1. Riunione della CNS1 a Bari
MEG physics analysis
Gianluca Cavoto
INFN Roma
Riunione della CNS1 a Bari
19-24 settembre 2011
Sep 20th 2011
Gianluca Cavoto

2. Outline Update since Feb analysis 2009 and 2010 data analysis
See A.Baldini talk at CSN1 in Apr.
Changes to data reconstruction and PDF definitions.
2009 and 2010 data analysis
Combined unblind result
Perspectives
MEG physics reach in 2011/2012
Sep 20th 2011
Gianluca Cavoto

3. MEG analysis in a nutshell
Extended maximum likelihood analysis
Determine number of signal S in signal region
Constraints on background rate (Accidental [A] and radiative [R]) from sidebands
Observables :Eγ,Ee,Teγ,θeγ,φeγ
Probability Density Function (PDF) from data
Measured resolutions in sideband or dedicated samples
Different PDF implementations (as cross-check)
Event by event information or averaged, different functional forms…
PDFs
Background rate
constraints
Sep 20th 2011
Gianluca Cavoto

4. The general method This is a blind analysis procedure
A portion of analysis fit region is kept hidden [BLIND BOX]
Use sidebands to determine background shape
Use ONLY sideband or alternative samples to determine resolutions
Reprocess data whenever a better reconstruction algorithm or calibration is available
This is a blind analysis procedure
Sep 20th 2011
Gianluca Cavoto

5. MEG dataset 2009 muons on target 2010 muons on target 0.65 1014
DAQ time : 35 days
DAQ time : 56 days
Optimized beam (degrader)
Improved electronics timing Slightly worse DCH conditions
Fully efficient detector
Stable conditions (DCH, LXe light)
Sep 20th 2011
Gianluca Cavoto

6. Summary of updates Improvements in data analysis since Feb
DC alignment (internal + global alignment)  
B field reconstructed with measured Bz      
More detailed implementation of positron observables correlations
Data analysis crosscheck
Two likelihood analysis implementations
Crosscheck on data and toy MC ensemble.
Fit to 2009 data
same configuration as for 2010
Data analysis follows a checklist that was agreed by the Collaboration
When it is fulfilled, the 2010 box is opened
Sep 20th 2011
Gianluca Cavoto

7. Our check-list
Sep 20th 2011
Gianluca Cavoto

8. Reconstructed B field Reanalysis of B field map measured in 2006
Br, BΦ reconstructed from measured Bz using Maxwell eq.
Account for possible Hall probe misalignment and general offset
Better performance in positron reconstruction together with improved DC alignment
DCH
region
~0.2 % difference (100 KeV on signal)
Use this map as default for any new data analysis (In Feb analysis we used the B field calculated from the measured COBRA coils current)
Sep 20th 2011
Gianluca Cavoto

9. DCH alignment General principle
Position of DCH wires in COBRA reference system determined using an optical survey (global alignment in COBRA) plus cosmics and Michel positron tracks (fine internal align. among layers).
Target position determined with optical survey and cross-checked with measured target holes position (Michel tracks).
Problems with previous methods
Global: did not include effect of nitrogen bag insertion
Internal: too dependent on starting point of align. algorithm
Use new optical survey [Mar 2011] for global alignment
Use Millepede technique [developed in spring ’11] for internal alignment
Sep 20th 2011
Gianluca Cavoto

10. Sep 20th 2011
Gianluca Cavoto

11. 2009 data reprocessing
Those were the most important changes to the data calibration and analysis.
The rest (timing, photon reconstruction) was not substantially changed (some small tweaking of the calib. parameters, resolution unchanged).
We therefore fit the 2009 data after the full recalibration and reprocessing.
Sep 20th 2011
Gianluca Cavoto

12. Time sideband fits and Sensitivity
Analysis region
48<Eγ<58 MeV
50<Ee<56 MeV
|Teγ|<0.7ns
|θeγ|,|φeγ|<50 mrad
Time
sidebands
[1.3,2.7] ns
[-2.7,-1.3] ns
NSIG
-7 +5 -3
-2+5 -2
NBG
278+17
-17
-16
NRMD
-5 +17
-14
-15
Upper limit on Nsig
3.8
3.0
Average UL on Nsig: 4.0
Error from MINOS [1.645 σ]
Results from sideband fits are consistent with expectation from toy exp’s
Sep 20th 2011
Gianluca Cavoto

13. Fit to 2009 data Teγ Ee Eγ θeγ φeγ UL at 90% C.L on NSIG : 10.4 Total
Accidental
Radiative
signal
Teγ Ee Eγ
Param
Best fit
MINOS
[1.645 σ]
NSIG
3.4
+6.6
- 4.3
NBG
273
+12
-12
NRMD
26.9
+4.5
- 4.5
Piu o meno
θeγ
φeγ
UL at 90% C.L on NSIG : 10.4
Sep 20th 2011
Gianluca Cavoto

14. 2010 data
2010 data were calibrated and studied with the same procedure as 2009 data.
In late June PDF were ready to fit 2010 signal region as well
Sep 20th 2011
Gianluca Cavoto

15. Signal photon energy PDF
Detailed calibration based on CEX runs
Energy scale stability cross-checked with several frequent calibration
Use event-by-event correction, here average PDF
Fraction =63%
W>2cm
W<2cm
Sep 20th 2011
Gianluca Cavoto

16. Signal positron resolutions
From double turn track studies
Nominal Positron momentum PDF from fit to Michel edge
After DCH alignment and B field optimization
residual bias accounted for in systematics
High quality and low quality tracks treated in different categories
Sep 20th 2011
Gianluca Cavoto

17. Positron variables correlation
Positron PDF takes into account correlations among angles and momentum
Geometrical effect due to target constraint
Very good agreement data with model prediction
Data 2009
Sep 20th 2011
Gianluca Cavoto

18. A note on phi resolution
Sep 20th 2011
Gianluca Cavoto

19. Signal relative timing
Fit to radiative events Teγ in Eγ sidebands
Clear signal, used to find center of signal region and signal resolution (scaled by Eγ)
Resolution does not depend significantly
on TC bar
Significant improvement
compared to 2009
Sep 20th 2011
Gianluca Cavoto

20. Sep 20th 2011
Gianluca Cavoto

21. Background ACC PDF MEASURED directly on data sideband
Use same selection as for final signal region, detailed dependence (on gamma position)
Positron E : Michel spectrum
Relative angles: effective distributions
Gamma E : Combination of RD +AIF + pileup+ cosmics
Consistent with signal resolution
ACC
Gamma
energy
Sep 20th 2011
Gianluca Cavoto

22. Summary of MEG performance
2009
2010
Gamma E
[sR, w>2cm – 63%]
1.9%
Relative timing Teγ (RMD)
150ps
130ps
Positron E [Michel edge]
330 keV(82% core)
330 keV (79% core)
Positron θ
9.4 mrad
11.0 mrad
Positron φ [at zero]
6.7 mrad
7.2 mrad
Positron Z/Y
1.5/1.1(core) mm
2.0/1.1(core)mm
Gamma position
5(u,v)6(w) mm
Trigger efficiency
91%
92%
Gamma efficiency
58%
59%
Positron efficiency
40%
34%
Muon stopping rate
s-1
DAQtime/real time
35/43 days
56/67 days
SES [analysis region]
Sep 20th 2011
Gianluca Cavoto

23. Sensitivity on 2009-2010 Combined: 1.6 10-12 2009: 3.3 10-12
Including systematics
2009
Mean:4.0
RMS: 2.2
2010
Mean: 5.4
RMS: 3.0
Nsig
Nsig
Use median to define expected UL on BF at 90%C.L. (“sensitivity”)
2009:
2010:
Combined:
Sep 20th 2011
Gianluca Cavoto

24. Fit to Teγ sidebands (2010)
Acc background counted in large region (1<|Teγ|<3.7ns) and scaled
[1.3,2.7] ns
[-2.7,-1.3] ns
NSIG
-4 +6 -2
2+8 -5
NBG
595+24
-24
-25
NRMD
14 +24
-21
Upper limit on Nsig
2.7
5.2
Error from MINOS [1.645 σ]
Upper limits well within the expectation
Sep 20th 2011
Gianluca Cavoto

25. Fit to signal region 2010 Ee Eγ Teγ φeγ θeγ Total Accidental Radiative
Param
Best fit
MINOS
[1.645σ]
NSIG
-2.2
+5.0
-1.9
NBG
609
+19
-19
NRMD
50.2
+9.2
- 9.2
θeγ
φeγ
Sep 20th 2011
Gianluca Cavoto

26. Systematics effects estimate
Sep 20th 2011
Gianluca Cavoto

27. Confidence intervals 2009 2010 Including systematics
Limit on Branching fraction (90%C.L.)
Lower Limit:
Upper Limit:
CL at 0 :
Upper Limit :
Sep 20th 2011
Gianluca Cavoto

28. Alternative analysis
UL extraction on 2010 data was done with alternative analysis: averaged PDF (no event-by-event information) different functional forms for PDFs
Done by Italian groups independently from our Jap colleagues
Both Bayesian and frequentist approach
Very useful to understand problems or develop studies UL
Compared with toy MC distribution
(freq)
2010 limit (Bayes)
<
Sep 20th 2011
Gianluca Cavoto

29. Analyses crosschecks
We generate an ensemble of toy experiments fit with both the likelihood definitions
Exp-by-Exp difference of Nsig best fit and UL on Nsig
We a priori declare the two analysis compatible on data if the difference
falls within 95% of the distribution of that difference
Sep 20th 2011
Gianluca Cavoto

30. Angular sidebands [2009]
Fit in angle sidebands θeγ- φeγ [-150,-50] mrad and [50,150]mrad
Check for presence of time-correlated bkg
Θ [mrad]
ϕ[mrad]
NBG
NRMD
NSIG
[-50,50]
[50,150]
-16
13 +15
-12
-1.3
[-150,-50]
-15
24 +15
-13
-2.7
-14
3 +13
-10
-1.1
-3 +11
- 7
-2.2
No evidence
2009 data
Error from MINOS [1.645 σ]
Sep 20th 2011
Gianluca Cavoto

31. 2009+2010 event distribution 51<Eγ<55 MeV; 52.34<Ee<55 MeV
Selection: |Teγ|<0.278ns; cosΘeγ<
51<Eγ<55 MeV; <Ee<55 MeV
Cosine Relative angle (3D)
Rank of variables
In each sample
1, 1.64, 2 σ contours
Sep 20th 2011
Gianluca Cavoto

32. 2009 data events distribution
Selection: |Teγ|<0.278ns; cosΘeγ<
51<Eγ<55 MeV; <Ee<55 MeV
Rank of variables
In each sample
1, 1.64, 2 σ contours
Sep 20th 2011
Gianluca Cavoto

33. 2010 data events distribution
Selection: |Teγ|<0.278ns; cosΘeγ<
51<Eγ<55 MeV; <Ee<55 MeV
Rank of variables
In each sample
1, 1.64, 2 σ contours
Sep 20th 2011
Gianluca Cavoto

34. MEG result on BF(μ→eγ)
Combination of 2009 and 2010 data gives a UL at 90%C.L on BF
Combination
account for different resolution
in sub-dataset.
Constraints from sidebands used
for accidental
and radiative decay
backgrounds. .
Paper accepted by PRL for publication (
Sep 20th 2011
Gianluca Cavoto

35. Radiative decay rate stability
Sep 20th 2011
Gianluca Cavoto

36. Normalization evaluation
Sep 20th 2011
Gianluca Cavoto

37. Perspectives
Adding more data from 2011 and 2012 will reduced this limit
Estimate with toyMC exp. and likelihood analysis
Using a Bayesian approach
some numerical difference with standard frequentistic
Some improvements to detector performances can be foreseen.
Better resolutions: Eγ (1.5%), positron E (290 keV) and positron θ (8.5 mrad)
Sep 20th 2011
Gianluca Cavoto

38. Expected UL vs data sample
NULL Signal hypothesis
2010
2011
2012 20  ✚ 
Data (freq.) ✚
Sensitivity (freq.)
Sep 20th 2011
Gianluca Cavoto

39. Conclusions
No evidence of signal in MEG dataset Combination of 2009 and 2010 improves by a factor 5 the best MEGA limit
G.Isidori et al. Phys.Rev. D75 (2007)
Excluded by MEG
from SM in g-2 meas.
Observed deviation
We will continue to analyze data as they come.
With hopefully better detector conditions
(better resolution, uniformity)
Working on better algorithms to improve
resolutions.
from SM in g-2 meas.
Observed deviation
at 3 sigma in 2012 (21 times 2009 DAQ time)
B physics
constraint
Sep 20th 2011
Gianluca Cavoto

40. Backup
Sep 20th 2011
Gianluca Cavoto

41. Time versus 3D angle
2009
2010
Sep 20th 2011
Gianluca Cavoto

42. Rsig plots
2009
2010
Combined
Sep 20th 2011
Gianluca Cavoto