Error calculating transport properties using SERTA

[ghaithjamal_nadhreen]

Hello,

I have been trying to calculate the transport properties using the SERTA ELPH_SCATTERING_APPROX. The calculations crash with an error about failing to open the phelel_params.hdf5 file. When I checked the files, it turns out VASP did not generate the phelel_params.hdf5 file. I have checked the wiki and made sure to include the "ELPH_POT_GENERATE = True" tag to generate the file. Nevertheless, it was not created.

Next, I tried using the phelel package and generated the phelel_params.hdf5 file separately. However, when VASP reached the ELPH step it crashed again with no error in the standard output file but reported a bug in the standard error file and asked to file a bug report.

I have attached the my files. Maybe I am doing something wrong. Your support is greatly appreciated.

Thank you

[manuel_engel1]

Hello,

Thanks for reaching out. Running the electron-phonon code in VASP is at minimum a two-step process. A brief overview of the workflow is given on the electron-phonon caterogy page.

The generation of the phelel_params.hdf5 is the first step of the process and it requires a supercell (you will have to make sure that this cell is large enough by converging the final result with respect to the supercell size). You can do this either via ELPH_POT_GENERATE or via the phelel method, as detailed here. You must not set any of the electron-phonon tags related to transport. The transport calculation comes in the second step which uses a unit cell.

The second step is the actual electron-phonon calculation. This step requires the phelel_params.hdf5 file as input. In this step, we use a primitive cell (this primitive cell must match the unit cell found during the previous supercell calculation, or you specify it explicitly using ELPH_POT_LATTICE during the previous step). This is where you can follow the documentation on calculating transport coefficients.

Let me know if that works for you.

[ghaithjamal_nadhreen]

Thank you for the reply. I followed the instructions from your reply. However, in the first step after the finite difference stages were done I got this error message: "SmithNormalForm routine failed because the input matrix had a negative determinant". What does this error mean? Maybe my supercell is not large enough?

Thank you.

[manuel_engel1]

This is a bit unexpected. It might be due to a bug, but I will look into it. I'm sorry for the inconvenience caused by this issue.

In the meantime, could you send me your input files for this new calculation? I would like to reproduce it on my end, so that we can include a potential fix in the upcoming patch release.

[ghaithjamal_nadhreen]

Thank you for your support. I have attached my input files.

[ghaithjamal_nadhreen]

I have used the phelel package to generate the phelel_params.hdf5 file. After that, I have run the transport calculation step. However, the calculations crash with an error in the std.err file stating a bug in the code.

Code: Select all

 -----------------------------------------------------------------------------
|                     _     ____    _    _    _____     _                     |
|                    | |   |  _ \  | |  | |  / ____|   | |                    |
|                    | |   | |_) | | |  | | | |  __    | |                    |
|                    |_|   |  _ <  | |  | | | | |_ |   |_|                    |
|                     _    | |_) | | |__| | | |__| |    _                     |
|                    (_)   |____/   \____/   \_____|   (_)                    |
|                                                                             |
|     internal error in: wap.F  at line: 3342                                 |
|                                                                             |
|     HDF5 call in wap.F:3342 produced error: 3                               |
|                                                                             |
|     If you are not a developer, you should not encounter this problem.      |
|     Please submit a bug report.                                             |
|                                                                             |
 -----------------------------------------------------------------------------

Can you kindly check it out?
Note: when running the phelel package, you need to in include the --fft-mesh tag to generate the phelel_params.hdf5 file (i.e. --fft-mesh 18 18 18). Otherwise the file will not be created. For some reason this is not mentioned in their wiki nor examples on Github. I have found it while digging through the code after encountering this issue.
https://github.com/phonopy/phelel/blob/ ... vatives.py on line 119

Your support is highly appreciated

[manuel_engel1]

Hello again,

I have spent some time this week to investigate this issue. First, I have determined that the SmithNormalForm issue is due to not specifying the FFT grid of the primitive cell via ELPH_POT_FFT_MESH. This is not stated clearly on the wiki and it's obviously a problem. Therefore, I have created a patch for the current electron-phonon code that can automatically determine the FFT grid of the primitive cell inside VASP. You can still supply it via ELPH_POT_FFT_MESH, but now the error message no longer shows up when running ELPH_POT_GENERATE=True.

Next, I have included useful information in the standard output and the OUTCAR when running with ELPH_POT_GENERATE=True. This includes the primitive FFT mesh and the primitive lattice vectors. The latter can be supplied via ELPH_POT_LATTICE, but are otherwise taken from the primitive cell that is found by VASP.

Furthermore, the primitive cell is now written to a new primitive POSCAR file, but the exact naming is not yet decided. Documentation on the wiki will follow as soon as this feature is available.

While digging through the code, I have encountered a related bug that is currently causing some calculations to fail, which you can read about here at the top of the list.

Finally, all of this will be available in the next patch release of VASP (coming very very soon).

---

Now, to your most recent issue, I will relay the information about phelel to the developer, but please keep in mind that we do not maintain the phelel package.

The bug message regarding wap.F that you encountered could, however, be happening on our side. Therefore, could I ask you once again to provide the input and output files so that I may reproduce the issue on my side?

Thank you for your patience. I hope that we can quickly implement all these fixes so that you can enjoy the new electron-phonon code more smoothly.

Kind regards

[ghaithjamal_nadhreen]

Here are my files for the transport calculations. However, I could not attached the phelel_params.hdf5 because it is too large (~500 Mb). Is there a way I can send it?

[manuel_engel1]

Thanks for sharing the files. The INCAR file looks correct. The problem could indeed exist within phelel_params.hdf5, but sharing the whole file wont be necessary. Could you instead run the following command from the same directory where the phelel_params.hdf5 is located:

Code: Select all

h5ls phelel_params.hdf5

This command is provided by HDF5 tools, so if you have the HDF5 library available, you should also have access to h5ls. It should produce output that looks something like this:

Code: Select all

Dij                      Dataset {1, 2, 8, 8, 2}
FFT_mesh                 Dataset {3}
atom_indices_in_derivatives Dataset {2}
dDijdu                   Dataset {1, 2, 3, 8, 8, 8, 2}
dVdu                     Dataset {1, 2, 3, 25920, 2}
dqijdu                   Dataset {1, 2, 3, 8, 8, 8, 2}
force_constants          Dataset {8, 8, 3, 3}
grid_point               Dataset {25920, 3}
lattice_point            Dataset {4, 3}
p2s_map                  Dataset {2}
primitive_lattice        Dataset {3, 3}
primitive_masses         Dataset {2}
primitive_positions      Dataset {2, 3}
qij                      Dataset {1, 2, 8, 8, 2}
supercell_lattice        Dataset {3, 3}
supercell_matrix         Dataset {3, 3}
supercell_positions      Dataset {8, 3}
unitcell_lattice         Dataset {3, 3}

I would like to check all the array dimensions and see if the error lies there. Thank you!

[ghaithjamal_nadhreen]

This is the content and array sizes of the phelel_params.hdf5 file generated by phelel package:

Code: Select all

Dij                      Dataset {1, 3, 18, 18, 2}
FFT_mesh                 Dataset {3}
dDijdu                   Dataset {1, 3, 3, 324, 18, 18, 2}
dVdu                     Dataset {1, 3, 3, 2916000, 2}
displacements_atom_indices Dataset {6}
displacements_vectors    Dataset {6, 3}
dqijdu                   Dataset {1, 3, 3, 324, 18, 18, 2}
grid_point               Dataset {2916000, 3}
lattice_point            Dataset {108, 3}
p2s_map                  Dataset {3}
primitive_lattice        Dataset {3, 3}
primitive_masses         Dataset {3}
primitive_matrix         Dataset {3, 3}
primitive_numbers        Dataset {3}
primitive_positions      Dataset {3, 3}
qij                      Dataset {1, 3, 18, 18, 2}
s2p_map                  Dataset {324}
shortest_vector_multiplicities Dataset {324, 3}
shortest_vectors         Dataset {324, 3, 27, 3}
supercell_lattice        Dataset {3, 3}
supercell_masses         Dataset {324}
supercell_matrix         Dataset {3, 3}
supercell_numbers        Dataset {324}
supercell_positions      Dataset {324, 3}
unitcell_lattice         Dataset {3, 3}
unitcell_masses          Dataset {12}
unitcell_numbers         Dataset {12}
unitcell_positions       Dataset {12, 3}

I would also like to report that using the https://www.vasp.at/wiki/index.php/ELPH_POT_FFT_MESH tag did work and I was able to obtain the phelel_params.hdf5 file from vasp directly.

[manuel_engel1]

I would also like to report that using the https://www.vasp.at/wiki/index.php/ELPH_POT_FFT_MESH tag did work and I was able to obtain the phelel_params.hdf5 file from vasp directly.

This is good to know.
It seems what you are reporting is purely a phelel issue. For some reason, the interatomic force constants are missing from your phelel_params.hdf5. I'm not exactly sure what is going on but this is probably something that needs to be discussed with the developer of phelel. Could I ask you to provide the input files that you used to generate the (faulty) phelel_params.hdf5 using phelel? Specifically, I would like to get the following information:

• Did you use velph (the command-line tool that comes with phelel) or did you use phelel directly?

• If you used velph, could you send the velph.toml file?

• If you used phelel directly, could you send the command that you used to generate the phelel_params.hdf5?

• The POSCAR, INCAR, KPOINTS, OUTCAR of one of the displaced supercell calculations used in the phelel process.

This would allow me to reference this post on the forum to the developer of phelel and open an issue on github with all the information in one place.

[ghaithjamal_nadhreen]

Did you use velph (the command-line tool that comes with phelel) or did you use phelel directly?

If you used velph, could you send the velph.toml file?

No I used phelel package directly.

If you used phelel directly, could you send the command that you used to generate the phelel_params.hdf5?

Code: Select all

phelel -v --fft-mesh 30 30 30 --cd disp-00{0..6} > phelel_log

The POSCAR, INCAR, KPOINTS, OUTCAR of one of the displaced supercell calculations used in the phelel process.

Attached

Thank you