HSE06 calculations not working

[eva_pospisilova]

Dear VASP Community,

I have tried HSE06 calculations with W.cI2, Cu.cF4, WO3 (Pearson symbols aP32, mP16, mP32, tP8), WO3-aP32 example enclosed

HSE06-2.zip

, starting from PAW PBE-relaxed structures from WAVECAR (ISTART = 1).

In the cases of WO3 structures (nonmagnetic), HSE06 did not even started. As for W.cI2 and Cu.cF4, these elemental HSE06 relaxations converged well with identical INCAR file.

What am I doing wrong? Are you able to successfully HSE06-rerelax the file I enclosed?

Eva

[jonathan_lahnsteiner2]

Dear Eva,

I was checking the calculation you sent. When executing the simulation I receive the following output message

Code: Select all

 -----------------------------------------------------------------------------
|                     _     ____    _    _    _____     _                     |
|                    | |   |  _ \  | |  | |  / ____|   | |                    |
|                    | |   | |_) | | |  | | | |  __    | |                    |
|                    |_|   |  _ <  | |  | | | | |_ |   |_|                    |
|                     _    | |_) | | |__| | | |__| |    _                     |
|                    (_)   |____/   \____/   \_____|   (_)                    |
|                                                                             |
|     internal error in: fock_dbl.F  at line: 1440                            |
|                                                                             |
|     internal error in FOCK_FCC: number of k-points incorrect 27 125 2 2     |
|     2                                                                       |
|                                                                             |
|     If you are not a developer, you should not encounter this problem.      |
|     Please submit a bug report.                                             |
|                                                                             |
 -----------------------------------------------------------------------------

This problem can be overcome by removing the tag NKRED from your INCAR file.

All the best Jonathan

[eva_pospisilova]

I still encounter problems with HSE06 runs with binary W-O oxides after NKRED tag has been removed.

I tried to relax several dozens of oxides within HSE06 setup and here I choose sample with the smallest 36-atom periodic unit cell: O2W-oP36, where the calculation failed at the very beginning just after the warning (in output):

"found WAVECAR, reading the header
number of bands has changed, file: 256 present: 136
trying to continue reading WAVECAR, but it might fail
WAVECAR: different cutoff or change in lattice found"

I do not understand, why "different cutoff or change in lattice found", sice I did not change the ENCUT = 550eV and I copied PBE CONTCAR (*.PBE files enclosed) to HSE06 POSCAR (*.HSE files enclosed) and I copied and read also WAVECAR (ISTART = 1), I used the same KPOINTS and POTCAR. I could not enclose the WAVECAR, because it is too large a file for the VASP forum.

HSE06-3.zip

I also do not understand, how it happened that "number of bands has changed".

I arrived at exactly the same failure (before the first electronic step) in dozens of cases with distinct W-O stoichiometries.

What else (except for placing NKRED tag in INCAR) am I doing wrong?

[alex]

Hi Eva,

in your PBE calculation you do not set NPAR, which results in an automatic setting of NPAR to the number of processes. So far no harm done.
For performance reasons, VASP sets NBANDS to the number of processes (256?!), so you end up with 256 bands in your PBE simulation.

NPAR is not set (I'm not even sure if allowed) in HSE. You end up with 136 bands. The lowest number whoch you can divide by 8 and fulfills the requirements of your system.

Hence all the mess.

Cheers,

alex

[eva_pospisilova]

Hi Alex,

could you see any way out? I would like to run HSE from well relaxed PBE sample, but HSE fails at the very beginning. Is there anything I should try?

Eva

[henrique_miranda]

Hi Eva,

Alex has a point that the number of bands changed between PBE and HSE06 because of your parallelization settings changed: in both calculations you use 128 MPI ranks but in HSE06 you have NPAR=8 while in PBE you don't.
In principle, if you set NPAR=8 in both calculations it should work because you will get the same NBANDS.

If this does not work still, then I noticed that you have a slightly different POSCAR for the PBE and HSE run. Is this intentional?
If you want to compute HSE with the HSE lattice parameter then I would suggest calculating the WAVECAR with PBE using the HSE lattice parameters.

Hope this helps!

[alex]

Hello Eva,

please use the same NPAR in both INCARs, or set NBANDS to the highest observed value for starters and report back if it does not work, please.

Best regards,

alex

[eva_pospisilova]

Dear Henrique and Alex,

I have to start HSE06 calculations for all tungsten oxides from already relaxed PBE WAVECAR and structure, because otherwise, it fails at the very beginning. Therefore HSE06 POSCAR (POSCAR.HSE) is identical with PBE CONTCAR (CONTCAR.PBE = POSCAR.HSE) and PBE WAVECAR is copied and read in the subsequent HSE06 calculation (ISTART=1).

I have applied your advice and in half of the cases, it runs (although it performed just a couple of steps with 50 atoms per 2 days on 2 nodes, 128 cores, on Devana supercomputer in Bratislava, it is pretty slow), and in other cases, e.g. in what I enclose

HSE06-4.zip

, it failed anyway.

I have several tens of W-O oxides with different stoichiometries to relax (to compute convex hull). They all have been already PBE relaxed, but it is fairly difficult to proceed further with electronically more precise HSE06 relaxations.

Eva

P.S.: In order to PBE-relax binary W-O oxides with different stoichiometries including more than 50 atoms per periodic cell, I had to change W_pv POTCAR as of 06Nov2007 for W 08Apr2002. With the W_pv 06Nov2007 POTCAR, almost all calculations failed to converge in the standard PBE setup (the same as I enclose here).

[henrique_miranda]

Hi Eva,

If I understand it correctly, you are trying to relax the structures using PBE and then use the relaxed structures with HSE06.
If for some reason this is not working (I think it should work) then you can do one additional step:

1. relax with PBE and copy final CONTAR to POSCAR of step2
2. run SCF with PBE (IBRION=-1) and copy POSCAR and WAVECAR to step 3
3. run HSE06 starting from the WAVECAR

From the files attached, I don't understand the problem you are having, i.e. what exactly do you mean with "it failed anyway".
What is the error you are getting?

As for the calculations being slow, there are a few things you can try:

1. carefully check that if you can reduce ENCUT while retaining the precision of your calculations.
2. change from NPAR parallelism into KPAR parallelism. NPAR means NCORE and NCORE/=1 is not very efficient for HF/Hybrid calculations
3. Try the NKRED tag
   https://www.vasp.at/wiki/index.php/NKRED
   https://www.vasp.at/wiki/index.php/Down ... k_operator
4. Try using ALGO=Normal instead of "ALGO = D ; TIME = 0.3". This uses LFOCKACE=.true. which might be 2 to 4 times faster.

W_pv is a very old POTCAR that is no longer recommended.
Take a look at the current list of recommended potentials here:
https://www.vasp.at/wiki/index.php/Avai ... potentials.

[eva_pospisilova]

Thank you, Henrique.

Yes, correctly, I am trying to relax the structures using PBE and then use the relaxed structures and WAVECAR with HSE06, in order to enable HSE06 relaxation, since it is not possible to run HSE06 with WO3 (non)stoichiometric oxides from the scratch.

I have applied your advice and performed one additional PBE step with IBRION=-1 and reused CONTCAR and WAVECAR in the subsequent HSE06 run. The output was the same: I did not received any error message, only the output, which I already posted:

running 64 mpi-ranks, on 1 nodes
distrk: each k-point on 64 cores, 1 groups
distr: one band on 8 cores, 8 groups
vasp.6.5.0 dev (build Mar 24 2024 20:02:22) complex

POSCAR found type information on POSCAR O W
POSCAR found : 2 types and 46 ions
scaLAPACK will be used
LDA part: xc-table for (Slater+PW92), standard interpolation
found WAVECAR, reading the header
POSCAR, INCAR and KPOINTS ok, starting setup
FFT: planning ... GRIDC
FFT: planning ... GRID_SOFT
FFT: planning ... GRID

(e.g. for already posted stoichiometry O12W11-oC46 or O49W18-mP67, O21W8-oP116, O73W25-mP196)
That is all and I do not have a clue, what is going on.

In some other cases (e.g. O47W17-mP64), HSE06 relaxation interrupted abruptly after several electronic steps without any error message (and without solo ionic step performed).

All the examples I mention here were relaxed with PBE without any problems.

Eva

[henrique_miranda]

I have another suspicion of what might be happening.
There is sometimes a problem in hybrid calculations:

SET_INDPW_FULL: insufficient memory

See for example this thread:
https://ww.vasp.at/forum/viewtopic.php?t=17510

This error message is misleading because the problem is not caused by insufficient memory but by an insufficient memory allocation.
In the last post of the thread above, I suggest a few different ways to try to overcome the problem. Maybe you can try some of those in particular:

1. Don't set NPAR or NCORE in the INCAR file.
2. Turn off symmetries with ISYM=0 (I understand for production runs you will not want to turn off the symmetry but it can help to check what is the origin of this problem).

In any case, it is strange that the code simply stops without you getting any output. That should not happen.