Search for Bsμμ and Bdμμ with 2 fb-1 of data

Slava KrutelyovUCSBwith

D.Glenzinski, H.Held (KIT), C.-J.Lin (FNAL)T.Kamon, M.Weinberger (TAMU)

M.Herndon (UWisconsin)

SUSY-Group Meeting08/29/2007

• Motivation and history• Analysis

– Setup

– Event selection: signal discriminants• muon ID with LH• Neural Net

– (optimized) Expectations and results

• Effects on SUSY

CDF 8925

blessed on Aug 09

08/29/2007 S. Krutelyov Search for Bs,d-->mumu with 2 fb-1 2

08/29/2007 S. Krutelyov Search for Bs,d-->mumu with 2 fb-1 3

• MSSM

– With RPV: can proceed at tree level• couplings not constrained much otherwise

– In general case has a diagram ~ tan6/MH4

• tan4/MH4 come from MSSM is a 2HDM

• remaining tan2is from SUSY (stop L-R mixing)

Just need large tan not too-high MH, some stop mixing

SO(10) with Yukawa unification tan mSUGRA with large tan

• Other BSM: with no extra flavor structure the enhancement < ~ a factor of 2

• With improving limits start asking any theorist “Is it allowed by Bsμμ?”

• More in the end of the talk

Motivations: Models

’i23 i22

b

s

Bsmm appears to get into the “model exclusion hall of fame”

Near 200 citations (combined for CDF/DØ RunII) in the past ~3 years

08/29/2007 S. Krutelyov Search for Bs,d-->mumu with 2 fb-1 4

Bits of history• Search for Bsis as old as hadron colliders

– First looked at by UA1• Recent limits are all from Tevatron

– DØ is often close to CDF

Prior

08/29/2007 S. Krutelyov Search for Bs,d-->mumu with 2 fb-1 5

Bs(d)Ingredients•Normalize to B+J/ K+ decays

• Construct discriminant to select Bs and suppress bgd• MC sim for signal• Data mass sidebands for bgd estimate• Minimize expected 90% C.L. upper limit

)()(

)()(

KBBRf

f

N

NBBR

s

u

B

totalB

totalBs

Bss

Norm

aliz

ati

on

mod

e

Bd Bs

N(UU+UX) ~ 3×104

Sig

nal sa

mple

(base

line)

08/29/2007 S. Krutelyov Search for Bs,d-->mumu with 2 fb-1 6

x 2 cut use muon LH (LH>0.1) Combined dX, dPhi, dZ, EM, HAD

• add dE/dX cut– Z>-0.2

• Suppress combinatorics– ~x2 less events (after preselection)

• Bring Bhh below the horizon– Bhh fakes are ~0.05 for Bs and ~0.5 for Bd windows

• cf. ~ 0.3 (Bs) and ~2 (Bd) with x 2 cut

Improving sensitivity: muon ID

08/29/2007 S. Krutelyov Search for Bs,d-->mumu with 2 fb-1 7

Normalization mode

•B+J/ψK+ + c.c.

08/29/2007 S. Krutelyov Search for Bs,d-->mumu with 2 fb-1 8

Discriminating variables•Used as inputs to Neural Net

R=1

μ

μ

p

Bs

i

iTT

T

pp

pI

pT

pT

pT

08/29/2007 S. Krutelyov Search for Bs,d-->mumu with 2 fb-1 9

Preselections: signal+sidebandsVertex Requirements:

• 2(vtx) < 15• (L3D) < 150 m• L3D < 1 cm• 0 < < 3 mm ( = proper decay length)• > 2• pT(B)>4 GeV && |y(B)|<1• 4669 < M() < 5969 MeV

Bs Search Sample:N(CMU-CMU) = 15594N(CMU-CMX) = 15072

(completely Bgd dominated)

In addition:• Isolation>0.5• Pointing angle <0.7

Note: without muon LH & dE/dX~x2 more backgrounds here

08/29/2007 S. Krutelyov Search for Bs,d-->mumu with 2 fb-1 10

Signal MC•Use Pythia

–BGen for systematics

•Reweigh to follow pT distribution from data•Compare using B+ data and MC

08/29/2007 S. Krutelyov Search for Bs,d-->mumu with 2 fb-1 11

NN: inputs•After applying pre-selections use sidebands (bgd) and signal MC to train NN

•6 inputs to NN training–λ – lifetime–λ/σ – lifetime significance–Δα – 3D poining (vtx vs

momentum)– Iso

–pT(B)

–pT(μ) min – lower of the pair

08/29/2007 S. Krutelyov Search for Bs,d-->mumu with 2 fb-1 12

NN inputs: x-check w data B+

plots

• Input variables are well-modeled– including pT(μ) high used in the trial NN

08/29/2007 S. Krutelyov Search for Bs,d-->mumu with 2 fb-1 13

NN: output

•Compared to LH (used previously) get a better (~2 times) rejection at the same efficiency–LH had 3 vars: Iso, P(λ), Δα

–5 var NN: add λ/σ and pT(B)

–7 var NN includes both pT(μ)

•introduces mass biasnot used

08/29/2007 S. Krutelyov Search for Bs,d-->mumu with 2 fb-1 14

NN output: x-check for biases

•The correlation is negligible•No evidence of a bias here or in other x-checks

– checked w/ control samples, trained on part of sidebands, bootstrap–

use of sidebands is justifiable

08/29/2007 S. Krutelyov Search for Bs,d-->mumu with 2 fb-1 15

NN outputs: x-check w data B+

•Good agreement

08/29/2007 S. Krutelyov Search for Bs,d-->mumu with 2 fb-1 16

Control Samples•Independent bkg control samples to cross check bkg estimate procedure

•OS - : Opposite-sign dimuon sample with ct<0

•SS+ : Same-sign dimuon sample with ct>0

•SS- : Same-sign dimuon sample with ct<0

•FM : Fake muon sample ct>0– (require at least one muon leg to fail our muon likelihood and

dE/dX requirement)

08/29/2007 S. Krutelyov Search for Bs,d-->mumu with 2 fb-1 17

Control Samples: NN x-checks

• NN performance: observation is consistent with expectations• Using a wider ± 150 MeV signal window for cross-check• For obsv<pred, prob= probability for making an observation ≤ obsv

08/29/2007 S. Krutelyov Search for Bs,d-->mumu with 2 fb-1 18

Applying NN: getting expected bgd

•Shape in sidebands give contribution of combinatorial bgd to signal bins

•Separate fit in each NN bin: pol1 in bins 1, 2; pol0 in bin 3

CMU-CMU CMU-CMX

08/29/2007 S. Krutelyov Search for Bs,d-->mumu with 2 fb-1 19

Inputs to the (exp./obs.) limit

•Inputs to be used in the “main” formula

•The biggest uncertainty is from fu/fs (after bgd stat)

08/29/2007 S. Krutelyov Search for Bs,d-->mumu with 2 fb-1 20

Optimization: choice of NN bins

•Relative (unnormalized) numbers are shown•The choice (of 3 bins) is

–[0.8, 0.95]U(0.95, 0.995]U(0.995, 1]

08/29/2007 S. Krutelyov Search for Bs,d-->mumu with 2 fb-1 21

Bhh contributions

•The contribution to Bs is small, considerable for Bd

Contribution to Bdμμ

Contribution to Bsμμ

08/29/2007 S. Krutelyov Search for Bs,d-->mumu with 2 fb-1 22

Expectations and observations: Bs

08/29/2007 S. Krutelyov Search for Bs,d-->mumu with 2 fb-1 23

Expectations and observations: Bd

08/29/2007 S. Krutelyov Search for Bs,d-->mumu with 2 fb-1 24

Results: limits

•New world-best result

08/29/2007 S. Krutelyov Search for Bs,d-->mumu with 2 fb-1 25

Results: 2D plots

•P-value with bgd-only hypothesis, 10K pseudoexperimentsCMU-CMU CMU-CMX

Bs Bd•Observed events are consistent with expectation

08/29/2007 S. Krutelyov Search for Bs,d-->mumu with 2 fb-1 26

Results: 2D plot combined

08/29/2007 S. Krutelyov Search for Bs,d-->mumu with 2 fb-1 27

Results: 1D plots

•Bs–2 events in CMU-CMU channel–1 event in CMU-CMX channel

08/29/2007 S. Krutelyov Search for Bs,d-->mumu with 2 fb-1 28

Limits history

•Getting better and better– faster than 1/sqrt(Lumi), slower than 1/Lumi

•Unofficial Tevatron combination: Br(Bsμμ) < 4.5x10-8 @ 95%

08/29/2007 S. Krutelyov Search for Bs,d-->mumu with 2 fb-1 29

08/29/2007 S. Krutelyov Search for Bs,d-->mumu with 2 fb-1 30

SUSY contemplations (1)

•The importance of Bsμμ grows with tanβ

•SUSY with large tanβ becomes more popular–driven by a more significant aμ now at ~3.4 σ away from SM

•The SO(10) scenario by Raby et al. is practically excluded now

R. Dermisek et al., hep-ph/0507233 (2005)

[easy model to exclude]

SO(10)

08/29/2007 S. Krutelyov Search for Bs,d-->mumu with 2 fb-1 31

SUSY contemplations (2)•Light MSSM Higgs scenario is next in line for significant exclusion.

Belyaev et al.SUSY07

08/29/2007 S. Krutelyov Search for Bs,d-->mumu with 2 fb-1 32

CDF Limit

5.8x10-8

2 fb-1

SUSY contemplations (3)

•The limit eats further into the mSUGRA space now–not as significant with bsγ

constraint considered

08/29/2007 S. Krutelyov Search for Bs,d-->mumu with 2 fb-1 33

SUSY contemplations (4)

•Start getting beyond bsγ at high tanβ•Still a way to go 2x10-8 will be way more noticeable

08/29/2007 S. Krutelyov Search for Bs,d-->mumu with 2 fb-1 34

Summary•We have updated the Bmm analysis with 2 fb-1

• The limits improved significantly:– Br(Bsmm)<5.8×10-8 @ 95%CL– Br(Bdmm)<1.8×10-8 @ 95%CL

• Both limits are the best in the world!!!!

• Plan to prepare/publish in PRL–draft in progress ( to GPs this week?)

08/29/2007 S. Krutelyov Search for Bs,d-->mumu with 2 fb-1 35

backup