Izučavanje radijativnih raspada W bozona i unapredjenje sistema za očitavanje mionskog spektrometra detektora ATLAS

[E. Bakos]

151 str.

Physics - High energy physics / Fizika - Fizika visokih energija

This thesis incorporates three research topics, all related to the ATLAS detector at the Large Hadron Collider (LHC) at CERN. The first topic covers the contribution to the Data Acquisition (DAQ) of ATLAS, precisely the implementation of the MDT Read Out Driver (MROD) functionality to the Front-End Link eXchange (FELIX) readout system of the ATLAS detector. The second topic represents the search for the rare radiative decay of W boson to meson and a photon, analyzing data already collected with ATLAS during Run 2. The third topic outlines two machine learning algorithms to identify D mesons from radiative W boson decays at the LHC. ATLAS is one of the general purpose detector placed on the LHC. ATLAS consist of many layers of sub-detectors which are designed to detect the different particles. The outermost layer of ATLAS, the muon spectrometer is made from four detection systems, each one exploiting different technologies. The major part of the muon spectrometer consist of Monitored Drift Tube (MDT) chambers which measures the properties (momentum, angular distributions, electric charge) of muons. Within the DAQ system, during the previous data-taking periods the MDT cham- bers were readout via the MDT Read Out Driver (MROD). This readout system is dependent on MROD cards, which will reach end of their lifetime during Run 3. Since repairing the broken MROD cards or ordering new ones is not possible, a new implementation is needed which will use the new FELIX system. As part of the system, the swROD implements data fragment building and formatting. An addi- tional module of the swROD, the CSMProcessor has been developed, which mimics the MROD building mechanism and incorporated the sub-detector specific settings which allows the configuration of the different MDT chambers, stores information about them, like their number or position. Multiple tests have been performed using the CSMProcessor, and the results show that the modified FELIX readout chain will be able to process the increased data rates in Run 3. Using ATLAS, many experimental validations have been performed to investi- gate the accuracy and applicability of the Standard Model (SM). To test the SM as thoroughly as possible, searches for not (yet) detected decays and more precise measurements are necessary. Radiative decays of the W boson to W → M γ, where M is a meson, are sensitive to the coupling of the W boson with the pho- ton and, more importantly, probe the strongly coupled Quantum Chromodynamics (QCD) regime. The search for the W → ργ, where a W boson decaying to a ρ me- son and a photon, and other rare decays are experimentally challenging, due to the multijet background, which consists of hadronized quarks and gluons. No search for the W → ργ has been performed so far, so no previous bounds exist. During the decay the ρ meson and the γ are decaying ”back-to-back”, sharing the momentum of the W boson. Since the lifetime of the ρ meson is short, it de- cays further to charged and a neutral π meson. Due to the nature of the ρ meson, i ABSTRACT isolated prompt ρ and a τ lepton decaying hadronically with exactly one charged and one neutral pion are indistinguishable within the detector, therefore the algo- rithm used to detect the τ candidate can be used to identify the ρ meson without any modification. After the initial trigger and object selection, processes with the same or similar final states are considered as background. The main sources of background are events involving inclusive photon + jet and dijet processes, where a track is reconstructed within a hadronic jet. This background cannot be reliably modelled with Monte Carlo (MC) simulation due to the complicated mixture of con- tributing processes. Instead, this contribution to the total background is modelled with a data-driven non-parametric approach. To extract the limit on the branching fractions of the B (W → ργ) a binned maximum-likelihood is performed to the se- lected events. The search provides a limit of B (W → ργ) < 6.29 × 10−6 at 95% confidence level. This limit can be further improve if we include the search in the track-plus-photon final state where the ρ meson is identified as a track. Due to the different triggers and selection criteria, the tau-plus-photon and track-plus-photon final state are orthogonal, therefore the selected events can be combined in the final fit. This approach provides, for the first time, a limit of B (W → ργ) < 5.17 × 10−6 at 95% confidence level. Reducing the background processes is also possible using the meson proper- ties. However, it is possible to construct many of different variables, which makes it difficult to make a proper selection and measure their usefulness. To make things easier a machine learning algorithm can be developed specifically for meson tag- ging. One way to do this is to list many high-level variables, and using deep neural network create a classification algorithm, which is able to distinguish between signal and background mesons originating from quarks and gluons. An other approach is to develop a convolutional neural network using low level variables, such as the momentum of the particle and the energy deposited in the calorimeter. However, we can reach the best results if we combine the two exploiting the advantage of both models.The developed algorithm based on these models is able to identify jets orig- inating from Ds mesons in radiative W decays and shows a good efficiency of 47% for signal with a 100 times rejection of jets from quarks and gluons. This presents an opportunity to improve measurements related to Ds mesons, particularly in the context of the rare decays.

Ova teza obuhvata tri teme koje se odnose na istraˇzivanja na detektoru ATLAS na Velikom sudaraˇcu hadrona (LHC) u CERN-u. Prva tema pokriva doprinos sistemu za prikupljanje podataka (Data Acquisition (DAQ)) detektora ATLAS, konkretno im- plementaciju MDT Read Out Driver (MROD) funkcionalnosti u Front-End Link eX- change (FELIX) sistem oˇcitavanja. Druga tema predstavlja potragu za retkim radija- tivnim raspadom W bozona na mezon i foton. Analizirani su podaci iz proton-proton sudara na energiji √s = 13 TeV prikupljenih pomo´cu detektora ATLAS tokom Run 2 perioda (2015-2018). Tre´ca tema opisuje dva algoritma maˇsinskog uˇcenja za iden- tifikaciju D mezona nastalih u radijativnim raspadima W bozona na LHC-u. ATLAS je jedan od dva detektora opˇste namene na LHC-u. Sastoji se od viˇse pod-detektora koji su dizajnirani da detektuju tragove naelektrisanih ˇcestica, elek- trone, fotone i dˇzetove, kao i mione. Kombinuju´ci navedene podatke mogu se mer- iti energije i uglovne raspodele navedenih objekata i odrediti znak naelektrisanja naelektrisanih ˇcestica. Takodje, mogu´ce je identifikovati hadronske raspade tau lep- tona i dˇzetove koji potiˇcu od b- ili c-kvarkova, a moˇze se meriti nedostaju´ca energija. Najudaljeniji pod-detektor ATLAS-a, mionski spektrometar, napravljen je od ˇcetiri sistema za detekciju, od kojih svaki koristi razliˇcite tehnologije. Najve´ci deo mion- skog spektrometra ˇcine Monitored Drift Tube (MDT) komore. U okviru sistema DAQ, tokom prethodnih perioda uzimanja podataka, MDT ko- more su oˇcitavane preko MDT Read Out Driver-a (MROD). Ovaj sistem oˇcitavanja zavisi od MROD kartica, koje ´ce dosti´ci kraj svog ˇzivotnog veka tokom Run 3. Poˇsto popravka MROD kartica ili naruˇcivanje novih nije mogu´ca, potrebna je im- plementacija koja ´ce koristiti novodizajnirani sistem za oˇcitavanje podataka FELIX. Kao deo sistema, software ROD (swROD) implementira izgradnju i formatiranje frag- menata podataka. Razvijen je modul swROD-a, CSMProcessor, koji oponaˇsa meh- anizam izgradnje MROD-a i ugradjuje specifiˇcna podeˇsavanja pod-detektora koja omogu´cavaju konfiguraciju razliˇcitih MDT komora, ˇcuvanje informacije o njima, kao ˇsto su njihov broj ili poloˇzaj. Viˇse testova je obavljeno koriˇs´cenjem CSMProcessor-a, a rezultati pokazuju da ´ce modifikovani sistem za oˇcitanja FELIX moˇze da obradi pove´cane brzine prenosa podataka tokom Run 3. Jedna od osnovnih tema istraˇzivanja na eksperimentu ATLAS je testiranje Stan- dardnog modela (SM). Jedan od mogu´cih testova SM-a ˇcine potrage za (joˇs) neotkri- venim raspadima gradijentnih (kao i Higsovog) bozona. Radijativni raspad W bo- zona W → M γ (gde je M mezon) osetljiv je na sprezanje W bozona sa fotonima i, ˇsto je joˇs vaˇznije, omogu´cava ispitivanje reˇzima jako spregnute kvantne hromodi- namike. Potraga za W → ργ, gde se W bozon raspada na ρ mezon i foton, i drugi retki raspadi su eksperimentalno izazovni, zbog velikog fona koji postiˇce iz produk- cije dˇzetova. Na hadronskim sudaraˇcima do sada ovaj kanal raspada nije izuˇcavan, tako da ne postoje prethodne granice za faktor grananja W → ργ. iii SA ˇZETAK Tokom raspada, ρ mezon i γ se raspadaju ”back-to-back”, dele´ci impuls W bo- zona. Poˇsto je vreme ˇzivota ρ mezona kratko, on se dalje raspada na naelektrisane i neutralne π mezone. Zbog prirode ρ mezona, izolovani ρ i tau lepton (koji se ras- pada hadronski sa jednim naelektrisanim i jednim neutralnim pionom) se ne mogu eksperimentalno razlikovati, pa se algoritam koji se koristi za detekciju tau moˇze se koristiti za identifikaciju ρ mezona bez bilo kakvih modifikacija. Nakon inicijalnog trigera i selekcije ˇcestica, procesi sa istim ili sliˇcnim konaˇcnim stanjima se sma- traju za fon. Glavni izvori fona su dogadjaji koji ukljuˇcuju inkluzivne procese foton + dˇzet i didˇzet, gde se tragovi ˇcestica rekonstruiˇsu unutar hadronskog dˇzeta. Ovaj fon se ne moˇze pouzdano modelovati Monte Carlo (MC) simulacijama, ve´c se ko- riste data-driven tehnike. Da bi se izraˇcunao limit na faktor grananja B (W → ργ) koriˇs´cen je pristup zasnovan na metodu maksimalne verodostojnosti. Dobijen je limit od B (W → ργ) < 6.29 × 10−6 na nivou poverenja 95% sa uraˇcunatim sistem- atskim greˇskama. Ovo ograniˇcenje se moˇze dodatno poboljˇsava ako se u analizu ukljuˇci konaˇcno stanje trag-plus-foton gde je ρ mezon identifikovan kao trag unutar detektora. Zbog razliˇcitih kriterijuma selekcije dogadjaji iz dva navedena konaˇcna stanja su statistiˇcki nezavisni, pa se izabrani dogadjaji mogu kombinovati. Konaˇcni dobijeni B (W → ργ) < 5.17 × 10−6 na nivou poverenja 95%. Suzbijanje fonskih procesa u identifikaciji mezona koji sadrˇze charm kvark se moˇze ostvariti koriˇs´cenjem razliˇcitih svojstava ovih mezona. Mogu´ce je konstruisati veliki broj varijabli, ˇsto oteˇzava pravi izbor i merenje njihove efikasnosti. Stoga je razvijen algoritam zasnovan na tehnici maˇsinskog uˇcenja. Prvi izuˇcavani pristup je primena duboke neuronske mreˇze za kreiranje klasifikacionog algoritma koji je u stanju da razlikuje Ds od mezona koji potiˇcu od kvarkova i gluona. Drugi pristup je da se razvije konvoluciona neuronska mreˇza koriste´ci varijable niskog nivoa, kao ˇsto su impuls ˇcestice i deponovana energija u kalorimetrima. Optimalan rezultat se postiˇze kombinovanjem dve navedene neuronske mreˇze. Dobijeni algoritam je u stanju da identifikuje dˇzetove koji potiˇcu iz Ds mezona u radijativnim W raspadima sa efikasnoˇs´cu 47% za signal i faktorom potiskivanja dˇzetova iz kvarkova i gluona od 100. Pokazano je da se na taj naˇcin na LHC-u mogu poboljˇsati merenja vezana za Ds mezone, posebno u kontekstu retkih raspada.

539.129.3.074(043.3)

Tekst