Observations for this programme have been carried out at four observatories using various combinations of telescopes, spectrographs and detectors. Table 1 (click here) gives an overview of the observing runs partly or entirely devoted to this project (totalling 222 nights). The columns in this table are, in order:
Table 1: Observing runs and instrumental configurations
Table 2: Ap stars with resolved magnetically split lines
Table 2: continued
Table 3: Mean magnetic field modulus: number of measurements,
average value, standard deviation, rms deviation from a
mathematical fit, estimated uncertainty (see text)
At the end of this section, we also give some information about the reduction procedures.
The largest fraction of the spectra were taken at the European
Southern Observatory, using the Coudé Echelle Spectrograph (CES).
A description of this instrument has been given by Lindgren &
Gilliotte (1989; see also the update by Mathys 1994a).
We used its long camera (LC) in
most runs, with as detectors a Reticon or one of the following CCDs
(referred to by their internal ESO number): RCA #9, Ford #24, Ford
#30, Loral #34, and Loral #38. All these detectors have a pixel
size of 15 
m (for the Reticon, this is the size in the dispersion
direction, while the CCDs have square pixels).
For a few runs, the observations were performed with the
short camera (SC) of the CES, always in combination with CCD RCA #9.
In general, light was fed into the CES through the 3-mirror coudé
train of the Coudé Auxiliary Telescope (CAT). The corresponding
instrumental configurations in Table 1 (click here) are denoted CES LC
or CES SC. In those configurations, the width of the spectrograph
entrance slit was adjusted to achieve a resolving power
of 
(with the LC, except with CCD #38), or of
(with the SC). With CCD #38, the resolution is
limited to approximately 3 pixels due to charge diffusion within
the chip itself, so that when
this CCD was used with the LC, the resolving power was only
.
In October 1992, some observations were performed replacing the
coudé train of the CAT by an optical fibre linking its prime focus
to the CES. Various fibres were used:
two of 200 
m of diameter (corresponding to 90 on
the sky), respectively blue and red optimized (with similar response
at the wavelength of our observations, 6150 Å), one of 135 
m of
diameter, and one of 50 
m of diameter. The 200 
m fibres were
used in combination with the long camera of the CES, and their output
was fed into the spectrograph through a Bowen-Walraven image slicer
giving 11 slices. This configuration is denoted ``CES LC/F200'' in
Table 1 (click here). The other two fibres were used together with the short
camera of the CES. The light coming out of the 50 
m fibre was fed
directly into the spectrograph, while that emerging from the 135 
m
fibre first passed through a 4-slice Bowen-Walraven image slicer.
These combinations are labeled ``CES SC/F50'' and ``CES SC/F135'',
respectively. The
resolving power achieved when using the fibres is determined by the
associated image slicer (or by the fibre itself in the case of the
50 
m fibre). They are of the order of 
for the
configuration CES LC/F200, and of approximately 
for the
other two configurations.
On two nights, light was fed to the CES from the Cassegrain focus of
the 3.6 m telescope
through a 200 
m optical fibre and an 11-slice image slicer. A
description of this configuration has been given by D'Odorico et
al. (1989). The resolving power obtained using it is approximately
.
A few spectra were also taken at ESO using the 3.5 m New Technology
Telescope (NTT) and the ESO Multi-Mode Instrument (EMMI; Zijlstra et
al. 1996),
with the R4 echelle grating #14. This grating is described by Dekker
et al. (1994). We used
it in the standard EMMI configuration, with the f/5.2 camera and the
CCD Tektronix #36 (
pixels of
m
). With the entrance slit width set to 08,
this configuration
yields a resolving power of 
over a broad wavelength
range (more than 2000 Å). The wavelength coverage is determined by
the grism which is used as cross-disperser. The spectra discussed
in this paper were taken with grisms #4, #5, and #6. The
ranges covered by these grisms all encompass the region around
6150 Å, and for the purpose of diagnosing the mean magnetic field
modulus from the splitting of the line 
, they all are equivalent.
Therefore, in Table 1 (click here), we do not distinguish between them.
At the Observatoire de Haute-Provence (OHP), we mostly observed for this
programme with the 1.52 m telescope (152) and the AURELIE spectrograph
(Gillet et al. 1994). The detector was a Thomson double linear
array (``barrette''), which has a pixel size of 
m
. We
used the 1200 grooves mm
grating #5 in the second order to
obtain a resolving power of 
.
A few additional spectra were recorded at OHP with the 1.93 m
telescope (193) and the cross-dispersed echelle spectrograph ELODIE
(Baranne et al. 1996). With the fixed configuration of this
instrument, and its CCD Tektronix (
pixels of
m
) one gets a resolving power of 
.
At Kitt Peak National Observatory (KPNO), we observed with the 0.9 m
coudé feed telescope and the coudé spectrograph (Willmarth
1996). We used the cross-dispersing grism No. 770 and the echelle
grating (31.6 groovesmm
) with Camera 5 (f/3.6). With a slit
width of 0.3 mm, we achieved a resolving power of 
using the CCD
TI #5 (
m
pixels).
Finally, a few spectra were recorded with the Canada-France-Hawaii
Telescope (CFHT) at Mauna Kea, and the f/4 coudé spectrograph
(GECKO; Glaspey 1993). The appropriate order of
the mosaic of four 316 groovesmm
echelle gratings was
selected using an interference filter. The
detector was the CCD Loral #3: with its pixel size of 15 
m, a
resolving power of 
was obtained.
The reductions were carried out using the image processing packages MIDAS (for the ESO and OHP data) and IRAF (KPNO and CFHT). The applied procedures were mostly standard, involving the following steps: electronic bias and scattered light subtraction, division by the spectrum of a white lamp for flat fielding, spectrum extraction (for two-dimensional detectors only), normalization to the continuum (fitting a suitable function through the highest points), wavelength calibration. For the latter, whenever possible (see below) the standard MIDAS option of rebinning the spectra to a constant wavelength step was not used. Instead each pixel was assigned a wavelength, which was found slightly but significantly more accurate. This is important, because the determination of the mean magnetic field modulus relies on measurements of small wavelength differences.
Figure 2: Same spectral region as in Fig. 1 (click here), as observed in
the stars identified next to each tracing. For the sake of clarity, the
wavelengths have been reduced to the laboratory reference frame
Figure 3: Same spectral region as in Fig. 1 (click here), as observed in
the stars identified next to each tracing. For the sake of clarity, the
wavelengths have been reduced to the laboratory reference frame
Figure 4: Same spectral region as in Fig. 1 (click here), as observed in the
stars identified next to each tracing. For the sake of clarity, the
wavelengths have been reduced to the laboratory reference frame
When appropriate, a median filter was applied
to remove cosmic ray hits. Care was taken not to introduce any
degradation of the line profiles when performing this operation. A few
spectra, in which the diagnostic line 
was badly affected by
cosmic events, have been discarded from this study. It cannot be ruled
out, though, that in a small number of cases, cosmic ray hits in this
line may have remained unnoticed: this may plausibly explain a
few outlying measurements of the magnetic field modulus.
The reductions of the spectra obtained with some instrumental configurations deserve some additional, more specific comments.
In the few spectra taken using as detector a RETICON with the ESO CES, the contribution of the RETICON dark current had to be subtracted. As described by Mathys & Solanki (1989), this was done through linear interpolation between the masked pixels at both ends of the array. With such a one-dimensional detector, though, the contributions of scattered or parasitic background light cannot be identified, thus they cannot be properly removed. However, the resulting error should be less than 0.5% of the continuum level.
Qualitatively similar but potentially more serious limitations affect the reduction of data recorded at OHP with AURELIE. Again, the one-dimensional format of the detector does not provide any information on the scattered light within the spectrograph. However, comparison of AURELIE spectra of the bright Ap star HD 137909 with quasi-simultaneous observations of this star performed with the KPNO cross-dispersed echelle spectrograph (where background contribution can in principle be accurately removed) makes us feel confident that the contribution of scattered light to our AURELIE spectra is mostly negligible. Of more concern is the high dark current of the Thomson ``barrette'', which was unstable. Errors in its subtraction may reach 1 to 2% of the continuum level in long exposures of faint stars. This uncertainty, of course, would affect equivalent width determinations. But it has essentially no impact on the determination of the mean magnetic field modulus.
In spectra recorded at ESO with the CES fed from the 3.6 m telescope or from the CAT though an optical fibre and an image slicer, the projected image of the slit on the detector within each slice is not quite perpendicular to the dispersion direction and shows some curvature. As a result, techniques developed to process long-slit spectra of extended sources must be used for the reduction. In short, the wavelength calibration is carried out individually for each row of the spectrum, and the extraction is performed only after each row has been rebinned to a constant wavelength step. Rebinning was achieved through cubic spline interpolation; the wavelength step was chosen small enough to keep the line profile degradation to a minimum.
In the CFHT spectra, the dispersion direction is along the CCD rows, and offsets of a number of columns had to be manually corrected before proceeding to the rest of the reduction, which is standard.
Not surprisingly, given the large number of spectra obtained with many
different combinations of telescopes and instruments, the quality
finally achieved after reduction is uneven. The vast majority of the
spectra, including virtually all those recorded with the ESO CES and a
large fraction of those obtained at OHP with AURELIE, have
signal-to-noise (S/N) ratios between 70 and 200. Typically, the ratio is
higher for brighter stars, although there are deviations from this
trend. Spectra recorded with smaller telescopes also tend to be
noisier. For the faintest stars (V=7.7) that could be
observed with the 0.9 m Coudé feed at KPNO, the S/N ratio in the
continuum at 6149 Å was limited to about
50. This results not only from the small size of the telescope, but
also from the low instrumental
efficiency due to the fact that the line of interest,
, is far from the blaze in the two orders where it can be observed
(91 and 92).
At ESO, with the 1.4 m CAT and the long camera of the CES, on the
best nights, S/N=80 was achieved in 2 hours of exposure on
HD 119027, the faintest (V=10.0) star with resolved lines presently
known. The 1.5 m telescope feeding AURELIE at OHP has about the same
size of the ESO CAT, but the AURELIE magnitude limit is about one
magnitude lower than for the 
, due to the less good sky
transparency of OHP compared to ESO and to the high readout noise of
the Thomson ``barrette''.
