fe] as a function of increasing metallicity for both the Sgr stream and GSE. Most important, at a given value of [Fe/H], the median [α/Fe] of the Sgr stream (both leading and trailing arms) is lower than GSE's. This difference becomes more prominent at [Fe/H] ≳ −1.5, in agreement with the aforementioned high-resolution spectroscopy results from both H3 and APOGEE. Despite that, the low accuracy/precision of [α/Fe] in SEGUE still makes it difficult to attribute stars to certain populations on an individual basis.

Zoom In Zoom Out Reset image size

With the SEGUE low-metallicity data at hand, we also explore carbon abundances. In particular, we are interested in finding carbon-enhanced metal-poor (CEMP; [C/Fe] > +0.7 and [Fe/H] < −1; see Beers & Christlieb 2005; Aoki et al. 2007, and Placco et al. 2014) stars in the Sgr stream. The reasoning for that being the recent results by Chiti et al. (2020), and also Hansen et al. (2018) and Chiti & Frebel (2019), where these authors found no CEMP star in their sample of Sgr dSph members within −3.1 < [Fe/H] ≲ −1.5. Moreover, we utilize observations of the Sgr stream as a shortcut to check for potential differences in CEMP fractions between a dwarf galaxy and the Milky Way's stellar halo (see Venn et al. 2012; Kirby et al. 2015; Salvadori et al. 2015; Chiti et al. 2018) in a homogeneous setting. Given that the CEMP phenomenon, especially at [Fe/H] ≲ −2.5, is connected to nucleosynthesis events associated with the first generations of stars, perhaps Population III (e.g., Nomoto et al. 2013; Yoon et al. 2016; Chiaki et al. 2017), identifying such objects provide clues about the first chemical-enrichment processes that happened in a galaxy.

Throughout this section, we consider carbon abundances obtained for SEGUE spectra by (Lee et al. 2013, see also Carollo et al. 2012; Lee et al. 2017, 2019, and Arentsen et al. 2022). Inconveniently, this catalog also comes with slight variations of the stellar atmospheric parameters in comparison with the public SEGUE DR9 release. Therefore, in order to confidently identify CEMP stars, we first select candidates using only Lee et al. (2013) [C/Fe] and [Fe/H] (subscripts "L13" in Figure 7). Then, we compare [Fe/H]_L13 with [Fe/H] values from our standard DR9 sample ([Fe/H]_DR9 (the right panel of Figure 7) to confirm their low-metallicity nature.

The middle panel of Figure 7 ([C/Fe]–[Fe/H]) exhibits our selection of CEMP candidates (yellow box). Note that we take only those stars at [Fe/H]_L13< −1.5, for consistency with the metallicity range covered by Chiti et al. (2020). Expanding this boundary to [Fe/H]_L13< −1 would only include a couple of additional CEMP candidates. We discovered a total of 39 likely CEMP stars (33 at [Fe/H]_L13< −2). With this sample at hand, we looked for those candidates confidently (3σ in [C/Fe]_L13) encompassed by the CEMP criteria. We found seven such objects, shown as star symbols in Figure 7. Although two of these CEMP stars have discrepant metallicity determinations ([Fe/H]_L13 versus [Fe/H]_DR9 (the right panel of Figure 7), we can still confidently assert that there exist CEMP stars in the Sgr stream.

A possible explanation for the lack of CEMP stars in Chiti et al. (2020)'s sample could be their photometric target selection, which was based on SkyMapper DR1 (Wolf et al. 2018). The excess of carbon, hence the exquisite strength of the CH G band, is capable of depressing the continuum extending to the wavelength region of the Ca ii K/H lines, close to the center of SkyMapper's v filter (3825 Å; see Da Costa et al. 2019 and references therein), a phenomenon referred to as carbon veiling (Yoon et al. 2020). A scenario where, if confirmed, the surviving core of the Sgr dSph has a lower CEMP fraction than its outskirts/stream at a given metallicity could be similar to what potentially happens to the Milky Way's bulge and halo (Arentsen et al. 2021). Either way, the unbiased discovery of additional VMP stars in Sgr as well as other dSph satellites (e.g., Skúladóttir et al. 2021) will be paramount for us to advance our understanding of the earliest stages of chemical enrichment in these systems.

Finally, we also calculate the fraction of CEMP stars in the Sgr stream and compare it with the Milky Way. Arentsen et al. (2022) have recently demonstrated that various observational efforts focused on the discovery and analysis of metal-poor stars via low/medium-resolution (up to ${ \mathcal R }\sim 3000$) spectroscopy report inconsistent CEMP fractions among them (Lee et al. 2013; Placco et al. 2018; Aguado et al. 2019; Placco et al. 2019; Yuan et al. 2020b; Arentsen et al. 2020; Limberg et al. 2021a; Shank et al. 2022). However, we reinforce that it is not our goal to provide absolute CEMP fractions (e.g., Rossi et al. 2005; Lucatello et al. 2006; Yoon et al. 2018), but rather use the SEGUE/StarHorse low-metallicity sample to make a homogeneous comparison. For this reason, we do not perform any evolutionary corrections (as in Placco et al. 2014) to the carbon abundances of Lee et al. (2013). The overall fraction of CEMP stars in the whole sample at [Fe/H] < −2, but excluding Sgr, is 19% ± 1% ¹⁹ within the same $\mathrm{log}g$ range. For the whole Sgr stream, leading and trailing arms altogether, this number is 16% ± 5%. Limiting our analysis to only giants (T_eff < 5800 K and $\mathrm{log}g\lt 4$), we find 12% for both the stream and full sample. Therefore, we conclude the SEGUE carbon-abundance data does not provide evidence for variations in the CEMP frequency between Sgr (stream) and the Milky Way.