Implementation and performance of the event filter muon selection for the ATLAS experiment at LHC

The ATLAS trigger system is composed of three levels: an initial hardware trigger level (LVL1) followed by two software-based stages (LVL2 trigger and event filter) included in the high level trigger (HLT) and implemented on processor farms. The LVL2 trigger starts from LVL1 information concerning pointers to restricted so-called regions of interest (ROI) and performs event selection by means of optimized algorithms. If the LVL2 is passed, the full event is built and sent to the event filter (EF) algorithms for further selection and classification. After that, events are finally collected and put into mass storage for subsequent physics analysis. Even if many differences arise in the requirements and in the interfaces between the two HLT stages, they have a coherent approach to event selection. Therefore, the design of a common core software framework has been implemented in order to allow the HLT architecture to be flexible to changes (background conditions, luminosity, description of the detector, etc.). Algorithms working in the event filter are designed to work not only in a general purpose or exclusive mode, but they have been implemented in such a way to process given trigger hypotheses produced at a previous stage in the HLT dataflow (seeding concept). This is done by acting in separate steps, so that decisions to go further in the process are taken at every new step. An overview of the HLT processing steps is given and the working principles of the EF offline algorithms for muon reconstruction and identification (MOORE and MuId) are discussed in deeper detail. The reconstruction performances of these algorithms in terms of efficiency, momentum resolution, rejection power and execution times on several samples of simulated single muon events are presented, also taking into account the high background environment that is expected for ATLAS.