R help - Jan 2011 - Getting stable solution while applying 3 parameter model (tpm) on response data

I have started exploring potential of R in applying IRT to a dataset. I have
a data of about 20k students who took a Maths test, a diagnostic in nature.

I find that I don't get stable solution while using tpm() even after passing
an argument "start.val = RANDOM". What should be done in this case to
achieve stable solution?

The item parameters estimated would not be sensible when the stable solution
is not arrived at. However I find that discrimination parameter of one of
the item estimated is negative in that case.

(I used tpm() from ltm package.)

	[[alternative HTML version deleted]]