Heavy Flavor Production at the Large Hadron Collider: A Machine Learning Approach

Abstract

Charmonia suppression has been considered as a smoking gun signature of quark-gluon plasma. However, the Large Hadron Collider has observed a lower degree of suppression as compared to the Relativistic Heavy Ion Collider energies, due to regeneration effects in heavy-ion collisions. Though proton collisions are considered to be the baseline measurements to characterize a hot and dense medium formation in heavy-ion collisions, LHC proton collisions with its new physics of heavy-ion-like QGP signatures have created new challenges. To understand this, the inclusive charmonia production at the forward rapidities in the dimuon channel is compared with the corresponding measurements in the dielectron channel at the midrapidity as a function of final state charged particle multiplicity. None of the theoretical models quantitatively reproduce the experimental findings leaving out a lot of room for theory. To circumvent this and find a reasonable understanding, we use machine learning tools to separate prompt and nonprompt charmonia and open charm mesons using the decay daughter track properties and the decay topologies of the mother particles. Using PYTHIA8 data, we train the machine learning models and successfully separate prompt and nonprompt charm hadrons from the inclusive sample to study various directions of their production dynamics. This study enables a domain of using machine learning techniques, which can be used in the experimental analysis to better understand charm hadron production and build possible theoretical understanding.