[gmx-users] Optimal number of time origins in the calculation of block-averaged time correlation functions

[Francisco Garcia]

Dear users,

I am comparing different schemes for computing the self-diffusion of a
system. One of the approches
I am considering is using block averages.

I have a 50 ns long trajectory and I want to compute the
block-averaged diffusion constant as follows:

(1) Divide the trajectory into blocks of 2 ns (25 blocks of equal length)
(2) For each block compute the rmsd over all particles and Nt time
origins (25 diffusion constants in all)
(3) Compute the average and standard deviation of the 25 diffusion constants

Point (2) above is where things get tricky. When you read previous
works, authors are usually
not specific about the exact values of the parameters they used.
Currently I choose the number of origins
(Nt) and the correlation time (Tc) arbitrarily. It is obvious that the
maximum value of Nt is BL/2,
where BL is the length of each block and the maximum value of Tc =
delta_t * BL/2. My question
is how to determine the optimal number of time origins Nt and Tc per
block. Suggestions from
experienced users or any reference to an article which discusses a
scheme for choosing the Nt
and Tc would be much appreciated. I am a novice so kindly be as
specific as possible in your response.

Thank you very much