[gpaw-users] : gpaw.KohnShamConvergenceError: Did not converge!

Nichols A. Romero naromero at alcf.anl.gov
Thu Nov 11 21:17:51 CET 2010


I tried to different solvers 'cg' and 'rmmdiis' and they both fail with 
the same error. The problem is that you have a bad configuration of 
atoms. They are on top of each other. 

----- Original Message -----


How do I flag our scaLAPACK? 

The following is the complete script. I ran it in two ways with the same result. One is in python line by line. The other is to save it as test0.py and run "python test0.py". 

I also tried this, the result is the same. 
calc = GPAW(h=0.18, nbands=24, xc='PBE', txt='CMO.out') 

------------------------------------------------------------------ 
from gpaw import setup_paths 
setup_paths.insert(0, '.') 

from ase import Atoms, Atom 
from gpaw import GPAW 

from ase.visualize import view 
from ase.io import write 
a = 3.937 
d = a /2 
slab = Atoms(['Ca','Mn','O','O','O'], 
positions=[(d, d, d),(0, 0, 0),(d, 0, 0),(0, d, 0),(0, 0, d)], 
cell=(d, d, d),pbc=(1,1,1)) 

write('slab.xyz', slab) 
view(slab) 

# gpaw calculator: 
calc = GPAW(h=0.18, nbands=36, xc='PBE', eigensolver='cg', kpts=(6,6,6), txt='CMO.out') 
slab.set_calculator(calc) 

e1 = slab.get_potential_energy() 
calc.write('CMO.gpw') 
-------------------------------------------------------------------------------------------------------- 

On 11/11/2010 8:48, Nichols A. Romero wrote: 

How many bands are in your calculation? If it's less than 1000, there
is no point to using ScaLAPACK. 

How are you running this calculation? In particular, what are you
runtime flags?

----- Original Message ----- 

The test went thru with no problem. What's the next?



On 11/10/2010 1:41 PM, Nichols A. Romero wrote: 

You need to test the MPI version with 4 cores if I recall correctly.
You
don't
need 4 cores to run this test. But you will need MPI.

mpirun -np 4 gpaw-python <gpaw_dir>/gpaw/test/test.py

------------------------------------------------------------------------

    Nic,

    I also suspected that the first time I saw this so I tried the
    test
    (see the following) and it did not complain about anything.

    gpaw-python `which gpaw-test` 2>&1 | tee test.log

    Is this test sufficient?

    On 11/10/2010 13:24, Nichols A. Romero wrote:

        Looks like ScaLAPACK's inverse Cholesky failed.

        Did you run your regression tests in parallel? Did they all
        pass?

        ------------------------------------------------------------------------

            Using more bands now, I have the following error

             >>> e1 = slab.get_potential_energy()
            Traceback (most recent call last):
            File "<stdin>", line 1, in <module>
            File "/home/ylli/ase/ase/atoms.py", line 503, in
            get_potential_energy
            return self.calc.get_potential_energy(self)
            File
            "/usr/lib/python2.6/dist-packages/gpaw/aseinterface.py",
            line 32, in get_potential_energy
            self.calculate(atoms, converge=True)
            File "/usr/lib/python2.6/dist-packages/gpaw/paw.py",
            line
            265, in calculate
            self.occupations):
            File "/usr/lib/python2.6/dist-packages/gpaw/scf.py",
            line
            46, in run
            wfs.eigensolver.iterate(hamiltonian, wfs)
            File
            "/usr/lib/python2.6/dist-packages/gpaw/eigensolvers/eigensolver.py",
            line 65, in iterate
            wfs.orthonormalize()
            File
            "/usr/lib/python2.6/dist-packages/gpaw/wavefunctions/fdpw.py",
            line 190, in orthonormalize
            self.overlap.orthonormalize(self, kpt)
            File "/usr/lib/python2.6/dist-packages/gpaw/overlap.py",
            line 76, in orthonormalize
            self.ksl.inverse_cholesky(S_nn)
            File "/usr/lib/python2.6/dist-packages/gpaw/blacs.py",
            line
            620, in inverse_cholesky
            raise RuntimeError('Failed to orthogonalize: %d' % info)
            RuntimeError: Failed to orthogonalize: 20


            On 11/9/2010 12:54, Christian Glinsvad wrote:


                Hi

                Include more unoccupied bands in your calculation.
                35
                valence electrons
                barely fit into the 18 bands - it converges just
                fine
                with 24 bands.

                Regards
                Christian Glinsvad

                On Tue, 9 Nov 2010, Yuelin Li wrote:

                    I have a simple script but could not find how to
                    make the calculation
                    converge. When encountering such problem, what
                    is
                    the general way to get
                    around it? Thanks.
                    ------------------------------------------

                    from ase import Atoms, Atom
                    from gpaw import GPAW

                    from ase.visualize import view
                    from ase.io import write
                    a = 3.937
                    d = a /2
                    slab = Atoms(['Ca','Mn','O','O','O'],
                    positions=[(d, d, d),(0, 0, 0),(d, 0, 0),(0, d,
                    0),(0, 0, d)],
                    cell=(a, a, a),pbc=(1,1,1))

                    write('slab.xyz', slab)
                    view(slab)

                    # gpaw calculator:
                    calc = GPAW(h=0.18, nbands=18, xc='PBE',
                    kpts=(6,6,6), txt='CMO.out')
                    slab.set_calculator(calc)

                    e1 = slab.get_potential_energy()

                    _______________________________________________
                    gpaw-users mailing list gpaw-users at listserv.fysik.dtu.dk https://listserv.fysik.dtu.dk/mailman/listinfo/gpaw-users _______________________________________________
            gpaw-users mailing list gpaw-users at listserv.fysik.dtu.dk https://listserv.fysik.dtu.dk/mailman/listinfo/gpaw-users --
        Nichols A. Romero, Ph.D.
        Argonne Leadership Computing Facility
        Argonne National Laboratory
        Building 240 Room 2-127
        9700 South Cass Avenue
        Argonne, IL 60490
        (630) 252-3441




--
Nichols A. Romero, Ph.D.
Argonne Leadership Computing Facility
Argonne National Laboratory
Building 240 Room 2-127
9700 South Cass Avenue
Argonne, IL 60490
(630) 252-3441 


-- 
Nichols A. Romero, Ph.D. 
Argonne Leadership Computing Facility 
Argonne National Laboratory 
Building 240 Room 2-127 
9700 South Cass Avenue 
Argonne, IL 60490 
(630) 252-3441 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://listserv.fysik.dtu.dk/pipermail/gpaw-users/attachments/20101111/1797d7c6/attachment.html 


More information about the gpaw-users mailing list