Machine learning driven identification of heavy flavor decay leptons in proton-proton collisions at the Large Hadron Collider

Raghunath Sahoo, Kangkan Goswami, Suraj Prasad · 2025

Read Machine learning driven identification of heavy flavor decay leptons in proton-proton collisions at the Large Hadron Collider by Raghunath Sahoo, Kangkan Goswami, Suraj Prasad (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-00712.txt.

Introduction

Raghunath Sahoo, with two co-authors, Kangkan Goswami and Suraj Prasad, uses machine learning to identify leptons from heavy-flavor decays in proton-proton collisions at the Large Hadron Collider. Heavy-flavor hadrons, particles with heavy quarks, probe the early stages of collisions, but measuring their semi-leptonic decays is resource-intensive and leans on event generators.

Sahoo and his co-authors train an XGBoost model on events generated by PYTHIA8 at the collision energy stated in the record. Inputs are displacement measures and pseudo-rapidity, a direction measure. The model reaches ninety-eight percent accuracy for heavy-flavor decay electrons and almost one hundred percent for muons, offering an independent identification method for precision studies.

Catalog record

CREATORRaghunath Sahoo, Kangkan Goswami, Suraj Prasad
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE31 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2509-00712

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Machine learning driven identification of heavy flavor decay leptons in proton-proton collisions at the Large Hadron Collider Raghunath Sahoo, Kangkan Goswami, Suraj Prasad Abstract The study of heavy-flavor hadrons is topical in the era of precision measurements, which is useful to test theories based on pQCD. The heavy-flavor hadrons are produced initially during heavy-ion or hadronic collisions and are one of the best probes to understand the initial stages of the collisions as well as the system evolution. In experiments, the heavy-flavor sectors are studied directly via their decay to different hadrons or di-leptons or via their semi-leptonic decay, which is accompanied by additional neutrinos. However, their measurement in experiments is resource-intensive and requires input from different Monte-Carlo event generators. In this study, we provide an independent method based on Machine Learning algorithms to separate such leptons coming from heavy-flavor semi-leptonic decays. We use PYTHIA8 to generate events for this study, which gives a good qualitative and quantitative description of heavy-flavor production in p ​ p pp collisions. We use the XGBoost model for this study, which is trained with p ​ p pp collisions at s = 13.6 \sqrt{s}=13.6 TeV. We use DCA, pseudorapidity, and transverse momentum as the input to the machine. The ML model provides an accuracy of 98% for heavy-flavor decay electrons and almost 100% for heavy-flavor decay muons. I Introduction In an attempt to explore and understand the elementary building blocks of the universe, ultra-relativistic heavy ions are collided at facilities like the Large Hadron Collider (LHC) at CERN and the Relativistic Heavy-Ion Collider (RHIC) at BNL. These collisions allow us to access a deconfined and thermalized matter of quarks and gluons, namely, Quark Gluon Plasma (QGP). However, this transient state of matter can’t be detected directly as the produced partons hadronize within the timescale of 10^{-23} seconds. We often make use of probes, such as heavy-flavour hadrons consisting of heavy quarks (HQs), to understand the interactions within the deconfined medium. The heavy quarks are produced in the initial hard scattering, and their study can act as a good testing ground for perturbative Quantum Chromodynamics (pQCD). These HQs traverse the whole deconfined medium to form open-heavy flavor hadrons (or Quarkonium states) at the deconfinement phase boundary. This makes them an excellent probe for studying the deconfined medium and understanding the dynamics of the hadronic phase. In experiments, to understand the production mechanism of heavy-flavor hadrons, the self-normalized yield of the heavy-flavor hadron is studied as a function of self-normalized charged particle multiplicity. This offers valuable insights into the underlying partonic dynamics and the interplay between hard and soft processes involved in particle production. In experiments, for pp and p-Pb collision, self-normalized yields of J/\psi ALICE:2020msa ; STAR:2018smh ; ALICE:2017wet ; ALICE:2020eji , D meson ALICE:2015ikl , \Upsilon (nS) CMS:2013jsu have been estimated at different center of mass energies. Moreover, the angular correlation between heavy flavor hadrons and charged particles has been studied in experiments, as it is sensitive to the production mechanism and hadronization of heavy quarks Norrbin:2000zc . In principle, the near-side peak is due to fragmentation of the same parton that produced the trigger particle. However, the away-side peak is induced due to the fragmentation of the other partons. The azimuthal correlation of prompt D mesons with charged particles has been studied at ALICE ALICE:2021kpy ; ALICE:2019oyn for pp collisions at \sqrt{s}=13 TeV and for pp and p-Pb collisions at \sqrt{s_{\rm NN}}=5.02 TeV, respectively. Although these studies are often done via hadronic decays of the D meson. The open-heavy flavors (B & D) hadrons often decay via a semi-leptonic channel, B,D\rightarrow l^{\pm}+\nu_{l}+X , with a branching ratio of 0.11 ParticleDataGroup:2024cfk . These semi-leptonic decays can’t be reconstructed due to the neutrino carrying some of the parent’s momentum. Hence, the preferred way of measuring these heavy-flavor decay leptons is to subtract the non-heavy-flavor leptons from the inclusive lepton spectra ALICE:2020hdw ; STAR:2014yia ; ALICE:2016ywm ; ALICE:2023xiu ; ALICE:2019rmo ; ALICE:2019nuy . Owing to the large cross-section, this decay channel is widely explored in experiments to study heavy-flavor hadrons. The heavy flavor decay electrons are used in studying the momentum anisotropy in experiments like STAR and ALICE across various center-of-mass energies, from \sqrt{s_{\rm NN}}=39 GeV Au-Au collisions in STAR to \sqrt{s_{\rm NN}}=5.02 TeV Pb-Pb collisions in ALICE STAR:2014yia ; ALICE:2016ywm ; ALICE:2020hdw . Moreover, recently, production and nuclear modification factor of heavy flavor decay leptons have been estimated in ALICE ALICE:2023xiu ; ALICE:2019rmo ; ALICE:2019nuy ; ALICE:2016mpw . However, most of the experimental measurements of heavy-flavor decay electrons heavily rely on the input from the Monte Carlo models. Moreover, a track-level identification of heavy-flavor decay leptons using the traditional methods is not possible. On the other hand, machine learning techniques have been employed in high-energy physics for several decades Bowser-Chao:1992giy ; Chiappetta:1993zv . They have been successfully applied to a wide range of studies, including jet measurements and tagging ALICE:2023waz ; ATLAS:2020iwa ; Hassan:2025yhp ; Draguet:2025efc , particle reconstruction CMS:2020poo ; Graczykowski:2022zae ; Tamir:2023aiz , event shape estimation Basak:2025idd ; Ortiz:2020rwg ; Prasad:2025yfj ; Mallick:2021wop , impact parameter estimation Bass:1996ez ; Mallick:2021wop ; Zhang:2021zxd , and the measurement of flow coefficients Mallick:2022alr ; Mallick:2023vgi ; Hirvonen:2023lqy . Furthermore, ML algorithms have been successfully applied in designing beam-pipe material in particle accelerators Singh:2024ihp . More recently, classification algorithms have been effectively used to distinguish prompt and nonprompt charmed hadrons, like J/\psi at forward and mid rapidity Prasad:2023zdd , D^{0} meson Goswami:2024xrx , and \lambda_{\rm{c}} baryon Mallick:2025bsn . These studies show the capability of machine learning models to study the heavy-flavour sector of particle production, with track-level identification of particle species and an independent way of separating them from the background particles. In this article, we attempt to separate the heavy flavor decay electrons from the inclusive electron sample. We use a classification algorithm, specifically, the gradient boosting technique XGBoost, known for its high accuracy and efficiency. We primarily take advantage of the displaced distance of closest approach (DCA) for weakly decaying heavy-flavor decay electrons. Along with DCA, we use transverse momentum and rapidity of the electrons. Topological features such as DCA and decay length can be of great importance in separating daughter particles decaying via weak interaction. In our study, we separate the heavy flavor decay electrons from the inclusive electron sample, using XGBoost. We train the model for pp collision at \sqrt{s}=13 TeV and test the XGB model for different center of mass energies, such as 900 GeV and 7 TeV. In Section II , we discuss the methodology of this work and in details about the machine learning algorithm. Section III presents the results and discussion, and finally, in Section IV , we conclude our work. II Methodology This section presents a brief introduction to PYTHIA8 and XGBoost algorithms. Moreover, we discuss the background sources for the electrons and the training datasets. II.1 PYTHIA8 PYTHIA8 is a Monte Carlo event generator used to simulate leptonic, hadronic, and heavy-ion collisions. PYTHIA8 was developed as a significant upgrade over PYTHIA6, with improved tuning to the LHC data Sjostrand:2007gs . We have generated around 10^{9} events of proton-proton collisions at the center-of-mass energy of 13 TeV. To mimic an experiment-like environment, the Beams:allowVertexSpread = on is set

Read the complete text