In this section we estimate the accuracy of our results.
The fine 
mesh below E = 1.789 Ry,
for excitation to the lower-lying, even-parity
levels, fully resolves the important resonance features
associated with the sextet and quartet odd-parity levels
up to the 
levels.
Therefore, rate coefficients for this
type of transitions should be highly accurate, 
%.
For optically allowed transitions from the low-lying levels to the
sextet and quartet odd parity levels with energies below 1.789 Ry, the
rate coefficients should also be of the same accuracy since resonances
are relatively less important
and the collision strengths are large and
dominated by the higher partial waves, as seen from Fig. 2 (click here).
For the forbidden (and inter-combination) transitions
from the low-lying levels to intermediate energy levels, up to
the levels, and transitions between
these intermediate-energy levels, the accuracy of the rate coefficients
is expected to be less, 
%. For transitions corresponding to
the high-lying levels with threshold energies greater than 1.789 Ry,
the uncertainty could exceed 50%
since a coarse energy mesh was used and the resonances and coupling
effects due to higher terms were neglected.
We would also emphasize here that for all transitions the
maxwellian-averaged collision strengths for high temperatures
(roughly larger than the highest threshold energy included in the
target expansion, about 300000 K in the present case)
could have a larger uncertainty, since resonances due to higher
target states are not included. However, data for these temperatures
are of little astrophysical interest. These general criteria should apply
to all our earlier publications in this series.
We hope the present work will provide a reasonably complete collisional dataset for extensive astrophysical diagnostics using Fe IV spectra from various sources.
Acknowledgements
We wish to thank Manuel Bautista for his contribution in obtaining the target wavefunctions and Dr. David Hummer, the coordinator of the IRON Project, for his comments. This work was supported by a grant (PHY-9421898) from the U.S. National Science Foundation. We are also grateful to the Ohio Supercomputer Center in Columbus, Ohio, for their support. Some of the computational work was carried out on the Cray Y-MP8/64, and the asymptotic code, STGFJ, was entirely run on the massively parallel Cray T3D at this center.