2 Observations

   
2.1 Selection of stars

The stars were selected from the $uvby-\beta$ photometric catalogues of Olsen ([1983], [1993]) according to the criteria of $5800 \leq \mbox{$T_{\rm eff}$ }\leq 6400~\mathrm{K}$, $4.0 \leq \mbox{{\rm log}$g$ }\leq 4.5$ and $-1.0 \leq \mbox{\rm [Fe/H]}\leq +0.3$with approximately equal numbers of stars in every metallicity interval of 0.1 dex. In this selection, the temperature was determined from the b-y index with the calibration of Magain ([1987]), gravity was calculated from the c₁ index as described in EAGLNT, and metallicity was derived from the m₁ index using the calibrations of Schuster & Nissen ([1989]). The later redeterminations of the temperature with the calibration of Alonso et al. ([1996]) and the gravity from Hipparcos parallax lead to slight deviations from the selection criteria for some stars.

Based on the above selection, 104 F and G stars were observed, but 3 high-rotation ( $V \sin i \geq 25 \mbox{\rm\,km\,s$^{-1}$ }$) stars and 9 double-line spectroscopic binaries were excluded from the sample. Another 11 stars have radial velocity dispersions higher than the measurement error of the CORAVEL survey. HD106516A and HD97916 are suspected binaries (Carney et al. [1994]), and HD25998 and HD206301 are possibly variables (e.g. Petit [1990]; Morris & Mutel [1988]). These 15 stars (marked in the column "Rem'' of Table 3) are being carefully used in our study. The remaining stars are considered as single stars, but are checked for differences in iron abundances between Fe I and Fe II lines using gravities from Hipparcos parallaxes as suggested by Fuhrmann ([1998]). As described later, additional 2 stars were excluded during the analysis and thus the sample contains 90 stars for the final discussion and conclusions.

2.2 Observations and data reduction

The observations were carried out with the Coudé Echelle Spectrograph attached to the 2.16 m telescope at Beijing Astronomical Observatory (Xinglong, PR China). The detector was a Tek CCD ( $1024\times 1024$ pixels with $24\times 24~\mu {\rm m}^{2}$ each in size). The red arm of the spectrograph with a 31.6 grooves/mm grating was used in combination with a prism as cross-disperser, providing a good separation between the echelle orders. With a 0.5 mm slit (1.1 arcsec), the resolving power was of the order of 40000 in the middle focus camera system.

The program stars were observed during three runs: March 21-27, 1997 (56 stars), October 21-23, 1997 (27 stars) and August 5-13, 1998 (21 stars). The exposure time was chosen in order to obtain a signal-to-noise ratio of at least 150 over the entire spectral range. Most bright stars have $S/N~\sim~200 - 400$. Figure 1 shows the spectra in the region of the oxygen triplet for two representative stars HD142373 and HD106516A. In addition, the solar flux spectrum as reflected from the Moon was observed with a $S/N~\sim$ 250 and used as one of the "standard'' stars in determining oscillator strengths for some lines (see Sect. 4.2).

Figure 1: Examples of spectra obtained with the 2.16 m telescope at Xinglong Station for HD142373 ( $\mbox{$T_{\rm eff}$ }=5920$ K, $\mbox{{\rm log}$g$ }=4.27$, $\mbox{\rm [Fe/H]}=-0.39$) with a high S/N $\sim$ 400 and HD106516A ( $\mbox{$T_{\rm eff}$ }=6135$ K, $\mbox{{\rm log}$g$ }=4.34$, $\mbox{\rm [Fe/H]}=-0.71$) with a relatively low S/N $\sim$ 160

The spectra were reduced with standard MIDAS routines for order identification, background subtraction, flat-field correction, order extraction and wavelength calibration. Bias, dark current and scattered light corrections are included in the background subtraction. If an early B-type star could be observed close to the program stars, it was used instead of the flat-field in order to remove interference fringes more efficiently. The spectrum was then normalized by a continuum function determined by fitting a spline curve to a set of pre-selected continuum windows estimated from the solar atlas. Finally, correction for radial velocity shift, measured from at least 20 lines, was applied before the measurement of equivalent widths.

   
2.3 Equivalent widths and comparison with EAGLNT

The equivalent widths were measured by three methods: direct integration, Gaussian and Voigt function fitting, depending on which method gave the best fit of the line profile. Usually, weak lines are well fitted by a Gaussian, whereas stronger lines in which the damping wings contribute significantly to their equivalent widths need the Voigt function to reproduce their profiles. If unblended lines are well separated from nearby lines, direct integration is the best method. In the case of some intermediate-strong lines, weighted averages of Gaussian and Voigt fitting were adopted.

The accuracy of the equivalent widths is estimated by comparing them to the independent measurements by EAGLNT for 25 stars in common. Five of them were observed at the ESO Observatory ( $R \sim 60\,000$, $S/N~\sim$ 200) and 23 were observed at the McDonald Observatory ( $R \sim 30\,000$, $S/N~\sim~200 - 500$).

Figure 2: A comparison of equivalent widths measured in this work with ESO data in EAGLNT for 5 stars in common. The thick line is the linear fit to the points, whereas the dashed line is the one-to-one relation

The systematic difference between the two sets of measurements is small and a linear least squares fitting gives:

$$1.025 \left(\pm0.012\right) EW_{\mathrm{ESO}} + 0.89 \left(\pm0.56\right) \,\, ({\rm m}{\mbox{\AA}})$$ EW_Xl =

$$1.083 \left(\pm0.006\right) EW_{\mathrm{McD}} - 0.94 \left(\pm0.28\right) \,\, ({\rm m}{\mbox{\AA}}).$$ EW_Xl =

The standard deviations around the two relations are 3.8 mÅ (for 129 lines in common with ESO) and 4.3 mÅ (for 575 lines in common with McDonald). Given that the error of the equivalent widths in EAGLNT is around 2 mÅ, we estimate an rms error of about 3 mÅ in our equivalent widths. As shown by the comparison of our equivalent widths with ESO data in Fig. 2, the equivalent widths below 50 mÅ are consistent with the one-to-one relation. The deviations for the stronger lines may be due to the fact that all lines in EAGLNT were measured by Gaussian fitting, which leads to an underestimate of equivalent widths for intermediate-strong lines because of neglecting their wings. We conclude that the Xinglong data may be more reliable than the EAGLNT data for lines in the range of 50 < EW < 100 mÅ.