What contrast level did the VLT achieve to detect Betelgeuse B?

The combined ADI+SDI pipeline reached a contrast of 8 × 10⁻⁵ (≈10⁻⁴) at a separation of 0.5″, sufficient to reveal the faint companion.

Why are synthetic injection tests crucial in high‑contrast imaging pipelines?

They quantify self‑subtraction bias and throughput loss, providing realistic photometric uncertainties; without them, measured companion fluxes can be off by up to 30%.

How does modeling Betelgeuse’s limb darkening affect companion astrometry?

Accurate limb‑darkening models correct for the primary’s resolved surface, reducing astrometric errors from 10 mas to <5 mas, which is essential for reliable orbital fits.

High-Contrast Imaging: VLT SPHERE Reveals Stellar Companions

TL;DR: The VLT’s high‑contrast imaging pipeline finally resolved Betelgeuse’s faint companion, showing that mature data‑reduction workflows can turn a 0.1 % flux signal into a reliable detection.

Introduction: A Century‑Old Puzzle Solved by Modern Optics

The red supergiant Betelgeuse has been a laboratory for stellar physics for over a century, yet its most basic binary status remained ambiguous. In 2026, a team led by Miguel Montargès used the ESO Very Large Telescope (VLT) to capture the clearest image yet of a second object, now dubbed Betelgeuse B (Phys.org). The detection wasn’t a lucky snapshot; it was the product of a rigorously engineered high‑contrast imaging workflow that pushed contrast limits to better than 10⁻⁴ at separations of 0.5″.

The breakthrough matters because binary interactions can dramatically alter mass‑loss rates, supernova progenitor pathways, and nucleosynthetic yields. For developers building pipelines for adaptive‑optics (AO) instruments, the Betelgeuse case is a concrete benchmark: it demonstrates that a mature reduction stack can extract a sub‑percent signal from a glare‑dominated point spread function (PSF). The rest of this deep‑dive dissects the hardware, software, and interpretive steps that made the discovery possible, then translates those lessons into actionable guidance for teams handling high‑contrast astronomical data.

High‑Contrast Imaging Techniques Used on the VLT

The VLT’s SPHERE instrument (Spectro‑Polarimetric High‑contrast Exoplanet REsearch) combines extreme AO, a coronagraph, and differential imaging to suppress starlight. In the Betelgeuse campaign, the team deployed a 1.2‑m diameter apodized pupil Lyot coronagraph, achieving an on‑axis attenuation of ~10⁻³ (Science News). Simultaneously, the AO loop ran at 1.2 kHz, delivering a Strehl ratio of 0.85 in the H‑band, which is critical for stabilizing the residual speckle pattern.

Two differential strategies were layered on top of the coronagraph. First, Angular Differential Imaging (ADI) exploited field rotation during a 3‑hour sequence, allowing the static PSF to be modelled and subtracted. Second, Spectral Differential Imaging (SDI) leveraged the narrowband H2/H3 filters (central wavelengths 1.593 µm and 1.667 µm) to discriminate between achromatic speckles and the chromatic signature of a cool companion. The combined ADI+SDI pipeline reduced the residual noise floor to 3 × 10⁻⁵ at 0.4″, a contrast level previously only seen in exoplanet surveys.

The hardware choices mattered as much as the software. The VLT’s 8.2‑m primary mirrors provide a diffraction limit of ~40 mas at 1.6 µm, comfortably resolving the 0.5″ separation inferred for Betelgeuse B. Moreover, the instrument’s internal calibration unit injected a synthetic PSF every 10 minutes, giving the pipeline a real‑time reference for drift correction. Without this, the cumulative wave‑front error would have introduced a systematic bias that could masquerade as a companion.

Data Reduction Pipeline for VLT/SPHERE

The reduction workflow can be broken into four deterministic stages: preprocessing, PSF modelling, speckle subtraction, and astrometric/photometric extraction. Each stage is codified in the ESO pipeline (v2.4.1) and wrapped by the open‑source pySPHERE library (v0.3.2). Developers should treat these as reusable modules rather than monolithic scripts.

Preprocessing begins with dark subtraction, flat‑fielding, and bad‑pixel interpolation. The pipeline also corrects for detector non‑linearity using the calibration curves supplied with each night’s data set. In the Betelgeuse run, the median dark current was 0.12 e⁻/s, and the flat‑field variance stayed below 0.8 % across the detector—numbers that keep systematic errors below the 10⁻⁴ contrast threshold.

PSF modelling employs a Karhunen‑Loève Image Projection (KLIP) algorithm with 20 eigen‑modes, calibrated on the synthetic PSF injections. The KLIP basis captures the evolving speckle morphology while preserving astrophysical signals. The team experimented with 10‑30 modes and found that 20 offered the optimal trade‑off: fewer modes left residual speckles; more modes over‑fit and attenuated the companion flux by up to 30 %.

Speckle subtraction integrates ADI and SDI. The ADI component aligns each frame to a common centre, then rotates it back to a common sky orientation before constructing a median reference PSF. The SDI component rescales the H2 and H3 images to a common wavelength, then subtracts them to remove chromatic speckles. The final residual map reveals a point source at a contrast of 8 × 10⁻⁵ relative to Betelgeuse’s photosphere.

Extraction uses a forward‑modelling approach: synthetic companions of varying fluxes are injected at the detected location, propagated through the same KLIP pipeline, and compared to the observed residuals via chi‑square minimisation. This yields a photometric uncertainty of ±0.12 mag in the H‑band and an astrometric error of ±7 mas, both well within the requirements for orbital fitting.

Developers building similar pipelines should automate the injection‑recovery loop; it is the only robust way to quantify self‑subtraction bias in high‑contrast regimes.

Interpreting the Companion Signal: From Detection to Physical Parameters

Once the point source is isolated, the next step is to translate contrast into physical properties. Betelgeuse’s H‑band magnitude is 0.0 mag (Vega system). A contrast of 8 × 10⁻⁵ corresponds to an absolute H magnitude of ~7.5 mag for the companion, after correcting for the 197 pc distance (Gaia DR3). Using the BT‑Settl stellar models (2025 release), an H‑band magnitude of 7.5 maps to a late‑K or early‑M dwarf with a mass of 0.7 M☉ and an effective temperature near 3800 K.

Spectroscopic follow‑up with VLT/CRIRES+ confirmed a CO bandhead depth consistent with a 3800 K photosphere, strengthening the stellar‑companion hypothesis over a compact object scenario. Orbital constraints derived from the 0.5″ separation and a presumed circular orbit yield a period of ~1,300 years, implying a semi‑major axis of ~600 AU. This distance is large enough that tidal interaction is negligible today, but historic mass‑transfer episodes during Betelgeuse’s red‑supergiant phase could have shaped its current wind geometry.

The detection also resolves a long‑standing discrepancy between interferometric asymmetries seen in VLTI/GRAVITY data (2019) and the lack of a visual companion. Those asymmetries now make sense as scattering off the companion’s wind‑interaction zone, a hypothesis that can be tested with ALMA CO mapping at 0.1″ resolution.

Implications for Stellar Evolution Models and Future Observations

Binary evolution codes (e.g., MESA‑binary v15140) have traditionally treated Betelgeuse as a single star, leading to supernova progenitor predictions that ignore angular‑momentum transfer. Introducing a 0.7 M☉ companion forces a revision: mass‑loss rates may be enhanced by up to 20 % during periastron passages, and the final core mass could shift by ~0.05 M☉, enough to affect the type of core‑collapse event (II‑P vs II‑L).

From an observational standpoint, the Betelgeuse result sets a new benchmark for high‑contrast imaging of bright, extended sources. Most pipelines are tuned for point‑like exoplanets; Betelgeuse forces developers to handle a primary star whose own photosphere is partially resolved (angular diameter ~45 mas). This requires a PSF model that incorporates limb darkening and surface granulation, which the team achieved by fitting a 3‑parameter power‑law limb‑darkening model to contemporaneous VLTI/PIONIER visibilities.

Future campaigns should therefore integrate multi‑instrument data streams: interferometric visibilities for primary star modelling, high‑contrast AO images for companion detection, and millimeter interferometry for wind structure. A unified data‑fusion framework, perhaps built on the Apache Arrow columnar format, would allow simultaneous fitting of all observables, reducing systematic uncertainties by ~30 % (preliminary simulations by the Montargès group).

What This Actually Means

The real story isn’t that Betelgeuse finally has a companion; it’s that the community now has a proven end‑to‑end workflow for extracting sub‑percent signals from the glare of a luminous, resolved star. Teams that try to copy the detection without adopting the full KLIP‑ADI‑SDI stack will inevitably chase ghosts—false positives that arise when speckle noise is under‑modelled. In practice, that means any project that aims to resolve faint companions around bright giants must allocate at least 30 % of its schedule to pipeline validation via synthetic injection tests. Skipping that step will generate maintenance debt: re‑processing will be required whenever a new instrument upgrade changes the speckle statistics, and the technical debt will manifest as missed detections or, worse, published false companions.

I predict that within the next 24 months, at least three major observatories (Keck, Subaru, and the upcoming ELT) will publish companion detections around red supergiants using a similar workflow, and the software stack will converge on a shared open‑source library (likely a fork of pySPHERE). Teams that adopt that library early will gain a 2‑month lead on data‑analysis cycles, translating into faster science output and more competitive grant proposals.

Key Takeaways

✔️Adopt a KLIP‑based ADI+SDI pipeline with 20 eigen‑modes for bright, resolved primaries; fewer modes leave speckles, more modes over‑subtract.
✔️Inject‑recover tests must be automated; they are the only reliable way to quantify self‑subtraction bias and set photometric error bars.
✔️Model the primary star’s limb darkening and granulation using contemporaneous interferometric visibilities; ignoring this can shift companion astrometry by >10 mas.
✔️Use forward‑modelling for photometry rather than simple aperture photometry; it accounts for algorithmic throughput loss.
✔️Plan 30 % of project time for pipeline validation and cross‑instrument data fusion; this upfront investment prevents costly re‑analysis later.