The ATLAS experiment at CERN searched for events where two Higgs bosons are produced together and decay to a pair of bottom quarks (b overline{b}) and two photons (γγ). The analysis uses 308 inverse femtobarns (fb⁻¹) of proton–proton collision data: 140 fb⁻¹ at 13 TeV and 168 fb⁻¹ at 13.6 TeV collected between 2015 and 2024. The result is consistent with the Standard Model: the measured Higgs-pair signal strength is μHH = 0.9 +1.4 −1.1, and a 95% confidence-level upper limit of μHH < 3.7 is reported. The same data limit the Higgs trilinear self-coupling modifier to −1.6 < κλ < 6.6 at 95% confidence.

The search looks specifically for events with exactly two well-measured photons and at least two jets identified as coming from b-quarks (b‑jets). Selected events are split into categories based on the combined mass of the b overline{b}γγ system, the expected signal purity, and the data-taking period. In each category the analysis fits the distribution of the two-photon mass to look for a narrow peak from Higgs decays on top of a smoothly varying background. Backgrounds include single-Higgs production that decays to γγ, genuine di-photon production with extra jets, and events where jets fake one or two photons; the latter are constrained with data-driven methods.

This update improves on earlier searches in the same final state. The analysis benefits from a new b‑tagging algorithm based on transformer neural networks (called GN2) and a kinematic fit that sharpens the reconstructed masses of the b overline{b} and b overline{b}γγ systems. Signal and background processes were modelled with detailed Monte Carlo simulations. Tools mentioned include Powheg and MadGraph for signal generation, Pythia for Higgs decays and showering, and Sherpa for continuum di-photon backgrounds. Some simulated samples used a faster detector approximation; the paper states the difference versus full simulation is within the statistical uncertainty for the reported limits.