what does d E actually mean in the output?

[guoxingmiao]

I had assumed it means the change of energy between steps (as in the VASP tuitorial: "dE is the change in the total energy between the current and the last step"), until recently I noticed it is nowhere close to such value. For example, I have a hard time converging an ion relaxation: the F and E0 virtually stopped changing (some <10^-4 variaztion) while d E still reports 10^-2 each step. Not surprisingly, it does not converge as I set EDIFF=0.001. For example, as below, the change between the last two E0 is 0.000005, while d E reports -0.018 so it does not converge. Is this normal?

Code: Select all

                                                                                                                                                                   
  34 F= -.76747776E+02 E0= -.76739597E+02  d E =-.180560E-01
 curvature:  -0.00 expect dE=-0.572E-05 dE for cont linesearch -0.322E-06
 ZBRENT: interpolating
 opt :   7.7535  next Energy=   -76.748184 (dE=-0.185E-01)
 bond charge predicted
       N       E                     dE             d eps       ncg     rms          rms(c)
DAV:   1    -0.767478208820E+02   -0.67046E-03   -0.43908E-03  5256   0.551E-01    0.464E-02
DAV:   2    -0.767494366242E+02   -0.16157E-02   -0.56860E-03  3384   0.117E+00    0.864E-02
DAV:   3    -0.767477339355E+02    0.17027E-02   -0.37551E-03  3408   0.991E-01    0.167E-02
DAV:   4    -0.767478741776E+02   -0.14024E-03   -0.13694E-03  3216   0.963E-02
  35 F= -.76747874E+02 E0= -.76739709E+02  d E =-.181539E-01
 curvature:  -0.04 expect dE=-0.915E-04 dE for cont linesearch -0.244E-05
 ZBRENT: bisectioning
 opt :   7.7287  next Energy=   -76.747887 (dE=-0.182E-01)
 bond charge predicted
       N       E                     dE             d eps       ncg     rms          rms(c)
DAV:   1    -0.767478675938E+02   -0.13366E-03   -0.84346E-04  3192   0.509E-02    0.174E-02
DAV:   2    -0.767478686959E+02   -0.11021E-05   -0.85529E-05  3072   0.309E-02
  36 F= -.76747869E+02 E0= -.76739718E+02  d E =-.181484E-01
 curvature:  -0.21 expect dE=-0.671E-03 dE for cont linesearch -0.277E-03
 ZBRENT: interpolating
 opt :   7.7491  next Energy=   -76.747875 (dE=-0.182E-01)
 bond charge predicted
       N       E                     dE             d eps       ncg     rms          rms(c)
DAV:   1    -0.767478814259E+02   -0.13832E-04   -0.35047E-04  3192   0.113E-01    0.188E-02
DAV:   2    -0.767479990997E+02   -0.11767E-03   -0.51165E-04  3216   0.344E-01
  37 F= -.76747999E+02 E0= -.76739879E+02  d E =-.182788E-01
 curvature:  -0.00 expect dE=-0.223E-05 dE for cont linesearch -0.501E-06
 ZBRENT: bisectioning
 opt :   7.7389  next Energy=   -76.748027 (dE=-0.183E-01)
 bond charge predicted
       N       E                     dE             d eps       ncg     rms          rms(c)
DAV:   1    -0.767478884676E+02   -0.70417E-05   -0.34511E-04  3192   0.289E-01    0.607E-03
DAV:   2    -0.767479431266E+02   -0.54659E-04   -0.19609E-04  3096   0.208E-01
  38 F= -.76747943E+02 E0= -.76739809E+02  d E =-.182228E-01
 curvature:  -0.04 expect dE=-0.996E-04 dE for cont linesearch -0.204E-04
 ZBRENT: interpolating
 opt :   7.7436  next Energy=   -76.747946 (dE=-0.182E-01)
 bond charge predicted
       N       E                     dE             d eps       ncg     rms          rms(c)
DAV:   1    -0.767478865395E+02    0.19281E-05   -0.11446E-04  3096   0.167E-01    0.128E-03
DAV:   2    -0.767478924109E+02   -0.58715E-05   -0.54410E-05  3072   0.133E-02
  39 F= -.76747892E+02 E0= -.76739733E+02  d E =-.181721E-01
 curvature:  -0.01 expect dE=-0.156E-04 dE for cont linesearch -0.344E-05
 ZBRENT: bisectioning
 opt :   7.7463  next Energy=   -76.747896 (dE=-0.182E-01)
 bond charge predicted
       N       E                     dE             d eps       ncg     rms          rms(c)
DAV:   1    -0.767478967494E+02   -0.10210E-04   -0.19734E-05  2640   0.103E-02    0.252E-03
DAV:   2    -0.767478972856E+02   -0.53619E-06   -0.31254E-06  2904   0.442E-03
  40 F= -.76747897E+02 E0= -.76739738E+02  d E =-.181770E-01
 curvature:  -0.01 expect dE=-0.176E-04 dE for cont linesearch -0.408E-05
 ZBRENT: bisectioning
 opt :   7.7477  next Energy=   -76.747899 (dE=-0.182E-01)
 

[marie-therese.huebsch]

Dear guoxingmiao,

Sorry for the late response!

dE always refers to some energy difference, but as there are many selfconsistency loops and minimization algorithms, e.g., the ionic relaxation, electronic relaxation, line minimization, and the diagonalization of the Kohn–Sham Hamiltonian, it might not be obvious which energy difference it refers to. A more detailed summary of each relaxation step can be found in the OUTCAR file!

Also, note that the electronic convergence criterion is set by the EDIFF tag, and the ionic convergence criterion is controlled by EDIFFG! Please read their documentation on the VASP Wiki.

Please try to understand your calculation using the information written in the OUTCAR file. If you still have questions, you can upload your INCAR, KPOINTS, POTCAR, and POSCAR file, as well as the corresponding OUTCAR file here and I will be help you interpret the information.

Best regards,
Marie-Therese

[huangjs]

Hi guoxingmiao,

In your case given above, it looks like you were using IBRION=2. You need to take the the energy difference between n+1 and n step and then add the difference to the dE shown in the n step, to have the dE that agrees with the number shown in the output for the n+1 step.

huangjs