$J/\psi$ production with machine learning at the LHC

Abstract

$J/\psi$ particle, the ground state of charmonium, is one of the significant probes to understand formation of quark–gluon plasma, a state of deconfined quarks and gluons created in relativistic collisions. $J/\psi$ is identified by reconstructing the decay products from relevant decay modes with the application of sophisticated techniques mainly based on a high level of physics knowledge and complex computation skills requiring long process time for data quality assurance. For the measurement, a high-purity sample is needed which can be obtained by traditional cut-based methods to extract well-defined particle signal distribution, resulting in high systematic uncertainties. It is revealed that application of artificial intelligence-based machine learning models in various fields enhanced the speed, accuracy, and efficiency of human efforts. Therefore, in this study random forest classifier (RFC), one of the successful classification algorithms, was implemented in measurement of $J/\psi$ production from its dielectron decay channel. With the RFC model, identification of $J/\psi$ was studied in three different signal selection approaches: loose track-level analysis, loose pair-level analysis, and tight track-level analysis. The RFC analyses for the charmonium production were found to be compatible with the experimental measurements, and tight signal selection has 98.3% success for predicting the state with 92.9% sensitivity and 93.3% precision. The invariant mass spectrum of $J/\psi$ was also presented for each approach.