HF type parallel calculation fails with a large structure

[taizo]

Dear All,

I'm calculating TiO2 surface slab, which has 4 layers or 52 atoms.
When I run this parallel in one node, it stops before first electronic step.
However, if I reduce the slab size to 1 layer, it works. Also, serial calculation for the 4 layers works.
Is this known problem, or did I do something wrong?

Here is my INCAR and environment:

INCAR:

ISMEAR = 0
SIGMA = 0.05
LREAL=Auto
LHFCALC=.TRUE.
HFSCREEN=0.2
ALGO=All
TIME=0.4
PRECFOCK=L


vasp: 5.2.11
CPU:Xeon 5570 Quad * 2
Memory:70GB
MPI: Intel MPI 3.2.1
FC: ifort 10.0.023
BLAS:GotoBLAS2
FFTW:fftw3.2.2

This is the error message obtained by -gdb option (with vasp recompiled with -traceback and check bounds options).

0: running on 8 nodes
1-7: forrtl: severe (408): fort: (2): Subscript #1 of the array TBLI has value 1 which is greater than the upper bound of 0
1-7:
1-7: Image PC Routine Line Source
1-7: vasp 00000000017F8B41 Unknown Unknown Unknown
1-7: vasp 00000000017F7B15 Unknown Unknown Unknown
1-7: vasp 00000000017A432A Unknown Unknown Unknown
1-7: vasp 0000000001757CF2 Unknown Unknown Unknown
0: distr: one band on 1 nodes, 8 groups
1-7: vasp 0000000001756C0E Unknown Unknown Unknown
1-7: vasp 000000000154C614 mapset_ 311 fftmpi_map.f90
1-7: vasp 00000000004210DA MAIN__ 1786 main.f90
1-7: vasp 0000000000406A6C Unknown Unknown Unknown
1-7: libc.so.6 00000031D1A1EC5D Unknown Unknown Unknown
1-7: vasp 0000000000406969 Unknown Unknown Unknown
0: vasp.5.2.11 23Dec10 complex

Best Regards,

Taizo Shibuya

[taizo]

I recompiled vasp.5.2.11 with -heap-arrays option in OFLAG in makefile. This option seemed to cure the problem. Thank you!

Taizo Shibuya

[admin]

you are correct, vasp.5.2 loads all large arrays from stack. therefore, if your run crashes for a larger system, please first of all check your stack size:
ulimit -s
should return "unlimited"
if it does not, please set it explicitely in your .bashrc (ulimit -s unlimited)