problem in dielectric constant calc using hybrid functional

[jcconesa]

Hi,
I'm trying to compute with VASP 5.4.1 the dielectric constant for the cubic perovskite Ba2YMoO6 at fixed geometry using hybrid functional of PBE0 type, PAW potentials and spin polarized conditions. After two SCF runs are completed (the last one needing more tan 300 cycles to reach the EDIFF test, due to the stringent convergence condition) the program gets stuck without aborting. The last lines in OUTCAR are

------------------------ aborting loop because EDIFF is reached ----------------------------------------


CALCP: cpu time 19.6910: real time 19.7109

Ionic dipole moment: p[ion]=( 17.95341 30.31005 0.10297 ) electrons Angst

Spin resolved dipole moment: p[sp1]=( -0.20595 0.39204 -0.05981 ) electrons Angst
p[sp2]=( 2.95906 -1.34308 2.38728 ) electrons Angst

Total electronic dipole moment: p[elc]=( 2.75311 -0.95105 2.32747 ) electrons Angst


FORLOC: cpu time 0.0080: real time 0.0090
FORHF : cpu time 1149.0223: real time 1148.9027
FORNL : cpu time 3.5625: real time 3.5618

and standard output ends with producing about one dozen groups of lines of the type

PEAD_POLARIZATION_CALC: WARNING: reciprocal direction 1 spin channel 1
contribution from string 8 not well clustered around <|S|>_av = ( 0.43063 0.11001 )
Im(ln[|S|/<|S|>_av])/2pi = 0.49952 > 1/4

where the string index varies and the value >1/4 apperas positive or negative but has always absolute magnitude slightly below 0.5. This happens for two values of <|S|>_av . INCAR is

SYSTEM = opt perovsk
NCORE = 1
KPAR=8
ALGO = Damped ; TIME = 0.5
NELM=500
ISMEAR = -5
PREC = Normal
LREAL=.TRUE.
ENCUT=500
LASPH = .TRUE. ; ADDGRID = .TRUE. ; LMAXMIX = 4
ISPIN = 2
EDIFF=1E-04
ICHARG=1
!
LHFCALC = .TRUE.; AEXX = 0.24
PRECFOCK = Fast
NKRED=2
!
LCALCEPS=.TRUE. ; EFIELD PEAD = 0.002 0.002 0.002
!
IBRION = -1
LWAVE=.FALSE.

and KPOINTS is

gamma
0
G
6 6 6

The program, compiled with ifort 11.1.038 and mkl libraries, runs under OpenMPI in a box having 4 nodes with 12 Xeon cores each, connected with infiniband.

Please help

Regards and thanks in advance,
José Carlos Conesa

[admin]

The parameter KPAR>1 was not tested on all platforms.
Check if your calculation gets stuck when KPAR=1 .