NEB failure

[johnbrehm]

I am running an NEB.

A. When I DO include a angdeg_lastimg and a acell_lastimg, the program just stops with no error following the start of the 2/7 image on the first iteration. Literally, there is no listing of an error in the log or out file. On some machines I do get a core dump.

B. When I do NOT include angdeg_lastimg but DO include acell_lastimg, the program bombs with this error:
--- !BUG
src_file: m_fftcore.F90
src_line: 3440
mpi_rank: 0
message: |
The variables ikg, mkmem, and mpw must satisfy ikg<=(mkmem-1)*mpw,
while the arguments of the routine are
ikg =5192, mkmem =4, mpw =1728
Probable cause: Known error in invars1 for parallel spin-polarized case.
Temporary solution: Change the number of parallel processes.

So? I think I tried this suggestion correctly a couple of times.. got all the way down to 1 processor. But still got same error.

C. When I do NOT include angdeg_lastimg and do NOT include acell_lastimg, the program runs fine.

HOWEVER: I NEED to define the final image and it DOES have a specified acell and angdeg different from the initial image.

What is the solution to this problem?
I have 32 kpoints and 160 bands so I use 512 processors == npband 32 * npkpt 16.

Here is my input file (without the xred coord lines):
ntimimage 28
nimage 7
imgmov 5 #NEB
neb_algo 1 #Henkelsman
#neb_spring 0.05 0.05 #default why? who TF knows
mep_solver 1

chksymbreak 0
fxcartfactor 10.0
tolimg 1.d-3
istatimg 1 # need the energies -- a 0 gives no energies

#dynimage 0 6*1
#nsppol 1
#prtwf 0
iscf 17

# optcell 2
# ionmov 2
nstep 500
ecutsm 0.5
toldfe 1.0d-3

#need van der waals
vdw_xc 7

ixc 11
ecut 25.0
diemac 9.0
diemix 0.6
dilatmx 1.1
strprecon 0.5

chkprim 0

# kpoints

kptopt 1
ngkpt 4 4 4
nshiftk 1
shiftk 0.5 0.5 0.5

# computational optimization
#wfoptalg 4
#fft_opt_lob 2
#fftalg 401
paral_kgb 1
npband 32
npkpt 16

znucl [this info not needed for this help-discussion]
nband 160
natom [this info not needed for this help-discussion]
ntypat [this info not needed for this help-discussion]
typat [this info not needed for this help-discussion]

angdeg 9.0E+01 9.9E+01 9.0E+01
acell 1.155E+01 2.00E+01 2.474E+01

xred
........
(etc.)
........

angdeg_lastimg 8.9E+01 9.8E+01 8.9E+01
acell_lastimg 1.156E+01 2.01E+01 2.49E+01

xred_lastimg
...........
(etc.)
............

[johnbrehm]

adding some more details to the problem:

I grepped for mkmem in the log file on one of my attempts.
I get a mkmem== 5.

mpi_setup: mkmem undefined in the input file.Use default mkmem = nkpt
mpi_setup: With nkpt_me=5 and mkmem = 68, ground state wf handled in core.
Resetting mkmem to nkpt_me to save memory space.
- mband = 160 mffmem = 1 mkmem = 5
P mkmem 5

Then in the __ABI_MPIABORTFILE__ I see it has been changed to 3:

--- !BUG
src_file: m_fftcore.F90
src_line: 3440
mpi_rank: 433
message: |
The variables ikg, mkmem, and mpw must satisfy ikg<=(mkmem-1)*mpw,
while the arguments of the routine are
ikg =2776, mkmem =3, mpw =1387
Probable cause: Known error in invars1 for parallel spin-polarized case.
Temporary solution: Change the number of parallel processes.

[torrent]

Dear johnbrehm,

I think the NEB with variable cell is not implemented (or, at least buggy). Im pretty sure you'll not succeed in running it.
This should be forbidden by the code but apparently it's not.

Sorry for this.
There is a hope for the future: one of the developer is currently working on that subject...

Regards