EDWAV internal error in gamma-only HSE calculations

[xamnidar]

Hi, when running HSE calculations with ALGO = Damped, I get an EDWAV error after the fifth scf step, at which point the code seems to hang. Curiously, this error only seems to happen with the gamma-only version of vasp. The same input files work fine for the regular version of vasp, even if only one k-point is used. (This is all with vasp 5.3).

Does anyone know anything about this problem? This error message has been mentioned elsewhere on the forums, but I haven't seen any solutions. The problem seems to be related to the optimization of the subspace rotation matrix.

Here is the end of stdout:

entering main loop
N E dE d eps ncg rms ort
DAV: 1 0.618617185026E+04 0.61862E+04 -0.25280E+05 400 0.104E+03
DAV: 2 0.158034666741E+04 -0.46058E+04 -0.38839E+04 400 0.300E+02
DAV: 3 0.637788802584E+03 -0.94256E+03 -0.83663E+03 400 0.126E+02
DAV: 4 0.515718555323E+03 -0.12207E+03 -0.11643E+03 400 0.486E+01
DAV: 5 0.500147332378E+03 -0.15571E+02 -0.15006E+02 400 0.168E+01
EDWAV: internal error, the gradient is not orthogonal 1 1
2.923204711949180E-004

Here is my INCAR:

SYSTEM = defect

ISTART = 0
ISMEAR = 0
SIGMA = 0.1
LWAVE = .FALSE.
LREAL = Auto
NSW = 100
IBRION = 1
EDIFFG = -0.02

ENCUT = 400
LCHARG = .TRUE.
LVTOT = .TRUE.
LVHAR = .TRUE.

ISPIN = 2

NELM = 200

LHFCALC = .TRUE. ; HFSCREEN = 0.2
ALGO = Damped; TIME = 0.4
PRECFOCK = Normal ! used PRECFOCK = Normal for high quality calculations

NELECT = 498

<span class='smallblacktext'>[ Edited ]</span>

[xamnidar]

Update: I have found that reducing the optimization flag from O3 to O1 solves this problem.

[tfarmstrong]

[quote="xamnidar"]Update: I have found that reducing the optimization flag from O3 to O1 solves this problem.[/quote]

I am getting the same error running vasp on another universities cluster, so I was wondering:

Can the optimization flag be altered simply in the INCAR file or do I have to do a recompile? I do not have rights to compile on their cluster that could be tricky.

Thanks,
Tom

[kayahan]

I had the same problem, but I am working on large band gap semiconductors (~3 eV) and doing B3LYP calculations, so what I did was that I set NBANDS= # of electrons/2 and then used ALGO = All. Basically it should be fine if you strictly follow http://cms.mpi.univie.ac.at/vasp/vasp/T ... tions.html

It gives a message about increasing NBANDS for faster convergence in OUTCAR, but I think it should be ignored in this case, since damped MD (Car-Parrinello) is unstable for large NBANDS values.

I haven't done any calculations for metals or small band gap SC's but my guess is playing with TIME tag should solve the problem.
<span class='smallblacktext'>[ Edited Mon Oct 21 2013, 02:39AM ]</span>