LO-TO splitting and Born effective charges

[biglinn]

Hi, @mesonepigreco:
I have a doubt about LO-TO splitting.
I check the dyn1 file(gamma point). LO mode and TO mode are degenerate.

But LO and TO is splitted after I plot phonon spectrum with the script from tutorial. Main codes are:

dyn = CC.Phonons.Phonons("harmonic_dyn", 4)

qpath, data = CC.Methods.get_bandpath(dyn.structure.unit_cell,
                                      PATH,
                                      SPECIAL_POINTS,
                                      N_POINTS)
xaxis, xticks, xlabels = data # Info to plot correclty the x axis

dispersion = CC.ForceTensor.get_phonons_in_qpath(dyn, qpath)

I look up the CellConstructor Documentation, but I don't find "Methods.get_bandpath" and "ForceTensor.get_phonons_in_qpath".
Sometimes, phonon spectrum doesn't show LO-TO splitting if the dyn matrices are not very good.
So, I wonder that how the plot script leads to LO-TO splitting.

[biglinn]

Moreover, if I use dyn = dyn.Interpolate( (2,2,2), (3,3,3)) to give supercell, Born effective charges are ignored.
I find a code InterpolateMesh to replace Interpolate in the document.

However, there is a problem.

It seems that this parameter lo_to_splitting is not defined in this function.
I try to run dyn = dyn.InterpolateMesh((3,3,3), lo_to_splitting = True), but there are still not Born charges in dyn files.
So how to generate Born effective charges and open LO-TO splitting?

[mesonepigreco]

Do not use InterpolateMesh as it is still experimental and may lead to incorrect results.

[mesonepigreco]

The code automatically looks for Born Effective charges inside the first dynamical matrix (in quantum espresso format).
If you computed it with quantum espresso ph.x, with epsil = .true., then you should have the effective charges stored inside the dynamical matrix. Cellconstructor reads automatically those values. When interpolating the dispersion using the ForceTensor module, as done in:

dispersion = CC.ForceTensor.get_phonons_in_qpath(dyn, qpath)

The code automatically takes the effective charges and computes the LO-TO splitting for the given path.
However, at Gamma, LO-TO splitting is not well defined as it is nonanalytical (the eigenvector of the LO-TO modes depends on the direction in which you approach to Gamma). This is why it is not reported in the decomposition of eigenvalues and eigenvectors.

As you noticed, the Interpolate function does not exploit LO-TO splitting. This should be fixed as you noticed by InterpolateMesh, that employs the ForceTensor module as the get_phonons_in_qpath, which uses the info of the effective charges to calculate LO-TO. However, the InterpolateMesh method has still some bugs and may lead to wrong results, I do not recommend using it.

[biglinn]

Hi, @mesonepigreco:
I manually add effective charges to the dyn files for plotting phonon spectrum. But there is a little imaginary phonon near Gamma point.
Does this have something to do with ASR rule? Or maybe the parameters nk1, nk2, nk3 of ph.x calculations?

[mesonepigreco]

Hi, it may be related to many things, most likely the mesh of q points for the direct calculations. I believe there are advanced ASR implementations to cure these things in modern QE versions (but I do not know how to use them), or it could be a slightly non perfectly relaxed cell or phonon calculation.

[biglinn]

It's so complex. I'll make tests.
Thanks!

[biglinn]

Hi, @mesonepigreco:
I notice that the first tutorial use a method:

relax = sscha.Relax.SSCHA(minim, calculator, N_configs = N_CONFIGS, max_pop = MAX_ITERATIONS)
……
# Run the NVT simulation
relax.relax(get_stress = True)

This relax.relax call the function compute_ensemble to calculate energies and forces from ensembles with ASE calculators.
And if I'm likely to use NPT ensembles, I can use relax.vc_relax.

So, could I use this method with DFT instead of ASE?
Since the second tutorial "Setting up automatic SSCHA simulations on clusters" does not use sscha.Relax.SSCHA method, it just use ensemble.generate and ensemble.save to get configurations and do minimization.

[biglinn]

Oh, sorry! The second tutorial has introduced relax method in the back part, I'll study it.