Author Topic: gradient accuracy - general question  (Read 6741 times)

asaf

  • Jr. Member
  • **
  • Posts: 23
  • Karma: +0/-0
gradient accuracy - general question
« on: December 13, 2010, 01:53:24 PM »
I guess that in order to increase the gradients accuracy one needs to set $scfconv to be large as possible (to my knowledge,In turbumole the largest value is 9):
1)Is there other things that can be done in order to increase the gradients accuracy?
2)If $scfconv=9, what is the accuracy in gradients?
3)Is larger value of  9 in $scfconv is allowed?
thanks

asaf

  • Jr. Member
  • **
  • Posts: 23
  • Karma: +0/-0
Re: gradient accuracy - general question
« Reply #1 on: December 13, 2010, 01:56:06 PM »
I interested in rdgrad calculation, and off course - i used "weight derivatives" option

christof.haettig

  • Global Moderator
  • Sr. Member
  • *****
  • Posts: 291
  • Karma: +0/-0
    • Hattig's Group at the RUB
Re: gradient accuracy - general question
« Reply #2 on: January 20, 2011, 02:38:08 PM »
What do you mean precisely with accuracy? The numerical accuracy for a given basis set, method/functional and given grid? In that case $scfconv and $weight derivatives are the option that decide about it. One can go beyond $scfconv 9, but how much depends on the basis set and the molecule. At some point you will reach the limit of the numerical precision of double precision algebra and at the point the calculation will never fulfill the requested convergence threshold.

More important for the overall accuracy of gradients and geometries are of cause things like the method/functional and the (auxiliary) basis set....

Christof