QMMM problems

[Frank Uhlig]

Dear Cp2k'lers,

I am trying to calculate the ionization potential of water in bulk
using CP2K's QM/MM capabilities. The original goal was to get the IP
of an isolated QM water molecule in a periodic MM water box. I
attached a simplified input file which can be used to reproduce the
following issues. It contains a single QM water molecule in an
otherwise empty box (a set of MM atoms for a periodic water box are
commented out and can be used for further testing). The MM cell and
the QMMM cell are set to the same size for comparison to a fully
periodic DFT calculation.

First, what kind of periodicity is triggered by the setting in
%QMMM%CELL%PERIODIC? Should it be that the all parts of the system are
periodic, only the MM or only the QM subsystem? From the input
reference I would guess a combination of the
%QMMM%CELL%PERIODIC%MULTIPOLE section with proper settings plus a
%QMMM%CELL%PERIODIC set to XYZ I would couple an isolated QM system to
a periodic MM system. Is that correct?
With the attached setup, both PERIODIC NONE and PERIODIC XYZ give the
same total energy of the system (within SCF convergence). Calculations
including the MM atoms give the same total energy for the respective
system no matter what periodicity was specified.

Second, whether the section %QMMM%PERIODIC is included with default
settings or not makes a difference if the multipole solver is used. In
the input reference manual it states that this section "cannot be
optional", but at least with the CP2K version I used [1] it behaves
optional and changes some settings.
If one compares the attached setup to a fully periodic calculation
using the Quickstep module for a single water molecule in the same box
one finds that both setups give equivalent results for the energetics
of the system. All that, provided the %QMMM%PERIODIC section is
included for the QMMM calculation. If one excludes the %QMMM%PERIODIC
section one gets an IP corresponding to an open system. My suspicion
here is that the %QMMM%PERIODIC section somehow triggers xyz
periodicity in the QMMM settings for the QM subsystem. The resulting
energies do not change if the multipole decoupling is turned on/off.

Third, if using the Wavelet Poisson solver in the attached input I get
a different behavior with respect to the section %QMMM%PERIODIC. The
input of the section still behaves optional, but whether included or
not, it does not change anything. But now I always get the result of
an isolated QM system. Only if the periodicity in
%DFT%POISSON%PERIODIC then is set to XYZ one returns to fully periodic
system (as one probably should). All is still unaffected by the
%QMMM%CELL%PERIODIC settings, AFAICT.

I realize that this is quite a mess, but I'd appreciate any help
concerning the meaning of the keywords and sections, what settings
are/should be triggered by them. I am of course willing to help as as
much as possible with further testing, explanation of what I meant by
the mess above and so on.

Best,

Frank

[1] - r12191 compiled with openmpi v1.4.2 (compiled with gcc 4.5.2),
linked to mkl v10.2.6, all regtests passed
-------------- next part --------------
A non-text attachment was scrubbed...
Name: example.inp
Type: application/octet-stream
Size: 14042 bytes
Desc: not available
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20120413/5e82f0c8/attachment.obj>