Project Hephaistos II: Dyson sphere candidates from Gaia DR3, 2MASS, and WISE

Brief

Suazo et al. (2024) cross-matched Gaia DR3, 2MASS, and AllWISE to assemble a ~5-million-source catalog within 300 pc, then filtered it through grid-search fitting against 220,745 partial Dyson sphere models (temperatures 100–700 K, covering factors ≥ 0.1), a CNN image classifier achieving 95% test accuracy on 960 hand-labeled WISE images, and additional cuts on H-alpha emission, Gaia optical variability (Gvar < 2), RUWE < 1.4, and per-band SNR > 3.5. The pipeline collapsed ~320,000 W3/W4-detected sources to ~11,000 (RMSE < 0.2 mag), then to 5,732 non-nebular sources, and finally to 7 candidates after visual inspection. All 7 are M-dwarfs; the authors state no known astrophysical phenomenon readily explains their observed infrared excess. Photometric data alone cannot confirm a Dyson sphere, and spectroscopic follow-up is required.

Metadata

Category

Search

Venue

Monthly Notices of the Royal Astronomical Society

Type

Peer-reviewed

Year

2024

Authors

Matias Suazo, Erik Zackrisson, Priyatam K. Mahto, Fabian Lundell, Carl Nettelblad, Andreas J. Korn, Jason T. Wright, Suman Majumdar

DOI

10.1093/mnras/stae1186

arXiv

2405.02927

Access

Open access

Length

3.3 M

Programs

Project Hephaistos

Instruments

Gaia DR3, 2MASS, WISE, AllWISE

Data sources

Gaia DR3, Gaia EDR3, 2MASS Point Source Catalog, AllWISE

Tags

SETI, technosignature, Dyson sphere, infrared excess, M-dwarfs, machine learning, photometric survey

Key points

• Initial sample: ~5 million Gaia DR3-2MASS-AllWISE cross-matched sources within 300 pc; requiring W3 and W4 detections and passing WISE contamination flags reduces this to ~320,000 stars.p.2
• Grid search fits each of the ~320,000 sources against 220,745 Dyson sphere models (simulated from 265 main-sequence stars, Teff ~2,800–12,500 K); RMSE < 0.2 mag threshold isolates ~11,000 candidates.p.4
• CNN classifier trained on 960 hand-labeled WISE W3 images (420×420 px, 9.625 arcmin field) achieves 95% test-set accuracy, 0.975 recall and 0.93 precision on the non-nebular class; classifies 5,732 of ~11,000 sources as non-nebular.p.5
• H-alpha pseudo-equivalent widths from Gaia DR3 are used to reject young accreting stars; sources with Hα in emission detected at ≥ 3σ confidence are excluded.p.6
• Optical variability observable Gvar > 2 triggers rejection; RUWE > 1.4 flags unreliable astrometry (correlates with unresolved binaries capable of generating warm dust via planetary collisions).p.6
• After SNR > 3.5 filtering in W3 and W4 and visual inspection, 7 candidates survive, all M-dwarfs, for which astrophysical explanations of the infrared excess are described as not easily available.p.1
• The Dyson sphere photometric model is mathematically identical to optically thin blackbody debris disk models; covering factor γ maps directly onto fractional disk luminosity (L_Disk/L_star), making the two signatures photometrically degenerate.p.3
• Prior Project Hephaistos I work (Suazo et al. 2022) analyzed >10^8 stars and set upper limits as low as ~1 in 100,000 for Dyson sphere prevalence; the 7 candidates from 5 million sources here (~1 in 700,000) sit within that bound.p.1

Verbatim

• “the pipeline identifies 7 candidates deserving of further analysis. All of these objects are M-dwarfs, for which astrophysical phenomena cannot easily account for the observed infrared excess emission.”

  p.1

• “The exact upper limits on the fraction of stars that may host Dyson spheres reported by Suazo et al. (2022) are a function of distance, covering fraction and Dyson sphere temperature, but reach as low as ∼ 1 in 100,000 objects in the most constraining situation.”

  p.1

• “An interesting feature of this model is that it is identical to optically thin blackbody debris disk models, where the covering factor 𝛾 resembles the fractional luminosity (L Disk /L ★ ).”

  p.3

• “One of the advantages of searches based on "Dysonian" signatures is that it does not rely on the willingness of other civilizations to contact us.”

  p.1

Most interesting

• All 7 final candidates are M-dwarfs, the lowest-luminosity main-sequence stars, despite M-dwarfs' habitability challenges; no solar-type or higher-mass stars survived the full pipeline.
• The Dyson sphere spectral signature is mathematically indistinguishable from an optically thin debris disk; only spectroscopic or high-resolution imaging follow-up can separate the two interpretations.
• The CNN was trained on just 960 hand-labeled images yet achieved 95% accuracy, sufficient to classify thousands of WISE fields and eliminate the dominant false-positive class (nebula-embedded young stars).
• The search explicitly excludes Dyson swarm geometries with absorbing elements large enough to cause photometric variability, since such variability would trigger the Gvar > 2 rejection cut designed to remove pre-main-sequence stars.
• Proposed Dyson sphere hosts include not just main-sequence stars but also white dwarfs, pulsars, and black holes, this survey restricts to main-sequence only, leaving those classes unsearched.
• The project is fully archival: no dedicated observations were required; all filtering draws on publicly released Gaia DR3, 2MASS, and AllWISE data products, meaning the same data has been sitting in open repositories awaiting this analysis.