[gmx-users] RE: diffusion constant level off

[Dr. Vitaly Chaban]

- Expected values are estimated. For example 輋 the structure is reach to

> converge at last 100 ns ( all simulation time is 200 ns), then in theory,
> diffusion coefficients will dont change at last 100 ns.
> -I have only one diffusion coefficient for each 10 ns. I have 20 diffusion
> coefficients during all simulation time.
> -I controlled all msd.xvg. They are linear. The plot I sent is diffusion
> versus time. (Please see it again). It isnt msd vs time
> - I tried to increase sampling. The results are same approximately.
> Commands for trestart to 1000 ps:
> Command for 1.diffusion:g_msd -f traj.xtc -s topol.top -o msd1.xvg
> -trestart 1000 -beginfit -1 -endfit -1 -b 0 -e 10000
> # D[   Protein] = 0.1091 (+/- 0.0212) (1e-5 cm^2/s)=13.line from msd.xvg
> Command for 2.diffusion:g_msd -f traj.xtc -s topol.top -o msd2.xvg
> -trestart 1000 -beginfit -1 -endfit -1 -b 10000 -e 20000
> # D[   Protein] = 0.0679(+/- 0.0013) (1e-5 cm^2/s)
> Command for 1.diffusion:g_msd -f traj.xtc -s topol.top -o msd3.xvg
> -trestart 1000 -beginfit -1 -endfit -1 -b 20000 -e 30000
> ...
> Command for 3.diffusion:g_msd -f traj.xtc -s topol.top -o msd20.xvg
> -trestart 1000 -beginfit -1 -endfit -1 -b 190000 -e 200000
> 2013/3/29 Justin Lemkul <jalemkul at vt.edu>
>
>

To increase sampling means to run a simulation for more nanoseconds.

If your particle (protein?) is big enough, it may require a very large time
to issue a proper diffusion. Therefore, the question may arise if you
really want linear diffusion for this particle...

Your plot is not bad. If my purpose was to assess the scale of linear
diffusion, I would stop with the values you depicted.