[gmx-users] Re: Conserved energy ("Conserved En.") in NVT simulation

[Janne Hirvi]

Thanks for your comments Mark,

If I understood correctly, you are saying, that maybe linear drift in "Conserved En." could indicate conservation of
that quantity, which we actually should be interested.

However, I made meanwhile some additional calculations and noticed that drift in "Conserved En." with v-rescale
temperature coupling decreases if tau_t is increased. So, in my opinion it would indicate that drift in "Conserved En."
is more or less related to "disturbance" due to temperature coupling. If this is true, there shouldn't be drift in perfect
case - am I correct? 

Btw. drifts in my previous mail were per ns.

Regards,

Janne

------------

> Dear gmx-users,
>
> I have run some tests (especially) for pure water, being able to achieve pretty much perfect energy conservation in
> NVE ensemble (with PME-switch and shifted VDW potential).

Great.

> Then, just for the test, I continued to NVT ensemble by using previous .tpr and .cpt files, but the result was a quite
> surprise, because there seems to be constant drift in "Conserved En.".
>
> The drift (100kJ/mol = 0.3%) is not especially large, but it exists with v-rescale temperature coupling. On the other
> hand, there is no drift at all, if I change to nose-hoover temperature coupling.
>
> The behavior is similar also my other systems (liquid mixtures), even though then neither of temperature coupling
> methods will give exactly perfect conservation, but nose-hoover gives much better one (500 kJ/mol = 3% vs. 60
> kJ/mol = 0.4%).
>
> Actually, I was expecting to see no drift at all, after all systems behaved fine in NVE ensemble with same parameters
> with an exception of added temperature coupling (especially when I used double precision for test purpose).
>
> So, the main question is, if it is just normal that even such parameters, which give perfect energy conservation in NVE
> ensemble may introduce drift in "Conserved En.", when NVT ensemble is used? Or is it just purely (only reason for drift)
> indication that some temperature coupling methods"work better" than others in this sense?

In the case of v-rescale, the description of "Conserved En." is an
overbid. If you see the Bussi 2007 paper, there is a conserved
quantity H-tilde, but computing it requires knowledge of the whole of
history. IIRC their H is what mdrun reports as "Conserved En.". I
don't know whether it can be shown that a linear trend in H is
equivalent to conservation in H-tilde, but that would be a useful
result!

Mark
------------