MPI Application rank 3 killed before MPI_Finalize() with signal 11

[vasp16888]

Dear vasp master and user:
I am running vasp 4.6.35, mpi calculation.
I am calculation bcc iron system, and I deform bcc iron by 1% along 100 dirction. Because I add spin_orbit coupling, and I set SAXIS=0 0 1, MAGMOM=0 0 1. Other parameters were tested for convergence.
I set KPOINTS:
===========
Auto
0
M
27 27 27
0 0 0
===========
kpoints change from 9x9x9 to 31x31x31.

But when kpoint 27x27x27, it stoped at the begining of the calculation, errors are as follows:

===================
running on 12 nodes
distr: one band on 3 nodes, 4 groups
vasp.4.6.35 3Apr08 complex
POSCAR found : 1 types and 1 ions
scaLAPACK will be used
LDA part: xc-table for Pade appr. of Perdew

-----------------------------------------------------------------------------
| |
| W W AA RRRRR N N II N N GGGG !!! |
| W W A A R R NN N II NN N G G !!! |
| W W A A R R N N N II N N N G !!! |
| W WW W AAAAAA RRRRR N N N II N N N G GGG ! |
| WW WW A A R R N NN II N NN G G |
| W W A A R R N N II N N GGGG !!! |
| |
| Full k-point grid generated |
| Inversion symmetry is not applied |
| |
MPI Application rank 3 killed before MPI_Finalize() with signal 11

-----------------------------------------------------------------------------

MPI Application rank 3 killed before MPI_Finalize() with signal 11
=============

the OUTCAR stoped at the kpoints sampling:
=============
0.000000 -0.074074 0.222222 2.000000
0.000000 -0.037037 0.259259 2.000000
0.000000 0.000000 0.296296 1.000000
-0.072622 0.000000 0.148148 2.000000
-0.036311 -0.037037 0.148148 4.000000
-0.036311 0.000000 0.185185 2.000000
0.000000 -0.074074 0.148148 2.000000
0.000000 -0.037037 0.185185 2.000000
0.000000 0.000000 0.222222 1.000000
-0.036311 0.000000 0.111111 2.000000
0.000000 -0.037037 0.111111 2.000000
0.000000 0.000000 0.148148 1.000000
0.000000 0.000000 0.074074 1.000000

=============
I checked the IBZKPT, it had 5216 kpoints in IBZ.

My questions: Is there any limitation of the kpoints in IBZ ? If it is, what's the limitation of kpoints in IBZ ?
If not, please suggest what the reasno for this would be ?

Thanks a lot:)

[spakinfo]

Can you provide further info
1) Have you restarted your calculation form a previous job (without LS)
2) What are your settings in INCAR (if possible paste all contents)

[alex]

Too less memory problem? 5000 k-points, quite a bunch.

[vasp16888]

[quote author============
SYSTEM = Bcc_Fe
ICHARG = 11
ISPIN = 2 # spin : 1-off 2-on

ENCUT = 550 # cut-off energy
PREC = H # precision : medium, high low
NELM = 1000 # maximal ELM
EDIFF = 1E-07 # stopping-criterion for ELM
NELMDL = -5 # No. of inconsistent ELM
NELMIN = 4 # minimal ELM

ISMEAR = -5 # smearing method : -5-tet -1-fermi 0-gaus 1-MP
SIGMA = 0.1 # smearing factor in eV


ALGO = N # algorithm
NSIM = 4 # parameter used in RMM-DIIS
LREAL = .FALSE. # space where projection will be done
VOSKOWN = 1 # gga related

AMIX = 0.2
BMIX = 0.0001
AMIX_MAG = 0.8
BMIX_MAG = 0.0001
LMAXMIX = 4

LORBMOM = .TRUE.
LNONCOLLINEAR= .TRUE.
GGA_COMPAT = .FALSE.
LSORBIT = .TRUE.
SAXIS = 0 0 1 (magnetic direction)
MAGMOM = 0 0 1
NBANDS = 48
ISYM = 2
============

POSCAR: (deform in x direction)
============
BCC_Fe
2.8389
-0.510 0.500 0.500
0.510 -0.500 0.500
0.510 0.500 -0.500
1
Direct
0.0000000000000000 0.0000000000000000 0.0000000000000000
============

KPOINTS
============
Automatic generation
0
Monkhorst-Pack
27 27 27
0 0 0
============

when k is from 9x9x9 to 25x25x25, no problem. The kpt in IBZ is less than 5000.
But when k = 27x27x27,
it stoped at the begining, see the error information above.
OUTCAR stoped at the k sampling.

I also checked the memory, it's just use 8 GB, we have 16 GB for each nodes, it's enough.

Any suggestions, thanks:)
<span class='smallblacktext'>[ Edited Fri Jun 11 2010, 01:15AM ]</span>

[vasp16888]

[quote="alex"]Too less memory problem? 5000 k-points, quite a bunch.[/quote]


I also checked the memory, it's just use 8 GB, we have 16 GB for each nodes, it's enough.

Looking forward to your suggestions, thanks:)

[spakinfo]

I see. Then I guess you are having this problem because of you kpoints. Note that when you do non-collinear calculation, you need to remove all the symmetry constraints (ISYM=0) and hence include all the kpoins in BZ in you calculation. From your first post I can see that most of the KPOINTS in OUTCAR have weighting factors more thatn 1, which means still you have not fully turned off symmetry (you have set ISYM to 2!).
Do as follows:

1) do a quick calc with following inputs:
ISYM=0
LSORBIT=.TRUE.
LNONCOLLINEAR= .TRUE.
SAXIS = 0 0 1 (magnetic direction)

2) As soon as have IBZCAR, you can stop the above calculation. Copy IBZCAR to KPOINTS and do a self-consistent calculation without above LS tags, namely:

============
SYSTEM = Bcc_Fe
ISTART=0
ICHARG = 2
ISPIN = 2

ENCUT = 550 # cut-off energy
PREC = H # precision : medium, high low
EDIFF = 1E-05 # stopping-criterion for ELM

ISMEAR = 1
VOSKOWN = 1 # gga related
LMAXMIX = 4
NBANDS=24
============

3) next copy CONTCAR to POSCAR, and finally do a LS calculation with following INCAR

============
SYSTEM = Bcc_Fe
ISTART=1
ICHARG = 11
ISPIN = 2

ENCUT = 550 # cut-off energy
PREC = H # precision : medium, high low
EDIFF = 1E-05 # stopping-criterion for ELM

ISMEAR = 1
VOSKOWN = 1 # gga related
LMAXMIX = 4

#=============LS TAGS
LNONCOLLINEAR= .TRUE.
GGA_COMPAT = .FALSE.
LSORBIT = .TRUE.
SAXIS = 0 0 1
NBANDS=48
============

I think with above setting you can finish your LS calculations without any problem. Few additional points.

2-You have put too many tags in your INCAR, try to stick with VASP defaul values, unless you have to change them.

3-make sure that the number of bands (NBANDS) in your LS calculation is EXACTLY two times more than BNADS in your self-consistent calculation.

Good luck

[vasp16888]

[quote author=0) and hence include all the kpoins in BZ in you calculation. From your first post I can see that most of the KPOINTS in OUTCAR have weighting factors more thatn 1, which means still you have not fully turned off symmetry (you have set ISYM to 2!).
Do as follows:

1) do a quick calc with following inputs:
ISYM=0
LSORBIT=.TRUE.
LNONCOLLINEAR= .TRUE.
SAXIS = 0 0 1 (magnetic direction)

2) As soon as have IBZCAR, you can stop the above calculation. Copy IBZCAR to KPOINTS and do a self-consistent calculation without above LS tags, namely:

============
SYSTEM = Bcc_Fe
ISTART=0
ICHARG = 2
ISPIN = 2

ENCUT = 550 # cut-off energy
PREC = H # precision : medium, high low
EDIFF = 1E-05 # stopping-criterion for ELM

ISMEAR = 1
VOSKOWN = 1 # gga related
LMAXMIX = 4
NBANDS=24
============

3) next copy CONTCAR to POSCAR, and finally do a LS calculation with following INCAR

============
SYSTEM = Bcc_Fe
ISTART=1
ICHARG = 11
ISPIN = 2

ENCUT = 550 # cut-off energy
PREC = H # precision : medium, high low
EDIFF = 1E-05 # stopping-criterion for ELM

ISMEAR = 1
VOSKOWN = 1 # gga related
LMAXMIX = 4

#=============LS TAGS
LNONCOLLINEAR= .TRUE.
GGA_COMPAT = .FALSE.
LSORBIT = .TRUE.
SAXIS = 0 0 1
NBANDS=48
============

I think with above setting you can finish your LS calculations without any problem. Few additional points.

2-You have put too many tags in your INCAR, try to stick with VASP defaul values, unless you have to change them.

3-make sure that the number of bands (NBANDS) in your LS calculation is EXACTLY two times more than BNADS in your self-consistent calculation.

Good luck [/quote]why don't you just set ISYM=0 and do the 2nd step directly????[/color]

I also don't know why in your 3 step copy CONTCAR to POSCAR since there is no optimization ?

[spakinfo]

When doing non-collinear calculation, you have to remove all the symmetry constraints (i.e. ISYM=0). The reason is due to the fact that the non-collinearity of magnetization may result in a break in band degeneracy so that e(k) will be no longer equal to e(-k).

Therefore all the kpoints in the BZ has to be included in order to perform a real non-collinear calculations.

As why I asked you to do step #1 and #2 before doing the final non-collinear calculation is because if you just simply put ISYM=0 without spin-orbit tags, VASP will still keep the inversion symmetry while sampling the BZ. You can see that in this case some of the generated kpoints in IBZCAR have weighting factors larger than 1.0. When you do the first step, you ask the program to lift all the possible symmetries including inversion one. You just need the IBZCAR from that step. If you check the new IBZCAR, you will quickly notice that all the kpoints have same weights (1).

The aim of the second step is to produce a preconverged spin-polarized charge and wavefunction for your non-collinear calculation. This step can save your computational time considerably.

As for your last question, I supposed that in step 2 you do structural optimization. If no, then you can safely start your non-collinear calculation form POSCAR of step 2.

Good luck

<span class='smallblacktext'>[ Edited Sat Jun 19 2010, 08:44PM ]</span>

[panda]

"I also checked the memory, it's just use 8 GB, we have 16 GB for each nodes, it's enough." doesn't really mean anything - do you know how the memory is being distributed? What about the array size? Stack size? I have had similar problems and resolved them by c running on fewer nodes with more memory b/c I was having array, stack size, paging, etc. issues. Unless you read every line of the source code you are not going to know exactly how each subroutine is being implemented with MPI - better to do trial and error for each case or read the specific source code for that case - at any rate going to fewer processors with more memory has helped me resolve similar issues - don't know if this will help you, but good luck!