DØ Level 3
Algorithms Meeting: 25th September 2002 in the Farside
Talks
-
Status of L3 Missing ET: Talks by
Lee Sawyer
followed by
Markus Klute.
Lee Sawyer is maintaining the infrastructure and is now
being helped by Markus Klute and Joe
Steele in testing the performance of the Missing ET tool.
The rate of progress is very rapid at the moment.
Status:
-
Tool and example filters and triggers have been put into the trigger database.
-
Various bugs have been fixed and should be released into t02.35.
-
eta_max was not implemented
-
Missing ET phi calculation was incorrect for cell-based missing ET
-
E cut was incorrect (was being applied to ET).
Open questions:
-
Should neg. cells be included? (currently included)
-
Which parameters should be controled by RCP and which should be
trigger list parameters?
-
How should muon correction be handled? There was a general feeling
that the missing ET tool should calculate calorimeter missing ET and
that then the missing ET filter should find muons using a muon filter
and, if appropriate, make the correction.
Still to be done:
-
Lee will run on ~50K data events to check:
-
robustness
-
time consumption in maxopt
Time consumption is 4ms/evt on a 1GHz clued0 node with no maxopt.
-
memory consumption
-
L3 vs. reco comparisons need to be re-done with a reco version that
doesn't suffer from the bugs in applying zero suppression.
-
test efficiency of ET cut on sample of offline selected W->enu events.
(Marco is to select these using pick_event?)
-
test rejection on sample of CEM(1,10) M&P events
-
demonstrate effectiveness of L3 NADA in keeping Missing ET under
control on a run with calorimeter hot towers
(Tests shown by Markus were using offline NADA and the calorimeter
quality of the used run was
uncertain. We want it to be really bad!)
Hopefully this can be done on a timescale of order a week so that we
can go to the trigger board with a proposal for a W->enu filter very soon.
-
Some results on vertex stability (from vertex_examine) and comparisons
of L3 and offline tracking: Michiel Sanders.
-
Beam spot position and angles are stable within one fill at the
required level for impact parameter b-tag. We can start doing impact
parameter b-tag before event by event primary vertex is available.
(Even afterwards, beam spot may be a useful constraint to apply to the
event by event primary vertex.)
Suggestion is to take a very short run at the beginning of a fill to
determine beam spot.
-
Open questions:
-
How do we get a measurement of beam spot position and angles as
quickly as possible? Needs about 5000 tracks or 500 events. This
currently takes 5-10 minutes in vertex_examine.
In principle this would be possible in a few seconds in L3.
-
Speed of vertex_examine: is it really limited by data unpacking to ~3Hz?
-
Stability of beam spot at the very beginning of the store needs to be
looked at closely.
-
What is the best mechanism to get this information into the L3 nodes?
Use COOR to download at the same time the trigger list is being downloaded?
-
Comparisons
of L3 and offline tracking:
-
L3 tracks are much flatter in phi.
-
Applying cuts on gtr401 tracks to simulate L3 track cuts makes
distributions look similar, but there are some major differences.
-
In particular, gtr401 tracks seem to have a much better efficiency for
matching to SMT than L3 global tracks!
p12 and p13 status reports
We had short reports by almost all tool
authors on:
-
the status of their cross checks of p12 readiness for online
running
-
the status of their code in this week's test release, which is expected
to be iterated upon for two weeks and form the basis for p13.
Summary:
-
Roottuples and some global tests:
Lukas Phaf
Lukas made some roottuples for run 163172 and did some global checks
of consistency.
-
Electrons: Volker Buescher.
Road algorithm is new for p12 and has been extensively tested
(see previous talk).
The technical changes discussed there are in the process of being tested.
"Standard" electrons have been compared p11 vs. p12.
Some differences in distribution of e.m.fraction are observed at the
edge of cc which are thought very likely due to new
cal weights for ICD. Above 10 GeV the differences are not apparent.
What is status of refined treatment of electron shower shape in p13?
-
Muons: Martin Wegner and Martijn Mulders (by email).
Lots of changes to offline muons have gone in for p12. Martijn is in
the process of doing the L3Muon checks as part of the standard
routine testing.
Martin is studying the special run that included L3Muon with central
track-match run online. The performance looks OK.
-
Taus: Yann Coadou
p12 (trigsim) vs. p11 (online) results are checked and are OK.
As noted previously, fractional differences at the 10^-4 level
in candidate properties are observed between p11 online and p11 (or
p12) trigsim.
-
Calorimeter unpacker: Marumi Kado.
Marumi has implemented zero suppression at L3 with a threshold that
can be chosen by the user. This is currently in the head version, but
needs to be included in the p12.04 build that will happen this
weekend.
Time spent in the unpacker is not much reduced, because of the
overhead in reading the calorimeter pedestal files and applying the
threshold.
Time should be saved in clustering, but this has not been measured.
-
Track-based VTX: Chris Barnes
P12 primary vertex in z is a new algorithm,
which has been extensively tested by Chris
(see previous talks).
For p13 expect to have first version of 3D primary vertex tool.
Will be presented in detail at next week's meeting.
-
IP b-tag: Per Jonsson.
New code for p12 version has been extensively tested (see previous
talks).
d0mino-debug gave time < 5 ms per event.
The memory consumption has not been fully tested.
For p13 have improved paramterization for track errors from Daniel
Whiteson.
-
CFT tracking: Ray Beuselinck.
Ray has been able to re-run some analyses on the p12 version of the CFT tracker
using p11 Monte Carlo samples for ttbar and qcd events (several thousand events
of each). The efficiencies and purities look very comparable with results
presented previously. The only significant change is the big improvement in
cputime for the reorganised stereo code.
The code in this weeks test release is stable -- same as p12 apart from a few
minor cleanups to eliminate gcc compiler warnings.
Ray hopes to give a detailed update of the status at a Level3 meeting, including
improvements to the axial tracking (yet to be done), later this autumn.
-
CPS: Andre Turcot
-
CPS p12: Chunhui will be asked to run on a large data sample and test timing
and memory consumption.
-
FPS: Nothing expected until p14 at the earliest.
-
Thumbnail (p13): Peter Tamburello
-
Infrastructure to store physics_result base class information for each
tool is ready except for:
-
L3FMuon physics_results can not be used, because they are not filled
or are not persistent.
-
Peter is waiting for Lee to supply description of missing ET physics_results.
-
Wrapper to produce physics_results for standalone track filter not
yet done.
-
Quality information on candidates is needed on the TMB, but will be
stored only if tool authors supply Peter with a description of the
access methods of the quantities needed.
-
Some "physics_results" information is not
stored in the L3Chunk, but is re-created from more elementary
information when it is required.
This is a problem when the more elementary
information is not present on the thumbnail.
This is the case, for example, for electron transverse shower shape
variables.
Still to do before p12 can run online:
-
Check L3 global tracking (Daniel).
-
Once calorimeter unpack changes go in to p12.04 then the tests of
calorimeter-based tools need to be re-done. Perhaps Lukas can be
persuaded to re-do his checks?
-
Find and fix crashes in L3 monitoring (Jon).
-
Add L3 physical stream information into L3Chunk (Jon).
-
Implementation of inclusive streaming for M&P events (monitor stream)
would be desirable, but not absolutely necessary. (It depends on how long it takes to find and fix
crashes.) (Jon).
Scribe: Terry Wyatt.