The Physics of ALICE HLT Trigger Modes R. Bramm, T. Kollegger, C. Loizides and R. Stock, bramm,kollegge,loizides,stock @ikf.uni-frankfurt.de, { } Physics Department, University of Frankfurt/Main 3 0 0 Abstract 2 n We discuss different physics cases, mainly of the ALICE TPC, such a J as pile-up, jets in pp and PbPb, Bottonium and Charmonium spec- 7 troscopy, and there corresponding demands on the ALICE High Level 1 Trigger (HLT) System. We show that compression and filter strategies 2 can reduce the data volume by factors of 5 to 10. By reconstructing v 0 (sub)events with the HLT, background events can be rejected with a 5 0 factor of up to 100 while keeping the signal (low cross-section probes). 2 Altogether the HLT improves the discussed physics capabilities of AL- 1 2 ICE by a factor of 5-100 in terms of statistics. 0 / x e 1 Physics Motivation - p e h ThePhysics of ALICEcan bebroadly groupedinto two types ofphysics : v observables, which connect to the soft and hard sectors of QCD, respec- i X tively. Among the former group one finds the physics of bulk hadron r a production, multihadron correlations and interferometry, fluctuations of short and long ranges, resonance and string decay signals, as well as multistrange hyperons. Mostly these observables refer to aspects of nonperturbative QCD even if their initialization makes contact to the primordial partonic ”transport” dynamics in which (at √s = 5.5 TeV per participant nucleon pair in PbPb collisions) perturbative QCD obvious plays a role, too. Pertubative QCD (pQCD) aspects domi- nate, however, in the second group of observables consisting of various aspects of jet and heavy flavour production. It is the main point of such studies to connect the initial dynamics during interpenetration of the interacting primordial hadronic ground state matter (supposed to be well understood from modern pQCD application to elementary pp collisions) to the ensuing parton cascade era. In that era the initial pQCD ”seedlings” such as bb pairs or very energetic quarks and gluons 1 may act as ”tracers” of a dynamical phase that may approach the de- confined quark-gluon state envisaged in non perturbative Lattice QCD theory, before it hadronizes. The resulting final signals, e.g. jets, open charm and quarkonia should thus exhibit certain characteristic atten- uation properties (suppressions, enhancements, quenching etc.) that might serve as diagnostic tools for the intermediate, non perturbative QCDera of near equilibrium partonic matter. Itis clear fromthe above that all such observables also require systematic study of the more ele- mentary pp collisions at LHC energy, in order to establish the base line physics - a further research focus of the ALICE experiment. Clearly there is a coincidence of the ”mostly soft” and ”predomi- nantly hard” QCD physics sectors of ALICE with relatively large, and small cross sections, respectively. It turns out that analysis with the former type of physics should require relatively modest event statistics, of a few 106 PbPb and about 108 pp collisions, whereas the systematic analysis of hard signals calls for an additional one or two orders of mag- nitude both in PbPb and pp. For example, inclusive production of jets with total transversal energy of up to 200 GeV, or of the weaker states in the bottonium family, is expected to occur (within the ALICE track- ing acceptance) about once every 104 to 105 central PbPb collisions. It is the latter sector of ALICE physics which will be addressed below. Obviously some kind of on-line higher level trigger selectivity is being called for, in order not to be drowned in excessive demands placed both on DAQ bandwidth and off-line data handling format. More specifically, ALICE differs from the other LHC experiments by its insistence on microscopic track by track analysis of the events, both in the central barrel (tracking from the ITS through the TPC and the TRD into the outer TOF layer and the photon array) and in the dimuon spectrometer. Calorimeters with a more summative output are absent in this basic design although an augmentation, by an electro- magnetic calorimeter, remains as an upgrade option of special interest for jet spectroscopy. The consequence of the insistence on single track resolution leads to a raw data flux (after on-line zero suppression and Huffman compression in the digital front end electronics) amounting to about 40-80 Megabytes in a central PbPb collision, which presents us with up to about 20.000 charged tracks per event. Clearly, high statis- 2 tics demands as associated with the above small cross section signals meet with excessive raw data volume calling for higher on-line selectiv- ity based on track pattern recognition. The required on-line selectivity can be accomplished with relatively modest computing and connectivity power, using an High Level Trigger (HLT) complex of about 1000 CPUs equipped with fast interconnectiv- ity. A major part of the required processor power can be implemented already in the local data concentrator circuits (LDC) of the DAQ front end instrumentation that is devoted to the raw data flux from the AL- ICE barrel TPC detector which produces, by almost an order of mag- nitude, the predominating data input on the ALICE DAQ system. We shall illustrate below how the rarest physics signals (cross-section wise) provide for the most topologically distinct tracking signatures in the TPCsubdetector-recognizablewithrelatively modestCPU-networking effort. We thus describe the physics cases motivating on-line HLT trig- ger schemes based on on-line tracking of the ALICE TPC output. Fur- ther refinement may result from calling in early time information of the ITS and TRD systems. The dimuon-forward spectrometer works out- side the acceptance of the barrel detectors. Its HLT requires separate consideration but is of minor format as compared to the TPC HLT. The following sections describe the raw outline of an HLT system that will cope with an on-line degradation of a TPC event rate of 200 Hz for central PbPb collisions. There are three basic modes to such an HLT functionality. It may either generate selected candidate events accord- ing to the detected track topologies corresponding to certain specific trigger tasks. We will focus on jet spectroscopy at first, with a back- ground suppression efficiency of between 50 and 100: thus a 200 Hz TPC event rate will generate 2 to 4 candidate events per second, which are written onto tape in full raw data format. Alternatively, the HLT system might reduce the full TPC sector readout to region-of-interest readout. E.g. reading out 2 sectors only (out of 36) in conjunction with aTRD-generated pre-candidate for anear back-to-back e+e− track pair resulting from bottonium decay will lead to an 18-fold compression of data flux. Finally, in later years of ALICE operation one may consider the option of employing the TPC HLT circuitry only insofar as on-line cluster identification is concerned, which then would be recorded by 3 DAQ instead of the full raw data arrays leading to an overall TPC data compactification byabout10to20. Thelattermodewillbeemployedin the next years by the RHIC experiment STAR,which resembles ALICE in layout. In summary there are three HLT modes: selective trigger, region ofinterestreadout(filter)andglobalTPCrawdatacompression. We have constructed and tested an HLT on-line tracking prototype at the RHIC STAR experiment (called L3 there) with 54 processors confronting the STAR TPC data format, which is about 10% of the expected event size for ALICE. Both the selective trigger and overall compactification (by a factor of 6 in this prototype case) modes work well. In particular, the on-line momentum and specific ionization reso- lutions of that device operating at a maximum rate of 50 Hz for central collisions of Au+Au at RHIC was shown to approach the off-line reso- lutions within about 20%. In later sections we shall extrapolate from this L3 to the ALICE HLT to obtain a guide line concerning ALICE HLT processing power demands. In the following sections we shall at first specify the ALICE running conditions in terms of LHC luminosity with special emphasis on the TPC event rate, which is the slowest ALICE subdetector but creates, by far, the highest raw data flux. However, it also creates conditions of several superimposed events captured in one TPC output frame. That condition can be dealt with by selection of ”clean events” in PbPb collisions (by lower level triggers) but only by an HLT on-line filter procedure in pp collisions at higher luminosity. This will lead to the proposal that data taking with inclusion of the TPC (there are other optionsinALICE)canproceedat200HzincentralPbPb,andat1KHz in pp. With these options we arrive at an optimum TPC operation vis a vis the LHC luminosities. Both options would create an event rate that is necessary for jet and bottoniumspectroscopy,butcannotbewrittendirectlytotape,because of the resulting raw data flux of about 10 to 15 Gbyte per second from the TPC alone. Further sections thus deal with the required expected HLT speed and selectivity in various modes. For illustration we shall mostly focus on high transverse energy jet spectroscopy, both in PbPb and in pp collisions. A consideration of other HLT trigger modes ends these sections of the document. 4 1.1 Running Conditions of ALICE At the design luminosities concerning the ALICE experiment we en- counter high interaction rates which are fully satisfactory for all observ- ablesof concern. However 8KHzminimumbiasPbPbrateas wellaspp at 140 KHz would both create roughly 0.5 Terrabytes of potential raw data per second which we can not write to mass storage. This estimate already takes into account on-line zero suppression in all its high gran- ularity subdetectors. Any kind of on-line intelligence is thus welcome to 1. maintain event rates appropriate for low cross section physics sig- nals (charm and bottom spectroscopy, high E jets), which range T down to one in 107 pp or to one in 104 PbPb events and to 2. reduce the bandwidths of DAQ processing and writing to tape, and the magnitude of data deposited and handled in mass stor- age. This is important as neither the taping speed has increased significantly (in a fashion resembling other component’s ”Moore’s law” pattern), nor has the tape and robot media cost shown a marked decrease. One is eager, therefore, to keep these cost fac- tors withinreasonable bounds,whileincreasing theoverall physics content of the raw data flow. In view of point 2 it would be desirable to cut down to say 1.0 GB/s taping rate, i.e. to about 0.5% of the raw data rate correspond- ing to the luminosity limits. In order to achieve this we must employ any suitable method of front-end/DAQ intelligence, in particular with the data-intensive subdetectors TPC, TRD and Dimuon spectrometer. Such methods start with zero suppression and automatic, loss free data compactification (Huffmanetc.), thelatterexpectedtoreachacompres- sion by about two. These methods will not concern us here. Instead we concentrate on high level trigger- and filter-procedures (HLT) which are based on on-line tracking in large processor/switch/storage arrays placed at the front-end of the DAQ. We will comment on the physics driving such HLT-efforts, primarily, because we need first to assess the potential physics benefit, in order to conclude about the appropriate 5 efforts in HLT and DAQ. The first topics that we choose to illustrate the expected HLT will be 1. Jet physics in pp (”first year” and beyond) 2. Jet physics in PbPb 3. Y spectroscopy in TRD and TPC, and 4. Event filter procedures for TPC/ITS open charm analysis Some of these topics concern genuine triggers enabling us to scan up- ward of 200 Hz of PbPb and up to one KHz of pp, without exceeding a reasonable DAQ writing budget. In other cases the filter aspect domi- nates: recording the accessible physics with lower data volumes. 1.1.1 Technical Assumptions At average luminosity of 1027 the minimum bias rate (assuming 8 barn total cross section) for PbPb is 8 KHz, of which we may label 1 KHz as ”central”. The bunch spacing is 100 ns, thus we get 1.2 10−3 events · per bunch crossing, i.e. very rarely we indeed get two events within the crossingtimeofabout100ps. ForKrKrat5 1028 luminosityand4barn · the min. bias rate is 240 KHz, and the fraction of events unresolvable to our electronics rises to the 1 % level. In pp we will have full LHC energy at first of √s=14 TeV (unlike in PbPb). We assume a luminosity of 2 1030 here, which would lead · to 140 KHz minimum bias rate for a 70 mbarn non-diffractive cross section. The above luminosity is already non-trivially low for the LHC overall running conditions (in combination with the maximal luminosi- ties employed in the other experiments it would be simpler to run at far higher L). A lower luminosity for ALICE will be accomplished ei- ∗ ther by a different β or by de-tuning (or both). One concludes that we can not safely assume that the 2 σ size of the interaction domain (the ”diamond”) will initially be as low as 10 cm. Furthermore we will argue below that a further reduction of pp luminosity to the 1029 level is highly desirable: a challenge to LHC technique. 1.1.2 TPC Event Pileup and Rate At the above event rates the TPC (90 µs drift time) has a significant double event fraction within the drift time already for PbPb which gets 6 forbiddingly high for KrKr and CaCa at higher potential luminosities (no TPC physics possible). In pp at L=2 1030 we will have many · (but smaller) events in each frame, displaced by the drift velocity of 2.8 cm/µs. More precisely the average fraction of PbPb double events in the TPC at 8 KHz minimum bias rate is [1-exp(-2τ f)] = 0.76 drift · where we have taken 2τ because each TPC frame contains displaced tracks/events occurring both 90 µs before and after the trigger. This effect is specific to the TPC as all other subdetectors have drift - or integration times of up to 5 µs only. The ”clean” minimum bias PbPb event rate thus shrinks to 1.900 Hz, the central rate to 240 Hz as far as the TPC is concerned; both at L (average)=1027. This limitation of useful event rate roughly coincides with two other TPC limitations, the maximum possible TPC gating frequency (now estimated to be about 1 KHz), and the maximum data transfer rate from the TPC front-end electronics to the HLT/DAQ complex as implied by the 216 DDL links, whose individual bandwidth is assumed here to amount to 150 MB/s. Estimating a ”clean” central PbPb event to contain 66 MB after zero suppressionandaflatdistributionofdatatrafficoveralltheTPC-DDLs we would get a maximum event transmission rate of 400/s. However, we already get into double events with about 100 MB event size this way, and, furthermore, the traffic distribution may not be flat over all DDLs. Thus we take about 200 Hz as a, perhaps preliminary, technical limit for ”clean” central PbPb collisions and about 400 Hz for min. bias for any ALICE data taking mode involving the TPC. The HLT system will thus be presented in a version that copes with the corresponding raw data input of up to about 15-20 GB per second in PbPb collisions. 1.1.3 TPC Pileup in PP In pp running at L=2 1030 and √s = 14 TeV the TPC physics gets · quite challenging [3]. At 70 mb we get 140 KHz min. bias event rate. From prior and post 90 µs to the trigger 25 events fall into one TPC frame which are on average half-complete. Average spacing in time is about 7 µs but note that the events are not therefore ordered in drift distance because of the variation of the primary vertex position which will randomize distances, perhaps even inverting the ordering. This influences the capability to identify the sub-event that belongs to the 7 trigger, by means of on-line tracking in the HLT leading to an output to DAQ that contains the relevant sub-event only (HLT filter mode). The TPC data flux per event arising in a situation with 25 half- complete min. bias pp collisions at √s = 14 TeV is estimated to be about 4.5 MB; making allowance for a certain non-track-density-related noise (e.g. electronics) we conservatively estimate 5-6 MB, i.e. roughly 8% of a central PbPb event. ThemaximumestimatedTPCgatingfrequencyof1KHzwouldthus result in a front-end data output rate of about 5.5 GB/s, equivalent to the output of 85 Hz worth of central PbPb. From the above consider- ations this is well compatible with the DDL and HLT bandwidths but can not be written to tape. An HLT facility is thus required to filter out the relevant information contained in each piled-up TPC frame. 1.1.4 Summary of ALICE Running Conditions Consideration of luminosity, event pileup conditions, maximum TPC gating rate, as well as data rate capability of the TPC front-end elec- tronics including the DDL bandwidth leads to an estimate of maximum event frequencies as far as the TPC is concerned: 1 KHz for minimum bias pp collisions 400 Hz for pileup-free PbPb min. bias collisions 200 Hz for pileup-free PbPb central collisions The other ALICE subdetectors can run along with these event frequen- cies, but can not accept significantly higher event rates anyhow - per- haps going up to 1.5 KHz for min. bias and central PbPb collisions in the ITS, TRD and Dimuon detectors, thus capable of defining detector specific level 3 pre-triggers for a pre-scaling that involves the TPC at its lower appropriate rates. However, 1 KHz pp collisions yield about 5.5 GB/s front-end flow 400 Hz min. bias yield about 15 GB/s front-end flow 200 Hz central PbPb yield about15-20 GB/s front-end flow 8 from the TPC alone. We shall show below that the intended ALICE physics requires such rates, which we can not write to tape. This con- sideration makes the case for an intelligent processor front-end system attached to theDAQ, deriving specifictriggers and/or reducingthe raw event sizes by appropriate filtering procedures: the High Level Trigger (HLT) system. In the following we will first of all consider the task of jet spectroscopy in pp and PbPb. These physics observables will serve to illustrate, both the minimal (pp)and maximal (PbPb) requirements, placed on the HLT functions. A short sketch of other physics observ- ables concludes these considerations. 1.2 Jet Physics in pp and PbPb Collisions High transverse energy jet production at LHC energy is of interest both from the point of view of higher order QCD (e.g. twist and gluon saturation) [6] and from the program of quantifying jet attenuation in extended partonic matter, i.e. in nuclear collisions at ALICE. The overall goalofthelatter ideaistodeterminetheQCD”stoppingpower” acting on a colour charge traversing a medium of colour charges, in analogy to the Bethe-Bloch physics of QED. This effect will attenuate the energy of the observed jets. Predictions of perturbative QCD as to the mechanism of energetic parton propagation in hot and cold QCD matter address the borderline of present state of the art theory, thus receiving increasing attention. For a length L traversed in QCD matter the induced radiative energy loss is proportional to L2 and expected to be much higher in a parton plasma than in colder hadronic matter even at moderate plasma temperatures of T 200 MeV. The effect of ≈ ”jet quenching” can be estimated [4] to amount to ∆E 60GeV( L )2 30GeVforL = 7.5fm. − ≈ 10fm ≈ Here, L is the transverse radius of the PbPb ”fire-sausage”, which is the relevant geometrical length scale as we are observing jets with av- erage center of mass angle of 900 with the ALICE TPC positioned at η < 1. Thus the main task is to measure the inclusive jet produc- CM | | tion yield at mid-rapidity as a function of E both in pp and central T PbPb. Of course one is, more specifically, interested in the details of the jet fragmentation function [1]. The capability of ALICE to recon- 9 structindetailthejetfragmentationfunctioninthechargedtracksector is crucially important in view of ongoing theoretical study concerning the detailed consequences of primordial partonic mechanisms of energy loss. It turns out that the hadronization outcome, from the energy loss components of the leading jet parton, may also be partially contained within the typical jet emission cone, resulting in an increase of local E T emission but expressed by relatively softer hadrons. The picture of quantifying energy loss of the primordial jet may thus be an over-simplification. One may have to investigate, instead, the microscopic changes, at track-by-track level, occuring within the fragmentation cone of a high E jet created in PbPb collisions, in order T to appropriately capture the overall attenuation effect acting on a lead- ing parton traversing high energy density partonic matter sections of the interaction ”fireball”. In this view, the jet attenuation phenomena should reside in a characteristic softening of the jet cone fragmentation function, rather than in a simple jet energy loss scenario, as consid- ered above. The microscopic tracking approach of ALICE may thus prove advantageous, in comparison to the summative E inspection by T calorimeters. 1.2.1 HLT Function in PP for Low Cross Section Signals As stated above, we probably will get only √s = 14 TeV initially for the pp beam. For low cross sections signals such as jets, Ω and Y we need about 1010 events, in order to record 5 104 jets with E 100 GeV T · ≥ 105 Ω in acceptance 103 Y e+e− in acceptance. → The ideal solution would be to reduce the luminosity further (below 2 1030), which technically is very difficult. Assuming a luminosity of · 1029 would result in an average of 1.7 min. bias events per TPC frame at 7 kHz event rate. For that scenario an average TPC event content of 0.5 MBatamaximumtrigger rateof1000 Hzwillleadto0.5GB/s to ≈ DAQ. Thisisawritingspeedcompatiblewith”firstyear” DAQ without employing an HLT. If LHC can notgo lower than a luminosity of 2 1030 at ALICEfor tech- · nical reasons, we need to employ HLT functionality. Now the average 10