[gmx-users] g_energy averages

[David L. Bostick]

.. So, if I want to find the standard error in an average value, I just
take the fluctuation and divide it by sqrt(nsteps)?

i.e.
If I get from the energy file
Energy                      Average       RMSD     Fluct.
#Surf*SurfTen               147.805    2270.69    2263.84

and the run was over 250000 steps, the standard error is

2263.84/sqrt(250000)

even if nstenergy was say 500 steps?

David

-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
David Bostick					Office: 262 Venable Hall
Dept. of Physics and Astronomy			Phone:  (919)962-0165
Program in Molecular and Cellular Biophysics
UNC-Chapel Hill
CB #3255 Phillips Hall				dbostick at physics.unc.edu
Chapel Hill, NC 27599	           		http://www.unc.edu/~dbostick
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-

On Wed, 17 Jul 2002, Erik Lindahl wrote:

> David L. Bostick wrote:
> > HI,
> >
> > I use g_energy on an edr file from a constant surface tension simulation.
> > I printed out Box-X, Box-Y, and Box-Z lengths and Press-XX, Press-YY, and
> > Press-ZZ pressures.  I am trying to fine tune a bilayer/protein system in
> > order to preserve the xy-area of the box.  I parsed the xvg file using a
> > simple awk script in order to calculate area, surface tension, average
> > pressures.. etc, but when comparing the values to those generated by
> > g_energy, there is a small discrepancy.  I am pretty sure I am calculating
> > things  correctly.  Does g_energy use some other precision for it's
> > calculation of averages and so forth?
> >
>
> Well, not a different precison but a completely different algorithm :-)
> It is described in detail at the end of the manual.
>
> We actually store a couple of partial sums in the energy file, which
> makes it possible to calculate the REAL average and fluctuation from all
> the simulation steps - not only the average of the values written to the
> energy file.
>
> It's kind of complicated; the easiest way to do it (as you've probably
> seen in a statistics textbook) is to store the sum and sum-of-squared
> values, but that leads to numerical inaccuracies for longs trajectories,
> thus the need for partial sums.
>
> Cheers,
>
> Erik
>
>