Higgs Width Portfolio

A

Rediscovering H→ZZ*→4ℓ

Foundation for every task. Select four-lepton events in 36.6 fb⁻¹ of 13 TeV ATLAS Open Data, pair them into Z bosons and fit the on-shell signal strength μ_on, which anchors the width measurement.

Four-lepton invariant mass

Profile likelihood for μ_on

Binned fit in 110–140 GeV with pyhf: Higgs signal, ZZ* and reducible backgrounds with normalisation, luminosity and MC-statistics nuisances.

Selection: trigger + matched lepton, loose/medium ID and isolation, p_T > 20/15/10 GeV, two same-flavour opposite-sign pairs, Z₁ closest to m_Z with 50 < m₁₂ < 106 GeV, m₃₄ threshold rising from 12 to 50 GeV, ΔR > 0.1, J/ψ veto.
Reducible background (Z+jets, tt̄) has only a handful of high-weight MC events; its shape is smoothed per sample with a kernel estimate and given a 50% normalisation uncertainty.

selection.py fit_onshell.py tests

B

Width from the off-shell tail

Task: Higgs boson width determination using H→ZZ events. Above 2m_Z the gg→H*→ZZ rate scales as μ_off = μ_on·Γ_H/Γ_H^SM, while the on-shell rate does not depend on Γ_H. Comparing both constrains the width. The signal interferes destructively with the gg→ZZ continuum, so the yield is n = K(μ S + √μ I + B_gg) + μ_qq Q + O.

Off-shell m_4ℓ as the width changes

Γ_H/Γ_H^SM

Templates: MCFM 7.0 (JHUGen fork) gg→ZZ→2e2μ at LO with K = 1.7, folded with a per-bin efficiency calibrated on ATLAS Open Data qq→ZZ against MCFM NLO.

Likelihood scan in Γ_H/Γ_H^SM

Expected reach vs luminosity

offshell.py MCFM cards tests

C

How stable is the result?

Task: tests of the stability of the results with respect to various assumptions. Each scenario changes one assumption of the baseline fit. Pseudo-data are generated with the baseline model at Γ_H = Γ_H^SM and HL-LHC luminosity (3000 fb⁻¹), so any shift away from 1 is the bias the assumption would introduce.

Fitted width under each assumption (truth = 1)

Couplings that differ on and off shell

If (κ_gκ_Z)² off shell is c times its on-shell value (e.g. running couplings or new states in the loop), a fit assuming c = 1 returns c·Γ_H.

c

Relaxing the on/off-shell link

Why the indirect width needs B_BSM = 0

run_moduleC.py

D

EFT interpretation

Task: effective field theory interpretations. Anomalous HZZ couplings a₂ (CP-even) and a₃ (CP-odd) correspond to dimension-6 operators of the Higgs basis. Their amplitudes grow with the energy through the HZZ vertex, so the off-shell tail constrains them far better than the on-shell rate.

Off-shell signal for different a₂

Constraints: on-shell only vs on + off shell

With rates only, the on-shell measurement cannot separate a coupling change from the overall gg→H strength μ_F (its curve stays at zero). The off-shell tail breaks that degeneracy. For a₂ > 0 a second degeneracy remains: the larger signal is cancelled by the stronger destructive interference, which angular observables would resolve.

run_moduleD.py EFT cards MCFM patch

E

Electron shower shapes in a LAr calorimeter

Task: study of electron shower shape development (authorship qualification). A Geant4 application in C++ models an ATLAS-like barrel: presampler, fine-η strips, middle and back Pb/LAr layers with ATLAS cell sizes, plus an Fe/scintillator hadronic section. 32 000 showers of e⁻, γ, π⁰ and π⁻ were simulated at 10–100 GeV.

Shower-shape distributions at 50 GeV

Material in front of the calorimeter

Upstream material (inner detector, solenoid, cryostat) starts showers early. This is the main source of data/MC differences in shower shapes.

Particle identification power (25 GeV)

Longitudinal profile and resolution

g4calo (C++)showers.py tests

✓

Evidence and reproducibility

Stack

Data: ATLAS Open Data 13 TeV, 2025 education release, exactly4lep skim (36.6 fb⁻¹).
Generators: MCFM 7.0 (JHUGen fork) for gg→ZZ signal, interference, continuum and anomalous couplings; MCFM NLO qq→ZZ. Built with gfortran 16 on Apple silicon (two legacy array declarations and a run-time coupling reader patched).
Simulation: Geant4 11.4.3, C++17, FTFP_BERT.
Analysis: Python, uproot, awkward, pyhf, iminuit, scipy, pytest.

Reproduce

python scripts/download.py        # ~1 GB of skims
python scripts/run_moduleA.py --reprocess
generators/mcfm/run_mcfm.sh; generators/mcfm/run_eft.sh
python scripts/run_moduleB.py; python scripts/run_moduleC.py; python scripts/run_moduleD.py
g4calo/run_production.sh 1000; python scripts/run_moduleE.py
pytest

ROOT figures (PDF)

Limits of this study: education-level open data (36.6 fb⁻¹, two years), m_4ℓ only (no matrix-element discriminant), LO gg templates with a flat K-factor, no VBF off-shell component, a planar toy calorimeter. Results are an educational reproduction, not ATLAS results.