Please use this identifier to cite or link to this item: http://repository.ipb.ac.id/handle/123456789/165466
Title: Optimalisasi Parameter Breit-Wigner untuk Klasifikasi Partikel Resonansi dengan Physics-Informed Neural Network
Other Titles: 
Authors: Puspita, R. Tony Ibnu Sumaryada Wijaya
Yani, Sitti
Khoriah, Siti
Issue Date: 2025
Publisher: IPB University
Abstract: Resonansi partikel muncul sebagai lonjakan massa invarian akibat peluruhan partikel berumur pendek. Penelitian ini bertujuan mengestimasi parameter resonansi berupa massa pusat dan lebar peluruhan menggunakan physics-informed neural network (PINN), sekaligus mengklasifikasikan tiga partikel resonansi, yaitu J/?, ?, dan Z, beserta background. Data berasal dari eksperimen CMS, berupa dimuon dan dielektron pada rentang massa invarian 2-110 GeV. Tahapan penelitian meliputi pra-pemrosesan, transformasi data, pendefinisian data signal dan background, estimasi parameter berbasis distribusi Breit-Wigner, serta klasifikasi partikel resonansi. Hasil menunjukkan PINN mampu memprediksi parameter dengan galat relatif <1%, sesuai referensi PDG. Pada klasifikasi, PINN (? = 0,1) mencapai akurasi tertinggi 94,21% (data latih:uji = 90:10), melampaui model neural network dan random forest. Namun, model dengan PINN membutuhkan waktu pengujian yang lebih lama karena penambahan perhitungan physics loss. Pengintegrasian hukum fisika yang bersesuaian dengan dataset pada model machine learning dapat meningkatkan akurasi sehingga metode ini menjadi alternatif untuk analisis data fisika partikel.
Particle resonance is generated from invariant mass spikes due to the decay of short-lived particles. This study aims to estimate resonance parameters such as center mass and decay width using physics-informed neural network (PINN), as well as classify three resonance particles, namely J/?, ?, and Z, along with the background. The data comes from the CMS experiment, in the form of dimuons and dielectrons in the invariant mass range of 2-110 GeV. The research stages include data pre-processing, data transformation, defining signal and background data, Breit-Wigner distribution-based parameter estimation, and resonance particle classification. The results show that PINN is able to predict the resonance parameters with <1% relative error, as per the PDG reference. For classification, PINN (? = 0,1) achieved the highest accuracy of 94,21% (training:testing data ratio = 90:10), outperforming neural network and random forest models. However, testing with PINN takes longer due to the additional physics loss calculation. Integration of the laws of physics that correspond to the dataset in machine learning model can improve accuracy, providing this method an alternative for particle physics data analysis.
URI: http://repository.ipb.ac.id/handle/123456789/165466
Appears in Collections:UT - Physics

Files in This Item:
File Description SizeFormat 
cover_G7401211002_c7ed2b7621ce43b4aae37301b251f54a.pdfCover283.42 kBAdobe PDFView/Open
fulltext_G7401211002_243e963d279e4794aae82ebf284e35cd.pdf
  Restricted Access
Fulltext3.46 MBAdobe PDFView/Open


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.