Track Reconstruction in the STAR TPC with a CA Based Approach

STAR [1] is an active collider heavy-ion experiment at RHIC/BNL (Upton, USA). The main tracking detector of the experiment is the Time Projection Chamber (TPC). Due to the collision rate increasing (up to 40 KHz for Au+Au 200 GeV/n and up to 1 MHz for pp 500 GeV) and recording rate (up to 10 KHz) the STAR experiment requires a fast track reconstruction procedure which can deal with high track densities (up to 5000 tracks). A track reconstruction algorithm based on the Cellular Automaton (CA) was proposed for the STAR TPC reconstruction. The CA algorithm is based on local reconstruction and therefore is robust, fast and easily parallelizable, that makes it perfectly suitable for high track densities and collision rates. The algorithm has been already implemented in the ALICE HLT TPC CA track finder package [2]. The CA reconstruction is performed in two stages: the local reconstruction of tracks in each TPC sector (sector reconstruction) and the tracks merging between sectors (merger). It is parallelized at both data (using the SIMD instruction set) and task (using the Intel Threading Building Blocks technology) levels. The memory usage in the algorithm is optimized. The CA algorithm has been adapted for the STAR TPC geometry, which is similar to the ALICE TPC geometry. The sector reconstruction part has been divided in two iterations: reconstruction of high and low momentum tracks. More accurate criteria for merging of tracks have been implemented. Detailed optimization of the code of the algorithm has been done in order to make it more efficient, robust and flexible.