GULP optimization and dispersion of CNT

Dear GULP community,
I am working on obtaining the phonon dispersion for a carbon nanotube with chirality (10,10). I relaxed the structure under constant pressure using the optimized Tersoff potential (Lindsay 2010). Afterward, I calculated the phonon dispersion but encountered a negative frequency at the Gamma point.

  • Has my system truly reached its energy minimum? If not, how can I achieve that?

  • if the problem isn’t related to the optimization procedure, how can I address this issue?

optimise conp comp phonon  dispersion eigenvector  

name cnt  

vectors  
100 	 0 	 0   
 0 	 100 	 0   
 0 	 0 	 2.456  
   
# **************     system        ************* #  
  
cartesian region 1
C core  21.78000000  15.00000000   0.00000000 0.0 1.0
C core  21.63200000  16.41000000   0.00000000 0.0 1.0
C core  21.44800000  17.09500000   1.228 0.0 1.0
C core  20.87100000  18.39000000   1.228 0.0 1.0
C core  20.48500000  18.98500000   0.00000000 0.0 1.0
C core  19.53700000  20.03800000   0.00000000 0.0 1.0
C core  18.98500000  20.48500000   1.228 0.0 1.0
C core  17.75800000  21.19400000   1.228 0.0 1.0
C core  17.09500000  21.44800000   0.00000000 0.0 1.0
C core  15.70900000  21.74300000   0.00000000 0.0 1.0
C core  15.00000000  21.78000000   1.228 0.0 1.0
C core  13.59000000  21.63200000   1.228 0.0 1.0
C core  12.90500000  21.44800000   0.00000000 0.0 1.0
C core  11.61000000  20.87100000   0.00000000 0.0 1.0
C core  11.01500000  20.48500000   1.228 0.0 1.0
C core   9.96200000  19.53700000   1.228 0.0 1.0
C core   9.51500000  18.98500000   0.00000000 0.0 1.0
C core   8.80600000  17.75800000   0.00000000 0.0 1.0
C core   8.55200000  17.09500000   1.228 0.0 1.0
C core   8.25700000  15.70900000   1.228 0.0 1.0
C core   8.22000000  15.00000000   0.00000000 0.0 1.0
C core   8.36800000  13.59000000   0.00000000 0.0 1.0
C core   8.55200000  12.90500000   1.228 0.0 1.0
C core   9.12900000  11.61000000   1.228 0.0 1.0
C core   9.51500000  11.01500000   0.00000000 0.0 1.0
C core  10.46300000   9.96200000   0.00000000 0.0 1.0
C core  11.01500000   9.51500000   1.228 0.0 1.0
C core  12.24200000   8.80600000   1.228 0.0 1.0
C core  12.90500000   8.55200000   0.00000000 0.0 1.0
C core  14.29100000   8.25700000   0.00000000 0.0 1.0
C core  15.00000000   8.22000000   1.228 0.0 1.0
C core  16.41000000   8.36800000   1.228 0.0 1.0
C core  17.09500000   8.55200000   0.00000000 0.0 1.0
C core  18.39000000   9.12900000   0.00000000 0.0 1.0
C core  18.98500000   9.51500000   1.228 0.0 1.0
C core  20.03800000  10.46300000   1.228 0.0 1.0
C core  20.48500000  11.01500000   0.00000000 0.0 1.0
C core  21.19400000  12.24200000   0.00000000 0.0 1.0
C core  21.44800000  12.90500000   1.228 0.0 1.0
C core  21.74300000  14.29100000   1.228 0.0 1.0

element 
mass C 12.01 
end 


# ************** Optimized tersoff potential ************* # 

species 
C core 0.0

boonebody
C  core 0.0 0.00000015724 0.72751 0.72751 
botwobody
C  core C  core 1393.600 430.0 3.48790 2.2119  1.8 2.1
borepulsive
C  core 3 1.0 2.2119 
boattractive theta
C  core 3 1.0 2.2119  38049.0  4.3484 -0.930 
 
shrink 0 0 1

dispersion 1 10
0.0 0.0 0.0  to 0.0 0.0 0.5 

# **************     output        ************* #  

output xyz opt_sys 
output xr opt_sys
output phonon dos
output eig eigen_vec

# ********************************************** #

eigen_vec.eig (2.9 MB)

system.gout (3.4 MB)

Thanks & Regards,
Vivekkumar Panneerselvam

Hi Vivekkumar
An imaginary frequency (which GULP writes as negative for simplicity) means your structure is unstable and wants to lower its symmetry along the eigenvector direction. You can use the “lower” keyword to displace the atoms in a way to break the symmetry & then restart the optimisation using “rfo” which checks the Hessian structure for stability.
Regards,
Julian

1 Like

Dear Julian,
Thank you for your response. I used the lower_symmetry command along with RFO, and as expected, the negative frequency disappeared. Instead of breaking the symmetry, I modified the lattice vectors from (60, 60, 2.456) to (30, 30, 2.456). This change in the lattice vectors influence the frequency. My CNT’s z-axis is approximately aligned at (14 Å, 14 Å). I systematically calculated the final energy of my system for different lattice vectors, and noticed that the energy increases and then stabilizes at a constant value.

Lattice vector        totalenergy (eV)    frequency ( @ Gamma)
100 100 2.456        -316.9898896795      -0.061414
60 60 2.456          -316.9898896795      -0.061426
50 50 2.456          -316.9898896489      -0.000016
40 40 2.456          -316.9898896489      -0.000014
30 30 2.456          -316.9898896489      -0.000018

With lower_symmetry and rfo method
60 60 2.456          -316.9898896796      -0.000336
30 30 2.456          -316.9898896795      -0.000009

Here, it can be observed that the final enegry for the RFO method and lattice vectors above 60Å (without breaking symmetry )are identical, except for the frequency. Based on the above data, it can be concluded that the minimized structure was achieved using the RFO optimization.

  • However, my question is: despite lattice vectors larger than 60 Å having the minimum energy, why does the structure remain unstable?
  • My second question is related to the supercell method. Suppose I have a 1 x 1 x N supercell (where N represents the number of unit cells). During the dispersion calculation, which option should I enable to obtain 3n phonon modes (where n is the number of atoms per unit cell) instead of 3Nn modes?

Thanks & regards,
Vivek

Hi Vivek,
The problem with what you’re doing is varying the cell parameters for a model that is short-range only & so there is no interaction across the vacuum part between images. Therefore it’s not sensible to optimise these directions. It would be smarter to use 1-D boundary boundary conditions for a 1-D tube model.
Using a N times supercell you will always get 3Nn modes. If you want 3n modes then use a single cell & k points.
Regards,
Julian

1 Like

Dear Julian,
Thank you for your suggestion. Following your advice, I conducted the 1D relaxation for my quasi-1D system. After adjusting the ftol to 1e-6, the negative frequency was eliminated. The Total lattice energy is now -316.98988968 eV, which aligns with my earlier analysis using the RFO method.

system.gout (3.8 MB)

With regards,
Vivek

Dear Julian,
From my above 1D dispersion calculation (with the CNT being periodic along the x-direction), I noticed that if the eigenvectors along the x-direction are non-zero, then the eigenvectors along the y and z directions are zero, or vice versa. This pattern holds for all k-points except at k=0 and k=0.5. However, when I treat the system as 3D, this behavior doesn’t occur.

  • Could you please suggest which approach is more suitable for determining the dispersion and eigenvectors at various k-points for a quasi-1D system?

Thanks & regards,
Vivek

Dear Vivek,
It makes perfect sense that the motions along the tube axis and normal to it are relatively decoupled because one direction is bond stretching/compression & the other angle bending/out of plane motion. It’s also important to remember that when phonon modes are degenerate (typically happens at high symmetry points such as gamma) then the mixing of the phonon eigenvectors becomes arbitrary. Hopefully this will help you interpret the modes.
Regards,
Julian

Dear Julian,
Thank you for the clarification.

With regards,
Vivek