Struggling to get SCF Convergence in Cell_Opt run

[ASSIDUO Network]

Hi there everyone, hope you've had a great week.

I've been trying to run a cell optimization of bulk Au, and I am using the attached input file, but I'm not getting an inner loop SCF convergence. I've made many small changes, such as including/excluding OUTER_SCF,  changing the SCF convergence criterion, changing the number of cell optimization steps, changing the number of KPoints and changing the mixing method. Nothing has worked. I haven't tried a combination of the above though.

Do you perhaps have any suggestions to me on how to get convergence? Furthermore, I would also appreciate some tips to speed up my simulations (settings/flags) wise.

Thanks in advance,
Lenard 

[Marcella Iannuzzi]

Hi Lenard, 

I got it converging in 10 iterations.

The outer SCF with diagonalisation is useless, since there is no preconditioner.

With metals you need to use smearing. 

Here are some settings I used:

      ADDED_MOS 200

      &DIAGONALIZATION T

        ALGORITHM STANDARD

      &END DIAGONALIZATION

      &MIXING T

        METHOD BROYDEN_MIXING

        ALPHA 0.01

        BETA 0.5

        NBUFFER 8

      &END MIXING

      &SMEAR

        METHOD FERMI_DIRAC

        ELECTRONIC_TEMPERATURE 500

      &END SMEAR

    &XC

      &XC_FUNCTIONAL

          &LIBXC

            FUNCTIONAL XC_GGA_X_RPW86

          &END LIBXC

          &LIBXC

            FUNCTIONAL  XC_GGA_C_PBE

          &END LIBXC

      &END XC_FUNCTIONAL

      &VDW_POTENTIAL

       POTENTIAL_TYPE NON_LOCAL

        &NON_LOCAL

          CUTOFF 300

          TYPE RVV10

##          VERBOSE_OUTPUT

          KERNEL_FILE_NAME ${data}/rVV10_kernel_table.dat

        &END NON_LOCAL

      &END VDW_POTENTIAL

    &END XC

The results:

 Step     Update method      Time    Convergence         Total energy    Change

  ------------------------------------------------------------------------------

     1 NoMix/Diag. 0.10E-01   20.0     0.51067755      -133.2796876462 -1.33E+02

     2 Broy./Diag. 0.10E-01   17.6     0.00064724      -136.0334243526 -2.75E+00

     3 Broy./Diag. 0.10E-01   17.5     0.03257808      -134.7316158415  1.30E+00

     4 Broy./Diag. 0.10E-01   17.7     0.00019866      -133.0666478987  1.66E+00

     5 Broy./Diag. 0.10E-01   17.6     0.00228816      -133.1462861174 -7.96E-02

     6 Broy./Diag. 0.10E-01   17.6     0.00032933      -133.1654553845 -1.92E-02

     7 Broy./Diag. 0.10E-01   17.6     0.00000406      -133.1816175525 -1.62E-02

     8 Broy./Diag. 0.10E-01   17.7     0.00009047      -133.1825852315 -9.68E-04

     9 Broy./Diag. 0.10E-01   17.6     0.00000504      -133.1830490186 -4.64E-04

    10 Broy./Diag. 0.10E-01   17.6     0.00000031      -133.1828944498  1.55E-04

  *** SCF run converged in    10 steps ***

  Electronic density on regular grids:        -44.0000000000        0.0000000000

  Core density on regular grids:               43.9999999999       -0.0000000001

  Total charge density on r-space grids:       -0.0000000001

  Total charge density g-space grids:          -0.0000000001

  Overlap energy of the core charge distribution:               0.00000001219968

  Self energy of the core charge distribution:               -231.41335460772382

  Core Hamiltonian energy:                                     74.19344628639691

  Hartree energy:                                              45.27318657026385

  Exchange-correlation energy:                                -21.30000506759340

  Dispersion energy:                                            0.06400174824714

  Electronic entropic energy:                                  -0.00016939092888

  Fermi energy:                                                 0.34714684334798

  Total energy:                                              -133.18289444982969

Regards

Marcella

[ASSIDUO Network]

I should have added this earlier, the simulation must be performed at 0 K (my supervisor's request), it's a static run. Therefore, I cannot set the temperature to 500 K. Also, why the need for ADDED_MOS=200? In another thread, I was told that ADDED_MOS=100 was too much.

Also, why the use of  FUNCTIONAL  XC_GGA_C_PBE and  FUNCTIONAL XC_GGA_X_RPW86? I'm new to CP2K so just want to understand everything better.

[Marcella Iannuzzi]

Hi

The smearing with finite electronic temperature is needed in order to allow convergence of the metallic electronic structure, even if the structure is kept a 0K.  The number of added MOS depends on the density of states around Fermi.  Too many states do not hurt, in terms of convergence. I just chose a safe number, probably less are also OK, as a lower smearing temperature would also work. 

As a functional I took one that I often use. Anyway, I think that adding VDW is a good idea. 

All these aspects are more general on electronic structure theory and not specific of CP2K. Please read in the literature for better understanding.

Regards

Marcella

[ASSIDUO Network]

I'm going to follow your advice, but if I remember correctly, last time I tried smearing with 0 K, the stimulation ended with an error. If it does it again, I will ask for some feedback again.

[Marcella Iannuzzi]

No smearing with 0K does not help.

Smearing needs a finite temperature, can be lower than 500, for instance 300K, not 0K

[Lenard Carroll]

Then I cannot use smearing, I guess I'll have to turn it off or will have to talk to my supervisor about using a non-zero temperature.

--
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/a48a92ce-995f-489b-a822-67e2e83e95e2n%40googlegroups.com.

[Lenard Carroll]

Fantastic, thanks for the info. I've only recently started my journey into DFT, been doing MD simulations for years now.

On Fri, Jan 22, 2021 at 10:41 AM Travis <polla...@gmail.com> wrote:

Hi,

For small band gap materials, you need to apply a broadening or smearing function that smooths out the energy levels near the Fermi level so that you don't oscillate between filling a low lying unoccupied level on one SCF iteration, emptying it on the next, and so on and so on. There are multiple methods for this such (refer to VASP manual for example, https://www.vasp.at/wiki/index.php/ISMEAR), CP2K implements Fermi smearing. You apply an electronic temperature that describes the shape of the broadening function. It isn't a physical temperature.

-T

To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/c8e010ca-db0d-459a-8051-678bfb7f4ae0n%40googlegroups.com.