Abundances were determined using the exact curve-of-growth technique. For
this method, we measured the equivalent widths, 
, of Mn
absorption lines in the programme spectra and compared these values to the
calculated curves of growth for each line, which were generated (assuming
LTE) by our spectrum-synthesis code UCLSYN (Smith & Dworetsky
[1988]; Smith [1992]). The necessary atmospheric parameters
given in Table 1 - 
, 
, and microturbulence
(
) - were taken from SD93,
except for 112 Her where we used the
values given by Ryabchikova et al. ([1996]). The values of 
are
taken from Dworetsky, Jomaron & Smith ([1998]). In the cases of
the five binaries with double spectra, we adopted the light ratios cited
in Section 2 in order to correct for dilution effects using the BINSYN code (Smalley [1996]), an extension of UCLSYN. We
used a combined grid of ATLAS9 model atmospheres: the 2 km s
grid of Kurucz [1993], and the COLK95 grid of Castelli et al.
[1997] for cooler stars, interpolating to produce a model at the
chosen 
and 
of each star.
Table 1: Stellar Parameters. Microturbulent velocity 
and projected
equatorial rotational velocity 
in km s
There appears to be no other published data on laboratory oscillator
strengths for visible-region MnII lines except for Warner [1967a].
We adopt the calculations of Kurucz [1990] taken from CD23. If we
adjust the Warner 
-values by the prescription of Smith
[1976], we find that the Kurucz calculations tend to give values
about 0.4 dex smaller for our lines from higher excitation levels. It is
not possible to use one set of data to `check' the other; a modern set of
measured absolute oscillator strengths for Mn II would be required. For
the Mn I lines, we adopt the oscillator strengths of Martin, Fuhr &
Wiese [1988], which are the same as those given in CD23. For both
neutral and ionised Mn we adopt 
given by Kurucz but
use UCLSYN's internal algorithms to estimate the Stark damping
parameter 
for each line. Van der Waals contributions to line
broadening are expected to be small; a suitable approximation by Warner
[1967b] was used.
Rather than measuring many tens of lines, of which some may be partially
blended or have poorly known gf values, relying on statistical
averaging to give a valid result which may nonetheless be biased, and
rather than fully synthesizing the profiles of many lines to mitigate the
blending problem, the approach followed here was to find a small sample of
Mn lines, all of which could be shown to be unblended over a wide range of
and composition. A search for such lines was carried out by Allen
[1998]. The following lines were selected as being appropriate for
a high-accuracy curve-of-growth analysis: Mn II 
4478,
4365, 4363, 4326, 4206, 3917; Mn I 
4030 and 4034.
Allen used the criteria that lines should be strong enough to appear in
most HgMn stars, be blend-free, and have well-determined atomic
parameters. However, Allen noted that 
4206 systematically
produced anomalously large abundances; the severity of the anomaly was
proportional to the line strength.
Figure 1: The six Mn lines used in this abundance analysis. These spectra
are of HR 7361. Histograms represent observations; continuous lines
represent synthetic spectra.
