Page 1 of 1
high vmc_ave_period and correlation less than 1
Posted: Thu Nov 21, 2013 1:26 pm
by varelse
With high vmc_ave_period I get the correlation time less than one step. The standard error seems to decrease instead of increase with the number of reblocking transformations. Is the usage of reblocking method still valid in this case? Is the error overestimated? Should I still look for the plateau?
Re: high vmc_ave_period and correlation less than 1
Posted: Fri Nov 29, 2013 3:53 pm
by Neil Drummond
Dear Varelse,
If the correlation period is one step (it can't really be smaller) then you would expect the reblocked error bars to "start" in a plateau (i.e., you won't see the error bar rise with reblocking transformation before reaching the plateau). In principle it doesn't matter what block length you choose (until the error in the error starts to become significant).
The decrease you see is probably random noise. To get a better feeling for the behaviour of the reblocked error bar, choose the y-axis of your reblock plot to start at zero.
Best wishes,
Neil.