[gmx-users] Re[3]: g_msd vs g_velacc

[Vitaly Chaban]

Xavier,

Ones more. The length of the trajectory is 200ps (200 000 time steps). Certainly, both for
velocities and displacements. I can ensure this is enough to obtain
correct values, no matter Einstein or Green-Kubo formula one uses.
Moreover, if you carry out a simulation for 1000ps, nothing will change.

> I tried to determine the length of your velocity C(t) you used to extract
> the correlation time.
The length of the ACF is 5ps with a step 0.005ps. But it is
calculated 200/5 = 40 times and then averaged (just in compliance with its
definition). Pls. pay attention that ACF(t) at t > 5ps is fairly zero
for any molecular liquid in the ordinary state.

> It seems that you used 5 ps for the velocities and 200 ps for the msd!
> That might be a reason fot differences, no?
Pls. see above.
And as for your interest why
> frequency to write velocities = 5
> frequency to write coordinates = 100
For MSD we obtain a straight line. Is it not very important which
interval to use in this case: 10, 50, 75 or 100. The result (diffusion
coefficient) will be the same. Otherwise, ACF does. The most
meaningful deviations from zero are observed at t < 2ps. Of course, it is
very important to have a small enough step for the velocity output. It
is characteristic not only for methanol which I used only as an
example.
Is it clear now?

Again, I'm absolutely sure in all values used for the calculation till
the moment when vac.xvg was obtained. But then...

I obtain integral. I should divide it by 3 (according to Green-Kubo).
But I get quite inconsistent value. Maybe the problem is concerned
with units in which velocity is outputted...

Will anybody share his expertise in such calculation? :)


>>> $ g_msd
>>> 2.3 (and it is true)
> XP> How much of msd(t) did you use? The part of the curve you use to fit
> XP> to Einstein equation will affect enormously the result!
> 
> 200 ps of trajectory was used. It is quite enough.
>It depends on many factors: the length of your simulation and the
>convergence of your msd curve are two of them.
> XP> Here you use 1001 points to integrate! Which result to 1001*0.005fs=5 ps!