Multivariate side-band subtraction using probabilistic event weights

TL;DR

This paper introduces a probabilistic method to assign event-by-event weights for separating signal from background in high-dimensional data, generalizing side-band subtraction without binning.

Contribution

It presents a novel, multidimensional approach to calculate event weights (Q-factors) for signal-background separation, extending traditional side-band methods.

Findings

Enables more accurate signal extraction in complex datasets.

Allows direct access to true signal spectra without binning.

Improves analysis precision in experimental physics.

Abstract

A common situation in experimental physics is to have a signal which can not be separated from a non-interfering background through the use of any cut. In this paper, we describe a procedure for determining, on an event-by-event basis, a quality factor ($Q$-factor) that a given event originated from the signal distribution. This procedure generalizes the "side-band" subtraction method to higher dimensions without requiring the data to be divided into bins. The $Q$-factors can then be used as event weights in subsequent analysis procedures, allowing one to more directly access the true spectrum of the signal.