SCCS Continuum solvation model

[Nico Holmberg]

Dear CP2K users & developers,

I am interested in trying out the SCCS implicit solvent model by Andreussi et al. ("Revised self-consistent continuum solvation in electronic-structure calculations", http://dx.doi.org/10.1063/1.3676407). According to the online input reference manual, this feature should be implemented in CP2K version 2.6, revision 14482 (http://cp2k.web.psi.ch/manual/devel/CP2K_INPUT/FORCE_EVAL/DFT/SCCS/ANDREUSSI.html). However, when I downloaded the said version from the SVN repository, I found no mention of the method in the source code. I also tried looking at the SVN change logs for the bibliography and input parsing files (common/bibliography.F and input_cp2k_dft.F) but again I found nothing.

If anyone is familiar with the current status of the SCCS method implementation I would appreciate any details you can provide.


Regards,

Nico Holmberg
Doctoral candidate
Department of Chemistry, Aalto University
Finland

[Matthias Krack]

Dear Nico,

you are referring to the local manual of the CP2K development version at PSI in which the SCCS model is in fact implemented. This implementation is not yet committed to the main CP2K repository, but I hope this will be the case soon. Just follow the commits to the main CP2K trunk version.

Regards,

Matthias

[Nico Holmberg]

Dear Matthias,

I didn't realize the PSI development version was separate from the trunk version. Thank you for the clarification. I will continue following the commits.


Regards,

Nico Holmberg

[Matthias Krack]

Dear Nico,

for your information, an implementation of the SCCS model is now committed to the CP2K trunk version.

Regards,

Matthias

[Satish Kumar]

Hello CP2K users

I was trying to do a geometry optimization of a metallic surface using SCCS [Andreussi]. However I was getting the error related to Analytical stress tensor not implemented yet in CP2K 2.6. I am not relaxing the unit cell and I was not sure if not including stress tensor would cause unreasonable total energies calculated using SCCS. Any thoughts?

Even though I specify Fattebert-Gygi subsection in SCCS, the calculation corresponding to only Andreussi takes place. Any idea what might be happening?

Will CP2K 2.7 resolve any of these issues?

[Matthias Krack]

Hi,

Maybe you requested the calculation of the stress tensor, e.g. you specified the keyword STRESS_TENSOR.
The presence of the SCCS method section is not sufficient. Did you select the right method using the METHOD keyword?

Matthias

[Satish Kumar]

Thank you Matthias, I now use the method and FATTEBERT-GYGI works but even for just a water molecule in a cell, the calculation runs forever and does not converge. I first restarted calculation from wavefunction of gas phase calculation, then tried all the tags with the default values in SCCS section. Finally I increased EPS_SCF within SCCS to 1.0E-04 thinking it may improve convergence. But none of them helped. I am attaching the result of grep-ing "OT CG" from the output file. Till EPS_SCF reaches 1.0E-04, the convergence is fine but I think when the SCCS loop begins (I don't actually find any explicit mention with "Print medium" about when SCF loops corresponding to SCCS begins), the convergence is very oscillating. Is there anything I can do to improve convergence with SCCS.

grep "OT CG" output:

 OT CG       0.64E-01    1.2     0.38492654       -16.5223537330 -1.65E+01

     4 OT CG       0.47E-01    1.4     0.33467993       -16.8712267043 -3.49E-01

     7 OT CG       0.37E-01    1.4     0.18362126       -17.1265030091 -2.55E-01

    10 OT CG       0.31E-01    1.4     0.06956912       -17.1886589747 -6.22E-02

    13 OT CG       0.74E-01    1.3     0.04840266       -17.2094960351 -2.08E-02

    16 OT CG       0.72E-01    1.4     0.01997461       -17.2194004920 -9.90E-03

    19 OT CG       0.40E-01    1.4     0.01155289       -17.2203057880 -9.05E-04

    22 OT CG       0.87E-01    1.4     0.00522995       -17.2209729877 -6.67E-04

    25 OT CG       0.61E-01    1.4     0.00301797       -17.2210686037 -9.56E-05

    28 OT CG       0.51E-01    1.4     0.00192432       -17.2210954837 -2.69E-05

    31 OT CG       0.58E-01    1.4     0.00085921       -17.2211078888 -1.24E-05

    34 OT CG       0.72E-01    1.5     0.00045713       -17.2211109595 -3.07E-06                 #I think beyond this SCCS starts

    37 OT CG       0.12E+00   10.5     0.03513936       -16.9374989979  2.84E-01

    40 OT CG       0.19E+00   12.8     0.53438787       -13.0284958890  3.91E+00

    43 OT CG       0.15E+00   12.6     0.18217280       -16.4652852134 -3.44E+00

    46 OT CG       0.71E-01   11.4     0.07232781       -16.6107052961 -1.45E-01

    49 OT CG       0.67E-01   11.0     0.02449012       -16.6312618955 -2.06E-02

    52 OT CG       0.11E+00    9.7     0.00857426       -16.6349458760 -3.68E-03

    55 OT CG       0.48E+00    9.4     0.01771721       -16.6370477486 -2.10E-03

    58 OT CG       0.12E+01   11.8     0.06495594       -16.6625230341 -2.55E-02

    61 OT CG       0.10E-01   13.1     0.05789947       -16.6669359830 -4.41E-03

    64 OT CG       0.45E-01   11.0     0.03152062       -16.6795843687 -1.26E-02

    67 OT CG       0.12E+00    9.9     0.01221376       -16.6858425676 -6.26E-03

    70 OT CG       0.17E+00    9.6     0.01215730       -16.6872572020 -1.41E-03

    73 OT CG       0.53E+00   11.9     0.03221792       -16.6915824133 -4.33E-03

    76 OT CG       0.16E+00   10.5     0.04503294       -16.7005483657 -8.97E-03

    79 OT CG       0.65E+00   15.5     0.19838443       -16.6988538474  1.69E-03

    82 OT CG       0.10E+01   13.5     0.69320124       -11.0634009652  5.64E+00

    85 OT CG       0.78E+00   34.4     0.57682957       -11.1232371185 -5.98E-02

    88 OT CG       0.53E+00   34.2     0.66048561       -10.3342895296  7.89E-01

    91 OT CG       0.34E+00   12.4     0.41865566       -15.7784514964 -5.44E+00

    94 OT CG       0.10E+00   34.0     0.29143613       -13.0784219668  2.70E+00

    97 OT CG       0.88E-01   12.5     0.24158684       -16.5534274750 -3.48E+00

   100 OT CG       0.47E-01   12.2     0.05092026       -16.7441647536 -1.91E-01

     1 OT CG       0.64E-01   21.9     0.28419623       -16.7524473550 -8.28E-03

     4 OT CG       0.58E-01   10.5     0.18454413       -17.0578460723 -3.05E-01

     7 OT CG       0.18E-01    9.4     0.14071768       -17.1058023331 -4.80E-02

    10 OT CG       0.11E-01   10.0     0.11841442       -17.1118483343 -6.05E-03

    13 OT CG       0.89E-04    9.3     0.11828280       -17.1118213718  2.70E-05

    16 OT CG       0.14E-03    7.1     0.11807277       -17.1117763182  4.51E-05

    19 OT CG       0.11E-05    6.7     0.11807112       -17.1117758246  4.94E-07

    22 OT CG       0.18E-05    4.9     0.11806842       -17.1117753439  4.81E-07

    25 OT CG       0.29E-05    5.6     0.11806413       -17.1117743931  9.51E-07

    28 OT CG       0.47E-05    5.6     0.11805727       -17.1117728613  1.53E-06

    31 OT CG       0.75E-05    6.9     0.11804630       -17.1117704080  2.45E-06

    34 OT CG       0.12E-04    5.8     0.11802874       -17.1117665058  3.90E-06

    37 OT CG       0.19E-04    6.3     0.11800066       -17.1117602075  6.30E-06

    40 OT CG       0.31E-04    6.5     0.11795573       -17.1117500820  1.01E-05

    43 OT CG       0.49E-04    6.5     0.11788392       -17.1117336756  1.64E-05

    46 OT CG       0.78E-04    6.8     0.11776915       -17.1117069099  2.68E-05

After 92 optimization steps:

--------  Informations at step =    92 ------------

  Optimization Method        =                 BFGS

  Total Energy               =       -17.2378273186

  Real energy change         =        -0.0000433883

  Predicted change in energy =        -0.0000203875

  Scaling factor             =         0.7336561632

  Step size                  =         0.0195987283

  Trust radius               =         0.4724315332

  Decrease in energy         =                   NO

  Used time                  =              179.174

  Convergence check :

  Max. step size             =         0.0195987283

  Conv. limit for step size  =         0.0030000000

  Convergence in step size   =                   NO

  RMS step size              =         0.0103012527

  Conv. limit for RMS step   =         0.0015000000

  Convergence in RMS step    =                   NO

  Max. gradient              =         0.0043764738

  Conv. limit for gradients  =         0.0009725000

  Conv. for gradients        =                   NO

  RMS gradient               =         0.0023573863

  Conv. limit for RMS grad.  =         0.0003000000

  Conv. for gradients        =                   NO

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

Input file:

&GLOBAL

   PROJECT_NAME 4x4

   RUN_TYPE  GEO_OPT

 &END GLOBAL

 &MOTION

   &GEO_OPT

     OPTIMIZER  BFGS

     MAX_ITER  200

     MAX_FORCE     9.7249999999999995E-04

     STEP_START_VAL  90

   &END GEO_OPT

 &END MOTION

 &FORCE_EVAL

   METHOD  QS

   &DFT

     BASIS_SET_FILE_NAME ./GTH_BASIS_SETS_5-12-10

     POTENTIAL_FILE_NAME ./GTH_POTENTIALS_5-12-10

     WFN_RESTART_FILE_NAME x.wfn

     &SCF

       MAX_SCF  100

       EPS_SCF     1.0000000000000001E-05

       SCF_GUESS  ATOMIC

       &OT  T

         MINIMIZER  CG

         LINESEARCH  3PNT

         PRECONDITIONER  FULL_SINGLE_INVERSE

       &END OT

       &OUTER_SCF  T

         EPS_SCF     1.0000000000000001E-05

         MAX_SCF  20

       &END OUTER_SCF

     &END SCF

     &QS

       EXTRAPOLATION  ASPC

       EXTRAPOLATION_ORDER  3

     &END QS

     &MGRID

       NGRIDS  5

       CUTOFF     4.8000000000000000E+02

     &END MGRID

     &XC

       DENSITY_CUTOFF     1.0000000000000000E-10

       GRADIENT_CUTOFF     1.0000000000000000E-10

       TAU_CUTOFF     1.0000000000000000E-10

       &XC_FUNCTIONAL  NO_SHORTCUT

         &PBE  T

         &END PBE

       &END XC_FUNCTIONAL

     &END XC

     &PRINT

     &END PRINT

     &SCCS  T

       DERIVATIVE_METHOD  FFT

       DIELECTRIC_CONSTANT     7.8400000000000006E+01

       EPS_SCCS     9.9999999999999995E-07

       EPS_SCF     5.0000000000000001E-04

       MAX_ITER  100

       METHOD  FATTEBERT-GYGI

       MIXING     5.9999999999999998E-01

       &FATTEBERT-GYGI

         BETA     1.7000000000000000E+00

         RHO_ZERO     5.9999999999999995E-04

       &END FATTEBERT-GYGI

     &END SCCS

   &END DFT

   &SUBSYS

     &CELL

       A     1.2000000000000000E+01    0.0000000000000000E+00    0.0000000000000000E+00

       B     0.0000000000000000E+00    1.3000000000000000E+01    0.0000000000000000E+00

       C     0.0000000000000000E+00    0.0000000000000000E+00    1.4000000000000000E+01

       MULTIPLE_UNIT_CELL  1 1 1

     &END CELL

     &COORD

O   -7.5733907062610586E-02    8.9770405110133449E-03    1.2904624124972142E-01

H    6.5401889552783654E-01   -5.0597734685847906E-01    5.1647932522805295E-01

H   -8.1191163881804929E-01   -9.1336870679231297E-02    7.6138190132564698E-01

     &END COORD

{

basis and potential data

}

     &TOPOLOGY

       NUMBER_OF_ATOMS  3

       MULTIPLE_UNIT_CELL  1 1 1

     &END TOPOLOGY

   &END SUBSYS

 &END FORCE_EVAL

[Matthias Krack]

Hi,

Did you check, if the inner SCCS cycle for the  polarisation charge  converges in each SCF iteration step? (EPS_SCCS should be tight 1E-8-1E-10)
My further suggestions are: 

- the DERIVATIVE_METHOD FFT requires quite high cutoff values for numerical reasons, possibly 600-800 Ry. In this respect CD5 is often a better choice.

- drop the EPS_* thresholds e.g.in the XC section and use just EPS_DEFAULT 1.0E-14 and possibly MAP_CONSISTENT.

Matthias

Von meinem iPod gesendet

--
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.

[Satish Kumar]

Hi 

Thank you for your response. I tried what you suggested.

1) When I use default values (refer in.defaults and out.default) for all tags within SCCS section, I do see polarization energy converging to some value (which I do not if is right) but the electronic minimizations do not converge. 

2) How does the electronic minimization work? I can see that vacuum bases electronic minimization takes place till EPS_SCF criteria within SCCS section is met. Then SCCS bases electronic minimzation occurs. After this is it just EPS_SCCS that determines when electronic minimization finishes? If not how does it work.

3) I have attaches two more input (in.*) and output (out.*). When I use EPS_SCF within SCCS ~E-13, I do not see any polarization energy printed (corresponding files are *.pol_energy). However, when I use EPS_SCF as default value, as above, I do see polarization enegy printed but electronic minimization doesn't converge.

4) Also, when I did a 1000 step electronic minimization without outer-scf loop, the energy corresponding to the convergence column in the output does not change after reaching a certain value. Corresponding seetings were: EPS_SCF 1.0E-11 and within SCCS section: EPS_SCCS     1.0E-11, EPS_SCF     1.01E-05. I do not understand what is happening.

Thank you.

[Matthias Krack]

Hi,

you are using DZVP-MOLOPT-GTH basis sets instead of their SR (Short-Range) variants DZVP-MOLOPT-SR-GTH which are better suited for condensed phase simulations. I think this is the major reason for your convergence problems, because an explicit water simulation (box filled with H2O molecules) will most likely show a similar behavior.
You forgot to drop the EPS_* thesholds

    DENSITY_CUTOFF     1.0000000000000000E-10
    GRADIENT_CUTOFF     1.0000000000000000E-10
    TAU_CUTOFF     1.0000000000000000E-10

in the &XC section.

You may also add in the &DFT section
     &PRINT
      &SCCS on
       &EACH
        QS_SCF 1
       &END EACH
      &END SCCS
     &END PRINT
for a more detailed SCCS output.

Finally, I would use for MAX_SCF in the &SCF section a much smaller value and a larger threshold value for EPS_SCF, e.g.
    &SCF
       MAX_SCF 31
       EPS_SCF 3.0E-7
The large MAX_SCF suppresses the outer SCF cycle and thus an improved preconditioning for OT.

Matthias