ABINIT Discussion Forums

Hello,

We are trying to compile abinit in parallel for a class, but we are not very familiar with abinit ourselves.

The serial version works fine, but in parallel we get:

getng.F90:431:BUG
The second dimension of the FFT grid, ngfft(2), should be a multiple of the number of processors for the FFT, nproc_fft. However, ngfft(2)= 20 and nproc_fft= 3
Action : contact ABINIT group.

Across 12 nodes, we get files dx_LOG_0001 through 0011 (not 0012, if this makes any difference) and dx_status_P-0001 through 0011.

The input file (below) is the same one that was used last year in class on abinit 5.7.3 in parallel on 4 nodes, 1 core each.
Is there anything about the input file that we should change for version 7.2.2?

Thank you,
Tam
-----------------------------------------------------------------------------------------
# Crystalline silicon : computation of the total energy
# Convergence with respect to the number of k points.

ndtset 4
#Definition of the k-point grids
kptopt 1 # Option for the automatic generation of k points, taking
# into account the symmetry
nshiftk 4
shiftk 0.5 0.5 0.5 # These shifts will be the same for all grids
0.5 0.0 0.0
0.0 0.5 0.0
0.0 0.0 0.5

ngkpt1 2 2 2 # Definition of the different grids
ngkpt2 4 4 4
ngkpt3 6 6 6
ngkpt4 8 8 8

getwfk -1 # This is to speed up the calculation, by restarting
# from previous wavefunctions, transferred from the old
# to the new k-points.



#Definition of the unit cell
acell 3*10.18 # This is equivalent to 10.18 10.18 10.18
rprim 0.0 0.5 0.5 # FCC primitive vectors (to be scaled by acell)
0.5 0.0 0.5
0.5 0.5 0.0

#Definition of the atom types
ntypat 1 # There is only one type of atom
znucl 14 # The keyword "znucl" refers to the atomic number of the
# possible type(s) of atom. The pseudopotential(s)
# mentioned in the "files" file must correspond
# to the type(s) of atom. Here, the only type is Silicon.


#Definition of the atoms
natom 2 # There are two atoms
typat 1 1 # They both are of type 1, that is, Silicon.
xred # This keyword indicate that the location of the atoms
# will follow, one triplet of number for each atom
0.0 0.0 0.0 # Triplet giving the REDUCED coordinate of atom 1.
1/4 1/4 1/4 # Triplet giving the REDUCED coordinate of atom 2.

#Definition of the planewave basis set
ecut 8.0 # Maximal kinetic energy cut-off, in Hartree

#Definition of the SCF procedure
nstep 10 # Maximal number of SCF cycles
toldfe 1.0d-6 # Will stop when, twice in a row, the difference
# between two consecutive evaluations of total energy
# differ by less than toldfe (in Hartree)
diemac 12.0 # Although this is not mandatory, it is worth to
# precondition the SCF cycle. The model dielectric
# function used as the standard preconditioner
# is described in the "dielng" input variable section.
# Here, we follow the prescription for bulk silicon.

Hello Tam

It's probably because abinit has automatically taken care of the parallelization parameters and has set NPFFT to 3, which is not a multiple of NGFFT(2)=20. In abinit 5.x, abinit would parallelize only on kpoints, while the 7.2.2 version activates the 3-level parallelization scheme by default. You have several options:

1. Change the number of cores you're running abinit with, hoping for a chance that you'll get a distribution that works
2. Configure yourself the parallelization parameters by setting NPFFT, NPKPT, and NPBAND "by hand", using for instance NPFFT=1.
3. Disable paral_kgb by setting paral_kgb=0. You'll then parallelize on kpoints only, which will avoid any issue with the number of cpus dedicated to the FFT grid.

If I were you, I'd use option 3 first. If you have a large system (lot of bands, few kpoints) then go with option 2 and set NPFFT=1, NPKPT = NSPPOL, and NPBAND = multiple of NBANDS.

Hope this helps
Boris

Hello Boris,

Thank you very much! Option 3 worked perfectly, and we will forward this information to the class so the students are informed about 1&2 as well.

Thanks again,
Tam

Great!

You should mark the problem as solved. It should be available somewhere in your first post.

Boris