L-BFGS moves fixed atoms

[ehermes]

Hello,

I have noticed that occasionally and somewhat unpredictably,GL-BFGS (iopt=1) will move fixed atoms when doing NEB calculations. Below is a link to an example calculation that shows fixed atoms moving in images 01, 03, and 06.

https://www.dropbox.com/s/092oprda3a9aj ... ple.tar.gz

It may be worth mentioning that I am using a patch from the Grimme group that implements the DFT-D3 dispersion correction in this particular example, but this phenomenon has also been observed by a colleague who does not use this patch.

Is there a mistake in how I set up my calculation, or is this a bug in the VTST patches for VASP?

Thank you,
Eric Hermes

[graeme]

Ah, that sounds bad. We'll take a look.

[graeme]

I don't see any motion of the frozen atom (which are actually the middle two layers in the slab) over this NEB calculation. Can you tell me what you see that made you believe that frozen atoms were moving?

[ehermes]

I have apparently tarred up the wrong run. Here's the correct example (same URL, but I've updated the file):

https://www.dropbox.com/s/092oprda3a9aj ... ple.tar.gz

The new tarball should be 2.6 MB. In this case, images 01, 03, and 06 show movement of the inner two layers of the slab, which should be fixed.

It does not appear to be as simple as erroneously moving the atoms under the forces they are experiencing, as the atoms seem to move uniformly in an unphysical way.

Thank you for taking a look at this,
Eric

[graeme]

There is clearly something wrong here. It may, however, be fixed in vtst 3.1 (the latest version). I'll try to reproduce this, and if the problem still shows up, we'll need to fix it.

Note: I tried to reproduce this and did not see any motion of frozen atoms in the latest code. LBFGS was reworked a little to provide an autoscaling feature, so it is not surprising that this was fixed along the way.