Search strategy for SUSY et al. at LHC

How does nature behave at 1TeV ?
A search strategy for SUSY et al.
Sascha Caron
University of
Freiburg
Outline: 1st Motivation, 2nd Strategy, 3rd Questions
KET-BSM meeting Aachen, April 2006
View from the Schauinsland in Freiburg a couple of weeks ago
The situation in 2006
• We still don’t know the origin of EW symmetry breaking
 The Higgs boson is not discovered yet
• Even with the SM Higgs:
‘fine tuning’ is required in the model to remain valid to high energies?,
Gravity is not included?, Fermion masses? What is Dark Matter?,…
 typical solutions by increasing the number of
symmetries, dimensions, forces, …
Higgs ?
Sascha Caron
page 1
Something else?
The situation in 2006
Investigate if EW
symmetry breaking
is caused by a Higgs.
Part 1
Higgs working groups
at ATLAS and CMS
Sascha Caron
page 2
Investigate if there is
other physics beyond the
Standard Model
Part 2
This approach:
Data mining strategies
How to find anything potentially
interesting and previously
unexpected in the data?
Number of models
The situation in 2006
Worried about
your search strategy?
5
30
Number of Higgs doublets
hep-th 0411129 SUSY spectra from special string vacua
Sascha Caron
page 2b
The situation in 2006
What do we expect to find at the LHC?
Sascha Caron
page 3
One physicist's schematic view of particle physics in the 21st century
(Courtesy of Hitoshi Murayama)
The situation in 2006
CMSSM
MN2SSM
SUSY VERSIONS
OF THE SM
NMSSM
(an additional Higgs
singlet)
SUSY with extra Dim
SUSY with extra forces
SUSY+ little Higgs,
…
Choose this point,
look at the LHC data,
exclude or not!
Sascha Caron
page 4
We found no deviation
We have excluded this point/area which
is epsilon of the parameter space
We found a deviation
Does this mean that the ‘real’ model
is this parameter point?
 Is it efficient to work like this?
Finding the unexpected – explaining the origin
The other strategy: START FROM THE DATA
1) Search for deviations in (almost )all final states
(they are all interesting either as signal or to understand background)
2) Determine ‘deviation(s)’ or ‘inconsistencies’ (e.g. all muon final
states have problems)
3) Determine their origin (detector effect, Monte Carlo? , new physics?)
4) Re-determine expectation and
repeat step 1-4) until publication in journal
Examples:
General Search for new Phenomena at H1 (2004) and
D0 Sleuth approach (2002 but only top final states)
Sascha Caron
page 7
Example: H1 General Search
• Event yields for HERA 1
data
First time a HEP experiment
analyzed all final states
Channels which have
not been syst. studied before
Sascha Caron
page 8
We
investigated
all
Mall and ΣPT
distributions
We developed
a simple and
powerful algorithm
to find and
quantify
deviations
automatically
Sascha Caron
page 9
Is this
approach
sensitive to
New Physics?
1000
10
MC SM experiments with larger deviation
100000
Martin Wessels Ph.D. thesis RWTH Aachen
H1 tested various models and found compatible sensitivity
to direct searches in all of them (without tuning a cut)!
Next step for me: Sensitivity tests of such an approach for
CMSSM points at ATLAS
Sascha Caron
page 9
Is this possible at LHC?
Is this the best strategy for an
‘early discovery’?
What do we need for this search?
What can we learn from theory?
Is this possible at LHC ?
Yes ! (H1 has made the ‘proof
of principle’)
Sascha Caron
page 11
Is this the best strategy for
‘early discovery’?
Answer 1 : DEPENDS ON THE PHYSICS
Answer 2 : I’M NOT 100% SURE TO BE HONEST
We like to start from a ‘simpler’ scenario and to extend
(after we know some of the detector response and of physics at 1 TeV)
Our attempt :
Start from channels where one might expect something new
and you don’t know exactly what and where you can
predict some of the background from data
pT_miss channels (Dark Matter…?)
Idea: “less model dependent” SUSY/DM searches
What do we need for this?
SM prediction (with complete uncertainty) in (finally) all channels
(Multi purpose event generators)
A multi-purpose analysis framework (as in H1) (I thought it would
be nice to run a simple version of this even on-line)
Uncertainties and fudge factors from data
(calibration with data candles, use data without pt_miss, use fits to
fudge factors, use a global background determination strategy, make
‘fake data’ for each channel, use fast ways to go from 1-4)
Later: A way to learn what we have found
Sascha Caron
page 13
What can
we learn
from theory?
- What are ‘model independent’ the best variables to determine the
underlying physics (Et, mass, endpoints?, spin information, something
else?)
- What do you need to determine the nature@1TeV Lagrangian? Do you
know already how to do this?
Would it be helpful to publish a ‘pseudo’ ATLAS/CMS signal?
- Tune QCD radiation: Best MC tunes via fits to almost all published data
- How can we best use Jet+X events to determine Jet+Ptmiss+Y events?
(e.g. include fit procedure into Generators to determine some QCD
radiation weight factors instead of predicting e.g. W+jets events with
Z+jet events?)
What can
we learn
from theory?
Attempts to determine LHC signals:
- LHC olympics (a signal only ‘fun’ analysis)
- LHC inverse problem
- BARD (automizing ME calculations of Madgraph and fitting to
signals)
Any interest from german theory to start something better?
Determine a general LHC Standard Model:
Madgraph/event, Sherpa/Amegic, …. 
General BSM Model Generators to determine the efficiency of
such an approach for any model (can we be more general?)
Theory and ‘Going the way into the other direction’…
A General analysis of LHC data
Summary
I’ve tried to illustrate
what we like to do
and why
(build such a framework for ATLAS)
Somebody interested in joining a general
data analysis strategy in germany ?
A bit more motivation
The SUSY search strategy
Examples of SUSY searches at LHC:
jjjjv channel cuts optimized on specific CMSSM points
• 1 jet with pT >100 GeV, 4 jets (pT>50 GeV)
• ETMISS > max(100 GeV ,0.2Meff)
• Transverse sfericity ST>0.2
• No isolated muon or electron (pT>20 GeV)
• better signal to background with a extra lepton
• + scanning on E_T distributions
I think we can gain sensitivity by exploring more channels (or by
subdividing the data instead of cutting)
Does the true signal slip through our harsh cuts?
Sascha Caron
page ?
A significant danger is finding correlations and signals
that do not really exist.
Many examples in particle physics history
We are looking for deviations …
How surprised should we be to find some?
How likely is a 4-5 sigma deviation at LHC
even if there is nothing in the data?
Sascha Caron
page 20
 Unsolvable problem if you use 2000 PhD students
Quantify the deviations
Step 2: Count how many
times you find
deviations bigger
than in those in your
real data.
3% of the
“Pseudo H1 experiments”
have found
a bigger deviation
Sascha Caron
page 21
Number of channels
Step 1: Repeat the whole analysis
with a pseudo data experiment
(dice your own MC data) many times.
3%
1
10-1
10-2 Probability
to find
deviation in
this channel
I know that this is not a new idea, but we do not often use it
What are the numbers for ATLAS or CMS?