Feasibility study of true muonium observation with the existing Belle-II dataset

TL;DR

This study assesses the potential to observe true muonium, a leptonic bound state, using existing Belle-II data, employing machine learning techniques to distinguish signal from background and estimate significance.

Contribution

It demonstrates the feasibility of detecting true muonium in current Belle-II datasets through a detailed simulation and machine learning analysis, which is a novel approach.

Findings

Potential for true muonium observation at Belle-II with existing data

Machine learning effectively separates signal from background

Estimated significance suggests possible discovery under certain conditions

Abstract

True muonium ($\mu^+\mu^-$) is one of the cleanest bound states, being composed only of leptons, along with true tauonium and positronium. Unlike the latter, true muonium and true tauonium have not been observed so far. This article shows that the spin-0 state of true muonium (para-TM), decaying into two photons, can be observed at a discovery level of significance in the dataset already collected by the Belle-II experiment at the $\Upsilon (4S)$ peak, with certain assumptions on systematic uncertainties. Para-TM is produced via photon-photon fusion, and its observation is based on the detection of the photon pair resulting from its decay, on top of the continuum background due predominantly to light-by-light scattering. Trigger, acceptance and isolation cuts, along with calorimeter resolution and reconstruction efficiency, are taken into account during the Monte Carlo simulation of both signal and background. In order to separate signal and background, a machine learning method, based on extremely randomized trees, is trained on simulated events. Finally, the expected statistical significance of TM observation is evaluated, taking into account systematic uncertainties in a parametric fashion.