ABINIT Discussion Forums

Hello abinit users,
I haved used abinit to calculate the graphene's superconducting properties. I have optimized the cell and ions, and my k-mesh:36*36*1,q-mesh:9*9*1,ecut:40 ha. However,in anaddb calculation,when I use defaults of elphsmear(0.01 ha) and a2fsmear(2e-5 ha), in the outfile of phonon linewidth, there are some phonon linewidth values are negative. The values aren't all positive until elphsmear=0.2 ha. We know, in the function of phonon linewidth, it couldn't be negative. I don't know why. Can you help me ? The follow is anaddb inputfile:
elphflag 1
nqpath 4
qpath
0.0 0.0 0.0 # gamma
0.5 0.0 0.0 # M
1/3 1/3 0.0 # K
0.0 0.0 0.0 # gamma

ngqpt 9 9 1

asr 2
dipdip 0

brav 1

nqshft 1
q1shft 0.0 0.0 0.0
ifcflag 1
ifcana 1
natifc 2
atifc 1 2
ifcout 3
dieflag 0

eivec 1

symdynmat 0
elphsmear 0.01

Thank you very much! Best wishes!

The negative value comes from the Fourier transform interpolation of the linewidths... they are unphysical I agree. They come from the rapid variation of the e-p linewidths with q, which is not smooth and sinusoidal (ie easily FT-able)

You can try playing with ep_doscalprod - if you do the scalar product with the phonon displacement vectors before interpolation things behave a bit better, but there is an approximation in diagonalizing the e-p matrices afterwards (the eigenvectors that diagonalize the dynamical matrix are not necesarily the same as those that diagonalize the e-p matrices - the residual of the diagonalization is printed to the log.

The better way would be to use either linear interpolation of the matrices or, much better, wannier interpolation. We are working on this, but it doesn't work yet.

Increasing elphsmear is not very physical - you are including a large swath of bands around the Fermi level and have a huge effective temperature (0.2 Ha!).

Matthieu

mverstra wrote:The negative value comes from the Fourier transform interpolation of the linewidths... they are unphysical I agree. They come from the rapid variation of the e-p linewidths with q, which is not smooth and sinusoidal (ie easily FT-able)

You can try playing with ep_doscalprod - if you do the scalar product with the phonon displacement vectors before interpolation things behave a bit better, but there is an approximation in diagonalizing the e-p matrices afterwards (the eigenvectors that diagonalize the dynamical matrix are not necesarily the same as those that diagonalize the e-p matrices - the residual of the diagonalization is printed to the log.

The better way would be to use either linear interpolation of the matrices or, much better, wannier interpolation. We are working on this, but it doesn't work yet.

Increasing elphsmear is not very physical - you are including a large swath of bands around the Fermi level and have a huge effective temperature (0.2 Ha!).

Matthieu

Thank you for your reply. I just used the keyword'ep_scalprod=1'，there was an error : read_gkk : enter normsq_gkq normsq_gkq : calling nmsq_gam_sumFS p0_19597: p4_error: interrupt SIGSEGV:11
I don't why.And when I used 'elphsmear=0.2Ha', the linewidths were all positive.So can I choose 0.2Ha？
Best wishes！

For the segfault you need to follow the nettiquette to describe the bug

For the 0.2Ha read my previous post - it's not reasonable.