Sherlock

1
Sherlock
Bruce Knuteson Berkeley
Dave Toback Maryland
2
Quasi-Model-Independent Search at DØ Sherlock

• Assume nothing about the new physics except that
  is high mass
• High mass assumption ? Quasi-Model Independent
• Systematically look at events by the final state
  signature
• Search for new physics by looking for excesses in
  multi-dimensional data distributions

3
Overview of Sherlock

• Define final state signatures
• (Signature based search)
• A priori prescription for defining variables and
  regions of space in those variables
• A systematic look for regions with an excess
  (more events than expected) with large ET
• Find most interesting region
• Compare with the expectations from hypothetical
  similar experiments using background expectations
• Take into account systematic errors and the the
  large number of regions searched

4
Labeling Final State Signatures

• Final State particles
• e, m, t, g, j, b, c, MET, W or Z
• Each event is uniquely identified
• All events which contain the same number of each
  of these objects belong to the same final state
• Need standard object ID

5
General rule for picking variables

• Looking for new high mass physics
• Mass-Energy Relationship
• Decay to known Standard Model particles
• light in comparison
• High energy long lived particles in final state
• High Mass ? High ET
• Look at ET of the final state particles

6
Sherlock Algorithm

• Look at various regions
• Find most interesting region (largest excess)
• Run hypothetical similar experiments using
  background expectations
• Fraction of hypothetical similar experiments
  (from backgrounds alone) which have an excess
  more significant than the one observed

ET of Y
ET of X
Background expectation Example signal events
7
Using Sherlock on Run I Data

• Look in events with an electron and a muon for a
  excess which might indicate new physics
• Why em? (why not?)
• Lots of theory models
• Supersymmetry? Anomalous Top quarks?
• Backgrounds include good example of heavy
  particles to look for
• Top quarks, W bosons

8
tt and WW production
? High ET relative to other backgrounds
em 2 Jets
em 0 Jets
_P
_P
P
P
9
Testing Sherlock

• Both WW and top quark pair production are good
  examples of high ET events which might show up
  with Sherlock
• Run Mock Experiments pretending we dont know
  about WW andtt production
• 4 Samples
• em 0 Jets WW
• em 1 Jet
• em 2 Jetstt
• em 3 Jets

10
Mock data with no signal
Fraction of hypothetical similar experiments
(from backgrounds alone) which have an excess
more significant than the one observed
Probability is flat as expected
em 1 Jets
em 0 Jets
Small P is interesting Smallest bin is lt5 No
indication of anything interesting
em 3 Jets
em 2 Jets
11
Sherlock with WW andtt
Pretend we dont know about WW andtt Mock
experiments with WW andtt as part of the sample
Observe an excess in 0 Jets (WW
production) 2 Jets (tt
production) in the mock trials
Remember Small P is interesting Smallest
bin is lt5
em 0 Jets
em 1 Jets
em 2 Jets
em 3 Jets
12
Combining Results
Pick most significant excess Take into account
that we looked in 4 datasets
DØ Preliminary
All overflows in last bin
Bkg WW tt
Mock Experiments
/ / / /
Bkg only
Excesses corresponding to WW andtt found
Significance of excess in standard deviations
13
Sherlocktt
Include WW as a background Expect an excess in
2 Jets only tt production
em 1 Jets
em 0 Jets
em 3 Jets
em 2 Jets
14
Findingtt alone
Use all backgrounds excepttt and look for
excesses
DØ Preliminary
Bkg tt
All overflows in last bin
/ / / /
Mock Experiments
Bkg only
Excess corresponding tott
Significance of excess in standard deviations
15
Final Sherlock Results
Use all backgrounds and look for excesses
DØ Preliminary
/ / / /
We see no evidence for new physics at high PT in
the emX data
16
Thoughts Conclusions

• Sherlock is sensitive to finding new physics when
  it is there to be found
• Would have pointed out either WW ortt if we
  didnt know about them
• Would find events like the eeggMET naturally
• Should be sensitive to many SUSY signatures
• While it cant be as sensitive as a dedicated
  search, it may be our only shot if we guess wrong
  about where to look in our data.
• A natural complement to the standard searches
• PRD with collaboration now. Released in June(?)