The ABINIT Users Mailing List ( CLOSED )

[Geoffrey Stenuit <geoffrey.stenuit@tyndall.ie>]

Dear Francois and Guillaume,

Thank you for your quick reply and advices.
To my understand (please correct my if i am wrong), i though that npband 
has to be a multiple of nband and npfft  a multiple of  mgfft. In my 
case, since i impose npfft=1 and npband=2 (while nband=132), these 
conditions are fullfilled. In addition, the density file was generated 
without any parallelisation over the bands and fft grid.
When npband=2, nband=130, (2 cpus were then requested for this job, so 
equal to npfft*npband) i got the correct eigenvalues with a 
computational time 1.6 faster then in the serial mode. So the scaling 
with a factor 2 was almost achieved. The same application with nband=132 
gives "NaN" for the eigenvalues.
I also tried to parallelize the job with the following flags:
wfoptalg  1
nbdblock  2

but the job crashed with the following error:
0 - MPI_GROUP_INCL : Invalid rank in array: ranks[1] = 2 not between 0 and 2
[0]  Aborting program !
[0] Aborting program!

Do you have an idea about the origin of the problem ?
Regards,

Joe

Francois Bottin wrote:

> Dear Joe,
> As Guillaume said, this part is under development. Some 
> functionalities of Abinit are not yet available in the framework of 
> BandFFT as the Non-self-consistent calculations (thank you Guillaume 
> for this point), but also I/O, spin polarization... It remains a lot 
> of work. You can fix one of these points if you want :-) !
> I don't really understand your problem. You use a number of band 
> processors (npband) higher than the number of valence bands in your 
> system (nband). Note that, if it is the case, the number of band and 
> FFT processors (npband and npfft) HAVE TO BE COMMENSURATE with the 
> number of bands and FFT planes in each direction (nband and ngfft).
> Regards
> Francois
>
> Geoffrey Stenuit wrote:
>
> > Dear Abinit users,
> >
> > Running abinit-5.2.4 compiled with pathscale (pathf90) on AMD Opteron 
> > (64bit), i experience some instabilities with the calculations of the 
> > band structure at k=0 when the npfft or npband is different from 1 
> > and wfoptalg set to 4. In particular, i found that the eigenvalues 
> > become "NaN" when the number of band that i request at this Gamma 
> > point is larger than the number of Valence band + 2 (i.e 128+2=130 in 
> > my GaAsN 64-atom supercell). On the other hand, when i request nband 
> > 129 or 130, the programm is giving the correct eigenvalues (but i 
> > need to analyse more than 130 bands).
> > Do you think it is a bug in abinit, or an error when i compiled it, or ?
> >
> > Please find in attachement the input file i submitted.
> >
> > Thank you for your help,
> >
> > Regards,
> >
> > Joe
> >
> >
> > ------------------------------------------------------------------------
> >
> > # Convergence with respect to the number of k points.
> >
> > chkprim 0
> > # mgfft = 96 => nbr of processors: 1 or 2, 3, 4, 6, 8, 12, 16, 24,... 
> > ,48,96 ### Parallel FLAGS:
> > fftalg 401
> > wfoptalg 4 fft_opt_lob 2
> > npband 2
> > npfft 1 ## mgfft=40 => 1,2,4,5,10,20 and 40 are OK
> > iprcch 0
> >
> >
> > #Definition of the unit cell
> > #Definition of the unit cell
> > acell 3*2.1203311987E+01          # This is equivalent to
> > rprim  1.0  0.0  0.0   # FCC primitive vectors (to be scaled by acell)
> >       0.0  1.0  0.0
> >       0.0  0.0  1.0
> >
> > xcart              # This keyword indicate that the location of the 
> > atoms
> > -1.74374822972885E-19  1.80698275312510E-19  1.09334001556395E-19
> > -1.74374822972885E-19  1.80698275312510E-19  1.06016559933404E+01
> > -1.74374822972885E-19  1.06016559933404E+01  1.09334001556395E-19
> > -1.74374822972885E-19  1.06016559933404E+01  1.06016559933404E+01
> >  1.06016559933404E+01  1.80698275312510E-19  1.09334001556395E-19
> >  1.06016559933404E+01  1.80698275312510E-19  1.06016559933404E+01
> >  1.06016559933404E+01  1.06016559933404E+01  1.09334001556395E-19
> >  1.06016559933404E+01  1.06016559933404E+01  1.06016559933404E+01
> >  9.68225628470105E-03  5.16468510169365E+00  5.16468510169366E+00
> > -9.68225628470105E-03  5.16468510169365E+00  1.60386268849871E+01
> > -9.68225628470105E-03  1.60386268849870E+01  5.16468510169366E+00
> >  9.68225628470105E-03  1.60386268849870E+01  1.60386268849871E+01
> >  1.06013045001119E+01  5.29123892677001E+00  5.29123892677001E+00
> >  1.06020074865687E+01  5.29123892677001E+00  1.59120730599107E+01
> >  1.06020074865687E+01  1.59120730599107E+01  5.29123892677001E+00
> >  1.06013045001119E+01  1.59120730599107E+01  1.59120730599107E+01
> >  5.16468510169366E+00  9.68225628470109E-03  5.16468510169366E+00
> >  5.16468510169366E+00 -9.68225628470105E-03  1.60386268849871E+01
> >  5.29123892677001E+00  1.06013045001119E+01  5.29123892677001E+00
> >  5.29123892677001E+00  1.06020074865687E+01  1.59120730599107E+01
> >  1.60386268849871E+01 -9.68225628470105E-03  5.16468510169366E+00
> >  1.60386268849871E+01  9.68225628470109E-03  1.60386268849871E+01
> >  1.59120730599107E+01  1.06020074865687E+01  5.29123892677001E+00
> >  1.59120730599107E+01  1.06013045001119E+01  1.59120730599107E+01
> >  5.16468510169366E+00  5.16468510169365E+00  9.68225628470109E-03
> >  5.29123892677001E+00  5.29123892677001E+00  1.06013045001119E+01
> >  5.16468510169366E+00  1.60386268849870E+01 -9.68225628470106E-03
> >  5.29123892677001E+00  1.59120730599107E+01  1.06020074865687E+01
> >  1.60386268849871E+01  5.16468510169365E+00 -9.68225628470106E-03
> >  1.59120730599107E+01  5.29123892677001E+00  1.06020074865687E+01
> >  1.60386268849871E+01  1.60386268849870E+01  9.68225628470109E-03
> >  1.59120730599107E+01  1.59120730599107E+01  1.06013045001119E+01
> >  7.88922678674774E+00  2.64420766012477E+00  7.88922678674772E+00
> >  7.93032740139436E+00  2.64935519364036E+00  1.85539567930404E+01
> >  7.95231990242752E+00  1.32509920842532E+01  7.95231990242752E+00
> >  7.88922678674774E+00  1.33140851999330E+01  1.85591043265559E+01
> >  1.85539567930404E+01  2.64935519364036E+00  7.93032740139436E+00
> >  1.89681850257071E+01  2.23512696097355E+00  1.89681850257071E+01
> >  1.85591043265559E+01  1.33140851999330E+01  7.88922678674772E+00
> >  1.85539567930404E+01  1.32729845852864E+01  1.85539567930404E+01
> >  2.23512696097355E+00  2.23512696097355E+00  2.23512696097355E+00
> >  2.64935519364036E+00  2.64935519364036E+00  1.32729845852864E+01
> >  2.64935519364036E+00  1.32729845852864E+01  2.64935519364037E+00
> >  2.64420766012477E+00  1.33140851999330E+01  1.33140851999330E+01
> >  1.32729845852864E+01  2.64935519364036E+00  2.64935519364037E+00
> >  1.33140851999330E+01  2.64420766012477E+00  1.33140851999330E+01
> >  1.33140851999330E+01  1.33140851999330E+01  2.64420766012476E+00
> >  1.32509920842532E+01  1.32509920842532E+01  1.32509920842532E+01
> >  2.64420766012477E+00  7.88922678674775E+00  7.88922678674772E+00
> >  2.64935519364036E+00  7.93032740139436E+00  1.85539567930404E+01
> >  2.64935519364036E+00  1.85539567930404E+01  7.93032740139436E+00
> >  2.23512696097355E+00  1.89681850257071E+01  1.89681850257071E+01
> >  1.32509920842532E+01  7.95231990242758E+00  7.95231990242752E+00
> >  1.33140851999330E+01  7.88922678674775E+00  1.85591043265559E+01
> >  1.33140851999330E+01  1.85591043265559E+01  7.88922678674772E+00
> >  1.32729845852864E+01  1.85539567930404E+01  1.85539567930404E+01
> >  7.88922678674774E+00  7.88922678674775E+00  2.64420766012476E+00
> >  7.95231990242752E+00  7.95231990242758E+00  1.32509920842532E+01
> >  7.93032740139436E+00  1.85539567930404E+01  2.64935519364037E+00
> >  7.88922678674774E+00  1.85591043265559E+01  1.33140851999330E+01
> >  1.85539567930404E+01  7.93032740139436E+00  2.64935519364037E+00
> >  1.85591043265559E+01  7.88922678674775E+00  1.33140851999330E+01
> >  1.89681850257071E+01  1.89681850257071E+01  2.23512696097355E+00
> >  1.85539567930404E+01  1.85539567930404E+01  1.32729845852864E+01
> >
> > # Second: The BS...
> > #getcell2 -1   # start from the previous acell values
> > #getxred2 -1   # start from the previous xred values
> > #prtwf 0 # No wavefunction output is provided
> > iscf    -2   # no self-consistent calculations
> > getden  -1   # because start with the previous computed density
> > kptopt 0 # read directly nkpt, kpt, kptnrm and wtk  (corresponds to 
> > the usage before version 2.1) 
>
> > nband  132  # 6*32=192 ( 2atoms: 6 = 4 VB + 2 CB) en plus pour avoir 
> > la conduction
> > nkpt 1 #5 #nbre of kpts
> > kpt 0.00 0.00 0.00
> >
> > tolwfr  1.0d-12
> > enunit  1             # Will output the eigenenergies in eV
> > #Definition of the atom types in the cristal
> > ntypat 3          # There are three types of atom
> > znucl 7 33 31     # 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 64           # There are 64 atoms
> > typat 1 2 2 2 2 2 2 2
> >      2 2 2 2 2 2 2 2
> >      2 2 2 2 2 2 2 2
> >      2 2 2 2 2 2 2 2
> >      3 3 3 3 3 3 3 3
> >      3 3 3 3 3 3 3 3
> >      3 3 3 3 3 3 3 3
> >      3 3 3 3 3 3 3 3
> >
> > #Definition of the planewave basis set
> > #ecut 20.0         # Maximal kinetic energy cut-off, in Hartree
> > ecut 25.0
> > #ecut 27.5625
> > #ecut3 30.0
> > #ecut4 35.0
> > #ecut 32.0
> >
> > #Definition of the SCF procedure
> > nstep 60          # Maximal number of SCF cycles
> > #toldff 5.0d-6 # A value ten times smaller than tolmxf is suggested 
> > (for example 5.0d-6 hartree/bohr)
> > toldfe 1.0d-10    # Will stop when, twice in a row, the difference 
> > diemac 12.0       # Although this is not mandatory, it is worth to
> >
> >