About cutoff benchmarking in Cerium(III) chloride

[Aditya Goyal]

Dear CP2K users,

I'm trying to simulate a molten cerium chloride system with 40CeCl3 molecules. Before performing any extensive calculation, I'm trying to benchmark the cutoff energy used in the DFT section by performing single-point energy calculations. I'm using DZVP basis sets for Ce & Cl with PBE functional by taking 800 Ry as a benchmark for comparison. 

For a Relative cutoff of 60 Ry 

Cutoff          Energy(Ha)               Delta E (kJ/mol) wrt. 800 Ry

350 Ry -3357.32389476218      3.61119311934135
550 Ry -3357.3063307436      49.7255239006927
600 Ry -3357.27607335904     129.166287063426
650 Ry -3357.28237568966     112.619518020088
700 Ry -3357.31280473755      32.7280527853891
750 Ry -3357.31982535559      14.2954201205428
800 Ry -3357.32527019281       0.00000000000000

Similarly, For a Relative cutoff of 100 Ry

Cutoff          Energy(Ha)               Delta E (kJ/mol) wrt. 800 Ry

350 Ry -3357.31559413633    -49.9141589061937
550 Ry -3357.3107413093    -37.1730615379035
600 Ry -3357.29750596273      -2.42365911922866
650 Ry -3357.33819509308    -109.252970852155
700 Ry -3357.32224091494     -67.3652761456795
750 Ry -3357.30040482541     -10.0346230856023
800 Ry -3357.29658283985      0.00000000000000

So, I'm not finding a convergence of energy or any particular trend at any cutoff (for both relative cutoff cases). Can anyone suggest some parameter changes that may be used to achieve convergence? Any help would be appreciated. 

Regards

Aditya

[Lucas Lodeiro]

Hello Aditya,

It is not clear what basis set you are using, since there are two DZVP-MOLOPT-SR-GTH for Ce, one with 12 electrons (-q12) and another with 30 electrons (-q30).

In my experience the best approach to know a cutoff estimate is from the HIGHER_COEFFICIENT in the whole basis set. Taking the result from HIGHER_COEFFICIENT*REL_CUTOFF you can get an approximate CUTOFF for convergence.

From BASIS_MOLOPT_UCL file, the DZVP-MOLOPT-SR-GTH-q12 basis set has a  HIGHER_COEFFICIENT of 4.348 and DZVP-MOLOPT-SR-GTH-q30  basis set has a  HIGHER_COEFFICIENT of 18.922, so with a REL_CUTOFF of 60Ry, the convergence must be reached at 260Ry and 1135Ry respectively.

Also, if you are using smoothers (XC_GRID section), the convergence test behavior will not be as expected, with a monotonous convergence... in my experience always show more oscillations...

A last thing, the EPS_DEFAULT and EPS_SCF values are important to reach a good CUTOFF convergence, values of 1E-12 and 1E-06, respectively, are usually good.

Regards - Lucas

--
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/459d444b-78fd-4461-87e2-2ae7e4761a6dn%40googlegroups.com.

[Aditya Goyal]

Hi Lucas, 

Thanks for the response. 
I'm using DZVP-MOLOPT-SR-GTH for Ce with 12 electrons (-q12). In my calculations, I tried cut-off energies of 200, 250, 300, ...... till 1000 Ry (at REL_CUTOFF= 60 Ry and 100 Ry). However, I mentioned only a few of them in the original query.

So, I didn't find energy convergence near 260 Ry and above (see original query). I'm not using any smoother (XC_GRID section) in my calculation. Would you recommend some other changes in the parameters?

Regards

Aditya