[CP2K:1272] Re: conserved quantity in QM/MM run

[Teodoro Laino]

Hi,

this is an additional follow up on the topic of the energy  
conservation for QMMM runs.
If you want to see a perfect energy conservation remember to switch  
off the centering of the
QM fragment at each MD step (keywords: NOCENTER, NOCENTER0 in &QMMM).

In fact, keep in mind that at the working cutoff your QM/DFT  
calculation is NOT translational invariant.
I.e. if you compute energy and forces in two different spatial  
position you will get slightly different
energies and forces (even for a pure QM calculation).
By default in QMMM runs the QM system is always centered and this  
introduces a bias in the forces
and you will never get a conserved quantity.. even using the PERIODIC  
QMMM extension..

Cheers
Teo

On 4 Aug 2008, at 11:20, Teo wrote:

>
> Hi,
>
> qmmm interaction potential is computed within the first image cell.
> This by
> definition cannot led to conserved quantity if you are studying a
> condensed phase system
> (you can trivially understand why)!
> If you want that, you've to use the PERIODIC version of the qmmm
> module and
> first of course get acquainted with it (see papers and so).
>
> Teo
>
>
> On Aug 4, 11:02 am, Bart De Sterck <bartde... at gmail.com> wrote:
>> Hi,
>>
>> we are performing md simulations on water with the purpose of
>> eventually studying the solvation of our target molecules. For this
>> aim, we first performed an MM simulation (NVE, NVT using the nosé-
>> hoover thermostat) using the TIP3P model in which the conserved
>> quantity is conserved indeed.
>>
>> To get acquinted with the QMMM routine, we performed a simulation in
>> which one of the water molecules was treated on the BLYP level of
>> theory. In this simulation however, either using the NVE or NVT
>> ensemble, we were unable to obtain a constant conserved quantity.
>>
>> We added to the file repository an input file for the NVE simulation
>> using the 15/07/2008 version of cp2k, together with graphs of the
>> relevent parameters during our 5 ps MD run.
>>
>> What are we doing wrong? We are looking forward to all suggestions.
>>
>> best regards
>>
>> Reinout Declerck and Bart De Sterck
> >