The ABINIT Users Mailing List ( CLOSED )

[matthieu verstraete <mjv500@york.ac.uk>]

> Also, I decide to increase qpoint mesh to 4x4x4 than 2x2x2.
This may be sufficient, but probably not. Depends on the dispersion of 
the phonons and elphon quantities.

> In short, but, the result was terrible. the Tc was estimated to be ~ 
> 10e15 Kelvin.
Is this the McMillan or the (very alpha stage) Eliashberg equation 
value? The latter should never be trusted, and I have done very few 
tests. Btw if anyone does do tests and has comments or improvements, 
please don't hesitate to share. If the coulomb mu value is not physical 
or the coupling strength is too low (eg if your DOS is too low) then 
you'll get an exponential explosion of Tc. It doesn't mean much, it's 
too sensitive a value.

> My question is following,
>
> 1. To generate higher mesh qpoint, suppose simple fcc case, how many 
> datasets I should try with ? anaddb asked me to generate 64 qpoints in 
> general.
> But thanks to symmetry, we can recude it ( am I right?). I was not so 
> sure in this problem. I tried the smallest set of qpoints.
Yes. You can reduce the qpoints, but not the perturbations at a given 
qpoint. You can run abinit with a kpoint grid set to the qpoint grid you 
want to use - correct ngkpt, shiftk and everything. In the header 
information you will have the reduced nkpt and kpt coordinates to be 
used in the phonon generation.

> Anyhow, anaddb accepted it. 8 qpoints was reasonable number 8(3+1) + 1 
> = 33 dataset was okay. But was it enough ?
> Especially when generating qpoints, could you recommend a good 
> literature to follow ? not Monkhost and Pack paper.
As above: examine the phonon dispersion and how it changes going from 
2x2x2 to 4x4x4.
> 2. What could be a critical parameter to check the convergence of the 
> calculations ?
make sure the 2DTE are converged for each phonon also.
> As far as I know, in addition to some basic convergence tests such as 
> ecut , # of kpoints, tolerances, we can check spectral functions' 
> behaviors. something more ?
Are your phonons ok? Have you checked the convergence of the ground 
state wrt ecut and kpoints? The hgh are notoriously hard 
pseudopotentials. The parameters look ok, globally. Is your acell 
correct? Not high pressure stuff or anything?

Without more information from your anaddb output I can't say if things 
are really going wrong or not. Instead of looking at Tc you should tell 
us about the frequencies, the elphon coupling parameter, and linewidths. 
How do they compare to literature?


Matthieu