Phonon DOS problem  [SOLVED]

Phonons, DFPT, electron-phonon, electric-field response, mechanical response…

Moderators: mverstra, joaocarloscabreu

Locked
CHILL
Posts: 25
Joined: Sun Oct 08, 2017 1:37 pm

Phonon DOS problem

Post by CHILL » Wed Oct 11, 2017 11:14 pm

1. I calculated phonon band structure, that had an unstable part.(first band at Figure 1. around gamma point or 0 of x axis)
Logically enough, DOS file also contains negative frequencies.
2. Using different values of rifcsph I managed to get rid of this instability in band structure. (Figure 2)
But in DOS file there are still negative terms, though quite small.
To finally get rid of the unstable part I decided to change parameter dossmearing in gaussian smearing method and I encountered a peculiarity, namely, DOS does not change at all with smearing. Standard value is of -5 power of magnitude in Ha. I tried to lower it up to -8 power and increase up to just 1 and no difference at all.
Also, tetrahedron did not help me with the problem.
So, can you tell me what can I do wrong? or give an advice how to proceed.
Attachments
Figure 1.
Figure 1.
orig.png (24.52 KiB) Viewed 18653 times
Figure 2.
Figure 2.
12.png (24.1 KiB) Viewed 18653 times

mverstra
Posts: 655
Joined: Wed Aug 19, 2009 12:01 pm

Re: Phonon DOS problem

Post by mverstra » Thu Oct 12, 2017 7:36 am

1) if the dos still goes to negative frequencies with a finite value (not just the tails of the gaussian) you have unstable q points elsewhere in the bz which f di not show up in the bs

2) in the tetrahedron case there is no smearing at all that's the point.

As to why the dos does not change with the smearing this is strange indeed. Check if there are other input variables which might do something similar. Could be that dosmear is used for another part of the code
Matthieu Verstraete
University of Liege, Belgium

CHILL
Posts: 25
Joined: Sun Oct 08, 2017 1:37 pm

Re: Phonon DOS problem

Post by CHILL » Wed Oct 18, 2017 1:17 pm

Dear mverstra,

I checked smearing again, it works. I was confused because when I ran anaddb with different smearings a value of smearing in the heading of DOS file was always the same. (I guess the heading of the file is not to be overwrited )
I checked for instabilities around Gamma point. Here are my best results (Figure 1). Still instabilities are present. Whereas point Gamma lies exactly at 0.
Do you think it can be due to a probable lack of relaxation or the subgrid should be larger?

Best regards, Mikhail
Attachments
Figure 1
Figure 1
TETRAHEDRON.png (77.32 KiB) Viewed 18640 times

mverstra
Posts: 655
Joined: Wed Aug 19, 2009 12:01 pm

Re: Phonon DOS problem

Post by mverstra » Fri Oct 27, 2017 12:39 pm

Hi again,

those ifcsph are already quite small (18 bohr). From the ngqpt you use you can tell the maximum range of your force constants (1./(smallest q * gprim)), start with that and decrease progressively. If nothing works you probably really need more q points, make a denser grid or use the double grid technique (see qrefine and related variables) to add a systematic grid of q near to Gamma.

In the end, it could also be correct - try doing a single shot small q calculation to see of the frequencies are really positive.

cheers
Matthieu Verstraete
University of Liege, Belgium

CHILL
Posts: 25
Joined: Sun Oct 08, 2017 1:37 pm

Re: Phonon DOS problem

Post by CHILL » Wed Nov 01, 2017 11:22 pm

Dear mverstra,

Thank you for your response!

I managed to to get rid of any negative frequencies in my phonon structure calculation (the one before anaddb) . (higher ecut in combination with a better relaxation and ecutsm worked, though it is really sensitive)
Anyway, neither of my q points have negative frequencies and the grid is quite dense (20x20x1). Anaddb results shows good band structure, but there are still negatives in PHDOS file and I cannot understand where did they come from.
As I did above, I studied vicinity of the Gamma - point with qph1l list. Results are depicted on Figure 1. Sorry, it looks ugly, but there are no negatives there clearly.
So can you tell me please, where could those negative frequencies come from?

Best regards, Mikhail
Attachments
Figure 1
Figure 1
TTH.png (41.68 KiB) Viewed 18593 times

mverstra
Posts: 655
Joined: Wed Aug 19, 2009 12:01 pm

Re: Phonon DOS problem

Post by mverstra » Thu Nov 02, 2017 9:47 am

Hi again,

have you tried tetrahedron DOS? As before, if you use gaussians they have a tail and the DOS conservatively calculates these down to some offset below 0, even if there are no imaginary frequencies. This can be safely ignored

Otherwise you can print out the values on the full grid to determine where these modes are (forensics). Set up the n1qpt list with the coordinates of the points on a complete mesh (or at least a homogeneous irreducible wedge) dense enough to find your culprits.
Matthieu Verstraete
University of Liege, Belgium

CHILL
Posts: 25
Joined: Sun Oct 08, 2017 1:37 pm

Re: Phonon DOS problem

Post by CHILL » Thu Nov 02, 2017 10:53 am

Dear mverstra,

Yes, I used tetrahedron above.
I have the problem with first acoustic mode, so the only negative it can attain should be at/near the Gamma - point. That is why I studied it with very dense mesh grid around Gamma. (Figure 1 above) The lowest "square" of points lies exactly at zero due to the acoustic sum rule . And then it is just monotonous increasing function within the Brillouin zone. So, I cannot see any possible problems.

The headline of the DOS file gives:

#
# Phonon density of states and atom type projected DOS
#
# Energy in Ha , DOS in states/Ha
# Gaussian method with smearing = 2.27816763E-05Ha , nqibz =154
#
# omega PHDOS INT_PHDOS PJDOS[atom_type=1] INT_PJDOS[atom_type1] ...
#


And the problematic frequencies are :

-1.10000000E-05 3.56217872E+01 6.08643017E-05 1.84641241E+01 3.15482773E-05 1.63911795E+01 2.80063909E-05 7.66483538E-01 1.30963349E-06
-6.44366475E-06 4.13147132E+01 2.80481933E-04 2.14216431E+01 1.45419829E-04 1.90035285E+01 1.29023912E-04 8.89541565E-01 6.03819134E-06
-1.88732949E-06 4.70857004E+01 4.86081076E-04 2.44232704E+01 2.52063745E-04 2.16478479E+01 2.23548979E-04 1.01458210E+00 1.04683519E-05

So there is some Gaussian smearing. And the problematic frequencies are of the order of above mentioned smearing.
I always ignore the smearing in the file, because in the tetrahedron it is irrelevant. But probably I make a mistake?
On the Figure I plotted the first acoustic band on the whole BZ.

Best regards, Mikhail
Attachments
Figure 1. Acoustic mode
Figure 1. Acoustic mode
Band_1.png (24.11 KiB) Viewed 18591 times

mverstra
Posts: 655
Joined: Wed Aug 19, 2009 12:01 pm

Re: Phonon DOS problem

Post by mverstra » Thu Nov 02, 2017 11:00 am

"Gaussian method with smearing" = prtdos 1. Set it to 2 for the tetrahedron, and yes, the tail of the DOS is probably negligible.
Matthieu Verstraete
University of Liege, Belgium

CHILL
Posts: 25
Joined: Sun Oct 08, 2017 1:37 pm

Re: Phonon DOS problem

Post by CHILL » Thu Nov 02, 2017 1:55 pm

Dear mverstra,

Negatives are still there, but I will just ignore them then.

mverstra
Posts: 655
Joined: Wed Aug 19, 2009 12:01 pm

Re: Phonon DOS problem  [SOLVED]

Post by mverstra » Thu Nov 02, 2017 1:58 pm

with tetrahedron dos you still have DOS at negative frequencies? It would be best to print out all of the frequencies for points on the DOS grid, to understand what is going on. You may still have an instability elsewhere in the BZ, though it would be very slight (few cm-1).
Matthieu Verstraete
University of Liege, Belgium

CHILL
Posts: 25
Joined: Sun Oct 08, 2017 1:37 pm

Re: Phonon DOS problem

Post by CHILL » Thu Nov 02, 2017 3:08 pm

But that is what I did on Figure 1.(with blue dots) I used very dense mesh around Gamma and there is nothing negative at all.
And everything else is not even close to 0.
For DOS I used 40x40x1 grid:
#
# Phonon density of states and atom type projected DOS
#
# Energy in Ha , DOS in states/Ha
# Tetrahedron method, nqibz= 154
#
# omega PHDOS INT_PHDOS PJDOS[atom_type=1] INT_PJDOS[atom_type1] ...
#
-1.10000000E-05 3.99551488-142 0.00000000E+00 3.94753616-142 0.00000000E+00 2.85448797-144 0.00000000E+00 1.94338380-144 0.00000000E+00
-6.44366475E-06 1.49147208-140 0.00000000E+00 1.47355836-140 0.00000000E+00 1.06589482-142 0.00000000E+00 7.25476887-143 0.00000000E+00
-1.88732949E-06 5.45724642-139 0.00000000E+00 5.39168641-139 0.00000000E+00 3.90136667-141 0.00000000E+00 2.65463419-141 0.00000000E+00

And there is not a single negative number in my PHFRQ file. (I generated the very same 40x40x1 grid there )
It just gives me not 0 at zero frequency in the Eliashberg function and that causes further calculation problems.

Best regards, Mikhail

mverstra
Posts: 655
Joined: Wed Aug 19, 2009 12:01 pm

Re: Phonon DOS problem

Post by mverstra » Thu Nov 02, 2017 5:15 pm

Near gamma things are clear, and now so is your DOS:

the second column is actually 3.99551488E-142 = 0

everything is fine except the printing of that column, which can not handle exponents greater than E-99

The starting frequency is a fixed buffer for the Gaussian case.

ciao
Matthieu Verstraete
University of Liege, Belgium

CHILL
Posts: 25
Joined: Sun Oct 08, 2017 1:37 pm

Re: Phonon DOS problem

Post by CHILL » Mon Nov 06, 2017 11:24 am

Dear Matthieu,

Thank you for your reply!!
But I still have a question about Eliashberg function.
In A2F file the header says:

#
# ABINIT package : a2f file
#
# a2f function integrated over the FS. omega in a.u.
# number of kpoints integrated over : 1600
# number of energy points : 400
# between omega_min = 0.000000E+00 Ha and omega_max = 0.510175E-02 Ha
# and the smearing width for gaussians is 0.200000E-04
# The DOS at Fermi level for spin 1 is 0.200928E+02

It gives a Gaussian smearing value too, can I change it to tetrahedron somehow?

The problem is I used telphint = 0 and I thought, that it automatically gives all results in terms of Tetrahedron method. But apparently not.
Do these two variables prtdos and telphint contradict ?

Best regards, Mikhail

mverstra
Posts: 655
Joined: Wed Aug 19, 2009 12:01 pm

Re: Phonon DOS problem

Post by mverstra » Mon Nov 06, 2017 8:58 pm

telphint 0 should do it - the epc and DOS calculations are separate parts of the code, indeed (no need of the electron phonon coupling to calculate the phonon DOS).

The gaussian smearing variable is printed even though it is not used... ugly, I agree. Anaddb should also print out a PHDOS file, with the dos calculated using the same integration scheme as the a2F.
Matthieu Verstraete
University of Liege, Belgium

CHILL
Posts: 25
Joined: Sun Oct 08, 2017 1:37 pm

2D phonon acoustic sum rule

Post by CHILL » Thu Nov 23, 2017 3:53 pm

Dear Matthieu,

I have another question regarding phonons. I'm trying to calculate phonons for 2D structure using Acoustic sum rule.
I attached two phonon band-structures: with and without ASR.
On the picture with asr = 0, first band looks more like parabola in the vicinity of GAMMA point (First and last points on x-axis) what I would expect for 2D material, like in graphene for example.
But with acoustic sum rule the picture looks less like parabola. So my question is, whether the acoustic sum rule is suitable for 2D structures or not?

Best regards, Mikhail
Attachments
With ASR
With ASR
ASR.png (18.2 KiB) Viewed 18456 times
Without ASR
Without ASR
NoASR.png (17.29 KiB) Viewed 18456 times

mverstra
Posts: 655
Joined: Wed Aug 19, 2009 12:01 pm

Re: Phonon DOS problem

Post by mverstra » Fri Nov 24, 2017 10:05 am

Indeed this is an important issue we are working on. In 2D (or 1D) you have to impose the _rotational_ sum rule as well, and that gives the parabolic band. For the moment in abinit this is not done (there is a version but it has numerical issues converging). We have a beta version from Olle Hellman, which works perfectly and will go into the abinit source code soon.

NB: this effect is very material dependent, and in many cases if the ASR correction is not too strong (check your pseudos) the parabolic band is decently reproduced.

M.
Matthieu Verstraete
University of Liege, Belgium

CHILL
Posts: 25
Joined: Sun Oct 08, 2017 1:37 pm

Re: Phonon DOS problem

Post by CHILL » Fri Dec 01, 2017 2:40 pm

Dear Matthieu,

Thank you a lot for your reply!
Can you, please, answer two questions about abinit:
1. In anaddb there is a variable called ndivsm, that should be responsible for segmentation of paths for phonon band structure, but adding it to input file changes nothing and segmentation is still determined automatically.
Is there a way to make it work?
2. Also I have negative values for the Eliashberg function and electron-phonon coupling along different bands (teph_5.out_ep_LAMBDA file), which is by definition is not possible?
what can be wrong with that?


Best regards, Mikhail

mverstra
Posts: 655
Joined: Wed Aug 19, 2009 12:01 pm

Re: Phonon DOS problem

Post by mverstra » Thu Sep 06, 2018 1:40 pm

Hi Mikhail,

I believe we discussed this offline, but for the record:
1) the ndivsm variable works, at least in the latest abinit/anaddb versions.
2) The negative values come from the interpolation of the |g_kk'|^2 by a Fourier transform. Though the values on the FT grid are positive, the interpolation can go negative. Sometimes this can be ignored, if it is a small effect, and sometimes it's too strong, at which point you either need more q points in the initial grid, or another interpolation scheme (this is in the works in another part of abinit).
Matthieu Verstraete
University of Liege, Belgium

Locked