Higgs Searches at the LHC: An Experimenter

1
Higgs Searches at the LHCAn Experimenters
Perspective

• Robert Cousins, UCLA
• 31st Johns Hopkins Workshop on Current Problems
  in Particle Theory
• Heidelberg, 2 August 2007

2
Four Excellent Talks lt2 Weeks Ago at EPS
... And many more at SUSY07!
3

• So, in this talk, I will not attempt to archive
  more than can be absorbed in the time allotted
  for my talk.
• While giving a broad overview illustrated by
  official CMS and ATLAS results, I will attempt
  to emphasize some aspects of Higgs searches where
  more work might be useful. Focus here on first
  observation, but many more issues will follow re
  couplings, etc.
• Among the numerous general resources available
  which aided me in preparing this talk, I mention
  in particular the CMS Physics Technical Design
  Report, and lecture notes by D. Rainwater,
  http//arxiv.org/abs/hep-ph/0702124.
• Much more in A. Djouadi, arXivhep-ph/0503172,
  0503173. Also V. Buescher and K. Jakobs, Int.
  J. Mod. Physics A, Vol 20, Nr. 12 (2005),
  2523-2602. hep/ph-0504099

4
SM Higgs Production
pb
NLO

1. gluon-gluon Fusion
2. W,Z Boson Fusion
3. Associated WH, ZH prod.
4. t tbar H production

BSM can change this in many ways, e.g. ,gg?Hbb.
Note ?tot 1011 pb, ?b 109 pb, ?jet gt 100 GeV
ET gt 106 pb Need control regions in data to
understand bkgnd.
5
SM Higgs decay modes
BSM ??, ??, and bb changed in many ways, even
within MSSM (M. Carena et al., hep-ph/0202167).
For effective Lagrangian approach to BSM
gg?H???, see Manohar and Wise, hep-ph/0601212
can be dramatic.
6
The Experimental Challenge

• Production cross section times decay branching
  ratio for H ? g g is 10-13 of the pp inelastic
  cross section.
• For H ? Z Z() ? 4 leptons, it is even smaller.
• These are inhumanly small numbers, lower even
  than searched-for rare decays of kaons and muons
  (10-11 to 10-12 B.R.).
  
  Top quark discovery at Tevatron was lt10-10 level.
• Thus the challenge at the LHC is to push the
  state of the art in both hadron collider
  techniques and rare decay techniques.
• Experience from both is to rely on theory and
  M.C. as little as possible, tuning both to real
  data. Measured ratios of similar processes (so
  that unknown systematics cancel at least
  partially) are typically the most robust.
• Be prepared for unexpected backgrounds.

7
ATLAS
8
CMS
9
Cross Section of CMS
4T central B, -2T in return yoke
10
ATLAS
11
The Approximate State of the Art in M.C. Studies
With K-factors
with K factors
LHC 1 fb-1 in 2008, increasing to 100 fb-1/year
at design luminosity.
12
H ? g g

• B.R. 0.002 at MH115-140. Classic bump-hunting
  on smooth background but (!) S/B 1/20.
• Experimental challenges
• g energy and angle resolution
• Both CMS and ATLAS optimized for this
• Reduce fake photons, reduce photons from ?0s
• Preshower, isolation (form of veto).
• Beyond simple cuts
• CMS ANN, classify events by quality, combine
  with weights
• ATLAS include kinematic variables in likelihood
• How safe is this? How to control? How to
  convince skeptics with more info than a mass
  peak?
• Also in VBF. What is interplay between VBF and
  inclusive?
• Once established, mass measurement to fraction of
  1.

13
Discovery potential of H-gtgg
CMS optimized Artificial Neural Net with
kinematics and g isolation as input, s/b per
event ATLAS likelihood pT, angles
SM
Significance for SM Higgs MH130 GeV for 30 fb-1,
NLO CMS Physics TDR 6.0 cut-based, 8.2
optimized ATLAS 6.3 cut-based, 30-40 better
with likelihood
14
H ? Z Z() ? 4 leptons

• Studied and discussed for years, since relatively
  clean and sensitive over large MH range,
  especially 4µ.
• Background is so low that bkgnd statistical
  uncertainty from sidebands may be an issue
  profitable to do more work on measuring
  backgrounds using other sign/flavor combinations,
  relaxing cuts, etc.?
• At low MH, continuum ZZ() bkgnd peaks above the
  signal need to be sure off-shell extrapolation
  is reliable. (Typically one requires one on-shell
  Z.)
• How low in MH can one push this channel?
• Can other kinematic variables (e.g. pT) be used
  convincingly?
• What is best way to optimize cuts (robust yet
  powerful)?
• Separate cuts for leptons 1, 2, 3, and 4?
• How strongly should cuts depend on mass?
• Multi-variate? (Event generators...)

15
H ? Z Z() ? e e- µ µ- (CMS PTDR)
tt and Zbb bkgnds reduced by isolation, impact
parameter cuts both to be understood from data.
16
4l bonus Higgs JCP. Generalization of an old
idea...
...with much richer potential information.
... or with ?s measured in Z frames
See Rainwater (2007) and refs therein, incl. VBF
extension... CERN workshopshep-ph/0608079CERN-
2006-009. Not for the first year!
17
H ? W W() ? 2l 2?

• H ? W W() is dominant decay mode above 135
  GeV, dramatically increases width of H and
  reduces other modes to rare except ZZ().
• A data analysts dream (?) since no mass peak,
  uses about every trick in the book... and chance
  for early discovery if MH 2MW and bkgnds
  understood!
• ATLAS updating old PDTR result. CMS studied 2µ 2?
  as a benchmark channel for muons, also other 2l
  2?.
• Backgrounds (several still with 15 uncertainty
  or greater) higher order effects, spin
  correlations are important need full generators.
• Continuum WW (and WZ and ZZ)
• tt, tWb (jet veto) and some bb (impact
  parameter), isolation
• Drell-Yan dimuons (angle btw muons is large
  unless jet present)
• Events with jets faking electrons, in particular
  W jets
• Sensitivity in a variety of kinematic quantities,
  incl spin correlations, ?fµ µ muons tend to
  come together when WW from spin 0.
• Cuts vs multivariate? Discussion of background
  estimation from data. What is optimal way to
  combine µµ, eµ, ee channels?

18
H Production by (Weak) Vector Boson Fusion
No color string to snap in central region
ATLAS fig.

• In last few years, widely studied following
  earlier work (e.g., Rainwater Zeppenfeld, PRD
  60,113004 and dozen refs therein) H decay modes
  tt, ? ?, WW.
• ATLAS (Asai et al.) says VBF tt mode is more
  promising at low MH than (non-VBF) ? ?, and VBF
  WW mode better than non-VBF.
• MH measurement relies on resolving MET along two
  axes of (non-back-to-back) tt. How will this
  work in real data?
• Will central region be as quiet as predicted?
  Is some sort of veto (calo, track, combination?)
  adequate, or better off with multi-variate?
• How well can backgrounds be understood from data?
  See discussions in Rainwater (2007) and Asai et
  al. (2003), and CMS PTDR.

19
(Weak) Vector Boson Fusion (sim with ATLFAST)
Asai, et al., Eur Phys J C 32, s02, s19-s54
(2003).
20
(Weak) Vector Boson Fusion, tt ? lepton tau
jet ...
CMS Physics TDR, full sim and reconstruction
Asai, et al. (2003). ATLFAST.
... VBF needs further study in all modes.
21
ttH, H ? bb

• Proving to be a very tough channel.

J.Cammin and M.Schumacher ATL-PHYS-2003-024 S/sqr
t(B) 2.8, MH 120 GeV, 30 fb-1 , being
revisited.
CMS NOTE 2006/119
22
Higgs Beyond the Standard Model

• Vast literature by now, detailing many
  possibilities benchmarks in MSSM extensions
  beyond MSSM substitutes for fundamental scalar.
  (EPS and SUSY07.)
• I will not attempt to discuss all the plots in
  various parameter spaces, but rather focus on a
  couple novel experimental signatures with respect
  to SM Higgs.
• Now at least 5 states, including charged Higgs
  bosons, CP-odd state, (even doubly-charged state
  in 3-doublet model).
• Enhanced coupling to b quarks, tau in some
  scenarios other scenarios such as decays
  dominant to invisible particles.
  Re-emphasizes need to understand b, tau, missing
  ET.
• Possibility of H decaying to SUSY particles
  (e.g., for ATLAS, Hansen et al., hep-ph/0504216)
• Emphasizes need to measure quantum numbers and
  couplings (in both production and decay)

23
A Couple Slices in MSSM Parameter Space
A. Djouadi, arXivhep-ph/0503173
24
MSSM Charged Higgs H, H-

• Dominant production is at a tbH vertex. For heavy
  H

For lighter H, on-shell tt production following
by t?Hb. Decays mostly to ??? for mass lt 180 GeV
tb mode opens above but seems hopeless, so ???
remains the focus. Tau polarization opposite to
taus from W decay useful handle! Events are
complex, with complex backgrounds (tt, tW,
Wjets) b jets must be understood some current
search strategies are dominated by systematic
errors. Current effort is on how to reduce
systematic errors with subsidiary measurements,
ratios. (SM top, Z???, etc.) Refs CMS Physics
TDR Mohn et al., ATL-PHYS-PUB-2007-006
25
Scenarios with Increased Hb Coupling
(MSSM large tan?)
Subsequent decay modes studied µµ, tt

• Re-emphasizes importance of early SM studies of b
  quarks (in copious tt production) and taus (in
  Z???), and modes such as Zbb.

Status in CMS Physics TDR
ATLAS update for µµ S. Gentile, et al.,
arXiv0705.2801v1
26
Includes parity-violating sign
27
Karl Jakobs at SUSY07
Invisible Higgs decays ?
Possible searches tt H ? lnb qqb PTmiss
Z H ? ll
PTmiss
qq H ? qq PTmiss

PTmiss
- J.F. Gunion, Phys. Rev. Lett. 72 (1994) - D.
Choudhury and D.P. Roy, Phys. Lett. B322 (1994)
- O. Eboli and D. Zeppenfeld, Phys. Lett. B495
(2000)
All three channels have been studied key
signature excess of events above SM
backgrounds with large PTmiss ( gt 100 GeV/c)
Sensitivity

• Problems / ongoing work
• ttH and ZH channels have low rates
• More difficult trigger situation for qqH
• backgrounds need to be precisely known
• (partially normalization using ref. channels
• possible)
• non SM scenarios are being
• studied at present
• first example SUSY scenario

95 CL

ATLAS preliminary
28
Higgs Bosons in Non-Minimal Models

• Little Higgs
• Doubly charged Higgs Spectacular resonance in
  same-sign dimuons
• Extra dimensions
• Radions, Higgs in radion decays

• Experimental issues similar to the rest in this
  talk resolution, tag jets, photon ID and
  isolation, b-tagging, background measurement.

29
Discussion

• In the last 25 years, an enormous amount of
  effort has gone into developing Higgs search
  strategies and predicting how well they will
  perform. A lot of this effort involved reducing
  uncertainties in predicting background.
• As the exciting time of real LHC data approaches,
  uncertainties in predicting how well search
  strategies will perform are relevant only in
  deciding where to concentrate the search
  effort... Soon we will measure background rates,
  and refine the search strategies!
• So lets remind ourselves of some principles of
  experimental HEP. Techniques developed at the
  Tevatron, LEP, and B factories will help us a
  lot, but we still have work to do while
  anticipating first beam.

30
NNLO calculation is not always needed for initial
discovery of di-object resonance.
Nor do you initially need absolute rate to 5.
31
Vetoes

• Veto requiring the absence of some particle,
  signature, etc. Notoriously difficult to predict
  effect, going back to the days of NIM
  electronics.
• Example vetoes
• Jet activity in central region, for VBF
  signature.
• Too many b quarks, when background is enhanced in
  bs (e.g. when background is tt).
• Typical isolation criteria.
• Note Optimal criteria for defining object (e.g.,
  b quark) for veto are not necessarily the same as
  for positive ID.
• Especially with pile-up rates of 20 events per
  beam crossing, will require great care and
  creative ways to calibrate.

32
Likelihoods, Multivariate Techniques

• Neyman-Pearson Lemma Best discriminating
  variable for distinguishing two simple hypotheses
  (no fitted parameters) is the ratio of the
  likelihoods under the two hypotheses. If
  possible to write down correctly with all the
  correlations, etc., then thats it.
• Poor persons version multiply 1D or 2D
  likelihoods as if no correlations. At least one
  can see the plots entering the calculation.
• Machine-learning techniques (ANN, BDT, etc.) can
  sometimes do better when it is hard to write down
  likelihood ratio with full correlations.
  (Essentially that is what they are attempting to
  do see H. Prosper in http//www.ippp.dur.ac.uk/Wo
  rkshops/02/statistics/proceedings.shtml). More
  and more experience in HEP.
• Very powerful, but can be very hard to track down
  puzzling behavior.

33
Single-Top 5 Years into Tevatron Run II
CDF The question arises to which extent the
results of the Matrix Element (ME), the
Likelihood Function (LF), and the Neural Networks
(NN) techniques are compatible... our
compatibility measure ...is 0.65. same
data! http//www-cdf.fnal.gov/physics/new/top/top
.html
D0 3.4 ? first evidence

• How much does one want to rely on multi-variate
  techniques for early discovery physics at LHC?
• How to do the controls?

34
Higgs and SUSY searches share many issues...
Note multi-b production.
35
Beyond First Observation What is it? What else
is there?

• More precise measurements and more precise
  theoretical calculations move into spotlight.
• Challenge to compare theory and expt for
  production cross section, with effect of cuts on
  kinematic distributions, etc. Event generators
  to highest possible order (and with flexibility
  for model tuning) are welcome!
• Can we discern new physics interfering ( or -)
  with the top loop in gg?H ?
• Mass O(0.1) over wide range once detectors
  well-cablibrated. Width see discussion by
  Rainwater 2007.
• Spin angles, e.g., leptons from (spin 0) H?WW
  tend to be in same direction.
• Multiple production and decay modes if M130
  GeV, several to compare!

36
Statistics for LHC

• Will build on the considerable experience of
  Tevatron, LEP, B factories, et al.
• ATLAS and CMS already discussing common
  (multiple) methods for comparing and combining
  channels and experiments.
• Aim is to have supported tools in ROOT for
  various frequentist and Bayesian methods.
• Incorporating systematic uncertainties still a
  challenge! Talks at PhyStat 2005 at Oxford.

37
Conclusion

• Over many years, Higgs-hunting strategies have
  evolved from concepts, to generator-level
  studies, to full simulation with reconstruction,
  with data-driven background techniques.
• Now the focus is shifting even more from
  projections of how well will we be able to do
  to how precisely will we do it. Understanding
  and controlling systematic errors, in particular
  as the analyses become more complicated, is at
  the forefront. Real data will come soon!
• A general area where theorists can help is in
  guidance on what kinematic distributions are
  reliable discriminants, especially if fed into a
  multivariate soup. Similarly, which parts of
  phase space make reliable control regions for
  predicting background in signal regions.
• Work is underway to have coherence in (various)
  statistical techniques, combining channels, etc.

38
Thanks

• To many for discussions and references,
  including
• CMS Higgs physics analysis group conveners
  Alexandre Nikitenko and Yves Sirois, and CMS
  Physics Coordinator Paris Sphicas and Claudio
  Campagnari.
• ATLAS Higgs working group conveners Louis Fayard
  and Markus Schumacher and Karl Jakobs.