Author Topic: Controling linear dependence thresholding  (Read 548 times)

jfurness

  • Newbie
  • *
  • Posts: 2
  • Karma: +0/-0
Controling linear dependence thresholding
« on: July 09, 2019, 11:58:54 pm »
I am running single point calculations using DFT from dscf and CCSD from ccsdf12 of hydrogen chains with the series of basis sets {cc-pVDZ, cc-pVTZ, cc-pVQZ, cc-pV5Z} for basis set extrapolations.

When the chain is compressed to short internuclear separations (~1.0 Bohr) the calculations with larger basis sets explode after a few iterations.

I was expecting this behaviour as the basis sets become linearly dependent. Normally such linear dependence would be projected out of the basis set for overlap matrix eigenvalues < some threshold.

Is it possible to set this threshold in the control file?

I have found the option lindep in $ricc2 for the coupled cluster calculations, though I am not certain this is doing what I want, and it definitely doesn't apply for the DFT calculations.

Many thanks,
James

uwe

  • Global Moderator
  • Sr. Member
  • *****
  • Posts: 457
  • Karma: +0/-0
Re: Controling linear dependence thresholding
« Reply #1 on: July 10, 2019, 08:45:11 pm »
Hi,

Quote
Normally such linear dependence would be projected out of the basis set for overlap matrix eigenvalues < some threshold.

I wonder what 'Normally' does mean  :). Probably other QC programs do that by default, but it was never a standard in Turbomole. Automatically changing the number of basis functions can be helpful for such interesting calculations you plan to run. But for applications it could also lead to discontinuities e.g. on the potential energy surface.

riper has the option to remove linearly dependent basis functions using the option lsdiag. Search the documentation for this keyword to see where it has to be added. It was initially implemented for periodic systems, but it might also work for molecules (but I have never tried it so far).

Regards,

Uwe