Re: CCL:IRC on G03

[Steve Williams <willsd/at/appstate.edu>]

 At 01:29 PM 12/9/2004 -0600, Philip Hultin wrote:
>  Alessandro:
> We did a fair bit of IRC a while ago (look for upcoming papers in J
>  Phys
>  Chem A) and we found that sometimes a calculation would begin to have
>  convergence failures as it moved further from the initial structure
>  (i.e.
>  the TS).  This was because the force constants became less reliable the
>  further one got along the IRC.
> In such cases, if we wanted to move further along the IRC, we needed
>  to
>  specify CALCALL to get fresh force constants at each point.  This is a
>  bit
>  time-consuming but not nearly as slow as a failed calculation.  Using
>  small step sizes in the IRC also helped a bit.
> My understanding of the documentation is that CALCFC only does the
>  force
>  constant calculation at the initial structure, NOT at the beginning of
>  each point along the IRC that is what CALCALL does in this case.
> After doing this a few times, we concluded that what Gaussian should
>  consider building into the IRC was an option to specify when new force
>  constants should be calculated.  It would be nice, for example, to be
>  able
>  to calculate force constants at every Nth point along the IRC it would
>  keep the calculation on track but take much less time than calculating
>  them at each new point.  Another useful variation would be to move N
>  points along the IRC using extrapolated force constants and only then
>  start doing calculated force constants.  After all, in most cases one
>  would expect force constants to be drastically different near the
>  starting
>  material and product states than they are near the TS, but that the
>  differences would initially be small.  Any thoughts about this
>  suggestion
>  would be welcome.
>  
I've long thought that a calcsometimes option would be useful for
 optimizations as well, particularly for transition state
 optimizations.  I've done this manually for some tricky transition
 state
 optimizations by limiting the number of steps, computing new force
 constants, doing several more steps, etc. Automatically computing the
 Hessian every 5 or 10 steps could be useful in several types of
 calculations.  Count me on supporting this as a future Gaussian option.
 (Mike, Doug, did you hear this?)
 Steve Williams
 Department of Chemistry
 Appalachian State University
 Boone, NC 28608
>  P. Hultin
>  Associate Professor of Chemistry
>  University of Manitoba