Observation of Bs-Bs Oscillations · Historic Review: A 20-Year Effort 1987 first evidence of B...
Transcript of Observation of Bs-Bs Oscillations · Historic Review: A 20-Year Effort 1987 first evidence of B...
Observation ofBs-Bs Oscillations
Christoph PausMassachusetts Institute of Technology
Wine and Cheese SeminarSeptember 22, 2006
Overview
Theory and some HistoryThe MethodEquipment Used for the MeasurementsOur Samplesb Flavor TaggingResults per SampleMixing Result SummaryConclusions
1
Theory and some History
2
Matter in the Standard ModelMatter build of families of fermion doublets
Leptons
�
νe
e−
�
L
�
νµµ−
�
L
�
νττ−
�L
Quarks
�
ud �
�
L
�
cs�
�L
�tb�
�L
Weak interaction through W ± bosons
cs
W+c
d
W+
In general: weak eigenstates ≠ strong eigenstatesmixing between families possiblelower quark doublet components absorb differenceneutrinos also mix
3
Cabibbo–Kobayashi–Maskawa MatrixExample: two families of quark pairs → one mixing angle�
d �
s�
�
=
�
cos θ sin θ− sin θ cos θ
��
ds
�
rotation matrix
Matrix has to be unitary: V �V = 1
Describe mixing between three quark-pair families���
d �
s�
b�
��� = V ×
���
dsb
��� with V =
���
Vud Vus Vub
Vcd Vcs Vcb
Vtd Vts Vtb
���
V is Cabbibo–Kobayashi–Maskawa matrix
Three families → 4 degrees of freedom3 angles1 complex phase → CP violation
4
CKM MatrixMatrix structure
mostly diagonalcrossing of families suppressedthe further the less probablevalues not predicted
Particles are conserved:V �V = 1
→ unitarity condition
u
d
c
t
s b
Wolfenstein parametrization (λ = 0.2272 ± 0.0010):
V =
���
1 − λ 2/2 λ Aλ 3(ρ − iη)−λ 1 − λ 2/2 Aλ 2
Aλ 3(1 − ρ − iη) −Aλ 2 1
��� + O(λ 4)
Least known parameters: ρ and η5
Unitarity Triangle
Unitarity condition: V �V = 1 V =
���
Vud Vus Vub
Vcd Vcs Vcb
Vtd Vts Vtb
���
→ VudV ∗ub + VcdV ∗
cb + VtdV ∗tb = 0
→ 1 + VudV ∗ub/VcdV ∗
cb + VtdV ∗tb/VcdV ∗
cb = 0
1
η
γ
≅*
*
*
* λ *
*
λ = sin θV V
V V
V
V V
V
ud
cd cb
ubcd cb
tbtd
c
ts
tdVV
α
βρ
6
Neutral B Meson Mixing
t⎯
t
(c, u)
(c⎯
, u⎯
)
W+ W−
s⎯
b
b⎯
s
Bs Bs
Vts
Vts W+
W−
s⎯
b
b⎯
s
(c⎯
, u⎯
)t
⎯
(c, u)tBs Bs
V
Vts
ts
Quark mixing → non diagonal Hamiltonian for ⟨B|H |B⟩
H =
�
M M12
M∗12 M
�
− i2
�Γ Γ12
Γ∗12 Γ
�
Diagonalizing the Hamiltonian results intwo masses: mH and mL and ∆m = mH − mL
two decay widths: ΓH and ΓL and ∆Γ = ΓH − ΓL
remember: Γ = 1/τ
Mass and decay width (lifetime) are measurable!!7
Theoretical Predictions - ∆mTheory prediction for B0/B0
s mix through box diagram
∆mq � mBqBBqf2Bq
��VtbV ∗tq
��2 q = s, d
Lattice QCD calculations
BBdf 2Bd
= (246 ± 11 ± 25) MeV2
Hadronic uncertainties limit|Vtd | determination to ≈ 11%
t⎯
t
(c, u)
(c⎯
, u⎯
)
W+ W−
s⎯
b
b⎯
s
Bs Bs
Vts
Vts
In ratio theory uncertainties are reduced∆ms∆md
= mBsmBd
ξ 2 |Vts|2
|Vtd |2 with ξ = 1.21 +0.047−0.035
Determine |Vts||Vtd | to ≈ 3.4%
8
Unitarity Triangle - Status EPS 2005Apex (ρ, η)
Squeezing alongside b
sin 2βVub/Vcb
Squeezing alongside c
∆md
∆ms
γ
CKM fit result:∆ms = 18.3 +6.5
−1.5 ps−1 -1.5
-1
-0.5
0
0.5
1
1.5
-1 -0.5 0 0.5 1 1.5 2
sin 2β
sol. w/ cos 2β < 0(excl. at CL > 0.95)
excluded at CL > 0.95
γ
γ
α
α
∆md
∆ms & ∆md
εK
εK
|Vub/Vcb|
sin 2β
sol. w/ cos 2β < 0(excl. at CL > 0.95)
excluded at CL > 0.95
α
βγ
ρ
η
excluded area has CL > 0.95
CK Mf i t t e r
EPS 2005
cb
9
First Measurement of B Mixing from UA1
Signature: like sign high pT leptons PLB 186 (1987) 247
Resulttime integratedχ = 0.121 ± 0.047implied heavy top
For Bs
too fast: χs = 0.5
Later Argus/Cleo/LEP/SLD/Tevatron/BaBar/Belle ....10
Historic Review: A 20-Year Effort1987
first evidence of B mixing from UA1 PLB 186 (1987) 247
Argus observes B0 mixing: UA1 implies large Bs mixing, mt > 50 GeV/c2PLB 192 (1987) 245
1989CLEO confirms Argus result PRL 62 (1989) 2233
1990sinclusive measurements of B0 mixing from LEP establish Bs mixing
1993first time dependent measurement of ∆md from Aleph PLB 313 (1993) 498
first lower limit on ∆ms from Aleph: ∆ms > 12 ⋅ 10−4 eV/c2PLB 322 (1994) 441
1999CDF Run I result on ∆ms: ∆ms > 5.8 ps−1
PRL 82 (1999) 3576
2005DØ first result on ∆ms: ∆ms > 5.0 ps−1
CDF Run II first result on ∆ms: ∆ms > 7.9 ps−1
2006D0 reports interval: ∆ms ∈ [17, 21] ps−1 at 90% CL PRL 97 (2006) 021802
CDF Run II first measurement ∆ms = 17.31 +0.33−0.18 ± 0.07 ps−1
PRL 97 (2006) 062003
11
The Method
12
A Picture Book Event
L xy
ct = L xymp
B
T
����������
����������
opposite side
K
K
π
π
0sB
Ds
+
+
K+
same side (vertexing)
opposite
D meson
fragmentationkaon
−l
side lepton
B hadronB jet
Collision Point
Creation of bb
side kaonopposite
K−
typically 1 mm
Ingredients to measure mixingproper decay time ct , B rest frameB flavor at decay, final stateB flavor at production, flavor tagging
13
An Event Display
14
B Mixing Phenomenology
Behavior in proper timeP(t)B0→B0 = 1
2τe−t /τ(1 + cos ∆mt)
P(t)B0→B0 = 1
2τe−t /τ(1 − cos ∆mt)
Determine asymmetry
A0(t) = N(t)unmixed−N(t)mixedN(t)unmixed+N(t)mixed
= cos ∆m t
In a perfect world
0
0.2
0.4
0.6
0.8
0 1 2 3proper decay time, t [ps]
prob
abili
ty d
ensi
ty
unmixedmixedtotal
-1
-0.5
0
0.5
1
0 1 2 3 4proper decay time, t [ps]
Mix
ed A
sym
met
ry
Perfect tag D=1
15
What Do We See in the End?
Flavor tagging Vertex Vertex and Momentum
-1
-0.5
0
0.5
1
0 1 2 3proper decay time, t [ps]
asym
met
ry
Realistic tag D=0.2
-1
-0.5
0
0.5
1
0 1 2 3proper decay time, t [ps]
asym
met
ry
Realistic resolution: vtx: 50 µm
-1
-0.5
0
0.5
1
0 1 2 3proper decay time, t [ps]
asym
met
ry
Realistic resolutions: vtx: 50 µm pt: σ(p)/p = 5%
1/σA =
�
nSεD2
2�
nSnS+nB
exp(−(∆msσct)22 )
σct =�
(σ0ct)2 +
�
ct σpp
2
16
Perfect to Realistic
-1
-0.5
0
0.5
1
0 1 2 3proper decay time, t [ps]
asym
met
ry
Perfect tag and resolutions
-1
-0.5
0
0.5
1
0 1 2 3proper decay time, t [ps]
asym
met
ry
Realistic tag D=0.2Realistic resolutions: vtx: 50 µm pt: σ(p)/p = 5%
Unbinned likelihood fit: p ∼ exp(−t /τ)(1 ± AD cos ∆mt)scan ∆m for signal: determine amplitude, Ameasure ∆ms with A = 1
17
Status: Scanning for Bs Oscillation SignalBefore this analysis
-1.5
-1
-0.5
0
0.5
1
1.5
2
2.5
0 2.5 5 7.5 10 12.5 15 17.5 20 22.5 25
∆ms (ps-1)
Am
plitu
de
data ± 1 σ 95% CL limit 14.4 ps-1
1.645 σ sensitivity 18.2 ps-1
data ± 1.645 σdata ± 1.645 σ (stat only)
Average for PDG 2006
before Spring 2005
]-1 [pssm∆0 5 10 15 20 25
Am
pli
tud
e
-4
-2
0
2
4 (stat.)σ 1.645 ±data
syst.)⊕ (stat. σ 1.645 ±data
σ 1 ±data
-195% CL limit: 14.8ps -1Expected limit: 14.1ps
DØ Run II
-11 fb
recent DØ resultPRL 97 (2006) 21802
17 ps−1 < ∆ms < 21 ps−1 at 90% CL
p-value about 5.0%
18
Last CDF Result PRL 97 (2006) 062003
]-1 [pssm∆0 10 20 30
Am
plitu
de
-2
0
2
σ 1 ± data
σ 1.645
σ 1.645 ± data
(stat. only)σ 1.645 ± data
95% CL limit
sensitivity
-116.7 ps-125.8 ps
-π +π +π _ s D→ 0
s, B+π _ s D→ 0
s X, B_ s D+ l→ 0
sB
CDF Run II -1L = 1.0 fb
A = 1.03 ± 0.28(stat) compatible with 1 for ∆ms ∼ 17.3 ps−1
19
Last CDF Result PRL 97 (2006) 062003
Signature: minimum of log(L(A = 1)) − log(L(A = 0))
)-1
(pssm∆15 16 17 18 19 20
log(
L)∆-
-5
0
5
10
15
20hadronicsemileptoniccombined
CDF Run II Preliminary -11 fb
Minimum: -6.75
p-value: 0.2%
Analysis was blinded
∆ms = 17.31 +0.33−0.18(stat) ± 0.07(syst) ps−1
20
Plan After April
Close box again
Refine analysis methods
Go for 5σ
21
What We Improved
Unchangedsame data: 1 fb−1
Improvementsflavor tagging
added opposite side kaon taggerapplied NN to combine all opposite side taggersapplied NN to same side tagger
signal yieldsadded partially reconstructed decaysused particle identification in selectionused NN for hadronic selectionadded trigger path in �D−
s
Added effective statistics of factor of 2.522
Equipment Used for theMeasurements
23
CDF II Detector
24
CDF II Detector - Key Features
’Deadtimeless’ trigger system3 level, pipelined, flexible systemSilicon Vertex Trigger (SVT) at 2nd level (≈25 kHz)
Charged particle reconstructionredundancy for pattern reco in busy environmentexcellent momentum resolution: R = 1.4m,B = 1.4Texcellent vertex resolution: L00 at 1.5cm
Particle identificationenergy loss in drift chamber (dE /dx)Time-of-Flight system at 1.4 m radiuselectron and muon identification
25
Our Samples
26
Sample Luminosity
Store Number
Tot
al L
umin
osity
(pb
-1)
0
250
500
750
1000
1250
1500
1750
2000
1000 1500 2000 2500 3000 3500 4000 4500
1 4 7 10 1 4 7 101 4 7 1 4 7102002 2003 2004 2005 2006Year
Month
DeliveredTo tape
Bs Mixing Analysis uses: 1 fb−1
now: 1.8 fb−1 delivered, 1.6 fb−1 on tape27
Samples Used in the Analysis
Sample to tune flavor taggers (largest)
�+track sample (track is required to be displaced)
Calibration samples (taggers, vertex resolutions)B+ → J/ψK +, D
0π+(π+π−), D0
�
+X ,B0 → J/ψK ∗0, D−π+(π+π−), D−
�
+X , D∗−
�
+XBs → J/ψφwith D∗− → D0π−, D0 → K −π+(π+π−), D− → K +π−π−
Signal sampleshadronic: Bs → D−
sπ+(π+π−)semileptonic: Bs → D−
s�
+Xwith D−
s → φπ−, K ∗0K −, π+π−π−
28
Samples - Semileptonic Versus Hadronic
]2
) [GeV/cπ,πφmass(5.2 5.4 5.6 5.8
2ca
ndid
ates
per
10
MeV
/c
0
100
200
300
400
data
fit+/K+π -
s D→ s B
+ρ-s D→ s B
+/K+π*-s D→ s B
X-s D→ b
+π - D→ 0 B-π +
cΛ → bΛ
comb. bkg.
]2
) [GeV/cπ,πφmass(5.2 5.4 5.6 5.8
-1CDF Run II L = 1 fb
Hadronic Dsπ(ππ)reconstruct: ct = Lxy
mB
pBT
great ct , mass resolutionsample is cleansmall branching ratio
]2
mass [GeV/c--l+π φ
3 4 5
2ca
ndid
ates
per
35
MeV
/c
0
500
1000
1500
2000 data
fit
signal0sB
false lepton & physics
comb. bkg.
]2
mass [GeV/c--l+π φ
3 4 5
]2
mass [GeV/c+π φ1.94 1.96 1.98 2.00
2ca
nd. p
er 1
MeV
/c
0
1000
2000
3000
4000
]2
mass [GeV/c+π φ1.94 1.96 1.98 2.00
Semileptonic �DsXreconstruct: ct∗ = Lxy
mB
p�DT
large branching ratioinferior ct , mass resolutionsample composition issue
29
b Flavor Tagging
30
Flavor Tagging Introduction
L xy
ct = L xymp
B
T
����������
����������
opposite side
K
K
π
π
0sB
Ds
+
+
K+
same side (vertexing)
opposite
D meson
fragmentationkaon
−l
side lepton
B hadronB jet
Collision Point
Creation of bb
side kaonopposite
K−
typically 1 mm
Crucial parametersefficiency: ε, dilution: D = 1−2pw (pw prob. for wrong decision)εD2 expresses statistical power
31
Flavor Tagging IntroductionTagging algorithms used
opposite side: electron, muon, jet charge, kaonsame side: pion (B+, B0), kaon (Bs)
Tuning the algorithmbased on huge heavy flavour rich �+track samplefind dependencies and determine parametrizationsfind optimal point for the algorithm
Calibration of the taggerverify algorithms on B0 and B+ semileptonic/hadronic samplesdetermine scale factor S inB+ : p ∼ exp(−t /τ)(1 ± SD)B0 : p ∼ exp(−t /τ)(1 ± SD cos ∆mdt)use single scale factor: should be consistent with 1
Scheme for combination of taggers32
OST Flavor TaggersIndividual performance and combination (�+track sample)
tagger [%] efficiency dilution εD2
Muon 4.6 ± 0.0 34.7 ± 0.5 0.58 ± 0.02Electron 3.2 ± 0.0 30.3 ± 0.7 0.29 ± 0.01JQT 95.5 ± 0.1 9.7 ± 0.2 0.90 ± 0.03Kaon 18.1 ± 0.1 11.1 ± 0.9 0.23 ± 0.02OST old 95.6 ± 0.1 11.9 ± 0.1 1.34 ± 0.03OST NN 95.8 ± 0.1 12.7 ± 0.2 1.54 ± 0.04
Opposite Side Taggersnew kaon taggernot mutually exclusive → hierarchical schemenew NN combination: tag decisions as input
33
OST Flavor Taggers - Neural Network
predicted dilution0 0.05 0.1 0.15 0.2 0.25 0.3
mea
sure
d di
lutio
n
0
0.05
0.1
0.15
0.2
0.25
0.3
/ ndf 2χ 45.06 / 9S 0.008463± 0.9917
/ ndf 2χ 45.06 / 9S 0.008463± 0.9917 CDF Run II Preliminary
combined OS tagger
0.01±S = 0.99
= 96 %∈
Linear parametrization works wellImprovement of εD2
hadronic 1.81 ± 0.10% from 1.51%semileptonic 1.82 ± 0.04% from 1.54%
34
Same Side Kaon TaggingFragmentation
Bd /u likely accompanied by π+/π−
Bs likely accompanied by a K +
processes differno direct transfer B+,B0 → Bs
need MC to measure tagger dilution
Strategytune MC with B+ and B0
apply PID to de-weight pionsuse MC to parametrize dilution
ParticleID very importantsignificantly improves εD2
reduces systematic uncertainty
→TOF (dE /dx) very important!
+
b b
b b
b b
d
u s
u
s
d s
u d
B0
B0
B
K+
0∗K
K0∗
K
d
u s
d s
u
s
u d
}
}
}
}
}
}
s
π
π
35
Flavor Tagging - SSKT Calibration: B+, B0
max PID dilution D [%]
5 10 15 20 25 30
max PID dilution D [%]
5 10 15 20 25 30
CDF Run II Preliminary -1 355 pb≈L
+ Kψ J/→ +B
+π 0
D → +B
π 30
D → +B
*0 Kψ J/→ 0B
+π - D→ 0B
π 3- D→ 0B
data
MC
syst.
find good agreement: particle Id and kinematic variablesuse kinematic variables to improve tagger
36
SSKT Particle Id Algorithm
log(LH(PID))
-20 -10 0 10 20
entr
ies
per
bin
0
200
400
Pythia Data MC pions MC kaons MC protons
log(LH(PID))
-20 -10 0 10 20
entr
ies
per
bin
0
200
400
+π -s D→ sB
CDF Run II Preliminary -1 1 fb≈L
[GeV/c]Tp0 1 2 3
entr
ies
per
bin
0
100
200
Pythia Data MC pions MC kaons MC protons
[GeV/c]Tp0 1 2 3
entr
ies
per
bin
0
100
200
+π -s D→ sB
CDF Run II Preliminary -1 1 fb≈L
[GeV/c]Tp0 1 2 3
entr
ies
per
bin
0
10
20
Pythia Data MC pions MC kaons MC protons
PID > 1 [GeV/c]Tp
0 1 2 3
entr
ies
per
bin
0
10
20
+π -s D→ sB
CDF Run II Preliminary -1 1 fb≈L
+πs- -> DsB
log(LH(PID)) of tagging track-20 -10 0 10
dilu
tion
[%] (
agre
emen
t)
-10
0
10
20
30
40
50
60
70
CDF Run II Monte Carlo
Pid algorithm works well
Kinematic of tag track not
used
algorithm)relL [GeV/c] (max pTtagging track p
1 2 3 4 5 6
dilu
tion
(agr
eem
ent)
0.1
0.2
0.3
0.4
0.5
algorithm)relL [GeV/c] (max pTtagging track p
1 2 3 4 5 6
dilu
tion
(agr
eem
ent)
0.1
0.2
0.3
0.4
0.5
(MC)+π -s D→ s B
CDF Run II Monte Carlo
ex.: max. prelL algorithm shows dependencies → use it
37
Neural Network SSKTAlgorithm
variables: pid, ∆R, pT , prelL , prel
T , b (bool tags have same charge)train: signal - RS kaons, bg - WS kaons, pions and protonsdecision: track charge of highest NN tag candidate
Expected improvementMC ≈ 6% relative
Measured improvement0% relative hadronic→ εD2 = 3.5%8% relative semileptonic→ now εD2 = 4.8%
Neural Network output0 0.2 0.4 0.6 0.8 1
Dilu
tion
[%] (
agre
emen
t)
0
0.2
0.4
0.6
0.8
1CDF Run II Monte Carlo
)πφ(s D→ sB
38
Results per Sample
39
Samples - Semileptonic SelectionUse of particle identification in selection
standard practice at B factoriesapplied in hadronic and semileptonic analysesusing combined Time-of-Flight and dE/dx (see SSKT)
Effectsstrongest in Ds → K ∗0Kno explicit D+ → K −π+π+
rejection (+35%)combinatorial backgrounddominated by π
Yields: 62k (was 37k)S/Bx2 for D−
s → K ∗0K −, φπ−
added trigger paths]
2 mass [GeV/c--l+ K*0K
3 4 5
2ca
ndid
ates
per
35
MeV
/c
0
2000
4000
data
fit
signal0sB
false lepton & physics reflection+ D
comb. bkg.
]2
mass [GeV/c--l+ K*0K3 4 5
]2
mass [GeV/c+ K*0K1.95 2.00
2ca
nd. p
er 2
MeV
/c
0
2000
4000
6000
]2
mass [GeV/c+ K*0K1.95 2.00
40
Semileptonic Amplitude Scan
]-1 [pssm∆
0 5 10 15 20 25 30 35
Am
plitu
de
-4
-3
-2
-1
0
1
2
3
4σ 1 ± data
σ 1.645
σ 1.645 ± data
(stat. only)σ 1.645 ± data
95% CL limit
sensitivity
-116.5 ps-119.3 ps
X_ s D+ l→ 0
sB
A compatible with 1 for ∆ms ∼ 17.75 ps−1
A/σA(∆ms = 17.75 ps−1) ≈ 2, can set two sided 95% CL limit
41
Semileptonic Likelihood Profile
Minimum17.89 ps−1
Depthsets 95% CL interval1σ equiv. to ± 0.3 ps−1
]-1 [pssm∆15 16 17 18 19 20
-log(
L)
-2
-1
0
1
2
3
4
42
Samples - Hadronic: Bs → D−sπ+(π+π−)
Bs Modes Signal S/B(φπ)π 2000(1600) +13%partial 3100(–) –(K ∗0K )π 1400(800) +35%(3π)π 700(600) +22%(φπ)3π 700(500) +92%(K ∗0K )3π 600(200) +110%(3π)3π 200(–) –total 8700(3700) –
Golden signatureBs → D−
s(φπ−)π+
]2
) [GeV/cπ,πφmass(5.2 5.4 5.6 5.8
2ca
ndid
ates
per
10
MeV
/c
0
100
200
300
400
data
fit+/K+π -
s D→ s B
+ρ-s D→ s B
+/K+π*-s D→ s B
X-s D→ b
+π - D→ 0 B-π +
cΛ → bΛ
comb. bkg.
]2
) [GeV/cπ,πφmass(5.2 5.4 5.6 5.8
-1CDF Run II L = 1 fb
Improvements:partial reconstruction, particle Id, NN in selection
43
How Good are Partially Reconstructed Decays?
sB
T/pReconstructedT = pκ
0.4 0.6 0.8 1.0
prob
abili
ty d
ensi
ty
5
10
ν l (*)s D→ 0
sBall 2 3.1 GeV/c≤ lsD 2.0 < m2 4.5 GeV/c≤ lsD 4.3 < m2 5.1 GeV/c≤ lsD 4.9 < m
π *s D→ 0
sB
ρ s D→ 0sB
CDF Run II
sB
T/pReconstructedT = pκ
0.4 0.6 0.8 1.0
Proper decay time [ps]0 1 2 3
Pro
per
deca
y tim
e re
solu
tion
[fs]
0
200
400
600
800
-1Osc. / 18 ps
π s D→ 0sB
ρ s / Dπ *s D→ 0
sB
2 5.1 GeV/c≤ lsD, 4.9 < mν l (*)s D→ 0
sB
2 4.5 GeV/c≤ lsD, 4.3 < mν l (*)s D→ 0
sB
2 3.1 GeV/c≤ lsD, 2.0 < mν l (*)s D→ 0
sB
CDF Run II
Decay modes: Bs → D∗−s (Dsγ(π0))π+, Bs → D−
sρ+(π0π−)need to correct for missing momentum, as semileptoniclarge reconstructed mass → missing momentum small
Partial reco almost as good as full reco44
Amplitude Scan Partial Reconstruction Only
]-1 [pssm∆0 10 20 30
Am
plitu
de
-4
-2
0
2
4 σ 1 ± data
σ 1.645
σ 1.645 ± data
(stat. only)σ 1.645 ± data
95% CL limit
sensitivity
-117.0 ps-118.3 ps
+π_*
s,D+ρ_ s D→ 0
sB
CDF Run II Preliminary -1L = fb
A = 1.02 ± 0.57(stat)consistent with 1 for ∆ms ∼ 17.75 ps−1
45
Amplitude Scan Golden Mode Only
]-1 [pssm∆0 10 20 30
Am
plitu
de
-4
-2
0
2
4 σ 1 ± data
σ 1.645
σ 1.645 ± data
(stat. only)σ 1.645 ± data
95% CL limit
sensitivity
-117.0 ps-122.7 ps
+π_*
s,D+ρ_ s,D+π
_ s D→ 0
sB
CDF Run II Preliminary -1L = fb
A = 1.27 ± 0.34(stat)consistent with 1 for ∆ms ∼ 17.75 ps−1
46
Likelihood Curves - Golden Channel Only
Golden LikelihoodMinimum: −7.4p-value: 0.15%
→ 3.2σ
Spring analysisMinimum: −6.4p-value: 0.2%
Golden sample
∆ms = 18.01 +0.17−0.18 ps−1
]-1 [pssm∆16 18 20
-ln(L
)
-8
-6
-4
-20
2
4
6
8
10
12
14Total golden
Part.Reco’d
Full.Reco’d
Consistent results: full and partial reconstruction47
Non-Golden Channels, Bs → Dsπ
Particle Idopt. D+ rejectionrelax kinematiccuts
Neural Netlarger signal atsame bgadd new modeBs → Ds(3π)3π
]2 mass [GeV/cBπ -) K-π +(K*K
5.4 5.6 5.82
Can
dida
tes
per
5 M
eV/c
0
50
100
150
data
fit+π -
s D→ s B
background
π - D→ 0 B
π cΛ → bΛ
]2 mass [GeV/cBπ -) K-π +(K*K
5.4 5.6 5.8
/ NDF = 83.54 / 60, Prob = 2.40%2χ
]2 mass [GeV/cBπ -π -π +π5.4 5.6 5.8
2C
andi
date
s pe
r 5
MeV
/c
0
50
100 data
fit
+π -s D→ s B
background
π cΛ → bΛ
]2 mass [GeV/cBπ -π -π +π5.4 5.6 5.8
/ NDF = 71.40 / 64, Prob = 24.55%2χ
proper time [cm]0 0.2 0.4
mµC
andi
date
s pe
r 50
1
10
210
data
fit+π -
s D→ s B
background
π - D→ 0 Bπ cΛ → bΛ
proper time [cm]0 0.2 0.4
/ NDF = 46.61 / 29, Prob = 2.04%2χ
proper time [cm]0 0.2 0.4
mµC
andi
date
s pe
r 50
1
10
210
data
fit+π -
s D→ s B
backgroundπ cΛ → bΛ
proper time [cm]0 0.2 0.4
/ NDF = 35.48 / 25, Prob = 7.99%2χ
48
Non-Golden Channels, Bs → Ds3π
]2 mass [GeV/cBπ 3-π) - K+
(Kφ5.4 5.6 5.8
2C
andi
date
s pe
r 5
MeV
/c
0
50
100
data
fitπ 3s D→ s B
background
π 3- D→ 0 Bπ 3cΛ → bΛ
satellites
]2 mass [GeV/cBπ 3-π) - K+
(Kφ5.4 5.6 5.8
/ NDF = 41.73 / 45, Prob = 61.14%2χ
]2 mass [GeV/cBπ 3-
) K-π +(K*K
5.4 5.6 5.82
Can
dida
tes
per
5 M
eV/c
0
50
100 data
fitπ 3s D→ s B
background
π 3- D→ 0 Bπ 3cΛ → bΛ
satellites
]2 mass [GeV/cBπ 3-
) K-π +(K*K
5.4 5.6 5.8
/ NDF = 41.67 / 39, Prob = 35.54%2χ
]2
mass [GeV/c-π +π + π) π (3sD5.4 5.6 5.8
2C
andi
date
s pe
r 5
MeV
/c
0
10
20
30
data
fitπ 3s D→ s B
backgroundπ 3cΛ → bΛ
satellites
]2
mass [GeV/c-π +π + π) π (3sD5.4 5.6 5.8
/ NDF = 24.54 / 24, Prob = 43.13%2χ
proper time [cm]0 0.2 0.4
mµC
andi
date
s pe
r 50
1
10
210 data
fitπ 3s D→ s B
background
π 3- D→ 0 Bπ 3cΛ → bΛ
proper time [cm]0 0.2 0.4
/ NDF = 28.01 / 26, Prob = 35.82%2χ
proper time [cm]0 0.2 0.4
mµC
andi
date
s pe
r 50
1
10
210 data
fitπ 3s D→ s B
background
π 3- D→ 0 Bπ 3cΛ → bΛ
proper time [cm]0 0.2 0.4
/ NDF = 40.82 / 24, Prob = 1.74%2χ
proper time [cm]0 0.2 0.4
mµC
andi
date
s pe
r 50
1
10
data
fitπ 3s D→ s B
backgroundπ 3cΛ → bΛ
proper time [cm]0 0.2 0.4
/ NDF = 19.38 / 18, Prob = 36.86%2χ
49
Performance of Neural Network Selection
2ca
ndid
ates
per
20.
0 M
eV/c
500
1000
1500
2000
2500
3000
CDF Run II Preliminary
Neural Network
-1L = 1.0 fb
Cut Based
]2candidate mass [GeV/c4.8 5 5.2 5.4 5.6 5.8 6
2ca
ndid
ates
per
20.
0 M
eV/c
0
500
1000
1500
2000
2500
3000
]2candidate mass [GeV/c4.8 5 5.2 5.4 5.6 5.8 6
2ca
ndid
ates
per
20.
0 M
eV/c
0
500
1000
1500
2000
2500
3000 Neural Network, butnot Cut Based
]2candidate mass [GeV/c4.8 5 5.2 5.4 5.6 5.8 6
Cut Based, butnot Neural Network
50
Amplitude Scan Non-Golden Modes
]-1 [pssm∆
0 5 10 15 20 25 30 35
Am
plit
ud
e
-3
-2
-1
0
1
2
3σ 1 ± data
σ 1.645
σ 1.645 ± data
(stat. only)σ 1.645 ± data
95% CL limit
sensitivity
-117.1 ps-128.3 ps
A = 1.29 ± 0.29(stat)consistent with 1 for ∆ms ∼ 17.75 ps−1
51
Likelihood Curves - Non-Golden Channels
Non-Golden LikelihoodMinimum: −9.6
Spring analysisMinimum: −6.4p-value: 0.2%
Non-Golden sample
∆ms = 17.66 ± 0.11 ps−1
]-1 [pssm∆15 16 17 18 19 20
log(
L)∆
-10
-5
0
5
10
15
CDF Run II Preliminary -1L = 1.0 fb
Consistent result with Golden Only
52
Combined Amplitude ScansSemileptonic Hadronic
]-1 [pssm∆
0 5 10 15 20 25 30 35
Am
plitu
de
-4
-3
-2
-1
0
1
2
3
4σ 1 ± data
σ 1.645
σ 1.645 ± data
(stat. only)σ 1.645 ± data
95% CL limit
sensitivity
-116.5 ps-119.3 ps
X_ s D+ l→ 0
sB
]-1 [pssm∆
0 5 10 15 20 25 30 35
Am
plitu
de
-2
-1.5
-1
-0.5
0
0.5
1
1.5
2σ 1 ± data
σ 1.645
σ 1.645 ± data
(stat. only)σ 1.645 ± data
95% CL limit
sensitivity
-117.1 ps-130.7 ps
Signal at ≈ 17.75 in hadronic analysis
Noteworld best semileptonic analysis: 19.3 ps−1
hadronic analysis in different league: 30.7 ps−1
53
Combined Amplitude Scan
]-1 [pssm∆
0 5 10 15 20 25 30 35
Am
plitu
de
-2
-1.5
-1
-0.5
0
0.5
1
1.5
2σ 1 ± data
σ 1.645
σ 1.645 ± data
(stat. only)σ 1.645 ± data
95% CL limit
sensitivity
-117.2 ps-131.3 ps
A = 1.21 ± 0.20(stat) compatible with 1 for ∆ms ∼ 17.75 ps−1
A/σA(∆ms = 17.75 ps−1) = 6.05, but what is the p-value?
54
Likelihood Profile
Difference, −∆ log(L)
log(L(A = 1)) − log(L(A = 0))
Minimum: −17.26
]-1 [pssm∆0 5 10 15 20 25 30 35
log(
L)∆
-30
-20
-10
0
10
20
30 data
expected no signal
expected signal
Key question:How often can random tags produce a minimum at least as deep?
55
Likelihood Significance
min log(L)∆
-30 -25 -20 -15 -10 -5 00
2
4
6
8
10
12
14
610×
observed
randomly tagged
=17.77sm∆signal
min log(L)∆-20 -15 -10 -5 0p-
valu
e
-910
-810
-710
-610
-510
-410
-310
-210
-110
1
σ5
28 trials out of 350 millionp-value ≈ 8 × 10−8 corresponding to 5.4σ
(5 standard deviations is = 5.7 × 10−7)→ passed observation criterion
56
∆ms Measurement∆ms = 17.77 ± 0.10(stat) ± 0.07(syst) ps−1
was submitted to PRL on Monday
]-1 [pssm∆15 16 17 18 19 20
log(
L)∆
-10
0
10
20
30combinedhadronic
semileptonic
Systematicwell behavedct scale uncertaintyrest: small
Agrees with SM18.3 +6.5
−1.5 ps−1EPS 2005
Agrees with 1st result17.31 +0.33
−0.18(stat) ± 0.07(syst) ps−1
PRL 97 (2006) 62003
57
Visualizing the Result
∆ms = 17.77 ± 0.10(stat) ± 0.07(syst) ps−1
[ps]sm∆/πDecay Time Modulo 20 0.05 0.1 0.15 0.2 0.25 0.3 0.35
Fitt
ed A
mpl
itude
-2
-1
0
1
2
data
cosine with A=1.28
CDF Run II Preliminary -1L = 1.0 fb
Proper decay time folded onto a 2π interval
58
Direct Measure of |Vtd /Vts|Relation between ∆mq and Vtq
∆ms∆md
= mBsmBd
ξ 2|VtsVtd
|2
InputsmBdmBs
= 0.98390 PRL 96 (2006) 202001
ξ = 1.21 +0.047−0.035 M.Okamoto hep-lat/0510113
∆md = 0.507 ± 0.005 PDG 2006
|Vtd ||Vts| = 0.2060 ± 0.0007(exp) +0.0081
−0.0060(theo)
Best so far Belle: PRL 96 221601 (2006)
|Vtd ||Vts| = 0.199 +0.026
−0.025(exp) +0.018−0.016(theo)
59
ConclusionsLong journey ended
19 years of search to see Bs-Bs oscillationsfound signal consistent with Bs-Bs oscillationssignificance: 5.4σ corresponding to p = 8 × 10−8
]-1 [pssm∆
0 5 10 15 20 25 30 35
Am
plitu
de
-2
-1.5
-1
-0.5
0
0.5
1
1.5
2σ 1 ± data
σ 1.645
σ 1.645 ± data
(stat. only)σ 1.645 ± data
95% CL limit
sensitivity
-117.2 ps-131.3 ps
∆ms = 17.77 ± 0.10(stat) ± 0.07(syst)60
Accomplishments in the Course of the AnalysisPhD Theses
C.Chen UPenn Charm Cross Section MeasurementI.Furic MIT Reconstruction of Bs → D−
sπ+
A.Bolshov MIT Reconstruction of Bs → D−s3π
S.DaRonco Padova Lifetimes from Exclusive ReconstructionJ.Piedra Cantabria B0 Mixing and Tagger CalibrationG.Giurgu CMU Muon TaggingV.Tiwari CMU Electron TaggingD.Usynin UPenn Charged Particle Composition Around B MesonsG.Salamanna Rome I Opposite Side Kaon TaggingA.Belloni MIT SSKT and Neural Network for Hadronic DecaysN.Leonardo MIT Likelihood FrameworkJ.Miles MIT Proper Time Resolution and Partial DecaysG.Di Giovanni Paris Same Side Kaon Tagging (next gen. analysis)B.Casal Cantabria Neural Network (next gen. analysis)
2006 Tollestrup AwardG.Ceballos (Cantabria), I.Furic (Chicago), S.Menzemer (MIT/Cantabria)
61
The End
62