Researchers from the Belle and Belle II collaborations searched for a very rare particle decay, B+ → K+ τ+ τ−, and found no evidence for it. They used data equivalent to 1.2 billion Υ(4S) mesons produced in electron–positron collisions. From that sample they set an upper bound on the decay probability (branching fraction) of B(B+→K+τ+τ−) < 0.56 × 10−3 at 90% confidence, improving the previous limit by about a factor of four.

The decay is interesting because it changes the type, or “flavor,” of a quark without changing its electric charge. In the Standard Model of particle physics, these flavor-changing neutral-current processes are very rare and happen only through looped quantum effects. The Standard Model prediction (after removing contributions from charmonium states) is about (1.0–2.0) × 10−7, far below current experimental reach. Some proposed non‑Standard-Model explanations for other anomalies could boost the rate by up to a thousand times, making this decay a promising place to look for new physics.

Finding this decay is hard because each τ (tau) lepton usually decays to at least one neutrino. Neutrinos escape the detector unseen, so the decay does not leave a simple, clean signature. To deal with that, the teams fully reconstructed the other B meson produced in the same collision. This “tagging” lets them infer the missing energy and better separate signal from background. They then searched the remaining particles for a charged kaon (K+) and two oppositely charged leptons that come from τ→ℓνℓντ decays, where ℓ is an electron or muon. They focused on events with very little leftover energy in the calorimeter after accounting for the two reconstructed B mesons.

The analysis used detailed detector simulations and a machine‑assisted full-event reconstruction algorithm (FEI). Selection rules include cuts on two kinematic quantities, Mbc and ΔE, that test how well the reconstructed partner B matches expectations. Topological variables were used to reduce background from non‑B collisions. In simulated signal events the partner-B reconstruction step kept about 0.8% of Belle events and 0.5% of Belle II events; when the partner B was retained it was correctly reconstructed in about 29% (Belle) and 44% (Belle II) of cases. The teams generated 5×107 simulated signal events per experiment and used large simulated background samples to design and test their selection.