Stellar Loci VI: An Updated Catalog of the Best and Brightest Metal-poor Stars

The Gaia criterion is very straightforward. We simply select candidate giant stars with [Fe/H] $<-2$ from Xu et al. (2022). But it has two limitations. One is that it cannot include stars with $|b|<10^{\circ}$. The other is contamination from more metal-rich stars due to the effects of binary or phot_bp_rp_excess_factor, as mentioned in Section 2.

Following Schlaufman & Casey (2014), we also use the WISE colors to select VMP candidates. However, differing from Schlaufman & Casey (2014), we have performed reddening corrections and use an updated criterion to improve the selection efficiency.

To determine the new WISE criterion, we use a selected sample of giant stars from LAMOST DR8 (Wang et al. 2022) that have high-quality 2MASS and ALLWISE photometry with errors smaller than 0.025 mag and reddening values smaller than 0.01 mag. These stars are dereddened using the Schlegel, Finkbeiner, & Davis (1998, hereafter SFD) dust reddening map. The reddening coefficients for the $K-W1$ and $W1-W2$ colors are empirically determined by Zhang et al. (2022, in preparation), which are 0.11 and 0.056, respectively. With the sample above, by considering both the selection efficiency and completeness, we set the new criterion to be ${(W1-W2)}_{0}>-0.392\times{(K-W1)}_{0}-0.017$, as shown in Figure 1; it yields a success rate of 46$\%$ and a completeness of 84$\%$ for the sample used.

With the new criterion, we cross-match the Gaia EDR3, 2MASS, and ALLWISE catalogs to first select giant stars, then the subset of these that are candidate VMP stars. For stars with $|b|\geq 10^{\circ}$, the giant stars are from Xu et al. (2022). and the SFD reddening map is used. For stars with $|b|<10^{\circ}$, the three-dimensional dust map from Chen et al. (2019) is used to do the reddening correction, which provides $E(G-K_{S})$, $E(BP-RP)$ and $E(H-K_{S})$. The giant stars are selected as those with $M_{K_{S}}<-1.5\times(BP-RP)_{0}^{2}+6.5\times(BP-RP)_{0}-3.8$, assuming that $A_{K_{S}}=1.987\times E(H-K_{S})$, as given by Yuan et al. (2013). Note there is also a cut of $d<6$ kpc, because of the distance limit of Chen et al. (2019). To avoid low success rates caused by large photometric errors, we also require that errors of the $Ks,W1$, and $W2$ bands are lower than 0.03, 0.025, and 0.025 mag, respectively. Note that the $W1-W2$ color can also be used to select M giant stars and estimate their metallicities (Li et al. 2016).

As mentioned above, some metal-rich stars are mis-classified as metal-poor stars with the Gaia photometric metallicities. Most metal-rich stars are disk stars, while most VMP ones are halo stars. Since the majority of disk stars and halo stars have different motions, we explore kinematics to exclude metal-rich stars from candidate Gaia VMP stars.

Figure 2 shows the Gaia tangential velocities, as a function of [Fe/H], for stars over different Galactic longitude ranges. Note that the unit for tangential velocities is mas kpc year^-1, which can easily be converted to km s^-1 by multiplying by 4.74. Here the stars are from the giant test sample of Xu et al. (2022). One can see that most metal-rich ([Fe/H] $>-1$) stars have tangential velocities lower than 40 mas kpc year^-1, but metal-poor stars ([Fe/H] $<-1$) have a much wider range, from 0 to 140 mas kpc year^-1. Therefore, one can make a cut on tangential velocities to remove of the most metal-rich stars. The limits (red lines in Figure 2) are different: 30 mas kpc year^-1 for sources within $-60^{\circ}<l<60^{\circ}$, 20 mas kpc year^-1 for sources with $120^{\circ}<l<240^{\circ}$, and 25 mas kpc year^-1 for sources with $60^{\circ}<l<120^{\circ}$ or $240^{\circ}<l<300^{\circ}$. Such a kinematic criterion can be used together with the Gaia and WISE criteria to select cleaner VMP samples, albeit with the introduction of a kinematic bias. Note that a small number of VMP stars with disk-like orbits are likely excluded by this criterion.

There are three criteria in our work, as described above. For sources with $|b|>10^{\circ}$, we define a Gold sample, two Silver samples, and two Bronze samples, corresponding to the confidence we place in the selection of VMP stars. The Gold sample satisfies all three criteria. The Silver GK sample satisfies the Gaia and Kinematic criteria, and the Silver GW sample satisfies the Gaia and WISE criteria. The Bronze G sample satisfies the Gaia criterion only. The Bronze WK sample satisfies the WISE and Kinematic criteria. For sources with $|b|<10^{\circ}$, we use the WISE and kinematic criteria to select VMP stars, and refer to them as the Low $b$ sample. The samples and their selection criteria are summarized in Table 1.

By cross-matching with results derived from medium-resolution ($R\sim$ 1800) from LAMOST DR8, we establish that the success rate for identifying VMP stars is 60.1$\%$ for the Gold sample, 39.2$\%$ for the Silver GW sample, 41.3$\%$ for the Silver GK sample, 15.4$\%$ for the Bronze G sample, 31.7$\%$ for the Bronze WK sample, and 16.6$\%$ for the Low $b$ sample, respectively.

The $G$-magnitude distributions for the six samples are shown in Figure 3. All of the candidate stars are brighter than 15 in the $G$-band, which are relatively easy to access with high-resolution follow-up spectroscopic observations. There is a peak around $G\sim 14$ in the Gold, Silver GW, and Bronze WK samples, due to the cut on photometric errors in the WISE criterion.

Spatial distributions of the six samples in the Galactic coordinate system are shown in Figure 4. The sources are roughly evenly distributed over high Galactic latitude areas, and slightly increase toward the Galactic center direction. For the Bronze G sample, there is a strong over-density in the dense Galactic center and disk regions, mainly because there are more metal-rich stars mis-classified as VMP stars. Such an over-density is significantly reduced by application of the Wise and Kinematic criteria. For the Gold sample, there is still a weak over-density in the Galactic center direction, which is likely real to some extent, because we expect to have more VMP stars there compared to those near the Sun’s position in the Galaxy.

We cross-match our six samples with LAMOST DR8 (Wang et al. 2022) in order to test the success rates of our various criteria. LAMOST DR8 has provided values of stellar atmospheric parameters for 5.16 million unique stars, including reliable estimates of [Fe/H] down to $\sim-3.5$ (see Figure A4 of Wang et al. 2022).

There are 2764, 4719, 6153, 18353, 8377, and 193 stars found in common for the Gold, Silver GW, Silver GK, Bronze G, Bronze WK, and Low $b$ samples, respectively, after a signal-to-noise cut in the $g$-band ($S/N_{g}>20$). Their [Fe/H] distributions are plotted in black in Figure 5. The Bronze G, Silver GW, and Silver GK samples show two clear peaks, one for the VMP stars, the other for the metal-rich stars. The metal-rich peak is most prominent in the Bronze G sample, as explained in Section 2. The metal-rich peak becomes weaker and weaker in the Silver GW and GK samples, and disappears in the Gold sample. The result suggests that the WISE and Kinematic criterion, particularly the latter, work well in removing contamination from metal-rich stars. There are quite large fractions of metal-poor stars ($-2<$ [Fe/H] $<-1$) in the Bronze WK and Low $b$ samples, both of which have used the WISE criterion. This is perhaps not surprising, given the relatively large errors in the WISE colors. Note that there is a dip at [Fe/H] $\sim-1.5$ in all samples, most prominent in the Bronze WK sample. This arises because, in LAMOST DR8, the metallicities for stars with [Fe/H] $>-1.5$ and $<-1.5$ are measured by two different pipelines (see Section 5 of Wang et al. 2022).

For a given sample, we define its success rate for selecting stars with [Fe/H] lower than a given limit as the ratio of DR8 stars in common with [Fe/H] lower than the limit, relative to the total number of stars in common. The results are listed in Table 4. For the Gold sample, the success rate is 93.0%, 83.1%, 60.1%, and 16.2% for stars with [Fe/H] lower than $-1$, $-1.5$, $-2$, and $-2.5$, respectively. For the two Silver samples, the success rate is somewhat lower, about 40% for stars with [Fe/H] $<-2$. It decreases to 31.8% for the Bronze WK sample, 15.4% for the Bronze G sample, and 16.6% for the Low $b$ sample, respectively.

The Bronze G sample shows the lowest success rate in all cases, due to the much larger number of metal-rich binary stars compared to VMP stars. To increase its success rate, we use the Renormalized Unit Weight Error (RUWE; Lindegren et al. 2021b) to exclude binaries with poor astrometric solutions. The sources with larger RUWE values are more likely binaries. We find that an empirical cut of $RUWE<1.1$ (red lines in Figure 5) can well-remove metal-rich binaries for the Gaia criterion³³3 Note the cut on RUWE is only used to remove binaries, regardless of their metallicites. However, as most binaries are metal-rich in terms of numbers, it seems that the cut preferentially removes metal-rich binaries.. With this simple additional criterion, the success rate of VMP stars increases to 63.1$\%$ for the Gold sample, about 50$\%$ for the Silver samples, and 29.3$\%$ for the Bronze G sample, as listed in Table 4.

Figure 6 plots the success rate of VMP stars identified in different sky areas. The success rate is relatively larger in the higher Galactic latitude regions due to less contamination. For samples using the Gaia criterion, the success rate in the Southern Galactic Hemisphere is lower than that in the Northern Galactic Hemisphere, likely due to the different systematics in the SFD reddening map (Schlafly et al. 2010; Sun et al. 2022). Systematic errors in the SFD map result in systematic errors in the Gaia metallicities.

We also cross-match our six samples with SEGUE DR12 (Alam et al. 2015) in order to test the success rates of our various criteria. The numbers of stars in common are limited, due to the bright limit for SDSS ($g\sim 14$) – only 49 stars for the Gold sample and 323 stars for the Bronze G sample. We find that the overall success rates are slightly better than those with LAMOST DR8, because most sources from SEGUE DR12 have $|b|>20^{\circ}$, where higher success rates are found in the test with LAMOST DR8.

To analyse the success rate over the full sky, metallicities from the Apache Point Observatory Galactic Evolution Experiment(APOGEE; Majewski et al. 2017) Data Release 17 (Abdurro’uf et al. 2022) are also used. There is an offset between the LAMOST and APOGEE metallicities, so we cross-match LAMOST DR8 giants of $S/N_{g}>20$ and $T_{eff}>4400$ K and those of APOGEE DR17 with $S/N>100$ to correct for this. Figure 7 plots [Fe/H] _LAMOST vs. Fe/H] _APOGEE. The linear fitting result of [Fe/H] _LAMOST = 0.62 $\times$ Fe/H] _APOGEE $-$ 0.94 is over-plotted as the black line. The APOGEE metallicities are converted into LAMOST metallicities using the above relation.

We cross-match our six samples with APOGEE DR17, and only sources with $S/N>50$ are used in subsequent studies. The result is plotted in Figure 8. The trend is similar to Figure 6. Note the much lower success rate in the region $120^{\circ}<l<240^{\circ}$ and $-40^{\circ}<b<-10^{\circ}$, due to systematics in the SFD map.