Disco in the Dust: Reflected light at the bow shock around Betelgeuse's companion explains its observed luminosity

Recent detections of α Orionis B, the putative companion to Betelgeuse, have been reported using multiple instruments and techniques. These include, most recently, a > 6σ detection using VLT/SPHERE reported by M. Montargès et al. (2026). The authors infer a bright companion (∼ 10−3 × Betelgeuse’s luminosity) with a mass of 2 − 3 M⊙ and Teff ≈ 10 − 12,000 K, loosely consistent with the uppermost bounds of S. B. Howell et al. (2025)’s recent mass estimate from speckle-imaging. However, it is in tension by a factor of two with the reported mass-exclusion limits via non-detection from a recent HST far-UV campaign (J. A. Goldberg et al. 2025). While mass identification from isochrone-fitting (conducted in all cases) is prone to uncertainty, the restrictive HST upper limit on FUV flux precludes the hot, blue emission of a ≳ 2 M⊙ main-sequence star. This discrepancy is reconciled by the following hypothesis: these detections observe the reflection of Betelgeuse’s own luminosity scattered off the companion’s bow shock and wake as it traverses the dusty circumstellar medium. To evaluate the feasibility of this scenario, we draw from custom, filter-specific stellar models computed with MESA, as well as Athena++ simulations of the companion’s bow shock in idealized conditions. Importantly, the expansive bow shock from a stellar-mass object necessarily subtends a sizable portion of Betelgeuse’s outgoing flux. We find that optical luminosity ratios of ∼ 10−3 − 10−4 are straightforward to resolve within reasonable assumptions about the companion mass and circumstellar dust, and disfavor a hot 3 M⊙ companion. We thereby reconcile competing mass hypotheses across observational campaigns through proper attribution of the observed brightness, motivating future multi-wavelength campaigns to further characterize this enigmatic system.

Improvements to AI systems:

1. Multi-wavelength observational reconciliation module – The AI can now integrate conflicting detection limits (e.g., HST FUV non-detection vs. VLT/SPHERE optical detection) into a unified physical model, automatically testing alternative explanations (e.g., scattered light vs. intrinsic emission) rather than defaulting to the most recent detection.

2. Bow-shock scattering physics emulator – The AI can simulate and quantify the fraction of a primary star’s flux scattered off a companion’s bow shock and wake, using dust properties and geometry, to predict apparent brightness ratios across filters (optical, FUV, NIR) for any stellar mass and separation.

3. Isochrone-fitting uncertainty propagation – The AI can flag mass estimates from isochrone fitting when they conflict with flux-based upper limits, and automatically re-run fits under alternative luminosity sources (e.g., reflected vs. intrinsic) to produce a probability distribution of plausible companion masses.

4. Cross-campaign hypothesis ranking – The AI can rank competing hypotheses (e.g., hot main-sequence companion vs. cool companion with scattered light) by computing Bayesian evidence from all available datasets, including non-detections, and output the most physically consistent scenario with quantified confidence.

5. Filter-specific stellar model interpolation – The AI can generate custom MESA-based synthetic photometry for arbitrary companion masses, temperatures, and dust environments, enabling rapid prediction of detectability in any planned or existing observation (e.g., JWST, HST, VLT).

What the improved AI system can do:

• Given a set of contradictory observational claims about a binary companion, it can autonomously propose and test a scattering-based explanation, then output a reconciled mass range and recommended future observations (e.g., specific filters and epochs) to break the degeneracy.

• It can predict whether a bow-shock reflection would be visible in a given instrument’s bandpass, and if so, what apparent magnitude and astrometric offset to expect, aiding in target selection for follow-up campaigns.

• It can flag when a “detection” is more likely an artifact of scattered primary light rather than a genuine companion, preventing false-positive planet/star discoveries in dusty environments.

• It can produce a decision tree for observers: if FUV flux is below X and optical flux is above Y, then the companion is likely low-mass with a bow shock; otherwise, a hot companion is preferred.

Recent detections of alpha Orionis B, the putative companion to Betelgeuse, have been reported using multiple instruments and techniques. These include, most recently, a >6 sigma detection using VLT/SPHERE reported by Montarges+2026. The authors infer a bright companion (about10-3 times Betelgeuse's luminosity) with a mass of 2-3,M and T eff about10-12, 000,K, loosely consistent with the uppermost bounds of Howell+2025's recent mass estimate from speckle-imaging. However, it is in tension by a factor of two with the reported mass-exclusion limits via non-detection from a recent HST far-UV campaign (Goldberg+2025). While mass identification from isochrone-fitting (conducted in all cases) is prone to uncertainty, the restrictive HST upper limit on FUV flux precludes the hot, blue emission of a 2,M main-sequence star. This discrepancy is reconciled by the following hypothesis: these detections observe the reflection of Betelgeuse's own luminosity scattered off the companion's bow shock and wake as it traverses the dusty circumstellar medium. To evaluate the feasibility of this scenario, we draw from custom, filter-specific stellar models computed with MESA, as well as Athena++ simulations of the companion's bow shock in idealized conditions. Importantly, the expansive bow shock from a stellar-mass object necessarily subtends a sizable portion of Betelgeuse's outgoing flux. We find that optical luminosity ratios of about10-3-10-4 are straightforward to resolve within reasonable assumptions about the companion mass and circumstellar dust, and disfavor a hot 3,M companion. We thereby reconcile competing mass hypotheses across observational campaigns through proper attribution of the observed brightness, motivating future multi-wavelength campaigns to further characterize this enigmatic system.