From 6cc8e9df63be481f6d444d7e074fec5538956f72 Mon Sep 17 00:00:00 2001 From: Morten Hjorth-Jensen Date: Wed, 4 Oct 2017 14:48:26 +0200 Subject: [PATCH] Update README.md --- doc/Programs/Sampling/README.md | 4 ++++ 1 file changed, 4 insertions(+) diff --git a/doc/Programs/Sampling/README.md b/doc/Programs/Sampling/README.md index b2078e441..2fe5140e4 100644 --- a/doc/Programs/Sampling/README.md +++ b/doc/Programs/Sampling/README.md @@ -13,3 +13,7 @@ Ouput is located into the `FILENAME/` folder. If more than 10⁵ lines are specified the autocorrelation function won't be computed, as it would take too long. The `gaussian.dat` dataset has been generated with numpy, as a proof of concept. It represents a normally distributed set of 5x10⁵ elements with `std = 0.05`. One will notice that the estimate on the error of the central value is greatly improved by all resampling methods. + +energy.dat is an autocorrelated data set, with autocorrelation time of roughly 200. It is useful to see the use of blocking on this dataset as a convenient method to estimate the autocorrelation time (compare the elapsed time on the different methods). + +In the plaquette.dat file there is a small data set (just 1000 samples) and it shows the strenght of using resampling methods to better estimate the error on the central value as opposed to the standard deviation.