Towards a generic implementation of matrix-element maximisation as a classifier in particle physics

TL;DR

This paper explores a more efficient matrix-element maximisation technique for particle physics classification tasks, demonstrating its advantages over traditional methods in speed and providing estimates of invisible particle momenta.

Contribution

It applies matrix-element maximisation to a complex process, compares optimization algorithms, and discusses potential for integrating momentum estimates into advanced analysis tools.

Findings

Maximisation is more CPU-efficient than traditional MEM.

Slight reduction in classification performance with maximisation.

Provides estimates of invisible particle momenta for further analysis.

Abstract

The so-called matrix-element method (MEM) has long been used successfully as a classification tool in particle physics searches. In the presence of invisible final state particles, the traditional MEM typically assigns probabilities to an event -- based on whether it is more signal or background-like -- through a phase space integration over all degrees of freedom of the invisible particles in the process(es). One inherent shortcoming of the traditional MEM is that the phase space integration can be slow, and therefore impractical for high multiplicity final states and/or large data sets. The recent alternative of matrix-element maximisation has recently been introduced to circumvent this problem, since maximising a highly-dimensional function can be a far more CPU-efficient task than that of integration. In this work, matrix-element maximisation is applied to the process of fully-leptonic top associated Higgs production, where the Higgs boson decays to two $b$-quarks. A variety of optimisation algorithms are tested in terms of their performance and speed, and it is explicitly found that the maximisation technique is far more CPU-efficient than the traditional MEM at the cost of a slight reduction in performance. An interesting consequence of using matrix-element maximisation is that the result of the procedure gives an estimate of the four-momenta for the invisible particles in the event. As a result, the idea of using these estimates as input information for more complicated tools is discussed with potential prospects for future developments of the method.