Gmx msd - Is a weighted linear regression done to calculate the diffusivity constant?

User discussions
1

GROMACS version: 2020.3
GROMACS modification: No

In the description of the gmx msd command it says:

The diffusion coefficient is determined by linear regression of the MSD, where, unlike for the normal output of D, the times are weighted according to the number of reference points, i.e. short times have a higher weight.

I don’t understand what is meant by weights here.
After running the gmx msd program, I do an unweighted linear regression on the MSD versus lag time data points, and I get the same value of the regression coefficient as printed by gmx msd. I used both gnuplot and statsmodels (python package)

This clearly indicates that an unweighted linear regression is done. So then, what is meant by “the times are weighted according to the number of reference points”?

Thanks for your help.

2

You are right, no weighting is used. I think that in older versions weighting was done with the number of starting points, but this got removed without updating the documentation.

3

I updated the documentation for the 2021 release.