Origin of $\alpha$-rich young stars: clues from C, N and O

The crucial aspect in deriving accurate elemental abundances is the selection of suitable stellar lines with good quality atomic and molecular data²²2see for example, http://kurucz.harvard.edu/linelists.html. An infrared solar spectral atlas, with a resolution of 0.05 Å (Livingston & Wallace, 1991) observed with the Fourier Transform Spectrometer at the McMath/Pierce Solar Telescope on Kitt Peak (Pierce, 1964) and degraded to the instrumental resolution of the APOGEE spectrograph, was used as a reference to visually confirm clean, unblended, relatively isolated and symmetric spectral lines due to various atomic and molecular species. The suitable spectral lines of an element are those lines which contain only contributions from the element at the central wavelength specified by the transition. The equivalent width (EW) of the selected atomic absorption lines in red giants were measured manually using the cursor commands in splot package of IRAF by fitting often a Gaussian profile. For a few lines, i.e. the 15060 Å and 16215 Å lines of Si and the 16719 Å and 16763 Å lines of Al, which are typically strong with EW > 200 mÅ and have a Lorentzian shaped profile, a direct integration was performed for the best measure of EW. The abundances of carbon, nitrogen, oxygen were derived by matching the synthetic spectra to selected regions around CO (15575 - 15585 Å ), CN (15370 - 15420 Å ; 15462 - 15472 Å ; 15560 - 15590 Å ) and OH (16367 - 16369 Å ) lines in the observed spectrum.

The oscillator strengths ($f$) for the elements Fe, Ni, Mg, Al, Si, Ca and Ti in the infrared H-band were extracted from Meléndez & Barbuy (1999). These oscillator strengths were derived by Meléndez & Barbuy (1999) from an iterative fit of synthetic spectra to the solar spectrum. The typical accuracies in the adopted log $gf$-values (where $g$ is the statistical weight of an atomic energy level) are better than 0.05 dex. The molecular data for OH were extracted from Brooke et al. (2016), CO and CN molecular data were taken from the Kurucz coxx.asc³³3http://kurucz.harvard.edu/linelists/linesmol/ line list and Sneden et al. (2014), respectively.

Although the spectra of red giants at the spectral resolution of the APOGEE spectrograph in the H-band are heavily crowded with absorption lines, our line selection results in a number of unblended lines for each element. The final list of neutral atomic lines selected for deriving chemical abundances includes Fe(28), Ni(6), Mg(2), Al(3), Si(6), Ca(3), Ti(3). Here, the numbers in the parentheses indicate the number of lines of an element considered in the abundance analysis. The full line list used in this work is shown in Table 1, while the measured EWs for all stars are listed in Tables 7 - 13. Additionally, we used a set of molecular lines due to CN (at wavelength 15374, 15400, 15466, 15576, 15581 Å ), OH at 16368 Å and a CO molecular band head at 15578 Å. The molecular line list employed here involves the dominant isotopes of each element and is composed of ¹²C¹⁶O, ¹²C¹⁴N and ¹⁶O¹H molecules.

The solar abundances were derived using solar EWs, measured off the infrared solar spectral atlas (Livingston & Wallace, 1991), and the ATLAS9 theoretical photosphere computed adopting the solar abundances from Asplund et al. (2006) with parameters T_eff⊙ = 5777 K, log ${g}_{\odot}$ = 4.44 dex, [Fe/H]=0.0 dex to establish a reference abundance scale. Lines with individual iron abundances falling outside of $\pm 0.05$ dex of the mean log $\epsilon$(Fe)=7.47 dex were rejected assuming either the selected Fe lines have inadequate atomic data or are blended with unidentified lines. This procedure is repeated for lines of other atomic species. We obtained a microturbulence velocity of $\xi_{t}$ = 0.97 km s^-1, a value in agreement with the published results (Asplund et al., 2006), by requiring that the Fe abundance from Fe 1 lines is independent of the EW of the line.

The derived solar abundances are reported in Tables 4 and 5. We refer to our solar abundances when determining the stellar abundances, [X/H] and [X/Fe], i.e., our analysis is essentially a differential one relative to the Sun.

The EW of a spectral line is influenced by the physical conditions and number density of absorbers in the stellar atmosphere. Therefore, it is essential to predetermine the stellar parameters to obtain a robust estimate of chemical abundances. Starting with a theoretical model with T_eff derived from the ASPCAP pipeline (Holtzman et al., 2018), asteroseismic log ${g}$ (Pinsonneault et al., 2014; Pinsonneault et al., 2018, for the stars analysed by Jofré et al. (2016) and the newly selected stars in the current work, respectively) and measured EWs, the individual line abundances were obtained by matching the computed EWs to the observed ones (Sneden, 1973) while satisfying the following constraints simultaneously. First, we obtain the microturbulence, $\xi_{t}$, assumed to be isotropic and depth independent, by requiring that the Fe abundance from Fe 1 lines is independent of the EW of the line. Second, we estimate the effective temperature by requiring that the abundance from Fe 1 lines is independent of the lower excitation potential of the line (excitation equilibrium). As the stellar parameters T_eff, log ${g}$ and $\xi_{\rm t}$ are interdependent, several iterations are needed to choose a suitable model from the grid. If required for convergence the asteroseismic log ${g}$ values were also iterated over within their uncertainties to satisfy that the Fe i lines are independent of both the EW and the lower excitation potential simultaneously. No Fe ii lines are present in the APOGEE wavelength regime and hence ionisation balance could not be enforced. The stellar parameters ($T_{\rm eff}$ and log $g$) used in the current analysis are listed in Table 3 together with the masses, radii and ages as presented by Pinsonneault et al. (2018).

We performed a differential abundance analysis relative to the Sun using the abfind driver of MOOG and adopting 1D model atmospheres (Castelli & Kurucz, 2003) based on the derived stellar parameters and the EWs of the absorption lines. In table 4, we list abundance results for individual stars averaged over all available lines of given species relative to solar abundances derived from the adopted $gf$-values. In Fig. 3 we show the trends of [X/Fe] vs. [Fe/H] and in Tables 7 - 13 we provide EW values for individual lines measured.

The computed synthetic spectra based on the derived stellar parameters were matched to the observed spectra by adjusting the abundances of the derived C, N and O abundances from combinations of vibration-rotation lines of CO and OH along with electronic transitions of CN. The adopted dissociation energies (D₀) in the spectrum synthesis of OH, CO and CN are 4.411 eV (Brooke et al., 2016), 11.092 eV (Huber & Herzberg, 1979) and 7.724 eV (Sneden et al., 2014), respectively. Following Smith et al. (2013), we measured the C, N and O abundances by fitting all three types of molecular lines consistently, i.e. we required molecular equilibrium: we first measured the oxygen abundance from the OH line, then derived the carbon abundance from CO in an iterative fashion. This process is repeated until the abundances of C and O via fits to the CO and OH lines yield consistent values. Then keeping these C and O abundances fixed, we derived the nitrogen abundance through the synthesis of multiple CN lines. The final adopted abundances of C, N and O are those values that provide self-consistent results from OH, CO and CN lines. We provide CNO abundance results for individual stars as well as the solar values in table 5 .

Stars in our sample are analysed identically and span small ranges in atmospheric parameters. Thus, systematic uncertainties affecting the abundances and abundance ratios [X/Fe] are expected to be consistent (and small) across the sample which spans a range in metallicity of $\sim$ $-$0.6 to $+$0.05 dex. Most of the elements analysed for chemical abundances are represented by more than one atomic or molecular line, and the internal consistency in the average abundance of an element is given by the $\pm$1$\sigma$ standard deviation about the mean abundance (later referred to as $\sigma_{1}$), which is typically less than 0.04 dex. As the multiple lines of each element have varying line strengths and excitation potentials, the respective 1$\sigma$ uncertainties also provide an indication of how well the 1D LTE line analysis can reproduce all spectral lines in the observed spectra.

Following the discussion in Smith et al. (2013), we evaluated the magnitude of uncertainties introduced in the measured chemical abundances by the uncertainties in stellar parameters ($T_{\rm eff}$, log $g$, $\xi_{t}$) and metallicity by varying each parameter separately by an amount equal to its uncertainty, while keeping the other parameters unchanged. As the sample of red giants analysed in this paper spans a limited range in T_eff, log $g$ and $\xi_{t}$, the abundance sensitivity to stellar parameters is carried out for a representative star KIC 6664950 with $T_{\rm eff}$ = 4800 K, log $g$ = 2.4 dex and $\xi_{t}$ = 1.7 km s^-1. The incremental changes about the average abundance of an element caused by varying T_eff, log $g$, $\xi_{t}$ and the model metallicity by $\pm$ 50 K, 0.2 dex, 0.2 km s^-1 and 0.1 dex (see for details of these choices Reddy & Lambert, 2015), respectively, with respect to the chosen model parameters are summarised in Table 6. The square root of the quadratic sum of the mean values of all four contributors are presented in the column headed $\sigma_{2}$. The total error $\sigma_{\rm tot}$ in the measured elemental abundance is the square root of the quadratic sum of $\sigma_{1}$ and $\sigma_{2}$. Entries in Table 6 reveal that the observed dispersions in [El/Fe] for many of the elements are in the range 0.02$-$0.06 dex comparable to their respective $\pm$1$\sigma$ measurement uncertainties. Exceptions include the element oxygen whose uncertainty of 0.1 dex is dominated by the sensitivity of O abundance to the overall metallicity.

We compare our results with the ones obtained by Hawkins et al. (2016) and the APOGEE collaboration as presented by Holtzman et al. (2018). The BACCHUS code (Hawkins et al., 2016) relies on the radiative transfer code Turbospectrum (Alvarez & Plez, 1998; Plez, 2012) and the MARCS model atmosphere grid (Gustafsson et al., 2008). The abundance are derived by comparing the observed spectrum with a set of convolved synthetic spectra characterised by different abundances. A $\chi^{2}$ diagnostic was used to obtain a robust abundance measurement. The ASPCAP pipeline (García Pérez et al., 2016; Holtzman et al., 2018) determines stellar parameters and abundances by finding the best match between observed spectra and a large grid of synthetic spectra. The synthetic spectral grid is multi-dimensional including $T_{\rm eff}$, $\log(g)$, [M/H], [$\alpha$/M], [C/M], [N/M], and microturbulent velocity. In the $T_{\rm eff}$ range that we are interested in, ASPCAP uses a set of model atmospheres specifically generated for APOGEE: the APOGEE ATLAS9 models (Mészáros et al., 2012; Zamora et al., 2015), which are based on the ATLAS9 model atmosphere code from Castelli & Kurucz (2004). These model atmospheres were constructed with the same C, N and $\alpha$ abundances, as well as [Fe/H], that were used to generate the synthetic spectrum at that grid point.

In this work we adopted $T_{\rm eff}$ and log $g$ from Hawkins et al. (2016) as starting values for the analysis of the set of 26 stars presented by Jofré et al. (2016). These values are the ones provided by APOGEE DR12. For the set of 25 stars that we newly selected in this work we used $T_{\rm eff}$ and log $g$ from the APOGEE DR14 as starting values for the analysis. In Fig. 1 we show the comparisons between the metallicities, the $\alpha$ abundances as well as $T_{\rm eff}$ and log $g$. For log $g$ all comparisons are close to the one-to-one line. For $T_{\rm eff}$ there is a group of orange dots that are located below the one-to-one line. These are the stars selected by Jofré et al. (2016) for which we used the Hawkins et al. (2016), i.e. APOGEE DR12 values of $T_{\rm eff}$ and log $g$, and this offset is reminiscent of the offset in $T_{\rm eff}$ between APOGEE DR12 and DR14 values.

There exists a constant offset with small dispersion in [Fe/H] between the literature and values derived in our study. The typical mean difference between our and APOGEE DR14 metallicities is $+$0.09$\pm$0.08 dex (51 stars) and between our and Hawkins et al. (2016) [Fe/H] estimates is $+$0.15$\pm$0.07 dex (26 stars), see also Table 2. Both dispersions about the mean differences are comparable to the line-to-line abundance dispersion observed for Fe lines. The comparable offsets of $+$0.15 dex and $+$0.09 dex in [Fe/H] between our study and the literature results provide an opportunity to assess the consistency of our abundance estimates. For stars in common between the APOGEE DR14 and BACCHUS analyses, the typical mean difference in [Fe/H] is $-$0.12 dex with a dispersion of 0.05 dex, corresponding to the line-to-line abundance dispersion for Fe. Therefore, the association of very similar offset (left panel in Figure 1) of our [Fe/H] values from Hawkins et al. (2016) and the calibrated [M/H] from ASPCAP strengthens the assumption that our measured metallicities are self-consistent across the sample of 51 giants.

For [$\alpha$/Fe] we show a comparison in the central top panel of Fig. 1. For both our work and for the BACCHUS pipeline the [$\alpha$/Fe] ratio is obtained by taking the average of [X/Fe] for the elements Mg, Si, Ca and Ti. For the APOGEE comparison we use the provided [$\alpha$/Fe] that results from the multi-dimensional grid matching (see above). Note that we do have stars with [$\alpha$/Fe] < 0.1 in the sample mostly originating from the Jofré et al. (2016) selection. In Tables 3, 4 and 5 we categorise stars with [$\alpha$/Fe] > 0.1 as $\alpha$-rich stars (21 young stars and 15 old stars) and stars with [$\alpha$/Fe] $\leq$ 0.1 as $\alpha$-normal stars (15 stars).

We compare in Figure 2 our LTE measured abundance ratios of [C/H], [N/H], [O/H] with results presented by Hawkins et al. (2016) (black) for 26 giants in common between the studies and with results from the APOGEE DR14 catalogue (orange, Holtzman et al., 2018) for 51 stars. We find general agreement with the BACCHUS and APOGEE DR14 results. However for the APOGEE results we find significantly more scatter. Because the [$\alpha$/Fe] measurement for APOGEE is derived by searching for the best match within a grid of synthetic spectra, it is most sensitive to the elements that have the most effect on the spectra. For the H-band, this element is O, instead of the elements used in the other [$\alpha$/Fe] calculations, which may account for part of the scatter. Additionally, we re-normalised the continuum in this work and used some additional features.

Summarising, Figs 1 and 2 highlight satisfactory agreement of chemical abundances for all the elements despite some biases in [Fe/H]. As we perform a homogeneous analysis such biases will not impact on our final results.