Si example from Quantum Espresso fails with Seg Fault

[jmullen]

I am stumped. After having resolved the issue I posted over in the P2Y forum, I have moved to running the GW calculation on Si as taken from Quantum Espresso. I am using QE 4.0.4 and 4.0.5 for testing (and have tried 4.1 for completeness) and all generate the same error. NOTE: each time, yambo has been compiled against the IOTK in the espresso version directory.

I am attaching input and output files in a tarball, but the short version is, running on one processor
p2y
yambo -i
yambo
yambo -x -p p -g g -V 2
yambo

with all the defaults taken except for plane waves because, at this point, I need it to complete before running convergences.
The error generated is a segmentation fault and core dump when loading, or immediately after loading, the WF's.

<---> [WF loader] Wfs (re)loading |####################| [100%] --(E) --(X)[medusa:25722] *** Process received signal ***
[medusa:25722] Signal: Segmentation fault (11)
[medusa:25722] Signal code: Address not mapped (1)


I am confused because I do not see this is other systems I am running at the moment. I am sure that, like my problem with P2Y, it is trivial and probably a typo, but I cannot see it. Any help is appreciated.

Thanks

[Conor Hogan]

Dear Jeff,
Yes, you are still doing some odd things with PW... !
Make sure you run things in two steps: scf and non-scf. The nscf ("bands") should read the potential from the scf, and then create the bandstructure for both occupied and unoccupied states. I think in your case yambo failed because you have no empty states! (8 electrons, 4 bands). Ideally we should have a check on this to make the code die peacefully.

My si_scf.in is:
&control
calculation='scf'
verbosity = 'high',
prefix='si_bulk-out'
pseudo_dir = '/gpfs/scratch/userinfm/cne0fm2h/PSEUDO/UPF',
wfcdir='./tmp'
/
&system
ibrav= 2, celldm(1)= 10.2, nat= 2, ntyp= 1,
ecutwfc =18.0
/
&electrons
conv_thr = 1.0d-8
mixing_beta = 0.7
/
ATOMIC_SPECIES
Si 28.086 Si.vbc.UPF
ATOMIC_POSITIONS
Si 0.00 0.00 0.00
Si 0.25 0.25 0.25
K_POINTS {automatic}
4 4 4 1 1 1


and my si_nscf.in is:

&control
calculation='bands'
verbosity = 'high',
pseudo_dir = '/gpfs/scratch/userinfm/cne0fm2h/PSEUDO/UPF',
wfcdir='./tmp'
prefix='si_bulk-out'
wf_collect = .true.
/
&system
ibrav= 2, celldm(1) =10.20, nat= 2, ntyp= 1,
ecutwfc =18.0, nbnd = 20,
force_symmorphic = .true.
/
&electrons
diago_full_acc = .true.
diago_thr_init = 1.0d-6
/
ATOMIC_SPECIES
Si 28.086 Si.vbc.UPF
ATOMIC_POSITIONS
Si 0.00 0.00 0.00
Si 0.25 0.25 0.25
K_POINTS {automatic}
4 4 4 1 1 1


Note the use and position of nbnd, force_symmorphic, wf_collect, diago_full_acc, diago_thr_init.

Hope it helps!
Conor

[jmullen]

Thanks,

I will do this. As for empty states, I thought that Quantum espresso did occupied + 4 for scf calculation and this seems to have worked fine for my other test systems. Apparently, I am doing this wrong for Yambo. I will recheck my other systems using the NSCF calculation.

Thanks