Transcript Flavour Tagging and GLoBal event cuts

1
DISCRETE '08
Symposium on Prospects in the Physics
of Discrete Symmetries
11–16 December 2008, IFIC, Valencia,
Spain
Flavour Tagging
performance in LHCb
Marc Grabalosa Gándara (14/12/08)
on behalf of the LHCb collaboration
Motivation
2



LHCb is a 2nd generation
precision experiment coming
after
B-Factories
and
Tevatron
Improve precision on g and
other CKM parameters
Many measurments require
the knowledge of the initial
flavour of the B meson
Unitarity Triangles
Bd0  p+ pBd0  r p
Bd0  DK*0
BS0  DSK
Bd0  D* p, 3p
BS0  DS p
Bd0  J/y KS0
BS0  J/y f
Importance of tagging
3


Bs oscillation frequency, phase and ΔΓs ( BsDsp, J/ΨΦ, J/Ψη, ηcΦ )
Measure the CKM parameters





 from Bdp0p–p+
 with BdJ/yKS as a proof of principle (  from bs penguin )
 in various channels, with different sensitivity to new physics:
 Time-dependent CP asymmetry of BsDs-K+ and Ds+K Time dependent CP asymmetries of Bd π+ π- and Bs  K+K Comparison of decay rates in the BdD0(K+π-,K-π+,K+K-)K*0 system
 Comparison of decay rates in the B-D0(K+π-,K+π-π+π-)K- system
 Dalitz analysis of B-  D0(KSπ-π+)K- and Bd  D0(KSπ-π+)K*0
Rare B decays
 Radiative penguin Bd  K* , Bs  Φ, Bd  
 Electroweak penguin Bd  K*0 + Gluonic penguin Bs  ΦΦ, Bd  ΦKs
 Rare box diagram Bs +Bc , b-baryon physics + unexpected !
LHCb Overview
4
B flight path of the order 5-10mm
Requirements for CP measurements in
B the sector are
Good particle Id,
excellent tracking and vertexing
Tracking:
δp/p= 0.35% to 0.55%
Vertexing:
~10μm transverse plane and ~60μm in z
Expected Impact parameter resolution
σIP =14μm + 35μm/pT
Calorimeter resolution:
σE/E = 1% + 10%√E (E in GeV)
RICH:
Particle identification, important to
distinguish between Kaon and pions.
Flavour Tagging
5
K+ (p)
Signal B
Same Side
p
Opposite Side
p
Opposite B
Kl-
Qvrtx
Tagging efficiency
ε tag =
Wrong tag fraction
ω=
Effective efficiency
N R + NW
N R + NW + NU
NW
N R + NW
ε
ε
1
2
ω
)2
e
ff =
ta
g(
If several candidates for the same
tagger exist  Select the one with
highest Pt.
Taggers make individual
decisions about the flavour
with varying accuracy, which
is evaluated by a NNet.
Opposite-side tagger (OS)
6
 Tagging objects from b → c → s chain.
pT of m’s:
 Kinematic and geometrical variables
(IPS, P, Pt,...) show a dependence in
purity of right vs wrong tags  CUTS
IP of K’s:
OS Vertex Charge tagger
7
 Use long tracks to build a 2-seed vertex after some kinematic cuts
 Use a NN to select good candidate (2-seed) to SV
All vtx
Good vtx
SV
 Other tracks are added iteratively
 Weighted charge can be used as a tagger
Typical performance: e = 43%  = 42% eeff = 1.14%
All trakcs
Tracks coming from b
Same-side tagger (SS)
8
Hadron from fragmentation (K )
or B** decay ()
• Particle selection cuts
• Proximity to signal b
Typical performance: e = 25.5%  = 35.6%
eeff = 2.13%
K from fragmentation
Other sources
Wrong tagging
Bs → Ds K
 from B*
 from fragmentation
Other sources
Wrong tagging
B0 → + -
Taggers
9
 The tag (b or bbar) is decided by the charge of the tagging object
 Combine the taggers to obtain a final decision of the tag
 Sort in 5 categories depending on the probability of the tag to be
correct
Neural Net
• Obtain a wrong tag fraction ()
for each event from the NN
output
• Has a higher efficency
Combine Particle IDentification
(PID)
• Sort events based on the PID of
the track ordering them in 
• NN independent. Simple method
• Has a lower efficency
 Each method will give a tag and a category
(related with the reliability)
Neural Net method
10
 For each event, each tagger will give an  as a function of the NN
output.
 The wrong tag fraction is fit linearly on the Neural Net output.
Opposite kaon
Bs  J/ 
right
wrong
tagger(K) (NNet) = a0 + a1 NNet
Combination of taggers
11
 Each tagger will have its own  tagger (NNet).
 The final probability for the event will be a combination of the tagger wrong tag
fractions:
Cat 5
Cat 4
P(B) = 1 - P(B)
Cat 3
P-1 =  k (1-  e) …
Cat 2
+1
P(B) = P
Cat 1
P+1 = (1-k)  e …
P
+1
+ P -1 
If P(B) < 0.55 events
is left untagged
To calculate the final combined
effective efficiency, we bin the
events in 5 categories (and treat
them separately in the CP fits).

P(B)
PID based combination of taggers
12

Form possible combinations according:
• Particle Identification (PID)
Muons, electrons , kaons, kaons
or pions SS , vertex charge
Untagged
= 43.0%
Smaller 
• Sum of the individual tagger
decisions (sum of charges)
abs(sum) > 1
Cat 1
Cat 2
Bin events in 5 categories
Cat 3

Cat 4
Sort all possible combinations, including the 
case when abs(sum)>1, according to the 
estimated on a control channel (62 possible
combinations, but only 41 non empty)
Cat 5

= 32.3%
PID combination
Results, ex. Bs  J/ 
13

Performance of taggers:
εtag
muons
electrons
kaons
SS kaons
vtx charge

6,15 ± 0,08
2,78 ± 0,05
15,33 ± 0,12
25,56 ± 0,14
32,79 ± 015
32,5 ± 0,6
29,9 ± 0,9
34,4 ± 0,4
35,6 ±0,3
40,8 ± 0,3
εeff
0,76 ± 0,05
0,45 ± 0,04
1,49 ± 0,07
2,13 ± 0,09
1,11 ± 0,07
Combine all taggers to obtain the global effective efficency, which is
the direct sum of εeff in the 5 tagging categories.
εtag
Using Nnet
PID combination



εeff
53,96 ± 0,16 33,13 ± 0,21 6,14 ± 0,14
56,65 ± 0,17 35,33 ± 0,22 4,89 ± 0,14
NNet εeff increases by ~20%
Performances for a few channels
14
eeff %
e%
%
8.85 ± 0.18
60.7
30.9
Bd→ J/ψ K* 4.29 ± 0.09
53.2
35.8
5.52 ± 0.16
56.8
34.4
Bu→ J/ψ K+ 4.11 ± 0.11
53.1
36.1
Bs→ Dsp
Bd→ pp
Differences can be due to different signal B spectra, trigger…
Control channels
15



Accumulate high statistics in various flavour-specific modes
 can be extracted by:
 B±: just comparing tagging with observed flavour
 Bd and Bs need fit of oscillation
Channel
Yield/2 fb-1
B/S
d /  ( 2fb-1 )
estimate
B+J/y(mm)K+
1.7 M
0.4
0.15%
B+D0p+
0.7 M
0.8
0.25%
B0J/y(mm)K*0(K+-)
0.7 M
0.2
0.2%
Bs Ds+ p-
0.08 M
0.3
0.7%
Bd0 D* - m+ n
9M
0.4
0.05%
B+ D0 (*) m + n
3.5 M
0.6
0.1%
Bs Ds(*) m + n
2M
0.1
0. 5%
Taggers can be calibrated using these control channels.
Topology close
to that of signal
channels
Semileptonics:
• High statistics
• More difficult
topology
Use of control channels
16
 B+ J/ K+ is a flavour specific channel
 No true MC information needed
 The  obtained in a given tagger for B+ J/ K+ can be used the same
taggers in other channels
B+  J/ K+
Opposite kaon
right
wrong
Bd  J/ Ks
right
wrong
tagger(K) (NNet)=a0 + a1 NNet
 Control channels will allow to measure  directly from data, with the
statistical accuracy required for physics measurements
Mistag extraction for
0
B
→ J/ Ks
17
One of the first measurements requiring flavour tagging of the B will be
sin 2 from B0J/ymmKS as a benchmark to demonstrate LHCb
capability in CP-asymmetry measurements
For the evaluation of the mistag rate, the following strategy, using B+J/ymmK+
and B0J/ymmK*0 as control channels, is foreseen:
•
With B+J/ymmK+ events determine for each tagger the dependence of the
mistag rate on the kinematical properties of the tagger.
Combine these probabilities into a single probability per event.
•
Use this function to subdivide B0J/ymmK*0 and B0J/ymmKS events into
5 samples of decreasing mistag-rate (tag categories).
•
Fit to flavour oscillations of B0J/ymmK*0 events, as a function of propertime,
in each of the 5 samples, to measure the mistag rate per category. Use these 5
mistag rates in the CP fit of B0J/ymmKS events, also subdivided into 5
categories.
Fit to flavour oscillations of B0  J/ K*0
in 5 categories
18
Cat 5
Cat 4
Only signal events
considered here
Cat 3
Cat 2
Cat 1
Control channel check
19
from MC truth from propertime fit
 Results from propertime fit are compatible to MC truth.
 In one year, 2/fb, with 215k events, s(sin2)~0.02
Background on control channels
20
 Control channels will be used with data events, where full account of
background has to be taken.
 We have devised the strategies to cope with it.
 For Bd+ J/ K*
Conclusions
21



Flavour tagging is a fundamental ingredient for
B physics measurements in LHCb.
Control channels will allow to measure  directly
from data, with the required statistical accuracy,
taking into account many possible effects
(backgrounds, trigger, etc.)
Expected effective tagging efficiency at LHCb is
~ 6 – 9 % for Bs and ~ 4 – 5 % for Bd,u channels