My Community

I am encountering a problem when attempting to generate Wannier90 files using VASP with spin-orbit coupling (SOC) enabled (LSORBIT = .TRUE.). Without incorporating SOC, the calculations run smoothly and I get the required .amn, .mmn, .eig files. However, when I enable SOC, I encounter the following error after VASP completes projecting:

"Projection 16 done.
Projection 17 done.
Projection 18 done.
Projection 19 done.
Projection 20 done.
Projection 21 done.

===================================================================================
= BAD TERMINATION OF ONE OF YOUR APPLICATION PROCESSES
= RANK 0 PID 215263 RUNNING AT cmslabserver2
= KILLED BY SIGNAL: 11 (Segmentation fault)"

I would appreciate any insights or suggestions to resolve this issue. Here are some additional details:

VASP version: 5.4.4
Wannier90 version: 2.1.0
projections and bands specified as
"
begin projections
Pt : d
Mn : d
Ga : p
end projections

guiding_centres=.true.
spinors = .true."

I have already tried various troubleshooting steps, such as adjusting the calculation parameters and checking the memory of the system, reducing KPOINTS, but the problem persists. I have also attached the input and the output files along with this query. If anyone has encountered a similar issue or has insights on how to resolve it, I would greatly appreciate your input.

Please provide a complete set of input files. We cannot reproduce the issue with the files you provided.

Based on the input you provided, I would guess that you run out of memory. Please check if a simple system with less atoms works with/without SOC.

Also note that you get a big warning message that the number of bands changed. While this is unlikely to be the cause of the segmentation fault, it is not recommended to increase the number of bands without relaxing the additional bands. Please include the same number of bands for all steps of this calculation.

Thank you for your prompt response.

The error occurs for an 8 atom system as well as for an 1 atom system. Here I have attached the input files for the 1-atom system. Since it was a non-scf run, I have used pre-converged CHGCAR and WAVECAR for the calculations, which I could not attach here. I was trying to recreate the results of example 4 from this github repository https://github.com/Chengcheng-Xiao/VASP ... e/example4 . The input files were taken from there.

I am very new to wannier90 and yet to fully understand the projection of bands. Since I was trying to reproduce the results, I did not change any tags from the input files.

I have also tried running different calculations for different systems using "random" projections and all of these calculations fail showing the same segmentation fault while SOC is implemented. Non-SOC calculations were completed without error.

I have attached the input files for your reference.

The INCAR file you attached seems to contain multiple steps. Can you please split the INCAR file in a sequence of INCAR.1, INCAR.2, ... to reproduce the issue? If there is something other than updating the INCAR file required to reproduce the issue, please add these instructions as well.

Thanks again for the response.

The INCAR file I have provided should reproduce the issue. Since I was trying to reproduce the results from the mentioned GitHub repository (https://github.com/Chengcheng-Xiao/VASP ... e/example4), I did not change the INCAR tags they provided. However, I have attached here the input files for an 8 atom system with a simpler INCAR file that also shows the same segmentation fault. The CHGCAR and WAVECAR files required for non-scf calculation could not be attached here due to size restrictions. Even though these input files should also reproduce the issue (without using pre-converged CHGCAR and WAVECAR), it might be better to produce CHGCAR and WAVECAR beforehand by running an SCF calculation. The required files for SCF calculation are inside the directory "scf" and the exact input and output files for which the error was encountered are inside the directory "error-files".

Unfortunately, I could not reproduce the segmentation fault that you report. After moving to a machine with large enough memory, the example you gave me attached works. It does however print several warning messages that may indicate that your input is not ideal, e.g, that the number of projections is different.

Now, this could point towards a issue with the memory available on your machine or improvements to VASP since you seem to use a very old version. If you cannot access a newer version of VASP, here are my suggestions to work towards understanding the VASP + Wannier90 interface which may help you to resolve the issue.

• Run the VASP + Wannier tutorial. You could use PBE instead of HSE if you want to make this faster and try to reproduce the PBE band structure using Wannier interpolation.
• Move to a lightly more complex material where the inclusion of SOC changes the band structure e.g. GaAs and repeat the steps above.
• Try the same for a simple magnetic material like MnO. The ground state is AFM, but even the FM state may already be more difficult.
• At this stage, you can move to the material you used.

I will try to see if I can get a nice Wannierization for this material, too.

Thank you so much for your detailed suggestions. I will try to follow the above steps for a better understanding.

I had a bit more time to look into this. Here are a couple suggestions specific to your system

• Your POSCAR file is the conventional cell with two formula units. You can create a smaller cell with half the size.
• At least in the beginning, I would not set ISYM=-1, because it results in much bigger k-points meshes than necessary.
• Carefully check that the orbitals you use as projectors are the one you include in the Wannierization. Specifically, you can use LORBIT=11 to produce DOS or band structure with orbital resolved quantities, to see which bands to include. Alternatively, you can also look at the POTCAR files. You can exclude bands you are not interested.