Phonon dispersion convergence in Bi2Te3 with s-tdep

[Vahid999]

I have run s-tdep with QE on Bi2Te3 at 300K up to 13 iterations. Each iteration has 50 configurations. I have mixed iterations 13 and12, 12 and 11, and 11 and 10 and so on for force constant calculations. My cutoff is converged. However, the phonon dispersion does not converge as attached. The supercells have 215 atoms and the k-grid is 2x2x2. Increasing it to 3x3x3 does not improve the dispersion convergence. Is this behaviour expected or there are tricks in order to converge the dispersion?

Thanks,
Vahid

phband10+11.pdf

phband11+12.pdf

phband12+13.pdf

[mjv500]

Hi Vahid, Not clear exactly what you expect, the difference is small. This said you have a huge number of configurations, so indeed it should be fully converged. I would expect that a few dozen (in total) would suffice. You might check if you are close to the correct T dependent minimum of the PES, with the first order flag. Or, you can get rid of the oldest configurations, to improve the sampling. Anyhow, your phonons don't look so bad Matthieu ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Professor Matthieu J Verstraete Fellow, American Physical Society <https://www.aps.org> Chair, Steering Committee, European Theoretical Spectroscopy Facility <https://www.etsf.eu> Alumnus Fellow, Young Academy of Europe <https://www.yacadeuro.org> <http://yacadeuro.org/> Institute for Theoretical Physics (ITP) Department of Physics Buys Ballot Gebouw/Building, Princetonplein 5, office 7.85 University of Utrecht, 3584 CC Utrecht Phone: +31 30 253 5926 Secretariat: +31 30 253 5929 and ***@***.*** Group web page (Liège): http://www.nanomat.ulg.ac.be/ Nanomat lab, Q-Mat center, Université de Liège Département de Physique, Bat. B5a, office 4/50 Allée du 6 août, 19 B-4000 Sart Tilman, Liège Belgium Phone : +32 4 366 90 17 European Theoretical Spectroscopy Facility <https://www.etsf.eu> (ETSF) Mail : ***@***.*** ***@***.*** ***@***.***

…

On Thu, Dec 11, 2025, 20:04 Vahid999 ***@***.***> wrote: *Vahid999* created an issue (tdep-developers/tdep#162) <#162> I have run s-tdep with QE on Bi2Te3 at 300K up to 13 iterations. Each iteration has 50 configurations. I have mixed iterations 13 and12, 12 and 11, and 11 and 10 and so on for force constant calculations. My cutoff is converged. However, the phonon dispersion does not converge as attached. The supercells have 215 atoms and the k-grid is 2x2x2. Increasing it to 3x3x3 does not improve the dispersion convergence. Is this behaviour expected or there are tricks in order to converge the dispersion? Thanks, Vahid phband10+11.pdf <https://github.com/user-attachments/files/24110217/phband10%2B11.pdf> phband11+12.pdf <https://github.com/user-attachments/files/24110218/phband11%2B12.pdf> phband12+13.pdf <https://github.com/user-attachments/files/24110220/phband12%2B13.pdf> — Reply to this email directly, view it on GitHub <#162>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABWOT362ITG2VKV6ZYLHFL34BG53BAVCNFSM6AAAAACOYSJHG2VHI2DSMVQWIX3LMV43ASLTON2WKOZTG4ZDANRRGU2TQMY> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[Vahid999]

Hi Professor Verstraete, Thank you for the quick response. I was expecting the dispersions to match from one iteration to the next. Also, since LO-TO is not included, I expect the top two phonons at Gamma to be degenerate. The input structure is actually not relaxed with QE but rather the experimental structure at 300 K with atoms relaxed. Using the first order flag changed the unit cell from the experimental value of cell 1.000000000000 2.19309263036978 -1.26618262050178 10.15655859709168 0.00000000000000 2.53236524100356 10.15655859709168 -2.19309263036978 -1.26618262050178 10.15655859709168 Bi Te 2 3 Direct coordinates 0.40035000000000 0.40035000000000 0.40035000000000 site 1 species 1: Bi 0.59965000000000 0.59965000000000 0.59965000000000 site 2 species 1: Bi 0.20882000000000 0.20882000000000 0.20882000000000 site 3 species 2: Te 0.79118000000000 0.79118000000000 0.79118000000000 site 4 species 2: Te 0.00000000000000 0.00000000000000 0.00000000000000 site 5 species 2: Te to cell 10.467500000000 0.20951446194123 -0.12096323100089 0.97029458773267 0.00000000000000 0.24192646200177 0.97029458773267 -0.20951446194123 -0.12096323100089 0.97029458773267 Bi Te 2 3 Direct coordinates 0.39947891296370 0.39947891296370 0.39947891296370 site 1 species 1: Bi 0.60052108703630 0.60052108703630 0.60052108703630 site 2 species 1: Bi 0.21088656501776 0.21088656501776 0.21088656501776 site 3 species 2: Te 0.78911343498224 0.78911343498224 0.78911343498224 site 4 species 2: Te 0.00000000000000 0.00000000000000 0.00000000000000 site 5 species 2: Te The largest difference is the change in the z-component of the lattice vectors. I am running iteration 14 with 30 configurations and the new unit cells from the first order flag. Thanks, Vahid On Dec 11, 2025, at 4:13 PM, Matthieu Verstraete ***@***.***> wrote: CAUTION: The Sender of this email is not from within Dalhousie. [https://avatars.githubusercontent.com/u/7137775?s=20&v=4]mjv500 left a comment (tdep-developers/tdep#162)<#162 (comment)> Hi Vahid, Not clear exactly what you expect, the difference is small. This said you have a huge number of configurations, so indeed it should be fully converged. I would expect that a few dozen (in total) would suffice. You might check if you are close to the correct T dependent minimum of the PES, with the first order flag. Or, you can get rid of the oldest configurations, to improve the sampling. Anyhow, your phonons don't look so bad Matthieu ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Professor Matthieu J Verstraete Fellow, American Physical Society <https://www.aps.org> Chair, Steering Committee, European Theoretical Spectroscopy Facility <https://www.etsf.eu> Alumnus Fellow, Young Academy of Europe <https://www.yacadeuro.org> <http://yacadeuro.org/> Institute for Theoretical Physics (ITP) Department of Physics Buys Ballot Gebouw/Building, Princetonplein 5, office 7.85 University of Utrecht, 3584 CC Utrecht Phone: +31 30 253 5926 Secretariat: +31 30 253 5929 and ***@***.*** Group web page (Liège): http://www.nanomat.ulg.ac.be/ Nanomat lab, Q-Mat center, Université de Liège Département de Physique, Bat. B5a, office 4/50 Allée du 6 août, 19 B-4000 Sart Tilman, Liège Belgium Phone : +32 4 366 90 17 European Theoretical Spectroscopy Facility <https://www.etsf.eu> (ETSF) Mail : ***@***.*** ***@***.*** ***@***.***

On Thu, Dec 11, 2025, 20:04 Vahid999 ***@***.***> wrote: *Vahid999* created an issue (tdep-developers/tdep#162) <#162> I have run s-tdep with QE on Bi2Te3 at 300K up to 13 iterations. Each iteration has 50 configurations. I have mixed iterations 13 and12, 12 and 11, and 11 and 10 and so on for force constant calculations. My cutoff is converged. However, the phonon dispersion does not converge as attached. The supercells have 215 atoms and the k-grid is 2x2x2. Increasing it to 3x3x3 does not improve the dispersion convergence. Is this behaviour expected or there are tricks in order to converge the dispersion? Thanks, Vahid phband10+11.pdf <https://github.com/user-attachments/files/24110217/phband10%2B11.pdf> phband11+12.pdf <https://github.com/user-attachments/files/24110218/phband11%2B12.pdf> phband12+13.pdf <https://github.com/user-attachments/files/24110220/phband12%2B13.pdf> — Reply to this email directly, view it on GitHub <#162>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABWOT362ITG2VKV6ZYLHFL34BG53BAVCNFSM6AAAAACOYSJHG2VHI2DSMVQWIX3LMV43ASLTON2WKOZTG4ZDANRRGU2TQMY> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

— Reply to this email directly, view it on GitHub<#162 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/ARHYUBCPMYTLPL5BMXBMYGL4BHF7JAVCNFSM6AAAAACOYSJHG2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTMNBTGYYTMMZTGQ>. You are receiving this because you authored the thread.Message ID: ***@***.***>

[Vahid999]

I just realized that the multiplication factors (line 2 in POSCAR) are different for the two POSCARs and the lattice vectors are unchanged. Only the atoms are slightly moved. Vahid On Dec 11, 2025, at 4:52 PM, Vahid Askarpour ***@***.***> wrote: Hi Professor Verstraete, Thank you for the quick response. I was expecting the dispersions to match from one iteration to the next. Since LO-TO is not included, I was also expecting the top two phonon modes at Gamma to be degenerate. The input structure is actually not relaxed with QE but rather the experimental structure at 300 K with atoms relaxed. Using the first order flag changed the unit cell from the experimental value of cell 1.000000000000 2.19309263036978 -1.26618262050178 10.15655859709168 0.00000000000000 2.53236524100356 10.15655859709168 -2.19309263036978 -1.26618262050178 10.15655859709168 Bi Te 2 3 Direct coordinates 0.40035000000000 0.40035000000000 0.40035000000000 site 1 species 1: Bi 0.59965000000000 0.59965000000000 0.59965000000000 site 2 species 1: Bi 0.20882000000000 0.20882000000000 0.20882000000000 site 3 species 2: Te 0.79118000000000 0.79118000000000 0.79118000000000 site 4 species 2: Te 0.00000000000000 0.00000000000000 0.00000000000000 site 5 species 2: Te to cell 10.467500000000 0.20951446194123 -0.12096323100089 0.97029458773267 0.00000000000000 0.24192646200177 0.97029458773267 -0.20951446194123 -0.12096323100089 0.97029458773267 Bi Te 2 3 Direct coordinates 0.39947891296370 0.39947891296370 0.39947891296370 site 1 species 1: Bi 0.60052108703630 0.60052108703630 0.60052108703630 site 2 species 1: Bi 0.21088656501776 0.21088656501776 0.21088656501776 site 3 species 2: Te 0.78911343498224 0.78911343498224 0.78911343498224 site 4 species 2: Te 0.00000000000000 0.00000000000000 0.00000000000000 site 5 species 2: Te The largest difference is the change in the z-component of the lattice vectors. I am running iteration 14 with 30 configurations and the new unit cells from the first order flag. Thanks, Vahid On Dec 11, 2025, at 4:13 PM, Matthieu Verstraete ***@***.***> wrote: CAUTION: The Sender of this email is not from within Dalhousie. [https://avatars.githubusercontent.com/u/7137775?s=20&v=4]mjv500 left a comment (tdep-developers/tdep#162)<#162 (comment)> Hi Vahid, Not clear exactly what you expect, the difference is small. This said you have a huge number of configurations, so indeed it should be fully converged. I would expect that a few dozen (in total) would suffice. You might check if you are close to the correct T dependent minimum of the PES, with the first order flag. Or, you can get rid of the oldest configurations, to improve the sampling. Anyhow, your phonons don't look so bad Matthieu ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Professor Matthieu J Verstraete Fellow, American Physical Society <https://www.aps.org> Chair, Steering Committee, European Theoretical Spectroscopy Facility <https://www.etsf.eu> Alumnus Fellow, Young Academy of Europe <https://www.yacadeuro.org> <http://yacadeuro.org/> Institute for Theoretical Physics (ITP) Department of Physics Buys Ballot Gebouw/Building, Princetonplein 5, office 7.85 University of Utrecht, 3584 CC Utrecht Phone: +31 30 253 5926 Secretariat: +31 30 253 5929 and ***@***.*** Group web page (Liège): http://www.nanomat.ulg.ac.be/ Nanomat lab, Q-Mat center, Université de Liège Département de Physique, Bat. B5a, office 4/50 Allée du 6 août, 19 B-4000 Sart Tilman, Liège Belgium Phone : +32 4 366 90 17 European Theoretical Spectroscopy Facility <https://www.etsf.eu> (ETSF) Mail : ***@***.*** ***@***.*** ***@***.***

On Thu, Dec 11, 2025, 20:04 Vahid999 ***@***.***> wrote: *Vahid999* created an issue (tdep-developers/tdep#162) <#162> I have run s-tdep with QE on Bi2Te3 at 300K up to 13 iterations. Each iteration has 50 configurations. I have mixed iterations 13 and12, 12 and 11, and 11 and 10 and so on for force constant calculations. My cutoff is converged. However, the phonon dispersion does not converge as attached. The supercells have 215 atoms and the k-grid is 2x2x2. Increasing it to 3x3x3 does not improve the dispersion convergence. Is this behaviour expected or there are tricks in order to converge the dispersion? Thanks, Vahid phband10+11.pdf <https://github.com/user-attachments/files/24110217/phband10%2B11.pdf> phband11+12.pdf <https://github.com/user-attachments/files/24110218/phband11%2B12.pdf> phband12+13.pdf <https://github.com/user-attachments/files/24110220/phband12%2B13.pdf> — Reply to this email directly, view it on GitHub <#162>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABWOT362ITG2VKV6ZYLHFL34BG53BAVCNFSM6AAAAACOYSJHG2VHI2DSMVQWIX3LMV43ASLTON2WKOZTG4ZDANRRGU2TQMY> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

— Reply to this email directly, view it on GitHub<#162 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/ARHYUBCPMYTLPL5BMXBMYGL4BHF7JAVCNFSM6AAAAACOYSJHG2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTMNBTGYYTMMZTGQ>. You are receiving this because you authored the thread.Message ID: ***@***.***>

[OrbitalC]

Hello Vahid

What you are observing is probably just a problem of noise in computing average. With 50 configurations per iterations you probably have a lot of variances. So your phonons are probably converged, but with a lot of uncertainty.
To reduce this uncertainty, you can simply increase the number of configurations.

A way to guarantee convergence with respect to the number of configurations is by doubling the number of configuration at each iteration.
If you start with just 1 configuration, you have 2 at the second iteration, then 4, 8, 16, 32, 64, ...
So as your phonons are getting better, you increase the number of configurations. This avoid throwing too much configurations at the beginning when the IFC are far from convergence.
Of course you can start with more configurations (it can help if your system is very anharmonic with unstabilities or if you start from already good phonons, like some computed at a different temperature)

Also, the mixing of the successive iterations is a good idea as it help the convergence of the full cycle.

Aloïs