PC GAMESS/Firefly-related discussion club

Learn how to ask questions correctly

Alex Granovsky
gran@classic.chem.msu.su

>>However, it is still possible to perform PHVA - the simplest way
>>is just to assign very large masses (e.g. 10⁶)
>>to the atoms you'd like to fix.

>That is exactly what I am doing now. :)
>I hope that it is formally correct, but still have some doubts.
>Because the geometry optimization is done with frozen atoms and
>calculating the Hessian (now with all atoms) we do it in the point
>with nonzero first derivatives.

This approach is absolutely correct regardless on
whether the geometry was fully or only partially optimized;
and there are no parts in Firefly assuming that gradient
is exactly zero when calculating second derivatives.


>>This type of analysis is not explicitly implemented in Firefly
>>at present. On of the reasons is that in the most cases, an overhead
>>due to calculation of the full Hessian matrix is quite acceptable,
>>especially for the semi-numerical calculations.

>Well, it depends. I apply Firefly not only for isolated molecules, but also
>for molecules coupled to metal leads and on metal surfaces. Especially
>in the last case the large number of metal atoms should (or may) be frozen.
>The "lost time" in this case can be longer than useful time. :(
>

Let us assume the Hessian has the following structure:

|AA|AF|
|FA|FF|

where A is for active (i.e., non-frozen) and F is for frozen atoms.
Even with infinitely large masses of frozen fragments,
one still needs not only AA block, but also the entire
off-diagonal blocks (namely |AF| and |FA| where |AF| = transpose(|FA|));
and only |FF| block is not needed.
This means that all these additional vibs are not completely useless at all.

Just another interesting point - assuming the same basis on each
atom, with half of atoms being frozen - then only 1/16 = (1-1/2)^4
of all two electron integrals does not depend on the coordinates
of active atoms (and hence does not depend on the changes of
their geometry).

regards,
Alex Granovsky