ABINIT Discussion Forums

Hello,
I have tried to generate the phonon band structure by following the RF2 instruction.
I obtained a set of q-points(12 points) from GS calculation and set 'nband' to 15. (It gave me errors with less than 15)
A material is a rhombohedral structure with 4 basis, and I acquired 'acell' and 'rprim' from the convergence studies.
I guess my problems are caused by the definition of k-point grid. First I tried with

ngkpt 4 4 4
nshifk 1
shiftk 0.5 0.5 0.5

because it's neither FCC nor BCC. It took more than 30 hrs to generate the dynamical matrices at the q-points and eventually
produced negative phonon frequencies at some wave vectors. In addition to the negative frequencies, the shape of the phonon
dispersion curves were fluctuating more than expected. I tried to use a different definition on the grid like

nkpt 5
nshifk 1
shiftk 0.5 0.5 0.5
kptrlatt -2 2 2
2 -2 2
2 2 -2

but it kept giving me errors with wrong nkpt values. I've tried more than 10 times with different inputs and pseudopotential data
to correct this problem. Any help on this problem will be very appreciated. I looked up the similar posting in the forum, but
the answers to the previous posting didn't help. Other parameters in my input files are

For the q-points other than gamma point
getwfk 1
kptopt 3
rfphon 1
rfatpol 1 5
rfdir 1 1 1
tolvrs 1.0d-8

#Common input variables
ecut 10
nband 15
ixc 1
iscf 5
nstep 100

The rest of the parameters are the same as the inputs in RF2 tutorial.
Thank you.

I think you should tighten the convergence parameter together with a larger nstep. I use "tolwfr 1.0d-22" and "nstep 1000".

Additionally, is you geometrical structure fully relaxed? I use "tolmxf 1.85d-6" in such calculation.

Sincerely,
Guangfu Luo

Your ecut is quite low, especially if you're using norm-conversing pseudo. Are you sure you are converged?

Also, your k-grid seems small (it's hard to tell without rprim and acell). Increasing it might get rid of these imaginary frequencies.

Simon

Thank you, Guangfu and Simon.

My geometrical structure is fully relaxed, and 'ecut' in my simulations was actually 20.
My 'ngkpt 4 4 4' was from the convergence study as following the tutorial#4,
and my simulation took more than 100 hrs to get the final results. Do I still need to try with larger k-grid?

I just got another error message from the latest simulation.

scfcge : ERROR -
Potential-based CG line minimization not converged after 13 restarts.
Action : read the eventual warnings about lack of convergence.
Some might be relevant. Otherwise, raise nband.

Would it be due to the insufficient 'ntstep'? I am still not sure if it needs to be greater than 200.
Someone in the forum suggested to use iscf 7 instead of iscf 5 to recover the error above.
Is it also okay for the response calculation?

I'll try to run a new simulation with your suggestions, reduced the tolerances and larger nstep.
I really appreciate your helps.

Sincerely,
Jung