Reconstructing galaxy histories offers clues to the scarcity of the earliest galaxies

The earliest galaxies cannot yet be studied simply by pointing a telescope far enough back in time. In this preprint, Laure Ciesla and colleagues instead reconstruct the histories of galaxies seen somewhat later, asking how much stellar mass they had already assembled and how bright they might previously have been.

Their results suggest that star formation became less common toward earlier times, with a sharper decline beyond redshift 17. But these are model-based reconstructions, not new detections of galaxies at those distances.

Reading a galaxy’s past from its light

Redshift, written as z, measures how much cosmic expansion has stretched light during its journey to us. Higher redshift means an earlier cosmic epoch. The paper places redshift 15 within the universe’s first 270 million years.

A galaxy’s light contains clues about its stellar populations, but those clues do not reveal a unique history. Different combinations of stellar ages and star formation rates can produce similar measurements. Young, luminous stars can also make an older population difficult to distinguish.

The researchers developed a model that allows star formation to rise and fall in bursts rather than follow a smooth curve. Its mathematical basis is an Ornstein–Uhlenbeck process: random fluctuations retain some memory of their recent state but tend back toward an underlying trend. Here, the strength of those fluctuations can change over time.

Implemented in the CIGALE galaxy-modeling software, the method compares predicted light across different wavelength bands with observations. It is a statistical description of burstiness, not a simulation identifying the physical cause of each burst.

Testing the reconstruction

The team first fitted simulated observations from SPHINX20, where the galaxies’ actual histories were available for comparison. The method recovered stellar masses and recent star formation rates reasonably well, and reproduced the population’s cumulative buildup of stars.

Recovering the earliest stages was harder. At redshift 8, the model compressed a broad range of true formation times into a narrower interval. Allowing earlier formation in the model introduced spurious very old galaxies rather than resolving the problem. At redshift 9, formation estimates still showed biases, although the oldest simulated galaxies were recovered better. Results were stronger at redshift 10.

These tests support population-level reconstruction more clearly than precise biographies of individual galaxies. Individual formation estimates remained uncertain by about one redshift unit.

How early had the stars assembled?

The researchers applied the model to 400 galaxies selected at redshifts between 9 and 15 in the UNICORN catalog of the GOODS-South field. Most distances were estimated from light measured through different filters, rather than confirmed by spectroscopy, which separates light into a detailed spectrum.

To define an early assembly milestone, they used the point when a galaxy had formed 20% of its stellar mass at the time it was observed. This is not the moment its first star appeared.

Among 218 galaxies observed between redshifts 10 and 15, only 28 had reached that milestone by redshift 15. For those 28, the median milestone occurred at redshift 16.9. Changing the threshold to 50% shifted the median to redshift 14.5—a substantial reminder that a galaxy’s reported formation time depends on its definition.

Across the full sample, 56% had formed some stars by redshift 15, while 24% had assembled at least ten million solar masses in stars. Thus, failing the 20% threshold does not mean a galaxy did not exist.

Looking farther back

Of the 28 early-assembling galaxies, 11 had reached the 20% milestone by redshift 17. Only one had done so by redshift 20.

The team reconstructed their ultraviolet light at those epochs, assuming no dust attenuation, and estimated the survey areas needed to encounter such sources. At redshift 20, the single qualifying reconstruction suggested roughly one object per 200 square arcminutes at the modeled brightness. The authors describe this as tentative, and possibly a lower limit on the area required.

Together, the estimates favor a relatively slow decline in star formation from redshift 15 to 17, followed by a faster decline toward redshift 20.

The central limitation is selection: these are ancestors of galaxies bright enough to enter a later survey. Earlier galaxies whose descendants remained too faint are missing. Model assumptions and burst-driven brightness changes add uncertainty, especially when the earliest estimate rests on one object. The study offers an indirect constraint on the observational frontier, not proof of a sharp boundary where galaxies began.