Flat phonon distribution problem at high frequencies

I’m doing calculations on 3D Ag8GeSn6. I calculated force constants via phonopy. it must be dynamically stable due to a previously synthesized material. however, there are negative phonons in phonon calculations and I obtained stable phonon dispersion with hiphive (after hiphive, there are no negative phonons), but this time the phonons at frequencies of 5 THz and above come out as flat lines and do not appear in phonon PDOS. I tried with many different cutoff parameters. Although a dynamically stable structure can be obtained, I don’t understand how high-frequency phonons often behave this way.

Can you clarify what you how you constructed the force constants with hiphive? Did you use force-constant sensing and enforce rotational sum-rules? Or did you do it in any other way?

It would also be easier to understand the flat phonon-mode problem if you attached figures of the phonopy and hiphive dispersions.

this is my setup.py
import numpy as np
import matplotlib.pyplot as plt
from hiphive import ForceConstantPotential, StructureContainer
from hiphive import enforce_rotational_sum_rules
from hiphive.fitting import Optimizer
from hiphive import ForceConstants
from hiphive import ClusterSpace
from ase.io import read
from hiphive.utilities import extract_parameters
cutoffs = [3.7]
supercell = read(‘SPOSCAR’)
fcs = ForceConstants.read_phonopy(supercell, ‘fc2.hdf5’)
cs = ClusterSpace(supercell, cutoffs)
parameters = extract_parameters(fcs, cs)
parameters_rot = enforce_rotational_sum_rules(cs, parameters, [‘Huang’, ‘Born-Huang’])
fcp_rot = ForceConstantPotential(cs, parameters_rot)
fcp_rot.write(‘fcp_rot.fcp’)
fcs2 = fcp_rot.get_force_constants(supercell)
fcs2.write_to_phonopy(‘FORCE_CONSTANTS’)".

and I will attach figures but I do not how ?

I think you should be able to upload/attach files to your comment.

Here you’re using a very small cutoff (3.7). When using extract_parameters it is recommended to use the longest possible cutoff in the supercell.

I attached figures. thank you for your help. How much big ?

The longest possible cutoff in the supercell, which you can find by calling hiphive.cutoffs.estimate_maximum_cutoff(atoms).

I was tried bigger cutoff parameters and I got to obtain exception: Found cluster (0, 2) twice, check cutoff!

Is this a bulk system and the maximum cutoff is about 3.7Å?

Then you probably need much larger supercell for your phonon-calculation.

exactly yes. in my unitcell there are 15 atoms and after creating a supercell I get the very big system and calculation is too much expensive.
thank you very much again. I will try with bigger supercell