DFPT for phonon dispersion of monolayer 2H-WS2

[xinwuchn]

Dear all,

I hope you're well.

I'm using DFPT method to calculate the phonon dispersion of monolayer 2H-WS2.
But here are some troubles. The forces seem to be wrong.
The calculation files are attached below.
The POSCAR_uc file was after relaxation (First, ISIF=3, EDIFF=1E-06; Then, ISIF=2, EDIFF=1E-08).
And the supercell size was 6x6x1.

I'd appreciated if you could give me some advices.

Best,
Xin

[alex]

Hello Xin,

you are looking at big forces in c, e.g., for many atoms. It looks like your W-S bond is with 204 pm a bit short. Did you relax also the lattice constants? Have you thought of vdW forces and corrections here? Then, of course, DFPT, is not usable any more...

Hth

alex

PS: your forces:

POSITION TOTAL-FORCE (eV/Angst)
-----------------------------------------------------------------------------------
1.59387 0.92022 10.16021 -0.000000 -0.000000 21.586309
4.78160 0.92022 10.16021 -0.000000 -0.000000 21.586309
7.96933 0.92022 10.16021 -0.000000 -0.000000 21.586309
11.15706 0.92022 10.16021 -0.000000 -0.000000 21.586309
14.34479 0.92022 10.16021 -0.000000 -0.000000 21.586309
17.53252 0.92022 10.16021 -0.000000 -0.000000 21.586309
-0.00000 3.68087 10.16021 -0.000000 -0.000000 21.586309
3.18773 3.68087 10.16021 -0.000000 -0.000000 21.586309
6.37546 3.68087 10.16021 -0.000000 -0.000000 21.586309
9.56319 3.68087 10.16021 -0.000000 -0.000000 21.586309
12.75092 3.68087 10.16021 -0.000000 -0.000000 21.586309
15.93865 3.68087 10.16021 -0.000000 -0.000000 21.586309
-1.59387 6.44153 10.16021 -0.000000 -0.000000 21.586309
1.59387 6.44153 10.16021 -0.000000 -0.000000 21.586309
4.78160 6.44153 10.16021 -0.000000 -0.000000 21.586309
7.96933 6.44153 10.16021 -0.000000 -0.000000 21.586309
11.15706 6.44153 10.16021 -0.000000 -0.000000 21.586309
14.34479 6.44153 10.16021 -0.000000 -0.000000 21.586309
-3.18773 9.20218 10.16021 -0.000000 -0.000000 21.586309
-0.00000 9.20218 10.16021 -0.000000 -0.000000 21.586309
3.18773 9.20218 10.16021 -0.000000 -0.000000 21.586309
6.37546 9.20218 10.16021 -0.000000 -0.000000 21.586309
9.56319 9.20218 10.16021 -0.000000 -0.000000 21.586309
12.75092 9.20218 10.16021 -0.000000 -0.000000 21.586309
-4.78160 11.96284 10.16021 -0.000000 -0.000000 21.586309
-1.59387 11.96284 10.16021 -0.000000 -0.000000 21.586309
1.59387 11.96284 10.16021 -0.000000 -0.000000 21.586309
4.78160 11.96284 10.16021 -0.000000 -0.000000 21.586309
7.96933 11.96284 10.16021 -0.000000 -0.000000 21.586309
11.15706 11.96284 10.16021 -0.000000 -0.000000 21.586309
-6.37546 14.72349 10.16021 -0.000000 -0.000000 21.586309
-3.18773 14.72349 10.16021 -0.000000 -0.000000 21.586309
-0.00000 14.72349 10.16021 -0.000000 -0.000000 21.586309
3.18773 14.72349 10.16021 -0.000000 -0.000000 21.586309
6.37546 14.72349 10.16021 -0.000000 -0.000000 21.586309
9.56319 14.72349 10.16021 -0.000000 -0.000000 21.586309
1.59387 0.92022 8.37034 0.000000 0.000000 -21.586309
4.78160 0.92022 8.37034 0.000000 0.000000 -21.586309
7.96933 0.92022 8.37034 0.000000 0.000000 -21.586309
11.15706 0.92022 8.37034 0.000000 0.000000 -21.586309
14.34479 0.92022 8.37034 0.000000 0.000000 -21.586309
17.53252 0.92022 8.37034 0.000000 0.000000 -21.586309
-0.00000 3.68087 8.37034 0.000000 0.000000 -21.586309
3.18773 3.68087 8.37034 0.000000 0.000000 -21.586309
6.37546 3.68087 8.37034 0.000000 0.000000 -21.586309
9.56319 3.68087 8.37034 0.000000 0.000000 -21.586309
12.75092 3.68087 8.37034 0.000000 0.000000 -21.586309
15.93865 3.68087 8.37034 0.000000 0.000000 -21.586309
-1.59387 6.44153 8.37034 0.000000 0.000000 -21.586309
1.59387 6.44153 8.37034 0.000000 0.000000 -21.586309
4.78160 6.44153 8.37034 0.000000 0.000000 -21.586309
7.96933 6.44153 8.37034 0.000000 0.000000 -21.586309
11.15706 6.44153 8.37034 0.000000 0.000000 -21.586309
14.34479 6.44153 8.37034 0.000000 0.000000 -21.586309
-3.18773 9.20218 8.37034 0.000000 0.000000 -21.586309
-0.00000 9.20218 8.37034 0.000000 0.000000 -21.586309
3.18773 9.20218 8.37034 0.000000 0.000000 -21.586309
6.37546 9.20218 8.37034 0.000000 0.000000 -21.586309
9.56319 9.20218 8.37034 0.000000 0.000000 -21.586309
12.75092 9.20218 8.37034 0.000000 0.000000 -21.586309
-4.78160 11.96284 8.37034 0.000000 0.000000 -21.586309
-1.59387 11.96284 8.37034 0.000000 0.000000 -21.586309
1.59387 11.96284 8.37034 0.000000 0.000000 -21.586309
4.78160 11.96284 8.37034 0.000000 0.000000 -21.586309
7.96933 11.96284 8.37034 0.000000 0.000000 -21.586309
11.15706 11.96284 8.37034 0.000000 0.000000 -21.586309
-6.37546 14.72349 8.37034 0.000000 0.000000 -21.586309
-3.18773 14.72349 8.37034 0.000000 0.000000 -21.586309
-0.00000 14.72349 8.37034 0.000000 0.000000 -21.586309
3.18773 14.72349 8.37034 0.000000 0.000000 -21.586309
6.37546 14.72349 8.37034 0.000000 0.000000 -21.586309
9.56319 14.72349 8.37034 0.000000 0.000000 -21.586309
-9.56319 16.56393 9.26527 0.000000 0.000000 0.000000
-6.37546 16.56393 9.26527 0.000000 0.000000 0.000000
-3.18773 16.56393 9.26527 0.000000 -0.000000 0.000000
-0.00000 16.56393 9.26527 0.000000 0.000000 0.000000
3.18773 16.56393 9.26527 0.000000 0.000000 0.000000
6.37546 16.56393 9.26527 0.000000 0.000000 0.000000
-1.59387 2.76066 9.26527 0.000000 0.000000 0.000000
1.59387 2.76066 9.26527 -0.000000 0.000000 0.000000
4.78160 2.76066 9.26527 0.000000 -0.000000 0.000000
7.96933 2.76066 9.26527 -0.000000 -0.000000 0.000000
11.15706 2.76066 9.26527 -0.000000 -0.000000 0.000000
14.34479 2.76066 9.26527 -0.000000 0.000000 0.000000
-3.18773 5.52131 9.26527 0.000000 0.000000 0.000000
-0.00000 5.52131 9.26527 -0.000000 0.000000 0.000000
3.18773 5.52131 9.26527 -0.000000 0.000000 0.000000
6.37546 5.52131 9.26527 -0.000000 0.000000 0.000000
9.56319 5.52131 9.26527 0.000000 0.000000 0.000000
12.75092 5.52131 9.26527 -0.000000 0.000000 0.000000
-4.78160 8.28197 9.26527 -0.000000 -0.000000 0.000000
-1.59387 8.28197 9.26527 0.000000 0.000000 0.000000
1.59387 8.28197 9.26527 0.000000 -0.000000 0.000000
4.78160 8.28197 9.26527 0.000000 -0.000000 0.000000
7.96933 8.28197 9.26527 0.000000 0.000000 0.000000
11.15706 8.28197 9.26527 0.000000 -0.000000 0.000000
-6.37546 11.04262 9.26527 -0.000000 -0.000000 0.000000
-3.18773 11.04262 9.26527 0.000000 -0.000000 0.000000
-0.00000 11.04262 9.26527 -0.000000 0.000000 0.000000
3.18773 11.04262 9.26527 -0.000000 -0.000000 0.000000
6.37546 11.04262 9.26527 0.000000 -0.000000 0.000000
9.56319 11.04262 9.26527 -0.000000 -0.000000 0.000000
-7.96933 13.80328 9.26527 0.000000 -0.000000 0.000000
-4.78160 13.80328 9.26527 0.000000 -0.000000 0.000000
-1.59387 13.80328 9.26527 0.000000 -0.000000 0.000000
1.59387 13.80328 9.26527 0.000000 -0.000000 0.000000
4.78160 13.80328 9.26527 0.000000 0.000000 0.000000
7.96933 13.80328 9.26527 0.000000 -0.000000 0.000000
-----------------------------------------------------------------------------------
total drift: -0.000050 0.000030 0.000001

[xinwuchn]

Hello, Alex,

Thanks so much for your reply!

I checked my input files.
I defined the wrong vacuum layer thickness, which may lead to the wrong forces in initial run.

Thanks again!

Best,
Xin

[michael_wolloch]

Dear xinwuchn,

You write that you used ISIF = 3 initially for relaxation. This is ill-advised for a system containing vacuum since the cell might collapse. Please use ISIF = 4 instead, which keeps the cell volume constant, but allows you to still relax your lateral lattice parameters. This might be the reason why your vacuum thickness was wrong initially.

If you still have troubles after fixing your vacuum and double-checking your lattice parameters, please get back to us, and I will try to replicate your results.

Cheers, Michael

[xinwuchn]

Dear Michael,

Thanks for your advice! I really didn't pay attention to this issue.

Fortunately, after modifying the vacuum thickness, the calculation seems to be correct!
But when the supercell was 6x6x1, there were some imaginary frequency at the Gamma point.
Expanding the supercell to 8x8x1 will meet the memory limit trouble...

Thanks again!

Best,
Xin WU

[michael_wolloch]

Dear Xin WU,

I am glad my comments were helpful and your system is set up correctly now.

I would suggest using finite differences instead of DFTP if you have memory issues.

If you have not worked through our phonon tutorials yet, this is also a great resource to consult, as is our how-to page on the VASP wiki on computing the phonon dispersion and DOS.

Let me know if you have trouble with the finite differences for the 8x8x1 cell. In that case, please send me the updated input and output files, so I can try to reproduce the problem and give better advice.

Cheers, Michael

[xinwuchn]

Dear Michael,

Thanks for your great advice.
In the phonon tutorials that you recommended, I found the phonon band at Gamma in 5x5x1 supercell case seems to be better than that in 6x6x1.
Is there any basis for judgment here, or is testing the only way?

For the WS2's 8x8x1 supercell case, I used finite differences method with EDIFF=1E-08 and NELM=150, but it cannot be well converged.
The files are attached here.

Thanks again! :-)

Best,
Xin WU

[michael_wolloch]

Dearn Xin WU,

In the phonon tutorials that you recommended, I found the phonon band at Gamma in 5x5x1 supercell case seems to be better than that in 6x6x1. Is there any basis for judgment here, or is testing the only way?

Unfortunately, testing is the only way. You just have to increase the super cell size until the phonons are converged. Please note that this is especially tricky for 2D materials, as is also stated in the tutorial:

Near the Γ point, a saddle point appears when the supercell size is too small. These slightly imaginary modes might appear again for larger supercell sizes but tend to disappear when the supercell size is large enough. The convergence of these quadratic 'soft' phonon modes in 2D materials is particularly difficult and a lot of care should be taken when performing these calculations.

For the WS2's 8x8x1 supercell case, I used finite differences method with EDIFF=1E-08 and NELM=150, but it cannot be well converged.
The files are attached here.

Your calculations are very well converged, but after around 40 iterations, you get some small oscillations in energy. I would advise you to try and adjust your settings a bit:

• Try switching on Gaussian smearing with ISMEAR = 0. The default smearing width of SIGMA = 0.2 is a good starting point.

• With a large supercell, the Γ point alone should be enough for k-point sampling. This should speed up your calculations and allow you to use the faster gamma-only version of VASP. Of course, you have to test that the phonons are converged with respect to k-point sampling. Note that this is not possible for tetrahedron smearing (i.e. ISMEAR = -5).

• The largest ENMAX value in your POTCAR is 258.689. This is much smaller than the ENCUT value of 650 eV that you are using. It is highly unlikely that you need to increase ENCUT by more than 30-50% even for cell relaxations. I advise you to lower ENCUT to about 350-400 eV. This will speed up calculations, allow you to go to larger supercells, and might even help with convergence.

• If convergence is still an issue, you might try and adjusting the density mixer to initial linear mixing setting: AMIX = 0.2 and BMIX = 0.0001

In summary:
Try 9x9x1 and 10x10x1 (larger if needed) supercells with lower energy cutoff and possibly less k-points until you are not observing any more imaginary frequencies. Double check the results with higher k-point density and possibly a larger ENCUT and smaller SIGMA.

As a sidenote:
I saw that in your band.conf DIM = 6 6 1. I have not used phonopy in a while, and I am not sure if that is a problem, but it should be 8 8 1 for the current set of calculations I imagine.

Cheers, Michael

[xinwuchn]

Dear Michael,

Thanks for your quick reply and great advices, which is so helpful for me.
I'll do sufficient tests following your advices, and update the new results.

By the way, when I used Phonopy, I set the DIM=8 8 1 in command line, which will override the band.conf settings.

Best,
Xin Wu

[xinwuchn]

Dear Michael,

Thanks for your great guidance! The convergence problem has been solved.
I did some tests following your advice.

With the ENCUT=380, ISMEAR=0, SIGMA=0.2, and Γ point alone sampling, I calculated the three sets of supercell (8x8x1, 10x10x1, and 16x16x1) by DFPT. But in terms of the imaginary frequency at the Γ point, the effect of 8x8x1 is the best, which is different from what I thought. Could you pls give some insights?

Best,
Xin

[michael_wolloch]

Dear Xin,

Unfortunately, it is impossible to draw definite conclusions from your test data. It is possible that you would have to use a larger cell still to get rid of the imaginary pocket, but since you also changed the smearing, the plane-wave cutoff, and the k-point sampling at the same time, these could also have an effect.

I fear you have to test these factors separately to gain more information.

I would also suggest that you think carefully about the final goal of your calculation. Other than the pocket around the Γ-point the bands seem to be very well converged (You could plot all data into the same pdf to make sure). It is possible that the small instability is not relevant to your problem at all, e.g. if you are interested in the optical branches.

I hope this helps,
cheers, Michael

[xinwuchn]

Dear Michael,

Thanks for your guidance!
Actually, this result has been basically met my requirements.
If necessary, I will test parameter by parameter as you suggested before.

Thanks again! :-)

Best,
Xin