Online Data Reduction using Track and Vertex Reconstruction on GPUs for the Mu3e Experiment

Abstract

Abstract.The Mu3e experiment searches for the lepton flavour violating decay μ + → e + e − e + , aiming to achieve a sensitivity of 2 · 10 −15 in its first phase and ultimately aspiring to a final sensitivity of 10 −16 . During the first phase of the experiment, a muon rate of ∼ 10 8 μ/s will be available, resulting in a data rate of ∼ 80 Gbit/s. The trigger-less readout system is based on optical links and switching FPGAs sending the complete detector data for a time slice to one node of the filter farm. A full online reconstruction is necessary to reduce the data rate to a manageable amount to be written to disk. Graphics processing units (GPUs) are used to fit tracks with a non-iterative 3D tracking algorithm for multiple scattering dominated resolution. In addition, a three track vertex selection is performed by calculating the vertex position from the intersections of the tracks. Together with kinematic cuts, this allows for a reduction of the output data rate to below 100 MB/s using 12 DAQ PCs. The Mu3e ExperimentThe Mu3e experiment [1] is designed to search for the lepton flavour violating decay μ + → e + e − e + . In the Standard Model (SM) this process is only allowed via neutrino mixing in loops, and it is heavily suppressed to below a branching fraction of 10 −54 [2]. Therefore any observation of lepton flavour violation in the charged lepton sector is a clear indication for new physics. Various models beyond the SM predict charged lepton flavour violation at a level to which future detectors are sensitive. The current limit on the μ + → e + e − e + branching fraction was set by the SINDRUM experiment at 10 −12 [3]. The Mu3e experiment aims to reach a single event sensitivity of 2 · 10 −15 in a first phase of the experiment with an existing beamline at the Paul Scherrer Institute in Switzerland (PSI), and aspires to a final sensitivity of 10 −16 with an upgraded beamline. This will improve the limit by four orders of magnitude compared to the last experiment. In these proceedings, the focus is on the experimental setup planned for the existing beamline.The detector design is driven by the requirements to distinguish the signal decay from background processes. Within the detector volume, muons will be stopped in a target and decay at rest. In the case of a signal event, two positrons and one electron are coincident in time and originate from one single vertex as shown in figure 1a. The combined energy of the three particles is equal to the rest mass of the muon and their combined momentum is zero. One source of background is radiative muon decay a

The search for Lepton-Flavour Violation with the Mu3e Experiment

“…Selected data frames get written to disk with a reduction factor ∼100. We showed that with commercial GPUs available today, and an optimized reconstruction algorithm [8], this reduction factor can be achieved with a 98% reconstruction efficiency [9] for signal events in the acceptance of the detector.…”

Track reconstruction for theMu3eexperiment based on a novel Multiple Scattering fit

“…As it is not possible to store all the data produced by the detector, the online filter farm will be used to reduce the data rate by performing a full reconstruction of each frame [5]. To cope with such high rates and be able to fully reconstructed each frame Mu3e will use the novel 3-dimensional Multiple Scattering (MS) triplet fit [7].…”

Allen: A High-Level Trigger on GPUs for LHCb

“…All other R&D efforts are intended for future experiments or upgrades. In most proposals, data from a single sub-detector is analyzed on the GPU at a significantly lower data rate than 40 Tbit/s [9][10][11]. For some, the GPU coprocessor performs track reconstruction, but event selection or data reduction occur on x86 CPUs [10].…”

“…For some, the GPU coprocessor performs track reconstruction, but event selection or data reduction occur on x86 CPUs [ 10 ]. In other cases, event selection for a single physics signature runs on the GPU [ 9 , 11 ]. For Run 3, ALICE plans to perform track reconstruction of more than one subdetector and data compression on the GPU, at a data rate of 5 Tbit/s [ 12 ].…”