EDWAV error only when using VTST (VASP 5.4.1)

[arosen]

If I try to optimize a structure using a VASP 5.4.1 build without VTST, my job runs fine. However, if I instead try to run the optimization using a VTST-enabled VASP 5.4.1 build, I get "EDWAV: internal error, the gradient is not orthogonal" within the first few electronic steps (generally within step 1-5). This is odd because I don't know why the electronic energy minimization would be any different between builds. I searched the forums for "EDWAV" but don't see any relevant questions (a more general Google search mainly shows issues when using the HSE06 functional, which I'm not using). This issue occurs for volume relaxations as well as ionic relaxations. It does not matter if I use the gamma-point or standard builds. Deleting the prior WAVECAR before running doesn't fix the issue. I believe it occurs whether I use a VTST optimizer or a standard VASP optimizer.

I have run into this scenario for a number of materials and on multiple computing environments. However, I don't know the details of the VASP builds and so unfortunately don't know if there are any compiler flags that differ between the standard VASP build and the VTST-linked VASP builds. (If this is the anticipated culprit, I will continue to work with IT staff for additional clarification about how the codes were compiled).

[graeme]

I'm not sure what is causing the error but I also can not see how it can be related to the vtstcode. The error is in the electronic convergence, which is not changed in any way by the vtstcode. I suggest trying a different electronic optimizer (IALGO) and perhaps also delete the WAVECAR in case there is some problem with an initial wavefunction.

[arosen]

Thanks for the reply. That's what I figured as well but wanted to confirm. I'll continue trying to isolate the issue.

[arosen]

To anyone that may run into a similar issue, the issue was resolved when switching ALGO keywords. This is bizarre because the job runs fine with ALGO = 'All' with VASP 5.4.1 but switching to the VTST-enabled build resulted in EDWAV errors unless I switched ALGO. As previously noted, VTST should not be impacting the electronic energy minimization, so I'm certain this has something to do with the way VASP was compiled. Anyway, just sharing this in case it helps someone one day...

[weiyidan]

According to my experience, this is related to your compilation, and you should change another intel compiler.