Using projections and correlations to approximate probability distributions

Abstract

A method to approximate continuous multidimensional probability-density functions (PDFs) using their projections and correlations is described. The method is particularly useful for event classification when estimates of systematic uncertainties are required and for the application of an unbinned maximum-likelihood analysis when an analytic model is not available. A simple goodness-of-fit test of the approximation can be used, and simulated event samples that follow the approximate PDFs can be generated efficiently. The source code for a Fortran-77 implementation of this method is available. © 1998 American Institute of Physics.