[Octopus-users] Potential Energy fluctuation during Ehrenfest-TDDFT MD simulation

[Alberto Castro]

Hi,

As Matthieu points out, the presence of the edges of the simulation
box may lead to unphysical forces, or in other words, to a violation
of the translational invariance. The other source of violation of
translational invariance is the grid itself. If you compute the total
energy of the system at differences point in space, you will see that
there is a small oscillation whose period coincides with the grid
spacing. This causes unnatural forces, and therefore one should make
sure that they are small enough (by using a small enough grid spacing,
and by making use of a filter -- use the variable FilterPotentials).

Cheers, Alberto.


On Tue, 4 Aug 2009, matthieu verstraete wrote:

>> potential energies fluctuated slightly during the simulation. (I think the
>> total energy was conserved enough.)
>> And I found this fluctuation corresponds to the stretching vibration of O2.
>> And the center of mass did not move from (0,0,0).
> excellent.
>
>>
>> (2) When I set the center of mass of O2 at (0,0, 3.78) initially, O2 started
>> the translational motion toward the center of simulation box (0,0,0).
>> I cannot understand the reason of this!
>> When I changed the size of box from (Lsize=13.0 Bohr) to (Lsize=15.0 Bohr),
>> this  translational motion still occurred with the same velocity.
> There is probably a force which propagates from the edges, due to
> numerical or derivative errors there. As your COM is not symmetric
> there is a net force on the molecule, and it is pushed (in this case
> towards the center). As you are trying to simulate an isolated system,
> you should always center it to minimize edge effects. A hard core
> octopusser might want to have a look at the details of the drift.
> These types of drift and rotation forces always appear due to
> numerical errors if nothing else, and are usually just removed from
> the total motion (COM fixed and ellipsis of inertia as well).
>
>> (3) It was found that when a molecule moves translationally, potential
>> energies fluctuated largely.
> I suspect the constant drift force is giving quite a bit of kinetic
> energy (perhaps your scale is incorrect in the first KE vs PE plots)
> and that this is feeding the PE.
>
>> (4) I think that the reason of this fluctuation derives from the difference
>> of the time scale between nuclear and electronic movement..?
> normally if your electrons are well integrated the ions are slow so
> there is no problem.
>
>
>> (5) Potential energies fluctuated around a certain value but never diverged
>> when molecule moves translationally.
>> So is it okay to consider the average value of these as the true potential
>> energy?
> well... the PE can take a constant offset, but if it looks
> qualitatively different from the case centered in 0,0,0 I would not
> trust it.
>
> M.