1. Max Baak 1 Overview VBF H  WW (ll) Max Baak, NIKHEF Atlas Nijmegen Uitje 22 April ‘08

2. Max Baak 2 People On behalf of:  Egge v/d Poel – supervision Peter Kluit  John Ottersbach – supervision MB / Peter Kluit  Marcel Raas – supervision Frank Filthaut  Barbara Millan (master) – supervision MB  Rikard Sandstrom (postdoc)  MB (post-doc) → start work for cern May 1 st  Gijs van den Oord (aio) – theory

3. Max Baak 3 Overview  Brief introduction VBF H → WW (ll)  (CSC) Sensitivity studies  Plan  Conclusions

4. Max Baak 4 Vector boson fusion: H  W + W -  ll VBF H  WW (ll)  Clean events: color-coherence between initial and final state W-radiating quarks  suppressed hadronic activity in central region  Spin zero Higgs: charged leptons prefer to point in same direction.  Two forward, high-P t jets from WW fusion process (“tagging jets”).

5. Max Baak 5 Higgs production x-sections (NLO)  H  WW: signifcant discovery potential over wide mass range (>130)  Gluon fusion H  WW channel suffers from large, uncertain QCD WW background contribution.  VBF: second significant production mechanism for Higgs at LHC VBF H  WW gg  H  WW VBF Hgluon fusion H

6. Max Baak 6 VBF H →WW (ll) x-section  Some leakage of gluon fusion Higgs through VBF selection  Expect ~ 40 reconstructed Higgs events / fb (@ 170 GeV/c 2 ) VBF H  WW  ll gg  H  WW  ll requirement: lepton → e / mu

7. Max Baak 7 Did you know?  “No-lose” theorem applies to W-W scattering: something must show up below m WW < 500 GeV/c 2 to avoid unitarity violation. → See talk by Gijs van den Oord for x-sec calculations.  Main background components ttbar production, W(W) + jets, QCD & EW   Great synergy with top and Z (muon) reconstruction

8. Max Baak 8 Higgs fitter (Sensitivity studies) Strategy:  Multi-dimensional fit for optimal signal significance Include all observables most sensitive to VBF H → WW (ll) signature Combined fit to signal sample and background control samples  Background shapes and normalization from (data) control samples. Minimal dependence on MC shape information  For now, did not focus on how background is exactly described. Use multi-dimensional Keys pdf for correct model of correlations, etc.  Parameterized signal model Use simple and generic shapes obtained from Monte Carlo.

9. Max Baak 9 Sample categories  Fit variables (5-dim):  (ll),  (ll), m T (H),  (jj), m(jj)  Higgs events mostly end up in BVeto-sigbox  Use other boxes to extrapolate bkg description into BVeto-sigbox.  Four possible background sample approximations: 1  3, or 1  2 1  3  correction_factor(2/4), or 1  2  correction_factor(3/4) BTag sampleBVeto sample sigbox sideband  (ll)  (ll) 1 2 3 4

10. Max Baak 10 One fit example m H (true) = 170 GeV/c 2 background signal + bkg Transverse Higgs mass (GeV/c 2 ) ParameterValueGl. Corrl.Input m(higgs)168 ± 812%170 n(bkg)90.5 ± 7.493%86 n(higgs;2j)18.6 ± 5.525% n(higgs;3j)9.6 ± 7.938% 1/fb ATLAS CSC BOOK 27

11. Max Baak 11 Bkg-only samples Sig+bkg samples m(higgs)=180 GeV Entries per bin 2 x  2 Significance determination  Generate many pseudo-experiments (using grid): 1.Background-only samples 2.Background + signal samples, for various Higgs mass.  Fit each sample with background-only and signal+bkg hypothesis  Plot  2 between the fits.  Extrapolate fraction of bkg-only sample to fake average signal sample. Bkg-only samples faking signal

12. Max Baak 12 Significance results ATLAS CSC BOOK Results with 1/fb of data:  If higgs mass = 170 GeV/c 2 : Close to 2.5 sigma signal sensitivity 9 GeV/c 2 mass resolution.  For m H < 140 GeV/c 2, similar sensitivity to gluon fusion analysis. Background shapes and normalization obtained fully from data control samples.

13. Max Baak 13 Global plan Preparation! Focus first on data-control samples. (Higgs: not a day-1 analysis.)  Control sample studies  Signal optimization  Redo Higgs analysis, focus on systematics.

14. Max Baak 14 Perspectives “Strong” points:  Knowledge of muon reconstruction  Use of advanced fitter techniques  Top quark knowledge Muon ID  Loosened muon selection  Connection of Higgs to Z sample Sample selection  Optimization of signal selection  Collaboration with UC Irvine Background studies  ttbar background extrapolation  Others may follow Fit  Fit optimization / tweakage  Studies of dominant systematic effects

15. Max Baak 15 Control sample: Z + n-jets Z + 2j  Important control sample in VBF H  WW(ll) for MET, muon reconstruction and alignment studies.  Link to first data Egge vd Poel 1Alignment studies Impact on m(ll) mass resolution 2Missing Et resolution In n-jet sample, compare with VBF H  WW (ll) 3Selection: loosened muon id Combined, stand-alone, segment tagged, calorimeter tagged Obvious: direct link with  reconstruction Plan: incorporation into Higgs signal modeling.

16. Max Baak 16 Control sample: ttbar background ttbar background  Dominant background contribution in VBF H → WW analysis  Link to first data Marcel Raas  Extrapolation of di-lepton ttbar background into signal box using b-tag information.  Extrapolation of di-lepton background from semi-leptonic bkg. And more by MR!  MC generator dependency of signal and background selection  Fast vs full simulation, vs different generators

17. Max Baak 17 Higgs signal modeling  Start out with modelling of transverse mass  Use empirical description for background transverse mass John P. Ottersbach  No well-described Higgs transverse mass (m T ) model available  Decomposition of reconstructed m T into: Kinematic piece Experimental piece  MC generator independence  Find motivated description working for entire Higgs mass range  Extend description to other observables.

18. Max Baak 18 Signal optimization (& Fitter)  CSC studies used “common” lepton and (tagging) jet selection.  Signal selection not yet optimized... Barbara Millan  Signal optimization studies using TMVA (master’s studies) Rikard Sandstrom / MB  Optimization of signal sensitivity using neuro-evolution Automated feature selection Genetic evolution of neural network’s internal structure  Optimization study with signal significance obtained from fitter. “Simple” 2-dimensional fit, with transverse mass and neural network output.  Started collaboration with UC Irvine (Whiteson brothers)

19. Max Baak 19 Fit studies MB  John and Marcel have been learning how to setup simultaneous fits (using background control samples).  CSC 5 dimensional fit: too slow for optimization studies → Need to develop fast 2 dimensional fit (previous slide) 5d fit, to be done: lots!  CSC Sensitivity study has not been extremely thorough. No systematics at all. Missing check of fit model assumptions Validation: unexplained fit biases present. Fit not yet optimized. → Leave for now … focus on studies with 2-dim fitter.

20. Max Baak 20 Conclusions  Group is acquiring mass. Great! (must be Higgs field ;-)  CSC studies: VBF H → WW (ll) as sensitive as gluon fusion. In comparison: background much better under control. Global plan: preparation!  Focus on control sample studies, from first data Z’s, muon reconstruction ttbar background extrapolation  Signal optimization with neuro-evolution

21. Max Baak 21 Backup

22. Max Baak 22 VBF: Higgs  W + W -  ll ATLAS TDR VBF H  WW  VBF H  WW: signifcant discovery potential over wide mass range  “No-lose” theorem applies to W-W scattering: something must show below m WW < 500 GeV/c 2 to avoid unitarity violation.  Great synergy with Z and top reconstruction

23. Max Baak 23 Keys pdf  Kernal estimation pdf : provides unbinned, unbiased estimate pdf for arbitrary set of data K. Cranmer, hep-ex/0011057  E.g. 1-dim keys pdf heavily used in BaBar.  I extended this to n-dim keys pdf to model any bkg distribtion. To be included in HEAD of RooFit  Automatically includes correct correlations between all observables

24. Max Baak 24 Higgs transverse mass  m T (H) : calculated like normal transverse mass Assume that : m( ) = m(ll)  Modelled with: double-sided exponential, with 2 different lifetimes, convoluted with Gaussian  In fit to data, only float Higgs mass (hopefully) m(H) TRUE = 170 GeV/c 2 tauL : missing momentum in z tauR : m(ll) = m( ) sigmaC:missing Et

25. Max Baak 25 Higgs (ll), (ll)  (ll)  (ll)  WW comes from spin zero Higgs: charged leptons prefer to point in same direction. Define angle in transverse plane  ll.  Significant fraction of various backgrounds does not have (anti-)correlated W spins. Higgs W–W– W+W+ l+l+ l–l–

26. Max Baak 26 Higgs (ll), (ll)  Modelled with two Gaussians (multiplied in decorrelated frame), and uniform “bkg” term.  “Bkg” is kinematical, not mis- reconstruction.  Small positive correlation angle  Signal defined as number of events in Gaussians.  In fit to data, shapes are fixed

27. Max Baak 27 Higgs m(jj), (jj)  Still need to find generic parametrization... looks complicated  For now modelled with keys pdf  (MC dependency...)  In fit to data, shape is fixed