Convergence with CUTOFF

[coko312]

Dear all,

I am a cp2k-DFT beginner so the answer to this question may seem obvious to you, but it would help me a lot!

I try to converge the total energy of a single Na+ ion as a function of the energy CUTOFF, but the convergence is very slow and even using 900 Ry is not fully satisfactory for what I want to do. Even with no experience with this system, I would not expect the computation to last several hours on a single processor for only 1 atom. I have attached my input-output files. What should I modify to be able to use a more reasonable CUTOFF? 

NB : the computation is non-periodic because I would like, next, to add more atoms around to compute interaction energies between these little "clusters" and various types of molecules.

Best,

[Augustin Bussy]

Additionally, I would suggest that you tighten the EPS_SCF threshold (to 1.E-7 or 1.E-8). Even using a really high CUTOFF, you might not observe converged results if the SCF cycle itself is not fully converged. 

Best,

Augustin

On Monday, 27 January 2020 09:54:34 UTC+1, Travis wrote:

Hi,

Vacuum is expensive with plane wave methods. Your box only needs to be large enough that the density goes to zero at the boundaries (6-10 Angstroms) with the wavelet solver. You'll see a warning about non-zero density if you don't use a large enough box. The rule of thumb for the cutoff is to multiply the largest exponent in the basis by REL_CUTOFF. For your basis set and REL_CUTOFF, that's about 1500 Ry. This puts the tightest orbital on the finest grid.

-T

[coko312]

Thank you very much both of you for these answers! 

Indeed I can reduce the size of my box to 10 Angstroms ; for one Na+ ion and a CO2 molecule, I need to use 12 Angstroms. So I am confident now that nothing is crazy with my simulation, but I guess I need a little more computing power and parallel simulations.

For a single Na+ ion, it works also very well with lower EPS_SCF (1.E-8). However, with this EPS_SCF I have some difficulties in making the SCF cycle converge with one Na+ ion and a CO2 molecule. I suspect it is due to the initial position of my CO2 molecule. What would you use as a reasonable distance/orientation, and how many iterations of the geometry optimization procedure would you expect (50, 100, 500 ... )? 

[Augustin Bussy]

Hi,

you can run a first geometry optimization run with a lower EPS_SCF so that it converges faster to an approximate solution. Once you are there you can restart with a tighter threshold. As for the initial position of the molecules, I am not familiar with your system myself, I am afraid.

Best,

Augustin

--
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+uns...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/0c081a4b-40b1-42fd-a558-f7826dc99030%40googlegroups.com.

-- 
Augustin Bussy
PhD Student
--------------------------

University of Zurich
Augustin Bussy
Department of Chemistry
Research Group Hutter
Winterthurerstrasse 190
CH-8057 Zurich

Tel: +41 44 635 4490

www.chem.uzh.ch
augusti...@chem.uzh.ch
--------------------------