Failure in GW0 calculations

[giacomo giorgi]

Dear Sir

I am nowadays using VASP 5.2 for GW0 on small systems (8 atoms)

In particular I am studying systems in which GaAs is replaced with Ge.
Pseudopotentials include d orbitals in the semicore ONLY for Ga (those of Ge are in the core)

I did the initial DFT calculation.

Then I used this INCAR for the subsequent GW0 calculation

SYSTEM = GaAs_PAW_LDA
ISMEAR = 0
SIGMA = 0.05
LOPTICS = .TRUE.
ALGO = GW0
ENCUTGW=90
ENCUT=283
ENAUG=491
NBANDS=150
ISYM=2
NOMEGA = 48
NELM = 4

please note that the system lattice parameter are a=b ≠ c.

If I give 4x4x4 Gamma centered k-point sampling I get the following segmentation fault message after the first iteration

Everything is OK if differently I give 3x3x3 Gamma centered k-point sampling....


calculate QP shifts <psi_nk| G(iteration)W_0 |psi_nk>: iteration 1
calculate exact exchange contribution
[vespasianus:01652] *** Process received signal ***
[vespasianus:01652] Signal: Segmentation fault (11)
[vespasianus:01652] Signal code: Address not mapped (1)
[vespasianus:01652] Failing at address: 0x2aa3e00010
[vespasianus:01652] [ 0] /lib64/tls/libpthread.so.0 [0x399400c5b0]
[vespasianus:01652] [ 1] /opt/intel/cmkl/9.1.023/lib/em64t/libmkl_p4n.so(mkl_blas_zcopy+0x203) [0x2a96359987]
[vespasianus:01652] *** End of error message ***
[vespasianus:01649] *** Process received signal ***
[vespasianus:01649] Signal: Segmentation fault (11)
[vespasianus:01649] Signal code: Address not mapped (1)
[vespasianus:01649] Failing at address: 0x2aa3e00010
[vespasianus:01650] *** Process received signal ***
[vespasianus:01649] [ 0] /lib64/tls/libpthread.so.0 [0x399400c5b0]
[vespasianus:01649] [ 1] /opt/intel/cmkl/9.1.023/lib/em64t/libmkl_p4n.so(mkl_blas_zcopy+0x203) [0x2a96359987]
[vespasianus:01649] *** End of error message ***
[vespasianus:01650] Signal: Segmentation fault (11)
[vespasianus:01650] Signal code: Address not mapped (1)
[vespasianus:01650] Failing at address: 0x2aa3e56010
[vespasianus:01650] [ 0] /lib64/tls/libpthread.so.0 [0x399400c5b0]
[vespasianus:01650] [ 1] /opt/intel/cmkl/9.1.023/lib/em64t/libmkl_p4n.so(mkl_blas_zcopy+0x203) [0x2a96359987]
[vespasianus:01650] *** End of error message ***
[vespasianus:01651] *** Process received signal ***
[vespasianus:01651] Signal: Segmentation fault (11)
[vespasianus:01651] Signal code: Address not mapped (1)
[vespasianus:01651] Failing at address: 0x2aa3ddb010
[vespasianus:01651] [ 0] /lib64/tls/libpthread.so.0 [0x399400c5b0]
[vespasianus:01651] [ 1] /opt/intel/cmkl/9.1.023/lib/em64t/libmkl_p4n.so(mkl_blas_zcopy+0x203) [0x2a96359987]
[vespasianus:01651] *** End of error message ***
--------------------------------------------------------------------------
mpirun noticed that process rank 3 with PID 1652 on node vespasianus exited on signal 11 (Segmentation fault).
--------------------------------------------------------------------------
allocating 12 responsefunctions rank= 360
allocating 48 responsefunctions rank= 360
Doing 12 frequencies in blocks of 48
NQ= 1 0.0000 0.0000 0.0000,




I can not think that is a memory requirement failure. I am working with vasp mpi on 4 nodes biprocessor. I am treating a 8 atom cell with only 44 electrons....

Moreover I have no problems in calculating at GW0 GaAs with d electrons of Ga in the semicore that means 72 electrons, and also with a 6x6x6 (Gamma centered) sampling

Any suggestions or idea for fixing this problem?
Thanks in advance

G

[dpj]

Hi Giacomo,

We have a similar problem - GW0 calculations that terminate after the first iteration. Simlar to your experience, for some systems we find no problems and for other systems the code stops with a segmentation fault after one iteration. We have tried the calculations on more than one computer system with the same result.

Did you manage to solve the problem?

Thank you

Daniel