To appear in The Astrophysical Journal Boron in Very Metal-Poor Stars1 2 3 4 4 Ram´on J. Garc´ıa L´opez , David L. Lambert , Bengt Edvardsson , Bengt Gustafsson , Dan 5 2 Kiselman , and Rafael Rebolo 8 9 9 1 n a J 6 1 1 v 7 6 1 1 0 8 9 / h p - o r t s a : v i X r a 1Based on observations obtained with the NASA/ESA Hubble Space Telescope (HST) at the Space Telescope Science Institute, which is operatedby the Associationof Universities for Researchin Astronomy, Inc., under NASA contract NAS5-26555 2Instituto de Astrof´ısica de Canarias,E-38200 La Laguna, Tenerife, Spain 3 Department of Astronomy, University of Texas, RLM 15.308,Austin, TX 78712, USA 4Uppsala Astronomical Observatory,Box 515, SE-751 20 Uppsala, Sweden 5The Royal Academy of Sciences, Stockholm Observatory, SE-133 36 Saltsjo¨baden, Sweden – 2 – ABSTRACT We have observed the B I λ 2497 ˚A line to derive the boron abundances of two very metal-poor stars selected to help in tracing the origin and evolution of ◦ this element in the early Galaxy: BD +23 3130 and HD 84937. The observations were conducted using the Goddard High Resolution Spectrograph on board the Hubble Space Telescope. A very detailed abundance analysis via spectral synthesis has been carried out for these two stars, as well as for two other metal-poor objects with published spectra, using both Kurucz and OSMARCS model photospheres, and taking into account consistently the NLTE effects on the line formation. We have also re-assessed all published boron abundances of old disk and halo unevolved stars. Our analysis shows that the combination of high effective temperature (Teff ∼> 6000 K, for which boron is mainly ionized) and low metallicity ([Fe/H] ∼< −1) makes it difficult to obtain accurate estimates of boron abundances from the B I λ 2497 ˚A line. This is the case of HD 84937 and three other published objects (including two stars with [Fe/H] ∼ −3), for which only upper limits can be established. BD +23◦3130, with [Fe/H] ∼ −2.9 and log N(B)NLTE = 0.05±0.30, appears then as the most metal-poor star for which a firm measurement of the boron abundance presently exists. The evolution of the boron abundance with metallicity that emerges from the seven remaining stars with Teff < 6000 K and [Fe/H] < −1, for which beryllium abundances were derived using the same stellar parameters, shows a linear increase with a slope ∼ 1. Furthermore, the B/Be ratio found is constant at a value ∼ 20 for stars in the range −3 < [Fe/H] < −1. These results point to spallation reactions of ambient protons and α particles with energetic particles enriched in CNO as the origin of boron and beryllium in halo stars. Subject headings: Galaxy: evolution — nuclear reactions, nucleosynthesis, ◦ abundances — stars: abundances — stars: individual (BD +23 3130, HD 84937) — stars: late-type — stars: Population II – 3 – 1. Introduction Understanding the origins of the chemical elements is a fundamental goal of contemporary astrophysics. As this understanding has advanced in recent years, a few elements have been rather neglected. Boron is one such element. A primary reason for its neglect has been that stellar absorption lines of the boron atom and ions are not detectable in spectra obtainable from the ground; the low abundance of boron means that this element is detectable only through its strong resonance lines which are in the ultraviolet. The advance in ultraviolet stellar spectroscopy made possible by the Hubble Space Telescope (HST) and its Goddard High Resolution Spectrograph (GHRS) has led to new data on the boron abundance of cool and hot stars. Pioneering work on stellar boron abundances was presented by Boesgaard & Heacox (1978) who analyzed the B II λ 1362 ˚A resonance line in spectra of early-type stars observed with the Copernicus satellite. To map the evolution of boron with metallicity it is necessary to observe unevolved metal-poor stars. Since these are cool, boron detection has to be made using the atomic resonance lines. No advance in our knowledge of the boron abundance came with the International Ultraviolet Explorer (IUE) satellite (an attempt to detect the B I resonance lines near λ 2497 ˚A in a metal-poor star was made by Molaro 1987). Duncan, Lambert, & Lemke’s (1992) detection and analysis of the B I λ 2497 ˚A lines in three metal-poor stars observed with the GHRS has been followed by other studies of these lines – see Edvardsson et al. (1994), Rebull et al. (1996), and Duncan et al. (1997a, 1998). Kiselman (1994) and Kiselman & Carlsson (1996) showed that non-local thermodynamic equilibrium (NLTE) effects have to be considered in analyses of the 2497 ˚A lines. Standard big bang nucleosynthesis calculations do not predict a significant production 6 9 10 11 of Li, Be, B, and B, which are thought to be produced by spallation reactions between cosmic rays and CNO nuclei (see Ramaty, Kozlovsky, & Lingenfelter 1996 and Ramaty et al. 1997 for recent discussions on the theoretical yields). Models of inhomogeneous big bang nucleosynthesis suggest that observable amounts of Be and B could be produced (e.g. Malaney & Fowler 1989; Kajino & Boyd 1990; Jedamzik et al. 1994), but this possibility has been questioned on the theoretical grounds (e.g. Terasawa & Sato 1990; Reeves et al. 1990). In this paper, we derive the boron abundances of two metal-poor stars observed by ◦ us to fill in specific places of the study of the evolution of boron. BD +23 3130 is a cool very metal-poor star ([Fe/H]∼ −2.9) selected to extend B abundance determinations to a star of a factor of two more deficient than HD 140283 ([Fe/H]∼ −2.7; previously studied by Duncan et al. 1992 and Edvardsson et al. 1994), and is probably the most metal-poor slightly evolved star for which boron can be measured with HST. Duncan et al. (1997a) – 4 – has also observed two stars (BD −13◦3442 and BD +3◦740) with metallicities [Fe/H]∼ −3; however, they are much warmer and, hence, boron is mainly ionized making more difficult to obtain a precise measurement of its abundance from the B I lines. Our second star, HD 6 84937, is of special interest because Li (another expected product of spallation and fusion reactions driven by cosmic rays) has been traced in its atmosphere (Smith, Lambert, & Nissen 1993; Hobbs & Thorburn 1994, 1997). Both stars also have previous measurements 7 of Li in the literature, and have been included in the samples for deriving beryllium and oxygen (from UV OH lines) abundances in metal-poor stars by Garc´ıa L´opez et al. (1998) and Israelian, Garc´ıa L´opez, & Rebolo (1998), respectively, using the same stellar parameters. To better study the dependence of the derived abundances on the tools usually employed in the analyses, we present here the first detailed comparison of boron abundances derived using the two sets of model photospheres widely employed in the analyses of stellar spectra: those calculated with the ATLAS program, constructed by R. L. Kurucz, and those from the MARCS program, developed at Uppsala Astronomical Observatory. In both cases there have been considerable recent developments in these codes and their underlying input data. This careful analysis of boron abundances in very metal-poor stars, in which we also re-assess published abundances of old disk and halo unevolved stars, is aimed at better constraining observationally the nucleosynthetic origins of boron. 2. Observations ◦ Observations of the stars BD +23 3130 and HD 84937 were carried out using the GHRS in the low resolution G270M/SSA mode (λ/∆λ ∼ 25000). The G270M grating was centered at λ 2497 ˚A to observe the B I doublet located at λλ 2496.772 and 2497.723 ˚A (air values). The standard FP-SPLIT mode, which provides spectra in slightly different carrousel positions, was used to obtain high signal-to-noise ratios (S/N) and minimize the effects of the photocathode glitches and the geomagnetically-induced image motion problem (GIMP). Exposure times were estimated from the UV fluxes measured by IUE at λ 2500 ˚A ◦ for our targets. BD +23 3130 is a cool metal-poor giant with B = 9.59 and was observed with HST during ten orbits (21434 sec exposure time) on 1995 August 4, while HD 84937 is a hotter metal-poor dwarf with B = 8.69 which required three orbits (5440 sec) on 1996 March 21. Data reductions were performed by standard procedures using the “stsdas” package of – 5 – 6 the IRAF suite of programs. Final S/N values of 100 and 125 per diode (50 and 63 per ◦ pixel) were achieved for BD +23 3130 and HD 84937, respectively. Their corresponding measured continuum fluxes at λ 2500 ˚A were 2.56×10−13 and 1.46×10−12 erg cm−2 s−1 ˚A−1. These values are about 30–40% larger than those measured by IUE at that wavelength, and these differences are reduced to 20–25% when the HST data are degraded to match the IUE spectral resolution. Differences in the flux calibrations of both instruments should explain these discrepancies. The wavelength calibration provided a dispersion of 0.024 ˚A pixel−1, and the final spectra were shifted in wavelength to correct for stellar and Earth motions using the Fe I λ 2496.533 ˚A line as the reference. Continuum normalization is a difficult task in a crowded spectral region like this one. In principle, the HST spectra have been corrected for the spectrograph efficiency versus wavelength. A soft curve was fitted to the upper points of the spectra, assigning a value of 1 to those pixels corresponding to the region just shortwards of λ 2500 ˚A, considered by Duncan et al. (1992; 1997b) and Edvardsson et al. (1994) as the most line-free region in the data. Spectral syntheses performed afterwards showed that this criterion agrees well with the predictions of synthetic spectra. Figure 1 shows a portion of the final spectra for both stars. 3. Stellar Parameters Effective temperatures (Teff) were estimated initially using the Alonso, Arribas, & Mart´ınez-Roger (1996a) calibrations versus V − K and b −y colors, which were derived applying the Infrared Flux Method (Blackwell et al. 1990), and cover a wide range in ◦ spectral types and metal content. Both stars, BD +23 3130 and HD 84937, belong to a large sample of metal-poor stars for which Garc´ıa L´opez et al. (1998) have derived beryllium abundances. All the details on photometric measurements, as well as on the selection of metallicity values for these two stars are provided in that work. Surface gravities (logg) were determined by comparing the observed Str¨omgren b−y and c1 indices with synthetic ones generated using the corresponding filter transmissions and a grid of Kurucz (1992) blanketed model atmospheres fluxes. This is also explained in detail by Garc´ıa L´opez et al. (1998), where a comparison is given with the spectroscopic estimates by other authors. In particular, Nissen, Høg, & Schuster (1997) have recently derived surface gravities for a sample of 54 metal-poor stars using Hipparcos parallaxes to 6IRAFisdistributedbyNationalOpticalAstronomyObservatories,whichisoperatedbytheAssociation ofUniversities for Researchin Astronomy,Inc., under contractwith the NationalScience Foundation,USA. – 6 – determine luminosities. For ten stars in common, the gravities derived by Garc´ıa L´opez et al. are slightly lower systematically, with a mean difference of 0.22±0.13 dex in logg. As we will see below, a difference of this magnitude does not affect the boron abundances derived from the B I λ 2497 ˚A lines in metal-poor stars. BD +23◦3130 had been generally considered in the past as a subgiant star with logg ∼ 3 −3.5, but the comparison of its Str¨omgren indices with our synthetic uvby photometry indicates that it is a giant star with logg ∼ 2.5. Garc´ıa L´opez et al. (1998) computed synthetic spectra in the region around the Be II doublet at λ 3131 ˚A (used to derive beryllium abundances) for different combinations of the stellar parameters, within the error bars provided by the photometric calibrations, and Table 1 lists the values which best reproduce the observations. These values also reproduce ◦ well the B I region observed in BD +23 3130 and HD 84937, with the exception of the derived metallicity which is slightly higher than that found from optical and near-UV analyses, as explained in the next section. Apart from our observations, we have also derived boron abundances for Procyon and for the other metal-poor stars with available measurements of the B I λ 2497 ˚A doublet. For Procyon we choose Teff = 6750 K, logg = 4.0 and [Fe/H]= 0.0, to be consistent with the analysis of Duncan et al. (1997b; see below). For consistency, we also estimated the stellar parameters for the other stars by the methods applied to our stars. Most of them were studied by Garc´ıa L´opez et al. (1998). All metal-poor stars considered for boron abundances analysis in this work are listed in Table 1. 4. Abundances Analysis 4.1. Oscillator Strengths Due to the general lack of accurate oscillator strengths for the lines in the B I region, we used a line list computed by comparing LTE synthetic spectroscopy with observed spectra of some stars. The initial line list was taken from Duncan et al. (1997b) who analyzed dwarfs and giants of approximately solar metallicity. The list was drawn from the Kurucz (1993) LOWLINES list with wavelengths measured in the laboratory but the loggf values were adjusted to reproduce the spectra of Procyon, the Sun, α Cen A & B, and the metal-poor star HD 140283. Several Fe I lines predicted by Kurucz as well as other lines of the same element studied by Johansson & Cowley (1988) and Johansson (1992) were added. Duncan et al.’s (1997b) line list is very similar to that compiled by Duncan et al. (1992). Synthetic spectra were obtained by us using the code WITA2, a UNIX-based version – 7 – of ABEL6 code (Pavlenko 1991), which computes LTE atomic and molecular synthetic profiles for a given model atmosphere. We employed initially a grid of models provided by Kurucz (1992), which are interpolated for given values of Teff, logg, and [Fe/H]. To test the behavior of the Duncan et al. (1997b) line list using WITA2 and a high-resolution HST spectrum, we compared a λ/∆λ ∼ 90000 spectrum of Procyon (observed by Lemke, Lambert, & Edvardsson 1993) with synthetic spectra obtained using the stellar parameters cited above, a microturbulent velocity ξ = 1.0 km s−1, a wavelength step of 0.009 ˚A, and convolved with a gaussian with a FWHM of 0.027 ˚A to reproduce the observed spectral resolution. The stronger B I line at λ 2497.723 ˚A is seriously blended with other lines, and the abundance determination relies only on the B I line at λ 2496.772 ˚A, more isolated and weaker, which is blended with lines of Co I and Fe I at λλ 2496.708 and 2496.870 ˚A, respectively. O’Brian & Lawler (1992) and Johansson et al. (1993) provide accurate values (which differ by only 0.03 dex) for the oscillator strength of this B I line. To improve the fit of the synthetic to the observed spectrum in the region surrounding the weaker B I line, we varied the loggfs of 13 of the 20 lines present in the 2496.4–2497.4 ˚A region until a good fit was obtained. In particular, we adopted the loggf value provided by O’Brian et al. (1991) for the Fe I λ 2496.533 ˚A line (with an uncertainty of 9%), the Fe I line λ 2496.933 ˚A was not taken into account, and we included two weak Fe I lines at λλ 2497.153 and 2497.420 ˚A from the list used by Duncan et al. (1992), modifying also their loggfs by 0.1 and 0.25 dex, respectively. Figure 2 shows the comparison between the observed spectrum of Procyon and five synthetic spectra computed for different values of the boron abundance. The set of loggfs employed to compute them reproduces the spectrum of Procyon with a similar quality to those illustrated by Lemke et al. (1993) and Duncan et al. (1997b). The boron abundance which best matches the observed spectrum is a value between log N(B)= 2.0 and 2.1 (in the customary scale where log N(B)=log (B/H)+12). The sensitivity of the observed feature to the boron abundance of the computed line is high, as can be seen in Figure 2b: changes of 0.1 dex around log N(B)∼ 2.0 can be appreciated. Taking into account the uncertainty in the continuum placement and a change of 50 K in the assigned effective temperature, the boron abundance provided by this analysis is log N(B)= 2.05±0.20. Since the possibility of unidentified lines located close to the B I λ 2496.772 ˚A line cannot be excluded, this abundance should be considered strictly as an upper limit. Our LTE boron abundance for Procyon is 0.15 dex smaller and 0.15 dex higher than those obtained by Duncan et al. (1997b) and Lemke et al. (1993), respectively. In the latter case, the model photosphere employed was computed using the MARCS code with the opacity sampling method (OSMARCS) as described by Edvardsson et al. (1993). Using the LTE code SPECTRUM included in the Uppsala Synthetic Spectrum Package, – 8 – we have computed synthetic spectra for Procyon with the line list previously compiled and an OSMARCS model atmosphere. In this case, the boron abundance which best matches the observed spectrum is between log N(B)= 1.9 and 2.0, just 0.1 dex smaller than the value obtained using a Kurucz model. Figure 3 shows the fits to the observed spectrum using both sets of model photospheres and a common boron abundance log N(B)= 2.0. It can be seen that they are very similar, being the larger differences in the wavelength range 2496.8 to 2497.2 ˚A (very slight changes in the gf values used would be needed to obtain a similar good fit with the OSMARCS model), but not affecting significantly the abundance derived from the B I λ 2496.772 ˚A line. Kurucz models used in our analyses were computed with a mixing length parameter α = 1.25, and we have checked (by running the program ATLAS9 using the UNIX-based version assembled by M. Lemke and distributed through 7 the CCP7 database ) that these models do not include overshooting. OSMARCS models do not include overshooting either and were computed with α = 1.5. A very good agreement between spectra computed with both sets of models was found also by Garc´ıa L´opez, Severino, & Gomez (1995) for the analysis of the Be II λ 3131 ˚A region, showing negligible differences in beryllium abundances (using MARCS models treating the line blanketing by means of opacity distribution functions and applying wavelength-dependent scaling factors − to the total H opacity). The very good agreement between these independent analyses in a metal-rich star like Procyon gives us a certain level of confidence to proceed to derive boron abundances in other stars. It should be noted, however, that gross similarities among the analyses does not preclude the possibility of systematic errors. Furthermore, the presence of greater uncertainties in the model photospheres for metal-poor stars (e.g. Gratton, Caretta, & Castelli 1996) will constrain also the validity of abundances derived in those stars. 4.2. BD +23◦3130 and HD 84937 4.2.1. Spectral Synthesis Using our line list and the stellar parameters listed in Table 1, we computed synthetic spectra of the two stars observed by us using both Kurucz and OSMARCS models. Computations were made with a step of 0.024 ˚A in wavelength and convolved with a gaussian to account for the spectral resolution of the observations. We had initially assigned a metallicity of −2.9 to BD +23◦3130 following the Be analysis by Garc´ıa L´opez 7http://star.arm.ac.uk/ccp7/ – 9 – et al. (1998); however, a better overall fit to the B I region is obtained when values of [Fe/H]= −2.6 and −2.7 are used for Kurucz and OSMARCS models, respectively. Figure 4 shows the comparison between observed and synthetic spectra (computed for different boron abundances) of this star using a Kurucz model. The major features surrounding the B I λ 2496.772 ˚A line are reproduced, although they are not fully fitted. The sensitivity of the synthetic spectra to boron abundance is seen in detail in Figure 4b. The boron abundance of BD +23◦3130 is log N(B)= −0.3, and changes of 0.1 dex around this value are clearly appreciated. The spectrum computed with log N(B)=−3 (no boron) shows much smaller features for both lines of the B I doublet, indicating that we have a positive detection of boron in this star. The errors in the boron abundance associated with variations of ±150 K in Teff, ±0.3 dex in logg, and ±0.3 dex in [Fe/H] are ±0.2, 0, and ±0.2, respectively. The B I line is too weak to be sensitive to the uncertainty of the adopted microturbulence −1 (ξ = 1.0 km s ). Placement of the continuum was checked by computing a synthetic spectrum in the 2499–2501 ˚A region using the Duncan et al. (1997b) line list at those wavelengths. At λ 2499.96 ˚A both synthetic and observed spectra reach a value of 1. Taking into account the S/N and the possible errors in normalizing the observed spectrum, an additional error of ±0.1 dex can be considered. Combining all these uncertainties, the final LTE boron abundance of BD +23◦3130 is log N(B)= −0.30±0.30. The analysis carried out with the same stellar parameters (except [Fe/H]) using OSMARCS models provided an abundance log N(B)= −0.40, showing the same sensitivity to changes in boron abundance as well as similar uncertainties related to the choice of the stellar parameters and continuum location. A comparison between the best fits to the observed spectrum obtained with both sets of models is shown in Figure 5. A similar procedure was followed to obtain the boron abundance of HD 84937. A comparison between observed and synthetic spectra computed with a Kurucz model is shown in Figure 6. The B I line is weaker for a given abundance in HD 84937 than in ◦ BD +23 3130 because boron is more ionized. The observed spectrum is reproduced with a value between log N(B)= 0.2 and 0.0 (Figure 6b) but, at this S/N level and with the uncertainties present in the parameters of the lines surrounding the B I line, a fit to the synthetic spectrum computed without boron (log N(B)= −3), or with other intermediate boron abundance, cannot be ruled out. This means that log N(B)= 0.1 must be considered an upper limit to the boron abundance in this star. As in the previous case, it was necessary to increase the adopted metallicity from [Fe/H]= −2.25 to −2.1 to obtain an overall good fit. Figure 7 shows the comparison between the fits to the observed spectrum obtained using Kurucz and OSMARCS models with the same stellar parameters and boron abundance. It can be seen how the overall behavior is very similar in both cases, as well as the upper limit at log N(B)= 0.1. The uncertainties of ±120 K in Teff, ±0.3 dex in logg, ±0.15 in – 10 – [Fe/H], and continuum placement will change this value within ±0.25 dex. Taking these uncertainties into account, Table 1 reflects a conservative upper limit log N(B)< 0.35 for this star. ◦ Spectral lines appear to be broader in BD +23 3130 with respect to HD 84937. The instrumental profile should be the dominant contributor to the line widths and this ought ◦ to be the same for the two stars. Additional broadening would arise if BD +23 3130 were a spectroscopic binary such that a weaker line was present slightly shifted to lower wavelengths. However, this star does not appear as a binary or suspected binary star in the literature, especially in the survey of proper motions of about 500 stars studied by Carney et al. (1994). On the other hand, HD 84937 is listed as a suspected binary in the catalog of Carney et al. and classified as a certain binary by Stryker et al. (1985) from radial velocity measurements, although is not considered as a possible binary by Nissen et al. (1997). Duncan et al. (1997b) observed two giant K-type stars belonging to the Hyades, as well as the field giant β Gem. The composite spectrum of the Hyades stars and the spectrum of β Gem show similar broad lines which are, in general, deeper than those observed in ◦ BD +23 3130 due to the big differences in metallicity. In any case, the spectral syntheses carried out by us and by Duncan et al. (1997b) show a very high sensitivity of the B I feature to the boron abundance in giant stars. 4.2.2. Stellar Fluxes A complementary test to check the validity of the model photospheres employed in the analysis consists in comparing predicted and observed stellar fluxes at the wavelength region under study. This is especially important in the ultraviolet, where an enormous number of lines are present such that the models may not account for the line blanketing. We exploit the absolute flux calibration provided with the HST spectra by following the Duncan et al.’s (1992) procedure, using the stellar magnitude at the K (2.2 µm) band to ◦ normalize the fluxes. K magnitudes for our stars (6.98 for BD +23 3130 and 7.10 for HD 84937) were taken from Laird, Carney, & Latham (1988). Theoretical fluxes corresponding to the Kurucz models were obtained by interpolating the grid of models currently available at the CCP7 database, while they were computed individually for the OSMARCS models used in the analyses. Below λ 4500 ˚A OSMARCS models are computed using opacity sampling techniques, while at longer wavelengths opacity distribution functions (ODF) are employed. The latter method is used for computing Kurucz models at all wavelengths. Comparisons between both set of models show a very good agreement for IR wavelengths (λ ∼> 1 µm), which