A neural network searched roughly two million DESI spectra and flagged 33,983 candidate Lyman-alpha emitters—galaxies with a strong hydrogen emission line. Subsequent spectral filters removed misidentified lines and separated likely active galactic nuclei, leaving 19,685 galaxies selected for emission associated with star formation.[1]
The accepted Astrophysical Journal manuscript, revised on 18 October, examines objects at redshifts around 2–3.5. DESI’s Redrock pipeline only included galaxy templates below 1.7, leaving many of these more distant objects mislabelled.[1]
Trained using visually inspected spectra and additional comparison objects, the model locates the emission line, then fits its peak to estimate redshift. Its held-out test recovered 472 of 492 emitters, with 24 false positives.[1]
On an independent set of 982 emitters matched to the HETDEX survey, recovery fell to 81.3%. Training favoured narrow, asymmetric lines; broad lines and unusually strong blue peaks sometimes escaped the model.[1]
Sources & further reading
- Unveiling Hidden Lyman Alpha Emitters in the DESI DR1 Data — accepted author manuscript, version 2 Chan et al. / arXiv · 18 Oct 2026 · Accessed 29 Oct 2026