CCL: IRC in Gaussian 98 stops at false minimum

From: Petr Toman <ccl-.at.-linus.imcm.cas.cz>
Subject: CCL: IRC in Gaussian 98 stops at false minimum
Date: Tue, 2 Mar 2004 13:18:40 +0100 (CET)

Dear Colleagues,
 I would like to calculate the reaction path from a transition structure of
 a merocyanine molecule (37 atoms).
 Transition point was verified by a Freq calculation, imaginary frequency
 is 43 cm-1.
 The problem is that IRC calculation stops after very few points reporting
 "Minimum found on this side of the potential".
 However, usual geometry optimization (OPT) starting at this point finds
 the actual minimum far away.
 My question is what is the criterion for a minimum in Gaussian IRC
 calculation?
 Thank you for your help.
 Best regards,
 Petr Toman