Force relaxation and its dependence on EDIFF

[simha]

Dear Vasp Users

I am doing a surface calculation and firstly relaxing the surface.

My INCAR setting are the following

PREC = Accurate
ISPIN = 2
EDIFF = 1E-04
LREAL = AUTO
ALGO = VeryFast
IALGO = 48
NELMIN = 3
LCHARG = .TRUE.
EDIFFG = -0.05
NSW = 37
IBRION = 1
ISIF = 0

When relaxing the ouput looks like this

** I am listing the last few iterations of first 2 ionic steps*****

RMM: 48 -0.701690649710E+03 0.21327E-02 -0.23235E-03 8988 0.631E-02 0.166E-01
RMM: 49 -0.701688606579E+03 0.20431E-02 -0.26740E-03 8931 0.658E-02 0.188E-01
RMM: 50 -0.701692660158E+03 -0.40536E-02 -0.87018E-04 8782 0.413E-02 0.163E-01
RMM: 51 -0.701695590506E+03 -0.29303E-02 -0.20824E-03 8963 0.636E-02 0.221E-01
RMM: 52 -0.701696999570E+03 -0.14091E-02 -0.64194E-04 8398 0.323E-02 0.223E-01
RMM: 53 -0.701692708804E+03 0.42908E-02 -0.39089E-03 8983 0.864E-02 0.221E-01
RMM: 54 -0.701681022916E+03 0.11686E-01 -0.15572E-02 9122 0.164E-01 0.471E-01
RMM: 55 -0.701691004750E+03 -0.99818E-02 -0.15943E-03 9013 0.499E-02 0.372E-01
RMM: 56 -0.701693018930E+03 -0.20142E-02 -0.26764E-03 8991 0.736E-02 0.260E-01
RMM: 57 -0.701689097077E+03 0.39219E-02 -0.33080E-02 9292 0.238E-01 0.516E-01
RMM: 58 -0.701682122258E+03 0.69748E-02 -0.74224E-02 9431 0.345E-01 0.159E+00
RMM: 59 -0.701714711424E+03 -0.32589E-01 -0.69837E-03 9635 0.116E-01 0.131E+00
RMM: 60 -0.701708370028E+03 0.63414E-02 -0.34600E-03 9060 0.902E-02
1 F= -.70170837E+03 E0= -.70153590E+03 d E =-.701708E+03 mag= 72.0027
BRION: g(F)= 0.148E+01 g(S)= 0.000E+00
bond charge predicted

RMM: 26 -0.701590723695E+03 -0.25532E-04 -0.22483E-05 6117 0.658E-03 0.221E-02
RMM: 27 -0.701590712942E+03 0.10752E-04 -0.41896E-05 6191 0.806E-03 0.166E-02
RMM: 28 -0.701590730411E+03 -0.17468E-04 -0.52186E-05 6699 0.916E-03 0.174E-02
RMM: 29 -0.701590782666E+03 -0.52255E-04 -0.10676E-05 5548 0.486E-03 0.892E-03
RMM: 30 -0.701590800100E+03 -0.17434E-04 -0.65469E-06 5422 0.396E-03 0.813E-03
RMM: 31 -0.701590808623E+03 -0.85224E-05 -0.27878E-06 5000 0.234E-03
2 F= -.70159081E+03 E0= -.70141433E+03 d E =0.117561E+00 mag= 72.0002
BRION: g(F)= 0.439E+01 g(S)= 0.000E+00 retain N= 1 mean eig= 0.44
eig: 0.444
bond charge predicted

In the first step, the electronic step 60 is reached and the criteria of convergence for electronic step is not reached i.e. E-04 but E-03 was present. If it continues to the next step the convergence is reached within the 60 iterations.

Is it necessary to have E-04 electronic convergence in each ionic step to have accurate forces?

Or Can we accept the E-03 convergence in first step and as in the subsequent steps I see a good electronic convergence of E-05 and also forces.

(I am aware of the fact that I fixed the parameter of EDIFF=E-04.
)

I will thankful if anyone can share his/her experience.

Regards
Narsimham

[admin]

no, even 10-4 is a very crude criterium, I doubt whether this is accurate enough to obtain reasonable forces. If even thi s convergence criterium was not reached, the results on the forces cannot be fully trusted. If you look at the total energy of the follow-up step, you see that it is less negative by 0.12 eV, i.e., the system relaxed into the wrong direction at the first step. You can of course continue from this run, but please carefully check if the final results look reasonable (Sometimes the final results are ok even if the first step was not fully converged, but this cannot be taken for granted).
It would be safer however to increase NELM and decrease AMIX, BMIX, AMIX_MAG and BMIX_MAG (the charge densities and magnetic densities are not converged either for the first step, the second one looks ok) and to restart