32 he online muon identification with the ALAS exeriment at the LHC Abstract he Large Hadron Collider (LHC) at CERN is a roton-roton collider roviding the highest energy and the highest instantaneous luminosity ever achieved at a hadron collider. During 22 runs, bunch crossings occurred every 5 ns. he online event selection system should reduce the event recording rate down to a few hundred Hz in the harsh conditions with many overlaing collisions occurring in one bunch crossing. Muons rovide a clear signature for many hysics rocesses. hey therefore often lay an imortant role in hysics analyses such as the recent discovery of a Higgs boson. he ALAS exeriment deloys a three-level online rocessing scheme. he Level (L) muon trigger system gets its inut from fast muon trigger detectors. Fast sector logic boards select muon candidates, which are assed via an interface board to the central trigger rocessor and then to the High Level rigger (HL). he muon HL is urely software-based and encomasses a Level 2 (L2) and an Event Filter (EF) trigger for a staged trigger aroach. It has access to the data of the recision muon detectors and other detector elements to refine the muon hyothesis. At L2, trigger-secific algorithms are used to increase rocessing seed by making use of look-u tables and simle algorithms. he EF muon triggers benefit from offline reconstruction software to obtain a recise determination of the track arameters. Algorithms based on two different aroaches, namely insideout and outside-in tracking, are used with a conditional OR to obtain maximal efficiency with minimal rocessing time. A full overview of the ALAS muon trigger system is given, the three years of running exerience are summarised and a reort about online erformance such as rocessing time, trigger rates, efficiency and resolution is given. I. I NRODUCION 28/5/23 AL-DAQ-PROC-23-4 Marilyn Marx, University of Manchester, UK on behalf of the ALAS Collaboration HE ALAS detector is a general urose exeriment built at the LHC and is described in detail elsewhere []. It has collected collision data at centre of mass energies of s = 7 ev in 2 and 2 and s = 8 ev in 22. he maximum instantaneous luminosity reached is 7.7 33 cm 2 s in conditions of large ileu with an average number of interactions er bunch crossing, hi, as large as 38. Identifying interesting and tyically rare hysics rocesses therefore becomes more and more of a challenge at the LHC. Events with muons in the final state are distinctive signatures of many rocesses, such as searches for the Higgs boson and hysics beyond the Standard Model (SM) but also recision measurements of the SM. Common features often include isolated muons with large transverse momentum. Figure shows an event dislay for a H ZZ 4 candidate event where the four muons give an invariant mass of 24.6 GeV [2]. For such hysics analyses, it is crucial to trigger efficiently on muons and to recisely determine the general erformance of the ALAS muon trigger. Fig.. Event dislay of a candidate H ZZ 4 event [2]. II. M UON RIGGER S YSEM AND R ECONSRUCION A LGORIHMS he largest sub-detector of the ALAS exeriment is the Muon Sectrometer (MS). Figure 2 shows a quarter-section of the muon system in a lane containing the beam axis. Muons are triggered within a raidity range of η < 2.4. Resistive Plate Chambers (RPC) and hin Ga Chambers (GC) in the barrel ( η <.5) and endca ( η >.5) regions, resectively, are used to trigger whereas Monitored Drift ubes (MD) and Cathode Stri Chambers (CSC) are used for recision measurements of the η coordinate of the muon tracks. he ALAS trigger uses a three-level system which reduces the initial bunch crossing rate of 2 MHz down to an average rate of 4 Hz which is written to tae for offline reconstruction and hysics analysis. A further 2 Hz are written to tae for delayed reconstruction and analysis. A total of 5-8 Hz is allocated to the muon trigger signatures. he muon system can reconstruct muons in standalone (SA) mode and in combination with the Inner Detector (ID) tracking systems. he selection of events with muons reflects this staged aroach of L, L2 and EF triggers with increasing comlexity and recision for each level. he L muon trigger system is hardware-based and receives inut from fast muon trigger detectors (RPC and GC) using low granularity signals. As shown in Figure 2, there are three lanes of dedicated trigger detectors. A L muon trigger can be formed out of either a 2-station or a 3-station coincidence from hits in time and sace in those lanes. Fast sector logic boards select muon candidates, based on their, using so-
32 2 RPC 3 RPC 2 RPC low MD MD MD high GC EI M D GC 2 GC M D GC 3 low Entries / 2.5 ms 3 2 ALAS rigger Oerations Data 2 s = 7 ev mean = 7 ms ile Calorimeter GC FI M D 5 5 m high XX-LLV4 2 4 6 8 ime [ms] Fig. 2. Quarter-section view of the ALAS muon detectors in a lane containing the beam axis [3] called regions of interest (RoI) which will be assed to the Central rigger Processor and then used to seed the subsequent software-based L2 and EF trigger levels, which are collectively called HL. At L2, each L muon hyothesis is refined by including the recise information from the MDs in the RoI. his allows to reconstruct MS-only (or SA) muons. Combined (CB) muons are reconstructed using ID and MS tracks. An isolation algorithm is alied at this stage using ID track and calorimeter information. he EF uses modified offline muon reconstruction algorithms with access to the full event information. his gives the ossibility to imlement comlex selection criteria. wo different algorithms oerate at the EF to reconstruct CB muons, an inside-out (IO) and an outside-in (OI) aroach. he IO aroach starts with the ID track information and extraolates its trajectory to the MS to build a CB muon, while the OI aroach starts from the MS track to find matches in the ID. In 22, these two aroaches were merged so that the OI algorithm is ran first and the IO algorithm is only ran if the revious stage was not able to reconstruct a muon. In 2, the threshold of the rimary trigger, which is the lowest unrescaled single muon trigger, was set to GeV for MS-only muons. As the LHC luminosity increased, the latter requirement was changed to CB muons. In 2, the rimary trigger threshold was increased from 3 GeV to 8 GeV. he L seed was initially based on a 2-station coincidence with a GeV threshold (L MU) and later on a 3-station coincidence with a GeV threshold (L MU). Finally, in 22 two different rimary triggers were used. he first trigger chain (EF mu24i tight) required isolated muons with a threshold of 24 GeV and a 3-station coincidence at L while the second trigger chain (EF mu36) did not require isolation and a threshold of 36 GeV. III. PROCESSING IME he rocessing times of the HL algorithms have been determined on Intel c Core M 2 Duo CPU E85 comuters Entries / 3 ms 3 2 ALAS rigger Oerations Data 2 s = 7 ev EF outside in: mean = 267 ms EF inside out: mean = 9 ms 5 5 2 25 3 35 ime [ms] Fig. 3. Processing times of the HL algorithms er RoI for the L2 CB reconstruction chain and the EF trigger chains [3]. with clock seed of 3.6 GHz by running on raw data from the events selected by jet, τ or missing transverse energy triggers at a luminosity of 3. 33 cm 2 s. In 2, the HL comuting nodes consisted mainly of quad-core CPUs running at 2.5 GHz. Figure 3 shows the rocessing time for the L2 CB muon trigger chain. he mean rocessing time was 7 ms out of which the L2 ID tracking consumed 63%, the L2 SA about 9% and the remainder was sent on unacking and decoding the data. he EF rocessing times for the OI and IO algorithms, shown in Figure 3, were 267 ms and 9 ms, resectively. he rocessing time for the OI algorithm was comosed of 65% for the SA algorithm, 29% for the CB algorithm and the rest for data unacking and decoding. he IO algorithm has a longer rocessing time as it needs to extraolate all the ID tracks, which have a much higher multilicity comared to the MS esecially in high occuancy events, in an RoI to the MS. he execution times of the HL chains have been measured to be well within the time restrictions of about 4 ms and 4 s for L2 and EF, resectively, and so allowed a stable trigger oeration.
32 3.9.8.7.6.5.4.3.2. ALAS Preliminary Data 2 η <.5 3 4 5 6 7 8 2 3 4 s = 7 ev L_MU (2 station coincidence) L_MU (3 station coincidence) 2 Rate [Hz] 8 sloe (total)= 6. ±.5 ALAS Preliminary sloe (fake)= 2. ±. Data 22 s = 8 ev 6 4 2 8 6 4 2 L_MU5 Barrel L_MU5 Barrel fake Linear Fit 2 3 4 5 6 7 32 Inst. Lum. [ cm 2 s ] Fig. 4. L trigger efficiency with resect to isolated offline CB muons as a function of in the barrel region [3]. IV. LEVEL PERFORMANCE All muon trigger efficiencies have been evaluated using a tag-and-robe method on Z + data samles. he trigger efficiencies of the L MU and L MU chains with resect to offline muons were measured in 2 data and found to be aroximately 8% and 7%, resectively, in the barrel region ( η <.5) as shown as a function of the offline muon in Figure 4. In the endca regions (.5 < η < 2.4) the efficiency is above 9% for both triggers. he L inefficiencies are mainly due to losses in geometrical accetance. hese are articularly relevant in the barrel region due to inactive regions that allow for services and due to the resence of a feet region where the detector stands on the ground. he L muon trigger rates have been determined in 22 data as a function of instantaneous luminosity. Figure 5 shows the barrel and endca rates for a > 5 GeV selection (L MU5). he dots reresent the total measured rate whereas the triangles indicate the fake rates. A fake trigger is defined as a trigger that does not match with an offline reconstructed muon within a cone of R =.4. he fake fraction is higher in the endcas due to tracking outside the magnetic field, measuring only the deviation from the direction ointing towards the nominal IP osition, and the smaller level arm with resect to the barrel layout. Fake triggers are understood to be mainly due to secondary articles, for examle rotons roduced in dense materials such as the magnets. A linear fit is alied to both the total and fake rate curves, the resulting sloes are shown in Figure 5. A. Resolution V. HL PERFORMANCE he erformance and in articular the resolution of the EF algorithms have been studied in 2 fb of 2 data using Z + samles with a > 8 GeV cut alied to R = η 2 + φ 2 Rate [Hz] 9 sloe (total)= 4 ± sloe (fake)= 66 ± 8 L_MU5 Endcas 7 L_MU5 Endcas fake 6 Linear Fit 5 4 3 2 ALAS Preliminary Data 22 s = 8 ev 2 3 4 5 6 7 32 Inst. Lum. [ cm 2 s ] Fig. 5. L muon trigger rates as a function of instantaneous luminosity in the barrel and end ca regions measured in.8 fb of 8 ev data. All uncertainties are statistical [4]. the offline muons. he residuals between the EF and offline muon track arameters (/ ) were evaluated in bins of. he widths of the residual distributions were then extracted in each bin with a Gaussian fit. he residuals were comuted by comaring the / values of SA and CB muon EF tracks to the offline CB muon candidates, namely σ((/ ) EF (/ ) offline ). Figure 6 shows the resolution of the inverse for the EF SA (triangles) as well as the EF CB OI (circles) and IO (squares) algorithms for muons in the endca regions. Similar results are obtained for the barrel region. he inclusion of ID tracks gives a clear imrovement in the resolution, articularly at lower values. he resolution uncertainties are equal to the σ value from the Gaussian fit in each slice. hese results illustrate the good agreement between online and offline track arameters. he level of agreement is comarable for the IO and OI algorithms. B. Isolation he EF outut rate can be controlled by alying isolation criteria to the muons. During 22 data taking, it was ossible
32 4 ] resolution [GeV /.3.25.2.5..5 ALAS Preliminary Data 2 s= 7 ev η >.5 EF SA EF inside out EF outside in 2 3 4 5 6 7 8 9 Fig. 6. Resolution of the inverse of the EF SA and CB algorithms in Z + samles as a function of the offline muon [3]. to kee the single muon trigger threshold below 25 GeV by choosing an isolation of / <.2, where is the sum of tracks that have > GeV found in a R <.2 cone around the muon candidate whose transverse momentum is. Figure 7 shows the rejection for inclusive L2 muon candidates as a function of the efficiency for muons from Z boson decays obtained by selecting muons with > 8 GeV and alying otimised track and calorimeter isolation criteria. he rejection is defined as the ratio between the number of L2 CB muons rejected by the L2 isolated trigger algorithm and the total number of L2 CB muons fed into the algorithm. At a given rejection value, the isolation criteria are changed to achieve the best efficiency for muons from Z boson decays. he dots on the curve indicate two different sets of isolation criteria, the tight isolation (E <.4 GeV and < 5.7 GeV) and the loose isolation (E < 2.7 GeV). he amount of data used for this study corresonds to 8 b. Figure 8 shows the mean value of the track variable / as a function of ile-u for the robe muons. he alication of the isolation cut decreases the trigger efficiency by less than.% as shown in Figure 9. he amount of data used to roduce these lots corresonds to 3.79 fb and the vertical error bars reresent statistical uncertainties. C. For 2 runs, both the OI and IO muon EF trigger algorithms were running in arallel to maximise the online selection efficiency. In 22, the increase in instantaneous luminosity made it necessary to otimise the EF steering configuration to attain higher efficiency while not increasing the required data acquisition bandwidth. he OI and IO algorithms have therefore been merged as described in section 2. his combination has increased the overall efficiency by % as can be seen in Figure as a function of. he Rejection.9.8.7.6.5 ALAS Preliminary Data 2 (tight isolation) s = 7 ev L2 Isolation (loose isolation).5.6.7.8.9 Fig. 7. Background rejection for L2 CB muon candidates versus signal efficiency for muons from Z boson decays [3]. he oerating oints for tight and loose isolation triggers are shown. Mean of track isolation variable.25.2.5..5 ALAS Preliminary Data 22 ( s = 8 ev) Ldt = 3.79 fb 5 5 2 25 3 35 <> Fig. 8. EF track isolation variable as a function of average number of interactions er bunch crossing [4]. circles indicate the OI and IO combination while the crosses and diamonds indicate the efficiency of the searate OI and IO algorithms, resectively. Figures and 2 show the resulting efficiency of the EF mu24i tight trigger chain as a function of the offline muon and, resectively, in the barrel and endca regions. he total efficiency is higher in the endcas ( 9%) than in the barrel ( 7%) due to reviously mentioned geometrical accetance losses at L. A stable efficiency lateau is reached quickly for > 24GeV and the efficiency is robust against ile-u. VI. CONCLUSION he ALAS muon trigger is crucial for many hysics analyses and its erformance in the last three years of data taking
32 5..999.998.997.996.995 ALAS Preliminary Data 22 ( s = 8 ev) / () <.2 Ldt = 3.79 fb 25 3 35 4 45 5 HL.2.98.96.94.92.9.88.86.84.82.8 ALAS Preliminary Data 22 ( s = 8 ev) Ldt =.74 fb η <.5 outside in inside out 3 4 5 6 7 8 9..999.998.997.996 ALAS Preliminary Data 22 ( s = 8 ev) / () <.2 Ldt = 3.79 fb.995 5 5 2 25 3 35 <> HL.2.98.96.94.92.9.88.86.84.82.8 ALAS Preliminary Data 22 ( s = 8 ev) Ldt =.74 fb η >.5 outside in inside out 3 4 5 6 7 8 9 Fig. 9. of the trigger isolation requirement with resect to offline isolated muons from Z boson decays as a function of and average number of interactions er bunch crossing [4]. Fig.. HL trigger efficiency of the EF mu24i tight chain with resect to isolated offline muons assing the L trigger as a function of in the barrel and endca regions [4]. has been resented. he trigger algorithm rocessing times have allowed for stable trigger oeration at high luminosities. he L muon trigger efficiency with resect to offline varies between 7-8% and 9% in the barrel and endca regions, resectively. It is mostly limited by the geometrical accetance and hit efficiency of the L detectors. he HL efficiency with resect to L is close to unity. Single muon triggers have been used as rimary triggers throughout data taking while gradually raising the thresholds and adding muon isolation requirements to deal with increasing instantaneous luminosities. he overall good erformance of the muon trigger system with stable resolution and little and ile-u deendence have made it ossible to ublish many imortant hysics analyses. A robust erformance of the ALAS muon trigger system and the ability to maintain a high efficiency with large background rejection will be vital for future runs at even higher instantaneous luminosity and ile-u conditions at the LHC. REFERENCES [] ALAS Collaboration, he ALAS Exeriment at the CERN Large Hadron Collider, JINS 3 S83 (28). [2] htts://twiki.cern.ch/twiki/bin/view/atlaspublic/ EventDislaysFromHiggsSearches. [3] ALAS Collaboration, Performance of the ALAS muon trigger in 2, ALAS-CONF-22-99, htts://cds.cern.ch/record/4626. [4] htts://twiki.cern.ch/twiki/bin/view/atlaspublic/ MuonriggerPublicResults
32 6.8 ALAS Preliminary.8 ALAS Preliminary.6.4.2 Data 22 ( s = 8 ev) 2 3 4 5 6 7 8 9 η <.5.6.4.2 5 5 2 25 3 35 Data 22 ( s = 8 ev) η <.5 <>.8 ALAS Preliminary.8 ALAS Preliminary.6.4.2 Data 22 ( s = 8 ev) 2 3 4 5 6 7 8 9 η >.5.6.4.2 Data 22 ( s = 8 ev) 5 5 2 25 3 35 η >.5 <> Fig.. of the EF mu24i tight chain with resect to isolated offline muons as a function of muon in the barrel and endca regions [4]. Fig. 2. of the EF mu24i tight chain with resect to isolated offline muons as a function of the average number of interactions er bunch crossing in the barrel and endca regions [4].