anaddb is extremely slow for phonon interpolation

MULTIBINIT, aTDEP, ANADDB, optics, cut3d, mrg* postprocessors

Moderators: MMNSchmitt, gonze

Locked
zhiwu
Posts: 2
Joined: Sun Aug 16, 2020 5:30 pm

anaddb is extremely slow for phonon interpolation

Post by zhiwu » Sun Aug 16, 2020 5:55 pm

I am using PAW and DFPT to carry out phonon band structure calculations for NiTi in B2 symmetry. See the attached file. The attached file shall be niti.tar. But .tar extension is not allowed (why?) and I changed its name to niti.in, which contains the DDB (combined) and the input files.

The problem is that anaddb takes for ever to compute IFC and phonon frequencies. It should generate results within a few minutes; instead, I have to use 28 cores to do such calculation (anaddb < phonon_band.files). anaddb used 35 minutes to construct the IFC, then for each q point, it took 4 minutes.

It seems to me that anaddb does not interpolate phonon frequencies (though I explicitly set ifcflag=1); instead it calculates phonons one by one using some other methods, since the computed phonon bands are not very smooth, though it looks essentially correct.

Please let me know what is wrong and how to fix it. I think it is caused by a bug in anaddb.
Attachments
niti.in
(765 KiB) Downloaded 310 times

zhiwu
Posts: 2
Joined: Sun Aug 16, 2020 5:30 pm

Re: anaddb is extremely slow for phonon interpolation

Post by zhiwu » Mon Aug 17, 2020 5:31 pm

I solved the issue.

Should not do electric field perturbation for a metal (NiTi) at the gamma q-point.

ebousquet
Posts: 469
Joined: Tue Apr 19, 2011 11:13 am
Location: University of Liege, Belgium

Re: anaddb is extremely slow for phonon interpolation

Post by ebousquet » Wed Aug 19, 2020 9:53 am

Dear zhiwu,
Is your system metallic or insulator? I would suspect that it is metallic and thus dipdip=1 has no sense here since the electronic dielectric constant (epsilon^infty) is infinite for metals. For the same reason, using the second list of wavevector to estimate the LO/TO splitting has no meaning too. If it is the case (NiTi=metallic) then what about running your input by putting dipdip=0 and removing nph2l and qph2l?
If it is insulating, then how large is the bandgap and how big is the electronic dielectric constant?
Best wishes,
Eric

Locked