ATLAS X->SH->bbgg

Summary

The paper presents a search for a hypothetical scalar boson X decaying into another scalar S and a SM Higgs boson in the bb+gamma-gamma (bbgg) final state. No significant deviation from the SM expectation is observed.

This search is a natural extension of the important HH->bbgg search. There have been several results in the field targeting the X->SH signature in various final states. The bbgg channel offers unique sensitivity in the low-mass region.

The results presented here are an extension of previously published Run 2 results (JHEP 11 (2024) 047). The new result has improved analysis techniques and adds data taken in 22 and 23. Relative to the previous analysis, the object reconstruction, identification, and calibration were improved. There was also a reoptimization of the size of the mgg SR. This result improves the sensitivity (by ~15–75%), where only 9-30% of improvement is expected from the additional data alone.

The background is relatively small and can be reliably estimated from mgg sidebands.

My only significant comment is that more figures should be made public to aid in presenting the material, e.g.: mgg plots, PNN in the CR and distribution of the key PNN inputs: CR, SR and high PNN output. Plots of PNN and inputs before and after reweighting.

General coments/questions

Section 5.3 Normalisation factors

"These normalisation factors are used to correct the normalisation of the γγ+jets background components in the SRs and CRs before the likelihood fit described in Section 7. They are then allowed to float in the fit."

Are these normalisation factors constrained in the fit by the CR uncertainties? If not, why not? Seems justified. If so, say it here.

Section 5.3 Residual gg+jets discrepancy

"After applying those factors, a residual discrepancy between the PNN distributions of the 𝛾𝛾+jets processes in data and simulation is still observed in the CR. This discrepancy is most prominent in the mbb*gg and mb*gg distributions."

Confused by the statement that the discrepancy is in the PNN distributions. The next sentence you say it is mbbgg and mbgg, which are inputs to the PNN. What PNN distribution shows this discrepancy? I guess it's the PNN in the CR, as you can't judge background mismodelling in the SR. If so, should say that.

Why not correct the input gg+jets variables with the event weight before training the PNN? I would at least use that as a systematic uncertainty. e.g.: how does the change in PNN output with and without this re-weighting compare to the change in PNN output with Sherpa vs Madgraph, which you use to assess systematic uncertainties?

Section 7: Cross section discusion

"The signal strength μ is treated as the parameter of interest and is defined as the ratio of the observed event rate to the expected event rate, where the latter is computed assuming a 13 TeV cross-section equal to the observed upper limits from the previous analysis [18], and the 13.6 TeV cross-section as introduced in Section 3. The choice of the 13 TeV cross-section value is arbitrary, as it does not impact the limits or observed significance values."

The first sentence talks about using both the cross-sections at 13 and 13.6 TeV. The next sentence then says the choice of using (presumably only) the 13 TeV cross-section is arbitrary. Please clarify the text.

Section 7: Track Reconstruction changes

"Most of them fail to pass the current 2 b-tagged preselection, either because a photon is now reconstructed as an electron, or because a jet is no longer tagged as a b-jet"

Sounds like something has also changed in the track reconstruction as photons are now being reconstructed as electrons. It would be good to also discuss this when talking about changes to the analysis.

Minor:

  • Abstract: "lighter" scale S is confusing. S is not restricted to be lighter than the higgs mS: 15-500 I think you mean S is lighter than X, but I had to reread to get the point.
  • For the reader it would nice to have a rough idea of relative importance of the different backgrounds before 5.2, where you talk about MVA training.
  • Section 6: It would be good for the reader to have a high-level statement about the relative importance of statistical vs systematic uncertainties at the beginning; this would help to put this section in context. (e.g.: your last sentence of this section "Even for the mass point with the largest systematic uncertainty, the results are dominated by the statistical uncertainty." should be in the 1st paragraph.)
  • "adjacent bins are iteratively merged until the rightmost bin contains at least one expected background event." I was under the impression that you needed 3 bins for the assumptions of the profile likelihood to be valid.

Comments:

  • Everyting in a ML, only see ML output distributionsin the SR and Limits.

My Notes

  • S and a lighter scalar S (WHy lighter?)
  • H->gg Run2+22+23
  • X: 170 - 1TeV S: 15 - 500
  • Extension of Run2 results from ATLAS (JHEP 11 (2024) 047
    • Preivous: 39 fb to 0.09 fb
      • This result: 9 fb to 0.06 fb. Factor of ~4 improvment

Whats new?

22+23 data Relative to the previous analysis the object reconstruction, identification and calibration was improved. Btagging Improvement of 20%, going up to about 65% at low-mass.

Analysis

  • Dominant background gg+jets data-driven Small background form H+jets
  • SR: mgg ~ 125
  • Sidebands define CR to fix gg+jets normalizaion
  • Analysis with Parameterised neural networks (mS and mX)

PLB-D-25-01868R1

Reviewer Invitation for Revised Version PLB-D-25-01868R1

Backlinks