ABINIT Discussion Forums

Dear all,

I'm trying to calculate the phonon spectrum of sc phase of As. However, when I start the rf calculation, the code stops after the 1st iteraion of the 1st dataset with a bug

ETOT 1 -7.6937995875092 -7.694E+00 9.704E-03 1.013E+01
scprqt: <Vxc>= -4.1658462E-01 hartree

dielmt : 15 largest eigenvalues of the hermitian RPA dielectric matrix
1-5 : NaN NaN NaN NaN NaN
6-10 : NaN NaN NaN NaN NaN
11-15: NaN NaN NaN NaN NaN

dielmt : 5 smallest eigenvalues of the hermitian RPA dielectric matrix
1-5 : NaN NaN NaN NaN NaN

Simple mixing update:
residual square of the potential : NaN

ITER STEP NUMBER 2
vtorho : nnsclo_now= 2, note that nnsclo,dbl_nnsclo,istep= 0 0 2

Non-SCF iterations; k pt # 1 k= 0.12500 0.12500 0.12500 band residuals:
res: NaN NaN NaN NaN NaN
ene: NaN NaN NaN NaN NaN
fxphas : BUG -
The eigenvector number 1 has zero norm.
Action : contact ABINIT group.



Has anyone encountered this bug before? What does it mean and how could I solve it? I've seen that there were a few topics about this bug before, but there was no (confirmed) solution. I'm using abinit 6.10.3. My input file is given bellow.

Thank you in advance.

Yours,

Igor Lukacevic


Input file

# Crystalline As - sc phase
# computation of the response to homogeneous
# electric field and atomic displacements, at q=0
# and phonon dispersions

ndtset 11

#Set 1 : ground state self-consistency

getwfk1 0 # Cancel default
kptopt1 1 # Automatic generation of k points, taking
# into account the symmetry
nqpt1 0 # Cancel default
tolvrs1 1.0d-18 # SCF stopping criterion (modify default)
rfphon1 0 # Cancel default

#Q vectors for all datasets

#Complete set of symmetry-inequivalent qpt chosen to be commensurate
# with kpt mesh so that only one set of GS wave functions is needed.
#Generated automatically by running GS calculation with kptopt=1,
# nshift=0, shiftk=0 0 0 (to include gamma) and taking output kpt set
# file as qpt set. Set nstep=1 so only one iteration runs.

nqpt 1 # One qpt for each dataset (only 0 or 1 allowed)
# This is the default for all datasets and must
# be explicitly turned off for dataset 1.

qpt2 0.00000000E+00 0.00000000E+00 0.00000000E+00
qpt3 2.50000000E-01 0.00000000E+00 0.00000000E+00
qpt4 5.00000000E-01 0.00000000E+00 0.00000000E+00
qpt5 2.50000000E-01 2.50000000E-01 0.00000000E+00
qpt6 5.00000000E-01 2.50000000E-01 0.00000000E+00
qpt7 5.00000000E-01 5.00000000E-01 0.00000000E+00
qpt8 2.50000000E-01 2.50000000E-01 2.50000000E-01
qpt9 5.00000000E-01 2.50000000E-01 2.50000000E-01
qpt10 5.00000000E-01 5.00000000E-01 2.50000000E-01
qpt11 5.00000000E-01 5.00000000E-01 5.00000000E-01

#Sets 2-11 : Finite-wave-vector phonon calculations (defaults for all datasets)

getwfk 1 # Use GS wave functions from dataset1
kptopt 3 # Need full k-point set for finite-Q response
rfphon 1 # Do phonon response
rfatpol 1 1 # Treat displacements of all atoms
rfdir 1 1 1 # Do all directions (symmetry will be used)
tolvrs 1.0d-8 # This default is active for sets 2-11

#######################################################################
#Common input variables

#Definition of the unit cell - p = 0 GPa
acell 3*4.9175745006
rprim 1.0 0.0 0.0
0.0 1.0 0.0
0.0 0.0 1.0

#Definition of the atom types
ntypat 1
znucl 33

#Definition of the atoms
natom 1
typat 1
xred 0.25 0.25 0.25

#Gives the number of band, explicitely (do not take the default)
nband 5
occopt 3
tsmear 0.001

#Exchange-correlation functional
ixc 1

#Definition of the planewave basis set
ecut 60
ecutsm 0.5
dilatmx 1.05

#Definition of the k-point grid
ngkpt 3*4
nshiftk 1
shiftk 0.5 0.5 0.5

#Definition of the SCF procedure
iscf 7
iprcel 45
nstep 100

#Timing options
timopt 2

Guess who....

1) if you remove the shiftk (to 000) the crash goes away
2) I don't know where it's coming from, but it's in the calculation of the dielectric susceptibility for iprec 45 (if you remove that the crash also goes away).

More later if I can find it...

Matthieu

Me again - the problem arises for too sparse k-point grids, with widely spaced energies. In the calculation of the dielectric preconditioning you have to divide by the derivative of the Fermi Dirac delta d/de (f_FD), and this is 0.0 everywhere to machine precision. At Gamma there is a finite value so a centered k-point grid will make the error go away. Otherwise you just need to use a more realistic k-grid (you have metallic occupations etc...)

I will add a check in abinit 6.13 so the code will not crash...

Matthieu

Been busy yesterday, eh?

Bless you! I'll keep your notices in mind for the future.

Best,

Igor