Cosmic-ray boosts sharpen model-dependent limits on lightweight dark matter

A possible boost for otherwise invisible particles

Underground dark matter detectors look for the recoil of an atomic nucleus struck by an incoming particle. For dark matter below about one gigaelectronvolt, or GeV, ordinary Galactic motion often supplies too little energy to clear the detectors’ nuclear-recoil thresholds. That limits these searches without establishing that lightweight dark matter is absent.

The proposed cosmic-ray mechanism changes the incoming particle’s energy. An energetic cosmic-ray nucleus could collide with a dark matter particle in the Galaxy, accelerating it enough to produce a detectable recoil later on Earth. The study calculates the consequences of that possibility; it does not report observing such particles.

Yang Yu and colleagues examine this scenario in the second arXiv version of their paper, posted on November 6, 2026. This account follows that preprint version; the supplied metadata also lists a journal DOI.

Four ways to transmit an interaction

The researchers assume dark matter is a Dirac fermion, a particular type of particle, and introduce a hypothetical mediator that carries its interaction with protons and neutrons. They compare four mediator types: scalar, vector, pseudoscalar and axial-vector.

In their framework, scalar and vector interactions are spin-independent. These can benefit from coherent scattering, in which contributions from nucleons within a nucleus reinforce one another. Pseudoscalar and axial-vector interactions are spin-dependent and have different target sensitivities.

The calculation assumes equal couplings to protons and neutrons. It explores mediator masses from one millionth of a GeV to one GeV, and presents coupling constraints for dark matter masses of 1, 10, 100 and 1,000 megaelectronvolts, or MeV. These are benchmark choices, not measurements of either particle’s mass.

Following particles from the Galaxy underground

The analysis connects three stages: cosmic rays accelerate dark matter, the boosted particles lose energy while approaching a detector, and surviving particles scatter inside its target.

For production, the researchers combine cosmic-ray spectra with their model-specific scattering probabilities, including cosmic-ray nuclei with atomic numbers up to 28. They assume a local dark matter density of 0.3 GeV per cubic centimetre and a Navarro–Frenk–White Galactic density profile.

An effective distance summarizes how much Galactic material contributes to the boosted flux. The paper compares one and ten kiloparsecs in its constant-cross-section analysis, then adopts ten kiloparsecs for the light-mediator calculations.

Propagation matters because a stronger interaction does not always yield a stronger underground signal. Particles that scatter too readily can lose their energy in the overlying material before reaching the detector. The researchers numerically integrate an elastic-scattering energy-loss equation, using representative atmospheric or crustal compositions and nuclear form factors that describe finite-size effects.

They compare predicted recoil rates with published experimental limits rather than reanalysing raw detector events. Their comparisons use LZ, XENON1T, XENONnT and Borexino, with MiniBooNE also included in the constant-cross-section study.

Why a light mediator changes the answer

The central result concerns the momentum transferred during a collision. The mediator’s propagator contributes a squared factor proportional to the inverse square of the sum of its mass squared and the momentum transfer squared. Neglecting the momentum term is appropriate only when the mediator is sufficiently heavy.

Keeping both terms produces a turnover in the coupling constraints around mediator masses of roughly 1–10 MeV. Below the relevant momentum scale, simply reducing the mediator mass no longer changes the interaction as predicted by the heavy-mediator approximation. The detailed outcome also depends on the mediator’s other momentum-dependent interaction factors.

For the light scalar and vector cases examined, the authors find LZ provides the strongest constraints among the experiments compared. They attribute this to low-threshold sensitivity and coherent nuclear scattering. Borexino becomes competitive for heavier mediators and supplies the spin-dependent coupling constraints shown for the pseudoscalar and axial-vector cases.

MiniBooNE illustrates another complementarity: its shallow location allows sensitivity to stronger interactions that would stop boosted particles before they reached deeper detectors.

Conditional exclusions, not a discovery

These limits depend on the assumed Galactic distribution, effective production distance, particle model and treatment of energy loss. The propagation calculation uses an elastic energy-loss description with representative material compositions, so its exclusions inherit those modeling choices.

Comparisons with meson decays, stellar cooling and the supernova SN1987A require additional assumptions: those observations generally constrain the mediator’s coupling to ordinary matter, whereas this analysis constrains a product involving its dark matter coupling too.

The result is therefore a set of conditional exclusions. Its main lesson is methodological: interpreting cosmic-ray-boosted dark matter searches through a light mediator requires retaining momentum dependence throughout the calculation.