W H → W W W → Search for Standard Model Higgs Boson lν.jj.jj A. Podkowa Production in the WH → WWW → lν.jj.jj Background Channel at DØ. SM Higgs Search MVA Progress Anthony Podkowa What I Did W Reco. SIST 2011 Multijet MVA Final MVA August 9, 2011 Results 1 / 21
Outline W H → W W W → lν.jj.jj 1 Background A. Podkowa The Standard Model & High Energy Physics How Do We Look for a Higgs? Background SM Machine Learning and Multivariate Analysis Higgs Search MVA 2 Progress This Summer Progress What I Did 3 What I Did W Reco. W Reconstruction Multijet MVA Final MVA Reducing Multijet Background Results Training a Final MVA 4 Preliminary Results 2 / 21
The Standard Model & the Higgs Boson W H → W W W → lν.jj.jj A. Podkowa Background SM Higgs Search MVA Progress What I Did W Reco. Multijet MVA Final MVA Results Models particles and their interactions Higgs Boson is the only missing piece of the Standard Model 3 / 21
What Exactly Are We Looking For? W H → W W W → lν.jj.jj WH → WWW → lν.jj.jj A. Podkowa Background SM Higgs Search MVA Progress What I Did W Reco. Multijet MVA Final MVA Results Involves searching for a small Signal in about 1400 × as much Background! 4 / 21
How Do We Detect Particles? W H → W W W → lν.jj.jj A. Podkowa Background SM Higgs Search MVA Progress What I Did W Reco. Multijet MVA Tracker: Hadronic Calorimeter: Final MVA For tracking charged Mostly absorbs energy from Results particles quarks and gluons (jets) EM Calorimeter: Mostly absorbs energy from Muon System : electrons and photons Mainly Muons make it here. 5 / 21
How Do We Look for This Process? W H → W W W → lν.jj.jj A. Podkowa Overview: 1 Use a C++ code framework (wh cafe) Background SM 2 Generate Monte Carlo Simulations corresponding to the Higgs Search signal and background processes MVA 3 Process kinematic properties of the data & MC Progress 4 Train Multivariate Classifiers using Computer Learning What I Did Techniques W Reco. 5 Apply Multivariate Classifiers to the data & MC Multijet MVA 6 Search for excesses corresponding to the signal Final MVA 7 Run statistical analyses to determine the significance of the Results findings (COLLIE) 6 / 21
Machine Learning & Multivariate Analysis W H → Many Moderately Significant Variables into One Very W W W → lν.jj.jj Significant One A. Podkowa We use Machine Learning techniques to perform Multivariate Analyses Background SM Machine Learning occurs in two phases: Higgs Search MVA Training: Progress What I Did Computer analyzes two data samples (signal & background W Reco. MC) for differences based off of a list of variables Multijet MVA Final MVA Results Classification: Computer uses what it “learned” to classify data as signal or background 7 / 21
Progress This Summer W H → W W W → lν.jj.jj Where We Began A. Podkowa Only Electron subchannel Working Background Small amount of selections SM Small subset of the data. Higgs Search No WWW specific variables MVA Progress What I Did W Reco. Multijet MVA Final MVA Results 8 / 21
Progress This Summer W H → W W W → lν.jj.jj Where We Began A. Podkowa Only Electron subchannel Working Background Small amount of selections SM Small subset of the data. Higgs Search No WWW specific variables MVA Progress Where We Are Now What I Did Both Electron and Muon subchannels Working W Reco. Added WWW variables Multijet MVA MVA Training Final MVA MVA Application Results COLLIE Input Generation → Preliminary Sensitivity Plots Added more Data (Up to 7.5 fb -1 ) 8 / 21
What I Did W H → W W W → lν.jj.jj A. Podkowa Background Maintained and Administrated a fork of wh cafe SM Integrated W → jj Reconstruction Code into wh cafe Higgs Search MVA Developed C++ code for: Progress Multijet MVA What I Did Final MVA W Reco. Statistical Inputs to COLLIE (Sensitivity Plots) Multijet MVA Final MVA Debugging Results 9 / 21
W Reconstruction W H → W W W → lν.jj.jj A. Podkowa Background To be able to analyze the intermediate state of the SM channel, we needed to reconstruct the W ’s Higgs Search MVA Need to appropriately combine the jet, lepton and neutrino Progress 4-vectors to obtain W ’s What I Did Thankfully, W → lν was already defined in wh cafe W Reco. Multijet MVA W → jj : required a little thought Final MVA Results 10 / 21
W → jj W H → W W W → 1 Generate each jet combination (12 34, 13 24, 14 23) lν.jj.jj A. Podkowa 2 Calculate the mass of each jet pair. Background 3 Calculate Error in each W mass by using SM ∆ m ij = m ij − m W , Higgs Search MVA where m W = 80 . 399 GeV (PDG) Progress 4 Sum the errors together: What I Did � � � � W Reco. E [ m ij kl ] = � ∆ m ij � + � ∆ m kl � � � � Multijet MVA � Final MVA 5 Select the combination with the lowest summed error Results 6 Label lower mass W as W 1 and the Higher Mass W 2 Allowed Us to Add 25 New Variables! 11 / 21
W Variables–Example W H → W W W → lν.jj.jj A. Podkowa Background SM Higgs Search MVA Progress What I Did W Reco. Multijet MVA Final MVA Results Angle between W lν & W 1 12 / 21
Reducing Multijet Background W H → W W W → lν.jj.jj A. Podkowa Background SM Higgs Search MVA Progress What I Did W Reco. Multijet MVA Final MVA Results Multijet Background is dominant Occurs when we have 5 jets with one “faking” a lepton Solution: Perform a Multivariate Analysis! 13 / 21
Multijet MVA W H → W W W → lν.jj.jj A. Podkowa Background SM Higgs Search MVA Progress What I Did W Reco. Multijet MVA Final MVA Results Train an MVA using just Multijet Background and Signal Reject all events with Multijet MVA Output ≤ –0.5 14 / 21
Multijet MVA W H → W W W → lν.jj.jj A. Podkowa Background SM Higgs Search MVA Progress What I Did W Reco. Multijet MVA Final MVA Results Removes 72% of Multijet Background at a cost of 0.2 Signal Events (9.5%) Results in a 47.1% improvement in the Signal to Background Ratio 15 / 21
Training a Final MVA W H → W W W → lν.jj.jj A. Podkowa Background SM In order to best discriminate between signal and Higgs Search background, we trained Final MVA’s for our channel MVA Progress Utilized many of our new WWW Variables What I Did Trained on all backgrounds, not just Multijet W Reco. Multijet MVA Final MVA Results 16 / 21
Final MVA W H → W W W → lν.jj.jj A. Podkowa Background SM Higgs Search MVA Progress What I Did W Reco. Multijet MVA Final MVA Results Preliminary Stages: Further Optimizations to Come! 17 / 21
Results W H → W W W → lν.jj.jj A. Podkowa Background SM Higgs Search MVA Progress What I Did W Reco. Multijet MVA Final MVA Results Sensitive to WH → WWW → lν.jj.jj to 20 × SM from 150 − 180 GeV This will only get better as we continue to optimize our MVA’s 18 / 21
Recap W H → W W W → lν.jj.jj A. Podkowa Background Much has been accomplished this summer. SM Higgs Search Majority of Analysis Code Working: MVA Both Electron and Muon SubChannels. Progress Multivariate Analysis Code. What I Did Preliminary Sensitivity Plots. W Reco. On our way to building a publication. Multijet MVA Final MVA Results 19 / 21
Acknowledgements W H → W W W → Supervisors: lν.jj.jj Dr. Ryuji Yamada A. Podkowa Dr. Mike Cooke Background Mentors: SM Higgs Search Jamieson Olsen MVA Elliott McCrory Progress Summer Students: What I Did Alex Abbinante (IMSA Graduate) W Reco. Multijet MVA Youssef Sarkis Mobarak (IPM) Final MVA Stephanie Hamilton (SIST) Results WH Group DØ Collaboration SIST Committee 20 / 21
Questions? W H → W W W → lν.jj.jj A. Podkowa Background SM Higgs Search MVA Questions? Progress What I Did W Reco. Multijet MVA Final MVA Results 21 / 21
Recommend
More recommend
