HSE error: EDWAV: internal error, the gradient is not orthogonal

[ronald_cohen]

I am getting this error when I rotate the initial magnetic moment away from the c axis:
EDWAV: internal error, the gradient is not orthogonal
-----------------------------------------------------------------------------
| _ ____ _ _ _____ _ |
| | | | _ \ | | | | / ____| | | |
| | | | |_) | | | | | | | __ | | |
| |_| | _ < | | | | | | |_ | |_| |
| _ | |_) | | |__| | | |__| | _ |
| (_) |____/ \____/ \_____| (_) |
| |
| internal error in: rot.F at line: 822 |
| |
| EDWAV: internal error, the gradient is not orthogonal 1 1 -1.703e-3 |
| |
| If you are not a developer, you should not encounter this problem. |
| Please submit a bug report. |
| |
-----------------------------------------------------------------------------
Input and output are attached. Thank you,

Ron

[christopher_sheldon1]

Hi Ron,

Thank you for your question and for attaching the input/ output files. I've repeated your calculation with the same version of VASP without getting an error. The error occurs before the calculation begins, implying that there is an issue with the setup. Taking a look at your INCAR, I have a few suggestions that you can try:

• I recommend using fewer CPUs. The calculation ran when I used 22 cores and turned off KPAR and NCORE. 1120 is a lot of cores and this will cause issues with the parallelisation for small systems like yours. I would recommend using a lower value for NCORE. NCORE should be tested, increasing incrementally to find what the best settings are for your computer setup (e.g., NCORE = 4, 8, 16, etc and KPAR = 1, 2, 4, 8, ...). There may well be a combination of NCORE and KPAR that enables your HSE calculation to run on 1120 cores but it is best to gradually scale up to this number. I would suggest starting with 56 cores, then increasing to 112, 224, etc. There is a more detailed description about parallelisation in VASP on our wiki.

• You should use LNONCOLLINEAR = .TRUE. ISPIN = 2 runs spin-polarised calculations, where the spin is treated as up and down, rather as spinors which the SAXIS tag requires (it defines the relative orientation of spinor space spanned by the Pauli matrices), i.e., noncollinear calculations need to be performed, i.e.

• Use PBE WAVECAR to speed up HSE06. I am unsure if you have done this but, generally, when performing hybrid calculations, you can save a lot of time by using the WAVECAR from a previously converged GGA (e.g., PBE) calculation as input. This reduces the number of SCF steps required for the more expensive HSE calculation. There is an example calculation available in our tutorials.

Does this answer your question? Let me know if there's anything further that I can help you with.

Best wishes,

Chris