CP2K exercise for converging the cutoff

[Dragoon]

Hi all,

As I'm still struggling with converging the plane wave cutoff, I tried this CP2K exercise for converging the cutoff for a water system (the link is https://www.cp2k.org/events:2018_summer_school:converging_cutoff ). This system has 32 H2O, using PBE functional (no XC smoothing) and MOLOPT-SR basis. REL_CUTOFF is 60 Ry.

I ran this exercise, and extended the cutoff to 2000Ry, yet I'm not sure at which cutoff either the energy or force is converged. I attached the result below. 

I do notice that the most coarse grid (#4) has the most grid levels starting from 800 Ry. I guess that means the cutoff is probably already too high, and that might be related to a slow convergence. Yet I'm not sure how to improve that, nor further explain this behavior for this simple system.

I really appreciate any explanations and helps.

Zhuoran

[Thomas Kühne]

Dear Zhuoran, 

you can find more detailed informations at: https://www.cp2k.org/howto:converging_cutoff

A very rough a priori estimate for the cutoff can be found at: https://www.cp2k.org/faq:cutoff

Best, 

Thomas Kühne

Am 19.12.2022 um 06:26 schrieb Dragoon <longdr...@gmail.com>:

Hi all,

As I'm still struggling with converging the plane wave cutoff, I tried this CP2K exercise for converging the cutoff for a water system (the link is https://www.cp2k.org/events:2018_summer_school:converging_cutoff ). This system has 32 H2O, using PBE functional (no XC smoothing) and MOLOPT-SR basis. REL_CUTOFF is 60 Ry.

I ran this exercise, and extended the cutoff to 2000Ry, yet I'm not sure at which cutoff either the energy or force is converged. I attached the result below. <Screenshot from 2022-12-19 00-01-47.png>

I do notice that the most coarse grid (#4) has the most grid levels starting from 800 Ry. I guess that means the cutoff is probably already too high, and that might be related to a slow convergence. Yet I'm not sure how to improve that, nor further explain this behavior for this simple system.

I really appreciate any explanations and helps.

Zhuoran

--
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/aa6d141f-3e25-4365-8cbd-f482217bedf0n%40googlegroups.com.
<Screenshot from 2022-12-19 00-01-47.png>

==============================

Thomas D. Kühne

Dynamics of Condensed Matter

Chair of Theoretical Chemistry

University of Paderborn

Warburger Str. 100

D-33098 Paderborn

Germany

thomas...@upb.de

+49/(0)5251/60-5726

[DMT]

In many publications a CUTOFF of 400 is normal and acceptable. I haven't seen anyone to have to go beyond 600, so far. Also, a very good value for NGRIDS is 5.

Dobromir

[Dragoon]

Hi Dobromir and Thomas,

Thank you all for the replies. 

As I dug through the threads in this group. I realized many problems of the CP2k Howto and exercise about PW cutoff. Please correct me for my misunderstandings. 

1. Based on one previous discussion, SCF needs to be converged with a large enough REL_CUTOFF, instead of a 1 step calculation like in the Howto. 

2. EPS_PGF_ORB 1.0E-07 or lower is usually needed for converging SCF.

3. 1e-3 ev / 3e-5 Hartree energy convergence is good enough for most system? For this convergence the exercise system need to go to 800 Ry cutoff.

4. GAPW sometimes can achieve a much smaller cutoff than GPW method. The default parameters in CP2k are good enough for GAPW.

Thank you again,

Zhuoran