[gmx-users] autocorrelation time of dVpot/dlambda?

[Shun Zhu]

Hi Michael:

            Thanks for the reply! I did the following command: g_analyze -f
energy.xvg -ac out.xvg

and in out.xvg I saw the following lines, it seems like a autocorrelation
decrease over time. Does that mean the autocorrelation time here is
infinite? Thanks

Steve

# This file was created Fri Mar 11 16:46:01 2011
# by the following command:
# g_analyze -f energy.1.xvg -ac crap.xvg
#
# g_analyze is part of G R O M A C S:
#
# God Rules Over Mankind, Animals, Cosmos and Such
#
@    title "Autocorrelation"
@    xaxis  label "Time (ps)"
@    yaxis  label "C(t)"
@TYPE xy
     0.000     1.00000
     0.100     0.12191
     0.200     0.04696
     0.300     0.05139
     0.400     0.08079
     0.500     0.06953
     0.600     0.07639
     0.700     0.07060
     0.800     0.07970
     0.900     0.07947
     1.000     0.06976
     1.100     0.07309
     1.200     0.06973
     1.300     0.07161
     1.400     0.07008
     1.500     0.06880
     1.600     0.06747
     1.700     0.06574
     1.800     0.06383
     1.900     0.06175
     2.000     0.06125
     2.100     0.06316
     2.200     0.06383
     2.300     0.06076
     2.400     0.05812
     2.500     0.05635
     2.600     0.05565
     2.700     0.05727
     2.800     0.05663
     2.900     0.05746
     3.000     0.05641
     3.100     0.05375
     3.200     0.05202
     3.300     0.05316
     3.400     0.05255
     3.500     0.05047
     3.600     0.05019
     3.700     0.05199
     3.800     0.05144

......




On Fri, Mar 11, 2011 at 4:18 PM, Michael Shirts <mrshirts at gmail.com> wrote:

> > I am doing free energy calculation in Gromacs and want to get an error
> > estimate of my results. Is it possible to compute the autocorrelation
> time
> > of dVpot/dlambda in Gromacs using a certain length of trajectory such as
> 10
> > ns? Thanks a lot
>
> The amount of simulation time required to compute the autocorrelation
> time of dVpot/dlambda will depend on the timescale of the system.  You
> should be able to use g_analyze to analyze the energy.xvg output, and
> get an autocorrelation time.  If your simulation time is 50x longer
> than the autocorrelation time, then you are very likely converged.
> However, if there are very long correlation times, you still might not
> have captured all of the time dependent variations in  Vpot/dlambda.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-users/attachments/20110311/f0bcb755/attachment.html>