Bachelor Thesis

PUBDB-2024-06640

http://join2-wiki.gsi.de/foswiki/pub/Main/Artwork/join2_logo100x88.png The Influence of Parton Shower and Hadronization Models on Machine LearningJet Taggers

Parra Asensio, D. (Corresponding author)Extern* ; Kogler, R. (Thesis advisor)DESY* ; Haller, J. (Thesis advisor)Extern*

2024

Abstract: In simulating high-energy particle collisions, Monte Carlo event generatorsemploy different models to create parton showers and replicate hadronization.The choice of model can influence the final state of the simulation and,consequently, the subsequent analysis of the generated event. This study utilizeda machine learning algorithm to discern the differences between quark and gluonjets generated through different approaches to showering and hadronization.Initially, particle jets from dijet events generated using Pythia, Sherpa and Herwigwere compared across various jet- and particle-level properties. Both Pythia andSherpa utilize parton showers ordered by transverse momentum, while Herwigemploys angular-ordered showers. However, Sherpa and Herwig share similarcluster hadronization models, while Pythia uses the Lund string model. Thecomparison revealed that Pythia and Sherpa were the most similar across mostproperties, due to a different minimum transverse mometum of jets compared toHerwig. To better understand the jet tagger’s decisions, the characteristics ofquark and gluon jets were analyzed for each generator. It was demonstrated thatgluon jets generally contain more particles, and thus less momentum and energy,to varying degrees, depending on the generator. A machine learning algorithmwas then trained on the previously generated Pythia and Sherpa samples andsubsequently tested on data from all generators under consistent conditions. Themodel trained on Pythia became specialized to Pythia and showed reducedaccuracy when tested on Herwig and Sherpa, whereas the model trained onSherpa performed almost equally well across all three test datasets. While therewere indications that the different hadronization models influenced this outcome,the parton shower model appeared to have minimal impact.

Note: Bachelorarbeit, University of Hamburg, 2024

---

Contributing Institute(s):

1. LHC/CMS Experiment (CMS)

Research Program(s):

1. 611 - Fundamental Particles and Forces (POF4-611) (POF4-611)

Experiment(s):

1. No specific instrument

Appears in the scientific report 2024

---