Problem with anaddb phonon interpolation

Hi everyone,

I’m sorry to bother again with questions about anaddb and phonons. I’m trying to calculate phonons for RuO2 with rutile structure, space group #136. I calculated the phonons at every irreducible q-point on a 4x4x4 grid.

If I check the results of the dfpt calculation at the point q=(0.25,0.25,0), I can see that all the phonon frequencies are real.

Later, when I use anaddb to Fourier interpolate the phonons on a finer q-path, I somehow end up getting imaginary phonons at the point q=(0.25,0.25,0), in contrast to the results of the dfpt calculation.

Can anybody suggest what I’m doing wrong?

I’m attaching the abo output file of the dfpt calculation and my anaddb input file.


ANA.ddb.abi (19.8 KB)
Ph_QPT4.abo (325.7 KB)

Hi Luca,

It might be that you are not doing anything wrong, but to make sure, can you compare the frequencies of Gamma and QPT4 at the end of the output file? And how many imaginary frequencies do you start to have after interpolation?
Also, can you provide the phonon dispersion between Gamma and (1/2,1/2,0)?

Let me know!

Hi Bogdan,

Sure, here are the phonon frequencies at Gamma (in Hartrees)

4.918776E-04 4.918776E-04 5.182654E-04 8.758738E-04 1.036207E-03
1.420605E-03 1.420605E-03 1.928247E-03 2.009786E-03 2.009786E-03
2.358521E-03 2.358521E-03 2.599143E-03 2.730200E-03 2.749015E-03
2.924342E-03 2.924342E-03 3.290083E-03

while these are the frequencies at (1/4,1/4,0)

1.689820E-04 3.906673E-04 7.974226E-04 8.503129E-04 9.439819E-04
1.058038E-03 1.377293E-03 1.650206E-03 1.907909E-03 2.031972E-03
2.201436E-03 2.338573E-03 2.611831E-03 2.698082E-03 2.781963E-03
2.900153E-03 2.980013E-03 3.131823E-03

I’m also attaching figures of the interpolated phonon frequencies w/ and w/o the acoustic sum rule turned on. Do you think it could just be a convergence issue? Isn’t it weird that interpolation turns the lowest (1/4,1/4,0) frequencies imaginary even w/o the asr imposed?

On a side note, I have calculated the dft ground state wavefunction with (AF)magnetism turned on, but with dfpt I got an error when trying to impose nspden=2 so I had to use nspden=1. Could this (using dfpt with nspden=2 on a wfk created with nspden=1) be creating false instabilities? Do you know if it is possible to use DFPT with magnetism turned on?

Sorry for the many questions and thanks for the help,

PHBST.pdf (169.9 KB)
PHBST_noASR.pdf (169.7 KB)

Hi Luca,

Indeed I was trying to figure out how the phonons would look before and after ASR.
Looking at the dispersion, nothing looks suspicious with the data that you are processing.
I would say it’s not convergence issues. It just looks to me that you have a strong instability at (1/2,1/2,1/2) in the geometry that you are studying.
If convergence would have been a problem, you would have seen phonon branches completely in the imaginary portion of the spectrum since you wouldn’t have had properly sampled the lowest energy around individual atomic perturbations.

Regarding the magnetism part, it is possible to use DFPT with collinear magnetism for both PAW and Normconserving pseudos using the following instructions alongside your spinat instruction :

nspden  2
nsppol  2
nspinor 1

In the case of noncollinear magnetism, it works only for normconserving pseudos, i.e. if you are trying to use the following set of instructions:

nspden  4
nsppol  1
nspinor 2

Regarding the use of WFK, I would try to keep consistent in the treatment of the WFK in general, not only in this particular case. What error do you get that you weren’t able to proceed with your calculation?

Maybe some preemptive questions for yourself would help finding if there is a problem or not:

  1. how does the relaxed structure that you use compare with the experimental structure?
  2. how does the band gap compare with experimental findings if there is a bandgap? otherwise how does the Fermi surface compare with any experimental indication?
  3. why do you expect not to have instabilities?
  4. if you systematically increase the convergence parameters, how the phonon dispersion changes?

Let me know!

Quick note: you can not use the afm trick of saving only one type of wave function (up and not down spin) with dfpt. You have to do the whole stack with nsppol =nspden = 2.

In my experience you get garbage phonons, but it does not crash. I thought we had put in an error message.

This may not be your problem, though, and in any event anaddb should interpolate back to exactly the same frequencies at the input DDB q points…

Could be due to the treatment of the asr, but it would have to be quite broken

Hi both,

after doing some more tests, I am now even more confused. First of all, I switched off magnetism, I am now just trying to calculate the phonons in paramagnetic RuO2.

I fully relaxed the structure and then calculated the phonons on a 4x4x4 q point grid, using a 16x16x20 k point grid and 550 eV energy cutoff. I still see unstable phonon modes at A and Z, which is in conflict with previous results found in the literature (PhysRevB.75.092301 and PhysRevLett.125.147001) that found all the phonons to be real under no applied strain. Notice that I am using a denser k-point grid and larger energy cutoff than what is used in those two papers, and that my phonon bands look really similar to those two in the GM-X-M line and only start to diverge in the other directions.

At this point I’m not sure what is wrong, do you have any ideas? I am also attaching a convergence test I did on the lowest gamma phonon frequency.

ConvTest.pdf (42.9 KB)
PHBST_nm.pdf (167.5 KB)
PHBST_nm_noASR.pdf (167.5 KB)

You have tons of empty bands, not useful (you need a few, but not so many).

And you are insanely over converging the scf cycle: Converge on toldfe to 1.e-10 and you should be fine.

Hi Luca,

Before proceeding any further, I would recommend a quick test where you replace the JTH pseudos with GBRV ones, in a quick non-magnetic, 4x4x3 k-points grid with a 4x4x3 q-point grid, and relatively low ecut ~400-500 eV.
You can find the pseudos that you need here:

Please let me know how your phonons look like in the two sets of pseudos.

If the problem persists, please try the previous calculation with ABINIT version 9.2.2, that you can find here

Keep us posted about the progress of your phonon calculations.