RMV, thanks again for your support and Fortran code (which I have successfully ported to python with the help of the cited papers [Fortran code is not that straight forward when you are used to python :-) ]).
Still I am wondering about one thing. The long time diffusion should be calculated from a region where the r2com is more or less linear.
From the attached picture (r2com for 86 POPC) I would tend to use frames between ~6000 and ~13000.
Is this reasonable? I am somewhat unsure if it is okay to discard so much points.
Currently I am calculating r2com for different time windows to get a clue how this effects the shape of the curve.
I tend to use the averaged data to make a judgment about the slope change between the two diffusion time scales for lipids in a bilayer. It's not always clear from the r^2 vs t plots for the individual lipids. Also, try plotting just 0-5000 ps to look for the slope change.