Re: CCL:Second negative frequency

Dear Adel,
 > I think your problem is with the numerical derivatives. I think using the
 > analytical would solve you problem. You do not you try small basis set
 nice try, but in the (yet) absence of this capability in the program
 I use, i.e. Turbomole I am left with that problem.
 > with analytical or if you can the smallest of your molecules with
 > analytical. There are so many paprameters also to vary with the numerical.
 > Why do not try to vary these (I really do not think it will help but you
 > may try). Why do not you try tighter geometry optimization. I think that
 well, sure I do all that kind stuff, tighten SCF-convergence etc., change
 the stepsize of the finite differences but you see, you do not know
 the answer in advance after 6n SCF steps and gradients. In my case I may
 have the answer just after 2-10 days etc. depending on the size of the system
  (50-130 atoms 500-1500 bfs) and numbers of CPU I use (though it
 runs perfectly parallel, calculations are done on 8-20 CPUs).  Checking
 for / fighting the numerical noise is not
 just gratifying considering error bars of at least +-20 cm-1 - so should
 I stop and relax with a second small imaginary frequency of -20 to -30
 cm-1?
 Peter
 P.S. Also I forget to point out that a number of people seem to be just
      happy and get away with the updated approximate hessian they get
      during the optimization process - is that any better? I do not
      think so.