Unable to reproduce phonon dispersion in QuantumATK

[krabidix]

Dear QuantumATK experts,

I tried to replicate the phonon band structure of a simple system (hexagonal monolayer of Cu) as done in https://doi.org/10.1155/2016/8429510 (see Fig 3 or attached image Phonon_bs_ref.PNG)
The results from QuantumATK have imaginary frequencies and even after tuning parameters (e.g: max_interaction_range, atomic_displacement) that control the force constants, I am not able to vanish the imaginary frequencies (see attached Phonon_bs_qatk.PNG).

Please, comment on how can I rid of these imaginary frequencies.


The input file is also attached.


Thanks!
krabidix

[krabidix]

Is there anyone who can comment?

[Tue Gunst]

Hi Krabidix,
Did you use a force-field or DFT? Also I assume that you optimized the structure and used the same 7 × 7 supercell?
Tue

[krabidix]

Hi Tue,
Thanks for the comment!
I used the DFT-PW method, and I have also shared the .py file already.
First I optimized the structure and then use repetitions=(7, 7, 1).
You may consider a look at the script.


Best Regards,
krabidix

[Tue Gunst]

Attaching a quick check with LCAO settings. Takes a few minutes and reproduces the result from the paper well.
I am running the PW calculation but that will take some hours.
One thing that I notice is that you have set use_wigner_seitz_scheme=False.
This is non-default and not recommended since you will need very large supercell if not allowing the symmetry conserving copying of periodic data in the supercell.
Tue

[krabidix]

Hi,
Thanks for working on the script.
Indeed the results are reproduced.
By tuning the max_interaction_range we need to set 'use_wigner_seitz_scheme=False'. QATK allows setting use_wigner_seitz_scheme=True only when we consider the max_interaction_range (Default) which includes all atoms. And with this (True) setting when one checks repetitions = checkNumberOfRepetitions(bulk_configuration) the automatic algorithm of ATK suggest a larger supercell.
So, one can limit the max_intr_range and use a smaller supercell as ATK suggests with
repetitions = checkNumberOfRepetitions(bulk_configuration,    max_interaction_range=4.0*Angstrom).
The parameters used in the dynamical matrix calculations affect too much on phonon calculations.
One thing I also noticed is that during optimization the symmetry gets broken for some 2D structures, which leads to the imaginary phonon frequencies also.
At the same time when we enforce the symmetry using constraints, the next issue came up with force and stress convergence. Which also affects the phonon energy relations.

Is the tuning of parameters the only way to check or there is a robust rule of thumb on which one can rely and move on to study the next structure if imaginary frequencies showed up?

 

[Tue Gunst]

I recommend to not set automatic repetitions, use_wigner_seitz_scheme=True, and not set a max_interaction_range.
If the max_interaction_range is lower than the range of physical coupling, the crystal symmetry is broken which probably results in negative phonons.
In that case, the wigner_seitz_scheme cannot restore it (hence it is disabled by default in the scriptor but can be set to true still in the script if needed).
Instead, use_wigner_seitz_scheme=True and converge the supercell size until physical interactions are included (converged dispersion and no negative modes).
As shown by the example, 7 repetitions seem to be sufficient with the wigner_seitz_scheme.

[krabidix]

Thanks a lot for your nice input. I will keep these things for further calculations.
I encountered numerical artifacts during optimization, for example:

Consider 2D square lattice with two atoms (A&B) basis (side of square lattice = a*sqrt(2)) where 'a' is the bond length, the atomic position in fractional coordinates A(0,0,0.5) & B(0.5,0.5,0.5). Now upon geometry optimization, the fractional coordinates remain the same but the cartesian coordinates show some non-zero values in of x and y plane for atom A, of the order of 10^-10 or even smaller time. Like atom A being at ( -10^-12, -10^9, 0.5).

Is this numerical noise could be the cause of imaginary modes?

[krabidix]

Hi,
I also want to add one more thing.
Using 'use_wigner_seitz_scheme=True', and invoking supercell sizes 3, and 5 for testing convergence.
What I got the supercell size of 3 is giving positive energies and with a larger supercell size the imaginary modes appeared. This is something unusual?
I have shared the script.

[Tue Gunst]

The numerical noise should not be the cause of the negative modes.
If you obtain negative modes at larger supercells it indicates that the supercell size was not converged yet.
One can "accidentally" have fairly good phonon bands in smaller supercells if interactions are cut off in a manner that fulfills or almost fulfills the lattice symmetry.
Tue