1 / 16

Update on Higgs self-coupling

Update on Higgs self-coupling. Tomáš Laštovička FZU AV CR WG6 Meeting 29 /11/2011. Outlook. Quick reminder Reconstruction of the Higgs mass Secondary vertex assisted jet finding A different approach to neural net training Higgs presence in 595 sample – cross-check

winter
Download Presentation

Update on Higgs self-coupling

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Update on Higgs self-coupling Tomáš Laštovička FZU AV CR WG6 Meeting 29/11/2011

  2. Outlook • Quick reminder • Reconstruction of the Higgs mass • Secondary vertex assisted jet finding • A different approach to neural net training • Higgs presence in 595 sample – cross-check • Summary and next steps

  3. Analysis chain • FastJetkT algorithm, R = 0.7 in exclusive mode requesting 4 jets • MarlinKinFit • The combinations of 2+2 jets with highest probability of having the same invariant mass selected and re-fitted. • Analysis chain essentially follows the chain for h→bB analysis • Neural net selection input variables added • pTmax, pTmin, MarlinKinFit probability, ymin, event invariant mass • which is nice, on the other hand, one can not use the momentum conservation due to neutrinos in the final state of hhnunu (and qqqqnunu).

  4. SiD Samples with 60BX Overlay – Full Sim/Rec i.e. qqqq, qqqqνee, qqqqee are not included, nor is the HHee channel (ZZ fusion)

  5. Higgs invariant mass reconstruction I • One of the most powerful quantities to select the signal events. • The reconstruction was rather poor, done in the following way: • Make 4 jets in 2 pairs → 3 combinations • MarlinKinFit constraint – both pairs have the same invariant mass, re-fit jets • Choose the combination with the highest probability (taken from MarlinKinFit).

  6. Higgs invariant mass reconstruction II • A different approach: • Use MarlinKinFit’s probability only to find the jet combination • Take measured jet four-vectors, not re-fitted ones • Plot the combination with higher mass (and the one with lower mass too) In HHnunu events there is most often something missing… Looks much more reasonable

  7. Higgs invariant mass reconstruction III • Adding max(M1inv,M2inv) to neural net inputs indeed improves the signal separation • Although not as much as one would expect • Cross section uncertainty: from 19-21% to around 18% • min(M1inv,M2inv) does not appear very useful

  8. Secondary vertex assisted jet finding I • Red and green are particles from different Hs, blue squares are secondary vertices and black points are directions of Hs (no RPs and jets overlaid) φ η

  9. Secondary vertex assisted jet finding II • Use secondary vertex information in the following way • Find jets, find secondary vertices (LCFI) • For each secondary vertex add a particle (neutral B-meson) to container of particles • Find jets again • Carefully remove added particles from containers and from jets • Do all the rest of the LCFI package • After implementing this idea, I run into a couple of issues • Some jets consisted of one added particle alone, i.e. no particles • Higgs invariant mass reconstruction is not improved. • Consequently, the results on self-coupling are actually worse. • This approach would require further work.

  10. A different approach to neural net training • The usual way • take variables, feed them to one neural net, flag signal as 1, background as 0. • Another way • take the same variables • feed them to two neural nets • train 1st to separate 4-jet events from 2-jet events • train 2nd to separate signal events from 4-jet background • merge outputs from both neural nets in 3rd neural net with 2 inputs, 1 output • This further improves the signal separation. from ~18 % to 16-16.5%

  11. Higgs presence in 595 sample (ee→ qqqqνν) • 595 sample (qqqqnunu) contains 120GeV Higgs • It was subtracted statistically so far = “subtract 527 from 595” • On the MC level there are no intermediate particles • combine quarks with electron/positron parent particles • If there is a combination where both pairs have Minv ≈ 120GeV reject event • In statistical subtraction, the “signal sample”, present in the background, was effectively used to train the neural net to recognize signal as background • i.e. signal was assigned to signal and background 1:1 • Despite tiny signal amount, one would expect improvement • Nevertheless, when removing 2-Higgs events, the result is about the same as for the statistical subtraction method (on the level of fluctuations) • i.e. the cross section uncertainty is around 16-16.5% → 14% on HHH

  12. Summary and next steps • The best numbers so far: • Cross section uncertainty: 16% • HHH stat. uncertainty: 13.5% • Nsignal = 135 Nbkg = 327 • only 8% signal selection efficiency • no qqnunu, qqnue, qqee , qq • surprising contribution from h→bb (36) • With all the 2-q backgrounds from h→bb analysis plus qqqqνν background. • Other 4-q backgrounds not included/simulated • Stephane generated 130GeV, 140GeV signal samples. • Lower CLIC energies: 1.4TeV Side-comment: There should be no useful information hidden in this plot, since it was used as a NN input.

  13. BACKUP

  14. Control Plots I • Where is the problem? • Higgs invariant mass reconstruction is poor, so is the jet flavour tag it’s the jets, they are poorly reconstructed

  15. Control Plots II • ymin and ymax should be the most useful quantities in order to separate 4Q events from 2Q events

  16. Control Plots III • Invariant mass plot after neural net selection • Working point = lowest statistical uncertainty Signal: ~200 of 1240 Background: ~1440 (2Q contribution is around 400)

More Related