FBC2 Meetings

• Long discussion about what to do

• Truth study
• Semi-Resolved
• Luka systematics suggestion:
  • 2->4 as nominal + fouirer modes
  • constrain the fourimodes (the number of modes and their coefficeints in 2->mixed)
  • can do, variance, bias and spurious signal tests

https://indico.cern.ch/event/1708261/

• Why do we have more background in the low-pT analysis?
  • Should think about background as 3b+ISRb
  • Made it clear that the semi-resolved is a 3b analysis.
  • Gave good spin on Semi-Resolved.
• Gave good presentation of SvB studies… (sensitrivity probably overestimated)
• ATLAS HH paper https://cds.cern.ch/record/2925853/files/ATL-PHYS-PUB-2025-006.pdf

• Semi-resolved
  • Orthogonal to boosted and resolved ?
  • BTagging ?

https://indico.cern.ch/event/1699809/

• Long meeting: Good updates Evan on synthetic data
• Apdate from Lei Zhao … good feedback

• Discussion

• 1st look at 2024

• Chat about Hbb presentation strategy

• Jet-veto studies

https://indico.cern.ch/event/1678220/

• Lots of contributions
• Gave MvD slides… Looks liek our SvB_Feynnet is not right
• [>>] Making 3m+1T selection ?
• [>>] Make / send Friends for vX sampels
• Lukas talk: Why does the lower threshold preserve the mass shape ?
• Send mixed_all …
• I also ran into the nans… they were do to numerical instablities when boosting frames… now fixed.
• would be intersting to know why the 10e4 event weight… From event with same hemi
• how do they do the morphing… classifier w/weights …effectivily MvD ?

• Evan: btagging in pairing vs SvB Using btagging in the pairing is less selstive to Bkg mismodelleing than in the SvB Failiure mode is if SR and CR have differnet 4/3 ratios Using in SvB is first order effect Failiure mode is if 4 nad 3b is different ratios are dif

• AE: https://indico.cern.ch/event/1673100/

  • Gave an update of low pt study with first look at limits
    • Fine tune the model / Joint Training ?

  Low Pt HH4b

• 1st look at Low-pT from AE: Low Pt HH4b … this will work! https://indico.cern.ch/event/1665695/

• FSR Recovery

  • Truth tag FSR: dR vs pT-ratio
  • Calibrate the 15 GeV jets?
  • Starting to Add ttbar cut flows using weights

"In processor_HH4b.py we have the option of plotting the ttbar (fourtag and threetag) processes from data by weighted with events with the corresponding FvT  probablities (option enabled with 'plot_ttbar_with_weights') When we do this I would also like to compute and output the corresponding ttbar cutflow where the event counts are summed with the corresponding weight."

• Decent amount of iteration…

"Oh wait, I see whats wrong.  We dont want separate _cutFlow_ttbar4b and _cutFlow_ttbar3b cutflows. We want one _cutFlow_ttbar but the cutFlow_ttbar._cutFlowThreeTag should be filled with threeTag events using weight_d3_to_t3 weights and cutFlow_ttbar._cutFlowFourTag shoudl be filled with threeTag events using weight_d3_to_t4 weights "

https://indico.cern.ch/event/1655563/

• Gave talk: people not as excited as I was …
• Suggestion to plot (limit ) the delta y
• Semi-merged analysis … is really a low-pt analysis

• DIscussion of pairing algorithm
• [>>] Slides re:synthetic data

• Lei talk:
  • Better choice for clustering algo ?
  • Better way to move PF cands

• https://indico.cern.ch/event/1639802/
• Gave talk .. Run3 Hemisphere Mixing Went well
  • [ ] How to get dressed bs in output files ?
  • [X] What is the selection for output jets in mixing and synthetic datasets ? Sent the following to Marina A: For mixed data all the input jets are mixed so the only preselection is the cut to make the nanoAOD (This is b/c the performance of mixing algorithm doesnt really scale with the number of jets) For the synthetic data I do use. a preselection to speed up the DeClustering algorithm The preselection is (event.Jet.pt >= 20) & (event.Jet.jetId >= 2) & event.Jet.lepton_cleaned & (np.abs(event.Jet.eta) <= 4.7) That preselection is on the input jets used to cluster (and compute the templates) when I sample the template to make the synthetic (fake) jets I make up new pts and eta. IF the fake jets fail the pt or eta acceptance I try again with a new random number. But I only allow 8 attempts, after which point I just accept the jets so here that can be some (hopefully small) number of jets that have pt<20 (For the b-jets I require pt > 30 and eta < 2.5)
• Lei gave a good talk
  • [X] Comment on signal suppression is a good thing
  • [X] Should make the animation pdf friendly
  • [X] Add another step in his movie

• https://indico.cern.ch/event/1621376/
• https://cernbox.cern.ch/files/spaces/eos/user/m/mkolosov/Run3/Run3HHto4b.docx

• Talk about AE talk
• Update on mass decorelation

• Gave Synthetic dataset slides… Went well

• Contrastive learning… Not clear what it is good for…
• Loukas seems pumped

• Sindu Talk
• Show mHH with an intermideate cut in SvB
• Double ratio correction in the intermediate bin?
• https://cmsweb.cern.ch/das/request?input=dataset%3D%2FGluGluToBulkGravitonToHHTo4B_M-1000_narrow_TuneCP5_13TeV-madgraph-pythia8%2FRunIISummer20UL16MiniAOD-106X_mcRun2_asymptotic_v13-v2%2FMINIAODSIM&instance=prod/global
  • https://indico.cern.ch/event/1523858/contributions/6540826/attachments/3077457/5446299/CMS_Presentation_29May.pdf

• Paper meeting with Marko / Marina / Stephane
• Marko will write the minutes

• Marina meets Stephane and Marko re:paper
• Discussing details

• Really pushing ZZ and ZH Run2+3 combination (We should prepare to add the 5/fb?)
• Not much… chat Jaco and Loukas about continuing to work together.
• They are happy to have me.
• Want me to reach out to ETH see what their plans are

• How hard is it to do the combination of ZZ and ZH ?

• Discussion of proceeding to the unblinding step 3
• Just waiting on the boosted

• https://indico.cern.ch/event/1558329/
• Unblinding stage 2
• Synthetic data improves in mass-decorrelation study

• Discusion re:paper
• Run-3 Resolved unblinded
• We will also unblind!

• https://indico.cern.ch/event/1551592/
• Looking into preaproval slides
• Luka using synthetic data for background!

• Preapproval May 27th
• Run3 new ntuples
• Chat Boosted
• AE slides for the boosted
• Naive Limit Combination

• ZZ and ZH for run3 results look good
• Discussion of paper strategy
• Showed bootstrap results

• [<<] Make Run3 CR cut and make a print of event/run numbers

• [>>] Print out first 100 events in 2022_preEE Data and Synthetic data

• Not Much

• Not much

• Synthetic datasets still look bad in CR
• Discussed Marina's USCMS talk on Friday
• (JA thinks: … Vivek will not be happy)

• [>>] Closure on the Synthetic datasets
• Discussion re:nJets and ttbar

• More evidence that Reclustering might be the way to go

• got the comments on Run 3… Not bad

• Not much

• Synthetic data still factor of 2 too low! [>>] Reskim and try again ?

• Not much

• Nice studies by CL:
• https://indico.cern.ch/event/1481600/

• Discussion of jet energy correction procedure
• Training discussion … Bailed
• Evan's email

• "JCM-like" correction degrades FvT closure in Run-3

• Discussion re:jet-veto maps
• [>>] Request presentation
• Gave talk on Classifier Studies… Summary: doesnt matter much which SB to use

• Need to push on update Run-2 status
• Boosted overlap numbers

• not much, discussion of new jet calbiration

• Not much
• Will implement Run-3 selection

• Not much… to much emphasis on reconstruction efficiency. What really matters is if the signal events end up in the SR. The SvB classifier will take care of the rest.

• Mail by Marko…
• Test the FvT from the fitting OR of SB ZZ and ZH.

• Not much: Jaco and Lukas away

• https://indico.cern.ch/event/1445702/
• Not much…
• Lukas on mass decorrelation
• Sent Loukas link to Low Pt HH4b

Not much.

Should Boosted do ZZ/ZH -> 4b ?

• Claim they already setup to do it for ZH, will look into making the signal

Will also try ZZ

First meeting in two week.

B-regression is a difference between the two resolved analyses.

Javier will send us ntuples with boosted BDT for the Run2 combination