The last CMS week of the year was held two weeks ago, summarizing all the upgrades and changes that happened in 2017, but also the plans of the groups for 2018. Since my service work concerns the L1 muon trigger performance, I was asked by the data performance group conveners to give a talk about the muon trigger algorithms and the improvements that happened last year.
This gave me the perfect excuse to bug all the L1 muon trigger experts, ask a lot of questions and learn all the details on how the different muon trigger algorithms work. So, I decided to share this here, as the trigger algorithms play a very important role in the quality of the data taking by CMS.
The CMS L1 trigger system consists of custom hardware processors and combines information coming from the calorimeter and the muon detectors, to select the most interesting events in a very short, fixed time interval of 3.8 μs. The L1 trigger selects events at a rate of 100 kHz, a factor of 400 less than the bunch crossing rate of 40 MHz.
In 2016, the L1 Trigger underwent a major upgrade in order to cope with the increased instantaneous luminosity and the higher trigger rates. In the context of the upgrade the L1 muon trigger moved from a detector-susbsystem based setup to a regional based setup, where the three muon track finders combine information from the detector systems (resistive plate chambers (RPCs), drift tubes (DT), cathodic strip chambers (CSCs)) and provide muon candidates.
The three muon track finders are separated based on the pseudorapidity (η) coverage of the detector. The Barrel Muon track Finder (BMTF) covers the central region of the detector, in low η regions, and receives track segments from the DTs and the RPCs. The Endcap (EMTF) covers the high η regions and uses informations provided by CSCs and RPCs. The region between the BMTF and the EMTF is covered by the Overlap Track Finder (OMTF), which uses all three detector systems for track reconstruction. The best 36 muon candidates from every track finder are send to the Global Muon Trigger (GMT), which sorts them, removes duplicates and sends the 8 best muons to the Global Trigger.
In more detail, the BMTF algorithm uses TwinMux (concentrator system) to receive track segments from the DTs and RPCs systems and follows 3 steps for the track reconstruction:
- Extrapolator unit: creates pairs of track segments based on their position in the detector.
- Track assembler unit: combines the pairs of the track segments to reconstruct a track and assign a quality to it.
- Assignment unit: computes the spacial coordinates and the transverse momentum of the track.
Before sending the best candidates to the GMT, the BMTF algorithm sorts them and chooses the 36 best ones.
The OMTF combines segments from the DTs, the RPCs and the CSCs in order to fit the best muon candidates. Due to the complex detector geometry in this region, the algorithm uses 52 golden patterns (GP), each corresponding to a transverse momentum range – sign combination. For every hit in all the layers of the detector in OMTF, the difference in the azimuthal direction between the hit and a predefined reference hit is calculated. This difference is used in every GP in order to calculate the log-likelihood, given a transverse momentum – sign hypothesis.
The GP with the maximum non-zero layers log-likelihood and the larger sum of log-likelihoods in all the layers is chosen by the algorithm. The same procedure is repeated 4 times for different reference hits, so that the algorithm can recognize more muons in the same event.
The last part of the L1 muon track finder covers the outer edges of the detector at high η. The EMTF receives track segments from the CSCs called local charge tracks (LCTs) and cluster RPC hits. The algorithm works with pattern formation, based on the values of the azimuthal angle of the LCTs.
In the next step, the LCT and RPC hits are matched to the patterns in the azimuthal direction in order to form tracks. Tracks that are found to be very scattered in polar angles are removed, together with tracks found twice (duplicates).
In the last step, the algorithm sorts the tracks and calculates the transverse momentum, azimuthal and bending angle, and charge of the best tracks, before sending them to the GMT.
Even though 2017 was not an easy year for the L1 muon trigger groups, due to technical problems, the groups managed to achieve a very good performance with higher overall efficiency as compared to 2016, due to the improvements and the upgrades imposed on the algorithms.
The L1 muon trigger rate has been decreased by 20-80% compared to the legacy rate (before the 2016 upgrade) due to the improved transverse momentum assignment, especially for low transverse momentum muons.
I hope that wasn’t too long or too technical and you enjoyed reading about steps that CMS is using, on a first stage, to select the most interesting muons for physics analysis.