Neural Network Analysis in Higgs Search using $t\overline{t}H,H \to b\overline{b}$ and TAG Database Development for ATLAS

The Large Hadron Collider, LHC, at Conseil Europeen pour la Recherche Nucleaire, CERN, in Geneva, Switzerland, is an international physics project of unprecedented scale. First proton beams were circulated in the LHC in 2008. The ATLAS Collaboration, an international group of 2000 analysts, scientists, software developers and hardware experts, seeks to push the boundaries of our current understanding of the Universe, and our ability to undertake such studies. A central physics focus of the ATLAS experiment is study of a Higgs boson, a theoretically predicted particle, as yet unobserved in nature. In this thesis, a Neural Network is adopted and developed as an analysis method in a study of a Standard Model Higgs boson in the low mass Higgs range, using the physics channel ttH;H ! bb and Higgs mass mH = 120 GeV. The Neural Network analysis shows that a neural network method can give an improvement in sensitivity of the ttH;H ! bb channel. A set of Event Characteristics, associated with a topology where the existence of a reconstructed Higgs boson is not required in each event are dened and it is demonstrated that these characteristics, when used in a neural network, can improve the sensitivity of the channel by improving separation of signal and background events. The neural network analysis uses a collection of Generic Event Characteristics, a neural network of layout 36 : 8 : 4 : 1, 1000 learning cycles and 734033 ttH;H ! bb simulated signal and background events, for an integrated luminosity of 1fb􀀀1, to give an output sensitivity of 4:74. We see that the neural network analysis method as described in this analysis improves the sensitivity of the channel from that of the Cuts-Based Analysis performed in previous studies. In the quest for new and multipurpose physics searches and studies, ATLAS will produce data of unprecedented volume and rate in Particle Physics. As analysts are internationally located, data must be accessible across worldwide collaborating institutions. A signicant challenge for the ATLAS collaboration lies in developing the capacity in computing terms to manage an unprecedented data challenge in a uid, sound and transparent way. The ATLAS Event Level Metadata System, TAG Database, is a central part of the ATLAS Computing system. The Event Level Metadata system captures information about ATLAS physics events on an event by event basis, and oers later access to the events for analysis. In this thesis, developments and implementation of the Event Level Metadata system are presented in terms of three studies, these are Feasability, Scalability and Accessibility. Feasibility studies demonstrate that an Event Level Metadata system can operate within the larger ATLAS software system and gathered information on the implications for Event Level Metadata system development. Scalability studies present implementation and performance of a realistic terabyte scale relational TAG Database and demonstrate that an Event Level Metadata system at terabyte scale is achievable. Accessibilty studies present the development of a web interface to the Event Level Metadata system. Studies in this thesis therefore demonstrate that an Event Level Metadata can be integrated with the ATLAS software system, develop solutions for integration, prove that an Event Level Metadata relational database can scale to ATLAS terabyte size, present performance results for a realistic ATLAS scale system and develop a user interface to the Event Level Metadata system.