HSE06 Freezes
Posted: Tue Jan 17, 2012 12:41 am
Hi all,
I'm having a strange problem with an HSE06 calculation. On the Ranger supercomputer, I am unable to get even a simple electronic relaxation of a monoclinic HfO2 unit cell (12 atoms) to converge.
Procedure:
I start with a standard PBE calculation, then use the output CHGCAR file in the HSE06 calculation. VASP then proceeds to run as expected, until it gets to the third or fourth electronic iteration. At this point, all apparent progress seems to stop, although no errors are written out by VASP. Instead, the job hangs until it is killed by Ranger.
I have heard that the full exchange potential used in HSE06 isn't applied until a few iterations in, so this hangup could be happening when VASP begins to apply the exchange potential. However, I have tried running this job on 32, 64, and 96 cores, so it seems unlikely that this problem is a memory problem.
Do you have any idea what I may be doing wrong, or how I can remedy this problem? My INCAR file is at the bottom of this post, and any help is greatly appreciated.
Regards,
Dan
###INCAR###
SYSTEM = Monoclinic HfO2
Start parameter for this Run:
ISTART = 1
ICHARG = 1
# Enabling hybrid functional (HSE06) calculations
LHFCALC = .TRUE.; HFSCREEN = 0.2 # Enables Hartree-Fock-style calculation
# Should be 0.2-0.3A. HSE06 requires 0.2 Angstroms
#ALGO = All # Advised for semiconductors and insulators
#TIME = 0.4
PRECFOCK = A # Set to Normal for high-quality calculations
# LMAXFOCK = 4 # 4 for d-electrons. May require 6 for f-electrons
NBANDS = 80
Electronic Relaxation 1
# IALGO = 48 # Can be specified by the choice of ALGO
NELM = 40
NELMIN = 0
NELMDL = 3
EDIFF = 1E-05 # Tolerance for electron convergence
Ionic Relaxation
EDIFFG = -1E-04 # Relaxation tolerance for forces (positive number would be for energy)
NSW = 0 # Number of ionic convergence steps
IBRION = -1 # Selects ionic relaxation algorithm. See VASP manual
ISIF = 0 # All lattice parameters vary
PREC = M
ISPIN = 2
NPAR = 32 # 32 cores used
RWIGS = 1.924 0.725
LPLANE = .TRUE.
LWAVE = .TRUE.
LCHARG = .TRUE.
I'm having a strange problem with an HSE06 calculation. On the Ranger supercomputer, I am unable to get even a simple electronic relaxation of a monoclinic HfO2 unit cell (12 atoms) to converge.
Procedure:
I start with a standard PBE calculation, then use the output CHGCAR file in the HSE06 calculation. VASP then proceeds to run as expected, until it gets to the third or fourth electronic iteration. At this point, all apparent progress seems to stop, although no errors are written out by VASP. Instead, the job hangs until it is killed by Ranger.
I have heard that the full exchange potential used in HSE06 isn't applied until a few iterations in, so this hangup could be happening when VASP begins to apply the exchange potential. However, I have tried running this job on 32, 64, and 96 cores, so it seems unlikely that this problem is a memory problem.
Do you have any idea what I may be doing wrong, or how I can remedy this problem? My INCAR file is at the bottom of this post, and any help is greatly appreciated.
Regards,
Dan
###INCAR###
SYSTEM = Monoclinic HfO2
Start parameter for this Run:
ISTART = 1
ICHARG = 1
# Enabling hybrid functional (HSE06) calculations
LHFCALC = .TRUE.; HFSCREEN = 0.2 # Enables Hartree-Fock-style calculation
# Should be 0.2-0.3A. HSE06 requires 0.2 Angstroms
#ALGO = All # Advised for semiconductors and insulators
#TIME = 0.4
PRECFOCK = A # Set to Normal for high-quality calculations
# LMAXFOCK = 4 # 4 for d-electrons. May require 6 for f-electrons
NBANDS = 80
Electronic Relaxation 1
# IALGO = 48 # Can be specified by the choice of ALGO
NELM = 40
NELMIN = 0
NELMDL = 3
EDIFF = 1E-05 # Tolerance for electron convergence
Ionic Relaxation
EDIFFG = -1E-04 # Relaxation tolerance for forces (positive number would be for energy)
NSW = 0 # Number of ionic convergence steps
IBRION = -1 # Selects ionic relaxation algorithm. See VASP manual
ISIF = 0 # All lattice parameters vary
PREC = M
ISPIN = 2
NPAR = 32 # 32 cores used
RWIGS = 1.924 0.725
LPLANE = .TRUE.
LWAVE = .TRUE.
LCHARG = .TRUE.