Hi，everyone
I’m new to gulp ,recently, I have constructed nv-diamond through vacancy and impurity. When I tried to output its phonon dispersion and density of states, the results turned out to be those of pure diamond. Furthermore, I would like to ask how to use Alamode to calculate it. As mentioned earlier, the output of Alamode is still the thermal conductivity of pure diamond
Does anyone have any related experience ?
jack
Hi Jack,
The vacancy and impurity options compute defect properties using the Mott-Littleton method, so it would be worth looking this up. This approach is aperiodic and considers defects at infinite dilution & so there is no such thing as phonon dispersion etc, which are for periodic systems. Hence if you request this, it has to be performed on the bulk & not the defective system. If you want phonon related properties for defects then you need to use the supercell method, where you place the defects in the supercell to obtain the concentration you want & then you’ll be able to compute periodic properties for the defective system.
Regards,
Julian
Hi Julian，
Thank you very much for your guidance. Since I plan to study the thermal conductivity of nv-diamond, I intend to learn and use the “super cell method” you mentioned. However, in the Example, I couldn’t find the corresponding example, and I don’t quite understand from the help that how “supercell” relates to produce defects. Can I just add the keyword ‘supercell’ while using ‘impurity’ and ‘vacancy’ ? If not how could i “place the defects in the supercell” Really be grateful for your guidance.
Jack
Hi Jack,
There is no special keyword or anything for defects via the supercell method. You just input a normal bulk unit cell in which you’ve created whatever defect structure you like. It’s called the supercell method because you may need to expand the regular unit cell to be a larger one (i.e. a supercell) in order to get the concentration of defects you want since the answer will depend on how far apart the defects are. If you want the equivalent of the Mott-Littleton method (i.e. infinite dilution) then you have to run a series of calculations with increasing sizes of cells until the defect energy converges. Of course this means that the effect on any bulk properties tends to zero as well, so if you’re looking for this then you have to work out what the real concentration of defects is in the material.
Regards,
Julian
Dear Professor,
Hello,
Following your suggestion, I wrote the following gin file（in the bottom）. May I ask if this is what you referred to as the “supercell method”? In the fractional coordinates (frac), I took a conventional unit cell of diamond, removed one C atom, and converted a neighboring C atom to an N atom, resulting in a 0.25 defect concentration of NV diamond for testing.
If I understand correctly, to obtain a larger defect system, user only need to use a 2x2x2 diamond conventional unit cell and repeat the previous operation.
- I’m thinking, if this is correct, wouldn’t the content in frac become increasingly lengthy to achieve a larger system?
- Additionally, it took me 6 hours to calculate the thermal conductivity, and in the anphon.log file,there is a warning below the phonon calculations : “OpenBLAS Warning: Detect OpenMP Loop and this application may hang. Please rebuild the library with USE_OPENMP=1 option.” I used the “runalamode” script and did not use parallel computing.
- The final calculation result was only 73.5985 W/mK (at room temperature). It is obviously much smaller than that of diamond.
May I ask if this is abnormal, or if I missed using some simplification methods in the gin file, because when I tried to calculate a bigger nv-diamond（C62N1） converted from a 2x2x2 diamond, I couldn’t even obtain _cubic.xml and cubic.fcs.
Jack
opti conp prop phon nodens alamode
name Diamond
title
test nv-dia
end
cell
3.566790 3.566790 3.566790 90.000000 90.000000 90.000000 1 1 1 1 1 1
fractional
C1 core 0.000000 0.000000 0.000000 0.000000 1.000000 0.0 1 1 1
C1 core 0.000000 0.500000 0.500000 0.000000 1.000000 0.0 1 1 1
C1 core 0.500000 0.500000 0.000000 0.000000 1.000000 0.0 1 1 1
N1 core 0.750000 0.250000 0.750000 0.000000 1.000000 0.0 1 1 1
C1 core 0.250000 0.250000 0.250000 0.000000 1.000000 0.0 1 1 1
C1 core 0.250000 0.750000 0.750000 0.000000 1.000000 0.0 1 1 1
C1 core 0.750000 0.750000 0.250000 0.000000 1.000000 0.0 1 1 1
Species
C1 core C
N1 core N
spacegroup
P 1
library Libraries/tersoff
temperature 200 10 19
ala_mode pattern
ala_disp 0.01 0.04
ala_cutoff 4.4
ala_shrink 10 10 10
ala_processor 4
shrink
40 40 40
dispersion 1 500
0.0 0.0 0.0 to 0.375 0.375 0.75 to 0.5 0.0 0.5 to 0.0 0.0 0.0 to 0.5 0.5 0.5
dump every Diamond.res
output phonon Diamond
output cif nv-Diamond.cif
Hi Jack,
Just putting the defect into a cell of diamond of your preferred size is exactly right.
- The fractional coordinates are always between 0 and 1, so the range never changes. Of course the corresponding Cartesian coordinates are proportional to cell size!
- Thermal conductivity is expensive to calculate since it’s a third derivative property, so it’s not surprising it takes time. Alamode uses symmetry and so for high symmetry systems the calculation can be fast, but the moment you put defects in you’re lowering the symmetry and so the cost will increase rapidly. Obviously using a faster computer and more processors is the best solution. You’ll need to ask the OpenBLAS people about their warning, but looks like you just need to disable openMP during the compilation and stick to using MPI for parallelism.
- The results obviously depend on your model - the more accurately it reproduces the phonon properties of the system, the better the answer will be. If the answer isn’t good enough then you need to derive or find a better model.
When running with a 2x2x2 supercell, the calculation will take up to 8 x 8 x 8 times longer as you have 8 times as many atoms and you’re computing cubic force constants (symmetry aside). So I guess the calculation is just still running & that’s why there is no file yet. 
Regards,
Julian