DETECTION OF NEUTRAL PHOSPHORUS IN THE NEAR-ULTRAVIOLET SPECTRA OF LATE-TYPE STARS^*

Phosphorus (P, Z = 15) is one of the few light elements whose nucleosynthetic origins in ancient stars remain unexamined by observations. Models by Woosley & Weaver (1995) and Kobayashi et al. (2006) predict the phosphorus yields of massive supernovae. These yields are incorporated into chemical evolution models like those of Timmes et al. (1995) and Cescutti et al. (2012) to predict the Galactic evolution of phosphorus from the first stars until today. At present, however, no observations are capable of verifying these predictions in low-metallicity Galactic environments.

One neutron-rich isotope of phosphorus, ³¹P, is found in solar system material. Previous derivations of the phosphorus abundance in the Sun show remarkable agreement among different studies. The earliest solar photospheric estimates by Goldberg et al. (1960) and Lambert & Warner (1968) derived log  (P) = 5.34 and 5.43, respectively, on the scale where log  (H) ≡ 12.00. These and other estimates from the last four decades are in agreement with modern estimates by Asplund et al. (2009) and Caffau et al. (2007), 5.41 and 5.46, respectively. These values are also in agreement with the recommended meteoritic abundance, 5.45 (Lodders et al. 2009).

Meléndez et al. (2009) derived phosphorus abundances in 10 solar twins and Caffau et al. (2011) studied the evolution of phosphorus in 20 stars with −1.0 < [Fe/H] <+0.3. Phosphorus ions have been detected in the atmospheres of normal O- and B-type stars (e.g., Struve 1930; Crowther et al. 2002), the F-type star Procyon (Kato et al. 1996), chemically peculiar A- and B-type stars on the upper main sequence or horizontal branch (e.g., Bidelman 1960; Sargent & Searle 1967; Leckrone et al. 1999; Castelli & Hubrig 2004) and hot stars that have evolved beyond the asymptotic giant branch (AGB; e.g., Vennes et al. 1996; Marcolino et al. 2007; Reiff et al. 2007). P-bearing molecules—including CP, PN, PO, HCP, C₂P, and PH₃—have been detected in cold, dense clouds (e.g., Turner & Bally 1987; Ziurys 1987) and the carbon- and oxygen-rich circumstellar envelopes of stars in the AGB phase of evolution (e.g., Guelin et al. 1990; Halfen et al. 2008; Milam et al. 2008; Tenenbaum & Ziurys 2008). P ii has been detected in neutral gas in the interstellar medium (ISM) of the Milky Way (e.g., Morton 1975; Jenkins et al. 1986; Savage & Sembach 1996; Lehner et al. 2003). P ii has been detected in the ISM of external galaxies, including the Large Magellanic Cloud (Friedman et al. 2000), M33 (Lebouteiller et al. 2006), and several more distant low-metallicity star-forming galaxies (e.g., Lebouteiller et al. 2013). P ii has also been detected in several high-redshift damped Lyman-α systems (DLAs; Outram et al. 1999; Molaro et al. 2001; Lopez et al. 2002; Levshakov et al. 2002), and P v has been detected in the broad-absorption-line region of a quasar (Junkkarinen et al. 1997). The common theme of these extra-solar phosphorus detections is that they do not offer opportunities to confront nucleosynthesis predictions in Galactic environments with metallicity [Fe/H] <−1.0.

In contrast, extensive observations are available for the neighboring elements silicon (Si, Z = 14) and sulfur (S, Z = 16). Observations of Si i lines across the optical spectrum of late-type stars with −4.0 ⩽ [Fe/H] ⩽+0.3 have led to consensus regarding the cosmic origins and evolution of this element (e.g., Luck & Bond 1981; Peterson 1981; Tomkin et al. 1985; Magain 1987; Gratton & Sneden 1988; Edvardsson et al. 1993; Ryan et al. 1996; Fulbright 2000; Cayrel et al. 2004; Reddy et al. 2006; Bensby et al. 2014). Sulfur has also been systematically investigated using a few multiplets of S i in the optical and near-infrared (NIR; e.g., Clegg et al. 1981; François 1987, 1988; Chen et al. 2002; Takada-Hidai et al. 2002, 2005; Reddy et al. 2003; Ecuvillon et al. 2004; Nissen et al. 2004, 2007; Ryde & Lambert 2004; Caffau et al. 2005, 2010; Nissen et al. 2007; Jönsson et al. 2011; Takeda & Takada-Hidai 2011; Spite et al. 2011; Matrozis et al. 2013). Most studies have found that both [Si/Fe] and [S/Fe] exhibit an enhanced "plateau" at low metallicity. This echos familiar trends seen in other elements produced along the α chain, most notably oxygen (O, Z = 8), magnesium (Mg, Z = 12), and calcium (Ca, Z = 20). Silicon and sulfur are formed primarily through α-captures during hydrostatic and explosive oxygen burning, and the enhanced [Si/Fe] and [S/Fe] plateaus result from nucleosynthesis in core-collapse supernovae early in the history of the Galaxy (e.g., Matteucci & François 1989; Timmes et al. 1995; Thielemann et al. 1996; Samland 1998; Goswami & Prantzos 2000; Kobayashi et al. 2006; Tominaga et al. 2007).

In stars like the Sun, no optical lines of P i are available, and only a few weak P i multiplets are present in the NIR. These NIR lines have been known in the solar spectrum for a long time (Moore et al. 1934). Only the studies of Kato et al. (1996), Meléndez et al. (2009), and Caffau et al. (2011) have derived phosphorus abundances in stars useful for examining the nucleosynthetic fossil record. The NIR lines are too weak to detect in late-type stars with [Fe/H] <−1.0. Here we show that this observational limitation can be overcome through the use of near-ultraviolet (NUV) spectroscopy of late-type stars. We examine P i absorption lines present in the near NUV spectra of late-type stars in the solar neighborhood spanning −3.8 < [Fe/H] <−0.1. Our results extend the metallicity range of stars with phosphorus abundances by nearly 3 dex.

We use standard definitions of elemental abundances and ratios. For element X, the logarithmic abundance is defined as the number of atoms of element X per 10¹² hydrogen atoms, log (X) ≡ log₁₀(N_X/N_H) + 12.0. For elements X and Y, the logarithmic abundance ratio relative to the solar ratio of X and Y is defined as [X/Y] ≡ log₁₀(N_X/N_Y) − log₁₀(N_X/N_Y)_☉. Abundances or ratios denoted with the ionization state indicate the total elemental abundance as derived from that particular ionization state. We adopt the Asplund et al. (2009) photospheric abundances for all elements studied.

We obtained observations covering the NUV spectral range from the Barbara A. Mikulski Archive for Space Telescopes (MAST). These observations were taken using the medium- or high-resolution echelle gratings in the Space Telescope Imaging Spectrograph (STIS; Kimble et al. 1998; Woodgate et al. 1998) on board the Hubble Space Telescope (HST). Spectra downloaded from the MAST have been reduced by the calstis pipeline and combined by Ayres (2010). Spectra obtained previously by our own observing programs were reduced by the calstis pipeline and processed as described in Roederer et al. (2012, 2014c) and Placco et al. (2014).

Some stars were discarded from our analysis because their spectra had low signal-to-noise ratios (S/N's). Cooler or more metal-rich stars were discarded because we could not reliably identify the continuum. Table 1 presents a list of the data sets and observational characteristics for the 14 stars retained for our study. Figure 1 illustrates the STIS spectra of 13 of these stars for a region surrounding the P i 2135.46 and 2136.18 Å lines. No spectrum covering these lines exists for BD+44 493, so this star is not shown in Figure 1.

Zoom In Zoom Out Reset image size

We supplement these NUV spectra with optical spectra taken with ground-based telescopes and found in online archives. Table 2 provides details about these optical spectra. These data were taken using the High Accuracy Radial velocity Planet Searcher (HARPS; Mayor et al. 2003) on the ESO 3.6 m Telescope at La Silla, the Ultraviolet and Visual Echelle Spectrograph (UVES; Dekker et al. 2000) on the 8.2 m Very Large Telescope (VLT UT2, Kueyen) at Cerro Paranal, or the High Resolution Echelle Spectrometer (HIRES; Vogt et al. 1994) on the 10 m Keck I Telescope at Mauna Kea. We obtained pipeline-reduced individual exposures from the archives, normalized the continua, and shifted each exposure to laboratory-zero velocity.

Some stars were not present in these archives at the time of our query, and we have collected new optical data instead. Some observations were made using the Magellan Inamori Kyocera Echelle (MIKE) spectrograph (Bernstein et al. 2003) on the 6.5 m Landon Clay Telescope (Magellan II) at Las Campanas. These data were obtained and reduced using procedures described in detail in Roederer et al. (2014b). Other observations were obtained using the Robert G. Tull Coudé Spectrograph (Tull et al. 1995) on the 2.7 m Harlan J. Smith Telescope at McDonald Observatory. These data were reduced and combined using the REDUCE software package (Piskunov & Valenti 2002) and standard IRAF procedures. Characteristics of these observations are also given in Table 2.

Although most stars in this sample have stellar parameters available in the literature, we carry out a homogeneous analysis based on archival optical spectra. We determine stellar parameters using a classical spectroscopic analysis starting with the measurement of equivalent widths, W_λ. These measurements are presented in Table 3. We use the ATLAS9 model atmosphere grid (Castelli & Kurucz 2003) and a recent version of MOOG (Sneden 1973) that includes the contribution of Rayleigh scattering from H i in the source function (Sobeck et al. 2011).

We use the line list compiled in Roederer et al. (2010b). However, we found few Fe i and Fe ii lines from this list were measurable in the spectrum of HD 155646. For this star, we use the line list from Ramírez et al. (2013). We verify that these two line lists yield similar stellar parameters and element abundances in the analysis of other metal-rich stars, and no systematic errors are introduced.

The derived model parameters are listed in Table 4. Effective temperatures (T_eff) and microturbulent velocities (v_t) are determined by minimizing trends of Fe i lines with excitation potential (E.P.) and reduced equivalent width [log (W_λ/λ)]. Surface gravities (log g) are determined by forcing the mean iron abundances derived from Fe i and Fe ii lines to agree. Once excitation and ionization balance are achieved, we apply the empirical temperature calibration described in Frebel et al. (2013). Stars with [Fe/H] >−2.5 fall outside the metallicity range over which this calibration was tested. We note, however, that studies of stars more metal-rich than this have found similar discrepancies between photometric and spectroscopic T_eff values (e.g., Johnson 2002), indicating that a calibration such as this is appropriate. Furthermore, at the relatively warm temperatures of these stars (>5200 K), the correction is small and comparable to the uncertainty (<150 K). We have three stars in common with Nissen et al. (2007) and two in common with Asplund et al. (2006), who derived T_eff by fitting the wings of the Hβ and Hα lines, respectively. The mean differences (−95 ± 57 K and −72 K, respectively) are comparable to the systematic zero point uncertainties of these different methods, suggesting that our results are not wildly in error.

Roederer et al. (2014b) derived stellar parameters and abundances for a large sample of metal-poor stars. That study derived T_eff and v_t by requiring no correlation between derived Fe i abundances and E.P. and log (W_λ/λ), as we have done. That study found typical statistical uncertainties of ≈50 K in T_eff and ≈0.1 km s^-1 in v_t for stars in similar evolutionary states to those in our sample. The uncertainties in T_eff and v_t when eliminating correlations with E.P. and log (W_λ/λ) are similar in our work here. We adopt these values as the statistical uncertainties in T_eff and v_t. The dispersion among Fe i and Fe ii lines suggests that ≈0.3 dex is a fair estimate of the statistical uncertainty in log g. We assume that the systematic uncertainty in T_eff could be as large as the correction applied to convert the spectroscopic T_eff to the photometric (V−K) scale (Equation (1) and Figure 2 of Frebel et al. 2013). Roederer et al. made comparisons for the derived model atmosphere parameters for large numbers of stars (see Table 9 there). The dispersions in log g and v_t values were found to be ≈0.35 dex and ≈0.4 km s^-1 for stars in class "SG," which have similar evolutionary states to the stars in our sample. We adopt these values as the systematic uncertainties in log g and v_t.

We have not rederived stellar parameters or abundances for BD+44 493, and it is not listed in Table 2. This star has been studied extensively by Ito et al. (2009, 2013), Placco et al. (2014), and Roederer et al. (2014a). We do not detect P i in this star, so we simply adopt the Ito et al. (2013) model parameters and abundances. We apply the same +0.23 dex offset to the phosphorus upper limit to correct for differences in the optical and NUV abundance scales (see Section 6.1).

The high first-ionization potential of phosphorus (10.49 eV) ensures that neutral phosphorus is the dominant ionization state in the atmospheres of late-type stars. Several NUV P i lines are listed in the Atomic Spectral Database (ASD) hosted by the National Institute of Standards and Technology (NIST; Kramida et al. 2013). Among the stars in our sample, HD 140283 has the most extensive spectral coverage at R ∼ 110,000 and high S/N. We use this star to identify which P i lines might be reliable abundance indicators.

The P i 2135.465 Å line is clearly detected in the HD 140283 spectrum. It is blended with a Cr ii line at 2135.42 Å. This line is marginally useful as an abundance indicator in HD 140283. It is illustrated in Figure 2, which compares the observed spectrum of HD 140283 with our synthetic spectra. The abundances derived from this line and all others in HD 140283 are presented in Table 5.

Zoom In Zoom Out Reset image size

Table 5. Atomic Data and Abundances for P i Lines Detected in HD 140283

Wavelength	Upper	Eupper a	Lower	Elower a	log gf	log Γ6/NH b	log
(Å)	level	(cm−1)	level	(cm−1)		(rad cm3 s−1)	HD 140283 c
2135.465	(3P)4s2P3/2	58174.366	3p3 2Do3/2	11361.02	−1.24	−7.42	2.67
2136.182	(3P)4s2P3/2	58174.366	3p3 2Do5/2	11376.63	−0.11	−7.42	2.58
2149.145	(3P)4s2P1/2	57876.574	3p3 2Do3/2	11361.02	−0.36	−7.42	⋅⋅⋅
2152.940	(1D)4s2D3/2	65156.242	3p3 2Po1/2	18722.71	−0.87	−7.22	2.49
2154.080	(1D)4s2D5/2	65157.126	3p3 2Po3/2	18748.01	−0.62	−7.22	2.46
2154.121	(1D)4s2D3/2	65156.242	3p3 2Po3/2	18748.01	−1.32	−7.22	d
2533.976	(3P)4s2P3/2	58174.366	3p3 2Po1/2	18722.71	−1.11	−7.42	2.61
2535.603	(3P)4s2P3/2	58174.366	3p3 2Po3/2	18748.01	−0.44	−7.42	⋅⋅⋅
2553.262	(3P)4s2P1/2	57876.574	3p3 2Po1/2	18722.71	−0.86	−7.42	⩽2.76
2554.915	(3P)4s2P1/2	57876.574	3p3 2Po3/2	18748.01	−1.23	−7.42	2.49

Notes. ^aSvendenius (1980). ^bCalculated for T = 10⁴ K. ^cCorrected for optical–NUV offset; see Section 6.1. ^dLine forms a blended, single feature with P i 2154.08 Å line.

Download table as:  ASCIITypeset image

The P i 2136.182 Å line is approaching saturation in HD 140283. We see no evidence of line asymmetries or hints of absorption by other species in the observed line profile of HD 140283. Saturated lines are less sensitive to the abundance, but otherwise this line appears to be a reliable abundance indicator. Figure 2 illustrates this line in HD 140283.

The P i 2149.145 Å line is detected but blended with an Fe i line at 2149.18 Å. This composite feature is saturated in HD 140283 and is not useful as a phosphorus abundance indicator. This line is shown in Figure 2, but the synthetic spectrum that has been "fit" to the observed one has not been used to derive an abundance.

The P i 2152.940 Å line is located in the wing of a much stronger Fe i line at 2153.01 Å. It can be used as an abundance indicator in HD 140283, as illustrated in Figure 2.

The P i 2154.080 and 2154.121 Å transitions blend together to form one line, although the absorption is dominated by the 2154.080 Å line. They are blended with an Fe i line at 2154.13 Å of comparable strength. NIST reports a "C" accuracy (25%) for the transition probability of this Fe i line. We adjust its log gf value in our synthesis (to −2.57) to derive an approximate phosphorus abundance in HD 140283. Figure 2 shows our best attempt to fit this blended feature in HD 140283.

The P i 2533.976 Å line is clean, unsaturated, and easily detected in HD 140283. An unidentified absorption feature is present at 2533.89 Å. We model this feature (and all other unidentified absorption lines) as an Fe i line with an E.P. of 1.5 eV and log gf value constrained by the observed line profile. A weak Mg i absorption feature may be present at 2534.07 Å. The P i 2533.976 Å line appears to be a reliable abundance indicator in HD 140283, as shown in Figure 2. Tests conducted using R  ∼ 30,000 spectra of HD 108317 and HD 122563 (obtained in program GO-12268) indicate, unfortunately, that the P i 2533.976 Å line is too blended to serve as a reliable abundance indicator in lower-resolution data. Both HD 108317 and HD 122563 have [Fe/H] <−2.4, so the P i 2533.976 Å line will not be a reliable abundance indicator in higher-metallicity stars, either.

The P i 2535.603 Å line cannot be discerned amid the absorption from a much stronger Fe i line at 2535.61 Å. Our synthesis suggests that this line would be detectable in the absence of the Fe i line. We cannot use this P i line as an abundance indicator, and it is not illustrated in Figure 2.

The P i 2553.262 Å line is detected but blended with several features of comparable strength. An unidentified feature at 2553.18 Å and a Co i feature at 2553.34 Å can be fit to match the observed line profile. Our synthesis also suggests that absorption from Mg i and Mn ii could be present at 2553.26 Å, and both features are coincident with the P i line. The log gf values of these blending features are not known from laboratory measurements. A synthesis with both the Mg i and Mn ii transitions removed can generally reproduce the P i line profile. Since the Mg i and Mn ii transitions could account for an unknown fraction of the absorption at this wavelength, we derive only an upper limit on the phosphorus abundance. This upper limit is larger than the phosphorus abundances derived from other, more secure, abundance indicators. Figure 2 illustrates this fit, which is computed assuming that neither Mg i nor Mn ii contribute to the absorption.

Absorption is detected at the P i 2554.915 Å line. A neighboring line of Fe ii at 2554.94 Å has comparable strength, and it blends with the P i line. NIST reports a "D" accuracy (50%) for the transition probability of this Fe ii line. This P i line is marginally useful as an abundance indicator in HD 140283 and is illustrated in Figure 2.

In summary, there are six marginally useful P i abundance indicators in the NUV spectrum of HD 140283. Our final phosphorus abundance in HD 140283 is a mean of these six abundance derivations. In stars that are more metal-rich or cooler than HD 140283, or in spectra taken at lower resolution or with lower S/N levels, most of these lines are not useful as abundance indicators. The phosphorus abundances we derive for the other stars in our sample rely only on the P i 2136 Å line. No existing NUV spectrum of BD+44 493, however, covers the P i 2136 Å line. We derive an upper limit on the phosphorus abundance in BD+44 493 from the P i 2553 Å line assuming that no Mg i or Mn ii contribute to the absorption.

5.1. Transition Probabilities

5.2. The Damping Constant for the 2136 Å Transition

We employ standard methods to derive abundances of phosphorus and other elements of interest. Abundances of iron, sodium, magnesium, aluminum, silicon, and calcium are derived using equivalent widths (measured from only the optical spectra), model atmospheres interpolated from the ATLAS9 grid, and the line analysis code MOOG, managed through the interactive software package described in Casey (2014). These results are presented in Table 6. The equivalent width, wavelength, E.P., and log gf value of each line are presented in Table 3. All damping constants are taken from Barklem et al. (2000) and Barklem & Aspelund-Johansson (2005), when available, otherwise we resort to the Unsöld (1955) approximation for these lines.

All phosphorus abundances are derived by spectrum synthesis matching, and we use MOOG to generate the synthetic spectra. The best fits are determined by minimizing the residuals between the observed and synthetic spectra. There is a relatively clean region of continuum from 2136.6–2136.8 Å in most of our spectra. Our fits match the observed and synthetic spectra levels here. Only in the most metal-rich stars does this region begin to show weak absorption, but our synthetic spectra adequately reproduce this region. The derived phosphorus abundances are reported in Table 7.

Table 7. Phosphorus Abundances

Star	log  (P)	[P/Fe]
BD+44493	<1.83	< +0.25
G64-12	2.28 (0.34)	+0.15 (0.26)
HD 2454	5.06 (0.27)	+0.07 (0.21)
HD 16220	4.91 (0.26)	−0.19 (0.20)
HD 43318	5.03 (0.26)	−0.14 (0.20)
HD 76932	4.93 (0.29)	+0.44 (0.20)
HD 94028	4.41 (0.29)	+0.61 (0.21)
HD 107113	5.06 (0.26)	+0.24 (0.20)
HD 108317	2.83 (0.35)	−0.14 (0.20)
HD 128279	3.08 (0.36)	+0.04 (0.22)
HD 140283	2.55 (0.30)	−0.21 (0.19)
HD 155646	5.11 (0.25)	−0.14 (0.19)
HD 160617	3.56 (0.27)	+0.07 (0.17)
HD 211998	3.78 (0.33)	−0.13 (0.21)

Download table as:  ASCIITypeset image

We assess the sensitivity of the phosphorus abundances to uncertainties in the model atmosphere parameters by altering each parameter by a fixed amount. Three sets of numbers are presented in Table 8. One set corresponds to the metal-rich stars observed with moderately high spectral resolution (R ∼ 30,000), one set corresponds to the metal-poor stars observed with moderately high spectral resolution (R ∼ 30,000), and one set corresponds to the metal-poor stars observed with high spectral resolution (R ∼ 110,000). The stars in each group are identified in the footnotes to Table 8. The uncertainties reported in Table 8 are conservative since covariances among the model parameters are not taken into account and could reduce the abundance uncertainties by small amounts. Regardless, the abundance uncertainties are not very large (≲0.1 dex), and they do not dominate the error budget.

Table 8. Abundance Sensitivities to Model Atmosphere Uncertainties

Group	Abundance	$\Delta {T_{\rm eff}}=$	Δlog  g =	Δvt =
	or Ratio	±100 K	±0.3 dex	±0.2 km s-1
Metal-rich,	[Na/Fe]	∓0.02	±0.03	∓0.03
R ∼ 30,000a	[Mg/Fe]	∓0.01	±0.01	∓0.02
	[Al/Fe]	∓0.04	±0.01	∓0.05
	[Si/Fe]	∓0.04	∓0.02	∓0.04
	[P/Fe]	∓0.01	∓0.02	∓0.02
	[Ca/Fe]	∓0.01	±0.03	∓0.01
	log (P)	±0.07	∓0.03	∓0.04
	log (Fe)	±0.08	∓0.03	∓0.06
Metal-poor,	[Na/Fe]	∓0.01	±0.04	∓0.01
R ∼ 30,000b	[Mg/Fe]	∓0.02	±0.02	∓0.01
	[Al/Fe]	∓0.01	±0.03	±0.02
	[Si/Fe]	∓0.05	∓0.02	∓0.03
	[P/Fe]	∓0.05	±0.02	±0.03
	[Ca/Fe]	∓0.04	±0.01	∓0.02
	log (P)	±0.06	∓0.05	∓0.08
	log (Fe)	±0.11	∓0.03	∓0.05
Metal-poor,	[Na/Fe]	∓0.02	±0.01	∓0.02
R ∼ 110,000c	[Mg/Fe]	∓0.02	±0.03	∓0.03
	[Al/Fe]	±0.01	±0.05	∓0.01
	[Si/Fe]	∓0.04	±0.01	∓0.03
	[P/Fe]	∓0.03	±0.01	∓0.01
	[Ca/Fe]	∓0.03	±0.01	∓0.01
	log (P)	±0.07	∓0.05	∓0.05
	log (Fe)	±0.10	∓0.04	∓ 0.05

Notes. ^aHD 2454, HD 16220, HD 43318, HD 107113, HD 155646. ^bG64-12, HD 108317, HD 128279. ^cHD 76932, HD 94028, HD 140283, HD 160617, HD 211998.

Download table as:  ASCIITypeset image

We have also considered the effect of macroturbulence in the line profiles and derived abundances. We adopt macroturbulent velocities from Figure 3 of Valenti & Fischer (2005) and compute the macroturbulence profiles using the radial-tangential formalism presented by Gray (1992). These provide a reasonable fit to the line profiles observed in our high-resolution NUV spectra. Gaussian smoothing of our synthetic spectra also provides a reasonable fit. The derived abundances are not sensitive to these values.

6.1. The Offset between the Optical and NUV Abundance Scales

Previous studies of abundances derived from NUV spectra of late-type stars have uncovered evidence that the optical and NUV abundance scales may differ by small amounts. For example, the [Fe/H] ratio derived from Fe i lines in the optical is frequently higher than the [Fe/H] ratio derived from Fe i lines in the NUV. Furthermore, the effect is most pronounced at wavelengths in the blue near the transition from the Paschen continuum to the Balmer continuum. Roederer et al. (2010a, 2012, 2014c), Lawler et al. (2013), Wood et al. (2013, 2014), and Placco et al. (2014) have examined and discussed this issue. Presently, the origin of the discrepancy is unknown, although missing sources of continuous opacity and departures from local thermodynamic equilibrium (LTE) may each be responsible to some degree.

We have derived abundances from 5–17 Fe i and ii lines in the NUV in six stars to get a sense of the magnitude of this offset in our data. These data are compared with the mean abundance derived from Fe i and ii lines in the optical. Table 9 presents these results. These six stars are selected to represent the range of metallicities, stellar parameters, and data quality of our full sample. We did not derive abundances from sufficient numbers of Fe i and ii lines in the NUV in the other stars to warrant inclusion in this sample.

Table 9. Mean Fe i Abundances Derived from Optical and NUV Lines

Star	〈log (Fe i)〉opt	Nlines	〈log (Fe i)〉NUV	Nlines	Δ(Fe i)opt-NUV	〈log (Fe ii)〉opt	Nlines	〈log (Fe ii)〉NUV	Nlines	Δ(Fe ii)opt-NUV	Δ(Fe i−Fe ii)
HD 2454	7.08	172	6.69	16	+0.39	7.09	18	6.75	12	+0.34	+0.05
HD 16220	7.19	121	6.94	11	+0.25	7.20	13	6.88	7	+0.32	−0.07
HD 43318	7.26	189	6.99	13	+0.27	7.26	20	7.05	7	+0.21	+0.06
HD 108317	5.06	243	4.73	17	+0.33	5.07	25	4.81	13	+0.26	+0.07
HD 140283	4.85	170	4.58	5	+0.27	4.87	15	4.78	5	+0.09	+0.18
HD 155646	7.34	65	7.02	11	+0.32	7.35	5	7.15	7	+0.20	+0.12
				〈Δ〉 =	+0.30				〈Δ〉 =	+0.23	+0.07
				s.e.m. =	0.02				s.e.m. =	0.03	0.03

Download table as:  ASCIITypeset image

On average, the Fe i lines in the optical yield abundances higher by +0.30 ± 0.02 dex (σ = 0.05 dex). The Fe ii lines in the optical yield abundances higher by +0.23 ± 0.04 dex (σ = 0.09 dex). The difference between these two values, +0.07 dex, suggests that non-LTE effects could be preferentially affecting low-lying levels of the minority species, Fe i, which are commonly found in the NUV. We observe an offset between the optical and NUV abundances derived from Fe ii lines, too. The Fe ii electronic level populations should not deviate substantially from their LTE values. This suggests that non-LTE effects are probably not the dominant source of the offset between the optical and NUV abundance scales.

Roederer et al. (2012, 2014c) also performed a detailed analysis of the Fe i and ii offset for HD 108317. Figure 3 illustrates the differences in the derived abundances between that study and the present one. As expected, the agreement is good for the optical lines (σ = 0.06 dex from 122 Fe i lines; σ = 0.07 dex from 9 Fe ii lines). The zero point offset (+0.09 dex for Fe i; −0.01 dex for Fe ii) can be explained by the slightly warmer T_eff derived in the present study. The dispersion increases considerably for the NUV lines near 2000 Å (σ = 0.39 dex from 14 Fe i lines; σ = 0.22 dex from 9 Fe ii lines). Both sets of measurements show an offset in the abundances derived from lines in the NUV, but the offset derived by Roederer et al. (0.08 dex for Fe i; 0.04 dex for Fe ii) is smaller than that found here (0.32 dex for Fe i; 0.18 dex for Fe ii).

Zoom In Zoom Out Reset image size

To minimize any potential influences from non-LTE effects on Fe i lines, we focus on the difference in Fe ii, 0.14 dex (i.e., 0.18 – 0.04 dex). The analysis presented in Roederer et al. (2014c) was performed by I.U.R., while the analysis presented here was performed by H.R.J. and T.T. Both analyses used the same observed spectrum and tools for analysis. This 0.14 dex difference reflects the uncertainty that may be expected when the analysis is performed by different individuals.

We test our underlying assumption that the phosphorus and iron abundances will track each other if the continuous opacity in the NUV is altered. We artificially adjust the NUV continuous opacity in our calculations so that the abundances derived from NUV Fe i and ii lines agree with the abundances derived from optical Fe i and ii lines. When using the adjusted opacities, the resulting [P/Fe] ratios change by less than 0.04 dex on average, which is less than the statistical measurement uncertainty. We conclude that our assumption is justified.

We apply a correction to our NUV abundances equal to the mean offset derived from Fe ii for the six stars in Table 9, +0.23 dex. This correction is reflected in the abundances presented in Table 7. We adopt the standard error of the mean (s.e.m.) of this correction, 0.03 dex, as its statistical uncertainty. We adopt the zero point uncertainty in the optical–NUV abundance offset, 0.14 dex, as the systematic uncertainty in this correction.

We emphasize that this is an empirical correction designed to place abundances derived from lines in the NUV on the same scale as abundances of numerous other species derived from lines in the optical. We explicitly assume that an unknown parameter(s) is affecting the Fe ii and P i lines similarly. Attempts to understand and account for the physical origin of this offset are challenging and beyond the scope of the present study. The uncertainty in this empirical correction is a substantial fraction of the total error budget, so future investigations into its origin are greatly encouraged.

6.2. Departures from LTE

6.3. Statistical Uncertainties

6.4. Systematic Uncertainties

6.5. Comparison with Caffau et al. (2011)

Our results are illustrated in Figure 4. The left-hand panel of Figure 4 shows the odd-Z elements ([Na/Fe], [Al/Fe], and [P/Fe]), and the right-hand panel of Figure 4 shows the even-Z elements ([Mg/Fe], [Si/Fe], and [Ca/Fe]). Several sets of abundances derived by other studies are shown for comparison. The even-Z α-elements exhibit the familiar pattern marked by a plateau at low metallicity and a gradual decline to the solar ratios at [Fe/H] >−1.0 or so. The [Na/Fe] ratios scatter about the solar value. The [Al/Fe] ratios are sub-solar at low metallicity and solar or super-solar at high metallicity. The derived sodium and aluminum abundances may be affected by departures from LTE (Gehren et al. 2004; Andrievsky et al. 2007, 2008; Lind et al. 2011). We have made no attempt to correct for this in our data, and only sodium in the Roederer et al. (2014b) comparison sample has been corrected. These corrections are negative and are large enough to explain the offset between the [Na/Fe] ratios in our study and those in the Roederer et al. study at low metallicity. Otherwise, the results for the 13 stars in our sample with phosphorus detections generally overlap with the comparison samples.

Zoom In Zoom Out Reset image size

The weighted mean [P/Fe] for all 13 stars in our sample is +0.04 ± 0.07 (σ = 0.25). The scatter about the mean, 0.25 dex, is comparable to the typical uncertainties in the [P/Fe] ratios, which range from 0.17 dex to 0.26 dex. A one-sided Kolmogorov–Smirnov test of the null hypothesis that these 13 stars are drawn from a normal distribution with a mean of +0.04 and standard deviation of 0.25 dex returns a p-value of 0.48. The null hypothesis cannot be excluded with any measure of confidence, and there is no compelling evidence for cosmic scatter in the [P/Fe] ratios.

Dividing the sample at [Fe/H] =−1.0, however, reveals possible differences in the high- and low-metallicity groups of stars. For the stars with [Fe/H] < −1.0, excluding BD+44 493, we derive the weighted mean ratios as follows: 〈[P/Fe]〉 = +0.04 ± 0.10 (σ = 0.28), 〈[Na/Fe]〉 = +0.23 ± 0.06 (σ = 0.20), 〈[Mg/Fe]〉 = +0.34 ± 0.02 (σ = 0.06), 〈[Al/Fe]〉 = −0.85 ± 0.09 (σ = 0.19), 〈[Si/Fe]〉 = +0.35 ± 0.06 (σ = 0.16), and 〈[Ca/Fe]〉 = +0.41 ± 0.02 (σ = 0.05). The mean [P/Fe] ratio in the low metallicity stars is about 0.2 dex lower than that predicted by one of the models by Cescutti et al. (2012), [P/Fe] ≈+0.2 to +0.3 dex, based on their corrected yields for massive stars. At high metallicity, [Fe/H] >−1.0, a trend of decreasing [P/Fe] with increasing [Fe/H] may be present, and is given by [P/Fe] =−0.86(±0.32) [Fe/H] − 0.33(±0.16). A similar but shallower trend is also seen in the Caffau et al. (2011) data at high metallicity, [P/Fe] ≈−0.4 [Fe/H] + 0.0. Type Ia supernovae are predicted to produce much smaller quantities of ³¹P (Iwamoto et al. 1999), accounting for the [P/Fe] decrease at high metallicity.

Phosphorus' behavior is consistent with that of a primary element in the stars in our sample, in the sense that it may be synthesized from hydrogen in proportions roughly equal to, e.g., the iron yield. As shown in Figure 4, this behavior is similar to that of the α-elements, with a constant [P/Fe] at low metallicity. This behavior is also found in several odd-Z elements not found on the α chain, including potassium. Potassium has two stable isotopes, ³⁹K and ⁴¹K, and its solar abundance is dominated by ³⁹K (93%). We have not derived potassium abundances for our sample; however, an analysis of 38 subgiants with −3.0 ⩽ [Fe/H] ⩽−1.0 by Roederer et al. (2014b) finds 〈[K/Fe]〉 = +0.30 ± 0.02 (σ = 0.15). The supernova models of Woosley & Weaver (1995) predict that ³¹P and ³⁹K are both primary isotopes. ³¹P is produced by neutron-capture on the products of the oxygen- and neon-burning shells in their models, and its ejection is not very sensitive to the explosion mechanism. ³⁹K is also produced by oxygen burning. Our results support predictions that phosphorus, like potassium, may have a primary origin at low metallicity.

Figure 4 also shows the [Si/Fe] and [P/Fe] ratios derived from detections of Si ii and P ii in un- (or minimally) depleted gas in four DLAs (Outram et al. 1999; Molaro et al. 2001; Levshakov et al. 2002; Lopez et al. 2002). These authors conclude that neither phosphorus nor iron are significantly depleted from the gas phase on dust grains in these systems. The absorbing clouds span redshifts 2.33 ⩽z ⩽ 3.39. No abundance information is available for sodium, magnesium, aluminum, or calcium in these DLAs. Our stellar abundances overlap the range of the DLA abundances, suggesting that the Galactic stars and DLAs could have been enriched by metals produced by similar nucleosynthesis channels in the early universe.

We defer more detailed comparisons of our data with supernova yield predictions and chemical evolution models to a companion paper (Jacobson et al. 2014).

We have detected several P i absorption lines in the NUV spectra of late-type metal-poor stars. All of these detections are made using publicly available STIS observations. We have mainly used the P i 2136.18 Å line to derive phosphorus abundances in 13 stars with −3.3 < [Fe/H] <−0.1. The P i 2553.26 Å line has been used to derive an upper limit in one star with [Fe/H] =−3.8. Departures from LTE are likely to have a negligible impact on the derived phosphorus abundances. We also estimate the uncertainties resulting from an offset between the optical and NUV abundance scales, revealed by examination of Fe i and Fe ii lines, whose cause is currently undetermined.

The derived [P/Fe] ratios compare well with previous work by Caffau et al. (2011), who used several NIR P i multiplets as abundance indicators. Our results also overlap the range of [P/Fe] ratios found in five undepleted DLA systems at high redshift. The six stars in our sample with [Fe/H] >−1.0 show a decreasing trend of [P/Fe] with increasing [Fe/H], as Caffau et al. also found. The stars with [Fe/H] <−1.0 show a constant [P/Fe] ratio, +0.04 ± 0.10, which is in marginal agreement with predictions made by the chemical evolution models of Cescutti et al. (2012).

To improve the abundance precision, the dominant sources of uncertainty must be addressed. A significant fraction of the uncertainty in the log gf value of this line comes from the measurement uncertainty of the lifetime of its upper level (known to 13%); modern techniques routinely reach a precision of 5% (e.g., Den Hartog et al. 2011) or better (e.g., Moehring et al. 2006). Another large source of uncertainty is the offset between the optical and NUV abundance scales. A differential analysis of the same (set of) P i line(s) in stars with similar stellar parameters would minimize the uncertainty in relative abundances. Practical limitations imposed by the stiff competition for HST observing time will limit future work to small numbers of bright stars, but such observations provide the only demonstrated approach to studying the nucleosynthetic origins of phosphorus in the early Galaxy.

We wish to express our appreciation to the many investigators whose observations have made this project possible. We thank R. Cooke for providing an unpublished DLA abundance that appears in Figure 4. We also thank the referee for several excellent suggestions that have improved this manuscript. Some of the data presented in this paper were obtained from the Mikulski Archive for Space Telescopes (MAST), the ESO Science Archive Facility, and the Keck Observatory Archive. STScI is operated by the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS5-26555. This research has made use of NASA's Astrophysics Data System Bibliographic Services, the arXiv pre-print server operated by Cornell University, the SIMBAD and VizieR databases hosted by the Strasbourg Astronomical Data Center, and the ASD hosted by NIST. IRAF is distributed by the National Optical Astronomy Observatories, which are operated by the Association of Universities for Research in Astronomy, Inc., under cooperative agreement with the National Science Foundation. Generous support for Program AR-13246 has been provided by NASA through a grant from the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy, Incorporated, under NASA contract NAS 5-26555. The work of T.T. and E.T. was supported by the MIT UROP program. A.F. is supported by NSF CAREER grant AST-1255160.

Facilities: ESO:3.6m (HARPS), HST (STIS), Keck:I (HIRES), Magellan:Clay (MIKE), Smith (Tull), VLT:Kueyen (UVES)