Unspecified crashes with ALGO=All and METAGGA (with tau) in final diagonalization

[marvin_poul]

Dear Forum,

I experience reproducible crashes without an error message for some small cells of Mg that I perform static calculations on. Here's a sample INCAR

Code: Select all

SYSTEM=structure_15 #jobname
PREC=Accurate
ALGO=A
LREAL=.FALSE.
LWAVE=.FALSE.
LORBIT=0
ENCUT=550
NBANDS=13
KSPACING=0.1
EDIFF=1e-06
LCHARG=.FALSE.
NCORE=2
KPAR=10
METAGGA=R2SCAN
LASPH=.TRUE.

I've set ALGO=A as recommended by the wiki.

The calculations run for a while and then crash with this message

Code: Select all

 final diagonalization occupied
Abort(1) on node 2 (rank 2 in comm 0): application called MPI_Abort(MPI_COMM_WORLD, 1) - process 2
Abort(1) on node 3 (rank 3 in comm 0): application called MPI_Abort(MPI_COMM_WORLD, 1) - process 3

It does not appear to be a memory issue, judging by the initial memory consumption reported by VASP. I also do not see a top spike in top just prior to the crash.

I can reproduce this behavior with version 6.4.0 and 6.5.1 and both the R2SCAN and LAK meta ggas.

The following variations fix the problem:
1. changing ALGO=Normal
2. changing to plain GGA

It seems the problem is confined to kinetic density parametrized meta GGAs, or at least the problem disappears under R2SCANL and OFR2 functionals.

I've attached three runs that reproduce the problem. I've tested the variations only on the structure_15.tar.gz run, but I would assume it should not matter.
I can work around the issue by changing ALGO, but a fix or an error message would nice to have in the future.

Best Regards,
Marvin

[marvin_poul]

When I remove parallelization over k points via KPAR=1, I obtain this error

Code: Select all

 -----------------------------------------------------------------------------
|                                                                             |
|     EEEEEEE  RRRRRR   RRRRRR   OOOOOOO  RRRRRR      ###     ###     ###     |
|     E        R     R  R     R  O     O  R     R     ###     ###     ###     |
|     E        R     R  R     R  O     O  R     R     ###     ###     ###     |
|     EEEEE    RRRRRR   RRRRRR   O     O  RRRRRR       #       #       #      |
|     E        R   R    R   R    O     O  R   R                               |
|     E        R    R   R    R   O     O  R    R      ###     ###     ###     |
|     EEEEEEE  R     R  R     R  OOOOOOO  R     R     ###     ###     ###     |
|                                                                             |
|     ERROR in subspace rotation PDSYEVX/ PZHEEVX: I2,IFAIL= 1 2              |
|                                                                             |
|       ---->  I REFUSE TO CONTINUE WITH THIS SICK JOB ... BYE!!! <----       |
|                                                                             |
 -----------------------------------------------------------------------------

Maybe that's just getting swallowed somewhere in the parallel run?

Marvin

[christopher_sheldon1]

Hi Marvin,

Thank you for your question. I have been able to repeat your first error for structures 15, 280, and 829. Setting KPAR=1 solved the problem for me for structure 829, which implies that this is an issue with the k-point mesh. I found that all calculations converged with a 6x6x6 k-point mesh and that issues ("ERROR in subspace rotation PDSYEVX/ PZHEEVX") only appear for the larger meshes that you have used (defined with KSPACING=0.1 or explicitly in KPOINTS). I have not found a reason for this crash yet but will continue to look into it.

Best wishes,

Chris