Unfortunately the strategy proposed by Uwe (thank you so much for your quick and kind reply anyway) doesn't seem to work, or, at least, I was not able to make it work.
Indeed aoforce finds five atoms for the NH3Q system instead of four (NH3), so it needs all 9 internal coordinates and -since it found only six internals- exits with the following warning:
[...]
WARNING: The internal coordinate analysis
needs all 3N-6/3N-5 internal coordinates!
I ignore the keyword!
MODTRACE: no modules on stack
Huch
force ended abnormally
force ended abnormally
program stopped.
I tried to define three more internal coords assigning them to 'f' or 'i' types, but I did not success in getting normal modes and obtained the above warning again (aoforce is not 'stupid', of course! :-)
I also defined three more 'k', internal coordinates and run a dscf/grad step, followed by a full 'aoforce' computation and an 'aoforce analysis only' run, but I obtained 9 normal modes, one of which corresponding to an imaginary frequency; moreover all the frequencies previously computed were dramatically modified by the new computation.
I add, for the sake of completeness, that the dscf/grad steps with the dummy hydrogen atom included and the key 'internal' switched 'on' in $optimize went fine. In fact the cartesian gradient for the coordinates of dummy H were zero as expected, while for the 'real' coordinates of NH3 I obtained the same values of the earlier cycles of optimization, i.e. less than 10^-5.
Suggestions or tips?
amedeo