ATLAS finds evidence for one of the Higgs boson's rarest decays
The ATLAS Collaboration has presented first evidence of the Higgs boson decaying into a virtual photon and a real photon, a process that occurs in roughly 1 in 10,000 Higgs decays, reaching 3.4σ significance by combining Run-2 and Run-3 data.
The ATLAS Collaboration presented a new result at ICHEP 2026 this week: first evidence that the Higgs boson decays into a virtual photon and a real photon, which then produce a pair of leptons and a photon in the final state. The decay mode, written H→γ*γ→llγ, is one of the rarest processes predicted for the Higgs, occurring in just one in about 10,000 Higgs decays — rare enough that seeing it at all requires pushing the experiment to its statistical limits.
ATLAS got there by combining its full Run-2 and Run-3 datasets and applying machine-learning techniques to identify the distinctive signature of electrons that have merged together in the detector, a topology that would have been largely invisible to the analysis methods available in earlier runs. The combination achieved 3.4σ statistical significance, clear evidence by the particle physics community’s conventional threshold, though short of the 5σ required to claim a discovery.
The measured signal rate came in at 1.03 ± 0.34 times the Standard Model expectation — in other words, the Higgs is behaving exactly as the theory predicts it should, even in this faintly visible corner of its decay pattern. That agreement is not a surprise, but it is useful: confirming the rate matches predictions means the result can be turned around and used as a tool rather than just a measurement, tightening constraints on how the Higgs interacts with photons at a level of detail previous data couldn’t reach.
What makes this particular decay channel worth the effort is its sensitivity to CP violation in the Higgs sector. The Standard Model predicts the Higgs is a pure CP-even scalar — it interacts the same way under a combined charge and parity reversal as it does normally — but several beyond-the-Standard-Model scenarios allow small CP-odd admixtures. H→γ*γ→llγ is one of the channels where those admixtures would leave an observable imprint, because the angular distributions of the final-state leptons carry information about the CP structure of the Higgs-photon-photon vertex that other decay modes can’t access.
The result, which will be published in the Journal of High Energy Physics, also serves as a proof of concept for a class of analyses that will become significantly more powerful in the coming years. The High-Luminosity LHC, scheduled to begin delivering data in the late 2020s, will increase the total dataset by roughly a factor of twenty compared to what ATLAS and CMS have collected so far. Rare Higgs decays that are barely visible now — or in some cases not yet measurable at all — will start producing statistically meaningful samples, and the machine-learning methods ATLAS deployed here for merged-electron identification will be central to extracting them.
For now, 3.4σ sits in the liminal space that particle physics has made familiar: enough to say the signal is almost certainly there, not yet enough to write it into the textbooks as established fact. But the Standard Model has survived another precision test in a regime where it was under no obligation to be right, and the door to probing the Higgs’s fundamental symmetries has opened a little wider.