The COMPASS Setup for Physics with Hadron Beams
Ph. Abbon, C. Adolph, R. Akhunzyanov, Yu. Alexandrov, M.G. Alexeev, G.D. Alexeev, A. Amoroso, V. Andrieux, V. Anosov, A. Austregesilo, B. Badelek, F. Balestra, J. Barth, G. Baum, R. Beck, Y. Bedfer, A. Berlin, J. Bernhard, K. Bicker, E.R. Bielert, J. Bieling, R. Birsa, J. Bisplinghoff, M. Bodlak, M. Boer, P. Bordalo, F. Bradamante, C. Braun, A. Bressan, M. Buechele, E. Burtin, L. Capozza, P. Ciliberti, M. Chiosso, S.U. Chung, A. Cicuttin, M. Colantoni, D. Cotte, M.L. Crespo, Q. Curiel, T. Dafni, S. Dalla Torre, S.S. Dasgupta, S. Dasgupta, O.Yu. Denisov, D. Desforge, A.M. Dinkelbach, S.V. Donskov, N. Doshita, V. Duic, W. Duennweber, D. Durand, M. Dziewiecki, A. Efremov, C. Elia, P.D. Eversheim, W. Eyrich, M. Faessler, A. Ferrero, M. Finger, M. Finger jr., H. Fischer, C. Franco, N. du Fresne von Hohenesche, J.M. Friedrich, V. Frolov, L. Gatignon, F. Gautheron, O.P. Gavrichtchouk, S. Gerassimov, R. Geyer, A. Giganon, I. Gnesi, B. Gobbo, S. Goertz, M. Gorzellik, S. Grabmueller, A. Grasso, M. Gregori, B. Grube, T. Grussenmeyer, A. Guskov, F. Haas, D. von Harrach, D. Hahne, R. Hashimoto, F.H. Heinsius, F. Herrmann, F. Hinterberger, Ch. Hoeppner, N. Horikawa, N. d'Hose, S. Huber, S. Ishimoto, A. Ivanov, Yu. Ivanshin, T. Iwata, R. Jahn, V. Jary, P. Jasinski, et al. (132 additional authors not shown)
EEUROPEAN ORGANIZATION FOR NUCLEAR RESEARCH
COMPASS
CERN-PH-EP-2014–24705 October 2014
The COMPASS Setup for Physics with Hadron Beams
The COMPASS Collaboration
Abstract
The main characteristics of the COMPASS experimental setup for physics with hadron beams aredescribed. This setup was designed to perform exclusive measurements of processes with severalcharged and/or neutral particles in the final state. Making use of a large part of the apparatus that waspreviously built for spin structure studies with a muon beam, it also features a new target systemas well as new or upgraded detectors. The hadron setup is able to operate at the high incidenthadron flux available at CERN. It is characterised by large angular and momentum coverages, largeand nearly flat acceptances, and good two and three-particle mass resolutions. In 2008 and 2009it was successfully used with positive and negative hadron beams and with liquid hydrogen andsolid nuclear targets. This article describes the new and upgraded detectors and auxiliary equipment,outlines the reconstruction procedures used, and summarises the general performance of the setup. key words: fixed target experiment, hadron spectroscopy, silicon microstrip detectors , GEM detector,drift chambers, RICH detector, calorimetry, front-end electronics, data acquisition, data reconstruc-tion, Monte-Carlo simulationPACS 13.85.-t, 07.05.Fb, 07.05.Hd, 07.05.Kf, 29.25.Pj, 29.30.-h, 29.40.Cs, 29.40.Gx, 29.40.Ka,29.40.Mc, 29.40.Vj, 29.40.Wk, 29.27.Fh, 29.85.Ca (to be submitted to Nucl. Instr. and Meth. A) a r X i v : . [ phy s i c s . i n s - d e t ] O c t he COMPASS Setup for Physics with Hadron Beams 1 The COMPASS Collaboration
Ph. Abbon , C. Adolph , R. Akhunzyanov , Yu. Alexandrov , M.G. Alexeev , G.D. Alexeev ,A. Amoroso , V. Andrieux , V. Anosov , A. Austregesilo , B. Badełek , F. Balestra ,J. Barth , G. Baum , R. Beck , Y. Bedfer , A. Berlin , J. Bernhard , K. Bicker , E. R. Bielert ,J. Bieling , R. Birsa , J. Bisplinghoff , M. Bodlak , M. Boer , P. Bordalo , F. Bradamante ,C. Braun , A. Bressan , M. Büchele , E. Burtin , L. Capozza , P. Ciliberti , M. Chiosso ,S.U. Chung , A. Cicuttin , M. Colantoni , D. Cotte , M.L. Crespo , Q. Curiel , T. Dafni ,S. Dalla Torre , S.S. Dasgupta , S. Dasgupta , O.Yu. Denisov , D. Desforge , A.M. Dinkelbach ,S.V. Donskov , N. Doshita , V. Duic , W. Dünnweber , D. Durand , M. Dziewiecki ,A. Efremov , C. Elia , P.D. Eversheim , W. Eyrich , M. Faessler , A. Ferrero , M. Finger ,M. Finger jr. , H. Fischer , C. Franco , N. du Fresne von Hohenesche , J.M. Friedrich ,V. Frolov , L. Gatignon , F. Gautheron , O.P. Gavrichtchouk , S. Gerassimov , R. Geyer ,A. Giganon , I. Gnesi , B. Gobbo , S. Goertz , M. Gorzellik , S. Grabmüller , A. Grasso ,M. Gregori , B. Grube , T. Grussenmeyer , A. Guskov , F. Haas , D. von Harrach , D. Hahne ,R. Hashimoto , F.H. Heinsius , F. Herrmann , F. Hinterberger , Ch. Höppner , N. Horikawa ,N. d’Hose , S. Huber , S. Ishimoto , A. Ivanov , Yu. Ivanshin , T. Iwata , R. Jahn , V. Jary ,P. Jasinski , P. Jörg , R. Joosten , E. Kabuß , B. Ketzer , G.V. Khaustov , Yu.A. Khokhlov ,Yu. Kisselev , F. Klein , K. Klimaszewski , J.H. Koivuniemi , V.N. Kolosov , K. Kondo ,K. Königsmann , I. Konorov , V.F. Konstantinov , A.M. Kotzinian , O. Kouznetsov ,M. Krämer , Z.V. Kroumchtein , N. Kuchinski , R. Kuhn , F. Kunne , K. Kurek , R.P. Kurjata ,A.A. Lednev , A. Lehmann , M. Levillain , S. Levorato , J. Lichtenstadt , A. Maggiora ,A. Magnon , N. Makke , G.K. Mallot , C. Marchand , J. Marroncle , A. Martin ,J. Marzec , J. Matousek , H. Matsuda , T. Matsuda , G. Menon , G. Meshcheryakov , W. Meyer ,T. Michigami , Yu.V. Mikhailov , Y. Miyachi , M.A. Moinester , A. Nagaytsev , T. Nagel ,F. Nerling , S. Neubert , D. Neyret , V.I. Nikolaenko J. Novy , W.-D. Nowak , A.S. Nunes ,A.G. Olshevsky , I. Orlov , M. Ostrick , R. Panknin , D. Panzieri , B. Parsamyan , S. Paul ,G. Pesaro , V. Pesaro , D.V. Peshekhonov , C. Pires , S. Platchkov , J. Pochodzalla ,V.A. Polyakov , J. Pretz , M. Quaresma , C. Quintans , S. Ramos , C. Regali , G. Reicherz ,J-M. Reymond , E. Rocco , N.S. Rossiyskaya , J.-Y. Rousse , D.I. Ryabchikov , A. Rychter ,A. Samartsev , V.D. Samoylenko , A. Sandacz , S. Sarkar , I.A. Savin , G. Sbrizzai ,P. Schiavon , C. Schill , T. Schlüter , K. Schmidt , H. Schmieden , K. Schönning ,S. Schopferer , M. Schott , O.Yu. Shevchenko , L. Silva , L. Sinha , S. Sirtl , M. Slunecka ,S. Sosio , F. Sozzi , A. Srnka , L. Steiger , M. Stolarski , M. Sulc , R. Sulej , H. Suzuki ,A. Szabelski , T. Szameitat , P. Sznajder , S. Takekawa , J. ter Wolbeek , S. Tessaro ,F. Tessarotto , F. Thibaud , V. Tskhay , S. Uhl , I. Uman , M. Virius , L. Wang , T. Weisrock ,Q. Weitzel , M. Wilfert , R. Windmolders , H. Wollny , K. Zaremba , M. Zavertyaev ,E. Zemlyanichkina , M. Ziembicki and A. Zink Universität Bielefeld, Fakultät für Physik, 33501 Bielefeld, Germany i2 Universität Bochum, Institut für Experimentalphysik, 44780 Bochum, Germany ip3
Universität Bonn, Helmholtz-Institut für Strahlen- und Kernphysik, 53115 Bonn, Germany i4 Universität Bonn, Physikalisches Institut, 53115 Bonn, Germany i5 Institute of Scientific Instruments, AS CR, 61264 Brno, Czech Republic j6 Matrivani Institute of Experimental Research & Education, Calcutta-700 030, India k7 Joint Institute for Nuclear Research, 141980 Dubna, Moscow region, Russia l8 Universität Erlangen–Nürnberg, Physikalisches Institut, 91054 Erlangen, Germany i9 Universität Freiburg, Physikalisches Institut, 79104 Freiburg, Germany ip10
CERN, 1211 Geneva 23, Switzerland Technical University in Liberec, 46117 Liberec, Czech Republic j The COMPASS Collaboration LIP, 1000-149 Lisbon, Portugal m13
Universität Mainz, Institut für Kernphysik, 55099 Mainz, Germany i14
University of Miyazaki, Miyazaki 889-2192, Japan n15
Lebedev Physical Institute, 119991 Moscow, Russia Ludwig-Maximilians-Universität München, Department für Physik, 80799 Munich, Germany io17
Technische Universität München, Physik Department, 85748 Garching, Germany io18
Nagoya University, 464 Nagoya, Japan n19
Charles University in Prague, Faculty of Mathematics and Physics, 18000 Prague, Czech Republic j20
Czech Technical University in Prague, 16636 Prague, Czech Republic j21
State Scientific Center Institute for High Energy Physics of National Research Center ‘KurchatovInstitute’, 142281 Protvino, Russia CEA IRFU/SPhN Saclay, 91191 Gif-sur-Yvette, France p23
Tel Aviv University, School of Physics and Astronomy, 69978 Tel Aviv, Israel q24
University of Trieste, Department of Physics, 34127 Trieste, Italy Trieste Section of INFN, 34127 Trieste, Italy Abdus Salam ICTP, 34151 Trieste, Italy University of Turin, Department of Physics, 10125 Turin, Italy University of Eastern Piedmont, 15100 Alessandria, Italy Torino Section of INFN, 10125 Turin, Italy National Centre for Nuclear Research, 00-681 Warsaw, Poland r31
University of Warsaw, Faculty of Physics, 00-681 Warsaw, Poland r32
Warsaw University of Technology, Institute of Radioelectronics, 00-665 Warsaw, Poland r33
Yamagata University, Yamagata, 992-8510 Japan na Also at Instituto Superior Técnico, Universidade de Lisboa, Lisbon, Portugal b Also at Department of Physics, Pusan National University, Busan 609-735, Republic of Korea andat Physics Department, Brookhaven National Laboratory, Upton, NY 11973, U.S.A. c Supported by the DFG Research Training Group Programme 1102 “Physics at Hadron Accelera-tors” d Also at Chubu University, Kasugai, Aichi, 487-8501 Japan ne Also at KEK, 1-1 Oho, Tsukuba, Ibaraki, 305-0801 Japan f Present address: Universität Bonn, Helmholtz-Institut für Strahlen- und Kernphysik, 53115 Bonn,Germany g Also at Moscow Institute of Physics and Technology, Moscow Region, 141700, Russia h present address: RWTH Aachen University, III. Physikalisches Institut, 52056 Aachen, Germany i Supported by the German Bundesministerium für Bildung und Forschung j Supported by Czech Republic MEYS Grants ME492 and LA242 k Supported by SAIL (CSR), Govt. of India l Supported by CERN-RFBR Grants 08-02-91009 and 12-02-91500 m Supported by the Portuguese FCT - Fundação para a Ciência e Tecnologia, COMPETE and QREN,Grants CERN/FP/109323/2009, CERN/FP/116376/2010 and CERN/FP/123600/2011 n Supported by the MEXT and the JSPS under the Grants No.18002006, No.20540299 and No.18540281;Daiko Foundation and Yamada Foundation o p Supported by EU FP7 (HadronPhysics3, Grant Agreement number 283286) q Supported by the Israel Science Foundation, founded by the Israel Academy of Sciences and Hu-manities r Supported by the Polish NCN Grant DEC-2011/01/M/ST2/02350 * Deceasedhe COMPASS Setup for Physics with Hadron Beams 3
The goal of the COMPASS experiment at CERN is a better understanding of the structure and dynamicsof hadrons. At the relevant length scales of ∼ − m the strong coupling constant α s approaches unity,which is the domain of non-perturbative Quantum Chromodynamics (QCD). Using a 160 −
200 GeV /c muon beam, COMPASS studies the nucleon spin structure by deep inelastic scattering off a polarised LiD or NH target [1]. Experiments with hadron beams of 190 GeV /c , which started in 2008, aimat precision spectroscopy of light mesons and baryons with masses up to 3 GeV /c , the identificationand systematic study of possible exotic configurations with gluonic degrees of freedom or multi-quarksystems, as well as the study of processes governed by chiral dynamics and tests of predictions of chiralperturbation theory.These experiments require a state-of-the-art spectrometer with high acceptance and high resolution forcharged and neutral particles in order to perform exclusive measurements of multi-particle final statesover a wide kinematic range. Three different mechanisms contribute to the production of a system X ,as shown in Fig. 1: diffractive dissociation and central production, which can be described to proceedvia the exchange of one or two Reggeons R , respectively, between beam hadron and target nucleus N ,and photo-production in the Coulomb field of a nucleus at very low values of momentum transfer. Inall these processes the final-state particles are emitted mostly in forward direction, which requires anexcellent angular resolution of the spectrometer very close to and even within the beam envelope. Forthe interpretation of the data using partial-wave analysis (PWA) tools, a large and uniform acceptanceover the whole kinematic domain under study is mandatory.The relative contributions of the above-mentioned processes to a data sample can be varied by applyingdifferent trigger conditions and by adjusting kinematic selection criteria in the analysis. At intermediate-to-large values of momentum transfer, the cross section for reactions mediated by Pomerons, i.e. Reggeonswith vacuum quantum numbers, is large, of the order of 1 − X that do not fit into the naive constituent quark model but areallowed by QCD, like glueballs or hybrids which carry gluonic degrees of freedom, or multi-quark sys-tems. States with gluonic degrees of freedom are generally believed to be enhanced in reactions in whichPomerons are exchanged. A small but significant contribution of a spin-exotic partial wave with non- qq (cid:48) quantum numbers J P C = − + , consistent with the π ( ) , was confirmed by COMPASS usingdata taken in 2004 [2]. However, an unambiguous understanding of the underlying structure of this andmany other light-hadron states requires experiments with higher statistical accuracy, employing differ-ent production mechanisms and observation of the same system X in different decay channels. At verysmall values of momentum transfer, the cross section is dominated by Primakoff reactions, i.e. Coulombscattering of pions or kaons off quasi-real photons emitted from a nuclear target. The dynamics of thescattering of a beam π into πγ , π − π , 3 π , etc., at low-energy, i.e. from threshold up to a few pion masses,is predicted by chiral perturbation theory (ChPT). COMPASS can thus scrutinise ChPT predictions of Fig. 1:
Production mechanisms employed in COMPASS for (left) diffractive dissociation, (middle) central pro-duction, (right) photo-production by quasi-real photons γ , with π denoting the beam particle (can be also p , K ),and N the target nucleon or nucleus. The COMPASS Collaborationchiral dynamics [3] and of fundamental low-energy parameters such as the polarisabilities of mesons.Compared to previous experiments, the main advantages of the COMPASS setup are the possibility tostudy reactions with different projectiles in high-intensity beams with up to 10 part ./ s and to reconstructfinal states containing both neutral and charged particles. Different charges and types of beam particles,e.g. π ± , K ± , and (anti)protons, can be selected by tuning the COMPASS beam line and by taggingthem with differential Cherenkov counters. The possibility to switch between pion and muon beamsof the same momentum is a unique asset for the measurement of pion polarisabilities at COMPASS,where the systematic error of the measurement can be significantly reduced through regular referencemeasurements with incident muons, i.e. point-like particles. As target material either liquid hydrogen orvarious solid-state nuclear targets are used. A recoil proton detector (RPD) is installed around the targetto ensure the exclusivity of the final state. A set of double-sided silicon microstrip detectors positionedupstream and downstream of the target is used to reconstruct the interaction vertex and the angles of theoutgoing particles. Here, the required angular resolution is dictated by Primakoff reactions, where pionsor muons scattered by angles of a few hundred µ rad have to be detected. A high momentum resolution forcharged particles is provided by a two-stage magnetic spectrometer. For the tracking in the beam regionnew pixelised Gas Electron Multiplier (GEM) detectors with a minimised material budget along the beamwere built, in replacement of the thicker scintillating fibre detectors. For the tracking at small angles,the existing Micromegas trackers were adapted to the hadron beam conditions. A major upgrade of theRing-Imaging Cherenkov (RICH) counter was carried out, which largely improved the performance ofparticle identification at high rates. Photons are detected in two electromagnetic calorimeters, whichhave been optimised for stability and uniformity in order to achieve good resolution. Several new triggerelements were built and implemented into the trigger system.The present paper describes the modifications and upgrades of the experimental setup required for thehadron programme of COMPASS. Some of these upgrades were already realised for the nucleon spinprogramme after 2005. After a brief overview of the layout of the spectrometer in Section 2, the beamline and associated detectors are described in Section 3 and the target region in Section 4. The newlyinstalled tracking detectors are discussed in Section 5. Section 6 deals with the systems used for particleidentification, namely the RICH counter and the two electromagnetic calorimeters. The various triggersystems are explained in Section 7 and the data acquisition in Section 8. The algorithms for eventreconstruction and the performance of individual detector components are summarised in Section 9,while the global spectrometer performance and Monte Carlo simulations of the apparatus are discussedin Section 10.Throughout this paper, the following kinematic variables will be used: the squared four-momentumtransfer from the incident beam particle to the recoiling target nucleus t = ( p beam − p X ) ; the reducedsquared 4-momentum transfer t (cid:48) = | t | − | t | min to the recoiling target nucleon beyond the kinematic min-imum | t | min ; the Gottfried-Jackson angle θ GJ , defined as the polar angle of the three-momentum of theisobar (i.e. di-pion) from the decay of X , and the corresponding azimuthal Treiman-Young angle φ TY .These two angles are calculated in the centre-of-momentum frame of X with the z -axis along the beamdirection and the y -axis perpendicular to the production plane, formed by the momentum vectors oftarget and recoil particles. The main features of the COMPASS experimental setup and most of the detectors as used until 2004 aredescribed in Ref. [1]. In this section a short overview of the apparatus is given, with particular emphasison detectors that are either specific to the data taking with hadron beams in 2008 and 2009, or were addedto the setup after 2005 to be used in both muon and hadron programmes.The COMPASS setup can be divided into four parts along the beam, starting with the beam line sectionhe COMPASS Setup for Physics with Hadron Beams 5
Fig. 2:
Three-dimensional view of the COMPASS setup for measurements with hadron beams. The beam comesfrom the left side. The upstream part of the setup (beam line) is not shown here. The different colours indicatedifferent detector types. and the detectors that identify the incoming beam particles. It is followed by the target region, whichis specific for each of the COMPASS physics programmes. It comprises the target and the detectorslocated in its near vicinity. The third part, called Large Angle Spectrometer (LAS) includes the firstdipole magnet, SM1, the tracking detectors around it, and the RICH-1 counter. The fourth part, calledSmall Angle Spectrometer (SAS), occupies the downstream part of the setup. It is built around theSM2 dipole magnet and includes several tracking detectors. Both LAS and SAS comprise a pair ofelectromagnetic and hadron calorimeters, and a muon filter. Figures 2 and 3 show the three-dimensionaland top views of the COMPASS setup, respectively.
The COMPASS setup is located at the end of the M2 beam line of the CERN SPS accelerator. The M2beam line can be tuned for beams of different particles, including secondary hadron beams and tertiarymuon or electron beams. Hadron and muon beams can be either of negative or positive charge. Switchingbetween beams takes typically thirty minutes.During data taking with hadron beams only the trajectory of the incident beam particle is measured. TheBeam Momentum Station (BMS), which is used for the determination of the incident momentum duringmeasurements with a muon beam, is moved out of the beam in order to minimise the material budgetalong the beam path. However, the muon beam is also used during Primakoff measurements in order tocomplement the data taken with pions. The BMS is then moved back into the beam line. Downstreamof the BMS location, two differential Cherenkov counters identify the hadrons (pions, kaons, or protons)that are present in the hadron beam. The COMPASS Collaboration
StrawsSilicons
RICH
ECAL1 HO4RICHWALL SM2SM1 GEMStrawsBC MuonFilter 1Muon GEMDCSciFi top view
DCMicromegasRPD MWPCGEMGEMMW1MWPC
Large area DCs
Filter 2ECAL2HCAL2TriggerVeto SandwichMC SciFiPixelGEMPixelGEM PixelGEMSciFi GEMMW2MWPCHCAL1MWPCGEM BK1 MWPCBK2
50 m10 20 30 40 x z BeamCOMPASS setup 2009
Fig. 3:
Top view of the COMPASS setup for data taking with hadron beams. The labels indicate the variousdetectors, as referenced throughout this paper. The vertical scale is only indicative of the relative detector sizes.The colour code follows that of Fig. 2.
Most of the data with hadron beams were collected using a liquid hydrogen target. The target regioncomprises the target itself and the detectors around it (Fig. 4). The target is surrounded by a time-of-flight detector that is called Recoil Proton Detector (RPD). Measuring the recoil protons from thetarget, this detector ensures the exclusivity of the processes under investigation. The RPD covers themomentum transfer range down to | t | = .
07 GeV /c . Three silicon stations operating at a temperatureof 200 K are mounted upstream of the target. Together with a scintillating fibre counter, these detectorsdetermine the trajectory of the beam particle before it enters the target. Two other silicon stations arelocated immediately downstream of the target, inside the RPD. A scintillator/iron sampling detector,called Sandwich Veto, is installed downstream of the RPD. Used as part of the trigger, this detectorvetoes particles detected outside of the LAS acceptance. A dedicated Multiplicity Counter (MC) ispositioned downstream of the RPD, behind the Sandwich Veto. This counter, which measures the numberof charged particles in the final state, extends the momentum transfer range towards values smaller than | t | = .
07 GeV /c .The liquid hydrogen target can be easily removed and replaced with a specially designed solid-targetholder. Up to 16 solid targets with different atomic numbers and different thicknesses can be mountedon the holder and used simultaneously during data taking. The large angle spectrometer includes the detectors located both upstream and downstream of the SM1magnet. The LAS tracking detectors measure scattered particles with polar angles of up to 180 mrad.In the region near the beam, a PixelGEM detector with low material budget was installed in replace-ment of the thicker scintillating fibre counter (SciFi) previously used with the muon beam. The designof the Micromegas detectors that are located upstream of SM1 was modified in order to minimise theirdischarge rate in the hadron beam. A new large-size drift chamber, DC4, is installed downstream ofthe SM1 magnet, in order to improve the resolution of the tracking at large angles. A major upgrade ofhe COMPASS Setup for Physics with Hadron Beams 7 !"
Fig. 4:
Side view of the target region with the liquid hydrogen target system. the RICH-1 counter was accomplished [4], which considerably improves its performance. The centralregion of RICH-1 was instrumented with multi-anode photomultipliers, in replacement of the previouslyused CsI photodetectors. A new analog readout with a reduced dead time was implemented in its pe-ripheral region. The tracking downstream of RICH-1 was supplemented with an additional drift-tubedetector, called Rich Wall (RW). A new electromagnetic calorimeter, ECAL1, was added to the LASsetup. ECAL1 extends the coverage of ECAL2 for detection of photons and electrons to larger angles.Its position was chosen with the aim of achieving a continuous angular coverage for both ECAL1 andECAL2. Installed since 2006, DC4, ECAL1, RICH-1 and RW are part of the apparatus that is commonto both hadron and muon physics programmes.
The SAS detectors are essentially identical to the detectors used during the data taking with muonbeam [1]. In order to minimize the material budget along the beam path, two new PixelGEM detectorsreplace two SciFi counters. In ECAL2, the inner-most lead glass blocks were replaced with radiation-hard Shashlik-type lead/scintillator modules of the same transverse size. In order to maximise the photondetection acceptance near the beam, the size of the ECAL2 central hole was reduced. The central holeof the hadron calorimeter HCAL2, located immediately behind ECAL2, was reduced accordingly.
The trigger system for hadron beam was designed to select the processes listed in Section 1. Severalnew trigger counters were built and combined with those already available [1]. A beam counter (BC)was installed upstream of the target, as a part of the beam-definition trigger. Both RPD and ECAL2detectors were included in the trigger. The information from the RPD is used to identify diffractivescattering events. High energy photons, particularly important for the Primakoff reaction, are selected by The COMPASS Collaborationthe central part of ECAL2. Triggering in the region of the lowest momentum transfer values is providedby a new Multiplicity Counter (MC). The existing veto system was extended to veto non-interactingbeam particles by adding two beam-killer (BK1 and BK2) scintillator counters along the beam path.With the nominal hadron beam intensity the trigger rate reaches values of up to 30 kHz. The correspond-ing data flow is as high as 350 MB/s. The COMPASS data acquisition system was upgraded to meetthese conditions.
The CERN SPS M2 beam was originally built as a high-energy, high-intensity muon beam. For theCOMPASS experiment, the beam line was partly rebuilt to include a high-intensity hadron beam optionas well as the possibility to use low-intensity electron beams. For beam particle identification, twoCEDARs were added just before the COMPASS spectrometer. Modifications relevant to the muon modeof operation were described in [1]. In this section, the hadron and electron beam modes of operation aresummarised. In addition, the detectors used for the identification of the particles in the hadron beam aredescribed.
In order to produce a secondary hadron or tertiary muon beam, 400 GeV /c protons from the CERN SPSare slowly extracted onto a primary production target (T6). These protons arrive during a time period of9 . · s − , whiche.g. for a negative beam of 190 GeV /c central momentum is achieved by using 9 · protons/cycle onT6 and the 500 mm target.The M2 beam line (Fig. 5) starts with a series of six high-gradient quadrupoles (Q1–Q6) for maximumacceptance and a set of three dipoles (BEND 1). The highest-acceptance optical mode works only upto 225 GeV/ c due to limited quadrupole gradients. The beam optics is optimised to achieve highestpossible momentum resolution.A pair of massive collimators, which allows a first momentum selection of the beam particles or is alsoused to dump the beam in case of access to the experiment (TAX 1,2), is located downstream of the firstdipoles (BEND 1). The particles passing through theses collimators are transported to an about 430 mlong array of alternately focusing and defocusing (FODO) quadrupoles (Q12–Q18). Before entering theFODO array, two pairs of horizontal (H1, H3) and vertical (V2, V4) collimators define the angular andmomentum acceptances of the beam. At the end of the FODO section, the beam is focused into a set offour vertical dipoles (BEND 4,5). The Be absorbers (ABS), which are used in the muon beam operation,are moved out in the hadron mode. After these dipoles, the beam is transported into a second 250 mlong FODO channel (Q22–Q30) that contains a second momentum-defining collimator. The beam isthen bent horizontally (BEND 7) and parallelised at the location of the two CEDARs (for more details,see Section 3.3), while restricting the momentum spread to below 1%. Behind the CEDARs, the beamis focused (Q34–Q36) onto the entrance of the electromagnetic calorimeter in the second spectrometerstage, which is located 33 m downstream of the target. The main beam characteristics are listed in Table 1.Negative beams are mainly composed of pions, while for momenta larger than 150 GeV /c the positivebeam have a dominant proton component. In both cases, kaons and electrons may be present at a levelof a few percent, depending on the energy chosen. The particle composition of the hadron component of A different optical mode is available for higher momenta, but the angular acceptance of that mode is about 40% lower. he COMPASS Setup for Physics with Hadron Beams 9
Fig. 5:
The CERN M2 beam line.
Table 1:
The main parameters of the M2 hadron beam.
Parameter Value
Length of beam line from primary target to COMPASS target 1131 . /c Maximum beam momentum (normal mode) 225 GeV /c Angular acceptance: Horizontal ± . ± . . µ srMomentum acceptance ± σ x × σ y ) 7 × Divergence at COMPASS target ( σ x × σ y ) 80 µ rad × µ rad Table 2:
The relative composition of the hadron beam at the COMPASS target for some typical momenta. Itdoes not include the e ± component, which is still present at 100 GeV /c but rapidly decreasing at higher momentadue to synchrotron radiation. The composition values are calculated from measured values [5] and their relativeuncertainties amount to 1% for pions and proton, and 2–3% for kaons and antiprotons.Momentum Positive beams Negative beams(GeV /c ) π + K + p π − K − ¯ p
100 0.618 0.015 0.367 0.958 0.018 0.024160 0.360 0.017 0.623 0.966 0.023 0.011190 0.240 0.014 0.746 0.968 0.024 0.008200 0.205 0.012 0.783 0.969 0.024 0.007 the beam is given in Table 2 for a few typical beam momenta.
A tertiary electron beam can be produced on demand. For this purpose, a 5 mm thick lead plate (alsocalled “electron target") equivalent to 90% of a radiation length is moved into the beam line at the end ofthe first FODO section, about 680 m downstream of the primary production target (see Fig. 5). A high-intensity negative hadron beam of 100 GeV /c , which contains electrons at the level of 10%, is directed tothe electron target. The hadrons mostly traverse the lead target, as its thickness is equivalent to only about3% of an interaction length. In contrast, most of the electrons of the beam lose energy by bremsstrahlung.The outgoing electrons have a momentum spectrum that extends up to the momentum of the parent beambut with very low intensity yielding useful electron energies of up to 50 GeV. The required electronmomentum is selected with the beam line magnets located downstream of the lead target. For the nominalenergies of 15, 20 and 40 GeV, which are used for the calibration of the electromagnetic calorimeters,intensities up to a few 10 electrons per spill are routinely reached. Two CEDAR detectors are installed 30 m before the COMPASS target region. They were designed toprovide fast beam particle identification at high rates for particle momenta up to 300 GeV /c [6]. The principle of operation of a CEDAR detector is illustrated in Fig. 6. For a beam containing particlesof different types but the same momentum, the angles of the emitted Cherenkov photons differ due to thedifferent masses. The Cherenkov photons are focused onto the photon detectors using a a mirror and asystem of lenses (lens, corrector, condenser). This results in rings of photons at the focal plane wherebycompensating for the chromatic aberration in the gas, which is mandatory for a proper separation of thehe COMPASS Setup for Physics with Hadron Beams 11
PMTPMT vapour-deposit mirrorcorrectordiaphragmcondenserquartz window lenselight pathhelium vessel
Fig. 6:
The basic principle of a CEDAR counter. Two particles with the same momentum but with different masses(here red and green lines) radiate Cherenkov photons at different angles, resulting in rings with different radii. Adiaphragm selects the rings from the required particle type. passive voltage dividercorrectorcondenser/diaphragmquartz windowsthermal insulation alignment tablepressure vessellense/vapour-deposit mirrorphotomultipliers
Fig. 7:
A cut through one of the CEDAR detectors. rings. A ring shaped diaphragm, which is located in the focal plane perpendicular to the beam direction,selects photon rings with a fixed radius. The radius of the photon ring is matched to the radius of thediaphragm by adjusting the pressure of the helium gas in the vessel.COMPASS operates two CEDAR detectors. Each consists of a 6 m long vessel containing pressurisedHe gas, a mirror, a lens system and a diaphragm (Fig. 7). The nominal pressure at 190 GeV/ c beammomentum is 10.5 bar. The photons are detected with eight PMTs (Thorn-EMI-9820) equipped withpassive voltage dividers.The photon rings are smeared by several effects, e.g. temperature changes, beam divergence and limitedprecision of alignment. In order to keep the density constant along the 6 m long vessels and thus therefractive index, good thermal insulation and conduction is mandatory. The vessel is covered with copperfilaments for thermal conduction and surrounded by a 10 cm thick polyethylene foam layer for insulation.In addition, the PMT voltage dividers are mounted outside the vessel. Particles travelling not parallel tothe optical axis will produce shifted photon rings that do not match the diaphragm. A tilt of the beamwith respect to the principal axis of the optical system can be corrected by adjusting the detector positionwith the help of a motorised base. The beam divergence could only be compensated by opening thediaphragm at the expense of a lower purity of the particle identification.2 The COMPASS Collaboration p/T (mbar/K)34 34.5 35 35.5 36 36.5 37 N o r m a li s e d r a t e ‡ ‡ + /K + p Fig. 8:
Pressure scan with CEDAR1 for a positive hadron beam with at least 4, 6 or 8 PMTs in coincidence. Thekaon peak cannot be distinguished from the pion peak.
As the parallelism of beam tracks is of great importance for an efficient operation of the CEDARs, thebeam divergence is monitored using pairs of single scintillating fibre detectors (one horizontal, one ver-tical) that were installed upstream (FISC1,2) and downstream (FISC3,4) of the CEDARs. Their positionin the beam can be adjusted to measure the track angles by a coincidence between an upstream and adownstream fibre hit. Furthermore, two scintillating discs (TRIG 2) are installed as beam counters. Theyare used to normalise the CEDAR count rates during so-called pressure scans. While taking physics data,the discs and single-fibre detectors are moved out of the beam in order to reduce the material budget inthe beam line.As the ratio of pressure over temperature, p/T , is proportional to the refractive index, the working pointof the CEDAR detectors is determined by performing pressure scans. In a pressure scan, the count ratenormalised to the rate in the FISC counters is determined as a function of the pressure in the vessel andthe multiplicity of PMT signals. Using the known beam composition, this yields also an online estimatefor the particle identification efficiency. A more refined offline method will be discussed in Section 9.6.During data taking the He pressure in the CEDARs is regularly adjusted to compensate for He leakageand to keep p/T constant.
In the high-energy positive hadron beam, the proton component is dominant. For the CEDARs, a differ-ence of 1 mm is expected between the ring radii of protons and kaons at 190 GeV /c . The plateau of theefficiency is reached with a slit width of 1 . ≥ p/T (mbar/K)33.8 34 34.2 34.4 34.6 N o r m a li s e d r a t e -4 -3 -2 -1
10 1 6 ‡ - p - K Fig. 9:
Pressure scan with CEDAR1 for a negative hadron beam with at least 6, or with 8 PMTs in coincidence. identify protons, the other to identify pions.
Negative hadron beams contain mainly pions with a small admixture of kaons and antiprotons. In thiscase, the CEDARs are used to identify kaons. Although the difference between the mean radii of thephotons rings of kaons and pions at 190 GeV /c is less than 0 . . ≥ ≥
6. Such a low efficiency is due to the high beam divergence of the very long beam line, in combinationwith the narrow slit width of the diaphragm. The loss due to the beam divergence is illustrated in Fig. 10which shows the distributions of beam track angles as measured by the Silicon detectors upstream of thetarget and propagated back to the CEDAR position. The distribution for all beam particles are comparedto those for beam particles identified online by CEDAR1 or CEDAR2.In order to reduce and to separate the background that mainly originates from pions, both CEDARsare set to detect kaons. The background can be measured by setting one CEDAR to detect kaons andperforming a pressure scan with the other CEDAR. As illustrated in Fig. 11, the pion background isbelow 7%.
The target region comprises the target systems, the Recoil proton detector, the Sandwich veto detectorand the Silicon detectors (Fig. 4). Either liquid hydrogen or solid targets can be used during the measure-ments performed with hadron beams. The hydrogen and lead targets are used for diffractive dissociationand central production measurements. A nickel target is used for the study of the Primakoff reaction. TheSilicon detectors, which are also located in the target region, will be described along with other trackingdetectors in Section 5.4 The COMPASS Collaboration
Horizontal track angle (mrad)-0.2 -0.1 0 0.1 0.2 E n t r i e s · all eventsCEDAR 1CEDAR 2 Vertical track angle (mrad)-0.2 -0.1 0 0.1 0.2 E n t r i e s · all eventsCEDAR 1CEDAR 2 Fig. 10:
Horizontal (left) and vertical (right) track angles at the CEDARs. The angles for all tracks measured bythe Silicon beam telescope and propagated back to the CEDAR positions are compared to the angles of the tracksaccepted by CEDAR1 or CEDAR2. The acceptance of the CEDARs is reduced significantly for very divergentbeam tracks. p/T (mbar/K)34 34.1 34.2 34.3 34.4 34.5 34.6 34.7 N o r m a li s e d r a t e -3 · Fig. 11:
Count rate of coincident events recorded with CEDAR1 and CEDAR2. The pressure of CEDAR2 wasscanned while CEDAR1 was set to detect kaons.
For scattering on protons, a liquid hydrogen target is used. The target cell has a cylindrical shape witha length of 400 mm along the beam and a diameter of 35 mm, which corresponds to a volume of 0 . .
5% of a radiation length ( X )and 5 .
5% of a nuclear interaction length ( λ I ). The diameter of the target is matched to the dimensions ofthe beam spot ( σ ≈ µ m thickness (5 · − X , · − λ I ). Theliquid hydrogen inlet and gas outlets are constructed from stainless steel pipes, which are connected to astainless steel ring surrounding the target Mylar cell. The hydrogen cell and the stainless steel pipes arewrapped with 10 layers of heat superinsulation foils (with thickness of ≤ µ m/foil).The target system with target cell, cryostat, and refrigerator is shown in Fig. 12. The target cell issurrounded by a cryostat tube made from aluminium. The cryostat has a diameter of 185 mm and isterminated towards the spectrometer by a 250 µ m thin Mylar window. Its diameter was chosen largehe COMPASS Setup for Physics with Hadron Beams 15 !" : ; < ' (( * ) & ( * ' =*3$'3>0+';5/3"'?01@ Fig. 12:
Side view of the liquid hydrogen target system. A closer view of the cylindrical Mylar cell and hydrogenpiping is shown in the inset. enough so that forward going particles detected by the COMPASS spectrometer pass through the windowand not through the aluminium cryostat itself. In order to reduce the amount of material traversed by therecoil particles, the thickness of the aluminium tube surrounding the target cell is 1 . µ m Mylar window of the cryostat has a diameter of 80 mm.During operation, the target cell is filled with about 0 . . . − Table 3:
Overview of target materials used during the measurements with hadron beams in 2008/2009.
Material Number Thickness x (elements) ( mm ) ( g / cm ) ( λ I ) ( X ) Liquid H .
5% 4 . .
250 0.284 0 .
14% 4 . .
125 0.142 0 .
07% 2 . .
025 0.028 0 .
01% 0 . .
050 0.057 0 .
03% 0 . .
050 0.097 0 .
05% 1 . .
025 0.048 0 .
03% 0 . . .
8% 29 . .
025 0.048 0 .
03% 0 . .
050 0.097 0 .
05% 1 . . For measurements with nuclear targets, a light-weight target holder made of carbon fibre rods and thinframes of fibreglass reinforced epoxy (FR4) was used. Housing up to 16 target disks, the target holderis inserted into the RPD instead of the liquid hydrogen target. Figure 13 shows a schematic view of thetarget holder and the frames onto which the foils were glued.The specifications of all targets used are listed in Table 3. Two different sets of nuclear targets weremounted on the target holder. The first set consisted of 16 thin disks made of Pb and W of naturalisotopic composition. The thicknesses of the disks and the distance between them was chosen such thatrecoil protons from each individual disk with momenta above 200 MeV /c could be detected over the fullacceptance of the RPD. The more downstream disks were made thinner in order to minimise the effectof multiple scattering and conversion for events originating from the more upstream targets. The 12Pb targets were used for diffractive dissociation measurements, while the four W targets were used forfeasibility studies of a measurement of the π lifetime.The second set of target disks consisted of one 4 . The Recoil Proton Detector (RPD) measures the velocity and energy loss of the recoiling particles emit-ted at large angles. For particles produced in the middle of the target, the region of full acceptance coverspolar angles from 50 ◦ to 90 ◦ (see Fig. 4). The energy loss of protons in the target walls and in the innerring limits the lowest detectable momentum to 270 MeV /c in the case of the liquid hydrogen target. TheRPD is also used in the trigger system (see Section 7) for the identification of protons.he COMPASS Setup for Physics with Hadron Beams 17 Fig. 13:
Schematic view of the target holder used for measurements with nuclear targets.
The design of the RPD closely follows the design of the detector used for the GAMS NA12/2 experimentat CERN [7]. It is made of two concentric cylindrical barrels of plastic scintillators that surround thetarget and are referred to as “rings" in the following. The inner ring is segmented in 12 slabs of BC404 R (cid:13) scintillator of dimensions 50 × . × . , which are positioned at a radius of 12 cm. Light guides forthe inner ring are made of Plexiglas and have a fish-tail geometry. They are tilted at an angle of 15 ◦ withrespect to the longitudinal axis in order to stay outside of the acceptance in the forward region.The outer ring is segmented in 24 slabs of plastic scintillators with dimensions of 115 × × ,produced at IHEP Protvino [8]. Each slab is made of a single piece of material that also includes a 29 cmlong light guide on each side. The ends are cut, twisted and molded to fit into a 3 . ◦ . In order to optimise the azimuthal angle resolution, the outer ring is positioned such thateach inner ring counter faces three outer ring slabs as viewed from the target centre (see Fig. 54).Each scintillator is read-out at both sides using EMI 9813B photomultiplier tubes. The PMTs areequipped with active voltage dividers to cope with the high rate and high light output. The PMT signalsare split using 8-fold active splitters [9] and sent to ADCs (2 dynamic ranges) and TDCs (2 thresholdlevels). The remaining outputs are used for the trigger system. Two outputs are connected to leadingedge discriminators with two different thresholds. Furthermore, the signals from the inner ring down-stream PMTs have the smallest time jitter with respect to the incoming track since light in the scintillatorpropagates in the same direction as the scattered particle. The signals from these are sent to ConstantFraction Discriminators to preserve their good timing properties. All logic signals are then fed into aFPGA-based system for triggering (see Section 7).The properties of each individual counter were measured during earlier tests using muons from the beamhalo with the RPD positioned transversely to the beam. The resolutions obtained are σ ( t ) =
200 psand σ ( z ) = . σ ( t ) =
400 ps and σ ( z ) = . Proton momentum (GeV/c)0.3 0.4 0.5 0.6 0.7 0.8 0.9 M o m e n t u m r e s o l u ti on ( % ) Fig. 14:
Momentum resolution of the RPD for protons detected at an angle of 70 ◦ relative to the beam axis. elastic pp scattering. For velocities of up to β = .
34 the protons are stopped in the outer ring. Abovethis value the protons escape the scintillator and deposit in it only part of their energy. The figure forpions would be similar and the energy loss for stopping pions would reach a maximum value of 10 MeVfor β = .
4. Therefore, proton particle identification is ensured only for β < .
4. In Fig. 15 there is noindication for presence of pions, as expected in pp elastic scattering.
The role of the Sandwich veto detector [10] lies in vetoing events in which photons or charged particlesreach the acceptance gap between RPD and LAS (see Fig. 4). This detector is a 2 m × E n t r i e s b E ( M e V ) D R i ng B pp fi pp Fig. 15:
Energy loss ∆ E in the outer ring of the RPD as a function of the velocity of the particle in elastic ppscattering. he COMPASS Setup for Physics with Hadron Beams 19 Fig. 16:
Sketch of the Sandwich veto detector. The active area of the detector (depicted in grey) has dimensions of200 ×
200 cm . five layers of steel-covered lead plates and scintillators with a total thickness of 5.1 radiation lengths.Segmented in 12 elements (Fig. 16), the detector has a central hole that matches the acceptance of thespectrometer. Each lead layer consists of 5 mm Pb plates, with 1 mm steel plates on each side to insurethe stiffness of the assembly. Each scintillator layer is formed of a pair of 80 ×
20 cm scintillator barslying side-by-side. The first three layers are 1 cm thick, the last two 0 . The tracking system of COMPASS is composed of many tracking stations, each consisting of a set ofplanar tracking detectors of a given type located at approximately the same z -coordinate along the beam.Many different detector technologies with different sizes, granularities and resolutions are in use. Farfrom the beam in the outer region, large areas of several square meters have to be covered in order todetect low-momentum particles scattered at large angles. Close to the beam in the inner region, theparticle rates quickly increase with decreasing distance to the beam, requiring fast detectors with goodresolution. The large-area tracking is provided by several variants of wire-based gas detectors such asMultiwire Proportional Chambers (MWPC), Drift Chambers (DC), and Straw Tube Chambers. The re-gion closer to the beam, where the particle rates are too high for wire-based detectors, is covered bytwo types of Micropattern Gaseous Detectors with strip readout, namely the Micromegas and Gas Elec-tron Multiplier (GEM) detectors. The beam region itself, where rates above 10 mm − s − are observed,is equipped with Scintillating Fibre Detectors and novel GEM detectors with pixel readout, the Pixel-GEMs. Tracking immediately upstream and downstream of the target is performed by silicon microstripdetectors.This section focuses on the upgrades of the tracking system for the hadron program as compared tothe setup used for muon beams, detailed in [1]. For some detectors, like the straw tube chambers,0 The COMPASS Collaboration Fig. 17:
The conical cryostat with the upstream beam window dismounted. The height of the (green) PCB framethat holds the detector (sensor) is about 100 mm, the length of the full cryostat about 400 mm. The bent coolingcapillary is fixed to the PCB close to the sensitive area of the detector. Inside the cryostat, the readout cables aredirectly soldered to the detector module and plugged to vacuum-sealed feedthrough connectors also visible on theouter surface of the cryostat. the multiwire proportional chambers, and the large area drift chambers, no mentionable changes wereintroduced, and therefore they are not described here. These detectors are however discussed in detail in[1]. The wire and strip detectors measure different projections of a particle penetration point. They areare called X and Y -planes when measuring horizontal and vertical coordinates, respectively. Detectorplanes measuring coordinates that are rotated clock or and anti-clockwise by a given angle with respectto the x -axis, are called U and V -planes, respectively. The COMPASS silicon microstrip tracking system consists of three stations upstream of the target, whichare used as a beam telescope, and two stations downstream of the target, which are used for vertex re-construction. As these detectors are traversed by the beam particles and by the forward-boosted reactionproducts, they are prone to radiation damage. The damage affects the bulk material in terms of change ofdoping, and the surface in terms of decrease of insulation, resulting in an increase of the depletion voltageand of the leakage current, respectively. In order to minimise these effects, the detectors are cooled withliquid nitrogen. Since the leakage current decreases with temperature, noise caused by radiation damageis suppressed. In addition to this, the cooling leads to a significant improvement of the spatial and timeresolution compared to room-temperature operation, as discussed below. While the system was designedto cool the detectors down to 130 K, the desired performance is already achieved at 200 K, which reducesthe thermal stress on the modules.One station comprises two Silicon detectors with a stereo angle of 5 ◦ between their respective striporientations to resolve multi-track hit ambiguities. Each detectors consist of a 300 µ m thick siliconsensor with an active area of 50 ×
70 mm . The signals are picked up on both sides, by 1280 strips on the n -side and 1024 perpendicularly-oriented strips on the p -side. The sensors are glued onto two L-shapedhe COMPASS Setup for Physics with Hadron Beams 21 Valve Box XYUVSI01
PhaseseparatorSI02SI03CCCentralnitrogensupplyExhaust E x h a u s t HeaterLevelmeterForepumpTM pumpFlowmeterValveFlow-regulator
Fig. 18:
Block diagram of the valve box and the first upstream cryostat labelled SI01. The other two upstreamcryostats SI02 and SI03 are equipped analogously. The downstream conical cryostat (CC) is shown in Fig. 19. Thephase separators are integrated in the cryostats near the detectors, but outside the acceptance.
FR4 printed circuit boards (L-boards) that hold the APV25-S1 [11] based readout electronics. There arethree cryostats for the beam stations upstream of the target and one conically shaped cryostat housingthe two stations downstream of the target (see Fig. 4).The cooling system of the Silicon detectors has to fulfil the requirement of a minimal amount of materialwithin the acceptance of the spectrometer. This prevents a solution, in which the detectors are connectedto a massive cold head to dissipate the electronic heat. The technology developed for these detectors isbased on the evaporation of liquid nitrogen in thin capillaries on the PCBs. The schematic layout of theSilicon cooling system is shown in Figs. 18 and 19. In order to dissipate about 8 W from each detector,purely liquid nitrogen must be provided to the capillary. For this purpose, a dedicated phase separatorthat removes the gaseous nitrogen is incorporated in each cryostat. The whole cooling infrastructureincreases the material thickness of the PCB on average by 0 .
1% of a radiation length.The nitrogen arrives from a central liquid nitrogen dewar located in the vicinity of the experimental hall.It is transferred by a 100 m long vacuum-isolated transfer line to a valve box near the Silicon stations.The valve box (Fig. 18) also acts as a buffer for the liquid nitrogen that is kept at 1 . Phaseseparator
XYUVSI04 XYUVSI05
Valve Box E x h a u s t CC Fig. 19:
Block diagram of the conical cryostat (CC), symbols as in Fig. 18. The phase separator is mounted in anextra housing outside the spectrometer acceptance with a vacuum connection to the cryostat. each detector. The thermal contact to the L-boards is made by soldering dots. The temperature of thedetector is regulated through the gaseous exhaust flow with a feedback time in the order of one second.All components are operated by a Programmable Logic Controller (PLC, SIMATIC S7 300), utilising aProportional-Integral-Derivative algorithm for the temperature regulation. The software used is a Java TM -based application called Muscade[12] which provides real-time monitoring, remote control, data storage,and an alarm system.In 2009, all Silicon stations were cooled to 200 K. The temperature of the system was stabilised to within ±
1K for all detectors of the upstream stations. Slightly larger variations were observed for the detectorsin the conical cryostat, where a partly blocked capillary prevented good cooling for one of the detectors,limiting the temperature to ∼
220 K only. This detector also exhibited slow drifts following the dailytemperature variations.The spatial resolution of the cold Silicon detectors is in the range 4 − µ m for clusters when two stripsare hit and amplitude weighting can be employed to determine the track position [13]. When only onestrip is hit, the resolution is in the range 7 − µ m. This spatial resolution is illustrated in Fig. 20 for oneof the detectors. It represents an improvement of 15-20% compared to room-temperature operation [1].The reduction of the leakage current and the increase of the signal each contribute of about 10% to thisimprovement. The time resolution, displayed in Fig. 21, is improved for the same reason and is in therange 1 . − . m) m Space residual (-50 -40 -30 -20 -10 0 10 20 30 40 50 E n t r i e s · Cluster size 1Cluster size 2 m m RMS1 = 7.1 m m RMS2 = 4.0
Fig. 20:
Spatial resolution as determined for a singleSilicon detector plane. “RMS1” and “RMS2” refer tothe cases of clusters with one and two hit strips, respec-tively.
Time residual [ns]-30 -20 -10 0 10 20 30 E n t r i e s · Cluster size 1Cluster size 2 = 1.6 ns s = 1.4 ns s Fig. 21:
Time resolution of a single Silicon detector pro-jection. x (cm)-4 -3 -2 -1 0 1 2 3 4 y ( c m ) -3-2-10123 E ff i c i e n c y Fig. 22:
Two-dimensional efficiency distribution for a plane in the beam telescope. The stereo-angle tilt of thesensitive area is visible. expected hit position on the detector. The presence of a hit is then checked within a ± σ window aroundthe expected position. The measured efficiency is above 99% as shown in Fig. 22 for one of the planes.Similar results were also obtained for operation with hadron beams. In order to minimise the material from detectors directly exposed to the hadron beam, some of thescintillating fibre detectors that were used with the muon beam were replaced by thinner detectors basedon Gas Electron Multiplier (GEM) foils [14]. Starting with the first hadron run in 2008, five GEMdetectors with a novel kind of readout and a thickness in the beam region of 0 .
26 % of a radiation length( X ) and 0 . λ I ) were installed, thereby reducing the materialbudget of the whole system by a factor of 5–10 compared to the scintillating fibre detectors.GEM detectors with a two-dimensional strip readout have been used in COMPASS since its start-up[15]. These gaseous detectors have proved to be able to cope with the high particle fluxes in the beamcentre, but the strip readout makes it impossible to separate individual hits close to the beam due to atoo high occupancy. In order to overcome this limitation, a novel read-out structure has been realised ona polyimide basis using the GEM patterning and wet-etching printed-circuit board (PCB) technologies4 The COMPASS Collaboration Fig. 23:
Schematic view of the pixel and strip region ofthe readout circuit. Note that the pixel region consistsof 32 ×
32 pixels of 1 mm size each, while only 4 × Fig. 24:
The PixelGEM read-out foil. The inner 10 ×
10 cm darkest part is the active area. The symmetricwires connecting the pads and the strips to the read-outelectronics surround this part. [16]. The central are of 32 ×
32 mm with 1024 pixels of 1 × size each are patterned on one side of50 µ m thick polyimide foil. The signal traces from the pixels to the readout electronics are routed on theother side of the foil, with an extremely small width of only 50 µ m and a pitch of 100 µ m. The rest of thetotal active area of 100 ×
100 mm , where the occupancy is sufficiently low, is covered by two orthogonalsets of 512 strips with a pitch of 400 µ m, realised on a second 50 µ m thick polyimide foil. The strip foilis then glued onto the one with the pixels, with the central area completely removed from the strip foilin order to open the pixels for charge collection. The strips are split in the middle and read out on bothsides in order to equalise their capacitances, also for the ones not ending at the pixel region. In Fig. 23,the pixel and strip regions are displayed schematically, while Fig. 24 shows a photograph of the completereadout foil. The readout foil is glued onto a light honeycomb sandwich panel of 610 ×
610 mm size,which serves as support plate and also carries the front-end electronics, the high voltage distributioncircuit, and the GEM stack.The GEM stack consists of three GEM foils of 10 ×
10 cm active area, stretched and glued onto largerframes of fibre glass material with 316 ×
316 cm inner dimensions. These frames with a thickness of2 mm are piled up and glued on top of each other. The active part of a GEM foil is sectorised on oneside into four parallel sectors of equal size, and a fifth sector in the centre matching the pixel area of32 ×
32 mm . The foils are mounted such that the segmented sides face the drift cathode. The potentialson the foils are defined through an external resistive divider. They are adjusted such that the largest gain isprovided by the first foil and it is stepwise decreasing for the second and third foil. The segmented sidesof a foil are supplied through individual 10 M Ω loading resistors, while there are no loading resistors forthe non-segmented side. This configuration allows for an operation of the detector even with a potentialpermanent short circuit in one of the sectors, and avoids a high electric field between the last foil and thereadout circuit in case of a discharge. The central sector of the third GEM foil is powered by a separatesupply through a 1 M Ω serial resistor, which allows an independent adjustment of the gain for the centralregion. This takes into account the fact that a smaller effective gain is necessary for the central regionbecause the signal is induced on pads instead of two sets of strips for the peripheral region. Efficiencyscans performed with prototype detectors showed that an effective gain of 8000 is required for the stripregion for fully efficient detection of minimum ionising particles, while a gain of 6000 is sufficient for thepixel region. The triple amplification together with the non-uniform gain distribution and the segmentedhe COMPASS Setup for Physics with Hadron Beams 25 Fig. 25:
A fully assembled PixelGEM detector, equipped with 16 APVfront-end cards. The digitisation of the analog signals from the APVs isdone at an external ADC card, which is connected via the grey cables.
Fig. 26:
Front-end card carrying(from top to bottom) the 130-pin con-nector, the protection network, a ce-ramic pitch adaptor, and the APV25-S1 ASIC for analog sampling of thesignals induced on the readout elec-trodes.
GEM foils, which were already used for the large-size COMPASS GEM detectors [17], ensures operationof the PixelGEM detectors without electrical discharges even in a high-intensity hadron beam.In order to minimise the material in the region near the beam, the gas-filled volume extends to cover atotal area of 316 ×
316 mm . It is enclosed by a frame defining the conversion volume, and a smallerhoneycomb panel of 330 ×
330 mm size, which carries the cathode foil made of Cu-coated polyimide.The material exposed to the beam is minimised by central holes of 30 mm diameter in both honeycombpanels and by reducing the thickness of each of the Cu layers on the drift cathode and the GEM foilsfrom originally 5 µ m to about 1 µ m.Figure 25 shows a top view of an assembled detector, with the high voltage distribution board (lowerright corner) and the 16 front-end electronics cards mounted upside-down. During operation the wholedetector is shielded from external electronic noise by a thin aluminium-coated Mylar foil. As the large-area GEM detectors, the PixelGEM detectors are operated in a gas mixture of Ar/CO (70%/30%).A total of 2048 channels per detector are read out using the APV25-S1 preamplifier/shaper ASIC [11],which samples the input signal at a frequency of 38 .
88 MHz into an analog pipeline with a depth of 160samples. Each chip is mounted onto a separate front-end card connected to the readout circuit usinghigh-density 130-pin connectors, of which two pins are used to connect the ground level of the chipto the detector ground. In contrast to the large-area detectors, where the front-end cards were directlywire-bonded to the readout circuit, the connector solution provides much more reliability and allows fora simple replacement of faulty cards. The front-end cards also contain an external protection network6 The COMPASS Collaboration
Space residual (cm)-0.2 -0.1 0 0.1 0.2 E n t r i e s · m m = 137 s Space residual (cm)-0.2 -0.1 0 0.1 0.2 E n t r i e s · m m = 64 s Fig. 27:
Residual distribution (difference between measured cluster position and track penetration point) in x -direction for (left) the pixel region and (right) the strip region of a PixelGEM detector. The quoted residual widthsare obtained from fits of a sum of two Gaussians. When corrected for the track uncertainties, spatial resolutions of106 µ m (pixels) and 54 µ m (strips) are obtained for this particular detector. consisting of a pair of high-speed switching diodes (BAV99) and an AC coupling using a 220 pF capacitorfor each channel, and a ceramics pitch adaptor. Figure 26 shows a photograph of the front-end card.Three signal amplitudes per channel are multiplexed onto a single differential line for each APV25-S1chip and digitised by a pipelined 12-bit differential ADC at a sampling rate of 40 MHz. The signals fromsixteen APV25-S1 chips are digitised on a custom-made ADC card. This card also includes a VirtexFPGA [18], which performs pedestal subtraction with individual values for each channel, common modenoise correction and zero suppression by applying individual thresholds for each channel.After a successfully operated prototype, which was tested in a muon beam with a flux up to 1 . · µ + / (cid:0) mm s (cid:1) [19, 20], five PixelGEM detectors were installed in the spectrometer in 2008. Onedetector was placed about 2 . ◦ with respect to the first, were installed around SM2, at 19 m and 24 mdownstream of the target, respectively (see Fig. 3).In the offline analysis, a pulse-shape analysis technique is used to extract the signal amplitude and timefor each channel, a feature of great importance in a high-intensity environment. Signals from neigh-bouring pixel channels on the detector are then weighted by their amplitudes and grouped into clusters.Corrections for a non-linear charge sharing between pixels are applied. These corrections have been de-termined in a dedicated test beam experiment, using high-resolution silicon microstrip detectors [19]. Atthis stage also a small (percent-level) cross-talk between channels, arising from the narrow and long PCBtraces between the pixels and the front-end cards, is removed. For the strip signals a simpler clusteringalgorithm based on a centre-of-gravity method is applied.In the following, the performance of the PixelGEM detectors at a hadron beam flux density of 2 . · π − / (cid:0) mm s (cid:1) (total flux of 6 . · π − / s), used for data taking, is shown. Figure 27 shows the resid-ual distribution, i.e. the difference between the measured cluster position and the extrapolated penetrationpoint of a reference track, for the pixel region (left) and the strip region (right). As for the Silicon detec-tors, the detector under investigation has been excluded from the track reconstruction, such that unbiasedresiduals are obtained. After deconvolving the uncertainty on the reconstructed tracks, one obtains, forall five PixelGEM detectors, spatial resolutions distributed around an average value of 125 µ m with astandard deviation of 13 µ m for the pixel regions, and an average value of 65 µ m with a standard de-viation of 12 µ m for the strip regions. From the pulse-shape analysis of the three samples read out perchannel per event, one can extract the time of the signal and thus efficiently remove background hits dueto pile-up. The time resolution is then determined by comparing the time extracted that way with the onemeasured by scintillation detectors for a given track, as shown in Figure 28 for one particular detector.The time resolutions obtained for the five PixelGEM detectors are distributed around an average valuehe COMPASS Setup for Physics with Hadron Beams 27 Time residual (ns)-100 -50 0 50 100 E n t r i e s · = 11.2 ns s Time residual (ns)-100 -50 0 50 100 E n t r i e s · = 9.0 ns s Fig. 28:
Time residual distribution (difference between measured cluster time and track time) for (left) the pixelregion and (right) the strip region ( x -direction) of a PixelGEM detector. of 11 . . . . (cid:15) of a detector in a high-background environment, one has to takeinto account the presence of uncorrelated background hits that may fall within the road width around atrack with a probability b and thus artificially increase the apparent efficiency (cid:15) app = (cid:15) + b ( − (cid:15) ) . Here,the background probability b at a given position on the detector is determined from hits that fall outsidethe road width around a given track used for the efficiency calculation. Figure 29 shows the background-corrected efficiency for a complete detector plane. Here, the pixel region is merged into the strip region,hence the complete active area of 10 ×
10 cm is shown. The lines of lower efficiency parallel to the x and y axes correspond to the boundaries between the HV sectors on the GEM foils. Few, or no tracksare reconstructed in the ring-shaped region when this particular detector is excluded from the trackingto obtain an unbiased efficiency determination. Background-corrected efficiencies for the PixelGEMdetectors were found to be above 97% for all detectors during data taking in 2008. The PixelGEMdetectors are also used for data taking with muon beams of intensities around 10 µ + / s. Twelve Micromegas (MicroMesh Gaseous Structure, or MM) detectors are used for tracking particlesemitted at small angles. Assembled in 3 stations of four detectors each, they are installed in the regionbetween the target and the first dipole magnet SM1. Each MM detector covers an active area of 40 ×
40 cm , except a central dead zone with a diameter of 5 cm, and measures a single projection of a particletrack crossing the detector. To this end, the anode plane is divided in three zones, a central zone with 512strips and pitch of 360 µ m, and two outer zones, each with 256 strips and pitch of 420 µ m. The detectors,of 1024 strips each, have a parallel plate electrode structure, with a volume separated into two regions:a 5 mm conversion gap with a moderate electric field (less than 1 kV / cm), where the ionising particleproduces primary electrons, and an amplification gap of 100 µ m with a much stronger field (typically40 kV / cm), where the primary electrons generate an avalanche. A 5 µ m thin metallic micro-mesh (grid),which captures most of the ions produced during the avalanche, separates the two regions. Another grid,which is used as a drift electrode, defines the conversion gap region.From 2006 onwards, the original MM detectors [1] were modified in order to satisfy two additionalrequirements: operate in a strong magnetic field, and withstand an increased flux of highly-ionisingparticles during data taking with hadron beam. The first requirement comes from the use of a super-conducting magnet with a 2 . E ff i c i e n c y y ( mm ) Fig. 29:
Efficiency of one of the PixelGEM detectors, measured in a high-intensity hadron beam. The horizontallines with reduced efficiency correspond to boundaries between GEM sectors. In the white region not enoughtracks are reconstructed when this particular detector is excluded from the tracking. were replaced with new 5 µ m thin, non ferromagnetic copper grid foils. The new mesh used for theamplification gap has 65 µ m diameter holes and a pitch of 90 µ m. The corresponding values of the driftelectrode are 300 µ m and 600 µ m.The second requirement comes from the use of hadron beams, which produce a large number of highlyionising secondary particles and generate nearly three orders of magnitude more discharges per incidentparticle. Since the discharge rate is proportional to the gain of the detector, the size of the conversion gapwas enlarged from 3 . H /CF with corresponding volume fractions of 85%/10%/5%. In comparison to data taking witha muon beam, the CF component was decreased from 10% to 5%, thereby further reducing the dischargerate at the expense of a slight decrease of the electron drift velocity. At the nominal hadron beam intensityof 5 × particles per second impinging on a 40 cm long liquid hydrogen target, each MM detector seesan integrated flux of up to 30 MHz, reaching 100 kHz per strip near the central dead zone. The meandischarge rate in such conditions varies between 0 .
026 Hz and 0 .
050 Hz, depending on the specific planeand HV settings. During a discharge the micromesh voltage decreases and thereby reduces the efficiencyof the detector. The decrease is recovered several microseconds after the discharge.For all MM detectors, a digital readout of the signal using the SFE16 chip is used. When recording theleading and the trailing edges of a signal, both the mean time and the amplitude of a hit can be calcu-lated, the latter by using a time-over-threshold technique. Adjacent hits are then combined to clusters.The average cluster size is 2 . . µ m and 420 µ m pitch, respectively. Themean value of the time resolution is 14 ns, compared to 8 . E ff i c i e n c y y ( c m ) -20-15-10-505101520 Fig. 30:
Two-dimensional efficiency of a Micromegas detector. The empty region in the middle is the 5 cm centraldead zone. detectors. This loss in performance is due to the combined effect of the increased conversion gap size,the decrease of the drift-gap electric field, and the use of a smaller fraction of CF in the gas mixture.The efficiencies of the MM detectors are determined using the same method as described in Section 5.2for the PixelGEM detectors. A two-dimensional representation of the efficiency of one of the detectorsis shown in Fig. 30. The mean efficiency values obtained for the other 11 MM detectors are all in therange 97.5% - 98.5%. Note that the use of copper grids with a less favourable geometric transparencythan that of the nickel grids has negligible effect on the final efficiency values.Figure 31 shows the space residual distribution of a MM detector for nominal beam intensity and runningconditions. After subtracting the contribution of the track uncertainty, the intrinsic spatial resolutionobtained is 105 µ m. The resolution value is a weighted average of the resolution in the central zone(360 µ m pitch), and that of the two outer zones (420 µ m pitch). The four detectors of the third MMstation operate in the fringe field of the SM1 dipole magnet, which exerts a Lorentz force on the driftingelectrons. The resolution of these detectors varies from 110 to 145 µ m, depending on the orientation ofthe strips relative to the SM1 field lines. A new large-size multiwire drift chamber (DC4) was installed already in 2006 in the LAS part of thesetup. The design of DC4 closely follows that of the medium-size DC trackers [1] already operationalin the COMPASS set-up, while the overall dimensions were enlarged to match the angular acceptancedownstream of the SM1 magnet. The distance between active wires was increased by 1 mm and theangle of the inclined wires was decreased. The DC4 chamber also features a water-cooling system,0 The COMPASS Collaborationwhich ensures a good temperature stability of its frame.The external dimensions of the DC4 detector are 294 × ×
17 cm with an active gas area of 248 ×
208 cm . The detector has eight layers of wires and four wire orientations: two vertical layers ( X - and X (cid:48) -plane), two horizontal layers ( Y - and Y (cid:48) -plane), two layers with wires inclined with respect to thevertical axis by + ◦ ( U - and U (cid:48) -plane) and two others by − ◦ ( V - and V (cid:48) -plane). The configurationof the detector along the beam is U U (cid:48) , V V (cid:48) , XX (cid:48) , Y Y (cid:48) . Every second layer is staggered by 4 mm (halfof the cell dimension) in order to minimise track ambiguities. Each layer consists of 256 active wiresmade of gold plated tungsten and 257 alternating potential wires made of beryllium, with diameters of20 µ m and 100 µ m, respectively. To avoid sagging, two nylon wires per plane are fixed perpendicularlyto the active and potential wires. The distance between every two active wires is 8 mm. Each wire layeris enclosed between two 25 µ m thick Mylar cathode foils, at a distance of ± × , which are small enough to cope with counting rates ashigh as 250 kHz per wire. During operation of the detector, the active wires are kept at 0 V, whereas bothMylar foils and potential wires are set at values close to -1700 V.Central zones with a diameter of 28 . H and CF with volume fractions of 45%, 45% and 10% respectively.This gas ensures a fast charge collection (drift velocity is 77 µ m / ns) while preserving a good spatialresolution. Full efficiency is reached for gain values close to 10 , corresponding to HV settings ofapproximately 1750 V.The read-out electronics of the detector is identical to the electronics used for the already installed DCtrackers. A single front-end card with 64 channels consists of eight pre-amplifier/amplifier/discriminatorchips[21], called ASD8. Each ASD8 card is connected to a 64-channel F1-TDC board. The thresholdson the ASD8 card are remotely controlled. The nominal threshold is set at values between 0 . . Space residual (cm)-0.2 -0.1 0 0.1 0.2 E n t r i e s · m m = 118 s Fig. 31:
Space residual distribution of a Micromegas detector. The quoted residual width is obtained from a fit ofa sum of two Gaussians. he COMPASS Setup for Physics with Hadron Beams 31 y ( c m ) -100-50050100 E ff i c i e n c y Fig. 32:
Two-dimensional representation of the efficiency for one of DC4 layers. The half horizontal lines withreduced efficiency indicate the position of the power supply lines of the beam killer.
Space residual (mm)-0.4 -0.2 0 0.2 0.4 E n t r i e s · m m = 319 DR s Fig. 33:
Double residual (see text) distribution of the DC4 chamber for one of its doublets. The quoted width isfrom the fit of a simple Gaussian. system was installed. The system consists of copper pipes running on both sides of the detector frame.The copper pipes are part of a closed secondary circuit filled with demineralised water and maintained atconstant temperature. Several probes, which are installed at various locations, continuously measure theactual detector temperature. The cooling system limits the temperature variation of the frame to within2 K.2 The COMPASS Collaboration !"
Fig. 34:
Sketch of a Mini Drift Tube module.
The efficiencies of the eight DC4 planes were measured at nominal running conditions (hadron beamintensity of 5 × / s). They were found to be in the range of 95% to 97%. Figure 32 shows theefficiency of the first vertical plane (Y1) of the DC4 chamber.The spatial resolution of the DC4 drift chamber planes is determined by taking advantage of the stag-gered layers with the same orientation (doublet). The difference between the positions of the hits in thetwo planes, x and x of a doublet (double residual, or DR ), is independent of the track uncertaintyassociated with the other planes or detectors. It includes a correction ∆ x ( θ track ) , which accounts fora non perpendicular incidence of the track, i.e. a position shift when going from one plane to another.The distance between two planes being 8 mm, this correction is small. The double residual is therefore DR = x − x − ∆ x ( θ track ) . Figure 33 shows the double residual distribution for the X -doublet ofDC4, measured under nominal beam conditions. Except for a shift of half a drift-cell length, the twolayers have identical characteristics; therefore the resulting DR resolution is σ DR = σ x + σ x = σ x .The position resolution for a single DC4 plane is consequently σ x = µ m. This result is obtained inthe central region of the detector, which corresponds to about one tenth of the total detector area. The Rich Wall detector is a large-area tracker that is positioned between RICH-1 and ECAL1. Thedetector was built to improve the tracking accuracy at large angles (150 < θ <
300 mrad) downstream ofRICH-1. The additional track points measured by the detector provide a better determination of large-angle particle trajectories through RICH-1 and, as a consequence, improve the accuracy of Cherenkovring reconstruction.The detector has dimensions of 5 . × .
91 m with a central hole of 1 . × .
51 m . It consists of eightplanes of Mini Drift Tubes (MDT) made up of MDT modules. An MDT module consists of an eight-cellaluminium comb extrusion with a wall thickness of 0 .
44 mm, which is covered on the top by a 0 .
15 mmthick stainless steel foil. Gold-plated tungsten wires of 50 µ m diameter are strung in the centre of thecells. The wire pitch is 10 mm. A Noryl R (cid:13) plastic envelope with a thickness of around 1 mm encapsulatesthe module. The wires are thermally glued to polyethylene plastic spacers (not shown in Fig. 34) at equaldistances of 1 m along the length of the MDT to provide electrostatic stability. A sketch of one MDTmodule is shown in Fig. 34.Figure 35 shows a front view of an X -plane. It consists of 2 ×
25 long MDT modules (length 3910 mm),and 2 ×
12 short modules (length 1700 mm) above and below the central hole. Similarly, a Y -planecomprises 2 ×
20 long MDT modules (length 5270 mm), and 2 × X - or two Y -planes. The twoplanes within one group are staggered by 2 . !" &'($ % ( % ( % "("!% "("!%($"$% !($ % ("%("% "!%"!% Fig. 35:
Front view of an X -plane of the Rich Wall detector. The large-size numbers correspond to the number ofMDT modules in each sector, the small numbers indicate the dimensions in units of mm. !" Fig. 36:
Schematic view of the Rich Wall readout chain.
The readout electronics consists of front-end (FE) cards fixed on the detector frame and digital (DG)cards plugged into the FE cards. The FE cards are connected to the MDT signal wires via short shieldedcables. Each FE card houses 16 MAD4 chips [22], a threshold digital-to-analog converter (DAC), a test-pulse generation circuit, power-supply filters and regulators. The FE card is connected to the DG cardsthrough a high-speed card edge connector. The DG card houses eight F1 TDC [23] chips, a high-speed(40 MHz) HOTLink chip, and initialisation circuits. The card reads out 64 TDC channels in parallel.Two 8-bit Analog Devices DAC8841 chips per DG card are used to independently set the threshold ofeach MAD4 chip (common threshold for four channels) and a third one is used to generate a variable-charge test pulse. The readout chain shown in Fig. 36 is completed with FPGA-based HotGeSiCA cards(see Section 8.1) programmed in two different ways. In the first stage, the data from eight DG cardsare multiplexed onto a single connection. Eight such multiplexers are then connected to the secondmultiplexing stage consisting of one HotGeSiCA card equipped with additional random access memory(RAM), which sends the data to the readout buffers.The gas mixture used in the Rich Wall detector is Ar/CO (70/30). For this gas mixture an operatingHV of 2050 V was chosen. Ageing tests performed with this gas mixture have shown no degradationeffects for incident charges of up to 1 C per cm of anode wire length. The beam-induced MDT charge,4 The COMPASS Collaboration Space residual (mm)-5 -4 -3 -2 -1 0 1 2 3 4 5 E v e n t s s Fig. 37:
Rich Wall residual distribution, showing the difference between reconstructed cluster position and extrap-olated track position along the axis perpendicular to the wire layer. The quoted sigma is extracted by fitting a sumof two Gaussians. integrated over the lifetime of the COMPASS experiment, is comparable to this value. The Rich Walldetector is operated in the drift mode where the coordinate of a crossing track is calculated from the drifttime in the MDT cell, using the measured RT relation. Operating the detector in the drift mode allowsus to obtain a single-plane coordinate resolution of the order of 1 . The hadron physics programme at COMPASS requires the reconstruction of final states with chargedand/or neutral particles in a large angular range. Several types of particle identification detectors (PID)are used to achieve this goal (see Fig. 3). Charged pions and kaons, as well as protons, with momentaof up to 50 GeV /c are identified in the RICH-1 detector, while their energy is measured in the twohadron calorimeters, HCAL1 and HCAL2. Photons emitted during the interaction and decay photonsare detected in two electromagnetic calorimeters, ECAL1 and ECAL2. Scattered muons are identified inthe two muon identification systems, consisting of drift tubes detectors (MW1 and MW2) and absorberwalls made of iron (Muon Filter 1) or concrete (Muon Filter 2).Since the publication of Ref. [1], the PID part of the setup was significantly upgraded. New photondetectors were installed in the central region of RICH-1, and a new readout system was implemented inits peripheral region. The new ECAL1 calorimeter was added, which extends the acceptance for photondetection to large angles. The ECAL2 calorimeter was upgraded with radiation-hard Shashlik modulesin its central region and with fully pipelined electronics. For both calorimeters, the calibration procedureand the monitoring of the individual modules were significantly improved. The hadron calorimeters andhe COMPASS Setup for Physics with Hadron Beams 35 Track angle (mrad)50 100 150 200 250 300 R i ng r e s o l u ti on ( m r a d ) RW OFFRW ON
Fig. 38:
Resolution of the reconstructed Cherenkov ring for pions as a function of the track angle. The twodifferent trends in the curve below and above ∼
175 mrad are due to the different RICH-1 photon detector types(see Section 6.1). the muon identification systems remained unchanged since their description in Ref. [1] and are hence notdiscussed here.
The RICH-1 detector [24] covers the horizontal and vertical angular acceptances downstream of theSM1 magnet (250 mrad ×
180 mrad). Its 3 m long vessel is filled with C F gas as a radiator [25].The refractive index of the radiator material corresponds to Cherenkov thresholds of about 2 .
5, 9, and17 GeV /c for pions, kaons, and protons, respectively. A steel pipe with a radius of 5 cm and thicknessof 0 .
15 mm separates the vessel from the beam path. Cherenkov photons produced along the path ofa hadron are reflected by a 21 m surface that consists of 116 spherical UV mirror elements that aregrouped into two spherical surfaces [26]. The mirrors are designed such that the photons are focusedonto two arrays of photon detectors (see Fig. 39), located outside of the spectrometer acceptance. Until 2004, Cherenkov photons were detected in Multiwire Proportional Chambers (MWPC) equippedwith solid-state CsI photocathodes that limit the MWPC operation to gains below 5 × . The first stageof the electronics readout system [27] was characterised by a long integration time; this was a limitingfactor in the COMPASS environment, where a high-rate uncorrelated background is present due to thelarge muon beam halo. High rates and large correlated background are also typical for measurementswith a hadron beam. In addition, the long base-line restoration time (about 3 . µ s) generated a non-negligible dead time.In order to overcome these limitations, a major upgrade of the RICH-1 detector was undertaken. Detailscan be found in Refs [28–30]. Two different technologies were chosen in order to minimise the overallcost of the project. In the peripheral regions that cover 75% of the photo-detection surface, where thelevel of the uncorrelated background is small, the MWPC/CsI photon detectors were kept. However,their front-end electronics was replaced by a new system [28] that is based on the 128 channel APV256 The COMPASS Collaboration upp e r d e t ec t o r l o w e r d e t ec t o r
576 mm
Fig. 39:
A typical event display during hadron data taking. The 16 squares represent the detector areas; the fourcentral ones are equipped with MAPMTs. The small squares represent the hits detected in the photon detectors. chip [11]. The new system provides two major improvements. First, it reduces the effective time windowfrom 3 µ s to 400 ns and decreases the dead-time losses of the readout system to values close to 5%.Second, the APV25 chip performs a triple sampling of the MWPC signal, which results in a muchimproved time resolution and in an increase in the signal-to-background ratio [28] from 0.35 with theold system to 2.13 with the new one.The central region of RICH-1, which covers 25% of the photo-detection surface, is instrumented witha detection system based on Multi-Anode PhotoMultiplier Tubes (MAPMTs) [30]. The MAPMTs arecoupled to individual telescopes of fused silica lenses which consist of a prismatic field lens followed by aconcentrator lens, thereby enlarging the effective active area of the photon detectors by a factor of seven.The effective pad size that results from the MAPMT pixel-size and the lens telescope magnification isabout 12 ×
12 mm . The new system detects about four times more Cherenkov photons than the old oneand reaches values as high as 60 photons per ring. The MAPMT detectors are intrinsically fast and havetime resolutions better than 1 ns. They are coupled to a readout system [29] based on the MAD4 highsensitivity amplifier/discriminators and the standard COMPASS F1 TDCs. A dedicated software package, called RICHONE [31], was developed for the RICH-1 data reduction. Itperforms pattern recognition and particle identification, and characterises the detector response. Figure 39he COMPASS Setup for Physics with Hadron Beams 37shows an example of a RICH-1 event in the hadron beam environment showing many rings in the centraldetectors. The time windows applied are the same as used in data reconstruction, namely 10 ns for theMAPMT part and 250 ns in the MWPC part. Each visible ring belongs to a detected particle. A majordifference between the use of RICH-1 with muon and with hadron beam is the different particle popu-lation in the events, which is due to the different event multiplicity and particle phase space. The muonbeam is characterised by a wide halo, which extends over all photon detectors and has a flux comparableto that of the focused beam. The Cherenkov photons that are emitted by the halo particles travellingparallel to the beam are focused into the central zone of the RICH-1 photon detectors, which results in alarge background. The hadron beam can be better focused and has hence much less halo. Nevertheless,due to the higher interaction rate in hadron scattering, a large number of particles is emitted at small polarangles, i.e. in the very forward region. These particles also populate the central region of the RICH-1photon detectors. A map of the integrated hit distribution in the central part of the RICH-1 photon de-tectors is shown in Fig. 40 for data taken with muon and positive hadron beam. The distributions arenormalised to the number of entries and the same scale is used for the comparison. Both distributionsshow large occupancies for photons emitted from particles traversing RICH-1 under small polar angles.The ring images in the muon environment have more overlap since they are mostly produced by the par-allel halo particles, while in the hadron case the particles have a slightly broader polar angle distribution.Figure 41 shows the projection of the hit distribution in the lower photon detectors onto the horizontalaxis, for both the muon and the hadron environment. Even if the origin of the background is differentin the two environments, the overall background distributions are similar. The same was observed whenchanging to another hadron beam or target, so that the general properties of the detector response remainthe same as the ones measured with a muon beam [31]. The uncertainties in the reconstructed angle ofthe individual Cherenkov photons is 2 mrad in the central region and 2 . . . /c . The average number of photons per ring at saturation, i.e. for β →
1, is 56 in the central and14 in the peripheral region. The dependence of the mean number of detected photons per ring versus thecorresponding Cherenkov angle is shown in Fig. 42 for the detectors equipped with MAPMTs. x (cm) y ( c m ) -50 0 50-50050 x (cm) y ( c m ) -50 0 50-50050 Fig. 40:
Two-dimensional hit distributions in the central part of the RICH-1 photon detectors for (left) data takenwith a muon beam and (right) data taken with a positive hadron beam.
Part of the data taking in 2009 was devoted to a test of the Primakoff measurement. For this test, RICH-1was filled with N gas in order to have a smaller material budget in the acceptance region. The responseis largely different in this case as the refractive index of N is lower than that of C F . In particular, thenumber of emitted photons at saturation is expected to be lower by a factor of 4.8 for the N radiator.The number of detected photons is then sufficient to allow for particle identification only in the central8 The COMPASS Collaboration x (cm)-60 -40 -20 0 20 40 60 E n t r i e s ( a . u . ) -3 · Muon dataHadron data
Fig. 41:
Horizontal axis projection of the integrated hit distributions for the lower photon detectors. Both centraland peripheral parts of RICH-1 are included. The shaded histogram refers to the muon environment, the open tothe hadron one. The small dips in the hit distributions correspond to the dead zones between the detector partsequipped with MAPMTs and with MWPCs.
Cherenkov angle (mrad)20 25 30 35 40 45 50 55 P ho t on s p e r r i ng < 90 mrad q
30 mrad <
Fig. 42:
Mean number of detected photons per reconstructed ring as a function of the corresponding Cherenkovangle θ Ch in the central region of the RICH-1 detector for track angles θ between 30 mrad and 90 mrad. The line isa fit with the functional form N = N sin ( θ Ch ) . part of RICH-1, which is equipped with MAPMTs. In this region, the average number of photons perring at saturation is 11.7, which has to be compared with 56 for the C F radiator (Fig. 42). The lowernumber of detected photons leads to an uncertainty in the determination of the ring angle, which is largerby a factor 2.2 with respect to the operation with the C F radiator. Nevertheless, the upper momentumlimit for pion-kaon separation is very similar for the two radiators as the poorer resolution of N iscompensated by a larger difference between the corresponding Cherenkov angles. The thresholds of theCherenkov effect are 5 .
6, 20, 38 GeV /c for pions, kaons, and protons respectively. Thus, in comparisonto the values for C F quoted above, the momentum range for pion-kaon separation is severely reduced. The ECAL1 calorimeter is part of the Large Angle Spectrometer. It consists of 1500 lead glass (LG)modules. For reasons of availability and cost, three types with different dimensions are used, see Table4. The calorimeter ECAL1 has a width of 3 .
97 m and a height of 2 .
86 m, which corresponds to the angularhe COMPASS Setup for Physics with Hadron Beams 39acceptance for photons coming from the centre of the liquid hydrogen target of 37 mrad to 136 mrad inthe horizontal direction and of 21 mrad to 98 mrad in the vertical direction. The central hole has a size of1 . × .
61 m . The ECAL1 calorimeter is installed on a motorised platform that allows horizontal andvertical movements orthogonal to the beam direction, which is used mainly for calibration purposes. A front view of the ECAL1 calorimeter is shown in Fig. 43. The central part of ECAL1 consists of608 LG modules of transverse dimensions 3 . × .
83 cm , which are denoted GAMS modules [32].They are arranged in a 44 ×
24 matrix with its central 28 ×
16 array left empty. Above and below thiscentral part, two 22 ×
13 matrices of “MAINZ” modules [33] are installed, which contain in total 572LG modules. One MAINZ module has the size of nearly four GAMS modules. In order to compensatefor the small difference in size, 1 . ×
20 large-size “OLGA” modules [34]. Each OLGA module has the size of nearly four MAINZ modules. Table 4summarises all relevant parameters of the LG modules used. It also contains the type of PMT that detecttheir Cherenkov light. The analogue signals coming from the PMTs pass through shaper modules. Theshaper modules preserve the integral value of a signal and enlarge its width to 80 ns FWHM in order tomatch with the SADC sampling rate of 77.76 million samples per second.In the offline event reconstruction the SADC information is used to extract the amplitude and time of asignal relative to the trigger time. After subtracting the ADC-baseline that is determined for even and
Mainz OLGA GAMS c m Fig. 43:
Configuration of ECAL1. The central area is equipped with GAMS modules. The MAINZ modules areinstalled above and below the GAMS area. The OLGA modules cover the outer left and right regions.
Table 4:
Parameters of the ECAL1 lead glass modules.
Parameter Units GAMS MAINZ OLGALG type TF1 SF57 SF5Density g/cm .
51 4 . X ) cm 2.74 1 .
55 2 . X . . . . .
61 4 . .
65 1 .
89 1 . . × .
83 7 . × . . × . A n and A n + , which surround the position in time that represents one-half of the maximumamplitude A max . In order to improve timing accuracy, the time at which the signal is reaching 50% of A max is derived from an interpolation between the samples n and n + t = (cid:18) n + A max / − A n A n + − A n (cid:19) × .
86 ns , (1)where 12.86 ns is the sampling period. For photon energies larger than 1 GeV the resulting time resolutionis about 1 ns. For the calibration of ECAL1 LG modules the field of SM1 is set to zero. A 15 GeV electron beam isused, which is a compromise that accounts for the different dynamic ranges of the three types of modules.An automatised calibration procedure changes the position of the calorimeter between two consecutivespills, so that every module is exposed during calibration. Up to several thousands electrons per moduleare collected within each spill. The total cluster charge deposited, i.e. the sum of the charges of themodule being calibrated and its neighbouring modules, is compared to the incident electron energy.Several iterations are necessary to determine the HV settings for all modules.The calibration coefficients, which relate the charge measured by each SADC to the energy depositedin the corresponding module, are calculated taking into consideration the energy range of the photonsdetected in that module during the experiment. Since the energy of the photons decreases as the anglebetween the photon direction and the beam axis increases, three different HV settings are applied. Forthe incident beam energy of 190 GeV, the settings are chosen such that the corresponding dynamic rangesfor the three types of modules extend up to 60, 30, and 20 GeV for GAMS, MAINZ and OLGA modules.The whole calibration procedure is applied once or twice per data taking period of several months.In order to control the light collection efficiency and the photomultiplier gains of all 1500 LG modules,the ECAL1 calorimeter is equipped with a dedicated monitoring system that is based on the design ofRef. [35]. It uses a single laser source, namely a Minilite-1 model from Continuum [36]. The use of asingle light source allows the detection of possible light collection or PMT instabilities individually ineach ECAL1 channel. The laser light is transmitted to the LG modules through one primary and eightsecondary optical fibre bundles. The fibre bundles are interconnected using light diffusers that guaranteea uniform distribution of the light in the fibres. Each secondary bundle consists of 240 fibres from whichbetween 160 and 200 fibres are connected to the LG modules.he COMPASS Setup for Physics with Hadron Beams 41 C on c r e t e w a ll D1 ECAL1
Laser
FEM
To SADC
D2 D2 D2 D2 D2 D2 D2 D2 (x8 fibres)
Fig. 44:
Schematic view of the LASER monitoring system for ECAL1. The laser beam is distributed to the ECAL1modules using one primary (D1) and eight secondary (D2) light diffusion spheres. For clarity, only one of the 8primary fibres dispatching the light to D2, only one of the secondary 1500 fibres transmitting it to the LG modules,and only one of the 8 front-end-monitoring (FEM) modules are explicitly shown.
Time (days)0 1 2 3 4 5 6 7 R e l a ti v e a m p lit ud e R e l a ti v e a m p lit ud e Fig. 45:
ECAL1 module responses as monitored during a period of one week for (left) a stable module and (right)an unstable module. The vertical scale is normalised to the SADC charge measured in the beginning of the period.
A simplified drawing of the monitoring system is shown in Fig. 44. The laser injects 532 nm light pulseswith 5 ns FWHM into all calorimeter modules at a frequency of 1 Hz in the SPS inter-spill periods. Thelaser energy per pulse is tuned to an amount that matches the photomultiplier signal amplitudes. Sincethe light output of the laser source itself may vary between two consecutive pulses, an independentreference measurement of the pulse amplitudes is required. Nineteen fibres, to make the available lightsignal strong enough, from each secondary bundle are plugged into eight reference photodiodes. Eachphotodiode is connected to a temperature-stabilised electronics circuitry [35], which is enclosed in aFront-End Monitoring (FEM) module. The eight FEM signals are read out by the same SADC electronicsas the calorimeter modules, thus providing an eight-fold normalisation of the laser pulse amplitude. Theamplitude of the signals from the LG modules is determined as the peak value of the SADC samples asobtained after pedestal subtraction.The recorded laser monitoring amplitudes are used to correct the responses of all ECAL1 modules on arun-by-run basis. The electron beam calibration provides the starting values of the calibration coefficients2 The COMPASS Collaboration C i . These coefficients convert the photomultiplier signal amplitude from ADC channels to GeV using A iGeV ( t ) = A iADC ( t ) × C i × L i ( t ) L i ( ) . (2)Here, A iADC ( t ) is the ADC amplitude for the module i measured at a given time t , L i ( t ) is the ADCamplitude of the monitoring amplitude of the module i at time t , and L i ( ) is the monitoring amplitudeof the same module, but recorded during the electron beam calibration run. The time t is taken as thetime of the run, for which the correction is to be applied. A display of two LG modules (a good one andan unstable one) as a function of time is shown in Fig. 45. ECAL2 is a part of the Small Angle Spectrometer. It consists of 3068 calorimeter modules of threedifferent types, all with the same transverse dimensions (3 . × .
83 cm ). With its dimensions of2 . × .
83 m , ECAL2 covers angular ranges between 1 . . The ECAL2 modules are arranged in a 64 ×
48 matrix, as shown in Fig. 46. For data taking with hadronbeams, its central hole with respect to the nominal beam directions is set to 2 × ×
10 cm . The outermost part of ECAL2 isequipped with 1332 TF1 lead glass [32] modules, which are identical to the GAMS modules used forECAL1. The intermediate part of ECAL2 is filled with 848 radiation-hardened modules (GAMS-R)made out of TF101 material [37]. This material is a standard TF1 type LG, which is enriched with 0.2%of cerium. The innermost part is equipped with 888 Shashlik type modules (see Fig. 47). The 39 cmlong Shashlik modules are composed of 154 double layers, each consisting of a 0 . .
55 mm thick scintillator plate. The photons from the Shashlik modules are collected by 16wavelength-shifting light fibres and guided onto FEU-84-3 photomultipliers.The different ECAL2 modules have identical transverse dimensions, but different radiation hardnessproperties. Calculations for the present ECAL2 configuration have shown that with the COMPASSnominal hadron beam intensity and duty cycle the most exposed modules, i.e., those located closest tothe beam, would stand radiation doses corresponding to several years of data taking for GAMS andGAMS-R and nearly 20 years for Shashlik, without significant degradation of their response.The photomultiplier signals coming from the 3068 ECAL2 modules are first transferred to a shaper,which increases the signal width to 120 ns FWHM. The signals are then digitised by sampling ADCs.The ECAL2 readout was upgraded with a new sampling ADC system, which provides a dynamic rangeof 12 bit and allows more elaborate data processing. The basic building block is a compact MezzanineSampling ADC card (MSADC), which performs a digitisation of the 16 analogue input channels at77.76 million cycles per second, with two interleaved multichannel ADCs [38]. Data processing isimplemented by a Xilinx Virtex4 FPGA [18]. The MSADC firmware includes a digital ring bufferto compensate the trigger latency, a pipelined logic for pedestal correction, zero suppression and dataformatting. In addition, an independent processing chain is implemented on the FPGA to determine timeand amplitude information for the calorimeter trigger described in Section 7.5. As shown in Fig. 48,four MSADCs are combined on a 9U VME carrier card, which merges the data from 64 calorimeterchannels and provides a serial 40 Mbit/s HOTLink interface to the HotGeSiCA multiplexer moduleshe COMPASS Setup for Physics with Hadron Beams 43
Shashlik GAMS GAMS-‐R c m Fig. 46:
Configuration of ECAL2. The outer and intermediate regions are equipped with GAMS and radiation-hardened GAMS modules respectively. The inner region is equipped with Shashlik sampling modules. The trans-verse sizes of all three types of modules are identical. The central hole of 2 × Fig. 47:
Photographs of a Shashlik-type calorimeter module. Left part: the upstream face of the module with itsfour central rods and 16 light fibres. Right part: the module itself with the fibres guide at the downstream face. in the next readout stage. In order to reduce the power dissipation, all MSADC supply voltages aregenerated centrally on the carrier card with DC/DC converters. The resulting noise on the MSADCchannels is below 1.5 least significant bit.The information from the MSADCs is also used to calculate the time for each event. For each ECAL2module, the algorithm interpolates between the times of the two samples around the one-half value ofthe maximum sampled amplitude (see Section 6.2). The time resolution for ECAL2 is shown in Fig. 49.For energies higher than 2 GeV, resolutions of 1 ns or better are achieved.
The ECAL2 calorimeter is calibrated by exposing all its modules to a 40 GeV electron beam. Apart fromdifferent geometry and different number of modules, the calibration procedure is identical to that used4 The COMPASS Collaboration
HOTLINK connectorVirtex-4 LX25 FPGAVirtex-4 LX25 FPGABackplane connectorfor triggering 40 MHz ADCs
MSADC card
Fig. 48:
VME carrier card with four mounted MSADC modules. for ECAL1 (see Section 6.2.2). The charge deposited in each cluster of LG modules (as measured inthe MSADCs) is calculated and compared to the incident electron energy. After the data for all modulesare collected, few iterations are necessary to determine the calibration coefficients for all Shashlik andLG modules. The final HV settings are calculated after taking into account the energy ranges of thephotons detected in the different parts of ECAL2. The high voltages of the PMTs of the innermost16 ×
16 modules are set to measure energies of up to 200 GeV. In the surrounding part, which representsa 48 ×
48 matrix, the maximal energy is set at 150 GeV. Finally, in the two outermost parts with 8 × Energy (GeV)0 2 4 6 8 10 12 14 T i m e r e s o l u ti on ( n s ) Fig. 49:
Standard deviation σ for the ECAL2 time resolution as a function of the photon energy E. The solid curveis a fit to the data points using the expression: σ ( E ) = . /E + . /E + . he COMPASS Setup for Physics with Hadron Beams 45 Time (days)0 1 2 3 4 5 6 7 R e l a ti v e a m p lit ud e R e l a ti v e a m p lit ud e Fig. 50:
ECAL2 module responses as monitored during a period of one week for (left) a stable module and (right)an unstable module. The vertical scale is normalised to the SADC charge measured in the beginning of the period. use of six LEDs increases the available light intensity and minimises possible intensity fluctuations byaveraging out the individual LED instabilities. The system is activated using a calibration trigger witha frequency of 1 Hz. A display of two ECAL2 modules with a stable and with an unstable response isshown in Fig. 50. The information from the monitoring system is used to correct short and long termdrifts of individual cells on a spill by spill basis.
The COMPASS trigger system for hadron beams is designed to select events that carry all the informationneeded for exclusive measurements. A fast response is needed to provide the time reference for thereadout of all detectors. A physics trigger consists of three subsystems: beam-defining elements to selectbeam particles crossing the target, veto detectors to reject events containing particles produced outsideof the target or outside of the spectrometer acceptance, and specific detector systems that account for theparticular physics case. The latter are: i) the proton trigger (Section 7.3) that is used for measurements ofdiffractive scattering and central production processes with momentum transfers t < − .
07 GeV /c (seeSection 4.3), ii) the multiplicity trigger that completes the coverage in t for reactions with higher chargedtrack multiplicities (Section 7.4), and the calorimeter trigger (Section 7.5) that is used for Primakoff datataking. Figure 51 shows schematically the location of the trigger elements in the spectrometer. BeamKiller2BeamKiller1BeamCounterHodoscopeVetosCEDARs RPD SandwichVetoMulitplicity CounterSciFi1 (Beam trigger) SM1 SM2 HCAL2ECAL2HCAL1ECAL1
Fig. 51:
Arrangement of trigger elements in the spectrometer (schematic side view, not to scale).
The beam trigger selects incoming beam particles and is used to define the reference time of an event.In addition, it reduces the geometric acceptance of the beam in the transverse plane to match the targetgeometry. It consists of a coincidence of a scintillating fibre detector, SciFi1, with a beam counter. SciFi1is located 7 m upstream of the liquid hydrogen target. It has one vertical and one horizontal plane. Eachplane is read out by six multi-anode photomultiplier tubes (PMT) with 16 channels each. In addition, thePMTs are read out at the last dynode stage, thus providing six analogue sums for each of two planes ofthe detector.The beam counter is a small scintillator disc that is located 50 cm downstream of SciFi1. It has a diameterof 3.2 cm, a thickness of 4 mm, and is centred at the beam. It is surrounded by a thin, black PVC tubecovered inside by aluminised Mylar foil with an internal reflection of better than 92%. A 35 cm longtube used as an air light-guide is connected to a single EMI 9813KB PMT. The PMT is equipped with avoltage divider that stands beam rates of up to 10 MHz. The efficiency of the beam counter was measuredto be 99.5% all over the surface of the disk, as shown in Fig. 52. E ff i c i e n c y x (mm)-15 -10 -5 0 5 10 15 y ( mm ) -15-10-5051015 Fig. 52:
Beam counter efficiency distribution in transverse coordinates.
The beam trigger is defined by the coincidence of the beam counter signal and the logical OR of the 6analogue sums of the SciFi1 X plane. The time resolution is measured to be 450 ps ±
50 ps (Fig. 53). Itis used as a time reference of the trigger system.
Time (ns)-10 -5 0 5 10 E n t r i e s · RMS = 0.45 ns
Fig. 53:
Time residual of the beam trigger. he COMPASS Setup for Physics with Hadron Beams 47
The veto system consists of two scintillation counters (“beam killers"), a ‘sandwich" veto detector (seeSection 4.4) and a hodoscope veto system. It inhibits false physics triggers. The overall dead time of theveto system was measured to be 13%-16% for the nominal beam intensity of 5 · s − . For Primakoffdata taking, it is reduced to 8%-10% due to the lower beam intensity. Two scintillating counters are positioned along the beam axis of the spectrometer at z = +25 m (BK1)and z = +33 m (BK2). Both counters have a diameter of 3.5 cm and a thickness of 0.5 cm. Their functionis to inhibit a trigger signal coming from non-interacting beam particles. The use of the beam killersintroduces an angular cut-off of 0.97 mrad with respect to the nominal beam axis. Including the beamkillers in the diffractive trigger (see Table 5), reduces its trigger rate by about a factor of 2. A Sandwich veto detector (described in Section 4.4) is used to veto charged and neutral particles thatare detected outside of the angular acceptance of the spectrometer and the RPD. Such particles aredominantly produced in inelastic, non-diffractive reactions or in reactions in which the target protonsare diffractively excited. Including the Sandwich veto in the trigger improves the purity of the physicstriggers by a factor of about 3.5.
The hodoscope veto system is the same as the one used for the muon programme. It consists of threeparts: a beam line hodoscope veto system (V BL ) installed at z = −
20 m, a Veto1 system located at z = − . z = − . Fig. 54:
Allowed combinations for target pointing in the RPD part of the proton trigger.
The proton trigger selects events with recoiling protons from the target. The RPD (described in Sec-tion 4.3) information is used for two purposes: target pointing and discrimination of protons from pionsand delta-electrons by measuring the energy loss in each ring of the RPD. Target pointing is implementedby allowing only for combinations, where hits in one scintillator of the inner ring are followed by a signalin one of the three corresponding outer ring scintillators, as shown in Fig. 54.For a particle traversing the RPD, its energy losses in the inner and in the outer rings are strongly cor-related. This is used to reject electrons coming from the target as well as part of the low-energy pions.8 The COMPASS CollaborationFig. 55 shows the calculated energy losses for both protons and pions, and for the minimum and max-imum polar angles (50 ◦ and 90 ◦ ) of the RPD acceptance (see Section 4.3). The area to be rejected isdefined using the two levels of discriminator thresholds in both rings. The coincidence of low-thresholdsignals for upstream and downstream PMTs of the inner (outer) ring is denoted by A Lowi ( B Lowj ), where i and j are the respective scintillator elements. Similarly, the coincidence of the two high-threshold sig-nals is denoted by the superscript “High". The trigger logic function for recoil protons has the followingexpression: RP D = (cid:95) i = A Lowi,down ∧ i + (cid:95) j = i − (cid:16) A Lowi B Highj ∨ A Highi B Lowj (cid:17) . (3)Here, the signals from the downstream PMTs of the inner ring, A Lowi,down , are used to minimise the timejitter with respect to the beam trigger. The trigger logic is set to reject the electrons that cross both ringsas well as pions that cross ring A but leave less than few MeV in ring B. In Fig. 55, the region rejectedby the proton trigger is indicated by the shaded area.In order to be able to measure the time-of-flight of all recoil protons, irrespective of their velocities, alarge time window of 50 ns is required for the coincidence between any of the inner-ring downstreamPMTs with the geometrically allowed outer-ring PMTs. The trigger logic function (3) is implemented ina single FPGA module that is fed by the logic signals from all PMTs of the RPD.
The multiplicity triggers were built to extend the measurements to events with momentum transfers − t smaller than − t < .
07 GeV /c that are outside the acceptance of the proton trigger. It uses themultiplicity counter to estimate the charged-particle multiplicities in the beam region or tag events withat least one (or two, see Table 5) track at large angles.The multiplicity counter (Fig. 56) consists of 12 trapezoidally shaped scintillator slabs with a centralhole of 20 mm diameter. It covers the charged-particle acceptance of the spectrometer at 1.7 m, whichprojects to a disk with a radius of 310 mm. The light is read out by one photomultiplier per slab. Thecounter was upgraded in 2009 with a scintillator disk with a diameter of 32 mm, which is centered atthe hole and is read out by two photomultipliers. In order to minimise photon conversion in the activearea, all scintillators have a thickness of 3 mm, which corresponds to 0 .
71 % of a radiation length. The
Energy loss ring B (MeV)5 10 15 20 25 30 35 E n e r gy l o ss r i ng A ( M e V ) protonspions (cid:176) protons 50 (cid:176) protons 90 (cid:176) pions 50 (cid:176) pions 90 Fig. 55:
Correlation between the energy losses of protons and pions traversing ring A and stopping (or traversing)ring B of the RPD. For each particle type the minimum and the maximum polar angles (50 ◦ and 90 ◦ ) are shown.The shaded area corresponds to the region rejected by the trigger logic. he COMPASS Setup for Physics with Hadron Beams 49 Fig. 56:
The multiplicity counter. All dimensions are in mm. photomultipliers for the inner disk are connected through an 83 cm long air light guide made of a tubeskeleton of 15 µ m aluminised Mylar inside a 150 µ m thick plastic coating.The two components of the multiplicity detector, the outer and the inner counters, are used to build twoindependent triggers, MT1 and MT2. The MT1 multiplicity trigger requires one hit or more in eachelement of the outer multiplicity counter. For this purpose, the threshold per element is set to reject noiseonly, thereby selecting charged particle multiplicities of one or larger. A logical OR of all elements isthen used as the trigger signal. The MT2 multiplicity trigger requires an energy deposit correspondingto 1.6 MIPs or higher in the inner counter. Data were also taken in stricter conditions for both MT1and MT2. For MT1 this was achieved by using a multiplicity logic instead of the logical OR, therebyselecting events where two or more slabs of the outer counter are hit. For MT2 a higher energy depositequivalent to 2.5 MIPs was required to select multiplicities of three or larger. These conditions enrichevents that have final states with higher multiplicities and therefore higher masses. The calorimeter trigger selects high-energy photons detected by ECAL2 within 12 ×
12 cells, whereby 8cells surrounding the beam hole are excluded, as depicted in Fig. 57. The trigger logic is implementedin the existing ECAL2 readout module described in Section 6.3.1. At the first stage, FPGAs mounted onthe MSADC cards detect a signal and extract, on a cell-by-cell basis, amplitude and time information.The time information is obtained using a digitally implemented constant fraction algorithm. In order to minimum ionising particle σ = .
97 GeV for a 60 GeV threshold, determined byfitting an error function, as shown in Fig. 58. The main contributions to this precision are the accuracyof the per-channel thresholds and the preliminary calibration constants used in the trigger system. Theoverall time resolution could be reduced to about 1 ns by digital signal processing, as illustrated in Fig. 59.
X (cells)0 10 20 30 40 50 60 Y ( ce ll s ) Fig. 57:
The active area of the ECAL2 trigger (shown in blue). The cells shown in orange are rejected due to highrates.
Energy (GeV)40 60 80 100 120 140 E ff i c i e n c y s Fig. 58:
Efficiency of the ECAL2 trigger as a function of the energy. The solid line is a fit to the data with an errorfunction. he COMPASS Setup for Physics with Hadron Beams 51
Time (ns)36 38 40 42 44 46 48 50 52 E n t r i e s · = 0.96 ns s Fig. 59:
Time resolution of the CFD algorithm for a representative cell in the centre and signal amplitudes above800 MeV.
The final physics triggers are summarised in Table 5 together with their typical rates. A standard physicstrigger is generated by a combination of the beam trigger, the veto system, and one of the specialisedtriggers described above.The diffractive trigger (DT0) is the main physics trigger for spectroscopy data taking. Based on theproton trigger, it selects events with recoiling protons from the target. Besides the low angle cut-off ofthe beam killers, DT0 introduces only a minimum bias on the angular acceptance of forward particles.The "low- t " triggers, LT1 and LT2, are especially important for measurements with solid-state targets. Inthis case, the recoil proton has to pass dense material that can lead to large uncertainties due to multiplescattering or protons stopping in the material. This means that for heavy targets the DT0 trigger is not asefficient as for the hydrogen target. Therefore, a large part of the solid-state target data is recorded withprescaled LT1 and LT2.The Primakoff trigger (Prim1) uses the calorimeter trigger with a 60 GeV threshold. A secondary Pri-makoff trigger (Prim2) is based on a calorimeter trigger with a threshold of 40 GeV and a prescalingfactor of two. Its purpose is to monitor the Prim1 trigger threshold.The kaon trigger (KT) makes use of the CEDAR detectors in the beam line, which are set to detectbeam kaons (see Section 3.3). Signals from both CEDARs need to be present for the trigger in order tomaximise its purity. It is used as a kaon-enriched beam trigger for luminosity monitoring via K → π decays and for systematic studies.Further auxiliary triggers are set up for monitoring purposes, systematic studies and alignment purposes(see Section 9.2). They include an additional beam trigger with a transverse acceptance of 3 . × . ,which is required for the alignment procedure. The Veto Inner trigger and Halo triggers make use ofthe hodoscope veto system to detect straight halo tracks for muon data taking, which is utilised in thealignment procedure, as well.All inputs to the trigger system and the signals of the individual sub-triggers and triggers themselves2 The COMPASS Collaboration Table 5:
Overview of trigger subsystems, vetos and physics triggers used for data taking.
Trigger subsystem Logical composition
Beam trigger (BT) SciFi1 ∧ beam counterBeam killer veto beam killer 1 ∧ beam killer 2Veto Sandwich ∨ veto hodoscopes ∨ beam killerProton trigger see Eq. 3Multiplicity trigger MT1 1 (later 2) el. of outer ring counterMultiplicity trigger MT2 amp. inner disk > (cid:80) × cell amplitude > thresholdCEDAR trigger CEDAR1 multiplicity ∧ CEDAR2 multiplicity
Physics trigger Logical composition Rate / 10 s spill
Diffractive trigger DT0 BT ∧ proton trigger ¯ ∧ veto 180kLow- t trigger LT1 BT ∧ MT1 ¯ ∧ veto 370k (140k)Low- t trigger LT2 BT ∧ MT2 ¯ ∧ veto 620K (260K)Primakoff trigger Prim1 BT ∧ calorimeter trigger ( >
60 GeV) ¯ ∧ veto 260kPrimakoff trigger Prim2 BT ∧ calorimeter trigger ( >
40 GeV) ¯ ∧ veto 450kKaon trigger KT BT ∧ CEDAR trigger ¯ ∧ veto 30k are monitored with TDCs and scalers. In addition, the individual signals of the multiplicity counters aremonitored by sampling ADCs. The COMPASS data acquisition system (DAQ) has been designed to cope with high trigger rates andlarge data flow. For data taking with hadron beams both interaction rate and particle multiplicity perinteraction are higher than for a muon beam, making these requirements even more important. The DAQis based on a pipelined architecture, which was fully implemented for the 2008/2009 data taking. Itis complemented with a Detector Control System (DCS), which permanently monitors all parametersrelevant for the operation of the setup. The general structure of the COMPASS DAQ and DCS systemswere described in Ref. [1]. In this section their main characteristics are shortly reviewed; only the mostimportant improvements and modifications are discussed.
For typical hadron beam intensities of up to 5 · particles per spill, the various COMPASS triggerscombine to a total trigger rate of more than 30 kHz. The overall number of electronic channels is largerthan ∼
250 000, and the generated event size has a mean value of 40 kB. Accordingly, a data rate of upto 1.2 GB/s is acquired during the 9 . Event size (Bytes) ) L T r i gg e r R a t e ( s
10 COMPASS CMS ATLASALICELHCbKTeV D0CDFOPAL BaBarH1BELLE I
Fig. 60:
Trigger rate versus event size. The COMPASS DAQ system is compared to several large-scale experi-ments. The comparison is done for first-level (L1) triggers or their equivalent. tors: CATCH and GeSiCA, including a more recent version, called HotGeSiCA. All modules are housedin VME crates. Spill and event numbers, reference clock and synchronisation signals are provided to theconcentrator modules by an optical link coming from the TCS. The HOTLink interface is also used totransmit this information to the front-end cards.Each 9U CATCH module [41] houses four CATCH mezzanine cards (CMC) that receive detector signalscoming from Micromegas, scintillating fibres, wire chambers, and hodoscope detectors. The CATCHfirmware merges the data from the mezzanine cards and transmits them to the central readout buffercomputers through a S-LINK optical link [42].For detectors with low occupancies the data of up to 4 CATCH modules are multiplexed by an S-LINKMUX card before being transmitted to the readout computers. The S-LINK MUX card houses oneS-LINK source card and is mounted on P3 connectors on the backside of a VME crate.The 9U GeSiCA modules can read up to four 12-channel GEM or Silicon SADC cards. The more re-cent 6U version named HotGeSiCA is also able to read SADC, MSADC, and APV data from the RICHand PixelGEM detectors. Although smaller in size, the HotGeSiCA module has eight RJ45 or opticalHOTLink ports, instead of four for GeSiCA. In addition, the HotGeSiCA module can be equipped with500 MB of memory and a HOTLink output interface for cascading HotGeSiCA modules and concentrat-ing the data from up to 64 front-end cards. This readout scheme is used for the Rich Wall detector (seeFig. 36). Similar to the CATCH modules, the GeSiCA and HotGeSiCA modules send data through a S-LINK interface to the readout buffers, while the information coming from the TCS receiver is transmittedto the front-end electronics through a HOTLink connection.The DAQ system is composed of two main types of computers, called Readout Buffers (ROB) and Event4 The COMPASS Collaboration
Concentration modules (~x160)
GeSiCA or HotGeSiCA ROB + PCI-‐e spill buffers
Detector Front-Ends (~x1500….)
TCS TCS
S-LINK Multiplexer Read-out buffers PCs (x30) Spill buffers (up to 4/PC)
Gigabit Switch
Ethernet Switch
Central Data Recording
Event Builders Online filter 10 Gbit/s link
MUX
S-LINKs HOTLinks
CATCH+ CMC
SADC TDC/Scalers
CondiLons Database
Fig. 61:
Overview of the COMPASS DAQ system. Data coming from the detectors are first digitised in the front-end cards and then merged in the concentrator modules, either CATCH or GeSiCA(HotGeSiCA). The data fromthe concentrator modules are first sent to the Readout Buffers and then transmitted to the Event Builders. The dataare temporarily saved on disk, before being migrated to the Central Data Recording facility. he COMPASS Setup for Physics with Hadron Beams 55Builders (EB). All computers run Linux operating system (see Table 6). Each ROB is equipped with up tofour spill-buffer PCI cards. The PCI cards collect the information from the corresponding concentratormodules via S-LINK optical fibres. The data are temporarily stored in a daughter SDRAM card of512 MB or 1 GB memory during the spill, before being fully transmitted to the computer. All ROBs areconnected to all EBs through a Gigabit Ethernet interface. The role of each EB is to build a completeevent using the information from all the ROBs, to split the data into files of 1 GB each (chunks), andto store these files on its internal disks. In addition, the EBs run an on-line filtering software used forboth data filtering and data quality monitoring. The software used for the data acquisition is the DATEacquisition framework [43], developed for the CERN experiment ALICE.From the EBs disks, the data files are transferred to the CERN computer centre into the CASTOR hierar-chical storage system [44]. Files are copied to CASTOR disk pools by multiple TCP/IP streams througha 10 GB/s optical link and then stored to tapes. Up to 20 TB of data per day can be stored on tape whenthe experiment is running.
In order to achieve high trigger rates with reasonable dead times, several improvements of the dataacquisition chain were necessary. The dead time, as defined by the Trigger Control System, depends onthree minimum time intervals. These are: the time interval between two consecutive triggers, the timeinterval for three successive triggers, and the time interval for ten successive triggers. During data takingwith a muon beam these values were set to 5, 75 and 250 µ s, respectively.For data taking with hadron beams, the minimum time between two consecutive triggers was decreasedto 3 µ s. Smaller values were prevented due to a noise correlated with the previous trigger and appearingon the front-end cards of the MWPC detectors. The minimal time interval for no more than three triggerswas set at 30 µ s. Smaller values could lead to an overflow in the TDC multi-event buffer. A third minimaltime interval is required by the internal buffer of the APV chip, which can store up to 10 events. Thistime is set according to the speed of the analogue output signal sent by the APV chip to the SADC cards.Analogue values are sent by the APV chip at a frequency of 20 MHz, corresponding to a digitisationtime (including overheads) of 21 µ s per event. More than 200 µ s are then necessary to read ten events,justifying the 250 µ s limit. Dedicated tests have shown that in the future the ten events interval can bereduced to 125 µ s if the APV read-out frequency is increased to 40 MHz and the SADC card firmware ismodified accordingly.Figure 62 shows a comparison between the dead times resulting from the different trigger settings. Theimprovement of the time interval for three triggers results in a significant decrease of the dead time. Forthe nominal hadron data taking trigger rates of 30 kHz the new settings generate a dead time of 15%instead of 28%. The COMPASS DAQ system operates with a large number of hardware (Table 6) and software com-ponents, which are controlled through various parameters. These parameters are produced by on-lineprocesses, operator entries, slow-control of the detectors, or result from specific run conditions. Togetherwith other monitored quantities, such as trigger rates, run and spill information, they are stored in severalMySQL [45] and Oracle [46] databases.The front-end configuration database incorporates all information relevant to the front-end modules andprocessors. The logbook database collects a large number of experimental parameters and operatorcomments relevant for a specific data-taking period, usually defined as a run. Parameters that are likelyto change more frequently, such as beam information, beam line settings, scalers and some monitoringvalues, are stored in a spill database . The
DATE configuration database is used for the description of the6 The COMPASS Collaboration
Attempt Rate (kHz)10 20 30 40 50 60 70 s ) m D ea d ti m e ( Fig. 62:
Data acquisition dead time for three different TCS settings, as measured as a function of the attemptedtrigger rate. The settings used in 2008/2009 are shown in red triangles.
Table 6:
Summary of the COMPASS data acquisition hardware, as used in 2009. Dual and quad-core processorsare indicated in the parenthesis near the processor name. Both memory and disk sizes are given in GB.
Service Nb Processor Memory size Disk size
Event builders 12 Xeon, 3 GHz ( ×
2) 4 11008 Xeon, 2.5 GHz ( ×
4) 4 5600Read-out buffers 5 Xeon, 2 GHz ( ×
4) 4 5008 Xeon, 3 GHz ( ×
2) 4 25016 Pentium-3, 866 MHz 1 18File servers 2 Xeon, 3 GHz ( ×
2) 4 1100Gateways computers 2 Pentium-4, 3 GHz ( ×
2) 2 82Database servers 3 Xeon, 3 GHz ( ×
2) 4 1100Run control 4 Pentium 4, 3 GHz ( ×
2) 2 82Front-end CPUs 26 Celeron, 336 MHz 0.256on-line computers (ROBs, EBs, FSs,..) as well as for parameters relevant for the data acquisition. Thisdatabase also includes the configuration of the on-line filter software. Another database, called
DATEmessage log database , collects all process logs and messages.The 130 GB of data from all MySQL databases are hosted on two physical servers, synchronised througha master–master replication. Clients connect to the database through a virtual address pointing to a thirdserver. The third server runs a MySQL Proxy software [47] that monitors the communication betweenthe client and the database. Besides the proxy, the third server also hosts a web server Apache [48]and a monitoring service (Nagios) [49]. The web service provides interfaces to run logbook, databaseadministration programs, and diagnostic tools. Nagios monitors the availability of database servers andthe state of replication.A specific database table is used for the Detector Control System (DCS), e.g. for monitoring of ECAL1and ECAL2 modules. On the other hand, read-out values of parameters obtained by the DCS indepen-he COMPASS Setup for Physics with Hadron Beams 57dently are copied to a dedicated database table.Full MySQL database backup is being executed regularly, whereby the binary log that is created duringthe replication is regarded as incremental backup. Furthermore, the databases are periodically replicatedinto the CERN computer centre.
The Detector Control System (DCS) [50] collects data from the various detectors, hardware devices,and data acquisition elements with programmable reading cycles. For the COMPASS experiment theactual cycle times range between 2 seconds and 30 minutes. It provides a user-friendly interface, whichis used to set remotely most parameters relevant for operating the experimental setup. When predefinedconditions are met, namely if monitored values go beyond predefined thresholds or settings, it displayson-line warnings and alerts in the user interface, sounds acoustic alarms in the control room, notifiespredefined recipients by SMS and email and, when necessary, switches off sensitive detector channels.All values and alerts are stored in a centralised Oracle database with a frequency of typically few minutesper monitored parameter. Queries on the database are executed regularly for storage of data, or ondemand.The DCS architecture consists of three layers: the supervisory layer, the front-ends layer and the deviceslayer. The supervisory layer of the DCS is based on a commercial SCADA system (Supervisory Controland Data Acquisition), PVSS-II [51], adopted by CERN. On top of PVSS-II, a package of software toolscalled Joint COntrol Project (JCOP) Framework [52] is also used. Developed at CERN, this packageis specific for high-energy physics applications. The front-ends layer includes the drivers necessary forthe hardware devices and provides the communication protocol between the supervisory layer and thedevices layer. The devices layer comprises all hardware elements and sensors.The system is flexible enough to easily incorporate new detectors and monitored parameters. For the datataking with hadron beams in 2008/2009, a number of new detectors were included in the system: RPD,Sandwich veto, Multiplicity Counter, Beam Killers, Beam Counter, PixelGEMs, liquid hydrogen target,and the new Silicon detectors with their cryogenic devices. In addition, the monitoring of the CEDARsand the two electromagnetic calorimeters was considerably improved.New high voltage and low voltage channels and VME crates were used and integrated in the DCS forthe hadron data taking. They are monitored and controlled by OPC or DIM servers [53, 54], with whichthey communicate by use of the CAN and CAENet field buses [55].The monitoring of CEDAR parameters (pressures, temperatures, HVs and motors) is done via a DIPserver maintained by CERN [56]. The ratio of the pressure to the temperature is calculated for everyspill. If the value is found outside the appropriate range, a warning signal requesting a correction of theCEDAR gas pressure is generated.The monitoring of the two electromagnetic calorimeters required a substantial extension in the numberof monitored channels, namely 1500 for ECAL1 and 3068 for ECAL2. Between spills, the calorime-ter modules of ECAL1 and ECAL2 are flashed by a laser and LED light pulses, respectively (see Sec-tion 6.2.2 and Section 6.3.2). The DAQ on-line filtering software collects the responses from all modules,calculates the average amplitudes for each spill, and stores them in the conditions database. The DCSreads them, compares them to the reference values, and defines its state of alert. The voltage and thecurrent of the powering system of ECAL1 and ECAL2, the power supply of the LED monitoring systemand the status of the laser of ECAL1 were added to the list of controlled parameters.The monitoring of the liquid hydrogen target and of the cryogenic system of the Silicon detectors is doneusing a dedicated Programmable Logic Controller (PLC) and a control system. Pressures, temperatures, Fig. 63:
Implementation of supervision, front-end and device layers of the Detector Control System. vacuum gauge values and liquid levels are transmitted to the DCS using a MODbus [57] server.The data generated by the DCS are temporarily saved on a local disk whose contents are transmitted toa centralised CERN Oracle database with a cycle of a few seconds. Parameters that are relevant for thephysics analysis are regularly copied to a COMPASS MySQL database or are provided to users as ROOTtrees or ASCII files.
The main reconstruction software is called CORAL (COmpass Reconstruction and AnaLysis); it is de-tailed in Ref. [1]. CORAL outputs mDST (“mini Data Summary Trees”) files that contain reconstructedevents and information related to the detectors. The mDST files are organized in tree-like structuresbased on the ROOT [58] package. The information stored in the mDST is analysed using a dedicatedsoftware called PHAST (PHysics Analysis Software Tool). Besides accessing the mDST data, PHASTprovides an environment for physics analysis and includes tools for mDST creation, further data process-ing and filtering. The event reconstruction in CORAL comprises all detectors except the RPD and theCEDARs, which are included at the PHAST level, i.e. during the mDST processing stage.This section describes the tracking method, the detector alignment procedure, and the vertex reconstruc-tion technique used in the analysis. It gives also details on the analysis flow for detectors introduced intoor upgraded for the hadron setup, namely for RPD, RICH-1, CEDARs, and ECALs.
The track reconstruction software reconstructs trajectories of charged particles, thereby determining suchproperties as their momentum and total radiation length traversed. It uses the measurements from thetracking detectors and combines them with the description of the magnetic fields and material distribu-tion in the setup. For the material distribution the ROOT geometry package [59] is used. The includeddetectors are all tracking detectors, trigger hodoscopes and the beam telescope (SciFi and Silicon detec-tors before the target in Fig. 3). Prior to the reconstruction process, a time cut relative to the trigger timeis applied to all hits.The track finding algorithm is subdivided into two steps. First, it searches for straight track segments inthe zones that are free of strong magnetic field or of large material thickness. In these zones the particlehe COMPASS Setup for Physics with Hadron Beams 59trajectories can be approximated by straight lines. In a second step, called “bridging", the straight tracksegments from different zones are combined over dipole magnets and hadron absorbers. In order toaccount for the deviation from a perfect straight line, as caused by fringe fields and multiple scattering,an iterative approach is used with progressively wider search roads. This approach is motivated by theidea of first solving the case of the straighter tracks, which have higher momenta, and turning to themore difficult case of the lower-momentum tracks only after the hits used in the first iterations have beenremoved from the search procedure. This scheme still yields many ghost tracks, particularly in the lateriterations that have wider “roads". To filter these out, candidate tracks are checked against a lookup tableof all tracks within the accessible phase space, which was produced in a dedicated MC simulation.The track fitting procedure is based on a Kalman filter [60]. It comprises the treatment of multiplescattering, which is based on a prior estimate of the track momentum. A “forward" fit, which startsfrom the most upstream tracking plane, gives the best estimate of the track parameters in the plane ofthe detector with the last hit. A “backward" fit, which starts from the most downstream tracking planeusing the same hits, gives the best estimate at the first hit. A process combining the results of the two fits,which is known in the Kalman formalism as “smoothing", is used to determine the local best estimate atany position along the track. For each of these estimates, the procedure also determines the uncertaintyin terms of the covariance matrix of the parameter vector. Outlier detection and elimination as well asthe resolution of left/right ambiguities in drift detectors are also done within this framework.For the data taken with hadron beams, the sequential three-step structure of straight zone track finding,bridging and fitting was adapted to the high rate. A search for straight track projections in the verticalplane spanning several zones of the spectrometer is attempted at an early stage. A re-evaluation ofthe hit patterns is undertaken after bridging, once the momenta are fairly well known and the numberof competing candidate tracks is reduced. These modifications improve the tracking through the driftchambers that are located in high fringe fields around the SM1 magnet and in the Silicon vertex detectorsthat are placed in the close vicinity of the target and are densely packed with hits.The event reconstruction was further improved by allowing some update of the track information aftervertex reconstruction (see Section 9.3). If the weighted mean time of the hits associated to the incidentparticle of a primary vertex deviates from the trigger time more than expected by statistical fluctuations,in the order of a few ns, the latter is re-evaluated. After correcting it and updating the hits in the driftdetectors, the tracks are either refitted or the search is restarted from scratch.For the evaluation of the performances of the tracking package, a GEANT3-based simulation of theCOMPASS setup is used, as described in Section 10.1. The evaluation is based on criteria of association,reconstruction, and reconstructibility. A track is associated to a MC particle if more than a fraction f of the hits originate from this particle. Here, pions decaying into muons are considered the same singleparticle. A MC particle is declared reconstructed if a track fulfilling some ad hoc requirements can beassociated to it. Not reconstructible (or not worth to be reconstructed) are those particles that fall outsidethe acceptance or are not relevant for the physics process being under study. In the examples presentedbelow, the fraction f is set at 75%, only primary particles are considered, the reconstructed tracks mustbe bridged over one or both magnets and must originate from the primary vertex.Several MC samples were evaluated, corresponding to different final states of the COMPASS hadronprogramme. They give similar performance values. As an example, Figs. 64, 65 show the results for thecase of the dissociation of 190 GeV /c pions on a hydrogen target into five charged pions with an invariantmass in the range 1 GeV /c ≤ M ( π ) ≤ /c . This process was selected, because it has a largerangular coverage compared to other processes.The efficiency is defined as the ratio of number of reconstructed reconstructible particles over the numberof reconstructible ones. With the criteria defined above, this corresponds to the fraction of primary par-ticles reconstructed with momentum and connected to a vertex. The efficiency represents the combined0 The COMPASS Collaborationperformance of all involved detectors of the reconstruction software: a particle may fail to be recon-structed because it decays, re-interacts, re-scatters, because of inefficiencies of the trackers or because ofdeficiencies of the algorithm. In order to isolate the contributions of the software, the efficiency that anideally performing algorithm would reach is computed in a special mode by exceptionally making use ofthe Monte Carlo truth information. The usual track finding steps are bypassed and the track hit patternsare determined instead by accumulating hits along the known trajectories of the generated particles upto a point where multiple scattering or reinteractions become dominant. The overall efficiency is thenfactorised into this ideal efficiency characterising the setup and a software contribution. The two factors,as well as their product, are shown as a function of momentum in Fig. 64. While beyond 10 GeV /c theoverall efficiency is nearly flat, it starts to decrease below that value. The software contribution is stableat 95% down to about 1 . /c . Momentum (GeV/c)1 10 E ff i c i e n c y ( % ) Fig. 64:
Efficiency of tracking and vertexing as a function of momentum with the efficiency of tracking software(red, solid line), setup efficiency (hatched, green area), and overall efficiency (crossed, blue area).
The momentum resolution is obtained from the statistical distribution of the momentum residual for thereconstructed sample. The distribution is first binned as a function of the momentum, and in each binit is fitted with a double Gaussian and the average standard deviation is taken as the resolution. This isdone for particles bridged over SM2 (and possibly also over SM1), those bridged over SM1 only, andthose only tracked in the fringe field of SM1. The latter have a very poor resolution, but can nonethelessbe useful to reject unwanted final states. The simultaneously obtained angular resolution is dominatedby the contribution from multiple Coulomb scattering in the target material (5% X in case of the liquidhydrogen target). In order to achieve the optimal reconstruction performance, a precise knowledge of the position andorientation in space of the more than 200 tracking detector planes of the COMPASS spectrometer ismandatory. In many cases, the geometrical survey of the experimental setup does not reach a precisionthat is comparable to the spatial resolution of the detectors. It is used as the starting point for an alignmentprocedure, which uses a sample of reconstructed tracks. The whole procedure is done in three steps withdifferent sets of data. Each step is repeated until the corrections become negligible compared to thedetector resolution.he COMPASS Setup for Physics with Hadron Beams 61
Momentum (GeV/c) 0 20 40 60 80 100 120 140 160 P / P ) ( % ) D ( s P / P ) ( % ) D ( s F r i ng e f i e l d SM1 fringe fieldSM1SM2
Fig. 65:
Relative momentum resolution as a function of track momentum. The standard deviation of the recon-struction error is shown for tracks deflected by the SM2 magnet alone or by both SM1 and SM2 (squares), by theSM1 magnet alone (circles) and for those deflected by the fringe field of SM1 only (triangles, right scale).
The first step uses data recorded with a muon beam with the spectrometer magnets switched off. There-fore, straight trajectories can safely be assumed and all spectrometer arms including the beam telescopecontribute to the reconstruction of a particle track. In order to reach a broad illumination of all spec-trometer parts, these data are recorded with a widely defocused muon beam and by using both beam (seeSection 7.1) and veto counters (see Section 7.2) as trigger. The alignment is performed by minimising thetotal χ of all tracks in the sample, keeping four detector planes (GM04XY, GM10XY) fixed. For thesepivotal points, the positions determined by an optical survey of the experimental setup has to be used inorder to keep the coordinate system fixed in space. For all other planes, corrections for a translation alongthe measured coordinate, a rotation around the beam axis, and the effective pitch are introduced. The ef-fective pitch takes into account a possible inclination of the detector plane with respect to the beam axis.The position along the beam axis is normally fixed to the position determined by the geometrical survey;a fit is only attempted if the residual distribution with respect to the beam axis of a given plane indicatesa possible problem. The minimisation is done by the Millipede program [61], which analytically invertsa large but sparsely populated matrix.For the second step of the procedure, the detector planes downstream of the target are aligned with thespectrometer magnets switched on. The magnetic field not only shifts the positions of the mechanicalsupport of some detectors, but also influences the internal processes of charge propagation in gaseousdetectors. The effect is strongest for some of the small-area trackers in the fringe field of SM1 (MM03and GM01), where the Lorentz-force acting on drifting and amplified charges results in an apparenttranslation of the detector planes of up to 400 µ m. For these detectors, a correction in form of an effectiveshift is applied, since the distortion is uniform over the active area of the respective detector within thespatial resolution.As the third step, the beam telescope upstream of the target is aligned with respect to the spectrometerthat is kept fixed in space. This step is essential to optimise the reconstruction of vertices in the target.As a result, a primary vertex is reconstructed for up to 90% of the triggered events.The alignment of the tracking stations is completed by an separate procedure for the silicon stationsto fully exploit their high resolution. For these stations, displacements of up to 50 µ m were observed,2 The COMPASS Collaboration Time (h)0 5 10 15 20 25 30 35 40 45 m ) m P o s iti on c o rr ec ti on ( -505101520 SI01USI02USI03USI04USI05U
Temperature C ) (cid:176) T e m p e r a t u r e ( Fig. 66:
Run-by-run alignment correction applied to the silicon detector positions and correlation with ambienttemperature. which are caused by variations in the temperature of the support structure. Therefore, a separate align-ment for the silicon micro-strip telescope was produced for each run to account for these variations. Asan example, Fig. 66 shows the corrections applied to the five silicon tracker stations in the horizontalplane as a function of time. The effect of this time-dependent alignment on the resolution is illustrated inFig. 67 for the distribution of the scattering angle vs. the vertex position. For the events with Primakoffkinematics, in which the one outgoing track has a a very small scattering angle, the improvement issubstantial. The background can therefore be reduced by a considerable fraction. As a result the distri-bution in Fig. 67 (right) matches the simulation, in which a perfect alignment is assumed (see Fig. 89 inSection 10.1).
Position along beam axis (cm)-90 -80 -70 -60 -50 -40 -30 -20 -10 0 S ca tt e r i ng a ng l e (r a d ) -3 · Ni WW Si Position along beam axis (cm)-90 -80 -70 -60 -50 -40 -30 -20 -10 0 S ca tt e r i ng a ng l e (r a d ) -3 · Ni WW Si
Fig. 67:
Distribution of scattering angle of the outgoing pion vs the position of primary vertex along the beamaxis from Primakoff data, illustrating the improvement of the vertex resolution between (left) standard alignmentand (right) run-by-run alignment. The structures correspond to interactions in the different targets used in themeasurement (see Table 3) and in the first Silicon station downstream of the targets.
The calorimeters are aligned with respect to the tracking detectors by using a separate procedure thatassociates charged particle tracks with signals in the calorimeters. Residuals are computed betweenthe expected impact point of the track and the reconstructed shower position in the calorimeter. Thecalorimeter positions are adjusted accordingly in the plane transverse to the beam.he COMPASS Setup for Physics with Hadron Beams 63
The vertex reconstruction uses as input the charged tracks reconstructed in the spectrometer and in thebeam telescope (see Section 9.1). Only two kinds of vertex topologies are considered: primary vertexand secondary vertex. The former designates the association of one beam track with any finite numberof spectrometer tracks, whereas the latter corresponds to a combination between two oppositely chargedtracks with a common origin.Tracks are fed into the vertexing procedure as vectors of parameter estimates and their correspondingcovariance matrices. Only tracks with momentum are accepted. Also, a cut is applied on the differenceof the track times for the incoming and outgoing tracks, except for those reconstructed only in driftdetectors. The contruction of a primary vertx is achieved in an iterative procedure that starts with the set of all trackscompatible with a given beam track and progressively removes outliers using an inverse Kalman filter.This procedure is prone to failing if the initial set of tracks contains a large number of fakes, because thepreliminary estimate of the vertex position may then be too far from the truth. A recovery mechanismis therefore applied in order to reconnect one by one tracks that were unduly discarded. The overallprocedure provides a good vertex-finding efficiency (see Fig. 64).The vertex resolution for the hadron setup is is found to be significantly better than that of the muonsetup described in [1]. The improvement is due to the reduced multiple scattering in the thinner targetsemployed with the hadron beam, as well as to the use of precise silicon microstrip detectors at both endsof the targets. For example, on the 5 π sample already used to evaluate the tracking performances andfor fully reconstructed 5 π final states, the resolution along the beam axis varies from 0 .
75 to 4 . π invariant mass, while the resolution across the beam axis lies in the 13 to 16 µ mrange. The vertex resolution achieved is illustrated in Fig. 67 for a single charged particle final state.Similarly, Fig. 68 shows the system of nuclear targets described in Section 4.2 as reconstructed for athree-particle final state. The sixteen lead and tungsten targets are all clearly separated. Thanks tothe good resolution, the various details of the liquid hydrogen target are distinctly visible in the two-dimensional xy and xz distributions shown in Fig. 69 and Fig. 70.Secondary vertices are reconstructed again using the Kalman filter for any pair of oppositely chargedtracks that satisfy a cut on the minimum distance of approach. Any track can thus be associated withseveral secondary vertices and the primary vertex. Reconstructed neutral particles can then be testedagainst different particle hypotheses and the neutral particles can again be combined with other chargedtracks to study heavier hadrons (see Section 10.1). The 12 PMTs of the inner ring and the 24 PMTs of the outer ring of the RPD (described in Section 4.3)provide information about the integrated charge and a set of time hits. Each possible combination ofupstream and downstream PMTs is used to determine a coordinate along the longitudinal direction of thescintillator and the time at which the particle crossed it. Hits are discarded if their reconstructed positionis outside the fiducial dimensions of the scintillators with a safety margin of 20 cm. Reconstructed hitsfor the inner ring elements are associated to hits in the three corresponding outer ring elements to formtracks. For each track, the momentum is determined from the time of flight using the proton masshypothesis and the calculated position of the hits. The track is extrapolated backwards to the vertexthat is reconstructed using the beam track and the tracks of the scattered particles. A correction on themomentum is determined by accounting for the amount of the material crossed by the recoil particle.After reconstruction, a set of RPD tracks is available for event selection and physics analysis. For each reconstructed vertex, the output comprises its Cartesian coordinates and the list of its associated particles togetherwith their reduced track parameters at the common origin, which encode only directional and momentum information.
Position along beam axis (cm)-70 -60 -50 -40 -30 -20 E n t r i e s ·
250 Pb
250 Pb 250 Pb 250 Pb 250 Pb 250 Pb 125 Pb125 Pb50Pb 25Pb 50W 25W 50Pb25Pb 50W 25W
Fig. 68:
Distribution of reconstructed interaction vertices with three outgoing charged particles along the beamdirection for exclusive events. For each solid state target the thickness is indicated (in µ m). E n t r i e s V e r t e x po s iti on y ( c m ) -4-3-2-101234 Fig. 69:
Vertex distributions for the liquid hydrogen target ( xy projection) for events with three charged tracks. The calibration of the RPD is done using proton-proton elastic scattering events. The impact pointof the scattered proton in the scintillator and its momentum can be predicted from the kinematics ofthis reaction. Matching of measurement and prediction allows for tuning the position offsets on eachindividual counter and the global offset of the RPD position in the COMPASS reference system. Thecorrelation of the predicted longitudinal vertex position and the one determined using the informationfrom the RPD is shown in Fig. 71. The momentum calibration is done by adjusting time offsets betweeneach possible pair of scintillators in ring A and ring B. The energy loss is calibrated using the featuresof the energy loss distribution as a function of the velocity of the proton. The maximum in the energyloss distribution is adjusted to agree with a Monte Carlo simulation (see Fig. 15). For the inner ring, thecorresponding distribution does not show the rising part seen in Fig. 15, hence the maximum energy loss ∆ E is used for calibration.he COMPASS Setup for Physics with Hadron Beams 65 E n t r i e s V e r t e x po s iti on x ( c m ) -4-3-2-101234 LH in/outlet MylarRohacell SI04 UV/XY Fig. 70:
Vertex distributions for the liquid hydrogen target ( xz projection) for events with three charged tracks.For the explanation of the structures, see also Figure 12. For elastic pp scattering, the correlation (difference by 180 ◦ ) between the azimuthal angles φ Spec de-termined from tracking the scattered proton in the spectrometer and φ RP D measured on the recoilingproton with the RPD is shown in Fig. 72. The value of the corresponding resolution, of about 80 mrad,is a consequence of the 24-fold segmentation of the outer ring barrel and the multiple scattering encoun-tered by the recoiling particle in the target. The measured momentum transfer | t | as determined fromforward and from RPD tracks is presented in Fig. 73. A clear correlation between the two measurementsover the covered range of momentum transfer is observed. E n t r i e s Spec z-70 -60 -50 -40 -30 ( c m ) R P D z -70-60-50-40-30 pp fi pp Fig. 71:
Correlation between the longitudinal vertex position z determined with the RPD and the one determinedwith the spectrometer. (rad) p - Spec f - RPD f -1 -0.5 0 0.5 1 E n t r i e s · pp fi pp RMS = 80 mrad
Fig. 72:
Correlation between the azimuthal angles of the recoil proton detected in the RPD and the scattered protondetected in the spectrometer. E n t r i e s /c (GeV Spec t0.1 0.2 0.3 0.4 0.5 ) / c ( G e V R P D t pp fi pp Fig. 73:
Momentum transfer correlation between the recoil proton detected in the RPD and the scattered protondetected in the spectrometer.
The separation between the different hadron types in RICH-1 is illustrated in Fig. 74, where the Cherenkovangles for reconstructed rings are shown as a function of the particle momenta. The four clearly visiblebands correspond to electrons, pions, kaons and protons. For comparison, the same picture for the N radiator (see Section 6.1.2) is shown in Fig. 75. In this case, the Cherenkov angle at saturation reachesonly 24 . F radiator.The particle identification (PID) efficiency was evaluated on samples of pions and kaons from the decayof φ and K S mesons, respectively. The PID relies on an extended maximum-likelihood method. Foreach particle, different likelihood functions corresponding to the relevant mass hypotheses are computedand then compared. The likelihood function parametrises the photon distribution taking into accountboth the photons emitted by the considered particle (the Cherenkov signal) and the photons emittedby other particles in the event (the background). For the background parametrisation, the map of theintegrated hits (see Section 6.1) in the photon detector is used. The PID probabilities (efficiency andmis-identification probabilities) are shown in Fig. 76 as a function of the particle momentum. The effi-ciency is larger than 90% in the region below 30 GeV /c , where the Cherenkov angles for different masshe COMPASS Setup for Physics with Hadron Beams 67 Momentum (GeV/c)10 20 30 40 50 60 C h e r e nkov a ng l e ( m r a d ) E n t r i e s p K p
Fig. 74:
Cherenkov angle for reconstructed rings as a function of the particle momentum for the C F radiator. E n t r i e s C h e r e nkov a ng l e ( m r a d ) p K p
Fig. 75:
Cherenkov angle for reconstructed rings as a function of the particle momentum for the N radiator. hypotheses are well separated. Correspondingly, the mis-identification probabilities are close to zero.Above 30 GeV /c , the Cherenkov angle starts to saturate, and as a consequence the efficiency decreasesand the mis-identification probability increases. Moreover, the high momentum region corresponds tosmall polar angle values, and thus to a region with larger hadron multiplicity in the events. Momentum (GeV/c)5 10 15 20 25 30 35 40 45 P I D p r ob a b ilit y p fi p K fi p Momentum (GeV/c)10 15 20 25 30 35 40 45 P I D p r ob a b ilit y K fi K p fi K Fig. 76:
Identification efficiency and mis-identification probabilities as a function of the particle momentum for(left) a pion sample and (right) a kaon sample.
As detailed in Section 3.3, two CEDARs are used to select the particle type in the hadron beam. Beamparticles can be identified by requiring a minimum number of hits in the eight PMTs attached to each ofthe two detectors. For protons, this method achieves high efficiency and purity due to the good separationof proton and pion rings as discussed in Section 3.3. On the other hand, an online efficiency for kaonidentification of only 35% is obtained using hit multiplicities, which is due to the large beam divergenceand the small difference between kaon and pion ring radii. In the offline analysis this efficiency increasesto 48% for physics events with a vertex in the target.For measurements with negative hadron beams both CEDARs are set on kaon identification, requiringgood kaon efficiency. In order to further improve the kaon identification, a different method was devel-oped for offline analysis. It is based on beam particles reconstructed in the beam telescope before thetarget and makes use of the response of each PMT individually to improve PID for particles that do nottravel parallel to the CEDAR optical axis. In a first step, the response of the PMTs for kaons and pionsis determined as a function of the horizontal and vertical angles between track and CEDAR optical axis, θ x and θ y . These angles are obtained from tracks measured in the beam telescope, which are traced backto the CEDAR position using the known beam optics.A clean kaon sample is obtained from data taken with the CEDAR kaon trigger plus a beam trigger byselecting decays of beam kaon into three charged pions K − → π − π + π − outside the target region. Theprobability for a kaon to produce a signal in one of the photomultipliers is: P ( θ x ,θ y ) ( signal | K ) = Number of beam particles with signals in PMTNumber of beam particles in kaon sample . (4)In order to identify a particle, the probability is needed that a signal is produced by a kaon. This proba-bility can be calculated using Bayes’ theorem: P ( θ x ,θ y ) ( K | signal ) = P ( θ x ,θ y ) ( signal | K ) · P ( θ x ,θ y ) ( K ) P ( θ x ,θ y ) ( signal ) . (5)Here, P ( K ) and P ( signal ) are the probabilities to have a kaon with ( θ x , θ y ) in the beam and to get asignal from any beam particle with ( θ x , θ y ) , respectively. Similar equations hold for a pion sample, whichis obtained using diffractive production of three charged pions on liquid hydrogen, π − p → π − π + π − p .Since the beam divergences for pions and kaons are the same, the probabilities P ( θ x ,θ y ) ( K ) and P ( θ x ,θ y ) ( π ) can be dropped together with the common P ( θ x ,θ y ) ( signal ) . The only quantities needed are the proba-bilities (Eq. (4)) for kaons and pions to produce a signal in a PMT. In order to avoid regions with lowstatistics, a cut ( θ x + θ y ) / <
200 mrad is applied to the data before further analysis. As an example, theprobability distributions P ( signal | π ) for all eight PMTs of CEDAR 2 are shown in Fig. 77. The insetsin the centre of the figure illustrate the position of a pion and a kaon ring relative to the PMT positionsfor θ x = θ y = θ x > θ y = θ x = θ y = θ x are shown in the right inset. The photon ring from the pion illuminates the PMTs on the leftside, thus reducing the kaon identification efficiency.Using the probabilities for all PMTs, the log-likelihood for a beam particle being a kaon is calculatedaccording to:log L ( K ) = (cid:88) signal log P ( θ x ,θ y ) ( signal | K ) + (cid:88) no signal log (cid:2) − P ( θ x ,θ y ) ( signal | K ) (cid:3) , (6)he COMPASS Setup for Physics with Hadron Beams 69 -200 -100 0 100 200-200-1000100200 00.20.40.60.81-200 -100 0 100 200-200-1000100200 00.20.40.60.81-200 -100 0 100 200-200-1000100200 00.20.40.60.81-200 -100 0 100 200-200-1000100200 00.20.40.60.81-200 -100 0 100 200-200-1000100200 00.20.40.60.81 -200 -100 0 100 200-200-1000100200 00.20.40.60.81 -200 -100 0 100 200-200-1000100200 00.20.40.60.81-200 -100 0 100 200-200-1000100200 00.20.40.60.81 θ x θ y Fig. 77:
Dependence of P ( signal | π ) on θ x (horizontal) and θ y (vertical) for the eight PMTs of CEDAR 2 (arrangedaccording to the CEDAR geometry). The range for both angles is from − µ rad to 250 µ rad. The insets in thecentre illustrate the position of a pion (dashed, red) and a kaon (green) ring relative to the PMT positions for θ x = θ y = θ x > θ y = where the first sum only counts photomultipliers with a signal and the second sum only those without asignal. A corresponding equation holds for the log-likelihood for a beam particle to be a pion. Figure 78shows the distribution of log L ( K ) vs. log L ( π ) for (a) the kaon and (b) the pion sample, while (c) showsthe results for an unbiased beam sample. The intensity in (c) reflects the beam composition, namely thatthe kaon component is nearly two orders of magnitude smaller than the pion component.Kaons and pions are identified requiring a certain difference between log L ( K ) and log L ( π ) . The par-ticle is identified as kaon If log L ( K ) > log L ( π ) + A , and as a pion if log L ( π ) > log L ( K ) + B . In allother cases no PID is given. The likelihood differences A and B are chosen by maximising purity andefficiency simultaneously. A good balance between high efficiency and high purity is achieved for the0 The COMPASS Collaboration Table 7:
Efficiencies and purities for the likelihood method ( A = B =
1) in comparison with the multiplicitymethod. Only statistical errors are given.
Kaon efficiency Kaon purityMultiplicity method ( . ± . ) % ( . ± . ) %Likelihood method ( . ± . ) % ( . ± . ) %choice of A = B = E n t r i e s p log L(-20 -15 -10 -5 0 l og L ( K ) -20-15-10-50 (a) kaon sample E n t r i e s p log L(-20 -15 -10 -5 0 l og L ( K ) -20-15-10-50 (b) pion sample E n t r i e s p log L(-20 -15 -10 -5 0 l og L ( K ) -20-15-10-50 (c) beam sample Fig. 78:
Values for the log-likelihoods function for different samples obtained from CEDAR 2 calculated for (a) thekaon sample, (b) the pion sample and (c) an unbiased beam sample. The red line indicates log L ( π ) = log L ( K ) . In order to determine the purity of the CEDAR identification, the reactions π − p → K − K S p and K − p → π − ¯ K S p are used. Due to conservation of strangeness, the incoming hadron is tagged by the outgoinghadron. The K S and ¯ K S are reconstructed using the two-pion invariant mass distribution. The negativelycharged outgoing particle is identified using RICH-1 information. After selecting incoming kaons withthe CEDAR, its purity is determined by the ratio of identified pions in RICH-1 divided by the totalnumber of identified particles. Thus the kaon purity p ( K ) is given by p ( K ) = N RICH ( π ) N RICH ( K ) + N RICH ( π ) . (7)The purity for pions is obtained in the same way.In order to determine the efficiency for pions and kaons, their numbers as obtained from the CEDARsare divided by the respective numbers of pions and kaons assuming the known beam decomposition (seeSection 3.1). The values for kaon efficiency and purity are given in Table 7.The kaon identification efficiency is improved by almost a factor of two for the likelihood method in com-parison with the multiplicity method when applied offline, while the corresponding purities are nearlyhe COMPASS Setup for Physics with Hadron Beams 71identical. With the multiplicity method pion identification is not possible as the pressure was adjustedfor kaon identification for the data taken with the negative hadron beam. The likelihood method allowsfor pion identification as well. The values obtained for pions are similar to those obtained for kaons.In the analysis of Primakoff data, the CEDAR information is needed for an efficient kaon rejection. Usingan optimisation of the likelihood method method described above, pions are identified with an efficiencyhigher than 95%, while the kaon component is suppressed by more than a factor of 20. Event reconstruction in ECAL1 and ECAL2 is performed by using time and signal amplitude informa-tion as directly extracted from the SADC samples. The signal amplitude for each module is convertedinto energy applying conversion coefficients that were derived from the electron beam calibration. Thevariation of the amplitudes over the data taking period is accounted for by using the information providedby the Laser and LED monitoring systems. Details about signal extraction, electron calibration, and datamonitoring are given in Section 6.2.The energy calibration of each module is further improved by using the data derived from an analysisof the π → γγ decay process. The π calibration is performed prior to the final data reconstruction ona fraction of the collected events. The two decay photons are singled out after having defined clustersof deposited energy and performed fits based on the definition of a shower profile. During the finaldata analysis, additional corrections are applied according to the specific data set, namely diffractivedissociation or Primakoff scattering. For both ECALs, the event reconstruction consists of associating an energy deposit in one or severaladjacent modules to a single incident particle. A set of energy deposits that is assumed to originatefrom a single particle is called in the following a shower ; the full energy deposit and hit position of theparticle are calculated from it. In many cases, two or more showers overlap and form a cluster . Thusprecise knowledge of the shower profile facilitates the separation of overlapping showers. In addition,it improves spatial and energy resolutions and limits the impact of inefficient or noisy cells. Clusters oftwo or more particles can result from electromagnetic showers initiated in the material upstream of theECAL or from decay photons that hit the ECAL at a distance smaller than the lateral shower size.The data analysis procedure starts by defining a cluster of neighbouring modules, in which the depositedcharges are larger than a pre-defined threshold (see Section 9.7.3). The cluster is then split into showersemploying a parametrisation for the lateral spread of the shower profile [62]. The shower parametersfor the lead glass and Shashlik modules are determined using electrons from a dedicated calibrationbeam (see Section 3.2). For Primakoff data taking in which mainly high-energy photons are detected, noelectron beam with the corresponding energy is available. The shower parameters are therefore derivedusing single photons from real data events.
The shower profiles used in the reconstruction are based on an empirical cumulative function, as definedin [62]. If the energy deposited by a shower is projected onto a transverse axis with the shower center at0, the fraction of the total shower energy accumulated between −∞ and a position x on this axis can bedescribed by: F ( x ) = + π (cid:88) i a i · arctan xb i . (8)2 The COMPASS CollaborationIn addition to providing a good description of this ratio, Eq. 8 is conveniently related to the energydeposited in each module [62]. Up to three contributing shower components (denoted by the index i )are summed up, with parameters a i and b i describing the relative weight and width of each component,respectively.In order to obtain the shower profile parameters, the following procedure is applied. A column-wisecalculation of the ratio of the energy accumulated so far over the total energy of clusters, taking thesimple centre-of-gravity as the central position, yields a distribution that is fitted with the cumulativefunction Eq. 8 describing the shower. This is illustrated in Fig. 79 (left), which represents the fractionof the total energy deposited up to a particular column at a given distance from the shower center. Thefraction of the total energy deposited in a column as a function of its distance from the shower center isshown in Fig. 79 (right). Distance from center (mm)-80 -60 -40 -20 0 20 40 60 80 R a ti o R a ti o Fig. 79:
Shower reconstruction: (left) fraction of total shower energy collected from −∞ up to a particular distancefrom the shower center (Eq. 8); (right) fraction of the total energy deposited in a column as a function of its distancefrom the shower center. This concept can be extended to two dimensions. In this case, the ratio between the accumulated energyup to a point ( x , y ) and the total energy of the shower is given by: F ( x, y ) = + π (cid:88) i a i · arctan xb i + arctan yb i + arctan x · yb i (cid:113) b i + x + y . (9)The first two terms account for the ratio along x and y projections, while the third term adds an asym-metry along the diagonal.Different sets of parameters are used for each module type. In ECAL1 the same profile with three con-tributions is used for all modules. The parameters were obtained for the GAMS-2000 spectrometer [62].Since COMPASS makes use of the same lead glass modules, the parameters are unchanged.In ECAL2, the profiles of both lead glass and radiation-hardened lead glass modules are described by twocontributions, with parameters derived from the electron beam calibration. For the Shashlik modules, theprofile consists of three contributions. The corresponding parameters are obtained from Primakoff dataevents that contain a single high-energy cluster. In both cases it has to be assumed that a cluster containsonly one shower.The two-dimensional ratio defined in Eq. 9 is used to calculate the relative energy deposited by a showerat the position ( u j , v j ) in each module: G j ( x, y ) = F ( u j + ∆ , v j + ∆ ) − F ( u j + ∆ , v j − ∆ ) he COMPASS Setup for Physics with Hadron Beams 73 − F ( u j − ∆ , v j + ∆ ) + F ( u j − ∆ , v j − ∆ ) . (10)Here u j = X j − x and v j = Y j − y are local coordinates relative to the centre ( X j , Y j ) of each module, j denotes an index over all modules and ∆ is half the transverse size of a module. In a first step, signals from adjacent modules are combined to form a cluster. Starting from the firstmodule not yet used in the cluster, each of the (vertical, horizontal, and diagonal) neighbours is checkedfor a measured energy above a threshold of 100 MeV for ECAL1, and of 200 MeV for ECAL2. For eachnew module added to the cluster, its neighbours are treated the same way.In a second step, a fit of the shower profiles to the cluster data is performed. The fit improves thespatial resolution of the calorimeter and separates overlapping showers. The fit is first done with a singleshower. Further showers are added one by one with a new fit being performed after each added shower.The parameters of the first shower are initialised to those of the module with the highest energy in thecluster and its neighbours. The energy is set to the sum of the energies of those modules, the position andtime are set to the mean of the respective information weighted with the energy of each of those modules.When adding more showers, the module searched is the one with the largest relative discrepancy betweenits measured energy and the energy predicted to be deposited by all showers fitted to the data so far. Thecentre of the new shower candidate must be located at a distance larger than √ / e i at impact point ( x i , y i ) in a moduleat position ( X j , Y j ) can be calculated as: E pred j,i = e i · G j ( x i , y i ) . (11)As several showers might be fitted into the same cluster, the total energy of all showers in the module isgiven by: E pred j = (cid:88) i E pred j,i = (cid:88) i e i · G j ( x i , y i ) . (12)In addition to the energy, the time information is also used. The time t i of a shower is defined as themean value of the times of all modules contributing to the shower, weighted with the energy depositedin each module. Similarly to the predicted energy deposit in a block, the predicted time is calculated as: T pred j = (cid:88) i e i · G j ( x i , y i ) (cid:80) i e i · G j ( x i , y i ) · t i (cid:80) i e i · G j ( x i , y i ) . (13)The predicted energy and time are compared to the measured energy E meas j and time T meas j in eachmodule. The Minuit fitter from the ROOT package [58] is used to optimise the shower parameters tomaximise the likelihood: − log L = (cid:88) j (cid:16) E meas j − E pred j (cid:17) σ j,E + (cid:16) T meas j − T pred j (cid:17) σ j,T . (14)4 The COMPASS CollaborationThe errors on the measured energies σ j,E and times σ j,T (Fig. 49) are calculated from an energy de-pendent parametrisation that has been determined from data beforehand. With this procedure the fitdistinguishes between in-time showers and pile-up events.The procedure of trying to add a new shower is stopped if the fit describes the data well or if the maximumnumber of showers in a cluster has been reached. The decision, whether the last shower added improvesthe fit, is based on a comparison of the log-likelihood normalised to the number of degrees of freedom.It is also checked that all showers have energies above the energy threshold and that they fulfill therequirements on the distance between two showers described above.This fitting procedure returns the energy e i , the position ( x i , y i ) and the time t i of each shower fitted intoa cluster.The number of modules contributing to the total shower energy reaches 5 × Cluster size0 5 10 15 20 25 30 35 40 E n t r i e s · Showers per cluster0 1 2 3 4 5 6 7 E n t r i e s · Fig. 80:
ECAL2 fit results for (left) number of modules per cluster and (right) number of fitted showers per cluster.
For ECAL1 a simplified version of the procedure described above is employed. For the larger MAINZand OLGA modules, an improvement of the performance by using shower profiles was not found. Eachmodule with a deposit larger than that of any of its neighbours is used as a starting point for a new shower.Energy and position of this module are used to initialise the parameters of the shower. If larger than theenergy threshold, the information contained in the neighbouring modules is then used to improve theposition of the shower by calculating its centre of gravity. π → γγ decays The π calibration procedure is performed, prior to the final analysis, using a fraction (equivalent to 1 to 2days of data taking) of the physics events. The reconstruction of the incident and outgoing particle tracksis required, with a definition of a primary interaction vertex. Only showers with energies E γ larger than1 GeV for ECAL1 and 3 GeV for ECAL2 are taken into account. Showers associated with charged tracksare discarded. In order to minimise combinatorial background, only events with less than 5 showers areused.The two-photon invariant mass M γγ is calculated for every pair of showers, assuming that both pho-tons originate from the interaction vertex. Only pairs with invariant masses within ±
50 MeV /c aroundthe nominal π mass, M π , are considered as valid π candidates. For each of two showers, a two-dimensional histogram E γ vs ( M γγ − M π ) is filled, which is associated with the shower’s central mod-ule. The central module is defined as the module which contains the highest fraction of the depositedhe COMPASS Setup for Physics with Hadron Beams 75 E n t r i e s (MeV/c p - M gg M-40 -20 0 20 40 [ G e V ] g E E n t r i e s (MeV/c p - M gg M-40 -20 0 20 40 [ G e V ] g E Fig. 81:
Energy deposition in two ECAL2 modules as a function of the difference between reconstructed andnominal π mass for (left) a module with typical behaviour and (right) a module with an unusual behaviour. energy. The values stored in the two-dimensional histograms are used as a starting point for the calibra-tion procedure.For most ECAL modules a slight energy dependence is observed, as illustrated in Fig. 81 (left). However,various types of unusual behaviours may also be present, e.g. as the one shown in Fig. 81 (right), whichresuls from a saturated photomultiplier tube. The variations are accounted for by introducing correctionfactors that depend on the photon energy. The correction factors are calculated in energy slices of 2 GeV.In each slice the spectrum is fitted with a Gaussian for the π peak and with a first order polynomial forthe background, in an interval of ±
20 MeV /c around the π peak. The results of the fits are displayedin Fig. 81 as black crosses; its horizontal and vertical tick marks represent the 3 σ fit error and the binsize, respectively. The fitted mass differences are then used to calculate the correction factor α i for eachenergy slice i , α i = ( + ∆ M i M π ) , (15)where ∆ M i is the fitted mass offset. The correction factor for each module is calculated assuming thatthe energy of the second decay photon is measured precisely. Since this is not the case, the π calibrationis done iteratively, each iteration adding corrections to the result from the previous iteration. Typically,after 8 to 10 iterations the procedure converges. The result is a significant improvement of the π massresolution and of the π mass offset, as shown in Fig. 82. After calibration, the mean value of thepeak position shifts from 9 . /c to 0 . /c . The mass resolution improves from 7 . /c to4 . /c .The calibration significantly improves the response of the individual ECAL2 modules, as illustrated inFig. 83. A similar improvement is observed for ECAL1. For most modules, the reconstructed pionmass after calibration agrees within less than 1 MeV /c with the nominal π mass for π energies up to160 GeV.The resulting calibration is used to correct the individual module responses during the event reconstruc-tion procedure. The precise value of the correction factor α , which corresponds to the actual energydeposited in each module, is determined by interpolation. Additional corrections, which are evaluated independently for the different data sets, are applied on topof the shower fit result. For the diffractive dissociation data, the reconstructed energy E as measured ina Shashlik module for an electron is compared with the corresponding charged track momentum p . Aposition dependence of the ratio p/E is then observed, as indicated in Fig. 84. This dependence reflects6 The COMPASS Collaboration ) M (MeV/c D -60 -40 -20 0 20 40 60 E n t r i e s · = 7.6 MeV/c s ) M (MeV/c D -60 -40 -20 0 20 40 60 E n t r i e s · = 4.6 MeV/c s = 4.6 MeV/c s Fig. 82:
Difference ∆ M between reconstructed and nominal π masses in ECAL2 for (left) before calibration and(right) after calibration. slight inefficiencies in the vicinity of the four central rods. It is accounted for by using the hit position asdetermined by the shower fit.The photons detected in the calorimeters cover an energy domain that extends from less than 1 GeV forECAL1 to more than 120 GeV for ECAL2. The measured cluster times for both ECAL1 and ECAL2show a slight energy dependence, mainly for low photon energies. This dependence, which is alwayssmaller than 1 ns, is fitted to the data and accounted for. For photon energies above 80 GeV, the kinematics of the Primakoff-Compton reaction, π − + ( A, Z ) → π − + γ + ( A, Z ) , constrain the detected photons to the central 4 × π calibration alone this cannot be achieved. For higher energies a different methodis applied. During the Primakoff data-taking period, muon beam data for systematic studies are periodi-cally collected. These data also contain Primakoff-Compton events with photon energies nearly as highas the beam energy. Moreover, since the Beam Momentum Station (BMS) is present in the beam line, themuon incident momentum is known. Since the scattered muon momentum is also measured, the energy x (cm)-100 -50 0 50 100 y ( c m ) -100-50050100 ) M ( M e V / c D -10-50510 x (cm)-100 -50 0 50 100 y ( c m ) -100-50050100 ) M ( M e V / c D -10-50510 Fig. 83:
Difference between reconstructed and nominal π masses as a function of the impact position for theECAL2 modules for (left) before π calibration and (right) after π calibration. The difference is calculated usingthe mean value of the fitted (with a Gaussian) X-projection of E γ vs ( M γγ − M π ) histograms. The grey rows atthe top and bottom ends and on the right side of ECAL2 are located beyond the angular acceptance for photonscoming from the target (see Section 6.3.) he COMPASS Setup for Physics with Hadron Beams 77 R a ti o y ( c m ) -1.5-1-0.500.511.5 Fig. 84:
Ratio of track momentum over calorimeter energy as a function of the impact position in a Shashlikmodule relative to its centre. The four central spots with a ratio larger than one correspond to the four module rods. E n t r i e s E n e r gy b a l a n ce ( G e V ) -20-15-10-50510 Fig. 85:
Difference between the beam energy and the total measured energy as a function of the photon energy. conservation (exclusivity) in the process provides an independent prediction of the energy of the emittedphoton. The comparison with the actual ECAL2 measurement based on the π mass calibration exhibitsa slightly falling slope as a function of the photon energy, as shown in Fig. 85.This trend does not only depend on the photon energy but also on the actual hit position within theShashlik modules. The observed dependence is fitted with a three-dimensional function that includesboth intra-cell coordinates and the shower energy. The correction reaches values of up to + −
12 GeV, as shown in Fig. 86. The major part of this correction is due to the steel rods that tie the Shash-lik stack together, as previously explained in Section 9.7.5. During data taking with a pion beam, theBMS is removed and no measurement of the incident pion momentum can be performed. It is assumedthat the corrections to the ECAL2 calibration for muon and pion beams are identical. This assumption issupported by the data itself; after applying the above intra-cell corrections, both the position and the stan-dard deviation of the exclusivity peak with a pion beam improve, similarly to the improvement achievedwith a muon beam. Fig. 87 illustrates the effect of these corrections on the central modules of ECAL2.8 The COMPASS Collaboration E n t r i e s E n e r gy b a l a n ce ( G e V ) -20-15-10-50510 Fig. 86:
Intra-cell energy variation as a function of the distance to the cell centre.
Energy difference (GeV)-30 -20 -10 0 10 20 30 E n t r i e s · =3.7 GeV RMS =3.3 GeV RMSMuon beam
Energy difference (GeV)-30 -20 -10 0 10 20 30 E n t r i e s · =4.0 GeV RMS =3.9 GeV RMSPion beam
Fig. 87:
Difference between beam and measured energies (energy balance) for Primakoff-Compton scattering (left)with a muon beam and (right) with a pion beam. The distributions are displayed with the standard π calibrationonly (dashed curve) and with linearity and intra-cell position corrections (solid curve); the corresponding RMS and RMS values are indicated. The efficiency for the reconstruction of single photons is defined as the fraction of photons that originatefrom the target reconstructed in one of the calorimeters. It thus includes effects of geometric acceptancefor photons like dead material, and the intrinsic performance of the calorimeters related to thresholds,etc. The reconstruction efficiency is evaluated by a Monte-Carlo procedure using diffractive events forthe π − π π channel (see Section 10.1) with 0 . < t (cid:48) < . /c . The thresholds applied correspondto those in the analysis of physics data (0 . . ∆ y/ ∆ z = ∆ x/ ∆ z = . ∆ y/ ∆ z ≈ ± .
02, the shadowof HCAL1 and SM2 on ECAL2 is visible as a horizontal line, while the vertical lines at ∆ x/ ∆ z ≈ ± . Photon energy (GeV)0 20 40 60 80 100 120 140 160 180 E ff i c i e n c y ( % ) E ff i c i e n c y ( % ) D X/ D -0.15 -0.1 -0.05 0 0.05 0.1 0.15 Z D Y / D -0.1-0.0500.050.1 Fig. 88:
Simulated photon efficiency (left) as a function of the photon energy and (right) as a function of the photondirection in the laboratory system.
10 Monte Carlo simulation and performance of the setup
The interpretation of physics processes that involve hadron beams and several particles in the final-stateparticles requires a thorough understanding of the experimental setup. This requirement can only beachieved through a realistic simulation of the apparatus and a detailed knowledge of its acceptance asfunction of any of the kinematic variables that are relevant in a particular physics process. The Monte-Carlo code used to describe the setup and to determine its acceptance is described below. It is followedby a selection of characteristic experimental results, for each of the two beam polarities and for vari-ous particles in the final state. All results were obtained with the nominal hadron beam momentum of190 GeV /c . The simulation of the COMPASS setup is performed using a dedicated Monte-Carlo (MC) code calledCOMGEANT. The code can be linked to external event generators specific to the reaction mechanismthat is dominant in a given channel. Final-state particles are then propagated through the setup. Thedigitisation of the MC data and the subsequent reconstruction are carried out with the same software thatis used for reconstructing the measured events.Three different event generators are used to simulate diffractive reactions, central production, and Pri-makoff reactions. The partial-wave analysis method employed for diffractively and centrally produced n -body final states takes into account the acceptance of the apparatus using Monte-Carlo pseudo data,where the final-state hadrons are distributed isotropically in the n -body phase space. In addition to the de-cay phase space, the generators also simulate the production kinematics. Diffractive events are generatedwith a t (cid:48) distribution that is tuned to the data. The central-production generator [63] simulates exponen-tial t (cid:48) distributions for both beam and target vertices. For Primakoff-Compton scattering, the generatorcalculates differential cross sections with contributions due to polarisability, first-order Compton vertexcorrections, and soft photon emission [64].For all generators, the beam phase space spanned by the positions and angles of the incoming particlesis generated using parametrizations extracted from real data. Since the incident energy is not measured,it is reconstructed from the kinematics of fully exclusive events. Interactions in the target volume aredistributed in the target material according to the target positions. Primary interactions of the beamparticles in materials and detectors surrounding the target are not generated.Scattered and secondary particles are propagated through the spectrometer by the simulation code COMGEANT,based on GEANT 3.21 [65]. Multiple scattering, energy loss, shower development, and secondary inter-actions are taken into account. This includes interactions of electrons and photons with detector materialand creation of electromagnetic showers by these particles. Additional physics processes like hadron0 The COMPASS Collaborationinteractions and in-flight decays are also taken into account. Furthermore, pile-up events due to two ormore particles occurring in the same time window can be generated.The digitisation of the simulated events is performed in CORAL (see Section 9.1). Dead and activematerials along the tracks are accounted for with the ROOT geometry package [59]. Charged-particletracks are reconstructed from the simulated hits in the tracking detectors using the same procedure as forthe real data (see Section 9.1). Detector properties such as efficiencies and resolutions are implementedin the reconstruction software using information from the experimental data. For RICH-1, the purityand efficiency of the detector are determined from the measured events and separately unfolded from thesimulated data. For the electromagnetic calorimeters, shower profiles are extracted from the ECAL MCdata, and a π calibration is performed as for real data (see Section 9.7). The RPD information is passeddirectly to the PHAST physics analysis software. Position along beam axis (cm)-90 -80 -70 -60 -50 -40 -30 -20 -10 0 S ca tt e r i ng a ng l e (r a d ) -3 · Ni Fig. 89:
Monte Carlo simulation of the Primakoff-Compton reaction, showing the reconstructed position of theprimary vertex along the beam direction as a function of the scattering angle of the outgoing pion. Note thatinteractions outside the target material are not simulated.
An example of the good MC description achieved for the Primakoff-Compton reaction is shown in Fig. 89for the reconstructed primary vertex as a function of the pion scattering angle. A qualitative comparisonwith Fig. 67 (see Section 9.3) shows that the agreement between MC and data is good. Backgroundeffects, e.g. from interactions with the detectors downstream of the target, are minimized by applyingselection cuts identical to those used for the specific physics process.Other observables relevant for the Primakoff-Compton reaction are discussed in Section 10.2. The mo-mentum distributions of the electromagnetic component for pion and muon interactions with a solid tar-get are well reproduced by the MC simulations (see Fig. 97). The π − π + π − decay of the K − mesons inthe beam, a process used for flux normalisation, is accurately simulated as illustrated by the momentumtransfer distributions in Fig. 96.The MC simulation of the photon reconstruction efficiency for channels with final-state photons is val-idated by comparing the acceptance-corrected particle decay yields for different decay channels. Forexample, the resulting branching ratios of ω → π − π + π and ω → π γ agree within 5% with the PDGvalues.The acceptance of the apparatus is determined by comparing the reconstructed and generated MC events.In the Partial Wave Analysis (PWA) formalism employed for diffractive scattering, the full multidimen-sional acceptance for a given final state is used. For example, in three-body analyses (like π − π + π − )for fixed four-momentum transfer and three-body mass the acceptance depends on five kinematic vari-ables. The acceptance determined by the simulated phase-space events is then used as an input to the fitsperformed in the PWA formalism.he COMPASS Setup for Physics with Hadron Beams 81 ) /c (GeV - p + p - p M0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.2 2.4 A cce p t a n ce ( % ) p - p + p - p fi p - p GJ q cos -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 A cce p t a n ce ( % ) p - p + p - p fi p - p Fig. 90:
Acceptance for the diffractively produced π − π + π − final state (left) as a function of the 3 π invariant massand (right) as a function of the polar angle of the π + π − isobar in the Gottfried-Jackson frame. For the purpose of illustration, the acceptance for a particular variable can be determined by projectingthe multi-dimensional acceptance onto this variable. In Fig. 90, the acceptance for the π − π + π − finalstate of diffractive dissociation is evaluated in the t (cid:48) range between 0.1 and 1 GeV /c . It is a fairly flatfunction of the invariant three-pion mass from near threshold up to 2 . /c and of the polar angle ofthe π + π − isobar in the Gottfried-Jackson frame (see definition in Section 1).More pronounced modulations of the acceptance are observed for channels where one or several final-state particles are identified by the RICH-1 detector. The impact of the particle identification on theacceptance in the K − π + π − channel is depicted in Fig. 91 for both the Kππ invariant mass distributionand the Gottfried-Jackson angle of the π + K − isobar. The reduction of the acceptance is mainly due tothe limited momentum range available for kaon identification (see Section 9.5).Figure 92 shows the corresponding acceptance plots for the π − π π final state. Compared to the charged-pion channel, the acceptance for the channel containing neutral pions is smaller and its dependence onthe three-pion mass is more pronounced. The decrease of the acceptance is mainly due to the photondetection efficiency, which is lower than that for charged particles as photons may get absorbed in pas-sive materials before reaching the calorimeters. The largest absorption is caused by the beam pipe ofthe RICH-1 detector (see Section 6.1), an effect which is mainly important for forward-going photons.Nevertheless, the detection of four photons smears the effect, so that the angular modulation of the ac-ceptance remains weak and similar to that of the charged-pion case. ) (GeV/c - p + p - K M1 1.2 1.4 1.6 1.8 2 2.2 2.4 A cce p t a n ce ( % ) p - p + p - K fi p - K ) p (K GJ Q cos-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 A cce p t a n ce ( % ) p - p + p - K fi p - K Fig. 91:
Acceptance for the diffractively produced K − π + π − final state (left) as a function of the Kππ invariantmass, and (right) as a function of the polar angle of the π + K − isobar in the Gottfried-Jackson frame. ) (GeV/c p p - p M0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.2 2.4 A cce p t a n ce ( % ) p p p - p fi p - p GJ q cos -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 A cce p t a n ce ( % ) p p p - p fi p - p Fig. 92:
Acceptance for the diffractively produced π − π π final state (left) as a function of the 3 π invariant massand (right) as a function of the polar angle of the π π isobar in the Gottfried-Jackson frame. Energy balance (GeV)-20 -15 -10 -5 0 5 E n t r i e s · p - p + p - p fi p - p = 1.9 GeV s Energy balance (GeV)-20 -15 -10 -5 0 5 E n t r i e s · Ni g - p fi Ni - p data simulation = 2.3 GeV s Fig. 93:
Energy balance between outgoing and incoming particles for (left) diffractive dissociation with threecharged pions in the final state and (right) for Primakoff scattering.
The performances of the individual detectors and of the reconstruction software were presented in theprevious sections. Here, the main characteristics of the setup are presented with examples from variousphysics processes.
The selection of exclusive events with a primary vertex inside the target is a prerequisite for most analysesperformed on the data with hadron beams. Exclusive events are selected by requiring energy conservationand transverse momentum balance between incoming and outgoing particles. Figure 93 shows the distri-butions of the difference between the energy of the outgoing particles and of the incoming beam particle,for the diffractive dissociation into three charged pions and for the Primakoff pion Compton scattering.The beam momentum station, which is used to determine the momentum of each incoming beam particlewhen operating the muon beam, is removed for hadron beams in order to reduce the amount of materialin the beam. Therefore, the beam energy is set to the value determined by the beam line settings. In thediffractive process, the energy of the outgoing pions is combined with the recoil proton energy measuredby the RPD. In the Primakoff reaction, the incident pion energy is shared between the scattered pionand the emitted photon, whereby the contribution of the target recoil remains negligible. The widths ofthe energy balance distributions shown in Fig. 93 are dominated by the momentum spread of the beamparticles (see Section 3.1), with a smaller contribution from the finite momentum resolution for chargedparticles and also from the finite energy resolution for photons in the case of the Primakoff reaction.he COMPASS Setup for Physics with Hadron Beams 83 ) /c t' (GeV0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 E n t r i e s p - p + p - p fi p - p Fig. 94:
Squared four-momentum transfer for π − π + π − events produced by a pion beam impingingon a liquid hydrogen target, and selected by the DT0trigger. ) /c t' (GeV0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 E n t r i e s Pb - p + p - p fi Pb - p Fig. 95:
Squared four-momentum transfer for π − π + π − events produced by pions hitting a lead tar-get, and selected by the multiplicity trigger. The physics processes studied in COMPASS can be identified via their characteristic dependence onthe reduced squared four-momentum transfer t (cid:48) , which is calculated from the four-momenta of the in-coming beam particle and the outgoing particles according to the equations given in Section 1. WhilePrimakoff reactions proceeding through the exchange of quasi-real photons dominate the cross sectionat t (cid:48) < .
001 GeV /c , diffractive and central production reactions prevail at larger values of t (cid:48) . Here weshow measured t (cid:48) distributions for the different physics triggers mentioned in Section 7.6.Figure 94 shows the t (cid:48) distribution recorded with the diffractive trigger DT0, determined from eventswith three charged pions in the final state. The cut at t (cid:48) ≈ .
07 GeV /c is due to the requirement of asignal in the RPD. The DT0 trigger thus enhances events with high t (cid:48) . The small leakage of events with t (cid:48) < .
07 GeV /c presumably originates from δ electrons or pions accidentally firing the RPD.The multiplicity triggers LT1 and LT2 are used to also include events with lower values of t (cid:48) , since norecoil proton is required. The corresponding t (cid:48) distribution measured with a pion beam and solid nucleartargets (Section 4.2), shown in Fig. 95, exhibits an exponential increase of the number of events towardslow values of t (cid:48) . In the first few bins also Primakoff events contribute, in addition to those generated bystrong interaction.A good resolution on the measurement of t (cid:48) is important in order to distinguish between Primakoff anddiffractive scattering. The resolution at very small values of t (cid:48) is determined using the decay of beamkaons into two or three charged pions. For free-particle decays, t (cid:48) is by definition zero, and the measuredwidth of the t (cid:48) distribution, shown in Fig. 96, gives a direct estimate of the resolution. A width of3 . · − GeV /c is obtained from the data, in good agreement with the resolution from Monte Carlosimulations. At higher values of t (cid:48) , the resolution can only be determined from Monte Carlo simulations.A value of 7 · − GeV /c is obtained for 0 . /c < t (cid:48) < . /c , from the simulation ofdiffractive production of three charged pions.For the measurement of the pion polarisability, exclusive π − γ events are selected from the data samplecollected with the calorimeter trigger (see Section 7.5). The left panel of Figure 97 shows the distributionof the four-momentum transfer | Q | = √ t (cid:48) , chosen here to emphasize its shape at small values. The peakat | Q | ≈ .
02 GeV /c mainly contains quasi-real photoproduction events. The fact that the interaction ispurely electromagnetic at very low values of t (cid:48) , which correspond to large impact parameters, becomesclear when comparing it to the right panel of Figure 97 that shows the corresponding distribution takenunder the same conditions, but with a µ − beam instead of a π − beam. For the pion beam, the stronginteraction dominates at | Q | values above 0 .
05 GeV /c , resulting in typical diffractive structures. TheMonte Carlo simulation, superimposed as solid line in both panels, describes both cases very well.4 The COMPASS Collaboration ) /c t' (GeV0 0.5 1 1.5 2 2.5 3 -3 · E n t r i e s · - p + p - p fi - K data simulation
Fig. 96:
Squared four-momentum transfer of reconstructed beam kaons (data points) compared to the MonteCarlo simulation of purely electromagnetic interaction (solid lines). The dashed line is an exponential fit, used todetermine the resolution. |Q| [GeV/c]0 0.05 0.1 0.15 0.2 0.25 0.3 E n t r i e s · Ni g - p fi Ni - p data simulation |Q| [GeV/c]0 0.05 0.1 0.15 0.2 0.25 0.3 E n t r i e s · Ni g - m fi Ni - m data simulation Fig. 97:
Momentum transfer distributions for exclusive (left) π − γ and (right) µ − γ events. The data (dotted lines)are compared to the MC simulation (solid lines). The mass resolution of the spectrometer is determined using known narrow states that are reconstructedin the spectrometer via their decay into neutral and/or charged particles. Here we show distributions fortwo-photon decays ( π , η ), for decays into final states with charged particles only ( K S , φ , Λ , Ξ ), and fordecays into final states containing both charged and neutral particles ( η , ω , η (cid:48) , f ).The invariant mass distributions of photon pairs in the π and η mass regions, as measured by ECAL1 andECAL2, are shown in Fig. 98. The distributions are obtained from diffractive interactions of a 190 GeVnegative hadron beam impinging on a liquid hydrogen target. Apart from the standard event selectionmentioned in Section 10.2.1, only clusters with energies larger than 0 . . . /c and 3 . /c are obtained for the two-photon π peaks detected in ECAL1 and in ECAL2, respectively. The corresponding values for the η meson are 19 . /c and 11 . /c .Hadrons reconstructed from decay modes that contain only charged particles are shown in Fig. 99. In aninclusive selection, the following resolutions are obtained: 5 .
90 MeV /c for the K S , 1 . /c for the φ ( ) , 1 .
99 MeV /c for the Λ and ¯ Λ and 2 .
80 MeV /c for the Ξ ± .Complex resonance decays with more than three particles in the final state are reconstructed e.g. bycombining a π or η in the γγ channel with a neutral pair of pions ( π + π − ) leaving the primary vertex.The left panel of Fig. 100 shows the invariant mass spectrum of the π − π + π final state in the ω ( ) masshe COMPASS Setup for Physics with Hadron Beams 85 ) (GeV/c gg M0.1 0.11 0.12 0.13 0.14 0.15 0.16 0.17 E n t r i e s · ) = 3.9 MeV/c p ( s ) (GeV/c gg M0.48 0.5 0.52 0.54 0.56 0.58 0.6 0.62 E n t r i e s · ) = 11.6 MeV/c h ( s Fig. 98:
Two-photon invariant mass distribution as measured in ECAL2, in the (left) π mass region and (right) η mass region. The solid curves are fits to the signal and to the background. The values of the resolution achievedare indicated in each plot. ) (GeV/c - p + p M0.44 0.46 0.48 0.5 0.52 0.54 0.56 0.58 E n t r i e s · ) = 5.9 MeV/c s0 (K s ) (GeV/c - K + K M0.98 1 1.02 1.04 1.06 1.08 1.1 1.12 E n t r i e s · p - K + p K fi p p ) = 1.9 MeV/c f ( s ) (GeV/c p p M1.10 1.11 1.12 1.13 E n t r i e s · ) = 2.0 MeV/c L ( s (GeV) pL M1.26 1.28 1.3 1.32 1.34 1.36 1.38 1.4 E n t r i e s · ) = 2.8 MeV/c – X ( s Fig. 99:
Reconstructed invariant masses for charged particles in the final state. The peaks shown are for (top left) K S ( ) , (top right) φ ( ) , (bottom left) Λ ( ) , and (bottom right) Ξ ± . The K S , Λ , and Ξ ± particles areproduced in inclusive reactions. The dashed curve in the φ ( ) plot is a fit to the background. ) (GeV/c p - p + p M0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1 E v e n t s · p p - p + p p fi p p ) = 6.6 MeV/c w ( s ) (GeV/c h + p - p M0.9 1.0 1.1 1.2 1.3 1.4 E n t r i e s · p h + p - p - p fi p - p ') = 6.1 MeV/c h ( s Fig. 100:
Invariant mass spectra for (left) π − π + π and (right) π − π + η systems. The full line in the left panel is afit to the ω peak only; the dashed line includes also the background. E n t r i e s /c (GeV - p + p M0 0.5 1 1.5 2 2.5 3 ) / c ( G e V - p + p M p - p + p - p fi p - p Fig. 101:
Dalitz plot for three diffractively produced charged pions after a cut of ±
130 MeV /c around the π ( ) mass. region from central production reactions of a proton beam with the liquid hydrogen target. As shownin the right panel of Fig. 100, selecting the η instead of the π gives access to the decays η (cid:48) ( ) → π − π + η and f ( ) → π − π + η , which are reconstructed from diffractive π − p → π − π + π − pγγ events.Deconvoluting the natural width of the ω , a resolution of the spectrometer of 6 . /c is obtained. Thenatural width of the η (cid:48) is negligible, so the width of the peak directly gives a resolution of 6 . /c inthis mass range.Three-body decays of short-lived resonances with correspondingly larger widths can be studied in Dalitzplots or by using the technique of PWA. A high-statistics Dalitz plot for the π − π + π − final state (5 · events) is depicted in Fig. 101, where the invariant mass of the 3 π system was required to be within ±
130 MeV /c around the nominal mass of the π ( ) resonance. The bands correspond to the decays π ( ) → ρπ and π ( ) → f ( ) π .he COMPASS Setup for Physics with Hadron Beams 87
11 Summary
In this paper, a detailed description of the COMPASS experimental setup as used for the physics pro-gramme with hadron beams is given. Operational since 2002, the setup was designed for both hadronstructure and hadron spectroscopy studies. It makes use of the various beams available at the CERNM2 beam line, namely positive and negative hadrons, positive and negative muons, and electrons. Theapparatus operates with beams in the energy range of 100 to 200 GeV and is able to detect charged andneutral hadrons in the final state. Two large-aperture dipole magnets provide wide angular and momen-tum acceptances. While the major part of the setup remains essentially unchanged, its target region isreconfigurable as a function of the specific experimental programme.For several years, the COMPASS setup was successfully used with muon beams and with a large-sizepolarised target ,or spin structure studies. After a week of exploratory data taking period in 2004, animportant part of the COMPASS hadron programme was conducted in 2008 and 2009. Over the years,several new components were added to the setup, according to the requirements of the hadron physicsprogramme and also to improve the overall performance of the apparatus.Immediately upstream of the COMPASS setup, two CEDAR detectors were installed into the M2 beamline. Based on the Cherenkov effect, the CEDARs identify the hadron beam particle, separating kaons,pions and protons. A new target system, consisting of either a solid-state target holder or a liquid hy-drogen target was built. A Recoil Proton Detector, surrounding the target, provides access to exclusivemeasurements. An accurate vertex resolution was achieved by adding nitrogen-cooled Silicon microstripdetectors upstream and downstream of the target.Several new PixelGEM detectors were positioned along the setup for particle tracking at very smallangles. Modified Micromegas detectors were used for tracking immediately downstream of the target inthe presence of high hadron fluxes. Two additional large-size drift chamber detectors were also installedin order to improve the detection at large polar angles.Both charged and neutral particle identifications were considerably improved. An important upgrade ofthe RICH-1 detector was carried out, resulting in higher efficiency and increased rate capability. TheECAL1 calorimeter was completed and added to the setup, while ECAL2 was modified to withstand thehigh flux in the case of hadron beams.The main part of the trigger system was rebuilt for use with hadron beams. Several new trigger and vetoelements such as recoil proton detector, multiplicity counter, and sandwich veto were added, therebyoptimizing the system for diffractive scattering. A new digital calorimeter trigger was developed forselecting Primakoff reactions. The data acquisition system was further tuned in order to stand hightrigger rates with low dead time. The detector control system was adapted to include the new detectorsand upgraded with new monitoring features.All new detectors were successfully included in the full software analysis chain. The tracking, recon-struction, simulation, and analysis tools were updated and adapted to the use with hadron beams. Theacceptance of the apparatus covers large angular and momentum ranges and is nearly uniform for allkinematical variables. The overall characteristics of the setup illustrate its important potential for hadronspectroscopy studies. Invariant masses of up to 3 GeV /c are covered with statistical accuracies sig-nificantly better than in previous experiments. The good energy resolutions achieved allow access toa large number of meson and baryon resonances. In summary, the upgraded COMPASS setup is fullyoperational for use with the various hadron beams available at CERN.8 The COMPASS Collaboration Acknowledgements
We gratefully acknowledge the CERN laboratory and the CERN BE, EN, IT, TE and PH departmentsfor providing constant and efficient support during the upgrade phase of our experimental setup andduring data taking. We express our gratitude to the numerous engineers and technicians from our homeinstitutions, who have contributed to the construction and later to the maintenance of our detectors andequipment. We are also grateful to A. Altingün for help in the preparation of the numerous figures.We acknowledge support from MEYS Grants ME492 and LA242 (Czech Republic), CEA (France),Bundesministerium für Bildung und Forschung, DFG cluster of excellence “Origin and Structure of theUniverse” and DFG Research Training Group Programme 1102 “Physics at Hadron Accelerators” (Ger-many), CERN-RFBR Grants 08-02-91009 and 12-02-91500, Israel Science Foundation, founded by theIsrael Academy of Sciences and Humanities (Israel), INFN and MIUR (Italy), MEXT and JSPS GrantsNos. 18002006, 20540299 and 18540281, Daiko Foundation and Yamada Foundation (Japan), SAIL(CSR) (Government of India), NCN Grant DEC-2011/01/M/ST2/02350 (Poland), Fundaç¯ao para a Ciên-cia e Tecnologia, COMPETE and QREN, Grants CERN/FP/109323/2009, CERN/FP/116376/2010 andCERN/FP/123600/2011 (Portugal) and from European Union FP7 (HadronPhysics3, Grant Agreementnumber 283286).
References [1] COMPASS Collaboration, P. Abbon, et al., Nucl. Instr. and Meth. A 577 (2007) 455.[2] COMPASS Collaboration, M. Alekseev, et al., Phys. Rev. Lett. 104 (2010) 241803.[3] COMPASS Collaboration, C. Adolph, et al., Phys.Rev.Lett. 108 (2012) 192001.[4] P. Abbon, et al., Nucl. Instr. and Meth. A 567 (2006) 114.[5] H. Atherton, et al., Precise measurements of particle production by 400 GeV/c protons on berylliumtargets, CERN Yellow Report, CERN 80-07 (1980).[6] C. Bovet, et al., The CEDAR counters for particle identification in the SPS secondary beams, CERNYellow Report, CERN 82-13 (1982).[7] GAMS NA-12/2 Collaboration, D. Alde, et al., Nucl. Instr. and Meth. A 342 (1994) 389.[8] T. Alimova, et al., IFVE-86-35 .[9] J. Bernhard, Aufbau des inneren Rings eines Recoildetektors am COMPASS Experiment, p. 36-38,Master’s thesis, Johannes-Gutenberg Universität Mainz (2007).[10] T. Schlüter, et al., Nucl. Instr. and Meth. A 654 (2011) 219.[11] M. J. French, et al., Nucl. Instr. and Meth. A 466 (2001) 359.[12] MUSCADE(r), µ π − pb → π − π − π ++