Higgs physics at Hadron Colliders Peter J Bussey
Higgs physics at Hadron Colliders Peter J Bussey Higgs-Maxwell Workshop 2008
Some reminders:
G. Rolandi
Cross sections for Higgs production at the Tevatron pb Note high qq cross sections. Suppressed at LHC!
S. M. Higgs Branching Ratios
Some analysis methods: (A) Cuts method (1) Plot various measured quantities. (2) See which quantities show biggest differences between signal and background in MC (3) Cut out those regions which are mainly background, aiming to preserve the signal. (4) Choose a convenient parameter as discriminant. Unblind the data if it was blind. Do a statistical analysis of remaining data vs MC background.
(B) “Matrix element” technique (not a very good name) (1) Choose a set of kinematic parameters that are measured in an event (x), and select a point in x-space. (2) Evaluate the net cross section for each background process to give an entry at this point, by integrating theoretical cross sections (“matrix element squared”) over all the unmeasured parameters. Call the probability for the ith background Pi(x). (3) Likewise evaluate PS(x) for the desired signal process.
Then the quantity we are interested in is the likelihood ratio defined as LR = PS(x) + PB(x) where PB(x) = total sum of all the backgrounds Pi(x). Thus, if it were all signal, LR would be unity. Background tends to lower LR values, signal to higher. The LR is used as the discriminant to separate signal from background.
(C) “Neural Net” technique The Neural Net is a program which can be “trained” to discriminate between signal and background by giving it designated MC samples of each. It outputs a discriminant whose value is low for background, and ideally unity for signal. There are many different such programs and different detailed ways of running them.
In practice, methods can be combined. E. g. apply some cuts before performing a ME or NN technique. Do an ME technique and input the LR as an input parameter into an NN. In none of these methods is there any relaxation of the need to understand the data and MC properly! Especially, we must be sure that the MC really does describe the data.
Final step, a Confidence Level analysis. Bayesian and frequentist techniques are used to calculate (usually) 95% CL on presence of a Higgs signal, usually as a ratio to SM expectations. Two important numbers: (1) the expectation limit based on given integrated luminosity and MC calculations (2) the actual limit obtained from the data. So (2) should be a statistical fluctuation on (1) if there is no signal.
The Tevatron CDF ←p D 0 ←d
LHC
ZH channel. Sensitive at the Tevatron at low Higgs masses. Signature: Missing ET ≥ 2 jets at least one b-tagged. Method: Cuts analysis Discriminator: invariant dijet mass distribution of the final state.
More details: Require ≥ 2 opposing jets with ET >20 Ge. V, Require overall ET ≥ 50 Ge. V Summed ET < 240 Ge. V No isolated leptons Remove events with noisy calorimeter signals. Apply NN to tag b-jets. Two b-tags of which one must be “tight”.
130 events predicted and 140 observed
Results from D 0 for two ZH channels Still some way to go
CDF Selections: ----- one high-ET lepton + 2 nd of same type, opp sign Z mass window 76 -106 Ge. V 2 or more cone jets, ET > 15 Ge. V fairly central 1 tight or 2 loose secondary vertex tags Train a NN on ZH vs Z + jets or The analysis uses a 2 -D NN score
(left) dijet mass spectrum (right) 95% limits wrt SM Higgs
Selections: ------ 2 jets, ET > 75, 35 Ge. V fairly central ET(vec)/(ET 1 + ET 2) >0. 45 >=1 secondary vertex tags Missing ET > 60 Ge. V Δφ between MET and 1 st jet > 0. 8 Cuts based analysis, followed by NN Note that there is a WH as well as a ZH component to this observed channel.
CDF ZH MET + jj Similar limits to D 0 - see later for better results
D 0: HW channel. The most sensitive at the Tevatron at low Higgs masses. Signature: 1 high-PT lepton ≥ 2 jets at least one b-tagged. Method: Cuts analysis: Discriminator: _ invariant dijet mass distribution of the WHlνbb final state.
More details: 2, 3 jet final states examined, but in the end just the 2 -jet channel is used. NN used to tag the jets as b jets. ET > 20 Ge. V Cone jets used, radius 0. 5 Limits calculated from the individual e and µ channels with single, double b tagging.
Dijet masses, single and double b tag. Backgrounds compared to data points
D 0: HW channel. Signature: 1 high-PT lepton ≥ 2 jets at least one b-tagged. Method: ME-LR analysis: Selection: basically similar to the cuts analysis. There is a rejection of badly measured jets by _ requiring missing ET to be not aligned or antialigned wrt lepton or jet.
Examples of LR discriminant. Single, double b-tagged events.
(1) (2) (3) (1)Cuts based limits (2) ME-LR based limits (3) NN (better statistics)
Selections: -- one isolated high-ET e or µ (>20 Ge. V) -- 2 or more cone jets, ET > 15 Ge. V fairly central -- Missing ET (>20 Ge. V) Train a NN on jets to give b tag Cuts based analysis superseded by NN. Dijet mass is an important variable
Two jets b tagged with NN. Then another NN
The WW challenge Two diagrams for the WW Channel: quark Annihilation from gluons, Vector Boson Fusion (VBF).
H→WW* using cuts method Method: select events with two isolated leptons (µ, e) and missing transverse energy. Missing ET cut designed to eliminate Drell-Yan (mainly Z production) Apply cuts on ET of leptons and on direction of missing ET of event. Restrictions on jets. Higgs WW leptons prefer to point in same direction, but the dominant WW background leptons do not, so their angular separation is a discriminating variable.
As the Higgs mass is scanned the selections change slightly and so does the CL Dominant WW background
Cuts limits summer 2007.
H→WW* using ME-LR method Method: select events with two isolated leptons (µ, e) and missing transverse energy. Missing ET cut designed to eliminate Drell-Yan (mainly Z production) Evaluate LR distribution and CL for Higgs.
Examples of LR distributions
H→WW* using NN method Method: select events with two isolated leptons (µ, e) and missing transverse energy. Missing ET cut designed to eliminate Drell-Yan (mainly Z production) Evaluate LR distribution and CL for Higgs. The events are categorised into two classes: high and low signal/bgd
Different multivariate analysis methods compared
Different NN training techniques compared.
Example of NN discriminant after removing Z’s
CDF WW limits: (left) ME-LR (right) NN Similar expectations, ME did better with data; group now combines the two methods (see next)
(left) ME + LR method with higher statistics (right) include LR data into NN method
μτ channel from D 0
lp 07 CDF combination Absolutely necessary to combine results.
Combined limits 2007 from D 0: WH, ZH, H, bb, WW
CDF’s Combined Limit
Tevatron Combined Limit December 2007 – improvements being prepared!
The CDF Combined Limit 1. 9 pb-1 m. H (Ge. V) 110 115 120 140 160 180 200 Median Expected limit/SM 4. 6 4. 8 5. 7 5. 6 2. 8 3. 9 8. 1 Observed limit/SM 5. 9 5. 8 6. 3 4. 2 1. 6 3. 0 11. 7
D 0 search for SUSY Higgs → τ τ 1 fb-1 The τ τ channel is too low for a competitive SM Higgs search. However there are SUSY Higgs’s (A, H, h) that can decay in this way and could have a detectable rate. Search uses τ →µ for trigger with the second τ decaying to e or hadronically. -- one µ with pt > 15 Ge. V, no other high-pt µ -- τ candidate with 1 or 3 tracks and appropriate calorimeter signals. Use a NN to tighten up the second τ identification. -- NN to provide Higgs discriminant.
(left) Visible mass of muon-tau system (right) Neural-Net output variable, MH=160 Ge. V.
Cross section limits in pb No sign of deviations above MC expectations.
Excluded regions in tanβ vs MA plane under various model assumptions. green = LEP red = CDF (325 pb-1) black, blue shade = present analysis (expected, found)
search for SUSY Higgs → τ τ 1 fb-1 Cuts based analysis. Trigger on high-p. T e, µ. Hadronic τ candidates selected by cone based algorithm. τ pairs with at least one lepton are selected. Overall Higgs acceptance 1 -3% Lepton p. T > 6 Ge. V/c typically Total summed transverse energy >45 -55 Ge. V. Cut on missing energy. Reject Z’s Plot visible mass.
Last year’s 2 -σ excess above MC expectation seems to have gone away!
Again, exclusion bounds on SUSY parameters, updating previous CDF limits and complementing LEP
Latest Higgs physics studies at LHC
Higgs Prospects at LHC (based on G Tartarelli’s Aspen talk) According to current schedule: Beam commissioning starts May 2008 First 14 Te. V collisions July 2008 Foreseen: L=2 x 1032 cm-2 s-1 by end of 2008 with integrated luminosity up to ~100 pb-1. 2009: should reach 2 x 1033, “low luminosity”. About 4 pile-up (minimum bias) events per bunch crossing (at nominal 25 ns bunch spacing) Probably >2010: should reach design luminosity 1034 cm, ‘‘high luminosity’’, ~20 events per bunch crossing
Prerequisites for a Higgs discovery Good understanding of detector and trigger: Electrons, Photons, Muons, b-tagging, Jets, ETmiss, t’s Demonstrate SM physics (top, W, Z, etc) Understand the background processes
From the Tevatron to the LHC…. Higgs cross section rises, but so do other channels.
H→γγ ATLAS Issues: Excellent photon ID needed. Distinguish from isolated π0 Understanding of shower shapes in calorimeter. Photon conversion useful? ~50% convert before the calorimeter. Reconstruct using tracker.
H→γγ CMS Lumi needed for 5 -σ discovery Significance for given lumi. Improvement obtained using optimised analysis, 6 variables. Note effect of systematic uncertainty.
H→ZZ*→ 4 l Clean channel; interesting for 130<M<160 Ge. V and >180 Ge. V. Good acceptance and lepton reconstruction important. Irreducible backgrounds: qq, gg→ZZ*/g* 4 l Reducible: Zbb→ 4 l, tt→ 4 l CMS
H→WW*→llνν CMS Discussed earlier Important around 160 Ge. V. Tevatron may already have eliminated this region.
Vector boson fusion Signatures: 1. Two forward “tag” jets (high-p. T) with large Mjj 2. No jet activity in the central region Typical cuts : • Tag jets are highest ET jets in opposite hemispheres, • ET>~40 Ge. V, Dhjj>~4, Mjj>500 -1000 Ge. V. • Higgs decay products between tag jets in h • No additional jet activity in the event
H→ττ→l+ had +ν’s in VBF CMS study Need to recognise the tau’s Need good missing ET resolution
Other search channels pp → WH, ZH, tt. H with H→different final states tt. H → ttbb (one t decaying semileptonically) Complex, needs good b tagging
Combined ATLAS+CMS
CONCLUSIONS Much progress but still a long way to go. We hope for: -- even more Tevatron luminosity -- improved experimental methods We are on the point of excluding the SM Higgs around 160 Ge. V at the Tevatron. Lower is harder. Brilliant prospects for LHC.
The Limit In Combination
Channels included in latest round of combination CDF lumi [fb-1] m. H range Channels Note# new new 1. 9 1. 7 1. 0 1. 9 110 -150 SS, JS, 1 SNN 9136 110 -150 SS, JS 9166 100 -150 SS, 1 S 8742 120 -200 ME+NN 9163 1. 7 0. 9* 0. 9 1. 7 1. 1 105 -145 120 -200 DØ Note: The mass grid doesn’t line up! ST, DT, NN 5472 DT 5506 ST, DT 5482 NN 5489, 5537 2 DNN 5485 *previously erroneously reported as 1. 7 fb-1
The CDF+DØ Limit In Combination m. H (Ge. V) 110 115 120 140 160 180 200 Median Expected limit/SM** old new 3. 7 3. 5 4. 1 3. 8 4. 9 4. 8 4. 0 3. 9 1. 9 2. 8 5. 9 Observed limit/SM 4. 3 4. 2 6. 8 5. 8 1. 1 2. 5 9. 2
- Slides: 76