We see Betelgeuse through its past light cone. The newest high-resolution data show a living, dynamic red-supergiant atmosphere - not an observed supernova.
Current verdictNo core collapse has been observed. The inaccessible remote state is unknowable until a causal signal arrives.
Earth receiver time2026the date information reaches us
Photon emission epochabout 1465 CEfor the default 172 pc model
Causality scenario tool
Distance changes the past epoch we see.
It does not subtract from a remaining-lifetime prediction conditioned on that observed state.
CalculatedPhoton emission epoch1465 CE
ModelCoordinate collapse year1765 CE
CalculatedSignal arrival year2326 CE
D/c = 172 pc × 3.26156 yr pc−1 = 561.0 yr
temit = 2026 - 561.0 = 1465.0
tcollapse = 1465.0 + 300 = 1765.0
tarrival = 1465.0 + 300 + 561.0 = 2326.0
tarrival = 2026 + 300 = 2326
∂tarrival/∂D = −1/c + 1/c = 0
Qualitative reading: distance shifts the two Betelgeuse-coordinate dates together. It does not shift the signal-arrival date when Δt is conditioned on the state received in 2026.
Change the distance: the emission and coordinate-collapse years move, but the Earth waiting time remains 300 years.
01
What we actually see
Latest received evidence
A living atmosphere, resolved across time and wavelength.
Measured
Persistent inner-atmosphere structure
Bands 6, 7, and 8 resolve a roughly 2300 K sub-millimetre photosphere, with hotter north-east and south-west regions and changed molecular gas since 2015.
These are different measurements, not colour filters.
Frequency changes optical depth, resolution, emitting layer, and instrument response. Every comparison must carry that metadata.
Mode
Instrument
Approximate layer
Product status
UV
HST STIS/GHRS
chromosphere and wind
archive spectrum
Optical
AAVSO / resolved imaging
photosphere and dust attenuation
photometry / imaging
Near-IR
interferometry
molecular layers and photosphere
archive-dependent
223.55 GHz
ALMA Band 6
larger sub-mm optical-depth surface
observed Aug 2023
337.99 GHz
ALMA Band 7
inner sub-mm atmosphere
observed Aug 2023
485.22 GHz
ALMA Band 8
deeper, smaller sub-mm surface
observed 1 Aug 2023
22-136 GHz
VLA + ALMA
extended radio atmosphere
observed 2021-2026
03
Quantitative evidence
Published measurements + reproducible equations
Measure the trend. State what it does—and does not—imply.
The plots below use the three published ALMA 2023 integrated-continuum measurements. Error bars are measurement uncertainties; fitted curves are calculated quantities.
Measured + fitted
Integrated continuum spectrum
Weighted in log space
Sν = A (ν/ν0)α
A = 0.518 Jy at ν0 = 337.99 GHz; α = 1.494 ± 0.133.
Qualitative reading: the integrated flux rises with frequency, but more slowly than a fixed-area, optically thick Rayleigh–Jeans ν2 reference. The changing apparent size and atmospheric optical depth prevent a one-parameter physical interpretation.
Measured geometry
Frequency-dependent diameter
Uniform elliptical-disc fit
L (AU) = θ (mas) × D (pc) / 1000
At 172 pc, Band 8 gives L = 9.34 AU and R ≈ 1004 R☉. Distance uncertainty scales these values linearly.
Qualitative reading: higher observing frequency reaches a smaller τν ≈ 1 surface. The reported mean brightness temperatures remain close—2342, 2294, and 2266 K—so diameter, opacity, and temperature must be interpreted together.
Herschel's 1836–1840 comparison estimates are shown as individual points. From 1893 onward, AAVSO visual estimates and Johnson V measurements remain separate and are summarized with robust time bins. No photographs or interpolated observations are inserted.
Herschel reconstruction AAVSO visual median + IQR AAVSO Johnson V median + IQR
F/F0 = 10−0.4(m−m0)
A one-magnitude increase means a factor 2.512 decrease in flux. The relative-flux switch is an exact magnitude conversion with m0 = 0.5; it is not a physical forecast.
m̃k = median(mi∈k), IQRk = Q3−Q1
The overview uses 90-day bins; the recent panel uses 30-day bins. Medians suppress isolated transcription errors, while the shaded quartiles retain within-bin observer scatter and intrinsic change.
Series
Epoch
Underlying records
Plot treatment
Herschel / Lloyd
1836–1840
27 estimates
Individual reconstructed visual magnitudes
AAVSO visual
1893–2026
48,378 detections
Separate robust time bins; no interpolation across gaps
AAVSO Johnson V
1964–2026
3,082 detections
Separate filter series; never merged into visual estimates
Sources: Lloyd (2020) and the AAVSO International Database, AUID 000-BBK-383, retrieved 31 August 2026. The AAVSO series includes only detections within the documented −0.5 to 3.0 magnitude quality screen; 86 upper-limit or implausible rows were excluded. Visual and V passbands are related but not identical measurements.
Published-fit reconstruction
Contour and radial-diagnostic laboratory
Figure conventions follow Dent et al. (2026); values are generated from the published component fits, not extracted from the editorial PNG.
Solid warm contours: +2,+4 mJy beam⁻¹ · dashed cool contours: −2,−4 mJy beam⁻¹. Residual morphology is illustrative, not a calibrated difference image.
Matched-beam method
Azimuthally averaged profiles and spectral index
α(r) = ln[I₈(r)/I₇(r)] / ln(ν₈/ν₇)
Radius (mas)
Band 7 (Jy beam⁻¹)
Band 8 (Jy beam⁻¹)
α
0
4.320e-2
8.046e-2
1.72
21
4.227e-2
6.737e-2
1.29
30
1.583e-2
2.361e-2
1.10
45
8.147e-4
1.087e-3
0.80
60
3.718e-4
4.439e-4
0.49
Iνmodel(x,y) = Idisk + ΣkAk exp[−½rk²]
Published disk size, component locations, widths, axial ratios, position angles, and the Band 8 beam set the reconstruction. Calibrated FITS and visibility-domain modelling remain required to reproduce the paper's observed contours or uncertainty map.
04
Prediction workbench
Spectra + image sequences
Forecast observables, then test them on held-out epochs.
The workbench separates a quantitative interpolation that can be run now from image and line diagnostics that require calibrated, registered archive products. It never converts atmospheric novelty into an unsupported explosion probability.
Conditional prediction
Continuum interpolation laboratory
Predict integrated flux only inside the measured 223.55–485.22 GHz interval, assuming the fitted power law remains valid.
Modelled integrated flux0.666 Jy0.651–0.681 Jy from slope uncertainty alone
Sν(ν*) = 0.518 [ν*/337.99]1.494 Jy
This band varies only α by ±0.133. It excludes normalization covariance, calibration systematics, intrinsic variability, and model discrepancy; it is not a full posterior-predictive interval.
In-sample standardized residuals
ν
Observed
Model
(O−M)/σ
223.55 GHz
0.28 Jy
0.279 Jy
0.06
337.99 GHz
0.51 Jy
0.518 Jy
-0.19
485.22 GHz
0.90 Jy
0.888 Jy
0.13
Measurements: Dent et al. 2026. This interpolation describes the received sub-mm atmosphere; it does not determine the core-burning stage.
Local visual comparator
Large-sequence blink comparator
Load up to 5,000 aligned PNG, JPEG, or WebP frames. Files remain in this browser session and are not transmitted. The century-scale record is photometry, not a sequence of resolved photographs; use the historical light curve above for that evidence.
Unzip it, then select all three numbered PNG files above. These August 2023 Bands 6/7/8 paper panels test visual blinking only; they are not a calibrated time sequence.
Quantitative residuals require common astrometry, pixel scale, units, flux scale, and PSF/synthesised beam. This viewer intentionally performs visual blinking only; numerical differencing belongs in the FITS pipeline.
The placeholder is an attributed paper figure, not a time-series datum. Reduced-motion mode disables automatic blinking; arrow controls remain available.
Spectral lines
Velocity and shape diagnostics
v ≈ c(ν₀ − ν)/ν₀
v̄ = Σ Iᵢvᵢ / Σ Iᵢ
Compare centroid, FWHM, skewness, line ratios, and channel-map morphology. A line shift is not automatically bulk stellar motion; opacity and asymmetric flows must be modelled.
Likelihood
Prediction with discrepancy
ln L = −½ Σ[(yᵢ−mᵢ)²/(σᵢ²+s²) + ln 2π(σᵢ²+s²)]
The nuisance term s represents excess scatter or model discrepancy. Without it, a precise instrument can make an incomplete physical model look falsely certain.
Forecast
Posterior predictive check
p(y*|y,M) = ∫ p(y*|θ,M)p(θ|y,M)dθ
Hold out whole epochs, not random pixels or adjacent channels. A useful forecast must reproduce unseen epochs and remain calibrated under instrument and wavelength changes.
05
Evolutionary clock
Assumption switch
One star. Two incompatible clocks.
The disagreement is scientifically useful only when the pulsation assignments and companion interpretation stay visible.
Model-dependent
Scenario A
Core helium burning
~400/416 d fundamental · ~2100/2170 d companion/LSP
Changing the pulsation-mode assignment changes the inferred evolutionary stage. Neither panel is a direct measurement of the core.
Asteroseismic scaling
Pn = Qn √[R³/(GM)]
The mode constant Qn depends on radial order and stellar structure. Assigning 416 d or 2200 d to the fundamental therefore changes the inferred radius and evolutionary state; the periods cannot be swapped without recomputing the stellar model.
Observed ratio
PLSP/PFM ≈ 2170/416 = 5.22
A companion orbit naturally supplies a slow clock independent of radial acoustic pulsation. This is why identifying every long period as a pulsation is not a neutral assumption.
06
Statistical inference
Inference begins with a generative assumption
Compatibility is conditional, not clairvoyant.
01Received observables
02Provenance + units
03Null hypothesis
04Likelihood + residuals
05Posterior checks
06Sensitivity analysis
Defensible statement
“These measurements are more compatible with scenario A under the stated likelihood, priors, and pulsation assignment.”
Not identifiable
“A remote core-collapse event occurred at a coordinate time from which no causal signal has reached Earth.”
p(Mk|y) ∝ p(y|Mk) p(Mk)
Model-family weights are conditional on the prior, likelihood, data selection, and discrepancy model. Report sensitivity to each; never print a scenario percentage without those ingredients.
07
Core-collapse watch
Neutrino watch
The final-hours messenger comes from the core.
Surface light constrains atmosphere and pulsation. Evolutionary models provide conditional long-range inference. A genuine late-stage warning is more directly tied to rising pre-supernova neutrino emission.
Detector counts fold this flux through cross-section, target mass, efficiency, threshold, and background. A pre-supernova alert is a near-term messenger calculation, not a centuries-long optical forecast.
08
Supernova laboratory
GPU-accelerated conditional forward model
Evolution through 3026 CE, with every assumption exposed.
The default branch renders atmospheric evolution without inserting a supernova through 3026 CE. A separate disputed branch lets the observer impose an arrival epoch from 2056–2326 CE and follow shock breakout, free expansion, wind-driven deceleration, and the transition from young to evolved supernova remnant.
Physics-coupled GPU rendering
Betelgeuse: conditional evolution through 3026 CE
Three-dimensional gas-flow-driven rendering for year 2026. Received-state extrapolation · red-supergiant atmosphere. Synthetic near-infrared continuum photosphere with a representative temperature of 3690 kelvin and root-mean-square gas velocity 0.00 kilometres per second.
2026 CEReceived-state extrapolation · red-supergiant atmosphereAccelerated surface time 0 d · step 0 · drag to orbit
Selected emitting layer3690 ± 0 KSimulated mean ± surface RMS around 3690 K
Surface-flow state0.00 km s⁻¹ RMScs = 6.25 km s⁻¹ · Mach 0.00 · cell crossing ≈ — yr
Measured geometric scaleθ = 42.49 mas · R ≈ 3.65 AUNear-IR R ≈ 786 R☉ at 172 pc
Representative gravity scalelog g = -0.11 · vesc ≈ 92 km s⁻¹M = 17.5 M☉ midpoint · L ≈ 103 × 10³ L☉
Observational anchors used by the simulation
Continuum surface42.49 ± 0.06 mas · 3690 ± 54 K
A limb-darkened global disk with few-percent H-band asymmetry and resolved bright structure—not a perfectly smooth surface. Ohnaka et al. 2011 · Haubois et al. 2009
Sub-mm atmosphere≈2300 K · 1.1–1.3 R★
Band 7/8 structure includes an ≈800 K hot enhancement, radius deviations up to ±6%, and weak emission to ≈2.5 R★. Dent et al. 2026
Flow constraintCells ≈0.6 R★ · local flows ≈20 km s⁻¹
The renderer targets a lower disk-integrated RMS; observed local upflows and downflows can be much faster. López Ariste et al. 2018
Rotation disciplineNo rigid rotation imposed
Large convective cells can mimic a velocity dipole, and the 2023 ALMA data show no clear rotation signature. Ma et al. 2024
Surface-layer solver
Gas dynamics and thermal transport
Coefficients are nondimensional surface-layer closures. They alter the solved flow field rather than merely changing colour. The worker maps one numerical step to 0.25 accelerated surface day; it is not synchronized to the compressed 1000-year evolutionary slider.
2026205621762326252627763026
Mass conservation
∂ρ/∂t + ∇·(ρv) = 0
Semi-Lagrangian transport advances density while a divergence term supplies compressional response. Weak relaxation prevents secular mass drift in the reduced surface layer.
Momentum equation
Dv/Dt = −∇p/ρ + αgT′r̂ + ν∇²v
Spherical pressure gradients, buoyancy, drag and viscous diffusion evolve the horizontal and radial velocity fields. Rotation is not imposed because the current observational interpretation is disputed.
Thermal energy
DT/Dt = −(γ−1)T∇·v + κ∇²T + Qconv − T′/τrad
Compression heats, expansion cools, thermal diffusion smooths grid scales, buoyant transport sustains cells, and radiative relaxation returns perturbations toward the 3690 K near-infrared continuum anchor.
Equation of state + scale
R = θD/2 · g = GM/R² · L = 4πR²σTeff⁴
θ = 42.49 mas and D = 172 pc give R ≈ 3.65 AU. A representative M = 17.5 M☉ gives log g ≈ -0.11 and vesc ≈ 92 km s⁻¹.
The surface shape now stays within observed near-IR and sub-mm asymmetry scales, while colour and radius change with the selected emitting layer. It remains a reduced spherical surface solver—not a three-dimensional radiation-hydrodynamics calculation. The radio view is explicitly false-colour; opacity, line formation, ionisation, deep stratification, shock capturing, magnetic fields, dust radiative transfer and calibrated boundary conditions remain unresolved. The preferred model does not predict core collapse anywhere on this timeline through 3026 CE.
09
Author synthesis
Signed conditional forecast · Biswajit Jana
My working prediction: Earth is more likely to see Betelgeuse’s supernova on a 105-year horizon than a 102-year horizon.
Representative point forecast≈ 202,026 CE
I adopt Δt = 200,000 years as one representative value inside the preferred “hundreds of thousands of years” model family. It is not a fitted mean, median, confidence interval, or countdown.
tsee = treceive + Δt2026 + 200,000 = 202,026 CEThe distance cancels because Δt is conditioned on the state received in 2026.
Preferred interpretation
~400/416 d fundamental mode; ~2170 d companion/LSP
This branch is compatible with early core-helium burning and the independent companion interpretation. The literature horizon is hundreds of thousands of years. My representative 200,000-year point maps to approximately 202,026 CE at Earth.
~2200 d fundamental mode; late core-carbon burning
This branch gives a literature horizon of several dozen to several hundred years: an illustrative Δt = 30–300 years maps to roughly 2056–2326 CE. The required large radius is in tension with angular- diameter constraints, and the companion weakens the need for this mode assignment.
It predicts when a causal supernova messenger would reach Earth under my preferred model interpretation. It does not establish Betelgeuse’s inaccessible present state, and atmospheric images or spectra alone cannot turn this order-of-magnitude synthesis into an exact date.
— Biswajit Jana, 30 August 2026
Scientific boundary
Inference can compare received evidence. It cannot cross our past light cone.
Every later model in this repository is required to separate measured observables, calculated quantities, simulations, model-dependent claims, and speculation.
