An analysis of COSMOS-Web galaxies finds that the link between stellar mass and star formation is slightly steeper than a straight one-to-one rise for lower-mass galaxies, then flattens above a turnover that changes with redshift. Researchers fitted the low-mass slope at γ = 1.215 ± 0.028. When they used this measured star-forming main sequence to evolve the galaxy stellar mass function—the distribution of galaxies across stellar masses—the model followed the observed progressive flattening at the low-mass end substantially better than the literature prescriptions they tested.
The work is a preprint: its front matter identifies it as an Astronomy & Astrophysics manuscript, arXiv:2608.25671v1, dated 26 August 2026. Its central question was whether a mass-complete star-forming main sequence measured out to redshift about 8 was consistent with the observed evolution of the galaxy stellar mass function.
From a very large catalog to a cleaner sample
The starting point was the COSMOS-Web v1.1 master catalog, which contained 784,016 items. The mass-complete parent sample narrowed that to 268,276 sources: 251,717 were classified as star-forming and 16,559 as quiescent. The researchers removed 255 ALMA-identified starbursts before stacking, leaving 251,462 star-forming galaxies in the final main-sequence sample.
To estimate star formation across those galaxies, the team stacked their map cutouts in Herschel and JCMT imaging. It corrected far-infrared blending using forward-modeled simulated maps, then combined infrared luminosity with unobscured ultraviolet emission to derive total star-formation rates.
Those binned star-formation-rate and stellar-mass measurements were fitted with a smoothly broken relation, a curve that permits one behavior at low mass and a flattening at higher mass. The fit used a Gaussian likelihood and an affine-invariant ensemble MCMC with flat bounded priors, and treated the low-mass slope as independent of redshift. The resulting slope, 1.215 ± 0.028, is mildly super-linear: the fitted star-formation rate rises a little faster than stellar mass in that regime, before the relation bends over.
A test of galaxy buildup
The researchers then used a continuity-equation model to evolve the mass distribution from redshift about 8. The measured main sequence supplied in-situ growth, the quiescent fraction reduced population-averaged growth, stellar-mass return reduced the mass retained by galaxies, and mergers were included together with removal of the donor galaxies.
Quiescent fractions were calculated in redshift intervals and 0.5-dex stellar-mass bins using NUV-r-J criteria, without adding a separate specific-star-formation-rate threshold. The fitted fraction rose strongly with stellar mass at low redshift, while its maximum was lower at higher redshift.
The comparison was made against observed COSMOS-Web galaxy stellar mass functions and against versions evolved with literature main-sequence prescriptions. The measured relation reproduced the observed progressive flattening of the low-mass end substantially better than those alternatives.
The analysis also showed why main-sequence normalization cannot be treated as a minor detail. Small differences in that normalization accumulated over several Gyr and grew into large discrepancies between the evolved mass functions.
Mergers had a more limited effect in the modeled low-mass regime. Including them changed low-mass number densities by at most about 0.2 dex by redshift about 0.65, without substantially changing the slope.
Where the model still falls short
The agreement was not uniform across the full mass range. The evolved mass function still showed some high-mass discrepancies; possible contributors included extrapolating the merger prescription beyond its calibrated range at redshifts above 4 and residual starburst contamination.
That second caveat is tied to the way the sample was cleaned. The ALMA data used to remove starbursts were heterogeneous and did not define a luminosity-complete selection, so some starbursts may have gone unidentified and remained in the stacking sample. The 255 removed objects therefore mark identified cases, not a complete census of all starbursts.
Taken together, the preprint offers a consistency result for the selected COSMOS-Web population: under its continuity-equation assumptions, the measured main sequence gives a substantially better account of the observed low-mass mass-function trend than the comparison prescriptions, while the high-mass mismatch leaves room for the model and sample assumptions to matter.
Machine-readable tables are available through the project GitHub repository, with supplementary figures and diagnostic plots in the associated Zenodo record. The research was funded in whole or in part by the National Science Centre, Poland, including grant 2023/50/E/ST9/00383; K. Lisiecki also acknowledged PRELUDIUM grant UMO-2023/49/N/ST9/00746.
Paper data and sources
Original title: Stellar mass growth in COSMOS-Web: a mass-complete main sequence to z ~ 8 and its consistency with GSMF evolution
Authors: M. P. Koprowski, K. Lisiecki, P. Sawant, J. V. Wijesekera
Journal/Repository: arXiv
Status: Preprint, not yet peer-reviewed
First online: 2026-08-26
DOI: Not available
Original paper · Full text