METAL-POOR LITHIUM-RICH GIANTS IN THE RADIAL VELOCITY EXPERIMENT SURVEY^*

The abundance determinations were achieved with the MOOG analysis program (Sneden 1973), using one-dimensional, plane-parallel Kurucz model atmospheres²² under the assumption of static equilibrium and LTE. Stellar parameters were derived following a variation of the methods described in Ruchti et al. (2010, 2011). The initial effective temperature, T_ini, was set by using the excitation temperature method based on Fe i lines. The initial value of the surface gravity (log g_ini) was set using the ionization equilibrium criterium utilizing the iron abundance derived by both Fe i and Fe ii lines. The initial [Fe/H] value of the stellar atmosphere for each star was chosen to match the [Fe ii/H] value derived from the analysis. The value of the microturbulent velocity (v_t) was set to minimize the magnitude of the slope of the relationship between the iron abundance derived from Fe i lines and the value of the reduced width of the line.

In Ruchti et al. (2010, 2011), we found during our analysis of the spectra for several globular cluster giants and giant stars selected from the Fulbright (2000) sample that the effective temperature estimate from the excitation method showed an offset that correlated with [Fe/H]_ini, when compared to photometric temperature estimates (using the Two Micron All Sky Survey, 2MASS, color–temperature transformations of González Hernández & Bonifacio 2009). We have since found similar results using the color–temperature transformations of L. Casagrande et al. (2011, private communication). We therefore applied the temperature corrections described in Ruchti et al. (2011) to the Li-rich candidates in our current sample. The analysis described above was then performed again, but with the effective temperature now forced to equal the corrected temperature estimate, T_eff. The final adopted values of the stellar parameters for our Li-rich stars are given in Table 2. The error in effective temperature, $\sigma _{T_{\rm eff}}=140$ K, and [Fe/H], σ_[Fe/H] = 0.10 dex, were adopted from Ruchti et al. (2010, 2011). We estimated an error in surface gravity of 0.3 dex; however, the error could be larger for the lowest gravity stars (as is described below).

The method of using ionization equilibrium to derive the surface gravity is believed to be unreliable in VMP stars due to non-LTE effects (Thevenin & Idiart 1999; Kraft & Ivans 2003). The abundance derived from the Li i line, however, is nearly independent of the adopted surface gravity. In Ruchti et al. (2010, 2011), it was assumed that no giants lie above the RGB tip, but we do not make this assumption here since the evolutionary stage of our giants will affect the interpretation of our abundance results. We therefore did not apply any of the corrections described in Ruchti et al. (2011) to the surface gravity of our Li-rich candidates. Most of our Li-rich candidates have log g > 1.0, values which were not corrected in Ruchti et al. (2010, 2011). Three stars have log g = 0.6–0.8, for which the correction would only be 0.1–0.2 dex, well within our errors. The star T9112-00430-1 has the lowest gravity, suggesting it lies far above the RGB tip (see Figure 1). The derived values of the stellar parameters (specifically gravity) of this star are most likely affected by large non-LTE effects. If we were to increase its gravity estimate by 0.5 dex, as prescribed by Ruchti et al. (2011), it would lie close to the RGB tip.

Zoom In Zoom Out Reset image size

The majority of our Li-rich candidates have [Fe/H] < -1.8. The most metal-rich star, T5496-00376-1, has [Fe/H] = -0.63, which is much more metal-rich than the rest of the candidates but is at the lower end of most previous studies. We include it here because of its Li-rich nature. Our estimates of T_eff and [Fe/H] for M3-IV101 strongly resemble those found by Kraft et al. (1999), showing differences of only 36 K and 0.02 dex in T_eff and [Fe/H], respectively. Our log g value, however, is about 0.2 dex lower than their value. This offset is very similar to the correction that would be made to our log g value if we were to follow the analysis in Ruchti et al. (2011). Our [Fe/H] estimate for M68-A96 also agrees within ∼0.02 dex with metallicity estimates for the M68 cluster (Lee et al. 2005).

The luminosity of each star was estimated by fitting to Padova isochrones (Marigo et al. 2008; Girardi et al. 2010). The Z-metallicity of each star was derived by combining [Fe/H] and alpha-enhancement using the transformation of Salaris et al. (1993). We assumed an alpha-enhancement equal to [Mg/Fe] measured for each star (see Section 4.3). We then fit each star to the isochrone with the closest matching metallicity in a grid of 12 Gyr isochrones with metallicity steps of Z = 0.00002. It is possible that some of our stars are younger (especially T5496-00376-1), but the error in the luminosity due to our uncertainty in log g far outweighs this. Figure 1 shows each star and the isochrone of the same Z-metallicity in the gravity–temperature plane.

The luminosity at the point on the isochrone with the same log g value as the star being fitted was chosen as the luminosity of the star, except for T9112-00430-1. This star has a gravity above the limits of the isochrone (see Figure 1). We therefore adopted the luminosity at the point of lowest gravity on the isochrone. We linearly interpolated the luminosity values versus log g for stars with log g values lying between the points on the corresponding isochrone. We further estimated two extreme luminosity values for each star by adding and subtracting 0.3 dex to each star's log g value and then fitting again. Errors were then estimated by taking the difference between these extreme values and the value taken from the original fit. This resulted in a typical error in the luminosity of each star of $\sigma _{\log (L/L_{\odot })}\sim 0.3$. Most stars lie between the RGB and AGB branches of the isochrones, making it difficult to label them as one or the other from inspection of Figure 1 alone. The estimated luminosity from fitting to either branch showed no differences. We discuss the phase of evolution of each star in more detail later (see Section 5). Our values for the luminosity of each star are listed in Table 2.

Given these luminosities, we followed the methodology described in Ruchti et al. (2011) to determine in which Galactic population each of the Li-rich giants most likely belongs. The final population assignment for each Li-rich candidate is given in Table 2. Note that the same population assignment was found using the stellar parameters derived in this work for the four stars (J142546.2-154629, J195244.9-600813, T5496-00376-1, and T6953-00510-1) that were also analyzed in Ruchti et al. (2010, 2011). All stars were assigned to the thick disk, halo, or thick/halo intermediate population (see Ruchti et al. 2011 for more details). This implies that the Li-rich giants in our sample are old (>10 Gyr) and metal-poor. Thus, they most likely have low masses, M < 1 M_☉.

The Li abundances for each star were derived assuming that all the Li was from the ⁷Li isotope. EWs were measured for both the 6708 Å and 6103 Å Li i lines. We then derived the abundance of Li from both lines in each of LTE and non-LTE following the methods described in Lind et al. (2009a). These values can be found in Table 3.

The derived A(Li) values for our Li-rich sample are shown in Figure 2 as a function of the stars' estimated luminosity (calculated above). Note that, for stars that had A(Li) estimates from both Li i lines, the value in the figure is the mean of the two values. Further, we found a typical error of ∼0.20 dex from the difference between that found for the 6708 Å and 6103 Å lines. Figure 2 also includes the Li abundances we derive for a sample of 58 RAVE very metal-poor (hereafter RAVE-VMP) stars with [Fe/H] <−2 from Fulbright et al. (2010). Note that 26 measurements are only upper limits. We followed the same analysis procedure given above for all RAVE-VMP data, and the full results for the entire sample will be published in a later paper.

Zoom In Zoom Out Reset image size

Note that both J043154.1-063210 and J195244.9-600813 have luminosities and Li abundances that place them along the trend of "Li-normal" giants near the RGB bump. We therefore no longer classify them as Li-rich. The remaining Li-rich giants have Li abundances that clearly separate them from the Li-normal giants (see Figure 2). As was found in previous studies, the majority of our Li-rich giants separate into two regions: the lower (near the RGB bump) and upper RGB, separated at log (L/L_☉) ∼ 2.4 (see also Figure 1). T6953-00510-1, however, appears to lie between these two regions.

We determined the CNO abundances for our giants using MOOG, under the assumption of molecular equilibrium, since the CNO atoms can be partly bound together in molecules, especially for cooler stars (see, e.g., Gratton & Sneden 1990).

Oxygen abundances were first determined from the EWs of the O i forbidden lines at 6300 Å and 6363 Å, and are given in Table 3. The gf-values of the lines were taken from Lambert (1978), and the solar abundance, A(O) = 8.69, was selected from Asplund et al. (2009). We corrected the O i 6300 Å line for the weak Ni i 6300.34 Å line (see Allende Prieto et al. 1999) following the same methodology as Fulbright & Johnson (2003).

We next determined [C/Fe], [N/Fe], and the ¹²C/¹³C ratios for our giants by spectral syntheses of CH and CN lines using the Plez line lists and oscillator strengths (B. Plez 2011, private communication; see also Hill et al. 2002; Gustafsson et al. 2008) combined with the Vienna Atomic Line Database (VALD;²³ Kupka et al. 2000). We adopted dissociation energies of 3.47 eV (Huber & Herzberg 1979) and 7.66 eV (Lambert 1978) for CH and CN, respectively. The CH lines between 4320 and 4328 Å were used to estimate the C abundance of each star given the value of A(O) found above. Given these values of A(O) and A(C), the N abundance was determined using CN lines at the 3883 Å band (and the 4216 Å band, see below). Finally, the ¹²C/¹³C isotopic ratio was constrained by combining information from the above lines with that derived from fits to CH and CN lines in the wavelength range 4200–4220 Å. We then iterated this procedure with the input CNO abundances equal to the previous iteration until we obtain a self-consistent solution to all three abundances.

During the syntheses of C and N, the Li-rich candidates fell into two classes: those with a strong CN-3883 band and those with a weak one. The final syntheses are shown in Figures 3 and 4, respectively. The lines in the CN-strong stars could very well be saturated (most obvious in the M3-IV101 spectrum). We therefore included the 4216 Å CN band in our syntheses to constrain the N abundance for these stars. Although the CN-4216 band is very weak for the CN-weak stars, we include it in Figure 4 for comparison.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The resultant [C/Fe] and [N/Fe] values from the syntheses are given in Table 3 and plotted versus log g in Figure 5. Note that we adopted the solar abundances for C and N from Asplund et al. (2009). The scatter in individual [C/Fe] and [N/Fe] values is due to variations between stars, not the quality of the determination (we investigate the interpretation of this scatter in a separate paper). We estimated the internal errors to be ∼0.1 dex in [C/Fe] and ∼0.1–0.2 dex in [N/Fe], depending on the quality of the spectrum and the strength of the CN bands. We further varied the estimated values of the stellar parameters of each star according to their 1σ errors to investigate the sensitivity of our syntheses to the chosen values of the stellar parameters for each star. Our stars consistently show errors of ∼0.1 and ∼0.1–0.2 dex in [C/Fe] and [N/Fe], respectively, and show the highest sensitivity to changes in the effective temperature of a star. Using the same error analysis as for C and N, we found that our errors in [O/Fe] are about 0.10 dex. It is important to note, however, that the EWs for C1012252-203007, J043154.1-063210, M68-A96, and T8448-00121-1 are less than 10 mÅ for both lines, and so the error in [O/Fe] increases to ∼0.15–0.20 dex. The ¹²C/¹³C isotopic ratio was very difficult to constrain since most of our stars were fairly deficient in carbon (see below) and so the ¹³CH features were very weak. The best-fit value for most stars subsequently had a fairly flat likelihood peak. We therefore give a range of values, as well as the best-fit value, in Table 3.

Zoom In Zoom Out Reset image size

The CNO abundances of M3-IV101 have been analyzed previously in the literature, which makes it useful for comparisons. We only compare the abundance values, since there will be offsets in the element ratios with iron due to differences in reference solar abundances. Kraft et al. (1999) found using CH lines that A(C) ∼ 6.55 for this star, while Pilachowski et al. (2003) found a value of ∼6.31 using CO lines. Our value of [C/Fe] = -0.5 corresponds to A(C) = 6.4, which lies between the Kraft et al. and Pilachowski et al. values. The variation in the values could be attributed to differences in atomic and molecular line data. On the other hand, we found values of A(O) = 7.41 and A(N) = 7.61, which are only 0.01 dex and 0.06 dex higher than those found by Kraft et al. (1999), respectively. Note that Pilachowski et al. (2003) did not measure N abundances in their work. They did, however, measure the carbon isotopic ratio, finding ¹²C/¹³C =11, which is very close to our best-fit value given in Table 3.

The final value of [C/Fe] for the Li-rich giants ranges from −0.7 for C1012254-203007 and T9112-00430-1 to +0.3 for J043154.1-063210, while all stars have [N/Fe] and [O/Fe] values greater than zero. In Figure 5, we also plot the ratios found for the RAVE-VMP comparison sample described in Section 4.1. As illustrated in the figure, all Li-rich giants fall within the general trend of the RAVE-VMP stars. The [C/Fe] and [O/Fe] ratios appear to possibly increase with gravity, while the [N/Fe] ratio may decrease with gravity. We computed Spearman's rank correlation coefficient, r_s, to investigate the level of correlation between all three ratios and log g. We found that [C/Fe] weakly correlates with gravity, with a value of r_s ∼ +0.5. However, we found values of r_s ∼ +0.3 and −0.3 for the [O/Fe] and [N/Fe] ratios, respectively, which implies no significant correlation. The weak trend in [C/Fe] is a sign that the lower gravity stars have been affected by slightly more CNO cycling and internal mixing (see also Section 5). The Li-rich giants at large gravity (log g > 1.5) show the possibility of a slight enhancement in [N/Fe] as compared to the RAVE-VMP stars, but this is not conclusive given the errors. We will discuss these trends in more detail for the entire RAVE-VMP sample in a later paper, but the main point to take away is that the Li-rich giants and RAVE-VMP stars appear to have experienced very similar CNO cycling.

All of the Li-rich giants have a ratio of the number of C atoms to O atoms, log (C/O) < 0, which can be an indicator of the nature of their evolution should any of them be AGB stars (see Section 5.3 for more details). M3-IV101 and T6953-00510-1, however, have log (N/O) ≳ 0. Further, like in [N/Fe], the Li-rich giants at large log g show indications of a slight enhancement in log (N/O) as compared to the RAVE-VMP stars. Note that T6953-00510-1 is the most enhanced in [N/Fe] and [O/Fe] and J043154.1-063210 has the highest enhancement in [C/Fe] among the Li-rich giants.

We have used the line lists of Fulbright (2000) and Johnson (2002) to measure the abundances of other elements for these stars, including Mg, Na, and several neutron-capture elements, for which the ratio with iron is given in Table 4. Hyperfine splitting effects were taken into account for the Na i D lines and the lines of Ba ii and Eu ii. Solar values for all elements were again selected from Asplund et al. (2009). The neutron-capture elements can be especially important since they can indicate the possible presence of dredge-up in AGB stars.

The derived abundance ratios for these elements are shown versus [Fe/H] and log g in Figures 6 and 7, respectively, for our Li-rich candidates. We further plot the abundance ratios of our Li-normal VMP giants. Most of our stars have ratios that resemble that of the Li-normal VMP giants of like metallicity and gravity. The Li-rich giant T6953-00510-1, however, shows enhancement in the s-process elements (Sr, Y, Zr, Ba, La, Pb) as compared to the other stars. Further evidence for s-process enhancement is found in the top plots of Figure 8, in which T6953-00510-1 shows enhancement in [Ba/Eu], which gives the ratio of s-process to r-process in a star. Other ratios, such as [Y/Zr], [Y/Ba], and [Ba/Eu] (see Figure 8), are similar between the Li-rich giants and Li-normal VMP stars.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Since T6953-00510-1 is the only star with enhanced s-process abundances, this enhancement is probably not connected to the mechanism that is enriching the Li in our stars. Further, this star is not located near the TP (thermally pulsing)-AGB evolutionary phase (see Figure 1) where we would expect dredge-up of s-process enriched material. The simplest explanation is that this star is a part of a binary system in which its atmosphere has been polluted by a higher mass companion that has already gone through its TP-AGB phase. We do not detect any variation in the radial velocity of this star (see Section 4.5), but this scenario is still possible if the binary system is wide. It is also possible that T6953-00510-1 formed from s-process enriched material left by a star of an early generation that had gone through its TP-AGB phase, but this scenario would require extremely incomplete mixing of the interstellar medium prior to the next generation of stars.

Lee et al. (2005) measured abundances for seven stars in M68, but they did not include M68 A-96. If we use their line list and follow their analysis methods, we measure abundances for this star nearly identical to their mean cluster values.

Some have found that many of their metal-rich Li-rich giants had high projected rotational velocities, vsini (e.g., Guillout et al. 2009). Further, Drake et al. (2002) suggested that the fraction of Li-rich stars can be as high as 50% among rapidly rotating giants. We therefore computed vsini for the Li-rich giants in our sample to determine if any are rapid rotators. We derived vsini following the methodology of Fekel (1997) and Hekker & Meléndez (2007). First, the measured stellar broadening, X_meas, was estimated as the average of the full width at half-maximum (FWHM) of several Fe i lines near 6750 Å. The FWHM of several ThAr lines (from arc spectra taken during the night of each observation) in the same wavelength region as the Fe i lines were averaged to estimate the instrumental broadening, X_inst. The intrinsic broadening can then be estimated as

$$\begin{equation} X_{\rm intr}=\sqrt{X_{\rm meas}^2-X_{\rm inst}^2}. \end{equation} \tag{ 1 }$$

We then determined the total broadening, X_tot, given our value of X_intr for each star, using the second-order polynomial fit to X_intr vs. X_tot from Hekker & Meléndez (2007):

$$\begin{equation} X_{\rm intr}=0.10963 + 0.002758X_{\rm tot} + 0.001278X_{\rm tot}^2. \end{equation} \tag{ 2 }$$

The projected rotational velocity, vsini, of a given star can then be computed as $\ssty \sqrt{X_{\rm tot}^2-v_{\rm m}^2}$, where v_m is the macroturbulent velocity of that star. We adopted the relations between v_m and T_eff in Hekker & Meléndez (2007) for different luminosity classes to estimate v_m for each Li-rich giant. We assumed that the Li-rich giants in our sample near the RGB tip were class II and those near the RGB bump were class III. Note that those stars with estimated values of v_m greater than the total broadening are assumed to have a non-measurable rotational velocity.

The majority of our Li-rich giants showed no measurable rotational velocity. We found non-negligible values of ∼3 km s^-1 for T5496-00376-1 and ∼4 km s^-1 for T8448-00121-1, but these values are only slightly larger than the expected projected rotation of low-mass giants, vsini ⩽ 2 km s^-1 (de Medeiros et al. 1996). C1012254-203007 and T9112-00430-1, however, have vsini = 8–10 km s^-1. This implies that they very well could be rapidly rotating.

What would cause these stars to rapidly rotate? It is possible that extra angular momentum, dredged-up as mass is redistributed during increased convection, could induce increased rotation in these stars (Fekel & Balachandran 1993). Another possibility is that these stars have accreted material from a planet or companion star, which would also contribute extra angular momentum to the stars. If indeed the two rapidly rotating stars are in a binary system, signatures may be present in their spectra; this is discussed below.

None of the echelle data shows obvious line profile asymmetries or other spectral signs of having a bright secondary star in the system, but the light from a low-luminosity main-sequence or white dwarf companion would be swamped by the light from the bright giant star in the optical spectrum.

For most of our stars, we only have two radial velocity measurements: the initial RAVE DR3 (Siebert et al. 2011) observation (with a systematic uncertainty of about 3 km s^-1) and the echelle observation discussed here. RAVE observed two of the stars twice, and we have three echelle observations of C1012254-203007. All together, we have 21 radial velocity measurements of the eight RAVE Li-rich stars. The mean difference (RAVE−echelle) in the heliocentric radial velocities is 0.8 ± 1.8 km s^-1, which lies within the RAVE RV measurement error.

Both T9112-00430-1 and T6953-00510-1 shared the largest difference of 3.8 km s^-1 and 3.4 km s^-1, respectively, between two observations of the same star. These differences, however, are only ∼0.5–1.0 km s^-1 larger than the systematic uncertainty of the RAVE radial velocities. All other repeat observations have a velocity difference of less than twice the internal uncertainties.

We further searched the spectra of our candidate Li-rich giants for signs of possible mass loss, inspecting the strong photospheric lines in each spectrum, such as Hα, the Na D lines, and Ca ii H and K lines (Balachandran et al. 2000; Drake et al. 2002). The only two stars in our sample to show possible evidence of mass loss are T9112-00430-1 and J142546.2-154629. In both cases, we identified emission in the wings of Hα. However, the emission was less pronounced in the spectrum of J142546.2-154629. Both stars showed no other signs (in the Na D or Ca ii H and K lines) of mass loss in their spectra. The emission features could be an indicator that these stars are evolving on the AGB (see Section 5.3). Further, this mass loss could be a possible trigger for Li production in these stars (De la Reza et al. 1996, 2000); however, an investigation of infrared excess is needed to confirm this.

Early on in a low-mass star's climb up the RGB, it undergoes the first dredge-up on the RGB after its outer convective envelope reaches the shell-burning regions. Standard theories of low-mass stellar evolution predict that the ¹²C/¹³C ratio drops from over 60 to about 40 around the point of the first dredge-up (Gratton et al. 2000). The first dredge-up mixes fresh H on the surface layers to the interior of the star, which causes a strong molecular weight discontinuity at the deepest extent of the convective envelope. It has been suggested that this molecular weight discontinuity, combined with rotation, induces extra mixing, such as thermohaline mixing, that triggers Li enrichment at the RGB bump (Charbonnel & Balachandran 2000; Ulrich 1972; Kippenhahn et al. 1980; Charbonnel & Zahn 2007; Eggleton et al. 2008). Further, Charbonnel & Zahn (2007) found that for low-mass (M < 0.9 M_☉) metal-poor ([Fe/H] < -0.5) stars, carbon and ¹²C/¹³C are also reduced while the nitrogen abundance increases. These trends predicted by thermohaline mixing were shown by Angelou et al. (2011) to agree with observational data in the study of the CNO abundances in M3 stars.

The two Li-normal giants (J043154.1-063210 and J195244.9-600813) are most likely evolving along the RGB. J043154.1-063210 lies at a higher gravity than the RGB luminosity bump (see Figure 1). Should it be on the RGB, it is most likely going through its first dredge-up. J195244.9-600813, on the other hand, appears to lie on the RGB bump. It is possible that both stars are at the beginning of the RGB-bump mixing phase. This would also explain the intermediate (best-fit) value of the ¹²C/¹³C ratio (∼12) of J043154.1-063210 since the mixing has not yet reduced the ratio (Charbonnel & Do Nascimento 1998; Gratton et al. 2004, and references therein).

The picture is more complicated, however, for our Li-rich giants. The majority of these stars have best-fit ¹²C/¹³C values greater than 10. This is unexpected beyond the RGB bump, given the above scenario, especially once a star has reached the RGB tip. One possibility is that our best-fit ¹²C/¹³C values are overestimated due to a flat likelihood peak. For example, T5496-00376-1, the most metal-rich ([Fe/H] = -0.63) giant in our sample, is the only Li-rich giant with stellar parameter values consistent with the RGB bump. The best-fit carbon isotopic ratio for this star, however, is a bit larger (∼15) than expected if Li enrichment took place as described above. Our lower limit to ¹²C/¹³C, however, is 5, which is much closer to that predicted by Li enrichment at the RGB bump. Alternatively, this star may not have finished this mixing episode, which would imply that the ¹²C/¹³C ratio is still decreasing and that its Li abundance is still increasing. However, this solution would not explain those stars near the RGB tip that would have finished the Li-enrichment phase at the RGB bump.

The Li enrichment obtained at the RGB bump is also expected to decrease as a star evolves beyond the RGB bump due to normal Li dilution. Indeed, the Li-rich phase at the RGB bump is quite short. According to Denissenkov & Herwig (2004), the phase should last no longer than a few million years. This implies that the Li should be (at least partially) depleted by the time a star reaches the RGB tip. According to the H-R diagram, T6953-00510-1 has a gravity and temperature consistent with it being on the RGB, but it has most likely evolved past the RGB bump. Its Li abundance is less than that of T5496-00376-1, as well as those found for solar metallicity stars (cf., Kumar et al. 2011). It could therefore have ended the extra mixing phase at the RGB bump and its Li abundance is now being depleted. Also note that it has a low carbon isotopic ratio, ¹²C/¹³C ∼5, suggesting that it has gone through thermohaline mixing.

It should be noted that this star is the only one in our sample that shows consistent s-process enhancement (as discussed in Section 4.3). We also argue that it may be "CNO-increased," in that its CNO abundances are somewhat enhanced in comparison to the other Li-rich stars in our sample. This star is not evolved enough to have produced (and dredged-up) the s-process and CNO abundances in its atmosphere. Most likely these enhancements are due to pre-enrichment at birth or mass transfer in a long-period binary. Although we found no conclusive disparities in the radial velocity of T6953-00510-1 (see Section 4.5), radial velocity variation from a long-period binary would be virtually undetectable without a dedicated radial velocity study.

The high-luminosity Li-rich giants in our sample, however, have Li abundances equal to or greater than the stars in our sample near the RGB bump (as well as that of solar-metallicity stars in previous samples), contrary to what was predicted for Li enrichment at the RGB bump (e.g., Charbonnel & Primas 2005; Kumar et al. 2011). A solution to this discrepancy is that these stars have undergone Li enrichment via extra-deep mixing combined with CBP. Sackmann & Boothroyd (1999) show that CBP can take place anywhere along the RGB, while the amount of Li enrichment depends strongly on the rate of mixing in the star. They further predict that the maximum Li abundance attained by a star can occur before significant amounts of ¹³C lower the ¹²C/¹³C ratio. Thus, Li-rich giants could have values of ¹²C/¹³C ranging from ∼4 up to ∼30, which is consistent with the values of ¹²C/¹³C found for our Li-rich giants. Note also that the projected rotational velocities (vsini = 8–10 km s^-1) of C1012254-203007 and T9112-00430-1 may be large enough to induce the high extra mixing rates needed to achieve Li enhancement via CBP.

Another possibility is the so-called Li flash of Palacios et al. (2001). This model also predicts that the Li-rich phase begins at the RGB luminosity function bump when ⁷Be, transported from the interior, decays into ⁷Li and burns in a Li-burning shell. This extra energy rapidly increases the star's luminosity and forces extra mixing, including the transport of more ⁷Be (which decays into ⁷Li) to the surface. This mixing initially does not change the surface carbon isotope ratios, but the enhanced convection eventually reaches the depth where ¹²C is converted to ¹³C and the temperature is high enough to burn the freshly minted Li (⁷Li(p, α)α). Both the surface Li abundance and ¹²C/¹³C ratio are then lowered.

Pilachowski et al. (2003) analyzed the previously discovered Li-rich giant M3 IV-101 and found that the intermediate value of the star's ¹²C/¹³C ratio (for which we found a similar value) and its high luminosity were consistent with the Li-flash model. Our analysis shows that several of the Li-rich giants are much brighter than the RGB bump and indeed that these stars have an estimated luminosity that is about 15 times the RGB-bump luminosity. The Li-flash model of Palacios et al. (2001) predicts a factor of ∼5.5 increase in luminosity, but based on an assumed 1.5 M_☉ solar-metallicity star, rather than the lower-mass VMP stars studied here. Further, recall that Denissenkov & Herwig (2004) showed that canonical extra mixing cannot be responsible for the Li flash. However, if this scenario were possible in low-mass stars, then it could explain the luminous Li-rich giants in our sample, but cannot provide a solution for the less luminous giants.

Kumar et al. (2011) found that their stars close to the RGB bump were more likely associated with the theoretical position of the red clump, in which metal-rich (or young) stars have begun their core He-burning. Since it is highly unlikely that the Li-rich phase at the RGB bump would last until the red clump, they postulated that the Cameron–Fowler mechanism may also play a role during the He-core flash for stars with M < 2.25 M_☉.

This relies on the fact that enough ³He has survived mixing along the RGB (e.g., thermohaline mixing) for the Cameron–Fowler mechanism to take effect. They suggest that stars experiencing this mechanism would survive as Li-rich giants for about 1% of their horizontal branch lifetime, which corresponds to a few Myr. Our stellar parameter values for T5496-00376-1 show that it is also consistent with the horizontal branch (see Figure 1). It is therefore possible that T5496-00376-1 could actually be a horizontal branch (core He-burning) star that has undergone Li production at the He-core flash.

The high-luminosity stars in our sample also appear to be consistent with the early TP-AGB phase of evolution. T9112-00430-1, for example, is most likely already evolving on the AGB. Its position far above the RGB tip (see Figure 1) suggests that the derived values of the stellar parameters are being affected by strong departures from hydrostatic equilibrium. More importantly, T9112-00430-1 has been found to be a variable star according to the All Sky Automated Survey (ASAS) Catalog of Variable Stars (Pojmanski 2002). This is further corroborated by the presence of emission in the wings of Hα in its spectrum (see Section 4.6), which suggests that it could be surrounded by a circumstellar envelope.

Hot bottom burning is not expected for our stars. They are metal-poor, which implies that they could be old, and so have masses low enough (M ≲ 1 M_☉) that the convective envelope and H-burning shell are not connected. However, given that enough ³He remains after the stars' ascent of the RGB, CBP (as described in Section 1) can occur on the AGB. Extra deep mixing mechanisms, similar to that of an RGB star (e.g., thermohaline mixing), in the radiative layer of a low-mass AGB star can connect the H-burning shell with the convective envelope, which will then drive the production of ⁷Li by the Cameron–Fowler mechanism. It is possible, however, that the Li enrichment from CBP on the AGB might be less than that on the RGB since the amount of ³He is expected to be depleted from the RGB. As noted in Section 5.1, C1012254-203007 and T9112-00430-1 have projected rotation velocities large enough that efficient mixing speeds could also be achieved to produce Li enhancement via CBP on the AGB.

During the TP-AGB phase, third dredge-up episodes enrich the envelope with s-process elements that are synthesized during the period between thermal pulses, when ¹³C in the star's interior is burned and supplies neutrons (Straniero et al. 1995). Our high-luminosity Li-rich giants show, however, no signs of s-process enrichment. This implies that if they are AGB stars, they have not gone through enough third dredge-up episodes to drive up the s-process. Further, Girardi et al. (2010) found that a low-mass (M_ini ≲ 1.0 M_☉) low-metallicity AGB star would remain O-rich (log (C/O) < 0) for its entire TP-AGB evolution, which is consistent with the CNO abundances found for the high-luminosity Li-rich giants in our sample.

We conclude that those stars in our sample that are evolving on the AGB (e.g., T9112-00430-1) should be (early) TP-AGB objects of low mass where very efficient CBP can occur, while the core is not massive enough to drive the third dredge-up.

Denissenkov & Herwig (2004) suggest that the influence of a nearby companion (giant planet or star) could induce the extra mixing necessary to bring ⁷Be up to the cooler surface where Li is produced, by the reaction ⁷Be(e⁻, ν)⁷Li, and does not suffer rapid destruction from proton captures. Indeed, Li-rich dwarf stars that show possible lithium pollution from a companion star have been discovered in globular clusters (see, e.g., Koch et al. 2011; Monaco et al. 2011a). In particular, Koch et al. (2011) found a Li-rich star that showed no enrichment in s-process to be in a binary system. They suggested that the Li enrichment arose from mass transfer from a companion giant star which had undergone CBP.

The high projected rotational velocity found for C1012254-203007 and T9112-00430-1 could have been produced by the transfer of angular momentum from a binary companion. While we did not have the observations to detect extrasolar planets around our stars, we checked the Schneider et al. (2011) extrasolar planet database but did not find coincidences. We further looked for radial velocity variations that might indicate a low-luminosity stellar-mass companion. The lack of large radial velocity variations over the whole sample strongly suggests that a close stellar-mass companion is not required for the Li-rich phase to occur. The abundances of T6953-00510-1, however, suggest that it may have been enriched through mass transfer by an evolved stellar companion in a wide binary. For this reason, recent Li enrichment (so that the Li has not yet burned away) from an evolved companion cannot be ruled out for this star.