Author Topic: Convergence problems in the case of constrained optimization  (Read 8537 times)

evgeniy

  • Sr. Member
  • ****
  • Posts: 110
  • Karma: +0/-0
Dear All,

I encountered quite severe divergence in geometry optimization when trying
to perform constrained geometry optimization. The constraints I have
are quite complex: I have one frozen internal coordinate and many
frozen cartesians. Optimization is perfromed of course in internal/redundent
coordinates. Only one, frozen, internal coordinate is defined.
Any idea/recipe on improving the convergence will be greatly apreciated!

Best regards,
Evgeniy

christof.haettig

  • Global Moderator
  • Sr. Member
  • *****
  • Posts: 291
  • Karma: +0/-0
    • Hattig's Group at the RUB
Re: Convergence problems in the case of constrained optimization
« Reply #1 on: November 18, 2010, 06:24:47 PM »
In TURBOMOLE frozen cartesian coordinates are only recognized if you disable the use of internal coordinates.
It's presently not possible to use a mix of cartesian and non-cartesian constraints.

Christof

uwe

  • Global Moderator
  • Hero Member
  • *****
  • Posts: 558
  • Karma: +0/-0
Re: Convergence problems in the case of constrained optimization
« Reply #2 on: November 19, 2010, 11:22:46 AM »
Hi,

fixed Cartesian coordinates are always taken care of in the module statpt (the default in Turbomole 6.x when running jobex), even if one uses internal coordinates.

So in principle fixing both should work - but the optimization algorithm does get problems if many internals and/or Cartesians are fixed.

You could try to reduce the step size with $coordinateupdate, use a different update algorithm or alter the guess Hessian (which really is an expert's option). I guess there is no general solution.

Regards,

Uwe