The specified ibrav 14 is not supported

[songxq1994]

Dear SSCHA developers,

I am very interested in SSCHA software. Recently, I have met such a problem about symmetry, as below:

raise ValueError("Error, the specified ibrav %d is not supported." % ibrav)
ValueError: Error, the specified ibrav 14 is not supported.

I am researching some low-symmetry structures, but the calculations of Triclinic crystal are likely not supported in the SSCHA code.
Are the restrictions originated from Quantum Espresso or SSCHA?
Is this based on physical or procedural considerations?
What can I do to resolve this issue?

Best regards,
X. Song.

[mesonepigreco]

Hi, I tried to implement ibrav 14 right now, can you update cellconstructor to the last version in the master and try again? It would be good if you can check if it is correct by visualizing the structure by running the following python code (replace dynmat with the path to your espresso dynamical matrix (excluding the final number):

import ase, ase.visualize, cellconstructor as CC, cellconstructor.Phonons
ase.visualize.view(CC.Phonons.Phonons("dynmat").structure.get_ase_atoms())

Otherwise, you could use ibrav 0 which allow you to specify all the atomic position and lattice version directly, which is the safest option for the SSCHA code.

Let us know if the implementation works,
bests,
Lorenzo

[songxq1994]

if the anharmonicity is converged in that cell, then, if you need dynamical matrices on a denser grid, you can interpolate.

Could you give me more detailed advice about how to judge the convergence of anharmonicity and how to interpolate for obtaining dynamical matrices on a denser grid based on the SSCHA results?

How big (number of atoms) is your system?

My simulated structure contains 24 atoms with space group P-1 and I allocated 4GB RAM to the ubuntu virtual machine on my PC.

[mesonepigreco]

Hi, it is very difficult to judge if the anharmonicity is well converged. In principle, one should study how the observable that you want to study changes as a function of the mesh grid but, in practice, simulating too large cells could be really challenging. In general, supercells containing about 100 atoms are good to converge, so your 2x2x2 mesh with 24 atoms contains already 192 atoms which should be enough.

To interpolate, you can load the dynamical matrix:

import cellconstructor as CC, cellconstructor.Phonons
dyn_222 = CC.Phonons.Phonons("my_sscha_dyn", nqirr = X) # X is the nqirr of the 2x2x2 mesh
new_dyn_444 = dyn.Interpolate((2,2,2), (4,4,4))  # Interpolate the dynamical matrix on a 4x4x4 mesh
new_dyn_444.save_qe("my_new_sscha_dyn") # The final SSCHA dynamical matrix on a 4x4x4

I do not understand why you are running out of memory. 4 Gb of RAM should be enough even to store your dynamical matrix in real space. Does the code crashes already when loading the dynamical matrix?

[songxq1994]

Hi, Lorenzo,
Sorry for my late response. Thanks for your detailed reply. I have not met similar problems again about running out of memory.

Does the code crashes already when loading the dynamical matrix?

The code is terminated immediately without any data interchange between PC and clusters.
#-----
I tried my best to perform the calculations of Raman, IR, and Superconductively (SC), which have related descriptions in the discussions "Infrared Response from SSCHA? #46 and Superconductivity calculations with phonons obtained from SSCHA code #103" but I still met some questions as below:

For SC calculations:
(1) I carried out SSCHA to get the new structure and anharmonic matdyn* files;
(2) I performed electron and phonon calculations using the new structures and retained the files " ./_ph0/prefix.dv, ./ph0/prefix.q*, and ./_ph0/prefix.phsave/patterns* ";
(3) I replaced the harmonic matdyn* files with anharmonic matdyn* files from SSCHA and calculated the e-p coupling and Tc in QE code.
The error in ph.out is

 Reading dVscf from file dv   
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%     
 Error in routine readmat (2):
     inconsistent data 

For the calculations of IR and Raman:

In #46, the easiest way is to perform a computation of the dielectric tensor and effective charges within a DFT code and copy them inside the first dynamical matrix file (as done in the usual output of quantum espresso). Then you can use the quantum espresso dynmat.x code on the converged SSCHA dynamical matrix to get IR activity of the modes.

My questions are:
(1) Only NC and LDA pseudopotential are likely supported for the calculations of IR and Raman in QE code. How should I handle the results based on the USPP and GGA pseudopotential;
(2) Does it mean that I perform a computation of the dielectric tensor and effective charges using the new structures from SSCHA and copy them inside the first anharmonic SSCHA dynamical matrix file?

Best Wishes,

Xianqi.

[mesonepigreco]

Sorry for my late reply, since there was a marked answer I did not get the notification.
Anyway, there is a new discussion for how to compute superconductivity at #103 , I suggest you can look at that.

Regarding the IR and Raman, yes, you should use NC pseudos with LDA to get the Raman tensor. However, usually, this is a very good approximation for the Raman intensities. For IR, effective charges and the dielectric tensor should also be implemented for USPP and GGA.
You can use the structure from the SSCHA and copy effective charges, dielectric tensor, and Raman tensor inside the last anharmonic SSCHA dynamical matrix file.

[songxq1994]

Hi,
Thanks for your kind reply and helpful hints. I will continue these tests and provide further feedback.

Bests,
Xianqi.