Researchers have developed an artificial intelligence technique that significantly improves the ability of liquid xenon detectors to search for a theorized but never-observed radioactive decay process that could reshape our understanding of neutrinos and matter itself.

The study, posted as a preprint on arXiv (arXiv:2603.23549) and not yet peer-reviewed, describes an 'augmented convolutional neural network' (A-CNN) designed to sharpen the search for neutrinoless double-beta decay (0νββ) — a hypothetical nuclear process in which two neutrons simultaneously transform into two protons and two electrons, releasing no neutrinos. If confirmed, the process would prove that neutrinos are their own antiparticles, a property known as being 'Majorana particles,' and would violate a fundamental symmetry called lepton number conservation.

The primary obstacle in such searches is background noise: gamma rays emitted by detector materials that can mimic the signal researchers are looking for. The A-CNN model was trained to extract additional geometric and topological information from detector event data to distinguish true signals from this background.

Testing the model using simulation and calibration data from XENONnT — a leading liquid xenon time projection chamber (TPC) experiment operating at the Gran Sasso National Laboratory in Italy — the researchers reported that the A-CNN achieved over 60% rejection of background events while retaining 90% of genuine signal candidates. The authors project this translates to approximately a 40% improvement in XENONnT's sensitivity for detecting 0νββ decay in xenon-136.

Liquid xenon TPCs were originally designed to detect dark matter particles through rare collisions, but their large, ultra-pure xenon targets make them dual-purpose instruments also capable of probing rare nuclear decays. Natural xenon contains approximately 8.9% of the isotope xenon-136, which is the relevant isotope for this decay search.

The researchers also suggest the A-CNN framework could be applied to next-generation liquid xenon observatories, including the proposed XLZD consortium detector, which would operate at even larger scales and push sensitivity further.

Important limitations apply. The study is a preprint and has not undergone peer review. The sensitivity improvements are based on simulations and calibration data rather than a full analysis of live physics data from XENONnT. Real-world performance may differ, and independent validation will be necessary before the method's full impact can be assessed.

Source: https://arxiv.org/abs/2603.23549