Cherenkov light as a mechanism for light flashes seen by astronauts in space

The paper investigates the mechanism behind light flashes (LFs) seen by astronauts in space, first reported during the Apollo 11 mission. The authors use Geant4 simulations of cosmic ray interactions with a simplified model of the human eye, combined with published models of human visual perception, to compare simulated light flash rates with experimental observations.

The study develops several Geant4 eye models, ranging from simple cubic water blocks to more realistic spherical models with diameters of 2.0 cm and 2.5 cm, and more complex models including vacuum, shielding materials (aluminium, kevlar), and human head tissues (bone, brain). The simulations use protons, heavier nuclei (from helium to oxygen and iron), and muons as primary particles with energies from 1 MeV to 50 GeV.

Key results show that Cherenkov radiation generated in the eye produces retinal responses consistent with the observed frequency of astronaut light flashes in interplanetary space and in low Earth orbit outside the South Atlantic Anomaly. The dominant contributing primaries are high Z nuclei, primarily iron, with additional contributions from oxygen and carbon nuclei. The paper states: "The model predicts that light flashes observed by astronauts in interplanetary space and in low Earth orbit, except the South Atlantic Anomaly, are produced primarily by heavy nuclei, particularly iron, with additional contributions from carbon and oxygen."

For the Cherenkov light production model, the authors use the Frank–Tamm formula with a water refractive index of 1.35 and an average chord length of 1.667 cm in a 2.5 cm water sphere. The model was verified against Geant4 simulations for protons, helium, lithium, and carbon nuclei, showing precise agreement. The number of photons produced ranges from a few hundred for hydrogen nuclei to more than three hundred thousand for zinc nuclei.

The analysis incorporates visual perception thresholds based on Hecht et al. (1942), requiring 5–14 absorbed photons in areas of approximately 500 rods (A500 areas) for visual sensation. The model applies a rod quantum efficiency of 29% for photon absorption. Results show that for the lower threshold (Ap ≥ 5), the predicted light flash rate from all elements from hydrogen to oxygen is 5.51 flashes per minute, while for the higher threshold (Ap ≥ 14), it is 1.86 flashes per minute. The paper notes: most light flashes seen in space are created by Carbon and Oxygen.

For iron nuclei, the model predicts approximately 0.17 light flashes per minute, with most originating from nuclei with energies above 40 GeV. Iron nuclei produce cloud-like light flashes due to the large number of activated retinal areas (hundreds for Ap ≥ 5, and over 50 for Ap ≥ 14). The paper suggests this may explain the approximately 5–10% of astronaut light flashes with diffuse cloud or blob morphologies.

The study also examines the influence of the heliosphere, magnetosphere, and Earth shadow. The light flash rate changes by approximately a factor of 10 between low and high cut-off rigidity regions, consistent with Sileye-2 experiment results. However, the model predicts negligible light flash rates in the South Atlantic Anomaly, indicating that light flashes observed in the SAA region are caused by another mechanism, not by Cherenkov light production.

Regarding shielding effects, simulations with ISS-like shielding (aluminium and kevlar) show that a 50 GeV iron nucleus loses only about 5% of its kinetic energy (2.7 GeV) while traversing the walls, and the resulting Cherenkov photon distribution and retinal activation remain largely unchanged. However, simulations with surrounding head tissues show that iron nuclei lose energy rapidly and may fall below the Cherenkov threshold before reaching the eye, whereas lighter nuclei like oxygen retain sufficient energy.

For muons at Earth's surface, the paper explains why light flashes are not observed. Although GeV muons produce approximately one third more visible Cherenkov photons than equally energetic protons (640 photons for 1 GeV muon vs. 312 for 1 GeV proton), the spatial distribution of those photons on the retina is insufficient to satisfy the modeled perception criteria. The predicted occurrence rate is more than four orders of magnitude lower than the light flash rates reported by astronauts, indicating that muon induced Cherenkov flashes are effectively unobservable at ground level. The upper limit estimate is approximately one muon induced light flash per 24 hours of continuous observation.

The paper concludes that Cherenkov radiation generated in the eye is the mechanism for astronaut light flashes in interplanetary space and low Earth orbit (outside the South Atlantic Anomaly), with iron, carbon, and oxygen nuclei as the dominant contributors. The absence of light flashes at Earth's surface is consistent with insufficient retinal stimulation from muon induced Cherenkov radiation under typical geometrical and flux conditions.

Improvements to AI Systems:

1. Physics-Aware Vision Simulation: Enhance computer vision models by incorporating Cherenkov radiation physics and retinal photon-absorption thresholds (e.g., Hecht’s 5–14 photon criteria) to simulate how biological eyes perceive high-energy particle tracks, enabling more accurate predictions of visual phenomena in extreme environments.

2. Multi-Scale Radiation Transport Modeling: Integrate Geant4-style Monte Carlo simulations into AI frameworks for real-time prediction of particle energy loss and photon generation in heterogeneous media (e.g., eye tissue, shielding), allowing AI to optimize spacecraft shielding designs or astronaut safety protocols without exhaustive simulation runs.

3. Threshold-Based Event Classification: Train AI classifiers to distinguish between light flash morphologies (e.g., point-like vs. cloud-like) based on simulated retinal activation patterns (e.g., number of activated A500 areas), improving automated detection of cosmic-ray-induced visual artifacts in astronaut monitoring systems.

4. Environmental Context-Aware Prediction: Develop AI models that incorporate heliospheric, magnetospheric, and geomagnetic cut-off rigidity data to predict light flash rates as a function of orbit and solar activity, enabling mission planners to forecast astronaut visual disturbances and schedule high-risk activities accordingly.

5. Anomaly Detection for Non-Cherenkov Mechanisms: Use the model’s prediction of negligible light flashes in the South Atlantic Anomaly to train anomaly-detection AI that flags unexpected visual events in that region, prompting investigation into alternative mechanisms (e.g., direct retinal ionization) and improving space medicine diagnostics.

6. Shielding Optimization via Surrogate Modeling: Replace full Geant4 simulations with a neural network surrogate trained on the paper’s energy-loss and photon-yield data, enabling rapid evaluation of shielding materials (aluminium, kevlar) and thicknesses for minimizing light flash incidence while balancing mass constraints in spacecraft design.

7. Muon Background Rejection in Detectors: Apply the finding that muon-induced Cherenkov flashes are unobservable (rate < 1 per 24h) to refine AI-based particle detectors at ground level, reducing false positives by filtering out muon events that would not trigger human-like visual perception criteria.

8. Retinal Response Calibration for VR/AR: Use the rod quantum efficiency (29%) and spatial summation rules (A500 areas) to calibrate AI-driven virtual reality displays, ensuring that simulated light intensities match biological perception thresholds for low-light or high-energy scenarios.

What the Improved AI System Can Do:

• Predict astronaut light flash rates and morphologies in real time for any orbit, shielding configuration, or solar condition, aiding mission control in crew health management.

• Automatically classify cosmic-ray events from onboard sensors as Cherenkov-induced or anomalous, triggering alerts for unusual radiation exposure.

• Optimize spacecraft and spacesuit designs by rapidly evaluating thousands of shielding configurations for their impact on visual perception disturbances.

• Provide physics-consistent synthetic training data for computer vision models operating in radiation-rich environments (e.g., space telescopes, nuclear facilities).

• Enhance human-in-the-loop simulations for astronaut training by accurately reproducing the visual experience of cosmic-ray interactions, including rare cloud-like flashes from iron nuclei.