1.4. Planteamiento del problema
2.2.2. Memoria como apuesta para el proceso de escritura
To calculate the rate of a specific trigger in a data-driven manner2
, 2
Since the simulation of the trigger was not yet working at the time it was not possible to do the rate estimation from MC.
one needs to consider a dataset that does not contain any trigger bias. This can for example be achieved by using only random trig- gers. However, since the cross-section of physically interesting event topologies is many orders of magnitude smaller than the total cross- section, an unreasonably large dataset would be required. To pre- vent this, an enhanced bias dataset is taken, in which the num- ber of physically relevant topologies3
are overrepresented. Events 3
For example the number of events containing one or more high-pT lep- tons.
are selected using L1 triggers of different signatures and thresholds and are stored without running any HLT algorithms4
. The trigger 4
The enhanced-bias data taking is based on five trigger chains that are seeded by 10–30 L1 items, includ- ing one random chain that selects events at random. The collection of enhanced-bias data is done while regu- lar physics data is taken, which means L1 prescales have to be taken into account. Enhanced-bias data is col- lected at a rate of ≈ 300 Hz and one collection run lasts approximately one hour. Thus, about1million events are recorded to base rate estimations on. menu for enhanced bias datasets is designed to allow to derive event-
weights, which restore the desired zero-bias spectrum for rate predic- tions [336]. The specific beam parameters of the collider (like optics or pileup) are reflected in the enhanced-bias data and a new dataset has to be recorded should the parameters change.
Three types of weights need to be considered for every event: a weight wEB which corrects the effect of the trigger selection on the dataset and restores the zero bias spectrum, a weight wC that ac- counts for the trigger chain and includes possible prescale correc- tions and lastly a weight wL, which is a weight to extrapolate the
event rate from the instantaneous luminosity at recording of the data to a desired reference luminosity. The rateRof a specific trigger can now be estimated through
R= ∑
N
e=1wEB(e)wC(e)wL(e)
∆t , (6.1)
where the sum runs over allNrecorded eventse, and∆tis the length of data collection.
Undergoing this calculation one arrives at an expected rate of the L1 muon trigger with threshold pT = 10 GeV of (224.6±1.3) kHz and at a rate of(8.38±0.04)kHz for a trigger onEmissT >50 GeV for luminosityL=2.0×1034cm−2s−15
. In a first step the expected rate 5
The quoted rates are based on enhanced-bias data taken in Septem- ber2016with2208colliding bunches in the machine. 1.5 million events were recorded.
of the late-muon trigger is estimated by neglecting the late compo- nent and the muon is treated as being in-time. The combined rate, which could be seen as an upper bound, is calculated as random coincidence between both components according to
Rcomb.=
pmuon·pEmiss
T ·b
wherepmuonand pEmiss
T are the probability of the muon orE miss T trig- ger firing in an event, b is the number of bunches in the machine andT=88.924µs the revolution time of a proton bunch around the
accelerator ring. Assuming a full ring, one arrives at a combined rate of
Rcomb. = (96.4±0.7)Hz (6.3) hinting at a very low rate when considering the total L1bandwidth of 100 kHz6
.
6
The main limitation for the late-muon trigger is unfortunately not the L1 bandwidth, but the computing power needed for the track fit done on HLT level. A low L1rate can thus neverthe- less cause problems later on if the CPU costs for the HLT are too high.
The effect of the delayed muon on the trigger rate cannot be as- sessed directly from data because of the dead time after any L1A was issued. Instead, the bunch structure within the collider can be ex- ploited. Recall the LHC filling scheme described in chapter3.1.2. In between the proton bunch trains there are empty buckets for which no collisions occur within ATLAS. Neglecting the influence of pos- sible signal7
the trigger only fires if a muon is wrongly attributed
7
which would be small in any case and
can therefore safely be ignored. to the following BC. To estimate how frequent this happens, those BCs which immediately follow the end of a bunch train, i.e. the first empty buckets, are considered. A special trigger menu is used to record events from empty BCs. One long 2016 data run corre- sponding to 277 pb−1 of integrated luminosity is considered here for a rough estimate. The filling scheme delivers 46 empty buckets following a filled one. An empty BC trigger with a muon thresh- old of 4 GeV is used8
. With the total time of data recording known
8
The implemented variety of trigger thresholds is limited. The motivation is to estimate upper bounds to the rate of the late-muon trigger by choosing a trigger with a low threshold.
(27556 s), one can estimate the rate of a trigger with>50 GeVEmissT in the first BC and a delayed muon withpT >4 GeV in the following BC to be
Rcomb.,delayed=120.03 Hz (6.4)
confirming the low expected rate from the previous estimate. Three incarnations of the late-muon trigger have been activated in mid-2017for data-taking. A 10 GeV delayed muon is combined with either a 50 GeV jet or withETmiss of either 40 or 50 GeV. The actual rate of the late-muon trigger in configuration as for the rate estimate above was measured in October2017to be
Rtruth=44.39 Hz (6.5)
and is therefore in good agreement with the derived upper rate lim- its.
6.4
Expected efficiency
0.2 0.4 0.6 0.8 1 1.2 β 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 Efficiency MC simulationFigure 6.4: Efficiency of the L1 late- muon triggers as function of the parti- cleβ.
The expected efficiency has been estimated for gluinoR-hadron sam- ples. Figure6.4shows the efficiency of the L1late-muon trigger as function ofβof the particle firing the trigger. The efficiency is highest
for which the single-muon triggers are not efficient anymore (com- pare to figure6.1). The loss of efficiency for particles with β < 0.5 is due to arrivals at the MS trigger stations with delays larger than1 BC. Figure6.5shows the efficiency of the jet-based late-muon trigger and the lower-threshold Emiss
T -based trigger as function of the sim- ulated gluino masses. The efficiency is dictated by the underlying
βspectrum. The increase in signal efficiency is ranging from 4% at
mg˜=1200 GeV to 2% atmg˜=3000 GeV. For stau and chargino sam- ples the efficiency is expected to be much higher since no neutral particles are produced (except for ˜χ±1χ˜01 production) and because a
larger portion of the βspectrum is located inside the region of max-
imal efficiency of the late-muon trigger since the possible expected
signal masses are smaller. 500 1000 1500 2000g~ mass [GeV]2500 3000
0 0.05 0.1 0.15 0.2 0.25 Trigger efficiency -based trigger miss T E jet-based trigger MC simulation
Figure 6.5: Efficiency of the L1 late- muon triggers as function of the sim- ulated gluino mass.
6.5
Summary
Single-muon triggers are inefficient for slow particles since arrival delays cause the trigger signal to be shifted to one of the follow- ing bunch crossings. The late-muon trigger is a fundamentally new trigger that aims to recover efficiency for slow particles by consid- ering two consecutive events. The target signature expects a soft jet with 50 GeV transverse momentum or missing-transverse energy of 40 GeV in the first bunch crossing and a muon in the immediately following one. The muon acts as the slow-particle candidate that ar- rives with delay at the trigger chambers in the Muon Spectrometer.
The trigger thresholds were chosen to allow for a largest possi- ble signal efficiency while still retaining a manageable data-taking rate. To this end the signal efficiency for various thresholds has been evaluated. The expected trigger rate was estimated in enhanced-bias data assuming random coincidence between trigger elements while for the late component the first empty bucket after a filled bunch train in the collider was used. The late-muon trigger performs within expectation and extends the coverage of muon triggers toβ>0.5.