Neutrino Mass Below 0.45 eV
KATRIN limits the neutrino mass to below 0.45 eV in April 2025, under a millionth of the electron mass; cosmology pushes the sum below 0.064 eV.
Open in the interactive tree →Neutrinos have mass (Nobel Prize 2015 for oscillations), but how much is unknown; the Standard Model predicted zero. KATRIN in Karlsruhe measures the energy spectrum of tritium beta decay with a 70-m beamline and a 23-m spectrometer, the most precise way to weigh neutrinos directly.
As of October 2026
Published in Science on 10 April 2025: upper limit 0.45 eV at 90% confidence, half the previous best (0.8 eV), from about 250 measuring days in 2019-2021, roughly a quarter of the planned data set. KATRIN completed its 1,000 measuring days on 26 October 2025 (19 campaigns, about six times the statistics of the published result); the analysis of the full data set, aiming near 0.3 eV, is still under way, and the TRISTAN detector for sterile-neutrino searches is being installed in 2026. Cosmological data (DESI DR2 plus CMB) indirectly limit the sum of masses to below 0.064 eV at 95% in the standard cosmological model, close to the roughly 0.06 eV minimum that oscillation experiments require, but only below 0.16 eV if dark energy evolves. The 2026 Physics Nobel Prize went to Francis Halzen for IceCube's discovery of high-energy astrophysical neutrinos.
Open steps
- KATRIN 1000-day final analysis Medium AI leverageAnalyse the full 1000-day data set with tight control of source, energy-loss and background systematics, aiming for a limit near 0.3 eV.
- Resolve the negative-mass preference Medium AI leverageFind whether the cosmological preference for neutrino mass sums below the oscillation minimum comes from excess CMB lensing, systematics or new physics.
- A technique beyond KATRIN Low AI leverageDevelop frequency-based beta-spectroscopy (cyclotron radiation) and atomic tritium sources to push direct mass sensitivity toward 0.04 eV, below KATRIN's reach.
- Is the neutrino its own antiparticle? Medium AI leverageReach half-lives beyond 10^28 years in tonne-scale neutrinoless double-beta decay detectors, which also constrains the mass and bears on the matter surplus.
Where AI could help
Medium AI leverage. AI speeds spectrum fits and signal reconstruction, but the mass limit is set by counting statistics, instrument systematics and the next experiments.
- Neural-network spectrum models that make fits with many systematic uncertainties fast
- Deep-learning reconstruction of single-electron signals in new techniques such as cyclotron radiation spectroscopy
- Joint fits of oscillation, cosmology and beta-decay data to track the allowed mass range
Shown so far
- In January 2022 KATRIN researchers showed a neural network that reproduces the model of the tritium beta spectrum about three orders of magnitude faster, with relative errors below 1e-4. source
Prerequisites
- Nuclear Physics1938
- Standard Model of Particles1973
- Neutrino Oscillations1998Oscillations proved neutrinos have mass; the absolute scale is measured now