HSE03 stops. Gradient not orthogonal

[sponza]

Hallo,
I am having some trouble in making calculations with hybrids (namely HSE03 on bulk ZnO).
I made three calculations : t2, t3 and t4.
To reduce the number of thinks to check I actually do not start from a preconverged WAVECAR, so the only input files I give are KPOINTS, POTCAR, POSCAR and INCAR.

Here the INCAR file of t2:

Code: Select all

Start?parameters?:
????ISTART?=?0
???ICHARG?=?2
Electronic?relaxation?:

???EDIFF??=?1E-6
???NELM???=?140?;?NELMIN?=?2?;?NELMDL?=?2
???PREC???=?Accurate
???LREAL??=?.FALSE.

DOS?related?values:
???ISMEAR?=?0??;?SIGMA?=?0.05

????LHFCALC?=?.TRUE.??;?HFSCREEN?=?0.2
????ALGO?=?Damped?;?TIME=0.4
????PRECFOCK=?F

The INCAR of t3 coincides with t2 and t4 differs only on the variable TIME=0.2

t2 and t4 uses Pseudopotentials with semicore (20 electrons for Zn), whereas the t3 calculation has 12 electrons for Zn.


What I get is that simulations t2 and t4 stops after some iterations. They are not killed though and the node just stay idle without updating the OUTCAR. Indeed making top in the node, I can see that some node is still working whereas other are not. For example to run t2 I reserved 16 processors and indeed 16 procs were used at the beginning. After 11 selfconsistent cycles the number of processors running reduced to 8 and the OUTCAR stopped being updated. Once I killed the calculation by hand, the standard output reported the following error

Code: Select all

EDWAV:?internal?error,?the?gradient?is?not?orthogonal??????????27???????????1
??-6.69360314135514
?EDWAV:?internal?error,?the?gradient?is?not?orthogonal??????????27???????????1
??-11.9172502538239
?EDWAV:?internal?error,?the?gradient?is?not?orthogonal??????????27???????????1
?-0.246382766126383
?EDWAV:?internal?error,?the?gradient?is?not?orthogonal??????????27???????????1
?-0.520519658375470
?EDWAV:?internal?error,?the?gradient?is?not?orthogonal??????????27???????????1
?-0.551160876860794
?EDWAV:?internal?error,?the?gradient?is?not?orthogonal??????????27???????????1
?-2.080130512674336E-002
?EDWAV:?internal?error,?the?gradient?is?not?orthogonal??????????27???????????1
?-6.868727224090194E-002
?EDWAV:?internal?error,?the?gradient?is?not?orthogonal??????????27???????????1
??-1.04424772690277

I guess that the 8 processors blocked actually encountered some problem.

I launched t4 with TIME=0.2 . The result is essentially the same but it stopped after the 25th iteration. The displayed error is the same.

Finally I run t3, which coincides with t2 but uses pseudos without semicore. Everything worked properly.


Now the questions:
What does this error mean?
In the forum I found that it may be a problem of compilation (here), do you know if there are other factors that may lead to this error?
Is it indeed a problem of pseudopotentials or I should be allowed to use semicore electrons with HSE03?
In principle I would like to compute G0W0 on top of the hybrid calculation, where semicore electrons are of vital importance, so if I can not use semicore with hybrids, how can I overcome this problem?

Thanks a lot for your help!

[sponza]

I have to add that I actually finally managed to perform a calculation with 20 electrons in Zn. That will be called t5 which is identical to t4 except for the fact that I added NPAR=16 in the input (I'm running on 16 processors).
For me this is not a good news because, on the contrary, it puzzles me even more. NPAR = N.procs should be the default, so why the calculation differ from the t4 simulation?
Why this time it didn't stop?

Some one has ideas?

[hipertrofia]

That is odd, because if you are doing a hybrid calculation VASP does not allow to set the NPAR option. Are you sure this last calculation is uses a hybrid?