[gmx-users] time-block analysis

[Christopher Neale]

Technically, in order to show that your simulation has reached complete convergence, you would need to
show that every type of data has converged. Obviously nobody does this. You should start with looking at the
convergence of any data type that you analyze in any other way. i.e. if you show a histogram of the distribution
of a dihedral angle as Figure 1, then Figure S1 might be some analysis of the convergence of that histogram.

Look at papers where people are doing similar things to what you are doing. Unfortunately, lots of otherwise
good papers don't even consider convergence so you may need to do a little digging.

PS: g_analyze sounds like a good idea. I've always scripted my own because it is more flexible, but if there is a 
gromacs tool to do this then use it.

Chris.

-- original message --

Very thanks for your reply.


In addition of your suggested script, in the gromacs, -ee option of
g_analyze produces error estimates using block averaging.

A person suggest me to do time-block analysis to find that my simulation
has converged or not.

You mentioned to "2-column data file that has time in the first column and
some other data in the second column".

To investigate convergence of simulation, what should be as other data in
mentioned file (2-column data file)?

Any help will highly appreciated.