The performance of the ATLAS Inner Detector Trigger Algorithms in pp collisions at the LHC

The ATLAS Inner Detector trigger algorithms has been running online during data taking with proton-proton collisions at the Large Hadron Collider (LHC) since December 2009. We will present preliminary results on the performance of the algorithms in collisions at centre-of-mass energies of 900GeV and 7TeV, including comparisons to the ATLAS offline tracking algorithms and to simulations. The ATLAS trigger performs the online event selection in three stages. The Inner Detector information is used in the second and third triggering stages, called Level-2 trigger (L2) and Event Filter (EF) respectively, and collectively the High Level Triggers (HLT). The HLT runs software algorithms in a large farm of commercial CPUs and is designed to reject collision events in real time, keeping the most interesting few in every thousand. The average execution time per event at L2(/EF) is about 40ms(/4s) and the Inner Detector trigger algorithms can take only a fraction of that. Within this time, the data from interesting regions of the Inner Detector have to be accessed from central buffers through the network, unpacked, clustered and converted to the ATLAS global coordinates, then pattern recognition follows to identify the trajectories of charged particles (tracks), and finally these tracks are used in combination with other information to accept or reject events, according to whether they satisfy certain trigger signatures. The various client s of the Inner Detector trigger information impose different constraints in the performance of the pattern recognition, in terms of efficiency and fake rate for tracks. We will give an overview of the different uses of the Inner Detector trigger algorithms and exemplify their online performance with results from the use of L2 tracks for the online determination of the LHC beam position.