Mixing Method Recommendation

[Chris Campbell]

Hello CP2k Group,

Should first state that I am new the DFT field.

Had a query on which mixing procedure to selected to be applied to the density matrix.  I have selected the BROYDEN_MIXING_NEW method in my system but without much basis for this other than I observed the first version of the method being selected in the cp2k tutorial.

Could someone please advise me on the suitability of this mixing method for my system or suggest possible reading material on this topic.

The system I am modelling is the geometry optimization of the MOF HKUST-1, my cell is 156 atoms made up of copper, carbon, oxygen and hydrogen. I will be using the m06-l functional.

Currently my single energy point calculations are converging. 

Any information would be greatly appreciated,

Chris

Also, apologies if this amount of information is insufficient for my query and will happily add additional if needed.

[S Ling]

Hi

HKUST-1 has a reasonable gap, and you may use &OT instead of direct diagonalization to converge SCF. Please try to replace the whole &SCF subsection of your input with the following:

      &SCF
         EPS_SCF     1.000E-06
         SCF_GUESS  RESTART
         MAX_SCF 20
         &OT T
            MINIMIZER  DIIS
            PRECONDITIONER  FULL_ALL
         &END OT
         &OUTER_SCF T
            EPS_SCF     1.000E-06
            MAX_SCF  10
         &END OUTER_SCF
      &END SCF

Please also note HKUST-1 is a magnetic system. If you don't specify MULTIPLICITY and/or &BS subsection in your input, your calculation will most likely converge to a closed-shell singlet state, which is not the ground state of HKUST-1. The triplet/ferromagnetic state, which is slightly higher in energy than the antiferromagnetic ground state, can be obtained by setting the MULTIPLICITY keyword properly. To get the antiferromagnetic ground state, you will need to set the &BS subsection in &KIND section for the two types of copper atoms.

SL

--
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 post to this group, send email to cp...@googlegroups.com.
Visit this group at http://groups.google.com/group/cp2k.
For more options, visit https://groups.google.com/d/optout.

[Chris Campbell]

Hello S Ling,

Thanks for the quick reply.

I have inserted this new scf, I had gotten rid of direct diagonalization already but my convergece still takes around 117 steps.

Regarding the &BS section I was wondering why two columns were required (please see https://github.com/misteliy/cp2k/blob/master/tests/QS/regtest-bs/cu2cl6_m3_clp6cud9.inp)?

I would expect just a number per keyword, i.e., NEL +1 and NEL -1 for Cu1 and Cu2, respectively. Then L should be 0 in both cases and N should be 4 (unpaired 4s electron in [Ar] 3d10 4s1 configuration).

Thanks for the help and previous replies to other posts which have been invaluable,

Chris

[S Ling]

Hi

Please refer to the following webpage for explanation on L & N:

http://manual.cp2k.org/trunk/CP2K_INPUT/FORCE_EVAL/SUBSYS/KIND/BS/ALPHA.html

The first column in the example corresponds to 4s orbital of Cu, and the second column corresponds to the 3d orbital of Cu.

SL

[Chris Campbell]

Hi S Ling,

Thanks for the explanation that helped.

I have completed my input and was just wondering if I could ask you to have a quick glance at it for any last advice? I understand if this is asking too much. My only changes to your scf were to increase MAX_SCF as convergence did appear to be occurring but taking longer than 20 and I lowered the tolerance. 

Kindest Regards,

Chris

[S Ling]

Hi

Several suggestions:

1. set MAX_SCF to 20 instead of 200 (my setting given in the previous post is more efficient)

2. set FUNCTIONAL to FUNCTIONAL XC_MGGA_X_M06_L XC_MGGA_C_M06_L (you don't have correlation functional in your input)

3. MULTIPLICITY 3? If you would like to consider a triplet/ferromagnetic state, then the total spin multiplicity depends on the total number of unpaired electrons; if you would like to consider a singlet/antiferromagnetic state, then set it to 1

4. As I said in my previous post, you don't need to set the &BS section for a triplet/ferromagnetic state. If you would like to consider a singlet/antiferromagnetic state, then for Cu2, you will need to swap the contents of &ALPHA and &BETA sections. With the current setting, you don't break the symmetry of Cu1 and Cu2 atoms

5. for H, use DZVP basis set instead of SZV

6. please check the convergence of your CUTOFF with respect to your target property, 280 Ry might be too small to give converged results on energetics and geometries

SL

--

[Chris Campbell]

Will make all the changes suggested above.

Thanks you again for all the help and best wishes,

Chris