Title: D0 Run II Trigger
1L2Cal
Review
(Algorithms)
Nikos Varelas University of Illinois at
Chicago L2Cal Group at UIC Mark Adams Bob
Hirosky Rob Martin Nikos Varelas Marc Buehler
(graduate student) James Heinmiller
(undergraduate) Mike Klawitter (part time
engineer)
1
2/6/99
L2 Review
NV/UIC
2Overview
- L2Cal Crate and I/O
- Status of L2Cal Algorithms
- Jets
- Electrons
- Missing ET
- Timing of Algorithms
- Summary
3L2Cal Inputs
- From Calorimeter via FIC/MBT
10 input cables with 304 Bytes/cable - Header
- L1 Seed Tower Bit Masks for EM and Total
- L1 Tower ET data for EM and Total
- From SCL via MBT
- L1 accept (3 Bytes)
- L1 Qualifiers (2 Bytes)
- L2Jet Needed
- L2Em Needed
- L2Etmiss Needed
- 3 Spare Bits
- Unbiased Sample
- Forced Write
- Collect Status
- L2Global accepts
4L2Cal Outputs
- To L3
- For normal events send the L2Global output
- For UBS or Forced Write events the full L1 input
and L2Cal output will be sent - To L2Global
- About 136 Bytes/event (including headers)
- Will be fine tuned when algorithms are finalized
based on input from physics/Id groups - Each worker will preface its data with a 12Byte
header - Header will include information about the
processing status (i.e., format errors, timeouts
etc) of the event - Each worker will complete transmission with a 4
Byte trailer
5L2Cal Algorithms
All L2Cal algorithms will use a low-threshold
reference set of L1 0.2x0.2 trigger towers as
input for clustering. L1 EM ET - Rounded in
0.25 GeV steps L1 Tot ET - Sum of EM and HAD
truncated in 0.5 GeV steps
What are the efficiencies of these algorithms?
Can we reduce the trigger rate w/o significant
cost in efficiency?
Can we do all these stuff in less than 100 ms???
6L2 Jets
- The algorithm
- Event Samples used in simulations
- Data W-gt JJ triggers from Run 1C Global
Runs with Lum 17E30 - MC UPG GEANT inclusive jet events w/
generated thresholds (2,5,10,20,40,80 GeV)
overlapped with (1,3,5,7) additional MB
interactions (not overlapped with noise)
- Start w/ list of jet seed towers from L1
- For each seed tower, cluster ET of the
surrounding 5x5 (or 3x3) tower array - Add to Jet list all clusters whose ET sum
exceeds a min threshold - ET order the Jet list (descending order)
- Eliminate Jets failing overlap restriction
If ET(A) gt ET(C) keep A,B If ET(C) gt ET(A)
keep C,B
7W-gtJJ DataTower Seed Distributions
For a high-ET (gt350GeV) jet data sample
m 20, RMS 6 for L1(1,2)
8W-gtJJ DataL2 Jet Distributions
9L2Jet Efficiency and Rate Studies
Measured w/ MC - UPG Geant sample
Pjet and Cal Jet matching methods 1) Projection
Method
Project PJet axis into calorimeter. Does
corresponding seed/cluster ET pass imposed cuts?
2) Matching Method (run L2Jet algorithm)
Compare L2Jets to PJets Look for matches
10L2Jet Efficiency for seeds/clusters
Reference algorithm L1(1,2) L2(1,10)
Central jets
Effs. for seed/cluster cuts and Algorithm(seed
cut,cluster cut)
11L2Jet Rate Estimates
- Method
- First weight MC events appropriately
- 1) use JETRAD to bridge all h PJet Cross
Section to central inclusive jet CS in
data - 2) estimate total MC event cross section for
PJet ETgt5 GeV 1/11 Min Bias cross section - Calculate trigger rate as fraction of MB events
passing imposed threshold(s) - Plot L2Jet Efficiency vs Rate for 20 and 100 GeV
PJets. - Compare L2 3x3 jet algorithm to 5x5 version
- Measure Eff. vs Rate w/ and w/o L2 clustering
- Examine the effects of 0.5 GeV truncation to
trigger-tower ETs
123 3x3 algorithm
5 5x5 algorithm
Eff. vs Rate at 20 GeV
- factor of 3 rate reduction w/ 20 eff. cost
- no strong cluster size preference
- need to tune the MC further so we can
study/improve - the algorithm for low-ET jets
133 3x3 algorithm
5 5x5 algorithm
Eff. vs Rate at 100 GeV
L2 thresholds (1,60)(1,50)(1,40)(1,30)(none)
L1(1,7)
L1(1,9)
L1 only
- order of magnitude rate reduction easily
attainable - at L2 w/o loss in efficiency
- no strong cluster size preference
14Effects of L1 Total-ET Truncation for 20 GeV Jets
L2 (1,4)
w/ 0.5 GeV truncation
L1 (1,2)
L2 (1,8)
L2 (1,6)
L2 (1,10)
0.25 GeV rounding
L2 (1,12)
L2 (1,15)
the effect of L1 energy truncation can be
accommodated at L2 by choosing lower jet
thresholds
15L2Jet Rejection
An Example at Lum1E32
Level-0
Eff at 100 GeV
45 mb x 1E32 4.5 MHz
L1(1,9)
96
6.7 KHz
ltL2(1,30)gt
L1(1,12)
92
130 Hz
1800 Hz for same Eff
L3
16L2 Electrons
- The algorithm
- Event Samples used in simulations
- Single electrons uniformly distributed in f, in
the forward region 1.9lthlt2.3 - ISAJET dijet events with various thresholds
starting at 2 GeV - Events were processed through UPG_GEANT with two
(on - average) additional interactions (not overlapped
with noise)
- Start w/ list of EM seed towers from L1
- For each seed tower, determine nearest neighbor
w/ the largest ET - Calculate the following summed ET quantities
- 1) ET(EM) of seed tower largest neighbor
- 2) ET(Total) of seed tower largest neighbor
- 3) sum ET(Total) of 3x3 trigger towers centered
on seed tower - Order surviving candidates in descending ET(EM)
Apply cuts on ET(EM), EM fraction, and Isolation
17L2 Electrons Efficiency and Rate Studies
Courtesy Mrinmoy Bhattacharjee
- L1 Cuts
- FPS 0.3 MIPs (upstream)
- 5 MIPs (downstream)
- U view matching
- V view matching
- CAL EM trigger tower above threshold
- Match FPS with CAL L1 Tower in Quadrant
- L2 Cuts
- FPS Require downstream U and V view matching -gt
convert to h, f in 0.2 x 0.2 bins - CAL Find EM cluster using NN algorithm. Apply
EM fraction and Isolation cuts. - Match FPS track to EM cluster within Df x Dh
0.3 x 0.3 - No rounding/truncation applied to L1 tower
energies
18Forward electrons
Eff. vs Background Rate at 18 GeV
preliminary
L2 thresholds (1,15)(1,12)(1,10)(none)
L1 (1,7)
L1 (1,7)
19Forward electrons
Eff. vs Background Rate at 18 GeV
preliminary
L2 thresholds (1,17)(1,15)(1,12)(none)
L1 (1,10)
L1 (1,10)
20Forward electrons
Eff. vs Background Rate at 30 GeV
preliminary
L2 thresholds (1,15)(1,12)(1,10)(none)
L1 (1,7)
L1 (1,7)
w/ FPS match
order of magnitude rate reduction attainable
at L2 w/ small cost in efficiency
21Forward electrons
Eff. vs Background Rate at 30 GeV
preliminary
L2 thresholds (1,17)(1,15)(1,12)(none)
L1 (1,10)
L1 (1,10)
w/ FPS match
order of magnitude rate reduction attainable
at L2 w/ small cost in efficiency
22L2Jet Output
- Candidates will be sorted in descending ET order
- Information per candidate
- eta (1 Byte)
- phi (1 Byte)
- ET (2 Bytes)
- eta center (1 Byte)
- phi center (1 Byte)
- eta leading TT (1 Byte)
- phi leading TT (1 Byte)
- Spare (4 Bytes)
- Total 12 Bytes/object
23L2Em Output
- Candidates will be sorted in descending ET order
- Information per candidate
- eta (1 Byte)
- phi (1 Byte)
- ET (2 Bytes)
- EM fraction (1 Byte)
- Isolation Fraction (1 Byte)
- eta leading TT (1 Byte)
- phi leading TT (1 Byte)
- eta other TT (1 Byte)
- phi other TT (1 Byte)
- Spare (2 Bytes)
- Total 12 Bytes/object
24L2Etmiss Output
- Need input from physics groups
- Information per event
- Missing ETX (2 Bytes)
- Missing ETY (2 Bytes)
- Scalar ET (2 Bytes)
- Spare (10 Bytes)
- Total 16 Bytes/event
25L2Etmiss
- The algorithm
- Possible Enhancements
- Calculate Scalar ET using the same cuts as for
Vector ET - Calculate ET for more than one set of Tower cuts
- Calculate ET using different threshold for each
Tower
Loops over all towers within prescribed h range,
calculating the vector ET sum of all towers with
ET gt Min_Tow_ET. It returns the X and Y
components of the Missing ET.
26L2Cal Algorithm Timing
- Code
- written in C
- compiled with C or C compiler on DEC Alpha
workstation running UNIX (timing results roughly
the same) - Executable down-loaded and run on UIC PC164
evaluation board containing DEC 21164 Alpha
processor with 500MHz clock - Event Sample
- MC Dijet data generated with ISAJET
- Data block Structure as planned for hardware
- 10 cable blocks containing
- EM Tower Seed Mask
- Total Tower Seed Mask
- EM Tower ET data
- Total Tower ET data
27L2Jet Algorithm Timing
Average seed range
Time (ms) 2.5 1.12 x ( seeds)
28L2Em Algorithm Timing
Time (ms) 2 2.3 x ( seeds)
29L2Etmiss Algorithm Timing
The average time for 0.5 GeV Tower ET threshold
is 33 ms
All Towers above threshold
30Summary
- We have a fully designed L2Cal Preprocessor
system which has sufficient CPU power to execute
reasonable L2 algorithms with lt few deadtime - if more power needed, can add up to two Workers
for parallel processing - We have working versions of Jets/Electron/Missing
ET algorithms which offer acceptable rate
reduction - The data movement architecture is complete and
the monitoring path has been established (see
previous talks) - We request TDR approval