We have taken into account vignetting corrections as well as the
March 1992 version of the PSPC response matrix (see Paper I for further
details about using this matrix). Moreover, we have taken into account
the systematic errors of the adopted response matrix computed in the
Appendix of Paper I by adding them in quadrature to the statistical
errors. This procedure is required because of the high number of counts
in the PSPC spectral channels. In the fitting process, we let 
and 
vary, as well as the normalization that provides directly a
measure of emission measure and, less directly (see below), an estimate
of plasma density. Moreover, we have included as a free parameter the
hydrogen column density N
. We have considered the PSPC channels in
the range 3-30, since channels 31-34 (corresponding to energies in the
range 2.02-2.48 keV) have too few counts for the validity
of the 
statistics in the fitting (see Figure 3 as an
example). With this choice, the number of degrees of freedom is 24.
This parametric fit approach is expected to give reliable results since: (i) the high statistics of our data allows us to determine parameters with quite small uncertainties (see Table 2); (ii) line strengths in the X-ray domain are strongly dependent on the ionization time (Figure 1; see also Brinkmann 1988) and this effect on the spectrum is measurable even with a low energy resolution detector like the ROSAT PSPC (Bocchino 1993). Since the results of the fitting procedure in some cases depend on the initial parameters guess, we have taken care of doing several trials with different starting points for the fitting to avoid convergence to local minima.
In summary, if we adopt the 95% confidence level (i.e. 
with 24 degrees of freedom), the model is acceptable in 21 spatial bins,
most of which are located around positions g3 and e6 (Region
E,F, and G in Paper I), whereas is rejected in the remaining 27 bins, like
d3, b5, d5 (Region C, H and A in Paper I). In the map
(Figure 2b), we show only those spatial bins where our STNEI
model is acceptable; the remaining bins are not displayed.
In Figure 3, we show spectra and best-fit STNEI models for two
sample bins where we found acceptable values (e6, 
counts; c7, 
2600 counts) but different parameter
values (T=0.15 keV, 
in e6 and T=0.21 keV,
in c7). The unfolded model spectra are also
shown in the right-hand panels, and indicate the different contribution
from the line emission to the total X-ray emissivity in the two cases.
In Figure 4, we show an example of spectrum which could
not be fitted successfully (d3, 15200 counts): in this bin our
model underestimates the observed spectrum in the 0.6-1.0 keV energy
range, and this characteristic is common also to other spectra with
unacceptable values.
In Table 2 we summarize the statistical errors associated
to T and 
, computed at the 99% confidence level
in the case of two interesting parameters (Lampton, Margon & Bowyer 1976).