1. Serge Sushkov ATLAS Event Filter IFAE Barcelona
The ATLAS High Level Trigger, designed for a broad discovery potential at LHC
Serge Sushkov
ATLAS Event Filter
IFAE Barcelona
Outline
LHC physics & trigger
ATLAS Trigger overview
Features of High Level Trigger
Additional functionality
A few words on status
Based on ATLAS
TDAQ overview talks

2. LHC Challenges: Trigger View
process
(pp), nb / rate, Hz
events/year
mini bias
~ 108
~ 1015
pT > 200 GeV
~ 100
~ 100 M
top pairs
0.85
~ 10 M
WW pairs
0.08
~ 1 M
ZZ pairs
0.011
~ 12 k
s = 14 TeV
L = 1034 cm-2 s-1
Bunch spacing = 25 ns  collision rate 40 MHz
Interactions / bunch crossing ~ 20
# readout ATLAS ~ 108
Event size ~ 1.5 MB
Writing to mass storage ~ 300 MB/s  accept rate 200 Hz
Corfu Workshop
S.Sushkov, ATLAS EF, IFAE Barcelona

3. Triggering Physics: “Basic Objects”
The foreseen computing power does not allow for a full
event reconstruction of all LVL1-accepted events
analyse & make decisions on part of event data
layered/stepwise architecture: refine at each next layer
Based on identified “physics object” (distinguished from
mini-bias signals) and thresholds:
type
reconstruction starts from…
resulting objects
jets
cluster of Calo cells above threshold
Jets, missing ET, electrons,
muons, taus, photons with
parameters above thresholds, improved by calibration corrections and cross-detector matching
missing ET
vector sum of E in Calo cells
electrons
narrow “jet” w/o hadronic E, matching track in Inner Tracker
muons
track in Inner Tr. & Muon Ch & E in Calo
taus
narrow “jet” with hadronic E & track (IT)
photons
narrow “jet” w/o track in In. Tr.
Corfu Workshop
S.Sushkov, ATLAS EF, IFAE Barcelona

4. ATLAS Trigger/DAQ: three level architecture
Level-1: Hardware
coarse granularity MU & CALO
simplified signature finding
determines Regions Of Interest
High Level Trigger:
Software-implemented
Level-2:
full detectors granularity
guided by & operates on ROI
fast => simplified analysis
Event Filter:
full event data is available
seeded by LVL2 results
1 s latency => more elaborate
offline reco & analysis tools
additional: calibr’n & monit’g
Corfu Workshop
S.Sushkov, ATLAS EF, IFAE Barcelona

5. Physics Channels & Trigger Elements
Identified & recon’d
“physics objects”
with certain thresholds:
Trigger Elements (TE)
Label: #<obj>[thresh]<iso>
Combination of TEs
corresponding to
physics channels:
Trigger Signatures
Set of signatures:
Trigger Menus
Requirements:
be as inclusive as
possible to include
unknown new physics
not to bias accepted
data samples
Corfu Workshop
S.Sushkov, ATLAS EF, IFAE Barcelona

6. HLT Concept: Steering Mechanism
HLT works using two types of algorithms:
Feature Extraction: find/identify “physics objects” (Trigger Elements) using detector-specific reco algorithms
Hypothesis Validation: requirements on TEs (obj type & thresholds)
Steering mechanism:
Repeat/refine TE reconstruction & Hypothesis Validations in iterative steps
next step starts only if requirements at previous step are satisfied
then next step is seeded by results of previous step and refines the event selection
Steering is organized by:
Trigger Element (TE): #<obj>[thresh]<iso>
Trigger Signature: combination of TEs
Trigger Menu: list of signatures for which an event
can be accepted at a given step
Sequence Table: specifies collection of algorithms to be run
on a given input TE (and specifies resulting TE)
Corfu Workshop
S.Sushkov, ATLAS EF, IFAE Barcelona

7. Example of Steering Mechanism: Z  e+e-
Iso
lation
pt>
15GeV
Cluster
shape
track
finding
EM15i +
e15i
e15 e
ecand
Signature 
STEP1
STEP 4
STEP 3
STEP2
t i m e
Advantages of Steering:
rejection of “bad” events occurs as
early as possible, thus only “good
physics events” reach full analysis
allows to use only ROI
LVL2, thus only “good” events go
through event building network and
are built as full events
each next step continues and refines
reconstr’n using results of previous
step (optimal use of time/resources)
LVL1 seed 
Corfu Workshop
S.Sushkov, ATLAS EF, IFAE Barcelona

8. Illustration of reco & analysis in HLT
Take the LVL1 EM RoI seed.
Use a LVL2 clustering algorithm for EM showers and compute shape variables:
Shower containment in sampling 2 (E3x7/E7x7).
Look for additional maximum in sampling 1 (E1-E2)/(E1+E2).
EM and Hadronic Energy.
Perform selection cuts on these variables.
Reconstruct Tracks in the Inner Detector inside the RoI.
Perform track/cluster matching criteria.
Use the refined RoI to seed the EF (better calibration constants, complete
event available, …) p0 g
Corfu Workshop
S.Sushkov, ATLAS EF, IFAE Barcelona

9. HLT Selection: based on Offline Algorithms
HLTSSW
Steering
ROBData
Collector
Data
Manager
HLT
Algorithms
Processing
Application
EventData
Model
Offline
EventDataModel
Reconstruction
StoreGate
Athena/
Gaudi
Event Filter
Level2
HLT Core Software
Offline Architecture & Core SW
Offline Reconstruction
HLT Algorithms
HLT Selection Software
HLT DataFlow Software
TDAQ DataFlow Software
Corfu Workshop
S.Sushkov, ATLAS EF, IFAE Barcelona

10. Features of using Offline algorithms in HLT
Advantages & design aims:
base on / use the same detector-specific offline reco/analysis algorithms both in offline data analysis & in online triggering (no code duplications)
selection/trigger algorithms can be developed in offline framework, without necessity to run TDAQ SW (emulating TDAQ in offline)
independent development works on TDAQ/Online Infrastructure
(emulating selection algorithms)
HLT (EF) has access to all external tools/lib of Offline world:
HLT can potentially run/use ANY Offline algorithm as additional functionality (calibration, physics & performance monitoring)
Consequences for HLT & Offline:
special interfaces between HLT & Offline components
different implementations of data access in HLT & offline
time/memory performance requirements for used Offline algorithms
versions compatibility issues: HLT SW should match both Offline and TDAQ/Online SW versions
Corfu Workshop
S.Sushkov, ATLAS EF, IFAE Barcelona

11. Sets of trigger algorithms
The work on trigger algorithms is argonized as:
grouped into “families” according to basic types of “physics objects”: photons, electrons, muons, taus, jets, missing ET
LVL2 & corresponding EF steering/trigger algorithms are combined into chains, called “vertical slices”, which are used
for efficiency & performance studies
within each “family/slice”, particular working groups optimise performance of selection/trigger algorithms, basing on
physics analyses for typical physics processes
cross-detector matching & calibration corrections are used for improving precision of reco parameters before applying thresholds
results of trigger reco/selection will be available to physics analysis as stored in special L2/EF Result & L2/EF Fragment data objects, appended to event (under development now)
Corfu Workshop
S.Sushkov, ATLAS EF, IFAE Barcelona

12. Trigger  Physics Analysis
Trigger efficiencies should be used in physics analysis
and using common offline tools/framework is important benefit !
Physics analyses may also be useful for tuning trigger:
sensitivity of physics channels to changing thresholds / lumi
understanding BG & necessary statistics for given cuts (QCD)
there are physics scenarios difficult for the trigger
SUSY spectrum with small mass gaps (previous)
… or long chains of consequent decays
models with many jets but no missET (RP-V SUSY)
perhaps, additional trigger may be useful & introduced ?
Additional functionality of HLT / Event Filter:
due to such benefits of Event Filter as
access to full event data
possibility to use potentially all/any tools/libs from Offline framework
very good modularity & flexibility of EF Farms design
it is possible to run in EF physics & detector performance monitoring algorithms based on Offline tools
more complicated reconstruction to monitor simplest typical physics channels (fit masses, reconstruct decays, etc)
Corfu Workshop
S.Sushkov, ATLAS EF, IFAE Barcelona

13. Status & Summary Implementation of HLT
(infrastructure) is nearly done…
ATLAS Combined Test Beam done
first test of all detectors & TDAQ components - combined
L2+EF trigger tested in real online readout/DataFlow
Tests of TDAQ+HLT SW are carried out well periodically
on dedicated test computer clusters (code development)
special realistic Large Scale Tests (HLT on up to 700 hosts)
TDAQ/HLT pre-commissioning in ATLAS P1 – recently started !!
HLT follows & is used for commissioning of ATLAS detectors
continue works on performance/stability & additional functionalities
now – it’s time to activate integration of
High Level Trigger with physics analyses
e/gamma & muon triggers are already well implemented and used in tests
tau and B-triggers are well on the way
jets/Emiss trigger works started / ongoing

14. BACK-UP SLIDES

15. Event Filter: Modularity & flexibility

16. Main idea & motivation Event Filter has unique features:
accessing full event data built by SFI
ability to run Athena offline algorithms within TDAQ DataFlow framework
On this basis EF can provide many additional
useful functionalities: online monitoring and
calibration
Athena
framework
PTIO
interf
EFHLT
interf
EF PSC
TileMon algo

17. RCD MobiDaq EF+Athena monitoring schema
add Atlantis Event
Display (LATER)
RodCrateDaq
MobiDaq
RCD
data file
debug only
RCD emu-EB
oh_display
(ONLINE !!)
GNAM
fileSampler OH
Ev. Mon. Svc
HistoSender
Athena
framework
PTIO
interf
EFHLT
interf
EF PSC
TileMon algo

18. Trigger rates per objects (TE)
Selection
2*1033 cm-2s-1
Rates (Hz)
Electron
e25i, 2e15i
~40
Photon
g60i, 2g20i
Muon
m20i, 2m10
Jets
j400, 3j165, 4j110
~25
Jet & ETmiss
j70 + xE70
~20
tau & ETmiss
t35 + xE45 ~5
b-physics
2m6 with mB /mJ/y
~10
Others
pre-scales, calibration, …
Total
~200