Our stellar sample is based upon that of SD93, who analysed IUE data on the UV resonance lines of 26 HgMn, 4 superficially normal and 10 normal stars. We observed 24 of the HgMn stars in the SD93 sample. The two excluded stars are 36 Lyn (which had been misclassified as a HgMn star) and HR 6000, which was not observed. Physical parameters of the 24 stars are given in Table 1.
Five stars are noted as double-lined spectroscopic binaries in
Table 1; one of these (HR 1800) is
better described as a
close visual binary in which we can see evidence of the secondary spectrum
as rotationally-broadened features. The parameters quoted in all five
cases are those adopted for the primary star. The adopted light ratios
(
) at 
4520 are from Harman [1997]: HR 4072,
5.45; 
Lup, 3.65; 
CrB, 2.70; or from Ryabchikova, Zakhorova
& Adelman [1996]: 112 Her, 6.3 from their Table 2. Suitable
adjustments for other wavelength regions were made. The adopted light
ratio for the visual binary HR 1800 (
arcsec) is 2.45,
based on 
mag from The Hipparcos Catalogue (ESA
[1997]). Another star, 33 Gem, is suspected of being double-lined
but there is not yet any information on the orbit or light ratio (Hubrig
& Launhardt [1993]); we treat it as a single star or `average
component.' We note that Adelman, Philip & Adelman [1996] also
treated 33 Gem as a single star, noting that the question of binarity
could not be conclusively resolved with their data.
Northern Hemisphere observations were taken with the Hamilton Échelle
Spectrograph (HES) - Vogt [1987] - at Lick Observatory, fed by the
0.6-m Coudé Auxilliary Telescope (CAT), during four runs in
1994-1997. Further details of the instrument can be found in Misch
[1997]. Stars south of 
were acquired as service
observations at the AAT using the UCL Échelle spectrograph. Shortly
before our observations in 1994, some of the HES optical components were
replaced, improving the resolution and instrumental profile, and making it
possible to use the full field of the 2048 
2048 CCDs to maximum
advantage. We used both the unthinned phosphor-coated Orbit CCD (Dewar
13) and from July 1995 the thinned Ford CCD (Dewar 6), depending on
availability as the latter is shared with the multi-object spectrograph on
the 3-metre telescope. The spectral range for the observations was
3800-9000Å, except for the AAT data which were only obtained in the
range 3700-4700Å, with the TEK2 CCD. Typical signal-to-noise (S/N)
per pixel in the centres of orders ranged from 75 to 250. The Orbit CCD
is cosmetically very clean, with very few bad pixels or columns, while the
thinned Ford CCD contains several column defects but offers a much higher
detector quantum efficiency in the blue. We used the Ford CCD whenever it
was available. With the slit settings used, the combination of
spectrographs and CCDs gave resolutions 
for the HES
and 
for the UCLES. Flat fields were made using polar
axis quartz lamps and wavelength calibrations were obtained with Th-Ar
comparisons.
The échelle spectra were extracted and calibrated using standard IRAF extraction packages (Churchill [1995]; Valdes [1990]), running on UCL's Starlink node. Previous measurements (Allen [1998]) showed that there were no measurable effects of parasitic light (residual scattered light) in the line profiles provided that general scattered light in the adjacent interorder spaces was taken as the subtracted background. In practice the residual scattered light was less than approximately 1 percent; we have therefore made no corrections for it. Allen's method is based on a direct comparison of the solar spectrum (as reflected from the roof of the CAT coelostat) observed with the HES, with the Kitt Peak Solar Flux Atlas (Kurucz et al [1984]). As the latter was obtained using a Fourier Transform Spectrometer, it has no measurable parasitic light. The KPNO spectrum is convolved with a suitable instrumental profile to match the HES data; both spectra must be normalised at the same points for a valid comparison. The ratio of summed equivalent widths of various features with good adjacent continuum points, in many different spectral orders, provides the measure of the amount of parasitic light.