Abundances and kinematics of carbon-enhanced metal-poor stars in the Galactic halo A new classification scheme based on Sr and Ba

Carbon-enhanced metal-poor (CEMP) stars span a wide range of stellar populations, from bona ﬁde second-generation stars to later-forming stars that provide excellent probes of binary mass transfer and stellar evolution. Here we analyse 11 metal-poor stars (8 of which are new to the literature)


Introduction
Like many types of living organisms, most of the oldest, most-Fe-poor stars, are carbon rich. This indicates that C has been produced in large amounts, from the earliest times in the very Tables A1 and A2 are only available at the CDS via anonymous ftp to cdsarc.u-strasbg.fr (130.79.128.5) or via http://cdsarc. u-strasbg.fr/viz-bin/qcat?J/A+A/623/A128 Based on observations obtained at ESO Paranal Observatory, programme 090.D-0321(A). first stars, right up until now. However, over time the dominant production sites may well have shifted. The demonstrated high frequency of carbon-enhanced metal-poor (CEMP) stars (up to 80% for [Fe/H] < -4, Yong et al. 2013;Placco et al. 2014;Yoon et al. 2018) seems to indicate that the first (likely massive) stars produced C, N, and O and possibly some Na and Mg, but not Fe or heavier elements in large amounts, keeping these stars Fe-poor.
To date, only two ultra-metal-poor ([Fe/H] < -4.5) stars without strong C enhancements have been identified (e.g. Caffau et al. 2011;Starkenburg et al. 2018). Most of the bona fide second-generation stars are CEMP-no stars, with low abundances of heavy elements on their surfaces (Yong et al. 2013), while the majority of CEMP stars remain those enhanced in slow neutron-capture elements; 80% of these are known to be members of binary systems (Starkenburg et al. 2014;Hansen et al. 2016a). Two much-less populated sub-groups are the CEMP-r and CEMP-r/s stars, which are also enhanced in rapid neutroncapture material, making their stellar spectra extremely crowded at most wavelengths. Some refer to CEMP-r/s as CEMP-i stars, as they appear to be enriched by the intermediate neutroncapture process (e.g. Abate et al. 2016;Hampel et al. 2016). Understanding how stars in the individual CEMP sub-groups become enriched in various elements provides important clues on their progenitor populations, their nucleosynthetic pathways, and their masses, which in turn can help constrain the initial mass function. Moreover, we can assess early binarity over a wide stellar mass range.
Many of the CEMP stars known today are faint and remote, and therefore they have been observed with larger telescopes with efficient, lower-resolution spectrographs, sufficient to derive accurate molecular abundances. However, offsets in atomic abundances might be introduced when comparing to abundances derived from high-resolution spectra of the same stars. Despite possible limitations in abundance accuracy owing to low-resolution spectra, a dichotomy in absolute C abundances has been shown to enable reliable separation of CEMP-no and CEMP-s stars in the A(C)-versus-[Fe/H] diagram (Rossi et al. 2005;Spite et al. 2013;Bonifacio et al. 2015;Hansen et al. 2015aHansen et al. , 2016bYoon et al. 2016).
Previous studies have suggested that the CEMP-no stars are typically associated with the outer halo, while the majority of CEMP-s stars reside in the inner halo (Carollo et al. 2012(Carollo et al. , 2014Beers et al. 2017;Lee et al. 2017;Yoon et al. 2018), but such dissections have so far mainly been based on distance estimates. To date, no kinematic study of these subclasses has been carried out (for large samples). This is vital for tracking the orbital histories of these stars to look for possible associations in phase space that could indicate a common origin, and to accurately trace the stars to their proper birth environment. This is now possible due to the recent advent of Gaia's second data release (DR2, Gaia Collaboration 2018), which underscores the need for additional observations of, in particular, relatively bright CEMP stars.
As shown by Hansen et al. (2015a), the CEMP-no stars, which may dominate the outer Galactic halo, are essentially single stars, while the CEMP-s stars are predominantly found in binary systems. This implies that the carbon in the CEMP-no stars was synthesised elsewhere, and was then implanted into the natal clouds of the very metal-poor stars of today. Realistic galaxy-formation models must take this enrichment process into account, whether the progenitors were fast-rotating massive stars (FRMS; Maeder & Meynet 2003;Hirschi 2007;Frischknecht et al. 2016;Choplin et al. 2016) or other firstgeneration stars that ended their lives as mixing and fallback supernovae (SNe).
Here we analyse a sample of metal-poor stars with different chemical enrichments and probe their kinematics to determine their membership in the inner-or outer-halo populations. Based on chemical abundances of only two heavy elements (Sr and Ba) we provide a new method for sub-classifying the CEMP stars. Moreover, we use their detailed chemical patterns to explore the nature and mass of some of the first (massive) stars that enriched these old CEMP stars.
The paper is organised as follows. Section 2 outlines the observations, Sect. 3 describes the stellar-parameter determination, Sect. 4 presents the derivation of stellar abundances, and Sect. 5 highlights our abundance results. Section 6 describes the use of the Sr/Ba ratio for discrimination of CEMP-s, CEMP-r/s, and CEMP-no stars and Sect. 7 details the kinematics derived using orbital parameters based on Gaia DR2. A brief summary of our conclusions is provided in Sect. 8.

Observations
Our programme sample was selected from the "Catalogue of carbon stars found in the Hamburg-ESO survey" (Christlieb et al. 2001) and the later studies by Placco et al. (2010Placco et al. ( , 2011; the likely most metal-poor stars (based on line indices calculated directly from the objective-prism spectra) were targeted. Except for one star, all stars turned out to be CEMP stars with [Fe/H] < −2.0 and [C/Fe] > 1.0. The 11 sample stars were observed between October 2012 and January 2013 with X-shooter/VLT (Vernet et al. 2011) using a nodding technique. The three arms UVB/VIS/NIS were used with slits widths of 1.0"/0.9"/0.9", resulting in resolving powers of R ∼ 5400/8900/5600, respectively, and covering the wavelength range 300−2500 nm. Stellar coordinates, exposure times, and heliocentric radial velocities are provided in Table 1. The raw echelle spectra were reduced using the X-shooter pipeline v. 2.6.5; the 1D spectra were radialvelocity shifted, co-added, and normalised. The radial velocity of HE 0002-1037 was measured from the Mg triplet and other strong lines, and subsequently the spectrum was shifted to zero velocity. The resulting spectrum was then used as a template for cross correlation to determine the radial velocities of the other programme stars.

Stellar parameter measurements
We follow the same approach for deriving stellar parameters and abundances as applied in Hansen et al. (2016b, Paper I), in order to make the samples as homogeneous as possible. The temperatures are based on V − K s colours, and are computed using the empirical infrared flux method (IRFM) relations from Alonso et al. (1999), adopting the mean IRSA 1 "S & F" reddening (Schlafly & Finkbeiner 2011). The E(B − V) was converted to E(V − K) following Alonso et al. (1996), and the necessary filter system corrections adopted according to Alonso et al. (1998) and Bessell (2005) before using the IRFM. Gravities were determined by fitting Padova isochrones (D. Yong priv. comm.), and the microturbulence was calculated using the empirical relation developed for the Gaia-ESO Survey 2 (M. Bergemann priv. comm). As is generally the case in low-to medium-resolution spectra of CEMP stars, determining the metallicity ([Fe/H]) is very challenging, due to the fact that these stars are metal-poor and exhibit weak Fe lines, which can suffer from the severe line blends from molecular bands (and in some cases also heavyelement atomic lines).
We therefore carefully vetted Fe lines that were clean in high-resolution spectra, and only included the ones that were useful in the X-shooter spectra. The Fe lines employed are listed in Table 2. This resulted in the stellar-atmospheric parameters 1 https://irsa.ipac.caltech.edu/applications/DUST/ 2 A public spectroscopic survey using the ESO facility FLAMES/VLT targeting >10 5 stars, https://www.gaia-eso.eu, Gilmore et al. (2012).
A128, page 2 of 13  Table 3; for comparison, in brackets we list the temperatures and gravities based on Gaia DR2 photometry and parallaxes (Gaia Collaboration 2018), respectively.
There is overall good agreement between our adopted temperature and gravity measurements and the Gaia-based ones. For most stars (8 out of 11 analysed in this work) our values agree with the Gaia estimates within 150 K and 0.3 dex for logg, respectively, while one star deviates in temperature by ∼300 K, and another one deviates in gravity by 0.5 dex. The uncertainties on the stellar parameters are indicated in Table 3.

Abundance analysis
Based on the derived stellar parameters, we interpolate ATLAS 9 atmospheric 1D models with new opacity distribution functions (Castelli et al. 2003). These, and a line list based on mainly Kurucz 3 and Sneden et al. (2016), were used together with MOOG (Sneden 1973(Sneden , v. 2014 to derive stellar abundances via spectrum synthesis, assuming local thermodynamic equilibrium. We synthesise the CH, C 2 , NH, CN, OH, and CO bands to obtain C, N, and O abundances, respectively (see Fig. 1) assuming molecular equilibrium. In most cases, the spectrum quality was too low to allow for a meaningful synthesis of OH (except 3 http://kurucz.harvard.edu/linelists.html for in a few stars with high S/N). Hence, the O abundances are therefore based on CO at 23 220 Å. When synthesising CN and CO bands, we use the already derived C abundances to derive N and O, respectively. The final abundances listed are based on an iterative process, which ceased when all C, N, and O bands were well-fit. Representative uncertainties on the abundances arising from uncertainties in the stellar parameters have been derived for HE 0059−6540 and are listed in Table 4.
Overall, there is a good agreement (∼0.2 dex) between the C abundances derived from CH and C 2 (see Table 5), and an even better agreement (0.1 dex) between the N abundance derived from NH and CN. A128, page 3 of 13 The oxygen abundances derived from OH may deviate by up to 0.3 dex, which we ascribe to the low S/N of the OH-band region in the UV and possible 3D effects (Dobrovolskas et al. 2013;Gallagher et al. 2016).
Abundances derived from our atomic line list (see Table  A.1 -available at the CDS) are listed in Table 5. Here we targeted atomic lines of 16 species between Na and Eu that fall in regions that are as little affected by molecular bands as possible. Hence, we mainly focus on lines in the wavelength regions: 5200-5400 Å, 5800-5900 Å, 6100-6200 Å, and 6630-6680 Å. The abundances and results are described below.

Abundance results
Here we compare our results to those presented in Hansen et al. 2016b (Paper I). All of our sample stars are giants, and therefore there is a chance that they have reached an evolutionary stage at which internal mixing processes have taken place, and altered their original composition. We therefore checked their C/N ratios, following the approach in Spite et al. (2005). Figure 2 shows that all of our stars have [C/N] > −0.6 and are unmixed, while two stars (HE 0221-3218 with [C/N] = −0.3 and HE 0020-1741 with [C/N] = −0.4 and a high O abundance) are on the verge of becoming mixed. This does not affect our results, as HE 0221-3218 is the most metal-rich star that does not fall into the CEMP class. We note that the CEMPno star (HE 0020-1741), with [C/N] = −0.4, is close to the mixing boundary in Fig. 2; it has a larger 13 C fraction than most of our other stars, and a very high oxygen abundance as well. This star is an evolved giant, but not yet at a point where we consider its surface composition to have been significantly altered. The isotopic abundance ratio for HE 0059-6540 exhibits a relatively large 13 C/ 12 C, however its atomic abundances seem to counter this, and with a gravity higher than that of HE 0020-1741, we do not consider HE 0059-6540 to be selfpolluted. However, we remain cautious due to the 13 C/ 12 C isotopic ratio of this star.
Hence, when comparing to yield predictions that might trace the source that produced the elemental abundances locked up in these CEMP giants, we assert that all our programme CEMP stars are not significantly affected by stellar-evolution processes such as gravitational settling, levitation, and other mixing processes.
Based on the classification criteria listed in Beers & Christlieb (2005), with updates from Aoki et al. (2007), we find Table 4. Uncertainties (σ) on derived abundances arising from the uncertainty on each of the stellar parameters which are added in quadrature to obtain the total uncertainty for HE 0059−6540. . The CEMP-s star HE 2258-4427 is however on the verge of being classified a CEMP-r/s star. Finally, we have one metalpoor, but non-C-enhanced star (HE 0221-3218). Our results and their sub-classifications can be seen in Fig. 3 and Table 6. Previous studies have used various elements or element pairs to sub-classify CEMP stars either into the four main classes or sub-groups thereof. In Masseron et al. (2010), the [Ba/C] ratio produced a linear trend as a function of [Fe/H] for CEMP-s stars but not for CEMP-r/s stars. However, Fig. 4 shows that the Ba/C ratio alone cannot separate CEMP-s from the CEMP-r/s stars. More recently, Yoon et al. (2016) split the CEMP-no stars into two sub-groups (their Group II and Group III stars), where, A(C), Mg, and Na were used as tracers. In Fig. 4, we explore whether A128, page 4 of 13  or not involving Mg in addition to Ba/C aids in the separation of CEMP-no, CEMP-s, and -r/s stars. As seen from Fig. 4, the combination of C, Mg, and Ba does not lead to a clear differentiation between the groups. Based on their cosmological models, Hartwig et al. (2018) suggested that [Mg/C] could be used to tell if a secondgeneration star was enriched by a single event (mono-enriched) or was the result of several pollution events (multi-enriched).
The bottom panel of Fig. 4 shows our programme data compared to Yoon et al. (2016), where some of their CEMP-no Gr.III stars appear to be mono-enriched, while others lie below the predicted 3σ confidence level. Surprisingly, some of the CEMP-s stars also seem to be mono-enriched, while our CEMP-no star (HE 0020-1741) at first glance appears to be multi-enriched already at the low metallicity of [Fe/H] = −3.6. However, we note that this star has a very high Mg abundance, and the cosmological predictions are likely not able to deal with or represent peculiar enhancements. Additional observations of CEMP stars would be interesting to help clarify the situation, as well as the inclusion of A128, page 5 of 13 A&A 623, A128 (2019) Carbon has already been shown to be a good separator between CEMP-no and CEMP-s stars, but currently there is no consistent way of sub-classifying all CEMP stars into their respective groups. A separation of CEMP-s and CEMP−r/s stars was attempted in Hollek et al. (2015) using [Y/Ba]; however, their application was limited to these two groups, and was shown not to apply to all CEMP groups.
In Paper I, a separation based only on heavy elements such as Sr and Ba was also suggested to sub-classify the CEMP stars, and at the same time learn about their progenitor site. The Sr/Ba ratio is different in AGB stars and FRMS, which makes Sr/Ba an efficient and useful descriptor to trace possible formation sources in the sense that only two elements/absorption features need to be analysed to derive this abundance ratio. Additionally, Choplin et al. (2017) showed that the Sr/Ba ratio, together with the O production in FRMS, is much higher than in AGB stars.
Here we show that Sr and Ba can be used to separate not only the various sub-groups of CEMP stars but also to distinguish C-normal stars from CEMP stars. Moreover, Sr is intrinsically a much stronger absorption feature than Y, and from a nucleosynthetic perspective, they both most likely originate from the same formation process. The elements Sr and Ba exhibit strong absorption lines, and are therefore detectable in lowerresolution, low-S/N spectra, making them useful features for large surveys. By comparison, Y and Eu are much weaker, and tend to disappear in stellar spectra around [Fe/H] = −3 (see, e.g. Hansen et al. 2014a).
Among the stars in the programme sample, the Sr/Ba ratio appears to be a very informative quantity. A very high [Sr/Ba] = 1.1 is found for our CEMP-no star (HE 0020-1741), which is in fair agreement with the FRMS yields (see Fig. 5). We note that this ratio is almost as high as the record high value found in the Sgr dSph galaxy (Hansen et al. 2018); only two stars in François et al. (2007) exceed [Sr/Ba] = 1.0. We note that those studies focussed on C-normal stars, hence the ratios and formation sites may differ. On the other hand, a low [Sr/Ba] is found in CEMP-r/s stars (Abate et al. 2016;Hampel et al. 2016), and we propose that [Sr/Ba] > −0.5 could be used to separate CEMP-s from CEMP-r/s stars, in instances for which Eu cannot be detected in the spectra (see the discussion below for more details).
Keeping the small sample size in mind (11 stars), our findings indicate that FRMS provide a good representation of the CEMP-no stars, and to a smaller extent some CEMP-s stars (see Fig. 5). Moreover, in  Choplin et al. (2017), contrasting with the abovedescribed AGB yields. Despite some of the stars falling slightly off the predictions, there is an overall good agreement between the results illustrated by the two panels.
The match of FRMS yields to CEMP-s abundances was also shown in Choplin et al. (2017) for seemingly single CEMP-s stars, indicating that a sub-group of CEMP-s stars could be polluted by fast-rotating massive stars. However, according to the simulations by Abate et al. (2018) all CEMP-s stars could be undetected binaries with periods > 10 4 days. The vast majority of CEMP-s stars (which are binaries) are well-reproduced by AGB stars. We discuss their mass range below.
A128, page 6 of 13 In all panels we show our CEMP-no stars as filled blue squares, CEMP-s stars as filled red triangles, CEMP-r/s as green circles, and C-normal metal-poor stars as filled black diamonds.
The best way to explore the origin of various CEMP subgroups is still their detailed abundance patterns, from which masses of the AGB donor star, as well as contributions from AGB stars, SNe, and neutron star mergers can be extracted.
For comparison to the yields (Fig. 6), we limit our consideration to the four likely mono-enriched stars. Here we have compared to the most metal-poor AGB yields from Lugaro et al. (2012), which have a total metallicity ([Fe/H]) of ∼−2.2. The magneto-hydrodynamical jet-driven supernova (MHD jet-SNe) yields are from Winteler et al. (2012), and finally, the dynamical ejecta from a neutron star merger (consisting of two neutron stars of 1.0 M each) are shown as well (Korobkin et al. 2012 andRosswog et al. 2013). We acknowledge that this is an incomplete representation of possible NSM yields.
Based on χ 2 fitting of the rare earth elements (REEs), we find that the majority of our CEMP-s stars fit the metal-poor, low-mass AGB yields from Lugaro et al. (2012) with 1.5 M , with a few stars slightly preferring 2.0 M . Our best fits typically result in χ 2 ∼ 1.01−1.07. The CEMP-r/s stars seem to favour the slightly more massive AGB donors with 2.0−5.0 M , where lowmass NSMs appear to have contributed to the REEs (56 < Z < 63), while the rare MHD core-collapse SNe may have enriched these stars in the lighter elements Sr and Y (see Fig. 6). We note that NSM disc ejecta could also have contributed material rich in Sr and Y, instead of or in addition to MHD jet-SNe.

A new classification scheme based on Sr and Ba
As already proposed in Paper I, the Sr/Ba ratio might be interesting for use in chemical tagging, since these elements are formed in different nucleosynthesis processes and astrophysical sites. Strontium is made in larger amounts than Ba in FRMS via the (weak) s-process. In contrast, the typical low-mass AGB star produces more Ba than Sr, yielding a low Sr/Ba ratio, while more massive AGB stars produce slightly larger, or equal amounts of Sr compared to Ba. This is the case for the metal-poor AGB yields from Lugaro et al. (2012), in agreement with Cristallo et al. (2011). The exact details and abundances vary depending on the metallicity of the model, as the AGB s-process yields are secondary, and hence metallicity (seed) dependent. ratio as listed in Table 6. The asterisk identifies stars where the old classification scheme, based on values in Table 5 following Beers & Christlieb (2005), and the new proposed Sr/Ba classification do not agree.
Right at the level of [Sr/Ba] = 0.5 a few stars may be misclassified however. Besides those, only one star (HE 0317-4705) would be incorrectly assigned as a CEMP-s instead of r/s. The lower bound on the CEMP-r/s class is set based on currently known [Sr/Ba] ratios and the MHD jet-SNe yield prediction (in order to separate it from CEMP-r stars, which are presently few in number, and not believed to be the product of AGB mass transfer). While the yields from Lugaro et al. (2012) cannot fully explain the CEMP-r/s stars, their light-to-heavy s-process ratio is seen to typically fall below −0.5 in their Fig. 7. When applying this to the CEMP sample in Paper I, all CEMP stars are well-classified, except for two CEMP-s stars, HE 0448-4806 and HE 2235-5055, for which Eu in our previous study could not be measured owing to low S/Ns, rendering the testing of the class challenging. These latter two stars are, according to this classification, CEMP-r/s stars. Clearly, this must be tested in a much larger sample; however, being able to sub-classify CEMP stars accurately, and to directly tie a site and its progenitor mass by only measuring abundances of two heavy elements, seems promising in the era of large surveys. Strontium and barium are the only two heavy elements beyond Fe that exhibit readily detectable absorption features in low-/moderate-resolution, low S/N spectra (Caffau et al. 2011;Hansen et al. 2013Hansen et al. , 2015bHansen et al. , 2016b).
An additional advantage of using Sr and Ba is their robust behaviour in LTE versus NLTE. Several studies have shown that the Sr II NLTE corrections are on average only ±0.1 dex (Bergemann et al. 2012;Hansen et al. 2013), and that they may only in a few cases increase to 0.2 dex, depending on stellar parameters and the Sr line used (Andrievsky et al. 2011). The Ba NLTE corrections are slightly higher (±0.1 − 0.3 dex), again A128, page 8 of 13  and Hansen et al. (2013), and the NLTE Ba abundance should be increased by 0.1 dex for the 5853 Å line (Andrievsky et al. 2009). This means that the Sr/Ba ratio would at most change by ±0.2 dex in NLTE versus LTE, which is in agreement with the metal-poor Sr/Ba NLTE study of C-normal stars by Andrievsky et al. (2011). A test of the 3D corrections for Sr indicated that the NLTE and 3D corrections would cancel out (Hansen et al. 2014b), which likely would bring the 1D LTE values closer to the fully 3D, NLTE-corrected Sr/Ba ratios. This makes this ratio a stable segregator that not only allows us to classify stars for statistical studies, but also provides information on the nature of the progenitors.
In comparison, C abundances are prone to large 3D corrections (on the order of −0.3 to −0.6 dex, especially the CEMPno stars with lower absolute A(C)); they may also be biased by the lack of exact O abundances (Dobrovolskas et al. 2013;Gallagher et al. 2016Gallagher et al. , 2017. This correction could ultimately push some CEMP-no stars out of the CEMP class, owing to the lowered (3D) C abundance. Oxygen is more difficult to derive than C, and is therefore missing for many CEMP stars, leaving an incomplete picture of the nature of the stars and their actual abundances. This could influence the fraction of metal-poor stars that are classified as CEMP stars.
Using Ba and Sr to sub-classify the CEMP stars, we note some separation from simple inspection of Fig. 7. The metal-poor, Cnormal sample from François et al. (2007) was NLTE corrected by Andrievsky et al. (2011), and these abundances (offset by a minor amount compared to our LTE values), clearly populate a distinct region of the diagram, despite overlapping perfectly in [Fe/H] with our CEMP sample (which ranges from [Fe/H] = −2 down to ∼−4). Except for one CEMP-s/no star (HE 0516-2515), the C-normal metal-poor region is cleanly separated from the CEMP stars. The cut may have to be adjusted in a larger sample, however, all CEMP stars appear to have higher [Ba/Fe] than C-normal stars, regardless of their sub-classes. The division at [Ba/Fe] = 0 is loosely set, and spreads around these values in agreement with an average Galactic chemical evolution value based on observations from Hansen et al. (2012) and Roederer et al. (2014). The blue CEMP-no panel is poorly populated, and would need more data points to confirm the bounds of this region. Here, C abundances may be crucial to separate a star with low Ba and normal C abundances from a CEMP-no star. The most metal-poor CEMP-no stars (with [Fe/H] < −4) may be viewed with caution, as these could fall slightly below the suggested cut (see Yong et al. 2013). Except for one CEMP-r/s star (HE 0317-4705), the CEMP-s region is cleanly separated, and shows a strong overlap with the CEMP-s stars in Caffau et al. (2018), while the CEMPr/s region is more contaminated by CEMP-s stars.
Additional i-process yields could help narrow this down. If the i-process is solely associated with AGB stars, and sets in at neutron densities that are only an order of magnitude larger than the classical AGB s-process (Karakas & Lattanzio 2014;Abate et al. 2016), some overlap between these two groups would be expected. These considerations are based on small number statistics, and the cuts between the CEMP classes need to be confirmed for a larger sample. The Sr/Ba ratio, however, clearly provides useful information about the nature of the individual stars and their progenitors, and helps us to understand the large star-to-star scatter seen both in LTE and NLTE-corrected samples (Andrievsky et al. 2011;Hansen et al. 2013).
The Sr/Ba ratio is also interesting from a purely nucleosynthetic point of view. Several studies have proposed that Sr could be formed by both a heavy and light process (e.g. a main and a weak process), while Ba, located beyond the second s-process peak, would mainly be formed by a main process (Qian & Wasserburg 2008;Andrievsky et al. 2011;Hansen et al. 2014a). Moreover, with the abundance Sr being much higher than that of Ba, a second or additional process or contribution appears to be required (François et al. 2007). This is in good agreement with Hansen et al. (2014a), where the abundance patterns in all but one of the most metal-poor stars could be well-explained A128, page 9 of 13 by two neutron-capture processes contributing to the abundances derived for very metal-poor stars (with [Fe/H] −2.5).
The most complete nucleosynthetic mapping still requires a rich abundance pattern, which in turn calls for either high-S/N, moderate-resolution spectra or high-resolution spectra. A note of caution when comparing abundances derived from spectra of various quality and resolution should be made. Several studies have dealt with both high-and moderate-resolution spectra and have found differences in abundances derived from these when analysing the same stars (Caffau et al. 2011;Cohen et al. 2013;Aguado et al. 2016). When comparing our abundances to those derived in for example Placco et al. (2016) for HE 0020-1741, we found that the [Fe/H] and other abundances differ by 0.3-0.4 dex, mainly due to (unresolved) blends. However, by using our list of clean Fe lines, this difference can be reduced. Alternatively, it might be worth reducing the stellar metallicities derived from moderate-resolution metal-poor spectra if their abundances are to be compared to those with high-resolution metallicities.

Kinematic analysis
In order to investigate the orbital histories of CEMP stars, we first cross-identified their coordinates with the second data release (DR2) of Gaia (Gaia Collaboration 2018), which yielded the required five-parameter astrometric solution in terms of position, proper motions, and parallaxes. The latter were considered in terms of the prior-free, Bayesian distance estimates of Bailer-Jones et al. (2018), in turn derived from the Gaia parallaxes (to avoid, e.g. negative parallaxes). Using the radial velocities derived above, we backwards-integrated the stellar orbits for 12 Gyr in a Galactic potential accounting for a logarithmic halo and spherical bulge (Fellhauer et al. 2008) and a disc model by Dehnen & Binney (1998). This neglects the warp and flare of the outer disc (e.g. Momany et al. 2006), which will have little impact on our distant stars.
For comparison purposes, we also performed the analysis in an identical matter for the CEMP-no, CEMP-s, and metalpoor stars from the studies of Hansen et al. (2015cHansen et al. ( , 2016a. Here, we note an overlap of three objects from our sample with the latter comparison samples, which naturally led to the same cross-match with Gaia. Accordingly, we found the same orbital parameters from the two data sets, except for slight modifications due to small differences in the adopted radial velocity 4 between either study (of 10 km s −1 at most). For the entire sample of 98 stars, 89% have relative parallax errors (σ ω /ω) below 40%, two thirds are better determined than 12%, and the median relative distance uncertainty, σ d /d, amounts to 12%. Figure 8 shows the resulting orbital parameters for our present sample and the comparison stars, namely peri-and apocentre distances (R peri , R apo ), maximum height above the plane (Z max ), and orbital eccentricity (e).
We also computed the total specific orbital energy (i.e. kinetic plus Galactic potential energy) and the specific orbital angular momentum, which we specify here in terms of the azimuthal action L z = −J ϕ .
As L z is a conserved quantity in axisymmetric potentials, its combination with the orbital energy (also a constant) in the Lindblad diagram of Fig. 9 offers an opportunity to identify groups of stars in phase space that are otherwise seemingly uncorrelated on the sky (Gómez et al. 2010). This proves particularly valuable if groups of stars are to be associated with an accretion origin from disrupted satellites (e.g. Roederer et al. 2018). As an insitu population of stars in binaries, the CEMP-s stars are unlikely to exhibit any correlations, and indeed no obvious clumping in Fig. 9 is seen. The same holds for the CEMP-no stars, arguing in favour of them originating from early, proto-halo enrichment phases without any coherent orbital histories. All of the stars are bound to the Milky Way, as their orbital energies are less than zero.
Finally, Fig. 10 shows a Toomre diagram, displaying the Galactocentric rotation velocity, V, and its perpendicular component, T = √ U 2 + W 2 . In this representation, an orbit is defined as retrograde for V < 0. This diagram is an often-used diagnostics tool to kinematically separate the Galactic components (e.g. Bensby et al. 2003), which we can use here to efficiently single out halo stars (see also Bonaca et al. 2017;Koppelman et al. 2018;Posti et al. 2018;Veljanoski et al. 2019).
Here, we adopted the criterion of Koppelman et al. (2018) to identify bona fide halo stars as |v − v LSR | > 210 km s −1 , where v designates the total velocity of the star, and v LSR refers to the local standard of rest, which we adopt here as 232 km s −1 with a solar peculiar motion of (U, V, W) = (11.1, 12.24, 7.25) km s −1 (Schönrich et al. 2010). This renders 70% of the entire sample halo stars (including our present work and the reference sets), while only 4 out of 11 of the stars from the current work would qualify as halo progeny via this strict criterion. Kordopatis et al. (2013) asserted that the low-metallicity tail of the metal-weak thick disc extends down to [Fe/H] = −2, while the stars in the (C)EMP samples at velocities between 100 and 210 km s −1 with A128, page 10 of 13 reliable distances (σ ω /ω < 13%) ranging from [Fe/H] = −2.2 to −3.9 are more likely to be halo stars. Moreover, only two of those stars (HE 0507−1430 and LP 624-44) lie in the range 1 kpc < Z max < 2 kpc (see Table A.2, available at the CDS). It is therefore likely that those metal-poor stars with disc-like orbits are either captured halo objects or could constitute an overlapping, inner-halo component, as their apocentres are also typically within ∼12 kpc. As for the sub-groups, there is a marginal preponderance of CEMP-s stars (9/26) at these velocities, while the other CEMP sub-classes are roughly represented in equal parts in the (kinematic) thick disc/halo transition. In order to investigate the origin and properties of the various classes of metal-poor stars, Table 7 lists the fractions of stars in each class satisfying certain orbital and kinematic constraints.
The median heliocentric distance of the entire sample (98 stars) and the stars of the present study (11 stars) are 3.4 and 4.4 kpc, respectively. It is worth noting that the entire sample of 98 stars, as well as each CEMP sub-class in itself, is kinematically unbiased with regards to the U and W components, with approximately half the stars moving on prograde or retrograde orbits. The CEMP-r and -r/s stars appear to have a slightly larger contribution of positive velocities, but this group also contains the lowest number of stars (10% of the sample), as manifested in the larger (Poisson) errors on their fractions, which holds for most of the arguments below. Likewise, there is a balance of prograde and retrograde motions, and the full sample displays only mild net rotation at a mean < V >= −23 ± 13 km s −1 and a velocity dispersion of 128 ± 10 km s −1 . The same holds when considering the CEMP sub-groups, albeit with a smaller dispersion (75 ± 20 km s −1 ) for the CEMP-r, −r/s stars. The values of these dispersions are also listed in Table 7. Overall, these values are broadly consistent with the kinematic properties of the Milky Way halo (e.g. Battaglia et al. 2005). From this aspect we can conclude that our entire sample is kinematically uncorrelated and is an in situ halo population, rather than a major, accreted component that would lead to rotation signatures (Deason et al. 2011). However, given the possible biases in target selection and overall sample size, these results should not be over-interpreted.
The majority of stars have eccentric orbits, with e in excess of 0.5, and the median eccentricity of our sample is 0.7, which confirms their membership in the halo. It is noteworthy that the most eccentric orbits are found among the C-normal, extremely metalpoor stars. About 60% of the stars are inhabitants of the inner halo, if we place the inner/outer halo transition via the stars' apocen- tres within ∼15 kpc (Carollo et al. 2010). This fraction is mostly independent on the CEMP sub-group, although we note that all of the CEMP-r and -r/s stars populate these inner regions. In turn, approximately one in four stars reaches apocentre distances exceeding 20 kpc regardless of CEMP sub-group, bringing them into the outer-halo regions. The metal-poor ([Fe/H] = −1.8) star HE 0854+0151 appears to have an apocentre of 290 kpc, which would place it outside the virial radius of the Galaxy. The orbital period is accordingly long, at ∼5 Gyr. This object has originally been classified as a CEMP-s star by Hansen et al. (2016a). Strictly, its more metal-rich nature defies this classification, along with three more candidates above [Fe/H] = −2 from that list. These should thus rather be labelled CH-stars -or a new, more stringent classification should be adopted. Despite their fundamentally different enrichment channels and purported origins (e.g. Bonifacio et al. 2015), the mean orbital parameters of CEMP-s and CEMPno stars are, on average, remarkably similar. In the following, we address a few cases with distinct kinematics.
HE 2158−5134. This newly analysed CEMP-s star, with a low metallicity of [Fe/H] = −3, exhibits the lowest velocity of the entire sample, at 68 ± 3 km s −1 relative to the LSR. Further, it has a moderate eccentricity (0.27) and a height above the plane, Z max , of 3 kpc. Kinematically, it may be a captured halo object or related to the metal-weak thick disc, despite its very low metallicity. Two additional CEMP-s and CEMP-no stars (HE 2238-4131 and HE 1300+0157) exhibit disc-like kinematics, if we take a velocity cut at 100 km s −1 as a discriminant.
Metal-rich stars. Five stars (HE 0408-1733, HE 2138-1616, HE 2141-1441, HE 2357-2718 have metallicities in the range −0.8 < [Fe/H] < −0.5, and thus are at the high-metallicity tail of the halo's metallicity distribution (Schörck et al. 2009). They exhibit a variety of orbital parameters, with reliable distance estimates to better than <38%, which are consistent with a halo origin, although we note that the perpendicular velocity component, T, is overall small and does not exceed 100 km s −1 (see Table A.2). It is feasible that these stars formed in the Galactic disc or bulge and were subsequently ejected.
High-velocity stars. Several stars in our sample have total velocities relative to the LSR exceeding 500 km s −1 . While some of them are hampered by larger parallax errors (on the order of 25-45%), the objects with the largest motions (HE 0854+0151 and HE 0058-3449) have distance estimates that are precise to better than 16%. These objects, with the highest values of A128, page 11 of 13  Close pericentres. About 10% of our sample have pericentric distances closer than 500 pc (Table A.2). Here, we highlight the CEMP-no star HE 0020-1741, with a large apocentric distance of 22 kpc. Despite a distance uncertainty of 25%, it has the most eccentric orbit of our sample e = 0.99, which brings it to a close Galactocentric passage, within ∼53 pc, and a period of ∼550 Myr (see Fig. 11). While we cannot unambiguously constrain its origin in the central Galactic regions, it is worth noting that the oldest and therefore possibly most metal-poor stars are believed to have formed in the innermost (R GC 3.5 kpc) halo regions (Brook et al. 2007;Tumlinson 2010), and in fact progressively more CEMP stars are being found toward these cen-tral parts of the Milky Way (Koch et al. 2016). Also noteworthy is the large height of this star above the plane, Z max = 10 kpc. This could also indicate that this star was once accreted into the Milky Way halo. This alternative scenario is further bolstered by its chemical composition (Table 4), which shows signatures of enrichment by faint SNe, as often seen in low-mass environments such as the dwarf spheroidal galaxies (Skúladóttir et al. 2015;Susmitha et al. 2017).

Conclusion
While the absolute C abundance, A(C), may provide a rough classification of CEMP stars into two groups (CEMP-s and CEMP-no), measured abundances of heavy elements beyond Fe are needed to better understand their origin and formation sites. In particular, if we want to know the mass of the associated AGB star or constrain the r-process site, a more complete abundance pattern is needed. The Sr/Ba ratio appears to be a good discriminant, and we suggest values and regions to sub-classify CEMPno, CEMP-r/s, and CEMP-s stars. The exact cuts may need to be adjusted based on a larger sample.
Here we show that the careful analysis of moderateresolution, high-S/N spectra provides precise and accurate abundance information to within ∼0.2 dex for 20 elements (including Fe) and two isotopes ( 12 C and 13 C). It is remarkable that moderate-resolution X-shooter data provide abundances that are sufficient in number and accuracy for exploring this class of metal-poor C-enriched stars. Compared to Paper I, we also showed that a S /N > 40 (at 4000 Å) is needed to obtain information on O and a number of heavy elements. Careful selection of Fe lines is crucial in order not to overestimate the [Fe/H] in moderate-resolution CEMP spectra. For this purpose, we provide a vetted line list. Alternatively, reducing the [Fe/H] by ∼0.3−0.4 dex from previously published lower-resolution spectra would be an option, if they are to be compared with highresolution ones. This might be important both now and in the future when comparing data across various high-and lowerresolution surveys.
A comparison to yield predictions showed that FRMS can reproduce our CEMP-no and a few (single) CEMP-s or A128, page 12 of 13 CEMP-r/s stars, while the majority of these are binary stars enhanced directly by an AGB companion star. A sub-division of the CEMP-s and CEMP-r/s stars may be made (based on small number statistics), in that low-mass (∼1.5 M ) AGB stars appear to lead to CEMP-s stars, while CEMP-r/s stars could be enriched by more massive AGB stars (∼2−5 M ). This could mean that early binary systems may favour low-mass AGB companions (in agreement with Abate et al. 2018). However, this is still speculative, and requires testing with a larger CEMP sample.
Our chemodynamical results indicate that all but two stars in the sample of 98 objects considered here belong to the halo populations, and that the CEMP-s and CEMP-no stars have remarkably similar kinematics. With the current Gaia DR2 data, they cannot easily be assigned to the inner/outer halo, as the properties of the CEMP-no and CEMP-s are only marginally different, but we estimate that 25% of the stars (CEMP and C-normal) reach the outer halo. With our sample alone we cannot confirm that the CEMP-no stars mainly belong to the outer halo, while CEMP-s stars dominate the inner halo (as proposed in Carollo et al. 2012). Most of the CEMP stars (this study, Paper I and literature CEMP studies) have an eccentricity of 0.7. The extremely metal-poor CEMP-no star, HE 0020-1741, stands out by having the most eccentric orbit with a close Galactocentric passage.
The moderate-resolution, high-S/N X-shooter spectra have again proved their worth in stellar and Galactic spectroscopynot only for very distant AGNs or GRBs, for which the instrument was designed. Combined with Gaia data, these spectra are very powerful in the analysis and classification of CEMP stars and in tracing their origins.