Problems with vibrational analysis with DFT and PCM

[sauer]

Hi everybody,

we are trying to do run a geometry optimization and vibrational analysis of a molecule in PCM at the DFT level, but have problems. Apparently one has to use numerical derivatives for the Hessian. But asking for this we get a
"MEMCHK ERROR, not a valid memget id in work(kalloc-1)". See the attached output file, which also includes in the input files.

But maybe we are using the wrong keywords to get this.

Any help would be appreciated.
Stephan

[magnus]

Looks like a bug which is possibly connected with switches between symmetries during numerical differentiation. Unfortunately, turning off symmetry doesn't work because of another bug.

In any case, I would run this in two steps. First do a geometry optimization and then perform vibrational analysis using the optimized geometry (and numerical Hessian).

[sauer]

Hi Magnus,

thanks for looking at this. Unfortunately, running it in two steps does not help either. It runs in the same bugs. It is a bit strange, because it must mean, that nobody does DFT/PCM geometry optimizations with Dalton.

Stephan

[magnus]

Sorry I didn't mean to imply that it works if you run out in two steps. That was to avoid the calculation of a numerical Hessian in the geometry optimization step where it is not needed.

The geometry optimization works if you do not ask for the numerical Hessian. However, it is done with a fixed cavity according to Kenneth and I'm not exactly sure what the implication of that is.

The error you get is because of a bug in the calculation of the numerical Hessian when you include PCM.

[magnus]

If the bug is related to switching symmetry as I suspect then perhaps it can be avoided if symmetry detection is turned off. Unfortunately specifying "NoSymmetry" in the molecule file is not enough.

[sauer]

And you do not happen to know where I could turn it off hard-code in the program?

[magnus]

And you do not happen to know where I could turn it off hard-code in the program?

Not very specific but somewhere in DALTON/abacus/abander.F.

[sauer]

Ok, thanks.