The Gaia RVS benchmark stars II. A sample of stars selected for their Gaia high radial velocity

Context. The Gaia satellite has already provided the astronomical community with three data releases, and the Radial Velocity Spectrometer (RVS) on board Gaia has provided the radial velocity for 33 million stars. Aims. When deriving the radial velocity from the RVS spectra, several stars are measured to have large values. To verify the credibility of these measurements, we selected some bright stars with the modulus of radial velocity in excess of 500kms − 1 to be observed with SOPHIE at OHP and UVES at VLT. This paper is devoted to investigating the chemical composition of the stars observed with UVES. Methods. We derived atmospheric parameters using Gaia photometry and parallaxes, and we performed a chemical analysis using the MyGIsFOS code. Results. We ﬁnd that the sample consists of metal-poor stars, although none have extremely low metallicities. The abundance patterns match what has been found in other samples of metal-poor stars selected irrespective of their radial velocities. We highlight the presence of three stars with low Cu and Zn abundances that are likely descendants of pair-instability supernovae. Two stars are apparently younger than 1Ga, and their masses exceed twice the turn-o ﬀ mass of metal-poor populations. This makes it unlikely that they are blue stragglers because it would imply they formed from triple or multiple systems. We suggest instead that they are young metal-poor stars accreted from a dwarf galaxy. Finally, we ﬁnd that the star RVS721 is associated with the Gjoll stream, which itself is associated with the Globular Cluster NGC3201.


Introduction
Since 2014 the Gaia satellite (Gaia Collaboration et al. 2016b) has provided crucial information (most important positions, parallaxes, proper motions, radial velocities, and stellar parameters) for a huge sample of stars.Since its launch, there have been three data releases (DR1, DR2, DR3; Gaia Collaboration et al. 2016aCollaboration et al. , 2018Collaboration et al. , 2022)).
The data provided in the Gaia releases have been used by the astronomical community to improve our knowledge on the Milky Way and the Local Group galaxies.There are still limitations on the data provided by Gaia, such as a limitation in brightness, and in this respect, there is nothing that can be done tra.The RVS spectra have a wavelength range of less than 30 nm, and several spectra have a low signal-to-noise ratio (S/N); in fact, a considerable fraction (about 25%) of V r determination in the Gaia DR3 derives from spectra with S/N < 5 (see Katz et al. 2023).These facts, especially the latter, introduce a limitation for a good V r determination.We also found in the Gaia DR 3 that there are stars with extreme V r , and it is unclear whether these values are real or just evidence of problematic spectra.To verify the extreme V r values and confirm or discard them, we added a sample of extreme V r stars in the project focused on observing the Gaia RVS benchmark stars.The kinematic investigation of the sample is to be presented by Katz et al. (in preparation).In this work, we show the results of our chemical investigation.
Another intriguing fact of this high V r sample is that these stars are surely high-speed stars, and it is indeed interesting to combine kinematic and chemical characteristics of high-speed stars.Caffau et al. (2020) investigated a sample of stars selected in Gaia DR 2 for their high transverse velocity, so still high-speed stars.In their sample, some stars happen to appear younger than 6 Ga.This is an intriguing finding that could be explained by a population of blue straggler stars.From a similar selection from Gaia DR3, Bonifacio et al. (2024) highlighted a metal-poor, apparently young population barely compatible with blue straggler stars because their masses are in excess of twice the mass of a typical turn-off metal-poor star and in excess of typical masses of blue stragglers in galactic globular clusters and in the field.We questioned whether we would also find this apparently young population in high V r stars.The answer we derived is positive, and the two young stars detected in this work are unlikely to be blue stragglers because their masses are in excess of 1.9M ⊙ , while blue straggler stars typically have masses < 1.2 to 1.5M ⊙ (see e.g.Carney et al. 2005;Fiorentino et al. 2014;Ferraro et al. 2023).

Target selection
The targets we selected from the Gaia EDR3 (Gaia Collaboration et al. 2021) and Gaia DR3 (Gaia Collaboration et al. 2022) catalogues are stars with an absolute radial velocity in excess of 500 kms −1 .Overall, we wanted to verify whether these large values are mistakes of the pipeline or that the stars really have a large radial velocity.
We made a similar selection of targets regarding a sample of stars observed with SOPHIE at Observatoire de Haute-Provence (OHP).The SOPHIE sample and the kinematic analysis of this sample are discussed in Katz et al. (in preparation).As stated in Katz et al. (in preparation), the stars observed with UVES and investigated in this work are high V r stars.The Gaia G magnitude in the sample is in the range from 10.2 to 13.6 mag.

Observations
A sample of 45 stars was observed in the ESO programme 0109.22XP.001with UVES (Dekker et al. 2000) at VLT in the setting DIC2 437+760 (wavelength ranges 373-499 and 565-946 nm), with slit 0 ′′ .4 (resolving power 90 000) in the blue arm and 0 ′′ .3 (resolving power 110 000) in the red arm.A single exposure of about half an hour was taken for each star.A sub-sample of 19 stars were re-observed in ESO programme 110.23V0.001.
Of the observations in P109, 42 were graded 'A' and three 'B'.In P110, of the 21 spectra investigated, 17 were graded 'A' and four 'B'.For the star RVS735, observed in P109, there is no blue spectrum.
For the stars of P109, we downloaded the UVES spectra reduced by ESO in order to have an homogeneous data reduction and to make data readily available to the community.The spectra of four stars are shown in Fig. 1 in the wavelength range of Gaia RVS.Few stars showed problems in the data reduction, and we reduced the spectra by using the ESO pipeline.For P110, we reduced the spectra to have quick access to them.
We performed the chemical analysis of 43 stars.Regarding the two stars for which there is no chemical analysis, the spectrum of RVS705 is noisy (S/N is too low) and crowded due to it being a carbon star (Alksnis et al. 2001).The second star, RVS724, is a variable and active star whose two spectra are different (see Fig. 2).For the stars with observations in P109 and P110, we added the spectra, except for RVS712 and RVS736, which showed clear differences in radial velocity.

Analysis
The kinematic investigation of this sample of stars is presented in Katz et al. (in preparation).The stars investigated in this work all belong to the Galactic halo according to the criteria by Bensby et al. (2014).
For the chemical investigation, we needed the spectrum to be at laboratory wavelength.We applied the radial velocity provided by Gaia DR3.All the stars we investigated have been confirmed to have a high radial velocity.For the majority of them, the velocity we derived is within a few kilometres per second from the value provided by Gaia DR3.Just one star, RVS739, shows a disagreement between the two velocities, by about 10 kms −1 , but the star is still a high V r star.Our pipeline MyGIsFOS for the chemical investigation was allowed to shift each feature by half of the total broadening of the grid used in the analysis of each star.We verified that the shift applied by MyGIsFOS was sufficient to investigate all the features.If it was not, we applied an extra shift to the spectrum.

Stellar parameters
To derive the stellar parameters, we used the Gaia DR3 photometry and parallax and the reddening from Vergely et al. (2022).We derived the parallax zero-point as suggested by Lindegren et al. (2021).We first de-reddened the Gaia G, G BP , and G RP photometry by using the maps provided by Vergely et al. (2022) and the theoretical extinction coefficients derived from model atmospheres (Mucciarelli & Bonifacio in preparation).The G BP −G RP Gaia DR3 colours were then compared to the synthetic photometry in order to derive the effective temperature.The gravity was derived by using the parallax, corrected by the zero-point, through the Stefan-Boltzmann equation.We first assumed a solar metallicity and derived for each star the effective temperature, T eff , and surface gravity, log g.
The initial stellar parameters were used to derive the metallicity of the stars by using MyGIsFOS (Sbordone et al. 2014).The metallicity derived in this way was then used to derive new stellar parameters.The iterations were repeated up to when the changes in T eff and log g were within 10 K and 0.02 dex, respectively.The micro-turbulence (ξ) was derived by using the calibration of Mashonkina et al. (2017).For the majority of the stars in the sample, the signal-to-noise ratio of the spectra is in fact too low to allow the use of lines with small equivalent widths in order to derive the micro-turbulence by the balance of A(Fe) derived from Fe i lines of different equivalent widths.
Regarding the derivation of the stellar parameters, the grid of synthetic photometry was limited, and we did not allow any extrapolation in T eff or log g beyond the grid.The lower limit accepted in T eff is 4000 K. Six stars converge at this lower-limit in T eff when deriving the stellar parameters, so one could suppose that these stars are in fact cooler.For all these stars (RVS708, RVS714, RVS719, RVS727, RVS730, and RVS734) the calibration of Mucciarelli et al. (2021) provides values close to 4000 K, so we expected that by adopting 4000 K, the stellar T eff would be well within the uncertainty.
If we had used the distance estimate of Bailer-Jones et al. (2021) provided in Gaia DR3 to derive the stellar parameters instead of the parallax corrected by the zero-point, we would have derived T eff as being, on average, 6 K hotter (with a difference always within 40 K) and an average difference in log g of -0.07 dex (with a maximum value at -0.21 dex).These changes in the stellar parameters would not have affected the metallicity derived from Fe i lines (generally well within 0.1 dex and 0.02 dex for T eff and log g, respectively) and would have only marginally affected (within 0.1 dex) the iron ionisation balance because the Fe abundance, when derived from Fe ii, is sensitive to the surface gravity.
When comparing our adopted stellar parameters (T eff and log g) to the parameters derived by using the calibration of Mucciarelli et al. (2021), we found an average difference in T eff of −73 ± 37 K 1 and in log g of −0.03 ± 0.02 dex.We adopted 100 K as the uncertainty in T eff and 0.1 dex in log g.
In Fig. 3, a comparison of the adopted stellar parameters to PARSEC isochrones (Bressan et al. 2012;Marigo et al. 2017) is shown.In the figure we noted that at least two stars (RVS725 and RVS726), which are both very metal poor, appear younger than expected when compared to the isochrones.When comparing the 1 Our values minus Mucciarelli et al. (2021) values.
G BP − G RP colour and the G absolute magnitude to isochrones of various ages, depending on the evolutionary stage, the possible expected masses and ages for RVS725 are 3.2M ⊙ with age of 230 Ma (as a sub-giant branch or a Hertzsprung gap for more intermediate and massive stars), 2.9M ⊙ with age of 280 Ma (still core He-burning, the blueward part of the Cepheid loop of intermediate and massive stars), 2.6M ⊙ with age of 390 Ma (the early asymptotic giant branch or a quick stage of red giant for massive stars2 ).For RVS726 we have just one solution in the interpolation, which is 1.9M ⊙ with age of 900 Ma (the early asymptotic giant branch or a quick stage of red giant for massive stars).A mass below about 1.8M ⊙ can in principle be acceptable for a star to be a blue straggler that is the result of the merging of two old metal-poor stars.However, one has to assume no mass was lost in the merging process.The mass expected for RVS725 is far too large.For both stars, to derive the masses and ages, we used a PARSEC isochrone of metallicity of −1.95 (we emphasise that PARSEC isochrones are not enhanced in α elements, and for this reason, we selected a metallicity that was slightly higher than the [Fe/H] derived for the two stars, which are α-enhanced) and ages from 100 Ma and 3 Ga with a step of 200 Ma.An isochrone at metallicity −1.65 dex would provide the same result or increase the masses by less than 0.1M ⊙ .We concluded that these two stars, but especially RVS725, can really be young stars and were probably formed in dwarf galaxies in a recent time and then accreted.
The stars analysed in this work are all metal poor ([Fe/H] < −1.0), but no star is extremely metal poor ([Fe/H] < −3).We recall that RVS705 is metal rich, but this star has not been anal-  ysed.The stellar parameters and the abundances derived are provided at CDS in an online table that is described in the appendix.

Abundances
The abundances were derived by MyGIsFOS (Sbordone et al. 2014).The solar reference abundances are listed in Table 1.The  synthetic spectra grids are the same as described in Caffau et al. (2021).
Lithium is clearly visible in the metal-poor turn-off star RVS738, providing A(Li) LTE = 2.34.Three evolved stars (RVS702, RVS706, RVS711) show a feature at the wavelength position of the Li 670 nm doublet, and the Li abundance derived from this feature is low (A(Li) LTE of -0.30, 1.07 and 0.89, respectively).For some elements, we compared the observed spectrum to a grid of synthetic spectra, computed with SYNTHE from an AT-LAS 9 model (Kurucz 2005) while only varying the abundance of one element.We minimised the χ 2 to derive the abundance.
We investigated the G-band at 430 nm formed by CH lines to verify if any of the stars was a carbon-enhanced metal-poor (CEMP) star.Through line profile fitting, we could derive the C abundance for 39 stars in the sample, and none appeared to be rich in carbon.We found nearly all stars but one have a negative or close to zero [C/Fe] ratio, in agreement with the results by Spite et al. (2005).These stars are all evolved, and the car- bon destruction is due to their stellar evolution.The turn-off star RVS738, as expected and described in Bonifacio et al. (2009), shows a C enhancement: [C/Fe] = +0.25 ± 0.25.In Fig. 4, the [C/Fe] ratio is plotted against the surface gravity.We expected that more evolved stars would have converted more C into N, but things are more complicated than that.The trend is not evident, and an extra-mixing should be present, though different in the various stars, in order to favour C destruction.We also expected at least a fraction of the stars to be CEMP stars, but we found none.In this metallicity regime (Bonifacio et al. 2015), we expected that CEMP stars would lie on the high C band described by Spite et al. (2013), with a carbon abundance just below solar.So [C/Fe] would still be positive even if the evolution of the CEMP stars had destroyed part of their carbon.
Oxygen was derived by line profile fitting because the region is contaminated by telluric absorptions.We derived A(O) from the [OI] lines for 26 stars in the sample.All the stars in the sample are enhanced in oxygen, with ⟨[O/Fe]⟩ = 0.60 ± 0.11, which is consistent with the results by Spite et al. (2005), and the star with the highest oxygen value was RVS733, [O/Fe] = 0.92.
The stars in the sample are enhanced in the α elements, as expected in metal-poor stars (see Fig. 5).The corrections for departures from local thermodynamic equilibrium (NLTE) for Mg and Si in this sample of stars are not large (see Matas Pinto et al. 2022;Caffau et al. 2023).The star RVS726 has a low [Si/Fe] ratio ([Si/Fe] = −0.21),lower than the other α elements, which are still low.The strong S i lines of Mult. 1 at 920 nm fall in the REDU CCD, but the region is unfortunately contaminated by telluric absorption.We then carefully inspected the analysis done by MyGIsFOS and rejected the problematic lines.We then derived A(S) for nine stars.After applying the NLTE corrections on [S/H] from Takeda et al. (2005), two stars were found to have a negative [S/Fe]: [S/Fe] = −0.05for RVS703 (based on one single line) and [S/Fe] = −0.25 for RVS712 (based on two lines with a relatively high scatter of 0.25 dex).We investigated several Ca i lines to derive the Ca abundance.The line-to-line scatter is generally small (⟨σ⟩ = 0.10 ± 0.03), as expected from the S/N ratios of the spectra.The NLTE corrections as provided by Mashonkina et al. (2016) are on the order of 0.1 dex.When deriving the NLTE corrections for the Ca i for each star for lines used and available in Mashonkina et al. (2007), we derived an average NLTE correction on A(Ca) of 0.10 dex.Titanium was derived from Ti ii lines for all stars in the sample, while for all but one star (RVS725), we could derive A(Ti) from Ti i lines.In the stellar parameter of the stars we investigated, Ti ii lines have small NLTE corrections, while the NLTE corrections for Ti i lines are non-negligible (see Sitnova et al. 2016Sitnova et al. , 2020)).It was thus safer if we relied on Ti ii lines for the Ti abundance.
We investigated the light elements (Na, Al).For Na we investigated the lines at 568.8, 589.0, 589.6, 616.0, 818.3, and 819.4 nm, and we derived A(Na) for 38 stars in the sample.We looked in the material provided by Takeda et al. (2003) to derive NLTE corrections.For the sample, we obtained ⟨[Na/Fe]⟩ = −0.13±0.18,with the lowest value being for RVS739 ([Na/Fe] = −0.57,which becomes even lower when applying the NLTE correction on [Na/H]).To investigate the Al abundance, we used the resonance lines at 394 and 396 nm just for the star RVS738, while for the other stars we used weak lines where the NLTE correction is not expected to be large (see e.g.Baumueller & Gehren 1997).The star RVS738 has [Al/Fe] = −1.07based on three lines and a line-to-line scatter below 0.02 dex.
The single K i line at 769.8 nm that we investigated to derive K abundances is contaminated by telluric absorption in the case of stars with large negative radial velocities.We thus decided to investigate this line through line profile fitting.The other reason supporting this choice was that the NLTE correction is large for this line in the stellar parameter range of the stars investigated here (Reggiani et al. 2019).An investigation with MyG-IsFOS fit the line with a model more metal rich by more than 0.5 dex.We derived the K abundance for 31 stars in the sam- ple.We derived the NLTE corrections in Reggiani et al. (2019), sometimes extrapolating in the tables provided.The large [K/Fe] ratios (⟨[K/Fe]⟩ = +0.58± 0.17) are due to large NLTE effects, but once they were taken into account in [K/Fe], we obtained ⟨[K/Fe]⟩ = +0.11± 0.14.
In 41 stars of the sample, Mn was detected, and we found that all the stars have negative [Mn/Fe] ratios (⟨[Mn/Fe]⟩ = −0.28 ± 0.11).This is due to NLTE effects on the Mn i lines (see e.g.Bergemann & Gehren 2008;Bergemann et al. 2019).We investigated the NLTE effects in Bergemann et al. (2019) and applied them to the lines we used, and we found corrections to always be positive, from about 0.1 to 0.43 dex.For all but one star, we could derive the Ni abundance, which is in excellent agreement with Fe (for the 42 stars ⟨[Ni/Fe]⟩ = −0.02± 0.05).
We derived the Cu abundance for 21 stars in the sample, and they all have a negative [Cu/Fe] ratio ([Cu/Fe] = −0.52 ± 0.14).Caffau et al. (2023) investigated the NLTE effects on Cu for a sample of stars of similar parameters and derived an NLTE correction on average slightly below +0.2 dex, so when applying these NLTE corrections, the stars in our sample would still keep a negative[Cu/Fe] ratio.
For 37 stars in the sample, Zn could be derived by investigating the Zn i lines at 472.2 and 481.0 nm (see Fig. 6).The NLTE corrections derived by interpolating in the values provided by Sitnova et al. (2022) are non-negative and small (not larger than 0.06 dex).When both lines are detected, the abundances derived from the two lines are generally in reasonable agreement.Just for one star (RVS734), the two lines provide abundances at almost 0.5 dex difference, but the S/N in this case is low (S/N=16 per pixel).Three stars in the sample (RVS709, RVS714, and RVS727) show a [Zn/Fe] < −0.3.The star RVS709 is also poor in Cu ([Cu/Fe] = −0.82; the lowest value in the sample), while the other two stars do not have any Cu detected.These three stars are good candidates as pair-instability supernova (PISN) descendants (see Salvadori et al. 2019;Caffau et al. 2023;Aguado et al. 2023a).The low Zn abundance could also point to the fact that these stars were accreted from a galaxy of such small mass to have allowed for a low enough star formation rate that the birth of massive stars was inhibited (see Mucciarelli et al. 2021).
We derived the barium abundances by using line profile fitting.We used three Ba ii lines: 585 nm, 614 nm, and 649 nm.For the three lines, as recommended by Heiter et al. (2021), we used the oscillator strengths of Davidson et al. (1992).Following Korotin et al. (2015), we adopted hyperfine splitting only for the 649 nm line, and we adopted that in Heiter et al. (2021).For all three lines, we interpolated the NLTE corrections in the table of Korotin et al. (2015).We were able to derive the barium abundance for 42 stars.We investigated Europium through line profile fitting of the Eu ii line at 381.9, 412.9, 664.5, and 742.6 nm.For 35 stars in the sample, we were able to derive A(Eu).The [Eu/Fe] ratio (with Fe from Fe ii lines) is always positive with ⟨[Eu/Fe]⟩ = 0.65 ± 0.21.
We decided to use the two stars with differences in radial velocity from the two spectra in order to check the precision of this set of data and separately analyse the two spectra.For RVS712, the average difference in the abundances we obtained from the two spectra is close to zero (0.02 ± 0.06 dex), with the differences in the range [−0.1, 0.1] dex.Also for RVS736, the average difference we obtained by analysing the two spectra is close to zero (0.02 ± 0.08 dex), with the differences in the range [−0.1, 0.15] dex.For Zn, whose abundance is based on the Zn i line at 481.0 nm, both spectra have an S/N around 15.We concluded that the precision of the chemical investigation is between 0.1 and 0.2 dex, according to the S/N of the spectra, which are summarised in Table 2.

Comparison with the RVS spectra
We retrieved from the Gaia archive the RVS spectra for two stars in our sample: RVS719 and RVS733.There are no astrophysical parameters for these two stars in Gaia DR3.Both spectra have an S/N per pixel between 20 and 30 at 855 nm, which are low for middle-resolution infra-red spectra.We analysed the RVS spectra to derive Fe and Ti abundances.
For RVS719, from four Fe i lines, we derived [Fe/H] = −1.26± 0.17, to be compared to [Fe/H] = −1.21± 0.22 from the complete UVES spectrum.From one single Ti i line, we derived [Ti/H] = −0.63, to be compared to [Ti/H] = −0.70 ± 0.12 from 25 lines from the UVES spectrum.The Ti i line investigated in the RVS spectrum was rejected in the UVES analysis because it is too strong, but it would provide [Ti/H]=-0.66.
For RVS733, we could only derive [Fe/H] = −1.32 ± 0.45 from four Fe i lines, to be compared to [Fe/H] = −1.67 ± 0.19 derived from the complete UVES spectrum.The agreement is well within the uncertainties.

Discussion
Three stars in the sample (RVS708, RVS709, RVS740) seem lower than the others in [Mg/Fe] and [Ca/Fe], while RVS726 has a low [Ca/Fe] but no Mg detection, and RVS745 has low [Mg/Fe] and [Ca/Fe] ratios, but this is consistent with other stars.This conjunction of low α elements and low [Zn/Fe] and [Cu/Fe] ratios makes RVS709 a perfect candidate as a PISN descendant.Salvadori et al. (2019) investigated from a theoretical point of view the chemical characteristics expected in a PISN descendant and highlighted that the effect is most evident at the metal-poor regime (with the most evident effect at −2 ≲ [Fe/H] ≲ −1) and overall on nitrogen, copper, and zinc.Since Cu and Zn abundance determinations are rare among chemical investigations in metal-poor stars, Aguado et al. (2023a) built a way to use all the abundances to select stars that can be PISN descendants and eventually observe them to derive A(Cu), A(Zn), and possibly also A(N).Recently, Xing et al. (2023) discovered another PISN descendant in the Galaxy.In Fig. 7, the [X/Fe] ratios (derived from neutral species) are compared for a Zn-poor star (RVS709), a Zn-normal star (RVS728), and a Zn-rich star (RVS740).The  star RVS709 shows a systematic [X/Fe] ratio that is lower with respect to the other two stars.In Fig. 7, we also show the theoretical yields of Heger & Woosley (2002) for three masses, 168, 195, and 242 M ⊙ , that end their lives as PISNs.It is clear from this comparison that none of these stars may have been formed from pure PISN ejecta.If PISN ejecta are diluted with gas polluted by less massive supernovae (SNe), then patterns like that of RVS709 can be reproduced (see Aguado et al. 2023b, for an extensive discussion).
Europium is a pure r−process element, while barium is mainly an s−process element (Arlandini et al. 1999;Prantzos et al. 2020).The ratio Eu/Ba can be used as a diagnostic of the relative importance of the two processes in any stellar population.In Fig. 8, we show [Eu/Ba] as a function of [Fe/H] for our programme stars as well as for the First Stars sample (Cayrel et al. 2004b;François et al. 2007;Siqueira Mello et al. 2013), complemented with two measurements from Sneden et al. (2009) and Hill et al. (2017), labelled FS+ in the plot.The FS+ stars show a constant [Eu/Ba] ratio of 0.52 with a tiny scatter of 0.09 dex.This strongly suggests that both Eu and Ba are synthesised in the same process.This value is not far from the prediction of Wanajo (2007) of 0.62 for the hot r−process (see also Siqueira Mello et al. 2013, Fig. 15).Of the five stars with metallicities above -1.3,three have [Eu/Ba]< 0. We tentatively interpreted this as a sign of Ba s−process production in AGB stars.For lower metallicity stars, however, the mean is very close to that of the FS+ sample, albeit with a larger scatter that is compatible with the lower S/N of our spectra with respect to those of the FS+ sample.This does suggest that in the most metal-poor part of the sample ([Fe/H] < −1.3), both Eu and Ba are formed only by the r−process, and by inference, we may extend this hypothesis to all the neutron capture elements.Caffau et al. (2020) and Bonifacio et al. (2024) investigated a sample of stars selected by their high speed with respect to the Sun.They required in Gaia DR2 a transverse velocity higher than 500 kms −1 .The sample of stars we investigated was selected according to their high radial velocity in order to verify their reliability (discussed in Katz et al. in preparation).All of these stars (of this paper and the stars discussed in Bonifacio et al. 2024) are then expected to be halo stars.All of these selections collected mainly metal-poor stars, with a paucity of extremely metal-poor stars, and in the case of the stars analysed in this work, the most metal-poor star (RVS713) has [Fe/H] = −2.94.It is true that extremely metal-poor stars ([Fe/H] < −3) are rare objects, but the samples containing the stars we investigated are quite large (about 400 stars), and just a handful of stars (in Bonifacio et al. 2024) have [Fe/H] < −3.
Another similarity with the investigations of Caffau et al. (2020) and Bonifacio et al. (2024) is the presence of stars compatible with a young age.The two stars RVS725 and RVS726 seem to be younger than 1 Ga, and they both are expected to have large masses.As discussed in Bonifacio et al. (2024), these stars can be evolved blue stragglers that resulted from the merger of three stars (in the case of RVS725, perhaps even four).The formation of a blue straggler by the merging of three stars is described by Meyer & Meyer-Hofmeister (1980), but as stated by Fiorentino et al. (2014), massive blue stragglers are very rare objects.To further support the small probability of forming a massive blue straggler from a triple system, one can first consider the low fraction of triple systems among solar-type stars (9% ± 2%, Raghavan et al. 2010) and the low probability that in a destabilised triple system, a collision involving all three stars takes place (0.6%, Toonen et al. 2022).On the other hand, collisions involving only the inner pair are relatively common, up to 24% (Toonen et al. 2022).But these latter numbers are only for a destabilised triple system, which are only a fraction of all triple systems.These considerations explain why blue stragglers with masses larger than twice the turn-off mass are rare.Thus, the possibility must be confronted that these stars are really young, formed in a recent accretion event at the moment of the infall of the accreted object (a dwarf galaxy or a cluster) in the Milky Way.If so, the age of these stars would provide us with the time of the accretion events.
The 43 stars we chemically investigated are all metal-poor ([Fe/H] < −1), but none is a CEMP star.This is surprising, as one would expect 21% of the stars to be CEMPs at the very metal-poor regime, [Fe/H] < −2 (Lucatello et al. 2006).Clearly the numbers are small, but one would expect to see at least one star rich in carbon.

Conclusion
The chemical analysis of our sample of stars shows, by and large, the typical patterns found in other samples of metal-poor stars.What stands out is (i) there are up to three candidate PISN descendants with low α elements, Cu and Zn, and (ii) two are apparently young stars that can hardly be interpreted as blue stragglers without invoking the merging of multiple systems.
We confirm the small number of extremely metal-poor stars among samples of high-velocity stars.We point out that RVS721 lies within 10 ′′ of the track of the Gjoll stream (Ibata et al. 2019) associated with the globular cluster NGC 3201.The metallicity of the star is −1.66±0.16dex, which is a good match with the catalogue metallicity of the cluster, −1.59 dex (Harris 1996), thus supporting the association.and the only significative peak is evident at 120.48 days, which is quite high for an RS CVn, since most of these stars have periods below 20 days (Martínez et al. 2022).We tried to phase the data with this period and were unable to obtain a smooth light curve.It is probable that the timeseries is too short and sparsely sampled to provide an accurate period.Under these conditions, it is impossible to understand in which phase our spectrum was observed.However, the whole light curve spans a range of 0.05 mag in G BP −G RP , corresponding to less than 100 K.This is an additional error to be considered on our adopted effective temperature.

A.2. RVS708
This star has been classified as a long period variable by Heinze et al. (2018), named as ATO J283.9396-17.3013,as well as by Gaia DR3.Heinze et al. (2018) provided a period of 315.971905 days from a Lomb-Scargle periodogram.We tried to phase the Gaia G epoch photometry (24 points) with this period but failed to obtain a smooth light curve.The G BP − G RP colour spans a range of 0.07 mag in the Gaia DR3 epoch photometry.This corresponds to about 150 K and has to be considered an additional error on our adopted effective temperature.Our spectrum shows clear emissions in the Ca ii H and K lines.

A.3. RVS719
For this star, there is no information in the literature.We computed the Lomb-Scargle periodogram for the G epoch photome- try but could not identify any statistically significant peak.The G BP −G RP colour spans a range of about 0.09 mag, corresponding to almost 200 K.This is an additional uncertainty on our adopted effective temperature.

A.4. RVS724
We computed the Lomb-Scargle periodogram using the G epoch photometry, and it is shown in Fig. A.2.There is only one peak higher than the false alarm probability of 10 −3 , corresponding to a period of 84.45946 days.We phased the G epoch photometry with this period, and the result is shown in Fig. A.3.While the result is plausible, it is inconsistent with the precision of the Gaia photometry.This suggests that a longer timeseries is required to determine an accurate period, and we do not exclude that the star may be multi-periodic.This discourages us to use this curve to estimate the phase at which we observed the spectrum.In the epoch photometry, the G BP − G RP colour spans a range slightly over 0.3 mag, which corresponds to over 600 K.This implies that the effective temperature that we derived from the mean colours is very uncertain.This is clear also from the two spectra at our disposal that provide different abundances, very likely because they were observed at different phases and Article number, page 10 of 14 E. Caffau et al.: Gaia RVS benchmark stars II different effective temperatures.Our spectra display a considerable level of activity with the presence of several emission lines.

E
Fig. 1.Four spectra shown in the wavelength range of the Gaia RVS.

Fig. 3 .
Fig. 3. Observed stars in the (T eff ,log g) diagram compared to PARSEC isochrones of metallicity -1.5 dex and of ages 12 Ga (blue crosses) and 1 Ga (red crosses), to guide the eye.The [Fe/H] of the stars are colour coded according the vertical bar on the left.

Fig. 5 .
Fig. 5. [Mg/Fe] versus [Fe/H] for the observed sample (blue circles).The young star is shown as a green dot inside a black circle, and the PISN candidates are shown as light-blue dots inside black squares.The comparison samples are the more metal-poor FS sample from (Cayrel et al. 2004a, red squares) and the sample of similar metallicity from Caffau et al. (2023, violet squares).
Fig. 8. [Eu/Ba] versus [Fe/H] diagram.For our stars, the NLTE corrections to Ba have been applied.For comparison, we added stars from the First Stars sample (see text), labelled FS+.The solid horizontal line corresponds to the mean value of all the FS+ values.

Fig
Fig. A.1.Lomb-Scargle periodogram of the G epoch photometry of RVS711.The dotted line shows the level of false alarm probability 1. −4

Fig
Fig. A.2. Lomb-Scargle periodogram of the G epoch photometry of RVS724 The dotted line shows the level of false alarm probability 1. −3

Table 2 .
Signal-to-noise ratio of the spectra.