prepfold sigma

When candidates are folded with the program prepfold, a summary plot is generated, along with a .pfd file. These are based on a datacube whose axes are pulse phase, frequency subband, and time subintegration. This datacube is shifted and added as necessary to search in DM, period, and period derivative. The final result, on the output plot, is a “significance” value, which is expressed in the equivalent number-of-sigma-away-from-the-mean for a normal distribution. This value has proved to be a fairly good indicator of candidate quality, in particular for the many very-short-period candidates that appear in search databases.

The value is based on an RMS amplitude divided by an estimate of the noise per bin in the final profile. As a result, it can be expected to follow chi-squared statistics, and these are used to convert values to “sigma” values. While measuring the RMS amplitude is very straightforward, estimating the background is a little trickier.

prepfold estimates the background in each time/frequency subintegration by, essentially, keeping a running standard deviation on all the raw data that goes into that subintegration. These are then stored in an auxiliary array (the .stats attribute) in the .pfd file, and used to estimate the amount of noise per bin in the final folded profile.