About "dE" in "OSZICAR"

[zhaoru_sun1]

Hello everyone,

As the Vaspwiki said

dE the change in the free energy from the last to the current step

But in my ion minimization calculation, I found that this is not the case:

Code: Select all

1 F= -.40565206E+03 E0= -.40562778E+03  d E =-.405652E+03
       N       E                     dE             d eps       ncg     rms          rms(c)
DAV:   1    -0.405657200457E+03   -0.51414E-02   -0.40483E+00  3288   0.245E+00    0.141E-01
DAV:   2    -0.405664323520E+03   -0.71231E-02   -0.72546E-02  4544   0.853E-01    0.189E-01
DAV:   3    -0.405664389471E+03   -0.65951E-04   -0.55461E-03  3320   0.904E-01    0.131E-01
DAV:   4    -0.405664216596E+03    0.17288E-03   -0.94294E-04  3488   0.376E-01    0.981E-02
DAV:   5    -0.405664234361E+03   -0.17765E-04   -0.68611E-04  4008   0.278E-01    0.834E-02
DAV:   6    -0.405664169467E+03    0.64894E-04   -0.75391E-04  4256   0.213E-01    0.245E-02
DAV:   7    -0.405664150767E+03    0.18700E-04   -0.83815E-05  4912   0.102E-01    0.145E-02
DAV:   8    -0.405664156230E+03   -0.54630E-05   -0.13978E-04  3368   0.550E-02    0.122E-02
DAV:   9    -0.405664155750E+03    0.48007E-06   -0.23013E-05  3208   0.230E-02    0.929E-03
DAV:  10    -0.405664155025E+03    0.72573E-06   -0.33263E-06  2512   0.149E-02
   2 F= -.40566416E+03 E0= -.40563999E+03  d E =-.120956E-01
       N       E                     dE             d eps       ncg     rms          rms(c)
DAV:   1    -0.405636339156E+03    0.27817E-01   -0.21267E+01  3280   0.563E+00    0.348E-01
DAV:   2    -0.405674944498E+03   -0.38605E-01   -0.39568E-01  4616   0.213E+00    0.471E-01
DAV:   3    -0.405674733535E+03    0.21096E-03   -0.28515E-02  3408   0.220E+00    0.299E-01
DAV:   4    -0.405673630326E+03    0.11032E-02   -0.52950E-03  3584   0.960E-01    0.231E-01
DAV:   5    -0.405673608558E+03    0.21769E-04   -0.42736E-03  3392   0.623E-01    0.199E-01
DAV:   6    -0.405673456292E+03    0.15227E-03   -0.49685E-03  4136   0.460E-01    0.905E-02
DAV:   7    -0.405673193007E+03    0.26328E-03   -0.72702E-04  4024   0.440E-01    0.312E-02
DAV:   8    -0.405673198195E+03   -0.51875E-05   -0.48210E-04  3712   0.970E-02    0.207E-02
DAV:   9    -0.405673204083E+03   -0.58887E-05   -0.15632E-04  3232   0.597E-02    0.195E-02
DAV:  10    -0.405673200652E+03    0.34310E-05   -0.21178E-05  3432   0.280E-02    0.114E-02
DAV:  11    -0.405673197184E+03    0.34682E-05   -0.13423E-05  3896   0.122E-02    0.412E-03
DAV:  12    -0.405673198201E+03   -0.10174E-05   -0.66401E-06  3264   0.126E-02    0.196E-03
DAV:  13    -0.405673198589E+03   -0.38772E-06   -0.76686E-07  1736   0.453E-03
   3 F= -.40567320E+03 E0= -.40564917E+03  d E =-.211392E-01

As above,

Code: Select all

3 F= -.40567320E+03
2 F= -.40566416E+03

but obviously,

Code: Select all

"3 F - 2 F= -0.00904" .not equal to. "d E =-.211392E-01"

What happened?

[merzuk.kaltak]

During Ionic minimization the free energy does not decrease monotonically.
Often, this is due to cell shape changes after an ionic step, such that the original basis set for a given energy cutoff ENCUT
is missing some Fourier modes in reciprocal space.
For relaxations, one should increase the energy cutoff slightly compared to the highest ENMAX value in the POTCAR file to prevent this kind of error.

A better measure for the ionic minimization is the force acting on the ions.
A convergence criterion of EDIFFG=-0.01 often suffices and yields tightly relaxed structures with an OUTCAR output like

Code: Select all

 POSITION                                       TOTAL-FORCE (eV/Angst)
 -----------------------------------------------------------------------------------
      1.46513      3.16890      1.41448         0.002536     -0.003502      0.000097
      8.16640      6.36170      1.41448        -0.002494      0.003595     -0.000054
      6.41213      1.41468      1.41447         0.004049      0.002728      0.000112
      3.21938      8.11592      1.41448        -0.003953     -0.002704     -0.000118

Nevertheless, you should always recalculate the ground state energy (single point calclulation) after an ionic relaxation to be sure that a minimum is reached.

The VASP community might tell you more with additional information (INCAR, POSCAR, POTCAR, OUTCAR, OSZICAR in form of a zip file).

[zhaoru_sun1]

During Ionic minimization the free energy does not decrease monotonically.
Often, this is due to cell shape changes after an ionic step, such that the original basis set for a given energy cutoff ENCUT
is missing some Fourier modes in reciprocal space.
For relaxations, one should increase the energy cutoff slightly compared to the highest ENMAX value in the POTCAR file to prevent this kind of error.

A better measure for the ionic minimization is the force acting on the ions.
A convergence criterion of EDIFFG=-0.01 often suffices and yields tightly relaxed structures with an OUTCAR output like

Code: Select all

 POSITION                                       TOTAL-FORCE (eV/Angst)
 -----------------------------------------------------------------------------------
      1.46513      3.16890      1.41448         0.002536     -0.003502      0.000097
      8.16640      6.36170      1.41448        -0.002494      0.003595     -0.000054
      6.41213      1.41468      1.41447         0.004049      0.002728      0.000112
      3.21938      8.11592      1.41448        -0.003953     -0.002704     -0.000118

Nevertheless, you should always recalculate the ground state energy (single point calclulation) after an ionic relaxation to be sure that a minimum is reached.

The VASP community might tell you more with additional information (INCAR, POSCAR, POTCAR, OUTCAR, OSZICAR in form of a zip file).

Thanks for your reply, I have attached my data file. But what I am confused about is the meaning of "d E", because I find that sometimes it is realy not equal to the difference of free energy "F" as mentioned above.

[merzuk.kaltak]

Thanks for your reply, I have attached my data file.

FYI, you haven't uploaded your data files.

[zhaoru_sun1]

FYI, you haven't uploaded your data files.

Sorry, I uploaded it again. And also shared a backup of GoogleDrive to avoid attachment failure. Thanks a lot!

[dorisV]

for IBRION=2 , dE of the initial, predictor and corrector steps are summed up, until the new geometry is chosen from these 3 steps
(please check the VASP Wiki concerning the algorithm of the conjugate gradient method)
therefore dE of steps 2,3, and 4 are
2: 2-1 (initial)
3: 3-1 (predictor)
4: 4-1 (corrector)
---
5: (5-4) ....

for IBRION=1, dE is the difference of the free energies between each subsequent ionic step.