CUTOFF convergence problem

[katarína stančiaková]

Dear CP2K users,
I am running my calculations and I have CUTOFF convergence problem - in fact I can not see any convergence in the total energy. I know that the relative differences in energies are important (not the absolute values of energies), therefore I am comparing the energy differences between two structures (I am working with zeolites, and my two structures differs in the position of hydrogen). Even with very high cutoff values I can not observe any convergence of the total energies as well as energy differences.
Do you have any idea where is the problem? I will be thankful for any help!

Here are my data:

CUTOFF [Ry]            Total Energy [a.u.]
                                structure 1                structure 2
600                           -3467.0403965424    -3467.0619253888
800                           -3467.0521866655    -3467.0622253178
1000                         -3467.0838714739    -3467.1096147317
1200                         -3467.0833499728    -3467.1103018678
1400                         -3467.0975445683    -3467.1102723936


ENERGY DIFFERENCES


CUTOFF [Ry]    E_str1 - E_str2 [kJ/mol]   

600                55.9750007029
800                26.1004960059
1000              66.932470253
1200              70.0749269477
1400              33.0923456248
   
Thank you
Katarina

[Matt W]

Hi Katrina,

At a given geometry (the one in your input) the total energy of the system converges quite OK with cutoff after about 800 Ry, according to me.

Are you comparing the energy differences after performing geometry optimization? Is it possible that a some of your runs are optimizing to subtly different structures. Even a tiny difference in the shape of the potential energy surface can lead you into a different minima, are you are starting from quite high energy structures it seems. Your structure 2 seems well behaved, but structure 1 gives a weird energy at the highest cutoff you used.

Matt

BTW, the basis sets you are using DZV-GTH are probably not really production quality.

[katarína stančiaková]

Dear Matt,
thank you very much your reply. I will try to use different basis sets to see if the results will change.

You are right, I am comparing the energy differences after geometry optimizations and it seems that with different cutoff values I am finding different minima. What is bothering me is that even CUTOFF 1200 Ry is obviously not enough as I found another minima for structure 1 with cutoff 1400 Ry. The using of such high cutoff is not very convinient for zeolite structures as far as I know. So I am looking for some "bug" in my setup.

Katarina

Dňa utorok, 10. novembra 2015 10:13:10 UTC+1 Matt W napísal(-a):

[Matt W]

Dear Katrina,

You are right, I am comparing the energy differences after geometry optimizations and it seems that with different cutoff values I am finding different minima. What is bothering me is that even CUTOFF 1200 Ry is obviously not enough as I found another minima for structure 1 with cutoff 1400 Ry. The using of such high cutoff is not very convinient for zeolite structures as far as I know. So I am looking for some "bug" in my setup.

energy minimisation can be essentially chaotic, tiny differences in the potential energy surface / initial conditions etc can rapidly lead you to minimise to different minima, especially if you start from very high energy structures. I think the 'bug' in your set up is that you start from a very high energy state and then minimize. If you did this for many different starting positions you should locate all the possible minima eventually. But better to start from close to a local minima and not ask so much from the optimiser.

It is not really a question of one being right and the other wrong in this case if you find different minima. Do both minima exist for both 1200 and 1400 Ry cutoff with consistent energy differences?

HTH,

Matt

[Guanna Li]

I also met the same problem for my zeolite system. I used an optimized structure as input and only calculated the single point energy with different CUTOFF values.
The energy differences were bigger than acceptable. 

[Guanna Li]

I reported my test results here:

https://groups.google.com/forum/#!topic/cp2k/UYaBsH6A7e0

All single point calculations used the same input structure. But you can see the enerny didn't converge even the CUTOFF value is big. 

[Matt W]

Hi,

your CUTOFF is still not converged. For PBE and a system with oxygen atoms to be fully converged in an individual energy you need to be at > 600 Ry, probably somewhat more.

300	-1752.615532
350	-1752.601872
400	-1752.596534
450	-1752.596534
500	-1752.595309
550	-1752.594747
600	-1752.594059
700	-1752.594543
800	-1752.594127
900	-1752.594142
1000	-1752.594152
1250	-1752.594154

Energies vs cutoff for your zeolite system. For a lot of uses you don't need to push that hard, but if you are comparing small relative energies, it might be needed. If the cost of the cutoff is too high, you can also try using the GAPW method or using pseudo potentials with non linear core correction. There are several threads mentioning these on the forum.

Matt