Upload
noel-cox
View
219
Download
0
Tags:
Embed Size (px)
Citation preview
ALICE Physics Forum, CERN - 07.03.07 Andrea Dainese 1
Vertex reconstructionVertex reconstructionin ALICEin ALICE
Andrea Dainesein collaboration with
E.Bruna, E.Crescio, S.Gorbunov, I.Kisel,
S.Masciocchi, M.Masera, F.Prinoalie
ve, M
.Tad
el
ALICE Physics Forum, CERN - 07.03.07 Andrea Dainese 2
OutlineInteraction diamond at LHC (pp)
Primary vertex reconstructions
along z with SPD tracklets
in 3D with SPD tracklets
in 3D with tracks
Secondary vertex reconstruction (heavy-flavour)
present algorithm
developments in progress
ALICE Physics Forum, CERN - 07.03.07 Andrea Dainese 3
Primary vertex in pp collisionsPrimary vertex resolution affects impact param. resolution:
E-by-E vertex position reconstructed in 3D using tracks (or SPD tracklets, see talk by F.Prino)
Then, interaction diamond profile from many events in fill
Nominal spread of the diamond at LHC:
vertextrackrd )(0
2
*
2,,
,yx
bunchyxdiamond
yx
pp 14TeV nominal (ALICE) pp 14 (0.9)TeV initial Pb-Pb
3.75 m 3.75 (3.75) m 3.75 m
0.55 (10) m(+displacement)
10 (10) m 0.5 m
12 (150) m 50 (198) m 11 m
~1034 (30) cm-2s-1 ~1030 (1028-29) cm-2s-1 ~1027 cm-2s-1
yx,
diamondyx,
L
*
ALICE Physics Forum, CERN - 07.03.07 Andrea Dainese 4
Algorithms for vertex reco
AliITSVertexerZ: z from SPD trackletsPROS: almost 100% efficient in triggered events;
CONS: only z; with large beam offsets has problems (loss of precision and accuracy);
AliITSVertexer3D: xyz from SPD trackletsPROS: 3D; no problem with beam offset;
CONS: not 100% efficient; cannot account for curvature;
AliVertexerTracks: xyz from tracksPROS: most precise (curvature known); gives cov matrix and 2;
CONS: can be done only after tracking;
Optimal performance by exploiting the pros of the different algorithms during event reconstruction (next slide)
ALICE Physics Forum, CERN - 07.03.07 Andrea Dainese 5
Barrel Tracking & Vertex Recons.1) Vertex (x,y,z) from SPD tracklets (AliITSVertexer3D)
only z with AliITSVertexerZ when AliITSVertexer3D fails
2) Barrel Tracking
3) Compute diamond transverse profile, as mean and RMS of vertices in high-multiplicity events of each “LHC fill” ➔ stored in OCDB
4) Repeat step 1)important, to have zvtx for all events (even with one tracklet)
5) Barrel Tracking6) Vertex with tracks (AliVertexerTracks), using info on
diamond profile, taken from OCDB (in progress; not in PDC06)
final (x,y) resolution better than diamond RMS for ALL events
1st p
ass
on
sub
sam
ple
2nd p
ass
full
stat
istic
s
ALICE Physics Forum, CERN - 07.03.07 Andrea Dainese 6
AliITSVertexerZ – the method
Beam axis
Layer 1
Layer 2
Build “tracklets” from SPD Clustersassociate each Cluster on layer1 to all the Clusters on layer2 within a window Δφ <0.01 rad
Calculate Zi = Z of closest approach of tracklet and nominal beam axis.
Fill a histogram of Zi with 100 m (200 in pp) bin size
define a z window (2 mm wide) around the “peak”
Calculate vertex position and errorZv = weighted average of the Zi of the tracklets in the window
after symmetrization around the peak
Error = propagation of errors on Clusters
also in case of just 1 tracklet
F.Prino, Münster
ALICE Physics Forum, CERN - 07.03.07 Andrea Dainese 7
AliITSVertexer3D – the method1) Build tracklets (= straight lines) from pairs of Clusters
First selection done w.r.t. nominal beam axis
Loose selection: Δφ < 0.5 rad, DCA to beam < 2.5 cm, |zINT|<5.3 cm
2) Combine tracklet pairs and select them according to:small DCA (< 1 mm) between the two tracklets
Tracklet intersection close to beam axis (rINT < 2.5 cm )
Tracklet intersection in the diamond region (zINT< 5.3 cm)
3) Get a first estimate of the vertex from selected trackletssame vertex finder algorithm used with tracks (AliVertexerTracks)
4) Re-build trackletsSelection done w.r.t. beam position from vertex estimate in previuos step
Tight selection: Δφ < 0.01 rad, DCA to beam < 0.5 cm, |zINT - zVERT|<0.5 cm
5) Combine tracklet pairs and select them according to:small DCA (<1 mm) between the two tracklets
Tracklet intersection close to the estimated vertex (r < 0.5 cm)
Tracklet intersection in the diamond region (z < 0.5 cm)
6) Calculate the vertex using the selected tracklets
F.Prino, Münster
ALICE Physics Forum, CERN - 07.03.07 Andrea Dainese 8
Vertexing with SPD: Efficiencypp 14 TeV events (PYTHIA minimum bias)
MC vertex: (0,0,0) with s = (50 m, 50 m, 5.3 cm) [*=10m]
No trigger selection
Efficiency = events with vertex / all eventsVertexer3D requires at least 2 selected tracklets
Events with just one tracklet are ≈10%
Z 3D
kPyMb 84% 75%
Non diffr. 98% 90%
SPDpp14TeV B = 0.5 T
SPDpp14TeV B = 0
Z 3D
kPyMb 84% 77%
Non diffr. 98% 92%
ALICE Physics Forum, CERN - 07.03.07 Andrea Dainese 9
Vertexing with SPD: Resolutions
Resolution improved for both vertexers in last weeks
New procedure for VertexerZ Cut optimization and improved track selection for Vertexer3D
SPDpp14TeV B = 0.5 T
SPDpp14TeV B = 0
Everything working with B=0better resolution in (x,y) because tracks are straight
ALICE Physics Forum, CERN - 07.03.07 Andrea Dainese 10
Vertexing with SPD:Beam offset in (x,y)
VertexerZ resolution dramatically worsens in case of large (and unknown) beam offset
The performance is completely recovered if the X,Y position of the beam is known
Vertexer3D performance not affected also in case of large and unknown beam offset
better resolutionwith large offsetdue to geom. effect
ALICE Physics Forum, CERN - 07.03.07 Andrea Dainese 11
Barrel Tracking & Vertex Recons.1) Vertex (x,y,z) from SPD tracklets (AliITSVertexer3D)
only z with AliITSVertexerZ when AliITSVertexer3D fails
2) Barrel Tracking
3) Compute diamond transverse profile, as mean and RMS of vertices in high-multiplicity events of each “LHC fill” ➔ stored in OCDB
4) Repeat step 1)important, to have zvtx for all events (even with one tracklet)
5) Barrel Tracking6) Vertex with tracks (AliVertexerTracks), using info on
diamond profile, taken from OCDB final (x,y) resolution better than diamond RMS for ALL events
1st p
ass
on
sub
sam
ple
2nd p
ass
full
stat
istic
s
ALICE Physics Forum, CERN - 07.03.07 Andrea Dainese 12
AliVertexerTracks
Three main “tasks” implemented:TRACKS SELECTION reject secondary tracks
VERTEX FINDINGgive first estimate of vertex positionmethod: approximate tracks as straight lines and calculate the
minimum-distance point among all the tracks at once
VERTEX FITTING:give optimal estimate of the position of the vertexgive vertex covariance matrix
give a (vertex quality)
ALICE Physics Forum, CERN - 07.03.07 Andrea Dainese 13
VertexerTracks::Vertex Fitter“Fast Vertex Fitting”, CMS Note 1997/051
Tracks are propagated to the point given by the vtx finder
Each track gives an independent measurement of the vertex position: a 2 is written as the sum of the single track
2s w.r.t. a generic vertex position rvtx:
where ri is the (x,y,z) position of the ith track and Vi = Wi-1 is
the cov matrix of ri, extracted from the track’s cov matrix
The solution that minimizes this 2 is analytic:
)rr( W)rr()r()r( 22ivtxi
T
iivtx
ivtxivtx
ii
ii
ivtx rWWr1
1
WV
ii
vertexvertex
covariancematrix
ALICE Physics Forum, CERN - 07.03.07 Andrea Dainese 14
Easy to add “beam constraint”
Same procedure as before + constraint:
iii
iivtx
rWrVWVr bb
1
b
1
b WVV
ii
vertex
vertexcovariance
matrix
zz
yyxy
xyxx
z
y
x
wgtd
wgtd
wgtd
00
0
0
V ;r bbwhere:
ALICE Physics Forum, CERN - 07.03.07 Andrea Dainese 15
VertexerTracks – the method
Primary vertex reconstruction: iterative procedureInput: interaction diamond (vtx0)
Select tracks with kITSrefit and 5 or 6 ITS clusters
1st iteration:TRACK SELECTION: |d0
r(vtx0)| < 15 d0
VERTEX FINDINGVERTEX FITTING vtx1
2nd iteration:TRACK SELECTION: |d0
r(vtx1)| < 3 d0
VERTEX FINDINGVERTEX FITTING vtx2 (final estimate)
ALICE Physics Forum, CERN - 07.03.07 Andrea Dainese 16
VertexerTracks: Efficiencypp 14 TeV events (PYTHIA minimum bias)
MC vertex: (0,0,0) with s = (50 m, 50 m, 5.3 cm) [*=10m]
No trigger selection
Vertex found if distance to MC vertex in (x,y) is < 1 mm
Efficiency: events with vertex found / all events
(with beam constraint; not yet in PDC, but soon)
ALICE Physics Forum, CERN - 07.03.07 Andrea Dainese 17
VertexerTracks:Resolutions & Pulls
res: from gaussian fit of residuals distribution
pulls: from gaussian fit of residual/error distribution
3 track-cut
(with beam constraint; not yet in PDC, but soon)
3 track-cut
ALICE Physics Forum, CERN - 07.03.07 Andrea Dainese 18
VertexerTracks: 2/ndf
ALICE Physics Forum, CERN - 07.03.07 Andrea Dainese 19
New development in VertexerTracks: possibility to remove tracks
Why remove tracks?
systematic underestimate of d0 for tracks used for vertex fit
primary vertex has to be reconstructed for each D/B candidate excluding its daughters: time-consuming!
Shortcut:
our fit is a “weighted mean over tracks”
➔ can calculate ri and Wi and subtract them from rvtx and V
without recalculating the vertex from scratch
implemented in AliVertexerTracks::RemoveTrackFromVertex()
ii
ii
ivtx rWWr1
1
WV
ii
ALICE Physics Forum, CERN - 07.03.07 Andrea Dainese 20
VertexerTracks w/o beam constraint(importance of event display)
tracks in TPC&ITS(≥5 pts)
problem spottedwith event display scanning(Y.Foka, M.Tadel)
trac
ks u
sed
for
vert
ex
“bad” verticesvertex
not found
ALICE Physics Forum, CERN - 07.03.07 Andrea Dainese 21
VertexerTracks w/o beam constraint(importance of event display)
vertexnot found
Ladies and Gentlemen, (in)famous event #24 !Ladies and Gentlemen, (in)famous event #24 !before after
ALICE Physics Forum, CERN - 07.03.07 Andrea Dainese 22
VertexerTracks- pp 0.9 TeV B = 0 = (198m,198m,7.9cm) -
pp14TeVB = 0.5 T
pp0.9TeVB = 0
SPDbarrel
ALICE Physics Forum, CERN - 07.03.07 Andrea Dainese 23
x
zpu
llpu
ll
x
z
reso
lutio
n [
m]
reso
lutio
n [
m]
Secondary vertex reconstruction AliVertexerTracks::VertexForSelectedTracks()
Finder+Fitter ➔ errors ➔ heavy flvr cuts on normalized varsTest on D0K vertex: same resolution as V0 vertexer
Test on D+K in progress (E.Bruna, F.Prino) Münster
ALICE Physics Forum, CERN - 07.03.07 Andrea Dainese 24
Implementation of a new vertexing package (Gorbunov, Kisel, Masciocchi)
Vertexing package based on the Kalman filterS. Gorbunov and I. Kisel, CBM-SOFT-note-2006-001, CBM-SOFT-note-2006-002
Used by HERA-B, CBM
Advanced fitting functionality included:Use of full track kinematics
Invariant mass constraint
Topological constraint (pointing to the primary)
Adapted to ALICE track format (AliESDtrack)
Tests in progress, using D0K as a benchmark
ALICE Physics Forum, CERN - 07.03.07 Andrea Dainese 25
New vertexing package- resolutions and pulls on D0 vertex -
x
z
x
z
Preliminary,
Work in progress
reso
lutio
n [
m]
reso
lutio
n [
m]
pull
pull
ALICE Physics Forum, CERN - 07.03.07 Andrea Dainese 26
Summary
3 complementary (not alternative) vertex reconstruction methods: all used during event reconstruction
AliITSVertexerZ: z from SPD trackletsPROS: almost 100% efficient in triggered events; error provided;
AliITSVertexer3D: xyz from SPD trackletsPROS: 3D; no problem with beam offset;
AliVertexerTracks: xyz from tracksPROS: most precise (curvature known); gives cov matrix and 2;
allows fast removal of tracks from fit; being tested for heavy-flavour vertexing (D0, D+,...): good
peformance
Kalman-filter vertex fitter on the way
ALICE Physics Forum, CERN - 07.03.07 Andrea Dainese 27
EXTRA SLIDES
ALICE Physics Forum, CERN - 07.03.07 Andrea Dainese 28
VertexerZ: average of residuals
Unbiased!The small bias (≈5 m) which was present in the previous releases of the code was due to a systematic error in finding the maximum of the histograms for few low multiplicity events
Corrected since Rev. 1.17 of AliITSVertexerZ.cxx
ALICE Physics Forum, CERN - 07.03.07 Andrea Dainese 29
Vertexer3D: average of residuals
No apparent bias
ALICE Physics Forum, CERN - 07.03.07 Andrea Dainese 30
VertexerZ vs. vertex position (II)
In events generated with AliRoot rev. after June 12th 2006 a bias (slope vs. ztrue) is observed
see Jan Fiete talk at first physics meeting on Jan 19th 2007
Major modifications in ITS geometry
ITS geometry changed to TGeo
SPD chip thickness reduced to 150 m
The bias is introduced by the SPD chip thickness
Bug found and fixed by L. Gaudichet on February 5th 2007
ALICE Physics Forum, CERN - 07.03.07 Andrea Dainese 31
VertexerZ and beam offset (I)
Beam offset up to 1 cm assumed unknown
Efficiency not affectedResolution dramatically worsens with increasing beam offset
ALICE Physics Forum, CERN - 07.03.07 Andrea Dainese 32
VertexerZ and beam offset (II)
Beam offset up to 1 cm assuming to know the beam position
Good performance for offsets up to 1 cm if the beam position is known
ALICE Physics Forum, CERN - 07.03.07 Andrea Dainese 33
Vertexer3D and beam offset
Beam offset up to 1 cm assumed unknown
Performance maintained
ALICE Physics Forum, CERN - 07.03.07 Andrea Dainese 34
Pile-upExpected interaction rate = 2×105 Hz at a luminosity of 5×1030 cm-2s-2
1 interaction every 200 bunch crossings
In case of SPD strobe duration of 100 ns 4 bunch crossings (0.02 interactions)
All events in the strobe are overlapped even if not belonging to the same bunch-cross
Caveat: high- multiplicity triggers will select piled-up events
First check on AliITSVertexerZ in the case of pile-up“Manual merging” of recpoints with an “ad hoc” macro
Results:Vertices with distances >600 μm: the vertex of the event with higher multiplicity is found
Vertices with distances <600 μm: an intermediate value of z is found
Under study: check if the vertexer can be used to “detect” the pile-up, searching for two peaks (possible in the case of well separated peaks)
Study to be performed also on the Vertexer3D
ALICE Physics Forum, CERN - 07.03.07 Andrea Dainese 35
VertexerTracks: Efficiency for triggered events
pp 14 TeV events (PYTHIA minimum bias)
MC vertex: (0,0,0) with s = (50 m, 50 m, 5.3 cm) [*=10m]
Select triggered events: [ (V0A or V0C) and SPDFO ]
Vertex found if distance to MC vertex in (x,y) is < 1 mm
Efficiency: events with vertex found / triggered events
(with beam constraint)
triggeredminimum bias
ALICE Physics Forum, CERN - 07.03.07 Andrea Dainese 36
Performance w/o beam constraint(as in PDC events)
with constraint without constraint
ALICE Physics Forum, CERN - 07.03.07 Andrea Dainese 37
Residuals vs zMC
Small bias in zrec (“pulled towards z=0”)
Larger bias observed in vertex with SPD (see talk by F.Prino)
It was due to bug in the SPD geometry, introduced in September, now corrected
However, for large z there can be acceptance-induced bias
AliVertexerTracks
ALICE Physics Forum, CERN - 07.03.07 Andrea Dainese 38
New development: possibility to remove (add) tracksOne-shot fit procedure
Select tracks with kITSrefit and 5 or 6 ITS clusters
Prpogate them to diamond centre (vtxmean)
TRACKS SELECTION: |d0r(vtxmean)| < 3 d0
VERTEX FITTING
standard procedure one-shot fitOK!
ALICE Physics Forum, CERN - 07.03.07 Andrea Dainese 39
Efficiencies vs zMC
VertexerZ(all events)
VertexerTracks(triggered events)
even
ts w
ith v
erte
x / a
ll ev
ents
even
ts w
ith v
erte
x / t
rigge
red
even
ts