resample_hankel: Resample bootstrap samples in Hankel effmass

In hadron: Analysis Framework for Monte Carlo Simulation Data in Physics

Resample bootstrap samples in Hankel effmass

Description

The bootstrap distribution in the Hankel effective mass can be quite broad due to outliers and long tails. These screw with proper error estimation. Therefore it can be useful to trim these tails. Just trimming a bootstrap distribution would lead to less samples, therefore we do a parametric resampling.

Usage

resample_hankel(hankel_effmass, distance = 5)

Arguments

hankel_effmass	Hankel effective mass from hankel2effmass.
distance	Numeric, threshold for marking outliers.

Details

The central values are also inferred from the distribution because they often are outliers themselves. The new central value is the middle between the upper and lower quantile, making the resulting distribution symmetric.

Half the distance between the quantiles is taken to be the error, therefore the quantiles are chosen at 16 and 84 percent to match the standard deviation. All points that are more than “distance” errors away from the new central value are taken to be outliers.

Value

The Hankel effmass object is returned with the same fields, the numbers have been changed.

Additionally there are the followi1ng fields:

• cov_full contains the full covariance matrix as determined from all the data. This will be skewed by the outliers.

• finite_count gives the number of non-outliers per time slice.

• complete_count gives the numbers of complete cases if all outliers are taken out. This number is often zero because the late time slices contain lots of outliers due to the noise.

• cov_3sigma_pairwise is the covariance matrix using only the non-outliers and removing NAs in a pairwise fashion, using the maximum of the data. This is the covariance matrix that is used for the resampling.

In case that no time slices had a finite error estimate, this function returns just NA.

hadron documentation built on Sept. 9, 2022, 5:06 p.m.