Strange error in double-relaxation job

[Nick Winner]

Hello all,

I am using the full MP suite of packages to run relaxation of charged defects. Using OptimizeFW from atomate as the starting point most calculations have run fine. However, I when running a double-relaxation the second relaxation returns with a smaller FFT grid (NGX=120 instead of 128). This is very important because the LOCPOT is only written at the very end of the process instead of once for each relaxation and it needs to have the FFT grid remain the same as my other calculations. To be clear: relaxation 1 has the proper grid density, relaxation 2 has a smaller grid, and the LOCPOT is only written from the second relaxation.

I'd like to continue using double-relaxations, but despite searching through the code I cannot find any instance as to where atomate/fireworks/custodian would intentionally change this, so I don't know how I can correct it.

Any suggestions on where to look?

Best,

Nick

[Anubhav Jain]

Hi Nick

My best guess is that NGX is set automatically by VASP based on your cell size since I am pretty sure we are not specifying any NGX in the INCAR (you can check the raw input files of your calculations - are they even specifying NGX?). After the first relaxation, your cell size changes. When VASP is run again to do the second relaxation, it is starting with a different cell size, and therefore your NGX will also change.

If you set NGX manually in your INCAR (e.g., using the modify_incar powerup), then I think VASP will respect that NGX without recalculating based on your cell size

Anubhav