SCF convergence

[Sihem AZIZI]

Dear all,

I ran a geometry optimization calculation for a 3-layer Ru slab. The calculation ended with the following warning:

 *** WARNING in qs_scf.F:542 :: SCF run NOT converged ***

I have attached the input and the output files.

Also, I would like to know how to add the dispersion and contribution of the VDW forces.

Thanks a lot in advance for your help

[Johann Pototschnig]

Your SCF shows convergence, but doesn't reach the threshold. 

It might be sufficient to increase the number of iterations: MAX_SCF 200

Otherwise you could try different convergence accelerators. 

https://manual.cp2k.org/trunk/CP2K_INPUT/FORCE_EVAL/DFT/SCF.html#CP2K_INPUT.FORCE_EVAL.DFT.SCF

For a geometry optimization you should increase the SCF threshold: EPS_SCF 1.0E-6

otherwise the gradients will be unreliable. 

You have a lot of Ru atoms, but only add 10 orbitals in the Fermi smearing, which might also cause problems in the SCF convergence. 

VdW correction are described in:

https://manual.cp2k.org/trunk/CP2K_INPUT/FORCE_EVAL/DFT/XC/VDW_POTENTIAL.html

[Sihem AZIZI]

Dear Johann,

Thanks a lot for your response.

I started a new job with the file attached. I paid attention to your comments, but I still get the same warning.

*** WARNING in qs-scf.F:542 :: SCF run NOT converged ***

I reduced the number of Ru atoms to reduce the calculation time, and now I don't know whether to stop the calculation or let it continue. Can you tell me what's wrong? Thanks.  ,

Sihem

[Frederick Stein]

Dear Sihem,

I would first try to reduce the mixing by adding

&MIXING

  # Lower values improve convergence but increase the number of SCF iterations

  # Difficult systems may require even lower values

  ALPHA 0.01

&END

to your SCF section. You can also try different mixing schemes (see here: https://manual.cp2k.org/trunk/CP2K_INPUT/FORCE_EVAL/DFT/SCF/MIXING.html ). Unfortunately, there is no standard recipe with difficult to converge systems.

HTH,

Frederick

[Johann Pototschnig]

The accuracy is probably not sufficient to get convergence. 

You need to increase the CUTOFF(900) and test which one works. 

It might also be necessary to increase EPS_DEFAULT(1.00E-016). 

https://manual.cp2k.org/trunk/CP2K_INPUT/FORCE_EVAL/DFT/QS.html#CP2K_INPUT.FORCE_EVAL.DFT.QS.EPS_DEFAULT

[Sihem AZIZI]

Thanks a lot