Applying Machine Learning Techniques To Intermediate-Length Cascade Decays

TL;DR

This paper explores the use of machine learning to optimize the analysis of intermediate-length cascade decays in collider experiments, improving discovery potential and property measurements of new particles.

Contribution

It introduces machine learning methods to identify optimal kinematic observables for cascade decay analysis, enhancing discovery and measurement accuracy.

Findings

$ ext{Delta}_4$ is confirmed as highly effective for analysis.

Machine learning improves identification of key observables.

Quantified accuracy for spin and mass measurements as a function of signal size.

Abstract

In the collider phenomenology of extensions of the Standard Model with partner particles, cascade decays occur generically, and they can be challenging to discover when the spectrum of new particles is compressed and the signal cross section is low. Achieving discovery-level significance and measuring the properties of the new particles appearing as intermediate states in the cascade decays is a longstanding problem, with analysis techniques for some decay topologies already optimized. We focus our attention on a benchmark decay topology with four final state particles where there is room for improvement, and where multidimensional analysis techniques have been shown to be effective in the past. Using machine learning techniques, we identify the optimal kinematic observables for discovery, spin determination and mass measurement. In agreement with past work, we confirm that the kinematic observable $\Delta_4$ is highly effective. We quantify the achievable accuracy for spin determination and for the precision for mass measurements as a function of the signal size.