Estudo da aplicabilidade de classificadores de aprendizado de máquinas na procura por acoplamentos quárticos anômalos no canal semileptônico da produção central exclusiva de WW usando CMS e PPS

• Autor(a) principal: Sousa, Matheus Pereira Macedo de
• Data de Publicação: 2021
• Outros Autores: [e-mail]
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Biblioteca Digital de Teses e Dissertações da UERJ
• Texto Completo: http://www.bdtd.uerj.br/handle/1/17086
• topic: Particles (Nuclear physics); Standard model (Nuclear physics); Machine learning - Particles (Nuclear physics); Couplings; Compact muon solenoid; Large Hadron Collider (France and Switzerland); Anomalous quartic couplings; Central exclusive production; CMS; Machine learning; Partículas (Física nuclear); Modelo padrão (Física nuclear); Aprendizado do computador – Partículas (Física nuclear); Acoplamentos; Solenóide de múon compacto; Grande colisor de hádrons (França e Suíça); Acoplamento quártico anômalo; Produção central exclusiva; CMS; Aprendizado de máquinas; CIENCIAS EXATAS E DA TERRA::FISICA::FISICA DAS PARTICULAS ELEMENTARES E CAMPOS::TEORIA GERAL DE PARTICULAS E CAMPOS

This text presents a comparative study between the cut-based method and machine learning for the search for anomalous quartic gauge coupling using protons identified in the CMS PPS subdetector. The theoretical motivation for the search for these couplings was developed by the Effective Field Theory and the description of new upper and lower limits. The entire experimental structure that was used in the reconstruction of the particles used in the analysis is narrated. The Precision Proton Spectrometer is fully nested within the central detector, enabling the observation of surviving protons from processes via Exclusive Central Production. The technology used in the detectors is based on silicon that is placed in mobile devices that approach the beam line called Romam Pots. The signal sample is composed by the production of the WW pair from the fusion of two photons for the CEP topology, while the background has the same final state but coming from different processes. The search was developed comparatively by two methods, cut-based and machine learning. For the second, we used LightGBM, a decision tree based algorithm. To compare data and Monte Carlo, the division into control regions was performed. The analysis was prepared for a luminosity of 9,792 fb⁻¹, which is somewhat low for physics searches beyond the Standard Model, so we used the blinded method of the signal region for the data to avoid bias. Finally, we count the number of events that passed the final selection of the two methods and compare the results of the main variables. We present the results for the sensitivity of the upper and lower bounds of the coupling constants of both methods.