Summary:Negative eigenvalues in IRC Calcn.

 Dear all,
 Here is a summary of the responses I received regarding my query posted
 a week ago.  Thank you to Patrik Johansson, Stefan Fau, Stephan Graf and
 Wibke Sudholt for your interest.  I was in fact unable to reproduce the
 error so maybe it was a one off error, albeit a somewhat peculiar one.
 The problem seems to occur at a change in symmetry of the TS/product.
 =========> Original Query <==========
 I have been doing some ab initio calculations recently on the ring
 opening of cyclobutene using GAUSSIAN94.  I have completed an
 optimisation of the transition state corresponding to the ring opening
 and formation of the TRANS product (trans-butadiene).  To check that the
 transition state corresponds to the trans conformation an IRC
 calculation was performed to follow the path in the direction TOWARDS
 (reverse direction) the trans product. The command line was:
 >>  # RHF/6-31G(d) IRC(CalcFC,Reverse,Stepsize=40,maxpoints=24)
 The calculation stopped on the 12th point along the path after 5
 optimisation steps having failed to converge with this unusual message:
 >>   Optimization stopped.
 >>      -- Wrong number of negative eigenvalues: Desired=  0 Actual=***
 This is very peculiar - I have never seen this before.  It is highly
 unlikely that the calculation has 'stumbled upon' another saddle point.
 Has anyone come across this before?  Any reasons/suggestions/ideas would
 be welcomed.
 RESPONSES
 ==========> 1 <==========
 Hi Dave
 Your problem sound like one I encountered about a year ago, maybe you've
 stumbled into a bifurcation point..
 If this is the case it would mean that your IRC calc cannot proceed
 since two
 alternative ways - truely equal from the calc point of view - are
 possible,
 this would for example be possible if the point where your calc crashes
 has a
 C2 or Cs symmetry ( or close to ). The way to check and to let the calc.
 continue is :
 1. to perform a freq calc at the IRC point immediately before your calc
 crashes
 ( perhaps with a smaller stepsize.)
 2. Check the resulting imaginary freqs and see which mode that
 corresponds to
 the path you would like to continue along.
 3. Move your geometry away from the bifurc point by using the mode
 above.
 4. Perform a new IRC calc from this new geometry.
 best wishes
 Patrik
 ==========> 2 <==========
 Hi Dave,
 something went seriously wrong with your IRC. The three stars indicate
 >999 negative eigenvalues! This is a problem of the actual run and might
 just vanish on repeating the calculation.
 By the way, it is advisable to start with the read in structure and
 hessian matrix of the confirmed TS using IRC(...,readfc,...) and
 geom=check since this avoids problems due to low numerical accuracy of
 the input TS structure.
 If you continue having problems, try to reduce the stepsize and increase
 the accuracy (using the tight option).
 Greetings, Stefan
 ==========> 3 <==========
 Hi Dave,
 I had a similar problem during the optimization of saddle points
 structures. Sometimes G94 stopped with the same message. It seems
 that there is a bug in the G94 routine which counts the number
 of negative eigenvalues. Because after testing the structure
 with a frequency job, I got the correct number of negative
 eigenvalues.
 With the OPT keyword is the NoEigenTest option
 available, which omits the counting of the negative eigenvalues.
 Unfortunately this option is not listed for the IRC keyword (at least on
 www.gaussian.com).
 But ther is maybe an IOp which does the same.
 Hope this helps,
 Stephan
 __________________________________________________________
 Dave Pender                                    Tel. 01483 259591
 Department of Chemistry                  Pager. 0336 785185
 University of Surrey
 Guildford
 Surrey
 GU2 5XH.
 ** "The World is a Stage, but the Play is Badly Cast" -- Oscar Wilde
 **