A slitless census of the COSMOS field: 12,275 emission-line galaxies from redshift 0 to 9, with the catalog released

The COSMOS field now has a homogeneous spectroscopic catalog of 12,275 emission-line galaxies whose redshifts run from 0 to 9. Earlier emission-line samples in this field came mostly from ground-based spectroscopy or narrow-band selection of a single line, so their redshift coverage and selection were not uniform. The sources were selected from JWST/NIRCam F444W slitless spectroscopy over a contiguous 0.33 square degree area, paired with deep multi-band imaging.

Each source carries a spectroscopic redshift, emission-line fluxes, and SED-derived physical properties, and is assigned a redshift quality rank from 1 to 3. Slitless spectroscopy has no slit, so one line can fall on several sources aligned along the dispersion direction, and each line must be assigned to the source that emitted it. Identification runs template cross-correlation first, then photometric-redshift constraints, followed by visual inspection, a Siamese convolutional network, and a large language model.

The F444W grism covers multiple emission lines at once, including [O III], Hα, He I, [S III], Paα, Paβ, Brα, PAH, and Brγ. A single detected line often admits several redshift solutions, most notably the degeneracy between Paα and Hα. Composite spectra stacked in several redshift bins reveal faint features that are hard to identify in individual spectra, including higher-order hydrogen recombination lines, CO absorption bands, and Fe II lines.

The catalog is aimed at readers studying galaxy evolution, star formation, and AGN activity, and is intended for statistical sample selection within the COSMOS field. The authors recommend using the redshift quality ranks and contamination flags when defining statistical samples, and excluding sources with IDs above 1,000,000 because they sit in bright-star masked regions and have unreliable photometric redshifts. The authors also note that parallel slitless programs such as SAPPHIRES and POPPIES overlap part of the COSMOS-3D field, and that future multi-filter surveys can further reduce the systematic uncertainty from single-line identification.

The authors report a 94% consistency rate among the 3,798 sources cross-matched with published ground-based spectroscopic redshifts; among 726 sources also detected in other slitless filters, 41 redshifts were revised and four were found to be spurious. The large language model agrees with the final human-validated redshift in about 90% of cases, with the most common error being Paα misclassified as Hα. The authors also note that single-line sources, multi-peaked photometric redshifts, and sources overlapping along the dispersion direction still produce ambiguous assignments; such sources are flagged as contaminated and capped at rank 2 or lower.