Got negative freq for dense q points

[shaoziji]

Dear users/developers,
I am doing the test work on abinit as a new user, I calculated the el-ph of the Im-3m BCC H3S with 4*4*4 q points, the results was ok an no negative freq appeared, but as I changed the q points to 8*8*8 and keep other input unchanged. I got a negative freq at one q points about -400 cm -1. The negative freq q point was generated by 8*8*8 q points but didn't appeared when q points was 4*4*4. Is there anything wrong with the calculation?

Sincerely thanks

Shao
the dataset of the q point next to the negative q point without negative freq

Code: Select all

ndtset 4
# DATASET 1 : make ground state wavefunctions and density
tolvrs1     1.0d-18
iscf1    7
nline1   8
rfphon1  0
nqpt1    0
prtwf1   1
getwfk1  0
kptopt1  1
dilatmx1    2.00
ecutsm1     0.5
strtarget1  -6.797858d-3  -6.797858d-3  -6.797858d-3  0  0  0
# DATASET 2 : calculate the Qpoints
qpt2         -3.75000000E-01 5.00000000E-01 1.25000000E-01
# DATASET 3 : calculate WF on full BZ
tolvrs3      1.0d-18
prtwf3       1
getwfk3      1
iscf3        7
nline3       1
nqpt3        0
rfphon3      0
# DATASET 5 : Compute the GKK correctly (no gauge problem)
# This step is very quick as NSCF+nstep 1
qpt4         -3.75000000E-01 5.00000000E-01 1.25000000E-01
prtgkk4      1            # print out gkk files directly
iscf4      -2
tolwfr4      1.0d-20
nline4       1
getwfk4      3
get1den4     2
# general data for all phonon calculations:
prepgkk 1        # flag to calculate all perturbations for el-phon calculations
use_nonscf_gkk 0 # enforce old default and scf-calculate all perturbations
nqpt 1           # 1 qpoint at a time
tolvrs 1.e-8     # tolerance on 2DTE convergence: potential^(1) is what we need
getwfk 1         # all other DS get wf from DS1
prtwf 0
rfatpol 1 4      # all atoms are perturbed-Notice
rfdir 1 1 1      # all directions of perturbation
rfphon 1
#  Common data for GS and perturbation runs
# The kpoint grid is minimalistic to keep the calculation manageable.
ngkpt        24 24 24
kptopt       3
# As the kinetic energy cutoff
ecut         35
# Use a centered grid for the kpoints
nshiftk    1
shiftk      0 0 0
#  Common data
acell      4.8813597008E+00  4.8813597008E+00  4.8813597008E+00
rprim     -5.7735026919E-01  5.7735026919E-01  5.7735026919E-01
           5.7735026919E-01 -5.7735026919E-01  5.7735026919E-01
           5.7735026919E-01  5.7735026919E-01 -5.7735026919E-01
nband        15
#  include metallic occupation function with a small smearing
occopt       6
tsmear       0.001
natom        4
ntypat       2
typat        1 1 1 2
xred      5.0000000000E-01  5.0000000000E-01  0.0000000000E+00
          0.0000000000E+00  5.0000000000E-01  5.0000000000E-01
          5.0000000000E-01  0.0000000000E+00  5.0000000000E-01
          0.0000000000E+00  0.0000000000E+00  0.0000000000E+00
nstep        100000
znucl        1 16

and the dataset of negative q point was nothing different from this but the coordinate of q : -0.375 -0.375 0.125

[bmatt]

Hello,
As a user, I get this problem too. In my experience, you must increase parameters like ngkpt and ecut (try ngkpt first) to help the situation. Sometimes though, if the structure itself is unstable, this negative frequency suggests instabilities in the crystal structure or phase. Increasing ngkpt makes the calculations expensive so make sure you can get away with using lower tolvrs when converging the scf phonon cycle in your rfphon dataset

[shaoziji]

Hello,
As a user, I get this problem too. In my experience, you must increase parameters like ngkpt and ecut (try ngkpt first) to help the situation. Sometimes though, if the structure itself is unstable, this negative frequency suggests instabilities in the crystal structure or phase. Increasing ngkpt makes the calculations expensive so make sure you can get away with using lower tolvrs when converging the scf phonon cycle in your rfphon dataset

Thank you very much for your answer, but my computation sources cannot afford the denser kpoints or larger ecut calculation. Is that ok to set a more dense kpoints during the scf calculation at first step and then a smaller kpoints for the following calculation about every qpoints? I want to calculate the el-ph porperties and the superconductive temperature(Tc) , if I set two ngkpt like this , will it influence the results?

Thank you very much

Shao

[bmatt]

Oh I see, when I first looked at the question I did not read thoroughly
I thought you were running a phonon band structure calculation and getting negative frequencies at specific q-points. My mistake, upon 2nd look, I have no real experience doing el-ph calcs

[shaoziji]

Oh I see, when I first looked at the question I did not read thoroughly
I thought you were running a phonon band structure calculation and getting negative frequencies at specific q-points. My mistake, upon 2nd look, I have no real experience doing el-ph calcs

Thank you all the same for your answer, I think you are right. I should use a denser k points for this calculation.