graphite in CP2K

[Michele Ceriotti]

Dear CP2K community,

     I have been trying for a few days to set up calculations of graphite as an exercise for a student but I am getting the weirdest results.  I am sure in the it will end up being a silly mistake in the input, but I can't get to see it so perhaps someone can help.

Despite using a fairly large FD smearing, the SCF cycle has a hard time converging, and when the maximum number of steps is reached

    50 Broy./Diag. 0.50E+00    1.2     0.00014351      -819.2909959720 -2.28E-09
  *** SCF run NOT converged ***

and the calculation carries on with what it has got, diagnostics are really strange.

For a start, eigenvalues show two weird very low-energy states
 MO EIGENVALUES AND MO OCCUPATION NUMBERS
# MO index          MO eigenvalue [a.u.]            MO occupation
         1                    -21.523076                 2.000000
         2                    -21.517160                 2.000000
         3                     -1.383548                 2.000000
         4                     -0.510418                 2.000000

forces on the atoms are insane
 ATOMIC FORCES in [a.u.]
 # Atom   Kind   Element          X              Y              Z
      1      1      C           4.68736611     4.63921800  -222.86999309
      2      1      C          -1.57720920    -4.48651652   124.33822840
      3      1      C          -5.01734639     3.10518104   176.16292919

and so is the stress tensor
 STRESS TENSOR [GPa]
            X               Y               Z
  X   -8233.84728056      8.05482610      0.90481573
  Y       8.05482610  -9552.76932222     -0.49445437
  Z       0.90481573     -0.49445437  -2526.59040628

I was thinking of an error in the structure or the cell parameters, but I checked it many times and everything seems in order. The same structure, with same functional and similar parameters in SIESTA converges like a stone, and gives no problem whatsoever.
 
Can you spot something obvious that I am missing? I'd really like to use CP2K for this exercise, but I can't seem to figure out what is going wrong.

Many thanks,
Michele

[Matthias Krack]

Hi Michele,

I would suggest to set EPS_DEFAULT in @QS section at least to 1.0E-12 or lower and for the ALPHA in &MIXING I would also use a smaller value like 0.2. Maybe this will help to converge your system properly.

Matthias

[Michele Ceriotti]

Hi Matthias,
   Thanks for the quick answer. I have already played with the obvious parameters, point is it is impossible to converge the scf properly.
Everything looks like what you get when the geometry is crazy, and indeed my student tried to enlarge the cell (by almost 10%!) and got  converged scf and more reasonable forces.
However this is inconsistent with the literature and the results from siesta.
We have been trying to fiddle with the scf parameters for days, but get consistently 100au forces on a geometry that should be very close to optimum.
Do you see something wrong with how we define the cell or the positions?
Best
Michele

--
You received this message because you are subscribed to a topic in the Google Groups "cp2k" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/cp2k/koy91UtxlQw/unsubscribe.
To unsubscribe from this group and all its topics, 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.

[Matt W]

Hi Michele,

please try using the MOLOPT basis sets provided with CP2K ($CP2K_root/tests/QS/BASIS_MOLOPT) that prehaps give a more suitable starting point for describing a "molecular" system like graphite than atomic optimization based ones.

Using a DZVP-MOLOPT-SR-GTH basis - I get SCF convergence in ~10 cycles and geometry converges in ~10 steps (with BFGS).

Matt

[Michele Ceriotti]

Dear Matt (and Marcella who basically replied the same in private),

   Thanks a million, using the molopt basis set does fix things. I had tried both the dzvp from basis_sets and gth_basis_set, and I was getting similar nonsense.

I am a bit scared seeing how much difference it makes changing the basis set. Any idea why graphite should be such a nasty beast? I had experimented with different basis sets for water and never saw such a dramatic effect.

All the best,
Michele

[Matthias Krack]

Dear Michele,

I would not consider graphite as a specifically nasty system. As Matt already wrote, the basis sets in GTH_BASIS_SETS resulted from atomic calculations using the actual GTH pseudopotential. Such an atomic basis set optimisation gives in the case of carbon only a set of exponents and contraction coefficients for a 2s and 2p function. The SZV (single-valence) basis sets in GTH_BASIS_SETS are the results of such optimisations. You may try the C SZV basis set and you will see that is behaves at least reasonably for your system (you may also add a prmitive d polarisation function -> SZVP)  However, experience has shown that such a minimal basis set is by far not accurate enough in most cases as it does not provide sufficient flexibility. Thus it is not suited for production runs. The double-zeta (DZV) GTH basis sets are derived from the corresponding SZV basis sets just by using the smallest exponent as a primitive function for the second valence function. This exponent is often rather small which results in a quite diffuse valence function without any nodal structure. These DZV basis sets were tested in molecular calculations in which they worked. You see, however, that the procedure is not based on any strict optimisation method. This deficiency may become apparent for condensed phase systems like your system. For such systems I would always recommend the use of the MOLOPT SR basis sets as already suggested which provide results closer to PW calculations. The generation procedure of the MOLOPT basis sets performs an optimisation of the contraction coefficients of all included valence functions. In this respect the MOLOPT SR basis sets can be considered as a kind of "second generation" CP2K basis sets.
Nevertheless, I would like to note that there are many cases in which the GTH_BASIS_SETS work fine and their use may become beneficial as they are computionally less demanding. I know that these basis set issues are quite annoying, especially for people coming from the PW community.

Best regards,

Matthias

[Michele Ceriotti]

Thanks for the clarification, Matthias.

   I must admit that as someone who started off with PWs it is quite surprising to see that changing basis set within the same "class of complexity" can mess up things to the point of not being able to converge the SCF. In the future I'll pay more attention to testing the basis set when using CP2K, and take due note that there is much more than the number of polarization functions to define the quality of the basis!

Best wishes,
Michele

[DMITRII Drugov]

Good day dear CP2K users, do you think its right to use  90 90 90 orthorhombic symmetry for Graphite cell?

Should it be hexagonal 90 90 120 alpha betta gamma?

Regrads,

Dmitrii 

[DMITRII Drugov]

Thank you for your reply! I am still running optimisation, I will share update once its done.

Regards,

Dmitrii

On Thursday, December 3, 2020 at 1:10:48 PM UTC+11 Travis wrote:

Hi,

Graphite unit cell is hexagonal (https://materialsproject.org/materials/mp-48/). You can orthogonalize any cell though. Do it by hand or use a program like atomsk (https://atomsk.univ-lille.fr/),

atomsk foo.cif -orthogonal-cell bar.cif

-T

[DMITRII Drugov]

Hi guys, could you suggest the right cell setting for graphite simulation. Should I use extra space in xy dimension to get right optimised geometry? I build slab from Avogadro with 384 atoms  and xyz 17.04000000  14.75707288  13.20000000 A. I froze two bottom layer and allowed remaining top two layer to equilibrate. 

I am using following parameters:

&GLOBAL

  PROJECT Graphite_geometry_optimization_moreMOs 

  RUN_TYPE GEO_OPT

  PRINT_LEVEL MEDIUM

&END GLOBAL

&FORCE_EVAL

  METHOD QS

  &DFT

    BASIS_SET_FILE_NAME BASIS_MOLOPT

    POTENTIAL_FILE_NAME GTH_POTENTIALS

    CHARGE 0

    MULTIPLICITY 1

    &MGRID

      CUTOFF 800

      NGRIDS 5

      REL_CUTOFF 70

    &END MGRID

    &QS

      EPS_DEFAULT 1.0E-12

      WF_INTERPOLATION ASPC

    &END QS

   &SCF

      SCF_GUESS ATOMIC

      EPS_SCF 1.0E-7

      MAX_SCF 1000

      CHOLESKY INVERSE

      ADDED_MOS 100

      &SMEAR ON

        METHOD FERMI_DIRAC

        ELECTRONIC_TEMPERATURE [K] 300

      &END SMEAR

      &DIAGONALIZATION

        ALGORITHM STANDARD

      &END DIAGONALIZATION

      &MIXING

        METHOD BROYDEN_MIXING

        ALPHA 0.4         

        NBROYDEN 8

      &END MIXING

    &END SCF

    &XC

      &XC_FUNCTIONAL

&PBE

&END PBE

      &END XC_FUNCTIONAL

      &vdW_POTENTIAL

    DISPERSION_FUNCTIONAL PAIR_POTENTIAL

    &PAIR_POTENTIAL

        PARAMETER_FILE_NAME dftd3.dat

        TYPE DFTD3

        REFERENCE_FUNCTIONAL PBE

        R_CUTOFF 15.0

    &END PAIR_POTENTIAL

     &END vdW_POTENTIAL

    &END XC

    &POISSON

      PERIODIC xy

      POISSON_SOLVER ANALYTIC

    &END POISSON

  &END DFT

  &SUBSYS

    &CELL

      ABC 18.46 13.53 50.0

      ALPHA_BETA_GAMMA 90.0 90.0 120.0

      PERIODIC xy

    &END CELL

    &COORD

      C    0.00000000   0.00000000   0.00000000

      ...

      C   17.75000000  13.52731681   9.90000000

    &END COORD

    &KIND C

      BASIS_SET DZVP-MOLOPT-GTH

      POTENTIAL GTH-PBE-q4

    &END KIND

   &END SUBSYS

&END FORCE_EVAL

&MOTION

  &CONSTRAINT

    &FIXED_ATOMS

     COMPONENTS_TO_FIX XYZ

     LIST 1..194

    &END FIXED_ATOMS

  &END CONSTRAINT

  &GEO_OPT

  OPTIMIZER LBFGS

  MAX_ITER 300

  &END GEO_OPT

  &END MOTION

I spend 34 hours with 36 CPU and 120 GB of memory and still didn't optimised it yet. It seems that energy does not decrease much. 

     384

 i =        1, E =     -2025.9045014398

     384

 i =        6, E =     -2075.1262328870

 If I change cell parameters to 

  &SUBSYS

    &CELL

      ABC 18.46 13.53 50.0

      PERIODIC xy

      SYMMETRY ORTHORHOMBIC

    &END CELL

     384

 i =        1, E =     -2132.8940479508

     384

 i =       11, E =     -2146.1458464543

It does not help much.

Regards,

Dmitrii

[fabia...@gmail.com]

Dear Dmitrii,

The correct cell dimensions depend on the slab you built. I have never used avogardo, but I assume it gives you this information somehow.

"It seems that energy does not decrease much."

The energy decreases by about 50 Hartree, or 3.5 eV per atom, in 6 iterations. This indicates that the initial structure was not even remotely close to reasonable.

In the orthorhombic case the change is smaller, but 14 hartree is still far too large.

I suggest you start with a smaller number of atoms to get a feeling for the system without having to wait so long. First, optimize bulk graphene (https://www.cp2k.org/howto:geometry_optimisation) and also try to relax the cell dimensions. When you have done that, use the result to build the slab.

Cheers,

Fabian

[DMITRII Drugov]

Thank you a lot for this! I will update you with optimisation results, once it's done.

Regards,

Dmitrii

On Sunday, December 6, 2020 at 7:33:33 AM UTC+11 Travis wrote:

Hi,

Attached are prepared CIF files for bulk graphite with 4 layers. The 6x4x2 ortho cell has 384 atoms which seems to be the one you're working with. Avogadro is not recommended for periodic systems (trying to open one of these CIF files on my PC takes ages). It often messes up the structure by either adding extra atoms or losing a few. I'd start with a cell optimization of the bulk cell - these lattice parameters are from VASP. Then you can implement slab boundaries and fix the bottom layer(s).

-T