Page 1 of 1

mBJ calculation is stuck

PostPosted: Sun Sep 02, 2018 4:21 pm
by jcconesa
I am trying to do calculations with METAGGA=MGJ. I first obtain the charge density file with a normal PBE calculation and then I do the MBJ one using this INCAR file:

PREC = Accurate
ALGO = Damped; TIME=0.3
IBRION = -1 ; NSW = 0
EMIN= -4.0; EMAX = 12.0; NEDOS = 1601

Results are quite reasonable with small unit cells, but when I use a cell of 55 atoms (with ISPIN=1) the MBJ calculation runs during a few hours and then it is stuck (without aborting) in one of the SCF cycles. I must then abort it manually. This occurs with different values of ALGO. I have seen that this happens when the first value in the OUTCAR line containing the words "total energy-change" is approaching the EDIFF value. The last lines in OUTCAR are:

--------------------------------------- Iteration 1( 83) ---------------------------------------

POTLOK: cpu time 1.6052: real time 1.6053
CMBJ = 1.2393

SETDIJ: cpu time 4.2327: real time 4.2330
EDDAV: cpu time 39.4296: real time 39.4313
BZINTS: Fermi energy: 9.570078;440.000000 electrons
Band energy:***********; BLOECHL correction: 0.000000
DOS: cpu time 0.0217: real time 0.0217

Can you please give some help?
Best regards,
José C. Conesa

Re: mBJ calculation is stuck

PostPosted: Fri Sep 07, 2018 9:24 am
by admin

using 80 cores (4 nodes a 2x10 cores) without any CHGCAR or WAVECAR read in (fresh start) we obtain a solution with your supplied INCAR in 77 steps and the calculation finishes without error.
The corresponding binary was compiled with INTEL-15.0.3 in combination with OpenMPI 1.8.10 and ScaLAPACK-2.0.2.

Nevertheless, we also obtain the following entry in the OUTCAR:
Code: Select all
 BZINTS: Fermi energy:  9.640039;440.000000 electrons
         Band energy:***********;  BLOECHL correction:  0.000000

which is related to the fact that the total band energy is too large for the expected format. This issue appears only for ISMEAR=-5 and is due to line 129 and 130 in ./src/tet.F of vasp.5.4.4 and reads:
Code: Select all
40 FORMAT(1X,'BZINTS: Fermi energy:',F10.6,';',F10.6,' electrons'/ &
     &       9X,'Band energy:',F11.6,';  BLOECHL correction:',F10.6)

This will be fixed in a future release by changing these lines to
Code: Select all
40 FORMAT(1X,'BZINTS: Fermi energy:',F10.6,';',F10.6,' electrons'/ &
     &       9X,'Band energy:',E13.6,';  BLOECHL correction:',F10.6)

This requires, however, to recompile vasp. Probably, the easiest way is to just switch to another k-point smearing, such as ISMEAR >=0.