[Feature Request]: Handle `FEXCF` error in VASP

[Andrew-S-Rosen]

Problem

The following is an error that @blaked8619 just received in a VASP calculation. This occurred when a cell relaxation was accidentally done on a molecule with vacuum space. So, unless there is a more general trend for this one, Custodian should simply catch/report the error and make no changes. It should probably warn the user to check out the structure though.

 -----------------------------------------------------------------------------
|                                                                             |
|     EEEEEEE  RRRRRR   RRRRRR   OOOOOOO  RRRRRR      ###     ###     ###     |
|     E        R     R  R     R  O     O  R     R     ###     ###     ###     |
|     E        R     R  R     R  O     O  R     R     ###     ###     ###     |
|     EEEEE    RRRRRR   RRRRRR   O     O  RRRRRR       #       #       #      |
|     E        R   R    R   R    O     O  R   R                               |
|     E        R    R   R    R   O     O  R    R      ###     ###     ###     |
|     EEEEEEE  R     R  R     R  OOOOOOO  R     R     ###     ###     ###     |
|                                                                             |
|     ERROR FEXCF: supplied exchange-correlation table                        |
|      is too small, maximal index : 9403                                     |
|                                                                             |
|       ---->  I REFUSE TO CONTINUE WITH THIS SICK JOB ... BYE!!! <----       |
|                                                                             |
 -----------------------------------------------------------------------------

Proposed Solution

TBD.

Alternatives

No response

[janosh]

i've gotten this error many times and feel haunted by it since i'm not a pro VASP user... yet none of the pro users i've shown it to had ever seen it before or knew how to handle it 👻 i'm glad it's finally surfacing for others too. if anyone can find the fix, it's you @Andrew-S-Rosen 😄

[Andrew-S-Rosen]

@janosh: Oh no, this was the first time I've ever seen it! It's a terrifying error though and makes no sense. It made me feel like there was something wrong with the xc functional or that I compiled VASP wrong!

In reality, a cell relaxation was accidentally done on an O2 molecule in a box with 20 A of vacuum in each dimension, and the error was associated with an enormous change in the forces likely due to unphysical cell stress/strain.

My guess is you probably have run into this on some "funky" structures, whether from sampling or from an MLIP. If that's the case, I fear there may not be a clear/consistent solution to this problem. But I am basing this off of only one sample point.

[janosh]

not sure if that's any hint but i also saw this error a few times on DFPT simulations for https://github.com/janosh/dielectrics

[Andrew-S-Rosen]

@janosh, @esoteric-ephemera: I have run into the ERROR FEXCF: supplied exchange-correlation table error a few more times. Empirically, in my cases, I have found it to be due to IMIX = 1 being set by Custodian.

This error came about in my cases on the first geometry step (i.e. the starting structure), following a brmix error. The brmix error caused IMIX to be switched from the default to a value of 1. This then caused the subsequent FEXCF issue. Removing IMIX and simply rerunning the job as-is made the calculation proceed.

I have no clue why.

Edit: May just be a coincidence...