residual force for EDIFFG < 0

[okuno]

Dear Vasp Users.

I'm relaxing the atom potision with EDIFFG < 0 option.

The manual said the when EDIFFG < 0, the relaxation will be stopped if all
forces are smaller than abs(EDIFFG).

So, I set EDIFFG=-0.005 (ev/Angust, rather severe condition) with IBIRON=2, ISFI=2, and tried, but the redisual forces are not
smaller than the EDIFFG like below,
POSITION TOTAL-FORCE (eV/Angst)
-----------------------------------------------------------------------------------
0.00000 0.00000 8.23045 0.000000 0.000000 -0.005814
0.00000 0.00000 4.37285 0.000000 0.000000 -0.003828
2.02102 2.02102 1.99133 0.000000 0.000000 -0.004440
2.02102 2.02102 6.09146 0.000000 0.000000 0.006101
2.02102 2.02102 7.88044 0.000000 0.000000 -0.000891
2.02102 2.02102 3.98478 0.000000 0.000000 0.006111
2.02102 0.00000 1.57380 0.000000 0.000000 -0.003879
0.00000 2.02102 1.57380 0.000000 0.000000 -0.003879
2.02102 0.00000 5.88869 0.000000 0.000000 0.005259
0.00000 2.02102 5.88869 0.000000 0.000000 0.005259
-----------------------------------------------------------------------------------

Why the total forces on ions are not smaller than 0.005?

My INCAR file is

SYSTEM = PZT_P4mm

Startparameter for this Run:
NWRITE = 2; LPETIM=F write-flag & timer
ISTART = 0 job : 0-new 1-cont 2-samecut
ICHARG = 2 initial charge 0-calc 1-from CHGCAR, 2-super potision


Electronic minimization
ALGO = NORMAL # default for IALGO=38
IALGO = 38 # default Davodason block iteration scheme
EDIFF = 1E-04 # stopping -criterion for ELM
PREC = High

relaxation
IBRION = 2
ISFI = 2
NSW=20
EDIFFG = -0.005

NELM =40 # maximum electronic SC steps (default=60)

GGA = 91 # GGA PB PW LM 91 PE RP

ISMEAR = -5
SIGMA = 0.1

[tjf]

Hi okuno

Firstly, look at how many optimisation steps are being made. Is it less than 20? If not, it's reaching the NSW limit that's terminating the job, not the EDIFFG limit.

Secondly, I'd be using an EDIFF smaller than 0.0001 and probably setting a NELMIN to something like 8 to get good forces smaller than 0.005.